diff --git a/build/make/iosbuild.sh b/build/make/iosbuild.sh
index fa63a4a..fb91b87 100755
--- a/build/make/iosbuild.sh
+++ b/build/make/iosbuild.sh
@@ -43,7 +43,7 @@
   mkdir "${target}"
   cd "${target}"
   eval "../../${LIBVPX_SOURCE_DIR}/configure" --target="${target}" \
-      --disable-docs ${devnull}
+      --disable-docs ${EXTRA_CONFIGURE_ARGS} ${devnull}
   export DIST_DIR
   eval make -j ${MAKE_JOBS} dist ${devnull}
   cd "${old_pwd}"
@@ -62,7 +62,7 @@
       echo "__ARM_ARCH_6__"
       ;;
     armv7-*)
-      echo "__ARM_ARCH_7__"
+      echo "__ARM_ARCH_7A__"
       ;;
     armv7s-*)
       echo "__ARM_ARCH_7S__"
@@ -206,6 +206,10 @@
 # Parse the command line.
 while [ -n "$1" ]; do
   case "$1" in
+    --extra-configure-args)
+      EXTRA_CONFIGURE_ARGS="$2"
+      shift
+      ;;
     --help)
       iosbuild_usage
       exit
@@ -235,6 +239,7 @@
 cat << EOF
   BUILD_ROOT=${BUILD_ROOT}
   DIST_DIR=${DIST_DIR}
+  EXTRA_CONFIGURE_ARGS=${EXTRA_CONFIGURE_ARGS}
   FRAMEWORK_DIR=${FRAMEWORK_DIR}
   HEADER_DIR=${HEADER_DIR}
   MAKE_JOBS=${MAKE_JOBS}
diff --git a/build/make/rtcd.pl b/build/make/rtcd.pl
index 5b0cefa..0872414 100755
--- a/build/make/rtcd.pl
+++ b/build/make/rtcd.pl
@@ -49,7 +49,7 @@
 
 my %config = ();
 while (<CONFIG_FILE>) {
-  next if !/^CONFIG_/;
+  next if !/^(?:CONFIG_|HAVE_)/;
   chomp;
   my @pair = split /=/;
   $config{$pair[0]} = $pair[1];
@@ -209,14 +209,16 @@
 #define RTCD_EXTERN extern
 #endif
 
+EOF
+
+process_forward_decls();
+print <<EOF;
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 EOF
-
-process_forward_decls();
-print "\n";
 declare_function_pointers("c", @ALL_ARCHS);
 
 print <<EOF;
@@ -388,7 +390,7 @@
   @REQUIRES = filter(keys %required ? keys %required : qw/media/);
   &require(@REQUIRES);
   arm;
-} elsif ($opts{arch} eq 'armv8') {
+} elsif ($opts{arch} eq 'armv8' || $opts{arch} eq 'arm64' ) {
   @ALL_ARCHS = filter(qw/neon/);
   arm;
 } else {
diff --git a/configure b/configure
index 32b70f1..7b9c211 100755
--- a/configure
+++ b/configure
@@ -281,6 +281,7 @@
     spatial_svc
     vp9_temporal_denoising
     fp_mb_stats
+    emulate_hardware_highbitdepth
 "
 CONFIG_LIST="
     external_build
diff --git a/examples.mk b/examples.mk
index bd38c41..fd67a44 100644
--- a/examples.mk
+++ b/examples.mk
@@ -114,7 +114,7 @@
 vpx_temporal_svc_encoder.SRCS        += video_writer.h video_writer.c
 vpx_temporal_svc_encoder.GUID        = B18C08F2-A439-4502-A78E-849BE3D60947
 vpx_temporal_svc_encoder.DESCRIPTION = Temporal SVC Encoder
-EXAMPLES-$(CONFIG_VP8_DECODER)     += simple_decoder.c
+EXAMPLES-$(CONFIG_DECODERS)        += simple_decoder.c
 simple_decoder.GUID                 = D3BBF1E9-2427-450D-BBFF-B2843C1D44CC
 simple_decoder.SRCS                += ivfdec.h ivfdec.c
 simple_decoder.SRCS                += tools_common.h tools_common.c
@@ -123,7 +123,7 @@
 simple_decoder.SRCS                += vpx_ports/mem_ops.h
 simple_decoder.SRCS                += vpx_ports/mem_ops_aligned.h
 simple_decoder.DESCRIPTION          = Simplified decoder loop
-EXAMPLES-$(CONFIG_VP8_DECODER)     += postproc.c
+EXAMPLES-$(CONFIG_DECODERS)        += postproc.c
 postproc.SRCS                      += ivfdec.h ivfdec.c
 postproc.SRCS                      += tools_common.h tools_common.c
 postproc.SRCS                      += video_common.h
@@ -132,7 +132,7 @@
 postproc.SRCS                      += vpx_ports/mem_ops_aligned.h
 postproc.GUID                       = 65E33355-F35E-4088-884D-3FD4905881D7
 postproc.DESCRIPTION                = Decoder postprocessor control
-EXAMPLES-$(CONFIG_VP8_DECODER)     += decode_to_md5.c
+EXAMPLES-$(CONFIG_DECODERS)        += decode_to_md5.c
 decode_to_md5.SRCS                 += md5_utils.h md5_utils.c
 decode_to_md5.SRCS                 += ivfdec.h ivfdec.c
 decode_to_md5.SRCS                 += tools_common.h tools_common.c
@@ -142,29 +142,34 @@
 decode_to_md5.SRCS                 += vpx_ports/mem_ops_aligned.h
 decode_to_md5.GUID                  = 59120B9B-2735-4BFE-B022-146CA340FE42
 decode_to_md5.DESCRIPTION           = Frame by frame MD5 checksum
-EXAMPLES-$(CONFIG_VP8_ENCODER)  += simple_encoder.c
+EXAMPLES-$(CONFIG_ENCODERS)     += simple_encoder.c
 simple_encoder.SRCS             += ivfenc.h ivfenc.c
 simple_encoder.SRCS             += tools_common.h tools_common.c
 simple_encoder.SRCS             += video_common.h
 simple_encoder.SRCS             += video_writer.h video_writer.c
 simple_encoder.GUID              = 4607D299-8A71-4D2C-9B1D-071899B6FBFD
 simple_encoder.DESCRIPTION       = Simplified encoder loop
-EXAMPLES-$(CONFIG_VP8_ENCODER)  += twopass_encoder.c
+EXAMPLES-$(CONFIG_VP9_ENCODER)  += vp9_lossless_encoder.c
+vp9_lossless_encoder.SRCS       += ivfenc.h ivfenc.c
+vp9_lossless_encoder.SRCS       += tools_common.h tools_common.c
+vp9_lossless_encoder.SRCS       += video_common.h
+vp9_lossless_encoder.SRCS       += video_writer.h video_writer.c
+vp9_lossless_encoder.GUID        = B63C7C88-5348-46DC-A5A6-CC151EF93366
+vp9_lossless_encoder.DESCRIPTION = Simplified lossless VP9 encoder
+EXAMPLES-$(CONFIG_ENCODERS)     += twopass_encoder.c
 twopass_encoder.SRCS            += ivfenc.h ivfenc.c
 twopass_encoder.SRCS            += tools_common.h tools_common.c
 twopass_encoder.SRCS            += video_common.h
 twopass_encoder.SRCS            += video_writer.h video_writer.c
 twopass_encoder.GUID             = 73494FA6-4AF9-4763-8FBB-265C92402FD8
 twopass_encoder.DESCRIPTION      = Two-pass encoder loop
-ifeq ($(CONFIG_DECODERS),yes)
-EXAMPLES-$(CONFIG_VP8_ENCODER)  += decode_with_drops.c
+EXAMPLES-$(CONFIG_DECODERS)     += decode_with_drops.c
 decode_with_drops.SRCS          += ivfdec.h ivfdec.c
 decode_with_drops.SRCS          += tools_common.h tools_common.c
 decode_with_drops.SRCS          += video_common.h
 decode_with_drops.SRCS          += video_reader.h video_reader.c
 decode_with_drops.SRCS          += vpx_ports/mem_ops.h
 decode_with_drops.SRCS          += vpx_ports/mem_ops_aligned.h
-endif
 decode_with_drops.GUID           = CE5C53C4-8DDA-438A-86ED-0DDD3CDB8D26
 decode_with_drops.DESCRIPTION    = Drops frames while decoding
 EXAMPLES-$(CONFIG_ENCODERS)        += set_maps.c
diff --git a/examples/decode_to_md5.c b/examples/decode_to_md5.c
index 1c56303..fbc0f4a 100644
--- a/examples/decode_to_md5.c
+++ b/examples/decode_to_md5.c
@@ -33,8 +33,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
-
 #include "vpx/vp8dx.h"
 #include "vpx/vpx_decoder.h"
 
diff --git a/examples/decode_with_drops.c b/examples/decode_with_drops.c
index a20fdac..9423e38 100644
--- a/examples/decode_with_drops.c
+++ b/examples/decode_with_drops.c
@@ -56,8 +56,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
-
 #include "vpx/vp8dx.h"
 #include "vpx/vpx_decoder.h"
 
diff --git a/examples/postproc.c b/examples/postproc.c
index 59c50b1..c74347c 100644
--- a/examples/postproc.c
+++ b/examples/postproc.c
@@ -43,8 +43,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
-
 #include "vpx/vp8dx.h"
 #include "vpx/vpx_decoder.h"
 
diff --git a/examples/set_maps.c b/examples/set_maps.c
index 2ee5bca..851adc4 100644
--- a/examples/set_maps.c
+++ b/examples/set_maps.c
@@ -47,7 +47,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vp8cx.h"
 #include "vpx/vpx_encoder.h"
 
diff --git a/examples/simple_decoder.c b/examples/simple_decoder.c
index 3318758..3f7d6aa 100644
--- a/examples/simple_decoder.c
+++ b/examples/simple_decoder.c
@@ -29,9 +29,7 @@
 // -----------------
 // For decoders, you only have to include `vpx_decoder.h` and then any
 // header files for the specific codecs you use. In this case, we're using
-// vp8. The `VPX_CODEC_DISABLE_COMPAT` macro can be defined to ensure
-// strict compliance with the latest SDK by disabling some backwards
-// compatibility features. Defining this macro is encouraged.
+// vp8.
 //
 // Initializing The Codec
 // ----------------------
@@ -81,8 +79,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
-
 #include "vpx/vp8dx.h"
 #include "vpx/vpx_decoder.h"
 
diff --git a/examples/simple_encoder.c b/examples/simple_encoder.c
index 30bb73a..f20c246 100644
--- a/examples/simple_encoder.c
+++ b/examples/simple_encoder.c
@@ -28,9 +28,7 @@
 // -----------------
 // For encoders, you only have to include `vpx_encoder.h` and then any
 // header files for the specific codecs you use. In this case, we're using
-// vp8. The `VPX_CODEC_DISABLE_COMPAT` macro can be defined to ensure
-// strict compliance with the latest SDK by disabling some backwards
-// compatibility features. Defining this macro is encouraged.
+// vp8.
 //
 // Getting The Default Configuration
 // ---------------------------------
@@ -101,7 +99,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vpx_encoder.h"
 
 #include "./tools_common.h"
diff --git a/examples/twopass_encoder.c b/examples/twopass_encoder.c
index 76d5a28..653ae94 100644
--- a/examples/twopass_encoder.c
+++ b/examples/twopass_encoder.c
@@ -28,9 +28,8 @@
 // Encoding A Frame
 // ----------------
 // Encoding a frame in two pass mode is identical to the simple encoder
-// example, except the deadline is set to VPX_DL_BEST_QUALITY to get the
-// best quality possible. VPX_DL_GOOD_QUALITY could also be used.
-//
+// example. To increase the quality while sacrificing encoding speed,
+// VPX_DL_BEST_QUALITY can be used in place of VPX_DL_GOOD_QUALITY.
 //
 // Processing Statistics Packets
 // -----------------------------
@@ -52,7 +51,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vpx_encoder.h"
 
 #include "./tools_common.h"
@@ -142,13 +140,13 @@
   // Calculate frame statistics.
   while (vpx_img_read(raw, infile)) {
     ++frame_count;
-    get_frame_stats(&codec, raw, frame_count, 1, 0, VPX_DL_BEST_QUALITY,
+    get_frame_stats(&codec, raw, frame_count, 1, 0, VPX_DL_GOOD_QUALITY,
                     &stats);
   }
 
   // Flush encoder.
   while (get_frame_stats(&codec, NULL, frame_count, 1, 0,
-                         VPX_DL_BEST_QUALITY, &stats)) {}
+                         VPX_DL_GOOD_QUALITY, &stats)) {}
 
   printf("Pass 0 complete. Processed %d frames.\n", frame_count);
   if (vpx_codec_destroy(&codec))
@@ -182,11 +180,11 @@
   // Encode frames.
   while (vpx_img_read(raw, infile)) {
     ++frame_count;
-    encode_frame(&codec, raw, frame_count, 1, 0, VPX_DL_BEST_QUALITY, writer);
+    encode_frame(&codec, raw, frame_count, 1, 0, VPX_DL_GOOD_QUALITY, writer);
   }
 
   // Flush encoder.
-  while (encode_frame(&codec, NULL, -1, 1, 0, VPX_DL_BEST_QUALITY, writer)) {}
+  while (encode_frame(&codec, NULL, -1, 1, 0, VPX_DL_GOOD_QUALITY, writer)) {}
 
   printf("\n");
 
diff --git a/examples/vp8_multi_resolution_encoder.c b/examples/vp8_multi_resolution_encoder.c
index 7c050fa..9f50dc7 100644
--- a/examples/vp8_multi_resolution_encoder.c
+++ b/examples/vp8_multi_resolution_encoder.c
@@ -24,7 +24,6 @@
 #include "third_party/libyuv/include/libyuv/scale.h"
 #include "third_party/libyuv/include/libyuv/cpu_id.h"
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vpx_encoder.h"
 #include "vpx/vp8cx.h"
 
diff --git a/examples/vp8cx_set_ref.c b/examples/vp8cx_set_ref.c
index 5f3f0a3..b0961a2 100644
--- a/examples/vp8cx_set_ref.c
+++ b/examples/vp8cx_set_ref.c
@@ -50,7 +50,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vp8cx.h"
 #include "vpx/vpx_encoder.h"
 
diff --git a/examples/vp9_lossless_encoder.c b/examples/vp9_lossless_encoder.c
new file mode 100644
index 0000000..3fcda0c
--- /dev/null
+++ b/examples/vp9_lossless_encoder.c
@@ -0,0 +1,144 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "vpx/vpx_encoder.h"
+#include "vpx/vp8cx.h"
+
+#include "./tools_common.h"
+#include "./video_writer.h"
+
+static const char *exec_name;
+
+void usage_exit() {
+  fprintf(stderr, "vp9_lossless_encoder: Example demonstrating VP9 lossless "
+                  "encoding feature. Supports raw input only.\n");
+  fprintf(stderr, "Usage: %s <width> <height> <infile> <outfile>\n", exec_name);
+  exit(EXIT_FAILURE);
+}
+
+static int encode_frame(vpx_codec_ctx_t *codec,
+                        vpx_image_t *img,
+                        int frame_index,
+                        int flags,
+                        VpxVideoWriter *writer) {
+  int got_pkts = 0;
+  vpx_codec_iter_t iter = NULL;
+  const vpx_codec_cx_pkt_t *pkt = NULL;
+  const vpx_codec_err_t res = vpx_codec_encode(codec, img, frame_index, 1,
+                                               flags, VPX_DL_GOOD_QUALITY);
+  if (res != VPX_CODEC_OK)
+    die_codec(codec, "Failed to encode frame");
+
+  while ((pkt = vpx_codec_get_cx_data(codec, &iter)) != NULL) {
+    got_pkts = 1;
+
+    if (pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
+      const int keyframe = (pkt->data.frame.flags & VPX_FRAME_IS_KEY) != 0;
+      if (!vpx_video_writer_write_frame(writer,
+                                        pkt->data.frame.buf,
+                                        pkt->data.frame.sz,
+                                        pkt->data.frame.pts)) {
+        die_codec(codec, "Failed to write compressed frame");
+      }
+      printf(keyframe ? "K" : ".");
+      fflush(stdout);
+    }
+  }
+
+  return got_pkts;
+}
+
+int main(int argc, char **argv) {
+  FILE *infile = NULL;
+  vpx_codec_ctx_t codec;
+  vpx_codec_enc_cfg_t cfg;
+  int frame_count = 0;
+  vpx_image_t raw;
+  vpx_codec_err_t res;
+  VpxVideoInfo info = {0};
+  VpxVideoWriter *writer = NULL;
+  const VpxInterface *encoder = NULL;
+  const int fps = 30;
+
+  exec_name = argv[0];
+
+  if (argc < 5)
+    die("Invalid number of arguments");
+
+  encoder = get_vpx_encoder_by_name("vp9");
+  if (!encoder)
+     die("Unsupported codec.");
+
+  info.codec_fourcc = encoder->fourcc;
+  info.frame_width = strtol(argv[1], NULL, 0);
+  info.frame_height = strtol(argv[2], NULL, 0);
+  info.time_base.numerator = 1;
+  info.time_base.denominator = fps;
+
+  if (info.frame_width <= 0 ||
+      info.frame_height <= 0 ||
+      (info.frame_width % 2) != 0 ||
+      (info.frame_height % 2) != 0) {
+    die("Invalid frame size: %dx%d", info.frame_width, info.frame_height);
+  }
+
+  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, info.frame_width,
+                                             info.frame_height, 1)) {
+    die("Failed to allocate image.");
+  }
+
+  printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
+
+  res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
+  if (res)
+    die_codec(&codec, "Failed to get default codec config.");
+
+  cfg.g_w = info.frame_width;
+  cfg.g_h = info.frame_height;
+  cfg.g_timebase.num = info.time_base.numerator;
+  cfg.g_timebase.den = info.time_base.denominator;
+
+  writer = vpx_video_writer_open(argv[4], kContainerIVF, &info);
+  if (!writer)
+    die("Failed to open %s for writing.", argv[4]);
+
+  if (!(infile = fopen(argv[3], "rb")))
+    die("Failed to open %s for reading.", argv[3]);
+
+  if (vpx_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0))
+    die_codec(&codec, "Failed to initialize encoder");
+
+  if (vpx_codec_control_(&codec, VP9E_SET_LOSSLESS, 1))
+    die_codec(&codec, "Failed to use lossless mode");
+
+  // Encode frames.
+  while (vpx_img_read(&raw, infile)) {
+    encode_frame(&codec, &raw, frame_count++, 0, writer);
+  }
+
+  // Flush encoder.
+  while (encode_frame(&codec, NULL, -1, 0, writer)) {}
+
+  printf("\n");
+  fclose(infile);
+  printf("Processed %d frames.\n", frame_count);
+
+  vpx_img_free(&raw);
+  if (vpx_codec_destroy(&codec))
+    die_codec(&codec, "Failed to destroy codec.");
+
+  vpx_video_writer_close(writer);
+
+  return EXIT_SUCCESS;
+}
diff --git a/examples/vp9_spatial_svc_encoder.c b/examples/vp9_spatial_svc_encoder.c
index 81d3800..9cd716b 100644
--- a/examples/vp9_spatial_svc_encoder.c
+++ b/examples/vp9_spatial_svc_encoder.c
@@ -46,10 +46,6 @@
     ARG_DEF("k", "kf-dist", 1, "number of frames between keyframes");
 static const arg_def_t scale_factors_arg =
     ARG_DEF("r", "scale-factors", 1, "scale factors (lowest to highest layer)");
-static const arg_def_t quantizers_arg =
-    ARG_DEF("q", "quantizers", 1, "quantizers for non key frames, also will "
-            "be applied to key frames if -qn is not specified (lowest to "
-            "highest layer)");
 static const arg_def_t passes_arg =
     ARG_DEF("p", "passes", 1, "Number of passes (1/2)");
 static const arg_def_t pass_arg =
@@ -68,10 +64,9 @@
 static const arg_def_t *svc_args[] = {
   &frames_arg,        &width_arg,         &height_arg,
   &timebase_arg,      &bitrate_arg,       &skip_frames_arg, &spatial_layers_arg,
-  &kf_dist_arg,       &scale_factors_arg, &quantizers_arg,  &passes_arg,
-  &pass_arg,          &fpf_name_arg,      &min_q_arg,       &max_q_arg,
-  &min_bitrate_arg,   &max_bitrate_arg,   &temporal_layers_arg,
-  NULL
+  &kf_dist_arg,       &scale_factors_arg, &passes_arg,      &pass_arg,
+  &fpf_name_arg,      &min_q_arg,         &max_q_arg,       &min_bitrate_arg,
+  &max_bitrate_arg,   &temporal_layers_arg,                 NULL
 };
 
 static const uint32_t default_frames_to_skip = 0;
@@ -119,6 +114,7 @@
   const char *fpf_file_name = NULL;
   unsigned int min_bitrate = 0;
   unsigned int max_bitrate = 0;
+  char string_options[1024] = {0};
 
   // initialize SvcContext with parameters that will be passed to vpx_svc_init
   svc_ctx->log_level = SVC_LOG_DEBUG;
@@ -169,9 +165,8 @@
       enc_cfg->kf_min_dist = arg_parse_uint(&arg);
       enc_cfg->kf_max_dist = enc_cfg->kf_min_dist;
     } else if (arg_match(&arg, &scale_factors_arg, argi)) {
-      vpx_svc_set_scale_factors(svc_ctx, arg.val);
-    } else if (arg_match(&arg, &quantizers_arg, argi)) {
-      vpx_svc_set_quantizers(svc_ctx, arg.val);
+      snprintf(string_options, 1024, "%s scale-factors=%s",
+               string_options, arg.val);
     } else if (arg_match(&arg, &passes_arg, argi)) {
       passes = arg_parse_uint(&arg);
       if (passes < 1 || passes > 2) {
@@ -185,9 +180,11 @@
     } else if (arg_match(&arg, &fpf_name_arg, argi)) {
       fpf_file_name = arg.val;
     } else if (arg_match(&arg, &min_q_arg, argi)) {
-      enc_cfg->rc_min_quantizer = arg_parse_uint(&arg);
+      snprintf(string_options, 1024, "%s min-quantizers=%s",
+               string_options, arg.val);
     } else if (arg_match(&arg, &max_q_arg, argi)) {
-      enc_cfg->rc_max_quantizer = arg_parse_uint(&arg);
+      snprintf(string_options, 1024, "%s max-quantizers=%s",
+               string_options, arg.val);
     } else if (arg_match(&arg, &min_bitrate_arg, argi)) {
       min_bitrate = arg_parse_uint(&arg);
     } else if (arg_match(&arg, &max_bitrate_arg, argi)) {
@@ -197,6 +194,10 @@
     }
   }
 
+  // There will be a space in front of the string options
+  if (strlen(string_options) > 0)
+    vpx_svc_set_options(svc_ctx, string_options + 1);
+
   if (passes == 0 || passes == 1) {
     if (pass) {
       fprintf(stderr, "pass is ignored since there's only one pass\n");
@@ -282,7 +283,7 @@
   int frame_duration = 1; /* 1 timebase tick per frame */
   FILE *infile = NULL;
   int end_of_stream = 0;
-  int frame_size;
+  int frames_received = 0;
 
   memset(&svc_ctx, 0, sizeof(svc_ctx));
   svc_ctx.log_print = 1;
@@ -304,12 +305,6 @@
   info.codec_fourcc = VP9_FOURCC;
   info.time_base.numerator = enc_cfg.g_timebase.num;
   info.time_base.denominator = enc_cfg.g_timebase.den;
-  if (vpx_svc_get_layer_resolution(&svc_ctx, svc_ctx.spatial_layers - 1,
-                                   (unsigned int *)&info.frame_width,
-                                   (unsigned int *)&info.frame_height) !=
-      VPX_CODEC_OK) {
-    die("Failed to get output resolution");
-  }
 
   if (!(app_input.passes == 2 && app_input.pass == 1)) {
     // We don't save the bitstream for the 1st pass on two pass rate control
@@ -325,6 +320,8 @@
 
   // Encode frames
   while (!end_of_stream) {
+    vpx_codec_iter_t iter = NULL;
+    const vpx_codec_cx_pkt_t *cx_pkt;
     if (frame_cnt >= app_input.frames_to_code || !vpx_img_read(&raw, infile)) {
       // We need one extra vpx_svc_encode call at end of stream to flush
       // encoder and get remaining data
@@ -337,18 +334,34 @@
     if (res != VPX_CODEC_OK) {
       die_codec(&codec, "Failed to encode frame");
     }
-    if (!(app_input.passes == 2 && app_input.pass == 1)) {
-      while ((frame_size = vpx_svc_get_frame_size(&svc_ctx)) > 0) {
-        vpx_video_writer_write_frame(writer,
-                                     vpx_svc_get_buffer(&svc_ctx),
-                                     frame_size, pts);
+
+    while ((cx_pkt = vpx_codec_get_cx_data(&codec, &iter)) != NULL) {
+      switch (cx_pkt->kind) {
+        case VPX_CODEC_CX_FRAME_PKT: {
+          if (cx_pkt->data.frame.sz > 0)
+            vpx_video_writer_write_frame(writer,
+                                         cx_pkt->data.frame.buf,
+                                         cx_pkt->data.frame.sz,
+                                         cx_pkt->data.frame.pts);
+
+          printf("SVC frame: %d, kf: %d, size: %d, pts: %d\n", frames_received,
+                 !!(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY),
+                 (int)cx_pkt->data.frame.sz, (int)cx_pkt->data.frame.pts);
+          ++frames_received;
+          break;
+        }
+        case VPX_CODEC_STATS_PKT: {
+          stats_write(&app_input.rc_stats,
+                      cx_pkt->data.twopass_stats.buf,
+                      cx_pkt->data.twopass_stats.sz);
+          break;
+        }
+        default: {
+          break;
+        }
       }
     }
-    if (vpx_svc_get_rc_stats_buffer_size(&svc_ctx) > 0) {
-      stats_write(&app_input.rc_stats,
-                  vpx_svc_get_rc_stats_buffer(&svc_ctx),
-                  vpx_svc_get_rc_stats_buffer_size(&svc_ctx));
-    }
+
     if (!end_of_stream) {
       ++frame_cnt;
       pts += frame_duration;
diff --git a/examples/vpx_temporal_svc_encoder.c b/examples/vpx_temporal_svc_encoder.c
index 5eac92c..1674804 100644
--- a/examples/vpx_temporal_svc_encoder.c
+++ b/examples/vpx_temporal_svc_encoder.c
@@ -18,7 +18,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "./vpx_config.h"
 #include "vpx_ports/vpx_timer.h"
 #include "vpx/vp8cx.h"
@@ -587,7 +586,7 @@
       vpx_codec_control(&codec, VP8E_SET_CPUUSED, speed);
       vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 3);
       vpx_codec_control(&codec, VP9E_SET_FRAME_PERIODIC_BOOST, 0);
-      vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, 0);
+      vpx_codec_control(&codec, VP9E_SET_NOISE_SENSITIVITY, 0);
       if (vpx_codec_control(&codec, VP9E_SET_SVC, 1)) {
         die_codec(&codec, "Failed to set SVC");
     }
diff --git a/libs.mk b/libs.mk
index c7c2748..f9f2d80 100644
--- a/libs.mk
+++ b/libs.mk
@@ -531,7 +531,6 @@
 	@echo "    [CREATE] $@"
 	@rm -f $@
 	@echo "INPUT += $^" >> $@
-	@echo "PREDEFINED = VPX_CODEC_DISABLE_COMPAT" >> $@
 	@echo "INCLUDE_PATH += ." >> $@;
 	@echo "ENABLED_SECTIONS += $(sort $(CODEC_DOC_SECTIONS))" >> $@
 
diff --git a/test/convolve_test.cc b/test/convolve_test.cc
index 1724db3..de947aa 100644
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -21,6 +21,9 @@
 #include "vpx_ports/mem.h"
 
 namespace {
+
+static const unsigned int kMaxDimension = 64;
+
 typedef void (*ConvolveFunc)(const uint8_t *src, ptrdiff_t src_stride,
                              uint8_t *dst, ptrdiff_t dst_stride,
                              const int16_t *filter_x, int filter_x_stride,
@@ -30,9 +33,10 @@
 struct ConvolveFunctions {
   ConvolveFunctions(ConvolveFunc h8, ConvolveFunc h8_avg,
                     ConvolveFunc v8, ConvolveFunc v8_avg,
-                    ConvolveFunc hv8, ConvolveFunc hv8_avg)
+                    ConvolveFunc hv8, ConvolveFunc hv8_avg,
+                    int bd)
       : h8_(h8), v8_(v8), hv8_(hv8), h8_avg_(h8_avg), v8_avg_(v8_avg),
-        hv8_avg_(hv8_avg) {}
+        hv8_avg_(hv8_avg), use_high_bd_(bd) {}
 
   ConvolveFunc h8_;
   ConvolveFunc v8_;
@@ -40,6 +44,7 @@
   ConvolveFunc h8_avg_;
   ConvolveFunc v8_avg_;
   ConvolveFunc hv8_avg_;
+  int use_high_bd_;  // 0 if high bitdepth not used, else the actual bit depth.
 };
 
 typedef std::tr1::tuple<int, int, const ConvolveFunctions *> ConvolveParam;
@@ -68,71 +73,66 @@
   const int kInterp_Extend = 4;
   const unsigned int intermediate_height =
       (kInterp_Extend - 1) + output_height + kInterp_Extend;
+  unsigned int i, j;
 
-  /* Size of intermediate_buffer is max_intermediate_height * filter_max_width,
-   * where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
-   *                                 + kInterp_Extend
-   *                               = 3 + 16 + 4
-   *                               = 23
-   * and filter_max_width = 16
-   */
-  uint8_t intermediate_buffer[71 * 64];
+  // Size of intermediate_buffer is max_intermediate_height * filter_max_width,
+  // where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
+  //                                 + kInterp_Extend
+  //                               = 3 + 16 + 4
+  //                               = 23
+  // and filter_max_width          = 16
+  //
+  uint8_t intermediate_buffer[71 * kMaxDimension];
   const int intermediate_next_stride = 1 - intermediate_height * output_width;
 
   // Horizontal pass (src -> transposed intermediate).
-  {
-    uint8_t *output_ptr = intermediate_buffer;
-    const int src_next_row_stride = src_stride - output_width;
-    unsigned int i, j;
-    src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
-    for (i = 0; i < intermediate_height; ++i) {
-      for (j = 0; j < output_width; ++j) {
-        // Apply filter...
-        const int temp = (src_ptr[0] * HFilter[0]) +
-                         (src_ptr[1] * HFilter[1]) +
-                         (src_ptr[2] * HFilter[2]) +
-                         (src_ptr[3] * HFilter[3]) +
-                         (src_ptr[4] * HFilter[4]) +
-                         (src_ptr[5] * HFilter[5]) +
-                         (src_ptr[6] * HFilter[6]) +
-                         (src_ptr[7] * HFilter[7]) +
-                         (VP9_FILTER_WEIGHT >> 1);  // Rounding
+  uint8_t *output_ptr = intermediate_buffer;
+  const int src_next_row_stride = src_stride - output_width;
+  src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
+  for (i = 0; i < intermediate_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      // Apply filter...
+      const int temp = (src_ptr[0] * HFilter[0]) +
+          (src_ptr[1] * HFilter[1]) +
+          (src_ptr[2] * HFilter[2]) +
+          (src_ptr[3] * HFilter[3]) +
+          (src_ptr[4] * HFilter[4]) +
+          (src_ptr[5] * HFilter[5]) +
+          (src_ptr[6] * HFilter[6]) +
+          (src_ptr[7] * HFilter[7]) +
+          (VP9_FILTER_WEIGHT >> 1);  // Rounding
 
-        // Normalize back to 0-255...
-        *output_ptr = clip_pixel(temp >> VP9_FILTER_SHIFT);
-        ++src_ptr;
-        output_ptr += intermediate_height;
-      }
-      src_ptr += src_next_row_stride;
-      output_ptr += intermediate_next_stride;
+      // Normalize back to 0-255...
+      *output_ptr = clip_pixel(temp >> VP9_FILTER_SHIFT);
+      ++src_ptr;
+      output_ptr += intermediate_height;
     }
+    src_ptr += src_next_row_stride;
+    output_ptr += intermediate_next_stride;
   }
 
   // Vertical pass (transposed intermediate -> dst).
-  {
-    uint8_t *src_ptr = intermediate_buffer;
-    const int dst_next_row_stride = dst_stride - output_width;
-    unsigned int i, j;
-    for (i = 0; i < output_height; ++i) {
-      for (j = 0; j < output_width; ++j) {
-        // Apply filter...
-        const int temp = (src_ptr[0] * VFilter[0]) +
-                         (src_ptr[1] * VFilter[1]) +
-                         (src_ptr[2] * VFilter[2]) +
-                         (src_ptr[3] * VFilter[3]) +
-                         (src_ptr[4] * VFilter[4]) +
-                         (src_ptr[5] * VFilter[5]) +
-                         (src_ptr[6] * VFilter[6]) +
-                         (src_ptr[7] * VFilter[7]) +
-                         (VP9_FILTER_WEIGHT >> 1);  // Rounding
+  src_ptr = intermediate_buffer;
+  const int dst_next_row_stride = dst_stride - output_width;
+  for (i = 0; i < output_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      // Apply filter...
+      const int temp = (src_ptr[0] * VFilter[0]) +
+          (src_ptr[1] * VFilter[1]) +
+          (src_ptr[2] * VFilter[2]) +
+          (src_ptr[3] * VFilter[3]) +
+          (src_ptr[4] * VFilter[4]) +
+          (src_ptr[5] * VFilter[5]) +
+          (src_ptr[6] * VFilter[6]) +
+          (src_ptr[7] * VFilter[7]) +
+          (VP9_FILTER_WEIGHT >> 1);  // Rounding
 
-        // Normalize back to 0-255...
-        *dst_ptr++ = clip_pixel(temp >> VP9_FILTER_SHIFT);
-        src_ptr += intermediate_height;
-      }
-      src_ptr += intermediate_next_stride;
-      dst_ptr += dst_next_row_stride;
+      // Normalize back to 0-255...
+      *dst_ptr++ = clip_pixel(temp >> VP9_FILTER_SHIFT);
+      src_ptr += intermediate_height;
     }
+    src_ptr += intermediate_next_stride;
+    dst_ptr += dst_next_row_stride;
   }
 }
 
@@ -159,16 +159,137 @@
                                 unsigned int dst_stride,
                                 unsigned int output_width,
                                 unsigned int output_height) {
-  uint8_t tmp[64 * 64];
+  uint8_t tmp[kMaxDimension * kMaxDimension];
 
-  assert(output_width <= 64);
-  assert(output_height <= 64);
+  assert(output_width <= kMaxDimension);
+  assert(output_height <= kMaxDimension);
   filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp, 64,
                      output_width, output_height);
   block2d_average_c(tmp, 64, dst_ptr, dst_stride,
                     output_width, output_height);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void high_filter_block2d_8_c(const uint16_t *src_ptr,
+                             const unsigned int src_stride,
+                             const int16_t *HFilter,
+                             const int16_t *VFilter,
+                             uint16_t *dst_ptr,
+                             unsigned int dst_stride,
+                             unsigned int output_width,
+                             unsigned int output_height,
+                             int bd) {
+  // Between passes, we use an intermediate buffer whose height is extended to
+  // have enough horizontally filtered values as input for the vertical pass.
+  // This buffer is allocated to be big enough for the largest block type we
+  // support.
+  const int kInterp_Extend = 4;
+  const unsigned int intermediate_height =
+      (kInterp_Extend - 1) + output_height + kInterp_Extend;
+
+  /* Size of intermediate_buffer is max_intermediate_height * filter_max_width,
+   * where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
+   *                                 + kInterp_Extend
+   *                               = 3 + 16 + 4
+   *                               = 23
+   * and filter_max_width = 16
+   */
+  uint16_t intermediate_buffer[71 * kMaxDimension];
+  const int intermediate_next_stride = 1 - intermediate_height * output_width;
+
+  // Horizontal pass (src -> transposed intermediate).
+  {
+    uint16_t *output_ptr = intermediate_buffer;
+    const int src_next_row_stride = src_stride - output_width;
+    unsigned int i, j;
+    src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
+    for (i = 0; i < intermediate_height; ++i) {
+      for (j = 0; j < output_width; ++j) {
+        // Apply filter...
+        const int temp = (src_ptr[0] * HFilter[0]) +
+                         (src_ptr[1] * HFilter[1]) +
+                         (src_ptr[2] * HFilter[2]) +
+                         (src_ptr[3] * HFilter[3]) +
+                         (src_ptr[4] * HFilter[4]) +
+                         (src_ptr[5] * HFilter[5]) +
+                         (src_ptr[6] * HFilter[6]) +
+                         (src_ptr[7] * HFilter[7]) +
+                         (VP9_FILTER_WEIGHT >> 1);  // Rounding
+
+        // Normalize back to 0-255...
+        *output_ptr = clip_pixel_high(temp >> VP9_FILTER_SHIFT, bd);
+        ++src_ptr;
+        output_ptr += intermediate_height;
+      }
+      src_ptr += src_next_row_stride;
+      output_ptr += intermediate_next_stride;
+    }
+  }
+
+  // Vertical pass (transposed intermediate -> dst).
+  {
+    uint16_t *src_ptr = intermediate_buffer;
+    const int dst_next_row_stride = dst_stride - output_width;
+    unsigned int i, j;
+    for (i = 0; i < output_height; ++i) {
+      for (j = 0; j < output_width; ++j) {
+        // Apply filter...
+        const int temp = (src_ptr[0] * VFilter[0]) +
+                         (src_ptr[1] * VFilter[1]) +
+                         (src_ptr[2] * VFilter[2]) +
+                         (src_ptr[3] * VFilter[3]) +
+                         (src_ptr[4] * VFilter[4]) +
+                         (src_ptr[5] * VFilter[5]) +
+                         (src_ptr[6] * VFilter[6]) +
+                         (src_ptr[7] * VFilter[7]) +
+                         (VP9_FILTER_WEIGHT >> 1);  // Rounding
+
+        // Normalize back to 0-255...
+        *dst_ptr++ = clip_pixel_high(temp >> VP9_FILTER_SHIFT, bd);
+        src_ptr += intermediate_height;
+      }
+      src_ptr += intermediate_next_stride;
+      dst_ptr += dst_next_row_stride;
+    }
+  }
+}
+
+void high_block2d_average_c(uint16_t *src,
+                            unsigned int src_stride,
+                            uint16_t *output_ptr,
+                            unsigned int output_stride,
+                            unsigned int output_width,
+                            unsigned int output_height,
+                            int bd) {
+  unsigned int i, j;
+  for (i = 0; i < output_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
+    }
+    output_ptr += output_stride;
+  }
+}
+
+void high_filter_average_block2d_8_c(const uint16_t *src_ptr,
+                                     const unsigned int src_stride,
+                                     const int16_t *HFilter,
+                                     const int16_t *VFilter,
+                                     uint16_t *dst_ptr,
+                                     unsigned int dst_stride,
+                                     unsigned int output_width,
+                                     unsigned int output_height,
+                                     int bd) {
+  uint16_t tmp[kMaxDimension * kMaxDimension];
+
+  assert(output_width <= kMaxDimension);
+  assert(output_height <= kMaxDimension);
+  high_filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp, 64,
+                          output_width, output_height, bd);
+  high_block2d_average_c(tmp, 64, dst_ptr, dst_stride,
+                         output_width, output_height, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
  public:
   static void SetUpTestCase() {
@@ -177,6 +298,13 @@
         vpx_memalign(kDataAlignment, kInputBufferSize + 1)) + 1;
     output_ = reinterpret_cast<uint8_t*>(
         vpx_memalign(kDataAlignment, kOutputBufferSize));
+#if CONFIG_VP9_HIGHBITDEPTH
+    input16_ = reinterpret_cast<uint16_t*>(
+        vpx_memalign(kDataAlignment,
+                     (kInputBufferSize + 1) * sizeof(uint16_t))) + 1;
+    output16_ = reinterpret_cast<uint16_t*>(
+        vpx_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
+#endif
   }
 
   static void TearDownTestCase() {
@@ -184,6 +312,12 @@
     input_ = NULL;
     vpx_free(output_);
     output_ = NULL;
+#if CONFIG_VP9_HIGHBITDEPTH
+    vpx_free(input16_ - 1);
+    input16_ = NULL;
+    vpx_free(output16_);
+    output16_ = NULL;
+#endif
   }
 
  protected:
@@ -191,7 +325,6 @@
   static const int kOuterBlockSize = 256;
   static const int kInputStride = kOuterBlockSize;
   static const int kOutputStride = kOuterBlockSize;
-  static const int kMaxDimension = 64;
   static const int kInputBufferSize = kOuterBlockSize * kOuterBlockSize;
   static const int kOutputBufferSize = kOuterBlockSize * kOuterBlockSize;
 
@@ -212,6 +345,12 @@
 
   virtual void SetUp() {
     UUT_ = GET_PARAM(2);
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ != 0)
+      mask_ = (1 << UUT_->use_high_bd_) - 1;
+    else
+      mask_ = 255;
+#endif
     /* Set up guard blocks for an inner block centered in the outer block */
     for (int i = 0; i < kOutputBufferSize; ++i) {
       if (IsIndexInBorder(i))
@@ -222,15 +361,25 @@
 
     ::libvpx_test::ACMRandom prng;
     for (int i = 0; i < kInputBufferSize; ++i) {
-      if (i & 1)
+      if (i & 1) {
         input_[i] = 255;
-      else
+#if CONFIG_VP9_HIGHBITDEPTH
+        input16_[i] = mask_;
+#endif
+      } else {
         input_[i] = prng.Rand8Extremes();
+#if CONFIG_VP9_HIGHBITDEPTH
+        input16_[i] = prng.Rand16() & mask_;
+#endif
+      }
     }
   }
 
   void SetConstantInput(int value) {
     memset(input_, value, kInputBufferSize);
+#if CONFIG_VP9_HIGHBITDEPTH
+    vpx_memset16(input16_, value, kInputBufferSize);
+#endif
   }
 
   void CheckGuardBlocks() {
@@ -240,20 +389,123 @@
     }
   }
 
-  uint8_t* input() const {
+  uint8_t *input() const {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ == 0) {
+      return input_ + BorderTop() * kOuterBlockSize + BorderLeft();
+    } else {
+      return CONVERT_TO_BYTEPTR(input16_ + BorderTop() * kOuterBlockSize +
+                                BorderLeft());
+    }
+#else
     return input_ + BorderTop() * kOuterBlockSize + BorderLeft();
+#endif
   }
 
-  uint8_t* output() const {
+  uint8_t *output() const {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ == 0) {
+      return output_ + BorderTop() * kOuterBlockSize + BorderLeft();
+    } else {
+      return CONVERT_TO_BYTEPTR(output16_ + BorderTop() * kOuterBlockSize +
+                                BorderLeft());
+    }
+#else
     return output_ + BorderTop() * kOuterBlockSize + BorderLeft();
+#endif
+  }
+
+  uint16_t lookup(uint8_t *list, int index) const {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ == 0) {
+      return list[index];
+    } else {
+      return CONVERT_TO_SHORTPTR(list)[index];
+    }
+#else
+    return list[index];
+#endif
+  }
+
+  void assign_val(uint8_t *list, int index, uint16_t val) const {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ == 0) {
+      list[index] = (uint8_t) val;
+    } else {
+      CONVERT_TO_SHORTPTR(list)[index] = val;
+    }
+#else
+    list[index] = (uint8_t) val;
+#endif
+  }
+
+  void wrapper_filter_average_block2d_8_c(const uint8_t *src_ptr,
+                                          const unsigned int src_stride,
+                                          const int16_t *HFilter,
+                                          const int16_t *VFilter,
+                                          uint8_t *dst_ptr,
+                                          unsigned int dst_stride,
+                                          unsigned int output_width,
+                                          unsigned int output_height) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ == 0) {
+      filter_average_block2d_8_c(src_ptr, src_stride, HFilter, VFilter,
+                                 dst_ptr, dst_stride, output_width,
+                                 output_height);
+    } else {
+      high_filter_average_block2d_8_c(CONVERT_TO_SHORTPTR(src_ptr), src_stride,
+                                      HFilter, VFilter,
+                                      CONVERT_TO_SHORTPTR(dst_ptr), dst_stride,
+                                      output_width, output_height,
+                                      UUT_->use_high_bd_);
+    }
+#else
+    filter_average_block2d_8_c(src_ptr, src_stride, HFilter, VFilter,
+                               dst_ptr, dst_stride, output_width,
+                               output_height);
+#endif
+  }
+
+  void wrapper_filter_block2d_8_c(const uint8_t *src_ptr,
+                                  const unsigned int src_stride,
+                                  const int16_t *HFilter,
+                                  const int16_t *VFilter,
+                                  uint8_t *dst_ptr,
+                                  unsigned int dst_stride,
+                                  unsigned int output_width,
+                                  unsigned int output_height) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (UUT_->use_high_bd_ == 0) {
+      filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter,
+                         dst_ptr, dst_stride, output_width, output_height);
+    } else {
+      high_filter_block2d_8_c(CONVERT_TO_SHORTPTR(src_ptr), src_stride,
+                              HFilter, VFilter,
+                              CONVERT_TO_SHORTPTR(dst_ptr), dst_stride,
+                              output_width, output_height, UUT_->use_high_bd_);
+    }
+#else
+    filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter,
+                       dst_ptr, dst_stride, output_width, output_height);
+#endif
   }
 
   const ConvolveFunctions* UUT_;
   static uint8_t* input_;
   static uint8_t* output_;
+#if CONFIG_VP9_HIGHBITDEPTH
+  static uint16_t* input16_;
+  static uint16_t* output16_;
+  int mask_;
+#endif
 };
+
 uint8_t* ConvolveTest::input_ = NULL;
 uint8_t* ConvolveTest::output_ = NULL;
+#if CONFIG_VP9_HIGHBITDEPTH
+uint16_t* ConvolveTest::input16_ = NULL;
+uint16_t* ConvolveTest::output16_ = NULL;
+#endif
 
 TEST_P(ConvolveTest, GuardBlocks) {
   CheckGuardBlocks();
@@ -272,7 +524,8 @@
 
   for (int y = 0; y < Height(); ++y)
     for (int x = 0; x < Width(); ++x)
-      ASSERT_EQ(out[y * kOutputStride + x], in[y * kInputStride + x])
+      ASSERT_EQ(lookup(out, y * kOutputStride + x),
+                lookup(in, y * kInputStride + x))
           << "(" << x << "," << y << ")";
 }
 
@@ -289,7 +542,8 @@
 
   for (int y = 0; y < Height(); ++y)
     for (int x = 0; x < Width(); ++x)
-      ASSERT_EQ(out[y * kOutputStride + x], in[y * kInputStride + x])
+      ASSERT_EQ(lookup(out, y * kOutputStride + x),
+                lookup(in, y * kInputStride + x))
           << "(" << x << "," << y << ")";
 }
 
@@ -306,7 +560,8 @@
 
   for (int y = 0; y < Height(); ++y)
     for (int x = 0; x < Width(); ++x)
-      ASSERT_EQ(out[y * kOutputStride + x], in[y * kInputStride + x])
+      ASSERT_EQ(lookup(out, y * kOutputStride + x),
+                lookup(in, y * kInputStride + x))
           << "(" << x << "," << y << ")";
 }
 
@@ -339,8 +594,18 @@
 TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
   uint8_t* const in = input();
   uint8_t* const out = output();
+#if CONFIG_VP9_HIGHBITDEPTH
+  uint8_t ref8[kOutputStride * kMaxDimension];
+  uint16_t ref16[kOutputStride * kMaxDimension];
+  uint8_t* ref;
+  if (UUT_->use_high_bd_ == 0) {
+    ref = ref8;
+  } else {
+    ref = CONVERT_TO_BYTEPTR(ref16);
+  }
+#else
   uint8_t ref[kOutputStride * kMaxDimension];
-
+#endif
 
   for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
     const InterpKernel *filters =
@@ -350,10 +615,10 @@
 
     for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
       for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
-        filter_block2d_8_c(in, kInputStride,
-                           filters[filter_x], filters[filter_y],
-                           ref, kOutputStride,
-                           Width(), Height());
+        wrapper_filter_block2d_8_c(in, kInputStride,
+                                   filters[filter_x], filters[filter_y],
+                                   ref, kOutputStride,
+                                   Width(), Height());
 
         if (filters == eighttap_smooth || (filter_x && filter_y))
           ASM_REGISTER_STATE_CHECK(
@@ -375,7 +640,8 @@
 
         for (int y = 0; y < Height(); ++y)
           for (int x = 0; x < Width(); ++x)
-            ASSERT_EQ(ref[y * kOutputStride + x], out[y * kOutputStride + x])
+            ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                      lookup(out, y * kOutputStride + x))
                 << "mismatch at (" << x << "," << y << "), "
                 << "filters (" << filter_bank << ","
                 << filter_x << "," << filter_y << ")";
@@ -387,16 +653,36 @@
 TEST_P(ConvolveTest, MatchesReferenceAveragingSubpixelFilter) {
   uint8_t* const in = input();
   uint8_t* const out = output();
+#if CONFIG_VP9_HIGHBITDEPTH
+  uint8_t ref8[kOutputStride * kMaxDimension];
+  uint16_t ref16[kOutputStride * kMaxDimension];
+  uint8_t* ref;
+  if (UUT_->use_high_bd_ == 0) {
+    ref = ref8;
+  } else {
+    ref = CONVERT_TO_BYTEPTR(ref16);
+  }
+#else
   uint8_t ref[kOutputStride * kMaxDimension];
+#endif
 
   // Populate ref and out with some random data
   ::libvpx_test::ACMRandom prng;
   for (int y = 0; y < Height(); ++y) {
     for (int x = 0; x < Width(); ++x) {
-      const uint8_t r = prng.Rand8Extremes();
+      uint16_t r;
+#if CONFIG_VP9_HIGHBITDEPTH
+      if (UUT_->use_high_bd_ == 0 || UUT_->use_high_bd_ == 8) {
+        r = prng.Rand8Extremes();
+      } else {
+        r = prng.Rand16() & mask_;
+      }
+#else
+      r = prng.Rand8Extremes();
+#endif
 
-      out[y * kOutputStride + x] = r;
-      ref[y * kOutputStride + x] = r;
+      assign_val(out, y * kOutputStride + x, r);
+      assign_val(ref, y * kOutputStride + x, r);
     }
   }
 
@@ -408,10 +694,10 @@
 
     for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
       for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
-        filter_average_block2d_8_c(in, kInputStride,
-                                   filters[filter_x], filters[filter_y],
-                                   ref, kOutputStride,
-                                   Width(), Height());
+        wrapper_filter_average_block2d_8_c(in, kInputStride,
+                                           filters[filter_x], filters[filter_y],
+                                           ref, kOutputStride,
+                                           Width(), Height());
 
         if (filters == eighttap_smooth || (filter_x && filter_y))
           ASM_REGISTER_STATE_CHECK(
@@ -433,7 +719,8 @@
 
         for (int y = 0; y < Height(); ++y)
           for (int x = 0; x < Width(); ++x)
-            ASSERT_EQ(ref[y * kOutputStride + x], out[y * kOutputStride + x])
+            ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                      lookup(out, y * kOutputStride + x))
                 << "mismatch at (" << x << "," << y << "), "
                 << "filters (" << filter_bank << ","
                 << filter_x << "," << filter_y << ")";
@@ -442,6 +729,103 @@
   }
 }
 
+TEST_P(ConvolveTest, FilterExtremes) {
+  uint8_t *const in = input();
+  uint8_t *const out = output();
+#if CONFIG_VP9_HIGHBITDEPTH
+  uint8_t ref8[kOutputStride * kMaxDimension];
+  uint16_t ref16[kOutputStride * kMaxDimension];
+  uint8_t *ref;
+  if (UUT_->use_high_bd_ == 0) {
+    ref = ref8;
+  } else {
+    ref = CONVERT_TO_BYTEPTR(ref16);
+  }
+#else
+  uint8_t ref[kOutputStride * kMaxDimension];
+#endif
+
+  // Populate ref and out with some random data
+  ::libvpx_test::ACMRandom prng;
+  for (int y = 0; y < Height(); ++y) {
+    for (int x = 0; x < Width(); ++x) {
+      uint16_t r;
+#if CONFIG_VP9_HIGHBITDEPTH
+      if (UUT_->use_high_bd_ == 0 || UUT_->use_high_bd_ == 8) {
+        r = prng.Rand8Extremes();
+      } else {
+        r = prng.Rand16() & mask_;
+      }
+#else
+      r = prng.Rand8Extremes();
+#endif
+      assign_val(out, y * kOutputStride + x, r);
+      assign_val(ref, y * kOutputStride + x, r);
+    }
+  }
+
+  for (int axis = 0; axis < 2; axis++) {
+    int seed_val = 0;
+    while (seed_val < 256) {
+      for (int y = 0; y < 8; ++y) {
+        for (int x = 0; x < 8; ++x) {
+#if CONFIG_VP9_HIGHBITDEPTH
+            assign_val(in, y * kOutputStride + x - SUBPEL_TAPS / 2 + 1,
+                       ((seed_val >> (axis ? y : x)) & 1) * mask_);
+#else
+            assign_val(in, y * kOutputStride + x - SUBPEL_TAPS / 2 + 1,
+                       ((seed_val >> (axis ? y : x)) & 1) * 255);
+#endif
+          if (axis) seed_val++;
+        }
+        if (axis)
+          seed_val-= 8;
+        else
+          seed_val++;
+      }
+      if (axis) seed_val += 8;
+
+      for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+        const InterpKernel *filters =
+            vp9_get_interp_kernel(static_cast<INTERP_FILTER>(filter_bank));
+        const InterpKernel *const eighttap_smooth =
+            vp9_get_interp_kernel(EIGHTTAP_SMOOTH);
+        for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+          for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+            wrapper_filter_block2d_8_c(in, kInputStride,
+                                       filters[filter_x], filters[filter_y],
+                                       ref, kOutputStride,
+                                       Width(), Height());
+            if (filters == eighttap_smooth || (filter_x && filter_y))
+              ASM_REGISTER_STATE_CHECK(
+                  UUT_->hv8_(in, kInputStride, out, kOutputStride,
+                             filters[filter_x], 16, filters[filter_y], 16,
+                             Width(), Height()));
+            else if (filter_y)
+              ASM_REGISTER_STATE_CHECK(
+                  UUT_->v8_(in, kInputStride, out, kOutputStride,
+                            kInvalidFilter, 16, filters[filter_y], 16,
+                            Width(), Height()));
+            else
+              ASM_REGISTER_STATE_CHECK(
+                  UUT_->h8_(in, kInputStride, out, kOutputStride,
+                            filters[filter_x], 16, kInvalidFilter, 16,
+                            Width(), Height()));
+
+            for (int y = 0; y < Height(); ++y)
+              for (int x = 0; x < Width(); ++x)
+                ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                          lookup(out, y * kOutputStride + x))
+                    << "mismatch at (" << x << "," << y << "), "
+                    << "filters (" << filter_bank << ","
+                    << filter_x << "," << filter_y << ")";
+          }
+        }
+      }
+    }
+  }
+}
+
 DECLARE_ALIGNED(256, const int16_t, kChangeFilters[16][8]) = {
     { 0,   0,   0,   0,   0,   0,   0, 128},
     { 0,   0,   0,   0,   0,   0, 128},
@@ -505,7 +889,8 @@
         kPixelSelected + ((kInitialSubPelOffset
             + kFilterPeriodAdjust * kInputPixelStep)
                           >> SUBPEL_BITS);
-    ASSERT_EQ(in[ref_x], out[x]) << "x == " << x << "width = " << Width();
+    ASSERT_EQ(lookup(in, ref_x), lookup(out, x))
+        << "x == " << x << "width = " << Width();
   }
 
   /* Test the vertical filter. */
@@ -520,7 +905,8 @@
         kPixelSelected + ((kInitialSubPelOffset
             + kFilterPeriodAdjust * kInputPixelStep)
                           >> SUBPEL_BITS);
-    ASSERT_EQ(in[ref_y * kInputStride], out[y * kInputStride]) << "y == " << y;
+    ASSERT_EQ(lookup(in, ref_y * kInputStride), lookup(out, y * kInputStride))
+        << "y == " << y;
   }
 
   /* Test the horizontal and vertical filters in combination. */
@@ -543,7 +929,8 @@
               + kFilterPeriodAdjustX * kInputPixelStep)
                             >> SUBPEL_BITS);
 
-      ASSERT_EQ(in[ref_y * kInputStride + ref_x], out[y * kOutputStride + x])
+      ASSERT_EQ(lookup(in, ref_y * kInputStride + ref_x),
+                lookup(out, y * kOutputStride + x))
           << "x == " << x << ", y == " << y;
     }
   }
@@ -570,7 +957,8 @@
 
       for (int y = 0; y < Height(); ++y) {
         for (int x = 0; x < Width(); ++x) {
-          ASSERT_EQ(in[y * kInputStride + x], out[y * kOutputStride + x])
+          ASSERT_EQ(lookup(in, y * kInputStride + x),
+                    lookup(out, y * kOutputStride + x))
               << "x == " << x << ", y == " << y
               << ", frac == " << frac << ", step == " << step;
         }
@@ -581,10 +969,480 @@
 
 using std::tr1::make_tuple;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+#if HAVE_SSE2 && ARCH_X86_64
+void wrap_convolve8_horiz_sse2_8(const uint8_t *src, ptrdiff_t src_stride,
+                                 uint8_t *dst, ptrdiff_t dst_stride,
+                                 const int16_t *filter_x,
+                                 int filter_x_stride,
+                                 const int16_t *filter_y,
+                                 int filter_y_stride,
+                                 int w, int h) {
+  vp9_high_convolve8_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+                                filter_x_stride, filter_y, filter_y_stride,
+                                w, h, 8);
+}
+
+void wrap_convolve8_avg_horiz_sse2_8(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const int16_t *filter_x,
+                                     int filter_x_stride,
+                                     const int16_t *filter_y,
+                                     int filter_y_stride,
+                                     int w, int h) {
+  vp9_high_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_vert_sse2_8(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x,
+                                int filter_x_stride,
+                                const int16_t *filter_y,
+                                int filter_y_stride,
+                                int w, int h) {
+  vp9_high_convolve8_vert_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_avg_vert_sse2_8(const uint8_t *src, ptrdiff_t src_stride,
+                                    uint8_t *dst, ptrdiff_t dst_stride,
+                                    const int16_t *filter_x,
+                                    int filter_x_stride,
+                                    const int16_t *filter_y,
+                                    int filter_y_stride,
+                                    int w, int h) {
+  vp9_high_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride, filter_x,
+                                   filter_x_stride, filter_y, filter_y_stride,
+                                   w, h, 8);
+}
+
+void wrap_convolve8_sse2_8(const uint8_t *src, ptrdiff_t src_stride,
+                           uint8_t *dst, ptrdiff_t dst_stride,
+                           const int16_t *filter_x,
+                           int filter_x_stride,
+                           const int16_t *filter_y,
+                           int filter_y_stride,
+                           int w, int h) {
+  vp9_high_convolve8_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_avg_sse2_8(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x,
+                               int filter_x_stride,
+                               const int16_t *filter_y,
+                               int filter_y_stride,
+                               int w, int h) {
+  vp9_high_convolve8_avg_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_horiz_sse2_10(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const int16_t *filter_x,
+                                  int filter_x_stride,
+                                  const int16_t *filter_y,
+                                  int filter_y_stride,
+                                  int w, int h) {
+  vp9_high_convolve8_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_avg_horiz_sse2_10(const uint8_t *src, ptrdiff_t src_stride,
+                                      uint8_t *dst, ptrdiff_t dst_stride,
+                                      const int16_t *filter_x,
+                                      int filter_x_stride,
+                                      const int16_t *filter_y,
+                                      int filter_y_stride,
+                                      int w, int h) {
+  vp9_high_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_vert_sse2_10(const uint8_t *src, ptrdiff_t src_stride,
+                                 uint8_t *dst, ptrdiff_t dst_stride,
+                                 const int16_t *filter_x,
+                                 int filter_x_stride,
+                                 const int16_t *filter_y,
+                                 int filter_y_stride,
+                                 int w, int h) {
+  vp9_high_convolve8_vert_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_avg_vert_sse2_10(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const int16_t *filter_x,
+                                     int filter_x_stride,
+                                     const int16_t *filter_y,
+                                     int filter_y_stride,
+                                     int w, int h) {
+  vp9_high_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_sse2_10(const uint8_t *src, ptrdiff_t src_stride,
+                            uint8_t *dst, ptrdiff_t dst_stride,
+                            const int16_t *filter_x,
+                            int filter_x_stride,
+                            const int16_t *filter_y,
+                            int filter_y_stride,
+                            int w, int h) {
+  vp9_high_convolve8_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_avg_sse2_10(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x,
+                                int filter_x_stride,
+                                const int16_t *filter_y,
+                                int filter_y_stride,
+                                int w, int h) {
+  vp9_high_convolve8_avg_sse2(src, src_stride, dst, dst_stride, filter_x,
+                              filter_x_stride, filter_y, filter_y_stride,
+                              w, h, 10);
+}
+
+void wrap_convolve8_horiz_sse2_12(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const int16_t *filter_x,
+                                  int filter_x_stride,
+                                  const int16_t *filter_y,
+                                  int filter_y_stride,
+                                  int w, int h) {
+  vp9_high_convolve8_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+                                filter_x_stride, filter_y, filter_y_stride,
+                                w, h, 12);
+}
+
+void wrap_convolve8_avg_horiz_sse2_12(const uint8_t *src, ptrdiff_t src_stride,
+                                      uint8_t *dst, ptrdiff_t dst_stride,
+                                      const int16_t *filter_x,
+                                      int filter_x_stride,
+                                      const int16_t *filter_y,
+                                      int filter_y_stride,
+                                      int w, int h) {
+  vp9_high_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+                                    filter_x_stride, filter_y, filter_y_stride,
+                                    w, h, 12);
+}
+
+void wrap_convolve8_vert_sse2_12(const uint8_t *src, ptrdiff_t src_stride,
+                                 uint8_t *dst, ptrdiff_t dst_stride,
+                                 const int16_t *filter_x,
+                                 int filter_x_stride,
+                                 const int16_t *filter_y,
+                                 int filter_y_stride,
+                                 int w, int h) {
+  vp9_high_convolve8_vert_sse2(src, src_stride, dst, dst_stride, filter_x,
+                               filter_x_stride, filter_y, filter_y_stride,
+                               w, h, 12);
+}
+
+void wrap_convolve8_avg_vert_sse2_12(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const int16_t *filter_x,
+                                     int filter_x_stride,
+                                     const int16_t *filter_y,
+                                     int filter_y_stride,
+                                     int w, int h) {
+  vp9_high_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride, filter_x,
+                                   filter_x_stride, filter_y, filter_y_stride, w, h, 12);
+}
+
+void wrap_convolve8_sse2_12(const uint8_t *src, ptrdiff_t src_stride,
+                            uint8_t *dst, ptrdiff_t dst_stride,
+                            const int16_t *filter_x,
+                            int filter_x_stride,
+                            const int16_t *filter_y,
+                            int filter_y_stride,
+                            int w, int h) {
+  vp9_high_convolve8_sse2(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 12);
+}
+
+void wrap_convolve8_avg_sse2_12(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x,
+                                int filter_x_stride,
+                                const int16_t *filter_y,
+                                int filter_y_stride,
+                                int w, int h) {
+  vp9_high_convolve8_avg_sse2(src, src_stride, dst, dst_stride, filter_x,
+                              filter_x_stride, filter_y, filter_y_stride, w, h, 12);
+}
+#endif  // HAVE_SSE2 && ARCH_X86_64
+
+void wrap_convolve8_horiz_c_8(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x,
+                              int filter_x_stride,
+                              const int16_t *filter_y,
+                              int filter_y_stride,
+                              int w, int h) {
+  vp9_high_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x,
+                             filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_avg_horiz_c_8(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const int16_t *filter_x,
+                                  int filter_x_stride,
+                                  const int16_t *filter_y,
+                                  int filter_y_stride,
+                                  int w, int h) {
+  vp9_high_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
+                                 filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_vert_c_8(const uint8_t *src, ptrdiff_t src_stride,
+                             uint8_t *dst, ptrdiff_t dst_stride,
+                             const int16_t *filter_x,
+                             int filter_x_stride,
+                             const int16_t *filter_y,
+                             int filter_y_stride,
+                             int w, int h) {
+  vp9_high_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x,
+                            filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_avg_vert_c_8(const uint8_t *src, ptrdiff_t src_stride,
+                                 uint8_t *dst, ptrdiff_t dst_stride,
+                                 const int16_t *filter_x,
+                                 int filter_x_stride,
+                                 const int16_t *filter_y,
+                                 int filter_y_stride,
+                                 int w, int h) {
+  vp9_high_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
+                                filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_c_8(const uint8_t *src, ptrdiff_t src_stride,
+                        uint8_t *dst, ptrdiff_t dst_stride,
+                        const int16_t *filter_x,
+                        int filter_x_stride,
+                        const int16_t *filter_y,
+                        int filter_y_stride,
+                        int w, int h) {
+  vp9_high_convolve8_c(src, src_stride, dst, dst_stride, filter_x,
+                       filter_x_stride, filter_y, filter_y_stride, w, h, 8);
+}
+
+void wrap_convolve8_avg_c_8(const uint8_t *src, ptrdiff_t src_stride,
+                            uint8_t *dst, ptrdiff_t dst_stride,
+                            const int16_t *filter_x,
+                            int filter_x_stride,
+                            const int16_t *filter_y,
+                            int filter_y_stride,
+                            int w, int h) {
+  vp9_high_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
+                           filter_x_stride, filter_y, filter_y_stride,
+                           w, h, 8);
+}
+
+void wrap_convolve8_horiz_c_10(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x,
+                               int filter_x_stride,
+                               const int16_t *filter_y,
+                               int filter_y_stride,
+                               int w, int h) {
+  vp9_high_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x,
+                             filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_avg_horiz_c_10(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const int16_t *filter_x,
+                                   int filter_x_stride,
+                                   const int16_t *filter_y,
+                                   int filter_y_stride,
+                                   int w, int h) {
+  vp9_high_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
+                                 filter_x_stride, filter_y, filter_y_stride,
+                                 w, h, 10);
+}
+
+void wrap_convolve8_vert_c_10(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x,
+                              int filter_x_stride,
+                              const int16_t *filter_y,
+                              int filter_y_stride,
+                              int w, int h) {
+  vp9_high_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x,
+                            filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_avg_vert_c_10(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const int16_t *filter_x,
+                                  int filter_x_stride,
+                                  const int16_t *filter_y,
+                                  int filter_y_stride,
+                                  int w, int h) {
+  vp9_high_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
+                                filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_c_10(const uint8_t *src, ptrdiff_t src_stride,
+                         uint8_t *dst, ptrdiff_t dst_stride,
+                         const int16_t *filter_x,
+                         int filter_x_stride,
+                         const int16_t *filter_y,
+                         int filter_y_stride,
+                         int w, int h) {
+  vp9_high_convolve8_c(src, src_stride, dst, dst_stride, filter_x,
+    filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_avg_c_10(const uint8_t *src, ptrdiff_t src_stride,
+                             uint8_t *dst, ptrdiff_t dst_stride,
+                             const int16_t *filter_x,
+                             int filter_x_stride,
+                             const int16_t *filter_y,
+                             int filter_y_stride,
+                             int w, int h) {
+  vp9_high_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
+                           filter_x_stride, filter_y, filter_y_stride, w, h, 10);
+}
+
+void wrap_convolve8_horiz_c_12(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x,
+                               int filter_x_stride,
+                               const int16_t *filter_y,
+                               int filter_y_stride,
+                               int w, int h) {
+  vp9_high_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x,
+                             filter_x_stride, filter_y, filter_y_stride,
+                             w, h, 12);
+}
+
+void wrap_convolve8_avg_horiz_c_12(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const int16_t *filter_x,
+                                   int filter_x_stride,
+                                   const int16_t *filter_y,
+                                   int filter_y_stride,
+                                   int w, int h) {
+  vp9_high_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
+                                 filter_x_stride, filter_y, filter_y_stride,
+                                 w, h, 12);
+}
+
+void wrap_convolve8_vert_c_12(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x,
+                              int filter_x_stride,
+                              const int16_t *filter_y,
+                              int filter_y_stride,
+                              int w, int h) {
+  vp9_high_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x,
+                            filter_x_stride, filter_y, filter_y_stride,
+                            w, h, 12);
+}
+
+void wrap_convolve8_avg_vert_c_12(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const int16_t *filter_x,
+                                  int filter_x_stride,
+                                  const int16_t *filter_y,
+                                  int filter_y_stride,
+                                  int w, int h) {
+  vp9_high_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
+                                filter_x_stride, filter_y, filter_y_stride,
+                                w, h, 12);
+}
+
+void wrap_convolve8_c_12(const uint8_t *src, ptrdiff_t src_stride,
+                         uint8_t *dst, ptrdiff_t dst_stride,
+                         const int16_t *filter_x,
+                         int filter_x_stride,
+                         const int16_t *filter_y,
+                         int filter_y_stride,
+                         int w, int h) {
+  vp9_high_convolve8_c(src, src_stride, dst, dst_stride, filter_x,
+                       filter_x_stride, filter_y, filter_y_stride,
+                       w, h, 12);
+}
+
+void wrap_convolve8_avg_c_12(const uint8_t *src, ptrdiff_t src_stride,
+                             uint8_t *dst, ptrdiff_t dst_stride,
+                             const int16_t *filter_x,
+                             int filter_x_stride,
+                             const int16_t *filter_y,
+                             int filter_y_stride,
+                             int w, int h) {
+  vp9_high_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
+                           filter_x_stride, filter_y, filter_y_stride,
+                           w, h, 12);
+}
+
+const ConvolveFunctions convolve8_c(
+    wrap_convolve8_horiz_c_8, wrap_convolve8_avg_horiz_c_8,
+    wrap_convolve8_vert_c_8, wrap_convolve8_avg_vert_c_8,
+    wrap_convolve8_c_8, wrap_convolve8_avg_c_8, 8);
+INSTANTIATE_TEST_CASE_P(C_8, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve8_c),
+    make_tuple(8, 4, &convolve8_c),
+    make_tuple(4, 8, &convolve8_c),
+    make_tuple(8, 8, &convolve8_c),
+    make_tuple(16, 8, &convolve8_c),
+    make_tuple(8, 16, &convolve8_c),
+    make_tuple(16, 16, &convolve8_c),
+    make_tuple(32, 16, &convolve8_c),
+    make_tuple(16, 32, &convolve8_c),
+    make_tuple(32, 32, &convolve8_c),
+    make_tuple(64, 32, &convolve8_c),
+    make_tuple(32, 64, &convolve8_c),
+    make_tuple(64, 64, &convolve8_c)));
+const ConvolveFunctions convolve10_c(
+    wrap_convolve8_horiz_c_10, wrap_convolve8_avg_horiz_c_10,
+    wrap_convolve8_vert_c_10, wrap_convolve8_avg_vert_c_10,
+    wrap_convolve8_c_10, wrap_convolve8_avg_c_10, 10);
+INSTANTIATE_TEST_CASE_P(C_10, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve10_c),
+    make_tuple(8, 4, &convolve10_c),
+    make_tuple(4, 8, &convolve10_c),
+    make_tuple(8, 8, &convolve10_c),
+    make_tuple(16, 8, &convolve10_c),
+    make_tuple(8, 16, &convolve10_c),
+    make_tuple(16, 16, &convolve10_c),
+    make_tuple(32, 16, &convolve10_c),
+    make_tuple(16, 32, &convolve10_c),
+    make_tuple(32, 32, &convolve10_c),
+    make_tuple(64, 32, &convolve10_c),
+    make_tuple(32, 64, &convolve10_c),
+    make_tuple(64, 64, &convolve10_c)));
+const ConvolveFunctions convolve12_c(
+    wrap_convolve8_horiz_c_12, wrap_convolve8_avg_horiz_c_12,
+    wrap_convolve8_vert_c_12, wrap_convolve8_avg_vert_c_12,
+    wrap_convolve8_c_12, wrap_convolve8_avg_c_12, 12);
+INSTANTIATE_TEST_CASE_P(C_12, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve12_c),
+    make_tuple(8, 4, &convolve12_c),
+    make_tuple(4, 8, &convolve12_c),
+    make_tuple(8, 8, &convolve12_c),
+    make_tuple(16, 8, &convolve12_c),
+    make_tuple(8, 16, &convolve12_c),
+    make_tuple(16, 16, &convolve12_c),
+    make_tuple(32, 16, &convolve12_c),
+    make_tuple(16, 32, &convolve12_c),
+    make_tuple(32, 32, &convolve12_c),
+    make_tuple(64, 32, &convolve12_c),
+    make_tuple(32, 64, &convolve12_c),
+    make_tuple(64, 64, &convolve12_c)));
+
+#else
+
 const ConvolveFunctions convolve8_c(
     vp9_convolve8_horiz_c, vp9_convolve8_avg_horiz_c,
     vp9_convolve8_vert_c, vp9_convolve8_avg_vert_c,
-    vp9_convolve8_c, vp9_convolve8_avg_c);
+    vp9_convolve8_c, vp9_convolve8_avg_c, 0);
 
 INSTANTIATE_TEST_CASE_P(C, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_c),
@@ -600,12 +1458,69 @@
     make_tuple(64, 32, &convolve8_c),
     make_tuple(32, 64, &convolve8_c),
     make_tuple(64, 64, &convolve8_c)));
+#endif
 
-#if HAVE_SSE2
+#if HAVE_SSE2 && ARCH_X86_64
+#if CONFIG_VP9_HIGHBITDEPTH
+const ConvolveFunctions convolve8_sse2(
+    wrap_convolve8_horiz_sse2_8, wrap_convolve8_avg_horiz_sse2_8,
+    wrap_convolve8_vert_sse2_8, wrap_convolve8_avg_vert_sse2_8,
+    wrap_convolve8_sse2_8, wrap_convolve8_avg_sse2_8, 8);
+INSTANTIATE_TEST_CASE_P(SSE2_8, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve8_sse2),
+    make_tuple(8, 4, &convolve8_sse2),
+    make_tuple(4, 8, &convolve8_sse2),
+    make_tuple(8, 8, &convolve8_sse2),
+    make_tuple(16, 8, &convolve8_sse2),
+    make_tuple(8, 16, &convolve8_sse2),
+    make_tuple(16, 16, &convolve8_sse2),
+    make_tuple(32, 16, &convolve8_sse2),
+    make_tuple(16, 32, &convolve8_sse2),
+    make_tuple(32, 32, &convolve8_sse2),
+    make_tuple(64, 32, &convolve8_sse2),
+    make_tuple(32, 64, &convolve8_sse2),
+    make_tuple(64, 64, &convolve8_sse2)));
+const ConvolveFunctions convolve10_sse2(
+    wrap_convolve8_horiz_sse2_10, wrap_convolve8_avg_horiz_sse2_10,
+    wrap_convolve8_vert_sse2_10, wrap_convolve8_avg_vert_sse2_10,
+    wrap_convolve8_sse2_10, wrap_convolve8_avg_sse2_10, 10);
+INSTANTIATE_TEST_CASE_P(SSE2_10, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve10_sse2),
+    make_tuple(8, 4, &convolve10_sse2),
+    make_tuple(4, 8, &convolve10_sse2),
+    make_tuple(8, 8, &convolve10_sse2),
+    make_tuple(16, 8, &convolve10_sse2),
+    make_tuple(8, 16, &convolve10_sse2),
+    make_tuple(16, 16, &convolve10_sse2),
+    make_tuple(32, 16, &convolve10_sse2),
+    make_tuple(16, 32, &convolve10_sse2),
+    make_tuple(32, 32, &convolve10_sse2),
+    make_tuple(64, 32, &convolve10_sse2),
+    make_tuple(32, 64, &convolve10_sse2),
+    make_tuple(64, 64, &convolve10_sse2)));
+const ConvolveFunctions convolve12_sse2(
+    wrap_convolve8_horiz_sse2_12, wrap_convolve8_avg_horiz_sse2_12,
+    wrap_convolve8_vert_sse2_12, wrap_convolve8_avg_vert_sse2_12,
+    wrap_convolve8_sse2_12, wrap_convolve8_avg_sse2_12, 12);
+INSTANTIATE_TEST_CASE_P(SSE2_12, ConvolveTest, ::testing::Values(
+    make_tuple(4, 4, &convolve12_sse2),
+    make_tuple(8, 4, &convolve12_sse2),
+    make_tuple(4, 8, &convolve12_sse2),
+    make_tuple(8, 8, &convolve12_sse2),
+    make_tuple(16, 8, &convolve12_sse2),
+    make_tuple(8, 16, &convolve12_sse2),
+    make_tuple(16, 16, &convolve12_sse2),
+    make_tuple(32, 16, &convolve12_sse2),
+    make_tuple(16, 32, &convolve12_sse2),
+    make_tuple(32, 32, &convolve12_sse2),
+    make_tuple(64, 32, &convolve12_sse2),
+    make_tuple(32, 64, &convolve12_sse2),
+    make_tuple(64, 64, &convolve12_sse2)));
+#else
 const ConvolveFunctions convolve8_sse2(
     vp9_convolve8_horiz_sse2, vp9_convolve8_avg_horiz_sse2,
     vp9_convolve8_vert_sse2, vp9_convolve8_avg_vert_sse2,
-    vp9_convolve8_sse2, vp9_convolve8_avg_sse2);
+    vp9_convolve8_sse2, vp9_convolve8_avg_sse2, 0);
 
 INSTANTIATE_TEST_CASE_P(SSE2, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_sse2),
@@ -621,13 +1536,14 @@
     make_tuple(64, 32, &convolve8_sse2),
     make_tuple(32, 64, &convolve8_sse2),
     make_tuple(64, 64, &convolve8_sse2)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 #endif
 
 #if HAVE_SSSE3
 const ConvolveFunctions convolve8_ssse3(
     vp9_convolve8_horiz_ssse3, vp9_convolve8_avg_horiz_ssse3,
     vp9_convolve8_vert_ssse3, vp9_convolve8_avg_vert_ssse3,
-    vp9_convolve8_ssse3, vp9_convolve8_avg_ssse3);
+    vp9_convolve8_ssse3, vp9_convolve8_avg_ssse3, 0);
 
 INSTANTIATE_TEST_CASE_P(SSSE3, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_ssse3),
@@ -645,11 +1561,11 @@
     make_tuple(64, 64, &convolve8_ssse3)));
 #endif
 
-#if HAVE_AVX2
+#if HAVE_AVX2 && HAVE_SSSE3
 const ConvolveFunctions convolve8_avx2(
     vp9_convolve8_horiz_avx2, vp9_convolve8_avg_horiz_ssse3,
     vp9_convolve8_vert_avx2, vp9_convolve8_avg_vert_ssse3,
-    vp9_convolve8_avx2, vp9_convolve8_avg_ssse3);
+    vp9_convolve8_avx2, vp9_convolve8_avg_ssse3, 0);
 
 INSTANTIATE_TEST_CASE_P(AVX2, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_avx2),
@@ -665,13 +1581,13 @@
     make_tuple(64, 32, &convolve8_avx2),
     make_tuple(32, 64, &convolve8_avx2),
     make_tuple(64, 64, &convolve8_avx2)));
-#endif
+#endif  // HAVE_AVX2 && HAVE_SSSE3
 
 #if HAVE_NEON_ASM
 const ConvolveFunctions convolve8_neon(
     vp9_convolve8_horiz_neon, vp9_convolve8_avg_horiz_neon,
     vp9_convolve8_vert_neon, vp9_convolve8_avg_vert_neon,
-    vp9_convolve8_neon, vp9_convolve8_avg_neon);
+    vp9_convolve8_neon, vp9_convolve8_avg_neon, 0);
 
 INSTANTIATE_TEST_CASE_P(NEON, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_neon),
@@ -693,7 +1609,7 @@
 const ConvolveFunctions convolve8_dspr2(
     vp9_convolve8_horiz_dspr2, vp9_convolve8_avg_horiz_dspr2,
     vp9_convolve8_vert_dspr2, vp9_convolve8_avg_vert_dspr2,
-    vp9_convolve8_dspr2, vp9_convolve8_avg_dspr2);
+    vp9_convolve8_dspr2, vp9_convolve8_avg_dspr2, 0);
 
 INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values(
     make_tuple(4, 4, &convolve8_dspr2),
diff --git a/test/datarate_test.cc b/test/datarate_test.cc
index a3d730a..1b7863a 100644
--- a/test/datarate_test.cc
+++ b/test/datarate_test.cc
@@ -126,6 +126,7 @@
   int denoiser_on_;
 };
 
+#if CONFIG_TEMPORAL_DENOISING
 // Check basic datarate targeting, for a single bitrate, but loop over the
 // various denoiser settings.
 TEST_P(DatarateTestLarge, DenoiserLevels) {
@@ -154,6 +155,7 @@
         << " The datarate for the file missed the target!";
   }
 }
+#endif  // CONFIG_TEMPORAL_DENOISING
 
 TEST_P(DatarateTestLarge, BasicBufferModel) {
   denoiser_on_ = 0;
@@ -238,6 +240,8 @@
     tot_frame_number_ = 0;
     first_drop_ = 0;
     num_drops_ = 0;
+    // Denoiser is off by default.
+    denoiser_on_ = 0;
     // For testing up to 3 layers.
     for (int i = 0; i < 3; ++i) {
       bits_total_[i] = 0;
@@ -311,6 +315,7 @@
                                   ::libvpx_test::Encoder *encoder) {
     if (video->frame() == 1) {
       encoder->Control(VP8E_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(VP9E_SET_NOISE_SENSITIVITY, denoiser_on_);
     }
     if (cfg_.ts_number_layers > 1) {
       if (video->frame() == 1) {
@@ -392,6 +397,7 @@
   int64_t bits_in_buffer_model_;
   vpx_codec_pts_t first_drop_;
   int num_drops_;
+  int denoiser_on_;
 };
 
 // Check basic rate targeting,
@@ -615,6 +621,36 @@
   }
 }
 
+#if CONFIG_VP9_TEMPORAL_DENOISING
+// Check basic datarate targeting, for a single bitrate, when denoiser is on.
+TEST_P(DatarateTestVP9Large, DenoiserLevels) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_dropframe_thresh = 1;
+  cfg_.rc_min_quantizer = 2;
+  cfg_.rc_max_quantizer = 56;
+  cfg_.rc_end_usage = VPX_CBR;
+  cfg_.g_lag_in_frames = 0;
+
+  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 140);
+
+  // For the temporal denoiser (#if CONFIG_VP9_TEMPORAL_DENOISING),
+  // there is only one denoiser mode: denoiserYonly(which is 1),
+  // but may add more modes in the future.
+  cfg_.rc_target_bitrate = 300;
+  ResetModel();
+  // Turn on the denoiser.
+  denoiser_on_ = 1;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.85)
+      << " The datarate for the file is lower than target by too much!";
+  ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.15)
+      << " The datarate for the file is greater than target by too much!";
+}
+#endif  // CONFIG_VP9_TEMPORAL_DENOISING
+
 VP8_INSTANTIATE_TEST_CASE(DatarateTestLarge, ALL_TEST_MODES);
 VP9_INSTANTIATE_TEST_CASE(DatarateTestVP9Large,
                           ::testing::Values(::libvpx_test::kOnePassGood,
diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index c38cc2e..d1ce109 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -20,12 +20,9 @@
 
 #include "./vp9_rtcd.h"
 #include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
 
-extern "C" {
-void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *output, int pitch);
-}
-
 using libvpx_test::ACMRandom;
 
 namespace {
@@ -258,42 +255,72 @@
   }
 }
 
-typedef void (*FdctFunc)(const int16_t *in, int16_t *out, int stride);
-typedef void (*IdctFunc)(const int16_t *in, uint8_t *out, int stride);
-typedef void (*FhtFunc)(const int16_t *in, int16_t *out, int stride,
+typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
+typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
                         int tx_type);
-typedef void (*IhtFunc)(const int16_t *in, uint8_t *out, int stride,
+typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
                         int tx_type);
 
-typedef std::tr1::tuple<FdctFunc, IdctFunc, int> Dct16x16Param;
-typedef std::tr1::tuple<FhtFunc, IhtFunc, int> Ht16x16Param;
+typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct16x16Param;
+typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht16x16Param;
 
-void fdct16x16_ref(const int16_t *in, int16_t *out, int stride,
+void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
                    int /*tx_type*/) {
   vp9_fdct16x16_c(in, out, stride);
 }
 
-void idct16x16_ref(const int16_t *in, uint8_t *dest, int stride,
+void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
                    int /*tx_type*/) {
   vp9_idct16x16_256_add_c(in, dest, stride);
 }
 
-void fht16x16_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
+void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
+                  int tx_type) {
   vp9_fht16x16_c(in, out, stride, tx_type);
 }
 
-void iht16x16_ref(const int16_t *in, uint8_t *dest, int stride, int tx_type) {
+void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
+                  int tx_type) {
   vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct16x16_256_add_c(in, out, stride, 10);
+}
+
+void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct16x16_256_add_c(in, out, stride, 12);
+}
+
+void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
+                      int tx_type) {
+  idct16x16_10(in, out, stride);
+}
+
+void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
+                      int tx_type) {
+  idct16x16_12(in, out, stride);
+}
+
+void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
+  vp9_high_iht16x16_256_add_c(in, out, stride, tx_type, 10);
+}
+
+void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
+  vp9_high_iht16x16_256_add_c(in, out, stride, tx_type, 12);
+}
+#endif
+
 class Trans16x16TestBase {
  public:
   virtual ~Trans16x16TestBase() {}
 
  protected:
-  virtual void RunFwdTxfm(int16_t *in, int16_t *out, int stride) = 0;
+  virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
 
-  virtual void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) = 0;
+  virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
 
   void RunAccuracyCheck() {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
@@ -302,23 +329,48 @@
     const int count_test_block = 10000;
     for (int i = 0; i < count_test_block; ++i) {
       DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
-      DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, kNumCoeffs);
+      DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
       DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
       DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+#endif
 
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        src[j] = rnd.Rand8();
-        dst[j] = rnd.Rand8();
-        test_input_block[j] = src[j] - dst[j];
+        if (bit_depth_ == VPX_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          test_input_block[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          test_input_block[j] = src16[j] - dst16[j];
+#endif
+        }
       }
 
       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
                                           test_temp_block, pitch_));
-      ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+#endif
+      }
 
       for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const uint32_t diff =
+            bit_depth_ == VPX_BITS_8 ?  dst[j] - src[j] : dst16[j] - src16[j];
+#else
         const uint32_t diff = dst[j] - src[j];
+#endif
         const uint32_t error = diff * diff;
         if (max_error < error)
           max_error = error;
@@ -326,10 +378,10 @@
       }
     }
 
-    EXPECT_GE(1u, max_error)
+    EXPECT_GE(1u  << 2 * (bit_depth_ - 8), max_error)
         << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
 
-    EXPECT_GE(count_test_block , total_error)
+    EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
         << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
   }
 
@@ -337,13 +389,13 @@
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 1000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j)
-        input_block[j] = rnd.Rand8() - rnd.Rand8();
+        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
 
       fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
@@ -359,21 +411,21 @@
     const int count_test_block = 1000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        input_block[j] = rnd.Rand8() - rnd.Rand8();
-        input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
+        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
+        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
       }
       if (i == 0) {
         for (int j = 0; j < kNumCoeffs; ++j)
-          input_extreme_block[j] = 255;
+          input_extreme_block[j] = mask_;
       } else if (i == 1) {
         for (int j = 0; j < kNumCoeffs; ++j)
-          input_extreme_block[j] = -255;
+          input_extreme_block[j] = -mask_;
       }
 
       fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
@@ -383,7 +435,7 @@
       // The minimum quant value is 4.
       for (int j = 0; j < kNumCoeffs; ++j) {
         EXPECT_EQ(output_block[j], output_ref_block[j]);
-        EXPECT_GE(4 * DCT_MAX_VALUE, abs(output_block[j]))
+        EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
             << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
       }
     }
@@ -394,39 +446,65 @@
     const int count_test_block = 1000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
 
     DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, ref, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, ref16, kNumCoeffs);
+#endif
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        input_block[j] = rnd.Rand8() - rnd.Rand8();
-        input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
+        if (bit_depth_ == VPX_BITS_8)
+          input_block[j] = rnd.Rand8() - rnd.Rand8();
+        else
+          input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
+        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
       }
       if (i == 0)
         for (int j = 0; j < kNumCoeffs; ++j)
-          input_extreme_block[j] = 255;
+          input_extreme_block[j] = mask_;
       if (i == 1)
         for (int j = 0; j < kNumCoeffs; ++j)
-          input_extreme_block[j] = -255;
+          input_extreme_block[j] = -mask_;
 
       fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
 
       // clear reconstructed pixel buffers
       vpx_memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
       vpx_memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
+#if CONFIG_VP9_HIGHBITDEPTH
+      vpx_memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
+      vpx_memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
+#endif
 
       // quantization with maximum allowed step sizes
       output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
       for (int j = 1; j < kNumCoeffs; ++j)
         output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
-      inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
-      ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
-
-      for (int j = 0; j < kNumCoeffs; ++j)
-        EXPECT_EQ(ref[j], dst[j]);
+      if (bit_depth_ == VPX_BITS_8) {
+        inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        inv_txfm_ref(output_ref_block, CONVERT_TO_BYTEPTR(ref16), pitch_,
+                     tx_type_);
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block,
+                                            CONVERT_TO_BYTEPTR(dst16), pitch_));
+#endif
+      }
+      if (bit_depth_ == VPX_BITS_8) {
+        for (int j = 0; j < kNumCoeffs; ++j)
+          EXPECT_EQ(ref[j], dst[j]);
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        for (int j = 0; j < kNumCoeffs; ++j)
+          EXPECT_EQ(ref16[j], dst16[j]);
+#endif
+      }
     }
   }
 
@@ -434,28 +512,52 @@
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 1000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, coeff, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+#endif
 
     for (int i = 0; i < count_test_block; ++i) {
       double out_r[kNumCoeffs];
 
       // Initialize a test block with input range [-255, 255].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        src[j] = rnd.Rand8();
-        dst[j] = rnd.Rand8();
-        in[j] = src[j] - dst[j];
+        if (bit_depth_ == VPX_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          in[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          in[j] = src16[j] - dst16[j];
+#endif
+        }
       }
 
       reference_16x16_dct_2d(in, out_r);
       for (int j = 0; j < kNumCoeffs; ++j)
         coeff[j] = round(out_r[j]);
 
-      ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
+                                            16));
+#endif
+      }
 
       for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const uint32_t diff =
+            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
         const uint32_t diff = dst[j] - src[j];
+#endif
         const uint32_t error = diff * diff;
         EXPECT_GE(1u, error)
             << "Error: 16x16 IDCT has error " << error
@@ -465,6 +567,8 @@
   }
   int pitch_;
   int tx_type_;
+  vpx_bit_depth_t bit_depth_;
+  int mask_;
   FhtFunc fwd_txfm_ref;
   IhtFunc inv_txfm_ref;
 };
@@ -479,17 +583,34 @@
     fwd_txfm_ = GET_PARAM(0);
     inv_txfm_ = GET_PARAM(1);
     tx_type_  = GET_PARAM(2);
+    bit_depth_ = GET_PARAM(3);
     pitch_    = 16;
     fwd_txfm_ref = fdct16x16_ref;
     inv_txfm_ref = idct16x16_ref;
+    mask_ = (1 << bit_depth_) - 1;
+#if CONFIG_VP9_HIGHBITDEPTH
+    switch (bit_depth_) {
+      case 10:
+        inv_txfm_ref = idct16x16_10_ref;
+        break;
+      case 12:
+        inv_txfm_ref = idct16x16_12_ref;
+        break;
+      default:
+        inv_txfm_ref = idct16x16_ref;
+        break;
+    }
+#else
+    inv_txfm_ref = idct16x16_ref;
+#endif
   }
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride);
   }
-  void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride);
   }
 
@@ -529,17 +650,34 @@
     fwd_txfm_ = GET_PARAM(0);
     inv_txfm_ = GET_PARAM(1);
     tx_type_  = GET_PARAM(2);
+    bit_depth_ = GET_PARAM(3);
     pitch_    = 16;
     fwd_txfm_ref = fht16x16_ref;
     inv_txfm_ref = iht16x16_ref;
+    mask_ = (1 << bit_depth_) - 1;
+#if CONFIG_VP9_HIGHBITDEPTH
+    switch (bit_depth_) {
+      case VPX_BITS_10:
+        inv_txfm_ref = iht16x16_10;
+        break;
+      case VPX_BITS_12:
+        inv_txfm_ref = iht16x16_12;
+        break;
+      default:
+        inv_txfm_ref = iht16x16_ref;
+        break;
+    }
+#else
+    inv_txfm_ref = iht16x16_ref;
+#endif
   }
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride, tx_type_);
   }
-  void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride, tx_type_);
   }
 
@@ -567,45 +705,78 @@
 
 using std::tr1::make_tuple;
 
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, Trans16x16DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0)));
+        make_tuple(&vp9_high_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, Trans16x16DCT,
+    ::testing::Values(
+        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_c, 0, VPX_BITS_8)));
+#endif
+
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, Trans16x16HT,
     ::testing::Values(
-        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0),
-        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1),
-        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2),
-        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3)));
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
+        make_tuple(&vp9_high_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, Trans16x16HT,
+    ::testing::Values(
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
+#endif
 
-#if HAVE_NEON_ASM
+#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     NEON, Trans16x16DCT,
     ::testing::Values(
         make_tuple(&vp9_fdct16x16_c,
-                   &vp9_idct16x16_256_add_neon, 0)));
+                   &vp9_idct16x16_256_add_neon, 0, VPX_BITS_8)));
 #endif
 
-#if HAVE_SSE2
+#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSE2, Trans16x16DCT,
     ::testing::Values(
         make_tuple(&vp9_fdct16x16_sse2,
-                   &vp9_idct16x16_256_add_sse2, 0)));
+                   &vp9_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
 INSTANTIATE_TEST_CASE_P(
     SSE2, Trans16x16HT,
     ::testing::Values(
-        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0),
-        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1),
-        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2),
-        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3)));
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0,
+                   VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1,
+                   VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2,
+                   VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3,
+                   VPX_BITS_8)));
 #endif
 
-#if HAVE_SSSE3
+#if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSSE3, Trans16x16DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_ssse3, 0)));
+        make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_ssse3, 0,
+                   VPX_BITS_8)));
 #endif
 }  // namespace
diff --git a/test/dct32x32_test.cc b/test/dct32x32_test.cc
index d2d437c..c7a1931 100644
--- a/test/dct32x32_test.cc
+++ b/test/dct32x32_test.cc
@@ -21,6 +21,7 @@
 #include "./vpx_config.h"
 #include "./vp9_rtcd.h"
 #include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
 
 using libvpx_test::ACMRandom;
@@ -71,10 +72,21 @@
   }
 }
 
-typedef void (*FwdTxfmFunc)(const int16_t *in, int16_t *out, int stride);
-typedef void (*InvTxfmFunc)(const int16_t *in, uint8_t *out, int stride);
+typedef void (*FwdTxfmFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*InvTxfmFunc)(const tran_low_t *in, uint8_t *out, int stride);
 
-typedef std::tr1::tuple<FwdTxfmFunc, InvTxfmFunc, int> Trans32x32Param;
+typedef std::tr1::tuple<FwdTxfmFunc, InvTxfmFunc, int, vpx_bit_depth_t>
+    Trans32x32Param;
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void idct32x32_10(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct32x32_1024_add_c(in, out, stride, 10);
+}
+
+void idct32x32_12(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct32x32_1024_add_c(in, out, stride, 12);
+}
+#endif
 
 class Trans32x32Test : public ::testing::TestWithParam<Trans32x32Param> {
  public:
@@ -84,12 +96,16 @@
     inv_txfm_ = GET_PARAM(1);
     version_  = GET_PARAM(2);  // 0: high precision forward transform
                                // 1: low precision version for rd loop
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
   }
 
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
   int version_;
+  vpx_bit_depth_t bit_depth_;
+  int mask_;
   FwdTxfmFunc fwd_txfm_;
   InvTxfmFunc inv_txfm_;
 };
@@ -100,23 +116,47 @@
   int64_t total_error = 0;
   const int count_test_block = 1000;
   DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+#endif
 
   for (int i = 0; i < count_test_block; ++i) {
-    // Initialize a test block with input range [-255, 255].
+    // Initialize a test block with input range [-mask_, mask_].
     for (int j = 0; j < kNumCoeffs; ++j) {
-      src[j] = rnd.Rand8();
-      dst[j] = rnd.Rand8();
-      test_input_block[j] = src[j] - dst[j];
+      if (bit_depth_ == 8) {
+        src[j] = rnd.Rand8();
+        dst[j] = rnd.Rand8();
+        test_input_block[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        src16[j] = rnd.Rand16() & mask_;
+        dst16[j] = rnd.Rand16() & mask_;
+        test_input_block[j] = src16[j] - dst16[j];
+#endif
+      }
     }
 
     ASM_REGISTER_STATE_CHECK(fwd_txfm_(test_input_block, test_temp_block, 32));
-    ASM_REGISTER_STATE_CHECK(inv_txfm_(test_temp_block, dst, 32));
+    if (bit_depth_ == VPX_BITS_8) {
+      ASM_REGISTER_STATE_CHECK(inv_txfm_(test_temp_block, dst, 32));
+#if CONFIG_VP9_HIGHBITDEPTH
+    } else {
+      ASM_REGISTER_STATE_CHECK(inv_txfm_(test_temp_block,
+                                         CONVERT_TO_BYTEPTR(dst16), 32));
+#endif
+    }
 
     for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+      const uint32_t diff =
+          bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
       const uint32_t diff = dst[j] - src[j];
+#endif
       const uint32_t error = diff * diff;
       if (max_error < error)
         max_error = error;
@@ -129,10 +169,10 @@
     total_error /= 45;
   }
 
-  EXPECT_GE(1u, max_error)
+  EXPECT_GE(1u << 2 * (bit_depth_ - 8), max_error)
       << "Error: 32x32 FDCT/IDCT has an individual round-trip error > 1";
 
-  EXPECT_GE(count_test_block, total_error)
+  EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
       << "Error: 32x32 FDCT/IDCT has average round-trip error > 1 per block";
 }
 
@@ -141,12 +181,12 @@
   const int count_test_block = 1000;
 
   DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
 
   for (int i = 0; i < count_test_block; ++i) {
     for (int j = 0; j < kNumCoeffs; ++j)
-      input_block[j] = rnd.Rand8() - rnd.Rand8();
+      input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
 
     const int stride = 32;
     vp9_fdct32x32_c(input_block, output_ref_block, stride);
@@ -170,21 +210,21 @@
 
   DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
 
   for (int i = 0; i < count_test_block; ++i) {
-    // Initialize a test block with input range [-255, 255].
+    // Initialize a test block with input range [-mask_, mask_].
     for (int j = 0; j < kNumCoeffs; ++j) {
-      input_block[j] = rnd.Rand8() - rnd.Rand8();
-      input_extreme_block[j] = rnd.Rand8() & 1 ? 255 : -255;
+      input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
+      input_extreme_block[j] = rnd.Rand8() & 1 ? mask_ : -mask_;
     }
     if (i == 0) {
       for (int j = 0; j < kNumCoeffs; ++j)
-        input_extreme_block[j] = 255;
+        input_extreme_block[j] = mask_;
     } else if (i == 1) {
       for (int j = 0; j < kNumCoeffs; ++j)
-        input_extreme_block[j] = -255;
+        input_extreme_block[j] = -mask_;
     }
 
     const int stride = 32;
@@ -201,9 +241,9 @@
         EXPECT_GE(6, abs(output_block[j] - output_ref_block[j]))
             << "Error: 32x32 FDCT rd has mismatched coefficients";
       }
-      EXPECT_GE(4 * DCT_MAX_VALUE, abs(output_ref_block[j]))
+      EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_ref_block[j]))
           << "Error: 32x32 FDCT C has coefficient larger than 4*DCT_MAX_VALUE";
-      EXPECT_GE(4 * DCT_MAX_VALUE, abs(output_block[j]))
+      EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
           << "Error: 32x32 FDCT has coefficient larger than "
           << "4*DCT_MAX_VALUE";
     }
@@ -214,26 +254,49 @@
   ACMRandom rnd(ACMRandom::DeterministicSeed());
   const int count_test_block = 1000;
   DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, coeff, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+#endif
 
   for (int i = 0; i < count_test_block; ++i) {
     double out_r[kNumCoeffs];
 
     // Initialize a test block with input range [-255, 255]
     for (int j = 0; j < kNumCoeffs; ++j) {
-      src[j] = rnd.Rand8();
-      dst[j] = rnd.Rand8();
-      in[j] = src[j] - dst[j];
+      if (bit_depth_ == VPX_BITS_8) {
+        src[j] = rnd.Rand8();
+        dst[j] = rnd.Rand8();
+        in[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        src16[j] = rnd.Rand16() & mask_;
+        dst16[j] = rnd.Rand16() & mask_;
+        in[j] = src16[j] - dst16[j];
+#endif
+      }
     }
 
     reference_32x32_dct_2d(in, out_r);
     for (int j = 0; j < kNumCoeffs; ++j)
       coeff[j] = round(out_r[j]);
-    ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, dst, 32));
+    if (bit_depth_ == VPX_BITS_8) {
+      ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, dst, 32));
+#if CONFIG_VP9_HIGHBITDEPTH
+    } else {
+      ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, CONVERT_TO_BYTEPTR(dst16), 32));
+#endif
+    }
     for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+      const int diff =
+          bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
       const int diff = dst[j] - src[j];
+#endif
       const int error = diff * diff;
       EXPECT_GE(1, error)
           << "Error: 32x32 IDCT has error " << error
@@ -244,39 +307,59 @@
 
 using std::tr1::make_tuple;
 
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, Trans32x32Test,
     ::testing::Values(
-        make_tuple(&vp9_fdct32x32_c, &vp9_idct32x32_1024_add_c, 0),
-        make_tuple(&vp9_fdct32x32_rd_c, &vp9_idct32x32_1024_add_c, 1)));
+        make_tuple(&vp9_high_fdct32x32_c,
+                   &idct32x32_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fdct32x32_rd_c,
+                   &idct32x32_10, 1, VPX_BITS_10),
+        make_tuple(&vp9_high_fdct32x32_c,
+                   &idct32x32_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_high_fdct32x32_rd_c,
+                   &idct32x32_12, 1, VPX_BITS_12),
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fdct32x32_rd_c,
+                   &vp9_idct32x32_1024_add_c, 1, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, Trans32x32Test,
+    ::testing::Values(
+        make_tuple(&vp9_fdct32x32_c,
+                   &vp9_idct32x32_1024_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fdct32x32_rd_c,
+                   &vp9_idct32x32_1024_add_c, 1, VPX_BITS_8)));
+#endif
 
-#if HAVE_NEON_ASM
+#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     NEON, Trans32x32Test,
     ::testing::Values(
         make_tuple(&vp9_fdct32x32_c,
-                   &vp9_idct32x32_1024_add_neon, 0),
+                   &vp9_idct32x32_1024_add_neon, 0, VPX_BITS_8),
         make_tuple(&vp9_fdct32x32_rd_c,
-                   &vp9_idct32x32_1024_add_neon, 1)));
+                   &vp9_idct32x32_1024_add_neon, 1, VPX_BITS_8)));
 #endif
 
-#if HAVE_SSE2
+#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSE2, Trans32x32Test,
     ::testing::Values(
         make_tuple(&vp9_fdct32x32_sse2,
-                   &vp9_idct32x32_1024_add_sse2, 0),
+                   &vp9_idct32x32_1024_add_sse2, 0, VPX_BITS_8),
         make_tuple(&vp9_fdct32x32_rd_sse2,
-                   &vp9_idct32x32_1024_add_sse2, 1)));
+                   &vp9_idct32x32_1024_add_sse2, 1, VPX_BITS_8)));
 #endif
 
-#if HAVE_AVX2
+#if HAVE_AVX2 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     AVX2, Trans32x32Test,
     ::testing::Values(
         make_tuple(&vp9_fdct32x32_avx2,
-                   &vp9_idct32x32_1024_add_sse2, 0),
+                   &vp9_idct32x32_1024_add_sse2, 0, VPX_BITS_8),
         make_tuple(&vp9_fdct32x32_rd_avx2,
-                   &vp9_idct32x32_1024_add_sse2, 1)));
+                   &vp9_idct32x32_1024_add_sse2, 1, VPX_BITS_8)));
 #endif
 }  // namespace
diff --git a/test/external_frame_buffer_test.cc b/test/external_frame_buffer_test.cc
index 44eba33..70b3009 100644
--- a/test/external_frame_buffer_test.cc
+++ b/test/external_frame_buffer_test.cc
@@ -71,6 +71,7 @@
     if (ext_fb_list_[idx].size < min_size) {
       delete [] ext_fb_list_[idx].data;
       ext_fb_list_[idx].data = new uint8_t[min_size];
+      memset(ext_fb_list_[idx].data, 0, min_size);
       ext_fb_list_[idx].size = min_size;
     }
 
diff --git a/test/fdct4x4_test.cc b/test/fdct4x4_test.cc
index 08a69ab..f803c8e 100644
--- a/test/fdct4x4_test.cc
+++ b/test/fdct4x4_test.cc
@@ -20,46 +20,71 @@
 
 #include "./vp9_rtcd.h"
 #include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
 
-extern "C" {
-void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *output, int pitch);
-}
-
 using libvpx_test::ACMRandom;
 
 namespace {
 const int kNumCoeffs = 16;
-typedef void (*FdctFunc)(const int16_t *in, int16_t *out, int stride);
-typedef void (*IdctFunc)(const int16_t *in, uint8_t *out, int stride);
-typedef void (*FhtFunc)(const int16_t *in, int16_t *out, int stride,
+typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
+typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
                         int tx_type);
-typedef void (*IhtFunc)(const int16_t *in, uint8_t *out, int stride,
+typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
                         int tx_type);
 
-typedef std::tr1::tuple<FdctFunc, IdctFunc, int> Dct4x4Param;
-typedef std::tr1::tuple<FhtFunc, IhtFunc, int> Ht4x4Param;
+typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct4x4Param;
+typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht4x4Param;
 
-void fdct4x4_ref(const int16_t *in, int16_t *out, int stride, int /*tx_type*/) {
+void fdct4x4_ref(const int16_t *in, tran_low_t *out, int stride,
+                 int tx_type) {
   vp9_fdct4x4_c(in, out, stride);
 }
 
-void fht4x4_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
+void fht4x4_ref(const int16_t *in, tran_low_t *out, int stride, int tx_type) {
   vp9_fht4x4_c(in, out, stride, tx_type);
 }
 
-void fwht4x4_ref(const int16_t *in, int16_t *out, int stride, int /*tx_type*/) {
+void fwht4x4_ref(const int16_t *in, tran_low_t *out, int stride,
+                 int tx_type) {
   vp9_fwht4x4_c(in, out, stride);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void idct4x4_10(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct4x4_16_add_c(in, out, stride, 10);
+}
+
+void idct4x4_12(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct4x4_16_add_c(in, out, stride, 12);
+}
+
+void iht4x4_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
+  vp9_high_iht4x4_16_add_c(in, out, stride, tx_type, 10);
+}
+
+void iht4x4_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
+  vp9_high_iht4x4_16_add_c(in, out, stride, tx_type, 12);
+}
+
+void iwht4x4_10(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_iwht4x4_16_add_c(in, out, stride, 10);
+}
+
+void iwht4x4_12(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_iwht4x4_16_add_c(in, out, stride, 12);
+}
+#endif
+
 class Trans4x4TestBase {
  public:
   virtual ~Trans4x4TestBase() {}
 
  protected:
-  virtual void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) = 0;
+  virtual void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) = 0;
 
-  virtual void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) = 0;
+  virtual void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) = 0;
 
   void RunAccuracyCheck(int limit) {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
@@ -68,23 +93,47 @@
     const int count_test_block = 10000;
     for (int i = 0; i < count_test_block; ++i) {
       DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, kNumCoeffs);
-      DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, kNumCoeffs);
+      DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, kNumCoeffs);
       DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
       DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+#endif
 
       // Initialize a test block with input range [-255, 255].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        src[j] = rnd.Rand8();
-        dst[j] = rnd.Rand8();
-        test_input_block[j] = src[j] - dst[j];
+        if (bit_depth_ == VPX_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          test_input_block[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          test_input_block[j] = src16[j] - dst16[j];
+#endif
+        }
       }
 
       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
                                           test_temp_block, pitch_));
-      ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block,
+                                            CONVERT_TO_BYTEPTR(dst16), pitch_));
+#endif
+      }
 
       for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const uint32_t diff =
+            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
         const uint32_t diff = dst[j] - src[j];
+#endif
         const uint32_t error = diff * diff;
         if (max_error < error)
           max_error = error;
@@ -105,13 +154,13 @@
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 5000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j)
-        input_block[j] = rnd.Rand8() - rnd.Rand8();
+        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
 
       fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
       ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
@@ -127,21 +176,21 @@
     const int count_test_block = 5000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_block, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, output_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_block, kNumCoeffs);
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        input_block[j] = rnd.Rand8() - rnd.Rand8();
-        input_extreme_block[j] = rnd.Rand8() % 2 ? 255 : -255;
+        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
+        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
       }
       if (i == 0) {
         for (int j = 0; j < kNumCoeffs; ++j)
-          input_extreme_block[j] = 255;
+          input_extreme_block[j] = mask_;
       } else if (i == 1) {
         for (int j = 0; j < kNumCoeffs; ++j)
-          input_extreme_block[j] = -255;
+          input_extreme_block[j] = -mask_;
       }
 
       fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
@@ -151,8 +200,8 @@
       // The minimum quant value is 4.
       for (int j = 0; j < kNumCoeffs; ++j) {
         EXPECT_EQ(output_block[j], output_ref_block[j]);
-        EXPECT_GE(4 * DCT_MAX_VALUE, abs(output_block[j]))
-            << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
+        EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
+            << "Error: 4x4 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
       }
     }
   }
@@ -161,24 +210,48 @@
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     const int count_test_block = 1000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, coeff, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+#endif
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < kNumCoeffs; ++j) {
-        src[j] = rnd.Rand8();
-        dst[j] = rnd.Rand8();
-        in[j] = src[j] - dst[j];
+        if (bit_depth_ == VPX_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          in[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          in[j] = src16[j] - dst16[j];
+#endif
+        }
       }
 
       fwd_txfm_ref(in, coeff, pitch_, tx_type_);
 
-      ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
+                                            pitch_));
+#endif
+      }
 
       for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const uint32_t diff =
+            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
         const uint32_t diff = dst[j] - src[j];
+#endif
         const uint32_t error = diff * diff;
         EXPECT_GE(static_cast<uint32_t>(limit), error)
             << "Error: 4x4 IDCT has error " << error
@@ -190,6 +263,8 @@
   int pitch_;
   int tx_type_;
   FhtFunc fwd_txfm_ref;
+  vpx_bit_depth_t bit_depth_;
+  int mask_;
 };
 
 class Trans4x4DCT
@@ -204,14 +279,16 @@
     tx_type_  = GET_PARAM(2);
     pitch_    = 4;
     fwd_txfm_ref = fdct4x4_ref;
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
   }
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride);
   }
-  void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride);
   }
 
@@ -247,15 +324,17 @@
     tx_type_  = GET_PARAM(2);
     pitch_    = 4;
     fwd_txfm_ref = fht4x4_ref;
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
   }
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride, tx_type_);
   }
 
-  void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride, tx_type_);
   }
 
@@ -291,14 +370,16 @@
     tx_type_  = GET_PARAM(2);
     pitch_    = 4;
     fwd_txfm_ref = fwht4x4_ref;
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
   }
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(const int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride);
   }
-  void RunInvTxfm(const int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride);
   }
 
@@ -323,57 +404,95 @@
 }
 using std::tr1::make_tuple;
 
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, Trans4x4DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct4x4_c, &vp9_idct4x4_16_add_c, 0)));
+        make_tuple(&vp9_high_fdct4x4_c, &idct4x4_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fdct4x4_c, &idct4x4_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_fdct4x4_c, &vp9_idct4x4_16_add_c, 0, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, Trans4x4DCT,
+    ::testing::Values(
+        make_tuple(&vp9_fdct4x4_c, &vp9_idct4x4_16_add_c, 0, VPX_BITS_8)));
+#endif
+
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, Trans4x4HT,
     ::testing::Values(
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 0),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 1),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 2),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 3)));
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_10, 1, VPX_BITS_10),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_10, 2, VPX_BITS_10),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_10, 3, VPX_BITS_10),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_12, 1, VPX_BITS_12),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_12, 2, VPX_BITS_12),
+        make_tuple(&vp9_high_fht4x4_c, &iht4x4_12, 3, VPX_BITS_12),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 3, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, Trans4x4HT,
+    ::testing::Values(
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 3, VPX_BITS_8)));
+#endif
+
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, Trans4x4WHT,
     ::testing::Values(
-        make_tuple(&vp9_fwht4x4_c, &vp9_iwht4x4_16_add_c, 0)));
+        make_tuple(&vp9_high_fwht4x4_c, &iwht4x4_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fwht4x4_c, &iwht4x4_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_fwht4x4_c, &vp9_iwht4x4_16_add_c, 0, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, Trans4x4WHT,
+    ::testing::Values(
+        make_tuple(&vp9_fwht4x4_c, &vp9_iwht4x4_16_add_c, 0, VPX_BITS_8)));
+#endif
 
-#if HAVE_NEON_ASM
+#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     NEON, Trans4x4DCT,
     ::testing::Values(
         make_tuple(&vp9_fdct4x4_c,
-                   &vp9_idct4x4_16_add_neon, 0)));
+                   &vp9_idct4x4_16_add_neon, 0, VPX_BITS_8)));
 INSTANTIATE_TEST_CASE_P(
     DISABLED_NEON, Trans4x4HT,
     ::testing::Values(
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 0),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 1),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 2),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 3)));
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 3, VPX_BITS_8)));
 #endif
 
-#if CONFIG_USE_X86INC && HAVE_MMX
+#if CONFIG_USE_X86INC && HAVE_MMX && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     MMX, Trans4x4WHT,
     ::testing::Values(
-        make_tuple(&vp9_fwht4x4_mmx, &vp9_iwht4x4_16_add_c, 0)));
+        make_tuple(&vp9_fwht4x4_mmx, &vp9_iwht4x4_16_add_c, 0, VPX_BITS_8)));
 #endif
 
-#if HAVE_SSE2
+#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSE2, Trans4x4DCT,
     ::testing::Values(
         make_tuple(&vp9_fdct4x4_sse2,
-                   &vp9_idct4x4_16_add_sse2, 0)));
+                   &vp9_idct4x4_16_add_sse2, 0, VPX_BITS_8)));
 INSTANTIATE_TEST_CASE_P(
     SSE2, Trans4x4HT,
     ::testing::Values(
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 0),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 1),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 2),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 3)));
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 3, VPX_BITS_8)));
 #endif
 
 }  // namespace
diff --git a/test/fdct8x8_test.cc b/test/fdct8x8_test.cc
index a694f0c..60d0be5 100644
--- a/test/fdct8x8_test.cc
+++ b/test/fdct8x8_test.cc
@@ -20,45 +20,96 @@
 
 #include "./vp9_rtcd.h"
 #include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
 
-extern "C" {
-void vp9_idct8x8_64_add_c(const int16_t *input, uint8_t *output, int pitch);
+const int kNumCoeffs = 64;
+const double kPi = 3.141592653589793238462643383279502884;
+void reference_8x8_dct_1d(const double in[8], double out[8], int stride) {
+  const double kInvSqrt2 = 0.707106781186547524400844362104;
+  for (int k = 0; k < 8; k++) {
+    out[k] = 0.0;
+    for (int n = 0; n < 8; n++)
+      out[k] += in[n] * cos(kPi * (2 * n + 1) * k / 16.0);
+    if (k == 0)
+      out[k] = out[k] * kInvSqrt2;
+  }
+}
+
+void reference_8x8_dct_2d(const int16_t input[kNumCoeffs],
+                          double output[kNumCoeffs]) {
+  // First transform columns
+  for (int i = 0; i < 8; ++i) {
+    double temp_in[8], temp_out[8];
+    for (int j = 0; j < 8; ++j)
+      temp_in[j] = input[j*8 + i];
+    reference_8x8_dct_1d(temp_in, temp_out, 1);
+    for (int j = 0; j < 8; ++j)
+      output[j * 8 + i] = temp_out[j];
+  }
+  // Then transform rows
+  for (int i = 0; i < 8; ++i) {
+    double temp_in[8], temp_out[8];
+    for (int j = 0; j < 8; ++j)
+      temp_in[j] = output[j + i*8];
+    reference_8x8_dct_1d(temp_in, temp_out, 1);
+    // Scale by some magic number
+    for (int j = 0; j < 8; ++j)
+      output[j + i * 8] = temp_out[j] * 2;
+  }
 }
 
 using libvpx_test::ACMRandom;
 
 namespace {
-typedef void (*FdctFunc)(const int16_t *in, int16_t *out, int stride);
-typedef void (*IdctFunc)(const int16_t *in, uint8_t *out, int stride);
-typedef void (*FhtFunc)(const int16_t *in, int16_t *out, int stride,
+typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
+typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
                         int tx_type);
-typedef void (*IhtFunc)(const int16_t *in, uint8_t *out, int stride,
+typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
                         int tx_type);
 
-typedef std::tr1::tuple<FdctFunc, IdctFunc, int> Dct8x8Param;
-typedef std::tr1::tuple<FhtFunc, IhtFunc, int> Ht8x8Param;
+typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct8x8Param;
+typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht8x8Param;
 
-void fdct8x8_ref(const int16_t *in, int16_t *out, int stride, int /*tx_type*/) {
+void fdct8x8_ref(const int16_t *in, tran_low_t *out, int stride, int tx_type) {
   vp9_fdct8x8_c(in, out, stride);
 }
 
-void fht8x8_ref(const int16_t *in, int16_t *out, int stride, int tx_type) {
+void fht8x8_ref(const int16_t *in, tran_low_t *out, int stride, int tx_type) {
   vp9_fht8x8_c(in, out, stride, tx_type);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void idct8x8_10(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct8x8_64_add_c(in, out, stride, 10);
+}
+
+void idct8x8_12(const tran_low_t *in, uint8_t *out, int stride) {
+  vp9_high_idct8x8_64_add_c(in, out, stride, 12);
+}
+
+void iht8x8_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
+  vp9_high_iht8x8_64_add_c(in, out, stride, tx_type, 10);
+}
+
+void iht8x8_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
+  vp9_high_iht8x8_64_add_c(in, out, stride, tx_type, 12);
+}
+#endif
+
 class FwdTrans8x8TestBase {
  public:
   virtual ~FwdTrans8x8TestBase() {}
 
  protected:
-  virtual void RunFwdTxfm(int16_t *in, int16_t *out, int stride) = 0;
-  virtual void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) = 0;
+  virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
+  virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
 
   void RunSignBiasCheck() {
     ACMRandom rnd(ACMRandom::DeterministicSeed());
     DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, 64);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, test_output_block, 64);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_output_block, 64);
     int count_sign_block[64][2];
     const int count_test_block = 100000;
 
@@ -67,7 +118,8 @@
     for (int i = 0; i < count_test_block; ++i) {
       // Initialize a test block with input range [-255, 255].
       for (int j = 0; j < 64; ++j)
-        test_input_block[j] = rnd.Rand8() - rnd.Rand8();
+        test_input_block[j] = ((rnd.Rand16() >> (16 - bit_depth_)) & mask_) -
+                              ((rnd.Rand16() >> (16 - bit_depth_)) & mask_);
       ASM_REGISTER_STATE_CHECK(
           RunFwdTxfm(test_input_block, test_output_block, pitch_));
 
@@ -82,7 +134,7 @@
     for (int j = 0; j < 64; ++j) {
       const int diff = abs(count_sign_block[j][0] - count_sign_block[j][1]);
       const int max_diff = 1125;
-      EXPECT_LT(diff, max_diff)
+      EXPECT_LT(diff, max_diff << (bit_depth_ - 8))
           << "Error: 8x8 FDCT/FHT has a sign bias > "
           << 1. * max_diff / count_test_block * 100 << "%"
           << " for input range [-255, 255] at index " << j
@@ -111,7 +163,7 @@
     for (int j = 0; j < 64; ++j) {
       const int diff = abs(count_sign_block[j][0] - count_sign_block[j][1]);
       const int max_diff = 10000;
-      EXPECT_LT(diff, max_diff)
+      EXPECT_LT(diff, max_diff << (bit_depth_ - 8))
           << "Error: 4x4 FDCT/FHT has a sign bias > "
           << 1. * max_diff / count_test_block * 100 << "%"
           << " for input range [-15, 15] at index " << j
@@ -127,16 +179,28 @@
     int total_error = 0;
     const int count_test_block = 100000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, 64);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, 64);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, 64);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, 64);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, src, 64);
+#if CONFIG_VP9_HIGHBITDEPTH
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, 64);
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, 64);
+#endif
 
     for (int i = 0; i < count_test_block; ++i) {
       // Initialize a test block with input range [-255, 255].
       for (int j = 0; j < 64; ++j) {
-        src[j] = rnd.Rand8();
-        dst[j] = rnd.Rand8();
-        test_input_block[j] = src[j] - dst[j];
+        if (bit_depth_ == VPX_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          test_input_block[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          test_input_block[j] = src16[j] - dst16[j];
+#endif
+        }
       }
 
       ASM_REGISTER_STATE_CHECK(
@@ -152,11 +216,23 @@
             test_temp_block[j] *= 4;
           }
       }
-      ASM_REGISTER_STATE_CHECK(
-          RunInvTxfm(test_temp_block, dst, pitch_));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+#endif
+      }
 
       for (int j = 0; j < 64; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const int diff =
+            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
         const int diff = dst[j] - src[j];
+#endif
         const int error = diff * diff;
         if (max_error < error)
           max_error = error;
@@ -164,11 +240,11 @@
       }
     }
 
-    EXPECT_GE(1, max_error)
+    EXPECT_GE(1 << 2 * (bit_depth_ - 8), max_error)
       << "Error: 8x8 FDCT/IDCT or FHT/IHT has an individual"
       << " roundtrip error > 1";
 
-    EXPECT_GE(count_test_block/5, total_error)
+    EXPECT_GE((count_test_block << 2 * (bit_depth_ - 8))/5, total_error)
       << "Error: 8x8 FDCT/IDCT or FHT/IHT has average roundtrip "
       << "error > 1/5 per block";
   }
@@ -180,37 +256,68 @@
     int total_coeff_error = 0;
     const int count_test_block = 100000;
     DECLARE_ALIGNED_ARRAY(16, int16_t, test_input_block, 64);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, test_temp_block, 64);
-    DECLARE_ALIGNED_ARRAY(16, int16_t, ref_temp_block, 64);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_temp_block, 64);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, ref_temp_block, 64);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, 64);
     DECLARE_ALIGNED_ARRAY(16, uint8_t, src, 64);
+#if CONFIG_VP9_HIGHBITDEPTH
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, 64);
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, 64);
+#endif
 
     for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-255, 255].
+      // Initialize a test block with input range [-mask_, mask_].
       for (int j = 0; j < 64; ++j) {
-        if (i == 0) {
-          src[j] = 255;
-          dst[j] = 0;
-        } else if (i == 1) {
-          src[j] = 0;
-          dst[j] = 255;
+        if (bit_depth_ == VPX_BITS_8) {
+          if (i == 0) {
+            src[j] = 255;
+            dst[j] = 0;
+          } else if (i == 1) {
+            src[j] = 0;
+            dst[j] = 255;
+          } else {
+            src[j] = rnd.Rand8() % 2 ? 255 : 0;
+            dst[j] = rnd.Rand8() % 2 ? 255 : 0;
+          }
+          test_input_block[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
         } else {
-          src[j] = rnd.Rand8() % 2 ? 255 : 0;
-          dst[j] = rnd.Rand8() % 2 ? 255 : 0;
+          if (i == 0) {
+            src16[j] = mask_;
+            dst16[j] = 0;
+          } else if (i == 1) {
+            src16[j] = 0;
+            dst16[j] = mask_;
+          } else {
+            src16[j] = rnd.Rand8() % 2 ? mask_ : 0;
+            dst16[j] = rnd.Rand8() % 2 ? mask_ : 0;
+          }
+          test_input_block[j] = src16[j] - dst16[j];
+#endif
         }
-
-        test_input_block[j] = src[j] - dst[j];
       }
 
       ASM_REGISTER_STATE_CHECK(
           RunFwdTxfm(test_input_block, test_temp_block, pitch_));
       ASM_REGISTER_STATE_CHECK(
           fwd_txfm_ref(test_input_block, ref_temp_block, pitch_, tx_type_));
-      ASM_REGISTER_STATE_CHECK(
-          RunInvTxfm(test_temp_block, dst, pitch_));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+#endif
+      }
 
       for (int j = 0; j < 64; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const int diff =
+            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
         const int diff = dst[j] - src[j];
+#endif
         const int error = diff * diff;
         if (max_error < error)
           max_error = error;
@@ -220,11 +327,11 @@
         total_coeff_error += abs(coeff_diff);
       }
 
-      EXPECT_GE(1, max_error)
+      EXPECT_GE(1 << 2 * (bit_depth_ - 8), max_error)
           << "Error: Extremal 8x8 FDCT/IDCT or FHT/IHT has"
           << "an individual roundtrip error > 1";
 
-      EXPECT_GE(count_test_block/5, total_error)
+      EXPECT_GE((count_test_block << 2 * (bit_depth_ - 8))/5, total_error)
           << "Error: Extremal 8x8 FDCT/IDCT or FHT/IHT has average"
           << " roundtrip error > 1/5 per block";
 
@@ -234,9 +341,97 @@
     }
   }
 
+  void RunInvAccuracyCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 1000;
+    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, uint8_t, dst, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, uint8_t, src, kNumCoeffs);
+#if CONFIG_VP9_HIGHBITDEPTH
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, src16, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, uint16_t, dst16, kNumCoeffs);
+#endif
+
+    for (int i = 0; i < count_test_block; ++i) {
+      double out_r[kNumCoeffs];
+
+      // Initialize a test block with input range [-255, 255].
+      for (int j = 0; j < kNumCoeffs; ++j) {
+        if (bit_depth_ == VPX_BITS_8) {
+          src[j] = rnd.Rand8() % 2 ? 255 : 0;
+          dst[j] = src[j] > 0 ? 0 : 255;
+          in[j] = src[j] - dst[j];
+#if CONFIG_VP9_HIGHBITDEPTH
+        } else {
+          src16[j] = rnd.Rand8() % 2 ? mask_ : 0;
+          dst16[j] = src16[j] > 0 ? 0 : mask_;
+          in[j] = src16[j] - dst16[j];
+#endif
+        }
+      }
+
+      reference_8x8_dct_2d(in, out_r);
+      for (int j = 0; j < kNumCoeffs; ++j)
+        coeff[j] = static_cast<tran_low_t>(round(out_r[j]));
+
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
+                                            pitch_));
+#endif
+      }
+
+      for (int j = 0; j < kNumCoeffs; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+        const uint32_t diff =
+            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+#else
+        const uint32_t diff = dst[j] - src[j];
+#endif
+        const uint32_t error = diff * diff;
+        EXPECT_GE(1u << 2 * (bit_depth_ - 8), error)
+            << "Error: 8x8 IDCT has error " << error
+            << " at index " << j;
+      }
+    }
+  }
+
+  void RunFwdAccuracyCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 1000;
+    DECLARE_ALIGNED_ARRAY(16, int16_t, in, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff_r, kNumCoeffs);
+    DECLARE_ALIGNED_ARRAY(16, tran_low_t, coeff, kNumCoeffs);
+
+    for (int i = 0; i < count_test_block; ++i) {
+      double out_r[kNumCoeffs];
+
+      // Initialize a test block with input range [-mask_, mask_].
+      for (int j = 0; j < kNumCoeffs; ++j)
+        in[j] = rnd.Rand8() % 2 == 0 ? mask_ : -mask_;
+
+      RunFwdTxfm(in, coeff, pitch_);
+      reference_8x8_dct_2d(in, out_r);
+      for (int j = 0; j < kNumCoeffs; ++j)
+        coeff_r[j] = static_cast<tran_low_t>(round(out_r[j]));
+
+      for (int j = 0; j < kNumCoeffs; ++j) {
+        const uint32_t diff = coeff[j] - coeff_r[j];
+        const uint32_t error = diff * diff;
+        EXPECT_GE(9u << 2 * (bit_depth_ - 8), error)
+            << "Error: 8x8 DCT has error " << error
+            << " at index " << j;
+      }
+    }
+  }
   int pitch_;
   int tx_type_;
   FhtFunc fwd_txfm_ref;
+  vpx_bit_depth_t bit_depth_;
+  int mask_;
 };
 
 class FwdTrans8x8DCT
@@ -251,15 +446,17 @@
     tx_type_  = GET_PARAM(2);
     pitch_    = 8;
     fwd_txfm_ref = fdct8x8_ref;
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
   }
 
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride);
   }
-  void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride);
   }
 
@@ -279,6 +476,14 @@
   RunExtremalCheck();
 }
 
+TEST_P(FwdTrans8x8DCT, FwdAccuracyCheck) {
+  RunFwdAccuracyCheck();
+}
+
+TEST_P(FwdTrans8x8DCT, InvAccuracyCheck) {
+  RunInvAccuracyCheck();
+}
+
 class FwdTrans8x8HT
     : public FwdTrans8x8TestBase,
       public ::testing::TestWithParam<Ht8x8Param> {
@@ -291,15 +496,17 @@
     tx_type_  = GET_PARAM(2);
     pitch_    = 8;
     fwd_txfm_ref = fht8x8_ref;
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
   }
 
   virtual void TearDown() { libvpx_test::ClearSystemState(); }
 
  protected:
-  void RunFwdTxfm(int16_t *in, int16_t *out, int stride) {
+  void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
     fwd_txfm_(in, out, stride, tx_type_);
   }
-  void RunInvTxfm(int16_t *out, uint8_t *dst, int stride) {
+  void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
     inv_txfm_(out, dst, stride, tx_type_);
   }
 
@@ -321,50 +528,81 @@
 
 using std::tr1::make_tuple;
 
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, FwdTrans8x8DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct8x8_c, &vp9_idct8x8_64_add_c, 0)));
+        make_tuple(&vp9_high_fdct8x8_c, &idct8x8_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fdct8x8_c, &idct8x8_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_fdct8x8_c, &vp9_idct8x8_64_add_c, 0, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, FwdTrans8x8DCT,
+    ::testing::Values(
+        make_tuple(&vp9_fdct8x8_c, &vp9_idct8x8_64_add_c, 0, VPX_BITS_8)));
+#endif
+
+#if CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     C, FwdTrans8x8HT,
     ::testing::Values(
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 0),
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 1),
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 2),
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 3)));
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_10, 0, VPX_BITS_10),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_10, 1, VPX_BITS_10),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_10, 2, VPX_BITS_10),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_10, 3, VPX_BITS_10),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_12, 0, VPX_BITS_12),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_12, 1, VPX_BITS_12),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_12, 2, VPX_BITS_12),
+        make_tuple(&vp9_high_fht8x8_c, &iht8x8_12, 3, VPX_BITS_12),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 3, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, FwdTrans8x8HT,
+    ::testing::Values(
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 3, VPX_BITS_8)));
+#endif
 
-#if HAVE_NEON_ASM
+#if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     NEON, FwdTrans8x8DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct8x8_neon, &vp9_idct8x8_64_add_neon, 0)));
+        make_tuple(&vp9_fdct8x8_neon, &vp9_idct8x8_64_add_neon, 0,
+                   VPX_BITS_8)));
 INSTANTIATE_TEST_CASE_P(
     DISABLED_NEON, FwdTrans8x8HT,
     ::testing::Values(
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 0),
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 1),
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 2),
-        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 3)));
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 3, VPX_BITS_8)));
 #endif
 
-#if HAVE_SSE2
+#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSE2, FwdTrans8x8DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct8x8_sse2, &vp9_idct8x8_64_add_sse2, 0)));
+        make_tuple(&vp9_fdct8x8_sse2, &vp9_idct8x8_64_add_sse2, 0,
+                   VPX_BITS_8)));
 INSTANTIATE_TEST_CASE_P(
     SSE2, FwdTrans8x8HT,
     ::testing::Values(
-        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 0),
-        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 1),
-        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 2),
-        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 3)));
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 3, VPX_BITS_8)));
 #endif
 
-#if HAVE_SSSE3 && ARCH_X86_64
+#if HAVE_SSSE3 && ARCH_X86_64 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSSE3, FwdTrans8x8DCT,
     ::testing::Values(
-        make_tuple(&vp9_fdct8x8_ssse3, &vp9_idct8x8_64_add_ssse3, 0)));
+        make_tuple(&vp9_fdct8x8_ssse3, &vp9_idct8x8_64_add_ssse3, 0,
+                   VPX_BITS_8)));
 #endif
 }  // namespace
diff --git a/test/idct8x8_test.cc b/test/idct8x8_test.cc
index 5f4c33a..f488cb4 100644
--- a/test/idct8x8_test.cc
+++ b/test/idct8x8_test.cc
@@ -109,7 +109,8 @@
   ACMRandom rnd(ACMRandom::DeterministicSeed());
   const int count_test_block = 10000;
   for (int i = 0; i < count_test_block; ++i) {
-    int16_t input[64], coeff[64];
+    int16_t input[64];
+    tran_low_t coeff[64];
     double output_r[64];
     uint8_t dst[64], src[64];
 
diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 50e7c23..b61d490 100644
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -118,7 +118,9 @@
   {1, "invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf"},
   {1, "invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf"},
   {1, "invalid-vp90-2-05-resize.ivf.s59293_r01-05_b6-.ivf"},
-  {1, "invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf"},
+  {1, "invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf"},
+  {1, "invalid-vp91-2-mixedrefcsp-444to420.ivf"},
+  {1, "invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf"},
 };
 
 VP9_INSTANTIATE_TEST_CASE(InvalidFileTest,
@@ -149,8 +151,8 @@
   {4, "invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm"},
   {4, "invalid-"
       "vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf"},
-  {2, "invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf"},
-  {4, "invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf"},
+  {2, "invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf"},
+  {4, "invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf"},
 };
 
 INSTANTIATE_TEST_CASE_P(
diff --git a/test/lpf_8_test.cc b/test/lpf_8_test.cc
new file mode 100644
index 0000000..abc4107
--- /dev/null
+++ b/test/lpf_8_test.cc
@@ -0,0 +1,587 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <cmath>
+#include <cstdlib>
+#include <string>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_integer.h"
+
+using libvpx_test::ACMRandom;
+
+namespace {
+// Horizontally and Vertically need 32x32: 8  Coeffs preceeding filtered section
+//                                         16 Coefs within filtered section
+//                                         8  Coeffs following filtered section
+const int kNumCoeffs = 1024;
+
+const int number_of_iterations = 10000;
+
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*loop_op_t)(uint16_t *s, int p, const uint8_t *blimit,
+                          const uint8_t *limit, const uint8_t *thresh,
+                          int count, int bd);
+typedef void (*dual_loop_op_t)(uint16_t *s, int p, const uint8_t *blimit0,
+                               const uint8_t *limit0, const uint8_t *thresh0,
+                               const uint8_t *blimit1, const uint8_t *limit1,
+                               const uint8_t *thresh1, int bd);
+#else
+typedef void (*loop_op_t)(uint8_t *s, int p, const uint8_t *blimit,
+                          const uint8_t *limit, const uint8_t *thresh,
+                          int count);
+typedef void (*dual_loop_op_t)(uint8_t *s, int p, const uint8_t *blimit0,
+                               const uint8_t *limit0, const uint8_t *thresh0,
+                               const uint8_t *blimit1, const uint8_t *limit1,
+                               const uint8_t *thresh1);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+typedef std::tr1::tuple<loop_op_t, loop_op_t, int> loop8_param_t;
+typedef std::tr1::tuple<dual_loop_op_t, dual_loop_op_t, int> dualloop8_param_t;
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+void wrapper_vertical_16_sse2(uint16_t *s, int p, const uint8_t *blimit,
+                              const uint8_t *limit, const uint8_t *thresh,
+                              int count, int bd) {
+  vp9_highbd_lpf_vertical_16_sse2(s, p, blimit, limit, thresh, bd);
+}
+
+void wrapper_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit,
+                           const uint8_t *limit, const uint8_t *thresh,
+                           int count, int bd) {
+  vp9_highbd_lpf_vertical_16_c(s, p, blimit, limit, thresh, bd);
+}
+
+void wrapper_vertical_16_dual_sse2(uint16_t *s, int p, const uint8_t *blimit,
+                                   const uint8_t *limit, const uint8_t *thresh,
+                                   int count, int bd) {
+  vp9_highbd_lpf_vertical_16_dual_sse2(s, p, blimit, limit, thresh, bd);
+}
+
+void wrapper_vertical_16_dual_c(uint16_t *s, int p, const uint8_t *blimit,
+                                const uint8_t *limit, const uint8_t *thresh,
+                                int count, int bd) {
+  vp9_highbd_lpf_vertical_16_dual_c(s, p, blimit, limit, thresh, bd);
+}
+#else
+void wrapper_vertical_16_sse2(uint8_t *s, int p, const uint8_t *blimit,
+                              const uint8_t *limit, const uint8_t *thresh,
+                              int count) {
+  vp9_lpf_vertical_16_sse2(s, p, blimit, limit, thresh);
+}
+
+void wrapper_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
+                           const uint8_t *limit, const uint8_t *thresh,
+                           int count) {
+  vp9_lpf_vertical_16_c(s, p, blimit, limit, thresh);
+}
+
+void wrapper_vertical_16_dual_sse2(uint8_t *s, int p, const uint8_t *blimit,
+                                   const uint8_t *limit, const uint8_t *thresh,
+                                   int count) {
+  vp9_lpf_vertical_16_dual_sse2(s, p, blimit, limit, thresh);
+}
+
+void wrapper_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
+                                const uint8_t *limit, const uint8_t *thresh,
+                                int count) {
+  vp9_lpf_vertical_16_dual_c(s, p, blimit, limit, thresh);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_SSE2
+
+class Loop8Test6Param : public ::testing::TestWithParam<loop8_param_t> {
+ public:
+  virtual ~Loop8Test6Param() {}
+  virtual void SetUp() {
+    loopfilter_op_ = GET_PARAM(0);
+    ref_loopfilter_op_ = GET_PARAM(1);
+    bit_depth_ = GET_PARAM(2);
+    mask_ = (1 << bit_depth_) - 1;
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  int bit_depth_;
+  int mask_;
+  loop_op_t loopfilter_op_;
+  loop_op_t ref_loopfilter_op_;
+};
+
+class Loop8Test9Param : public ::testing::TestWithParam<dualloop8_param_t> {
+ public:
+  virtual ~Loop8Test9Param() {}
+  virtual void SetUp() {
+    loopfilter_op_ = GET_PARAM(0);
+    ref_loopfilter_op_ = GET_PARAM(1);
+    bit_depth_ = GET_PARAM(2);
+    mask_ = (1 << bit_depth_) - 1;
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  int bit_depth_;
+  int mask_;
+  dual_loop_op_t loopfilter_op_;
+  dual_loop_op_t ref_loopfilter_op_;
+};
+
+TEST_P(Loop8Test6Param, OperationCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = number_of_iterations;
+#if CONFIG_VP9_HIGHBITDEPTH
+  int32_t bd = bit_depth_;
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_s, kNumCoeffs);
+#else
+  DECLARE_ALIGNED_ARRAY(8, uint8_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(8, uint8_t, ref_s, kNumCoeffs);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < count_test_block; ++i) {
+    int err_count = 0;
+    uint8_t tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, blimit[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, limit[16])  = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, thresh[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    int32_t p = kNumCoeffs/32;
+    int count = 1;
+
+    uint16_t tmp_s[kNumCoeffs];
+    int j = 0;
+    while (j < kNumCoeffs) {
+      uint8_t val = rnd.Rand8();
+      if (val & 0x80) {  // 50% chance to choose a new value.
+        tmp_s[j] = rnd.Rand16();
+        j++;
+      } else {  // 50% chance to repeat previous value in row X times
+        int k = 0;
+        while (k++ < ((val & 0x1f) + 1) && j < kNumCoeffs) {
+          if (j < 1) {
+            tmp_s[j] = rnd.Rand16();
+          } else if (val & 0x20) {  // Increment by an value within the limit
+            tmp_s[j] = (tmp_s[j - 1] + (*limit - 1));
+          } else {  // Decrement by an value within the limit
+            tmp_s[j] = (tmp_s[j - 1] - (*limit - 1));
+          }
+          j++;
+        }
+      }
+    }
+    for (j = 0; j < kNumCoeffs; j++) {
+      if (i % 2) {
+        s[j] = tmp_s[j] & mask_;
+      } else {
+        s[j] = tmp_s[p * (j % p) + j / p] & mask_;
+      }
+      ref_s[j] = s[j];
+    }
+#if CONFIG_VP9_HIGHBITDEPTH
+    ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, count, bd);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count, bd));
+#else
+    ref_loopfilter_op_(ref_s+8+p*8, p, blimit, limit, thresh, count);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+    for (int j = 0; j < kNumCoeffs; ++j) {
+      err_count += ref_s[j] != s[j];
+    }
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Loop8Test6Param, C output doesn't match SSE2 "
+         "loopfilter output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(Loop8Test6Param, ValueCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = number_of_iterations;
+#if CONFIG_VP9_HIGHBITDEPTH
+  const int32_t bd = bit_depth_;
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_s, kNumCoeffs);
+#else
+  DECLARE_ALIGNED_ARRAY(8, uint8_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(8, uint8_t, ref_s, kNumCoeffs);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < count_test_block; ++i) {
+    int err_count = 0;
+    uint8_t tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, blimit[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, limit[16])  = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, thresh[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    int32_t p = kNumCoeffs / 32;
+    int count = 1;
+    for (int j = 0; j < kNumCoeffs; ++j) {
+      s[j] = rnd.Rand16() & mask_;
+      ref_s[j] = s[j];
+    }
+#if CONFIG_VP9_HIGHBITDEPTH
+    ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit, limit, thresh, count, bd);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count, bd));
+#else
+    ref_loopfilter_op_(ref_s+8+p*8, p, blimit, limit, thresh, count);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit, limit, thresh, count));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+    for (int j = 0; j < kNumCoeffs; ++j) {
+      err_count += ref_s[j] != s[j];
+    }
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Loop8Test6Param, C output doesn't match SSE2 "
+         "loopfilter output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(Loop8Test9Param, OperationCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = number_of_iterations;
+#if CONFIG_VP9_HIGHBITDEPTH
+  const int32_t bd = bit_depth_;
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_s, kNumCoeffs);
+#else
+  DECLARE_ALIGNED_ARRAY(8,  uint8_t,  s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(8,  uint8_t,  ref_s, kNumCoeffs);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < count_test_block; ++i) {
+    int err_count = 0;
+    uint8_t tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, blimit0[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, limit0[16])  = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, thresh0[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, blimit1[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, limit1[16])  = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, thresh1[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    int32_t p = kNumCoeffs / 32;
+    uint16_t tmp_s[kNumCoeffs];
+    int j = 0;
+    const uint8_t limit = *limit0 < *limit1 ? *limit0 : *limit1;
+    while (j < kNumCoeffs) {
+      uint8_t val = rnd.Rand8();
+      if (val & 0x80) {  // 50% chance to choose a new value.
+        tmp_s[j] = rnd.Rand16();
+        j++;
+      } else {  // 50% chance to repeat previous value in row X times.
+        int k = 0;
+        while (k++ < ((val & 0x1f) + 1) && j < kNumCoeffs) {
+          if (j < 1) {
+            tmp_s[j] = rnd.Rand16();
+          } else if (val & 0x20) {  // Increment by a value within the limit.
+            tmp_s[j] = (tmp_s[j - 1] + (limit - 1));
+          } else {  // Decrement by an value within the limit.
+            tmp_s[j] = (tmp_s[j - 1] - (limit - 1));
+          }
+          j++;
+        }
+      }
+    }
+    for (j = 0; j < kNumCoeffs; j++) {
+      if (i % 2) {
+        s[j] = tmp_s[j] & mask_;
+      } else {
+        s[j] = tmp_s[p * (j % p) + j / p] & mask_;
+      }
+      ref_s[j] = s[j];
+    }
+#if CONFIG_VP9_HIGHBITDEPTH
+    ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1, bd);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1, bd));
+#else
+    ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+    for (int j = 0; j < kNumCoeffs; ++j) {
+      err_count += ref_s[j] != s[j];
+    }
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Loop8Test9Param, C output doesn't match SSE2 "
+         "loopfilter output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(Loop8Test9Param, ValueCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = number_of_iterations;
+#if CONFIG_VP9_HIGHBITDEPTH
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_s, kNumCoeffs);
+#else
+  DECLARE_ALIGNED_ARRAY(8,  uint8_t, s, kNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(8,  uint8_t, ref_s, kNumCoeffs);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  int err_count_total = 0;
+  int first_failure = -1;
+  for (int i = 0; i < count_test_block; ++i) {
+    int err_count = 0;
+    uint8_t tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, blimit0[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, limit0[16])  = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, thresh0[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, blimit1[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, limit1[16])  = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    tmp = rnd.Rand8();
+    DECLARE_ALIGNED(16, const uint8_t, thresh1[16]) = {
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
+        tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp
+    };
+    int32_t p = kNumCoeffs / 32;  // TODO(pdlf) can we have non-square here?
+    for (int j = 0; j < kNumCoeffs; ++j) {
+      s[j] = rnd.Rand16() & mask_;
+      ref_s[j] = s[j];
+    }
+#if CONFIG_VP9_HIGHBITDEPTH
+    const int32_t bd = bit_depth_;
+    ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1, bd);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit0, limit0,
+                       thresh0, blimit1, limit1, thresh1, bd));
+#else
+    ref_loopfilter_op_(ref_s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1);
+    ASM_REGISTER_STATE_CHECK(
+        loopfilter_op_(s + 8 + p * 8, p, blimit0, limit0, thresh0,
+                       blimit1, limit1, thresh1));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+    for (int j = 0; j < kNumCoeffs; ++j) {
+      err_count += ref_s[j] != s[j];
+    }
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Loop8Test9Param, C output doesn't match SSE2"
+         "loopfilter output. "
+      << "First failed at test case " << first_failure;
+}
+
+using std::tr1::make_tuple;
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    SSE2_C_COMPARE_SINGLE, Loop8Test6Param,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_lpf_horizontal_4_sse2,
+                   &vp9_highbd_lpf_horizontal_4_c, 8),
+        make_tuple(&vp9_highbd_lpf_vertical_4_sse2,
+                   &vp9_highbd_lpf_vertical_4_c, 8),
+        make_tuple(&vp9_highbd_lpf_horizontal_8_sse2,
+                   &vp9_highbd_lpf_horizontal_8_c, 8),
+        make_tuple(&vp9_highbd_lpf_horizontal_16_sse2,
+                   &vp9_highbd_lpf_horizontal_16_c, 8),
+        make_tuple(&vp9_highbd_lpf_vertical_8_sse2,
+                   &vp9_highbd_lpf_vertical_8_c, 8),
+        make_tuple(&wrapper_vertical_16_sse2,
+                   &wrapper_vertical_16_c, 8),
+        make_tuple(&vp9_highbd_lpf_horizontal_4_sse2,
+                   &vp9_highbd_lpf_horizontal_4_c, 10),
+        make_tuple(&vp9_highbd_lpf_vertical_4_sse2,
+                   &vp9_highbd_lpf_vertical_4_c, 10),
+        make_tuple(&vp9_highbd_lpf_horizontal_8_sse2,
+                   &vp9_highbd_lpf_horizontal_8_c, 10),
+        make_tuple(&vp9_highbd_lpf_horizontal_16_sse2,
+                   &vp9_highbd_lpf_horizontal_16_c, 10),
+        make_tuple(&vp9_highbd_lpf_vertical_8_sse2,
+                   &vp9_highbd_lpf_vertical_8_c, 10),
+        make_tuple(&wrapper_vertical_16_sse2,
+                   &wrapper_vertical_16_c, 10),
+        make_tuple(&vp9_highbd_lpf_horizontal_4_sse2,
+                   &vp9_highbd_lpf_horizontal_4_c, 12),
+        make_tuple(&vp9_highbd_lpf_vertical_4_sse2,
+                   &vp9_highbd_lpf_vertical_4_c, 12),
+        make_tuple(&vp9_highbd_lpf_horizontal_8_sse2,
+                   &vp9_highbd_lpf_horizontal_8_c, 12),
+        make_tuple(&vp9_highbd_lpf_horizontal_16_sse2,
+                   &vp9_highbd_lpf_horizontal_16_c, 12),
+        make_tuple(&vp9_highbd_lpf_vertical_8_sse2,
+                   &vp9_highbd_lpf_vertical_8_c, 12),
+        make_tuple(&wrapper_vertical_16_sse2,
+                   &wrapper_vertical_16_c, 12)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    SSE2_C_COMPARE_SINGLE, Loop8Test6Param,
+    ::testing::Values(
+        make_tuple(&vp9_lpf_horizontal_8_sse2, &vp9_lpf_horizontal_8_c, 8),
+        make_tuple(&vp9_lpf_horizontal_16_sse2, &vp9_lpf_horizontal_16_c, 8),
+        make_tuple(&vp9_lpf_vertical_8_sse2, &vp9_lpf_vertical_8_c, 8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    SSE2_C_COMPARE_DUAL, Loop8Test6Param,
+    ::testing::Values(
+        make_tuple(&wrapper_vertical_16_dual_sse2,
+                   &wrapper_vertical_16_dual_c, 8),
+        make_tuple(&wrapper_vertical_16_dual_sse2,
+                   &wrapper_vertical_16_dual_c, 10),
+        make_tuple(&wrapper_vertical_16_dual_sse2,
+                   &wrapper_vertical_16_dual_c, 12)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    SSE2_C_COMPARE_DUAL, Loop8Test6Param,
+    ::testing::Values(
+        make_tuple(&wrapper_vertical_16_sse2, &wrapper_vertical_16_c, 8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_SSE2
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    SSE_C_COMPARE_DUAL, Loop8Test9Param,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_lpf_horizontal_4_dual_sse2,
+                   &vp9_highbd_lpf_horizontal_4_dual_c, 8),
+        make_tuple(&vp9_highbd_lpf_horizontal_8_dual_sse2,
+                   &vp9_highbd_lpf_horizontal_8_dual_c, 8),
+        make_tuple(&vp9_highbd_lpf_vertical_4_dual_sse2,
+                   &vp9_highbd_lpf_vertical_4_dual_c, 8),
+        make_tuple(&vp9_highbd_lpf_vertical_8_dual_sse2,
+                   &vp9_highbd_lpf_vertical_8_dual_c, 8),
+        make_tuple(&vp9_highbd_lpf_horizontal_4_dual_sse2,
+                   &vp9_highbd_lpf_horizontal_4_dual_c, 10),
+        make_tuple(&vp9_highbd_lpf_horizontal_8_dual_sse2,
+                   &vp9_highbd_lpf_horizontal_8_dual_c, 10),
+        make_tuple(&vp9_highbd_lpf_vertical_4_dual_sse2,
+                   &vp9_highbd_lpf_vertical_4_dual_c, 10),
+        make_tuple(&vp9_highbd_lpf_vertical_8_dual_sse2,
+                   &vp9_highbd_lpf_vertical_8_dual_c, 10),
+        make_tuple(&vp9_highbd_lpf_horizontal_4_dual_sse2,
+                   &vp9_highbd_lpf_horizontal_4_dual_c, 12),
+        make_tuple(&vp9_highbd_lpf_horizontal_8_dual_sse2,
+                   &vp9_highbd_lpf_horizontal_8_dual_c, 12),
+        make_tuple(&vp9_highbd_lpf_vertical_4_dual_sse2,
+                   &vp9_highbd_lpf_vertical_4_dual_c, 12),
+        make_tuple(&vp9_highbd_lpf_vertical_8_dual_sse2,
+                   &vp9_highbd_lpf_vertical_8_dual_c, 12)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    SSE_C_COMPARE_DUAL, Loop8Test9Param,
+    ::testing::Values(
+        make_tuple(&vp9_lpf_horizontal_4_dual_sse2,
+                   &vp9_lpf_horizontal_4_dual_c, 8),
+        make_tuple(&vp9_lpf_horizontal_8_dual_sse2,
+                   &vp9_lpf_horizontal_8_dual_c, 8),
+        make_tuple(&vp9_lpf_vertical_4_dual_sse2,
+                   &vp9_lpf_vertical_4_dual_c, 8),
+        make_tuple(&vp9_lpf_vertical_8_dual_sse2,
+                   &vp9_lpf_vertical_8_dual_c, 8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif
+}  // namespace
diff --git a/test/partial_idct_test.cc b/test/partial_idct_test.cc
index 15f4e6c..9c24fee 100644
--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -26,8 +26,8 @@
 using libvpx_test::ACMRandom;
 
 namespace {
-typedef void (*FwdTxfmFunc)(const int16_t *in, int16_t *out, int stride);
-typedef void (*InvTxfmFunc)(const int16_t *in, uint8_t *out, int stride);
+typedef void (*FwdTxfmFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*InvTxfmFunc)(const tran_low_t *in, uint8_t *out, int stride);
 typedef std::tr1::tuple<FwdTxfmFunc,
                         InvTxfmFunc,
                         InvTxfmFunc,
@@ -74,8 +74,8 @@
       FAIL() << "Wrong Size!";
       break;
   }
-  DECLARE_ALIGNED_ARRAY(16, int16_t, test_coef_block1, kMaxNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, test_coef_block2, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_coef_block1, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_coef_block2, kMaxNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, dst1, kMaxNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, dst2, kMaxNumCoeffs);
 
@@ -83,7 +83,7 @@
   const int block_size = size * size;
 
   DECLARE_ALIGNED_ARRAY(16, int16_t, input_extreme_block, kMaxNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, output_ref_block, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, output_ref_block, kMaxNumCoeffs);
 
   int max_error = 0;
   for (int i = 0; i < count_test_block; ++i) {
@@ -153,8 +153,8 @@
       FAIL() << "Wrong Size!";
       break;
   }
-  DECLARE_ALIGNED_ARRAY(16, int16_t, test_coef_block1, kMaxNumCoeffs);
-  DECLARE_ALIGNED_ARRAY(16, int16_t, test_coef_block2, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_coef_block1, kMaxNumCoeffs);
+  DECLARE_ALIGNED_ARRAY(16, tran_low_t, test_coef_block2, kMaxNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, dst1, kMaxNumCoeffs);
   DECLARE_ALIGNED_ARRAY(16, uint8_t, dst2, kMaxNumCoeffs);
   const int count_test_block = 1000;
@@ -229,6 +229,7 @@
                    &vp9_idct4x4_16_add_c,
                    &vp9_idct4x4_1_add_c,
                    TX_4X4, 1)));
+
 #if HAVE_NEON_ASM
 INSTANTIATE_TEST_CASE_P(
     NEON, PartialIDctTest,
@@ -259,7 +260,7 @@
                    TX_4X4, 1)));
 #endif
 
-#if HAVE_SSE2
+#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSE2, PartialIDctTest,
     ::testing::Values(
@@ -293,7 +294,7 @@
                    TX_4X4, 1)));
 #endif
 
-#if HAVE_SSSE3 && ARCH_X86_64
+#if HAVE_SSSE3 && ARCH_X86_64 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSSE3_64, PartialIDctTest,
     ::testing::Values(
@@ -303,7 +304,7 @@
                    TX_8X8, 12)));
 #endif
 
-#if HAVE_SSSE3
+#if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
     SSSE3, PartialIDctTest,
     ::testing::Values(
diff --git a/test/svc_test.cc b/test/svc_test.cc
index 218f53d..fdde702 100644
--- a/test/svc_test.cc
+++ b/test/svc_test.cc
@@ -74,6 +74,7 @@
     const vpx_codec_err_t res =
         vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
     EXPECT_EQ(VPX_CODEC_OK, res);
+    vpx_codec_control(&codec_, VP8E_SET_CPUUSED, 4);  // Make the test faster
     codec_initialized_ = true;
   }
 
@@ -83,11 +84,23 @@
     codec_initialized_ = false;
   }
 
+  void GetStatsData(std::string *const stats_buf) {
+    vpx_codec_iter_t iter = NULL;
+    const vpx_codec_cx_pkt_t *cx_pkt;
+
+    while ((cx_pkt = vpx_codec_get_cx_data(&codec_, &iter)) != NULL) {
+      if (cx_pkt->kind == VPX_CODEC_STATS_PKT) {
+        EXPECT_GT(cx_pkt->data.twopass_stats.sz, 0U);
+        ASSERT_TRUE(cx_pkt->data.twopass_stats.buf != NULL);
+        stats_buf->append(static_cast<char*>(cx_pkt->data.twopass_stats.buf),
+                          cx_pkt->data.twopass_stats.sz);
+      }
+    }
+  }
+
   void Pass1EncodeNFrames(const int n, const int layers,
                           std::string *const stats_buf) {
     vpx_codec_err_t res;
-    size_t stats_size = 0;
-    const char *stats_data = NULL;
 
     ASSERT_GT(n, 0);
     ASSERT_GT(layers, 0);
@@ -104,22 +117,15 @@
       res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
                            video.duration(), VPX_DL_GOOD_QUALITY);
       ASSERT_EQ(VPX_CODEC_OK, res);
-      stats_size = vpx_svc_get_rc_stats_buffer_size(&svc_);
-      EXPECT_GT(stats_size, 0U);
-      stats_data = vpx_svc_get_rc_stats_buffer(&svc_);
-      ASSERT_TRUE(stats_data != NULL);
-      stats_buf->append(stats_data, stats_size);
+      GetStatsData(stats_buf);
       video.Next();
     }
 
     // Flush encoder and test EOS packet.
     res = vpx_svc_encode(&svc_, &codec_, NULL, video.pts(),
                          video.duration(), VPX_DL_GOOD_QUALITY);
-    stats_size = vpx_svc_get_rc_stats_buffer_size(&svc_);
-    EXPECT_GT(stats_size, 0U);
-    stats_data = vpx_svc_get_rc_stats_buffer(&svc_);
-    ASSERT_TRUE(stats_data != NULL);
-    stats_buf->append(stats_data, stats_size);
+    ASSERT_EQ(VPX_CODEC_OK, res);
+    GetStatsData(stats_buf);
 
     ReleaseEncoder();
   }
@@ -127,20 +133,27 @@
   void StoreFrames(const size_t max_frame_received,
                    struct vpx_fixed_buf *const outputs,
                    size_t *const frame_received) {
-    size_t frame_size;
-    while ((frame_size = vpx_svc_get_frame_size(&svc_)) > 0) {
-      ASSERT_LT(*frame_received, max_frame_received);
+    vpx_codec_iter_t iter = NULL;
+    const vpx_codec_cx_pkt_t *cx_pkt;
 
-      if (*frame_received == 0) {
-        EXPECT_EQ(1, vpx_svc_is_keyframe(&svc_));
+    while ((cx_pkt = vpx_codec_get_cx_data(&codec_, &iter)) != NULL) {
+      if (cx_pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
+        const size_t frame_size = cx_pkt->data.frame.sz;
+
+        EXPECT_GT(frame_size, 0U);
+        ASSERT_TRUE(cx_pkt->data.frame.buf != NULL);
+        ASSERT_LT(*frame_received, max_frame_received);
+
+        if (*frame_received == 0)
+          EXPECT_EQ(1, !!(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY));
+
+        outputs[*frame_received].buf = malloc(frame_size + 16);
+        ASSERT_TRUE(outputs[*frame_received].buf != NULL);
+        memcpy(outputs[*frame_received].buf, cx_pkt->data.frame.buf,
+               frame_size);
+        outputs[*frame_received].sz = frame_size;
+        ++(*frame_received);
       }
-
-      outputs[*frame_received].buf = malloc(frame_size + 16);
-      ASSERT_TRUE(outputs[*frame_received].buf != NULL);
-      memcpy(outputs[*frame_received].buf, vpx_svc_get_buffer(&svc_),
-             frame_size);
-      outputs[*frame_received].sz = frame_size;
-      ++(*frame_received);
     }
   }
 
@@ -386,11 +399,6 @@
 
 TEST_F(SvcTest, InitTwoLayers) {
   svc_.spatial_layers = 2;
-  vpx_svc_set_scale_factors(&svc_, "4/16,16*16");  // invalid scale values
-  vpx_codec_err_t res = vpx_svc_init(&svc_, &codec_, codec_iface_, &codec_enc_);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  vpx_svc_set_scale_factors(&svc_, "4/16,16/16");  // valid scale values
   InitializeEncoder();
 }
 
@@ -427,6 +435,16 @@
   res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
   EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
 
+  res = vpx_svc_set_options(&svc_, "scale-factors=1/3, 3*3");
+  EXPECT_EQ(VPX_CODEC_OK, res);
+  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
+  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
+
+  res = vpx_svc_set_options(&svc_, "scale-factors=1/3");
+  EXPECT_EQ(VPX_CODEC_OK, res);
+  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
+  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
+
   res = vpx_svc_set_options(&svc_, "scale-factors=1/3,2/3");
   EXPECT_EQ(VPX_CODEC_OK, res);
   InitializeEncoder();
@@ -434,12 +452,32 @@
 
 TEST_F(SvcTest, SetQuantizersOption) {
   svc_.spatial_layers = 2;
-  vpx_codec_err_t res = vpx_svc_set_options(&svc_, "quantizers=not-quantizers");
+  vpx_codec_err_t res = vpx_svc_set_options(&svc_, "max-quantizers=nothing");
   EXPECT_EQ(VPX_CODEC_OK, res);
   res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
   EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
 
-  vpx_svc_set_options(&svc_, "quantizers=40,45");
+  res = vpx_svc_set_options(&svc_, "min-quantizers=nothing");
+  EXPECT_EQ(VPX_CODEC_OK, res);
+  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
+  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
+
+  res = vpx_svc_set_options(&svc_, "max-quantizers=40");
+  EXPECT_EQ(VPX_CODEC_OK, res);
+  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
+  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
+
+  res = vpx_svc_set_options(&svc_, "min-quantizers=40");
+  EXPECT_EQ(VPX_CODEC_OK, res);
+  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
+  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
+
+  res = vpx_svc_set_options(&svc_, "max-quantizers=30,30 min-quantizers=40,40");
+  EXPECT_EQ(VPX_CODEC_OK, res);
+  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
+  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
+
+  res = vpx_svc_set_options(&svc_, "max-quantizers=40,40 min-quantizers=30,30");
   InitializeEncoder();
 }
 
@@ -459,42 +497,6 @@
   InitializeEncoder();
 }
 
-TEST_F(SvcTest, SetQuantizers) {
-  vpx_codec_err_t res = vpx_svc_set_quantizers(NULL, "40,30");
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_set_quantizers(&svc_, NULL);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  svc_.spatial_layers = 2;
-  res = vpx_svc_set_quantizers(&svc_, "40");
-  EXPECT_EQ(VPX_CODEC_OK, res);
-  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_set_quantizers(&svc_, "40,30");
-  EXPECT_EQ(VPX_CODEC_OK, res);
-  InitializeEncoder();
-}
-
-TEST_F(SvcTest, SetScaleFactors) {
-  vpx_codec_err_t res = vpx_svc_set_scale_factors(NULL, "4/16,16/16");
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_set_scale_factors(&svc_, NULL);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  svc_.spatial_layers = 2;
-  res = vpx_svc_set_scale_factors(&svc_, "4/16");
-  EXPECT_EQ(VPX_CODEC_OK, res);
-  res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_set_scale_factors(&svc_, "4/16,16/16");
-  EXPECT_EQ(VPX_CODEC_OK, res);
-  InitializeEncoder();
-}
-
 // Test that decoder can handle an SVC frame as the first frame in a sequence.
 TEST_F(SvcTest, OnePassEncodeOneFrame) {
   codec_enc_.g_pass = VPX_RC_ONE_PASS;
@@ -513,39 +515,6 @@
   FreeBitstreamBuffers(&outputs[0], 3);
 }
 
-TEST_F(SvcTest, GetLayerResolution) {
-  svc_.spatial_layers = 2;
-  vpx_svc_set_scale_factors(&svc_, "4/16,8/16");
-  vpx_svc_set_quantizers(&svc_, "40,30");
-
-  InitializeEncoder();
-
-  // ensure that requested layer is a valid layer
-  uint32_t layer_width, layer_height;
-  vpx_codec_err_t res = vpx_svc_get_layer_resolution(&svc_, svc_.spatial_layers,
-                                     &layer_width, &layer_height);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_get_layer_resolution(NULL, 0, &layer_width, &layer_height);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_get_layer_resolution(&svc_, 0, NULL, &layer_height);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_get_layer_resolution(&svc_, 0, &layer_width, NULL);
-  EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
-  res = vpx_svc_get_layer_resolution(&svc_, 0, &layer_width, &layer_height);
-  EXPECT_EQ(VPX_CODEC_OK, res);
-  EXPECT_EQ(kWidth * 4 / 16, layer_width);
-  EXPECT_EQ(kHeight * 4 / 16, layer_height);
-
-  res = vpx_svc_get_layer_resolution(&svc_, 1, &layer_width, &layer_height);
-  EXPECT_EQ(VPX_CODEC_OK, res);
-  EXPECT_EQ(kWidth * 8 / 16, layer_width);
-  EXPECT_EQ(kHeight * 8 / 16, layer_height);
-}
-
 TEST_F(SvcTest, TwoPassEncode10Frames) {
   // First pass encode
   std::string stats_buf;
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 84b13f9..e6114ab 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -683,10 +683,10 @@
 717da707afcaa1f692ff1946f291054eb75a4f06  screendata.y4m
 b7c1296630cdf1a7ef493d15ff4f9eb2999202f6  invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf
 0a3884edb3fd8f9d9b500223e650f7de257b67d8  invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
-fac89b5735be8a86b0dc05159f996a5c3208ae32  invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
-22e0ee8babe574722baf4ef6d7ff5d7cf80d386c  invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
-4506dfdcdf8ee4250924b075a0dcf1f070f72e5a  invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
-d3ea592c8d7b05d14c7ed48befc0a3aaf7709b7a  invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
+fac89b5735be8a86b0dc05159f996a5c3208ae32  invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf
+0a3884edb3fd8f9d9b500223e650f7de257b67d8  invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf.res
+4506dfdcdf8ee4250924b075a0dcf1f070f72e5a  invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf
+bcdedaf168ac225575468fda77502d2dc9fd5baa  invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf.res
 65e93f9653bcf65b022f7d225268d1a90a76e7bb  vp90-2-19-skip.webm
 368dccdde5288c13c25695d2eacdc7402cadf613  vp90-2-19-skip.webm.md5
 ffe460282df2b0e7d4603c2158653ad96f574b02  vp90-2-19-skip-01.webm
@@ -695,5 +695,9 @@
 0a3884edb3fd8f9d9b500223e650f7de257b67d8  invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf.res
 5e67e24e7f53fd189e565513cef8519b1bd6c712  invalid-vp90-2-05-resize.ivf.s59293_r01-05_b6-.ivf
 741158f67c0d9d23726624d06bdc482ad368afc9  invalid-vp90-2-05-resize.ivf.s59293_r01-05_b6-.ivf.res
-8b1f7bf7e86c0976d277f60e8fcd9539e75a079a  invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf
-fb79dcbbbb8c82d5a750e339acce66e39a32f15f  invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf.res
+8b1f7bf7e86c0976d277f60e8fcd9539e75a079a  invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf
+9c6bdf048fb2e66f07d4b4db5b32e6f303bd6109  invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf.res
+552e372e9b78127389fb06b34545df2cec15ba6d  invalid-vp91-2-mixedrefcsp-444to420.ivf
+a61774cf03fc584bd9f0904fc145253bb8ea6c4c  invalid-vp91-2-mixedrefcsp-444to420.ivf.res
+812d05a64a0d83c1b504d0519927ddc5a2cdb273  invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf
+1e472baaf5f6113459f0399a38a5a5e68d17799d  invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf.res
diff --git a/test/test.mk b/test/test.mk
index c839c92..df4969f 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -30,6 +30,7 @@
 LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += keyframe_test.cc
 
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += external_frame_buffer_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += invalid_file_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += user_priv_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += active_map_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += borders_test.cc
@@ -60,7 +61,6 @@
 endif
 
 LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += decode_api_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += invalid_file_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_DECODERS)    += test_vector_test.cc
 
 # Currently we only support decoder perf tests for vp9. Also they read from WebM
@@ -128,6 +128,8 @@
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += lpf_8_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9)         += vp9_intrapred_test.cc
 
 ifeq ($(CONFIG_VP9_ENCODER),yes)
 LIBVPX_TEST_SRCS-$(CONFIG_SPATIAL_SVC) += svc_test.cc
@@ -811,14 +813,18 @@
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm.res
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf.res
 
 ifeq ($(CONFIG_DECODE_PERF_TESTS),yes)
 # BBB VP9 streams
diff --git a/test/tools_common.sh b/test/tools_common.sh
index 15c7bce..4300738 100755
--- a/test/tools_common.sh
+++ b/test/tools_common.sh
@@ -200,11 +200,11 @@
   [ "$(vpx_config_option_enabled CONFIG_WEBM_IO)" = "yes" ] && echo yes
 }
 
-# Filters strings from positional parameter one using the filter specified by
-# positional parameter two. Filter behavior depends on the presence of a third
-# positional parameter. When parameter three is present, strings that match the
-# filter are excluded. When omitted, strings matching the filter are included.
-# The filtered string is echoed to stdout.
+# Filters strings from $1 using the filter specified by $2. Filter behavior
+# depends on the presence of $3. When $3 is present, strings that match the
+# filter are excluded. When $3 is omitted, strings matching the filter are
+# included.
+# The filtered result is echoed to stdout.
 filter_strings() {
   strings=${1}
   filter=${2}
@@ -253,6 +253,15 @@
     tests_to_filter=$(filter_strings "${tests_to_filter}" ${VPX_TEST_FILTER})
   fi
 
+  # User requested test listing: Dump test names and return.
+  if [ "${VPX_TEST_LIST_TESTS}" = "yes" ]; then
+    for test_name in $tests_to_filter; do
+      echo ${test_name}
+    done
+    return
+  fi
+
+  # Combine environment and actual tests.
   local tests_to_run="${env_tests} ${tests_to_filter}"
 
   check_git_hashes
@@ -283,6 +292,7 @@
     --prefix: Allows for a user specified prefix to be inserted before all test
               programs. Grants the ability, for example, to run test programs
               within valgrind.
+    --list-tests: List all test names and exit without actually running tests.
     --verbose: Verbose output.
 
     When the --bin-path option is not specified the script attempts to use
@@ -342,6 +352,9 @@
     --show-program-output)
       devnull=
       ;;
+    --list-tests)
+      VPX_TEST_LIST_TESTS=yes
+      ;;
     *)
       vpx_test_usage
       exit 1
@@ -401,6 +414,7 @@
   VP9_WEBM_FILE=${VP9_WEBM_FILE}
   VPX_TEST_EXE_SUFFIX=${VPX_TEST_EXE_SUFFIX}
   VPX_TEST_FILTER=${VPX_TEST_FILTER}
+  VPX_TEST_LIST_TESTS=${VPX_TEST_LIST_TESTS}
   VPX_TEST_OUTPUT_DIR=${VPX_TEST_OUTPUT_DIR}
   VPX_TEST_PREFIX=${VPX_TEST_PREFIX}
   VPX_TEST_RAND=${VPX_TEST_RAND}
diff --git a/test/variance_test.cc b/test/variance_test.cc
index f76402e..a438d17 100644
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
@@ -214,6 +214,99 @@
   EXPECT_EQ(expected, var);
 }
 
+#if CONFIG_VP8_ENCODER
+template<typename MseFunctionType>
+class MseTest
+    : public ::testing::TestWithParam<tuple<int, int, MseFunctionType> > {
+ public:
+  virtual void SetUp() {
+    const tuple<int, int, MseFunctionType>& params = this->GetParam();
+    log2width_  = get<0>(params);
+    width_ = 1 << log2width_;
+    log2height_ = get<1>(params);
+    height_ = 1 << log2height_;
+    mse_ = get<2>(params);
+
+    rnd(ACMRandom::DeterministicSeed());
+    block_size_ = width_ * height_;
+    src_ = reinterpret_cast<uint8_t *>(vpx_memalign(16, block_size_));
+    ref_ = new uint8_t[block_size_];
+    ASSERT_TRUE(src_ != NULL);
+    ASSERT_TRUE(ref_ != NULL);
+  }
+
+  virtual void TearDown() {
+    vpx_free(src_);
+    delete[] ref_;
+    libvpx_test::ClearSystemState();
+  }
+
+ protected:
+  void RefTest_mse();
+  void RefTest_sse();
+  void MaxTest_mse();
+  void MaxTest_sse();
+
+  ACMRandom rnd;
+  uint8_t* src_;
+  uint8_t* ref_;
+  int width_, log2width_;
+  int height_, log2height_;
+  int block_size_;
+  MseFunctionType mse_;
+};
+
+template<typename MseFunctionType>
+void MseTest<MseFunctionType>::RefTest_mse() {
+  for (int i = 0; i < 10; ++i) {
+    for (int j = 0; j < block_size_; j++) {
+      src_[j] = rnd.Rand8();
+      ref_[j] = rnd.Rand8();
+    }
+    unsigned int sse1, sse2;
+    ASM_REGISTER_STATE_CHECK(mse_(src_, width_, ref_, width_, &sse1));
+    variance_ref(src_, ref_, log2width_, log2height_, &sse2);
+    EXPECT_EQ(sse1, sse2);
+  }
+}
+
+template<typename MseFunctionType>
+void MseTest<MseFunctionType>::RefTest_sse() {
+  for (int i = 0; i < 10; ++i) {
+    for (int j = 0; j < block_size_; j++) {
+      src_[j] = rnd.Rand8();
+      ref_[j] = rnd.Rand8();
+    }
+    unsigned int sse2;
+    unsigned int var1;
+    ASM_REGISTER_STATE_CHECK(
+        var1 = mse_(src_, width_, ref_, width_));
+    variance_ref(src_, ref_, log2width_, log2height_, &sse2);
+    EXPECT_EQ(var1, sse2);
+  }
+}
+
+template<typename MseFunctionType>
+void MseTest<MseFunctionType>::MaxTest_mse() {
+  memset(src_, 255, block_size_);
+  memset(ref_, 0, block_size_);
+  unsigned int sse;
+  ASM_REGISTER_STATE_CHECK(mse_(src_, width_, ref_, width_, &sse));
+  const unsigned int expected = block_size_ * 255 * 255;
+  EXPECT_EQ(expected, sse);
+}
+
+template<typename MseFunctionType>
+void MseTest<MseFunctionType>::MaxTest_sse() {
+  memset(src_, 255, block_size_);
+  memset(ref_, 0, block_size_);
+  unsigned int var;
+  ASM_REGISTER_STATE_CHECK(var = mse_(src_, width_, ref_, width_));
+  const unsigned int expected = block_size_ * 255 * 255;
+  EXPECT_EQ(expected, var);
+}
+#endif
+
 #if CONFIG_VP9_ENCODER
 
 unsigned int subpel_avg_variance_ref(const uint8_t *ref,
@@ -343,12 +436,31 @@
 namespace vp8 {
 
 #if CONFIG_VP8_ENCODER
+typedef unsigned int (*vp8_sse_fn_t)(const unsigned char *src_ptr,
+    int source_stride, const unsigned char *ref_ptr, int  ref_stride);
+
+typedef MseTest<vp8_sse_fn_t> VP8SseTest;
+typedef MseTest<vp8_variance_fn_t> VP8MseTest;
 typedef VarianceTest<vp8_variance_fn_t> VP8VarianceTest;
 
+TEST_P(VP8SseTest, Ref_sse) { RefTest_sse(); }
+TEST_P(VP8SseTest, Max_sse) { MaxTest_sse(); }
+TEST_P(VP8MseTest, Ref_mse) { RefTest_mse(); }
+TEST_P(VP8MseTest, Max_mse) { MaxTest_mse(); }
 TEST_P(VP8VarianceTest, Zero) { ZeroTest(); }
 TEST_P(VP8VarianceTest, Ref) { RefTest(); }
 TEST_P(VP8VarianceTest, OneQuarter) { OneQuarterTest(); }
 
+const vp8_sse_fn_t get4x4sse_cs_c = vp8_get4x4sse_cs_c;
+INSTANTIATE_TEST_CASE_P(
+    C, VP8SseTest,
+    ::testing::Values(make_tuple(2, 2, get4x4sse_cs_c)));
+
+const vp8_variance_fn_t mse16x16_c = vp8_mse16x16_c;
+INSTANTIATE_TEST_CASE_P(
+    C, VP8MseTest,
+    ::testing::Values(make_tuple(4, 4, mse16x16_c)));
+
 const vp8_variance_fn_t variance4x4_c = vp8_variance4x4_c;
 const vp8_variance_fn_t variance8x8_c = vp8_variance8x8_c;
 const vp8_variance_fn_t variance8x16_c = vp8_variance8x16_c;
@@ -363,6 +475,16 @@
                       make_tuple(4, 4, variance16x16_c)));
 
 #if HAVE_NEON
+const vp8_sse_fn_t get4x4sse_cs_neon = vp8_get4x4sse_cs_neon;
+INSTANTIATE_TEST_CASE_P(
+    NEON, VP8SseTest,
+    ::testing::Values(make_tuple(2, 2, get4x4sse_cs_neon)));
+
+const vp8_variance_fn_t mse16x16_neon = vp8_mse16x16_neon;
+INSTANTIATE_TEST_CASE_P(
+    NEON, VP8MseTest,
+    ::testing::Values(make_tuple(4, 4, mse16x16_neon)));
+
 const vp8_variance_fn_t variance8x8_neon = vp8_variance8x8_neon;
 const vp8_variance_fn_t variance8x16_neon = vp8_variance8x16_neon;
 const vp8_variance_fn_t variance16x8_neon = vp8_variance16x8_neon;
@@ -375,6 +497,7 @@
                       make_tuple(4, 4, variance16x16_neon)));
 #endif
 
+
 #if HAVE_MMX
 const vp8_variance_fn_t variance4x4_mmx = vp8_variance4x4_mmx;
 const vp8_variance_fn_t variance8x8_mmx = vp8_variance8x8_mmx;
diff --git a/test/vp9_intrapred_test.cc b/test/vp9_intrapred_test.cc
new file mode 100644
index 0000000..7d08d9e
--- /dev/null
+++ b/test/vp9_intrapred_test.cc
@@ -0,0 +1,284 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <string>
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vpx_mem/vpx_mem.h"
+#include "test/util.h"
+
+namespace {
+
+using libvpx_test::ACMRandom;
+
+const int count_test_block = 100000;
+
+// Base class for VP9 intra prediction tests.
+class VP9IntraPredBase {
+ public:
+  virtual ~VP9IntraPredBase() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  virtual void Predict(PREDICTION_MODE mode) = 0;
+
+  void CheckPrediction(int test_case_number, int *error_count) const {
+    // For each pixel ensure that the calculated value is the same as reference.
+    for (int y = 0; y < block_size_; y++) {
+      for (int x = 0; x < block_size_; x++) {
+        *error_count += ref_dst_[x + y * stride_] != dst_[x + y * stride_];
+        if (*error_count == 1) {
+          ASSERT_EQ(ref_dst_[x + y * stride_], dst_[x + y * stride_])
+              << " Failed on Test Case Number "<< test_case_number;
+        }
+      }
+    }
+  }
+
+  void RunTest(uint16_t* left_col, uint16_t* above_data,
+               uint16_t* dst, uint16_t* ref_dst) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    left_col_ = left_col;
+    dst_ = dst;
+    ref_dst_ = ref_dst;
+    above_row_ = above_data + 16;
+    int error_count = 0;
+    for (int i = 0; i < count_test_block; ++i) {
+      // Fill edges with random data, try first with saturated values.
+      for (int x = -1; x <= block_size_*2; x++) {
+        if (i == 0) {
+          above_row_[x] = mask_;
+        } else {
+          above_row_[x] = rnd.Rand16() & mask_;
+        }
+      }
+      for (int y = 0; y < block_size_; y++) {
+        if (i == 0) {
+          left_col_[y] = mask_;
+        } else {
+          left_col_[y] = rnd.Rand16() & mask_;
+        }
+      }
+      Predict(DC_PRED);
+      CheckPrediction(i, &error_count);
+    }
+    ASSERT_EQ(0, error_count);
+  }
+
+  int block_size_;
+  uint16_t *above_row_;
+  uint16_t *left_col_;
+  uint16_t *dst_;
+  uint16_t *ref_dst_;
+  ptrdiff_t stride_;
+  int mask_;
+};
+
+typedef void (*intra_pred_fn_t)(
+      uint16_t *dst, ptrdiff_t stride, const uint16_t *above,
+      const uint16_t *left, int bps);
+typedef std::tr1::tuple<intra_pred_fn_t,
+                        intra_pred_fn_t, int, int> intra_pred_params_t;
+class VP9IntraPredTest
+    : public VP9IntraPredBase,
+      public ::testing::TestWithParam<intra_pred_params_t> {
+
+  virtual void SetUp() {
+    pred_fn_    = GET_PARAM(0);
+    ref_fn_     = GET_PARAM(1);
+    block_size_ = GET_PARAM(2);
+    bit_depth_  = GET_PARAM(3);
+    stride_     = block_size_ * 3;
+    mask_       = (1 << bit_depth_) - 1;
+  }
+
+  virtual void Predict(PREDICTION_MODE mode) {
+    const uint16_t *const_above_row = above_row_;
+    const uint16_t *const_left_col = left_col_;
+    ref_fn_(ref_dst_, stride_, const_above_row, const_left_col, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(pred_fn_(dst_, stride_, const_above_row,
+                                      const_left_col, bit_depth_));
+  }
+  intra_pred_fn_t pred_fn_;
+  intra_pred_fn_t ref_fn_;
+  int bit_depth_;
+};
+
+TEST_P(VP9IntraPredTest, IntraPredTests) {
+  // max block size is 32
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, left_col, 2*32);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, above_data, 2*32+32);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, dst, 3 * 32 * 32);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_dst, 3 * 32 * 32);
+  RunTest(left_col, above_data, dst, ref_dst);
+}
+
+using std::tr1::make_tuple;
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+                                       &vp9_high_dc_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+                                       &vp9_high_tm_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+                                       &vp9_high_tm_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 8)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 8),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 8),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 8),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 8),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 8)));
+#endif
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+                                       &vp9_high_dc_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+                                       &vp9_high_tm_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+                                       &vp9_high_tm_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                   &vp9_high_dc_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 10)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 10),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 10),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 10),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                   &vp9_high_tm_predictor_4x4_c, 4, 10),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 10)));
+#endif
+
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+                                       &vp9_high_dc_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+                                       &vp9_high_tm_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+                                       &vp9_high_tm_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 12)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
+                        ::testing::Values(
+                            make_tuple(&vp9_high_dc_predictor_4x4_sse,
+                                       &vp9_high_dc_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+                                       &vp9_high_dc_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+                                       &vp9_high_dc_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_4x4_sse,
+                                       &vp9_high_v_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_v_predictor_8x8_sse2,
+                                       &vp9_high_v_predictor_8x8_c, 8, 12),
+                            make_tuple(&vp9_high_v_predictor_16x16_sse2,
+                                       &vp9_high_v_predictor_16x16_c, 16, 12),
+                            make_tuple(&vp9_high_v_predictor_32x32_sse2,
+                                       &vp9_high_v_predictor_32x32_c, 32, 12),
+                            make_tuple(&vp9_high_tm_predictor_4x4_sse,
+                                       &vp9_high_tm_predictor_4x4_c, 4, 12),
+                            make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+                                       &vp9_high_tm_predictor_8x8_c, 8, 12)));
+#endif
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_SSE2
+}  // namespace
diff --git a/vp8/common/arm/neon/sixtappredict_neon.c b/vp8/common/arm/neon/sixtappredict_neon.c
index aad6133..4c2efc9 100644
--- a/vp8/common/arm/neon/sixtappredict_neon.c
+++ b/vp8/common/arm/neon/sixtappredict_neon.c
@@ -9,10 +9,7 @@
  */
 
 #include <arm_neon.h>
-
-#ifdef _MSC_VER
-#define __builtin_prefetch(x)
-#endif
+#include "vpx_ports/mem.h"
 
 static const int8_t vp8_sub_pel_filters[8][8] = {
     {0,  0,  128,   0,   0, 0, 0, 0},  /* note that 1/8 pel positionyys are */
diff --git a/vp8/common/arm/neon/variance_neon.c b/vp8/common/arm/neon/variance_neon.c
index afd2dc3..1b19790 100644
--- a/vp8/common/arm/neon/variance_neon.c
+++ b/vp8/common/arm/neon/variance_neon.c
@@ -9,10 +9,7 @@
  */
 
 #include <arm_neon.h>
-
-#ifdef _MSC_VER
-#define __builtin_prefetch(x)
-#endif
+#include "vpx_ports/mem.h"
 
 unsigned int vp8_variance16x16_neon(
         const unsigned char *src_ptr,
diff --git a/vp8/common/arm/neon/vp8_subpixelvariance_neon.c b/vp8/common/arm/neon/vp8_subpixelvariance_neon.c
index 6405bf2..8308d55 100644
--- a/vp8/common/arm/neon/vp8_subpixelvariance_neon.c
+++ b/vp8/common/arm/neon/vp8_subpixelvariance_neon.c
@@ -12,10 +12,6 @@
 #include "vpx_ports/mem.h"
 #include "vpx/vpx_integer.h"
 
-#ifdef _MSC_VER
-#define __builtin_prefetch(x)
-#endif
-
 static const uint16_t bilinear_taps_coeff[8][2] = {
     {128,   0},
     {112,  16},
diff --git a/vp8/common/postproc.c b/vp8/common/postproc.c
index e50d393..277f371 100644
--- a/vp8/common/postproc.c
+++ b/vp8/common/postproc.c
@@ -214,6 +214,7 @@
     x = 50 + (x - 50) * 10 / 8;
     return x * x / 3;
 }
+
 void vp8_mbpost_proc_across_ip_c(unsigned char *src, int pitch, int rows, int cols, int flimit)
 {
     int r, c, i;
@@ -226,14 +227,14 @@
         int sumsq = 0;
         int sum   = 0;
 
-        for (i = -8; i<0; i++)
+        for (i = -8; i < 0; i++)
           s[i]=s[0];
 
         /* 17 avoids valgrind warning - we buffer values in c in d
          * and only write them when we've read 8 ahead...
          */
-        for (i = cols; i<cols+17; i++)
-          s[i]=s[cols-1];
+        for (i = 0; i < 17; i++)
+          s[i+cols]=s[cols-1];
 
         for (i = -8; i <= 6; i++)
         {
@@ -264,7 +265,6 @@
     }
 }
 
-
 void vp8_mbpost_proc_down_c(unsigned char *dst, int pitch, int rows, int cols, int flimit)
 {
     int r, c, i;
@@ -284,8 +284,8 @@
         /* 17 avoids valgrind warning - we buffer values in c in d
          * and only write them when we've read 8 ahead...
          */
-        for (i = rows; i < rows+17; i++)
-          s[i*pitch]=s[(rows-1)*pitch];
+        for (i = 0; i < 17; i++)
+          s[(i+rows)*pitch]=s[(rows-1)*pitch];
 
         for (i = -8; i <= 6; i++)
         {
@@ -385,13 +385,13 @@
 }
 #endif
 
-#if !(CONFIG_TEMPORAL_DENOISING)
 void vp8_de_noise(VP8_COMMON                 *cm,
                   YV12_BUFFER_CONFIG         *source,
                   YV12_BUFFER_CONFIG         *post,
                   int                         q,
                   int                         low_var_thresh,
-                  int                         flag)
+                  int                         flag,
+                  int                         uvfilter)
 {
     int mbr;
     double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
@@ -412,18 +412,20 @@
             source->y_buffer + 16 * mbr * source->y_stride,
             source->y_buffer + 16 * mbr * source->y_stride,
             source->y_stride, source->y_stride, source->y_width, limits, 16);
-
-        vp8_post_proc_down_and_across_mb_row(
-            source->u_buffer + 8 * mbr * source->uv_stride,
-            source->u_buffer + 8 * mbr * source->uv_stride,
-            source->uv_stride, source->uv_stride, source->uv_width, limits, 8);
-        vp8_post_proc_down_and_across_mb_row(
-            source->v_buffer + 8 * mbr * source->uv_stride,
-            source->v_buffer + 8 * mbr * source->uv_stride,
-            source->uv_stride, source->uv_stride, source->uv_width, limits, 8);
+        if (uvfilter == 1) {
+          vp8_post_proc_down_and_across_mb_row(
+              source->u_buffer + 8 * mbr * source->uv_stride,
+              source->u_buffer + 8 * mbr * source->uv_stride,
+              source->uv_stride, source->uv_stride, source->uv_width, limits,
+              8);
+          vp8_post_proc_down_and_across_mb_row(
+              source->v_buffer + 8 * mbr * source->uv_stride,
+              source->v_buffer + 8 * mbr * source->uv_stride,
+              source->uv_stride, source->uv_stride, source->uv_width, limits,
+              8);
+        }
     }
 }
-#endif
 
 double vp8_gaussian(double sigma, double mu, double x)
 {
diff --git a/vp8/common/postproc.h b/vp8/common/postproc.h
index 33d0a7f..0fa12a7 100644
--- a/vp8/common/postproc.h
+++ b/vp8/common/postproc.h
@@ -39,7 +39,8 @@
                   YV12_BUFFER_CONFIG         *post,
                   int                         q,
                   int                         low_var_thresh,
-                  int                         flag);
+                  int                         flag,
+                  int                         uvfilter);
 
 void vp8_deblock(struct VP8Common           *oci,
                  YV12_BUFFER_CONFIG         *source,
diff --git a/vp8/common/rtcd_defs.pl b/vp8/common/rtcd_defs.pl
index a90c876..c73ecf9 100644
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -289,22 +289,19 @@
 $vp8_sub_pixel_variance16x16_neon_asm=vp8_sub_pixel_variance16x16_neon;
 
 add_proto qw/unsigned int vp8_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse";
-specialize qw/vp8_variance_halfpixvar16x16_h mmx sse2 media neon_asm/;
+specialize qw/vp8_variance_halfpixvar16x16_h mmx sse2 media neon/;
 $vp8_variance_halfpixvar16x16_h_sse2=vp8_variance_halfpixvar16x16_h_wmt;
 $vp8_variance_halfpixvar16x16_h_media=vp8_variance_halfpixvar16x16_h_armv6;
-$vp8_variance_halfpixvar16x16_h_neon_asm=vp8_variance_halfpixvar16x16_h_neon;
 
 add_proto qw/unsigned int vp8_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse";
-specialize qw/vp8_variance_halfpixvar16x16_v mmx sse2 media neon_asm/;
+specialize qw/vp8_variance_halfpixvar16x16_v mmx sse2 media neon/;
 $vp8_variance_halfpixvar16x16_v_sse2=vp8_variance_halfpixvar16x16_v_wmt;
 $vp8_variance_halfpixvar16x16_v_media=vp8_variance_halfpixvar16x16_v_armv6;
-$vp8_variance_halfpixvar16x16_v_neon_asm=vp8_variance_halfpixvar16x16_v_neon;
 
 add_proto qw/unsigned int vp8_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse";
-specialize qw/vp8_variance_halfpixvar16x16_hv mmx sse2 media neon_asm/;
+specialize qw/vp8_variance_halfpixvar16x16_hv mmx sse2 media neon/;
 $vp8_variance_halfpixvar16x16_hv_sse2=vp8_variance_halfpixvar16x16_hv_wmt;
 $vp8_variance_halfpixvar16x16_hv_media=vp8_variance_halfpixvar16x16_hv_armv6;
-$vp8_variance_halfpixvar16x16_hv_neon_asm=vp8_variance_halfpixvar16x16_hv_neon;
 
 #
 # Single block SAD
@@ -407,14 +404,12 @@
 $vp8_sub_pixel_mse16x16_sse2=vp8_sub_pixel_mse16x16_wmt;
 
 add_proto qw/unsigned int vp8_mse16x16/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse";
-specialize qw/vp8_mse16x16 mmx sse2 media neon_asm/;
+specialize qw/vp8_mse16x16 mmx sse2 media neon/;
 $vp8_mse16x16_sse2=vp8_mse16x16_wmt;
 $vp8_mse16x16_media=vp8_mse16x16_armv6;
-$vp8_mse16x16_neon_asm=vp8_mse16x16_neon;
 
 add_proto qw/unsigned int vp8_get4x4sse_cs/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride";
-specialize qw/vp8_get4x4sse_cs mmx neon_asm/;
-$vp8_get4x4sse_cs_neon_asm=vp8_get4x4sse_cs_neon;
+specialize qw/vp8_get4x4sse_cs mmx neon/;
 
 #
 # Block copy
diff --git a/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm b/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm
deleted file mode 100644
index f82af3e..0000000
--- a/vp8/encoder/arm/neon/vp8_mse16x16_neon.asm
+++ /dev/null
@@ -1,123 +0,0 @@
-;
-;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-;  Use of this source code is governed by a BSD-style license
-;  that can be found in the LICENSE file in the root of the source
-;  tree. An additional intellectual property rights grant can be found
-;  in the file PATENTS.  All contributing project authors may
-;  be found in the AUTHORS file in the root of the source tree.
-;
-
-
-    EXPORT  |vp8_mse16x16_neon|
-    EXPORT  |vp8_get4x4sse_cs_neon|
-
-    ARM
-    REQUIRE8
-    PRESERVE8
-
-    AREA ||.text||, CODE, READONLY, ALIGN=2
-;============================
-; r0    unsigned char *src_ptr
-; r1    int source_stride
-; r2    unsigned char *ref_ptr
-; r3    int  recon_stride
-; stack unsigned int *sse
-;note: in this function, sum is never used. So, we can remove this part of calculation
-;from vp8_variance().
-
-|vp8_mse16x16_neon| PROC
-    vpush           {q7}
-
-    vmov.i8         q7, #0                      ;q7, q8, q9, q10 - sse
-    vmov.i8         q8, #0
-    vmov.i8         q9, #0
-    vmov.i8         q10, #0
-
-    mov             r12, #8
-
-mse16x16_neon_loop
-    vld1.8          {q0}, [r0], r1              ;Load up source and reference
-    vld1.8          {q2}, [r2], r3
-    vld1.8          {q1}, [r0], r1
-    vld1.8          {q3}, [r2], r3
-
-    vsubl.u8        q11, d0, d4
-    vsubl.u8        q12, d1, d5
-    vsubl.u8        q13, d2, d6
-    vsubl.u8        q14, d3, d7
-
-    vmlal.s16       q7, d22, d22
-    vmlal.s16       q8, d23, d23
-
-    subs            r12, r12, #1
-
-    vmlal.s16       q9, d24, d24
-    vmlal.s16       q10, d25, d25
-    vmlal.s16       q7, d26, d26
-    vmlal.s16       q8, d27, d27
-    vmlal.s16       q9, d28, d28
-    vmlal.s16       q10, d29, d29
-
-    bne             mse16x16_neon_loop
-
-    vadd.u32        q7, q7, q8
-    vadd.u32        q9, q9, q10
-
-    ldr             r12, [sp, #16]              ;load *sse from stack
-
-    vadd.u32        q10, q7, q9
-    vpaddl.u32      q1, q10
-    vadd.u64        d0, d2, d3
-
-    vst1.32         {d0[0]}, [r12]
-    vmov.32         r0, d0[0]
-
-    vpop            {q7}
-    bx              lr
-
-    ENDP
-
-
-;=============================
-; r0    unsigned char *src_ptr,
-; r1    int  source_stride,
-; r2    unsigned char *ref_ptr,
-; r3    int  recon_stride
-|vp8_get4x4sse_cs_neon| PROC
-    vpush           {q7}
-
-    vld1.8          {d0}, [r0], r1              ;Load up source and reference
-    vld1.8          {d4}, [r2], r3
-    vld1.8          {d1}, [r0], r1
-    vld1.8          {d5}, [r2], r3
-    vld1.8          {d2}, [r0], r1
-    vld1.8          {d6}, [r2], r3
-    vld1.8          {d3}, [r0], r1
-    vld1.8          {d7}, [r2], r3
-
-    vsubl.u8        q11, d0, d4
-    vsubl.u8        q12, d1, d5
-    vsubl.u8        q13, d2, d6
-    vsubl.u8        q14, d3, d7
-
-    vmull.s16       q7, d22, d22
-    vmull.s16       q8, d24, d24
-    vmull.s16       q9, d26, d26
-    vmull.s16       q10, d28, d28
-
-    vadd.u32        q7, q7, q8
-    vadd.u32        q9, q9, q10
-    vadd.u32        q9, q7, q9
-
-    vpaddl.u32      q1, q9
-    vadd.u64        d0, d2, d3
-
-    vmov.32         r0, d0[0]
-
-    vpop            {q7}
-    bx              lr
-
-    ENDP
-
-    END
diff --git a/vp8/encoder/arm/neon/vp8_mse16x16_neon.c b/vp8/encoder/arm/neon/vp8_mse16x16_neon.c
new file mode 100644
index 0000000..f806809
--- /dev/null
+++ b/vp8/encoder/arm/neon/vp8_mse16x16_neon.c
@@ -0,0 +1,131 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+unsigned int vp8_mse16x16_neon(
+        const unsigned char *src_ptr,
+        int source_stride,
+        const unsigned char *ref_ptr,
+        int recon_stride,
+        unsigned int *sse) {
+    int i;
+    int16x4_t d22s16, d23s16, d24s16, d25s16, d26s16, d27s16, d28s16, d29s16;
+    int64x1_t d0s64;
+    uint8x16_t q0u8, q1u8, q2u8, q3u8;
+    int32x4_t q7s32, q8s32, q9s32, q10s32;
+    uint16x8_t q11u16, q12u16, q13u16, q14u16;
+    int64x2_t q1s64;
+
+    q7s32 = vdupq_n_s32(0);
+    q8s32 = vdupq_n_s32(0);
+    q9s32 = vdupq_n_s32(0);
+    q10s32 = vdupq_n_s32(0);
+
+    for (i = 0; i < 8; i++) {  // mse16x16_neon_loop
+        q0u8 = vld1q_u8(src_ptr);
+        src_ptr += source_stride;
+        q1u8 = vld1q_u8(src_ptr);
+        src_ptr += source_stride;
+        q2u8 = vld1q_u8(ref_ptr);
+        ref_ptr += recon_stride;
+        q3u8 = vld1q_u8(ref_ptr);
+        ref_ptr += recon_stride;
+
+        q11u16 = vsubl_u8(vget_low_u8(q0u8), vget_low_u8(q2u8));
+        q12u16 = vsubl_u8(vget_high_u8(q0u8), vget_high_u8(q2u8));
+        q13u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q3u8));
+        q14u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q3u8));
+
+        d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
+        d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16));
+        q7s32 = vmlal_s16(q7s32, d22s16, d22s16);
+        q8s32 = vmlal_s16(q8s32, d23s16, d23s16);
+
+        d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
+        d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
+        q9s32 = vmlal_s16(q9s32, d24s16, d24s16);
+        q10s32 = vmlal_s16(q10s32, d25s16, d25s16);
+
+        d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
+        d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
+        q7s32 = vmlal_s16(q7s32, d26s16, d26s16);
+        q8s32 = vmlal_s16(q8s32, d27s16, d27s16);
+
+        d28s16 = vreinterpret_s16_u16(vget_low_u16(q14u16));
+        d29s16 = vreinterpret_s16_u16(vget_high_u16(q14u16));
+        q9s32 = vmlal_s16(q9s32, d28s16, d28s16);
+        q10s32 = vmlal_s16(q10s32, d29s16, d29s16);
+    }
+
+    q7s32 = vaddq_s32(q7s32, q8s32);
+    q9s32 = vaddq_s32(q9s32, q10s32);
+    q10s32 = vaddq_s32(q7s32, q9s32);
+
+    q1s64 = vpaddlq_s32(q10s32);
+    d0s64 = vadd_s64(vget_low_s64(q1s64), vget_high_s64(q1s64));
+
+    vst1_lane_u32((uint32_t *)sse, vreinterpret_u32_s64(d0s64), 0);
+    return vget_lane_u32(vreinterpret_u32_s64(d0s64), 0);
+}
+
+unsigned int vp8_get4x4sse_cs_neon(
+        const unsigned char *src_ptr,
+        int source_stride,
+        const unsigned char *ref_ptr,
+        int recon_stride) {
+    int16x4_t d22s16, d24s16, d26s16, d28s16;
+    int64x1_t d0s64;
+    uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8;
+    int32x4_t q7s32, q8s32, q9s32, q10s32;
+    uint16x8_t q11u16, q12u16, q13u16, q14u16;
+    int64x2_t q1s64;
+
+    d0u8 = vld1_u8(src_ptr);
+    src_ptr += source_stride;
+    d4u8 = vld1_u8(ref_ptr);
+    ref_ptr += recon_stride;
+    d1u8 = vld1_u8(src_ptr);
+    src_ptr += source_stride;
+    d5u8 = vld1_u8(ref_ptr);
+    ref_ptr += recon_stride;
+    d2u8 = vld1_u8(src_ptr);
+    src_ptr += source_stride;
+    d6u8 = vld1_u8(ref_ptr);
+    ref_ptr += recon_stride;
+    d3u8 = vld1_u8(src_ptr);
+    src_ptr += source_stride;
+    d7u8 = vld1_u8(ref_ptr);
+    ref_ptr += recon_stride;
+
+    q11u16 = vsubl_u8(d0u8, d4u8);
+    q12u16 = vsubl_u8(d1u8, d5u8);
+    q13u16 = vsubl_u8(d2u8, d6u8);
+    q14u16 = vsubl_u8(d3u8, d7u8);
+
+    d22s16 = vget_low_s16(vreinterpretq_s16_u16(q11u16));
+    d24s16 = vget_low_s16(vreinterpretq_s16_u16(q12u16));
+    d26s16 = vget_low_s16(vreinterpretq_s16_u16(q13u16));
+    d28s16 = vget_low_s16(vreinterpretq_s16_u16(q14u16));
+
+    q7s32 = vmull_s16(d22s16, d22s16);
+    q8s32 = vmull_s16(d24s16, d24s16);
+    q9s32 = vmull_s16(d26s16, d26s16);
+    q10s32 = vmull_s16(d28s16, d28s16);
+
+    q7s32 = vaddq_s32(q7s32, q8s32);
+    q9s32 = vaddq_s32(q9s32, q10s32);
+    q9s32 = vaddq_s32(q7s32, q9s32);
+
+    q1s64 = vpaddlq_s32(q9s32);
+    d0s64 = vadd_s64(vget_low_s64(q1s64), vget_high_s64(q1s64));
+
+    return vget_lane_u32(vreinterpret_u32_s64(d0s64), 0);
+}
diff --git a/vp8/encoder/denoising.c b/vp8/encoder/denoising.c
index 0c98eb1..12f9734 100644
--- a/vp8/encoder/denoising.c
+++ b/vp8/encoder/denoising.c
@@ -68,6 +68,10 @@
     int adj_val[3] = {3, 4, 6};
     int shift_inc1 = 0;
     int shift_inc2 = 1;
+    int col_sum[16] = {0, 0, 0, 0,
+                       0, 0, 0, 0,
+                       0, 0, 0, 0,
+                       0, 0, 0, 0};
     /* If motion_magnitude is small, making the denoiser more aggressive by
      * increasing the adjustment for each level. Add another increment for
      * blocks that are labeled for increase denoising. */
@@ -98,11 +102,11 @@
             if (absdiff <= 3 + shift_inc1)
             {
                 running_avg_y[c] = mc_running_avg_y[c];
-                sum_diff += diff;
+                col_sum[c] += diff;
             }
             else
             {
-                if (absdiff >= 4 && absdiff <= 7)
+                if (absdiff >= 4 + shift_inc1 && absdiff <= 7)
                     adjustment = adj_val[0];
                 else if (absdiff >= 8 && absdiff <= 15)
                     adjustment = adj_val[1];
@@ -116,7 +120,7 @@
                     else
                         running_avg_y[c] = sig[c] + adjustment;
 
-                    sum_diff += adjustment;
+                    col_sum[c] += adjustment;
                 }
                 else
                 {
@@ -125,7 +129,7 @@
                     else
                         running_avg_y[c] = sig[c] - adjustment;
 
-                    sum_diff -= adjustment;
+                    col_sum[c] -= adjustment;
                 }
             }
         }
@@ -136,6 +140,23 @@
         running_avg_y += avg_y_stride;
     }
 
+    for (c = 0; c < 16; ++c) {
+      // Below we clip the value in the same way which SSE code use.
+      // When adopting aggressive denoiser, the adj_val for each pixel
+      // could be at most 8 (this is current max adjustment of the map).
+      // In SSE code, we calculate the sum of adj_val for
+      // the columns, so the sum could be upto 128(16 rows). However,
+      // the range of the value is -128 ~ 127 in SSE code, that's why
+      // we do this change in C code.
+      // We don't do this for UV denoiser, since there are only 8 rows,
+      // and max adjustments <= 8, so the sum of the columns will not
+      // exceed 64.
+      if (col_sum[c] >= 128) {
+        col_sum[c] = 127;
+      }
+      sum_diff += col_sum[c];
+    }
+
     sum_diff_thresh= SUM_DIFF_THRESHOLD;
     if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH;
     if (abs(sum_diff) > sum_diff_thresh) {
@@ -166,14 +187,14 @@
                 running_avg_y[c] = 0;
               else
                 running_avg_y[c] = running_avg_y[c] - adjustment;
-              sum_diff -= adjustment;
+              col_sum[c] -= adjustment;
             } else if (diff < 0) {
               // Bring denoised signal up.
               if (running_avg_y[c] + adjustment > 255)
                 running_avg_y[c] = 255;
               else
                 running_avg_y[c] = running_avg_y[c] + adjustment;
-              sum_diff += adjustment;
+              col_sum[c] += adjustment;
             }
           }
           // TODO(marpan): Check here if abs(sum_diff) has gone below the
@@ -182,6 +203,15 @@
           mc_running_avg_y += mc_avg_y_stride;
           running_avg_y += avg_y_stride;
         }
+
+        sum_diff = 0;
+        for (c = 0; c < 16; ++c) {
+          if (col_sum[c] >= 128) {
+            col_sum[c] = 127;
+          }
+          sum_diff += col_sum[c];
+        }
+
         if (abs(sum_diff) > sum_diff_thresh)
           return COPY_BLOCK;
       } else {
@@ -354,6 +384,7 @@
     denoiser->denoise_pars.pickmode_mv_bias = 100;
     denoiser->denoise_pars.qp_thresh = 0;
     denoiser->denoise_pars.consec_zerolast = UINT_MAX;
+    denoiser->denoise_pars.spatial_blur = 0;
   } else {
     denoiser->denoise_pars.scale_sse_thresh = 2;
     denoiser->denoise_pars.scale_motion_thresh = 16;
@@ -362,6 +393,7 @@
     denoiser->denoise_pars.pickmode_mv_bias = 60;
     denoiser->denoise_pars.qp_thresh = 100;
     denoiser->denoise_pars.consec_zerolast = 10;
+    denoiser->denoise_pars.spatial_blur = 20;
   }
 }
 
@@ -412,12 +444,27 @@
     vp8_denoiser_set_parameters(denoiser, mode);
     denoiser->nmse_source_diff = 0;
     denoiser->nmse_source_diff_count = 0;
+    denoiser->qp_avg = 0;
+    // QP threshold below which we can go up to aggressive mode.
+    denoiser->qp_threshold_up = 80;
+    // QP threshold above which we can go back down to normal mode.
+    // For now keep this second threshold high, so not used currently.
+    denoiser->qp_threshold_down = 128;
+    // Bitrate thresholds and noise metric (nmse) thresholds for switching to
+    // aggressive mode.
     // TODO(marpan): Adjust thresholds, including effect on resolution.
+    denoiser->bitrate_threshold = 200000;  // (bits/sec).
     denoiser->threshold_aggressive_mode = 35;
-    if (width * height > 640 * 480)
+    if (width * height > 640 * 480) {
+      denoiser->bitrate_threshold = 500000;
+      denoiser->threshold_aggressive_mode = 100;
+    } else if (width * height > 960 * 540) {
+      denoiser->bitrate_threshold = 800000;
       denoiser->threshold_aggressive_mode = 150;
-    else if (width * height > 1280 * 720)
+    } else if (width * height > 1280 * 720) {
+      denoiser->bitrate_threshold = 2000000;
       denoiser->threshold_aggressive_mode = 1400;
+    }
     return 0;
 }
 
diff --git a/vp8/encoder/denoising.h b/vp8/encoder/denoising.h
index e1844f1..fb7930b 100644
--- a/vp8/encoder/denoising.h
+++ b/vp8/encoder/denoising.h
@@ -19,7 +19,7 @@
 #endif
 
 #define SUM_DIFF_THRESHOLD (16 * 16 * 2)
-#define SUM_DIFF_THRESHOLD_HIGH (16 * 16 * 3)
+#define SUM_DIFF_THRESHOLD_HIGH (600)
 #define MOTION_MAGNITUDE_THRESHOLD (8*3)
 
 #define SUM_DIFF_THRESHOLD_UV (96)   // (8 * 8 * 1.5)
@@ -67,6 +67,8 @@
   unsigned int qp_thresh;
   // Threshold for number of consecutive frames for blocks coded as ZEROMV-LAST.
   unsigned int consec_zerolast;
+  // Threshold for amount of spatial blur on Y channel. 0 means no spatial blur.
+  unsigned int spatial_blur;
 } denoise_params;
 
 typedef struct vp8_denoiser
@@ -81,6 +83,10 @@
     int threshold_aggressive_mode;
     int nmse_source_diff;
     int nmse_source_diff_count;
+    int qp_avg;
+    int qp_threshold_up;
+    int qp_threshold_down;
+    int bitrate_threshold;
     denoise_params denoise_pars;
 } VP8_DENOISER;
 
diff --git a/vp8/encoder/encodemb.c b/vp8/encoder/encodemb.c
index 7ed2fe1..eb0619d 100644
--- a/vp8/encoder/encodemb.c
+++ b/vp8/encoder/encodemb.c
@@ -258,13 +258,6 @@
     b = &mb->block[ib];
     d = &mb->e_mbd.block[ib];
 
-    /* Enable this to test the effect of RDO as a replacement for the dynamic
-     *  zero bin instead of an augmentation of it.
-     */
-#if 0
-    vp8_strict_quantize_b(b, d);
-#endif
-
     dequant_ptr = d->dequant;
     coeff_ptr = b->coeff;
     qcoeff_ptr = d->qcoeff;
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 74e75c4..d8eff66 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -3370,33 +3370,43 @@
   if (total > 0 &&
       (num_blocks > (tot_num_blocks >> 4))) {
     // Update the recursive mean square source_diff.
-    if (cpi->denoiser.nmse_source_diff_count == 0)
+    if (cpi->denoiser.nmse_source_diff_count == 0) {
       // First sample in new interval.
       cpi->denoiser.nmse_source_diff = total;
-    else
+      cpi->denoiser.qp_avg = cm->base_qindex;
+    } else {
       // For subsequent samples, use average with weight ~1/4 for new sample.
       cpi->denoiser.nmse_source_diff = (int)((total >> 2) +
           3 * (cpi->denoiser.nmse_source_diff >> 2));
+      cpi->denoiser.qp_avg = (int)((cm->base_qindex >> 2) +
+          3 * (cpi->denoiser.qp_avg >> 2));
+    }
     cpi->denoiser.nmse_source_diff_count++;
   }
   // Check for changing the denoiser mode, when we have obtained #samples =
-  // num_mode_change.
+  // num_mode_change. Condition the change also on the bitrate and QP.
   if (cpi->denoiser.nmse_source_diff_count == num_mode_change) {
     // Check for going up: from normal to aggressive mode.
     if ((cpi->denoiser.denoiser_mode == kDenoiserOnYUV) &&
         (cpi->denoiser.nmse_source_diff >
-        cpi->denoiser.threshold_aggressive_mode)) {
+        cpi->denoiser.threshold_aggressive_mode) &&
+        (cpi->denoiser.qp_avg < cpi->denoiser.qp_threshold_up &&
+         cpi->target_bandwidth > cpi->denoiser.bitrate_threshold)) {
       vp8_denoiser_set_parameters(&cpi->denoiser, kDenoiserOnYUVAggressive);
     } else {
       // Check for going down: from aggressive to normal mode.
-      if ((cpi->denoiser.denoiser_mode == kDenoiserOnYUVAggressive) &&
+      if (((cpi->denoiser.denoiser_mode == kDenoiserOnYUVAggressive) &&
           (cpi->denoiser.nmse_source_diff <
-          cpi->denoiser.threshold_aggressive_mode)) {
+          cpi->denoiser.threshold_aggressive_mode)) ||
+          ((cpi->denoiser.denoiser_mode == kDenoiserOnYUVAggressive) &&
+          (cpi->denoiser.qp_avg > cpi->denoiser.qp_threshold_down ||
+           cpi->target_bandwidth < cpi->denoiser.bitrate_threshold))) {
         vp8_denoiser_set_parameters(&cpi->denoiser, kDenoiserOnYUV);
       }
     }
     // Reset metric and counter for next interval.
     cpi->denoiser.nmse_source_diff = 0;
+    cpi->denoiser.qp_avg = 0;
     cpi->denoiser.nmse_source_diff_count = 0;
   }
 }
@@ -4013,6 +4023,17 @@
 
     scale_and_extend_source(cpi->un_scaled_source, cpi);
 
+#if CONFIG_TEMPORAL_DENOISING && CONFIG_POSTPROC
+    // Option to apply spatial blur under the aggressive or adaptive
+    // (temporal denoising) mode.
+    if (cpi->oxcf.noise_sensitivity >= 3) {
+      if (cpi->denoiser.denoise_pars.spatial_blur != 0) {
+        vp8_de_noise(cm, cpi->Source, cpi->Source,
+            cpi->denoiser.denoise_pars.spatial_blur, 1, 0, 0);
+      }
+    }
+#endif
+
 #if !(CONFIG_REALTIME_ONLY) && CONFIG_POSTPROC && !(CONFIG_TEMPORAL_DENOISING)
 
     if (cpi->oxcf.noise_sensitivity > 0)
@@ -4045,11 +4066,11 @@
 
         if (cm->frame_type == KEY_FRAME)
         {
-            vp8_de_noise(cm, cpi->Source, cpi->Source, l , 1,  0);
+            vp8_de_noise(cm, cpi->Source, cpi->Source, l , 1,  0, 1);
         }
         else
         {
-            vp8_de_noise(cm, cpi->Source, cpi->Source, l , 1,  0);
+            vp8_de_noise(cm, cpi->Source, cpi->Source, l , 1,  0, 1);
 
             src = cpi->Source->y_buffer;
 
diff --git a/vp8/encoder/quantize.c b/vp8/encoder/quantize.c
index fda997f..9953bd6 100644
--- a/vp8/encoder/quantize.c
+++ b/vp8/encoder/quantize.c
@@ -16,57 +16,6 @@
 #include "quantize.h"
 #include "vp8/common/quant_common.h"
 
-#define EXACT_QUANT
-
-#ifdef EXACT_FASTQUANT
-void vp8_fast_quantize_b_c(BLOCK *b, BLOCKD *d)
-{
-    int i, rc, eob;
-    int zbin;
-    int x, y, z, sz;
-    short *coeff_ptr       = b->coeff;
-    short *zbin_ptr        = b->zbin;
-    short *round_ptr       = b->round;
-    short *quant_ptr       = b->quant_fast;
-    unsigned char *quant_shift_ptr = b->quant_shift;
-    short *qcoeff_ptr      = d->qcoeff;
-    short *dqcoeff_ptr     = d->dqcoeff;
-    short *dequant_ptr     = d->dequant;
-
-    vpx_memset(qcoeff_ptr, 0, 32);
-    vpx_memset(dqcoeff_ptr, 0, 32);
-
-    eob = -1;
-
-    for (i = 0; i < 16; i++)
-    {
-        rc   = vp8_default_zig_zag1d[i];
-        z    = coeff_ptr[rc];
-        zbin = zbin_ptr[rc] ;
-
-        sz = (z >> 31);                              /* sign of z */
-        x  = (z ^ sz) - sz;                          /* x = abs(z) */
-
-        if (x >= zbin)
-        {
-            x += round_ptr[rc];
-            y  = ((((x * quant_ptr[rc]) >> 16) + x)
-                 * quant_shift_ptr[rc]) >> 16;       /* quantize (x) */
-            x  = (y ^ sz) - sz;                      /* get the sign back */
-            qcoeff_ptr[rc] = x;                      /* write to destination */
-            dqcoeff_ptr[rc] = x * dequant_ptr[rc];   /* dequantized value */
-
-            if (y)
-            {
-                eob = i;                             /* last nonzero coeffs */
-            }
-        }
-    }
-    *d->eob = (char)(eob + 1);
-}
-
-#else
-
 void vp8_fast_quantize_b_c(BLOCK *b, BLOCKD *d)
 {
     int i, rc, eob;
@@ -100,9 +49,6 @@
     *d->eob = (char)(eob + 1);
 }
 
-#endif
-
-#ifdef EXACT_QUANT
 void vp8_regular_quantize_b_c(BLOCK *b, BLOCKD *d)
 {
     int i, rc, eob;
@@ -155,117 +101,6 @@
     *d->eob = (char)(eob + 1);
 }
 
-/* Perform regular quantization, with unbiased rounding and no zero bin. */
-void vp8_strict_quantize_b_c(BLOCK *b, BLOCKD *d)
-{
-    int i;
-    int rc;
-    int eob;
-    int x;
-    int y;
-    int z;
-    int sz;
-    short *coeff_ptr;
-    short *quant_ptr;
-    short *quant_shift_ptr;
-    short *qcoeff_ptr;
-    short *dqcoeff_ptr;
-    short *dequant_ptr;
-
-    coeff_ptr       = b->coeff;
-    quant_ptr       = b->quant;
-    quant_shift_ptr = b->quant_shift;
-    qcoeff_ptr      = d->qcoeff;
-    dqcoeff_ptr     = d->dqcoeff;
-    dequant_ptr     = d->dequant;
-    eob = - 1;
-    vpx_memset(qcoeff_ptr, 0, 32);
-    vpx_memset(dqcoeff_ptr, 0, 32);
-    for (i = 0; i < 16; i++)
-    {
-        int dq;
-        int rounding;
-
-        /*TODO: These arrays should be stored in zig-zag order.*/
-        rc = vp8_default_zig_zag1d[i];
-        z = coeff_ptr[rc];
-        dq = dequant_ptr[rc];
-        rounding = dq >> 1;
-        /* Sign of z. */
-        sz = -(z < 0);
-        x = (z + sz) ^ sz;
-        x += rounding;
-        if (x >= dq)
-        {
-            /* Quantize x. */
-            y  = ((((x * quant_ptr[rc]) >> 16) + x) * quant_shift_ptr[rc]) >> 16;
-            /* Put the sign back. */
-            x = (y + sz) ^ sz;
-            /* Save the coefficient and its dequantized value. */
-            qcoeff_ptr[rc] = x;
-            dqcoeff_ptr[rc] = x * dq;
-            /* Remember the last non-zero coefficient. */
-            if (y)
-                eob = i;
-        }
-    }
-
-    *d->eob = (char)(eob + 1);
-}
-
-#else
-
-void vp8_regular_quantize_b_c(BLOCK *b, BLOCKD *d)
-{
-    int i, rc, eob;
-    int zbin;
-    int x, y, z, sz;
-    short *zbin_boost_ptr = b->zrun_zbin_boost;
-    short *coeff_ptr      = b->coeff;
-    short *zbin_ptr       = b->zbin;
-    short *round_ptr      = b->round;
-    short *quant_ptr      = b->quant;
-    short *qcoeff_ptr     = d->qcoeff;
-    short *dqcoeff_ptr    = d->dqcoeff;
-    short *dequant_ptr    = d->dequant;
-    short zbin_oq_value   = b->zbin_extra;
-
-    vpx_memset(qcoeff_ptr, 0, 32);
-    vpx_memset(dqcoeff_ptr, 0, 32);
-
-    eob = -1;
-
-    for (i = 0; i < 16; i++)
-    {
-        rc   = vp8_default_zig_zag1d[i];
-        z    = coeff_ptr[rc];
-
-        zbin = zbin_ptr[rc] + *zbin_boost_ptr + zbin_oq_value;
-
-        zbin_boost_ptr ++;
-        sz = (z >> 31);                              /* sign of z */
-        x  = (z ^ sz) - sz;                          /* x = abs(z) */
-
-        if (x >= zbin)
-        {
-            y  = ((x + round_ptr[rc]) * quant_ptr[rc]) >> 16; /* quantize (x) */
-            x  = (y ^ sz) - sz;                      /* get the sign back */
-            qcoeff_ptr[rc]  = x;                     /* write to destination */
-            dqcoeff_ptr[rc] = x * dequant_ptr[rc];   /* dequantized value */
-
-            if (y)
-            {
-                eob = i;                             /* last nonzero coeffs */
-                zbin_boost_ptr = &b->zrun_zbin_boost[0]; /* reset zrl */
-            }
-        }
-    }
-
-    *d->eob = (char)(eob + 1);
-}
-
-#endif
-
 void vp8_quantize_mby_c(MACROBLOCK *x)
 {
     int i;
@@ -403,8 +238,6 @@
 };
 
 
-#define EXACT_QUANT
-#ifdef EXACT_QUANT
 static void invert_quant(int improved_quant, short *quant,
                          short *shift, short d)
 {
@@ -526,68 +359,6 @@
         }
     }
 }
-#else
-void vp8cx_init_quantizer(VP8_COMP *cpi)
-{
-    int i;
-    int quant_val;
-    int Q;
-
-    int zbin_boost[16] = {0, 0, 8, 10, 12, 14, 16, 20, 24, 28, 32, 36, 40, 44, 44, 44};
-
-    for (Q = 0; Q < QINDEX_RANGE; Q++)
-    {
-        /* dc values */
-        quant_val = vp8_dc_quant(Q, cpi->common.y1dc_delta_q);
-        cpi->Y1quant[Q][0] = (1 << 16) / quant_val;
-        cpi->Y1zbin[Q][0] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;
-        cpi->Y1round[Q][0] = (qrounding_factors[Q] * quant_val) >> 7;
-        cpi->common.Y1dequant[Q][0] = quant_val;
-        cpi->zrun_zbin_boost_y1[Q][0] = (quant_val * zbin_boost[0]) >> 7;
-
-        quant_val = vp8_dc2quant(Q, cpi->common.y2dc_delta_q);
-        cpi->Y2quant[Q][0] = (1 << 16) / quant_val;
-        cpi->Y2zbin[Q][0] = ((qzbin_factors_y2[Q] * quant_val) + 64) >> 7;
-        cpi->Y2round[Q][0] = (qrounding_factors_y2[Q] * quant_val) >> 7;
-        cpi->common.Y2dequant[Q][0] = quant_val;
-        cpi->zrun_zbin_boost_y2[Q][0] = (quant_val * zbin_boost[0]) >> 7;
-
-        quant_val = vp8_dc_uv_quant(Q, cpi->common.uvdc_delta_q);
-        cpi->UVquant[Q][0] = (1 << 16) / quant_val;
-        cpi->UVzbin[Q][0] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;;
-        cpi->UVround[Q][0] = (qrounding_factors[Q] * quant_val) >> 7;
-        cpi->common.UVdequant[Q][0] = quant_val;
-        cpi->zrun_zbin_boost_uv[Q][0] = (quant_val * zbin_boost[0]) >> 7;
-
-        /* all the ac values = ; */
-        for (i = 1; i < 16; i++)
-        {
-            int rc = vp8_default_zig_zag1d[i];
-
-            quant_val = vp8_ac_yquant(Q);
-            cpi->Y1quant[Q][rc] = (1 << 16) / quant_val;
-            cpi->Y1zbin[Q][rc] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;
-            cpi->Y1round[Q][rc] = (qrounding_factors[Q] * quant_val) >> 7;
-            cpi->common.Y1dequant[Q][rc] = quant_val;
-            cpi->zrun_zbin_boost_y1[Q][i] = (quant_val * zbin_boost[i]) >> 7;
-
-            quant_val = vp8_ac2quant(Q, cpi->common.y2ac_delta_q);
-            cpi->Y2quant[Q][rc] = (1 << 16) / quant_val;
-            cpi->Y2zbin[Q][rc] = ((qzbin_factors_y2[Q] * quant_val) + 64) >> 7;
-            cpi->Y2round[Q][rc] = (qrounding_factors_y2[Q] * quant_val) >> 7;
-            cpi->common.Y2dequant[Q][rc] = quant_val;
-            cpi->zrun_zbin_boost_y2[Q][i] = (quant_val * zbin_boost[i]) >> 7;
-
-            quant_val = vp8_ac_uv_quant(Q, cpi->common.uvac_delta_q);
-            cpi->UVquant[Q][rc] = (1 << 16) / quant_val;
-            cpi->UVzbin[Q][rc] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;
-            cpi->UVround[Q][rc] = (qrounding_factors[Q] * quant_val) >> 7;
-            cpi->common.UVdequant[Q][rc] = quant_val;
-            cpi->zrun_zbin_boost_uv[Q][i] = (quant_val * zbin_boost[i]) >> 7;
-        }
-    }
-}
-#endif
 
 #define ZBIN_EXTRA_Y \
     (( cpi->common.Y1dequant[QIndex][1] *  \
diff --git a/vp8/vp8cx_arm.mk b/vp8/vp8cx_arm.mk
index 551271e..ed19fd4 100644
--- a/vp8/vp8cx_arm.mk
+++ b/vp8/vp8cx_arm.mk
@@ -36,9 +36,9 @@
 #File list for neon
 # encoder
 VP8_CX_SRCS-$(HAVE_NEON_ASM)  += encoder/arm/neon/fastquantizeb_neon$(ASM)
-VP8_CX_SRCS-$(HAVE_NEON_ASM)  += encoder/arm/neon/vp8_mse16x16_neon$(ASM)
 
 VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/denoising_neon.c
-VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/vp8_shortwalsh4x4_neon.c
-VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/subtract_neon.c
 VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/shortfdct_neon.c
+VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/subtract_neon.c
+VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/vp8_mse16x16_neon.c
+VP8_CX_SRCS-$(HAVE_NEON)  += encoder/arm/neon/vp8_shortwalsh4x4_neon.c
diff --git a/vp9/common/vp9_alloccommon.c b/vp9/common/vp9_alloccommon.c
index 21ae8d5..8b3b9db 100644
--- a/vp9/common/vp9_alloccommon.c
+++ b/vp9/common/vp9_alloccommon.c
@@ -44,14 +44,8 @@
 static void setup_mi(VP9_COMMON *cm) {
   cm->mi = cm->mip + cm->mi_stride + 1;
   cm->prev_mi = cm->prev_mip + cm->mi_stride + 1;
-  cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
-  cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1;
 
   vpx_memset(cm->mip, 0, cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mip));
-
-  vpx_memset(cm->mi_grid_base, 0, cm->mi_stride * (cm->mi_rows + 1) *
-                                      sizeof(*cm->mi_grid_base));
-
   clear_mi_border(cm, cm->prev_mip);
 }
 
@@ -63,11 +57,6 @@
         (MODE_INFO *)vpx_calloc(mi_size, sizeof(MODE_INFO));
     if (cm->mip_array[i] == NULL)
       return 1;
-
-    cm->mi_grid_base_array[i] =
-        (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*));
-    if (cm->mi_grid_base_array[i] == NULL)
-      return 1;
   }
 
   cm->mi_alloc_size = mi_size;
@@ -78,8 +67,6 @@
 
   cm->mip = cm->mip_array[cm->mi_idx];
   cm->prev_mip = cm->mip_array[cm->prev_mi_idx];
-  cm->mi_grid_base = cm->mi_grid_base_array[cm->mi_idx];
-  cm->prev_mi_grid_base = cm->mi_grid_base_array[cm->prev_mi_idx];
 
   return 0;
 }
@@ -90,14 +77,10 @@
   for (i = 0; i < 2; ++i) {
     vpx_free(cm->mip_array[i]);
     cm->mip_array[i] = NULL;
-    vpx_free(cm->mi_grid_base_array[i]);
-    cm->mi_grid_base_array[i] = NULL;
   }
 
   cm->mip = NULL;
   cm->prev_mip = NULL;
-  cm->mi_grid_base = NULL;
-  cm->prev_mi_grid_base = NULL;
 }
 
 void vp9_free_ref_frame_buffers(VP9_COMMON *cm) {
@@ -224,12 +207,8 @@
   // Current mip will be the prev_mip for the next frame.
   cm->mip = cm->mip_array[cm->mi_idx];
   cm->prev_mip = cm->mip_array[cm->prev_mi_idx];
-  cm->mi_grid_base = cm->mi_grid_base_array[cm->mi_idx];
-  cm->prev_mi_grid_base = cm->mi_grid_base_array[cm->prev_mi_idx];
 
   // Update the upper left visible macroblock ptrs.
   cm->mi = cm->mip + cm->mi_stride + 1;
   cm->prev_mi = cm->prev_mip + cm->mi_stride + 1;
-  cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
-  cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1;
 }
diff --git a/vp9/common/vp9_blockd.c b/vp9/common/vp9_blockd.c
index dab8f96..e13445f 100644
--- a/vp9/common/vp9_blockd.c
+++ b/vp9/common/vp9_blockd.c
@@ -40,7 +40,7 @@
     const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
     foreach_transformed_block_visitor visit, void *arg) {
   const struct macroblockd_plane *const pd = &xd->plane[plane];
-  const MB_MODE_INFO* mbmi = &xd->mi[0]->mbmi;
+  const MB_MODE_INFO* mbmi = &xd->mi[0].src_mi->mbmi;
   // block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
   // 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
   // transform size varies per plane, look it up in a common way.
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 951e6e0..702efe0 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -21,6 +21,7 @@
 #include "vp9/common/vp9_common_data.h"
 #include "vp9/common/vp9_enums.h"
 #include "vp9/common/vp9_filter.h"
+#include "vp9/common/vp9_idct.h"
 #include "vp9/common/vp9_mv.h"
 #include "vp9/common/vp9_scale.h"
 #include "vp9/common/vp9_seg_common.h"
@@ -98,6 +99,9 @@
   int_mv as_mv[2];  // first, second inter predictor motion vectors
 } b_mode_info;
 
+// Note that the rate-distortion optimization loop, bit-stream writer, and
+// decoder implementation modules critically rely on the enum entry values
+// specified herein. They should be refactored concurrently.
 typedef enum {
   NONE = -1,
   INTRA_FRAME = 0,
@@ -139,7 +143,8 @@
   INTERP_FILTER interp_filter;
 } MB_MODE_INFO;
 
-typedef struct {
+typedef struct MODE_INFO {
+  struct MODE_INFO *src_mi;
   MB_MODE_INFO mbmi;
   b_mode_info bmi[4];
 } MODE_INFO;
@@ -176,7 +181,7 @@
 };
 
 struct macroblockd_plane {
-  int16_t *dqcoeff;
+  tran_low_t *dqcoeff;
   PLANE_TYPE plane_type;
   int subsampling_x;
   int subsampling_y;
@@ -202,8 +207,7 @@
 
   int mi_stride;
 
-  // A NULL indicates that the 8x8 is not part of the image
-  MODE_INFO **mi;
+  MODE_INFO *mi;
 
   int up_available;
   int left_available;
@@ -223,11 +227,17 @@
   /* mc buffer */
   DECLARE_ALIGNED(16, uint8_t, mc_buf[80 * 2 * 80 * 2]);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  /* Bit depth: 8, 10, 12 */
+  int bd;
+  DECLARE_ALIGNED(16, uint16_t, mc_buf_high[80 * 2 * 80 * 2]);
+#endif
+
   int lossless;
 
   int corrupted;
 
-  DECLARE_ALIGNED(16, int16_t, dqcoeff[MAX_MB_PLANE][64 * 64]);
+  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[MAX_MB_PLANE][64 * 64]);
 
   ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
   ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];
@@ -245,7 +255,7 @@
 
 static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type,
                                   const MACROBLOCKD *xd) {
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  const MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
 
   if (plane_type != PLANE_TYPE_Y || is_inter_block(mbmi))
     return DCT_DCT;
@@ -254,7 +264,7 @@
 
 static INLINE TX_TYPE get_tx_type_4x4(PLANE_TYPE plane_type,
                                       const MACROBLOCKD *xd, int ib) {
-  const MODE_INFO *const mi = xd->mi[0];
+  const MODE_INFO *const mi = xd->mi[0].src_mi;
 
   if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(&mi->mbmi))
     return DCT_DCT;
diff --git a/vp9/common/vp9_common.h b/vp9/common/vp9_common.h
index 5587192..8305e7f 100644
--- a/vp9/common/vp9_common.h
+++ b/vp9/common/vp9_common.h
@@ -65,6 +65,18 @@
 }
 
 #if CONFIG_VP9_HIGHBITDEPTH
+static INLINE uint16_t clip_pixel_high(int val, int bd) {
+  switch (bd) {
+    case 8:
+    default:
+      return (uint16_t)clamp(val, 0, 255);
+    case 10:
+      return (uint16_t)clamp(val, 0, 1023);
+    case 12:
+      return (uint16_t)clamp(val, 0, 4095);
+  }
+}
+
 #define CONVERT_TO_SHORTPTR(x) ((uint16_t*)(((uintptr_t)x) << 1))
 #define CONVERT_TO_BYTEPTR(x) ((uint8_t*)(((uintptr_t)x) >> 1 ))
 #endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/common/vp9_convolve.c b/vp9/common/vp9_convolve.c
index d8aaf32..ad70e59 100644
--- a/vp9/common/vp9_convolve.c
+++ b/vp9/common/vp9_convolve.c
@@ -282,3 +282,280 @@
     dst += dst_stride;
   }
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
+                                uint8_t *dst8, ptrdiff_t dst_stride,
+                                const InterpKernel *x_filters,
+                                int x0_q4, int x_step_q4,
+                                int w, int h, int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_q4 = x0_q4;
+    for (x = 0; x < w; ++x) {
+      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_x[k] * x_filter[k];
+      dst[x] = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+      x_q4 += x_step_q4;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static void high_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
+                                    uint8_t *dst8, ptrdiff_t dst_stride,
+                                    const InterpKernel *x_filters,
+                                    int x0_q4, int x_step_q4,
+                                    int w, int h, int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_q4 = x0_q4;
+    for (x = 0; x < w; ++x) {
+      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_x[k] * x_filter[k];
+      dst[x] = ROUND_POWER_OF_TWO(dst[x] +
+          clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
+      x_q4 += x_step_q4;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static void high_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
+                               uint8_t *dst8, ptrdiff_t dst_stride,
+                               const InterpKernel *y_filters,
+                               int y0_q4, int y_step_q4, int w, int h,
+                               int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+  for (x = 0; x < w; ++x) {
+    int y_q4 = y0_q4;
+    for (y = 0; y < h; ++y) {
+      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = clip_pixel_high(
+          ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+      y_q4 += y_step_q4;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
+static void high_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
+                                   uint8_t *dst8, ptrdiff_t dst_stride,
+                                   const InterpKernel *y_filters,
+                                   int y0_q4, int y_step_q4, int w, int h,
+                                   int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+  for (x = 0; x < w; ++x) {
+    int y_q4 = y0_q4;
+    for (y = 0; y < h; ++y) {
+      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
+          clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
+      y_q4 += y_step_q4;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
+static void high_convolve(const uint8_t *src, ptrdiff_t src_stride,
+                          uint8_t *dst, ptrdiff_t dst_stride,
+                          const InterpKernel *const x_filters,
+                          int x0_q4, int x_step_q4,
+                          const InterpKernel *const y_filters,
+                          int y0_q4, int y_step_q4,
+                          int w, int h, int bd) {
+  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
+  // 2d filtering proceeds in 2 steps:
+  //   (1) Interpolate horizontally into an intermediate buffer, temp.
+  //   (2) Interpolate temp vertically to derive the sub-pixel result.
+  // Deriving the maximum number of rows in the temp buffer (135):
+  // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
+  // --Largest block size is 64x64 pixels.
+  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
+  //   original frame (in 1/16th pixel units).
+  // --Must round-up because block may be located at sub-pixel position.
+  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
+  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
+  uint16_t temp[64 * 135];
+  int intermediate_height = (((h - 1) * y_step_q4 + 15) >> 4) + SUBPEL_TAPS;
+
+  assert(w <= 64);
+  assert(h <= 64);
+  assert(y_step_q4 <= 32);
+  assert(x_step_q4 <= 32);
+
+  if (intermediate_height < h)
+    intermediate_height = h;
+
+  high_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1),
+                      src_stride, CONVERT_TO_BYTEPTR(temp), 64,
+                      x_filters, x0_q4, x_step_q4, w,
+                      intermediate_height, bd);
+  high_convolve_vert(CONVERT_TO_BYTEPTR(temp) + 64 * (SUBPEL_TAPS / 2 - 1),
+                     64, dst, dst_stride, y_filters, y0_q4, y_step_q4,
+                     w, h, bd);
+}
+
+
+void vp9_high_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x, int x_step_q4,
+                                const int16_t *filter_y, int y_step_q4,
+                                int w, int h, int bd) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+  const int x0_q4 = get_filter_offset(filter_x, filters_x);
+  (void)filter_y;
+  (void)y_step_q4;
+
+  high_convolve_horiz(src, src_stride, dst, dst_stride, filters_x,
+                      x0_q4, x_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+                                    uint8_t *dst, ptrdiff_t dst_stride,
+                                    const int16_t *filter_x, int x_step_q4,
+                                    const int16_t *filter_y, int y_step_q4,
+                                    int w, int h, int bd) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+  const int x0_q4 = get_filter_offset(filter_x, filters_x);
+  (void)filter_y;
+  (void)y_step_q4;
+
+  high_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
+                     x0_q4, x_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h, int bd) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+  (void)filter_x;
+  (void)x_step_q4;
+
+  high_convolve_vert(src, src_stride, dst, dst_stride, filters_y,
+                     y0_q4, y_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const int16_t *filter_x, int x_step_q4,
+                                   const int16_t *filter_y, int y_step_q4,
+                                   int w, int h, int bd) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+  (void)filter_x;
+  (void)x_step_q4;
+
+  high_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
+                         y0_q4, y_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
+                          uint8_t *dst, ptrdiff_t dst_stride,
+                          const int16_t *filter_x, int x_step_q4,
+                          const int16_t *filter_y, int y_step_q4,
+                          int w, int h, int bd) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+  const int x0_q4 = get_filter_offset(filter_x, filters_x);
+
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+  high_convolve(src, src_stride, dst, dst_stride,
+                filters_x, x0_q4, x_step_q4,
+                filters_y, y0_q4, y_step_q4, w, h, bd);
+}
+
+void vp9_high_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h, int bd) {
+  // Fixed size intermediate buffer places limits on parameters.
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, temp, 64 * 64);
+  assert(w <= 64);
+  assert(h <= 64);
+
+  vp9_high_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), 64,
+                       filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
+  vp9_high_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), 64, dst, dst_stride,
+                          NULL, 0, NULL, 0, w, h, bd);
+}
+
+void vp9_high_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
+                              uint8_t *dst8, ptrdiff_t dst_stride,
+                              const int16_t *filter_x, int filter_x_stride,
+                              const int16_t *filter_y, int filter_y_stride,
+                              int w, int h, int bd) {
+  int r;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  (void)filter_x;
+  (void)filter_y;
+  (void)filter_x_stride;
+  (void)filter_y_stride;
+  (void)bd;
+
+  for (r = h; r > 0; --r) {
+    vpx_memcpy(dst, src, w * sizeof(uint16_t));
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+void vp9_high_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride,
+                             uint8_t *dst8, ptrdiff_t dst_stride,
+                             const int16_t *filter_x, int filter_x_stride,
+                             const int16_t *filter_y, int filter_y_stride,
+                             int w, int h, int bd) {
+  int x, y;
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  (void)filter_x;
+  (void)filter_y;
+  (void)filter_x_stride;
+  (void)filter_y_stride;
+  (void)bd;
+
+  for (y = 0; y < h; ++y) {
+    for (x = 0; x < w; ++x) {
+      dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+#endif
diff --git a/vp9/common/vp9_convolve.h b/vp9/common/vp9_convolve.h
index 6bf71fc..faf70b1 100644
--- a/vp9/common/vp9_convolve.h
+++ b/vp9/common/vp9_convolve.h
@@ -23,6 +23,14 @@
                               const int16_t *filter_y, int y_step_q4,
                               int w, int h);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*high_convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const int16_t *filter_x, int x_step_q4,
+                                   const int16_t *filter_y, int y_step_q4,
+                                   int w, int h, int bd);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vp9/common/vp9_debugmodes.c b/vp9/common/vp9_debugmodes.c
index d2522bb..3f16841 100644
--- a/vp9/common/vp9_debugmodes.c
+++ b/vp9/common/vp9_debugmodes.c
@@ -26,7 +26,8 @@
                           size_t member_offset) {
   int mi_row, mi_col;
   int mi_index = 0;
-  MODE_INFO **mi = cm->mi_grid_visible;
+  // TODO(hkuang): Fix this debug function.
+  MODE_INFO **mi = NULL;
   int rows = cm->mi_rows;
   int cols = cm->mi_cols;
   char prefix = descriptor[0];
@@ -51,7 +52,8 @@
   int mi_col;
   int mi_index = 0;
   FILE *mvs = fopen(file, "a");
-  MODE_INFO **mi = cm->mi_grid_visible;
+  // TODO(hkuang): Fix this debug function.
+  MODE_INFO **mi = NULL;
   int rows = cm->mi_rows;
   int cols = cm->mi_cols;
 
diff --git a/vp9/common/vp9_entropy.c b/vp9/common/vp9_entropy.c
index 3a54de2..c3fdeb4 100644
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -23,6 +23,26 @@
 const vp9_prob vp9_cat6_prob[] = {
     254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129
 };
+#if CONFIG_VP9_HIGHBITDEPTH
+const vp9_prob vp9_cat1_prob_high10[] = { 159 };
+const vp9_prob vp9_cat2_prob_high10[] = { 165, 145 };
+const vp9_prob vp9_cat3_prob_high10[] = { 173, 148, 140 };
+const vp9_prob vp9_cat4_prob_high10[] = { 176, 155, 140, 135 };
+const vp9_prob vp9_cat5_prob_high10[] = { 180, 157, 141, 134, 130 };
+const vp9_prob vp9_cat6_prob_high10[] = {
+    255, 255, 254, 254, 254, 252, 249, 243,
+    230, 196, 177, 153, 140, 133, 130, 129
+};
+const vp9_prob vp9_cat1_prob_high12[] = { 159 };
+const vp9_prob vp9_cat2_prob_high12[] = { 165, 145 };
+const vp9_prob vp9_cat3_prob_high12[] = { 173, 148, 140 };
+const vp9_prob vp9_cat4_prob_high12[] = { 176, 155, 140, 135 };
+const vp9_prob vp9_cat5_prob_high12[] = { 180, 157, 141, 134, 130 };
+const vp9_prob vp9_cat6_prob_high12[] = {
+    255, 255, 255, 255, 254, 254, 254, 252, 249,
+    243, 230, 196, 177, 153, 140, 133, 130, 129
+};
+#endif
 
 const uint8_t vp9_coefband_trans_8x8plus[1024] = {
   0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4,
diff --git a/vp9/common/vp9_entropy.h b/vp9/common/vp9_entropy.h
index 8a10f23..8cdfc5c 100644
--- a/vp9/common/vp9_entropy.h
+++ b/vp9/common/vp9_entropy.h
@@ -58,6 +58,21 @@
 DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat5_prob[5]);
 DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat6_prob[14]);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat1_prob_high10[1]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat2_prob_high10[2]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat3_prob_high10[3]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat4_prob_high10[4]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat5_prob_high10[5]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat6_prob_high10[16]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat1_prob_high12[1]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat2_prob_high12[2]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat3_prob_high12[3]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat4_prob_high12[4]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat5_prob_high12[5]);
+DECLARE_ALIGNED(16, extern const uint8_t, vp9_cat6_prob_high12[18]);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 #define EOB_MODEL_TOKEN 3
 extern const vp9_tree_index vp9_coefmodel_tree[];
 
@@ -70,8 +85,16 @@
 
 // indexed by token value
 extern const vp9_extra_bit vp9_extra_bits[ENTROPY_TOKENS];
+#if CONFIG_VP9_HIGHBITDEPTH
+extern const vp9_extra_bit vp9_extra_bits_high10[ENTROPY_TOKENS];
+extern const vp9_extra_bit vp9_extra_bits_high12[ENTROPY_TOKENS];
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 #define DCT_MAX_VALUE           16384
+#if CONFIG_VP9_HIGHBITDEPTH
+#define DCT_MAX_VALUE_HIGH10    65536
+#define DCT_MAX_VALUE_HIGH12   262144
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 /* Coefficients are predicted via a 3-dimensional probability table. */
 
@@ -191,7 +214,7 @@
 
 static INLINE const scan_order *get_scan(const MACROBLOCKD *xd, TX_SIZE tx_size,
                                          PLANE_TYPE type, int block_idx) {
-  const MODE_INFO *const mi = xd->mi[0];
+  const MODE_INFO *const mi = xd->mi[0].src_mi;
 
   if (is_inter_block(&mi->mbmi) || type != PLANE_TYPE_Y || xd->lossless) {
     return &vp9_default_scan_orders[tx_size];
diff --git a/vp9/common/vp9_frame_buffers.c b/vp9/common/vp9_frame_buffers.c
index 733b3a9..34795b7 100644
--- a/vp9/common/vp9_frame_buffers.c
+++ b/vp9/common/vp9_frame_buffers.c
@@ -61,6 +61,10 @@
     if (!int_fb_list->int_fb[i].data)
       return -1;
 
+    // This memset is needed for fixing valgrind error from C loop filter
+    // due to access uninitialized memory in frame border. It could be
+    // removed if border is totally removed.
+    vpx_memset(int_fb_list->int_fb[i].data, 0, min_size);
     int_fb_list->int_fb[i].size = min_size;
   }
 
diff --git a/vp9/common/vp9_idct.c b/vp9/common/vp9_idct.c
index 856d41e..b196fc5 100644
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -18,14 +18,47 @@
 #include "vp9/common/vp9_common.h"
 #include "vp9/common/vp9_idct.h"
 
-void vp9_iwht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
+#if CONFIG_EMULATE_HARDWARE_HIGHBITDEPTH
+// When CONFIG_EMULATE_HW_HIGHBITDEPTH is 1 the transform performs strict
+// overflow wrapping to match expected hardware implementations.
+// bd of 8 uses trans_low with 16bits, need to remove 16bits
+// bd of 10 uses trans_low with 18bits, need to remove 14bits
+// bd of 12 uses trans_low with 20bits, need to remove 12bits
+// bd of x uses trans_low with 8+x bits, need to remove 24-x bits
+#define WRAPLOW(x) ((((int32_t)x) << (24 - bd)) >> (24 - bd))
+#else
+#define WRAPLOW(x) (x)
+#endif  // CONFIG_EMULATE_HARDWARE_HIGHBITDEPTH
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE tran_low_t clamp_high(tran_high_t value, tran_low_t low,
+                                    tran_low_t high) {
+  return value < low ? low : (value > high ? high : value);
+}
+
+static INLINE tran_low_t clip_pixel_bd_high(tran_high_t dest,
+                                            tran_high_t trans, int bd) {
+  trans = WRAPLOW(trans);
+  switch (bd) {
+    case 8:
+    default:
+      return clamp_high(WRAPLOW(dest + trans), 0, 255);
+    case 10:
+      return clamp_high(WRAPLOW(dest + trans), 0, 1023);
+    case 12:
+      return clamp_high(WRAPLOW(dest + trans), 0, 4095);
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+void vp9_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
 /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
    0.5 shifts per pixel. */
   int i;
-  int16_t output[16];
-  int a1, b1, c1, d1, e1;
-  const int16_t *ip = input;
-  int16_t *op = output;
+  tran_low_t output[16];
+  tran_high_t a1, b1, c1, d1, e1;
+  const tran_low_t *ip = input;
+  tran_low_t *op = output;
 
   for (i = 0; i < 4; i++) {
     a1 = ip[0] >> UNIT_QUANT_SHIFT;
@@ -70,12 +103,12 @@
   }
 }
 
-void vp9_iwht4x4_1_add_c(const int16_t *in, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest, int dest_stride) {
   int i;
-  int a1, e1;
-  int16_t tmp[4];
-  const int16_t *ip = in;
-  int16_t *op = tmp;
+  tran_high_t a1, e1;
+  tran_low_t tmp[4];
+  const tran_low_t *ip = in;
+  tran_low_t *op = tmp;
 
   a1 = ip[0] >> UNIT_QUANT_SHIFT;
   e1 = a1 >> 1;
@@ -96,9 +129,9 @@
   }
 }
 
-static void idct4(const int16_t *input, int16_t *output) {
-  int16_t step[4];
-  int temp1, temp2;
+static void idct4(const tran_low_t *input, tran_low_t *output) {
+  tran_low_t step[4];
+  tran_high_t temp1, temp2;
   // stage 1
   temp1 = (input[0] + input[2]) * cospi_16_64;
   temp2 = (input[0] - input[2]) * cospi_16_64;
@@ -116,11 +149,11 @@
   output[3] = step[0] - step[3];
 }
 
-void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[4 * 4];
-  int16_t *outptr = out;
+void vp9_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
+  tran_low_t out[4 * 4];
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[4], temp_out[4];
+  tran_low_t temp_in[4], temp_out[4];
 
   // Rows
   for (i = 0; i < 4; ++i) {
@@ -140,10 +173,11 @@
   }
 }
 
-void vp9_idct4x4_1_add_c(const int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest,
+                         int dest_stride) {
   int i;
-  int a1;
-  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  tran_high_t a1;
+  tran_low_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 4);
 
@@ -156,9 +190,9 @@
   }
 }
 
-static void idct8(const int16_t *input, int16_t *output) {
-  int16_t step1[8], step2[8];
-  int temp1, temp2;
+static void idct8(const tran_low_t *input, tran_low_t *output) {
+  tran_low_t step1[8], step2[8];
+  tran_high_t temp1, temp2;
   // stage 1
   step1[0] = input[0];
   step1[2] = input[4];
@@ -201,11 +235,11 @@
   output[7] = step1[0] - step1[7];
 }
 
-void vp9_idct8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[8 * 8];
-  int16_t *outptr = out;
+void vp9_idct8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
+  tran_low_t out[8 * 8];
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[8], temp_out[8];
+  tran_low_t temp_in[8], temp_out[8];
 
   // First transform rows
   for (i = 0; i < 8; ++i) {
@@ -225,10 +259,10 @@
   }
 }
 
-void vp9_idct8x8_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct8x8_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i, j;
-  int a1;
-  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  tran_high_t a1;
+  tran_low_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 5);
   for (j = 0; j < 8; ++j) {
@@ -238,13 +272,13 @@
   }
 }
 
-static void iadst4(const int16_t *input, int16_t *output) {
-  int s0, s1, s2, s3, s4, s5, s6, s7;
+static void iadst4(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
 
-  int x0 = input[0];
-  int x1 = input[1];
-  int x2 = input[2];
-  int x3 = input[3];
+  tran_high_t x0 = input[0];
+  tran_high_t x1 = input[1];
+  tran_high_t x2 = input[2];
+  tran_high_t x3 = input[3];
 
   if (!(x0 | x1 | x2 | x3)) {
     output[0] = output[1] = output[2] = output[3] = 0;
@@ -280,7 +314,7 @@
   output[3] = dct_const_round_shift(s3);
 }
 
-void vp9_iht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride,
+void vp9_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride,
                          int tx_type) {
   const transform_2d IHT_4[] = {
     { idct4, idct4  },  // DCT_DCT  = 0
@@ -290,9 +324,9 @@
   };
 
   int i, j;
-  int16_t out[4 * 4];
-  int16_t *outptr = out;
-  int16_t temp_in[4], temp_out[4];
+  tran_low_t out[4 * 4];
+  tran_low_t *outptr = out;
+  tran_low_t temp_in[4], temp_out[4];
 
   // inverse transform row vectors
   for (i = 0; i < 4; ++i) {
@@ -311,17 +345,17 @@
                                   + dest[j * stride + i]);
   }
 }
-static void iadst8(const int16_t *input, int16_t *output) {
+static void iadst8(const tran_low_t *input, tran_low_t *output) {
   int s0, s1, s2, s3, s4, s5, s6, s7;
 
-  int x0 = input[7];
-  int x1 = input[0];
-  int x2 = input[5];
-  int x3 = input[2];
-  int x4 = input[3];
-  int x5 = input[4];
-  int x6 = input[1];
-  int x7 = input[6];
+  tran_high_t x0 = input[7];
+  tran_high_t x1 = input[0];
+  tran_high_t x2 = input[5];
+  tran_high_t x3 = input[2];
+  tran_high_t x4 = input[3];
+  tran_high_t x5 = input[4];
+  tran_high_t x6 = input[1];
+  tran_high_t x7 = input[6];
 
   if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
     output[0] = output[1] = output[2] = output[3] = output[4]
@@ -395,12 +429,12 @@
   { iadst8, iadst8 }   // ADST_ADST = 3
 };
 
-void vp9_iht8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride,
+void vp9_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride,
                          int tx_type) {
   int i, j;
-  int16_t out[8 * 8];
-  int16_t *outptr = out;
-  int16_t temp_in[8], temp_out[8];
+  tran_low_t out[8 * 8];
+  tran_low_t *outptr = out;
+  tran_low_t temp_in[8], temp_out[8];
   const transform_2d ht = IHT_8[tx_type];
 
   // inverse transform row vectors
@@ -421,11 +455,11 @@
   }
 }
 
-void vp9_idct8x8_12_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[8 * 8] = { 0 };
-  int16_t *outptr = out;
+void vp9_idct8x8_12_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
+  tran_low_t out[8 * 8] = { 0 };
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[8], temp_out[8];
+  tran_low_t temp_in[8], temp_out[8];
 
   // First transform rows
   // only first 4 row has non-zero coefs
@@ -446,9 +480,9 @@
   }
 }
 
-static void idct16(const int16_t *input, int16_t *output) {
-  int16_t step1[16], step2[16];
-  int temp1, temp2;
+static void idct16(const tran_low_t *input, tran_low_t *output) {
+  tran_low_t step1[16], step2[16];
+  tran_high_t temp1, temp2;
 
   // stage 1
   step1[0] = input[0/2];
@@ -611,11 +645,12 @@
   output[15] = step2[0] - step2[15];
 }
 
-void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[16 * 16];
-  int16_t *outptr = out;
+void vp9_idct16x16_256_add_c(const tran_low_t *input, uint8_t *dest,
+                             int stride) {
+  tran_low_t out[16 * 16];
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[16], temp_out[16];
+  tran_low_t temp_in[16], temp_out[16];
 
   // First transform rows
   for (i = 0; i < 16; ++i) {
@@ -635,25 +670,26 @@
   }
 }
 
-static void iadst16(const int16_t *input, int16_t *output) {
-  int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
+static void iadst16(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8;
+  tran_high_t s9, s10, s11, s12, s13, s14, s15;
 
-  int x0 = input[15];
-  int x1 = input[0];
-  int x2 = input[13];
-  int x3 = input[2];
-  int x4 = input[11];
-  int x5 = input[4];
-  int x6 = input[9];
-  int x7 = input[6];
-  int x8 = input[7];
-  int x9 = input[8];
-  int x10 = input[5];
-  int x11 = input[10];
-  int x12 = input[3];
-  int x13 = input[12];
-  int x14 = input[1];
-  int x15 = input[14];
+  tran_high_t x0 = input[15];
+  tran_high_t x1 = input[0];
+  tran_high_t x2 = input[13];
+  tran_high_t x3 = input[2];
+  tran_high_t x4 = input[11];
+  tran_high_t x5 = input[4];
+  tran_high_t x6 = input[9];
+  tran_high_t x7 = input[6];
+  tran_high_t x8 = input[7];
+  tran_high_t x9 = input[8];
+  tran_high_t x10 = input[5];
+  tran_high_t x11 = input[10];
+  tran_high_t x12 = input[3];
+  tran_high_t x13 = input[12];
+  tran_high_t x14 = input[1];
+  tran_high_t x15 = input[14];
 
   if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8
            | x9 | x10 | x11 | x12 | x13 | x14 | x15)) {
@@ -813,12 +849,12 @@
   { iadst16, iadst16 }   // ADST_ADST = 3
 };
 
-void vp9_iht16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride,
+void vp9_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest, int stride,
                             int tx_type) {
   int i, j;
-  int16_t out[16 * 16];
-  int16_t *outptr = out;
-  int16_t temp_in[16], temp_out[16];
+  tran_low_t out[16 * 16];
+  tran_low_t *outptr = out;
+  tran_low_t temp_in[16], temp_out[16];
   const transform_2d ht = IHT_16[tx_type];
 
   // Rows
@@ -839,11 +875,12 @@
   }
 }
 
-void vp9_idct16x16_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[16 * 16] = { 0 };
-  int16_t *outptr = out;
+void vp9_idct16x16_10_add_c(const tran_low_t *input, uint8_t *dest,
+                            int stride) {
+  tran_low_t out[16 * 16] = { 0 };
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[16], temp_out[16];
+  tran_low_t temp_in[16], temp_out[16];
 
   // First transform rows. Since all non-zero dct coefficients are in
   // upper-left 4x4 area, we only need to calculate first 4 rows here.
@@ -864,10 +901,10 @@
   }
 }
 
-void vp9_idct16x16_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct16x16_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i, j;
-  int a1;
-  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  tran_high_t a1;
+  tran_low_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 6);
   for (j = 0; j < 16; ++j) {
@@ -877,9 +914,9 @@
   }
 }
 
-static void idct32(const int16_t *input, int16_t *output) {
-  int16_t step1[32], step2[32];
-  int temp1, temp2;
+static void idct32(const tran_low_t *input, tran_low_t *output) {
+  tran_low_t step1[32], step2[32];
+  tran_high_t temp1, temp2;
 
   // stage 1
   step1[0] = input[0];
@@ -1244,11 +1281,12 @@
   output[31] = step1[0] - step1[31];
 }
 
-void vp9_idct32x32_1024_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[32 * 32];
-  int16_t *outptr = out;
+void vp9_idct32x32_1024_add_c(const tran_low_t *input, uint8_t *dest,
+                              int stride) {
+  tran_low_t out[32 * 32];
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[32], temp_out[32];
+  tran_low_t temp_in[32], temp_out[32];
 
   // Rows
   for (i = 0; i < 32; ++i) {
@@ -1265,7 +1303,7 @@
     if (zero_coeff[0] | zero_coeff[1])
       idct32(input, outptr);
     else
-      vpx_memset(outptr, 0, sizeof(int16_t) * 32);
+      vpx_memset(outptr, 0, sizeof(tran_low_t) * 32);
     input += 32;
     outptr += 32;
   }
@@ -1281,11 +1319,12 @@
   }
 }
 
-void vp9_idct32x32_34_add_c(const int16_t *input, uint8_t *dest, int stride) {
-  int16_t out[32 * 32] = {0};
-  int16_t *outptr = out;
+void vp9_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest,
+                            int stride) {
+  tran_low_t out[32 * 32] = {0};
+  tran_low_t *outptr = out;
   int i, j;
-  int16_t temp_in[32], temp_out[32];
+  tran_low_t temp_in[32], temp_out[32];
 
   // Rows
   // only upper-left 8x8 has non-zero coeff
@@ -1306,11 +1345,11 @@
   }
 }
 
-void vp9_idct32x32_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
+void vp9_idct32x32_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i, j;
-  int a1;
+  tran_high_t a1;
 
-  int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+  tran_low_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 6);
 
@@ -1322,7 +1361,8 @@
 }
 
 // idct
-void vp9_idct4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                     int eob) {
   if (eob > 1)
     vp9_idct4x4_16_add(input, dest, stride);
   else
@@ -1330,14 +1370,16 @@
 }
 
 
-void vp9_iwht4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                     int eob) {
   if (eob > 1)
     vp9_iwht4x4_16_add(input, dest, stride);
   else
     vp9_iwht4x4_1_add(input, dest, stride);
 }
 
-void vp9_idct8x8_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+void vp9_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
+                     int eob) {
   // If dc is 1, then input[0] is the reconstructed value, do not need
   // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
 
@@ -1354,7 +1396,7 @@
     vp9_idct8x8_64_add(input, dest, stride);
 }
 
-void vp9_idct16x16_add(const int16_t *input, uint8_t *dest, int stride,
+void vp9_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride,
                        int eob) {
   /* The calculation can be simplified if there are not many non-zero dct
    * coefficients. Use eobs to separate different cases. */
@@ -1367,7 +1409,7 @@
     vp9_idct16x16_256_add(input, dest, stride);
 }
 
-void vp9_idct32x32_add(const int16_t *input, uint8_t *dest, int stride,
+void vp9_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
                        int eob) {
   if (eob == 1)
     vp9_idct32x32_1_add(input, dest, stride);
@@ -1379,7 +1421,7 @@
 }
 
 // iht
-void vp9_iht4x4_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                     int stride, int eob) {
   if (tx_type == DCT_DCT)
     vp9_idct4x4_add(input, dest, stride, eob);
@@ -1387,7 +1429,7 @@
     vp9_iht4x4_16_add(input, dest, stride, tx_type);
 }
 
-void vp9_iht8x8_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                     int stride, int eob) {
   if (tx_type == DCT_DCT) {
     vp9_idct8x8_add(input, dest, stride, eob);
@@ -1396,7 +1438,7 @@
   }
 }
 
-void vp9_iht16x16_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                       int stride, int eob) {
   if (tx_type == DCT_DCT) {
     vp9_idct16x16_add(input, dest, stride, eob);
@@ -1404,3 +1446,1433 @@
     vp9_iht16x16_256_add(input, dest, stride, tx_type);
   }
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
+                               int stride, int bd) {
+  /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
+     0.5 shifts per pixel. */
+  int i;
+  tran_low_t output[16];
+  tran_high_t a1, b1, c1, d1, e1;
+  const tran_low_t *ip = input;
+  tran_low_t *op = output;
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  for (i = 0; i < 4; i++) {
+    a1 = ip[0] >> UNIT_QUANT_SHIFT;
+    c1 = ip[1] >> UNIT_QUANT_SHIFT;
+    d1 = ip[2] >> UNIT_QUANT_SHIFT;
+    b1 = ip[3] >> UNIT_QUANT_SHIFT;
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
+    op[0] = WRAPLOW(a1);
+    op[1] = WRAPLOW(b1);
+    op[2] = WRAPLOW(c1);
+    op[3] = WRAPLOW(d1);
+    ip += 4;
+    op += 4;
+  }
+
+  ip = output;
+  for (i = 0; i < 4; i++) {
+    a1 = ip[4 * 0];
+    c1 = ip[4 * 1];
+    d1 = ip[4 * 2];
+    b1 = ip[4 * 3];
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
+    dest[stride * 0] = clip_pixel_bd_high(dest[stride * 0], a1, bd);
+    dest[stride * 1] = clip_pixel_bd_high(dest[stride * 1], b1, bd);
+    dest[stride * 2] = clip_pixel_bd_high(dest[stride * 2], c1, bd);
+    dest[stride * 3] = clip_pixel_bd_high(dest[stride * 3], d1, bd);
+
+    ip++;
+    dest++;
+  }
+}
+
+static void high_idct4(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_low_t step[4];
+  tran_high_t temp1, temp2;
+  (void) bd;
+  // stage 1
+  temp1 = (input[0] + input[2]) * cospi_16_64;
+  temp2 = (input[0] - input[2]) * cospi_16_64;
+  step[0] = WRAPLOW(dct_const_round_shift(temp1));
+  step[1] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
+  temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
+  step[2] = WRAPLOW(dct_const_round_shift(temp1));
+  step[3] = WRAPLOW(dct_const_round_shift(temp2));
+
+  // stage 2
+  output[0] = WRAPLOW(step[0] + step[3]);
+  output[1] = WRAPLOW(step[1] + step[2]);
+  output[2] = WRAPLOW(step[1] - step[2]);
+  output[3] = WRAPLOW(step[0] - step[3]);
+}
+
+void vp9_high_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8,
+                              int dest_stride, int bd) {
+  int i;
+  tran_high_t a1, e1;
+  tran_low_t tmp[4];
+  const tran_low_t *ip = in;
+  tran_low_t *op = tmp;
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+  (void) bd;
+
+  a1 = ip[0] >> UNIT_QUANT_SHIFT;
+  e1 = a1 >> 1;
+  a1 -= e1;
+  op[0] = WRAPLOW(a1);
+  op[1] = op[2] = op[3] = WRAPLOW(e1);
+
+  ip = tmp;
+  for (i = 0; i < 4; i++) {
+    e1 = ip[0] >> 1;
+    a1 = ip[0] - e1;
+    dest[dest_stride * 0] = clip_pixel_bd_high(dest[dest_stride * 0], a1, bd);
+    dest[dest_stride * 1] = clip_pixel_bd_high(dest[dest_stride * 1], e1, bd);
+    dest[dest_stride * 2] = clip_pixel_bd_high(dest[dest_stride * 2], e1, bd);
+    dest[dest_stride * 3] = clip_pixel_bd_high(dest[dest_stride * 3], e1, bd);
+    ip++;
+    dest++;
+  }
+}
+
+void vp9_high_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
+                               int stride, int bd) {
+  tran_low_t out[4 * 4];
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[4], temp_out[4];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // Rows
+  for (i = 0; i < 4; ++i) {
+    high_idct4(input, outptr, bd);
+    input += 4;
+    outptr += 4;
+  }
+
+  // Columns
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = out[j * 4 + i];
+    high_idct4(temp_in, temp_out, bd);
+    for (j = 0; j < 4; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
+  }
+}
+
+void vp9_high_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest8,
+                              int dest_stride, int bd) {
+  int i;
+  tran_high_t a1;
+  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
+  a1 = ROUND_POWER_OF_TWO(out, 4);
+
+  for (i = 0; i < 4; i++) {
+    dest[0] = clip_pixel_bd_high(dest[0], a1, bd);
+    dest[1] = clip_pixel_bd_high(dest[1], a1, bd);
+    dest[2] = clip_pixel_bd_high(dest[2], a1, bd);
+    dest[3] = clip_pixel_bd_high(dest[3], a1, bd);
+    dest += dest_stride;
+  }
+}
+
+static void high_idct8(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_low_t step1[8], step2[8];
+  tran_high_t temp1, temp2;
+  // stage 1
+  step1[0] = input[0];
+  step1[2] = input[4];
+  step1[1] = input[2];
+  step1[3] = input[6];
+  temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
+  temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
+  step1[4] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[7] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
+  temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
+  step1[5] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[6] = WRAPLOW(dct_const_round_shift(temp2));
+
+  // stage 2 & stage 3 - even half
+  high_idct4(step1, step1, bd);
+
+  // stage 2 - odd half
+  step2[4] = WRAPLOW(step1[4] + step1[5]);
+  step2[5] = WRAPLOW(step1[4] - step1[5]);
+  step2[6] = WRAPLOW(-step1[6] + step1[7]);
+  step2[7] = WRAPLOW(step1[6] + step1[7]);
+
+  // stage 3 - odd half
+  step1[4] = step2[4];
+  temp1 = (step2[6] - step2[5]) * cospi_16_64;
+  temp2 = (step2[5] + step2[6]) * cospi_16_64;
+  step1[5] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[6] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[7] = step2[7];
+
+  // stage 4
+  output[0] = WRAPLOW(step1[0] + step1[7]);
+  output[1] = WRAPLOW(step1[1] + step1[6]);
+  output[2] = WRAPLOW(step1[2] + step1[5]);
+  output[3] = WRAPLOW(step1[3] + step1[4]);
+  output[4] = WRAPLOW(step1[3] - step1[4]);
+  output[5] = WRAPLOW(step1[2] - step1[5]);
+  output[6] = WRAPLOW(step1[1] - step1[6]);
+  output[7] = WRAPLOW(step1[0] - step1[7]);
+}
+
+void vp9_high_idct8x8_64_add_c(const tran_low_t *input, uint8_t *dest8,
+                               int stride, int bd) {
+  tran_low_t out[8 * 8];
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[8], temp_out[8];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // First transform rows.
+  for (i = 0; i < 8; ++i) {
+    high_idct8(input, outptr, bd);
+    input += 8;
+    outptr += 8;
+  }
+
+  // Then transform columns.
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j * 8 + i];
+    high_idct8(temp_in, temp_out, bd);
+    for (j = 0; j < 8; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(dest[j * stride + i],
+                                        ROUND_POWER_OF_TWO(temp_out[j], 5),
+                                        bd);
+  }
+}
+
+void vp9_high_idct8x8_1_add_c(const tran_low_t *input, uint8_t *dest8,
+                              int stride, int bd) {
+  int i, j;
+  tran_high_t a1;
+  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+  out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
+  a1 = ROUND_POWER_OF_TWO(out, 5);
+  for (j = 0; j < 8; ++j) {
+    for (i = 0; i < 8; ++i)
+      dest[i] = clip_pixel_bd_high(dest[i], a1, bd);
+    dest += stride;
+  }
+}
+
+static void high_iadst4(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
+
+  tran_high_t x0 = input[0];
+  tran_high_t x1 = input[1];
+  tran_high_t x2 = input[2];
+  tran_high_t x3 = input[3];
+  (void) bd;
+
+  if (!(x0 | x1 | x2 | x3)) {
+    vpx_memset(output, 0, 4 * sizeof(*output));
+    return;
+  }
+
+  s0 = sinpi_1_9 * x0;
+  s1 = sinpi_2_9 * x0;
+  s2 = sinpi_3_9 * x1;
+  s3 = sinpi_4_9 * x2;
+  s4 = sinpi_1_9 * x2;
+  s5 = sinpi_2_9 * x3;
+  s6 = sinpi_4_9 * x3;
+  s7 = x0 - x2 + x3;
+
+  x0 = s0 + s3 + s5;
+  x1 = s1 - s4 - s6;
+  x2 = sinpi_3_9 * s7;
+  x3 = s2;
+
+  s0 = x0 + x3;
+  s1 = x1 + x3;
+  s2 = x2;
+  s3 = x0 + x1 - x3;
+
+  // 1-D transform scaling factor is sqrt(2).
+  // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
+  // + 1b (addition) = 29b.
+  // Hence the output bit depth is 15b.
+  output[0] = WRAPLOW(dct_const_round_shift(s0));
+  output[1] = WRAPLOW(dct_const_round_shift(s1));
+  output[2] = WRAPLOW(dct_const_round_shift(s2));
+  output[3] = WRAPLOW(dct_const_round_shift(s3));
+}
+
+void vp9_high_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
+                              int stride, int tx_type, int bd) {
+  const high_transform_2d IHT_4[] = {
+    { high_idct4, high_idct4  },    // DCT_DCT  = 0
+    { high_iadst4, high_idct4 },    // ADST_DCT = 1
+    { high_idct4, high_iadst4 },    // DCT_ADST = 2
+    { high_iadst4, high_iadst4 }    // ADST_ADST = 3
+  };
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  int i, j;
+  tran_low_t out[4 * 4];
+  tran_low_t *outptr = out;
+  tran_low_t temp_in[4], temp_out[4];
+
+  // Inverse transform row vectors.
+  for (i = 0; i < 4; ++i) {
+    IHT_4[tx_type].rows(input, outptr, bd);
+    input  += 4;
+    outptr += 4;
+  }
+
+  // Inverse transform column vectors.
+  for (i = 0; i < 4; ++i) {
+    for (j = 0; j < 4; ++j)
+      temp_in[j] = out[j * 4 + i];
+    IHT_4[tx_type].cols(temp_in, temp_out, bd);
+    for (j = 0; j < 4; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
+  }
+}
+
+static void high_iadst8(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
+
+  tran_high_t x0 = input[7];
+  tran_high_t x1 = input[0];
+  tran_high_t x2 = input[5];
+  tran_high_t x3 = input[2];
+  tran_high_t x4 = input[3];
+  tran_high_t x5 = input[4];
+  tran_high_t x6 = input[1];
+  tran_high_t x7 = input[6];
+  (void) bd;
+
+  if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
+    vpx_memset(output, 0, 8 * sizeof(*output));
+    return;
+  }
+
+  // stage 1
+  s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
+  s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
+  s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
+  s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
+  s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
+  s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
+  s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
+  s7 = cospi_6_64  * x6 - cospi_26_64 * x7;
+
+  x0 = WRAPLOW(dct_const_round_shift(s0 + s4));
+  x1 = WRAPLOW(dct_const_round_shift(s1 + s5));
+  x2 = WRAPLOW(dct_const_round_shift(s2 + s6));
+  x3 = WRAPLOW(dct_const_round_shift(s3 + s7));
+  x4 = WRAPLOW(dct_const_round_shift(s0 - s4));
+  x5 = WRAPLOW(dct_const_round_shift(s1 - s5));
+  x6 = WRAPLOW(dct_const_round_shift(s2 - s6));
+  x7 = WRAPLOW(dct_const_round_shift(s3 - s7));
+
+  // stage 2
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 =  cospi_8_64  * x4 + cospi_24_64 * x5;
+  s5 =  cospi_24_64 * x4 - cospi_8_64  * x5;
+  s6 = -cospi_24_64 * x6 + cospi_8_64  * x7;
+  s7 =  cospi_8_64  * x6 + cospi_24_64 * x7;
+
+  x0 = s0 + s2;
+  x1 = s1 + s3;
+  x2 = s0 - s2;
+  x3 = s1 - s3;
+  x4 = WRAPLOW(dct_const_round_shift(s4 + s6));
+  x5 = WRAPLOW(dct_const_round_shift(s5 + s7));
+  x6 = WRAPLOW(dct_const_round_shift(s4 - s6));
+  x7 = WRAPLOW(dct_const_round_shift(s5 - s7));
+
+  // stage 3
+  s2 = cospi_16_64 * (x2 + x3);
+  s3 = cospi_16_64 * (x2 - x3);
+  s6 = cospi_16_64 * (x6 + x7);
+  s7 = cospi_16_64 * (x6 - x7);
+
+  x2 = WRAPLOW(dct_const_round_shift(s2));
+  x3 = WRAPLOW(dct_const_round_shift(s3));
+  x6 = WRAPLOW(dct_const_round_shift(s6));
+  x7 = WRAPLOW(dct_const_round_shift(s7));
+
+  output[0] = WRAPLOW(x0);
+  output[1] = WRAPLOW(-x4);
+  output[2] = WRAPLOW(x6);
+  output[3] = WRAPLOW(-x2);
+  output[4] = WRAPLOW(x3);
+  output[5] = WRAPLOW(-x7);
+  output[6] = WRAPLOW(x5);
+  output[7] = WRAPLOW(-x1);
+}
+
+static const high_transform_2d HIGH_IHT_8[] = {
+  { high_idct8,  high_idct8  },  // DCT_DCT  = 0
+  { high_iadst8, high_idct8  },  // ADST_DCT = 1
+  { high_idct8,  high_iadst8 },  // DCT_ADST = 2
+  { high_iadst8, high_iadst8 }   // ADST_ADST = 3
+};
+
+void vp9_high_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest8,
+                              int stride, int tx_type, int bd) {
+  int i, j;
+  tran_low_t out[8 * 8];
+  tran_low_t *outptr = out;
+  tran_low_t temp_in[8], temp_out[8];
+  const high_transform_2d ht = HIGH_IHT_8[tx_type];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // Inverse transform row vectors.
+  for (i = 0; i < 8; ++i) {
+    ht.rows(input, outptr, bd);
+    input += 8;
+    outptr += 8;
+  }
+
+  // Inverse transform column vectors.
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j * 8 + i];
+    ht.cols(temp_in, temp_out, bd);
+    for (j = 0; j < 8; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
+  }
+}
+
+void vp9_high_idct8x8_10_add_c(const tran_low_t *input, uint8_t *dest8,
+                               int stride, int bd) {
+  tran_low_t out[8 * 8] = { 0 };
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[8], temp_out[8];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // First transform rows.
+  // Only first 4 row has non-zero coefs.
+  for (i = 0; i < 4; ++i) {
+    high_idct8(input, outptr, bd);
+    input += 8;
+    outptr += 8;
+  }
+  // Then transform columns.
+  for (i = 0; i < 8; ++i) {
+    for (j = 0; j < 8; ++j)
+      temp_in[j] = out[j * 8 + i];
+    high_idct8(temp_in, temp_out, bd);
+    for (j = 0; j < 8; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
+  }
+}
+
+static void high_idct16(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_low_t step1[16], step2[16];
+  tran_high_t temp1, temp2;
+  (void) bd;
+
+  // stage 1
+  step1[0] = input[0/2];
+  step1[1] = input[16/2];
+  step1[2] = input[8/2];
+  step1[3] = input[24/2];
+  step1[4] = input[4/2];
+  step1[5] = input[20/2];
+  step1[6] = input[12/2];
+  step1[7] = input[28/2];
+  step1[8] = input[2/2];
+  step1[9] = input[18/2];
+  step1[10] = input[10/2];
+  step1[11] = input[26/2];
+  step1[12] = input[6/2];
+  step1[13] = input[22/2];
+  step1[14] = input[14/2];
+  step1[15] = input[30/2];
+
+  // stage 2
+  step2[0] = step1[0];
+  step2[1] = step1[1];
+  step2[2] = step1[2];
+  step2[3] = step1[3];
+  step2[4] = step1[4];
+  step2[5] = step1[5];
+  step2[6] = step1[6];
+  step2[7] = step1[7];
+
+  temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
+  temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
+  step2[8] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[15] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
+  temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
+  step2[9] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[14] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
+  temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
+  step2[10] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[13] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
+  temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
+  step2[11] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[12] = WRAPLOW(dct_const_round_shift(temp2));
+
+  // stage 3
+  step1[0] = step2[0];
+  step1[1] = step2[1];
+  step1[2] = step2[2];
+  step1[3] = step2[3];
+
+  temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
+  temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
+  step1[4] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[7] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
+  temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
+  step1[5] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[6] = WRAPLOW(dct_const_round_shift(temp2));
+
+  step1[8] = WRAPLOW(step2[8] + step2[9]);
+  step1[9] = WRAPLOW(step2[8] - step2[9]);
+  step1[10] = WRAPLOW(-step2[10] + step2[11]);
+  step1[11] = WRAPLOW(step2[10] + step2[11]);
+  step1[12] = WRAPLOW(step2[12] + step2[13]);
+  step1[13] = WRAPLOW(step2[12] - step2[13]);
+  step1[14] = WRAPLOW(-step2[14] + step2[15]);
+  step1[15] = WRAPLOW(step2[14] + step2[15]);
+
+  // stage 4
+  temp1 = (step1[0] + step1[1]) * cospi_16_64;
+  temp2 = (step1[0] - step1[1]) * cospi_16_64;
+  step2[0] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[1] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
+  temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
+  step2[2] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[3] = WRAPLOW(dct_const_round_shift(temp2));
+  step2[4] = WRAPLOW(step1[4] + step1[5]);
+  step2[5] = WRAPLOW(step1[4] - step1[5]);
+  step2[6] = WRAPLOW(-step1[6] + step1[7]);
+  step2[7] = WRAPLOW(step1[6] + step1[7]);
+
+  step2[8] = step1[8];
+  step2[15] = step1[15];
+  temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
+  temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
+  step2[9] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[14] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
+  temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
+  step2[10] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[13] = WRAPLOW(dct_const_round_shift(temp2));
+  step2[11] = step1[11];
+  step2[12] = step1[12];
+
+  // stage 5
+  step1[0] = WRAPLOW(step2[0] + step2[3]);
+  step1[1] = WRAPLOW(step2[1] + step2[2]);
+  step1[2] = WRAPLOW(step2[1] - step2[2]);
+  step1[3] = WRAPLOW(step2[0] - step2[3]);
+  step1[4] = step2[4];
+  temp1 = (step2[6] - step2[5]) * cospi_16_64;
+  temp2 = (step2[5] + step2[6]) * cospi_16_64;
+  step1[5] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[6] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[7] = step2[7];
+
+  step1[8] = WRAPLOW(step2[8] + step2[11]);
+  step1[9] = WRAPLOW(step2[9] + step2[10]);
+  step1[10] = WRAPLOW(step2[9] - step2[10]);
+  step1[11] = WRAPLOW(step2[8] - step2[11]);
+  step1[12] = WRAPLOW(-step2[12] + step2[15]);
+  step1[13] = WRAPLOW(-step2[13] + step2[14]);
+  step1[14] = WRAPLOW(step2[13] + step2[14]);
+  step1[15] = WRAPLOW(step2[12] + step2[15]);
+
+  // stage 6
+  step2[0] = WRAPLOW(step1[0] + step1[7]);
+  step2[1] = WRAPLOW(step1[1] + step1[6]);
+  step2[2] = WRAPLOW(step1[2] + step1[5]);
+  step2[3] = WRAPLOW(step1[3] + step1[4]);
+  step2[4] = WRAPLOW(step1[3] - step1[4]);
+  step2[5] = WRAPLOW(step1[2] - step1[5]);
+  step2[6] = WRAPLOW(step1[1] - step1[6]);
+  step2[7] = WRAPLOW(step1[0] - step1[7]);
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  temp1 = (-step1[10] + step1[13]) * cospi_16_64;
+  temp2 = (step1[10] + step1[13]) * cospi_16_64;
+  step2[10] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[13] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = (-step1[11] + step1[12]) * cospi_16_64;
+  temp2 = (step1[11] + step1[12]) * cospi_16_64;
+  step2[11] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[12] = WRAPLOW(dct_const_round_shift(temp2));
+  step2[14] = step1[14];
+  step2[15] = step1[15];
+
+  // stage 7
+  output[0] = WRAPLOW(step2[0] + step2[15]);
+  output[1] = WRAPLOW(step2[1] + step2[14]);
+  output[2] = WRAPLOW(step2[2] + step2[13]);
+  output[3] = WRAPLOW(step2[3] + step2[12]);
+  output[4] = WRAPLOW(step2[4] + step2[11]);
+  output[5] = WRAPLOW(step2[5] + step2[10]);
+  output[6] = WRAPLOW(step2[6] + step2[9]);
+  output[7] = WRAPLOW(step2[7] + step2[8]);
+  output[8] = WRAPLOW(step2[7] - step2[8]);
+  output[9] = WRAPLOW(step2[6] - step2[9]);
+  output[10] = WRAPLOW(step2[5] - step2[10]);
+  output[11] = WRAPLOW(step2[4] - step2[11]);
+  output[12] = WRAPLOW(step2[3] - step2[12]);
+  output[13] = WRAPLOW(step2[2] - step2[13]);
+  output[14] = WRAPLOW(step2[1] - step2[14]);
+  output[15] = WRAPLOW(step2[0] - step2[15]);
+}
+
+void vp9_high_idct16x16_256_add_c(const tran_low_t *input, uint8_t *dest8,
+                                  int stride, int bd) {
+  tran_low_t out[16 * 16];
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[16], temp_out[16];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // First transform rows.
+  for (i = 0; i < 16; ++i) {
+    high_idct16(input, outptr, bd);
+    input += 16;
+    outptr += 16;
+  }
+
+  // Then transform columns.
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j * 16 + i];
+    high_idct16(temp_in, temp_out, bd);
+    for (j = 0; j < 16; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
+  }
+}
+
+static void high_iadst16(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8;
+  tran_high_t s9, s10, s11, s12, s13, s14, s15;
+
+  tran_high_t x0 = input[15];
+  tran_high_t x1 = input[0];
+  tran_high_t x2 = input[13];
+  tran_high_t x3 = input[2];
+  tran_high_t x4 = input[11];
+  tran_high_t x5 = input[4];
+  tran_high_t x6 = input[9];
+  tran_high_t x7 = input[6];
+  tran_high_t x8 = input[7];
+  tran_high_t x9 = input[8];
+  tran_high_t x10 = input[5];
+  tran_high_t x11 = input[10];
+  tran_high_t x12 = input[3];
+  tran_high_t x13 = input[12];
+  tran_high_t x14 = input[1];
+  tran_high_t x15 = input[14];
+  (void) bd;
+
+  if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8
+           | x9 | x10 | x11 | x12 | x13 | x14 | x15)) {
+    vpx_memset(output, 0, 16 * sizeof(*output));
+    return;
+  }
+
+  // stage 1
+  s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
+  s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
+  s2 = x2 * cospi_5_64  + x3 * cospi_27_64;
+  s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
+  s4 = x4 * cospi_9_64  + x5 * cospi_23_64;
+  s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
+  s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
+  s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
+  s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
+  s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
+  s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
+  s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
+  s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
+  s13 = x12 * cospi_7_64  - x13 * cospi_25_64;
+  s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
+  s15 = x14 * cospi_3_64  - x15 * cospi_29_64;
+
+  x0 = WRAPLOW(dct_const_round_shift(s0 + s8));
+  x1 = WRAPLOW(dct_const_round_shift(s1 + s9));
+  x2 = WRAPLOW(dct_const_round_shift(s2 + s10));
+  x3 = WRAPLOW(dct_const_round_shift(s3 + s11));
+  x4 = WRAPLOW(dct_const_round_shift(s4 + s12));
+  x5 = WRAPLOW(dct_const_round_shift(s5 + s13));
+  x6 = WRAPLOW(dct_const_round_shift(s6 + s14));
+  x7 = WRAPLOW(dct_const_round_shift(s7 + s15));
+  x8  = WRAPLOW(dct_const_round_shift(s0 - s8));
+  x9  = WRAPLOW(dct_const_round_shift(s1 - s9));
+  x10 = WRAPLOW(dct_const_round_shift(s2 - s10));
+  x11 = WRAPLOW(dct_const_round_shift(s3 - s11));
+  x12 = WRAPLOW(dct_const_round_shift(s4 - s12));
+  x13 = WRAPLOW(dct_const_round_shift(s5 - s13));
+  x14 = WRAPLOW(dct_const_round_shift(s6 - s14));
+  x15 = WRAPLOW(dct_const_round_shift(s7 - s15));
+
+  // stage 2
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = x4;
+  s5 = x5;
+  s6 = x6;
+  s7 = x7;
+  s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
+  s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
+  s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
+  s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
+  s12 = -x12 * cospi_28_64 + x13 * cospi_4_64;
+  s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
+  s14 = -x14 * cospi_12_64 + x15 * cospi_20_64;
+  s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
+
+  x0 = WRAPLOW(s0 + s4);
+  x1 = WRAPLOW(s1 + s5);
+  x2 = WRAPLOW(s2 + s6);
+  x3 = WRAPLOW(s3 + s7);
+  x4 = WRAPLOW(s0 - s4);
+  x5 = WRAPLOW(s1 - s5);
+  x6 = WRAPLOW(s2 - s6);
+  x7 = WRAPLOW(s3 - s7);
+  x8 = WRAPLOW(dct_const_round_shift(s8 + s12));
+  x9 = WRAPLOW(dct_const_round_shift(s9 + s13));
+  x10 = WRAPLOW(dct_const_round_shift(s10 + s14));
+  x11 = WRAPLOW(dct_const_round_shift(s11 + s15));
+  x12 = WRAPLOW(dct_const_round_shift(s8 - s12));
+  x13 = WRAPLOW(dct_const_round_shift(s9 - s13));
+  x14 = WRAPLOW(dct_const_round_shift(s10 - s14));
+  x15 = WRAPLOW(dct_const_round_shift(s11 - s15));
+
+  // stage 3
+  s0 = x0;
+  s1 = x1;
+  s2 = x2;
+  s3 = x3;
+  s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
+  s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
+  s6 = -x6 * cospi_24_64 + x7 * cospi_8_64;
+  s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
+  s8 = x8;
+  s9 = x9;
+  s10 = x10;
+  s11 = x11;
+  s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
+  s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
+  s14 = -x14 * cospi_24_64 + x15 * cospi_8_64;
+  s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
+
+  x0 = WRAPLOW(s0 + s2);
+  x1 = WRAPLOW(s1 + s3);
+  x2 = WRAPLOW(s0 - s2);
+  x3 = WRAPLOW(s1 - s3);
+  x4 = WRAPLOW(dct_const_round_shift(s4 + s6));
+  x5 = WRAPLOW(dct_const_round_shift(s5 + s7));
+  x6 = WRAPLOW(dct_const_round_shift(s4 - s6));
+  x7 = WRAPLOW(dct_const_round_shift(s5 - s7));
+  x8 = WRAPLOW(s8 + s10);
+  x9 = WRAPLOW(s9 + s11);
+  x10 = WRAPLOW(s8 - s10);
+  x11 = WRAPLOW(s9 - s11);
+  x12 = WRAPLOW(dct_const_round_shift(s12 + s14));
+  x13 = WRAPLOW(dct_const_round_shift(s13 + s15));
+  x14 = WRAPLOW(dct_const_round_shift(s12 - s14));
+  x15 = WRAPLOW(dct_const_round_shift(s13 - s15));
+
+  // stage 4
+  s2 = (- cospi_16_64) * (x2 + x3);
+  s3 = cospi_16_64 * (x2 - x3);
+  s6 = cospi_16_64 * (x6 + x7);
+  s7 = cospi_16_64 * (-x6 + x7);
+  s10 = cospi_16_64 * (x10 + x11);
+  s11 = cospi_16_64 * (-x10 + x11);
+  s14 = (- cospi_16_64) * (x14 + x15);
+  s15 = cospi_16_64 * (x14 - x15);
+
+  x2 = WRAPLOW(dct_const_round_shift(s2));
+  x3 = WRAPLOW(dct_const_round_shift(s3));
+  x6 = WRAPLOW(dct_const_round_shift(s6));
+  x7 = WRAPLOW(dct_const_round_shift(s7));
+  x10 = WRAPLOW(dct_const_round_shift(s10));
+  x11 = WRAPLOW(dct_const_round_shift(s11));
+  x14 = WRAPLOW(dct_const_round_shift(s14));
+  x15 = WRAPLOW(dct_const_round_shift(s15));
+
+  output[0] = WRAPLOW(x0);
+  output[1] = WRAPLOW(-x8);
+  output[2] = WRAPLOW(x12);
+  output[3] = WRAPLOW(-x4);
+  output[4] = WRAPLOW(x6);
+  output[5] = WRAPLOW(x14);
+  output[6] = WRAPLOW(x10);
+  output[7] = WRAPLOW(x2);
+  output[8] = WRAPLOW(x3);
+  output[9] = WRAPLOW(x11);
+  output[10] = WRAPLOW(x15);
+  output[11] = WRAPLOW(x7);
+  output[12] = WRAPLOW(x5);
+  output[13] = WRAPLOW(-x13);
+  output[14] = WRAPLOW(x9);
+  output[15] = WRAPLOW(-x1);
+}
+
+static const high_transform_2d HIGH_IHT_16[] = {
+  { high_idct16,  high_idct16  },  // DCT_DCT  = 0
+  { high_iadst16, high_idct16  },  // ADST_DCT = 1
+  { high_idct16,  high_iadst16 },  // DCT_ADST = 2
+  { high_iadst16, high_iadst16 }   // ADST_ADST = 3
+};
+
+void vp9_high_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest8,
+                                 int stride, int tx_type, int bd) {
+  int i, j;
+  tran_low_t out[16 * 16];
+  tran_low_t *outptr = out;
+  tran_low_t temp_in[16], temp_out[16];
+  const high_transform_2d ht = HIGH_IHT_16[tx_type];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // Rows
+  for (i = 0; i < 16; ++i) {
+    ht.rows(input, outptr, bd);
+    input += 16;
+    outptr += 16;
+  }
+
+  // Columns
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j * 16 + i];
+    ht.cols(temp_in, temp_out, bd);
+    for (j = 0; j < 16; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
+  }
+}
+
+void vp9_high_idct16x16_10_add_c(const tran_low_t *input, uint8_t *dest8,
+                                 int stride, int bd) {
+  tran_low_t out[16 * 16] = { 0 };
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[16], temp_out[16];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // First transform rows. Since all non-zero dct coefficients are in
+  // upper-left 4x4 area, we only need to calculate first 4 rows here.
+  for (i = 0; i < 4; ++i) {
+    high_idct16(input, outptr, bd);
+    input += 16;
+    outptr += 16;
+  }
+
+  // Then transform columns.
+  for (i = 0; i < 16; ++i) {
+    for (j = 0; j < 16; ++j)
+      temp_in[j] = out[j*16 + i];
+    high_idct16(temp_in, temp_out, bd);
+    for (j = 0; j < 16; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
+  }
+}
+
+void vp9_high_idct16x16_1_add_c(const tran_low_t *input, uint8_t *dest8,
+                                int stride, int bd) {
+  int i, j;
+  tran_high_t a1;
+  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
+  a1 = ROUND_POWER_OF_TWO(out, 6);
+  for (j = 0; j < 16; ++j) {
+    for (i = 0; i < 16; ++i)
+      dest[i] = clip_pixel_bd_high(dest[i], a1, bd);
+    dest += stride;
+  }
+}
+
+static void high_idct32(const tran_low_t *input, tran_low_t *output, int bd) {
+  tran_low_t step1[32], step2[32];
+  tran_high_t temp1, temp2;
+  (void) bd;
+
+  // stage 1
+  step1[0] = input[0];
+  step1[1] = input[16];
+  step1[2] = input[8];
+  step1[3] = input[24];
+  step1[4] = input[4];
+  step1[5] = input[20];
+  step1[6] = input[12];
+  step1[7] = input[28];
+  step1[8] = input[2];
+  step1[9] = input[18];
+  step1[10] = input[10];
+  step1[11] = input[26];
+  step1[12] = input[6];
+  step1[13] = input[22];
+  step1[14] = input[14];
+  step1[15] = input[30];
+
+  temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64;
+  temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64;
+  step1[16] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[31] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64;
+  temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64;
+  step1[17] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[30] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64;
+  temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64;
+  step1[18] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[29] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64;
+  temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64;
+  step1[19] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[28] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64;
+  temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64;
+  step1[20] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[27] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64;
+  temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64;
+  step1[21] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[26] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64;
+  temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64;
+  step1[22] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[25] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64;
+  temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64;
+  step1[23] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[24] = WRAPLOW(dct_const_round_shift(temp2));
+
+  // stage 2
+  step2[0] = step1[0];
+  step2[1] = step1[1];
+  step2[2] = step1[2];
+  step2[3] = step1[3];
+  step2[4] = step1[4];
+  step2[5] = step1[5];
+  step2[6] = step1[6];
+  step2[7] = step1[7];
+
+  temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
+  temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
+  step2[8] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[15] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
+  temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
+  step2[9] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[14] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
+  temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
+  step2[10] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[13] = WRAPLOW(dct_const_round_shift(temp2));
+
+  temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
+  temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
+  step2[11] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[12] = WRAPLOW(dct_const_round_shift(temp2));
+
+  step2[16] = WRAPLOW(step1[16] + step1[17]);
+  step2[17] = WRAPLOW(step1[16] - step1[17]);
+  step2[18] = WRAPLOW(-step1[18] + step1[19]);
+  step2[19] = WRAPLOW(step1[18] + step1[19]);
+  step2[20] = WRAPLOW(step1[20] + step1[21]);
+  step2[21] = WRAPLOW(step1[20] - step1[21]);
+  step2[22] = WRAPLOW(-step1[22] + step1[23]);
+  step2[23] = WRAPLOW(step1[22] + step1[23]);
+  step2[24] = WRAPLOW(step1[24] + step1[25]);
+  step2[25] = WRAPLOW(step1[24] - step1[25]);
+  step2[26] = WRAPLOW(-step1[26] + step1[27]);
+  step2[27] = WRAPLOW(step1[26] + step1[27]);
+  step2[28] = WRAPLOW(step1[28] + step1[29]);
+  step2[29] = WRAPLOW(step1[28] - step1[29]);
+  step2[30] = WRAPLOW(-step1[30] + step1[31]);
+  step2[31] = WRAPLOW(step1[30] + step1[31]);
+
+  // stage 3
+  step1[0] = step2[0];
+  step1[1] = step2[1];
+  step1[2] = step2[2];
+  step1[3] = step2[3];
+
+  temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
+  temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
+  step1[4] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[7] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
+  temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
+  step1[5] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[6] = WRAPLOW(dct_const_round_shift(temp2));
+
+  step1[8] = WRAPLOW(step2[8] + step2[9]);
+  step1[9] = WRAPLOW(step2[8] - step2[9]);
+  step1[10] = WRAPLOW(-step2[10] + step2[11]);
+  step1[11] = WRAPLOW(step2[10] + step2[11]);
+  step1[12] = WRAPLOW(step2[12] + step2[13]);
+  step1[13] = WRAPLOW(step2[12] - step2[13]);
+  step1[14] = WRAPLOW(-step2[14] + step2[15]);
+  step1[15] = WRAPLOW(step2[14] + step2[15]);
+
+  step1[16] = step2[16];
+  step1[31] = step2[31];
+  temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64;
+  temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64;
+  step1[17] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[30] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64;
+  temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64;
+  step1[18] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[29] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[19] = step2[19];
+  step1[20] = step2[20];
+  temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64;
+  temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64;
+  step1[21] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[26] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64;
+  temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64;
+  step1[22] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[25] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[27] = step2[27];
+  step1[28] = step2[28];
+
+  // stage 4
+  temp1 = (step1[0] + step1[1]) * cospi_16_64;
+  temp2 = (step1[0] - step1[1]) * cospi_16_64;
+  step2[0] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[1] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
+  temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
+  step2[2] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[3] = WRAPLOW(dct_const_round_shift(temp2));
+  step2[4] = WRAPLOW(step1[4] + step1[5]);
+  step2[5] = WRAPLOW(step1[4] - step1[5]);
+  step2[6] = WRAPLOW(-step1[6] + step1[7]);
+  step2[7] = WRAPLOW(step1[6] + step1[7]);
+
+  step2[8] = step1[8];
+  step2[15] = step1[15];
+  temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
+  temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
+  step2[9] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[14] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
+  temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
+  step2[10] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[13] = WRAPLOW(dct_const_round_shift(temp2));
+  step2[11] = step1[11];
+  step2[12] = step1[12];
+
+  step2[16] = WRAPLOW(step1[16] + step1[19]);
+  step2[17] = WRAPLOW(step1[17] + step1[18]);
+  step2[18] = WRAPLOW(step1[17] - step1[18]);
+  step2[19] = WRAPLOW(step1[16] - step1[19]);
+  step2[20] = WRAPLOW(-step1[20] + step1[23]);
+  step2[21] = WRAPLOW(-step1[21] + step1[22]);
+  step2[22] = WRAPLOW(step1[21] + step1[22]);
+  step2[23] = WRAPLOW(step1[20] + step1[23]);
+
+  step2[24] = WRAPLOW(step1[24] + step1[27]);
+  step2[25] = WRAPLOW(step1[25] + step1[26]);
+  step2[26] = WRAPLOW(step1[25] - step1[26]);
+  step2[27] = WRAPLOW(step1[24] - step1[27]);
+  step2[28] = WRAPLOW(-step1[28] + step1[31]);
+  step2[29] = WRAPLOW(-step1[29] + step1[30]);
+  step2[30] = WRAPLOW(step1[29] + step1[30]);
+  step2[31] = WRAPLOW(step1[28] + step1[31]);
+
+  // stage 5
+  step1[0] = WRAPLOW(step2[0] + step2[3]);
+  step1[1] = WRAPLOW(step2[1] + step2[2]);
+  step1[2] = WRAPLOW(step2[1] - step2[2]);
+  step1[3] = WRAPLOW(step2[0] - step2[3]);
+  step1[4] = step2[4];
+  temp1 = (step2[6] - step2[5]) * cospi_16_64;
+  temp2 = (step2[5] + step2[6]) * cospi_16_64;
+  step1[5] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[6] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[7] = step2[7];
+
+  step1[8] = WRAPLOW(step2[8] + step2[11]);
+  step1[9] = WRAPLOW(step2[9] + step2[10]);
+  step1[10] = WRAPLOW(step2[9] - step2[10]);
+  step1[11] = WRAPLOW(step2[8] - step2[11]);
+  step1[12] = WRAPLOW(-step2[12] + step2[15]);
+  step1[13] = WRAPLOW(-step2[13] + step2[14]);
+  step1[14] = WRAPLOW(step2[13] + step2[14]);
+  step1[15] = WRAPLOW(step2[12] + step2[15]);
+
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64;
+  temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64;
+  step1[18] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[29] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64;
+  temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64;
+  step1[19] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[28] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64;
+  temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64;
+  step1[20] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[27] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64;
+  temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64;
+  step1[21] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[26] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[22] = step2[22];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[25] = step2[25];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+
+  // stage 6
+  step2[0] = WRAPLOW(step1[0] + step1[7]);
+  step2[1] = WRAPLOW(step1[1] + step1[6]);
+  step2[2] = WRAPLOW(step1[2] + step1[5]);
+  step2[3] = WRAPLOW(step1[3] + step1[4]);
+  step2[4] = WRAPLOW(step1[3] - step1[4]);
+  step2[5] = WRAPLOW(step1[2] - step1[5]);
+  step2[6] = WRAPLOW(step1[1] - step1[6]);
+  step2[7] = WRAPLOW(step1[0] - step1[7]);
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  temp1 = (-step1[10] + step1[13]) * cospi_16_64;
+  temp2 = (step1[10] + step1[13]) * cospi_16_64;
+  step2[10] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[13] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = (-step1[11] + step1[12]) * cospi_16_64;
+  temp2 = (step1[11] + step1[12]) * cospi_16_64;
+  step2[11] = WRAPLOW(dct_const_round_shift(temp1));
+  step2[12] = WRAPLOW(dct_const_round_shift(temp2));
+  step2[14] = WRAPLOW(step1[14]);
+  step2[15] = WRAPLOW(step1[15]);
+
+  step2[16] = WRAPLOW(step1[16] + step1[23]);
+  step2[17] = WRAPLOW(step1[17] + step1[22]);
+  step2[18] = WRAPLOW(step1[18] + step1[21]);
+  step2[19] = WRAPLOW(step1[19] + step1[20]);
+  step2[20] = WRAPLOW(step1[19] - step1[20]);
+  step2[21] = WRAPLOW(step1[18] - step1[21]);
+  step2[22] = WRAPLOW(step1[17] - step1[22]);
+  step2[23] = WRAPLOW(step1[16] - step1[23]);
+
+  step2[24] = WRAPLOW(-step1[24] + step1[31]);
+  step2[25] = WRAPLOW(-step1[25] + step1[30]);
+  step2[26] = WRAPLOW(-step1[26] + step1[29]);
+  step2[27] = WRAPLOW(-step1[27] + step1[28]);
+  step2[28] = WRAPLOW(step1[27] + step1[28]);
+  step2[29] = WRAPLOW(step1[26] + step1[29]);
+  step2[30] = WRAPLOW(step1[25] + step1[30]);
+  step2[31] = WRAPLOW(step1[24] + step1[31]);
+
+  // stage 7
+  step1[0] = WRAPLOW(step2[0] + step2[15]);
+  step1[1] = WRAPLOW(step2[1] + step2[14]);
+  step1[2] = WRAPLOW(step2[2] + step2[13]);
+  step1[3] = WRAPLOW(step2[3] + step2[12]);
+  step1[4] = WRAPLOW(step2[4] + step2[11]);
+  step1[5] = WRAPLOW(step2[5] + step2[10]);
+  step1[6] = WRAPLOW(step2[6] + step2[9]);
+  step1[7] = WRAPLOW(step2[7] + step2[8]);
+  step1[8] = WRAPLOW(step2[7] - step2[8]);
+  step1[9] = WRAPLOW(step2[6] - step2[9]);
+  step1[10] = WRAPLOW(step2[5] - step2[10]);
+  step1[11] = WRAPLOW(step2[4] - step2[11]);
+  step1[12] = WRAPLOW(step2[3] - step2[12]);
+  step1[13] = WRAPLOW(step2[2] - step2[13]);
+  step1[14] = WRAPLOW(step2[1] - step2[14]);
+  step1[15] = WRAPLOW(step2[0] - step2[15]);
+
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  step1[18] = step2[18];
+  step1[19] = step2[19];
+  temp1 = (-step2[20] + step2[27]) * cospi_16_64;
+  temp2 = (step2[20] + step2[27]) * cospi_16_64;
+  step1[20] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[27] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = (-step2[21] + step2[26]) * cospi_16_64;
+  temp2 = (step2[21] + step2[26]) * cospi_16_64;
+  step1[21] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[26] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = (-step2[22] + step2[25]) * cospi_16_64;
+  temp2 = (step2[22] + step2[25]) * cospi_16_64;
+  step1[22] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[25] = WRAPLOW(dct_const_round_shift(temp2));
+  temp1 = (-step2[23] + step2[24]) * cospi_16_64;
+  temp2 = (step2[23] + step2[24]) * cospi_16_64;
+  step1[23] = WRAPLOW(dct_const_round_shift(temp1));
+  step1[24] = WRAPLOW(dct_const_round_shift(temp2));
+  step1[28] = step2[28];
+  step1[29] = step2[29];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+
+  // final stage
+  output[0] = WRAPLOW(step1[0] + step1[31]);
+  output[1] = WRAPLOW(step1[1] + step1[30]);
+  output[2] = WRAPLOW(step1[2] + step1[29]);
+  output[3] = WRAPLOW(step1[3] + step1[28]);
+  output[4] = WRAPLOW(step1[4] + step1[27]);
+  output[5] = WRAPLOW(step1[5] + step1[26]);
+  output[6] = WRAPLOW(step1[6] + step1[25]);
+  output[7] = WRAPLOW(step1[7] + step1[24]);
+  output[8] = WRAPLOW(step1[8] + step1[23]);
+  output[9] = WRAPLOW(step1[9] + step1[22]);
+  output[10] = WRAPLOW(step1[10] + step1[21]);
+  output[11] = WRAPLOW(step1[11] + step1[20]);
+  output[12] = WRAPLOW(step1[12] + step1[19]);
+  output[13] = WRAPLOW(step1[13] + step1[18]);
+  output[14] = WRAPLOW(step1[14] + step1[17]);
+  output[15] = WRAPLOW(step1[15] + step1[16]);
+  output[16] = WRAPLOW(step1[15] - step1[16]);
+  output[17] = WRAPLOW(step1[14] - step1[17]);
+  output[18] = WRAPLOW(step1[13] - step1[18]);
+  output[19] = WRAPLOW(step1[12] - step1[19]);
+  output[20] = WRAPLOW(step1[11] - step1[20]);
+  output[21] = WRAPLOW(step1[10] - step1[21]);
+  output[22] = WRAPLOW(step1[9] - step1[22]);
+  output[23] = WRAPLOW(step1[8] - step1[23]);
+  output[24] = WRAPLOW(step1[7] - step1[24]);
+  output[25] = WRAPLOW(step1[6] - step1[25]);
+  output[26] = WRAPLOW(step1[5] - step1[26]);
+  output[27] = WRAPLOW(step1[4] - step1[27]);
+  output[28] = WRAPLOW(step1[3] - step1[28]);
+  output[29] = WRAPLOW(step1[2] - step1[29]);
+  output[30] = WRAPLOW(step1[1] - step1[30]);
+  output[31] = WRAPLOW(step1[0] - step1[31]);
+}
+
+void vp9_high_idct32x32_1024_add_c(const tran_low_t *input, uint8_t *dest8,
+                                   int stride, int bd) {
+  tran_low_t out[32 * 32];
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[32], temp_out[32];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // Rows
+  for (i = 0; i < 32; ++i) {
+    tran_low_t zero_coeff[16];
+    for (j = 0; j < 16; ++j)
+      zero_coeff[j] = input[2 * j] | input[2 * j + 1];
+    for (j = 0; j < 8; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+    for (j = 0; j < 4; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+    for (j = 0; j < 2; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+
+    if (zero_coeff[0] | zero_coeff[1])
+      high_idct32(input, outptr, bd);
+    else
+      vpx_memset(outptr, 0, sizeof(tran_low_t) * 32);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = out[j * 32 + i];
+    high_idct32(temp_in, temp_out, bd);
+    for (j = 0; j < 32; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
+  }
+}
+
+void vp9_high_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest8,
+                                 int stride, int bd) {
+  tran_low_t out[32 * 32] = {0};
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[32], temp_out[32];
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  // Rows
+  // Only upper-left 8x8 has non-zero coeff.
+  for (i = 0; i < 8; ++i) {
+    high_idct32(input, outptr, bd);
+    input += 32;
+    outptr += 32;
+  }
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = out[j * 32 + i];
+    high_idct32(temp_in, temp_out, bd);
+    for (j = 0; j < 32; ++j)
+      dest[j * stride + i] = clip_pixel_bd_high(
+          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
+  }
+}
+
+void vp9_high_idct32x32_1_add_c(const tran_low_t *input, uint8_t *dest8,
+                                int stride, int bd) {
+  int i, j;
+  int a1;
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
+  a1 = ROUND_POWER_OF_TWO(out, 6);
+
+  for (j = 0; j < 32; ++j) {
+    for (i = 0; i < 32; ++i)
+      dest[i] = clip_pixel_bd_high(dest[i], a1, bd);
+    dest += stride;
+  }
+}
+
+// idct
+void vp9_high_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                          int eob, int bd) {
+  if (eob > 1)
+    vp9_high_idct4x4_16_add(input, dest, stride, bd);
+  else
+    vp9_high_idct4x4_1_add(input, dest, stride, bd);
+}
+
+
+void vp9_high_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                          int eob, int bd) {
+  if (eob > 1)
+    vp9_high_iwht4x4_16_add(input, dest, stride, bd);
+  else
+    vp9_high_iwht4x4_1_add(input, dest, stride, bd);
+}
+
+void vp9_high_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
+                          int eob, int bd) {
+  // If dc is 1, then input[0] is the reconstructed value, do not need
+  // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
+
+  // The calculation can be simplified if there are not many non-zero dct
+  // coefficients. Use eobs to decide what to do.
+  // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
+  // Combine that with code here.
+  // DC only DCT coefficient
+  if (eob == 1) {
+    vp9_high_idct8x8_1_add(input, dest, stride, bd);
+  } else if (eob <= 10) {
+    vp9_high_idct8x8_10_add(input, dest, stride, bd);
+  } else {
+    vp9_high_idct8x8_64_add(input, dest, stride, bd);
+  }
+}
+
+void vp9_high_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride,
+                       int eob, int bd) {
+  // The calculation can be simplified if there are not many non-zero dct
+  // coefficients. Use eobs to separate different cases.
+  // DC only DCT coefficient.
+  if (eob == 1) {
+    vp9_high_idct16x16_1_add(input, dest, stride, bd);
+  } else if (eob <= 10) {
+    vp9_high_idct16x16_10_add(input, dest, stride, bd);
+  } else {
+    vp9_high_idct16x16_256_add(input, dest, stride, bd);
+  }
+}
+
+void vp9_high_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
+                       int eob, int bd) {
+  // Non-zero coeff only in upper-left 8x8
+  if (eob == 1) {
+    vp9_high_idct32x32_1_add(input, dest, stride, bd);
+  } else if (eob <= 34) {
+    vp9_high_idct32x32_34_add(input, dest, stride, bd);
+  } else {
+    vp9_high_idct32x32_1024_add(input, dest, stride, bd);
+  }
+}
+
+// iht
+void vp9_high_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
+                         uint8_t *dest, int stride, int eob, int bd) {
+  if (tx_type == DCT_DCT)
+    vp9_high_idct4x4_add(input, dest, stride, eob, bd);
+  else
+    vp9_high_iht4x4_16_add(input, dest, stride, tx_type, bd);
+}
+
+void vp9_high_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
+                         uint8_t *dest, int stride, int eob, int bd) {
+  if (tx_type == DCT_DCT) {
+    vp9_high_idct8x8_add(input, dest, stride, eob, bd);
+  } else {
+    vp9_high_iht8x8_64_add(input, dest, stride, tx_type, bd);
+  }
+}
+
+void vp9_high_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
+                           uint8_t *dest, int stride, int eob, int bd) {
+  if (tx_type == DCT_DCT) {
+    vp9_high_idct16x16_add(input, dest, stride, eob, bd);
+  } else {
+    vp9_high_iht16x16_256_add(input, dest, stride, tx_type, bd);
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/common/vp9_idct.h b/vp9/common/vp9_idct.h
index 7f595e1..694be3c 100644
--- a/vp9/common/vp9_idct.h
+++ b/vp9/common/vp9_idct.h
@@ -36,52 +36,69 @@
 #define dual_set_epi16(a, b) \
   _mm_set_epi16(b, b, b, b, a, a, a, a)
 
+// Note:
+// tran_low_t  is the datatype used for final transform coefficients.
+// tran_high_t is the datatype used for intermediate transform stages.
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef int64_t tran_high_t;
+typedef int32_t tran_low_t;
+#else
+typedef int32_t tran_high_t;
+typedef int16_t tran_low_t;
+#endif
+
 // Constants:
 //  for (int i = 1; i< 32; ++i)
 //    printf("static const int cospi_%d_64 = %.0f;\n", i,
 //           round(16384 * cos(i*M_PI/64)));
 // Note: sin(k*Pi/64) = cos((32-k)*Pi/64)
-static const int cospi_1_64  = 16364;
-static const int cospi_2_64  = 16305;
-static const int cospi_3_64  = 16207;
-static const int cospi_4_64  = 16069;
-static const int cospi_5_64  = 15893;
-static const int cospi_6_64  = 15679;
-static const int cospi_7_64  = 15426;
-static const int cospi_8_64  = 15137;
-static const int cospi_9_64  = 14811;
-static const int cospi_10_64 = 14449;
-static const int cospi_11_64 = 14053;
-static const int cospi_12_64 = 13623;
-static const int cospi_13_64 = 13160;
-static const int cospi_14_64 = 12665;
-static const int cospi_15_64 = 12140;
-static const int cospi_16_64 = 11585;
-static const int cospi_17_64 = 11003;
-static const int cospi_18_64 = 10394;
-static const int cospi_19_64 = 9760;
-static const int cospi_20_64 = 9102;
-static const int cospi_21_64 = 8423;
-static const int cospi_22_64 = 7723;
-static const int cospi_23_64 = 7005;
-static const int cospi_24_64 = 6270;
-static const int cospi_25_64 = 5520;
-static const int cospi_26_64 = 4756;
-static const int cospi_27_64 = 3981;
-static const int cospi_28_64 = 3196;
-static const int cospi_29_64 = 2404;
-static const int cospi_30_64 = 1606;
-static const int cospi_31_64 = 804;
+static const tran_high_t cospi_1_64  = 16364;
+static const tran_high_t cospi_2_64  = 16305;
+static const tran_high_t cospi_3_64  = 16207;
+static const tran_high_t cospi_4_64  = 16069;
+static const tran_high_t cospi_5_64  = 15893;
+static const tran_high_t cospi_6_64  = 15679;
+static const tran_high_t cospi_7_64  = 15426;
+static const tran_high_t cospi_8_64  = 15137;
+static const tran_high_t cospi_9_64  = 14811;
+static const tran_high_t cospi_10_64 = 14449;
+static const tran_high_t cospi_11_64 = 14053;
+static const tran_high_t cospi_12_64 = 13623;
+static const tran_high_t cospi_13_64 = 13160;
+static const tran_high_t cospi_14_64 = 12665;
+static const tran_high_t cospi_15_64 = 12140;
+static const tran_high_t cospi_16_64 = 11585;
+static const tran_high_t cospi_17_64 = 11003;
+static const tran_high_t cospi_18_64 = 10394;
+static const tran_high_t cospi_19_64 = 9760;
+static const tran_high_t cospi_20_64 = 9102;
+static const tran_high_t cospi_21_64 = 8423;
+static const tran_high_t cospi_22_64 = 7723;
+static const tran_high_t cospi_23_64 = 7005;
+static const tran_high_t cospi_24_64 = 6270;
+static const tran_high_t cospi_25_64 = 5520;
+static const tran_high_t cospi_26_64 = 4756;
+static const tran_high_t cospi_27_64 = 3981;
+static const tran_high_t cospi_28_64 = 3196;
+static const tran_high_t cospi_29_64 = 2404;
+static const tran_high_t cospi_30_64 = 1606;
+static const tran_high_t cospi_31_64 = 804;
 
 //  16384 * sqrt(2) * sin(kPi/9) * 2 / 3
-static const int sinpi_1_9 = 5283;
-static const int sinpi_2_9 = 9929;
-static const int sinpi_3_9 = 13377;
-static const int sinpi_4_9 = 15212;
+static const tran_high_t sinpi_1_9 = 5283;
+static const tran_high_t sinpi_2_9 = 9929;
+static const tran_high_t sinpi_3_9 = 13377;
+static const tran_high_t sinpi_4_9 = 15212;
 
-static INLINE int dct_const_round_shift(int input) {
-  int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-#if CONFIG_COEFFICIENT_RANGE_CHECKING
+static INLINE tran_low_t dct_const_round_shift(tran_high_t input) {
+  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+#if CONFIG_VP9_HIGHBITDEPTH
+  // For valid highbitdepth VP9 streams, intermediate stage coefficients will
+  // stay within the ranges:
+  // - 8 bit: signed 16 bit integer
+  // - 10 bit: signed 18 bit integer
+  // - 12 bit: signed 20 bit integer
+#elif CONFIG_COEFFICIENT_RANGE_CHECKING
   // For valid VP9 input streams, intermediate stage coefficients should always
   // stay within the range of a signed 16 bit integer. Coefficients can go out
   // of this range for invalid/corrupt VP9 streams. However, strictly checking
@@ -91,32 +108,59 @@
   assert(INT16_MIN <= rv);
   assert(rv <= INT16_MAX);
 #endif
-  return (int16_t)rv;
+  return (tran_low_t)rv;
 }
 
-typedef void (*transform_1d)(const int16_t*, int16_t*);
+typedef void (*transform_1d)(const tran_low_t*, tran_low_t*);
 
 typedef struct {
   transform_1d cols, rows;  // vertical and horizontal
 } transform_2d;
 
-void vp9_iwht4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob);
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*high_transform_1d)(const tran_low_t*, tran_low_t*, int bd);
 
-void vp9_idct4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob);
-void vp9_idct8x8_add(const int16_t *input, uint8_t *dest, int stride, int eob);
-void vp9_idct16x16_add(const int16_t *input, uint8_t *dest, int stride, int
+typedef struct {
+  high_transform_1d cols, rows;  // vertical and horizontal
+} high_transform_2d;
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                     int eob);
+void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                     int eob);
+void vp9_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
+                     int eob);
+void vp9_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride, int
                        eob);
-void vp9_idct32x32_add(const int16_t *input, uint8_t *dest, int stride,
+void vp9_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
                        int eob);
 
-void vp9_iht4x4_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                     int stride, int eob);
-void vp9_iht8x8_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                     int stride, int eob);
-void vp9_iht16x16_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                       int stride, int eob);
 
-
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                          int eob, int bd);
+void vp9_high_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
+                          int eob, int bd);
+void vp9_high_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
+                          int eob, int bd);
+void vp9_high_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride,
+                            int eob, int bd);
+void vp9_high_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
+                            int eob, int bd);
+void vp9_high_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
+                         uint8_t *dest, int stride, int eob, int bd);
+void vp9_high_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
+                         uint8_t *dest, int stride, int eob, int bd);
+void vp9_high_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
+                           uint8_t *dest, int stride, int eob, int bd);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vp9/common/vp9_loopfilter.c b/vp9/common/vp9_loopfilter.c
index 3b39d42..4d6d457 100644
--- a/vp9/common/vp9_loopfilter.c
+++ b/vp9/common/vp9_loopfilter.c
@@ -392,6 +392,107 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_filter_selectively_vert_row2(PLANE_TYPE plane_type,
+                                              uint16_t *s, int pitch,
+                                              unsigned int mask_16x16_l,
+                                              unsigned int mask_8x8_l,
+                                              unsigned int mask_4x4_l,
+                                              unsigned int mask_4x4_int_l,
+                                              const loop_filter_info_n *lfi_n,
+                                              const uint8_t *lfl, int bd) {
+  const int mask_shift = plane_type ? 4 : 8;
+  const int mask_cutoff = plane_type ? 0xf : 0xff;
+  const int lfl_forward = plane_type ? 4 : 8;
+
+  unsigned int mask_16x16_0 = mask_16x16_l & mask_cutoff;
+  unsigned int mask_8x8_0 = mask_8x8_l & mask_cutoff;
+  unsigned int mask_4x4_0 = mask_4x4_l & mask_cutoff;
+  unsigned int mask_4x4_int_0 = mask_4x4_int_l & mask_cutoff;
+  unsigned int mask_16x16_1 = (mask_16x16_l >> mask_shift) & mask_cutoff;
+  unsigned int mask_8x8_1 = (mask_8x8_l >> mask_shift) & mask_cutoff;
+  unsigned int mask_4x4_1 = (mask_4x4_l >> mask_shift) & mask_cutoff;
+  unsigned int mask_4x4_int_1 = (mask_4x4_int_l >> mask_shift) & mask_cutoff;
+  unsigned int mask;
+
+  for (mask = mask_16x16_0 | mask_8x8_0 | mask_4x4_0 | mask_4x4_int_0 |
+       mask_16x16_1 | mask_8x8_1 | mask_4x4_1 | mask_4x4_int_1;
+       mask; mask >>= 1) {
+    const loop_filter_thresh *lfi0 = lfi_n->lfthr + *lfl;
+    const loop_filter_thresh *lfi1 = lfi_n->lfthr + *(lfl + lfl_forward);
+
+    // TODO(yunqingwang): count in loopfilter functions should be removed.
+    if (mask & 1) {
+      if ((mask_16x16_0 | mask_16x16_1) & 1) {
+        if ((mask_16x16_0 & mask_16x16_1) & 1) {
+          vp9_highbd_lpf_vertical_16_dual(s, pitch, lfi0->mblim, lfi0->lim,
+                                          lfi0->hev_thr, bd);
+        } else if (mask_16x16_0 & 1) {
+          vp9_highbd_lpf_vertical_16(s, pitch, lfi0->mblim, lfi0->lim,
+                                     lfi0->hev_thr, bd);
+        } else {
+          vp9_highbd_lpf_vertical_16(s + 8 *pitch, pitch, lfi1->mblim,
+                                     lfi1->lim, lfi1->hev_thr, bd);
+        }
+      }
+
+      if ((mask_8x8_0 | mask_8x8_1) & 1) {
+        if ((mask_8x8_0 & mask_8x8_1) & 1) {
+          vp9_highbd_lpf_vertical_8_dual(s, pitch, lfi0->mblim, lfi0->lim,
+                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
+                                         lfi1->hev_thr, bd);
+        } else if (mask_8x8_0 & 1) {
+          vp9_highbd_lpf_vertical_8(s, pitch, lfi0->mblim, lfi0->lim,
+                                    lfi0->hev_thr, 1, bd);
+        } else {
+          vp9_highbd_lpf_vertical_8(s + 8 * pitch, pitch, lfi1->mblim,
+                                    lfi1->lim, lfi1->hev_thr, 1, bd);
+        }
+      }
+
+      if ((mask_4x4_0 | mask_4x4_1) & 1) {
+        if ((mask_4x4_0 & mask_4x4_1) & 1) {
+          vp9_highbd_lpf_vertical_4_dual(s, pitch, lfi0->mblim, lfi0->lim,
+                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
+                                         lfi1->hev_thr, bd);
+        } else if (mask_4x4_0 & 1) {
+          vp9_highbd_lpf_vertical_4(s, pitch, lfi0->mblim, lfi0->lim,
+                                    lfi0->hev_thr, 1, bd);
+        } else {
+          vp9_highbd_lpf_vertical_4(s + 8 * pitch, pitch, lfi1->mblim,
+                                    lfi1->lim, lfi1->hev_thr, 1, bd);
+        }
+      }
+
+      if ((mask_4x4_int_0 | mask_4x4_int_1) & 1) {
+        if ((mask_4x4_int_0 & mask_4x4_int_1) & 1) {
+          vp9_highbd_lpf_vertical_4_dual(s + 4, pitch, lfi0->mblim, lfi0->lim,
+                                         lfi0->hev_thr, lfi1->mblim, lfi1->lim,
+                                         lfi1->hev_thr, bd);
+        } else if (mask_4x4_int_0 & 1) {
+          vp9_highbd_lpf_vertical_4(s + 4, pitch, lfi0->mblim, lfi0->lim,
+                                    lfi0->hev_thr, 1, bd);
+        } else {
+          vp9_highbd_lpf_vertical_4(s + 8 * pitch + 4, pitch, lfi1->mblim,
+                                    lfi1->lim, lfi1->hev_thr, 1, bd);
+        }
+      }
+    }
+
+    s += 8;
+    lfl += 1;
+    mask_16x16_0 >>= 1;
+    mask_8x8_0 >>= 1;
+    mask_4x4_0 >>= 1;
+    mask_4x4_int_0 >>= 1;
+    mask_16x16_1 >>= 1;
+    mask_8x8_1 >>= 1;
+    mask_4x4_1 >>= 1;
+    mask_4x4_int_1 >>= 1;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void filter_selectively_horiz(uint8_t *s, int pitch,
                                      unsigned int mask_16x16,
                                      unsigned int mask_8x8,
@@ -419,7 +520,7 @@
         }
       } else if (mask_8x8 & 1) {
         if ((mask_8x8 & 3) == 3) {
-          // Next block's thresholds
+          // Next block's thresholds.
           const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
 
           vp9_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
@@ -448,7 +549,7 @@
         }
       } else if (mask_4x4 & 1) {
         if ((mask_4x4 & 3) == 3) {
-          // Next block's thresholds
+          // Next block's thresholds.
           const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
 
           vp9_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
@@ -488,6 +589,112 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_filter_selectively_horiz(uint16_t *s, int pitch,
+                                          unsigned int mask_16x16,
+                                          unsigned int mask_8x8,
+                                          unsigned int mask_4x4,
+                                          unsigned int mask_4x4_int,
+                                          const loop_filter_info_n *lfi_n,
+                                          const uint8_t *lfl, int bd) {
+  unsigned int mask;
+  int count;
+
+  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int;
+       mask; mask >>= count) {
+    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
+
+    count = 1;
+    if (mask & 1) {
+      if (mask_16x16 & 1) {
+        if ((mask_16x16 & 3) == 3) {
+          vp9_highbd_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim,
+                                       lfi->hev_thr, 2, bd);
+          count = 2;
+        } else {
+          vp9_highbd_lpf_horizontal_16(s, pitch, lfi->mblim, lfi->lim,
+                                       lfi->hev_thr, 1, bd);
+        }
+      } else if (mask_8x8 & 1) {
+        if ((mask_8x8 & 3) == 3) {
+          // Next block's thresholds.
+          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
+
+          vp9_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
+                                           lfi->hev_thr, lfin->mblim, lfin->lim,
+                                           lfin->hev_thr, bd);
+
+          if ((mask_4x4_int & 3) == 3) {
+            vp9_highbd_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
+                                             lfi->lim, lfi->hev_thr,
+                                             lfin->mblim, lfin->lim,
+                                             lfin->hev_thr, bd);
+          } else {
+            if (mask_4x4_int & 1) {
+              vp9_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
+                                          lfi->lim, lfi->hev_thr, 1, bd);
+            } else if (mask_4x4_int & 2) {
+              vp9_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
+                                          lfin->lim, lfin->hev_thr, 1, bd);
+            }
+          }
+          count = 2;
+        } else {
+          vp9_highbd_lpf_horizontal_8(s, pitch, lfi->mblim, lfi->lim,
+                                      lfi->hev_thr, 1, bd);
+
+          if (mask_4x4_int & 1) {
+            vp9_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
+                                        lfi->lim, lfi->hev_thr, 1, bd);
+          }
+        }
+      } else if (mask_4x4 & 1) {
+        if ((mask_4x4 & 3) == 3) {
+          // Next block's thresholds.
+          const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + 1);
+
+          vp9_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
+                                           lfi->hev_thr, lfin->mblim, lfin->lim,
+                                           lfin->hev_thr, bd);
+          if ((mask_4x4_int & 3) == 3) {
+            vp9_highbd_lpf_horizontal_4_dual(s + 4 * pitch, pitch, lfi->mblim,
+                                             lfi->lim, lfi->hev_thr,
+                                             lfin->mblim, lfin->lim,
+                                             lfin->hev_thr, bd);
+          } else {
+            if (mask_4x4_int & 1) {
+              vp9_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
+                                          lfi->lim, lfi->hev_thr, 1, bd);
+            } else if (mask_4x4_int & 2) {
+              vp9_highbd_lpf_horizontal_4(s + 8 + 4 * pitch, pitch, lfin->mblim,
+                                          lfin->lim, lfin->hev_thr, 1, bd);
+            }
+          }
+          count = 2;
+        } else {
+          vp9_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
+                                      lfi->hev_thr, 1, bd);
+
+          if (mask_4x4_int & 1) {
+            vp9_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim,
+                                        lfi->lim, lfi->hev_thr, 1, bd);
+          }
+        }
+      } else if (mask_4x4_int & 1) {
+        vp9_highbd_lpf_horizontal_4(s + 4 * pitch, pitch, lfi->mblim, lfi->lim,
+                                    lfi->hev_thr, 1, bd);
+      }
+    }
+    s += 8 * count;
+    lfl += count;
+    mask_16x16 >>= count;
+    mask_8x8 >>= count;
+    mask_4x4 >>= count;
+    mask_4x4_int >>= count;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 // This function ors into the current lfm structure, where to do loop
 // filters for the specific mi we are looking at. It uses information
 // including the block_size_type (32x16, 32x32, etc.), the transform size,
@@ -619,12 +826,12 @@
 // by mi_row, mi_col.
 // TODO(JBB): This function only works for yv12.
 void vp9_setup_mask(VP9_COMMON *const cm, const int mi_row, const int mi_col,
-                    MODE_INFO **mi, const int mode_info_stride,
+                    MODE_INFO *mi, const int mode_info_stride,
                     LOOP_FILTER_MASK *lfm) {
   int idx_32, idx_16, idx_8;
   const loop_filter_info_n *const lfi_n = &cm->lf_info;
-  MODE_INFO **mip = mi;
-  MODE_INFO **mip2 = mi;
+  MODE_INFO *mip = mi;
+  MODE_INFO *mip2 = mi;
 
   // These are offsets to the next mi in the 64x64 block. It is what gets
   // added to the mi ptr as we go through each loop. It helps us to avoid
@@ -652,28 +859,28 @@
                         cm->mi_cols - mi_col : MI_BLOCK_SIZE);
 
   vp9_zero(*lfm);
-  assert(mip[0] != NULL);
+  assert(mip != NULL);
 
   // TODO(jimbankoski): Try moving most of the following code into decode
   // loop and storing lfm in the mbmi structure so that we don't have to go
   // through the recursive loop structure multiple times.
-  switch (mip[0]->mbmi.sb_type) {
+  switch (mip->mbmi.sb_type) {
     case BLOCK_64X64:
-      build_masks(lfi_n, mip[0] , 0, 0, lfm);
+      build_masks(lfi_n, mip , 0, 0, lfm);
       break;
     case BLOCK_64X32:
-      build_masks(lfi_n, mip[0], 0, 0, lfm);
+      build_masks(lfi_n, mip, 0, 0, lfm);
       mip2 = mip + mode_info_stride * 4;
       if (4 >= max_rows)
         break;
-      build_masks(lfi_n, mip2[0], 32, 8, lfm);
+      build_masks(lfi_n, mip2, 32, 8, lfm);
       break;
     case BLOCK_32X64:
-      build_masks(lfi_n, mip[0], 0, 0, lfm);
+      build_masks(lfi_n, mip, 0, 0, lfm);
       mip2 = mip + 4;
       if (4 >= max_cols)
         break;
-      build_masks(lfi_n, mip2[0], 4, 2, lfm);
+      build_masks(lfi_n, mip2, 4, 2, lfm);
       break;
     default:
       for (idx_32 = 0; idx_32 < 4; mip += offset_32[idx_32], ++idx_32) {
@@ -683,23 +890,23 @@
         const int mi_32_row_offset = ((idx_32 >> 1) << 2);
         if (mi_32_col_offset >= max_cols || mi_32_row_offset >= max_rows)
           continue;
-        switch (mip[0]->mbmi.sb_type) {
+        switch (mip->mbmi.sb_type) {
           case BLOCK_32X32:
-            build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+            build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
             break;
           case BLOCK_32X16:
-            build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+            build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
             if (mi_32_row_offset + 2 >= max_rows)
               continue;
             mip2 = mip + mode_info_stride * 2;
-            build_masks(lfi_n, mip2[0], shift_y + 16, shift_uv + 4, lfm);
+            build_masks(lfi_n, mip2, shift_y + 16, shift_uv + 4, lfm);
             break;
           case BLOCK_16X32:
-            build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+            build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
             if (mi_32_col_offset + 2 >= max_cols)
               continue;
             mip2 = mip + 2;
-            build_masks(lfi_n, mip2[0], shift_y + 2, shift_uv + 1, lfm);
+            build_masks(lfi_n, mip2, shift_y + 2, shift_uv + 1, lfm);
             break;
           default:
             for (idx_16 = 0; idx_16 < 4; mip += offset_16[idx_16], ++idx_16) {
@@ -713,29 +920,29 @@
               if (mi_16_col_offset >= max_cols || mi_16_row_offset >= max_rows)
                 continue;
 
-              switch (mip[0]->mbmi.sb_type) {
+              switch (mip->mbmi.sb_type) {
                 case BLOCK_16X16:
-                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
                   break;
                 case BLOCK_16X8:
-                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
                   if (mi_16_row_offset + 1 >= max_rows)
                     continue;
                   mip2 = mip + mode_info_stride;
-                  build_y_mask(lfi_n, mip2[0], shift_y+8, lfm);
+                  build_y_mask(lfi_n, mip2, shift_y+8, lfm);
                   break;
                 case BLOCK_8X16:
-                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
                   if (mi_16_col_offset +1 >= max_cols)
                     continue;
                   mip2 = mip + 1;
-                  build_y_mask(lfi_n, mip2[0], shift_y+1, lfm);
+                  build_y_mask(lfi_n, mip2, shift_y+1, lfm);
                   break;
                 default: {
                   const int shift_y = shift_32_y[idx_32] +
                                       shift_16_y[idx_16] +
                                       shift_8_y[0];
-                  build_masks(lfi_n, mip[0], shift_y, shift_uv, lfm);
+                  build_masks(lfi_n, mip, shift_y, shift_uv, lfm);
                   mip += offset[0];
                   for (idx_8 = 1; idx_8 < 4; mip += offset[idx_8], ++idx_8) {
                     const int shift_y = shift_32_y[idx_32] +
@@ -749,7 +956,7 @@
                     if (mi_8_col_offset >= max_cols ||
                         mi_8_row_offset >= max_rows)
                       continue;
-                    build_y_mask(lfi_n, mip[0], shift_y, lfm);
+                    build_y_mask(lfi_n, mip, shift_y, lfm);
                   }
                   break;
                 }
@@ -903,14 +1110,53 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_filter_selectively_vert(uint16_t *s, int pitch,
+                                         unsigned int mask_16x16,
+                                         unsigned int mask_8x8,
+                                         unsigned int mask_4x4,
+                                         unsigned int mask_4x4_int,
+                                         const loop_filter_info_n *lfi_n,
+                                         const uint8_t *lfl, int bd) {
+  unsigned int mask;
+
+  for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int;
+       mask; mask >>= 1) {
+    const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
+
+    if (mask & 1) {
+      if (mask_16x16 & 1) {
+        vp9_highbd_lpf_vertical_16(s, pitch, lfi->mblim, lfi->lim,
+                                   lfi->hev_thr, bd);
+      } else if (mask_8x8 & 1) {
+        vp9_highbd_lpf_vertical_8(s, pitch, lfi->mblim, lfi->lim,
+                                  lfi->hev_thr, 1, bd);
+      } else if (mask_4x4 & 1) {
+        vp9_highbd_lpf_vertical_4(s, pitch, lfi->mblim, lfi->lim,
+                                lfi->hev_thr, 1, bd);
+      }
+    }
+    if (mask_4x4_int & 1)
+      vp9_highbd_lpf_vertical_4(s + 4, pitch, lfi->mblim, lfi->lim,
+                                lfi->hev_thr, 1, bd);
+    s += 8;
+    lfl += 1;
+    mask_16x16 >>= 1;
+    mask_8x8 >>= 1;
+    mask_4x4 >>= 1;
+    mask_4x4_int >>= 1;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void filter_block_plane_non420(VP9_COMMON *cm,
                                       struct macroblockd_plane *plane,
-                                      MODE_INFO **mi_8x8,
+                                      MODE_INFO *mi_8x8,
                                       int mi_row, int mi_col) {
   const int ss_x = plane->subsampling_x;
   const int ss_y = plane->subsampling_y;
-  const int row_step = 1 << ss_x;
-  const int col_step = 1 << ss_y;
+  const int row_step = 1 << ss_y;
+  const int col_step = 1 << ss_x;
   const int row_step_stride = cm->mi_stride * row_step;
   struct buf_2d *const dst = &plane->dst;
   uint8_t* const dst0 = dst->buf;
@@ -929,7 +1175,7 @@
 
     // Determine the vertical edges that need filtering
     for (c = 0; c < MI_BLOCK_SIZE && mi_col + c < cm->mi_cols; c += col_step) {
-      const MODE_INFO *mi = mi_8x8[c];
+      const MODE_INFO *mi = mi_8x8[c].src_mi;
       const BLOCK_SIZE sb_type = mi[0].mbmi.sb_type;
       const int skip_this = mi[0].mbmi.skip && is_inter_block(&mi[0].mbmi);
       // left edge of current unit is block/partition edge -> no skip
@@ -1001,12 +1247,32 @@
 
     // Disable filtering on the leftmost column
     border_mask = ~(mi_col == 0);
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (cm->use_highbitdepth) {
+      high_filter_selectively_vert(CONVERT_TO_SHORTPTR(dst->buf),
+                                   dst->stride,
+                                   mask_16x16_c & border_mask,
+                                   mask_8x8_c & border_mask,
+                                   mask_4x4_c & border_mask,
+                                   mask_4x4_int[r],
+                                   &cm->lf_info, &lfl[r << 3],
+                                   (int)cm->bit_depth);
+    } else {
+      filter_selectively_vert(dst->buf, dst->stride,
+                              mask_16x16_c & border_mask,
+                              mask_8x8_c & border_mask,
+                              mask_4x4_c & border_mask,
+                              mask_4x4_int[r],
+                              &cm->lf_info, &lfl[r << 3]);
+    }
+#else
     filter_selectively_vert(dst->buf, dst->stride,
                             mask_16x16_c & border_mask,
                             mask_8x8_c & border_mask,
                             mask_4x4_c & border_mask,
                             mask_4x4_int[r],
                             &cm->lf_info, &lfl[r << 3]);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
     dst->buf += 8 * dst->stride;
     mi_8x8 += row_step_stride;
   }
@@ -1030,13 +1296,32 @@
       mask_8x8_r = mask_8x8[r];
       mask_4x4_r = mask_4x4[r];
     }
-
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (cm->use_highbitdepth) {
+      high_filter_selectively_horiz(CONVERT_TO_SHORTPTR(dst->buf),
+                                    dst->stride,
+                                    mask_16x16_r,
+                                    mask_8x8_r,
+                                    mask_4x4_r,
+                                    mask_4x4_int_r,
+                                    &cm->lf_info, &lfl[r << 3],
+                                    (int)cm->bit_depth);
+    } else {
+      filter_selectively_horiz(dst->buf, dst->stride,
+                               mask_16x16_r,
+                               mask_8x8_r,
+                               mask_4x4_r,
+                               mask_4x4_int_r,
+                               &cm->lf_info, &lfl[r << 3]);
+    }
+#else
     filter_selectively_horiz(dst->buf, dst->stride,
                              mask_16x16_r,
                              mask_8x8_r,
                              mask_4x4_r,
                              mask_4x4_int_r,
                              &cm->lf_info, &lfl[r << 3]);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
     dst->buf += 8 * dst->stride;
   }
 }
@@ -1062,7 +1347,29 @@
       unsigned int mask_4x4_l = mask_4x4 & 0xffff;
       unsigned int mask_4x4_int_l = mask_4x4_int & 0xffff;
 
-      // Disable filtering on the leftmost column
+      // Disable filtering on the leftmost column.
+#if CONFIG_VP9_HIGHBITDEPTH
+      if (cm->use_highbitdepth) {
+        high_filter_selectively_vert_row2(plane->plane_type,
+                                          CONVERT_TO_SHORTPTR(dst->buf),
+                                          dst->stride,
+                                          mask_16x16_l,
+                                          mask_8x8_l,
+                                          mask_4x4_l,
+                                          mask_4x4_int_l,
+                                          &cm->lf_info, &lfm->lfl_y[r << 3],
+                                          (int)cm->bit_depth);
+      } else {
+        filter_selectively_vert_row2(plane->plane_type,
+                                     dst->buf, dst->stride,
+                                     mask_16x16_l,
+                                     mask_8x8_l,
+                                     mask_4x4_l,
+                                     mask_4x4_int_l,
+                                     &cm->lf_info,
+                                     &lfm->lfl_y[r << 3]);
+      }
+#else
       filter_selectively_vert_row2(plane->plane_type,
                                    dst->buf, dst->stride,
                                    mask_16x16_l,
@@ -1070,7 +1377,7 @@
                                    mask_4x4_l,
                                    mask_4x4_int_l,
                                    &cm->lf_info, &lfm->lfl_y[r << 3]);
-
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       dst->buf += 16 * dst->stride;
       mask_16x16 >>= 16;
       mask_8x8 >>= 16;
@@ -1100,12 +1407,35 @@
         mask_4x4_r = mask_4x4 & 0xff;
       }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+      if (cm->use_highbitdepth) {
+        high_filter_selectively_horiz(CONVERT_TO_SHORTPTR(dst->buf),
+                                      dst->stride,
+                                      mask_16x16_r,
+                                      mask_8x8_r,
+                                      mask_4x4_r,
+                                      mask_4x4_int & 0xff,
+                                      &cm->lf_info,
+                                      &lfm->lfl_y[r << 3],
+                                      (int)cm->bit_depth);
+      } else {
+        filter_selectively_horiz(dst->buf, dst->stride,
+                                 mask_16x16_r,
+                                 mask_8x8_r,
+                                 mask_4x4_r,
+                                 mask_4x4_int & 0xff,
+                                 &cm->lf_info,
+                                 &lfm->lfl_y[r << 3]);
+      }
+#else
       filter_selectively_horiz(dst->buf, dst->stride,
                                mask_16x16_r,
                                mask_8x8_r,
                                mask_4x4_r,
                                mask_4x4_int & 0xff,
-                               &cm->lf_info, &lfm->lfl_y[r << 3]);
+                               &cm->lf_info,
+                               &lfm->lfl_y[r << 3]);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
       dst->buf += 8 * dst->stride;
       mask_16x16 >>= 8;
@@ -1135,14 +1465,39 @@
         unsigned int mask_4x4_l = mask_4x4 & 0xff;
         unsigned int mask_4x4_int_l = mask_4x4_int & 0xff;
 
-        // Disable filtering on the leftmost column
+        // Disable filtering on the leftmost column.
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (cm->use_highbitdepth) {
+          high_filter_selectively_vert_row2(plane->plane_type,
+                                            CONVERT_TO_SHORTPTR(dst->buf),
+                                            dst->stride,
+                                            mask_16x16_l,
+                                            mask_8x8_l,
+                                            mask_4x4_l,
+                                            mask_4x4_int_l,
+                                            &cm->lf_info,
+                                            &lfm->lfl_uv[r << 1],
+                                            (int)cm->bit_depth);
+        } else {
+          filter_selectively_vert_row2(plane->plane_type,
+                                       dst->buf, dst->stride,
+                                       mask_16x16_l,
+                                       mask_8x8_l,
+                                       mask_4x4_l,
+                                       mask_4x4_int_l,
+                                       &cm->lf_info,
+                                       &lfm->lfl_uv[r << 1]);
+        }
+#else
         filter_selectively_vert_row2(plane->plane_type,
                                      dst->buf, dst->stride,
                                      mask_16x16_l,
                                      mask_8x8_l,
                                      mask_4x4_l,
                                      mask_4x4_int_l,
-                                     &cm->lf_info, &lfm->lfl_uv[r << 1]);
+                                     &cm->lf_info,
+                                     &lfm->lfl_uv[r << 1]);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
         dst->buf += 16 * dst->stride;
         mask_16x16 >>= 8;
@@ -1177,12 +1532,35 @@
         mask_4x4_r = mask_4x4 & 0xf;
       }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+      if (cm->use_highbitdepth) {
+        high_filter_selectively_horiz(CONVERT_TO_SHORTPTR(dst->buf),
+                                      dst->stride,
+                                      mask_16x16_r,
+                                      mask_8x8_r,
+                                      mask_4x4_r,
+                                      mask_4x4_int_r,
+                                      &cm->lf_info,
+                                      &lfm->lfl_uv[r << 1],
+                                      (int)cm->bit_depth);
+      } else {
+        filter_selectively_horiz(dst->buf, dst->stride,
+                                 mask_16x16_r,
+                                 mask_8x8_r,
+                                 mask_4x4_r,
+                                 mask_4x4_int_r,
+                                 &cm->lf_info,
+                                 &lfm->lfl_uv[r << 1]);
+      }
+#else
       filter_selectively_horiz(dst->buf, dst->stride,
                                mask_16x16_r,
                                mask_8x8_r,
                                mask_4x4_r,
                                mask_4x4_int_r,
-                               &cm->lf_info, &lfm->lfl_uv[r << 1]);
+                               &cm->lf_info,
+                               &lfm->lfl_uv[r << 1]);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
       dst->buf += 8 * dst->stride;
       mask_16x16 >>= 4;
@@ -1193,7 +1571,7 @@
   }
 }
 
-void vp9_loop_filter_rows(const YV12_BUFFER_CONFIG *frame_buffer,
+void vp9_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
                           VP9_COMMON *cm,
                           struct macroblockd_plane planes[MAX_MB_PLANE],
                           int start, int stop, int y_only) {
@@ -1204,7 +1582,7 @@
   int mi_row, mi_col;
 
   for (mi_row = start; mi_row < stop; mi_row += MI_BLOCK_SIZE) {
-    MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
+    MODE_INFO *mi = cm->mi + mi_row * cm->mi_stride;
 
     for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) {
       int plane;
@@ -1247,9 +1625,8 @@
                        y_only);
 }
 
-int vp9_loop_filter_worker(void *arg1, void *arg2) {
-  LFWorkerData *const lf_data = (LFWorkerData*)arg1;
-  (void)arg2;
+int vp9_loop_filter_worker(LFWorkerData *const lf_data, void *unused) {
+  (void)unused;
   vp9_loop_filter_rows(lf_data->frame_buffer, lf_data->cm, lf_data->planes,
                        lf_data->start, lf_data->stop, lf_data->y_only);
   return 1;
diff --git a/vp9/common/vp9_loopfilter.h b/vp9/common/vp9_loopfilter.h
index 6fa2773..0ede58a 100644
--- a/vp9/common/vp9_loopfilter.h
+++ b/vp9/common/vp9_loopfilter.h
@@ -89,7 +89,7 @@
 // by mi_row, mi_col.
 void vp9_setup_mask(struct VP9Common *const cm,
                     const int mi_row, const int mi_col,
-                    MODE_INFO **mi_8x8, const int mode_info_stride,
+                    MODE_INFO *mi_8x8, const int mode_info_stride,
                     LOOP_FILTER_MASK *lfm);
 
 void vp9_filter_block_plane(struct VP9Common *const cm,
@@ -111,13 +111,13 @@
                            int y_only, int partial_frame);
 
 // Apply the loop filter to [start, stop) macro block rows in frame_buffer.
-void vp9_loop_filter_rows(const YV12_BUFFER_CONFIG *frame_buffer,
+void vp9_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
                           struct VP9Common *cm,
                           struct macroblockd_plane planes[MAX_MB_PLANE],
                           int start, int stop, int y_only);
 
 typedef struct LoopFilterWorkerData {
-  const YV12_BUFFER_CONFIG *frame_buffer;
+  YV12_BUFFER_CONFIG *frame_buffer;
   struct VP9Common *cm;
   struct macroblockd_plane planes[MAX_MB_PLANE];
 
@@ -129,8 +129,8 @@
   int num_lf_workers;
 } LFWorkerData;
 
-// Operates on the rows described by LFWorkerData passed as 'arg1'.
-int vp9_loop_filter_worker(void *arg1, void *arg2);
+// Operates on the rows described by 'lf_data'.
+int vp9_loop_filter_worker(LFWorkerData *const lf_data, void *unused);
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vp9/common/vp9_loopfilter_filters.c b/vp9/common/vp9_loopfilter_filters.c
index 25d3311..5af52c2 100644
--- a/vp9/common/vp9_loopfilter_filters.c
+++ b/vp9/common/vp9_loopfilter_filters.c
@@ -17,6 +17,20 @@
   return (int8_t)clamp(t, -128, 127);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE int16_t signed_char_clamp_high(int t, int bd) {
+  switch (bd) {
+    case 10:
+      return (int16_t)clamp(t, -128*4, 128*4-1);
+    case 12:
+      return (int16_t)clamp(t, -128*16, 128*16-1);
+    case 8:
+    default:
+      return (int16_t)clamp(t, -128, 128-1);
+  }
+}
+#endif
+
 // should we apply any filter at all: 11111111 yes, 00000000 no
 static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit,
                                  uint8_t p3, uint8_t p2,
@@ -337,3 +351,390 @@
                                 const uint8_t *limit, const uint8_t *thresh) {
   mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16);
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+// Should we apply any filter at all: 11111111 yes, 00000000 no ?
+static INLINE int8_t high_filter_mask(uint8_t limit, uint8_t blimit,
+                                      uint16_t p3, uint16_t p2,
+                                      uint16_t p1, uint16_t p0,
+                                      uint16_t q0, uint16_t q1,
+                                      uint16_t q2, uint16_t q3, int bd) {
+  int8_t mask = 0;
+  int16_t limit16 = (uint16_t)limit << (bd - 8);
+  int16_t blimit16 = (uint16_t)blimit << (bd - 8);
+  mask |= (abs(p3 - p2) > limit16) * -1;
+  mask |= (abs(p2 - p1) > limit16) * -1;
+  mask |= (abs(p1 - p0) > limit16) * -1;
+  mask |= (abs(q1 - q0) > limit16) * -1;
+  mask |= (abs(q2 - q1) > limit16) * -1;
+  mask |= (abs(q3 - q2) > limit16) * -1;
+  mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit16) * -1;
+  return ~mask;
+}
+
+static INLINE int8_t high_flat_mask4(uint8_t thresh,
+                                     uint16_t p3, uint16_t p2,
+                                     uint16_t p1, uint16_t p0,
+                                     uint16_t q0, uint16_t q1,
+                                     uint16_t q2, uint16_t q3, int bd) {
+  int8_t mask = 0;
+  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
+  mask |= (abs(p1 - p0) > thresh16) * -1;
+  mask |= (abs(q1 - q0) > thresh16) * -1;
+  mask |= (abs(p2 - p0) > thresh16) * -1;
+  mask |= (abs(q2 - q0) > thresh16) * -1;
+  mask |= (abs(p3 - p0) > thresh16) * -1;
+  mask |= (abs(q3 - q0) > thresh16) * -1;
+  return ~mask;
+}
+
+static INLINE int8_t high_flat_mask5(uint8_t thresh,
+                                     uint16_t p4, uint16_t p3,
+                                     uint16_t p2, uint16_t p1,
+                                     uint16_t p0, uint16_t q0,
+                                     uint16_t q1, uint16_t q2,
+                                     uint16_t q3, uint16_t q4, int bd) {
+  int8_t mask = ~high_flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
+  mask |= (abs(p4 - p0) > thresh16) * -1;
+  mask |= (abs(q4 - q0) > thresh16) * -1;
+  return ~mask;
+}
+
+// Is there high edge variance internal edge:
+// 11111111_11111111 yes, 00000000_00000000 no ?
+static INLINE int16_t high_hev_mask(uint8_t thresh, uint16_t p1, uint16_t p0,
+                                    uint16_t q0, uint16_t q1, int bd) {
+  int16_t hev = 0;
+  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
+  hev |= (abs(p1 - p0) > thresh16) * -1;
+  hev |= (abs(q1 - q0) > thresh16) * -1;
+  return hev;
+}
+
+static INLINE void high_filter4(int8_t mask, uint8_t thresh, uint16_t *op1,
+                                uint16_t *op0, uint16_t *oq0, uint16_t *oq1,
+                                int bd) {
+  int16_t filter1, filter2;
+  // ^0x80 equivalent to subtracting 0x80 from the values to turn them
+  // into -128 to +127 instead of 0 to 255.
+  int shift = bd - 8;
+  const int16_t ps1 = (int16_t)*op1 - (0x80 << shift);
+  const int16_t ps0 = (int16_t)*op0 - (0x80 << shift);
+  const int16_t qs0 = (int16_t)*oq0 - (0x80 << shift);
+  const int16_t qs1 = (int16_t)*oq1 - (0x80 << shift);
+  const uint16_t hev = high_hev_mask(thresh, *op1, *op0, *oq0, *oq1, bd);
+
+  // Add outer taps if we have high edge variance.
+  int16_t filter = signed_char_clamp_high(ps1 - qs1, bd) & hev;
+
+  // Inner taps.
+  filter = signed_char_clamp_high(filter + 3 * (qs0 - ps0), bd) & mask;
+
+  // Save bottom 3 bits so that we round one side +4 and the other +3
+  // if it equals 4 we'll set to adjust by -1 to account for the fact
+  // we'd round 3 the other way.
+  filter1 = signed_char_clamp_high(filter + 4, bd) >> 3;
+  filter2 = signed_char_clamp_high(filter + 3, bd) >> 3;
+
+  *oq0 = signed_char_clamp_high(qs0 - filter1, bd) + (0x80 << shift);
+  *op0 = signed_char_clamp_high(ps0 + filter2, bd) + (0x80 << shift);
+
+  // Outer tap adjustments.
+  filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;
+
+  *oq1 = signed_char_clamp_high(qs1 - filter, bd) + (0x80 << shift);
+  *op1 = signed_char_clamp_high(ps1 + filter, bd) + (0x80 << shift);
+}
+
+void vp9_highbd_lpf_horizontal_4_c(uint16_t *s, int p /* pitch */,
+                                   const uint8_t *blimit, const uint8_t *limit,
+                                   const uint8_t *thresh, int count, int bd) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint16_t p3 = s[-4 * p];
+    const uint16_t p2 = s[-3 * p];
+    const uint16_t p1 = s[-2 * p];
+    const uint16_t p0 = s[-p];
+    const uint16_t q0 = s[0 * p];
+    const uint16_t q1 = s[1 * p];
+    const uint16_t q2 = s[2 * p];
+    const uint16_t q3 = s[3 * p];
+    const int8_t mask = high_filter_mask(*limit, *blimit,
+                                         p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    high_filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p, bd);
+    ++s;
+  }
+}
+
+void vp9_highbd_lpf_horizontal_4_dual_c(uint16_t *s, int p,
+                                        const uint8_t *blimit0,
+                                        const uint8_t *limit0,
+                                        const uint8_t *thresh0,
+                                        const uint8_t *blimit1,
+                                        const uint8_t *limit1,
+                                        const uint8_t *thresh1,
+                                        int bd) {
+  vp9_highbd_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1, bd);
+  vp9_highbd_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1, bd);
+}
+
+void vp9_highbd_lpf_vertical_4_c(uint16_t *s, int pitch, const uint8_t *blimit,
+                                 const uint8_t *limit, const uint8_t *thresh,
+                                 int count, int bd) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint16_t q0 = s[0],  q1 = s[1],  q2 = s[2],  q3 = s[3];
+    const int8_t mask = high_filter_mask(*limit, *blimit,
+                                         p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    high_filter4(mask, *thresh, s - 2, s - 1, s, s + 1, bd);
+    s += pitch;
+  }
+}
+
+void vp9_highbd_lpf_vertical_4_dual_c(uint16_t *s, int pitch,
+                                      const uint8_t *blimit0,
+                                      const uint8_t *limit0,
+                                      const uint8_t *thresh0,
+                                      const uint8_t *blimit1,
+                                      const uint8_t *limit1,
+                                      const uint8_t *thresh1,
+                                      int bd) {
+  vp9_highbd_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1, bd);
+  vp9_highbd_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1,
+                              thresh1, 1, bd);
+}
+
+static INLINE void high_filter8(int8_t mask, uint8_t thresh, uint8_t flat,
+                                uint16_t *op3, uint16_t *op2,
+                                uint16_t *op1, uint16_t *op0,
+                                uint16_t *oq0, uint16_t *oq1,
+                                uint16_t *oq2, uint16_t *oq3, int bd) {
+  if (flat && mask) {
+    const uint16_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0;
+    const uint16_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3;
+
+    // 7-tap filter [1, 1, 1, 2, 1, 1, 1]
+    *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3);
+    *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3);
+    *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3);
+    *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3);
+    *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3);
+    *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3);
+  } else {
+    high_filter4(mask, thresh, op1,  op0, oq0, oq1, bd);
+  }
+}
+
+void vp9_highbd_lpf_horizontal_8_c(uint16_t *s, int p, const uint8_t *blimit,
+                                   const uint8_t *limit, const uint8_t *thresh,
+                                   int count, int bd) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
+    const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
+
+    const int8_t mask = high_filter_mask(*limit, *blimit,
+                                         p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat = high_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    high_filter8(mask, *thresh, flat,
+                 s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p,
+                 s, s + 1 * p, s + 2 * p, s + 3 * p, bd);
+    ++s;
+  }
+}
+
+void vp9_highbd_lpf_horizontal_8_dual_c(uint16_t *s, int p,
+                                        const uint8_t *blimit0,
+                                        const uint8_t *limit0,
+                                        const uint8_t *thresh0,
+                                        const uint8_t *blimit1,
+                                        const uint8_t *limit1,
+                                        const uint8_t *thresh1,
+                                        int bd) {
+  vp9_highbd_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1, bd);
+  vp9_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1, bd);
+}
+
+void vp9_highbd_lpf_vertical_8_c(uint16_t *s, int pitch, const uint8_t *blimit,
+                                 const uint8_t *limit, const uint8_t *thresh,
+                                 int count, int bd) {
+  int i;
+
+  for (i = 0; i < 8 * count; ++i) {
+    const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
+    const int8_t mask = high_filter_mask(*limit, *blimit,
+                                         p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat = high_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    high_filter8(mask, *thresh, flat,
+                 s - 4, s - 3, s - 2, s - 1,
+                 s, s + 1, s + 2, s + 3,
+                 bd);
+    s += pitch;
+  }
+}
+
+void vp9_highbd_lpf_vertical_8_dual_c(uint16_t *s, int pitch,
+                                      const uint8_t *blimit0,
+                                      const uint8_t *limit0,
+                                      const uint8_t *thresh0,
+                                      const uint8_t *blimit1,
+                                      const uint8_t *limit1,
+                                      const uint8_t *thresh1,
+                                      int bd) {
+  vp9_highbd_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1, bd);
+  vp9_highbd_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1,
+                              thresh1, 1, bd);
+}
+
+static INLINE void high_filter16(int8_t mask, uint8_t thresh,
+                                 uint8_t flat, uint8_t flat2,
+                                 uint16_t *op7, uint16_t *op6,
+                                 uint16_t *op5, uint16_t *op4,
+                                 uint16_t *op3, uint16_t *op2,
+                                 uint16_t *op1, uint16_t *op0,
+                                 uint16_t *oq0, uint16_t *oq1,
+                                 uint16_t *oq2, uint16_t *oq3,
+                                 uint16_t *oq4, uint16_t *oq5,
+                                 uint16_t *oq6, uint16_t *oq7, int bd) {
+  if (flat2 && flat && mask) {
+    const uint16_t p7 = *op7;
+    const uint16_t p6 = *op6;
+    const uint16_t p5 = *op5;
+    const uint16_t p4 = *op4;
+    const uint16_t p3 = *op3;
+    const uint16_t p2 = *op2;
+    const uint16_t p1 = *op1;
+    const uint16_t p0 = *op0;
+    const uint16_t q0 = *oq0;
+    const uint16_t q1 = *oq1;
+    const uint16_t q2 = *oq2;
+    const uint16_t q3 = *oq3;
+    const uint16_t q4 = *oq4;
+    const uint16_t q5 = *oq5;
+    const uint16_t q6 = *oq6;
+    const uint16_t q7 = *oq7;
+
+    // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1]
+    *op6 = ROUND_POWER_OF_TWO(p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 +
+                              q0, 4);
+    *op5 = ROUND_POWER_OF_TWO(p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 +
+                              q0 + q1, 4);
+    *op4 = ROUND_POWER_OF_TWO(p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 +
+                              q0 + q1 + q2, 4);
+    *op3 = ROUND_POWER_OF_TWO(p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 +
+                              q0 + q1 + q2 + q3, 4);
+    *op2 = ROUND_POWER_OF_TWO(p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 +
+                              q0 + q1 + q2 + q3 + q4, 4);
+    *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 +
+                              q0 + q1 + q2 + q3 + q4 + q5, 4);
+    *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 +
+                              q0 + q1 + q2 + q3 + q4 + q5 + q6, 4);
+    *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 +
+                              q0 * 2 + q1 + q2 + q3 + q4 + q5 + q6 + q7, 4);
+    *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 +
+                              q0 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + q7 * 2, 4);
+    *oq2 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 +
+                              q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3, 4);
+    *oq3 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 +
+                              q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4);
+    *oq4 = ROUND_POWER_OF_TWO(p2 + p1 + p0 +
+                              q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4);
+    *oq5 = ROUND_POWER_OF_TWO(p1 + p0 +
+                              q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4);
+    *oq6 = ROUND_POWER_OF_TWO(p0 +
+                              q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4);
+  } else {
+    high_filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3,
+                 bd);
+  }
+}
+
+void vp9_highbd_lpf_horizontal_16_c(uint16_t *s, int p, const uint8_t *blimit,
+                                    const uint8_t *limit, const uint8_t *thresh,
+                                    int count, int bd) {
+  int i;
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < 8 * count; ++i) {
+    const uint16_t p3 = s[-4 * p];
+    const uint16_t p2 = s[-3 * p];
+    const uint16_t p1 = s[-2 * p];
+    const uint16_t p0 = s[-p];
+    const uint16_t q0 = s[0 * p];
+    const uint16_t q1 = s[1 * p];
+    const uint16_t q2 = s[2 * p];
+    const uint16_t q3 = s[3 * p];
+    const int8_t mask = high_filter_mask(*limit, *blimit,
+                                    p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat = high_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat2 = high_flat_mask5(
+        1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0,
+        q0, s[4 * p], s[5 * p], s[6 * p], s[7 * p], bd);
+
+    high_filter16(mask, *thresh, flat, flat2,
+                  s - 8 * p, s - 7 * p, s - 6 * p, s - 5 * p,
+                  s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p,
+                  s, s + 1 * p, s + 2 * p, s + 3 * p,
+                  s + 4 * p, s + 5 * p, s + 6 * p, s + 7 * p,
+                  bd);
+    ++s;
+  }
+}
+
+static void high_mb_lpf_vertical_edge_w(uint16_t *s, int p,
+                                        const uint8_t *blimit,
+                                        const uint8_t *limit,
+                                        const uint8_t *thresh,
+                                        int count, int bd) {
+  int i;
+
+  for (i = 0; i < count; ++i) {
+    const uint16_t p3 = s[-4];
+    const uint16_t p2 = s[-3];
+    const uint16_t p1 = s[-2];
+    const uint16_t p0 = s[-1];
+    const uint16_t q0 = s[0];
+    const uint16_t q1 = s[1];
+    const uint16_t q2 = s[2];
+    const uint16_t q3 = s[3];
+    const int8_t mask = high_filter_mask(*limit, *blimit,
+                                         p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat = high_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat2 = high_flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0,
+                                         q0, s[4], s[5], s[6], s[7], bd);
+
+    high_filter16(mask, *thresh, flat, flat2,
+                  s - 8, s - 7, s - 6, s - 5, s - 4, s - 3, s - 2, s - 1,
+                  s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6, s + 7,
+                  bd);
+    s += p;
+  }
+}
+
+void vp9_highbd_lpf_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit,
+                                  const uint8_t *limit, const uint8_t *thresh,
+                                  int bd) {
+  high_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8, bd);
+}
+
+void vp9_highbd_lpf_vertical_16_dual_c(uint16_t *s, int p,
+                                       const uint8_t *blimit,
+                                       const uint8_t *limit,
+                                       const uint8_t *thresh,
+                                       int bd) {
+  high_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/common/vp9_mvref_common.c b/vp9/common/vp9_mvref_common.c
index ab64d30..a09afff 100644
--- a/vp9/common/vp9_mvref_common.c
+++ b/vp9/common/vp9_mvref_common.c
@@ -21,9 +21,9 @@
   const int *ref_sign_bias = cm->ref_frame_sign_bias;
   int i, refmv_count = 0;
   const MODE_INFO *prev_mi = !cm->error_resilient_mode && cm->prev_mi
-        ? cm->prev_mi_grid_visible[mi_row * xd->mi_stride + mi_col]
+        ? cm->prev_mi[mi_row * xd->mi_stride + mi_col].src_mi
         : NULL;
-  const MB_MODE_INFO *const prev_mbmi = prev_mi ? &prev_mi->mbmi : NULL;
+  const MB_MODE_INFO *const prev_mbmi = prev_mi ? &prev_mi->src_mi->mbmi : NULL;
 
 
   const POSITION *const mv_ref_search = mv_ref_blocks[mi->mbmi.sb_type];
@@ -41,7 +41,7 @@
     const POSITION *const mv_ref = &mv_ref_search[i];
     if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
       const MODE_INFO *const candidate_mi = xd->mi[mv_ref->col + mv_ref->row *
-                                                   xd->mi_stride];
+                                                   xd->mi_stride].src_mi;
       const MB_MODE_INFO *const candidate = &candidate_mi->mbmi;
       // Keep counts for entropy encoding.
       context_counter += mode_2_counter[candidate->mode];
@@ -61,7 +61,7 @@
     const POSITION *const mv_ref = &mv_ref_search[i];
     if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
       const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row *
-                                                    xd->mi_stride]->mbmi;
+                                                    xd->mi_stride].src_mi->mbmi;
       different_ref_found = 1;
 
       if (candidate->ref_frame[0] == ref_frame)
@@ -87,7 +87,7 @@
       const POSITION *mv_ref = &mv_ref_search[i];
       if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
         const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row
-                                              * xd->mi_stride]->mbmi;
+                                              * xd->mi_stride].src_mi->mbmi;
 
         // If the candidate is INTRA we don't want to consider its mv.
         IF_DIFF_REF_FRAME_ADD_MV(candidate);
@@ -145,7 +145,7 @@
                                    int block, int ref, int mi_row, int mi_col,
                                    int_mv *nearest, int_mv *near) {
   int_mv mv_list[MAX_MV_REF_CANDIDATES];
-  MODE_INFO *const mi = xd->mi[0];
+  MODE_INFO *const mi = xd->mi[0].src_mi;
   b_mode_info *bmi = mi->bmi;
   int n;
 
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index 637867a..792e9d9 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -150,11 +150,6 @@
   MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */
   MODE_INFO *prev_mi;  /* 'mi' from last frame (points into prev_mip) */
 
-  MODE_INFO **mi_grid_base;
-  MODE_INFO **mi_grid_visible;
-  MODE_INFO **prev_mi_grid_base;
-  MODE_INFO **prev_mi_grid_visible;
-
   // Persistent mb segment id map used in prediction.
   unsigned char *last_frame_seg_map;
 
diff --git a/vp9/common/vp9_postproc.c b/vp9/common/vp9_postproc.c
index e4e6ce7..abe71da 100644
--- a/vp9/common/vp9_postproc.c
+++ b/vp9/common/vp9_postproc.c
@@ -19,6 +19,9 @@
 #include "vpx_scale/vpx_scale.h"
 #include "vpx_scale/yv12config.h"
 
+#if CONFIG_VP9_HIGHBITDEPTH
+#include "vp9/common/vp9_common.h"
+#endif
 #include "vp9/common/vp9_onyxc_int.h"
 #include "vp9/common/vp9_postproc.h"
 #include "vp9/common/vp9_systemdependent.h"
@@ -152,6 +155,84 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_post_proc_down_and_across_c(const uint16_t *src_ptr,
+                                            uint16_t *dst_ptr,
+                                            int src_pixels_per_line,
+                                            int dst_pixels_per_line,
+                                            int rows,
+                                            int cols,
+                                            int flimit) {
+  uint16_t const *p_src;
+  uint16_t *p_dst;
+  int row;
+  int col;
+  int i;
+  int v;
+  int pitch = src_pixels_per_line;
+  uint16_t d[8];
+
+  for (row = 0; row < rows; row++) {
+    // post_proc_down for one row.
+    p_src = src_ptr;
+    p_dst = dst_ptr;
+
+    for (col = 0; col < cols; col++) {
+      int kernel = 4;
+      int v = p_src[col];
+
+      for (i = -2; i <= 2; i++) {
+        if (abs(v - p_src[col + i * pitch]) > flimit)
+          goto down_skip_convolve;
+
+        kernel += kernel5[2 + i] * p_src[col + i * pitch];
+      }
+
+      v = (kernel >> 3);
+
+    down_skip_convolve:
+      p_dst[col] = v;
+    }
+
+    /* now post_proc_across */
+    p_src = dst_ptr;
+    p_dst = dst_ptr;
+
+    for (i = 0; i < 8; i++)
+      d[i] = p_src[i];
+
+    for (col = 0; col < cols; col++) {
+      int kernel = 4;
+      v = p_src[col];
+
+      d[col & 7] = v;
+
+      for (i = -2; i <= 2; i++) {
+        if (abs(v - p_src[col + i]) > flimit)
+          goto across_skip_convolve;
+
+        kernel += kernel5[2 + i] * p_src[col + i];
+      }
+
+      d[col & 7] = (kernel >> 3);
+
+    across_skip_convolve:
+      if (col >= 2)
+        p_dst[col - 2] = d[(col - 2) & 7];
+    }
+
+    /* handle the last two pixels */
+    p_dst[col - 2] = d[(col - 2) & 7];
+    p_dst[col - 1] = d[(col - 1) & 7];
+
+
+    /* next row */
+    src_ptr += pitch;
+    dst_ptr += dst_pixels_per_line;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static int q2mbl(int x) {
   if (x < 20) x = 20;
 
@@ -162,10 +243,46 @@
 void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
                                  int rows, int cols, int flimit) {
   int r, c, i;
-
   uint8_t *s = src;
   uint8_t d[16];
 
+  for (r = 0; r < rows; r++) {
+    int sumsq = 0;
+    int sum = 0;
+
+    for (i = -8; i <= 6; i++) {
+      sumsq += s[i] * s[i];
+      sum += s[i];
+      d[i + 8] = 0;
+    }
+
+    for (c = 0; c < cols + 8; c++) {
+      int x = s[c + 7] - s[c - 8];
+      int y = s[c + 7] + s[c - 8];
+
+      sum += x;
+      sumsq += x * y;
+
+      d[c & 15] = s[c];
+
+      if (sumsq * 15 - sum * sum < flimit) {
+        d[c & 15] = (8 + sum + s[c]) >> 4;
+      }
+
+      s[c - 8] = d[(c - 8) & 15];
+    }
+    s += pitch;
+  }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_mbpost_proc_across_ip_c(uint16_t *src, int pitch,
+                                        int rows, int cols, int flimit) {
+  int r, c, i;
+
+  uint16_t *s = src;
+  uint16_t d[16];
+
 
   for (r = 0; r < rows; r++) {
     int sumsq = 0;
@@ -196,6 +313,7 @@
     s += pitch;
   }
 }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
                             int rows, int cols, int flimit) {
@@ -229,6 +347,40 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_mbpost_proc_down_c(uint16_t *dst, int pitch,
+                                   int rows, int cols, int flimit) {
+  int r, c, i;
+  const int16_t *rv3 = &vp9_rv[63 & rand()];  // NOLINT
+
+  for (c = 0; c < cols; c++) {
+    uint16_t *s = &dst[c];
+    int sumsq = 0;
+    int sum = 0;
+    uint16_t d[16];
+    const int16_t *rv2 = rv3 + ((c * 17) & 127);
+
+    for (i = -8; i <= 6; i++) {
+      sumsq += s[i * pitch] * s[i * pitch];
+      sum += s[i * pitch];
+    }
+
+    for (r = 0; r < rows + 8; r++) {
+      sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
+      sum += s[7 * pitch] - s[-8 * pitch];
+      d[r & 15] = s[0];
+
+      if (sumsq * 15 - sum * sum < flimit) {
+        d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
+      }
+
+      s[-8 * pitch] = d[(r - 8) & 15];
+      s += pitch;
+    }
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG   *source,
                                        YV12_BUFFER_CONFIG   *post,
                                        int                   q,
@@ -239,6 +391,51 @@
   (void) low_var_thresh;
   (void) flag;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (source->flags & YV12_FLAG_HIGHBITDEPTH) {
+    vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->y_buffer),
+                                         CONVERT_TO_SHORTPTR(post->y_buffer),
+                                         source->y_stride, post->y_stride,
+                                         source->y_height, source->y_width,
+                                         ppl);
+
+    vp9_highbd_mbpost_proc_across_ip(CONVERT_TO_SHORTPTR(post->y_buffer),
+                                     post->y_stride, post->y_height,
+                                     post->y_width, q2mbl(q));
+
+    vp9_highbd_mbpost_proc_down(CONVERT_TO_SHORTPTR(post->y_buffer),
+                                post->y_stride, post->y_height,
+                                post->y_width, q2mbl(q));
+
+    vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->u_buffer),
+                                         CONVERT_TO_SHORTPTR(post->u_buffer),
+                                         source->uv_stride, post->uv_stride,
+                                         source->uv_height, source->uv_width,
+                                         ppl);
+    vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->v_buffer),
+                                         CONVERT_TO_SHORTPTR(post->v_buffer),
+                                         source->uv_stride, post->uv_stride,
+                                         source->uv_height, source->uv_width,
+                                         ppl);
+  } else {
+    vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
+                                  source->y_stride, post->y_stride,
+                                  source->y_height, source->y_width, ppl);
+
+    vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
+                              post->y_width, q2mbl(q));
+
+    vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
+                         post->y_width, q2mbl(q));
+
+    vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
+                                  source->uv_stride, post->uv_stride,
+                                  source->uv_height, source->uv_width, ppl);
+    vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
+                                  source->uv_stride, post->uv_stride,
+                                  source->uv_height, source->uv_width, ppl);
+  }
+#else
   vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
                                 source->y_stride, post->y_stride,
                                 source->y_height, source->y_width, ppl);
@@ -255,6 +452,7 @@
   vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
                                 source->uv_stride, post->uv_stride,
                                 source->uv_height, source->uv_width, ppl);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 }
 
 void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
@@ -272,9 +470,24 @@
   const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
 
   for (i = 0; i < MAX_MB_PLANE; ++i)
+#if CONFIG_VP9_HIGHBITDEPTH
+    assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
+           (dst->flags & YV12_FLAG_HIGHBITDEPTH));
+    if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+      vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(srcs[i]),
+                                           CONVERT_TO_SHORTPTR(dsts[i]),
+                                           src_strides[i], dst_strides[i],
+                                           src_heights[i], src_widths[i], ppl);
+    } else {
+      vp9_post_proc_down_and_across(srcs[i], dsts[i],
+                                    src_strides[i], dst_strides[i],
+                                    src_heights[i], src_widths[i], ppl);
+    }
+#else
     vp9_post_proc_down_and_across(srcs[i], dsts[i],
                                   src_strides[i], dst_strides[i],
                                   src_heights[i], src_widths[i], ppl);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 }
 
 void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
@@ -293,15 +506,32 @@
 
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     const int src_stride = src_strides[i];
-    const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
     const int src_width = src_widths[i] - 4;
     const int src_height = src_heights[i] - 4;
-
     const int dst_stride = dst_strides[i];
-    uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+    assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
+           (dst->flags & YV12_FLAG_HIGHBITDEPTH));
+    if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+      const uint16_t *const src = CONVERT_TO_SHORTPTR(srcs[i] + 2 * src_stride
+                                                      + 2);
+      uint16_t *const dst = CONVERT_TO_SHORTPTR(dsts[i] + 2 * dst_stride + 2);
+      vp9_highbd_post_proc_down_and_across(src, dst, src_stride, dst_stride,
+                                           src_height, src_width, ppl);
+    } else {
+      const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
+      uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
+
+      vp9_post_proc_down_and_across(src, dst, src_stride, dst_stride,
+                                    src_height, src_width, ppl);
+    }
+#else
+    const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
+    uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
     vp9_post_proc_down_and_across(src, dst, src_stride, dst_stride,
                                   src_height, src_width, ppl);
+#endif
   }
 }
 
@@ -405,6 +635,9 @@
 #if CONFIG_VP9_POSTPROC || CONFIG_INTERNAL_STATS
   if (vp9_realloc_frame_buffer(&cm->post_proc_buffer, cm->width, cm->height,
                                cm->subsampling_x, cm->subsampling_y,
+#if CONFIG_VP9_HIGHBITDEPTH
+                               cm->use_highbitdepth,
+#endif
                                VP9_DEC_BORDER_IN_PIXELS, NULL, NULL, NULL) < 0)
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate post-processing buffer");
diff --git a/vp9/common/vp9_pred_common.c b/vp9/common/vp9_pred_common.c
index 0146384..901a043 100644
--- a/vp9/common/vp9_pred_common.c
+++ b/vp9/common/vp9_pred_common.c
@@ -348,7 +348,7 @@
 // left of the entries corresponding to real blocks.
 // The prediction flags in these dummy entries are initialized to 0.
 int vp9_get_tx_size_context(const MACROBLOCKD *xd) {
-  const int max_tx_size = max_txsize_lookup[xd->mi[0]->mbmi.sb_type];
+  const int max_tx_size = max_txsize_lookup[xd->mi[0].src_mi->mbmi.sb_type];
   const MB_MODE_INFO *const above_mbmi = get_mbmi(get_above_mi(xd));
   const MB_MODE_INFO *const left_mbmi = get_mbmi(get_left_mi(xd));
   const int has_above = above_mbmi != NULL;
diff --git a/vp9/common/vp9_pred_common.h b/vp9/common/vp9_pred_common.h
index 2c96506..39774f1 100644
--- a/vp9/common/vp9_pred_common.h
+++ b/vp9/common/vp9_pred_common.h
@@ -19,11 +19,11 @@
 #endif
 
 static INLINE const MODE_INFO *get_above_mi(const MACROBLOCKD *const xd) {
-  return xd->up_available ? xd->mi[-xd->mi_stride] : NULL;
+  return xd->up_available ? xd->mi[-xd->mi_stride].src_mi : NULL;
 }
 
 static INLINE const MODE_INFO *get_left_mi(const MACROBLOCKD *const xd) {
-  return xd->left_available ? xd->mi[-1] : NULL;
+  return xd->left_available ? xd->mi[-1].src_mi : NULL;
 }
 
 int vp9_get_segment_id(const VP9_COMMON *cm, const uint8_t *segment_ids,
diff --git a/vp9/common/vp9_quant_common.c b/vp9/common/vp9_quant_common.c
index 3332e58..564a3eb 100644
--- a/vp9/common/vp9_quant_common.c
+++ b/vp9/common/vp9_quant_common.c
@@ -47,6 +47,78 @@
   1022, 1058, 1098, 1139, 1184, 1232, 1282, 1336,
 };
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static const int16_t dc_qlookup_10[QINDEX_RANGE] = {
+  4,     9,    10,    13,    15,    17,    20,    22,
+  25,    28,    31,    34,    37,    40,    43,    47,
+  50,    53,    57,    60,    64,    68,    71,    75,
+  78,    82,    86,    90,    93,    97,   101,   105,
+  109,   113,   116,   120,   124,   128,   132,   136,
+  140,   143,   147,   151,   155,   159,   163,   166,
+  170,   174,   178,   182,   185,   189,   193,   197,
+  200,   204,   208,   212,   215,   219,   223,   226,
+  230,   233,   237,   241,   244,   248,   251,   255,
+  259,   262,   266,   269,   273,   276,   280,   283,
+  287,   290,   293,   297,   300,   304,   307,   310,
+  314,   317,   321,   324,   327,   331,   334,   337,
+  343,   350,   356,   362,   369,   375,   381,   387,
+  394,   400,   406,   412,   418,   424,   430,   436,
+  442,   448,   454,   460,   466,   472,   478,   484,
+  490,   499,   507,   516,   525,   533,   542,   550,
+  559,   567,   576,   584,   592,   601,   609,   617,
+  625,   634,   644,   655,   666,   676,   687,   698,
+  708,   718,   729,   739,   749,   759,   770,   782,
+  795,   807,   819,   831,   844,   856,   868,   880,
+  891,   906,   920,   933,   947,   961,   975,   988,
+  1001,  1015,  1030,  1045,  1061,  1076,  1090,  1105,
+  1120,  1137,  1153,  1170,  1186,  1202,  1218,  1236,
+  1253,  1271,  1288,  1306,  1323,  1342,  1361,  1379,
+  1398,  1416,  1436,  1456,  1476,  1496,  1516,  1537,
+  1559,  1580,  1601,  1624,  1647,  1670,  1692,  1717,
+  1741,  1766,  1791,  1817,  1844,  1871,  1900,  1929,
+  1958,  1990,  2021,  2054,  2088,  2123,  2159,  2197,
+  2236,  2276,  2319,  2363,  2410,  2458,  2508,  2561,
+  2616,  2675,  2737,  2802,  2871,  2944,  3020,  3102,
+  3188,  3280,  3375,  3478,  3586,  3702,  3823,  3953,
+  4089,  4236,  4394,  4559,  4737,  4929,  5130,  5347,
+};
+
+static const int16_t dc_qlookup_12[QINDEX_RANGE] = {
+  4,    12,    18,    25,    33,    41,    50,    60,
+  70,    80,    91,   103,   115,   127,   140,   153,
+  166,   180,   194,   208,   222,   237,   251,   266,
+  281,   296,   312,   327,   343,   358,   374,   390,
+  405,   421,   437,   453,   469,   484,   500,   516,
+  532,   548,   564,   580,   596,   611,   627,   643,
+  659,   674,   690,   706,   721,   737,   752,   768,
+  783,   798,   814,   829,   844,   859,   874,   889,
+  904,   919,   934,   949,   964,   978,   993,  1008,
+  1022,  1037,  1051,  1065,  1080,  1094,  1108,  1122,
+  1136,  1151,  1165,  1179,  1192,  1206,  1220,  1234,
+  1248,  1261,  1275,  1288,  1302,  1315,  1329,  1342,
+  1368,  1393,  1419,  1444,  1469,  1494,  1519,  1544,
+  1569,  1594,  1618,  1643,  1668,  1692,  1717,  1741,
+  1765,  1789,  1814,  1838,  1862,  1885,  1909,  1933,
+  1957,  1992,  2027,  2061,  2096,  2130,  2165,  2199,
+  2233,  2267,  2300,  2334,  2367,  2400,  2434,  2467,
+  2499,  2532,  2575,  2618,  2661,  2704,  2746,  2788,
+  2830,  2872,  2913,  2954,  2995,  3036,  3076,  3127,
+  3177,  3226,  3275,  3324,  3373,  3421,  3469,  3517,
+  3565,  3621,  3677,  3733,  3788,  3843,  3897,  3951,
+  4005,  4058,  4119,  4181,  4241,  4301,  4361,  4420,
+  4479,  4546,  4612,  4677,  4742,  4807,  4871,  4942,
+  5013,  5083,  5153,  5222,  5291,  5367,  5442,  5517,
+  5591,  5665,  5745,  5825,  5905,  5984,  6063,  6149,
+  6234,  6319,  6404,  6495,  6587,  6678,  6769,  6867,
+  6966,  7064,  7163,  7269,  7376,  7483,  7599,  7715,
+  7832,  7958,  8085,  8214,  8352,  8492,  8635,  8788,
+  8945,  9104,  9275,  9450,  9639,  9832, 10031, 10245,
+  10465, 10702, 10946, 11210, 11482, 11776, 12081, 12409,
+  12750, 13118, 13501, 13913, 14343, 14807, 15290, 15812,
+  16356, 16943, 17575, 18237, 18949, 19718, 20521, 21387,
+};
+#endif
+
 static const int16_t ac_qlookup[QINDEX_RANGE] = {
   4,       8,    9,   10,   11,   12,   13,   14,
   15,     16,   17,   18,   19,   20,   21,   22,
@@ -82,15 +154,116 @@
   1597, 1628, 1660, 1692, 1725, 1759, 1793, 1828,
 };
 
-int16_t vp9_dc_quant(int qindex, int delta) {
+#if CONFIG_VP9_HIGHBITDEPTH
+static const int16_t ac_qlookup_10[QINDEX_RANGE] = {
+  4,     9,    11,    13,    16,    18,    21,    24,
+  27,    30,    33,    37,    40,    44,    48,    51,
+  55,    59,    63,    67,    71,    75,    79,    83,
+  88,    92,    96,   100,   105,   109,   114,   118,
+  122,   127,   131,   136,   140,   145,   149,   154,
+  158,   163,   168,   172,   177,   181,   186,   190,
+  195,   199,   204,   208,   213,   217,   222,   226,
+  231,   235,   240,   244,   249,   253,   258,   262,
+  267,   271,   275,   280,   284,   289,   293,   297,
+  302,   306,   311,   315,   319,   324,   328,   332,
+  337,   341,   345,   349,   354,   358,   362,   367,
+  371,   375,   379,   384,   388,   392,   396,   401,
+  409,   417,   425,   433,   441,   449,   458,   466,
+  474,   482,   490,   498,   506,   514,   523,   531,
+  539,   547,   555,   563,   571,   579,   588,   596,
+  604,   616,   628,   640,   652,   664,   676,   688,
+  700,   713,   725,   737,   749,   761,   773,   785,
+  797,   809,   825,   841,   857,   873,   889,   905,
+  922,   938,   954,   970,   986,  1002,  1018,  1038,
+  1058,  1078,  1098,  1118,  1138,  1158,  1178,  1198,
+  1218,  1242,  1266,  1290,  1314,  1338,  1362,  1386,
+  1411,  1435,  1463,  1491,  1519,  1547,  1575,  1603,
+  1631,  1663,  1695,  1727,  1759,  1791,  1823,  1859,
+  1895,  1931,  1967,  2003,  2039,  2079,  2119,  2159,
+  2199,  2239,  2283,  2327,  2371,  2415,  2459,  2507,
+  2555,  2603,  2651,  2703,  2755,  2807,  2859,  2915,
+  2971,  3027,  3083,  3143,  3203,  3263,  3327,  3391,
+  3455,  3523,  3591,  3659,  3731,  3803,  3876,  3952,
+  4028,  4104,  4184,  4264,  4348,  4432,  4516,  4604,
+  4692,  4784,  4876,  4972,  5068,  5168,  5268,  5372,
+  5476,  5584,  5692,  5804,  5916,  6032,  6148,  6268,
+  6388,  6512,  6640,  6768,  6900,  7036,  7172,  7312,
+};
+
+static const int16_t ac_qlookup_12[QINDEX_RANGE] = {
+  4,    13,    19,    27,    35,    44,    54,    64,
+  75,    87,    99,   112,   126,   139,   154,   168,
+  183,   199,   214,   230,   247,   263,   280,   297,
+  314,   331,   349,   366,   384,   402,   420,   438,
+  456,   475,   493,   511,   530,   548,   567,   586,
+  604,   623,   642,   660,   679,   698,   716,   735,
+  753,   772,   791,   809,   828,   846,   865,   884,
+  902,   920,   939,   957,   976,   994,  1012,  1030,
+  1049,  1067,  1085,  1103,  1121,  1139,  1157,  1175,
+  1193,  1211,  1229,  1246,  1264,  1282,  1299,  1317,
+  1335,  1352,  1370,  1387,  1405,  1422,  1440,  1457,
+  1474,  1491,  1509,  1526,  1543,  1560,  1577,  1595,
+  1627,  1660,  1693,  1725,  1758,  1791,  1824,  1856,
+  1889,  1922,  1954,  1987,  2020,  2052,  2085,  2118,
+  2150,  2183,  2216,  2248,  2281,  2313,  2346,  2378,
+  2411,  2459,  2508,  2556,  2605,  2653,  2701,  2750,
+  2798,  2847,  2895,  2943,  2992,  3040,  3088,  3137,
+  3185,  3234,  3298,  3362,  3426,  3491,  3555,  3619,
+  3684,  3748,  3812,  3876,  3941,  4005,  4069,  4149,
+  4230,  4310,  4390,  4470,  4550,  4631,  4711,  4791,
+  4871,  4967,  5064,  5160,  5256,  5352,  5448,  5544,
+  5641,  5737,  5849,  5961,  6073,  6185,  6297,  6410,
+  6522,  6650,  6778,  6906,  7034,  7162,  7290,  7435,
+  7579,  7723,  7867,  8011,  8155,  8315,  8475,  8635,
+  8795,  8956,  9132,  9308,  9484,  9660,  9836, 10028,
+  10220, 10412, 10604, 10812, 11020, 11228, 11437, 11661,
+  11885, 12109, 12333, 12573, 12813, 13053, 13309, 13565,
+  13821, 14093, 14365, 14637, 14925, 15213, 15502, 15806,
+  16110, 16414, 16734, 17054, 17390, 17726, 18062, 18414,
+  18766, 19134, 19502, 19886, 20270, 20670, 21070, 21486,
+  21902, 22334, 22766, 23214, 23662, 24126, 24590, 25070,
+  25551, 26047, 26559, 27071, 27599, 28143, 28687, 29247,
+};
+#endif
+
+int16_t vp9_dc_quant(int qindex, int delta, vpx_bit_depth_t bit_depth) {
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      return dc_qlookup[clamp(qindex + delta, 0, MAXQ)];
+    case VPX_BITS_10:
+      return dc_qlookup_10[clamp(qindex + delta, 0, MAXQ)];
+    case VPX_BITS_12:
+      return dc_qlookup_12[clamp(qindex + delta, 0, MAXQ)];
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
+  (void) bit_depth;
   return dc_qlookup[clamp(qindex + delta, 0, MAXQ)];
+#endif
 }
 
-int16_t vp9_ac_quant(int qindex, int delta) {
+int16_t vp9_ac_quant(int qindex, int delta, vpx_bit_depth_t bit_depth) {
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      return ac_qlookup[clamp(qindex + delta, 0, MAXQ)];
+    case VPX_BITS_10:
+      return ac_qlookup_10[clamp(qindex + delta, 0, MAXQ)];
+    case VPX_BITS_12:
+      return ac_qlookup_12[clamp(qindex + delta, 0, MAXQ)];
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
+  (void) bit_depth;
   return ac_qlookup[clamp(qindex + delta, 0, MAXQ)];
+#endif
 }
 
-
 int vp9_get_qindex(const struct segmentation *seg, int segment_id,
                    int base_qindex) {
   if (vp9_segfeature_active(seg, segment_id, SEG_LVL_ALT_Q)) {
diff --git a/vp9/common/vp9_quant_common.h b/vp9/common/vp9_quant_common.h
index d1545d9..b626605 100644
--- a/vp9/common/vp9_quant_common.h
+++ b/vp9/common/vp9_quant_common.h
@@ -11,6 +11,7 @@
 #ifndef VP9_COMMON_VP9_QUANT_COMMON_H_
 #define VP9_COMMON_VP9_QUANT_COMMON_H_
 
+#include "vpx/vpx_codec.h"
 #include "vp9/common/vp9_blockd.h"
 
 #ifdef __cplusplus
@@ -22,8 +23,8 @@
 #define QINDEX_RANGE (MAXQ - MINQ + 1)
 #define QINDEX_BITS 8
 
-int16_t vp9_dc_quant(int qindex, int delta);
-int16_t vp9_ac_quant(int qindex, int delta);
+int16_t vp9_dc_quant(int qindex, int delta, vpx_bit_depth_t bit_depth);
+int16_t vp9_ac_quant(int qindex, int delta, vpx_bit_depth_t bit_depth);
 
 int vp9_get_qindex(const struct segmentation *seg, int segment_id,
                    int base_qindex);
diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index 65c8759..faf5d65 100644
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -63,6 +63,53 @@
   } while (--b_h);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_build_mc_border(const uint8_t *src8, int src_stride,
+                                 uint16_t *dst, int dst_stride,
+                                 int x, int y, int b_w, int b_h,
+                                 int w, int h) {
+  // Get a pointer to the start of the real data for this row.
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *ref_row = src - x - y * src_stride;
+
+  if (y >= h)
+    ref_row += (h - 1) * src_stride;
+  else if (y > 0)
+    ref_row += y * src_stride;
+
+  do {
+    int right = 0, copy;
+    int left = x < 0 ? -x : 0;
+
+    if (left > b_w)
+      left = b_w;
+
+    if (x + b_w > w)
+      right = x + b_w - w;
+
+    if (right > b_w)
+      right = b_w;
+
+    copy = b_w - left - right;
+
+    if (left)
+      vpx_memset16(dst, ref_row[0], left);
+
+    if (copy)
+      memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t));
+
+    if (right)
+      vpx_memset16(dst + left + copy, ref_row[w - 1], right);
+
+    dst += dst_stride;
+    ++y;
+
+    if (y > 0 && y < h)
+      ref_row += src_stride;
+  } while (--b_h);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void inter_predictor(const uint8_t *src, int src_stride,
                             uint8_t *dst, int dst_stride,
                             const int subpel_x,
@@ -97,6 +144,42 @@
                   sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void high_inter_predictor(const uint8_t *src, int src_stride,
+                                 uint8_t *dst, int dst_stride,
+                                 const int subpel_x,
+                                 const int subpel_y,
+                                 const struct scale_factors *sf,
+                                 int w, int h, int ref,
+                                 const InterpKernel *kernel,
+                                 int xs, int ys, int bd) {
+  sf->high_predict[subpel_x != 0][subpel_y != 0][ref](
+      src, src_stride, dst, dst_stride,
+      kernel[subpel_x], xs, kernel[subpel_y], ys, w, h, bd);
+}
+
+void vp9_high_build_inter_predictor(const uint8_t *src, int src_stride,
+                                    uint8_t *dst, int dst_stride,
+                                    const MV *src_mv,
+                                    const struct scale_factors *sf,
+                                    int w, int h, int ref,
+                                    const InterpKernel *kernel,
+                                    enum mv_precision precision,
+                                    int x, int y, int bd) {
+  const int is_q4 = precision == MV_PRECISION_Q4;
+  const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
+                     is_q4 ? src_mv->col : src_mv->col * 2 };
+  MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
+  const int subpel_x = mv.col & SUBPEL_MASK;
+  const int subpel_y = mv.row & SUBPEL_MASK;
+
+  src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
+
+  high_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
+                       sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static INLINE int round_mv_comp_q4(int value) {
   return (value < 0 ? value - 2 : value + 2) / 4;
 }
@@ -179,7 +262,7 @@
                                    int x, int y, int w, int h,
                                    int mi_x, int mi_y) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  const MODE_INFO *mi = xd->mi[0];
+  const MODE_INFO *mi = xd->mi[0].src_mi;
   const int is_compound = has_second_ref(&mi->mbmi);
   const InterpKernel *kernel = vp9_get_interp_kernel(mi->mbmi.interp_filter);
   int ref;
@@ -222,8 +305,19 @@
     pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
            + (scaled_mv.col >> SUBPEL_BITS);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      high_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
+                           subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys,
+                           xd->bd);
+    } else {
+      inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
+                      subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
+    }
+#else
     inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
                     subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 }
 
@@ -241,7 +335,7 @@
     const int bw = 4 * num_4x4_w;
     const int bh = 4 * num_4x4_h;
 
-    if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
+    if (xd->mi[0].src_mi->mbmi.sb_type < BLOCK_8X8) {
       int i = 0, x, y;
       assert(bsize == BLOCK_8X8);
       for (y = 0; y < num_4x4_h; ++y)
@@ -277,7 +371,7 @@
                                        int x, int y, int w, int h,
                                        int mi_x, int mi_y) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  const MODE_INFO *mi = xd->mi[0];
+  const MODE_INFO *mi = xd->mi[0].src_mi;
   const int is_compound = has_second_ref(&mi->mbmi);
   const InterpKernel *kernel = vp9_get_interp_kernel(mi->mbmi.interp_filter);
   int ref;
@@ -393,16 +487,64 @@
           y0 < 0 || y0 > frame_height - 1 || y1 < 0 || y1 > frame_height - 1) {
         uint8_t *buf_ptr1 = ref_frame + y0 * pre_buf->stride + x0;
         // Extend the border.
-        build_mc_border(buf_ptr1, pre_buf->stride, xd->mc_buf, x1 - x0 + 1,
-                        x0, y0, x1 - x0 + 1, y1 - y0 + 1, frame_width,
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+          high_build_mc_border(buf_ptr1,
+                               pre_buf->stride,
+                               xd->mc_buf_high,
+                               x1 - x0 + 1,
+                               x0,
+                               y0,
+                               x1 - x0 + 1,
+                               y1 - y0 + 1,
+                               frame_width,
+                               frame_height);
+          buf_stride = x1 - x0 + 1;
+          buf_ptr = CONVERT_TO_BYTEPTR(xd->mc_buf_high) +
+              y_pad * 3 * buf_stride + x_pad * 3;
+        } else {
+          build_mc_border(buf_ptr1,
+                          pre_buf->stride,
+                          xd->mc_buf,
+                          x1 - x0 + 1,
+                          x0,
+                          y0,
+                          x1 - x0 + 1,
+                          y1 - y0 + 1,
+                          frame_width,
+                          frame_height);
+          buf_stride = x1 - x0 + 1;
+          buf_ptr = xd->mc_buf + y_pad * 3 * buf_stride + x_pad * 3;
+        }
+#else
+        build_mc_border(buf_ptr1,
+                        pre_buf->stride,
+                        xd->mc_buf,
+                        x1 - x0 + 1,
+                        x0,
+                        y0,
+                        x1 - x0 + 1,
+                        y1 - y0 + 1,
+                        frame_width,
                         frame_height);
         buf_stride = x1 - x0 + 1;
         buf_ptr = xd->mc_buf + y_pad * 3 * buf_stride + x_pad * 3;
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       }
     }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      high_inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
+                           subpel_y, sf, w, h, ref, kernel, xs, ys, xd->bd);
+    } else {
+      inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
+                    subpel_y, sf, w, h, ref, kernel, xs, ys);
+    }
+#else
     inter_predictor(buf_ptr, buf_stride, dst, dst_buf->stride, subpel_x,
                     subpel_y, sf, w, h, ref, kernel, xs, ys);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 }
 
@@ -419,7 +561,7 @@
     const int bw = 4 * num_4x4_w;
     const int bh = 4 * num_4x4_h;
 
-    if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
+    if (xd->mi[0].src_mi->mbmi.sb_type < BLOCK_8X8) {
       int i = 0, x, y;
       assert(bsize == BLOCK_8X8);
       for (y = 0; y < num_4x4_h; ++y)
diff --git a/vp9/common/vp9_reconinter.h b/vp9/common/vp9_reconinter.h
index 58c596e..e70cc4c 100644
--- a/vp9/common/vp9_reconinter.h
+++ b/vp9/common/vp9_reconinter.h
@@ -39,6 +39,17 @@
                                enum mv_precision precision,
                                int x, int y);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_build_inter_predictor(const uint8_t *src, int src_stride,
+                                    uint8_t *dst, int dst_stride,
+                                    const MV *mv_q3,
+                                    const struct scale_factors *sf,
+                                    int w, int h, int do_avg,
+                                    const InterpKernel *kernel,
+                                    enum mv_precision precision,
+                                    int x, int y, int bd);
+#endif
+
 static INLINE int scaled_buffer_offset(int x_offset, int y_offset, int stride,
                                        const struct scale_factors *sf) {
   const int x = sf ? sf->scale_value_x(x_offset, sf) : x_offset;
diff --git a/vp9/common/vp9_reconintra.c b/vp9/common/vp9_reconintra.c
index 471929a..7ebd2ea 100644
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -40,11 +40,289 @@
     type##_predictor(dst, stride, size, above, left); \
   }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+#define intra_pred_high_sized(type, size) \
+  void vp9_high_##type##_predictor_##size##x##size##_c( \
+      uint16_t *dst, ptrdiff_t stride, const uint16_t *above, \
+      const uint16_t *left, int bd) { \
+    high_##type##_predictor(dst, stride, size, above, left, bd); \
+  }
+
+#define intra_pred_allsizes(type) \
+  intra_pred_sized(type, 4) \
+  intra_pred_sized(type, 8) \
+  intra_pred_sized(type, 16) \
+  intra_pred_sized(type, 32) \
+  intra_pred_high_sized(type, 4) \
+  intra_pred_high_sized(type, 8) \
+  intra_pred_high_sized(type, 16) \
+  intra_pred_high_sized(type, 32)
+
+#else
+
 #define intra_pred_allsizes(type) \
   intra_pred_sized(type, 4) \
   intra_pred_sized(type, 8) \
   intra_pred_sized(type, 16) \
   intra_pred_sized(type, 32)
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE void high_d207_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  int r, c;
+  (void) above;
+  (void) bd;
+
+  // First column.
+  for (r = 0; r < bs - 1; ++r) {
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+  }
+  dst[(bs - 1) * stride] = left[bs - 1];
+  dst++;
+
+  // Second column.
+  for (r = 0; r < bs - 2; ++r) {
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
+                                         left[r + 2], 2);
+  }
+  dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
+                                              left[bs - 1] * 3, 2);
+  dst[(bs - 1) * stride] = left[bs - 1];
+  dst++;
+
+  // Rest of last row.
+  for (c = 0; c < bs - 2; ++c)
+    dst[(bs - 1) * stride + c] = left[bs - 1];
+
+  for (r = bs - 2; r >= 0; --r) {
+    for (c = 0; c < bs - 2; ++c)
+      dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
+  }
+}
+
+static INLINE void high_d63_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                      const uint16_t *above,
+                                      const uint16_t *left, int bd) {
+  int r, c;
+  (void) left;
+  (void) bd;
+  for (r = 0; r < bs; ++r) {
+    for (c = 0; c < bs; ++c) {
+      dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
+                                          above[r/2 + c + 1] * 2 +
+                                          above[r/2 + c + 2], 2)
+                     : ROUND_POWER_OF_TWO(above[r/2 + c] +
+                                          above[r/2 + c + 1], 1);
+    }
+    dst += stride;
+  }
+}
+
+static INLINE void high_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                      const uint16_t *above,
+                                      const uint16_t *left, int bd) {
+  int r, c;
+  (void) left;
+  (void) bd;
+  for (r = 0; r < bs; ++r) {
+    for (c = 0; c < bs; ++c) {
+      dst[c] = r + c + 2 < bs * 2 ?  ROUND_POWER_OF_TWO(above[r + c] +
+                                                        above[r + c + 1] * 2 +
+                                                        above[r + c + 2], 2)
+                                  : above[bs * 2 - 1];
+    }
+    dst += stride;
+  }
+}
+
+static INLINE void high_d117_predictor(uint16_t *dst, ptrdiff_t stride,
+                                       int bs, const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  int r, c;
+  (void) bd;
+
+  // first row
+  for (c = 0; c < bs; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+  dst += stride;
+
+  // second row
+  dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+  for (c = 1; c < bs; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+  dst += stride;
+
+  // the rest of first col
+  dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+  for (r = 3; r < bs; ++r)
+    dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
+                                               left[r - 1], 2);
+
+  // the rest of the block
+  for (r = 2; r < bs; ++r) {
+    for (c = 1; c < bs; c++)
+      dst[c] = dst[-2 * stride + c - 1];
+    dst += stride;
+  }
+}
+
+static INLINE void high_d135_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  int r, c;
+  (void) bd;
+  dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+  for (c = 1; c < bs; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+
+  dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+  for (r = 2; r < bs; ++r)
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
+                                         left[r], 2);
+
+  dst += stride;
+  for (r = 1; r < bs; ++r) {
+    for (c = 1; c < bs; c++)
+      dst[c] = dst[-stride + c - 1];
+    dst += stride;
+  }
+}
+
+static INLINE void high_d153_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  int r, c;
+  (void) bd;
+  dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+  for (r = 1; r < bs; r++)
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+  dst++;
+
+  dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+  dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+  for (r = 2; r < bs; r++)
+    dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
+                                         left[r], 2);
+  dst++;
+
+  for (c = 0; c < bs - 2; c++)
+    dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
+  dst += stride;
+
+  for (r = 1; r < bs; ++r) {
+    for (c = 0; c < bs - 2; c++)
+      dst[c] = dst[-stride + c - 2];
+    dst += stride;
+  }
+}
+
+static INLINE void high_v_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                    const uint16_t *above,
+                                    const uint16_t *left, int bd) {
+  int r;
+  (void) left;
+  (void) bd;
+  for (r = 0; r < bs; r++) {
+    vpx_memcpy(dst, above, bs * sizeof(uint16_t));
+    dst += stride;
+  }
+}
+
+static INLINE void high_h_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                    const uint16_t *above, const uint16_t *left,
+                                    int bd) {
+  int r;
+  (void) above;
+  (void) bd;
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, left[r], bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+                                     const uint16_t *above,
+                                     const uint16_t *left, int bd) {
+  int r, c;
+  int ytop_left = above[-1];
+  (void) bd;
+
+  for (r = 0; r < bs; r++) {
+    for (c = 0; c < bs; c++)
+      dst[c] = clip_pixel_high(left[r] + above[c] - ytop_left, bd);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
+                                         int bs, const uint16_t *above,
+                                         const uint16_t *left, int bd) {
+  int r;
+  (void) above;
+  (void) left;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, 128 << (bd - 8), bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
+                                          int bs, const uint16_t *above,
+                                          const uint16_t *left, int bd) {
+  int i, r, expected_dc, sum = 0;
+  (void) above;
+  (void) bd;
+
+  for (i = 0; i < bs; i++)
+    sum += left[i];
+  expected_dc = (sum + (bs >> 1)) / bs;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, expected_dc, bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
+                                         int bs, const uint16_t *above,
+                                         const uint16_t *left, int bd) {
+  int i, r, expected_dc, sum = 0;
+  (void) left;
+  (void) bd;
+
+  for (i = 0; i < bs; i++)
+    sum += above[i];
+  expected_dc = (sum + (bs >> 1)) / bs;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, expected_dc, bs);
+    dst += stride;
+  }
+}
+
+static INLINE void high_dc_predictor(uint16_t *dst, ptrdiff_t stride,
+                                     int bs, const uint16_t *above,
+                                     const uint16_t *left, int bd) {
+  int i, r, expected_dc, sum = 0;
+  const int count = 2 * bs;
+  (void) bd;
+
+  for (i = 0; i < bs; i++) {
+    sum += above[i];
+    sum += left[i];
+  }
+
+  expected_dc = (sum + (count >> 1)) / count;
+
+  for (r = 0; r < bs; r++) {
+    vpx_memset16(dst, expected_dc, bs);
+    dst += stride;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
                                   const uint8_t *above, const uint8_t *left) {
@@ -293,6 +571,14 @@
 static intra_pred_fn pred[INTRA_MODES][TX_SIZES];
 static intra_pred_fn dc_pred[2][2][TX_SIZES];
 
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride,
+                                   const uint16_t *above, const uint16_t *left,
+                                   int bd);
+static intra_high_pred_fn pred_high[INTRA_MODES][4];
+static intra_high_pred_fn dc_pred_high[2][2][4];
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 void vp9_init_intra_predictors() {
 #define INIT_ALL_SIZES(p, type) \
   p[TX_4X4] = vp9_##type##_predictor_4x4; \
@@ -315,9 +601,164 @@
   INIT_ALL_SIZES(dc_pred[1][0], dc_left);
   INIT_ALL_SIZES(dc_pred[1][1], dc);
 
-#undef INIT_ALL_SIZES
+#if CONFIG_VP9_HIGHBITDEPTH
+  INIT_ALL_SIZES(pred_high[V_PRED], high_v);
+  INIT_ALL_SIZES(pred_high[H_PRED], high_h);
+  INIT_ALL_SIZES(pred_high[D207_PRED], high_d207);
+  INIT_ALL_SIZES(pred_high[D45_PRED], high_d45);
+  INIT_ALL_SIZES(pred_high[D63_PRED], high_d63);
+  INIT_ALL_SIZES(pred_high[D117_PRED], high_d117);
+  INIT_ALL_SIZES(pred_high[D135_PRED], high_d135);
+  INIT_ALL_SIZES(pred_high[D153_PRED], high_d153);
+  INIT_ALL_SIZES(pred_high[TM_PRED], high_tm);
+
+  INIT_ALL_SIZES(dc_pred_high[0][0], high_dc_128);
+  INIT_ALL_SIZES(dc_pred_high[0][1], high_dc_top);
+  INIT_ALL_SIZES(dc_pred_high[1][0], high_dc_left);
+  INIT_ALL_SIZES(dc_pred_high[1][1], high_dc);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#undef intra_pred_allsizes
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void build_intra_predictors_high(const MACROBLOCKD *xd,
+                                        const uint8_t *ref8,
+                                        int ref_stride,
+                                        uint8_t *dst8,
+                                        int dst_stride,
+                                        PREDICTION_MODE mode,
+                                        TX_SIZE tx_size,
+                                        int up_available,
+                                        int left_available,
+                                        int right_available,
+                                        int x, int y,
+                                        int plane, int bd) {
+  int i;
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, left_col, 64);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, above_data, 128 + 16);
+  uint16_t *above_row = above_data + 16;
+  const uint16_t *const_above_row = above_row;
+  const int bs = 4 << tx_size;
+  int frame_width, frame_height;
+  int x0, y0;
+  const struct macroblockd_plane *const pd = &xd->plane[plane];
+  //  int base=128;
+  int base = 128 << (bd - 8);
+  // 127 127 127 .. 127 127 127 127 127 127
+  // 129  A   B  ..  Y   Z
+  // 129  C   D  ..  W   X
+  // 129  E   F  ..  U   V
+  // 129  G   H  ..  S   T   T   T   T   T
+
+  // Get current frame pointer, width and height.
+  if (plane == 0) {
+    frame_width = xd->cur_buf->y_width;
+    frame_height = xd->cur_buf->y_height;
+  } else {
+    frame_width = xd->cur_buf->uv_width;
+    frame_height = xd->cur_buf->uv_height;
+  }
+
+  // Get block position in current frame.
+  x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
+  y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
+
+  // left
+  if (left_available) {
+    if (xd->mb_to_bottom_edge < 0) {
+      /* slower path if the block needs border extension */
+      if (y0 + bs <= frame_height) {
+        for (i = 0; i < bs; ++i)
+          left_col[i] = ref[i * ref_stride - 1];
+      } else {
+        const int extend_bottom = frame_height - y0;
+        for (i = 0; i < extend_bottom; ++i)
+          left_col[i] = ref[i * ref_stride - 1];
+        for (; i < bs; ++i)
+          left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
+      }
+    } else {
+      /* faster path if the block does not need extension */
+      for (i = 0; i < bs; ++i)
+        left_col[i] = ref[i * ref_stride - 1];
+    }
+  } else {
+    // TODO(Peter): this value should probably change for high bitdepth
+    vpx_memset16(left_col, base + 1, bs);
+  }
+
+  // TODO(hkuang) do not extend 2*bs pixels for all modes.
+  // above
+  if (up_available) {
+    const uint16_t *above_ref = ref - ref_stride;
+    if (xd->mb_to_right_edge < 0) {
+      /* slower path if the block needs border extension */
+      if (x0 + 2 * bs <= frame_width) {
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, 2 * bs * sizeof(uint16_t));
+        } else {
+          vpx_memcpy(above_row, above_ref, bs * sizeof(uint16_t));
+          vpx_memset16(above_row + bs, above_row[bs - 1], bs);
+        }
+      } else if (x0 + bs <= frame_width) {
+        const int r = frame_width - x0;
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, r * sizeof(uint16_t));
+          vpx_memset16(above_row + r, above_row[r - 1],
+                       x0 + 2 * bs - frame_width);
+        } else {
+          vpx_memcpy(above_row, above_ref, bs * sizeof(uint16_t));
+          vpx_memset16(above_row + bs, above_row[bs - 1], bs);
+        }
+      } else if (x0 <= frame_width) {
+        const int r = frame_width - x0;
+        if (right_available && bs == 4) {
+          vpx_memcpy(above_row, above_ref, r * sizeof(uint16_t));
+          vpx_memset16(above_row + r, above_row[r - 1],
+                       x0 + 2 * bs - frame_width);
+        } else {
+          vpx_memcpy(above_row, above_ref, r * sizeof(uint16_t));
+          vpx_memset16(above_row + r, above_row[r - 1],
+                       x0 + 2 * bs - frame_width);
+        }
+      }
+      // TODO(Peter) this value should probably change for high bitdepth
+      above_row[-1] = left_available ? above_ref[-1] : (base+1);
+    } else {
+      /* faster path if the block does not need extension */
+      if (bs == 4 && right_available && left_available) {
+        const_above_row = above_ref;
+      } else {
+        vpx_memcpy(above_row, above_ref, bs * sizeof(uint16_t));
+        if (bs == 4 && right_available)
+          vpx_memcpy(above_row + bs, above_ref + bs, bs * sizeof(uint16_t));
+        else
+          vpx_memset16(above_row + bs, above_row[bs - 1], bs);
+        // TODO(Peter): this value should probably change for high bitdepth
+        above_row[-1] = left_available ? above_ref[-1] : (base+1);
+      }
+    }
+  } else {
+    vpx_memset16(above_row, base - 1, bs * 2);
+    // TODO(Peter): this value should probably change for high bitdepth
+    above_row[-1] = base - 1;
+  }
+
+  // predict
+  if (mode == DC_PRED) {
+    dc_pred_high[left_available][up_available][tx_size](dst, dst_stride,
+                                                        const_above_row,
+                                                        left_col, xd->bd);
+  } else {
+    pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col,
+                             xd->bd);
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
                                    int ref_stride, uint8_t *dst, int dst_stride,
                                    PREDICTION_MODE mode, TX_SIZE tx_size,
@@ -454,6 +895,14 @@
   const int y = loff * 4;
 
   assert(bwl >= 0);
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    build_intra_predictors_high(xd, ref, ref_stride, dst, dst_stride, mode,
+                                tx_size, have_top, have_left, have_right,
+                                x, y, plane, xd->bd);
+    return;
+  }
+#endif
   build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
                          have_top, have_left, have_right, x, y, plane);
 }
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 667e057..0e95141 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -6,6 +6,7 @@
 
 #include "vpx/vpx_integer.h"
 #include "vp9/common/vp9_enums.h"
+#include "vp9/common/vp9_idct.h"
 
 struct macroblockd;
 
@@ -45,6 +46,13 @@
   $avx_x86_64 = $avx2_x86_64 = '';
 }
 
+# optimizations which depend on multiple features
+if ((vpx_config("HAVE_AVX2") eq "yes") && (vpx_config("HAVE_SSSE3") eq "yes")) {
+  $avx2_ssse3 = 'avx2';
+} else {
+  $avx2_ssse3 = '';
+}
+
 #
 # RECON
 #
@@ -296,15 +304,15 @@
 $vp9_convolve_avg_neon_asm=vp9_convolve_avg_neon;
 
 add_proto qw/void vp9_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8 sse2 ssse3 neon_asm dspr2 avx2/;
+specialize qw/vp9_convolve8 sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";
 $vp9_convolve8_neon_asm=vp9_convolve8_neon;
 
 add_proto qw/void vp9_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_horiz sse2 ssse3 neon_asm dspr2 avx2/;
+specialize qw/vp9_convolve8_horiz sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";
 $vp9_convolve8_horiz_neon_asm=vp9_convolve8_horiz_neon;
 
 add_proto qw/void vp9_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_vert sse2 ssse3 neon_asm dspr2 avx2/;
+specialize qw/vp9_convolve8_vert sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";
 $vp9_convolve8_vert_neon_asm=vp9_convolve8_vert_neon;
 
 add_proto qw/void vp9_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
@@ -322,68 +330,415 @@
 #
 # dct
 #
-add_proto qw/void vp9_idct4x4_1_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct4x4_1_add sse2 neon_asm dspr2/;
-$vp9_idct4x4_1_add_neon_asm=vp9_idct4x4_1_add_neon;
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+  add_proto qw/void vp9_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct4x4_1_add/;
 
-add_proto qw/void vp9_idct4x4_16_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct4x4_16_add sse2 neon_asm dspr2/;
-$vp9_idct4x4_16_add_neon_asm=vp9_idct4x4_16_add_neon;
+  add_proto qw/void vp9_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct4x4_16_add/;
 
-add_proto qw/void vp9_idct8x8_1_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct8x8_1_add sse2 neon_asm dspr2/;
-$vp9_idct8x8_1_add_neon_asm=vp9_idct8x8_1_add_neon;
+  add_proto qw/void vp9_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_1_add/;
 
-add_proto qw/void vp9_idct8x8_64_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct8x8_64_add sse2 neon_asm dspr2/, "$ssse3_x86_64";
-$vp9_idct8x8_64_add_neon_asm=vp9_idct8x8_64_add_neon;
+  add_proto qw/void vp9_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_64_add/;
 
-add_proto qw/void vp9_idct8x8_12_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct8x8_12_add sse2 neon_asm dspr2/, "$ssse3_x86_64";
-$vp9_idct8x8_12_add_neon_asm=vp9_idct8x8_12_add_neon;
+  add_proto qw/void vp9_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_12_add/;
 
-add_proto qw/void vp9_idct16x16_1_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct16x16_1_add sse2 neon_asm dspr2/;
-$vp9_idct16x16_1_add_neon_asm=vp9_idct16x16_1_add_neon;
+  add_proto qw/void vp9_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_1_add/;
 
-add_proto qw/void vp9_idct16x16_256_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct16x16_256_add sse2 ssse3 neon_asm dspr2/;
-$vp9_idct16x16_256_add_neon_asm=vp9_idct16x16_256_add_neon;
+  add_proto qw/void vp9_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_256_add/;
 
-add_proto qw/void vp9_idct16x16_10_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct16x16_10_add sse2 ssse3 neon_asm dspr2/;
-$vp9_idct16x16_10_add_neon_asm=vp9_idct16x16_10_add_neon;
+  add_proto qw/void vp9_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_10_add/;
 
-add_proto qw/void vp9_idct32x32_1024_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct32x32_1024_add sse2 neon_asm dspr2/;
-$vp9_idct32x32_1024_add_neon_asm=vp9_idct32x32_1024_add_neon;
+  add_proto qw/void vp9_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_1024_add/;
 
-add_proto qw/void vp9_idct32x32_34_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct32x32_34_add sse2 neon_asm dspr2/;
-$vp9_idct32x32_34_add_neon_asm=vp9_idct32x32_1024_add_neon;
+  add_proto qw/void vp9_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_34_add/;
 
-add_proto qw/void vp9_idct32x32_1_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_idct32x32_1_add sse2 neon_asm dspr2/;
-$vp9_idct32x32_1_add_neon_asm=vp9_idct32x32_1_add_neon;
+  add_proto qw/void vp9_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_1_add/;
 
-add_proto qw/void vp9_iht4x4_16_add/, "const int16_t *input, uint8_t *dest, int dest_stride, int tx_type";
-specialize qw/vp9_iht4x4_16_add sse2 neon_asm dspr2/;
-$vp9_iht4x4_16_add_neon_asm=vp9_iht4x4_16_add_neon;
+  add_proto qw/void vp9_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+  specialize qw/vp9_iht4x4_16_add/;
 
-add_proto qw/void vp9_iht8x8_64_add/, "const int16_t *input, uint8_t *dest, int dest_stride, int tx_type";
-specialize qw/vp9_iht8x8_64_add sse2 neon_asm dspr2/;
-$vp9_iht8x8_64_add_neon_asm=vp9_iht8x8_64_add_neon;
+  add_proto qw/void vp9_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+  specialize qw/vp9_iht8x8_64_add/;
 
-add_proto qw/void vp9_iht16x16_256_add/, "const int16_t *input, uint8_t *output, int pitch, int tx_type";
-specialize qw/vp9_iht16x16_256_add sse2 dspr2/;
+  add_proto qw/void vp9_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
+  specialize qw/vp9_iht16x16_256_add/;
 
-# dct and add
+  # dct and add
 
-add_proto qw/void vp9_iwht4x4_1_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_iwht4x4_1_add/;
+  add_proto qw/void vp9_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_iwht4x4_1_add/;
 
-add_proto qw/void vp9_iwht4x4_16_add/, "const int16_t *input, uint8_t *dest, int dest_stride";
-specialize qw/vp9_iwht4x4_16_add/;
+  add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_iwht4x4_16_add/;
+} else {
+  add_proto qw/void vp9_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct4x4_1_add sse2 neon_asm dspr2/;
+  $vp9_idct4x4_1_add_neon_asm=vp9_idct4x4_1_add_neon;
+
+  add_proto qw/void vp9_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct4x4_16_add sse2 neon_asm dspr2/;
+  $vp9_idct4x4_16_add_neon_asm=vp9_idct4x4_16_add_neon;
+
+  add_proto qw/void vp9_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_1_add sse2 neon_asm dspr2/;
+  $vp9_idct8x8_1_add_neon_asm=vp9_idct8x8_1_add_neon;
+
+  add_proto qw/void vp9_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_64_add sse2 neon_asm dspr2/, "$ssse3_x86_64";
+  $vp9_idct8x8_64_add_neon_asm=vp9_idct8x8_64_add_neon;
+
+  add_proto qw/void vp9_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_12_add sse2 neon_asm dspr2/, "$ssse3_x86_64";
+  $vp9_idct8x8_12_add_neon_asm=vp9_idct8x8_12_add_neon;
+
+  add_proto qw/void vp9_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_1_add sse2 neon_asm dspr2/;
+  $vp9_idct16x16_1_add_neon_asm=vp9_idct16x16_1_add_neon;
+
+  add_proto qw/void vp9_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_256_add sse2 ssse3 neon_asm dspr2/;
+  $vp9_idct16x16_256_add_neon_asm=vp9_idct16x16_256_add_neon;
+
+  add_proto qw/void vp9_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_10_add sse2 ssse3 neon_asm dspr2/;
+  $vp9_idct16x16_10_add_neon_asm=vp9_idct16x16_10_add_neon;
+
+  add_proto qw/void vp9_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_1024_add sse2 neon_asm dspr2/;
+  $vp9_idct32x32_1024_add_neon_asm=vp9_idct32x32_1024_add_neon;
+
+  add_proto qw/void vp9_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_34_add sse2 neon_asm dspr2/;
+  $vp9_idct32x32_34_add_neon_asm=vp9_idct32x32_1024_add_neon;
+
+  add_proto qw/void vp9_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_1_add sse2 neon_asm dspr2/;
+  $vp9_idct32x32_1_add_neon_asm=vp9_idct32x32_1_add_neon;
+
+  add_proto qw/void vp9_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+  specialize qw/vp9_iht4x4_16_add sse2 neon_asm dspr2/;
+  $vp9_iht4x4_16_add_neon_asm=vp9_iht4x4_16_add_neon;
+
+  add_proto qw/void vp9_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+  specialize qw/vp9_iht8x8_64_add sse2 neon_asm dspr2/;
+  $vp9_iht8x8_64_add_neon_asm=vp9_iht8x8_64_add_neon;
+
+  add_proto qw/void vp9_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
+  specialize qw/vp9_iht16x16_256_add sse2 dspr2/;
+
+  # dct and add
+
+  add_proto qw/void vp9_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_iwht4x4_1_add/;
+
+  add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_iwht4x4_16_add/;
+}
+
+# High bitdepth functions
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+  #
+  # Intra prediction
+  #
+  add_proto qw/void vp9_high_d207_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d207_predictor_4x4/;
+
+  add_proto qw/void vp9_high_d45_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d45_predictor_4x4/;
+
+  add_proto qw/void vp9_high_d63_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d63_predictor_4x4/;
+
+  add_proto qw/void vp9_high_h_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_h_predictor_4x4/;
+
+  add_proto qw/void vp9_high_d117_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d117_predictor_4x4/;
+
+  add_proto qw/void vp9_high_d135_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d135_predictor_4x4/;
+
+  add_proto qw/void vp9_high_d153_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d153_predictor_4x4/;
+
+  add_proto qw/void vp9_high_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_v_predictor_4x4 neon/, "$sse_x86inc";
+
+  add_proto qw/void vp9_high_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_tm_predictor_4x4/, "$sse_x86inc";
+
+  add_proto qw/void vp9_high_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_predictor_4x4/, "$sse_x86inc";
+
+  add_proto qw/void vp9_high_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_top_predictor_4x4/;
+
+  add_proto qw/void vp9_high_dc_left_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_left_predictor_4x4/;
+
+  add_proto qw/void vp9_high_dc_128_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_128_predictor_4x4/;
+
+  add_proto qw/void vp9_high_d207_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d207_predictor_8x8/;
+
+  add_proto qw/void vp9_high_d45_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d45_predictor_8x8/;
+
+  add_proto qw/void vp9_high_d63_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d63_predictor_8x8/;
+
+  add_proto qw/void vp9_high_h_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_h_predictor_8x8/;
+
+  add_proto qw/void vp9_high_d117_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d117_predictor_8x8/;
+
+  add_proto qw/void vp9_high_d135_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d135_predictor_8x8/;
+
+  add_proto qw/void vp9_high_d153_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d153_predictor_8x8/;
+
+  add_proto qw/void vp9_high_v_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_v_predictor_8x8/, "$sse2_x86inc";
+
+  add_proto qw/void vp9_high_tm_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_tm_predictor_8x8/, "$sse2_x86inc";
+
+  add_proto qw/void vp9_high_dc_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_predictor_8x8/, "$sse2_x86inc";;
+
+  add_proto qw/void vp9_high_dc_top_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_top_predictor_8x8/;
+
+  add_proto qw/void vp9_high_dc_left_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_left_predictor_8x8/;
+
+  add_proto qw/void vp9_high_dc_128_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_128_predictor_8x8/;
+
+  add_proto qw/void vp9_high_d207_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d207_predictor_16x16/;
+
+  add_proto qw/void vp9_high_d45_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d45_predictor_16x16/;
+
+  add_proto qw/void vp9_high_d63_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d63_predictor_16x16/;
+
+  add_proto qw/void vp9_high_h_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_h_predictor_16x16/;
+
+  add_proto qw/void vp9_high_d117_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d117_predictor_16x16/;
+
+  add_proto qw/void vp9_high_d135_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d135_predictor_16x16/;
+
+  add_proto qw/void vp9_high_d153_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d153_predictor_16x16/;
+
+  add_proto qw/void vp9_high_v_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_v_predictor_16x16 neon/, "$sse2_x86inc";
+
+  add_proto qw/void vp9_high_tm_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_tm_predictor_16x16/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_dc_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_predictor_16x16/, "$sse2_x86inc";
+
+  add_proto qw/void vp9_high_dc_top_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_top_predictor_16x16/;
+
+  add_proto qw/void vp9_high_dc_left_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_left_predictor_16x16/;
+
+  add_proto qw/void vp9_high_dc_128_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_128_predictor_16x16/;
+
+  add_proto qw/void vp9_high_d207_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d207_predictor_32x32/;
+
+  add_proto qw/void vp9_high_d45_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d45_predictor_32x32/;
+
+  add_proto qw/void vp9_high_d63_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d63_predictor_32x32/;
+
+  add_proto qw/void vp9_high_h_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_h_predictor_32x32/;
+
+  add_proto qw/void vp9_high_d117_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d117_predictor_32x32/;
+
+  add_proto qw/void vp9_high_d135_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d135_predictor_32x32/;
+
+  add_proto qw/void vp9_high_d153_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_d153_predictor_32x32/;
+
+  add_proto qw/void vp9_high_v_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_v_predictor_32x32/, "$sse2_x86inc";
+
+  add_proto qw/void vp9_high_tm_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_tm_predictor_32x32/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_predictor_32x32/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_dc_top_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_top_predictor_32x32/;
+
+  add_proto qw/void vp9_high_dc_left_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_left_predictor_32x32/;
+
+  add_proto qw/void vp9_high_dc_128_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vp9_high_dc_128_predictor_32x32/;
+
+  #
+  # Sub Pixel Filters
+  #
+  add_proto qw/void vp9_high_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve_copy/;
+
+  add_proto qw/void vp9_high_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve_avg/;
+
+  add_proto qw/void vp9_high_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_horiz/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_vert/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_avg/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_avg_horiz/, "$sse2_x86_64";
+
+  add_proto qw/void vp9_high_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+  specialize qw/vp9_high_convolve8_avg_vert/, "$sse2_x86_64";
+
+  #
+  # Loopfilter
+  #
+  add_proto qw/void vp9_highbd_lpf_vertical_16/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
+  specialize qw/vp9_highbd_lpf_vertical_16 sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_vertical_16_dual/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
+  specialize qw/vp9_highbd_lpf_vertical_16_dual sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_vertical_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+  specialize qw/vp9_highbd_lpf_vertical_8 sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_vertical_8_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
+  specialize qw/vp9_highbd_lpf_vertical_8_dual sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_vertical_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+  specialize qw/vp9_highbd_lpf_vertical_4 sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_vertical_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
+  specialize qw/vp9_highbd_lpf_vertical_4_dual sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_horizontal_16/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+  specialize qw/vp9_highbd_lpf_horizontal_16 sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_horizontal_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+  specialize qw/vp9_highbd_lpf_horizontal_8 sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_horizontal_8_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
+  specialize qw/vp9_highbd_lpf_horizontal_8_dual sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_horizontal_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
+  specialize qw/vp9_highbd_lpf_horizontal_4 sse2/;
+
+  add_proto qw/void vp9_highbd_lpf_horizontal_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
+  specialize qw/vp9_highbd_lpf_horizontal_4_dual sse2/;
+
+  #
+  # post proc
+  #
+  if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
+    add_proto qw/void vp9_highbd_mbpost_proc_down/, "uint16_t *dst, int pitch, int rows, int cols, int flimit";
+    specialize qw/vp9_highbd_mbpost_proc_down/;
+
+    add_proto qw/void vp9_highbd_mbpost_proc_across_ip/, "uint16_t *src, int pitch, int rows, int cols, int flimit";
+    specialize qw/vp9_highbd_mbpost_proc_across_ip/;
+
+    add_proto qw/void vp9_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
+    specialize qw/vp9_highbd_post_proc_down_and_across/;
+
+    add_proto qw/void vp9_highbd_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
+    specialize qw/vp9_highbd_plane_add_noise/;
+  }
+
+  #
+  # dct
+  #
+  add_proto qw/void vp9_high_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct4x4_1_add/;
+
+  add_proto qw/void vp9_high_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct4x4_16_add/;
+
+  add_proto qw/void vp9_high_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct8x8_1_add/;
+
+  add_proto qw/void vp9_high_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct8x8_64_add/;
+
+  add_proto qw/void vp9_high_idct8x8_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct8x8_10_add/;
+
+  add_proto qw/void vp9_high_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct16x16_1_add/;
+
+  add_proto qw/void vp9_high_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct16x16_256_add/;
+
+  add_proto qw/void vp9_high_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct16x16_10_add/;
+
+  add_proto qw/void vp9_high_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct32x32_1024_add/;
+
+  add_proto qw/void vp9_high_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct32x32_34_add/;
+
+  add_proto qw/void vp9_high_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_idct32x32_1_add/;
+
+  add_proto qw/void vp9_high_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
+  specialize qw/vp9_high_iht4x4_16_add/;
+
+  add_proto qw/void vp9_high_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
+  specialize qw/vp9_high_iht8x8_64_add/;
+
+  add_proto qw/void vp9_high_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type, int bd";
+  specialize qw/vp9_high_iht16x16_256_add/;
+
+  # dct and add
+
+  add_proto qw/void vp9_high_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_iwht4x4_1_add/;
+
+  add_proto qw/void vp9_high_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_high_iwht4x4_16_add/;
+}
 
 #
 # Encoder functions below this point.
@@ -699,23 +1054,42 @@
 specialize qw/vp9_get_mb_ss/, "$sse2_x86inc";
 # ENCODEMB INVOKE
 
-add_proto qw/int64_t vp9_block_error/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size, int64_t *ssz";
-specialize qw/vp9_block_error avx2/, "$sse2_x86inc";
-
 add_proto qw/void vp9_subtract_block/, "int rows, int cols, int16_t *diff_ptr, ptrdiff_t diff_stride, const uint8_t *src_ptr, ptrdiff_t src_stride, const uint8_t *pred_ptr, ptrdiff_t pred_stride";
 specialize qw/vp9_subtract_block neon/, "$sse2_x86inc";
 
-add_proto qw/void vp9_quantize_fp/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-specialize qw/vp9_quantize_fp neon/, "$ssse3_x86_64";
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+# the transform coefficients are held in 32-bit
+# values, so the assembler code for  vp9_block_error can no longer be used.
+  add_proto qw/int64_t vp9_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
+  specialize qw/vp9_block_error/;
 
-add_proto qw/void vp9_quantize_fp_32x32/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-specialize qw/vp9_quantize_fp_32x32/, "$ssse3_x86_64";
+  add_proto qw/void vp9_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_fp/;
 
-add_proto qw/void vp9_quantize_b/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-specialize qw/vp9_quantize_b/, "$ssse3_x86_64";
+  add_proto qw/void vp9_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_fp_32x32/;
 
-add_proto qw/void vp9_quantize_b_32x32/, "const int16_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-specialize qw/vp9_quantize_b_32x32/, "$ssse3_x86_64";
+  add_proto qw/void vp9_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_b/;
+
+  add_proto qw/void vp9_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_b_32x32/;
+} else {
+  add_proto qw/int64_t vp9_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
+  specialize qw/vp9_block_error avx2/, "$sse2_x86inc";
+
+  add_proto qw/void vp9_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_fp neon/, "$ssse3_x86_64";
+
+  add_proto qw/void vp9_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_fp_32x32/, "$ssse3_x86_64";
+
+  add_proto qw/void vp9_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_b/, "$ssse3_x86_64";
+
+  add_proto qw/void vp9_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_quantize_b_32x32/, "$ssse3_x86_64";
+}
 
 #
 # Structured Similarity (SSIM)
@@ -729,44 +1103,86 @@
 }
 
 # fdct functions
-add_proto qw/void vp9_fht4x4/, "const int16_t *input, int16_t *output, int stride, int tx_type";
-specialize qw/vp9_fht4x4 sse2/;
 
-add_proto qw/void vp9_fht8x8/, "const int16_t *input, int16_t *output, int stride, int tx_type";
-specialize qw/vp9_fht8x8 sse2/;
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+  add_proto qw/void vp9_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_fht4x4/;
 
-add_proto qw/void vp9_fht16x16/, "const int16_t *input, int16_t *output, int stride, int tx_type";
-specialize qw/vp9_fht16x16 sse2/;
+  add_proto qw/void vp9_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_fht8x8/;
 
-add_proto qw/void vp9_fwht4x4/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fwht4x4/, "$mmx_x86inc";
+  add_proto qw/void vp9_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_fht16x16/;
 
-add_proto qw/void vp9_fdct4x4_1/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct4x4_1 sse2/;
+  add_proto qw/void vp9_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fwht4x4/;
 
-add_proto qw/void vp9_fdct4x4/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct4x4 sse2/;
+  add_proto qw/void vp9_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct4x4_1/;
 
-add_proto qw/void vp9_fdct8x8_1/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct8x8_1 sse2 neon/;
+  add_proto qw/void vp9_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct4x4/;
 
-add_proto qw/void vp9_fdct8x8/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct8x8 sse2 neon/, "$ssse3_x86_64";
+  add_proto qw/void vp9_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct8x8_1/;
 
-add_proto qw/void vp9_fdct16x16_1/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct16x16_1 sse2/;
+  add_proto qw/void vp9_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct8x8/;
 
-add_proto qw/void vp9_fdct16x16/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct16x16 sse2/;
+  add_proto qw/void vp9_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct16x16_1/;
 
-add_proto qw/void vp9_fdct32x32_1/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct32x32_1 sse2/;
+  add_proto qw/void vp9_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct16x16/;
 
-add_proto qw/void vp9_fdct32x32/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct32x32 sse2 avx2/;
+  add_proto qw/void vp9_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct32x32_1/;
 
-add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct32x32_rd sse2 avx2/;
+  add_proto qw/void vp9_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct32x32/;
+
+  add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct32x32_rd/;
+} else {
+  add_proto qw/void vp9_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_fht4x4 sse2/;
+
+  add_proto qw/void vp9_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_fht8x8 sse2/;
+
+  add_proto qw/void vp9_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_fht16x16 sse2/;
+
+  add_proto qw/void vp9_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fwht4x4/, "$mmx_x86inc";
+
+  add_proto qw/void vp9_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct4x4_1 sse2/;
+
+  add_proto qw/void vp9_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct4x4 sse2/;
+
+  add_proto qw/void vp9_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct8x8_1 sse2 neon/;
+
+  add_proto qw/void vp9_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct8x8 sse2 neon/, "$ssse3_x86_64";
+
+  add_proto qw/void vp9_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct16x16_1 sse2/;
+
+  add_proto qw/void vp9_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct16x16 sse2/;
+
+  add_proto qw/void vp9_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct32x32_1 sse2/;
+
+  add_proto qw/void vp9_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct32x32 sse2 avx2/;
+
+  add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_fdct32x32_rd sse2 avx2/;
+}
 
 #
 # Motion search
@@ -788,6 +1204,654 @@
 add_proto qw/void vp9_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
 specialize qw/vp9_temporal_filter_apply sse2/;
 
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+
+  # variance
+  add_proto qw/unsigned int vp9_high_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_variance4x4/;
+
+  add_proto qw/void vp9_high_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
+  specialize qw/vp9_high_get8x8var/;
+
+  add_proto qw/void vp9_high_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
+  specialize qw/vp9_high_get16x16var/;
+
+  add_proto qw/unsigned int vp9_high_10_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_10_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_10_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_10_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_10_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_10_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_10_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_10_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_10_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_10_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_10_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_10_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_10_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_variance4x4/;
+
+  add_proto qw/void vp9_high_10_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
+  specialize qw/vp9_high_10_get8x8var/;
+
+  add_proto qw/void vp9_high_10_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
+  specialize qw/vp9_high_10_get16x16var/;
+
+  add_proto qw/unsigned int vp9_high_12_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_12_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_12_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_12_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_12_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_12_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_12_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_12_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_12_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_12_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_12_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_12_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_12_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_variance4x4/;
+
+  add_proto qw/void vp9_high_12_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
+  specialize qw/vp9_high_12_get8x8var/;
+
+  add_proto qw/void vp9_high_12_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
+  specialize qw/vp9_high_12_get16x16var/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_sub_pixel_variance4x4/;
+
+  add_proto qw/unsigned int vp9_high_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_sub_pixel_avg_variance4x4/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_sub_pixel_variance4x4/;
+
+  add_proto qw/unsigned int vp9_high_10_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_10_sub_pixel_avg_variance4x4/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance64x64/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance32x64/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance64x32/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance32x16/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance16x32/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance32x32/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance16x16/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance8x16/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance16x8/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance8x8/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance8x4/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance4x8/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_sub_pixel_variance4x4/;
+
+  add_proto qw/unsigned int vp9_high_12_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, const uint8_t *second_pred";
+  specialize qw/vp9_high_12_sub_pixel_avg_variance4x4/;
+
+  add_proto qw/unsigned int vp9_high_sad64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad64x64/;
+
+  add_proto qw/unsigned int vp9_high_sad32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vp9_high_sad32x64/;
+
+  add_proto qw/unsigned int vp9_high_sad64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vp9_high_sad64x32/;
+
+  add_proto qw/unsigned int vp9_high_sad32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vp9_high_sad32x16/;
+
+  add_proto qw/unsigned int vp9_high_sad16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vp9_high_sad16x32/;
+
+  add_proto qw/unsigned int vp9_high_sad32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad32x32/;
+
+  add_proto qw/unsigned int vp9_high_sad16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad16x16/;
+
+  add_proto qw/unsigned int vp9_high_sad16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad16x8/;
+
+  add_proto qw/unsigned int vp9_high_sad8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad8x16/;
+
+  add_proto qw/unsigned int vp9_high_sad8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad8x8/;
+
+  add_proto qw/unsigned int vp9_high_sad8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vp9_high_sad8x4/;
+
+  add_proto qw/unsigned int vp9_high_sad4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vp9_high_sad4x8/;
+
+  add_proto qw/unsigned int vp9_high_sad4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride";
+  specialize qw/vp9_high_sad4x4/;
+
+  add_proto qw/unsigned int vp9_high_sad64x64_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad64x64_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad32x64_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad32x64_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad64x32_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad64x32_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad32x16_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad32x16_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad16x32_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad16x32_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad32x32_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad32x32_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad16x16_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad16x16_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad16x8_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad16x8_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad8x16_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad8x16_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad8x8_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad8x8_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad8x4_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad8x4_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad4x8_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad4x8_avg/;
+
+  add_proto qw/unsigned int vp9_high_sad4x4_avg/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, const uint8_t *second_pred";
+  specialize qw/vp9_high_sad4x4_avg/;
+
+  add_proto qw/void vp9_high_sad64x64x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad64x64x3/;
+
+  add_proto qw/void vp9_high_sad32x32x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad32x32x3/;
+
+  add_proto qw/void vp9_high_sad16x16x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad16x16x3/;
+
+  add_proto qw/void vp9_high_sad16x8x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad16x8x3/;
+
+  add_proto qw/void vp9_high_sad8x16x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad8x16x3/;
+
+  add_proto qw/void vp9_high_sad8x8x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad8x8x3/;
+
+  add_proto qw/void vp9_high_sad4x4x3/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad4x4x3/;
+
+  add_proto qw/void vp9_high_sad64x64x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad64x64x8/;
+
+  add_proto qw/void vp9_high_sad32x32x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad32x32x8/;
+
+  add_proto qw/void vp9_high_sad16x16x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad16x16x8/;
+
+  add_proto qw/void vp9_high_sad16x8x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad16x8x8/;
+
+  add_proto qw/void vp9_high_sad8x16x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad8x16x8/;
+
+  add_proto qw/void vp9_high_sad8x8x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad8x8x8/;
+
+  add_proto qw/void vp9_high_sad8x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad8x4x8/;
+
+  add_proto qw/void vp9_high_sad4x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad4x8x8/;
+
+  add_proto qw/void vp9_high_sad4x4x8/, "const uint8_t *src_ptr, int  src_stride, const uint8_t *ref_ptr, int  ref_stride, uint32_t *sad_array";
+  specialize qw/vp9_high_sad4x4x8/;
+
+  add_proto qw/void vp9_high_sad64x64x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad64x64x4d/;
+
+  add_proto qw/void vp9_high_sad32x64x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad32x64x4d/;
+
+  add_proto qw/void vp9_high_sad64x32x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad64x32x4d/;
+
+  add_proto qw/void vp9_high_sad32x16x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad32x16x4d/;
+
+  add_proto qw/void vp9_high_sad16x32x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad16x32x4d/;
+
+  add_proto qw/void vp9_high_sad32x32x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad32x32x4d/;
+
+  add_proto qw/void vp9_high_sad16x16x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad16x16x4d/;
+
+  add_proto qw/void vp9_high_sad16x8x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad16x8x4d/;
+
+  add_proto qw/void vp9_high_sad8x16x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad8x16x4d/;
+
+  add_proto qw/void vp9_high_sad8x8x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad8x8x4d/;
+
+  # TODO(jingning): need to convert these 4x8/8x4 functions into sse2 form
+  add_proto qw/void vp9_high_sad8x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad8x4x4d/;
+
+  add_proto qw/void vp9_high_sad4x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad4x8x4d/;
+
+  add_proto qw/void vp9_high_sad4x4x4d/, "const uint8_t *src_ptr, int  src_stride, const uint8_t* const ref_ptr[], int  ref_stride, unsigned int *sad_array";
+  specialize qw/vp9_high_sad4x4x4d/;
+
+  add_proto qw/unsigned int vp9_high_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_mse16x16/;
+
+  add_proto qw/unsigned int vp9_high_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_mse8x16/;
+
+  add_proto qw/unsigned int vp9_high_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_mse16x8/;
+
+  add_proto qw/unsigned int vp9_high_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_mse8x8/;
+
+  add_proto qw/unsigned int vp9_high_10_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_mse16x16/;
+
+  add_proto qw/unsigned int vp9_high_10_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_mse8x16/;
+
+  add_proto qw/unsigned int vp9_high_10_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_mse16x8/;
+
+  add_proto qw/unsigned int vp9_high_10_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_10_mse8x8/;
+
+  add_proto qw/unsigned int vp9_high_12_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_mse16x16/;
+
+  add_proto qw/unsigned int vp9_high_12_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_mse8x16/;
+
+  add_proto qw/unsigned int vp9_high_12_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_mse16x8/;
+
+  add_proto qw/unsigned int vp9_high_12_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
+  specialize qw/vp9_high_12_mse8x8/;
+
+  # ENCODEMB INVOKE
+
+  add_proto qw/int64_t vp9_high_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
+  specialize qw/vp9_high_block_error/;
+
+  add_proto qw/void vp9_high_subtract_block/, "int rows, int cols, int16_t *diff_ptr, ptrdiff_t diff_stride, const uint8_t *src_ptr, ptrdiff_t src_stride, const uint8_t *pred_ptr, ptrdiff_t pred_stride, int bd";
+  specialize qw/vp9_high_subtract_block/;
+
+  add_proto qw/void vp9_high_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_high_quantize_fp/;
+
+  add_proto qw/void vp9_high_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_high_quantize_fp_32x32/;
+
+  add_proto qw/void vp9_high_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_high_quantize_b/;
+
+  add_proto qw/void vp9_high_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, int zbin_oq_value, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/vp9_high_quantize_b_32x32/;
+
+  #
+  # Structured Similarity (SSIM)
+  #
+  if (vpx_config("CONFIG_INTERNAL_STATS") eq "yes") {
+    add_proto qw/void vp9_high_ssim_parms_8x8/, "uint16_t *s, int sp, uint16_t *r, int rp, uint32_t *sum_s, uint32_t *sum_r, uint32_t *sum_sq_s, uint32_t *sum_sq_r, uint32_t *sum_sxr";
+    specialize qw/vp9_high_ssim_parms_8x8/;
+
+    add_proto qw/void vp9_high_ssim_parms_8x8_shift/, "uint16_t *s, int sp, uint16_t *r, int rp, uint32_t *sum_s, uint32_t *sum_r, uint32_t *sum_sq_s, uint32_t *sum_sq_r, uint32_t *sum_sxr, unsigned int bd, unsigned int shift";
+    specialize qw/vp9_high_ssim_parms_8x8_shift/;
+  }
+
+  # fdct functions
+  add_proto qw/void vp9_high_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_high_fht4x4/;
+
+  add_proto qw/void vp9_high_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_high_fht8x8/;
+
+  add_proto qw/void vp9_high_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
+  specialize qw/vp9_high_fht16x16/;
+
+  add_proto qw/void vp9_high_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fwht4x4/;
+
+  add_proto qw/void vp9_high_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct4x4/;
+
+  add_proto qw/void vp9_high_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct8x8_1/;
+
+  add_proto qw/void vp9_high_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct8x8/;
+
+  add_proto qw/void vp9_high_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct16x16_1/;
+
+  add_proto qw/void vp9_high_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct16x16/;
+
+  add_proto qw/void vp9_high_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct32x32_1/;
+
+  add_proto qw/void vp9_high_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct32x32/;
+
+  add_proto qw/void vp9_high_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp9_high_fdct32x32_rd/;
+
+  add_proto qw/void vp9_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
+  specialize qw/vp9_highbd_temporal_filter_apply/;
+
+}
+# End vp9_high encoder functions
+
 }
 # end encoder functions
 1;
diff --git a/vp9/common/vp9_scale.c b/vp9/common/vp9_scale.c
index 2f58323..63e2b53 100644
--- a/vp9/common/vp9_scale.c
+++ b/vp9/common/vp9_scale.c
@@ -43,9 +43,16 @@
   return res;
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
+                                       int other_w, int other_h,
+                                       int this_w, int this_h,
+                                       int use_high) {
+#else
 void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
                                        int other_w, int other_h,
                                        int this_w, int this_h) {
+#endif
   if (!valid_ref_frame_size(other_w, other_h, this_w, this_h)) {
     sf->x_scale_fp = REF_INVALID_SCALE;
     sf->y_scale_fp = REF_INVALID_SCALE;
@@ -111,4 +118,48 @@
   // 2D subpel motion always gets filtered in both directions
   sf->predict[1][1][0] = vp9_convolve8;
   sf->predict[1][1][1] = vp9_convolve8_avg;
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (use_high) {
+    if (sf->x_step_q4 == 16) {
+      if (sf->y_step_q4 == 16) {
+        // No scaling in either direction.
+        sf->high_predict[0][0][0] = vp9_high_convolve_copy;
+        sf->high_predict[0][0][1] = vp9_high_convolve_avg;
+        sf->high_predict[0][1][0] = vp9_high_convolve8_vert;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg_vert;
+        sf->high_predict[1][0][0] = vp9_high_convolve8_horiz;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg_horiz;
+      } else {
+        // No scaling in x direction. Must always scale in the y direction.
+        sf->high_predict[0][0][0] = vp9_high_convolve8_vert;
+        sf->high_predict[0][0][1] = vp9_high_convolve8_avg_vert;
+        sf->high_predict[0][1][0] = vp9_high_convolve8_vert;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg_vert;
+        sf->high_predict[1][0][0] = vp9_high_convolve8;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg;
+      }
+    } else {
+      if (sf->y_step_q4 == 16) {
+        // No scaling in the y direction. Must always scale in the x direction.
+        sf->high_predict[0][0][0] = vp9_high_convolve8_horiz;
+        sf->high_predict[0][0][1] = vp9_high_convolve8_avg_horiz;
+        sf->high_predict[0][1][0] = vp9_high_convolve8;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg;
+        sf->high_predict[1][0][0] = vp9_high_convolve8_horiz;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg_horiz;
+      } else {
+        // Must always scale in both directions.
+        sf->high_predict[0][0][0] = vp9_high_convolve8;
+        sf->high_predict[0][0][1] = vp9_high_convolve8_avg;
+        sf->high_predict[0][1][0] = vp9_high_convolve8;
+        sf->high_predict[0][1][1] = vp9_high_convolve8_avg;
+        sf->high_predict[1][0][0] = vp9_high_convolve8;
+        sf->high_predict[1][0][1] = vp9_high_convolve8_avg;
+      }
+    }
+    // 2D subpel motion always gets filtered in both directions.
+    sf->high_predict[1][1][0] = vp9_high_convolve8;
+    sf->high_predict[1][1][1] = vp9_high_convolve8_avg;
+  }
+#endif
 }
diff --git a/vp9/common/vp9_scale.h b/vp9/common/vp9_scale.h
index ad6f5d7..2e923db 100644
--- a/vp9/common/vp9_scale.h
+++ b/vp9/common/vp9_scale.h
@@ -32,13 +32,23 @@
   int (*scale_value_y)(int val, const struct scale_factors *sf);
 
   convolve_fn_t predict[2][2][2];  // horiz, vert, avg
+#if CONFIG_VP9_HIGHBITDEPTH
+  high_convolve_fn_t high_predict[2][2][2];  // horiz, vert, avg
+#endif
 };
 
 MV32 vp9_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
+                                       int other_w, int other_h,
+                                       int this_w, int this_h,
+                                       int use_high);
+#else
 void vp9_setup_scale_factors_for_frame(struct scale_factors *sf,
                                        int other_w, int other_h,
                                        int this_w, int this_h);
+#endif
 
 static INLINE int vp9_is_valid_scale(const struct scale_factors *sf) {
   return sf->x_scale_fp != REF_INVALID_SCALE &&
diff --git a/vp9/common/x86/vp9_asm_stubs.c b/vp9/common/x86/vp9_asm_stubs.c
index 1b4904c..407573a 100644
--- a/vp9/common/x86/vp9_asm_stubs.c
+++ b/vp9/common/x86/vp9_asm_stubs.c
@@ -139,25 +139,172 @@
                            filter_x, x_step_q4, filter_y, y_step_q4, w, h); \
   } \
 }
-#if HAVE_AVX2
+
+#if CONFIG_VP9_HIGHBITDEPTH
+
+typedef void high_filter8_1dfunction (
+  const uint16_t *src_ptr,
+  const ptrdiff_t src_pitch,
+  uint16_t *output_ptr,
+  ptrdiff_t out_pitch,
+  unsigned int output_height,
+  const int16_t *filter,
+  int bd
+);
+
+#define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
+  void vp9_high_convolve8_##name##_##opt(const uint8_t *src8, \
+                                         ptrdiff_t src_stride, \
+                                         uint8_t *dst8, ptrdiff_t dst_stride, \
+                                         const int16_t *filter_x, \
+                                         int x_step_q4, \
+                                         const int16_t *filter_y, \
+                                         int y_step_q4, \
+                                         int w, int h, int bd) { \
+  if (step_q4 == 16 && filter[3] != 128) { \
+    uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
+    uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
+    if (filter[0] || filter[1] || filter[2]) { \
+      while (w >= 16) { \
+        vp9_high_filter_block1d16_##dir##8_##avg##opt(src_start, \
+                                                      src_stride, \
+                                                      dst, \
+                                                      dst_stride, \
+                                                      h, \
+                                                      filter, \
+                                                      bd); \
+        src += 16; \
+        dst += 16; \
+        w -= 16; \
+      } \
+      while (w >= 8) { \
+        vp9_high_filter_block1d8_##dir##8_##avg##opt(src_start, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 8; \
+        dst += 8; \
+        w -= 8; \
+      } \
+      while (w >= 4) { \
+        vp9_high_filter_block1d4_##dir##8_##avg##opt(src_start, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 4; \
+        dst += 4; \
+        w -= 4; \
+      } \
+    } else { \
+      while (w >= 16) { \
+        vp9_high_filter_block1d16_##dir##2_##avg##opt(src, \
+                                                      src_stride, \
+                                                      dst, \
+                                                      dst_stride, \
+                                                      h, \
+                                                      filter, \
+                                                      bd); \
+        src += 16; \
+        dst += 16; \
+        w -= 16; \
+      } \
+      while (w >= 8) { \
+        vp9_high_filter_block1d8_##dir##2_##avg##opt(src, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 8; \
+        dst += 8; \
+        w -= 8; \
+      } \
+      while (w >= 4) { \
+        vp9_high_filter_block1d4_##dir##2_##avg##opt(src, \
+                                                     src_stride, \
+                                                     dst, \
+                                                     dst_stride, \
+                                                     h, \
+                                                     filter, \
+                                                     bd); \
+        src += 4; \
+        dst += 4; \
+        w -= 4; \
+      } \
+    } \
+  } \
+  if (w) { \
+    vp9_high_convolve8_##name##_c(src8, src_stride, dst8, dst_stride, \
+                                  filter_x, x_step_q4, filter_y, y_step_q4, \
+                                  w, h, bd); \
+  } \
+}
+
+#define HIGH_FUN_CONV_2D(avg, opt) \
+void vp9_high_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
+                                   uint8_t *dst, ptrdiff_t dst_stride, \
+                                   const int16_t *filter_x, int x_step_q4, \
+                                   const int16_t *filter_y, int y_step_q4, \
+                                   int w, int h, int bd) { \
+  assert(w <= 64); \
+  assert(h <= 64); \
+  if (x_step_q4 == 16 && y_step_q4 == 16) { \
+    if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
+        filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, fdata2, 64 * 71); \
+      vp9_high_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \
+                                     CONVERT_TO_BYTEPTR(fdata2), 64, \
+                                     filter_x, x_step_q4, filter_y, y_step_q4, \
+                                     w, h + 7, bd); \
+      vp9_high_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2) + 192, \
+                                           64, dst, dst_stride, \
+                                           filter_x, x_step_q4, filter_y, \
+                                           y_step_q4, w, h, bd); \
+    } else { \
+      DECLARE_ALIGNED_ARRAY(16, uint16_t, fdata2, 64 * 65); \
+      vp9_high_convolve8_horiz_##opt(src, src_stride, \
+                                     CONVERT_TO_BYTEPTR(fdata2), 64, \
+                                     filter_x, x_step_q4, filter_y, y_step_q4, \
+                                     w, h + 1, bd); \
+      vp9_high_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2), 64, \
+                                           dst, dst_stride, \
+                                           filter_x, x_step_q4, filter_y, \
+                                           y_step_q4, w, h, bd); \
+    } \
+  } else { \
+    vp9_high_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
+                                filter_x, x_step_q4, filter_y, y_step_q4, w, \
+                                h, bd); \
+  } \
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_AVX2 && HAVE_SSSE3
 filter8_1dfunction vp9_filter_block1d16_v8_avx2;
 filter8_1dfunction vp9_filter_block1d16_h8_avx2;
 filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-#if (ARCH_X86_64)
+#if ARCH_X86_64
 filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
 filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
 filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
 #define vp9_filter_block1d8_v8_avx2 vp9_filter_block1d8_v8_intrin_ssse3
 #define vp9_filter_block1d8_h8_avx2 vp9_filter_block1d8_h8_intrin_ssse3
 #define vp9_filter_block1d4_h8_avx2 vp9_filter_block1d4_h8_intrin_ssse3
-#else
+#else  // ARCH_X86
 filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
 filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
 filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
 #define vp9_filter_block1d8_v8_avx2 vp9_filter_block1d8_v8_ssse3
 #define vp9_filter_block1d8_h8_avx2 vp9_filter_block1d8_h8_ssse3
 #define vp9_filter_block1d4_h8_avx2 vp9_filter_block1d4_h8_ssse3
-#endif
+#endif  // ARCH_X86_64 / ARCH_X86
 filter8_1dfunction vp9_filter_block1d16_v2_ssse3;
 filter8_1dfunction vp9_filter_block1d16_h2_ssse3;
 filter8_1dfunction vp9_filter_block1d8_v2_ssse3;
@@ -190,9 +337,9 @@
 //                          const int16_t *filter_y, int y_step_q4,
 //                          int w, int h);
 FUN_CONV_2D(, avx2);
-#endif
+#endif  // HAVE_AX2 && HAVE_SSSE3
 #if HAVE_SSSE3
-#if (ARCH_X86_64)
+#if ARCH_X86_64
 filter8_1dfunction vp9_filter_block1d16_v8_intrin_ssse3;
 filter8_1dfunction vp9_filter_block1d16_h8_intrin_ssse3;
 filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
@@ -204,14 +351,14 @@
 #define vp9_filter_block1d8_v8_ssse3 vp9_filter_block1d8_v8_intrin_ssse3
 #define vp9_filter_block1d8_h8_ssse3 vp9_filter_block1d8_h8_intrin_ssse3
 #define vp9_filter_block1d4_h8_ssse3 vp9_filter_block1d4_h8_intrin_ssse3
-#else
+#else  // ARCH_X86
 filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
 filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
 filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
 filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
 filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
 filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
-#endif
+#endif  // ARCH_X86_64 / ARCH_X86
 filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
@@ -270,7 +417,7 @@
 //                              int w, int h);
 FUN_CONV_2D(, ssse3);
 FUN_CONV_2D(avg_ , ssse3);
-#endif
+#endif  // HAVE_SSSE3
 
 #if HAVE_SSE2
 filter8_1dfunction vp9_filter_block1d16_v8_sse2;
@@ -336,4 +483,75 @@
 //                             int w, int h);
 FUN_CONV_2D(, sse2);
 FUN_CONV_2D(avg_ , sse2);
-#endif
+
+#if CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
+high_filter8_1dfunction vp9_high_filter_block1d16_v8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h8_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_v8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v8_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h8_avg_sse2;
+
+high_filter8_1dfunction vp9_high_filter_block1d16_v2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h2_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_v2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d16_h2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_v2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d8_h2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_v2_avg_sse2;
+high_filter8_1dfunction vp9_high_filter_block1d4_h2_avg_sse2;
+
+// void vp9_high_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                                    uint8_t *dst, ptrdiff_t dst_stride,
+//                                    const int16_t *filter_x, int x_step_q4,
+//                                    const int16_t *filter_y, int y_step_q4,
+//                                    int w, int h, int bd);
+// void vp9_high_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                                   uint8_t *dst, ptrdiff_t dst_stride,
+//                                   const int16_t *filter_x, int x_step_q4,
+//                                   const int16_t *filter_y, int y_step_q4,
+//                                   int w, int h, int bd);
+// void vp9_high_convolve8_avg_horiz_sse2(const uint8_t *src,
+//                                        ptrdiff_t src_stride,
+//                                        uint8_t *dst, ptrdiff_t dst_stride,
+//                                        const int16_t *filter_x,
+//                                        int x_step_q4,
+//                                        const int16_t *filter_y,
+//                                        int y_step_q4,
+//                                        int w, int h, int bd);
+// void vp9_high_convolve8_avg_vert_sse2(const uint8_t *src,
+//                                       ptrdiff_t src_stride,
+//                                       uint8_t *dst, ptrdiff_t dst_stride,
+//                                       const int16_t *filter_x, int x_step_q4,
+//                                       const int16_t *filter_y, int y_step_q4,
+//                                       int w, int h, int bd);
+HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
+HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
+HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
+HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
+                 sse2);
+
+// void vp9_high_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                              uint8_t *dst, ptrdiff_t dst_stride,
+//                              const int16_t *filter_x, int x_step_q4,
+//                              const int16_t *filter_y, int y_step_q4,
+//                              int w, int h, int bd);
+// void vp9_high_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
+//                                  uint8_t *dst, ptrdiff_t dst_stride,
+//                                  const int16_t *filter_x, int x_step_q4,
+//                                  const int16_t *filter_y, int y_step_q4,
+//                                  int w, int h, int bd);
+HIGH_FUN_CONV_2D(, sse2);
+HIGH_FUN_CONV_2D(avg_ , sse2);
+#endif  // CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
+#endif  // HAVE_SSE2
diff --git a/vp9/common/x86/vp9_high_intrapred_sse2.asm b/vp9/common/x86/vp9_high_intrapred_sse2.asm
new file mode 100644
index 0000000..ff45071
--- /dev/null
+++ b/vp9/common/x86/vp9_high_intrapred_sse2.asm
@@ -0,0 +1,476 @@
+;
+;  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+pw_4:  times 8 dw 4
+pw_8:  times 8 dw 8
+pw_16: times 4 dd 16
+pw_32: times 4 dd 32
+
+SECTION .text
+INIT_MMX sse
+cglobal high_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  movq                  m0, [aboveq]
+  movq                  m2, [leftq]
+  DEFINE_ARGS dst, stride, one
+  mov                 oned, 0x0001
+  pxor                  m1, m1
+  movd                  m3, oned
+  pshufw                m3, m3, 0x0
+  paddw                 m0, m2
+  pmaddwd               m0, m3
+  packssdw              m0, m1
+  pmaddwd               m0, m3
+  paddw                 m0, [GLOBAL(pw_4)]
+  psraw                 m0, 3
+  pshufw                m0, m0, 0x0
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal high_dc_predictor_8x8, 4, 5, 4, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m2, [leftq]
+  DEFINE_ARGS dst, stride, stride3, one
+  mov                 oned, 0x00010001
+  lea             stride3q, [strideq*3]
+  movd                  m3, oned
+  pshufd                m3, m3, 0x0
+  paddw                 m0, m2
+  pmaddwd               m0, m3
+  packssdw              m0, m1
+  pmaddwd               m0, m3
+  packssdw              m0, m1
+  pmaddwd               m0, m3
+  paddw                 m0, [GLOBAL(pw_8)]
+  psrlw                 m0, 4
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+  lea                 dstq, [dstq+strideq*8]
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal high_dc_predictor_16x16, 4, 5, 5, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m3, [aboveq+16]
+  mova                  m2, [leftq]
+  mova                  m4, [leftq+16]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 4
+  paddw                 m0, m2
+  paddw                 m0, m3
+  paddw                 m0, m4
+  movhlps               m2, m0
+  paddw                 m0, m2
+  punpcklwd             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  punpckldq             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  paddd                 m0, [GLOBAL(pw_16)]
+  psrad                 m0, 5
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+.loop:
+  mova   [dstq              ], m0
+  mova   [dstq           +16], m0
+  mova   [dstq+strideq*2    ], m0
+  mova   [dstq+strideq*2 +16], m0
+  mova   [dstq+strideq*4    ], m0
+  mova   [dstq+strideq*4 +16], m0
+  mova   [dstq+stride3q*2   ], m0
+  mova   [dstq+stride3q*2+16], m0
+  lea                 dstq, [dstq+strideq*8]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal high_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m2, [aboveq+16]
+  mova                  m3, [aboveq+32]
+  mova                  m4, [aboveq+48]
+  mova                  m5, [leftq]
+  mova                  m6, [leftq+16]
+  mova                  m7, [leftq+32]
+  mova                  m8, [leftq+48]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 8
+  paddw                 m0, m2
+  paddw                 m0, m3
+  paddw                 m0, m4
+  paddw                 m0, m5
+  paddw                 m0, m6
+  paddw                 m0, m7
+  paddw                 m0, m8
+  movhlps               m2, m0
+  paddw                 m0, m2
+  punpcklwd             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  punpckldq             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  paddd                 m0, [GLOBAL(pw_32)]
+  psrad                 m0, 6
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+.loop:
+  mova [dstq               ], m0
+  mova [dstq          +16  ], m0
+  mova [dstq          +32  ], m0
+  mova [dstq          +48  ], m0
+  mova [dstq+strideq*2     ], m0
+  mova [dstq+strideq*2+16  ], m0
+  mova [dstq+strideq*2+32  ], m0
+  mova [dstq+strideq*2+48  ], m0
+  mova [dstq+strideq*4     ], m0
+  mova [dstq+strideq*4+16  ], m0
+  mova [dstq+strideq*4+32  ], m0
+  mova [dstq+strideq*4+48  ], m0
+  mova [dstq+stride3q*2    ], m0
+  mova [dstq+stride3q*2 +16], m0
+  mova [dstq+stride3q*2 +32], m0
+  mova [dstq+stride3q*2 +48], m0
+  lea                 dstq, [dstq+strideq*8]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+%endif
+
+INIT_MMX sse
+cglobal high_v_predictor_4x4, 3, 3, 1, dst, stride, above
+  movq                  m0, [aboveq]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+  RET
+
+INIT_XMM sse2
+cglobal high_v_predictor_8x8, 3, 3, 1, dst, stride, above
+  mova                  m0, [aboveq]
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+  lea                 dstq, [dstq+strideq*8]
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+  RET
+
+INIT_XMM sse2
+cglobal high_v_predictor_16x16, 3, 4, 2, dst, stride, above
+  mova                  m0, [aboveq]
+  mova                  m1, [aboveq+16]
+  DEFINE_ARGS dst, stride, stride3, nlines4
+  lea             stride3q, [strideq*3]
+  mov              nlines4d, 4
+.loop:
+  mova    [dstq              ], m0
+  mova    [dstq           +16], m1
+  mova    [dstq+strideq*2    ], m0
+  mova    [dstq+strideq*2 +16], m1
+  mova    [dstq+strideq*4    ], m0
+  mova    [dstq+strideq*4 +16], m1
+  mova    [dstq+stride3q*2   ], m0
+  mova    [dstq+stride3q*2+16], m1
+  lea                 dstq, [dstq+strideq*8]
+  dec             nlines4d
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal high_v_predictor_32x32, 3, 4, 4, dst, stride, above
+  mova                  m0, [aboveq]
+  mova                  m1, [aboveq+16]
+  mova                  m2, [aboveq+32]
+  mova                  m3, [aboveq+48]
+  DEFINE_ARGS dst, stride, stride3, nlines4
+  lea             stride3q, [strideq*3]
+  mov              nlines4d, 8
+.loop:
+  mova [dstq               ], m0
+  mova [dstq            +16], m1
+  mova [dstq            +32], m2
+  mova [dstq            +48], m3
+  mova [dstq+strideq*2     ], m0
+  mova [dstq+strideq*2  +16], m1
+  mova [dstq+strideq*2  +32], m2
+  mova [dstq+strideq*2  +48], m3
+  mova [dstq+strideq*4     ], m0
+  mova [dstq+strideq*4  +16], m1
+  mova [dstq+strideq*4  +32], m2
+  mova [dstq+strideq*4  +48], m3
+  mova [dstq+stride3q*2    ], m0
+  mova [dstq+stride3q*2 +16], m1
+  mova [dstq+stride3q*2 +32], m2
+  mova [dstq+stride3q*2 +48], m3
+  lea                 dstq, [dstq+strideq*8]
+  dec             nlines4d
+  jnz .loop
+  REP_RET
+
+INIT_MMX sse
+cglobal high_tm_predictor_4x4, 5, 6, 5, dst, stride, above, left, bps, one
+  movd                  m1, [aboveq-2]
+  movq                  m0, [aboveq]
+  pshufw                m1, m1, 0x0
+  ; Get the values to compute the maximum value at this bit depth
+  mov                 oned, 1
+  movd                  m3, oned
+  movd                  m4, bpsd
+  pshufw                m3, m3, 0x0
+  DEFINE_ARGS dst, stride, line, left
+  mov                lineq, -2
+  mova                  m2, m3
+  psllw                 m3, m4
+  add                leftq, 8
+  psubw                 m3, m2 ; max possible value
+  pxor                  m4, m4 ; min possible value
+  psubw                 m0, m1
+.loop:
+  movq                  m1, [leftq+lineq*4]
+  movq                  m2, [leftq+lineq*4+2]
+  pshufw                m1, m1, 0x0
+  pshufw                m2, m2, 0x0
+  paddw                 m1, m0
+  paddw                 m2, m0
+  ;Clamp to the bit-depth
+  pminsw                m1, m3
+  pminsw                m2, m3
+  pmaxsw                m1, m4
+  pmaxsw                m2, m4
+  ;Store the values
+  movq    [dstq          ], m1
+  movq    [dstq+strideq*2], m2
+  lea                 dstq, [dstq+strideq*4]
+  inc                lineq
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal high_tm_predictor_8x8, 5, 6, 5, dst, stride, above, left, bps, one
+  movd                  m1, [aboveq-2]
+  mova                  m0, [aboveq]
+  pshuflw               m1, m1, 0x0
+  ; Get the values to compute the maximum value at this bit depth
+  mov                 oned, 1
+  pxor                  m3, m3
+  pxor                  m4, m4
+  pinsrw                m3, oned, 0
+  pinsrw                m4, bpsd, 0
+  pshuflw               m3, m3, 0x0
+  DEFINE_ARGS dst, stride, line, left
+  punpcklqdq            m3, m3
+  mov                lineq, -4
+  mova                  m2, m3
+  punpcklqdq            m1, m1
+  psllw                 m3, m4
+  add                leftq, 16
+  psubw                 m3, m2 ; max possible value
+  pxor                  m4, m4 ; min possible value
+  psubw                 m0, m1
+.loop:
+  movd                  m1, [leftq+lineq*4]
+  movd                  m2, [leftq+lineq*4+2]
+  pshuflw               m1, m1, 0x0
+  pshuflw               m2, m2, 0x0
+  punpcklqdq            m1, m1
+  punpcklqdq            m2, m2
+  paddw                 m1, m0
+  paddw                 m2, m0
+  ;Clamp to the bit-depth
+  pminsw                m1, m3
+  pminsw                m2, m3
+  pmaxsw                m1, m4
+  pmaxsw                m2, m4
+  ;Store the values
+  mova      [dstq          ], m1
+  mova      [dstq+strideq*2], m2
+  lea                 dstq, [dstq+strideq*4]
+  inc                lineq
+  jnz .loop
+  REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal high_tm_predictor_16x16, 5, 6, 8, dst, stride, above, left, bps, one
+  movd                  m2, [aboveq-2]
+  mova                  m0, [aboveq]
+  mova                  m1, [aboveq+16]
+  pshuflw               m2, m2, 0x0
+  ; Get the values to compute the maximum value at this bit depth
+  mov                 oned, 1
+  pxor                  m7, m7
+  pxor                  m8, m8
+  pinsrw                m7, oned, 0
+  pinsrw                m8, bpsd, 0
+  pshuflw               m7, m7, 0x0
+  DEFINE_ARGS dst, stride, line, left
+  punpcklqdq            m7, m7
+  mov                lineq, -8
+  mova                  m5, m7
+  punpcklqdq            m2, m2
+  psllw                 m7, m8
+  add                leftq, 32
+  psubw                 m7, m5 ; max possible value
+  pxor                  m8, m8 ; min possible value
+  psubw                 m0, m2
+  psubw                 m1, m2
+.loop:
+  movd                  m2, [leftq+lineq*4]
+  movd                  m3, [leftq+lineq*4+2]
+  pshuflw               m2, m2, 0x0
+  pshuflw               m3, m3, 0x0
+  punpcklqdq            m2, m2
+  punpcklqdq            m3, m3
+  paddw                 m4, m2, m0
+  paddw                 m5, m3, m0
+  paddw                 m2, m1
+  paddw                 m3, m1
+  ;Clamp to the bit-depth
+  pminsw                m4, m7
+  pminsw                m5, m7
+  pminsw                m2, m7
+  pminsw                m3, m7
+  pmaxsw                m4, m8
+  pmaxsw                m5, m8
+  pmaxsw                m2, m8
+  pmaxsw                m3, m8
+  ;Store the values
+  mova   [dstq             ], m4
+  mova   [dstq+strideq*2   ], m5
+  mova   [dstq          +16], m2
+  mova   [dstq+strideq*2+16], m3
+  lea                 dstq, [dstq+strideq*4]
+  inc                lineq
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal high_tm_predictor_32x32, 5, 6, 12, dst, stride, above, left, bps, one
+  movd                  m0, [aboveq-2]
+  mova                  m1, [aboveq]
+  mova                  m2, [aboveq+16]
+  mova                  m3, [aboveq+32]
+  mova                  m4, [aboveq+48]
+  pshuflw               m0, m0, 0x0
+  ; Get the values to compute the maximum value at this bit depth
+  mov                 oned, 1
+  pxor                 m10, m10
+  pxor                 m11, m11
+  pinsrw               m10, oned, 0
+  pinsrw               m11, bpsd, 0
+  pshuflw              m10, m10, 0x0
+  DEFINE_ARGS dst, stride, line, left
+  punpcklqdq           m10, m10
+  mov                lineq, -16
+  mova                  m5, m10
+  punpcklqdq            m0, m0
+  psllw                m10, m11
+  add                leftq, 64
+  psubw                m10, m5 ; max possible value
+  pxor                 m11, m11 ; min possible value
+  psubw                 m1, m0
+  psubw                 m2, m0
+  psubw                 m3, m0
+  psubw                 m4, m0
+.loop:
+  movd                  m5, [leftq+lineq*4]
+  movd                  m6, [leftq+lineq*4+2]
+  pshuflw               m5, m5, 0x0
+  pshuflw               m6, m6, 0x0
+  punpcklqdq            m5, m5
+  punpcklqdq            m6, m6
+  paddw                 m7, m5, m1
+  paddw                 m8, m5, m2
+  paddw                 m9, m5, m3
+  paddw                 m5, m4
+  ;Clamp these values to the bit-depth
+  pminsw                m7, m10
+  pminsw                m8, m10
+  pminsw                m9, m10
+  pminsw                m5, m10
+  pmaxsw                m7, m11
+  pmaxsw                m8, m11
+  pmaxsw                m9, m11
+  pmaxsw                m5, m11
+  ;Store these values
+  mova   [dstq           ], m7
+  mova   [dstq        +16], m8
+  mova   [dstq        +32], m9
+  mova   [dstq        +48], m5
+  paddw                 m7, m6, m1
+  paddw                 m8, m6, m2
+  paddw                 m9, m6, m3
+  paddw                 m6, m4
+  ;Clamp these values to the bit-depth
+  pminsw                m7, m10
+  pminsw                m8, m10
+  pminsw                m9, m10
+  pminsw                m6, m10
+  pmaxsw                m7, m11
+  pmaxsw                m8, m11
+  pmaxsw                m9, m11
+  pmaxsw                m6, m11
+  ;Store these values
+  mova   [dstq+strideq*2   ], m7
+  mova   [dstq+strideq*2+16], m8
+  mova   [dstq+strideq*2+32], m9
+  mova   [dstq+strideq*2+48], m6
+  lea                 dstq, [dstq+strideq*4]
+  inc                lineq
+  jnz .loop
+  REP_RET
+%endif
diff --git a/vp9/common/x86/vp9_high_loopfilter_intrin_sse2.c b/vp9/common/x86/vp9_high_loopfilter_intrin_sse2.c
new file mode 100644
index 0000000..32e4b20
--- /dev/null
+++ b/vp9/common/x86/vp9_high_loopfilter_intrin_sse2.c
@@ -0,0 +1,1119 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>  // SSE2
+
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_loopfilter.h"
+#include "vpx_ports/emmintrin_compat.h"
+
+static INLINE __m128i signed_char_clamp_bd_sse2(__m128i value, int bd) {
+    __m128i ubounded;
+    __m128i lbounded;
+    __m128i retval;
+
+    const __m128i zero = _mm_set1_epi16(0);
+    const __m128i one = _mm_set1_epi16(1);
+    const __m128i t80 = _mm_slli_epi16(_mm_set1_epi16(0x80), bd - 8);
+    const __m128i max = _mm_subs_epi16(
+        _mm_subs_epi16(_mm_slli_epi16(one, bd), one), t80);
+    const __m128i min = _mm_subs_epi16(zero, t80);
+    ubounded = _mm_cmpgt_epi16(value, max);
+    lbounded = _mm_cmplt_epi16(value, min);
+    retval = _mm_andnot_si128(_mm_or_si128(ubounded, lbounded), value);
+    ubounded = _mm_and_si128(ubounded, max);
+    lbounded = _mm_and_si128(lbounded, min);
+    retval = _mm_or_si128(retval, ubounded);
+    retval = _mm_or_si128(retval, lbounded);
+    return retval;
+}
+
+// TODO(debargha, peter): Break up large functions into smaller ones
+// in this file.
+static void highbd_mb_lpf_horizontal_edge_w_sse2_8(uint16_t *s,
+                                                   int p,
+                                                   const uint8_t *_blimit,
+                                                   const uint8_t *_limit,
+                                                   const uint8_t *_thresh,
+                                                   int bd) {
+  const __m128i zero = _mm_set1_epi16(0);
+  const __m128i one = _mm_set1_epi16(1);
+  const __m128i blimit = _mm_slli_epi16(
+      _mm_unpacklo_epi8(
+          _mm_load_si128((const __m128i *)_blimit), zero), bd - 8);
+  const __m128i limit = _mm_slli_epi16(
+      _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), bd - 8);
+  const __m128i thresh = _mm_slli_epi16(
+      _mm_unpacklo_epi8(
+          _mm_load_si128((const __m128i *)_thresh), zero), bd - 8);
+  __m128i q7, p7, q6, p6, q5, p5, q4, p4, q3, p3, q2, p2, q1, p1, q0, p0;
+  __m128i mask, hev, flat, flat2, abs_p1p0, abs_q1q0;
+  __m128i ps1, qs1, ps0, qs0;
+  __m128i abs_p0q0, abs_p1q1, ffff, work;
+  __m128i filt, work_a, filter1, filter2;
+  __m128i flat2_q6, flat2_p6, flat2_q5, flat2_p5, flat2_q4, flat2_p4;
+  __m128i flat2_q3, flat2_p3, flat2_q2, flat2_p2, flat2_q1, flat2_p1;
+  __m128i flat2_q0, flat2_p0;
+  __m128i flat_q2, flat_p2, flat_q1, flat_p1, flat_q0, flat_p0;
+  __m128i pixelFilter_p, pixelFilter_q;
+  __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0;
+  __m128i sum_p7, sum_q7, sum_p3, sum_q3;
+  __m128i t4, t3, t80, t1;
+  __m128i eight, four;
+
+  q4 = _mm_load_si128((__m128i *)(s + 4 * p));
+  p4 = _mm_load_si128((__m128i *)(s - 5 * p));
+  q3 = _mm_load_si128((__m128i *)(s + 3 * p));
+  p3 = _mm_load_si128((__m128i *)(s - 4 * p));
+  q2 = _mm_load_si128((__m128i *)(s + 2 * p));
+  p2 = _mm_load_si128((__m128i *)(s - 3 * p));
+  q1 = _mm_load_si128((__m128i *)(s + 1 * p));
+  p1 = _mm_load_si128((__m128i *)(s - 2 * p));
+  q0 = _mm_load_si128((__m128i *)(s + 0 * p));
+  p0 = _mm_load_si128((__m128i *)(s - 1 * p));
+
+  //  high_filter_mask
+  abs_p1p0 = _mm_or_si128(_mm_subs_epu16(p1, p0), _mm_subs_epu16(p0, p1));
+  abs_q1q0 = _mm_or_si128(_mm_subs_epu16(q1, q0), _mm_subs_epu16(q0, q1));
+
+  ffff = _mm_cmpeq_epi16(abs_p1p0, abs_p1p0);
+
+  abs_p0q0 = _mm_or_si128(_mm_subs_epu16(p0, q0), _mm_subs_epu16(q0, p0));
+  abs_p1q1 = _mm_or_si128(_mm_subs_epu16(p1, q1), _mm_subs_epu16(q1, p1));
+
+  //  high_hev_mask (in C code this is actually called from high_filter4)
+  flat = _mm_max_epi16(abs_p1p0, abs_q1q0);
+  hev = _mm_subs_epu16(flat, thresh);
+  hev = _mm_xor_si128(_mm_cmpeq_epi16(hev, zero), ffff);
+
+  abs_p0q0 =_mm_adds_epu16(abs_p0q0, abs_p0q0);  // abs(p0 - q0) * 2
+  abs_p1q1 = _mm_srli_epi16(abs_p1q1, 1);  // abs(p1 - q1) / 2
+  mask = _mm_subs_epu16(_mm_adds_epu16(abs_p0q0, abs_p1q1), blimit);
+  mask = _mm_xor_si128(_mm_cmpeq_epi16(mask, zero), ffff);
+  mask = _mm_and_si128(mask, _mm_adds_epu16(limit, one));
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p1, p0),
+                                    _mm_subs_epu16(p0, p1)),
+                       _mm_or_si128(_mm_subs_epu16(q1, q0),
+                                    _mm_subs_epu16(q0, q1)));
+  mask = _mm_max_epi16(work, mask);
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p1),
+                                    _mm_subs_epu16(p1, p2)),
+                       _mm_or_si128(_mm_subs_epu16(q2, q1),
+                                    _mm_subs_epu16(q1, q2)));
+  mask = _mm_max_epi16(work, mask);
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p3, p2),
+                                    _mm_subs_epu16(p2, p3)),
+                       _mm_or_si128(_mm_subs_epu16(q3, q2),
+                                    _mm_subs_epu16(q2, q3)));
+  mask = _mm_max_epi16(work, mask);
+
+  mask = _mm_subs_epu16(mask, limit);
+  mask = _mm_cmpeq_epi16(mask, zero);  // return ~mask
+
+  // lp filter
+  // high_filter4
+  t4 = _mm_set1_epi16(4);
+  t3 = _mm_set1_epi16(3);
+  t80 = _mm_slli_epi16(_mm_set1_epi16(0x80), bd - 8);
+  t1 = _mm_set1_epi16(0x1);
+
+  ps1 = _mm_subs_epi16(p1, t80);
+  qs1 = _mm_subs_epi16(q1, t80);
+  ps0 = _mm_subs_epi16(p0, t80);
+  qs0 = _mm_subs_epi16(q0, t80);
+
+  filt = _mm_and_si128(
+      signed_char_clamp_bd_sse2(_mm_subs_epi16(ps1, qs1), bd), hev);
+  work_a = _mm_subs_epi16(qs0, ps0);
+  filt = _mm_adds_epi16(filt, work_a);
+  filt = _mm_adds_epi16(filt, work_a);
+  filt = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, work_a), bd);
+  filt = _mm_and_si128(filt, mask);
+
+  filter1 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t4), bd);
+  filter2 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t3), bd);
+
+  // Filter1 >> 3
+  filter1 = _mm_srai_epi16(filter1, 0x3);
+  filter2 = _mm_srai_epi16(filter2, 0x3);
+
+  qs0 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs0, filter1), bd),
+      t80);
+  ps0 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps0, filter2), bd),
+      t80);
+  filt = _mm_adds_epi16(filter1, t1);
+  filt = _mm_srai_epi16(filt, 1);
+  filt = _mm_andnot_si128(hev, filt);
+
+  qs1 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs1, filt), bd),
+      t80);
+  ps1 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps1, filt), bd),
+      t80);
+  // end high_filter4
+  // loopfilter done
+
+  // high_flat_mask4
+  flat = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p0),
+                                    _mm_subs_epu16(p0, p2)),
+                       _mm_or_si128(_mm_subs_epu16(p3, p0),
+                                    _mm_subs_epu16(p0, p3)));
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(q2, q0),
+                                    _mm_subs_epu16(q0, q2)),
+                       _mm_or_si128(_mm_subs_epu16(q3, q0),
+                                    _mm_subs_epu16(q0, q3)));
+  flat = _mm_max_epi16(work, flat);
+  work = _mm_max_epi16(abs_p1p0, abs_q1q0);
+  flat = _mm_max_epi16(work, flat);
+  flat = _mm_subs_epu16(flat, _mm_slli_epi16(one, bd - 8));
+  flat = _mm_cmpeq_epi16(flat, zero);
+  // end flat_mask4
+
+  // flat & mask = flat && mask (as used in filter8)
+  // (because, in both vars, each block of 16 either all 1s or all 0s)
+  flat = _mm_and_si128(flat, mask);
+
+  p5 = _mm_load_si128((__m128i *)(s - 6 * p));
+  q5 = _mm_load_si128((__m128i *)(s + 5 * p));
+  p6 = _mm_load_si128((__m128i *)(s - 7 * p));
+  q6 = _mm_load_si128((__m128i *)(s + 6 * p));
+  p7 = _mm_load_si128((__m128i *)(s - 8 * p));
+  q7 = _mm_load_si128((__m128i *)(s + 7 * p));
+
+  // high_flat_mask5 (arguments passed in are p0, q0, p4-p7, q4-q7
+  // but referred to as p0-p4 & q0-q4 in fn)
+  flat2 = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p4, p0),
+                                     _mm_subs_epu16(p0, p4)),
+                        _mm_or_si128(_mm_subs_epu16(q4, q0),
+                                     _mm_subs_epu16(q0, q4)));
+
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p5, p0),
+                                    _mm_subs_epu16(p0, p5)),
+                       _mm_or_si128(_mm_subs_epu16(q5, q0),
+                                    _mm_subs_epu16(q0, q5)));
+  flat2 = _mm_max_epi16(work, flat2);
+
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p6, p0),
+                                    _mm_subs_epu16(p0, p6)),
+                       _mm_or_si128(_mm_subs_epu16(q6, q0),
+                                    _mm_subs_epu16(q0, q6)));
+  flat2 = _mm_max_epi16(work, flat2);
+
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p7, p0),
+                                    _mm_subs_epu16(p0, p7)),
+                       _mm_or_si128(_mm_subs_epu16(q7, q0),
+                                    _mm_subs_epu16(q0, q7)));
+  flat2 = _mm_max_epi16(work, flat2);
+
+  flat2 = _mm_subs_epu16(flat2, _mm_slli_epi16(one, bd - 8));
+  flat2 = _mm_cmpeq_epi16(flat2, zero);
+  flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
+  // end high_flat_mask5
+
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // flat and wide flat calculations
+  eight = _mm_set1_epi16(8);
+  four = _mm_set1_epi16(4);
+
+  pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6, p5),
+                                _mm_add_epi16(p4, p3));
+  pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6, q5),
+                                _mm_add_epi16(q4, q3));
+
+  pixetFilter_p2p1p0 = _mm_add_epi16(p0, _mm_add_epi16(p2, p1));
+  pixelFilter_p = _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0);
+
+  pixetFilter_q2q1q0 = _mm_add_epi16(q0, _mm_add_epi16(q2, q1));
+  pixelFilter_q = _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0);
+  pixelFilter_p = _mm_add_epi16(eight, _mm_add_epi16(pixelFilter_p,
+                                                      pixelFilter_q));
+  pixetFilter_p2p1p0 =   _mm_add_epi16(four,
+                                       _mm_add_epi16(pixetFilter_p2p1p0,
+                                                     pixetFilter_q2q1q0));
+  flat2_p0 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(p7, p0)), 4);
+  flat2_q0 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(q7, q0)), 4);
+  flat_p0 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
+                                         _mm_add_epi16(p3, p0)), 3);
+  flat_q0 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
+                                         _mm_add_epi16(q3, q0)), 3);
+
+  sum_p7 = _mm_add_epi16(p7, p7);
+  sum_q7 = _mm_add_epi16(q7, q7);
+  sum_p3 = _mm_add_epi16(p3, p3);
+  sum_q3 = _mm_add_epi16(q3, q3);
+
+  pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6);
+  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6);
+  flat2_p1 = _mm_srli_epi16(
+      _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p1)), 4);
+  flat2_q1 = _mm_srli_epi16(
+      _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q1)), 4);
+
+  pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2);
+  pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2);
+  flat_p1 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
+                                         _mm_add_epi16(sum_p3, p1)), 3);
+  flat_q1 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0,
+                                         _mm_add_epi16(sum_q3, q1)), 3);
+
+  sum_p7 = _mm_add_epi16(sum_p7, p7);
+  sum_q7 = _mm_add_epi16(sum_q7, q7);
+  sum_p3 = _mm_add_epi16(sum_p3, p3);
+  sum_q3 = _mm_add_epi16(sum_q3, q3);
+
+  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5);
+  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5);
+  flat2_p2 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(sum_p7, p2)), 4);
+  flat2_q2 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
+                                          _mm_add_epi16(sum_q7, q2)), 4);
+
+  pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1);
+  pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1);
+  flat_p2 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
+                                         _mm_add_epi16(sum_p3, p2)), 3);
+  flat_q2 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0,
+                                         _mm_add_epi16(sum_q3, q2)), 3);
+
+  sum_p7 = _mm_add_epi16(sum_p7, p7);
+  sum_q7 = _mm_add_epi16(sum_q7, q7);
+  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4);
+  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4);
+  flat2_p3 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(sum_p7, p3)), 4);
+  flat2_q3 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
+                                          _mm_add_epi16(sum_q7, q3)), 4);
+
+  sum_p7 = _mm_add_epi16(sum_p7, p7);
+  sum_q7 = _mm_add_epi16(sum_q7, q7);
+  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3);
+  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3);
+  flat2_p4 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(sum_p7, p4)), 4);
+  flat2_q4 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
+                                          _mm_add_epi16(sum_q7, q4)), 4);
+
+  sum_p7 = _mm_add_epi16(sum_p7, p7);
+  sum_q7 = _mm_add_epi16(sum_q7, q7);
+  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2);
+  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2);
+  flat2_p5 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(sum_p7, p5)), 4);
+  flat2_q5 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
+                                          _mm_add_epi16(sum_q7, q5)), 4);
+
+  sum_p7 = _mm_add_epi16(sum_p7, p7);
+  sum_q7 = _mm_add_epi16(sum_q7, q7);
+  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1);
+  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1);
+  flat2_p6 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
+                                          _mm_add_epi16(sum_p7, p6)), 4);
+  flat2_q6 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
+                                          _mm_add_epi16(sum_q7, q6)), 4);
+
+  //  wide flat
+  //  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+  //  high_filter8
+  p2 = _mm_andnot_si128(flat, p2);
+  //  p2 remains unchanged if !(flat && mask)
+  flat_p2 = _mm_and_si128(flat, flat_p2);
+  //  when (flat && mask)
+  p2 = _mm_or_si128(p2, flat_p2);  // full list of p2 values
+  q2 = _mm_andnot_si128(flat, q2);
+  flat_q2 = _mm_and_si128(flat, flat_q2);
+  q2 = _mm_or_si128(q2, flat_q2);  // full list of q2 values
+
+  ps1 = _mm_andnot_si128(flat, ps1);
+  //  p1 takes the value assigned to in in filter4 if !(flat && mask)
+  flat_p1 = _mm_and_si128(flat, flat_p1);
+  //  when (flat && mask)
+  p1 = _mm_or_si128(ps1, flat_p1);  // full list of p1 values
+  qs1 = _mm_andnot_si128(flat, qs1);
+  flat_q1 = _mm_and_si128(flat, flat_q1);
+  q1 = _mm_or_si128(qs1, flat_q1);  // full list of q1 values
+
+  ps0 = _mm_andnot_si128(flat, ps0);
+  //  p0 takes the value assigned to in in filter4 if !(flat && mask)
+  flat_p0 = _mm_and_si128(flat, flat_p0);
+  //  when (flat && mask)
+  p0 = _mm_or_si128(ps0, flat_p0);  // full list of p0 values
+  qs0 = _mm_andnot_si128(flat, qs0);
+  flat_q0 = _mm_and_si128(flat, flat_q0);
+  q0 = _mm_or_si128(qs0, flat_q0);  // full list of q0 values
+  // end high_filter8
+
+  // high_filter16
+  p6 = _mm_andnot_si128(flat2, p6);
+  //  p6 remains unchanged if !(flat2 && flat && mask)
+  flat2_p6 = _mm_and_si128(flat2, flat2_p6);
+  //  get values for when (flat2 && flat && mask)
+  p6 = _mm_or_si128(p6, flat2_p6);  // full list of p6 values
+  q6 = _mm_andnot_si128(flat2, q6);
+  //  q6 remains unchanged if !(flat2 && flat && mask)
+  flat2_q6 = _mm_and_si128(flat2, flat2_q6);
+  //  get values for when (flat2 && flat && mask)
+  q6 = _mm_or_si128(q6, flat2_q6);  // full list of q6 values
+  _mm_store_si128((__m128i *)(s - 7 * p), p6);
+  _mm_store_si128((__m128i *)(s + 6 * p), q6);
+
+  p5 = _mm_andnot_si128(flat2, p5);
+  //  p5 remains unchanged if !(flat2 && flat && mask)
+  flat2_p5 = _mm_and_si128(flat2, flat2_p5);
+  //  get values for when (flat2 && flat && mask)
+  p5 = _mm_or_si128(p5, flat2_p5);
+  //  full list of p5 values
+  q5 = _mm_andnot_si128(flat2, q5);
+  //  q5 remains unchanged if !(flat2 && flat && mask)
+  flat2_q5 = _mm_and_si128(flat2, flat2_q5);
+  //  get values for when (flat2 && flat && mask)
+  q5 = _mm_or_si128(q5, flat2_q5);
+  //  full list of q5 values
+  _mm_store_si128((__m128i *)(s - 6 * p), p5);
+  _mm_store_si128((__m128i *)(s + 5 * p), q5);
+
+  p4 = _mm_andnot_si128(flat2, p4);
+  //  p4 remains unchanged if !(flat2 && flat && mask)
+  flat2_p4 = _mm_and_si128(flat2, flat2_p4);
+  //  get values for when (flat2 && flat && mask)
+  p4 = _mm_or_si128(p4, flat2_p4);  // full list of p4 values
+  q4 = _mm_andnot_si128(flat2, q4);
+  //  q4 remains unchanged if !(flat2 && flat && mask)
+  flat2_q4 = _mm_and_si128(flat2, flat2_q4);
+  //  get values for when (flat2 && flat && mask)
+  q4 = _mm_or_si128(q4, flat2_q4);  // full list of q4 values
+  _mm_store_si128((__m128i *)(s - 5 * p), p4);
+  _mm_store_si128((__m128i *)(s + 4 * p), q4);
+
+  p3 = _mm_andnot_si128(flat2, p3);
+  //  p3 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_p3 = _mm_and_si128(flat2, flat2_p3);
+  //  get values for when (flat2 && flat && mask)
+  p3 = _mm_or_si128(p3, flat2_p3);  // full list of p3 values
+  q3 = _mm_andnot_si128(flat2, q3);
+  //  q3 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_q3 = _mm_and_si128(flat2, flat2_q3);
+  //  get values for when (flat2 && flat && mask)
+  q3 = _mm_or_si128(q3, flat2_q3);  // full list of q3 values
+  _mm_store_si128((__m128i *)(s - 4 * p), p3);
+  _mm_store_si128((__m128i *)(s + 3 * p), q3);
+
+  p2 = _mm_andnot_si128(flat2, p2);
+  //  p2 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_p2 = _mm_and_si128(flat2, flat2_p2);
+  //  get values for when (flat2 && flat && mask)
+  p2 = _mm_or_si128(p2, flat2_p2);
+  //  full list of p2 values
+  q2 = _mm_andnot_si128(flat2, q2);
+  //  q2 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_q2 = _mm_and_si128(flat2, flat2_q2);
+  //  get values for when (flat2 && flat && mask)
+  q2 = _mm_or_si128(q2, flat2_q2);  // full list of q2 values
+  _mm_store_si128((__m128i *)(s - 3 * p), p2);
+  _mm_store_si128((__m128i *)(s + 2 * p), q2);
+
+  p1 = _mm_andnot_si128(flat2, p1);
+  //  p1 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_p1 = _mm_and_si128(flat2, flat2_p1);
+  //  get values for when (flat2 && flat && mask)
+  p1 = _mm_or_si128(p1, flat2_p1);  // full list of p1 values
+  q1 = _mm_andnot_si128(flat2, q1);
+  //  q1 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_q1 = _mm_and_si128(flat2, flat2_q1);
+  //  get values for when (flat2 && flat && mask)
+  q1 = _mm_or_si128(q1, flat2_q1);  // full list of q1 values
+  _mm_store_si128((__m128i *)(s - 2 * p), p1);
+  _mm_store_si128((__m128i *)(s + 1 * p), q1);
+
+  p0 = _mm_andnot_si128(flat2, p0);
+  //  p0 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_p0 = _mm_and_si128(flat2, flat2_p0);
+  //  get values for when (flat2 && flat && mask)
+  p0 = _mm_or_si128(p0, flat2_p0);  // full list of p0 values
+  q0 = _mm_andnot_si128(flat2, q0);
+  //  q0 takes value from high_filter8 if !(flat2 && flat && mask)
+  flat2_q0 = _mm_and_si128(flat2, flat2_q0);
+  //  get values for when (flat2 && flat && mask)
+  q0 = _mm_or_si128(q0, flat2_q0);  // full list of q0 values
+  _mm_store_si128((__m128i *)(s - 1 * p), p0);
+  _mm_store_si128((__m128i *)(s - 0 * p), q0);
+}
+
+static void highbd_mb_lpf_horizontal_edge_w_sse2_16(uint16_t *s,
+                                                    int p,
+                                                    const uint8_t *_blimit,
+                                                    const uint8_t *_limit,
+                                                    const uint8_t *_thresh,
+                                                    int bd) {
+  highbd_mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh, bd);
+  highbd_mb_lpf_horizontal_edge_w_sse2_8(s + 8, p, _blimit, _limit, _thresh,
+                                         bd);
+}
+
+// TODO(yunqingwang): remove count and call these 2 functions(8 or 16) directly.
+void vp9_highbd_lpf_horizontal_16_sse2(uint16_t *s, int p,
+                                       const uint8_t *_blimit,
+                                       const uint8_t *_limit,
+                                       const uint8_t *_thresh,
+                                       int count, int bd) {
+  if (count == 1)
+    highbd_mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh, bd);
+  else
+    highbd_mb_lpf_horizontal_edge_w_sse2_16(s, p, _blimit, _limit, _thresh, bd);
+}
+
+void vp9_highbd_lpf_horizontal_8_sse2(uint16_t *s, int p,
+                                      const uint8_t *_blimit,
+                                      const uint8_t *_limit,
+                                      const uint8_t *_thresh,
+                                      int count, int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, flat_op2, 16);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, flat_op1, 16);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, flat_op0, 16);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, flat_oq2, 16);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, flat_oq1, 16);
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, flat_oq0, 16);
+  const __m128i zero = _mm_set1_epi16(0);
+  const __m128i blimit = _mm_slli_epi16(
+      _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero),
+      bd - 8);
+  const __m128i limit = _mm_slli_epi16(
+      _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero),
+      bd - 8);
+  const __m128i thresh = _mm_slli_epi16(
+      _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero),
+      bd - 8);
+  __m128i mask, hev, flat;
+  __m128i p3 = _mm_load_si128((__m128i *)(s - 4 * p));
+  __m128i q3 = _mm_load_si128((__m128i *)(s + 3 * p));
+  __m128i p2 = _mm_load_si128((__m128i *)(s - 3 * p));
+  __m128i q2 = _mm_load_si128((__m128i *)(s + 2 * p));
+  __m128i p1 = _mm_load_si128((__m128i *)(s - 2 * p));
+  __m128i q1 = _mm_load_si128((__m128i *)(s + 1 * p));
+  __m128i p0 = _mm_load_si128((__m128i *)(s - 1 * p));
+  __m128i q0 = _mm_load_si128((__m128i *)(s + 0 * p));
+  const __m128i one = _mm_set1_epi16(1);
+  const __m128i ffff = _mm_cmpeq_epi16(one, one);
+  __m128i abs_p1q1, abs_p0q0, abs_q1q0, abs_p1p0, work;
+  const __m128i four = _mm_set1_epi16(4);
+  __m128i workp_a, workp_b, workp_shft;
+
+  const __m128i t4 = _mm_set1_epi16(4);
+  const __m128i t3 = _mm_set1_epi16(3);
+  const __m128i t80 = _mm_slli_epi16(_mm_set1_epi16(0x80), bd - 8);
+  const __m128i t1 = _mm_set1_epi16(0x1);
+  const __m128i ps1 = _mm_subs_epi16(p1, t80);
+  const __m128i ps0 = _mm_subs_epi16(p0, t80);
+  const __m128i qs0 = _mm_subs_epi16(q0, t80);
+  const __m128i qs1 = _mm_subs_epi16(q1, t80);
+  __m128i filt;
+  __m128i work_a;
+  __m128i filter1, filter2;
+
+  (void)count;
+
+  // filter_mask and hev_mask
+  abs_p1p0 = _mm_or_si128(_mm_subs_epu16(p1, p0),
+                          _mm_subs_epu16(p0, p1));
+  abs_q1q0 = _mm_or_si128(_mm_subs_epu16(q1, q0),
+                          _mm_subs_epu16(q0, q1));
+
+  abs_p0q0 = _mm_or_si128(_mm_subs_epu16(p0, q0),
+                          _mm_subs_epu16(q0, p0));
+  abs_p1q1 = _mm_or_si128(_mm_subs_epu16(p1, q1),
+                          _mm_subs_epu16(q1, p1));
+  flat = _mm_max_epi16(abs_p1p0, abs_q1q0);
+  hev = _mm_subs_epu16(flat, thresh);
+  hev = _mm_xor_si128(_mm_cmpeq_epi16(hev, zero), ffff);
+
+  abs_p0q0 =_mm_adds_epu16(abs_p0q0, abs_p0q0);
+  abs_p1q1 = _mm_srli_epi16(abs_p1q1, 1);
+  mask = _mm_subs_epu16(_mm_adds_epu16(abs_p0q0, abs_p1q1), blimit);
+  mask = _mm_xor_si128(_mm_cmpeq_epi16(mask, zero), ffff);
+  // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
+  // So taking maximums continues to work:
+  mask = _mm_and_si128(mask, _mm_adds_epu16(limit, one));
+  mask = _mm_max_epi16(abs_p1p0, mask);
+  // mask |= (abs(p1 - p0) > limit) * -1;
+  mask = _mm_max_epi16(abs_q1q0, mask);
+  // mask |= (abs(q1 - q0) > limit) * -1;
+
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p1),
+                                    _mm_subs_epu16(p1, p2)),
+                       _mm_or_si128(_mm_subs_epu16(q2, q1),
+                                    _mm_subs_epu16(q1, q2)));
+  mask = _mm_max_epi16(work, mask);
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p3, p2),
+                                    _mm_subs_epu16(p2, p3)),
+                       _mm_or_si128(_mm_subs_epu16(q3, q2),
+                                    _mm_subs_epu16(q2, q3)));
+  mask = _mm_max_epi16(work, mask);
+  mask = _mm_subs_epu16(mask, limit);
+  mask = _mm_cmpeq_epi16(mask, zero);
+
+  // flat_mask4
+  flat = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p0),
+                                    _mm_subs_epu16(p0, p2)),
+                       _mm_or_si128(_mm_subs_epu16(q2, q0),
+                                    _mm_subs_epu16(q0, q2)));
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p3, p0),
+                                    _mm_subs_epu16(p0, p3)),
+                       _mm_or_si128(_mm_subs_epu16(q3, q0),
+                                    _mm_subs_epu16(q0, q3)));
+  flat = _mm_max_epi16(work, flat);
+  flat = _mm_max_epi16(abs_p1p0, flat);
+  flat = _mm_max_epi16(abs_q1q0, flat);
+  flat = _mm_subs_epu16(flat, _mm_slli_epi16(one, bd - 8));
+  flat = _mm_cmpeq_epi16(flat, zero);
+  flat = _mm_and_si128(flat, mask);  // flat & mask
+
+  // Added before shift for rounding part of ROUND_POWER_OF_TWO
+
+  workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1));
+  workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0);
+  workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3);
+  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+  _mm_store_si128((__m128i *)&flat_op2[0], workp_shft);
+
+  workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1);
+  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+  _mm_store_si128((__m128i *)&flat_op1[0], workp_shft);
+
+  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2);
+  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0);
+  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+  _mm_store_si128((__m128i *)&flat_op0[0], workp_shft);
+
+  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3);
+  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0);
+  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+  _mm_store_si128((__m128i *)&flat_oq0[0], workp_shft);
+
+  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3);
+  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1);
+  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+  _mm_store_si128((__m128i *)&flat_oq1[0], workp_shft);
+
+  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3);
+  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2);
+  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
+  _mm_store_si128((__m128i *)&flat_oq2[0], workp_shft);
+
+  // lp filter
+  filt = signed_char_clamp_bd_sse2(_mm_subs_epi16(ps1, qs1), bd);
+  filt = _mm_and_si128(filt, hev);
+  work_a = _mm_subs_epi16(qs0, ps0);
+  filt = _mm_adds_epi16(filt, work_a);
+  filt = _mm_adds_epi16(filt, work_a);
+  filt = _mm_adds_epi16(filt, work_a);
+  // (vp9_filter + 3 * (qs0 - ps0)) & mask
+  filt = signed_char_clamp_bd_sse2(filt, bd);
+  filt = _mm_and_si128(filt, mask);
+
+  filter1 = _mm_adds_epi16(filt, t4);
+  filter2 = _mm_adds_epi16(filt, t3);
+
+  // Filter1 >> 3
+  filter1 = signed_char_clamp_bd_sse2(filter1, bd);
+  filter1 = _mm_srai_epi16(filter1, 3);
+
+  // Filter2 >> 3
+  filter2 = signed_char_clamp_bd_sse2(filter2, bd);
+  filter2 = _mm_srai_epi16(filter2, 3);
+
+  // filt >> 1
+  filt = _mm_adds_epi16(filter1, t1);
+  filt = _mm_srai_epi16(filt, 1);
+  // filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;
+  filt = _mm_andnot_si128(hev, filt);
+
+  work_a = signed_char_clamp_bd_sse2(_mm_subs_epi16(qs0, filter1), bd);
+  work_a = _mm_adds_epi16(work_a, t80);
+  q0 = _mm_load_si128((__m128i *)flat_oq0);
+  work_a = _mm_andnot_si128(flat, work_a);
+  q0 = _mm_and_si128(flat, q0);
+  q0 = _mm_or_si128(work_a, q0);
+
+  work_a = signed_char_clamp_bd_sse2(_mm_subs_epi16(qs1, filt), bd);
+  work_a = _mm_adds_epi16(work_a, t80);
+  q1 = _mm_load_si128((__m128i *)flat_oq1);
+  work_a = _mm_andnot_si128(flat, work_a);
+  q1 = _mm_and_si128(flat, q1);
+  q1 = _mm_or_si128(work_a, q1);
+
+  work_a = _mm_loadu_si128((__m128i *)(s + 2 * p));
+  q2 = _mm_load_si128((__m128i *)flat_oq2);
+  work_a = _mm_andnot_si128(flat, work_a);
+  q2 = _mm_and_si128(flat, q2);
+  q2 = _mm_or_si128(work_a, q2);
+
+  work_a = signed_char_clamp_bd_sse2(_mm_adds_epi16(ps0, filter2), bd);
+  work_a = _mm_adds_epi16(work_a, t80);
+  p0 = _mm_load_si128((__m128i *)flat_op0);
+  work_a = _mm_andnot_si128(flat, work_a);
+  p0 = _mm_and_si128(flat, p0);
+  p0 = _mm_or_si128(work_a, p0);
+
+  work_a = signed_char_clamp_bd_sse2(_mm_adds_epi16(ps1, filt), bd);
+  work_a = _mm_adds_epi16(work_a, t80);
+  p1 = _mm_load_si128((__m128i *)flat_op1);
+  work_a = _mm_andnot_si128(flat, work_a);
+  p1 = _mm_and_si128(flat, p1);
+  p1 = _mm_or_si128(work_a, p1);
+
+  work_a = _mm_loadu_si128((__m128i *)(s - 3 * p));
+  p2 = _mm_load_si128((__m128i *)flat_op2);
+  work_a = _mm_andnot_si128(flat, work_a);
+  p2 = _mm_and_si128(flat, p2);
+  p2 = _mm_or_si128(work_a, p2);
+
+  _mm_store_si128((__m128i *)(s - 3 * p), p2);
+  _mm_store_si128((__m128i *)(s - 2 * p), p1);
+  _mm_store_si128((__m128i *)(s - 1 * p), p0);
+  _mm_store_si128((__m128i *)(s + 0 * p), q0);
+  _mm_store_si128((__m128i *)(s + 1 * p), q1);
+  _mm_store_si128((__m128i *)(s + 2 * p), q2);
+}
+
+void vp9_highbd_lpf_horizontal_8_dual_sse2(uint16_t *s, int p,
+                                           const uint8_t *_blimit0,
+                                           const uint8_t *_limit0,
+                                           const uint8_t *_thresh0,
+                                           const uint8_t *_blimit1,
+                                           const uint8_t *_limit1,
+                                           const uint8_t *_thresh1,
+                                           int bd) {
+  vp9_highbd_lpf_horizontal_8_sse2(s, p, _blimit0, _limit0, _thresh0, 1, bd);
+  vp9_highbd_lpf_horizontal_8_sse2(s + 8, p, _blimit1, _limit1, _thresh1,
+                                   1, bd);
+}
+
+void vp9_highbd_lpf_horizontal_4_sse2(uint16_t *s, int p,
+                                      const uint8_t *_blimit,
+                                      const uint8_t *_limit,
+                                      const uint8_t *_thresh,
+                                      int count, int bd) {
+  const __m128i zero = _mm_set1_epi16(0);
+  const __m128i blimit = _mm_slli_epi16(
+      _mm_unpacklo_epi8(
+          _mm_load_si128((const __m128i *)_blimit), zero), bd - 8);
+  const __m128i limit = _mm_slli_epi16(
+      _mm_unpacklo_epi8(
+          _mm_load_si128((const __m128i *)_limit), zero), bd - 8);
+  const __m128i thresh = _mm_slli_epi16(
+      _mm_unpacklo_epi8(
+          _mm_load_si128((const __m128i *)_thresh), zero), bd - 8);
+  __m128i mask, hev, flat;
+  __m128i p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
+  __m128i p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
+  __m128i p1 = _mm_loadu_si128((__m128i *)(s - 2 * p));
+  __m128i p0 = _mm_loadu_si128((__m128i *)(s - 1 * p));
+  __m128i q0 = _mm_loadu_si128((__m128i *)(s - 0 * p));
+  __m128i q1 = _mm_loadu_si128((__m128i *)(s + 1 * p));
+  __m128i q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
+  __m128i q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
+  const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu16(p1, p0),
+                                        _mm_subs_epu16(p0, p1));
+  const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu16(q1, q0),
+                                        _mm_subs_epu16(q0, q1));
+  const __m128i ffff = _mm_cmpeq_epi16(abs_p1p0, abs_p1p0);
+  const __m128i one = _mm_set1_epi16(1);
+  __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu16(p0, q0),
+                                  _mm_subs_epu16(q0, p0));
+  __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu16(p1, q1),
+                                  _mm_subs_epu16(q1, p1));
+  __m128i work;
+  const __m128i t4 = _mm_set1_epi16(4);
+  const __m128i t3 = _mm_set1_epi16(3);
+  const __m128i t80 = _mm_slli_epi16(_mm_set1_epi16(0x80), bd - 8);
+  const __m128i tff80 = _mm_slli_epi16(_mm_set1_epi16(0xff80), bd - 8);
+  const __m128i tffe0 = _mm_slli_epi16(_mm_set1_epi16(0xffe0), bd - 8);
+  const __m128i t1f = _mm_srli_epi16(_mm_set1_epi16(0x1fff), 16 - bd);
+  // equivalent to shifting 0x1f left by bitdepth - 8
+  // and setting new bits to 1
+  const __m128i t1 = _mm_set1_epi16(0x1);
+  const __m128i t7f = _mm_srli_epi16(_mm_set1_epi16(0x7fff), 16 - bd);
+  // equivalent to shifting 0x7f left by bitdepth - 8
+  // and setting new bits to 1
+  const __m128i ps1 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s - 2 * p)),
+                                     t80);
+  const __m128i ps0 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s - 1 * p)),
+                                     t80);
+  const __m128i qs0 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s + 0 * p)),
+                                     t80);
+  const __m128i qs1 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s + 1 * p)),
+                                     t80);
+  __m128i filt;
+  __m128i work_a;
+  __m128i filter1, filter2;
+
+  (void)count;
+
+  // filter_mask and hev_mask
+  flat = _mm_max_epi16(abs_p1p0, abs_q1q0);
+  hev = _mm_subs_epu16(flat, thresh);
+  hev = _mm_xor_si128(_mm_cmpeq_epi16(hev, zero), ffff);
+
+  abs_p0q0 =_mm_adds_epu16(abs_p0q0, abs_p0q0);
+  abs_p1q1 = _mm_srli_epi16(abs_p1q1, 1);
+  mask = _mm_subs_epu16(_mm_adds_epu16(abs_p0q0, abs_p1q1), blimit);
+  mask = _mm_xor_si128(_mm_cmpeq_epi16(mask, zero), ffff);
+  // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
+  // So taking maximums continues to work:
+  mask = _mm_and_si128(mask, _mm_adds_epu16(limit, one));
+  mask = _mm_max_epi16(flat, mask);
+  // mask |= (abs(p1 - p0) > limit) * -1;
+  // mask |= (abs(q1 - q0) > limit) * -1;
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p1),
+                                    _mm_subs_epu16(p1, p2)),
+                       _mm_or_si128(_mm_subs_epu16(p3, p2),
+                                    _mm_subs_epu16(p2, p3)));
+  mask = _mm_max_epi16(work, mask);
+  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(q2, q1),
+                                    _mm_subs_epu16(q1, q2)),
+                       _mm_or_si128(_mm_subs_epu16(q3, q2),
+                                    _mm_subs_epu16(q2, q3)));
+  mask = _mm_max_epi16(work, mask);
+  mask = _mm_subs_epu16(mask, limit);
+  mask = _mm_cmpeq_epi16(mask, zero);
+
+  // filter4
+  filt = signed_char_clamp_bd_sse2(_mm_subs_epi16(ps1, qs1), bd);
+  filt = _mm_and_si128(filt, hev);
+  work_a = _mm_subs_epi16(qs0, ps0);
+  filt = _mm_adds_epi16(filt, work_a);
+  filt = _mm_adds_epi16(filt, work_a);
+  filt = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, work_a), bd);
+  // (vp9_filter + 3 * (qs0 - ps0)) & mask
+  filt = _mm_and_si128(filt, mask);
+
+  filter1 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t4), bd);
+  filter2 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t3), bd);
+
+  // Filter1 >> 3
+  work_a = _mm_cmpgt_epi16(zero, filter1);  // get the values that are <0
+  filter1 = _mm_srli_epi16(filter1, 3);
+  work_a = _mm_and_si128(work_a, tffe0);  // sign bits for the values < 0
+  filter1 = _mm_and_si128(filter1, t1f);  // clamp the range
+  filter1 = _mm_or_si128(filter1, work_a);  // reinsert the sign bits
+
+  // Filter2 >> 3
+  work_a = _mm_cmpgt_epi16(zero, filter2);
+  filter2 = _mm_srli_epi16(filter2, 3);
+  work_a = _mm_and_si128(work_a, tffe0);
+  filter2 = _mm_and_si128(filter2, t1f);
+  filter2 = _mm_or_si128(filter2, work_a);
+
+  // filt >> 1
+  filt = _mm_adds_epi16(filter1, t1);
+  work_a = _mm_cmpgt_epi16(zero, filt);
+  filt = _mm_srli_epi16(filt, 1);
+  work_a = _mm_and_si128(work_a, tff80);
+  filt = _mm_and_si128(filt, t7f);
+  filt = _mm_or_si128(filt, work_a);
+
+  filt = _mm_andnot_si128(hev, filt);
+
+  q0 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs0, filter1), bd), t80);
+  q1 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs1, filt), bd), t80);
+  p0 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps0, filter2), bd), t80);
+  p1 = _mm_adds_epi16(
+      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps1, filt), bd), t80);
+
+  _mm_storeu_si128((__m128i *)(s - 2 * p), p1);
+  _mm_storeu_si128((__m128i *)(s - 1 * p), p0);
+  _mm_storeu_si128((__m128i *)(s + 0 * p), q0);
+  _mm_storeu_si128((__m128i *)(s + 1 * p), q1);
+}
+
+void vp9_highbd_lpf_horizontal_4_dual_sse2(uint16_t *s, int p,
+                                           const uint8_t *_blimit0,
+                                           const uint8_t *_limit0,
+                                           const uint8_t *_thresh0,
+                                           const uint8_t *_blimit1,
+                                           const uint8_t *_limit1,
+                                           const uint8_t *_thresh1,
+                                           int bd) {
+  vp9_highbd_lpf_horizontal_4_sse2(s, p, _blimit0, _limit0, _thresh0, 1, bd);
+  vp9_highbd_lpf_horizontal_4_sse2(s + 8, p, _blimit1, _limit1, _thresh1, 1,
+                                   bd);
+}
+
+static INLINE void highbd_transpose(uint16_t *src[], int in_p,
+                                    uint16_t *dst[], int out_p,
+                                    int num_8x8_to_transpose) {
+  int idx8x8 = 0;
+  __m128i p0, p1, p2, p3, p4, p5, p6, p7, x0, x1, x2, x3, x4, x5, x6, x7;
+  do {
+    uint16_t *in = src[idx8x8];
+    uint16_t *out = dst[idx8x8];
+
+    p0 = _mm_loadu_si128((__m128i *)(in + 0*in_p));  // 00 01 02 03 04 05 06 07
+    p1 = _mm_loadu_si128((__m128i *)(in + 1*in_p));  // 10 11 12 13 14 15 16 17
+    p2 = _mm_loadu_si128((__m128i *)(in + 2*in_p));  // 20 21 22 23 24 25 26 27
+    p3 = _mm_loadu_si128((__m128i *)(in + 3*in_p));  // 30 31 32 33 34 35 36 37
+    p4 = _mm_loadu_si128((__m128i *)(in + 4*in_p));  // 40 41 42 43 44 45 46 47
+    p5 = _mm_loadu_si128((__m128i *)(in + 5*in_p));  // 50 51 52 53 54 55 56 57
+    p6 = _mm_loadu_si128((__m128i *)(in + 6*in_p));  // 60 61 62 63 64 65 66 67
+    p7 = _mm_loadu_si128((__m128i *)(in + 7*in_p));  // 70 71 72 73 74 75 76 77
+    // 00 10 01 11 02 12 03 13
+    x0 = _mm_unpacklo_epi16(p0, p1);
+    // 20 30 21 31 22 32 23 33
+    x1 = _mm_unpacklo_epi16(p2, p3);
+    // 40 50 41 51 42 52 43 53
+    x2 = _mm_unpacklo_epi16(p4, p5);
+    // 60 70 61 71 62 72 63 73
+    x3 = _mm_unpacklo_epi16(p6, p7);
+    // 00 10 20 30 01 11 21 31
+    x4 = _mm_unpacklo_epi32(x0, x1);
+    // 40 50 60 70 41 51 61 71
+    x5 = _mm_unpacklo_epi32(x2, x3);
+    // 00 10 20 30 40 50 60 70
+    x6 = _mm_unpacklo_epi64(x4, x5);
+    // 01 11 21 31 41 51 61 71
+    x7 = _mm_unpackhi_epi64(x4, x5);
+
+    _mm_storeu_si128((__m128i *)(out + 0*out_p), x6);
+    // 00 10 20 30 40 50 60 70
+    _mm_storeu_si128((__m128i *)(out + 1*out_p), x7);
+    // 01 11 21 31 41 51 61 71
+
+    // 02 12 22 32 03 13 23 33
+    x4 = _mm_unpackhi_epi32(x0, x1);
+    // 42 52 62 72 43 53 63 73
+    x5 = _mm_unpackhi_epi32(x2, x3);
+    // 02 12 22 32 42 52 62 72
+    x6 = _mm_unpacklo_epi64(x4, x5);
+    // 03 13 23 33 43 53 63 73
+    x7 = _mm_unpackhi_epi64(x4, x5);
+
+    _mm_storeu_si128((__m128i *)(out + 2*out_p), x6);
+    // 02 12 22 32 42 52 62 72
+    _mm_storeu_si128((__m128i *)(out + 3*out_p), x7);
+    // 03 13 23 33 43 53 63 73
+
+    // 04 14 05 15 06 16 07 17
+    x0 = _mm_unpackhi_epi16(p0, p1);
+    // 24 34 25 35 26 36 27 37
+    x1 = _mm_unpackhi_epi16(p2, p3);
+    // 44 54 45 55 46 56 47 57
+    x2 = _mm_unpackhi_epi16(p4, p5);
+    // 64 74 65 75 66 76 67 77
+    x3 = _mm_unpackhi_epi16(p6, p7);
+    // 04 14 24 34 05 15 25 35
+    x4 = _mm_unpacklo_epi32(x0, x1);
+    // 44 54 64 74 45 55 65 75
+    x5 = _mm_unpacklo_epi32(x2, x3);
+    // 04 14 24 34 44 54 64 74
+    x6 = _mm_unpacklo_epi64(x4, x5);
+    // 05 15 25 35 45 55 65 75
+    x7 = _mm_unpackhi_epi64(x4, x5);
+
+    _mm_storeu_si128((__m128i *)(out + 4*out_p), x6);
+    // 04 14 24 34 44 54 64 74
+    _mm_storeu_si128((__m128i *)(out + 5*out_p), x7);
+    // 05 15 25 35 45 55 65 75
+
+    // 06 16 26 36 07 17 27 37
+    x4 = _mm_unpackhi_epi32(x0, x1);
+    // 46 56 66 76 47 57 67 77
+    x5 = _mm_unpackhi_epi32(x2, x3);
+    // 06 16 26 36 46 56 66 76
+    x6 = _mm_unpacklo_epi64(x4, x5);
+    // 07 17 27 37 47 57 67 77
+    x7 = _mm_unpackhi_epi64(x4, x5);
+
+    _mm_storeu_si128((__m128i *)(out + 6*out_p), x6);
+    // 06 16 26 36 46 56 66 76
+    _mm_storeu_si128((__m128i *)(out + 7*out_p), x7);
+    // 07 17 27 37 47 57 67 77
+  } while (++idx8x8 < num_8x8_to_transpose);
+}
+
+static INLINE void highbd_transpose8x16(uint16_t *in0, uint16_t *in1,
+                                        int in_p, uint16_t *out, int out_p) {
+  uint16_t *src0[1];
+  uint16_t *src1[1];
+  uint16_t *dest0[1];
+  uint16_t *dest1[1];
+  src0[0] = in0;
+  src1[0] = in1;
+  dest0[0] = out;
+  dest1[0] = out + 8;
+  highbd_transpose(src0, in_p, dest0, out_p, 1);
+  highbd_transpose(src1, in_p, dest1, out_p, 1);
+}
+
+void vp9_highbd_lpf_vertical_4_sse2(uint16_t *s, int p,
+                                    const uint8_t *blimit,
+                                    const uint8_t *limit,
+                                    const uint8_t *thresh,
+                                    int count, int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, t_dst, 8 * 8);
+  uint16_t *src[1];
+  uint16_t *dst[1];
+  (void)count;
+
+  // Transpose 8x8
+  src[0] = s - 4;
+  dst[0] = t_dst;
+
+  highbd_transpose(src, p, dst, 8, 1);
+
+  // Loop filtering
+  vp9_highbd_lpf_horizontal_4_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1,
+                                   bd);
+
+  src[0] = t_dst;
+  dst[0] = s - 4;
+
+  // Transpose back
+  highbd_transpose(src, 8, dst, p, 1);
+}
+
+void vp9_highbd_lpf_vertical_4_dual_sse2(uint16_t *s, int p,
+                                         const uint8_t *blimit0,
+                                         const uint8_t *limit0,
+                                         const uint8_t *thresh0,
+                                         const uint8_t *blimit1,
+                                         const uint8_t *limit1,
+                                         const uint8_t *thresh1,
+                                         int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, t_dst, 16 * 8);
+  uint16_t *src[2];
+  uint16_t *dst[2];
+
+  // Transpose 8x16
+  highbd_transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16);
+
+  // Loop filtering
+  vp9_highbd_lpf_horizontal_4_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0,
+                                        thresh0, blimit1, limit1, thresh1, bd);
+  src[0] = t_dst;
+  src[1] = t_dst + 8;
+  dst[0] = s - 4;
+  dst[1] = s - 4 + p * 8;
+
+  // Transpose back
+  highbd_transpose(src, 16, dst, p, 2);
+}
+
+void vp9_highbd_lpf_vertical_8_sse2(uint16_t *s, int p,
+                                    const uint8_t *blimit,
+                                    const uint8_t *limit,
+                                    const uint8_t *thresh,
+                                    int count, int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, t_dst, 8 * 8);
+  uint16_t *src[1];
+  uint16_t *dst[1];
+  (void)count;
+
+  // Transpose 8x8
+  src[0] = s - 4;
+  dst[0] = t_dst;
+
+  highbd_transpose(src, p, dst, 8, 1);
+
+  // Loop filtering
+  vp9_highbd_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1,
+                                   bd);
+
+  src[0] = t_dst;
+  dst[0] = s - 4;
+
+  // Transpose back
+  highbd_transpose(src, 8, dst, p, 1);
+}
+
+void vp9_highbd_lpf_vertical_8_dual_sse2(uint16_t *s, int p,
+                                         const uint8_t *blimit0,
+                                         const uint8_t *limit0,
+                                         const uint8_t *thresh0,
+                                         const uint8_t *blimit1,
+                                         const uint8_t *limit1,
+                                         const uint8_t *thresh1,
+                                         int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, t_dst, 16 * 8);
+  uint16_t *src[2];
+  uint16_t *dst[2];
+
+  // Transpose 8x16
+  highbd_transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16);
+
+  // Loop filtering
+  vp9_highbd_lpf_horizontal_8_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0,
+                                        thresh0, blimit1, limit1, thresh1, bd);
+  src[0] = t_dst;
+  src[1] = t_dst + 8;
+
+  dst[0] = s - 4;
+  dst[1] = s - 4 + p * 8;
+
+  // Transpose back
+  highbd_transpose(src, 16, dst, p, 2);
+}
+
+void vp9_highbd_lpf_vertical_16_sse2(uint16_t *s, int p,
+                                     const uint8_t *blimit,
+                                     const uint8_t *limit,
+                                     const uint8_t *thresh,
+                                     int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, t_dst, 8 * 16);
+  uint16_t *src[2];
+  uint16_t *dst[2];
+
+  src[0] = s - 8;
+  src[1] = s;
+  dst[0] = t_dst;
+  dst[1] = t_dst + 8 * 8;
+
+  // Transpose 16x8
+  highbd_transpose(src, p, dst, 8, 2);
+
+  // Loop filtering
+  highbd_mb_lpf_horizontal_edge_w_sse2_8(t_dst + 8 * 8, 8, blimit, limit,
+                                         thresh, bd);
+  src[0] = t_dst;
+  src[1] = t_dst + 8 * 8;
+  dst[0] = s - 8;
+  dst[1] = s;
+
+  // Transpose back
+  highbd_transpose(src, 8, dst, p, 2);
+}
+
+void vp9_highbd_lpf_vertical_16_dual_sse2(uint16_t *s,
+                                          int p,
+                                          const uint8_t *blimit,
+                                          const uint8_t *limit,
+                                          const uint8_t *thresh,
+                                          int bd) {
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, t_dst, 256);
+
+  //  Transpose 16x16
+  highbd_transpose8x16(s - 8, s - 8 + 8 * p, p, t_dst, 16);
+  highbd_transpose8x16(s, s + 8 * p, p, t_dst + 8 * 16, 16);
+
+  //  Loop filtering
+  highbd_mb_lpf_horizontal_edge_w_sse2_16(t_dst + 8 * 16, 16, blimit, limit,
+                                          thresh, bd);
+
+  //  Transpose back
+  highbd_transpose8x16(t_dst, t_dst + 8 * 16, 16, s - 8, p);
+  highbd_transpose8x16(t_dst + 8, t_dst + 8 + 8 * 16, 16, s - 8 + 8 * p, p);
+}
diff --git a/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm b/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm
new file mode 100644
index 0000000..4bdbb83
--- /dev/null
+++ b/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm
@@ -0,0 +1,962 @@
+;
+;  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;Note: tap3 and tap4 have to be applied and added after other taps to avoid
+;overflow.
+
+%macro HIGH_GET_FILTERS_4 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm7, [rdx]                 ;load filters
+    pshuflw     xmm0, xmm7, 0b              ;k0
+    pshuflw     xmm1, xmm7, 01010101b       ;k1
+    pshuflw     xmm2, xmm7, 10101010b       ;k2
+    pshuflw     xmm3, xmm7, 11111111b       ;k3
+    psrldq      xmm7, 8
+    pshuflw     xmm4, xmm7, 0b              ;k4
+    pshuflw     xmm5, xmm7, 01010101b       ;k5
+    pshuflw     xmm6, xmm7, 10101010b       ;k6
+    pshuflw     xmm7, xmm7, 11111111b       ;k7
+
+    punpcklwd   xmm0, xmm6
+    punpcklwd   xmm2, xmm5
+    punpcklwd   xmm3, xmm4
+    punpcklwd   xmm1, xmm7
+
+    movdqa      k0k6, xmm0
+    movdqa      k2k5, xmm2
+    movdqa      k3k4, xmm3
+    movdqa      k1k7, xmm1
+
+    movq        xmm6, rcx
+    pshufd      xmm6, xmm6, 0
+    movdqa      krd, xmm6
+
+    ;Compute max and min values of a pixel
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)      ;bps
+    movq        xmm0, rdx
+    movq        xmm1, rcx
+    pshufd      xmm0, xmm0, 0b
+    movdqa      xmm2, xmm0
+    psllw       xmm0, xmm1
+    psubw       xmm0, xmm2
+    pxor        xmm1, xmm1
+    movdqa      max, xmm0                  ;max value (for clamping)
+    movdqa      min, xmm1                  ;min value (for clamping)
+
+%endm
+
+%macro HIGH_APPLY_FILTER_4 1
+    punpcklwd   xmm0, xmm6                  ;two row in one register
+    punpcklwd   xmm1, xmm7
+    punpcklwd   xmm2, xmm5
+    punpcklwd   xmm3, xmm4
+
+    pmaddwd     xmm0, k0k6                  ;multiply the filter factors
+    pmaddwd     xmm1, k1k7
+    pmaddwd     xmm2, k2k5
+    pmaddwd     xmm3, k3k4
+
+    paddd       xmm0, xmm1                  ;sum
+    paddd       xmm0, xmm2
+    paddd       xmm0, xmm3
+
+    paddd       xmm0, krd                   ;rounding
+    psrad       xmm0, 7                     ;shift
+    packssdw    xmm0, xmm0                  ;pack to word
+
+    ;clamp the values
+    pminsw      xmm0, max
+    pmaxsw      xmm0, min
+
+%if %1
+    movq        xmm1, [rdi]
+    pavgw       xmm0, xmm1
+%endif
+    movq        [rdi], xmm0
+%endm
+
+%macro HIGH_GET_FILTERS 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm7, [rdx]                 ;load filters
+    pshuflw     xmm0, xmm7, 0b              ;k0
+    pshuflw     xmm1, xmm7, 01010101b       ;k1
+    pshuflw     xmm2, xmm7, 10101010b       ;k2
+    pshuflw     xmm3, xmm7, 11111111b       ;k3
+    pshufhw     xmm4, xmm7, 0b              ;k4
+    pshufhw     xmm5, xmm7, 01010101b       ;k5
+    pshufhw     xmm6, xmm7, 10101010b       ;k6
+    pshufhw     xmm7, xmm7, 11111111b       ;k7
+    punpcklqdq  xmm2, xmm2
+    punpcklqdq  xmm3, xmm3
+    punpcklwd   xmm0, xmm1
+    punpckhwd   xmm6, xmm7
+    punpckhwd   xmm2, xmm5
+    punpckhwd   xmm3, xmm4
+
+    movdqa      k0k1, xmm0                  ;store filter factors on stack
+    movdqa      k6k7, xmm6
+    movdqa      k2k5, xmm2
+    movdqa      k3k4, xmm3
+
+    movq        xmm6, rcx
+    pshufd      xmm6, xmm6, 0
+    movdqa      krd, xmm6                   ;rounding
+
+    ;Compute max and min values of a pixel
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)       ;bps
+    movq        xmm0, rdx
+    movq        xmm1, rcx
+    pshufd      xmm0, xmm0, 0b
+    movdqa      xmm2, xmm0
+    psllw       xmm0, xmm1
+    psubw       xmm0, xmm2
+    pxor        xmm1, xmm1
+    movdqa      max, xmm0                  ;max value (for clamping)
+    movdqa      min, xmm1                  ;min value (for clamping)
+%endm
+
+%macro LOAD_VERT_8 1
+    movdqu      xmm0, [rsi + %1]            ;0
+    movdqu      xmm1, [rsi + rax + %1]      ;1
+    movdqu      xmm6, [rsi + rdx * 2 + %1]  ;6
+    lea         rsi,  [rsi + rax]
+    movdqu      xmm7, [rsi + rdx * 2 + %1]  ;7
+    movdqu      xmm2, [rsi + rax + %1]      ;2
+    movdqu      xmm3, [rsi + rax * 2 + %1]  ;3
+    movdqu      xmm4, [rsi + rdx + %1]      ;4
+    movdqu      xmm5, [rsi + rax * 4 + %1]  ;5
+%endm
+
+%macro HIGH_APPLY_FILTER_8 2
+    movdqu      temp, xmm4
+    movdqa      xmm4, xmm0
+    punpcklwd   xmm0, xmm1
+    punpckhwd   xmm4, xmm1
+    movdqa      xmm1, xmm6
+    punpcklwd   xmm6, xmm7
+    punpckhwd   xmm1, xmm7
+    movdqa      xmm7, xmm2
+    punpcklwd   xmm2, xmm5
+    punpckhwd   xmm7, xmm5
+
+    movdqu      xmm5, temp
+    movdqu      temp, xmm4
+    movdqa      xmm4, xmm3
+    punpcklwd   xmm3, xmm5
+    punpckhwd   xmm4, xmm5
+    movdqu      xmm5, temp
+
+    pmaddwd     xmm0, k0k1
+    pmaddwd     xmm5, k0k1
+    pmaddwd     xmm6, k6k7
+    pmaddwd     xmm1, k6k7
+    pmaddwd     xmm2, k2k5
+    pmaddwd     xmm7, k2k5
+    pmaddwd     xmm3, k3k4
+    pmaddwd     xmm4, k3k4
+
+    paddd       xmm0, xmm6
+    paddd       xmm0, xmm2
+    paddd       xmm0, xmm3
+    paddd       xmm5, xmm1
+    paddd       xmm5, xmm7
+    paddd       xmm5, xmm4
+
+    paddd       xmm0, krd                   ;rounding
+    paddd       xmm5, krd
+    psrad       xmm0, 7                     ;shift
+    psrad       xmm5, 7
+    packssdw    xmm0, xmm5                  ;pack back to word
+
+    ;clamp the values
+    pminsw      xmm0, max
+    pmaxsw      xmm0, min
+
+%if %1
+    movdqu      xmm1, [rdi + %2]
+    pavgw       xmm0, xmm1
+%endif
+    movdqu      [rdi + %2], xmm0
+%endm
+
+;void vp9_filter_block1d4_v8_sse2
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d4_v8_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movq        xmm0, [rsi]                 ;load src: row 0
+    movq        xmm1, [rsi + rax]           ;1
+    movq        xmm6, [rsi + rdx * 2]       ;6
+    lea         rsi,  [rsi + rax]
+    movq        xmm7, [rsi + rdx * 2]       ;7
+    movq        xmm2, [rsi + rax]           ;2
+    movq        xmm3, [rsi + rax * 2]       ;3
+    movq        xmm4, [rsi + rdx]           ;4
+    movq        xmm5, [rsi + rax * 4]       ;5
+
+    HIGH_APPLY_FILTER_4 0
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d8_v8_sse2
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d8_v8_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 0, 0
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d16_v8_sse2
+;(
+;    unsigned char *src_ptr,
+;    unsigned int   src_pitch,
+;    unsigned char *output_ptr,
+;    unsigned int   out_pitch,
+;    unsigned int   output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d16_v8_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 0, 0
+    sub         rsi, rax
+
+    LOAD_VERT_8 16
+    HIGH_APPLY_FILTER_8 0, 16
+    add         rdi, rbx
+
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d4_v8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movq        xmm0, [rsi]                 ;load src: row 0
+    movq        xmm1, [rsi + rax]           ;1
+    movq        xmm6, [rsi + rdx * 2]       ;6
+    lea         rsi,  [rsi + rax]
+    movq        xmm7, [rsi + rdx * 2]       ;7
+    movq        xmm2, [rsi + rax]           ;2
+    movq        xmm3, [rsi + rax * 2]       ;3
+    movq        xmm4, [rsi + rdx]           ;4
+    movq        xmm5, [rsi + rax * 4]       ;5
+
+    HIGH_APPLY_FILTER_4 1
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d8_v8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 1, 0
+
+    lea         rdi, [rdi + rbx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_v8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rbx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rbx, [rbx + rbx]
+    lea         rdx, [rax + rax * 2]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+.loop:
+    LOAD_VERT_8 0
+    HIGH_APPLY_FILTER_8 1, 0
+    sub         rsi, rax
+
+    LOAD_VERT_8 16
+    HIGH_APPLY_FILTER_8 1, 16
+    add         rdi, rbx
+
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+    pop rbx
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d4_h8_sse2
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d4_h8_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm4,   [rsi + 2]
+    movdqa      xmm1, xmm0
+    movdqa      xmm6, xmm4
+    movdqa      xmm7, xmm4
+    movdqa      xmm2, xmm0
+    movdqa      xmm3, xmm0
+    movdqa      xmm5, xmm4
+
+    psrldq      xmm1, 2
+    psrldq      xmm6, 4
+    psrldq      xmm7, 6
+    psrldq      xmm2, 4
+    psrldq      xmm3, 6
+    psrldq      xmm5, 2
+
+    HIGH_APPLY_FILTER_4 0
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d8_h8_sse2
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d8_h8_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 0, 0
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void vp9_filter_block1d16_h8_sse2
+;(
+;    unsigned char  *src_ptr,
+;    unsigned int    src_pixels_per_line,
+;    unsigned char  *output_ptr,
+;    unsigned int    output_pitch,
+;    unsigned int    output_height,
+;    short *filter
+;)
+global sym(vp9_high_filter_block1d16_h8_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 0, 0
+
+    movdqu      xmm0,   [rsi + 10]           ;load src
+    movdqu      xmm1,   [rsi + 12]
+    movdqu      xmm2,   [rsi + 14]
+    movdqu      xmm3,   [rsi + 16]
+    movdqu      xmm4,   [rsi + 18]
+    movdqu      xmm5,   [rsi + 20]
+    movdqu      xmm6,   [rsi + 22]
+    movdqu      xmm7,   [rsi + 24]
+
+    HIGH_APPLY_FILTER_8 0, 16
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d4_h8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 7
+    %define k0k6 [rsp + 16 * 0]
+    %define k2k5 [rsp + 16 * 1]
+    %define k3k4 [rsp + 16 * 2]
+    %define k1k7 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define max [rsp + 16 * 5]
+    %define min [rsp + 16 * 6]
+
+    HIGH_GET_FILTERS_4
+
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm4,   [rsi + 2]
+    movdqa      xmm1, xmm0
+    movdqa      xmm6, xmm4
+    movdqa      xmm7, xmm4
+    movdqa      xmm2, xmm0
+    movdqa      xmm3, xmm0
+    movdqa      xmm5, xmm4
+
+    psrldq      xmm1, 2
+    psrldq      xmm6, 4
+    psrldq      xmm7, 6
+    psrldq      xmm2, 4
+    psrldq      xmm3, 6
+    psrldq      xmm5, 2
+
+    HIGH_APPLY_FILTER_4 1
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 7
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d8_h8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 1, 0
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_h8_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h8_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 8
+    %define k0k1 [rsp + 16 * 0]
+    %define k6k7 [rsp + 16 * 1]
+    %define k2k5 [rsp + 16 * 2]
+    %define k3k4 [rsp + 16 * 3]
+    %define krd [rsp + 16 * 4]
+    %define temp [rsp + 16 * 5]
+    %define max [rsp + 16 * 6]
+    %define min [rsp + 16 * 7]
+
+    HIGH_GET_FILTERS
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    lea         rax, [rax + rax]            ;bytes per line
+    lea         rdx, [rdx + rdx]
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+
+.loop:
+    movdqu      xmm0,   [rsi - 6]           ;load src
+    movdqu      xmm1,   [rsi - 4]
+    movdqu      xmm2,   [rsi - 2]
+    movdqu      xmm3,   [rsi]
+    movdqu      xmm4,   [rsi + 2]
+    movdqu      xmm5,   [rsi + 4]
+    movdqu      xmm6,   [rsi + 6]
+    movdqu      xmm7,   [rsi + 8]
+
+    HIGH_APPLY_FILTER_8 1, 0
+
+    movdqu      xmm0,   [rsi + 10]           ;load src
+    movdqu      xmm1,   [rsi + 12]
+    movdqu      xmm2,   [rsi + 14]
+    movdqu      xmm3,   [rsi + 16]
+    movdqu      xmm4,   [rsi + 18]
+    movdqu      xmm5,   [rsi + 20]
+    movdqu      xmm6,   [rsi + 22]
+    movdqu      xmm7,   [rsi + 24]
+
+    HIGH_APPLY_FILTER_8 1, 16
+
+    lea         rsi, [rsi + rax]
+    lea         rdi, [rdi + rdx]
+    dec         rcx
+    jnz         .loop
+
+    add rsp, 16 * 8
+    pop rsp
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm b/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm
new file mode 100644
index 0000000..b7d4a61
--- /dev/null
+++ b/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm
@@ -0,0 +1,494 @@
+;
+;  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro HIGH_GET_PARAM_4 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm3, [rdx]                 ;load filters
+    pshuflw     xmm4, xmm3, 11111111b       ;k3
+    psrldq      xmm3, 8
+    pshuflw     xmm3, xmm3, 0b              ;k4
+    punpcklwd   xmm4, xmm3                  ;k3k4
+
+    movq        xmm3, rcx                   ;rounding
+    pshufd      xmm3, xmm3, 0
+
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)       ;bps
+    movq        xmm5, rdx
+    movq        xmm2, rcx
+    pshufd      xmm5, xmm5, 0b
+    movdqa      xmm1, xmm5
+    psllw       xmm5, xmm2
+    psubw       xmm5, xmm1                  ;max value (for clamping)
+    pxor        xmm2, xmm2                  ;min value (for clamping)
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+%endm
+
+%macro HIGH_APPLY_FILTER_4 1
+
+    punpcklwd   xmm0, xmm1                  ;two row in one register
+    pmaddwd     xmm0, xmm4                  ;multiply the filter factors
+
+    paddd       xmm0, xmm3                  ;rounding
+    psrad       xmm0, 7                     ;shift
+    packssdw    xmm0, xmm0                  ;pack to word
+
+    ;clamp the values
+    pminsw      xmm0, xmm5
+    pmaxsw      xmm0, xmm2
+
+%if %1
+    movq        xmm1, [rdi]
+    pavgw       xmm0, xmm1
+%endif
+
+    movq        [rdi], xmm0
+    lea         rsi, [rsi + 2*rax]
+    lea         rdi, [rdi + 2*rdx]
+    dec         rcx
+%endm
+
+%if ARCH_X86_64
+%macro HIGH_GET_PARAM 0
+    mov         rdx, arg(5)                 ;filter ptr
+    mov         rsi, arg(0)                 ;src_ptr
+    mov         rdi, arg(2)                 ;output_ptr
+    mov         rcx, 0x00000040
+
+    movdqa      xmm6, [rdx]                 ;load filters
+
+    pshuflw     xmm7, xmm6, 11111111b       ;k3
+    pshufhw     xmm6, xmm6, 0b              ;k4
+    psrldq      xmm6, 8
+    punpcklwd   xmm7, xmm6                  ;k3k4k3k4k3k4k3k4
+
+    movq        xmm4, rcx                   ;rounding
+    pshufd      xmm4, xmm4, 0
+
+    mov         rdx, 0x00010001
+    movsxd      rcx, DWORD PTR arg(6)       ;bps
+    movq        xmm8, rdx
+    movq        xmm5, rcx
+    pshufd      xmm8, xmm8, 0b
+    movdqa      xmm1, xmm8
+    psllw       xmm8, xmm5
+    psubw       xmm8, xmm1                  ;max value (for clamping)
+    pxor        xmm5, xmm5                  ;min value (for clamping)
+
+    movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
+    movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
+    movsxd      rcx, DWORD PTR arg(4)       ;output_height
+%endm
+
+%macro HIGH_APPLY_FILTER_8 1
+    movdqa      xmm6, xmm0
+    punpckhwd   xmm6, xmm1
+    punpcklwd   xmm0, xmm1
+    pmaddwd     xmm6, xmm7
+    pmaddwd     xmm0, xmm7
+
+    paddd       xmm6, xmm4                  ;rounding
+    paddd       xmm0, xmm4                  ;rounding
+    psrad       xmm6, 7                     ;shift
+    psrad       xmm0, 7                     ;shift
+    packssdw    xmm0, xmm6                  ;pack back to word
+
+    ;clamp the values
+    pminsw      xmm0, xmm8
+    pmaxsw      xmm0, xmm5
+
+%if %1
+    movdqu      xmm1, [rdi]
+    pavgw       xmm0, xmm1
+%endif
+    movdqu      [rdi], xmm0                 ;store the result
+
+    lea         rsi, [rsi + 2*rax]
+    lea         rdi, [rdi + 2*rdx]
+    dec         rcx
+%endm
+
+%macro HIGH_APPLY_FILTER_16 1
+    movdqa      xmm9, xmm0
+    movdqa      xmm6, xmm2
+    punpckhwd   xmm9, xmm1
+    punpckhwd   xmm6, xmm3
+    punpcklwd   xmm0, xmm1
+    punpcklwd   xmm2, xmm3
+
+    pmaddwd     xmm9, xmm7
+    pmaddwd     xmm6, xmm7
+    pmaddwd     xmm0, xmm7
+    pmaddwd     xmm2, xmm7
+
+    paddd       xmm9, xmm4                  ;rounding
+    paddd       xmm6, xmm4
+    paddd       xmm0, xmm4
+    paddd       xmm2, xmm4
+
+    psrad       xmm9, 7                     ;shift
+    psrad       xmm6, 7
+    psrad       xmm0, 7
+    psrad       xmm2, 7
+
+    packssdw    xmm0, xmm9                  ;pack back to word
+    packssdw    xmm2, xmm6                  ;pack back to word
+
+    ;clamp the values
+    pminsw      xmm0, xmm8
+    pmaxsw      xmm0, xmm5
+    pminsw      xmm2, xmm8
+    pmaxsw      xmm2, xmm5
+
+%if %1
+    movdqu      xmm1, [rdi]
+    movdqu      xmm3, [rdi + 16]
+    pavgw       xmm0, xmm1
+    pavgw       xmm2, xmm3
+%endif
+    movdqu      [rdi], xmm0               ;store the result
+    movdqu      [rdi + 16], xmm2          ;store the result
+
+    lea         rsi, [rsi + 2*rax]
+    lea         rdi, [rdi + 2*rdx]
+    dec         rcx
+%endm
+%endif
+
+global sym(vp9_high_filter_block1d4_v2_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movq        xmm0, [rsi]                 ;load src
+    movq        xmm1, [rsi + 2*rax]
+
+    HIGH_APPLY_FILTER_4 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_v2_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;0
+    movdqu      xmm1, [rsi + 2*rax]         ;1
+
+    HIGH_APPLY_FILTER_8 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_v2_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu        xmm0, [rsi]               ;0
+    movdqu        xmm2, [rsi + 16]
+    movdqu        xmm1, [rsi + 2*rax]       ;1
+    movdqu        xmm3, [rsi + 2*rax + 16]
+
+    HIGH_APPLY_FILTER_16 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
+
+global sym(vp9_high_filter_block1d4_v2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_v2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movq        xmm0, [rsi]                 ;load src
+    movq        xmm1, [rsi + 2*rax]
+
+    HIGH_APPLY_FILTER_4 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_v2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_v2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;0
+    movdqu      xmm1, [rsi + 2*rax]         ;1
+
+    HIGH_APPLY_FILTER_8 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_v2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_v2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu        xmm0, [rsi]               ;0
+    movdqu        xmm1, [rsi + 2*rax]       ;1
+    movdqu        xmm2, [rsi + 16]
+    movdqu        xmm3, [rsi + 2*rax + 16]
+
+    HIGH_APPLY_FILTER_16 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
+
+global sym(vp9_high_filter_block1d4_h2_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqa      xmm1, xmm0
+    psrldq      xmm1, 2
+
+    HIGH_APPLY_FILTER_4 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_h2_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqu      xmm1, [rsi + 2]
+
+    HIGH_APPLY_FILTER_8 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_h2_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h2_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0,   [rsi]               ;load src
+    movdqu      xmm1,   [rsi + 2]
+    movdqu      xmm2,   [rsi + 16]
+    movdqu      xmm3,   [rsi + 18]
+
+    HIGH_APPLY_FILTER_16 0
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
+
+global sym(vp9_high_filter_block1d4_h2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d4_h2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM_4
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqa      xmm1, xmm0
+    psrldq      xmm1, 2
+
+    HIGH_APPLY_FILTER_4 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+%if ARCH_X86_64
+global sym(vp9_high_filter_block1d8_h2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d8_h2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 8
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0, [rsi]                 ;load src
+    movdqu      xmm1, [rsi + 2]
+
+    HIGH_APPLY_FILTER_8 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+global sym(vp9_high_filter_block1d16_h2_avg_sse2) PRIVATE
+sym(vp9_high_filter_block1d16_h2_avg_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 9
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    HIGH_GET_PARAM
+.loop:
+    movdqu      xmm0,   [rsi]               ;load src
+    movdqu      xmm1,   [rsi + 2]
+    movdqu      xmm2,   [rsi + 16]
+    movdqu      xmm3,   [rsi + 18]
+
+    HIGH_APPLY_FILTER_16 1
+    jnz         .loop
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+%endif
diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c
index b60f8a0..df60987 100644
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -3573,6 +3573,7 @@
                                  int stride) {
   const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
   const __m128i final_rounding = _mm_set1_epi16(1<<5);
+  const __m128i zero = _mm_setzero_si128();
 
   // idct constants for each stage
   const __m128i stg1_0 = pair_set_epi16(cospi_31_64, -cospi_1_64);
@@ -3635,7 +3636,6 @@
           stp2_30, stp2_31;
   __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
   int i, j, i32;
-  int zero_flag[2];
 
   for (i = 0; i < 4; i++) {
     i32 = (i << 5);
@@ -3710,13 +3710,7 @@
       zero_idx[13] = _mm_or_si128(zero_idx[10], zero_idx[11]);
       zero_idx[14] = _mm_or_si128(zero_idx[12], zero_idx[13]);
 
-      zero_idx[0] = _mm_unpackhi_epi64(zero_idx[14], zero_idx[14]);
-      zero_idx[1] = _mm_or_si128(zero_idx[0], zero_idx[14]);
-      zero_idx[2] = _mm_srli_epi64(zero_idx[1], 32);
-      zero_flag[0] = _mm_cvtsi128_si32(zero_idx[1]);
-      zero_flag[1] = _mm_cvtsi128_si32(zero_idx[2]);
-
-      if (!zero_flag[0] && !zero_flag[1]) {
+      if (_mm_movemask_epi8(_mm_cmpeq_epi32(zero_idx[14], zero)) == 0xFFFF) {
         col[i32 + 0] = _mm_setzero_si128();
         col[i32 + 1] = _mm_setzero_si128();
         col[i32 + 2] = _mm_setzero_si128();
@@ -3795,7 +3789,6 @@
       col[i32 + 31] = _mm_sub_epi16(stp1_0, stp1_31);
     }
   for (i = 0; i < 4; i++) {
-      const __m128i zero = _mm_setzero_si128();
       // Second 1-D idct
       j = i << 3;
 
diff --git a/vp9/common/x86/vp9_loopfilter_intrin_sse2.c b/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
index 448ad5a..320328e 100644
--- a/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
+++ b/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
@@ -12,6 +12,10 @@
 #include "vp9/common/vp9_loopfilter.h"
 #include "vpx_ports/emmintrin_compat.h"
 
+static INLINE __m128i abs_diff(__m128i a, __m128i b) {
+  return _mm_or_si128(_mm_subs_epu8(a, b), _mm_subs_epu8(b, a));
+}
+
 static void mb_lpf_horizontal_edge_w_sse2_8(unsigned char *s,
                                             int p,
                                             const unsigned char *_blimit,
@@ -46,15 +50,12 @@
 
   {
     __m128i abs_p1q1, abs_p0q0, abs_q1q0, fe, ff, work;
-    abs_p1p0 = _mm_or_si128(_mm_subs_epu8(q1p1, q0p0),
-                            _mm_subs_epu8(q0p0, q1p1));
+    abs_p1p0 = abs_diff(q1p1, q0p0);
     abs_q1q0 =  _mm_srli_si128(abs_p1p0, 8);
     fe = _mm_set1_epi8(0xfe);
     ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-    abs_p0q0 = _mm_or_si128(_mm_subs_epu8(q0p0, p0q0),
-                            _mm_subs_epu8(p0q0, q0p0));
-    abs_p1q1 = _mm_or_si128(_mm_subs_epu8(q1p1, p1q1),
-                            _mm_subs_epu8(p1q1, q1p1));
+    abs_p0q0 = abs_diff(q0p0, p0q0);
+    abs_p1q1 = abs_diff(q1p1, p1q1);
     flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
     hev = _mm_subs_epu8(flat, thresh);
     hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
@@ -68,10 +69,8 @@
     // mask |= (abs(p1 - p0) > limit) * -1;
     // mask |= (abs(q1 - q0) > limit) * -1;
 
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2p2, q1p1),
-                                     _mm_subs_epu8(q1p1, q2p2)),
-                        _mm_or_si128(_mm_subs_epu8(q3p3, q2p2),
-                                     _mm_subs_epu8(q2p2, q3p3)));
+    work = _mm_max_epu8(abs_diff(q2p2, q1p1),
+                        abs_diff(q3p3, q2p2));
     mask = _mm_max_epu8(work, mask);
     mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8));
     mask = _mm_subs_epu8(mask, limit);
@@ -125,10 +124,7 @@
 
     {
       __m128i work;
-      flat = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2p2, q0p0),
-                                       _mm_subs_epu8(q0p0, q2p2)),
-                          _mm_or_si128(_mm_subs_epu8(q3p3, q0p0),
-                                       _mm_subs_epu8(q0p0, q3p3)));
+      flat = _mm_max_epu8(abs_diff(q2p2, q0p0), abs_diff(q3p3, q0p0));
       flat = _mm_max_epu8(abs_p1p0, flat);
       flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8));
       flat = _mm_subs_epu8(flat, one);
@@ -142,21 +138,12 @@
       q6p6 = _mm_loadl_epi64((__m128i *)(s - 7 * p));
       q6p6 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q6p6),
                                            (__m64 *)(s + 6 * p)));
-
-      flat2 = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q4p4, q0p0),
-                                        _mm_subs_epu8(q0p0, q4p4)),
-                           _mm_or_si128(_mm_subs_epu8(q5p5, q0p0),
-                                        _mm_subs_epu8(q0p0, q5p5)));
+      flat2 = _mm_max_epu8(abs_diff(q4p4, q0p0), abs_diff(q5p5, q0p0));
 
       q7p7 = _mm_loadl_epi64((__m128i *)(s - 8 * p));
       q7p7 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q7p7),
                                            (__m64 *)(s + 7 * p)));
-
-      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q6p6, q0p0),
-                                       _mm_subs_epu8(q0p0, q6p6)),
-                          _mm_or_si128(_mm_subs_epu8(q7p7, q0p0),
-                                       _mm_subs_epu8(q0p0, q7p7)));
-
+      work = _mm_max_epu8(abs_diff(q6p6, q0p0), abs_diff(q7p7, q0p0));
       flat2 = _mm_max_epu8(work, flat2);
       flat2 = _mm_max_epu8(flat2, _mm_srli_si128(flat2, 8));
       flat2 = _mm_subs_epu8(flat2, one);
@@ -364,20 +351,41 @@
   }
 }
 
+static INLINE __m128i filter_add2_sub2(const __m128i *const total,
+                                       const __m128i *const a1,
+                                       const __m128i *const a2,
+                                       const __m128i *const s1,
+                                       const __m128i *const s2) {
+  __m128i x = _mm_add_epi16(*a1, *total);
+  x = _mm_add_epi16(_mm_sub_epi16(x, _mm_add_epi16(*s1, *s2)), *a2);
+  return x;
+}
+
+static INLINE __m128i filter8_mask(const __m128i *const flat,
+                                   const __m128i *const other_filt,
+                                   const __m128i *const f8_lo,
+                                   const __m128i *const f8_hi) {
+  const __m128i f8 = _mm_packus_epi16(_mm_srli_epi16(*f8_lo, 3),
+                                      _mm_srli_epi16(*f8_hi, 3));
+  const __m128i result = _mm_and_si128(*flat, f8);
+  return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result);
+}
+
+static INLINE __m128i filter16_mask(const __m128i *const flat,
+                                    const __m128i *const other_filt,
+                                    const __m128i *const f_lo,
+                                    const __m128i *const f_hi) {
+  const __m128i f = _mm_packus_epi16(_mm_srli_epi16(*f_lo, 4),
+                                     _mm_srli_epi16(*f_hi, 4));
+  const __m128i result = _mm_and_si128(*flat, f);
+  return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result);
+}
+
 static void mb_lpf_horizontal_edge_w_sse2_16(unsigned char *s,
                                              int p,
                                              const unsigned char *_blimit,
                                              const unsigned char *_limit,
                                              const unsigned char *_thresh) {
-  DECLARE_ALIGNED_ARRAY(16, unsigned char, flat2_op, 7 * 16);
-  DECLARE_ALIGNED_ARRAY(16, unsigned char, flat2_oq, 7 * 16);
-
-  DECLARE_ALIGNED_ARRAY(16, unsigned char, flat_op, 3 * 16);
-  DECLARE_ALIGNED_ARRAY(16, unsigned char, flat_oq, 3 * 16);
-
-  DECLARE_ALIGNED_ARRAY(16, unsigned char, ap, 8 * 16);
-  DECLARE_ALIGNED_ARRAY(16, unsigned char, aq, 8 * 16);
-
   const __m128i zero = _mm_set1_epi16(0);
   const __m128i one = _mm_set1_epi8(1);
   const __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
@@ -387,8 +395,14 @@
   __m128i p7, p6, p5;
   __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4;
   __m128i q5, q6, q7;
-  int i = 0;
 
+  __m128i op2, op1, op0, oq0, oq1, oq2;
+
+  __m128i max_abs_p1p0q1q0;
+
+  p7 = _mm_loadu_si128((__m128i *)(s - 8 * p));
+  p6 = _mm_loadu_si128((__m128i *)(s - 7 * p));
+  p5 = _mm_loadu_si128((__m128i *)(s - 6 * p));
   p4 = _mm_loadu_si128((__m128i *)(s - 5 * p));
   p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
   p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
@@ -399,58 +413,59 @@
   q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
   q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
   q4 = _mm_loadu_si128((__m128i *)(s + 4 * p));
-
-  _mm_store_si128((__m128i *)&ap[4 * 16], p4);
-  _mm_store_si128((__m128i *)&ap[3 * 16], p3);
-  _mm_store_si128((__m128i *)&ap[2 * 16], p2);
-  _mm_store_si128((__m128i *)&ap[1 * 16], p1);
-  _mm_store_si128((__m128i *)&ap[0 * 16], p0);
-  _mm_store_si128((__m128i *)&aq[4 * 16], q4);
-  _mm_store_si128((__m128i *)&aq[3 * 16], q3);
-  _mm_store_si128((__m128i *)&aq[2 * 16], q2);
-  _mm_store_si128((__m128i *)&aq[1 * 16], q1);
-  _mm_store_si128((__m128i *)&aq[0 * 16], q0);
-
+  q5 = _mm_loadu_si128((__m128i *)(s + 5 * p));
+  q6 = _mm_loadu_si128((__m128i *)(s + 6 * p));
+  q7 = _mm_loadu_si128((__m128i *)(s + 7 * p));
 
   {
-    const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
-                                          _mm_subs_epu8(p0, p1));
-    const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
-                                          _mm_subs_epu8(q0, q1));
+    const __m128i abs_p1p0 = abs_diff(p1, p0);
+    const __m128i abs_q1q0 = abs_diff(q1, q0);
     const __m128i fe = _mm_set1_epi8(0xfe);
-    const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-    __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
-                                    _mm_subs_epu8(q0, p0));
-    __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
-                                    _mm_subs_epu8(q1, p1));
+    const __m128i ff = _mm_cmpeq_epi8(zero, zero);
+    __m128i abs_p0q0 = abs_diff(p0, q0);
+    __m128i abs_p1q1 = abs_diff(p1, q1);
     __m128i work;
-    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-    hev = _mm_subs_epu8(flat, thresh);
-    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
+    max_abs_p1p0q1q0 = _mm_max_epu8(abs_p1p0, abs_q1q0);
 
     abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
     abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
     mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
     mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
     // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-    mask = _mm_max_epu8(flat, mask);
+    mask = _mm_max_epu8(max_abs_p1p0q1q0, mask);
     // mask |= (abs(p1 - p0) > limit) * -1;
     // mask |= (abs(q1 - q0) > limit) * -1;
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1),
-                                     _mm_subs_epu8(p1, p2)),
-                         _mm_or_si128(_mm_subs_epu8(p3, p2),
-                                      _mm_subs_epu8(p2, p3)));
+    work = _mm_max_epu8(abs_diff(p2, p1), abs_diff(p3, p2));
     mask = _mm_max_epu8(work, mask);
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1),
-                                     _mm_subs_epu8(q1, q2)),
-                         _mm_or_si128(_mm_subs_epu8(q3, q2),
-                                      _mm_subs_epu8(q2, q3)));
+    work = _mm_max_epu8(abs_diff(q2, q1), abs_diff(q3, q2));
     mask = _mm_max_epu8(work, mask);
     mask = _mm_subs_epu8(mask, limit);
     mask = _mm_cmpeq_epi8(mask, zero);
   }
 
-  // lp filter
+  {
+    __m128i work;
+    work = _mm_max_epu8(abs_diff(p2, p0), abs_diff(q2, q0));
+    flat = _mm_max_epu8(work, max_abs_p1p0q1q0);
+    work = _mm_max_epu8(abs_diff(p3, p0), abs_diff(q3, q0));
+    flat = _mm_max_epu8(work, flat);
+    work = _mm_max_epu8(abs_diff(p4, p0), abs_diff(q4, q0));
+    flat = _mm_subs_epu8(flat, one);
+    flat = _mm_cmpeq_epi8(flat, zero);
+    flat = _mm_and_si128(flat, mask);
+    flat2 = _mm_max_epu8(abs_diff(p5, p0), abs_diff(q5, q0));
+    flat2 = _mm_max_epu8(work, flat2);
+    work = _mm_max_epu8(abs_diff(p6, p0), abs_diff(q6, q0));
+    flat2 = _mm_max_epu8(work, flat2);
+    work = _mm_max_epu8(abs_diff(p7, p0), abs_diff(q7, q0));
+    flat2 = _mm_max_epu8(work, flat2);
+    flat2 = _mm_subs_epu8(flat2, one);
+    flat2 = _mm_cmpeq_epi8(flat2, zero);
+    flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
+  }
+
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // filter4
   {
     const __m128i t4 = _mm_set1_epi8(4);
     const __m128i t3 = _mm_set1_epi8(3);
@@ -459,23 +474,27 @@
     const __m128i t1f = _mm_set1_epi8(0x1f);
     const __m128i t1 = _mm_set1_epi8(0x1);
     const __m128i t7f = _mm_set1_epi8(0x7f);
+    const __m128i ff = _mm_cmpeq_epi8(t4, t4);
 
-    __m128i ps1 = _mm_xor_si128(p1, t80);
-    __m128i ps0 = _mm_xor_si128(p0, t80);
-    __m128i qs0 = _mm_xor_si128(q0, t80);
-    __m128i qs1 = _mm_xor_si128(q1, t80);
     __m128i filt;
     __m128i work_a;
     __m128i filter1, filter2;
 
-    filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
-    work_a = _mm_subs_epi8(qs0, ps0);
+    op1 = _mm_xor_si128(p1, t80);
+    op0 = _mm_xor_si128(p0, t80);
+    oq0 = _mm_xor_si128(q0, t80);
+    oq1 = _mm_xor_si128(q1, t80);
+
+    hev = _mm_subs_epu8(max_abs_p1p0q1q0, thresh);
+    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
+    filt = _mm_and_si128(_mm_subs_epi8(op1, oq1), hev);
+
+    work_a = _mm_subs_epi8(oq0, op0);
     filt = _mm_adds_epi8(filt, work_a);
     filt = _mm_adds_epi8(filt, work_a);
     filt = _mm_adds_epi8(filt, work_a);
     // (vp9_filter + 3 * (qs0 - ps0)) & mask
     filt = _mm_and_si128(filt, mask);
-
     filter1 = _mm_adds_epi8(filt, t4);
     filter2 = _mm_adds_epi8(filt, t3);
 
@@ -485,7 +504,7 @@
     work_a = _mm_and_si128(work_a, te0);
     filter1 = _mm_and_si128(filter1, t1f);
     filter1 = _mm_or_si128(filter1, work_a);
-    qs0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
+    oq0 = _mm_xor_si128(_mm_subs_epi8(oq0, filter1), t80);
 
     // Filter2 >> 3
     work_a = _mm_cmpgt_epi8(zero, filter2);
@@ -493,7 +512,7 @@
     work_a = _mm_and_si128(work_a, te0);
     filter2 = _mm_and_si128(filter2, t1f);
     filter2 = _mm_or_si128(filter2, work_a);
-    ps0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
+    op0 = _mm_xor_si128(_mm_adds_epi8(op0, filter2), t80);
 
     // filt >> 1
     filt = _mm_adds_epi8(filter1, t1);
@@ -503,345 +522,195 @@
     filt = _mm_and_si128(filt, t7f);
     filt = _mm_or_si128(filt, work_a);
     filt = _mm_andnot_si128(hev, filt);
-    ps1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
-    qs1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
+    op1 = _mm_xor_si128(_mm_adds_epi8(op1, filt), t80);
+    oq1 = _mm_xor_si128(_mm_subs_epi8(oq1, filt), t80);
     // loopfilter done
 
+    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    // filter8
     {
-      __m128i work;
-      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p0),
-                                       _mm_subs_epu8(p0, p2)),
-                           _mm_or_si128(_mm_subs_epu8(q2, q0),
-                                        _mm_subs_epu8(q0, q2)));
-      flat = _mm_max_epu8(work, flat);
-      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p3, p0),
-                                       _mm_subs_epu8(p0, p3)),
-                           _mm_or_si128(_mm_subs_epu8(q3, q0),
-                                        _mm_subs_epu8(q0, q3)));
-      flat = _mm_max_epu8(work, flat);
-      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p4, p0),
-                                       _mm_subs_epu8(p0, p4)),
-                           _mm_or_si128(_mm_subs_epu8(q4, q0),
-                                        _mm_subs_epu8(q0, q4)));
-      flat = _mm_subs_epu8(flat, one);
-      flat = _mm_cmpeq_epi8(flat, zero);
-      flat = _mm_and_si128(flat, mask);
+      const __m128i four = _mm_set1_epi16(4);
+      const __m128i p3_lo = _mm_unpacklo_epi8(p3, zero);
+      const __m128i p2_lo = _mm_unpacklo_epi8(p2, zero);
+      const __m128i p1_lo = _mm_unpacklo_epi8(p1, zero);
+      const __m128i p0_lo = _mm_unpacklo_epi8(p0, zero);
+      const __m128i q0_lo = _mm_unpacklo_epi8(q0, zero);
+      const __m128i q1_lo = _mm_unpacklo_epi8(q1, zero);
+      const __m128i q2_lo = _mm_unpacklo_epi8(q2, zero);
+      const __m128i q3_lo = _mm_unpacklo_epi8(q3, zero);
 
-      p5 = _mm_loadu_si128((__m128i *)(s - 6 * p));
-      q5 = _mm_loadu_si128((__m128i *)(s + 5 * p));
-      flat2 = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p5, p0),
-                                       _mm_subs_epu8(p0, p5)),
-                           _mm_or_si128(_mm_subs_epu8(q5, q0),
-                                        _mm_subs_epu8(q0, q5)));
-      _mm_store_si128((__m128i *)&ap[5 * 16], p5);
-      _mm_store_si128((__m128i *)&aq[5 * 16], q5);
-      flat2 = _mm_max_epu8(work, flat2);
-      p6 = _mm_loadu_si128((__m128i *)(s - 7 * p));
-      q6 = _mm_loadu_si128((__m128i *)(s + 6 * p));
-      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p6, p0),
-                                       _mm_subs_epu8(p0, p6)),
-                           _mm_or_si128(_mm_subs_epu8(q6, q0),
-                                        _mm_subs_epu8(q0, q6)));
-      _mm_store_si128((__m128i *)&ap[6 * 16], p6);
-      _mm_store_si128((__m128i *)&aq[6 * 16], q6);
-      flat2 = _mm_max_epu8(work, flat2);
+      const __m128i p3_hi = _mm_unpackhi_epi8(p3, zero);
+      const __m128i p2_hi = _mm_unpackhi_epi8(p2, zero);
+      const __m128i p1_hi = _mm_unpackhi_epi8(p1, zero);
+      const __m128i p0_hi = _mm_unpackhi_epi8(p0, zero);
+      const __m128i q0_hi = _mm_unpackhi_epi8(q0, zero);
+      const __m128i q1_hi = _mm_unpackhi_epi8(q1, zero);
+      const __m128i q2_hi = _mm_unpackhi_epi8(q2, zero);
+      const __m128i q3_hi = _mm_unpackhi_epi8(q3, zero);
+      __m128i f8_lo, f8_hi;
 
-      p7 = _mm_loadu_si128((__m128i *)(s - 8 * p));
-      q7 = _mm_loadu_si128((__m128i *)(s + 7 * p));
-      work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p7, p0),
-                                       _mm_subs_epu8(p0, p7)),
-                           _mm_or_si128(_mm_subs_epu8(q7, q0),
-                                        _mm_subs_epu8(q0, q7)));
-      _mm_store_si128((__m128i *)&ap[7 * 16], p7);
-      _mm_store_si128((__m128i *)&aq[7 * 16], q7);
-      flat2 = _mm_max_epu8(work, flat2);
-      flat2 = _mm_subs_epu8(flat2, one);
-      flat2 = _mm_cmpeq_epi8(flat2, zero);
-      flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
+      f8_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, four),
+                            _mm_add_epi16(p3_lo, p2_lo));
+      f8_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, f8_lo),
+                            _mm_add_epi16(p2_lo, p1_lo));
+      f8_lo = _mm_add_epi16(_mm_add_epi16(p0_lo, q0_lo), f8_lo);
+
+      f8_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, four),
+                            _mm_add_epi16(p3_hi, p2_hi));
+      f8_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, f8_hi),
+                            _mm_add_epi16(p2_hi, p1_hi));
+      f8_hi = _mm_add_epi16(_mm_add_epi16(p0_hi, q0_hi), f8_hi);
+
+      op2 = filter8_mask(&flat, &p2, &f8_lo, &f8_hi);
+
+      f8_lo = filter_add2_sub2(&f8_lo, &q1_lo, &p1_lo, &p2_lo, &p3_lo);
+      f8_hi = filter_add2_sub2(&f8_hi, &q1_hi, &p1_hi, &p2_hi, &p3_hi);
+      op1 = filter8_mask(&flat, &op1, &f8_lo, &f8_hi);
+
+      f8_lo = filter_add2_sub2(&f8_lo, &q2_lo, &p0_lo, &p1_lo, &p3_lo);
+      f8_hi = filter_add2_sub2(&f8_hi, &q2_hi, &p0_hi, &p1_hi, &p3_hi);
+      op0 = filter8_mask(&flat, &op0, &f8_lo, &f8_hi);
+
+      f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q0_lo, &p0_lo, &p3_lo);
+      f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q0_hi, &p0_hi, &p3_hi);
+      oq0 = filter8_mask(&flat, &oq0, &f8_lo, &f8_hi);
+
+      f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q1_lo, &q0_lo, &p2_lo);
+      f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q1_hi, &q0_hi, &p2_hi);
+      oq1 = filter8_mask(&flat, &oq1, &f8_lo, &f8_hi);
+
+      f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q2_lo, &q1_lo, &p1_lo);
+      f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q2_hi, &q1_hi, &p1_hi);
+      oq2 = filter8_mask(&flat, &q2, &f8_lo, &f8_hi);
     }
 
     // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    // flat and wide flat calculations
+    // wide flat calculations
     {
       const __m128i eight = _mm_set1_epi16(8);
-      const __m128i four = _mm_set1_epi16(4);
-      __m128i temp_flat2 = flat2;
-      unsigned char *src = s;
-      int i = 0;
-      do {
-        __m128i workp_shft;
-        __m128i a, b, c;
+      const __m128i p7_lo = _mm_unpacklo_epi8(p7, zero);
+      const __m128i p6_lo = _mm_unpacklo_epi8(p6, zero);
+      const __m128i p5_lo = _mm_unpacklo_epi8(p5, zero);
+      const __m128i p4_lo = _mm_unpacklo_epi8(p4, zero);
+      const __m128i p3_lo = _mm_unpacklo_epi8(p3, zero);
+      const __m128i p2_lo = _mm_unpacklo_epi8(p2, zero);
+      const __m128i p1_lo = _mm_unpacklo_epi8(p1, zero);
+      const __m128i p0_lo = _mm_unpacklo_epi8(p0, zero);
+      const __m128i q0_lo = _mm_unpacklo_epi8(q0, zero);
+      const __m128i q1_lo = _mm_unpacklo_epi8(q1, zero);
+      const __m128i q2_lo = _mm_unpacklo_epi8(q2, zero);
+      const __m128i q3_lo = _mm_unpacklo_epi8(q3, zero);
+      const __m128i q4_lo = _mm_unpacklo_epi8(q4, zero);
+      const __m128i q5_lo = _mm_unpacklo_epi8(q5, zero);
+      const __m128i q6_lo = _mm_unpacklo_epi8(q6, zero);
+      const __m128i q7_lo = _mm_unpacklo_epi8(q7, zero);
 
-        unsigned int off = i * 8;
-        p7 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[7 * 16] + off)),
-                               zero);
-        p6 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[6 * 16] + off)),
-                               zero);
-        p5 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[5 * 16] + off)),
-                               zero);
-        p4 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[4 * 16] + off)),
-                               zero);
-        p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[3 * 16] + off)),
-                               zero);
-        p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[2 * 16] + off)),
-                               zero);
-        p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[1 * 16] + off)),
-                               zero);
-        p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&ap[0 * 16] + off)),
-                               zero);
-        q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[0 * 16] + off)),
-                               zero);
-        q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[1 * 16] + off)),
-                               zero);
-        q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[2 * 16] + off)),
-                               zero);
-        q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[3 * 16] + off)),
-                               zero);
-        q4 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[4 * 16] + off)),
-                               zero);
-        q5 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[5 * 16] + off)),
-                               zero);
-        q6 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[6 * 16] + off)),
-                               zero);
-        q7 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(&aq[7 * 16] + off)),
-                               zero);
+      const __m128i p7_hi = _mm_unpackhi_epi8(p7, zero);
+      const __m128i p6_hi = _mm_unpackhi_epi8(p6, zero);
+      const __m128i p5_hi = _mm_unpackhi_epi8(p5, zero);
+      const __m128i p4_hi = _mm_unpackhi_epi8(p4, zero);
+      const __m128i p3_hi = _mm_unpackhi_epi8(p3, zero);
+      const __m128i p2_hi = _mm_unpackhi_epi8(p2, zero);
+      const __m128i p1_hi = _mm_unpackhi_epi8(p1, zero);
+      const __m128i p0_hi = _mm_unpackhi_epi8(p0, zero);
+      const __m128i q0_hi = _mm_unpackhi_epi8(q0, zero);
+      const __m128i q1_hi = _mm_unpackhi_epi8(q1, zero);
+      const __m128i q2_hi = _mm_unpackhi_epi8(q2, zero);
+      const __m128i q3_hi = _mm_unpackhi_epi8(q3, zero);
+      const __m128i q4_hi = _mm_unpackhi_epi8(q4, zero);
+      const __m128i q5_hi = _mm_unpackhi_epi8(q5, zero);
+      const __m128i q6_hi = _mm_unpackhi_epi8(q6, zero);
+      const __m128i q7_hi = _mm_unpackhi_epi8(q7, zero);
 
-        c = _mm_sub_epi16(_mm_slli_epi16(p7, 3), p7);  // p7 * 7
-        c = _mm_add_epi16(_mm_slli_epi16(p6, 1), _mm_add_epi16(p4, c));
+      __m128i f_lo;
+      __m128i f_hi;
 
-        b = _mm_add_epi16(_mm_add_epi16(p3, four), _mm_add_epi16(p3, p2));
-        a = _mm_add_epi16(p3, _mm_add_epi16(p2, p1));
-        a = _mm_add_epi16(_mm_add_epi16(p0, q0), a);
+      f_lo = _mm_sub_epi16(_mm_slli_epi16(p7_lo, 3), p7_lo);  // p7 * 7
+      f_lo = _mm_add_epi16(_mm_slli_epi16(p6_lo, 1),
+                           _mm_add_epi16(p4_lo, f_lo));
+      f_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, f_lo),
+                           _mm_add_epi16(p2_lo, p1_lo));
+      f_lo = _mm_add_epi16(_mm_add_epi16(p0_lo, q0_lo), f_lo);
+      f_lo = _mm_add_epi16(_mm_add_epi16(p5_lo, eight), f_lo);
 
-        _mm_storel_epi64((__m128i *)&flat_op[2 * 16 + i * 8],
-                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
-                                          , b));
+      f_hi = _mm_sub_epi16(_mm_slli_epi16(p7_hi, 3), p7_hi);  // p7 * 7
+      f_hi = _mm_add_epi16(_mm_slli_epi16(p6_hi, 1),
+                           _mm_add_epi16(p4_hi, f_hi));
+      f_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, f_hi),
+                           _mm_add_epi16(p2_hi, p1_hi));
+      f_hi = _mm_add_epi16(_mm_add_epi16(p0_hi, q0_hi), f_hi);
+      f_hi = _mm_add_epi16(_mm_add_epi16(p5_hi, eight), f_hi);
 
-        c = _mm_add_epi16(_mm_add_epi16(p5, eight), c);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[6 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      p6 = filter16_mask(&flat2, &p6, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 7 * p), p6);
 
-        a = _mm_add_epi16(q1, a);
-        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p3, p2)), p1);
-        _mm_storel_epi64((__m128i *)&flat_op[1 * 16 + i * 8],
-                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
-                                          , b));
+      f_lo = filter_add2_sub2(&f_lo, &q1_lo, &p5_lo, &p6_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q1_hi, &p5_hi, &p6_hi, &p7_hi);
+      p5 = filter16_mask(&flat2, &p5, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 6 * p), p5);
 
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p6)), p5);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[5 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q2_lo, &p4_lo, &p5_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q2_hi, &p4_hi, &p5_hi, &p7_hi);
+      p4 = filter16_mask(&flat2, &p4, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 5 * p), p4);
 
-        a = _mm_add_epi16(q2, a);
-        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p3, p1)), p0);
-        _mm_storel_epi64((__m128i *)&flat_op[i * 8],
-                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
-                                          , b));
+      f_lo = filter_add2_sub2(&f_lo, &q3_lo, &p3_lo, &p4_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q3_hi, &p3_hi, &p4_hi, &p7_hi);
+      p3 = filter16_mask(&flat2, &p3, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 4 * p), p3);
 
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p5)), p4);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[4 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q4_lo, &p2_lo, &p3_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q4_hi, &p2_hi, &p3_hi, &p7_hi);
+      op2 = filter16_mask(&flat2, &op2, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 3 * p), op2);
 
-        a = _mm_add_epi16(q3, a);
-        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p3, p0)), q0);
-        _mm_storel_epi64((__m128i *)&flat_oq[i * 8],
-                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
-                                          , b));
+      f_lo = filter_add2_sub2(&f_lo, &q5_lo, &p1_lo, &p2_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q5_hi, &p1_hi, &p2_hi, &p7_hi);
+      op1 = filter16_mask(&flat2, &op1, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 2 * p), op1);
 
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p4)), p3);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[3 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q6_lo, &p0_lo, &p1_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q6_hi, &p0_hi, &p1_hi, &p7_hi);
+      op0 = filter16_mask(&flat2, &op0, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 1 * p), op0);
 
-        b = _mm_add_epi16(q3, b);
-        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p2, q0)), q1);
-        _mm_storel_epi64((__m128i *)&flat_oq[16 + i * 8],
-                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
-                                          , b));
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q0_lo, &p0_lo, &p7_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q0_hi, &p0_hi, &p7_hi);
+      oq0 = filter16_mask(&flat2, &oq0, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s - 0 * p), oq0);
 
-        c = _mm_add_epi16(q4, c);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p3)), p2);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[2 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q1_lo, &p6_lo, &q0_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q1_hi, &p6_hi, &q0_hi);
+      oq1 = filter16_mask(&flat2, &oq1, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s + 1 * p), oq1);
 
-        b = _mm_add_epi16(q3, b);
-        b = _mm_add_epi16(_mm_sub_epi16(b, _mm_add_epi16(p1, q1)), q2);
-        _mm_storel_epi64((__m128i *)&flat_oq[2 * 16 + i * 8],
-                         _mm_packus_epi16(_mm_srli_epi16(_mm_add_epi16(a, b), 3)
-                                          , b));
-        a = _mm_add_epi16(q5, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p2)), p1);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q2_lo, &p5_lo, &q1_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q2_hi, &p5_hi, &q1_hi);
+      oq2 = filter16_mask(&flat2, &oq2, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s + 2 * p), oq2);
 
-        a = _mm_add_epi16(q6, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p1)), p0);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_op[i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q3_lo, &p4_lo, &q2_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q3_hi, &p4_hi, &q2_hi);
+      q3 = filter16_mask(&flat2, &q3, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s + 3 * p), q3);
 
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p7, p0)), q0);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q4_lo, &p3_lo, &q3_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q4_hi, &p3_hi, &q3_hi);
+      q4 = filter16_mask(&flat2, &q4, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s + 4 * p), q4);
 
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p6, q0)), q1);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q5_lo, &p2_lo, &q4_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q5_hi, &p2_hi, &q4_hi);
+      q5 = filter16_mask(&flat2, &q5, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s + 5 * p), q5);
 
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p5, q1)), q2);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[2 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
-
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p4, q2)), q3);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[3 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
-
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p3, q3)), q4);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[4 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
-
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p2, q4)), q5);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[5 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
-
-        a = _mm_add_epi16(q7, a);
-        c = _mm_add_epi16(_mm_sub_epi16(c, _mm_add_epi16(p1, q5)), q6);
-        workp_shft = _mm_srli_epi16(_mm_add_epi16(a, c), 4);
-        _mm_storel_epi64((__m128i *)&flat2_oq[6 * 16 + i * 8],
-                         _mm_packus_epi16(workp_shft, workp_shft));
-
-        temp_flat2 = _mm_srli_si128(temp_flat2, 8);
-        src += 8;
-      } while (++i < 2);
+      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q6_lo, &p1_lo, &q5_lo);
+      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q6_hi, &p1_hi, &q5_hi);
+      q6 = filter16_mask(&flat2, &q6, &f_lo, &f_hi);
+      _mm_storeu_si128((__m128i *)(s + 6 * p), q6);
     }
     // wide flat
     // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-    work_a = _mm_load_si128((__m128i *)&ap[2 * 16]);
-    p2 = _mm_load_si128((__m128i *)&flat_op[2 * 16]);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p2 = _mm_and_si128(flat, p2);
-    p2 = _mm_or_si128(work_a, p2);
-    _mm_store_si128((__m128i *)&flat_op[2 * 16], p2);
-
-    p1 = _mm_load_si128((__m128i *)&flat_op[1 * 16]);
-    work_a = _mm_andnot_si128(flat, ps1);
-    p1 = _mm_and_si128(flat, p1);
-    p1 = _mm_or_si128(work_a, p1);
-    _mm_store_si128((__m128i *)&flat_op[1 * 16], p1);
-
-    p0 = _mm_load_si128((__m128i *)&flat_op[0]);
-    work_a = _mm_andnot_si128(flat, ps0);
-    p0 = _mm_and_si128(flat, p0);
-    p0 = _mm_or_si128(work_a, p0);
-    _mm_store_si128((__m128i *)&flat_op[0], p0);
-
-    q0 = _mm_load_si128((__m128i *)&flat_oq[0]);
-    work_a = _mm_andnot_si128(flat, qs0);
-    q0 = _mm_and_si128(flat, q0);
-    q0 = _mm_or_si128(work_a, q0);
-    _mm_store_si128((__m128i *)&flat_oq[0], q0);
-
-    q1 = _mm_load_si128((__m128i *)&flat_oq[1 * 16]);
-    work_a = _mm_andnot_si128(flat, qs1);
-    q1 = _mm_and_si128(flat, q1);
-    q1 = _mm_or_si128(work_a, q1);
-    _mm_store_si128((__m128i *)&flat_oq[1 * 16], q1);
-
-    work_a = _mm_load_si128((__m128i *)&aq[2 * 16]);
-    q2 = _mm_load_si128((__m128i *)&flat_oq[2 * 16]);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q2 = _mm_and_si128(flat, q2);
-    q2 = _mm_or_si128(work_a, q2);
-    _mm_store_si128((__m128i *)&flat_oq[2 * 16], q2);
-
-    // write out op6 - op3
-    {
-      unsigned char *dst = (s - 7 * p);
-      for (i = 6; i > 2; i--) {
-        __m128i flat2_output;
-        work_a = _mm_load_si128((__m128i *)&ap[i * 16]);
-        flat2_output = _mm_load_si128((__m128i *)&flat2_op[i * 16]);
-        work_a = _mm_andnot_si128(flat2, work_a);
-        flat2_output = _mm_and_si128(flat2, flat2_output);
-        work_a = _mm_or_si128(work_a, flat2_output);
-        _mm_storeu_si128((__m128i *)dst, work_a);
-        dst += p;
-      }
-    }
-
-    work_a = _mm_load_si128((__m128i *)&flat_op[2 * 16]);
-    p2 = _mm_load_si128((__m128i *)&flat2_op[2 * 16]);
-    work_a = _mm_andnot_si128(flat2, work_a);
-    p2 = _mm_and_si128(flat2, p2);
-    p2 = _mm_or_si128(work_a, p2);
-    _mm_storeu_si128((__m128i *)(s - 3 * p), p2);
-
-    work_a = _mm_load_si128((__m128i *)&flat_op[1 * 16]);
-    p1 = _mm_load_si128((__m128i *)&flat2_op[1 * 16]);
-    work_a = _mm_andnot_si128(flat2, work_a);
-    p1 = _mm_and_si128(flat2, p1);
-    p1 = _mm_or_si128(work_a, p1);
-    _mm_storeu_si128((__m128i *)(s - 2 * p), p1);
-
-    work_a = _mm_load_si128((__m128i *)&flat_op[0]);
-    p0 = _mm_load_si128((__m128i *)&flat2_op[0]);
-    work_a = _mm_andnot_si128(flat2, work_a);
-    p0 = _mm_and_si128(flat2, p0);
-    p0 = _mm_or_si128(work_a, p0);
-    _mm_storeu_si128((__m128i *)(s - 1 * p), p0);
-
-    work_a = _mm_load_si128((__m128i *)&flat_oq[0]);
-    q0 = _mm_load_si128((__m128i *)&flat2_oq[0]);
-    work_a = _mm_andnot_si128(flat2, work_a);
-    q0 = _mm_and_si128(flat2, q0);
-    q0 = _mm_or_si128(work_a, q0);
-    _mm_storeu_si128((__m128i *)(s - 0 * p), q0);
-
-    work_a = _mm_load_si128((__m128i *)&flat_oq[1 * 16]);
-    q1 = _mm_load_si128((__m128i *)&flat2_oq[16]);
-    work_a = _mm_andnot_si128(flat2, work_a);
-    q1 = _mm_and_si128(flat2, q1);
-    q1 = _mm_or_si128(work_a, q1);
-    _mm_storeu_si128((__m128i *)(s + 1 * p), q1);
-
-    work_a = _mm_load_si128((__m128i *)&flat_oq[2 * 16]);
-    q2 = _mm_load_si128((__m128i *)&flat2_oq[2 * 16]);
-    work_a = _mm_andnot_si128(flat2, work_a);
-    q2 = _mm_and_si128(flat2, q2);
-    q2 = _mm_or_si128(work_a, q2);
-    _mm_storeu_si128((__m128i *)(s + 2 * p), q2);
-
-    // write out oq3 - oq7
-    {
-      unsigned char *dst = (s + 3 * p);
-      for (i = 3; i < 7; i++) {
-        __m128i flat2_output;
-        work_a = _mm_load_si128((__m128i *)&aq[i * 16]);
-        flat2_output = _mm_load_si128((__m128i *)&flat2_oq[i * 16]);
-        work_a = _mm_andnot_si128(flat2, work_a);
-        flat2_output = _mm_and_si128(flat2, flat2_output);
-        work_a = _mm_or_si128(work_a, flat2_output);
-        _mm_storeu_si128((__m128i *)dst, work_a);
-        dst += p;
-      }
-    }
   }
 }
 
@@ -893,14 +762,11 @@
     const __m128i fe = _mm_set1_epi8(0xfe);
     const __m128i ff = _mm_cmpeq_epi8(fe, fe);
     __m128i abs_p1q1, abs_p0q0, abs_q1q0, abs_p1p0, work;
-    abs_p1p0 = _mm_or_si128(_mm_subs_epu8(q1p1, q0p0),
-                            _mm_subs_epu8(q0p0, q1p1));
+    abs_p1p0 = abs_diff(q1p1, q0p0);
     abs_q1q0 =  _mm_srli_si128(abs_p1p0, 8);
 
-    abs_p0q0 = _mm_or_si128(_mm_subs_epu8(q0p0, p0q0),
-                            _mm_subs_epu8(p0q0, q0p0));
-    abs_p1q1 = _mm_or_si128(_mm_subs_epu8(q1p1, p1q1),
-                            _mm_subs_epu8(p1q1, q1p1));
+    abs_p0q0 = abs_diff(q0p0, p0q0);
+    abs_p1q1 = abs_diff(q1p1, p1q1);
     flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
     hev = _mm_subs_epu8(flat, thresh);
     hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
@@ -914,10 +780,8 @@
     // mask |= (abs(p1 - p0) > limit) * -1;
     // mask |= (abs(q1 - q0) > limit) * -1;
 
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2p2, q1p1),
-                                     _mm_subs_epu8(q1p1, q2p2)),
-                        _mm_or_si128(_mm_subs_epu8(q3p3, q2p2),
-                                     _mm_subs_epu8(q2p2, q3p3)));
+    work = _mm_max_epu8(abs_diff(q2p2, q1p1),
+                        abs_diff(q3p3, q2p2));
     mask = _mm_max_epu8(work, mask);
     mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8));
     mask = _mm_subs_epu8(mask, limit);
@@ -925,10 +789,8 @@
 
     // flat_mask4
 
-    flat = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2p2, q0p0),
-                                     _mm_subs_epu8(q0p0, q2p2)),
-                        _mm_or_si128(_mm_subs_epu8(q3p3, q0p0),
-                                     _mm_subs_epu8(q0p0, q3p3)));
+    flat = _mm_max_epu8(abs_diff(q2p2, q0p0),
+                        abs_diff(q3p3, q0p0));
     flat = _mm_max_epu8(abs_p1p0, flat);
     flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8));
     flat = _mm_subs_epu8(flat, one);
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index a9c03f0..4e85caf 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -195,7 +195,7 @@
   struct macroblockd_plane *const pd = &xd->plane[plane];
   if (eob > 0) {
     TX_TYPE tx_type = DCT_DCT;
-    int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+    tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
     if (xd->lossless) {
       tx_type = DCT_DCT;
       vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
@@ -249,7 +249,7 @@
   VP9_COMMON *const cm = args->cm;
   MACROBLOCKD *const xd = args->xd;
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  MODE_INFO *const mi = xd->mi[0];
+  MODE_INFO *const mi = xd->mi[0].src_mi;
   const PREDICTION_MODE mode = (plane == 0) ? get_y_mode(mi, block)
                                             : mi->mbmi.uv_mode;
   int x, y;
@@ -305,12 +305,14 @@
   const int offset = mi_row * cm->mi_stride + mi_col;
   int x, y;
 
-  xd->mi = cm->mi_grid_visible + offset;
-  xd->mi[0] = &cm->mi[offset];
-  xd->mi[0]->mbmi.sb_type = bsize;
+  xd->mi = cm->mi + offset;
+  xd->mi[0].src_mi = &xd->mi[0];  // Point to self.
+  xd->mi[0].mbmi.sb_type = bsize;
+
   for (y = 0; y < y_mis; ++y)
-    for (x = !y; x < x_mis; ++x)
-      xd->mi[y * cm->mi_stride + x] = xd->mi[0];
+    for (x = !y; x < x_mis; ++x) {
+      xd->mi[y * cm->mi_stride + x].src_mi = &xd->mi[0];
+    }
 
   set_skip_context(xd, mi_row, mi_col);
 
@@ -319,12 +321,12 @@
   set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
 
   vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
-  return &xd->mi[0]->mbmi;
+  return &xd->mi[0].mbmi;
 }
 
 static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                     int idx, int mi_row, int mi_col) {
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
   xd->block_refs[idx] = ref_buffer;
   if (!vp9_is_valid_scale(&ref_buffer->sf))
@@ -668,6 +670,15 @@
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffer");
   }
+  cm->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
+}
+
+static INLINE int valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth,
+                                          int ref_xss, int ref_yss,
+                                          vpx_bit_depth_t this_bit_depth,
+                                          int this_xss, int this_yss) {
+  return ref_bit_depth == this_bit_depth && ref_xss == this_xss &&
+         ref_yss == this_yss;
 }
 
 static void setup_frame_size_with_refs(VP9_COMMON *cm,
@@ -707,6 +718,18 @@
   if (!has_valid_ref_frame)
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Referenced frame has invalid size");
+  for (i = 0; i < REFS_PER_FRAME; ++i) {
+    RefBuffer *const ref_frame = &cm->frame_refs[i];
+    if (!valid_ref_frame_img_fmt(
+            ref_frame->buf->bit_depth,
+            ref_frame->buf->uv_crop_width < ref_frame->buf->y_crop_width,
+            ref_frame->buf->uv_crop_height < ref_frame->buf->y_crop_height,
+            cm->bit_depth,
+            cm->subsampling_x,
+            cm->subsampling_y))
+      vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+                         "Referenced frame has incompatible color space");
+  }
 
   resize_context_buffers(cm, width, height);
   setup_display_size(cm, rb);
@@ -723,6 +746,7 @@
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffer");
   }
+  cm->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
 }
 
 static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
@@ -938,9 +962,8 @@
   return vp9_reader_find_end(&tile_data->bit_reader);
 }
 
-static int tile_worker_hook(void *arg1, void *arg2) {
-  TileWorkerData *const tile_data = (TileWorkerData*)arg1;
-  const TileInfo *const tile = (TileInfo*)arg2;
+static int tile_worker_hook(TileWorkerData *const tile_data,
+                            const TileInfo *const tile) {
   int mi_row, mi_col;
 
   for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
@@ -1201,6 +1224,7 @@
     }
 
     setup_frame_size(cm, rb);
+    pbi->need_resync = 0;
   } else {
     cm->intra_only = cm->show_frame ? 0 : vp9_rb_read_bit(rb);
 
@@ -1224,6 +1248,7 @@
 
       pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
       setup_frame_size(cm, rb);
+      pbi->need_resync = 0;
     } else {
       pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
       for (i = 0; i < REFS_PER_FRAME; ++i) {
@@ -1242,16 +1267,30 @@
 
       for (i = 0; i < REFS_PER_FRAME; ++i) {
         RefBuffer *const ref_buf = &cm->frame_refs[i];
+#if CONFIG_VP9_HIGHBITDEPTH
+        vp9_setup_scale_factors_for_frame(&ref_buf->sf,
+                                          ref_buf->buf->y_crop_width,
+                                          ref_buf->buf->y_crop_height,
+                                          cm->width, cm->height,
+                                          cm->use_highbitdepth);
+#else
         vp9_setup_scale_factors_for_frame(&ref_buf->sf,
                                           ref_buf->buf->y_crop_width,
                                           ref_buf->buf->y_crop_height,
                                           cm->width, cm->height);
+#endif
         if (vp9_is_scaled(&ref_buf->sf))
           vp9_extend_frame_borders(ref_buf->buf);
       }
     }
   }
 
+  if (pbi->need_resync) {
+    vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+                       "Keyframe / intra-only frame required to reset decoder"
+                       " state");
+  }
+
   if (!cm->error_resilient_mode) {
     cm->refresh_frame_context = vp9_rb_read_bit(rb);
     cm->frame_parallel_decoding_mode = vp9_rb_read_bit(rb);
@@ -1337,11 +1376,11 @@
   int q;
 
   for (q = 0; q < QINDEX_RANGE; q++) {
-    cm->y_dequant[q][0] = vp9_dc_quant(q, cm->y_dc_delta_q);
-    cm->y_dequant[q][1] = vp9_ac_quant(q, 0);
+    cm->y_dequant[q][0] = vp9_dc_quant(q, cm->y_dc_delta_q, cm->bit_depth);
+    cm->y_dequant[q][1] = vp9_ac_quant(q, 0, cm->bit_depth);
 
-    cm->uv_dequant[q][0] = vp9_dc_quant(q, cm->uv_dc_delta_q);
-    cm->uv_dequant[q][1] = vp9_ac_quant(q, cm->uv_ac_delta_q);
+    cm->uv_dequant[q][0] = vp9_dc_quant(q, cm->uv_dc_delta_q, cm->bit_depth);
+    cm->uv_dequant[q][1] = vp9_ac_quant(q, cm->uv_ac_delta_q, cm->bit_depth);
   }
 }
 
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index 32e80f9..ef2dc80 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -103,7 +103,7 @@
                                  int mi_row, int mi_col,
                                  vp9_reader *r) {
   struct segmentation *const seg = &cm->seg;
-  const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
+  const BLOCK_SIZE bsize = xd->mi[0].src_mi->mbmi.sb_type;
   int segment_id;
 
   if (!seg->enabled)
@@ -120,7 +120,7 @@
 static int read_inter_segment_id(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                                  int mi_row, int mi_col, vp9_reader *r) {
   struct segmentation *const seg = &cm->seg;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
   int predicted_segment_id, segment_id;
 
@@ -160,10 +160,10 @@
 static void read_intra_frame_mode_info(VP9_COMMON *const cm,
                                        MACROBLOCKD *const xd,
                                        int mi_row, int mi_col, vp9_reader *r) {
-  MODE_INFO *const mi = xd->mi[0];
+  MODE_INFO *const mi = xd->mi[0].src_mi;
   MB_MODE_INFO *const mbmi = &mi->mbmi;
-  const MODE_INFO *above_mi = xd->mi[-cm->mi_stride];
-  const MODE_INFO *left_mi  = xd->left_available ? xd->mi[-1] : NULL;
+  const MODE_INFO *above_mi = xd->mi[-cm->mi_stride].src_mi;
+  const MODE_INFO *left_mi  = xd->left_available ? xd->mi[-1].src_mi : NULL;
   const BLOCK_SIZE bsize = mbmi->sb_type;
   int i;
 
@@ -519,7 +519,7 @@
                                        MACROBLOCKD *const xd,
                                        const TileInfo *const tile,
                                        int mi_row, int mi_col, vp9_reader *r) {
-  MODE_INFO *const mi = xd->mi[0];
+  MODE_INFO *const mi = xd->mi[0].src_mi;
   MB_MODE_INFO *const mbmi = &mi->mbmi;
   int inter_block;
 
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index 9106b0d..6ee3d70 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -60,6 +60,7 @@
   }
 
   cm->error.setjmp = 1;
+  pbi->need_resync = 1;
   initialize_dec();
 
   // Initialize the references to not point to any frame buffers.
@@ -238,6 +239,7 @@
   cm->new_fb_idx = get_free_fb(cm);
 
   if (setjmp(cm->error.jmp)) {
+    pbi->need_resync = 1;
     cm->error.setjmp = 0;
     vp9_clear_system_state();
 
diff --git a/vp9/decoder/vp9_decoder.h b/vp9/decoder/vp9_decoder.h
index 848d212..4f52bb9 100644
--- a/vp9/decoder/vp9_decoder.h
+++ b/vp9/decoder/vp9_decoder.h
@@ -58,6 +58,7 @@
 
   int max_threads;
   int inv_tile_order;
+  int need_resync;  // wait for key/intra-only frame
 } VP9Decoder;
 
 int vp9_receive_compressed_data(struct VP9Decoder *pbi,
diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index 91cdf38..5778748 100644
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -13,6 +13,7 @@
 
 #include "vp9/common/vp9_blockd.h"
 #include "vp9/common/vp9_common.h"
+#include "vp9/common/vp9_entropy.h"
 
 #include "vp9/decoder/vp9_detokenize.h"
 
@@ -31,33 +32,35 @@
 #define INCREMENT_COUNT(token)                              \
   do {                                                      \
      if (!cm->frame_parallel_decoding_mode)                 \
-       ++coef_counts[band][ctx][token];                     \
+       ++coef_counts[band][ctx][token];                      \
   } while (0)
 
-#define WRITE_COEF_CONTINUE(val, token)                  \
-  {                                                      \
-    v = (val * dqv) >> dq_shift;                         \
-    dqcoeff[scan[c]] = vp9_read_bit(r) ? -v : v;         \
-    token_cache[scan[c]] = vp9_pt_energy_class[token];   \
-    ++c;                                                 \
-    ctx = get_coef_context(nb, token_cache, c);          \
-    dqv = dq[1];                                         \
-    continue;                                            \
-  }
+static INLINE int read_coeff(const vp9_prob *probs, int n, vp9_reader *r) {
+  int i, val = 0;
+  for (i = 0; i < n; ++i)
+    val = (val << 1) | vp9_read(r, probs[i]);
+  return val;
+}
 
-#define ADJUST_COEF(prob, bits_count)                   \
-  do {                                                  \
-    val += (vp9_read(r, prob) << bits_count);           \
-  } while (0)
+static const vp9_tree_index coeff_subtree_high[TREE_SIZE(ENTROPY_TOKENS)] = {
+  2, 6,                                         /* 0 = LOW_VAL */
+  -TWO_TOKEN, 4,                                /* 1 = TWO */
+  -THREE_TOKEN, -FOUR_TOKEN,                    /* 2 = THREE */
+  8, 10,                                        /* 3 = HIGH_LOW */
+  -CATEGORY1_TOKEN, -CATEGORY2_TOKEN,           /* 4 = CAT_ONE */
+  12, 14,                                       /* 5 = CAT_THREEFOUR */
+  -CATEGORY3_TOKEN, -CATEGORY4_TOKEN,           /* 6 = CAT_THREE */
+  -CATEGORY5_TOKEN, -CATEGORY6_TOKEN            /* 7 = CAT_FIVE */
+};
 
 static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd, PLANE_TYPE type,
-                       int16_t *dqcoeff, TX_SIZE tx_size, const int16_t *dq,
-                       int ctx, const int16_t *scan, const int16_t *nb,
-                       vp9_reader *r) {
+                        tran_low_t *dqcoeff, TX_SIZE tx_size, const int16_t *dq,
+                        int ctx, const int16_t *scan, const int16_t *nb,
+                        vp9_reader *r) {
   const int max_eob = 16 << (tx_size << 1);
   const FRAME_CONTEXT *const fc = &cm->fc;
   FRAME_COUNTS *const counts = &cm->counts;
-  const int ref = is_inter_block(&xd->mi[0]->mbmi);
+  const int ref = is_inter_block(&xd->mi[0].src_mi->mbmi);
   int band, c = 0;
   const vp9_prob (*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] =
       fc->coef_probs[tx_size][type][ref];
@@ -69,11 +72,51 @@
   uint8_t token_cache[32 * 32];
   const uint8_t *band_translate = get_band_translate(tx_size);
   const int dq_shift = (tx_size == TX_32X32);
-  int v;
+  int v, token;
   int16_t dqv = dq[0];
+  const uint8_t *cat1_prob;
+  const uint8_t *cat2_prob;
+  const uint8_t *cat3_prob;
+  const uint8_t *cat4_prob;
+  const uint8_t *cat5_prob;
+  const uint8_t *cat6_prob;
+
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (cm->use_highbitdepth) {
+    if (cm->bit_depth == VPX_BITS_10) {
+      cat1_prob = vp9_cat1_prob_high10;
+      cat2_prob = vp9_cat2_prob_high10;
+      cat3_prob = vp9_cat3_prob_high10;
+      cat4_prob = vp9_cat4_prob_high10;
+      cat5_prob = vp9_cat5_prob_high10;
+      cat6_prob = vp9_cat6_prob_high10;
+    } else {
+      cat1_prob = vp9_cat1_prob_high12;
+      cat2_prob = vp9_cat2_prob_high12;
+      cat3_prob = vp9_cat3_prob_high12;
+      cat4_prob = vp9_cat4_prob_high12;
+      cat5_prob = vp9_cat5_prob_high12;
+      cat6_prob = vp9_cat6_prob_high12;
+    }
+  } else {
+    cat1_prob = vp9_cat1_prob;
+    cat2_prob = vp9_cat2_prob;
+    cat3_prob = vp9_cat3_prob;
+    cat4_prob = vp9_cat4_prob;
+    cat5_prob = vp9_cat5_prob;
+    cat6_prob = vp9_cat6_prob;
+  }
+#else
+  cat1_prob = vp9_cat1_prob;
+  cat2_prob = vp9_cat2_prob;
+  cat3_prob = vp9_cat3_prob;
+  cat4_prob = vp9_cat4_prob;
+  cat5_prob = vp9_cat5_prob;
+  cat6_prob = vp9_cat6_prob;
+#endif
 
   while (c < max_eob) {
-    int val;
+    int val = -1;
     band = *band_translate++;
     prob = coef_probs[band][ctx];
     if (!cm->frame_parallel_decoding_mode)
@@ -95,81 +138,63 @@
       prob = coef_probs[band][ctx];
     }
 
-    // ONE_CONTEXT_NODE_0_
     if (!vp9_read(r, prob[ONE_CONTEXT_NODE])) {
       INCREMENT_COUNT(ONE_TOKEN);
-      WRITE_COEF_CONTINUE(1, ONE_TOKEN);
-    }
-
-    INCREMENT_COUNT(TWO_TOKEN);
-
-    prob = vp9_pareto8_full[prob[PIVOT_NODE] - 1];
-
-    if (!vp9_read(r, prob[LOW_VAL_CONTEXT_NODE])) {
-      if (!vp9_read(r, prob[TWO_CONTEXT_NODE])) {
-        WRITE_COEF_CONTINUE(2, TWO_TOKEN);
+      token = ONE_TOKEN;
+      val = 1;
+    } else {
+      INCREMENT_COUNT(TWO_TOKEN);
+      token = vp9_read_tree(r, coeff_subtree_high,
+                            vp9_pareto8_full[prob[PIVOT_NODE] - 1]);
+      switch (token) {
+        case TWO_TOKEN:
+        case THREE_TOKEN:
+        case FOUR_TOKEN:
+          val = token;
+          break;
+        case CATEGORY1_TOKEN:
+          val = CAT1_MIN_VAL + read_coeff(cat1_prob, 1, r);
+          break;
+        case CATEGORY2_TOKEN:
+          val = CAT2_MIN_VAL + read_coeff(cat2_prob, 2, r);
+          break;
+        case CATEGORY3_TOKEN:
+          val = CAT3_MIN_VAL + read_coeff(cat3_prob, 3, r);
+          break;
+        case CATEGORY4_TOKEN:
+          val = CAT4_MIN_VAL + read_coeff(cat4_prob, 4, r);
+          break;
+        case CATEGORY5_TOKEN:
+          val = CAT5_MIN_VAL + read_coeff(cat5_prob, 5, r);
+          break;
+        case CATEGORY6_TOKEN:
+#if CONFIG_VP9_HIGHBITDEPTH
+          switch (cm->bit_depth) {
+            case VPX_BITS_8:
+              val = CAT6_MIN_VAL + read_coeff(cat6_prob, 14, r);
+              break;
+            case VPX_BITS_10:
+              val = CAT6_MIN_VAL + read_coeff(cat6_prob, 16, r);
+              break;
+            case VPX_BITS_12:
+              val = CAT6_MIN_VAL + read_coeff(cat6_prob, 18, r);
+              break;
+            default:
+              assert(0);
+              return -1;
+          }
+#else
+          val = CAT6_MIN_VAL + read_coeff(cat6_prob, 14, r);
+#endif
+          break;
       }
-      if (!vp9_read(r, prob[THREE_CONTEXT_NODE])) {
-        WRITE_COEF_CONTINUE(3, THREE_TOKEN);
-      }
-      WRITE_COEF_CONTINUE(4, FOUR_TOKEN);
     }
-
-    if (!vp9_read(r, prob[HIGH_LOW_CONTEXT_NODE])) {
-      if (!vp9_read(r, prob[CAT_ONE_CONTEXT_NODE])) {
-        val = CAT1_MIN_VAL;
-        ADJUST_COEF(vp9_cat1_prob[0], 0);
-        WRITE_COEF_CONTINUE(val, CATEGORY1_TOKEN);
-      }
-      val = CAT2_MIN_VAL;
-      ADJUST_COEF(vp9_cat2_prob[0], 1);
-      ADJUST_COEF(vp9_cat2_prob[1], 0);
-      WRITE_COEF_CONTINUE(val, CATEGORY2_TOKEN);
-    }
-
-    if (!vp9_read(r, prob[CAT_THREEFOUR_CONTEXT_NODE])) {
-      if (!vp9_read(r, prob[CAT_THREE_CONTEXT_NODE])) {
-        val = CAT3_MIN_VAL;
-        ADJUST_COEF(vp9_cat3_prob[0], 2);
-        ADJUST_COEF(vp9_cat3_prob[1], 1);
-        ADJUST_COEF(vp9_cat3_prob[2], 0);
-        WRITE_COEF_CONTINUE(val, CATEGORY3_TOKEN);
-      }
-      val = CAT4_MIN_VAL;
-      ADJUST_COEF(vp9_cat4_prob[0], 3);
-      ADJUST_COEF(vp9_cat4_prob[1], 2);
-      ADJUST_COEF(vp9_cat4_prob[2], 1);
-      ADJUST_COEF(vp9_cat4_prob[3], 0);
-      WRITE_COEF_CONTINUE(val, CATEGORY4_TOKEN);
-    }
-
-    if (!vp9_read(r, prob[CAT_FIVE_CONTEXT_NODE])) {
-      val = CAT5_MIN_VAL;
-      ADJUST_COEF(vp9_cat5_prob[0], 4);
-      ADJUST_COEF(vp9_cat5_prob[1], 3);
-      ADJUST_COEF(vp9_cat5_prob[2], 2);
-      ADJUST_COEF(vp9_cat5_prob[3], 1);
-      ADJUST_COEF(vp9_cat5_prob[4], 0);
-      WRITE_COEF_CONTINUE(val, CATEGORY5_TOKEN);
-    }
-    val = 0;
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[0]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[1]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[2]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[3]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[4]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[5]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[6]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[7]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[8]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[9]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[10]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[11]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[12]);
-    val = (val << 1) | vp9_read(r, vp9_cat6_prob[13]);
-    val += CAT6_MIN_VAL;
-
-    WRITE_COEF_CONTINUE(val, CATEGORY6_TOKEN);
+    v = (val * dqv) >> dq_shift;
+    dqcoeff[scan[c]] = vp9_read_bit(r) ? -v : v;
+    token_cache[scan[c]] = vp9_pt_energy_class[token];
+    ++c;
+    ctx = get_coef_context(nb, token_cache, c);
+    dqv = dq[1];
   }
 
   return c;
diff --git a/vp9/decoder/vp9_dthread.c b/vp9/decoder/vp9_dthread.c
index b82ea6a..62ea6c1 100644
--- a/vp9/decoder/vp9_dthread.c
+++ b/vp9/decoder/vp9_dthread.c
@@ -99,7 +99,7 @@
 
   for (r = start; r < stop; r += num_lf_workers) {
     const int mi_row = r << MI_BLOCK_SIZE_LOG2;
-    MODE_INFO **const mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
+    MODE_INFO *const mi = cm->mi + mi_row * cm->mi_stride;
 
     for (c = 0; c < sb_cols; ++c) {
       const int mi_col = c << MI_BLOCK_SIZE_LOG2;
@@ -121,10 +121,10 @@
 }
 
 // Row-based multi-threaded loopfilter hook
-static int loop_filter_row_worker(void *arg1, void *arg2) {
-  TileWorkerData *const tile_data = (TileWorkerData*)arg1;
+static int loop_filter_row_worker(TileWorkerData *const tile_data,
+                                  void *unused) {
   LFWorkerData *const lf_data = &tile_data->lfdata;
-  (void) arg2;
+  (void)unused;
   loop_filter_rows_mt(lf_data->frame_buffer, lf_data->cm, lf_data->planes,
                       lf_data->start, lf_data->stop, lf_data->y_only,
                       lf_data->lf_sync, lf_data->num_lf_workers);
@@ -145,15 +145,13 @@
   const int num_workers = MIN(pbi->max_threads & ~1, tile_cols);
   int i;
 
-  // Allocate memory used in thread synchronization.
-  // This always needs to be done even if frame_filter_level is 0.
+  if (!frame_filter_level) return;
+
   if (!lf_sync->sync_range || cm->last_height != cm->height) {
     vp9_loop_filter_dealloc(lf_sync);
-    vp9_loop_filter_alloc(cm, lf_sync, sb_rows, cm->width);
+    vp9_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width);
   }
 
-  if (!frame_filter_level) return;
-
   vp9_loop_filter_frame_init(cm, frame_filter_level);
 
   // Initialize cur_sb_col to -1 for all SB rows.
@@ -216,7 +214,7 @@
 }
 
 // Allocate memory for lf row synchronization
-void vp9_loop_filter_alloc(VP9_COMMON *cm, VP9LfSync *lf_sync, int rows,
+void vp9_loop_filter_alloc(VP9LfSync *lf_sync, VP9_COMMON *cm, int rows,
                            int width) {
   lf_sync->rows = rows;
 #if CONFIG_MULTITHREAD
diff --git a/vp9/decoder/vp9_dthread.h b/vp9/decoder/vp9_dthread.h
index 8b02ef7..b1fbdeb 100644
--- a/vp9/decoder/vp9_dthread.h
+++ b/vp9/decoder/vp9_dthread.h
@@ -42,8 +42,8 @@
 } VP9LfSync;
 
 // Allocate memory for loopfilter row synchronization.
-void vp9_loop_filter_alloc(struct VP9Common *cm, VP9LfSync *lf_sync,
-                           int rows, int width);
+void vp9_loop_filter_alloc(VP9LfSync *lf_sync, VP9_COMMON *cm, int rows,
+                           int width);
 
 // Deallocate loopfilter synchronization related mutex and data.
 void vp9_loop_filter_dealloc(VP9LfSync *lf_sync);
diff --git a/vp9/encoder/vp9_aq_complexity.c b/vp9/encoder/vp9_aq_complexity.c
index 33f9239..f7fca0c 100644
--- a/vp9/encoder/vp9_aq_complexity.c
+++ b/vp9/encoder/vp9_aq_complexity.c
@@ -23,9 +23,9 @@
 static const double aq_c_transitions[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
   {{1.0, 1.0, 1.0}, {1.0, 0.25, 0.0}, {1.0, 0.5, 0.25}};
 
-static int get_aq_c_strength(int q_index) {
+static int get_aq_c_strength(int q_index, vpx_bit_depth_t bit_depth) {
   // Approximate base quatizer (truncated to int)
-  int base_quant = vp9_ac_quant(q_index, 0) / 4;
+  const int base_quant = vp9_ac_quant(q_index, 0, bit_depth) / 4;
   return (base_quant > 20) + (base_quant > 45);
 }
 
@@ -40,7 +40,7 @@
       cpi->refresh_alt_ref_frame ||
       (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref)) {
     int segment;
-    const int aq_strength = get_aq_c_strength(cm->base_qindex);
+    const int aq_strength = get_aq_c_strength(cm->base_qindex, cm->bit_depth);
     const int active_segments = aq_c_active_segments[aq_strength];
 
     // Clear down the segment map.
@@ -70,7 +70,8 @@
     for (segment = 1; segment < active_segments; ++segment) {
       int qindex_delta =
           vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type, cm->base_qindex,
-                                     aq_c_q_adj_factor[aq_strength][segment]);
+                                     aq_c_q_adj_factor[aq_strength][segment],
+                                     cm->bit_depth);
 
       // For AQ complexity mode, we dont allow Q0 in a segment if the base
       // Q is not 0. Q0 (lossless) implies 4x4 only and in AQ mode 2 a segment
@@ -115,7 +116,7 @@
     // It is converted to bits * 256 units.
     const int target_rate = (cpi->rc.sb64_target_rate * xmis * ymis * 256) /
                             (bw * bh);
-    const int aq_strength = get_aq_c_strength(cm->base_qindex);
+    const int aq_strength = get_aq_c_strength(cm->base_qindex, cm->bit_depth);
     const int active_segments = aq_c_active_segments[aq_strength];
 
     // The number of segments considered and the transition points used to
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index e7f0daa..514ff7a 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -200,7 +200,7 @@
 
     // Rate target ratio to set q delta.
     const float rate_ratio_qdelta = 2.0;
-    const double q = vp9_convert_qindex_to_q(cm->base_qindex);
+    const double q = vp9_convert_qindex_to_q(cm->base_qindex, cm->bit_depth);
     vp9_clear_system_state();
     // Some of these parameters may be set via codec-control function later.
     cr->max_sbs_perframe = 10;
@@ -242,7 +242,8 @@
     // Set the q delta for segment 1.
     qindex_delta = vp9_compute_qdelta_by_rate(rc, cm->frame_type,
                                               cm->base_qindex,
-                                              rate_ratio_qdelta);
+                                              rate_ratio_qdelta,
+                                              cm->bit_depth);
     // TODO(marpan): Incorporate the actual-vs-target rate over/undershoot from
     // previous encoded frame.
     if (-qindex_delta > cr->max_qdelta_perc * cm->base_qindex / 100)
diff --git a/vp9/encoder/vp9_aq_variance.c b/vp9/encoder/vp9_aq_variance.c
index 56db95e..b96f00f 100644
--- a/vp9/encoder/vp9_aq_variance.c
+++ b/vp9/encoder/vp9_aq_variance.c
@@ -75,7 +75,7 @@
 void vp9_vaq_frame_setup(VP9_COMP *cpi) {
   VP9_COMMON *cm = &cpi->common;
   struct segmentation *seg = &cm->seg;
-  const double base_q = vp9_convert_qindex_to_q(cm->base_qindex);
+  const double base_q = vp9_convert_qindex_to_q(cm->base_qindex, cm->bit_depth);
   const int base_rdmult = vp9_compute_rd_mult(cpi, cm->base_qindex +
                                               cm->y_dc_delta_q);
   int i;
@@ -99,7 +99,8 @@
         continue;
       }
 
-      qindex_delta = vp9_compute_qdelta(&cpi->rc, base_q, base_q * Q_RATIO(i));
+      qindex_delta = vp9_compute_qdelta(&cpi->rc, base_q, base_q * Q_RATIO(i),
+                                        cm->bit_depth);
       vp9_set_segdata(seg, SEGMENT_ID(i), SEG_LVL_ALT_Q, qindex_delta);
       vp9_enable_segfeature(seg, SEGMENT_ID(i), SEG_LVL_ALT_Q);
 
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index b605248..f658dda 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -189,7 +189,7 @@
 // This function encodes the reference frame
 static void write_ref_frames(const VP9_COMMON *cm, const MACROBLOCKD *xd,
                              vp9_writer *w) {
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  const MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const int is_compound = has_second_ref(mbmi);
   const int segment_id = mbmi->segment_id;
 
@@ -329,11 +329,12 @@
 }
 
 static void write_mb_modes_kf(const VP9_COMMON *cm, const MACROBLOCKD *xd,
-                              MODE_INFO **mi_8x8, vp9_writer *w) {
+                              MODE_INFO *mi_8x8, vp9_writer *w) {
   const struct segmentation *const seg = &cm->seg;
-  const MODE_INFO *const mi = mi_8x8[0];
-  const MODE_INFO *const above_mi = mi_8x8[-xd->mi_stride];
-  const MODE_INFO *const left_mi = xd->left_available ? mi_8x8[-1] : NULL;
+  const MODE_INFO *const mi = mi_8x8;
+  const MODE_INFO *const above_mi = mi_8x8[-xd->mi_stride].src_mi;
+  const MODE_INFO *const left_mi =
+      xd->left_available ? mi_8x8[-1].src_mi : NULL;
   const MB_MODE_INFO *const mbmi = &mi->mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
 
@@ -372,8 +373,8 @@
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
   MODE_INFO *m;
 
-  xd->mi = cm->mi_grid_visible + (mi_row * cm->mi_stride + mi_col);
-  m = xd->mi[0];
+  xd->mi = cm->mi + (mi_row * cm->mi_stride + mi_col);
+  m = xd->mi;
 
   set_mi_row_col(xd, tile,
                  mi_row, num_8x8_blocks_high_lookup[m->mbmi.sb_type],
@@ -427,7 +428,7 @@
   if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
     return;
 
-  m = cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col];
+  m = cm->mi[mi_row * cm->mi_stride + mi_col].src_mi;
 
   partition = partition_lookup[bsl][m->mbmi.sb_type];
   write_partition(cm, xd, bs, mi_row, mi_col, partition, bsize, w);
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index b726383..767bd7f 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -28,8 +28,8 @@
 
 struct macroblock_plane {
   DECLARE_ALIGNED(16, int16_t, src_diff[64 * 64]);
-  int16_t *qcoeff;
-  int16_t *coeff;
+  tran_low_t *qcoeff;
+  tran_low_t *coeff;
   uint16_t *eobs;
   struct buf_2d src;
 
@@ -119,8 +119,12 @@
   // Used to store sub partition's choices.
   MV pred_mv[MAX_REF_FRAMES];
 
-  void (*fwd_txm4x4)(const int16_t *input, int16_t *output, int stride);
-  void (*itxm_add)(const int16_t *input, uint8_t *dest, int stride, int eob);
+  void (*fwd_txm4x4)(const int16_t *input, tran_low_t *output, int stride);
+  void (*itxm_add)(const tran_low_t *input, uint8_t *dest, int stride, int eob);
+#if CONFIG_VP9_HIGHBITDEPTH
+  void (*high_itxm_add)(const tran_low_t *input, uint8_t *dest, int stride,
+                        int eob, int bd);
+#endif
 };
 
 #ifdef __cplusplus
diff --git a/vp9/encoder/vp9_context_tree.c b/vp9/encoder/vp9_context_tree.c
index 9b7a932..12acc51 100644
--- a/vp9/encoder/vp9_context_tree.c
+++ b/vp9/encoder/vp9_context_tree.c
@@ -30,13 +30,13 @@
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     for (k = 0; k < 3; ++k) {
       CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(int16_t)));
+                      vpx_memalign(16, num_pix * sizeof(*ctx->coeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(int16_t)));
+                      vpx_memalign(16, num_pix * sizeof(*ctx->qcoeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->dqcoeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(int16_t)));
+                      vpx_memalign(16, num_pix * sizeof(*ctx->dqcoeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->eobs[i][k],
-                      vpx_memalign(16, num_pix * sizeof(uint16_t)));
+                      vpx_memalign(16, num_pix * sizeof(*ctx->eobs[i][k])));
       ctx->coeff_pbuf[i][k]   = ctx->coeff[i][k];
       ctx->qcoeff_pbuf[i][k]  = ctx->qcoeff[i][k];
       ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k];
diff --git a/vp9/encoder/vp9_context_tree.h b/vp9/encoder/vp9_context_tree.h
index 236389b..97f0741 100644
--- a/vp9/encoder/vp9_context_tree.h
+++ b/vp9/encoder/vp9_context_tree.h
@@ -19,15 +19,15 @@
 typedef struct {
   MODE_INFO mic;
   uint8_t *zcoeff_blk;
-  int16_t *coeff[MAX_MB_PLANE][3];
-  int16_t *qcoeff[MAX_MB_PLANE][3];
-  int16_t *dqcoeff[MAX_MB_PLANE][3];
+  tran_low_t *coeff[MAX_MB_PLANE][3];
+  tran_low_t *qcoeff[MAX_MB_PLANE][3];
+  tran_low_t *dqcoeff[MAX_MB_PLANE][3];
   uint16_t *eobs[MAX_MB_PLANE][3];
 
   // dual buffer pointers, 0: in use, 1: best in store
-  int16_t *coeff_pbuf[MAX_MB_PLANE][3];
-  int16_t *qcoeff_pbuf[MAX_MB_PLANE][3];
-  int16_t *dqcoeff_pbuf[MAX_MB_PLANE][3];
+  tran_low_t *coeff_pbuf[MAX_MB_PLANE][3];
+  tran_low_t *qcoeff_pbuf[MAX_MB_PLANE][3];
+  tran_low_t *dqcoeff_pbuf[MAX_MB_PLANE][3];
   uint16_t *eobs_pbuf[MAX_MB_PLANE][3];
 
   int is_coded;
diff --git a/vp9/encoder/vp9_dct.c b/vp9/encoder/vp9_dct.c
index 59222f0..eff8996 100644
--- a/vp9/encoder/vp9_dct.c
+++ b/vp9/encoder/vp9_dct.c
@@ -18,15 +18,17 @@
 #include "vp9/common/vp9_idct.h"
 #include "vp9/common/vp9_systemdependent.h"
 
-static INLINE int fdct_round_shift(int input) {
-  int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-  assert(INT16_MIN <= rv && rv <= INT16_MAX);
+static INLINE tran_high_t fdct_round_shift(tran_high_t input) {
+  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  // TODO(debargha, peter.derivaz): Find new bounds for this assert
+  // and make the bounds consts.
+  // assert(INT16_MIN <= rv && rv <= INT16_MAX);
   return rv;
 }
 
-static void fdct4(const int16_t *input, int16_t *output) {
-  int16_t step[4];
-  int temp1, temp2;
+static void fdct4(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t step[4];
+  tran_high_t temp1, temp2;
 
   step[0] = input[0] + input[3];
   step[1] = input[1] + input[2];
@@ -43,9 +45,9 @@
   output[3] = fdct_round_shift(temp2);
 }
 
-void vp9_fdct4x4_1_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fdct4x4_1_c(const int16_t *input, tran_low_t *output, int stride) {
   int r, c;
-  int16_t sum = 0;
+  tran_low_t sum = 0;
   for (r = 0; r < 4; ++r)
     for (c = 0; c < 4; ++c)
       sum += input[r * stride + c];
@@ -54,7 +56,7 @@
   output[1] = 0;
 }
 
-void vp9_fdct4x4_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fdct4x4_c(const int16_t *input, tran_low_t *output, int stride) {
   // The 2D transform is done with two passes which are actually pretty
   // similar. In the first one, we transform the columns and transpose
   // the results. In the second one, we transform the rows. To achieve that,
@@ -63,22 +65,23 @@
   // in normal/row positions).
   int pass;
   // We need an intermediate buffer between passes.
-  int16_t intermediate[4 * 4];
-  const int16_t *in = input;
-  int16_t *out = intermediate;
+  tran_low_t intermediate[4 * 4];
+  const int16_t *in_pass0 = input;
+  const tran_low_t *in = NULL;
+  tran_low_t *out = intermediate;
   // Do the two transform/transpose passes
   for (pass = 0; pass < 2; ++pass) {
-    /*canbe16*/ int input[4];
-    /*canbe16*/ int step[4];
-    /*needs32*/ int temp1, temp2;
+    tran_high_t input[4];      // canbe16
+    tran_high_t step[4];       // canbe16
+    tran_high_t temp1, temp2;  // needs32
     int i;
     for (i = 0; i < 4; ++i) {
       // Load inputs.
       if (0 == pass) {
-        input[0] = in[0 * stride] * 16;
-        input[1] = in[1 * stride] * 16;
-        input[2] = in[2 * stride] * 16;
-        input[3] = in[3 * stride] * 16;
+        input[0] = in_pass0[0 * stride] * 16;
+        input[1] = in_pass0[1 * stride] * 16;
+        input[2] = in_pass0[2 * stride] * 16;
+        input[3] = in_pass0[3 * stride] * 16;
         if (i == 0 && input[0]) {
           input[0] += 1;
         }
@@ -102,6 +105,7 @@
       out[1] = fdct_round_shift(temp1);
       out[3] = fdct_round_shift(temp2);
       // Do next column (which is a transposed row in second/horizontal pass)
+      in_pass0++;
       in++;
       out += 4;
     }
@@ -119,9 +123,9 @@
   }
 }
 
-static void fadst4(const int16_t *input, int16_t *output) {
-  int x0, x1, x2, x3;
-  int s0, s1, s2, s3, s4, s5, s6, s7;
+static void fadst4(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t x0, x1, x2, x3;
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
 
   x0 = input[0];
   x1 = input[1];
@@ -166,15 +170,15 @@
   { fadst4, fadst4 }   // ADST_ADST = 3
 };
 
-void vp9_fht4x4_c(const int16_t *input, int16_t *output,
+void vp9_fht4x4_c(const int16_t *input, tran_low_t *output,
                   int stride, int tx_type) {
   if (tx_type == DCT_DCT) {
     vp9_fdct4x4_c(input, output, stride);
   } else {
-    int16_t out[4 * 4];
-    int16_t *outptr = &out[0];
+    tran_low_t out[4 * 4];
+    tran_low_t *outptr = &out[0];
     int i, j;
-    int16_t temp_in[4], temp_out[4];
+    tran_low_t temp_in[4], temp_out[4];
     const transform_2d ht = FHT_4[tx_type];
 
     // Columns
@@ -199,10 +203,10 @@
   }
 }
 
-static void fdct8(const int16_t *input, int16_t *output) {
-  /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
-  /*needs32*/ int t0, t1, t2, t3;
-  /*canbe16*/ int x0, x1, x2, x3;
+static void fdct8(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
+  tran_high_t t0, t1, t2, t3;                  // needs32
+  tran_high_t x0, x1, x2, x3;                  // canbe16
 
   // stage 1
   s0 = input[0] + input[7];
@@ -251,9 +255,9 @@
   output[7] = fdct_round_shift(t3);
 }
 
-void vp9_fdct8x8_1_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fdct8x8_1_c(const int16_t *input, tran_low_t *output, int stride) {
   int r, c;
-  int16_t sum = 0;
+  tran_low_t sum = 0;
   for (r = 0; r < 8; ++r)
     for (c = 0; c < 8; ++c)
       sum += input[r * stride + c];
@@ -262,16 +266,16 @@
   output[1] = 0;
 }
 
-void vp9_fdct8x8_c(const int16_t *input, int16_t *final_output, int stride) {
+void vp9_fdct8x8_c(const int16_t *input, tran_low_t *final_output, int stride) {
   int i, j;
-  int16_t intermediate[64];
+  tran_low_t intermediate[64];
 
   // Transform columns
   {
-    int16_t *output = intermediate;
-    /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
-    /*needs32*/ int t0, t1, t2, t3;
-    /*canbe16*/ int x0, x1, x2, x3;
+    tran_low_t *output = intermediate;
+    tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
+    tran_high_t t0, t1, t2, t3;                  // needs32
+    tran_high_t x0, x1, x2, x3;                  // canbe16
 
     int i;
     for (i = 0; i < 8; i++) {
@@ -333,9 +337,9 @@
   }
 }
 
-void vp9_fdct16x16_1_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fdct16x16_1_c(const int16_t *input, tran_low_t *output, int stride) {
   int r, c;
-  int16_t sum = 0;
+  tran_low_t sum = 0;
   for (r = 0; r < 16; ++r)
     for (c = 0; c < 16; ++c)
       sum += input[r * stride + c];
@@ -344,7 +348,7 @@
   output[1] = 0;
 }
 
-void vp9_fdct16x16_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fdct16x16_c(const int16_t *input, tran_low_t *output, int stride) {
   // The 2D transform is done with two passes which are actually pretty
   // similar. In the first one, we transform the columns and transpose
   // the results. In the second one, we transform the rows. To achieve that,
@@ -353,37 +357,38 @@
   // in normal/row positions).
   int pass;
   // We need an intermediate buffer between passes.
-  int16_t intermediate[256];
-  const int16_t *in = input;
-  int16_t *out = intermediate;
+  tran_low_t intermediate[256];
+  const int16_t *in_pass0 = input;
+  const tran_low_t *in = NULL;
+  tran_low_t *out = intermediate;
   // Do the two transform/transpose passes
   for (pass = 0; pass < 2; ++pass) {
-    /*canbe16*/ int step1[8];
-    /*canbe16*/ int step2[8];
-    /*canbe16*/ int step3[8];
-    /*canbe16*/ int input[8];
-    /*needs32*/ int temp1, temp2;
+    tran_high_t step1[8];      // canbe16
+    tran_high_t step2[8];      // canbe16
+    tran_high_t step3[8];      // canbe16
+    tran_high_t input[8];      // canbe16
+    tran_high_t temp1, temp2;  // needs32
     int i;
     for (i = 0; i < 16; i++) {
       if (0 == pass) {
         // Calculate input for the first 8 results.
-        input[0] = (in[0 * stride] + in[15 * stride]) * 4;
-        input[1] = (in[1 * stride] + in[14 * stride]) * 4;
-        input[2] = (in[2 * stride] + in[13 * stride]) * 4;
-        input[3] = (in[3 * stride] + in[12 * stride]) * 4;
-        input[4] = (in[4 * stride] + in[11 * stride]) * 4;
-        input[5] = (in[5 * stride] + in[10 * stride]) * 4;
-        input[6] = (in[6 * stride] + in[ 9 * stride]) * 4;
-        input[7] = (in[7 * stride] + in[ 8 * stride]) * 4;
+        input[0] = (in_pass0[0 * stride] + in_pass0[15 * stride]) * 4;
+        input[1] = (in_pass0[1 * stride] + in_pass0[14 * stride]) * 4;
+        input[2] = (in_pass0[2 * stride] + in_pass0[13 * stride]) * 4;
+        input[3] = (in_pass0[3 * stride] + in_pass0[12 * stride]) * 4;
+        input[4] = (in_pass0[4 * stride] + in_pass0[11 * stride]) * 4;
+        input[5] = (in_pass0[5 * stride] + in_pass0[10 * stride]) * 4;
+        input[6] = (in_pass0[6 * stride] + in_pass0[ 9 * stride]) * 4;
+        input[7] = (in_pass0[7 * stride] + in_pass0[ 8 * stride]) * 4;
         // Calculate input for the next 8 results.
-        step1[0] = (in[7 * stride] - in[ 8 * stride]) * 4;
-        step1[1] = (in[6 * stride] - in[ 9 * stride]) * 4;
-        step1[2] = (in[5 * stride] - in[10 * stride]) * 4;
-        step1[3] = (in[4 * stride] - in[11 * stride]) * 4;
-        step1[4] = (in[3 * stride] - in[12 * stride]) * 4;
-        step1[5] = (in[2 * stride] - in[13 * stride]) * 4;
-        step1[6] = (in[1 * stride] - in[14 * stride]) * 4;
-        step1[7] = (in[0 * stride] - in[15 * stride]) * 4;
+        step1[0] = (in_pass0[7 * stride] - in_pass0[ 8 * stride]) * 4;
+        step1[1] = (in_pass0[6 * stride] - in_pass0[ 9 * stride]) * 4;
+        step1[2] = (in_pass0[5 * stride] - in_pass0[10 * stride]) * 4;
+        step1[3] = (in_pass0[4 * stride] - in_pass0[11 * stride]) * 4;
+        step1[4] = (in_pass0[3 * stride] - in_pass0[12 * stride]) * 4;
+        step1[5] = (in_pass0[2 * stride] - in_pass0[13 * stride]) * 4;
+        step1[6] = (in_pass0[1 * stride] - in_pass0[14 * stride]) * 4;
+        step1[7] = (in_pass0[0 * stride] - in_pass0[15 * stride]) * 4;
       } else {
         // Calculate input for the first 8 results.
         input[0] = ((in[0 * 16] + 1) >> 2) + ((in[15 * 16] + 1) >> 2);
@@ -406,9 +411,9 @@
       }
       // Work on the first eight values; fdct8(input, even_results);
       {
-        /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
-        /*needs32*/ int t0, t1, t2, t3;
-        /*canbe16*/ int x0, x1, x2, x3;
+        tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
+        tran_high_t t0, t1, t2, t3;                  // needs32
+        tran_high_t x0, x1, x2, x3;                  // canbe16
 
         // stage 1
         s0 = input[0] + input[7];
@@ -514,6 +519,7 @@
       }
       // Do next column (which is a transposed row in second/horizontal pass)
       in++;
+      in_pass0++;
       out += 16;
     }
     // Setup in/out for next pass.
@@ -522,17 +528,17 @@
   }
 }
 
-static void fadst8(const int16_t *input, int16_t *output) {
-  int s0, s1, s2, s3, s4, s5, s6, s7;
+static void fadst8(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
 
-  int x0 = input[7];
-  int x1 = input[0];
-  int x2 = input[5];
-  int x3 = input[2];
-  int x4 = input[3];
-  int x5 = input[4];
-  int x6 = input[1];
-  int x7 = input[6];
+  tran_high_t x0 = input[7];
+  tran_high_t x1 = input[0];
+  tran_high_t x2 = input[5];
+  tran_high_t x3 = input[2];
+  tran_high_t x4 = input[3];
+  tran_high_t x5 = input[4];
+  tran_high_t x6 = input[1];
+  tran_high_t x7 = input[6];
 
   // stage 1
   s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
@@ -600,15 +606,15 @@
   { fadst8, fadst8 }   // ADST_ADST = 3
 };
 
-void vp9_fht8x8_c(const int16_t *input, int16_t *output,
+void vp9_fht8x8_c(const int16_t *input, tran_low_t *output,
                   int stride, int tx_type) {
   if (tx_type == DCT_DCT) {
     vp9_fdct8x8_c(input, output, stride);
   } else {
-    int16_t out[64];
-    int16_t *outptr = &out[0];
+    tran_low_t out[64];
+    tran_low_t *outptr = &out[0];
     int i, j;
-    int16_t temp_in[8], temp_out[8];
+    tran_low_t temp_in[8], temp_out[8];
     const transform_2d ht = FHT_8[tx_type];
 
     // Columns
@@ -633,17 +639,18 @@
 
 /* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
    pixel. */
-void vp9_fwht4x4_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fwht4x4_c(const int16_t *input, tran_low_t *output, int stride) {
   int i;
-  int a1, b1, c1, d1, e1;
-  const int16_t *ip = input;
-  int16_t *op = output;
+  tran_high_t a1, b1, c1, d1, e1;
+  const int16_t *ip_pass0 = input;
+  const tran_low_t *ip = NULL;
+  tran_low_t *op = output;
 
   for (i = 0; i < 4; i++) {
-    a1 = ip[0 * stride];
-    b1 = ip[1 * stride];
-    c1 = ip[2 * stride];
-    d1 = ip[3 * stride];
+    a1 = ip_pass0[0 * stride];
+    b1 = ip_pass0[1 * stride];
+    c1 = ip_pass0[2 * stride];
+    d1 = ip_pass0[3 * stride];
 
     a1 += b1;
     d1 = d1 - c1;
@@ -657,7 +664,7 @@
     op[8] = d1;
     op[12] = b1;
 
-    ip++;
+    ip_pass0++;
     op++;
   }
   ip = output;
@@ -687,12 +694,12 @@
 }
 
 // Rewrote to use same algorithm as others.
-static void fdct16(const int16_t in[16], int16_t out[16]) {
-  /*canbe16*/ int step1[8];
-  /*canbe16*/ int step2[8];
-  /*canbe16*/ int step3[8];
-  /*canbe16*/ int input[8];
-  /*needs32*/ int temp1, temp2;
+static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
+  tran_high_t step1[8];      // canbe16
+  tran_high_t step2[8];      // canbe16
+  tran_high_t step3[8];      // canbe16
+  tran_high_t input[8];      // canbe16
+  tran_high_t temp1, temp2;  // needs32
 
   // step 1
   input[0] = in[0] + in[15];
@@ -715,9 +722,9 @@
 
   // fdct8(step, step);
   {
-    /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
-    /*needs32*/ int t0, t1, t2, t3;
-    /*canbe16*/ int x0, x1, x2, x3;
+    tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;  // canbe16
+    tran_high_t t0, t1, t2, t3;                  // needs32
+    tran_high_t x0, x1, x2, x3;                  // canbe16
 
     // stage 1
     s0 = input[0] + input[7];
@@ -828,25 +835,26 @@
   out[15] = fdct_round_shift(temp2);
 }
 
-static void fadst16(const int16_t *input, int16_t *output) {
-  int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
+static void fadst16(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7, s8;
+  tran_high_t s9, s10, s11, s12, s13, s14, s15;
 
-  int x0 = input[15];
-  int x1 = input[0];
-  int x2 = input[13];
-  int x3 = input[2];
-  int x4 = input[11];
-  int x5 = input[4];
-  int x6 = input[9];
-  int x7 = input[6];
-  int x8 = input[7];
-  int x9 = input[8];
-  int x10 = input[5];
-  int x11 = input[10];
-  int x12 = input[3];
-  int x13 = input[12];
-  int x14 = input[1];
-  int x15 = input[14];
+  tran_high_t x0 = input[15];
+  tran_high_t x1 = input[0];
+  tran_high_t x2 = input[13];
+  tran_high_t x3 = input[2];
+  tran_high_t x4 = input[11];
+  tran_high_t x5 = input[4];
+  tran_high_t x6 = input[9];
+  tran_high_t x7 = input[6];
+  tran_high_t x8 = input[7];
+  tran_high_t x9 = input[8];
+  tran_high_t x10 = input[5];
+  tran_high_t x11 = input[10];
+  tran_high_t x12 = input[3];
+  tran_high_t x13 = input[12];
+  tran_high_t x14 = input[1];
+  tran_high_t x15 = input[14];
 
   // stage 1
   s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
@@ -997,15 +1005,15 @@
   { fadst16, fadst16 }   // ADST_ADST = 3
 };
 
-void vp9_fht16x16_c(const int16_t *input, int16_t *output,
+void vp9_fht16x16_c(const int16_t *input, tran_low_t *output,
                     int stride, int tx_type) {
   if (tx_type == DCT_DCT) {
     vp9_fdct16x16_c(input, output, stride);
   } else {
-    int16_t out[256];
-    int16_t *outptr = &out[0];
+    tran_low_t out[256];
+    tran_low_t *outptr = &out[0];
     int i, j;
-    int16_t temp_in[16], temp_out[16];
+    tran_low_t temp_in[16], temp_out[16];
     const transform_2d ht = FHT_16[tx_type];
 
     // Columns
@@ -1028,19 +1036,21 @@
   }
 }
 
-static INLINE int dct_32_round(int input) {
-  int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-  assert(-131072 <= rv && rv <= 131071);
+static INLINE tran_high_t dct_32_round(tran_high_t input) {
+  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  // TODO(debargha, peter.derivaz): Find new bounds for this assert,
+  // and make the bounds consts.
+  // assert(-131072 <= rv && rv <= 131071);
   return rv;
 }
 
-static INLINE int half_round_shift(int input) {
-  int rv = (input + 1 + (input < 0)) >> 2;
+static INLINE tran_high_t half_round_shift(tran_high_t input) {
+  tran_high_t rv = (input + 1 + (input < 0)) >> 2;
   return rv;
 }
 
-static void fdct32(const int *input, int *output, int round) {
-  int step[32];
+static void fdct32(const tran_high_t *input, tran_high_t *output, int round) {
+  tran_high_t step[32];
   // Stage 1
   step[0] = input[0] + input[(32 - 1)];
   step[1] = input[1] + input[(32 - 2)];
@@ -1362,9 +1372,9 @@
   output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
 }
 
-void vp9_fdct32x32_1_c(const int16_t *input, int16_t *output, int stride) {
+void vp9_fdct32x32_1_c(const int16_t *input, tran_low_t *output, int stride) {
   int r, c;
-  int16_t sum = 0;
+  tran_low_t sum = 0;
   for (r = 0; r < 32; ++r)
     for (c = 0; c < 32; ++c)
       sum += input[r * stride + c];
@@ -1373,13 +1383,13 @@
   output[1] = 0;
 }
 
-void vp9_fdct32x32_c(const int16_t *input, int16_t *out, int stride) {
+void vp9_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
   int i, j;
-  int output[32 * 32];
+  tran_high_t output[32 * 32];
 
   // Columns
   for (i = 0; i < 32; ++i) {
-    int temp_in[32], temp_out[32];
+    tran_high_t temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
       temp_in[j] = input[j * stride + i] * 4;
     fdct32(temp_in, temp_out, 0);
@@ -1389,7 +1399,7 @@
 
   // Rows
   for (i = 0; i < 32; ++i) {
-    int temp_in[32], temp_out[32];
+    tran_high_t temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
       temp_in[j] = output[j + i * 32];
     fdct32(temp_in, temp_out, 0);
@@ -1401,13 +1411,13 @@
 // Note that although we use dct_32_round in dct32 computation flow,
 // this 2d fdct32x32 for rate-distortion optimization loop is operating
 // within 16 bits precision.
-void vp9_fdct32x32_rd_c(const int16_t *input, int16_t *out, int stride) {
+void vp9_fdct32x32_rd_c(const int16_t *input, tran_low_t *out, int stride) {
   int i, j;
-  int output[32 * 32];
+  tran_high_t output[32 * 32];
 
   // Columns
   for (i = 0; i < 32; ++i) {
-    int temp_in[32], temp_out[32];
+    tran_high_t temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
       temp_in[j] = input[j * stride + i] * 4;
     fdct32(temp_in, temp_out, 0);
@@ -1420,7 +1430,7 @@
 
   // Rows
   for (i = 0; i < 32; ++i) {
-    int temp_in[32], temp_out[32];
+    tran_high_t temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
       temp_in[j] = output[j + i * 32];
     fdct32(temp_in, temp_out, 1);
@@ -1428,3 +1438,61 @@
       out[j + i * 32] = temp_out[j];
   }
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_fdct4x4_c(const int16_t *input, tran_low_t *output, int stride) {
+  vp9_fdct4x4_c(input, output, stride);
+}
+
+void vp9_high_fht4x4_c(const int16_t *input, tran_low_t *output,
+                       int stride, int tx_type) {
+  vp9_fht4x4_c(input, output, stride, tx_type);
+}
+
+void vp9_high_fdct8x8_1_c(const int16_t *input, tran_low_t *final_output,
+                          int stride) {
+  vp9_fdct8x8_1_c(input, final_output, stride);
+}
+
+void vp9_high_fdct8x8_c(const int16_t *input, tran_low_t *final_output,
+                        int stride) {
+  vp9_fdct8x8_c(input, final_output, stride);
+}
+
+void vp9_high_fdct16x16_1_c(const int16_t *input, tran_low_t *output,
+                            int stride) {
+  vp9_fdct16x16_1_c(input, output, stride);
+}
+
+void vp9_high_fdct16x16_c(const int16_t *input, tran_low_t *output,
+                          int stride) {
+  vp9_fdct16x16_c(input, output, stride);
+}
+
+void vp9_high_fht8x8_c(const int16_t *input, tran_low_t *output,
+                  int stride, int tx_type) {
+  vp9_fht8x8_c(input, output, stride, tx_type);
+}
+
+void vp9_high_fwht4x4_c(const int16_t *input, tran_low_t *output, int stride) {
+  vp9_fwht4x4_c(input, output, stride);
+}
+
+void vp9_high_fht16x16_c(const int16_t *input, tran_low_t *output,
+                    int stride, int tx_type) {
+  vp9_fht16x16_c(input, output, stride, tx_type);
+}
+
+void vp9_high_fdct32x32_1_c(const int16_t *input, tran_low_t *out, int stride) {
+  vp9_fdct32x32_1_c(input, out, stride);
+}
+
+void vp9_high_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
+  vp9_fdct32x32_c(input, out, stride);
+}
+
+void vp9_high_fdct32x32_rd_c(const int16_t *input, tran_low_t *out,
+                             int stride) {
+  vp9_fdct32x32_rd_c(input, out, stride);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_denoiser.c b/vp9/encoder/vp9_denoiser.c
index c4cf5ee..681b2a5 100644
--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -89,9 +89,9 @@
   int total_adj = 0;
   int shift_inc = 1;
 
-  /* If motion_magnitude is small, making the denoiser more aggressive by
-   * increasing the adjustment for each level. Add another increment for
-   * blocks that are labeled for increase denoising. */
+  // If motion_magnitude is small, making the denoiser more aggressive by
+  // increasing the adjustment for each level. Add another increment for
+  // blocks that are labeled for increase denoising.
   if (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) {
     if (increase_denoising) {
       shift_inc = 2;
@@ -213,7 +213,7 @@
   int sse_diff = ctx->zeromv_sse - ctx->newmv_sse;
   MV_REFERENCE_FRAME frame;
   MACROBLOCKD *filter_mbd = &mb->e_mbd;
-  MB_MODE_INFO *mbmi = &filter_mbd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &filter_mbd->mi[0].src_mi->mbmi;
 
   MB_MODE_INFO saved_mbmi;
   int i, j;
diff --git a/vp9/encoder/vp9_denoiser.h b/vp9/encoder/vp9_denoiser.h
index a913add..fa714b1 100644
--- a/vp9/encoder/vp9_denoiser.h
+++ b/vp9/encoder/vp9_denoiser.h
@@ -18,7 +18,7 @@
 extern "C" {
 #endif
 
-#define MOTION_MAGNITUDE_THRESHOLD (8*3)
+#define MOTION_MAGNITUDE_THRESHOLD (8 * 3)
 
 typedef enum vp9_denoiser_decision {
   COPY_BLOCK,
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 72ced05..be5ee7b 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -130,8 +130,8 @@
                                         int mi_row,
                                         int mi_col) {
   const int idx_str = xd->mi_stride * mi_row + mi_col;
-  xd->mi = cm->mi_grid_visible + idx_str;
-  xd->mi[0] = cm->mi + idx_str;
+  xd->mi = cm->mi + idx_str;
+  xd->mi[0].src_mi = &xd->mi[0];
 }
 
 static void set_offsets(VP9_COMP *cpi, const TileInfo *const tile,
@@ -148,7 +148,7 @@
 
   set_modeinfo_offsets(cm, xd, mi_row, mi_col);
 
-  mbmi = &xd->mi[0]->mbmi;
+  mbmi = &xd->mi[0].src_mi->mbmi;
 
   // Set up destination pointers.
   vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
@@ -197,7 +197,7 @@
   for (j = 0; j < block_height; ++j)
     for (i = 0; i < block_width; ++i) {
       if (mi_row + j < cm->mi_rows && mi_col + i < cm->mi_cols)
-        xd->mi[j * xd->mi_stride + i] = xd->mi[0];
+        xd->mi[j * xd->mi_stride + i].src_mi = &xd->mi[0];
     }
 }
 
@@ -207,7 +207,7 @@
   if (cpi->common.mi_cols > mi_col && cpi->common.mi_rows > mi_row) {
     MACROBLOCKD *const xd = &cpi->mb.e_mbd;
     set_modeinfo_offsets(&cpi->common, xd, mi_row, mi_col);
-    xd->mi[0]->mbmi.sb_type = bsize;
+    xd->mi[0].src_mi->mbmi.sb_type = bsize;
     duplicate_mode_info_in_sb(&cpi->common, xd, mi_row, mi_col, bsize);
   }
 }
@@ -405,13 +405,13 @@
   if (cm->frame_type != KEY_FRAME) {
     vp9_setup_pre_planes(xd, 0, yv12, mi_row, mi_col, sf);
 
-    xd->mi[0]->mbmi.ref_frame[0] = LAST_FRAME;
-    xd->mi[0]->mbmi.sb_type = BLOCK_64X64;
+    xd->mi[0].src_mi->mbmi.ref_frame[0] = LAST_FRAME;
+    xd->mi[0].src_mi->mbmi.sb_type = BLOCK_64X64;
     vp9_find_best_ref_mvs(xd, cm->allow_high_precision_mv,
-                          xd->mi[0]->mbmi.ref_mvs[LAST_FRAME],
+                          xd->mi[0].src_mi->mbmi.ref_mvs[LAST_FRAME],
                           &nearest_mv, &near_mv);
 
-    xd->mi[0]->mbmi.mv[0] = nearest_mv;
+    xd->mi[0].src_mi->mbmi.mv[0] = nearest_mv;
     vp9_build_inter_predictors_sby(xd, mi_row, mi_col, BLOCK_64X64);
 
     d = xd->plane[0].dst.buf;
@@ -515,8 +515,8 @@
   struct macroblock_plane *const p = x->plane;
   struct macroblockd_plane *const pd = xd->plane;
   MODE_INFO *mi = &ctx->mic;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  MODE_INFO *mi_addr = xd->mi[0];
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
+  MODE_INFO *mi_addr = &xd->mi[0];
   const struct segmentation *const seg = &cm->seg;
 
   const int mis = cm->mi_stride;
@@ -527,6 +527,7 @@
   assert(mi->mbmi.sb_type == bsize);
 
   *mi_addr = *mi;
+  mi_addr->src_mi = mi_addr;
 
   // If segmentation in use
   if (seg->enabled && output_enabled) {
@@ -540,7 +541,7 @@
     // Else for cyclic refresh mode update the segment map, set the segment id
     // and then update the quantizer.
     if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
-      vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi,
+      vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0].src_mi->mbmi,
                                         mi_row, mi_col, bsize, 1);
     }
   }
@@ -566,7 +567,7 @@
     for (x_idx = 0; x_idx < mi_width; x_idx++)
       if ((xd->mb_to_right_edge >> (3 + MI_SIZE_LOG2)) + mi_width > x_idx
         && (xd->mb_to_bottom_edge >> (3 + MI_SIZE_LOG2)) + mi_height > y) {
-        xd->mi[x_idx + y * mis] = mi_addr;
+        xd->mi[x_idx + y * mis].src_mi = mi_addr;
       }
 
   if (cpi->oxcf.aq_mode)
@@ -654,13 +655,13 @@
 static void set_mode_info_seg_skip(MACROBLOCK *x, TX_MODE tx_mode, int *rate,
                                    int64_t *dist, BLOCK_SIZE bsize) {
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   INTERP_FILTER filter_ref;
 
   if (xd->up_available)
-    filter_ref = xd->mi[-xd->mi_stride]->mbmi.interp_filter;
+    filter_ref = xd->mi[-xd->mi_stride].src_mi->mbmi.interp_filter;
   else if (xd->left_available)
-    filter_ref = xd->mi[-1]->mbmi.interp_filter;
+    filter_ref = xd->mi[-1].src_mi->mbmi.interp_filter;
   else
     filter_ref = EIGHTTAP;
 
@@ -675,7 +676,7 @@
   mbmi->mv[0].as_int = 0;
   mbmi->interp_filter = filter_ref;
 
-  xd->mi[0]->bmi[0].as_mv[0].as_int = 0;
+  xd->mi[0].src_mi->bmi[0].as_mv[0].as_int = 0;
   x->skip = 1;
 
   *rate = 0;
@@ -717,7 +718,7 @@
   }
 
   set_offsets(cpi, tile, mi_row, mi_col, bsize);
-  mbmi = &xd->mi[0]->mbmi;
+  mbmi = &xd->mi[0].src_mi->mbmi;
   mbmi->sb_type = bsize;
 
   for (i = 0; i < MAX_MB_PLANE; ++i) {
@@ -799,7 +800,7 @@
 
 static void update_stats(VP9_COMMON *cm, const MACROBLOCK *x) {
   const MACROBLOCKD *const xd = &x->e_mbd;
-  const MODE_INFO *const mi = xd->mi[0];
+  const MODE_INFO *const mi = xd->mi[0].src_mi;
   const MB_MODE_INFO *const mbmi = &mi->mbmi;
 
   if (!frame_is_intra_only(cm)) {
@@ -1011,15 +1012,15 @@
 
 static void set_partial_b64x64_partition(MODE_INFO *mi, int mis,
     int bh_in, int bw_in, int row8x8_remaining, int col8x8_remaining,
-    BLOCK_SIZE bsize, MODE_INFO **mi_8x8) {
+    BLOCK_SIZE bsize, MODE_INFO *mi_8x8) {
   int bh = bh_in;
   int r, c;
   for (r = 0; r < MI_BLOCK_SIZE; r += bh) {
     int bw = bw_in;
     for (c = 0; c < MI_BLOCK_SIZE; c += bw) {
       const int index = r * mis + c;
-      mi_8x8[index] = mi + index;
-      mi_8x8[index]->mbmi.sb_type = find_partition_size(bsize,
+      mi_8x8[index].src_mi = mi + index;
+      mi_8x8[index].src_mi->mbmi.sb_type = find_partition_size(bsize,
           row8x8_remaining - r, col8x8_remaining - c, &bh, &bw);
     }
   }
@@ -1031,7 +1032,7 @@
 // may not be allowed in which case this code attempts to choose the largest
 // allowable partition.
 static void set_fixed_partitioning(VP9_COMP *cpi, const TileInfo *const tile,
-                                   MODE_INFO **mi_8x8, int mi_row, int mi_col,
+                                   MODE_INFO *mi_8x8, int mi_row, int mi_col,
                                    BLOCK_SIZE bsize) {
   VP9_COMMON *const cm = &cpi->common;
   const int mis = cm->mi_stride;
@@ -1050,8 +1051,8 @@
     for (block_row = 0; block_row < MI_BLOCK_SIZE; block_row += bh) {
       for (block_col = 0; block_col < MI_BLOCK_SIZE; block_col += bw) {
         int index = block_row * mis + block_col;
-        mi_8x8[index] = mi_upper_left + index;
-        mi_8x8[index]->mbmi.sb_type = bsize;
+        mi_8x8[index].src_mi = mi_upper_left + index;
+        mi_8x8[index].src_mi->mbmi.sb_type = bsize;
       }
     }
   } else {
@@ -1061,20 +1062,21 @@
   }
 }
 
-static void copy_partitioning(VP9_COMMON *cm, MODE_INFO **mi_8x8,
-  MODE_INFO **prev_mi_8x8) {
+static void copy_partitioning(VP9_COMMON *cm, MODE_INFO *mi_8x8,
+  MODE_INFO *prev_mi_8x8) {
   const int mis = cm->mi_stride;
   int block_row, block_col;
 
   for (block_row = 0; block_row < 8; ++block_row) {
     for (block_col = 0; block_col < 8; ++block_col) {
-      MODE_INFO *const prev_mi = prev_mi_8x8[block_row * mis + block_col];
+      MODE_INFO *const prev_mi =
+          prev_mi_8x8[block_row * mis + block_col].src_mi;
       const BLOCK_SIZE sb_type = prev_mi ? prev_mi->mbmi.sb_type : 0;
 
       if (prev_mi) {
         const ptrdiff_t offset = prev_mi - cm->prev_mi;
-        mi_8x8[block_row * mis + block_col] = cm->mi + offset;
-        mi_8x8[block_row * mis + block_col]->mbmi.sb_type = sb_type;
+        mi_8x8[block_row * mis + block_col].src_mi = cm->mi + offset;
+        mi_8x8[block_row * mis + block_col].src_mi->mbmi.sb_type = sb_type;
       }
     }
   }
@@ -1082,8 +1084,8 @@
 
 static void constrain_copy_partitioning(VP9_COMP *const cpi,
                                         const TileInfo *const tile,
-                                        MODE_INFO **mi_8x8,
-                                        MODE_INFO **prev_mi_8x8,
+                                        MODE_INFO *mi_8x8,
+                                        MODE_INFO *prev_mi_8x8,
                                         int mi_row, int mi_col,
                                         BLOCK_SIZE bsize) {
   VP9_COMMON *const cm = &cpi->common;
@@ -1103,7 +1105,7 @@
     for (block_row = 0; block_row < MI_BLOCK_SIZE; block_row += bh) {
       for (block_col = 0; block_col < MI_BLOCK_SIZE; block_col += bw) {
         const int index = block_row * mis + block_col;
-        MODE_INFO *prev_mi = prev_mi_8x8[index];
+        MODE_INFO *prev_mi = prev_mi_8x8[index].src_mi;
         const BLOCK_SIZE sb_type = prev_mi ? prev_mi->mbmi.sb_type : 0;
         // Use previous partition if block size is not larger than bsize.
         if (prev_mi && sb_type <= bsize) {
@@ -1112,18 +1114,18 @@
             for (block_col2 = 0; block_col2 < bw; ++block_col2) {
               const int index2 = (block_row + block_row2) * mis +
                   block_col + block_col2;
-              prev_mi = prev_mi_8x8[index2];
+              prev_mi = prev_mi_8x8[index2].src_mi;
               if (prev_mi) {
                 const ptrdiff_t offset = prev_mi - cm->prev_mi;
-                mi_8x8[index2] = cm->mi + offset;
-                mi_8x8[index2]->mbmi.sb_type = prev_mi->mbmi.sb_type;
+                mi_8x8[index2].src_mi = cm->mi + offset;
+                mi_8x8[index2].src_mi->mbmi.sb_type = prev_mi->mbmi.sb_type;
               }
             }
           }
         } else {
           // Otherwise, use fixed partition of size bsize.
-          mi_8x8[index] = mi_upper_left + index;
-          mi_8x8[index]->mbmi.sb_type = bsize;
+          mi_8x8[index].src_mi = mi_upper_left + index;
+          mi_8x8[index].src_mi->mbmi.sb_type = bsize;
         }
       }
     }
@@ -1149,7 +1151,7 @@
 
 static void set_source_var_based_partition(VP9_COMP *cpi,
                                            const TileInfo *const tile,
-                                           MODE_INFO **mi_8x8,
+                                           MODE_INFO *mi_8x8,
                                            int mi_row, int mi_col) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCK *const x = &cpi->mb;
@@ -1187,8 +1189,8 @@
         d16[j] = cpi->source_diff_var + offset + boffset;
 
         index = b_mi_row * mis + b_mi_col;
-        mi_8x8[index] = mi_upper_left + index;
-        mi_8x8[index]->mbmi.sb_type = BLOCK_16X16;
+        mi_8x8[index].src_mi = mi_upper_left + index;
+        mi_8x8[index].src_mi->mbmi.sb_type = BLOCK_16X16;
 
         // TODO(yunqingwang): If d16[j].var is very large, use 8x8 partition
         // size to further improve quality.
@@ -1209,8 +1211,8 @@
         d32[i].var = d32[i].sse - (((int64_t)d32[i].sum * d32[i].sum) >> 10);
 
         index = coord_lookup[i*4].row * mis + coord_lookup[i*4].col;
-        mi_8x8[index] = mi_upper_left + index;
-        mi_8x8[index]->mbmi.sb_type = BLOCK_32X32;
+        mi_8x8[index].src_mi = mi_upper_left + index;
+        mi_8x8[index].src_mi->mbmi.sb_type = BLOCK_32X32;
       }
     }
 
@@ -1221,8 +1223,8 @@
 
       // Use 64x64 partition
       if (is_larger_better) {
-        mi_8x8[0] = mi_upper_left;
-        mi_8x8[0]->mbmi.sb_type = BLOCK_64X64;
+        mi_8x8[0].src_mi = mi_upper_left;
+        mi_8x8[0].src_mi->mbmi.sb_type = BLOCK_64X64;
       }
     }
   } else {   // partial in-image SB64
@@ -1263,7 +1265,7 @@
   return this_sad < 2 * threshold;
 }
 
-static int sb_has_motion(const VP9_COMMON *cm, MODE_INFO **prev_mi_8x8,
+static int sb_has_motion(const VP9_COMMON *cm, MODE_INFO *prev_mi_8x8,
                          const int motion_thresh) {
   const int mis = cm->mi_stride;
   int block_row, block_col;
@@ -1271,7 +1273,8 @@
   if (cm->prev_mi) {
     for (block_row = 0; block_row < 8; ++block_row) {
       for (block_col = 0; block_col < 8; ++block_col) {
-        const MODE_INFO *prev_mi = prev_mi_8x8[block_row * mis + block_col];
+        const MODE_INFO *prev_mi =
+            prev_mi_8x8[block_row * mis + block_col].src_mi;
         if (prev_mi) {
           if (abs(prev_mi->mbmi.mv[0].as_mv.row) > motion_thresh ||
               abs(prev_mi->mbmi.mv[0].as_mv.col) > motion_thresh)
@@ -1288,15 +1291,17 @@
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCK *const x = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const struct segmentation *const seg = &cm->seg;
 
-  *(xd->mi[0]) = ctx->mic;
+  *(xd->mi[0].src_mi) = ctx->mic;
+  xd->mi[0].src_mi = &xd->mi[0];
+
 
   // For in frame adaptive Q, check for reseting the segment_id and updating
   // the cyclic refresh map.
   if ((cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) && seg->enabled) {
-    vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi,
+    vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0].src_mi->mbmi,
                                       mi_row, mi_col, bsize, 1);
     vp9_init_plane_quantizers(cpi, x);
   }
@@ -1353,9 +1358,9 @@
 
   if (bsize >= BLOCK_8X8) {
     const int idx_str = xd->mi_stride * mi_row + mi_col;
-    MODE_INFO ** mi_8x8 = cm->mi_grid_visible + idx_str;
+    MODE_INFO *mi_8x8 = cm->mi[idx_str].src_mi;
     ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
-    subsize = mi_8x8[0]->mbmi.sb_type;
+    subsize = mi_8x8[0].src_mi->mbmi.sb_type;
   } else {
     ctx = 0;
     subsize = BLOCK_4X4;
@@ -1408,7 +1413,7 @@
 
 static void rd_use_partition(VP9_COMP *cpi,
                              const TileInfo *const tile,
-                             MODE_INFO **mi_8x8,
+                             MODE_INFO *mi_8x8,
                              TOKENEXTRA **tp, int mi_row, int mi_col,
                              BLOCK_SIZE bsize, int *rate, int64_t *dist,
                              int do_recon, PC_TREE *pc_tree) {
@@ -1435,7 +1440,7 @@
   int64_t chosen_rd = INT64_MAX;
   BLOCK_SIZE sub_subsize = BLOCK_4X4;
   int splits_below = 0;
-  BLOCK_SIZE bs_type = mi_8x8[0]->mbmi.sb_type;
+  BLOCK_SIZE bs_type = mi_8x8[0].src_mi->mbmi.sb_type;
   int do_partition_search = 1;
   PICK_MODE_CONTEXT *ctx = &pc_tree->none;
 
@@ -1465,7 +1470,7 @@
       splits_below = 1;
       for (i = 0; i < 4; i++) {
         int jj = i >> 1, ii = i & 0x01;
-        MODE_INFO * this_mi = mi_8x8[jj * bss * mis + ii * bss];
+        MODE_INFO *this_mi = mi_8x8[jj * bss * mis + ii * bss].src_mi;
         if (this_mi && this_mi->mbmi.sb_type >= sub_subsize) {
           splits_below = 0;
         }
@@ -1489,7 +1494,7 @@
       }
 
       restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
-      mi_8x8[0]->mbmi.sb_type = bs_type;
+      mi_8x8[0].src_mi->mbmi.sb_type = bs_type;
       pc_tree->partitioning = partition;
     }
   }
@@ -1647,7 +1652,7 @@
 
   // If last_part is better set the partitioning to that.
   if (last_part_rd < chosen_rd) {
-    mi_8x8[0]->mbmi.sb_type = bsize;
+    mi_8x8[0].src_mi->mbmi.sb_type = bsize;
     if (bsize >= BLOCK_8X8)
       pc_tree->partitioning = partition;
     chosen_rate = last_part_rate;
@@ -1714,7 +1719,7 @@
 //
 // The min and max are assumed to have been initialized prior to calling this
 // function so repeat calls can accumulate a min and max of more than one sb64.
-static void get_sb_partition_size_range(MACROBLOCKD *xd, MODE_INFO **mi_8x8,
+static void get_sb_partition_size_range(MACROBLOCKD *xd, MODE_INFO *mi_8x8,
                                         BLOCK_SIZE *min_block_size,
                                         BLOCK_SIZE *max_block_size,
                                         int bs_hist[BLOCK_SIZES]) {
@@ -1726,7 +1731,7 @@
   // Check the sb_type for each block that belongs to this region.
   for (i = 0; i < sb_height_in_blocks; ++i) {
     for (j = 0; j < sb_width_in_blocks; ++j) {
-      MODE_INFO * mi = mi_8x8[index+j];
+      MODE_INFO *mi = mi_8x8[index+j].src_mi;
       BLOCK_SIZE sb_type = mi ? mi->mbmi.sb_type : 0;
       bs_hist[sb_type]++;
       *min_block_size = MIN(*min_block_size, sb_type);
@@ -1753,9 +1758,9 @@
                                     BLOCK_SIZE *max_block_size) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
-  MODE_INFO **mi = xd->mi;
-  const int left_in_image = xd->left_available && mi[-1];
-  const int above_in_image = xd->up_available && mi[-xd->mi_stride];
+  MODE_INFO *mi = xd->mi[0].src_mi;
+  const int left_in_image = xd->left_available && mi[-1].src_mi;
+  const int above_in_image = xd->up_available && mi[-xd->mi_stride].src_mi;
   const int row8x8_remaining = tile->mi_row_end - mi_row;
   const int col8x8_remaining = tile->mi_col_end - mi_col;
   int bh, bw;
@@ -1774,19 +1779,20 @@
     // passed in values for min and max as a starting point.
     // Find the min and max partition used in previous frame at this location
     if (cm->frame_type != KEY_FRAME) {
-      MODE_INFO **const prev_mi =
-          &cm->prev_mi_grid_visible[mi_row * xd->mi_stride + mi_col];
+      MODE_INFO *prev_mi =
+          cm->prev_mip + cm->mi_stride + 1 + mi_row * xd->mi_stride + mi_col;
+
       get_sb_partition_size_range(xd, prev_mi, &min_size, &max_size, bs_hist);
     }
     // Find the min and max partition sizes used in the left SB64
     if (left_in_image) {
-      MODE_INFO **left_sb64_mi = &mi[-MI_BLOCK_SIZE];
+      MODE_INFO *left_sb64_mi = mi[-MI_BLOCK_SIZE].src_mi;
       get_sb_partition_size_range(xd, left_sb64_mi, &min_size, &max_size,
                                   bs_hist);
     }
     // Find the min and max partition sizes used in the above SB64.
     if (above_in_image) {
-      MODE_INFO **above_sb64_mi = &mi[-xd->mi_stride * MI_BLOCK_SIZE];
+      MODE_INFO *above_sb64_mi = mi[-xd->mi_stride * MI_BLOCK_SIZE].src_mi;
       get_sb_partition_size_range(xd, above_sb64_mi, &min_size, &max_size,
                                   bs_hist);
     }
@@ -1848,10 +1854,10 @@
                                  BLOCK_SIZE *max_block_size) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
-  MODE_INFO **mi_8x8 = xd->mi;
-  const int left_in_image = xd->left_available && mi_8x8[-1];
+  MODE_INFO *mi_8x8 = xd->mi;
+  const int left_in_image = xd->left_available && mi_8x8[-1].src_mi;
   const int above_in_image = xd->up_available &&
-                             mi_8x8[-xd->mi_stride];
+                             mi_8x8[-xd->mi_stride].src_mi;
   int row8x8_remaining = tile->mi_row_end - mi_row;
   int col8x8_remaining = tile->mi_col_end - mi_col;
   int bh, bw;
@@ -1864,15 +1870,15 @@
   if (search_range_ctrl &&
       (left_in_image || above_in_image || cm->frame_type != KEY_FRAME)) {
     int block;
-    MODE_INFO **mi;
+    MODE_INFO *mi;
     BLOCK_SIZE sb_type;
 
     // Find the min and max partition sizes used in the left SB64.
     if (left_in_image) {
       MODE_INFO *cur_mi;
-      mi = &mi_8x8[-1];
+      mi = mi_8x8[-1].src_mi;
       for (block = 0; block < MI_BLOCK_SIZE; ++block) {
-        cur_mi = mi[block * xd->mi_stride];
+        cur_mi = mi[block * xd->mi_stride].src_mi;
         sb_type = cur_mi ? cur_mi->mbmi.sb_type : 0;
         min_size = MIN(min_size, sb_type);
         max_size = MAX(max_size, sb_type);
@@ -1880,9 +1886,9 @@
     }
     // Find the min and max partition sizes used in the above SB64.
     if (above_in_image) {
-      mi = &mi_8x8[-xd->mi_stride * MI_BLOCK_SIZE];
+      mi = mi_8x8[-xd->mi_stride * MI_BLOCK_SIZE].src_mi;
       for (block = 0; block < MI_BLOCK_SIZE; ++block) {
-        sb_type = mi[block] ? mi[block]->mbmi.sb_type : 0;
+        sb_type = mi[block].src_mi ? mi[block].src_mi->mbmi.sb_type : 0;
         min_size = MIN(min_size, sb_type);
         max_size = MAX(max_size, sb_type);
       }
@@ -1912,8 +1918,10 @@
   int idx, idy;
 
   MODE_INFO *mi;
-  MODE_INFO **prev_mi =
-      &cm->prev_mi_grid_visible[mi_row * cm->mi_stride + mi_col];
+  const int idx_str = cm->mi_stride * mi_row + mi_col;
+  MODE_INFO *prev_mi = (cm->prev_mip + cm->mi_stride + 1 + idx_str)->src_mi;
+
+
   BLOCK_SIZE bs, min_size, max_size;
 
   min_size = BLOCK_64X64;
@@ -1922,7 +1930,7 @@
   if (prev_mi) {
     for (idy = 0; idy < mi_height; ++idy) {
       for (idx = 0; idx < mi_width; ++idx) {
-        mi = prev_mi[idy * cm->mi_stride + idx];
+        mi = prev_mi[idy * cm->mi_stride + idx].src_mi;
         bs = mi ? mi->mbmi.sb_type : bsize;
         min_size = MIN(min_size, bs);
         max_size = MAX(max_size, bs);
@@ -1932,7 +1940,7 @@
 
   if (xd->left_available) {
     for (idy = 0; idy < mi_height; ++idy) {
-      mi = xd->mi[idy * cm->mi_stride - 1];
+      mi = xd->mi[idy * cm->mi_stride - 1].src_mi;
       bs = mi ? mi->mbmi.sb_type : bsize;
       min_size = MIN(min_size, bs);
       max_size = MAX(max_size, bs);
@@ -1941,7 +1949,7 @@
 
   if (xd->up_available) {
     for (idx = 0; idx < mi_width; ++idx) {
-      mi = xd->mi[idx - cm->mi_stride];
+      mi = xd->mi[idx - cm->mi_stride].src_mi;
       bs = mi ? mi->mbmi.sb_type : bsize;
       min_size = MIN(min_size, bs);
       max_size = MAX(max_size, bs);
@@ -2466,14 +2474,17 @@
     vp9_zero(cpi->mb.pred_mv);
     cpi->pc_root->index = 0;
 
+    // TODO(yunqingwang): use_lastframe_partitioning is no longer used in good-
+    // quality encoding. Need to evaluate it in real-time encoding later to
+    // decide if it can be removed too. And then, do the code cleanup.
     if ((sf->partition_search_type == SEARCH_PARTITION &&
          sf->use_lastframe_partitioning) ||
          sf->partition_search_type == FIXED_PARTITION ||
          sf->partition_search_type == VAR_BASED_PARTITION ||
          sf->partition_search_type == VAR_BASED_FIXED_PARTITION) {
       const int idx_str = cm->mi_stride * mi_row + mi_col;
-      MODE_INFO **mi = cm->mi_grid_visible + idx_str;
-      MODE_INFO **prev_mi = cm->prev_mi_grid_visible + idx_str;
+      MODE_INFO *mi = cm->mi + idx_str;
+      MODE_INFO *prev_mi = (cm->prev_mip + cm->mi_stride + 1 + idx_str)->src_mi;
       cpi->mb.source_variance = UINT_MAX;
       if (sf->partition_search_type == FIXED_PARTITION) {
         set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64);
@@ -2579,12 +2590,12 @@
 static void reset_skip_tx_size(VP9_COMMON *cm, TX_SIZE max_tx_size) {
   int mi_row, mi_col;
   const int mis = cm->mi_stride;
-  MODE_INFO **mi_ptr = cm->mi_grid_visible;
+  MODE_INFO *mi_ptr = cm->mi;
 
   for (mi_row = 0; mi_row < cm->mi_rows; ++mi_row, mi_ptr += mis) {
     for (mi_col = 0; mi_col < cm->mi_cols; ++mi_col) {
-      if (mi_ptr[mi_col]->mbmi.tx_size > max_tx_size)
-        mi_ptr[mi_col]->mbmi.tx_size = max_tx_size;
+      if (mi_ptr[mi_col].src_mi->mbmi.tx_size > max_tx_size)
+        mi_ptr[mi_col].src_mi->mbmi.tx_size = max_tx_size;
     }
   }
 }
@@ -2621,7 +2632,7 @@
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *mbmi;
   set_offsets(cpi, tile, mi_row, mi_col, bsize);
-  mbmi = &xd->mi[0]->mbmi;
+  mbmi = &xd->mi[0].src_mi->mbmi;
   mbmi->sb_type = bsize;
 
   if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cm->seg.enabled)
@@ -2652,27 +2663,27 @@
   switch (partition) {
     case PARTITION_NONE:
       set_modeinfo_offsets(cm, xd, mi_row, mi_col);
-      *(xd->mi[0]) = pc_tree->none.mic;
+      *(xd->mi[0].src_mi) = pc_tree->none.mic;
       duplicate_mode_info_in_sb(cm, xd, mi_row, mi_col, bsize);
       break;
     case PARTITION_VERT:
       set_modeinfo_offsets(cm, xd, mi_row, mi_col);
-      *(xd->mi[0]) = pc_tree->vertical[0].mic;
+      *(xd->mi[0].src_mi) = pc_tree->vertical[0].mic;
       duplicate_mode_info_in_sb(cm, xd, mi_row, mi_col, bsize);
 
       if (mi_col + hbs < cm->mi_cols) {
         set_modeinfo_offsets(cm, xd, mi_row, mi_col + hbs);
-        *(xd->mi[0]) = pc_tree->vertical[1].mic;
+        *(xd->mi[0].src_mi) = pc_tree->vertical[1].mic;
         duplicate_mode_info_in_sb(cm, xd, mi_row, mi_col + hbs, bsize);
       }
       break;
     case PARTITION_HORZ:
       set_modeinfo_offsets(cm, xd, mi_row, mi_col);
-      *(xd->mi[0]) = pc_tree->horizontal[0].mic;
+      *(xd->mi[0].src_mi) = pc_tree->horizontal[0].mic;
       duplicate_mode_info_in_sb(cm, xd, mi_row, mi_col, bsize);
       if (mi_row + hbs < cm->mi_rows) {
         set_modeinfo_offsets(cm, xd, mi_row + hbs, mi_col);
-        *(xd->mi[0]) = pc_tree->horizontal[1].mic;
+        *(xd->mi[0].src_mi) = pc_tree->horizontal[1].mic;
         duplicate_mode_info_in_sb(cm, xd, mi_row + hbs, mi_col, bsize);
       }
       break;
@@ -2751,7 +2762,7 @@
   if (partition_none_allowed) {
     nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col,
                         &this_rate, &this_dist, bsize, ctx);
-    ctx->mic.mbmi = xd->mi[0]->mbmi;
+    ctx->mic.mbmi = xd->mi[0].src_mi->mbmi;
     ctx->skip_txfm[0] = x->skip_txfm[0];
     ctx->skip = x->skip;
 
@@ -2836,7 +2847,7 @@
                         &this_rate, &this_dist, subsize,
                         &pc_tree->horizontal[0]);
 
-    pc_tree->horizontal[0].mic.mbmi = xd->mi[0]->mbmi;
+    pc_tree->horizontal[0].mic.mbmi = xd->mi[0].src_mi->mbmi;
     pc_tree->horizontal[0].skip_txfm[0] = x->skip_txfm[0];
     pc_tree->horizontal[0].skip = x->skip;
 
@@ -2848,7 +2859,7 @@
                           &this_rate, &this_dist, subsize,
                           &pc_tree->horizontal[1]);
 
-      pc_tree->horizontal[1].mic.mbmi = xd->mi[0]->mbmi;
+      pc_tree->horizontal[1].mic.mbmi = xd->mi[0].src_mi->mbmi;
       pc_tree->horizontal[1].skip_txfm[0] = x->skip_txfm[0];
       pc_tree->horizontal[1].skip = x->skip;
 
@@ -2880,7 +2891,7 @@
     nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col,
                         &this_rate, &this_dist, subsize,
                         &pc_tree->vertical[0]);
-    pc_tree->vertical[0].mic.mbmi = xd->mi[0]->mbmi;
+    pc_tree->vertical[0].mic.mbmi = xd->mi[0].src_mi->mbmi;
     pc_tree->vertical[0].skip_txfm[0] = x->skip_txfm[0];
     pc_tree->vertical[0].skip = x->skip;
     sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
@@ -2889,7 +2900,7 @@
       nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col + ms,
                           &this_rate, &this_dist, subsize,
                           &pc_tree->vertical[1]);
-      pc_tree->vertical[1].mic.mbmi = xd->mi[0]->mbmi;
+      pc_tree->vertical[1].mic.mbmi = xd->mi[0].src_mi->mbmi;
       pc_tree->vertical[1].skip_txfm[0] = x->skip_txfm[0];
       pc_tree->vertical[1].skip = x->skip;
       if (this_rate == INT_MAX) {
@@ -2954,7 +2965,7 @@
 
 static void nonrd_use_partition(VP9_COMP *cpi,
                                 const TileInfo *const tile,
-                                MODE_INFO **mi,
+                                MODE_INFO *mi,
                                 TOKENEXTRA **tp,
                                 int mi_row, int mi_col,
                                 BLOCK_SIZE bsize, int output_enabled,
@@ -2973,27 +2984,27 @@
   if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
     return;
 
-  subsize = (bsize >= BLOCK_8X8) ? mi[0]->mbmi.sb_type : BLOCK_4X4;
+  subsize = (bsize >= BLOCK_8X8) ? mi[0].src_mi->mbmi.sb_type : BLOCK_4X4;
   partition = partition_lookup[bsl][subsize];
 
   switch (partition) {
     case PARTITION_NONE:
       nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col, totrate, totdist,
                           subsize, &pc_tree->none);
-      pc_tree->none.mic.mbmi = xd->mi[0]->mbmi;
+      pc_tree->none.mic.mbmi = xd->mi[0].src_mi->mbmi;
       pc_tree->none.skip_txfm[0] = x->skip_txfm[0];
       pc_tree->none.skip = x->skip;
       break;
     case PARTITION_VERT:
       nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col, totrate, totdist,
                           subsize, &pc_tree->vertical[0]);
-      pc_tree->vertical[0].mic.mbmi = xd->mi[0]->mbmi;
+      pc_tree->vertical[0].mic.mbmi = xd->mi[0].src_mi->mbmi;
       pc_tree->vertical[0].skip_txfm[0] = x->skip_txfm[0];
       pc_tree->vertical[0].skip = x->skip;
       if (mi_col + hbs < cm->mi_cols) {
         nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col + hbs,
                             &rate, &dist, subsize, &pc_tree->vertical[1]);
-        pc_tree->vertical[1].mic.mbmi = xd->mi[0]->mbmi;
+        pc_tree->vertical[1].mic.mbmi = xd->mi[0].src_mi->mbmi;
         pc_tree->vertical[1].skip_txfm[0] = x->skip_txfm[0];
         pc_tree->vertical[1].skip = x->skip;
         if (rate != INT_MAX && dist != INT64_MAX &&
@@ -3006,13 +3017,13 @@
     case PARTITION_HORZ:
       nonrd_pick_sb_modes(cpi, tile, mi_row, mi_col, totrate, totdist,
                           subsize, &pc_tree->horizontal[0]);
-      pc_tree->horizontal[0].mic.mbmi = xd->mi[0]->mbmi;
+      pc_tree->horizontal[0].mic.mbmi = xd->mi[0].src_mi->mbmi;
       pc_tree->horizontal[0].skip_txfm[0] = x->skip_txfm[0];
       pc_tree->horizontal[0].skip = x->skip;
       if (mi_row + hbs < cm->mi_rows) {
         nonrd_pick_sb_modes(cpi, tile, mi_row + hbs, mi_col,
                             &rate, &dist, subsize, &pc_tree->horizontal[0]);
-        pc_tree->horizontal[1].mic.mbmi = xd->mi[0]->mbmi;
+        pc_tree->horizontal[1].mic.mbmi = xd->mi[0].src_mi->mbmi;
         pc_tree->horizontal[1].skip_txfm[0] = x->skip_txfm[0];
         pc_tree->horizontal[1].skip = x->skip;
         if (rate != INT_MAX && dist != INT64_MAX &&
@@ -3083,10 +3094,9 @@
     int dummy_rate = 0;
     int64_t dummy_dist = 0;
     const int idx_str = cm->mi_stride * mi_row + mi_col;
-    MODE_INFO **mi = cm->mi_grid_visible + idx_str;
-    MODE_INFO **prev_mi = cm->prev_mi_grid_visible + idx_str;
+    MODE_INFO *mi = cm->mi + idx_str;
+    MODE_INFO *prev_mi = (cm->prev_mip + cm->mi_stride + 1 + idx_str)->src_mi;
     BLOCK_SIZE bsize;
-
     x->in_static_area = 0;
     x->source_variance = UINT_MAX;
     vp9_zero(x->pred_mv);
@@ -3287,8 +3297,8 @@
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &x->e_mbd;
 
-  xd->mi = cm->mi_grid_visible;
-  xd->mi[0] = cm->mi;
+  xd->mi = cm->mi;
+  xd->mi[0].src_mi = &xd->mi[0];
 
   vp9_zero(cm->counts);
   vp9_zero(cpi->coef_counts);
@@ -3558,8 +3568,8 @@
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCK *const x = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MODE_INFO **mi_8x8 = xd->mi;
-  MODE_INFO *mi = mi_8x8[0];
+  MODE_INFO *mi_8x8 = xd->mi;
+  MODE_INFO *mi = mi_8x8;
   MB_MODE_INFO *mbmi = &mi->mbmi;
   const int seg_skip = vp9_segfeature_active(&cm->seg, mbmi->segment_id,
                                              SEG_LVL_SKIP);
@@ -3614,16 +3624,8 @@
 
     vp9_build_inter_predictors_sbuv(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
 
-    if (!x->skip) {
-      mbmi->skip = 1;
-      vp9_encode_sb(x, MAX(bsize, BLOCK_8X8));
-      vp9_tokenize_sb(cpi, t, !output_enabled, MAX(bsize, BLOCK_8X8));
-    } else {
-      mbmi->skip = 1;
-      if (output_enabled && !seg_skip)
-        cm->counts.skip[vp9_get_skip_context(xd)][1]++;
-      reset_skip_context(xd, MAX(bsize, BLOCK_8X8));
-    }
+    vp9_encode_sb(x, MAX(bsize, BLOCK_8X8));
+    vp9_tokenize_sb(cpi, t, !output_enabled, MAX(bsize, BLOCK_8X8));
   }
 
   if (output_enabled) {
@@ -3646,7 +3648,7 @@
       for (y = 0; y < mi_height; y++)
         for (x = 0; x < mi_width; x++)
           if (mi_col + x < cm->mi_cols && mi_row + y < cm->mi_rows)
-            mi_8x8[mis * y + x]->mbmi.tx_size = tx_size;
+            mi_8x8[mis * y + x].src_mi->mbmi.tx_size = tx_size;
     }
   }
 }
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 6678450..2eae149 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -103,13 +103,13 @@
   MACROBLOCKD *const xd = &mb->e_mbd;
   struct macroblock_plane *const p = &mb->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  const int ref = is_inter_block(&xd->mi[0]->mbmi);
+  const int ref = is_inter_block(&xd->mi[0].src_mi->mbmi);
   vp9_token_state tokens[1025][2];
   unsigned best_index[1025][2];
   uint8_t token_cache[1024];
-  const int16_t *const coeff = BLOCK_OFFSET(mb->plane[plane].coeff, block);
-  int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  const tran_low_t *const coeff = BLOCK_OFFSET(mb->plane[plane].coeff, block);
+  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   const int eob = p->eobs[block];
   const PLANE_TYPE type = pd->plane_type;
   const int default_eob = 16 << (tx_size << 1);
@@ -294,22 +294,33 @@
 }
 
 static INLINE void fdct32x32(int rd_transform,
-                             const int16_t *src, int16_t *dst, int src_stride) {
+                             const int16_t *src, tran_low_t *dst,
+                             int src_stride) {
   if (rd_transform)
     vp9_fdct32x32_rd(src, dst, src_stride);
   else
     vp9_fdct32x32(src, dst, src_stride);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE void high_fdct32x32(int rd_transform, const int16_t *src,
+                                  tran_low_t *dst, int src_stride) {
+  if (rd_transform)
+    vp9_high_fdct32x32_rd(src, dst, src_stride);
+  else
+    vp9_high_fdct32x32(src, dst, src_stride);
+}
+#endif
+
 void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block,
                         BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
   MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblock_plane *const p = &x->plane[plane];
   const struct macroblockd_plane *const pd = &xd->plane[plane];
   const scan_order *const scan_order = &vp9_default_scan_orders[tx_size];
-  int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
+  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   uint16_t *const eob = &p->eobs[block];
   const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
   int i, j;
@@ -357,9 +368,9 @@
   MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblock_plane *const p = &x->plane[plane];
   const struct macroblockd_plane *const pd = &xd->plane[plane];
-  int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
+  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   uint16_t *const eob = &p->eobs[block];
   const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
   int i, j;
@@ -405,9 +416,9 @@
   const struct macroblock_plane *const p = &x->plane[plane];
   const struct macroblockd_plane *const pd = &xd->plane[plane];
   const scan_order *const scan_order = &vp9_default_scan_orders[tx_size];
-  int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
+  tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   uint16_t *const eob = &p->eobs[block];
   const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
   int i, j;
@@ -458,7 +469,7 @@
   struct optimize_ctx *const ctx = args->ctx;
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   int i, j;
   uint8_t *dst;
   ENTROPY_CONTEXT *a, *l;
@@ -538,7 +549,7 @@
   MACROBLOCKD *const xd = &x->e_mbd;
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   int i, j;
   uint8_t *dst;
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
@@ -559,10 +570,15 @@
 void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
   MACROBLOCKD *const xd = &x->e_mbd;
   struct optimize_ctx ctx;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   struct encode_b_args arg = {x, &ctx, &mbmi->skip};
   int plane;
 
+  mbmi->skip = 1;
+
+  if (x->skip)
+    return;
+
   for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
     if (!x->skip_recode)
       vp9_subtract_plane(x, bsize, plane);
@@ -584,12 +600,12 @@
   struct encode_b_args* const args = arg;
   MACROBLOCK *const x = args->x;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  int16_t *coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-  int16_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *coeff = BLOCK_OFFSET(p->coeff, block);
+  tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  tran_low_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   const scan_order *scan_order;
   TX_TYPE tx_type;
   PREDICTION_MODE mode;
@@ -669,7 +685,7 @@
     case TX_4X4:
       tx_type = get_tx_type_4x4(pd->plane_type, xd, block);
       scan_order = &vp9_scan_orders[TX_4X4][tx_type];
-      mode = plane == 0 ? get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
+      mode = plane == 0 ? get_y_mode(xd->mi[0].src_mi, block) : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
@@ -716,7 +732,7 @@
 
 void vp9_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
   const MACROBLOCKD *const xd = &x->e_mbd;
-  struct encode_b_args arg = {x, NULL, &xd->mi[0]->mbmi.skip};
+  struct encode_b_args arg = {x, NULL, &xd->mi[0].src_mi->mbmi.skip};
 
   vp9_foreach_transformed_block_in_plane(xd, bsize, plane, encode_block_intra,
                                          &arg);
diff --git a/vp9/encoder/vp9_encodemv.c b/vp9/encoder/vp9_encodemv.c
index 9d42a12..0898395 100644
--- a/vp9/encoder/vp9_encodemv.c
+++ b/vp9/encoder/vp9_encodemv.c
@@ -242,7 +242,7 @@
 }
 
 void vp9_update_mv_count(VP9_COMMON *cm, const MACROBLOCKD *xd) {
-  const MODE_INFO *mi = xd->mi[0];
+  const MODE_INFO *mi = xd->mi[0].src_mi;
   const MB_MODE_INFO *const mbmi = &mi->mbmi;
 
   if (mbmi->sb_type < BLOCK_8X8) {
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 2ca91b9..bfd35d5 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -330,7 +330,8 @@
       seg->update_map = 1;
       seg->update_data = 1;
 
-      qi_delta = vp9_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875);
+      qi_delta = vp9_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875,
+                                    cm->bit_depth);
       vp9_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2);
       vp9_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
 
@@ -351,7 +352,8 @@
         seg->update_data = 1;
         seg->abs_delta = SEGMENT_DELTADATA;
 
-        qi_delta = vp9_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125);
+        qi_delta = vp9_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125,
+                                      cm->bit_depth);
         vp9_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2);
         vp9_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
 
@@ -411,15 +413,15 @@
 
 static void update_reference_segmentation_map(VP9_COMP *cpi) {
   VP9_COMMON *const cm = &cpi->common;
-  MODE_INFO **mi_8x8_ptr = cm->mi_grid_visible;
+  MODE_INFO *mi_8x8_ptr = cm->mi;
   uint8_t *cache_ptr = cm->last_frame_seg_map;
   int row, col;
 
   for (row = 0; row < cm->mi_rows; row++) {
-    MODE_INFO **mi_8x8 = mi_8x8_ptr;
+    MODE_INFO *mi_8x8 = mi_8x8_ptr;
     uint8_t *cache = cache_ptr;
     for (col = 0; col < cm->mi_cols; col++, mi_8x8++, cache++)
-      cache[0] = mi_8x8[0]->mbmi.segment_id;
+      cache[0] = mi_8x8[0].src_mi->mbmi.segment_id;
     mi_8x8_ptr += cm->mi_stride;
     cache_ptr += cm->mi_cols;
   }
@@ -556,6 +558,9 @@
 
   cm->profile = oxcf->profile;
   cm->bit_depth = oxcf->bit_depth;
+#if CONFIG_VP9_HIGHBITDEPTH
+  cm->use_highbitdepth = oxcf->use_highbitdepth;
+#endif
   cm->color_space = UNKNOWN;
 
   cm->width = oxcf->width;
@@ -599,6 +604,612 @@
                                            : maximum * bandwidth / 1000;
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+#define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF) \
+    cpi->fn_ptr[BT].sdf = SDF; \
+    cpi->fn_ptr[BT].sdaf = SDAF; \
+    cpi->fn_ptr[BT].vf = VF; \
+    cpi->fn_ptr[BT].svf = SVF; \
+    cpi->fn_ptr[BT].svaf = SVAF; \
+    cpi->fn_ptr[BT].sdx3f = SDX3F; \
+    cpi->fn_ptr[BT].sdx8f = SDX8F; \
+    cpi->fn_ptr[BT].sdx4df = SDX4DF;
+
+#define MAKE_BFP_SAD_WRAPPER(fnname) \
+static unsigned int fnname##_bits8(const uint8_t *src_ptr, \
+                                   int source_stride, \
+                                   const uint8_t *ref_ptr, \
+                                   int ref_stride) {  \
+  return fnname(src_ptr, source_stride, ref_ptr, ref_stride); \
+} \
+static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
+                                    int source_stride, \
+                                    const uint8_t *ref_ptr, \
+                                    int ref_stride) {  \
+  return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 2; \
+} \
+static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
+                                    int source_stride, \
+                                    const uint8_t *ref_ptr, \
+                                    int ref_stride) {  \
+  return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 4; \
+}
+
+#define MAKE_BFP_SADAVG_WRAPPER(fnname) static unsigned int \
+fnname##_bits8(const uint8_t *src_ptr, \
+               int source_stride, \
+               const uint8_t *ref_ptr, \
+               int ref_stride, \
+               const uint8_t *second_pred) {  \
+  return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred); \
+} \
+static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
+                                    int source_stride, \
+                                    const uint8_t *ref_ptr, \
+                                    int ref_stride, \
+                                    const uint8_t *second_pred) {  \
+  return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
+                second_pred) >> 2; \
+} \
+static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
+                                    int source_stride, \
+                                    const uint8_t *ref_ptr, \
+                                    int ref_stride, \
+                                    const uint8_t *second_pred) {  \
+  return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
+                second_pred) >> 4; \
+}
+
+#define MAKE_BFP_SAD3_WRAPPER(fnname) \
+static void fnname##_bits8(const uint8_t *src_ptr, \
+                           int source_stride, \
+                           const uint8_t *ref_ptr, \
+                           int  ref_stride, \
+                           unsigned int *sad_array) {  \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+} \
+static void fnname##_bits10(const uint8_t *src_ptr, \
+                            int source_stride, \
+                            const uint8_t *ref_ptr, \
+                            int  ref_stride, \
+                            unsigned int *sad_array) {  \
+  int i; \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+  for (i = 0; i < 3; i++) \
+    sad_array[i] >>= 2; \
+} \
+static void fnname##_bits12(const uint8_t *src_ptr, \
+                            int source_stride, \
+                            const uint8_t *ref_ptr, \
+                            int  ref_stride, \
+                            unsigned int *sad_array) {  \
+  int i; \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+  for (i = 0; i < 3; i++) \
+    sad_array[i] >>= 4; \
+}
+
+#define MAKE_BFP_SAD8_WRAPPER(fnname) \
+static void fnname##_bits8(const uint8_t *src_ptr, \
+                           int source_stride, \
+                           const uint8_t *ref_ptr, \
+                           int  ref_stride, \
+                           unsigned int *sad_array) {  \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+} \
+static void fnname##_bits10(const uint8_t *src_ptr, \
+                            int source_stride, \
+                            const uint8_t *ref_ptr, \
+                            int  ref_stride, \
+                            unsigned int *sad_array) {  \
+  int i; \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+  for (i = 0; i < 8; i++) \
+    sad_array[i] >>= 2; \
+} \
+static void fnname##_bits12(const uint8_t *src_ptr, \
+                            int source_stride, \
+                            const uint8_t *ref_ptr, \
+                            int  ref_stride, \
+                            unsigned int *sad_array) {  \
+  int i; \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+  for (i = 0; i < 8; i++) \
+    sad_array[i] >>= 4; \
+}
+#define MAKE_BFP_SAD4D_WRAPPER(fnname) \
+static void fnname##_bits8(const uint8_t *src_ptr, \
+                           int source_stride, \
+                           const uint8_t* const ref_ptr[], \
+                           int  ref_stride, \
+                           unsigned int *sad_array) {  \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+} \
+static void fnname##_bits10(const uint8_t *src_ptr, \
+                            int source_stride, \
+                            const uint8_t* const ref_ptr[], \
+                            int  ref_stride, \
+                            unsigned int *sad_array) {  \
+  int i; \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+  for (i = 0; i < 4; i++) \
+  sad_array[i] >>= 2; \
+} \
+static void fnname##_bits12(const uint8_t *src_ptr, \
+                            int source_stride, \
+                            const uint8_t* const ref_ptr[], \
+                            int  ref_stride, \
+                            unsigned int *sad_array) {  \
+  int i; \
+  fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
+  for (i = 0; i < 4; i++) \
+  sad_array[i] >>= 4; \
+}
+
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad32x16)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad32x16_avg)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad32x16x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad16x32)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad16x32_avg)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad16x32x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad64x32)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad64x32_avg)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad64x32x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad32x64)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad32x64_avg)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad32x64x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad32x32)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad32x32_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad32x32x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad32x32x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad32x32x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad64x64)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad64x64_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad64x64x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad64x64x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad64x64x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad16x16)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad16x16_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad16x16x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad16x16x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad16x16x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad16x8)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad16x8_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad16x8x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad16x8x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad16x8x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad8x16)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad8x16_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad8x16x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad8x16x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad8x16x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad8x8)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad8x8_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad8x8x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad8x8x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad8x8x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad8x4)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad8x4_avg)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad8x4x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad8x4x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad4x8)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad4x8_avg)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad4x8x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad4x8x4d)
+MAKE_BFP_SAD_WRAPPER(vp9_high_sad4x4)
+MAKE_BFP_SADAVG_WRAPPER(vp9_high_sad4x4_avg)
+MAKE_BFP_SAD3_WRAPPER(vp9_high_sad4x4x3)
+MAKE_BFP_SAD8_WRAPPER(vp9_high_sad4x4x8)
+MAKE_BFP_SAD4D_WRAPPER(vp9_high_sad4x4x4d)
+
+static void highbd_set_var_fns(VP9_COMP *const cpi) {
+  VP9_COMMON *const cm = &cpi->common;
+  if (cm->use_highbitdepth) {
+    switch (cm->bit_depth) {
+      case VPX_BITS_8:
+        HIGHBD_BFP(BLOCK_32X16,
+                   vp9_high_sad32x16_bits8,
+                   vp9_high_sad32x16_avg_bits8,
+                   vp9_high_variance32x16,
+                   vp9_high_sub_pixel_variance32x16,
+                   vp9_high_sub_pixel_avg_variance32x16,
+                   NULL,
+                   NULL,
+                   vp9_high_sad32x16x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_16X32,
+                   vp9_high_sad16x32_bits8,
+                   vp9_high_sad16x32_avg_bits8,
+                   vp9_high_variance16x32,
+                   vp9_high_sub_pixel_variance16x32,
+                   vp9_high_sub_pixel_avg_variance16x32,
+                   NULL,
+                   NULL,
+                   vp9_high_sad16x32x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_64X32,
+                   vp9_high_sad64x32_bits8,
+                   vp9_high_sad64x32_avg_bits8,
+                   vp9_high_variance64x32,
+                   vp9_high_sub_pixel_variance64x32,
+                   vp9_high_sub_pixel_avg_variance64x32,
+                   NULL,
+                   NULL,
+                   vp9_high_sad64x32x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_32X64,
+                   vp9_high_sad32x64_bits8,
+                   vp9_high_sad32x64_avg_bits8,
+                   vp9_high_variance32x64,
+                   vp9_high_sub_pixel_variance32x64,
+                   vp9_high_sub_pixel_avg_variance32x64,
+                   NULL,
+                   NULL,
+                   vp9_high_sad32x64x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_32X32,
+                   vp9_high_sad32x32_bits8,
+                   vp9_high_sad32x32_avg_bits8,
+                   vp9_high_variance32x32,
+                   vp9_high_sub_pixel_variance32x32,
+                   vp9_high_sub_pixel_avg_variance32x32,
+                   vp9_high_sad32x32x3_bits8,
+                   vp9_high_sad32x32x8_bits8,
+                   vp9_high_sad32x32x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_64X64,
+                   vp9_high_sad64x64_bits8,
+                   vp9_high_sad64x64_avg_bits8,
+                   vp9_high_variance64x64,
+                   vp9_high_sub_pixel_variance64x64,
+                   vp9_high_sub_pixel_avg_variance64x64,
+                   vp9_high_sad64x64x3_bits8,
+                   vp9_high_sad64x64x8_bits8,
+                   vp9_high_sad64x64x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_16X16,
+                   vp9_high_sad16x16_bits8,
+                   vp9_high_sad16x16_avg_bits8,
+                   vp9_high_variance16x16,
+                   vp9_high_sub_pixel_variance16x16,
+                   vp9_high_sub_pixel_avg_variance16x16,
+                   vp9_high_sad16x16x3_bits8,
+                   vp9_high_sad16x16x8_bits8,
+                   vp9_high_sad16x16x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_16X8,
+                   vp9_high_sad16x8_bits8,
+                   vp9_high_sad16x8_avg_bits8,
+                   vp9_high_variance16x8,
+                   vp9_high_sub_pixel_variance16x8,
+                   vp9_high_sub_pixel_avg_variance16x8,
+                   vp9_high_sad16x8x3_bits8,
+                   vp9_high_sad16x8x8_bits8,
+                   vp9_high_sad16x8x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_8X16,
+                   vp9_high_sad8x16_bits8,
+                   vp9_high_sad8x16_avg_bits8,
+                   vp9_high_variance8x16,
+                   vp9_high_sub_pixel_variance8x16,
+                   vp9_high_sub_pixel_avg_variance8x16,
+                   vp9_high_sad8x16x3_bits8,
+                   vp9_high_sad8x16x8_bits8,
+                   vp9_high_sad8x16x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_8X8,
+                   vp9_high_sad8x8_bits8,
+                   vp9_high_sad8x8_avg_bits8,
+                   vp9_high_variance8x8,
+                   vp9_high_sub_pixel_variance8x8,
+                   vp9_high_sub_pixel_avg_variance8x8,
+                   vp9_high_sad8x8x3_bits8,
+                   vp9_high_sad8x8x8_bits8,
+                   vp9_high_sad8x8x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_8X4,
+                   vp9_high_sad8x4_bits8,
+                   vp9_high_sad8x4_avg_bits8,
+                   vp9_high_variance8x4,
+                   vp9_high_sub_pixel_variance8x4,
+                   vp9_high_sub_pixel_avg_variance8x4,
+                   NULL,
+                   vp9_high_sad8x4x8_bits8,
+                   vp9_high_sad8x4x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_4X8,
+                   vp9_high_sad4x8_bits8,
+                   vp9_high_sad4x8_avg_bits8,
+                   vp9_high_variance4x8,
+                   vp9_high_sub_pixel_variance4x8,
+                   vp9_high_sub_pixel_avg_variance4x8,
+                   NULL,
+                   vp9_high_sad4x8x8_bits8,
+                   vp9_high_sad4x8x4d_bits8)
+
+        HIGHBD_BFP(BLOCK_4X4,
+                   vp9_high_sad4x4_bits8,
+                   vp9_high_sad4x4_avg_bits8,
+                   vp9_high_variance4x4,
+                   vp9_high_sub_pixel_variance4x4,
+                   vp9_high_sub_pixel_avg_variance4x4,
+                   vp9_high_sad4x4x3_bits8,
+                   vp9_high_sad4x4x8_bits8,
+                   vp9_high_sad4x4x4d_bits8)
+        break;
+
+      case VPX_BITS_10:
+        HIGHBD_BFP(BLOCK_32X16,
+                   vp9_high_sad32x16_bits10,
+                   vp9_high_sad32x16_avg_bits10,
+                   vp9_high_10_variance32x16,
+                   vp9_high_10_sub_pixel_variance32x16,
+                   vp9_high_10_sub_pixel_avg_variance32x16,
+                   NULL,
+                   NULL,
+                   vp9_high_sad32x16x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_16X32,
+                   vp9_high_sad16x32_bits10,
+                   vp9_high_sad16x32_avg_bits10,
+                   vp9_high_10_variance16x32,
+                   vp9_high_10_sub_pixel_variance16x32,
+                   vp9_high_10_sub_pixel_avg_variance16x32,
+                   NULL,
+                   NULL,
+                   vp9_high_sad16x32x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_64X32,
+                   vp9_high_sad64x32_bits10,
+                   vp9_high_sad64x32_avg_bits10,
+                   vp9_high_10_variance64x32,
+                   vp9_high_10_sub_pixel_variance64x32,
+                   vp9_high_10_sub_pixel_avg_variance64x32,
+                   NULL,
+                   NULL,
+                   vp9_high_sad64x32x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_32X64,
+                   vp9_high_sad32x64_bits10,
+                   vp9_high_sad32x64_avg_bits10,
+                   vp9_high_10_variance32x64,
+                   vp9_high_10_sub_pixel_variance32x64,
+                   vp9_high_10_sub_pixel_avg_variance32x64,
+                   NULL,
+                   NULL,
+                   vp9_high_sad32x64x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_32X32,
+                   vp9_high_sad32x32_bits10,
+                   vp9_high_sad32x32_avg_bits10,
+                   vp9_high_10_variance32x32,
+                   vp9_high_10_sub_pixel_variance32x32,
+                   vp9_high_10_sub_pixel_avg_variance32x32,
+                   vp9_high_sad32x32x3_bits10,
+                   vp9_high_sad32x32x8_bits10,
+                   vp9_high_sad32x32x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_64X64,
+                   vp9_high_sad64x64_bits10,
+                   vp9_high_sad64x64_avg_bits10,
+                   vp9_high_10_variance64x64,
+                   vp9_high_10_sub_pixel_variance64x64,
+                   vp9_high_10_sub_pixel_avg_variance64x64,
+                   vp9_high_sad64x64x3_bits10,
+                   vp9_high_sad64x64x8_bits10,
+                   vp9_high_sad64x64x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_16X16,
+                   vp9_high_sad16x16_bits10,
+                   vp9_high_sad16x16_avg_bits10,
+                   vp9_high_10_variance16x16,
+                   vp9_high_10_sub_pixel_variance16x16,
+                   vp9_high_10_sub_pixel_avg_variance16x16,
+                   vp9_high_sad16x16x3_bits10,
+                   vp9_high_sad16x16x8_bits10,
+                   vp9_high_sad16x16x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_16X8,
+                   vp9_high_sad16x8_bits10,
+                   vp9_high_sad16x8_avg_bits10,
+                   vp9_high_10_variance16x8,
+                   vp9_high_10_sub_pixel_variance16x8,
+                   vp9_high_10_sub_pixel_avg_variance16x8,
+                   vp9_high_sad16x8x3_bits10,
+                   vp9_high_sad16x8x8_bits10,
+                   vp9_high_sad16x8x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_8X16,
+                   vp9_high_sad8x16_bits10,
+                   vp9_high_sad8x16_avg_bits10,
+                   vp9_high_10_variance8x16,
+                   vp9_high_10_sub_pixel_variance8x16,
+                   vp9_high_10_sub_pixel_avg_variance8x16,
+                   vp9_high_sad8x16x3_bits10,
+                   vp9_high_sad8x16x8_bits10,
+                   vp9_high_sad8x16x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_8X8,
+                   vp9_high_sad8x8_bits10,
+                   vp9_high_sad8x8_avg_bits10,
+                   vp9_high_10_variance8x8,
+                   vp9_high_10_sub_pixel_variance8x8,
+                   vp9_high_10_sub_pixel_avg_variance8x8,
+                   vp9_high_sad8x8x3_bits10,
+                   vp9_high_sad8x8x8_bits10,
+                   vp9_high_sad8x8x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_8X4,
+                   vp9_high_sad8x4_bits10,
+                   vp9_high_sad8x4_avg_bits10,
+                   vp9_high_10_variance8x4,
+                   vp9_high_10_sub_pixel_variance8x4,
+                   vp9_high_10_sub_pixel_avg_variance8x4,
+                   NULL,
+                   vp9_high_sad8x4x8_bits10,
+                   vp9_high_sad8x4x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_4X8,
+                   vp9_high_sad4x8_bits10,
+                   vp9_high_sad4x8_avg_bits10,
+                   vp9_high_10_variance4x8,
+                   vp9_high_10_sub_pixel_variance4x8,
+                   vp9_high_10_sub_pixel_avg_variance4x8,
+                   NULL,
+                   vp9_high_sad4x8x8_bits10,
+                   vp9_high_sad4x8x4d_bits10)
+
+        HIGHBD_BFP(BLOCK_4X4,
+                   vp9_high_sad4x4_bits10,
+                   vp9_high_sad4x4_avg_bits10,
+                   vp9_high_10_variance4x4,
+                   vp9_high_10_sub_pixel_variance4x4,
+                   vp9_high_10_sub_pixel_avg_variance4x4,
+                   vp9_high_sad4x4x3_bits10,
+                   vp9_high_sad4x4x8_bits10,
+                   vp9_high_sad4x4x4d_bits10)
+        break;
+
+      case VPX_BITS_12:
+        HIGHBD_BFP(BLOCK_32X16,
+                   vp9_high_sad32x16_bits12,
+                   vp9_high_sad32x16_avg_bits12,
+                   vp9_high_12_variance32x16,
+                   vp9_high_12_sub_pixel_variance32x16,
+                   vp9_high_12_sub_pixel_avg_variance32x16,
+                   NULL,
+                   NULL,
+                   vp9_high_sad32x16x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_16X32,
+                   vp9_high_sad16x32_bits12,
+                   vp9_high_sad16x32_avg_bits12,
+                   vp9_high_12_variance16x32,
+                   vp9_high_12_sub_pixel_variance16x32,
+                   vp9_high_12_sub_pixel_avg_variance16x32,
+                   NULL,
+                   NULL,
+                   vp9_high_sad16x32x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_64X32,
+                   vp9_high_sad64x32_bits12,
+                   vp9_high_sad64x32_avg_bits12,
+                   vp9_high_12_variance64x32,
+                   vp9_high_12_sub_pixel_variance64x32,
+                   vp9_high_12_sub_pixel_avg_variance64x32,
+                   NULL,
+                   NULL,
+                   vp9_high_sad64x32x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_32X64,
+                   vp9_high_sad32x64_bits12,
+                   vp9_high_sad32x64_avg_bits12,
+                   vp9_high_12_variance32x64,
+                   vp9_high_12_sub_pixel_variance32x64,
+                   vp9_high_12_sub_pixel_avg_variance32x64,
+                   NULL,
+                   NULL,
+                   vp9_high_sad32x64x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_32X32,
+                   vp9_high_sad32x32_bits12,
+                   vp9_high_sad32x32_avg_bits12,
+                   vp9_high_12_variance32x32,
+                   vp9_high_12_sub_pixel_variance32x32,
+                   vp9_high_12_sub_pixel_avg_variance32x32,
+                   vp9_high_sad32x32x3_bits12,
+                   vp9_high_sad32x32x8_bits12,
+                   vp9_high_sad32x32x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_64X64,
+                   vp9_high_sad64x64_bits12,
+                   vp9_high_sad64x64_avg_bits12,
+                   vp9_high_12_variance64x64,
+                   vp9_high_12_sub_pixel_variance64x64,
+                   vp9_high_12_sub_pixel_avg_variance64x64,
+                   vp9_high_sad64x64x3_bits12,
+                   vp9_high_sad64x64x8_bits12,
+                   vp9_high_sad64x64x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_16X16,
+                   vp9_high_sad16x16_bits12,
+                   vp9_high_sad16x16_avg_bits12,
+                   vp9_high_12_variance16x16,
+                   vp9_high_12_sub_pixel_variance16x16,
+                   vp9_high_12_sub_pixel_avg_variance16x16,
+                   vp9_high_sad16x16x3_bits12,
+                   vp9_high_sad16x16x8_bits12,
+                   vp9_high_sad16x16x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_16X8,
+                   vp9_high_sad16x8_bits12,
+                   vp9_high_sad16x8_avg_bits12,
+                   vp9_high_12_variance16x8,
+                   vp9_high_12_sub_pixel_variance16x8,
+                   vp9_high_12_sub_pixel_avg_variance16x8,
+                   vp9_high_sad16x8x3_bits12,
+                   vp9_high_sad16x8x8_bits12,
+                   vp9_high_sad16x8x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_8X16,
+                   vp9_high_sad8x16_bits12,
+                   vp9_high_sad8x16_avg_bits12,
+                   vp9_high_12_variance8x16,
+                   vp9_high_12_sub_pixel_variance8x16,
+                   vp9_high_12_sub_pixel_avg_variance8x16,
+                   vp9_high_sad8x16x3_bits12,
+                   vp9_high_sad8x16x8_bits12,
+                   vp9_high_sad8x16x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_8X8,
+                   vp9_high_sad8x8_bits12,
+                   vp9_high_sad8x8_avg_bits12,
+                   vp9_high_12_variance8x8,
+                   vp9_high_12_sub_pixel_variance8x8,
+                   vp9_high_12_sub_pixel_avg_variance8x8,
+                   vp9_high_sad8x8x3_bits12,
+                   vp9_high_sad8x8x8_bits12,
+                   vp9_high_sad8x8x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_8X4,
+                   vp9_high_sad8x4_bits12,
+                   vp9_high_sad8x4_avg_bits12,
+                   vp9_high_12_variance8x4,
+                   vp9_high_12_sub_pixel_variance8x4,
+                   vp9_high_12_sub_pixel_avg_variance8x4,
+                   NULL,
+                   vp9_high_sad8x4x8_bits12,
+                   vp9_high_sad8x4x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_4X8,
+                   vp9_high_sad4x8_bits12,
+                   vp9_high_sad4x8_avg_bits12,
+                   vp9_high_12_variance4x8,
+                   vp9_high_12_sub_pixel_variance4x8,
+                   vp9_high_12_sub_pixel_avg_variance4x8,
+                   NULL,
+                   vp9_high_sad4x8x8_bits12,
+                   vp9_high_sad4x8x4d_bits12)
+
+        HIGHBD_BFP(BLOCK_4X4,
+                   vp9_high_sad4x4_bits12,
+                   vp9_high_sad4x4_avg_bits12,
+                   vp9_high_12_variance4x4,
+                   vp9_high_12_sub_pixel_variance4x4,
+                   vp9_high_12_sub_pixel_avg_variance4x4,
+                   vp9_high_sad4x4x3_bits12,
+                   vp9_high_sad4x4x8_bits12,
+                   vp9_high_sad4x4x4d_bits12)
+        break;
+
+      default:
+        assert(0 && "cm->bit_depth should be VPX_BITS_8, "
+                    "VPX_BITS_10 or VPX_BITS_12");
+    }
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 void vp9_change_config(struct VP9_COMP *cpi, const VP9EncoderConfig *oxcf) {
   VP9_COMMON *const cm = &cpi->common;
   RATE_CONTROL *const rc = &cpi->rc;
@@ -613,6 +1224,11 @@
     assert(cm->bit_depth > VPX_BITS_8);
 
   cpi->oxcf = *oxcf;
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (cpi->oxcf.use_highbitdepth) {
+    cpi->mb.e_mbd.bd = (int)cm->bit_depth;
+  }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
   rc->baseline_gf_interval = DEFAULT_GF_INTERVAL;
 
@@ -683,6 +1299,10 @@
   cpi->ext_refresh_frame_flags_pending = 0;
   cpi->ext_refresh_frame_context_pending = 0;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  highbd_set_var_fns(cpi);
+#endif
+
 #if CONFIG_VP9_TEMPORAL_DENOISING
   if (cpi->oxcf.noise_sensitivity > 0) {
     vp9_denoiser_alloc(&(cpi->denoiser), cm->width, cm->height,
@@ -981,8 +1601,10 @@
 
   // Default rd threshold factors for mode selection
   for (i = 0; i < BLOCK_SIZES; ++i) {
-    for (j = 0; j < MAX_MODES; ++j)
+    for (j = 0; j < MAX_MODES; ++j) {
       cpi->rd.thresh_freq_fact[i][j] = 32;
+      cpi->rd.mode_map[i][j] = j;
+    }
   }
 
 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF)\
@@ -1181,6 +1803,7 @@
 
 #endif
 }
+
 static int64_t get_sse(const uint8_t *a, int a_stride,
                        const uint8_t *b, int b_stride,
                        int width, int height) {
@@ -1222,6 +1845,63 @@
   return total_sse;
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
+                                    const uint8_t *b8, int b_stride,
+                                    int width, int height,
+                                    unsigned int input_shift) {
+  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+  int64_t total_sse = 0;
+  int x, y;
+  for (y = 0; y < height; ++y) {
+    for (x = 0; x < width; ++x) {
+      int64_t diff;
+      diff = (a[x] >> input_shift) - (b[x] >> input_shift);
+      total_sse += diff * diff;
+    }
+    a += a_stride;
+    b += b_stride;
+  }
+  return total_sse;
+}
+
+static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
+                              const uint8_t *b, int b_stride,
+                              int width, int height) {
+  int64_t total_sse = 0;
+  int x, y;
+  const int dw = width % 16;
+  const int dh = height % 16;
+  unsigned int sse = 0;
+  int sum = 0;
+  if (dw > 0) {
+    high_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
+                  dw, height, &sse, &sum);
+    total_sse += sse;
+  }
+  if (dh > 0) {
+    high_variance(&a[(height - dh) * a_stride], a_stride,
+                  &b[(height - dh) * b_stride], b_stride,
+                  width - dw, dh, &sse, &sum);
+    total_sse += sse;
+  }
+  for (y = 0; y < height / 16; ++y) {
+    const uint8_t *pa = a;
+    const uint8_t *pb = b;
+    for (x = 0; x < width / 16; ++x) {
+      vp9_high_mse16x16(pa, a_stride, pb, b_stride, &sse);
+      total_sse += sse;
+      pa += 16;
+      pb += 16;
+    }
+    a += 16 * a_stride;
+    b += 16 * b_stride;
+  }
+  return total_sse;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 typedef struct {
   double psnr[4];       // total/y/u/v
   uint64_t sse[4];      // total/y/u/v
@@ -1261,18 +1941,79 @@
                                   (double)total_sse);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
+                             const YV12_BUFFER_CONFIG *b,
+                             PSNR_STATS *psnr,
+                             unsigned int bit_depth,
+                             unsigned int in_bit_depth) {
+  const int widths[3] = {a->y_width,  a->uv_width,  a->uv_width };
+  const int heights[3] = {a->y_height, a->uv_height, a->uv_height};
+  const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer,  a->v_buffer };
+  const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
+  const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer,  b->v_buffer };
+  const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
+  int i;
+  uint64_t total_sse = 0;
+  uint32_t total_samples = 0;
+  const double peak = (double)((1 << in_bit_depth) - 1);
+  const unsigned int input_shift = bit_depth - in_bit_depth;
+
+  for (i = 0; i < 3; ++i) {
+    const int w = widths[i];
+    const int h = heights[i];
+    const uint32_t samples = w * h;
+    uint64_t sse;
+    if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
+      if (input_shift) {
+        sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
+                                   b_planes[i], b_strides[i], w, h,
+                                   input_shift);
+      } else {
+        sse = highbd_get_sse(a_planes[i], a_strides[i],
+                             b_planes[i], b_strides[i], w, h);
+      }
+    } else {
+      sse = get_sse(a_planes[i], a_strides[i],
+                    b_planes[i], b_strides[i],
+                    w, h);
+    }
+    psnr->sse[1 + i] = sse;
+    psnr->samples[1 + i] = samples;
+    psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
+
+    total_sse += sse;
+    total_samples += samples;
+  }
+
+  psnr->sse[0] = total_sse;
+  psnr->samples[0] = total_samples;
+  psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
+                                  (double)total_sse);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void generate_psnr_packet(VP9_COMP *cpi) {
   struct vpx_codec_cx_pkt pkt;
   int i;
   PSNR_STATS psnr;
+#if CONFIG_VP9_HIGHBITDEPTH
+  calc_highbd_psnr(cpi->Source, cpi->common.frame_to_show, &psnr,
+                   cpi->mb.e_mbd.bd, cpi->oxcf.input_bit_depth);
+#else
   calc_psnr(cpi->Source, cpi->common.frame_to_show, &psnr);
+#endif
+
   for (i = 0; i < 4; ++i) {
     pkt.data.psnr.samples[i] = psnr.samples[i];
     pkt.data.psnr.sse[i] = psnr.sse[i];
     pkt.data.psnr.psnr[i] = psnr.psnr[i];
   }
   pkt.kind = VPX_CODEC_PSNR_PKT;
-  vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
+  if (is_two_pass_svc(cpi))
+    cpi->svc.layer_context[cpi->svc.spatial_layer_id].psnr_pkt = pkt.data.psnr;
+  else
+    vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
 }
 
 int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags) {
@@ -1371,6 +2112,36 @@
   uint8_t *src = s->y_buffer;
   int h = cm->height;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (s->flags & YV12_FLAG_HIGHBITDEPTH) {
+    uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer);
+
+    do {
+      fwrite(src16, s->y_width, 2,  yuv_rec_file);
+      src16 += s->y_stride;
+    } while (--h);
+
+    src16 = CONVERT_TO_SHORTPTR(s->u_buffer);
+    h = s->uv_height;
+
+    do {
+      fwrite(src16, s->uv_width, 2,  yuv_rec_file);
+      src16 += s->uv_stride;
+    } while (--h);
+
+    src16 = CONVERT_TO_SHORTPTR(s->v_buffer);
+    h = s->uv_height;
+
+    do {
+      fwrite(src16, s->uv_width, 2, yuv_rec_file);
+      src16 += s->uv_stride;
+    } while (--h);
+
+    fflush(yuv_rec_file);
+    return;
+  }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
   do {
     fwrite(src, s->y_width, 1,  yuv_rec_file);
     src += s->y_stride;
@@ -1396,8 +2167,14 @@
 }
 #endif
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
+                                                YV12_BUFFER_CONFIG *dst,
+                                                int bd) {
+#else
 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
                                                 YV12_BUFFER_CONFIG *dst) {
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   // TODO(dkovalev): replace YV12_BUFFER_CONFIG with vpx_image_t
   int i;
   const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
@@ -1413,15 +2190,31 @@
   const int dst_heights[3] = {dst->y_crop_height, dst->uv_crop_height,
                               dst->uv_crop_height};
 
-  for (i = 0; i < MAX_MB_PLANE; ++i)
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+      vp9_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
+                              src_strides[i], dsts[i], dst_heights[i],
+                              dst_widths[i], dst_strides[i], bd);
+    } else {
+      vp9_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
+                       dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
+    }
+#else
     vp9_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
                      dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
-
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  }
   vp9_extend_frame_borders(dst);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
+                                   YV12_BUFFER_CONFIG *dst, int bd) {
+#else
 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
                                    YV12_BUFFER_CONFIG *dst) {
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   const int src_w = src->y_crop_width;
   const int src_h = src->y_crop_height;
   const int dst_w = dst->y_crop_width;
@@ -1445,10 +2238,24 @@
                                      src_stride + (x / factor) * src_w / dst_w;
         uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+          vp9_high_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+                             kernel[x_q4 & 0xf], 16 * src_w / dst_w,
+                             kernel[y_q4 & 0xf], 16 * src_h / dst_h,
+                             16 / factor, 16 / factor, bd);
+        } else {
+          vp9_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+                        kernel[x_q4 & 0xf], 16 * src_w / dst_w,
+                        kernel[y_q4 & 0xf], 16 * src_h / dst_h,
+                        16 / factor, 16 / factor);
+        }
+#else
         vp9_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
                       kernel[x_q4 & 0xf], 16 * src_w / dst_w,
                       kernel[y_q4 & 0xf], 16 * src_h / dst_h,
                       16 / factor, 16 / factor);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       }
     }
   }
@@ -1617,9 +2424,14 @@
                                cm->subsampling_x, cm->subsampling_y,
 #if CONFIG_VP9_HIGHBITDEPTH
                                cm->use_highbitdepth,
-#endif
+#endif  // CONFIG_VP9_HIGHBITDEPTH
                                VP9_ENC_BORDER_IN_PIXELS, NULL, NULL, NULL);
+#if CONFIG_VP9_HIGHBITDEPTH
+      scale_and_extend_frame(ref, &cm->frame_bufs[new_fb].buf,
+                             (int)cm->bit_depth);
+#else
       scale_and_extend_frame(ref, &cm->frame_bufs[new_fb].buf);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
     } else {
       cpi->scaled_ref_idx[ref_frame - 1] = idx;
@@ -1809,11 +2621,22 @@
            rc->this_key_frame_forced &&
            (rc->projected_frame_size < rc->max_frame_bandwidth)) {
         int last_q = q;
-        int kf_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+        int kf_err;
 
         int high_err_target = cpi->ambient_err;
         int low_err_target = cpi->ambient_err >> 1;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (cm->use_highbitdepth) {
+          kf_err = vp9_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm),
+                                        cm->bit_depth);
+        } else {
+          kf_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+        }
+#else
+        kf_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
         // Prevent possible divide by zero error below for perfect KF
         kf_err += !kf_err;
 
@@ -1984,7 +2807,11 @@
                                           YV12_BUFFER_CONFIG *scaled) {
   if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
       cm->mi_rows * MI_SIZE != unscaled->y_height) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
+#else
     scale_and_extend_frame_nonnormative(unscaled, scaled);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
     return scaled;
   } else {
     return unscaled;
@@ -2233,9 +3060,11 @@
 #endif
 
 #if CONFIG_INTERNAL_STATS
-  int i;
-  for (i = 0; i < MAX_MODES; ++i)
-    cpi->mode_chosen_counts[i] = 0;
+  {
+    int i;
+    for (i = 0; i < MAX_MODES; ++i)
+      cpi->mode_chosen_counts[i] = 0;
+  }
 #endif
 
   vp9_set_speed_features(cpi);
@@ -2272,7 +3101,17 @@
   // fixed interval. Note the reconstruction error if it is the frame before
   // the force key frame
   if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (cm->use_highbitdepth) {
+      cpi->ambient_err = vp9_highbd_get_y_sse(cpi->Source,
+                                              get_frame_new_buffer(cm),
+                                              cm->bit_depth);
+    } else {
+      cpi->ambient_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+    }
+#else
     cpi->ambient_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 
   // If the encoder forced a KEY_FRAME decision
@@ -2384,10 +3223,7 @@
 static void Pass2Encode(VP9_COMP *cpi, size_t *size,
                         uint8_t *dest, unsigned int *frame_flags) {
   cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
-
-  vp9_rc_get_second_pass_params(cpi);
   encode_frame_to_data_rate(cpi, size, dest, frame_flags);
-
   vp9_twopass_postencode_update(cpi);
 }
 
@@ -2401,13 +3237,19 @@
   }
 }
 
-static void check_initial_width(VP9_COMP *cpi, int subsampling_x,
-                                int subsampling_y) {
+static void check_initial_width(VP9_COMP *cpi,
+#if CONFIG_VP9_HIGHBITDEPTH
+                                int use_highbitdepth,
+#endif
+                                int subsampling_x, int subsampling_y) {
   VP9_COMMON *const cm = &cpi->common;
 
   if (!cpi->initial_width) {
     cm->subsampling_x = subsampling_x;
     cm->subsampling_y = subsampling_y;
+#if CONFIG_VP9_HIGHBITDEPTH
+    cm->use_highbitdepth = use_highbitdepth;
+#endif
 
     alloc_raw_frame_buffers(cpi);
     alloc_ref_frame_buffers(cpi);
@@ -2429,20 +3271,16 @@
   int res = 0;
   const int subsampling_x = sd->uv_width  < sd->y_width;
   const int subsampling_y = sd->uv_height < sd->y_height;
-
+#if CONFIG_VP9_HIGHBITDEPTH
+  const int use_highbitdepth = sd->flags & YV12_FLAG_HIGHBITDEPTH;
+  check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y);
+#else
   check_initial_width(cpi, subsampling_x, subsampling_y);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
   vpx_usec_timer_start(&timer);
 
-#if CONFIG_SPATIAL_SVC
-  if (is_two_pass_svc(cpi))
-    res = vp9_svc_lookahead_push(cpi, cpi->lookahead, sd, time_stamp, end_time,
-                                 frame_flags);
-  else
-#endif
-    res = vp9_lookahead_push(cpi->lookahead,
-                             sd, time_stamp, end_time, frame_flags);
-  if (res)
+  if (vp9_lookahead_push(cpi->lookahead, sd, time_stamp, end_time, frame_flags))
     res = -1;
   vpx_usec_timer_mark(&timer);
   cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
@@ -2571,11 +3409,12 @@
   MV_REFERENCE_FRAME ref_frame;
   int arf_src_index;
 
-  if (is_two_pass_svc(cpi) && oxcf->pass == 2) {
+  if (is_two_pass_svc(cpi)) {
 #if CONFIG_SPATIAL_SVC
-    vp9_svc_lookahead_peek(cpi, cpi->lookahead, 0, 1);
+    vp9_svc_start_frame(cpi);
 #endif
-    vp9_restore_layer_context(cpi);
+    if (oxcf->pass == 2)
+      vp9_restore_layer_context(cpi);
   }
 
   vpx_usec_timer_start(&cmptimer);
@@ -2594,13 +3433,7 @@
   if (arf_src_index) {
     assert(arf_src_index <= rc->frames_to_key);
 
-#if CONFIG_SPATIAL_SVC
-    if (is_two_pass_svc(cpi))
-      source = vp9_svc_lookahead_peek(cpi, cpi->lookahead, arf_src_index, 0);
-    else
-#endif
-      source = vp9_lookahead_peek(cpi->lookahead, arf_src_index);
-    if (source != NULL) {
+    if ((source = vp9_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
       cpi->alt_ref_source = source;
 
 #if CONFIG_SPATIAL_SVC
@@ -2638,13 +3471,7 @@
   if (!source) {
     // Get last frame source.
     if (cm->current_video_frame > 0) {
-#if CONFIG_SPATIAL_SVC
-      if (is_two_pass_svc(cpi))
-        last_source = vp9_svc_lookahead_peek(cpi, cpi->lookahead, -1, 0);
-      else
-#endif
-        last_source = vp9_lookahead_peek(cpi->lookahead, -1);
-      if (last_source == NULL)
+      if ((last_source = vp9_lookahead_peek(cpi->lookahead, -1)) == NULL)
         return -1;
     }
 
@@ -2711,6 +3538,12 @@
   cm->frame_bufs[cm->new_fb_idx].ref_count--;
   cm->new_fb_idx = get_free_fb(cm);
 
+  // For two pass encodes analyse the first pass stats and determine
+  // the bit allocation and other parameters for this frame / group of frames.
+  if ((oxcf->pass == 2) && (!cpi->use_svc || is_two_pass_svc(cpi))) {
+    vp9_rc_get_second_pass_params(cpi);
+  }
+
   if (!cpi->use_svc && cpi->multi_arf_allowed) {
     if (cm->frame_type == KEY_FRAME) {
       init_buffer_indices(cpi);
@@ -2749,10 +3582,17 @@
     RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - 1];
     ref_buf->buf = buf;
     ref_buf->idx = idx;
+#if CONFIG_VP9_HIGHBITDEPTH
+    vp9_setup_scale_factors_for_frame(&ref_buf->sf,
+                                      buf->y_crop_width, buf->y_crop_height,
+                                      cm->width, cm->height,
+                                      (buf->flags & YV12_FLAG_HIGHBITDEPTH) ?
+                                          1 : 0);
+#else
     vp9_setup_scale_factors_for_frame(&ref_buf->sf,
                                       buf->y_crop_width, buf->y_crop_height,
                                       cm->width, cm->height);
-
+#endif  // CONFIG_VP9_HIGHBITDEPTH
     if (vp9_is_scaled(&ref_buf->sf))
       vp9_extend_frame_borders(buf);
   }
@@ -2766,7 +3606,16 @@
   if (oxcf->pass == 1 &&
       (!cpi->use_svc || is_two_pass_svc(cpi))) {
     const int lossless = is_lossless_requested(oxcf);
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (cpi->oxcf.use_highbitdepth)
+      cpi->mb.fwd_txm4x4 = lossless ? vp9_high_fwht4x4 : vp9_high_fdct4x4;
+    else
+      cpi->mb.fwd_txm4x4 = lossless ? vp9_fwht4x4 : vp9_fdct4x4;
+    cpi->mb.high_itxm_add = lossless ? vp9_high_iwht4x4_add :
+                                       vp9_high_idct4x4_add;
+#else
     cpi->mb.fwd_txm4x4 = lossless ? vp9_fwht4x4 : vp9_fdct4x4;
+#endif  // CONFIG_VP9_HIGHBITDEPTH
     cpi->mb.itxm_add = lossless ? vp9_iwht4x4_add : vp9_idct4x4_add;
     vp9_first_pass(cpi, source);
   } else if (oxcf->pass == 2 &&
@@ -2819,7 +3668,12 @@
         YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
         YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
         PSNR_STATS psnr;
+#if CONFIG_VP9_HIGHBITDEPTH
+        calc_highbd_psnr(orig, recon, &psnr, cpi->mb.e_mbd.bd,
+                         cpi->oxcf.input_bit_depth);
+#else
         calc_psnr(orig, recon, &psnr);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
         cpi->total += psnr.psnr[0];
         cpi->total_y += psnr.psnr[1];
@@ -2839,7 +3693,12 @@
 #endif
           vp9_clear_system_state();
 
+#if CONFIG_VP9_HIGHBITDEPTH
+          calc_highbd_psnr(orig, recon, &psnr, cpi->mb.e_mbd.bd,
+                           cpi->oxcf.input_bit_depth);
+#else
           calc_psnr(orig, pp, &psnr2);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
           cpi->totalp += psnr2.psnr[0];
           cpi->totalp_y += psnr2.psnr[1];
@@ -2848,12 +3707,32 @@
           cpi->totalp_sq_error += psnr2.sse[0];
           cpi->totalp_samples += psnr2.samples[0];
 
+#if CONFIG_VP9_HIGHBITDEPTH
+          if (cm->use_highbitdepth) {
+            frame_ssim2 = vp9_highbd_calc_ssim(
+                orig, recon, &weight, xd->bd,
+                xd->bd - cpi->oxcf.input_bit_depth);
+          } else {
+            frame_ssim2 = vp9_calc_ssim(orig, recon, 1, &weight);
+          }
+#else
           frame_ssim2 = vp9_calc_ssim(orig, recon, &weight);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
           cpi->summed_quality += frame_ssim2 * weight;
           cpi->summed_weights += weight;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+          if (cm->use_highbitdepth) {
+            frame_ssim2 = vp9_highbd_calc_ssim(
+                orig, &cm->post_proc_buffer, &weight,
+                xd->bd, xd->bd - cpi->oxcf.input_bit_depth);
+          } else {
+            frame_ssim2 = vp9_calc_ssim(orig, &cm->post_proc_buffer, &weight);
+          }
+#else
           frame_ssim2 = vp9_calc_ssim(orig, &cm->post_proc_buffer, &weight);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
           cpi->summedp_quality += frame_ssim2 * weight;
           cpi->summedp_weights += weight;
@@ -2872,7 +3751,18 @@
 
       if (cpi->b_calculate_ssimg) {
         double y, u, v, frame_all;
+#if CONFIG_VP9_HIGHBITDEPTH
+        if (cm->use_high) {
+          frame_all = vp9_highbd_calc_ssimg(cpi->Source, cm->frame_to_show, &y,
+                                            &u, &v, xd->bd,
+                                            xd->bd - cpi->oxcf.input_bit_depth);
+        } else {
+          frame_all = vp9_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u,
+                                     &v);
+        }
+#else
         frame_all = vp9_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u, &v);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
         cpi->total_ssimg_y += y;
         cpi->total_ssimg_u += u;
         cpi->total_ssimg_v += v;
@@ -2882,6 +3772,12 @@
   }
 
 #endif
+
+  if (is_two_pass_svc(cpi) && cm->show_frame) {
+    ++cpi->svc.spatial_layer_to_encode;
+    if (cpi->svc.spatial_layer_to_encode >= cpi->svc.number_spatial_layers)
+      cpi->svc.spatial_layer_to_encode = 0;
+  }
   return 0;
 }
 
@@ -2963,8 +3859,11 @@
 int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
                          unsigned int height) {
   VP9_COMMON *cm = &cpi->common;
-
+#if CONFIG_VP9_HIGHBITDEPTH
+  check_initial_width(cpi, 1, 1, cm->use_highbitdepth);
+#else
   check_initial_width(cpi, 1, 1);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
   if (width) {
     cm->width = width;
@@ -3010,6 +3909,35 @@
                       a->y_crop_width, a->y_crop_height);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+int vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
+                         const YV12_BUFFER_CONFIG *b,
+                         vpx_bit_depth_t bit_depth) {
+  unsigned int sse;
+  int sum;
+  assert(a->y_crop_width == b->y_crop_width);
+  assert(a->y_crop_height == b->y_crop_height);
+  assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
+  assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      high_variance(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
+                    a->y_crop_width, a->y_crop_height, &sse, &sum);
+      return (int) sse;
+    case VPX_BITS_10:
+      high_10_variance(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
+                       a->y_crop_width, a->y_crop_height, &sse, &sum);
+      return (int) sse;
+    case VPX_BITS_12:
+      high_12_variance(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
+                       a->y_crop_width, a->y_crop_height, &sse, &sum);
+      return (int) sse;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 int vp9_get_quantizer(VP9_COMP *cpi) {
   return cpi->common.base_qindex;
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 0d3c4c1..9bd16bc 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -217,6 +217,9 @@
 
   vp8e_tuning tuning;
   vp9e_tune_content content;
+#if CONFIG_VP9_HIGHBITDEPTH
+  int use_highbitdepth;
+#endif
 } VP9EncoderConfig;
 
 static INLINE int is_lossless_requested(const VP9EncoderConfig *cfg) {
@@ -480,6 +483,11 @@
 }
 
 int vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
+#if CONFIG_VP9_HIGHBITDEPTH
+int vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
+                         const YV12_BUFFER_CONFIG *b,
+                         vpx_bit_depth_t bit_depth);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 void vp9_alloc_compressor_data(VP9_COMP *cpi);
 
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 8041b59..0282e9f 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -35,26 +35,28 @@
 #include "vp9/encoder/vp9_rd.h"
 #include "vp9/encoder/vp9_variance.h"
 
-#define OUTPUT_FPF 0
+#define OUTPUT_FPF          0
+#define ARF_STATS_OUTPUT    0
 
-#define IIFACTOR   12.5
-#define IIKFACTOR1 12.5
-#define IIKFACTOR2 15.0
-#define RMAX       512.0
-#define GF_RMAX    96.0
-#define ERR_DIVISOR   150.0
-#define MIN_DECAY_FACTOR 0.1
-#define SVC_FACTOR_PT_LOW 0.45
-#define FACTOR_PT_LOW 0.5
-#define FACTOR_PT_HIGH 0.9
-
-#define KF_MB_INTRA_MIN 150
-#define GF_MB_INTRA_MIN 100
+#define BOOST_FACTOR        12.5
+#define ERR_DIVISOR         100.0
+#define FACTOR_PT_LOW       0.5
+#define FACTOR_PT_HIGH      0.9
+#define FIRST_PASS_Q        10.0
+#define GF_MAX_BOOST        96.0
+#define INTRA_MODE_PENALTY  1024
+#define KF_MAX_BOOST        128.0
+#define MIN_DECAY_FACTOR    0.01
+#define MIN_GF_INTERVAL     4
+#define MIN_KF_BOOST        300
+#define NEW_MV_MODE_PENALTY 32
+#define SVC_FACTOR_PT_LOW   0.45
 
 #define DOUBLE_DIVIDE_CHECK(x) ((x) < 0 ? (x) - 0.000001 : (x) + 0.000001)
 
-#define MIN_KF_BOOST        300
-#define MIN_GF_INTERVAL     4
+#if ARF_STATS_OUTPUT
+unsigned int arf_count = 0;
+#endif
 
 static void swap_yv12(YV12_BUFFER_CONFIG *a, YV12_BUFFER_CONFIG *b) {
   YV12_BUFFER_CONFIG temp = *a;
@@ -62,8 +64,8 @@
   *b = temp;
 }
 
-static int gfboost_qadjust(int qindex) {
-  const double q = vp9_convert_qindex_to_q(qindex);
+static int gfboost_qadjust(int qindex, vpx_bit_depth_t bit_depth) {
+  const double q = vp9_convert_qindex_to_q(qindex, bit_depth);
   return (int)((0.00000828 * q * q * q) +
                (-0.0055 * q * q) +
                (1.32 * q) + 79.3);
@@ -297,9 +299,9 @@
   MV tmp_mv = {0, 0};
   MV ref_mv_full = {ref_mv->row >> 3, ref_mv->col >> 3};
   int num00, tmp_err, n;
-  const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
+  const BLOCK_SIZE bsize = xd->mi[0].src_mi->mbmi.sb_type;
   vp9_variance_fn_ptr_t v_fn_ptr = cpi->fn_ptr[bsize];
-  const int new_mv_mode_penalty = 256;
+  const int new_mv_mode_penalty = NEW_MV_MODE_PENALTY;
 
   int step_param = 3;
   int further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
@@ -360,11 +362,11 @@
   }
 }
 
-static int find_fp_qindex() {
+static int find_fp_qindex(vpx_bit_depth_t bit_depth) {
   int i;
 
   for (i = 0; i < QINDEX_RANGE; ++i)
-    if (vp9_convert_qindex_to_q(i) >= 30.0)
+    if (vp9_convert_qindex_to_q(i, bit_depth) >= FIRST_PASS_Q)
       break;
 
   if (i == QINDEX_RANGE)
@@ -414,7 +416,7 @@
   int mvcount = 0;
   int intercount = 0;
   int second_ref_count = 0;
-  int intrapenalty = 256;
+  const int intrapenalty = INTRA_MODE_PENALTY;
   int neutral_count = 0;
   int new_mv_count = 0;
   int sum_in_vectors = 0;
@@ -434,44 +436,54 @@
   vp9_clear_system_state();
 
   set_first_pass_params(cpi);
-  vp9_set_quantizer(cm, find_fp_qindex());
+  vp9_set_quantizer(cm, find_fp_qindex(cm->bit_depth));
 
   if (lc != NULL) {
-    MV_REFERENCE_FRAME ref_frame = LAST_FRAME;
     twopass = &lc->twopass;
 
-    if (cpi->common.current_video_frame == 0) {
-      cpi->ref_frame_flags = 0;
+    cpi->lst_fb_idx = cpi->svc.spatial_layer_id;
+    cpi->ref_frame_flags = VP9_LAST_FLAG;
+
+    if (cpi->svc.number_spatial_layers + cpi->svc.spatial_layer_id <
+        REF_FRAMES) {
+      cpi->gld_fb_idx =
+          cpi->svc.number_spatial_layers + cpi->svc.spatial_layer_id;
+      cpi->ref_frame_flags |= VP9_GOLD_FLAG;
+      cpi->refresh_golden_frame = (lc->current_video_frame_in_layer == 0);
     } else {
-    if (lc->current_video_frame_in_layer <
-        (unsigned int)cpi->svc.number_temporal_layers)
-        cpi->ref_frame_flags = VP9_GOLD_FLAG;
-      else
-        cpi->ref_frame_flags = VP9_LAST_FLAG | VP9_GOLD_FLAG;
+      cpi->refresh_golden_frame = 0;
     }
 
+    if (lc->current_video_frame_in_layer == 0)
+      cpi->ref_frame_flags = 0;
+
     vp9_scale_references(cpi);
 
     // Use either last frame or alt frame for motion search.
     if (cpi->ref_frame_flags & VP9_LAST_FLAG) {
       first_ref_buf = vp9_get_scaled_ref_frame(cpi, LAST_FRAME);
-      ref_frame = LAST_FRAME;
       if (first_ref_buf == NULL)
         first_ref_buf = get_ref_frame_buffer(cpi, LAST_FRAME);
-    } else if (cpi->ref_frame_flags & VP9_GOLD_FLAG) {
-      first_ref_buf = vp9_get_scaled_ref_frame(cpi, GOLDEN_FRAME);
-      ref_frame = GOLDEN_FRAME;
-      if (first_ref_buf == NULL)
-        first_ref_buf = get_ref_frame_buffer(cpi, GOLDEN_FRAME);
+    }
+
+    if (cpi->ref_frame_flags & VP9_GOLD_FLAG) {
+      const int ref_idx =
+          cm->ref_frame_map[get_ref_frame_idx(cpi, GOLDEN_FRAME)];
+      const int scaled_idx = cpi->scaled_ref_idx[GOLDEN_FRAME - 1];
+
+      gld_yv12 = (scaled_idx != ref_idx) ? &cm->frame_bufs[scaled_idx].buf :
+                 get_ref_frame_buffer(cpi, GOLDEN_FRAME);
+    } else {
+      gld_yv12 = NULL;
     }
 
     recon_y_stride = new_yv12->y_stride;
     recon_uv_stride = new_yv12->uv_stride;
     uv_mb_height = 16 >> (new_yv12->y_height > new_yv12->uv_height);
 
-    // Disable golden frame for svc first pass for now.
-    gld_yv12 = NULL;
-    set_ref_ptrs(cm, xd, ref_frame, NONE);
+    set_ref_ptrs(cm, xd,
+                 (cpi->ref_frame_flags & VP9_LAST_FLAG) ? LAST_FRAME: NONE,
+                 (cpi->ref_frame_flags & VP9_GOLD_FLAG) ? GOLDEN_FRAME : NONE);
 
     cpi->Source = vp9_scale_if_required(cm, cpi->un_scaled_source,
                                         &cpi->scaled_source);
@@ -483,8 +495,8 @@
   vp9_setup_pre_planes(xd, 0, first_ref_buf, 0, 0, NULL);
   vp9_setup_dst_planes(xd->plane, new_yv12, 0, 0);
 
-  xd->mi = cm->mi_grid_visible;
-  xd->mi[0] = cm->mi;
+  xd->mi = cm->mi;
+  xd->mi[0].src_mi = &xd->mi[0];
 
   vp9_frame_init_quantizer(cpi);
 
@@ -531,8 +543,8 @@
       xd->plane[1].dst.buf = new_yv12->u_buffer + recon_uvoffset;
       xd->plane[2].dst.buf = new_yv12->v_buffer + recon_uvoffset;
       xd->left_available = (mb_col != 0);
-      xd->mi[0]->mbmi.sb_type = bsize;
-      xd->mi[0]->mbmi.ref_frame[0] = INTRA_FRAME;
+      xd->mi[0].src_mi->mbmi.sb_type = bsize;
+      xd->mi[0].src_mi->mbmi.ref_frame[0] = INTRA_FRAME;
       set_mi_row_col(xd, &tile,
                      mb_row << 1, num_8x8_blocks_high_lookup[bsize],
                      mb_col << 1, num_8x8_blocks_wide_lookup[bsize],
@@ -545,8 +557,8 @@
 
       // Do intra 16x16 prediction.
       x->skip_encode = 0;
-      xd->mi[0]->mbmi.mode = DC_PRED;
-      xd->mi[0]->mbmi.tx_size = use_dc_pred ?
+      xd->mi[0].src_mi->mbmi.mode = DC_PRED;
+      xd->mi[0].src_mi->mbmi.tx_size = use_dc_pred ?
          (bsize >= BLOCK_16X16 ? TX_16X16 : TX_8X8) : TX_4X4;
       vp9_encode_intra_block_plane(x, bsize, 0);
       this_error = vp9_get_mb_ss(x->plane[0].src_diff);
@@ -581,7 +593,8 @@
       x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16) + BORDER_MV_PIXELS_B16;
 
       // Other than for the first frame do a motion search.
-      if (cm->current_video_frame > 0) {
+      if ((lc == NULL && cm->current_video_frame > 0) ||
+          (lc != NULL && lc->current_video_frame_in_layer > 0)) {
         int tmp_err, motion_error, raw_motion_error;
         // Assume 0,0 motion with no mv overhead.
         MV mv = {0, 0} , tmp_mv = {0, 0};
@@ -628,7 +641,9 @@
           }
 
           // Search in an older reference frame.
-          if (cm->current_video_frame > 1 && gld_yv12 != NULL) {
+          if (((lc == NULL && cm->current_video_frame > 1) ||
+               (lc != NULL && lc->current_video_frame_in_layer > 1))
+              && gld_yv12 != NULL) {
             // Assume 0,0 motion with no mv overhead.
             int gf_motion_error;
 
@@ -695,11 +710,11 @@
           mv.row *= 8;
           mv.col *= 8;
           this_error = motion_error;
-          xd->mi[0]->mbmi.mode = NEWMV;
-          xd->mi[0]->mbmi.mv[0].as_mv = mv;
-          xd->mi[0]->mbmi.tx_size = TX_4X4;
-          xd->mi[0]->mbmi.ref_frame[0] = LAST_FRAME;
-          xd->mi[0]->mbmi.ref_frame[1] = NONE;
+          xd->mi[0].src_mi->mbmi.mode = NEWMV;
+          xd->mi[0].src_mi->mbmi.mv[0].as_mv = mv;
+          xd->mi[0].src_mi->mbmi.tx_size = TX_4X4;
+          xd->mi[0].src_mi->mbmi.ref_frame[0] = LAST_FRAME;
+          xd->mi[0].src_mi->mbmi.ref_frame[1] = NONE;
           vp9_build_inter_predictors_sby(xd, mb_row << 1, mb_col << 1, bsize);
           vp9_encode_sby_pass1(x, bsize);
           sum_mvr += mv.row;
@@ -817,12 +832,18 @@
   vp9_clear_system_state();
   {
     FIRSTPASS_STATS fps;
+    // The minimum error here insures some bit alocation to frames even
+    // in static regions. The allocation per MB declines for larger formats
+    // where the typical "real" energy per MB also falls.
+    // Initial estimate here uses sqrt(mbs) to define the min_err, where the
+    // number of mbs is propotional to image area.
+    const double min_err = 200 * sqrt(cm->MBs);
 
     fps.frame = cm->current_video_frame;
     fps.spatial_layer_id = cpi->svc.spatial_layer_id;
-    fps.intra_error = (double)(intra_error >> 8);
-    fps.coded_error = (double)(coded_error >> 8);
-    fps.sr_coded_error = (double)(sr_coded_error >> 8);
+    fps.coded_error = (double)(coded_error >> 8) + min_err;
+    fps.sr_coded_error = (double)(sr_coded_error >> 8) + min_err;
+    fps.intra_error = (double)(intra_error >> 8) + min_err;
     fps.count = 1.0;
     fps.pcnt_inter = (double)intercount / cm->MBs;
     fps.pcnt_second_ref = (double)second_ref_count / cm->MBs;
@@ -893,7 +914,7 @@
 
   // Special case for the first frame. Copy into the GF buffer as a second
   // reference.
-  if (cm->current_video_frame == 0 && gld_yv12 != NULL) {
+  if (cm->current_video_frame == 0 && gld_yv12 != NULL && lc == NULL) {
     vp8_yv12_copy_frame(lst_yv12, gld_yv12);
   }
 
@@ -922,12 +943,13 @@
                                      double err_divisor,
                                      double pt_low,
                                      double pt_high,
-                                     int q) {
+                                     int q,
+                                     vpx_bit_depth_t bit_depth) {
   const double error_term = err_per_mb / err_divisor;
 
   // Adjustment based on actual quantizer to power term.
-  const double power_term = MIN(vp9_convert_qindex_to_q(q) * 0.0125 + pt_low,
-                                pt_high);
+  const double power_term =
+      MIN(vp9_convert_qindex_to_q(q, bit_depth) * 0.0125 + pt_low, pt_high);
 
   // Calculate correction factor.
   if (power_term < 1.0)
@@ -962,9 +984,11 @@
       const double factor =
           calc_correction_factor(err_per_mb, ERR_DIVISOR,
                                  is_svc_upper_layer ? SVC_FACTOR_PT_LOW :
-                                 FACTOR_PT_LOW, FACTOR_PT_HIGH, q);
+                                 FACTOR_PT_LOW, FACTOR_PT_HIGH, q,
+                                 cpi->common.bit_depth);
       const int bits_per_mb = vp9_rc_bits_per_mb(INTER_FRAME, q,
-                                                 factor * speed_term);
+                                                 factor * speed_term,
+                                                 cpi->common.bit_depth);
       if (bits_per_mb <= target_norm_bits_per_mb)
         break;
     }
@@ -1017,17 +1041,6 @@
                              10000000.0);
   }
 
-  // Calculate a minimum intra value to be used in determining the IIratio
-  // scores used in the second pass. We have this minimum to make sure
-  // that clips that are static but "low complexity" in the intra domain
-  // are still boosted appropriately for KF/GF/ARF.
-  if (!is_two_pass_svc) {
-    // We don't know the number of MBs for each layer at this point.
-    // So we will do it later.
-    twopass->kf_intra_err_min = KF_MB_INTRA_MIN * cpi->common.MBs;
-    twopass->gf_intra_err_min = GF_MB_INTRA_MIN * cpi->common.MBs;
-  }
-
   // This variable monitors how far behind the second ref update is lagging.
   twopass->sr_update_lag = 1;
 
@@ -1051,34 +1064,58 @@
 
   // Reset the vbr bits off target counter
   cpi->rc.vbr_bits_off_target = 0;
+
+  // Static sequence monitor variables.
+  twopass->kf_zeromotion_pct = 100;
+  twopass->last_kfgroup_zeromotion_pct = 100;
+}
+
+#define SR_DIFF_PART 0.0015
+#define MOTION_AMP_PART 0.003
+#define INTRA_PART 0.005
+#define DEFAULT_DECAY_LIMIT 0.75
+#define LOW_SR_DIFF_TRHESH 0.1
+#define SR_DIFF_MAX 128.0
+
+static double get_sr_decay_rate(const VP9_COMMON *cm,
+                                const FIRSTPASS_STATS *frame) {
+  double sr_diff = (frame->sr_coded_error - frame->coded_error) / cm->MBs;
+  double sr_decay = 1.0;
+  const double motion_amplitude_factor =
+    frame->pcnt_motion * ((frame->mvc_abs + frame->mvr_abs) / 2);
+  const double pcnt_intra = 100 * (1.0 - frame->pcnt_inter);
+
+  if ((sr_diff > LOW_SR_DIFF_TRHESH)) {
+    sr_diff = MIN(sr_diff, SR_DIFF_MAX);
+    sr_decay = 1.0 - (SR_DIFF_PART * sr_diff) -
+               (MOTION_AMP_PART * motion_amplitude_factor) -
+               (INTRA_PART * pcnt_intra);
+  }
+  return MAX(sr_decay, MIN(DEFAULT_DECAY_LIMIT, frame->pcnt_inter));
 }
 
 // This function gives an estimate of how badly we believe the prediction
 // quality is decaying from frame to frame.
-static double get_prediction_decay_rate(const VP9_COMMON *cm,
-                                        const FIRSTPASS_STATS *next_frame) {
-  // Look at the observed drop in prediction quality between the last frame
-  // and the GF buffer (which contains an older frame).
-  const double mb_sr_err_diff = (next_frame->sr_coded_error -
-                                     next_frame->coded_error) / cm->MBs;
-  const double second_ref_decay = mb_sr_err_diff <= 512.0
-      ? fclamp(pow(1.0 - (mb_sr_err_diff / 512.0), 0.5), 0.85, 1.0)
-      : 0.85;
-
-  return MIN(second_ref_decay, next_frame->pcnt_inter);
-}
-
-// This function gives an estimate of how badly we believe the prediction
-// quality is decaying from frame to frame.
-static double get_zero_motion_factor(const FIRSTPASS_STATS *frame) {
-  const double sr_ratio = frame->coded_error /
-                          DOUBLE_DIVIDE_CHECK(frame->sr_coded_error);
+static double get_zero_motion_factor(const VP9_COMMON *cm,
+                                     const FIRSTPASS_STATS *frame) {
   const double zero_motion_pct = frame->pcnt_inter -
                                  frame->pcnt_motion;
-
-  return MIN(sr_ratio, zero_motion_pct);
+  double sr_decay = get_sr_decay_rate(cm, frame);
+  return MIN(sr_decay, zero_motion_pct);
 }
 
+#define ZM_POWER_FACTOR 0.75
+
+static double get_prediction_decay_rate(const VP9_COMMON *cm,
+                                        const FIRSTPASS_STATS *next_frame) {
+  const double sr_decay_rate = get_sr_decay_rate(cm, next_frame);
+  const double zero_motion_factor =
+    (0.95 * pow((next_frame->pcnt_inter - next_frame->pcnt_motion),
+                ZM_POWER_FACTOR));
+
+  return MAX(zero_motion_factor,
+             (sr_decay_rate + ((1.0 - sr_decay_rate) * zero_motion_factor)));
+}
 
 // Function to test for a condition where a complex transition is followed
 // by a static section. For example in slide shows where there is a fade
@@ -1156,19 +1193,17 @@
   }
 }
 
-// Calculate a baseline boost number for the current frame.
-static double calc_frame_boost(const TWO_PASS *twopass,
+#define BASELINE_ERR_PER_MB 1000.0
+static double calc_frame_boost(VP9_COMP *cpi,
                                const FIRSTPASS_STATS *this_frame,
-                               double this_frame_mv_in_out) {
+                               double this_frame_mv_in_out,
+                               double max_boost) {
   double frame_boost;
 
-  // Underlying boost factor is based on inter intra error ratio.
-  if (this_frame->intra_error > twopass->gf_intra_err_min)
-    frame_boost = (IIFACTOR * this_frame->intra_error /
-                   DOUBLE_DIVIDE_CHECK(this_frame->coded_error));
-  else
-    frame_boost = (IIFACTOR * twopass->gf_intra_err_min /
-                   DOUBLE_DIVIDE_CHECK(this_frame->coded_error));
+  // Underlying boost factor is based on inter error ratio.
+  frame_boost = (BASELINE_ERR_PER_MB * cpi->common.MBs) /
+                DOUBLE_DIVIDE_CHECK(this_frame->coded_error);
+  frame_boost = frame_boost * BOOST_FACTOR;
 
   // Increase boost for frames where new data coming into frame (e.g. zoom out).
   // Slightly reduce boost if there is a net balance of motion out of the frame
@@ -1179,7 +1214,7 @@
   else
     frame_boost += frame_boost * (this_frame_mv_in_out / 2.0);
 
-  return MIN(frame_boost, GF_RMAX);
+  return MIN(frame_boost, max_boost);
 }
 
 static int calc_arf_boost(VP9_COMP *cpi, int offset,
@@ -1220,8 +1255,9 @@
                           ? MIN_DECAY_FACTOR : decay_accumulator;
     }
 
-    boost_score += decay_accumulator * calc_frame_boost(twopass, this_frame,
-                                                        this_frame_mv_in_out);
+    boost_score += decay_accumulator * calc_frame_boost(cpi, this_frame,
+                                                        this_frame_mv_in_out,
+                                                        GF_MAX_BOOST);
   }
 
   *f_boost = (int)boost_score;
@@ -1258,8 +1294,9 @@
                               ? MIN_DECAY_FACTOR : decay_accumulator;
     }
 
-    boost_score += decay_accumulator * calc_frame_boost(twopass, this_frame,
-                                                        this_frame_mv_in_out);
+    boost_score += decay_accumulator * calc_frame_boost(cpi, this_frame,
+                                                        this_frame_mv_in_out,
+                                                        GF_MAX_BOOST);
   }
   *b_boost = (int)boost_score;
 
@@ -1569,7 +1606,7 @@
     gf_group_err -= gf_first_frame_err;
 
   // Motion breakout threshold for loop below depends on image size.
-  mv_ratio_accumulator_thresh = (cpi->common.width + cpi->common.height) / 10.0;
+  mv_ratio_accumulator_thresh = (cpi->common.width + cpi->common.height) / 4.0;
 
   // Work out a maximum interval for the GF group.
   // If the image appears almost completely static we can extend beyond this.
@@ -1581,7 +1618,8 @@
    // At high Q when there are few bits to spare we are better with a longer
    // interval to spread the cost of the GF.
    active_max_gf_interval =
-     12 + ((int)vp9_convert_qindex_to_q(rc->last_q[INTER_FRAME]) >> 5);
+     12 + ((int)vp9_convert_qindex_to_q(rc->last_q[INTER_FRAME],
+                                        cpi->common.bit_depth) >> 5);
 
    if (active_max_gf_interval > rc->max_gf_interval)
      active_max_gf_interval = rc->max_gf_interval;
@@ -1612,11 +1650,13 @@
     if (!flash_detected) {
       last_loop_decay_rate = loop_decay_rate;
       loop_decay_rate = get_prediction_decay_rate(&cpi->common, &next_frame);
+
       decay_accumulator = decay_accumulator * loop_decay_rate;
 
       // Monitor for static sections.
-      zero_motion_accumulator = MIN(zero_motion_accumulator,
-                                    get_zero_motion_factor(&next_frame));
+      zero_motion_accumulator =
+        MIN(zero_motion_accumulator,
+            get_zero_motion_factor(&cpi->common, &next_frame));
 
       // Break clause to detect very still sections after motion. For example,
       // a static image after a fade or other transition.
@@ -1628,8 +1668,9 @@
     }
 
     // Calculate a boost number for this frame.
-    boost_score += decay_accumulator * calc_frame_boost(twopass, &next_frame,
-                                                        this_frame_mv_in_out);
+    boost_score += decay_accumulator * calc_frame_boost(cpi, &next_frame,
+                                                        this_frame_mv_in_out,
+                                                        GF_MAX_BOOST);
 
     // Break out conditions.
     if (
@@ -1638,38 +1679,21 @@
       (
         // Don't break out with a very short interval.
         (i > MIN_GF_INTERVAL) &&
-        ((boost_score > 125.0) || (next_frame.pcnt_inter < 0.75)) &&
         (!flash_detected) &&
         ((mv_ratio_accumulator > mv_ratio_accumulator_thresh) ||
          (abs_mv_in_out_accumulator > 3.0) ||
          (mv_in_out_accumulator < -2.0) ||
-         ((boost_score - old_boost_score) < IIFACTOR)))) {
+         ((boost_score - old_boost_score) < BOOST_FACTOR)))) {
       boost_score = old_boost_score;
       break;
     }
 
     *this_frame = next_frame;
-
     old_boost_score = boost_score;
   }
 
   twopass->gf_zeromotion_pct = (int)(zero_motion_accumulator * 1000.0);
 
-  // Don't allow a gf too near the next kf.
-  if ((rc->frames_to_key - i) < MIN_GF_INTERVAL) {
-    while (i < (rc->frames_to_key + !rc->next_key_frame_forced)) {
-      ++i;
-
-      if (EOF == input_stats(twopass, this_frame))
-        break;
-
-      if (i < rc->frames_to_key) {
-        mod_frame_err = calculate_modified_err(twopass, oxcf, this_frame);
-        gf_group_err += mod_frame_err;
-      }
-    }
-  }
-
   // Set the interval until the next gf.
   if (cpi->common.frame_type == KEY_FRAME || rc->source_alt_ref_active)
     rc->baseline_gf_interval = i - 1;
@@ -1696,10 +1720,7 @@
   // Should we use the alternate reference frame.
   if (allow_alt_ref &&
       (i < cpi->oxcf.lag_in_frames) &&
-      (i >= MIN_GF_INTERVAL) &&
-      // For real scene cuts (not forced kfs) don't allow arf very near kf.
-      (rc->next_key_frame_forced ||
-      (i <= (rc->frames_to_key - MIN_GF_INTERVAL)))) {
+      (i >= MIN_GF_INTERVAL)) {
     // Calculate the boost for alt ref.
     rc->gfu_boost = calc_arf_boost(cpi, 0, (i - 1), (i - 1), &f_boost,
                                    &b_boost);
@@ -1710,7 +1731,7 @@
       (cpi->multi_arf_allowed && (rc->baseline_gf_interval >= 6) &&
       (zero_motion_accumulator < 0.995)) ? 1 : 0;
   } else {
-    rc->gfu_boost = (int)boost_score;
+    rc->gfu_boost = MAX((int)boost_score, 125);
     rc->source_alt_ref_pending = 0;
   }
 
@@ -1723,7 +1744,8 @@
   // Calculate the extra bits to be used for boosted frame(s)
   {
     int q = rc->last_q[INTER_FRAME];
-    int boost = (rc->gfu_boost * gfboost_qadjust(q)) / 100;
+    int boost =
+        (rc->gfu_boost * gfboost_qadjust(q, cpi->common.bit_depth)) / 100;
 
     // Set max and minimum boost and hence minimum allocation.
     boost = clamp(boost, 125, (rc->baseline_gf_interval + 1) * 200);
@@ -1764,6 +1786,9 @@
   }
 }
 
+// TODO(PGW) Re-examine the use of II ration in this code in the light of#
+// changes elsewhere
+#define KF_II_MAX 128.0
 static int test_candidate_kf(TWO_PASS *twopass,
                              const FIRSTPASS_STATS *last_frame,
                              const FIRSTPASS_STATS *this_frame,
@@ -1793,11 +1818,11 @@
 
     // Examine how well the key frame predicts subsequent frames.
     for (i = 0; i < 16; ++i) {
-      double next_iiratio = (IIKFACTOR1 * local_next_frame.intra_error /
+      double next_iiratio = (BOOST_FACTOR * local_next_frame.intra_error /
                              DOUBLE_DIVIDE_CHECK(local_next_frame.coded_error));
 
-      if (next_iiratio > RMAX)
-        next_iiratio = RMAX;
+      if (next_iiratio > KF_II_MAX)
+        next_iiratio = KF_II_MAX;
 
       // Cumulative effect of decay in prediction quality.
       if (local_next_frame.pcnt_inter > 0.85)
@@ -1852,7 +1877,9 @@
   FIRSTPASS_STATS next_frame;
   FIRSTPASS_STATS last_frame;
   int kf_bits = 0;
+  int loop_decay_counter = 0;
   double decay_accumulator = 1.0;
+  double av_decay_accumulator = 0.0;
   double zero_motion_accumulator = 1.0;
   double boost_score = 0.0;
   double kf_mod_err = 0.0;
@@ -2006,42 +2033,38 @@
   // Reset the first pass file position.
   reset_fpf_position(twopass, start_position);
 
-  // Scan through the kf group collating various stats used to deteermine
+  // Scan through the kf group collating various stats used to determine
   // how many bits to spend on it.
   decay_accumulator = 1.0;
   boost_score = 0.0;
-  for (i = 0; i < rc->frames_to_key; ++i) {
+  for (i = 0; i < (rc->frames_to_key - 1); ++i) {
     if (EOF == input_stats(twopass, &next_frame))
       break;
 
     // Monitor for static sections.
-    zero_motion_accumulator =MIN(zero_motion_accumulator,
-                                 get_zero_motion_factor(&next_frame));
+    zero_motion_accumulator =
+      MIN(zero_motion_accumulator,
+          get_zero_motion_factor(&cpi->common, &next_frame));
 
-    // For the first few frames collect data to decide kf boost.
-    if (i <= (rc->max_gf_interval * 2)) {
-      double r;
-      if (next_frame.intra_error > twopass->kf_intra_err_min)
-        r = (IIKFACTOR2 * next_frame.intra_error /
-             DOUBLE_DIVIDE_CHECK(next_frame.coded_error));
-      else
-        r = (IIKFACTOR2 * twopass->kf_intra_err_min /
-             DOUBLE_DIVIDE_CHECK(next_frame.coded_error));
-
-      if (r > RMAX)
-        r = RMAX;
+    // Not all frames in the group are necessarily used in calculating boost.
+    if ((i <= rc->max_gf_interval) ||
+        ((i <= (rc->max_gf_interval * 4)) && (decay_accumulator > 0.5))) {
+      const double frame_boost =
+        calc_frame_boost(cpi, this_frame, 0, KF_MAX_BOOST);
 
       // How fast is prediction quality decaying.
       if (!detect_flash(twopass, 0)) {
-        const double loop_decay_rate = get_prediction_decay_rate(&cpi->common,
-                                                                 &next_frame);
+        const double loop_decay_rate =
+          get_prediction_decay_rate(&cpi->common, &next_frame);
         decay_accumulator *= loop_decay_rate;
         decay_accumulator = MAX(decay_accumulator, MIN_DECAY_FACTOR);
+        av_decay_accumulator += decay_accumulator;
+        ++loop_decay_counter;
       }
-
-      boost_score += (decay_accumulator * r);
+      boost_score += (decay_accumulator * frame_boost);
     }
   }
+  av_decay_accumulator /= (double)loop_decay_counter;
 
   reset_fpf_position(twopass, start_position);
 
@@ -2053,14 +2076,12 @@
       calculate_section_intra_ratio(start_position, twopass->stats_in_end,
                                     rc->frames_to_key);
 
+  // Apply various clamps for min and max boost
+  rc->kf_boost = (int)(av_decay_accumulator * boost_score);
+  rc->kf_boost = MAX(rc->kf_boost, (rc->frames_to_key * 3));
+  rc->kf_boost = MAX(rc->kf_boost, MIN_KF_BOOST);
+
   // Work out how many bits to allocate for the key frame itself.
-  rc->kf_boost = (int)boost_score;
-
-  if (rc->kf_boost  < (rc->frames_to_key * 3))
-    rc->kf_boost  = (rc->frames_to_key * 3);
-  if (rc->kf_boost   < MIN_KF_BOOST)
-    rc->kf_boost = MIN_KF_BOOST;
-
   kf_bits = calculate_boost_bits((rc->frames_to_key - 1),
                                   rc->kf_boost, twopass->kf_group_bits);
 
@@ -2134,6 +2155,10 @@
       break;
   }
   if (is_two_pass_svc(cpi)) {
+    if (cpi->svc.temporal_layer_id > 0) {
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 0;
+    }
     if (cpi->svc.layer_context[cpi->svc.spatial_layer_id].gold_ref_idx < 0)
       cpi->refresh_golden_frame = 0;
     if (cpi->alt_ref_source == NULL)
@@ -2198,11 +2223,6 @@
 
   vp9_clear_system_state();
 
-  if (lc != NULL && twopass->kf_intra_err_min == 0) {
-    twopass->kf_intra_err_min = KF_MB_INTRA_MIN * cpi->common.MBs;
-    twopass->gf_intra_err_min = GF_MB_INTRA_MIN * cpi->common.MBs;
-  }
-
   if (cpi->oxcf.rc_mode == VPX_Q) {
     twopass->active_worst_quality = cpi->oxcf.cq_level;
   } else if (cm->current_video_frame == 0 ||
@@ -2214,7 +2234,7 @@
                                                 section_target_bandwidth);
     twopass->active_worst_quality = tmp_q;
     rc->ni_av_qi = tmp_q;
-    rc->avg_q = vp9_convert_qindex_to_q(tmp_q);
+    rc->avg_q = vp9_convert_qindex_to_q(tmp_q, cm->bit_depth);
   }
   vp9_zero(this_frame);
   if (EOF == input_stats(twopass, &this_frame))
@@ -2268,6 +2288,18 @@
     rc->frames_till_gf_update_due = rc->baseline_gf_interval;
     if (lc != NULL)
       cpi->refresh_golden_frame = 1;
+
+#if ARF_STATS_OUTPUT
+    {
+      FILE *fpfile;
+      fpfile = fopen("arf.stt", "a");
+      ++arf_count;
+      fprintf(fpfile, "%10d %10d %10d %10ld\n",
+              cm->current_video_frame, rc->kf_boost, arf_count, rc->gfu_boost);
+
+      fclose(fpfile);
+    }
+#endif
   }
 
   configure_buffer_updates(cpi);
@@ -2307,6 +2339,7 @@
   if (cpi->common.frame_type != KEY_FRAME &&
       !vp9_is_upper_layer_key_frame(cpi)) {
     twopass->kf_group_bits -= bits_used;
+    twopass->last_kfgroup_zeromotion_pct = twopass->kf_zeromotion_pct;
   }
   twopass->kf_group_bits = MAX(twopass->kf_group_bits, 0);
 
diff --git a/vp9/encoder/vp9_firstpass.h b/vp9/encoder/vp9_firstpass.h
index aaa6b03..0b82d32 100644
--- a/vp9/encoder/vp9_firstpass.h
+++ b/vp9/encoder/vp9_firstpass.h
@@ -93,8 +93,6 @@
   double modified_error_min;
   double modified_error_max;
   double modified_error_left;
-  double kf_intra_err_min;
-  double gf_intra_err_min;
 
 #if CONFIG_FP_MB_STATS
   uint8_t *frame_mb_stats_buf;
@@ -110,6 +108,7 @@
   int sr_update_lag;
 
   int kf_zeromotion_pct;
+  int last_kfgroup_zeromotion_pct;
   int gf_zeromotion_pct;
 
   int active_worst_quality;
diff --git a/vp9/encoder/vp9_lookahead.h b/vp9/encoder/vp9_lookahead.h
index 2786193..a33d300 100644
--- a/vp9/encoder/vp9_lookahead.h
+++ b/vp9/encoder/vp9_lookahead.h
@@ -30,10 +30,6 @@
   int64_t             ts_start;
   int64_t             ts_end;
   unsigned int        flags;
-
-#if CONFIG_SPATIAL_SVC
-  vpx_svc_parameters_t svc_params[VPX_SS_MAX_LAYERS];
-#endif
 };
 
 // The max of past frames we want to keep in the queue.
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index b8e7164..42981d8 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -63,8 +63,8 @@
         &distortion, &sse, NULL, 0, 0);
   }
 
-  xd->mi[0]->mbmi.mode = NEWMV;
-  xd->mi[0]->mbmi.mv[0].as_mv = *dst_mv;
+  xd->mi[0].src_mi->mbmi.mode = NEWMV;
+  xd->mi[0].src_mi->mbmi.mv[0].as_mv = *dst_mv;
 
   vp9_build_inter_predictors_sby(xd, mb_row, mb_col, BLOCK_16X16);
 
@@ -141,7 +141,7 @@
   for (mode = DC_PRED; mode <= TM_PRED; mode++) {
     unsigned int err;
 
-    xd->mi[0]->mbmi.mode = mode;
+    xd->mi[0].src_mi->mbmi.mode = mode;
     vp9_predict_intra_block(xd, 0, 2, TX_16X16, mode,
                             x->plane[0].src.buf, x->plane[0].src.stride,
                             xd->plane[0].dst.buf, xd->plane[0].dst.stride,
@@ -247,7 +247,7 @@
   xd->plane[0].dst.stride  = buf->y_stride;
   xd->plane[0].pre[0].stride  = buf->y_stride;
   xd->plane[1].dst.stride = buf->uv_stride;
-  xd->mi[0] = &mi_local;
+  xd->mi[0].src_mi = &mi_local;
   mi_local.mbmi.sb_type = BLOCK_16X16;
   mi_local.mbmi.ref_frame[0] = LAST_FRAME;
   mi_local.mbmi.ref_frame[1] = NONE;
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index d6f6b25..5366c3c 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -320,24 +320,24 @@
     switch (whichdir) {
       case 0:
         CHECK_BETTER(left, tr, tc - hstep);
-        CHECK_BETTER(up, tr - hstep, tc);
-        CHECK_BETTER(diag, tr - hstep, tc - hstep);
-        break;
-      case 1:
-        CHECK_BETTER(right, tr, tc + hstep);
-        CHECK_BETTER(up, tr - hstep, tc);
-        CHECK_BETTER(diag, tr - hstep, tc + hstep);
-        break;
-      case 2:
-        CHECK_BETTER(left, tr, tc - hstep);
         CHECK_BETTER(down, tr + hstep, tc);
         CHECK_BETTER(diag, tr + hstep, tc - hstep);
         break;
-      case 3:
+      case 1:
         CHECK_BETTER(right, tr, tc + hstep);
         CHECK_BETTER(down, tr + hstep, tc);
         CHECK_BETTER(diag, tr + hstep, tc + hstep);
         break;
+      case 2:
+        CHECK_BETTER(left, tr, tc - hstep);
+        CHECK_BETTER(up, tr - hstep, tc);
+        CHECK_BETTER(diag, tr - hstep, tc - hstep);
+        break;
+      case 3:
+        CHECK_BETTER(right, tr, tc + hstep);
+        CHECK_BETTER(up, tr - hstep, tc);
+        CHECK_BETTER(diag, tr - hstep, tc + hstep);
+        break;
     }
   } else {
     FIRST_LEVEL_CHECKS;
@@ -484,6 +484,52 @@
 #define MAX_PATTERN_CANDIDATES      8  // max number of canddiates per scale
 #define PATTERN_CANDIDATES_REF      3  // number of refinement candidates
 
+// Calculate and return a sad+mvcost list around an integer best pel.
+static INLINE void calc_int_sad_cost_list(MACROBLOCK *x,
+                                          const MV *ref_mv,
+                                          int sadpb,
+                                          const vp9_variance_fn_ptr_t *fn_ptr,
+                                          const MV *best_mv,
+                                          int *cost_list) {
+  static const MV neighbors[4] = {{0, -1}, {1, 0}, {0, 1}, {-1, 0}};
+  const struct buf_2d *const what = &x->plane[0].src;
+  const struct buf_2d *const in_what = &x->e_mbd.plane[0].pre[0];
+  const MV fcenter_mv = {ref_mv->row >> 3, ref_mv->col >> 3};
+  int br = best_mv->row;
+  int bc = best_mv->col;
+  MV this_mv;
+  int i;
+
+  this_mv.row = br;
+  this_mv.col = bc;
+  cost_list[0] = fn_ptr->sdf(what->buf, what->stride,
+                             get_buf_from_mv(in_what, &this_mv),
+                             in_what->stride) +
+      mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
+  if (check_bounds(x, br, bc, 1)) {
+    for (i = 0; i < 4; i++) {
+      const MV this_mv = {br + neighbors[i].row,
+        bc + neighbors[i].col};
+      cost_list[i + 1] = fn_ptr->sdf(what->buf, what->stride,
+                                     get_buf_from_mv(in_what, &this_mv),
+                                     in_what->stride) +
+          mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
+    }
+  } else {
+    for (i = 0; i < 4; i++) {
+      const MV this_mv = {br + neighbors[i].row,
+        bc + neighbors[i].col};
+      if (!is_mv_in(x, &this_mv))
+        cost_list[i + 1] = INT_MAX;
+      else
+        cost_list[i + 1] = fn_ptr->sdf(what->buf, what->stride,
+                                       get_buf_from_mv(in_what, &this_mv),
+                                       in_what->stride) +
+            mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
+    }
+  }
+}
+
 // Generic pattern search function that searches over multiple scales.
 // Each scale can have a different number of candidates and shape of
 // candidates as indicated in the num_candidates and candidates arrays
@@ -648,11 +694,11 @@
   // Returns the one-away integer pel sad values around the best as follows:
   // sad_list[0]: sad at the best integer pel
   // sad_list[1]: sad at delta {0, -1} (left)   from the best integer pel
-  // sad_list[2]: sad at delta {-1, 0} (top)    from the best integer pel
+  // sad_list[2]: sad at delta { 1, 0} (bottom) from the best integer pel
   // sad_list[3]: sad at delta { 0, 1} (right)  from the best integer pel
-  // sad_list[4]: sad at delta { 1, 0} (bottom) from the best integer pel
+  // sad_list[4]: sad at delta {-1, 0} (top)    from the best integer pel
   if (sad_list) {
-    static const MV neighbors[4] = {{0, -1}, {-1, 0}, {0, 1}, {1, 0}};
+    static const MV neighbors[4] = {{0, -1}, {1, 0}, {0, 1}, {-1, 0}};
     sad_list[0] = bestsad;
     if (check_bounds(x, br, bc, 1)) {
       for (i = 0; i < 4; i++) {
@@ -660,7 +706,10 @@
                             bc + neighbors[i].col};
         sad_list[i + 1] = vfp->sdf(what->buf, what->stride,
                                    get_buf_from_mv(in_what, &this_mv),
-                                   in_what->stride);
+                                   in_what->stride) +
+            (use_mvcost ?
+             mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit) :
+             0);
       }
     } else {
       for (i = 0; i < 4; i++) {
@@ -671,7 +720,300 @@
         else
           sad_list[i + 1] = vfp->sdf(what->buf, what->stride,
                                      get_buf_from_mv(in_what, &this_mv),
+                                     in_what->stride) +
+              (use_mvcost ?
+               mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit) :
+               0);
+      }
+    }
+  }
+  best_mv->row = br;
+  best_mv->col = bc;
+  return bestsad;
+}
+
+// A specialized function where the smallest scale search candidates
+// are 4 1-away neighbors, and sad_list is non-null
+// TODO(debargha): Merge this function with the one above. Also remove
+// use_mvcost option since it is always 1, to save unnecessary branches.
+static int vp9_pattern_search_sad(const MACROBLOCK *x,
+                                  MV *ref_mv,
+                                  int search_param,
+                                  int sad_per_bit,
+                                  int do_init_search,
+                                  int *sad_list,
+                                  const vp9_variance_fn_ptr_t *vfp,
+                                  int use_mvcost,
+                                  const MV *center_mv,
+                                  MV *best_mv,
+                                  const int num_candidates[MAX_PATTERN_SCALES],
+                                  const MV candidates[MAX_PATTERN_SCALES]
+                                                     [MAX_PATTERN_CANDIDATES]) {
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  static const int search_param_to_steps[MAX_MVSEARCH_STEPS] = {
+    10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
+  };
+  int i, s, t;
+  const struct buf_2d *const what = &x->plane[0].src;
+  const struct buf_2d *const in_what = &xd->plane[0].pre[0];
+  int br, bc;
+  int bestsad = INT_MAX;
+  int thissad;
+  int k = -1;
+  const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
+  int best_init_s = search_param_to_steps[search_param];
+  // adjust ref_mv to make sure it is within MV range
+  clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+  br = ref_mv->row;
+  bc = ref_mv->col;
+  if (sad_list != NULL) {
+    sad_list[0] = sad_list[1] = sad_list[2] = sad_list[3] = sad_list[4] =
+        INT_MAX;
+  }
+
+  // Work out the start point for the search
+  bestsad = vfp->sdf(what->buf, what->stride,
+                     get_buf_from_mv(in_what, ref_mv), in_what->stride) +
+      mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
+
+  // Search all possible scales upto the search param around the center point
+  // pick the scale of the point that is best as the starting scale of
+  // further steps around it.
+  if (do_init_search) {
+    s = best_init_s;
+    best_init_s = -1;
+    for (t = 0; t <= s; ++t) {
+      int best_site = -1;
+      if (check_bounds(x, br, bc, 1 << t)) {
+        for (i = 0; i < num_candidates[t]; i++) {
+          const MV this_mv = {br + candidates[t][i].row,
+                              bc + candidates[t][i].col};
+          thissad = vfp->sdf(what->buf, what->stride,
+                             get_buf_from_mv(in_what, &this_mv),
+                             in_what->stride);
+          CHECK_BETTER
+        }
+      } else {
+        for (i = 0; i < num_candidates[t]; i++) {
+          const MV this_mv = {br + candidates[t][i].row,
+                              bc + candidates[t][i].col};
+          if (!is_mv_in(x, &this_mv))
+            continue;
+          thissad = vfp->sdf(what->buf, what->stride,
+                             get_buf_from_mv(in_what, &this_mv),
+                             in_what->stride);
+          CHECK_BETTER
+        }
+      }
+      if (best_site == -1) {
+        continue;
+      } else {
+        best_init_s = t;
+        k = best_site;
+      }
+    }
+    if (best_init_s != -1) {
+      br += candidates[best_init_s][k].row;
+      bc += candidates[best_init_s][k].col;
+    }
+  }
+
+  // If the center point is still the best, just skip this and move to
+  // the refinement step.
+  if (best_init_s != -1) {
+    int do_sad = (num_candidates[0] == 4 && sad_list != NULL);
+    int best_site = -1;
+    s = best_init_s;
+
+    for (; s >= do_sad; s--) {
+      if (!do_init_search || s != best_init_s) {
+        if (check_bounds(x, br, bc, 1 << s)) {
+          for (i = 0; i < num_candidates[s]; i++) {
+            const MV this_mv = {br + candidates[s][i].row,
+                                bc + candidates[s][i].col};
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        } else {
+          for (i = 0; i < num_candidates[s]; i++) {
+            const MV this_mv = {br + candidates[s][i].row,
+                                bc + candidates[s][i].col};
+            if (!is_mv_in(x, &this_mv))
+              continue;
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        }
+
+        if (best_site == -1) {
+          continue;
+        } else {
+          br += candidates[s][best_site].row;
+          bc += candidates[s][best_site].col;
+          k = best_site;
+        }
+      }
+
+      do {
+        int next_chkpts_indices[PATTERN_CANDIDATES_REF];
+        best_site = -1;
+        next_chkpts_indices[0] = (k == 0) ? num_candidates[s] - 1 : k - 1;
+        next_chkpts_indices[1] = k;
+        next_chkpts_indices[2] = (k == num_candidates[s] - 1) ? 0 : k + 1;
+
+        if (check_bounds(x, br, bc, 1 << s)) {
+          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
+            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
+                                bc + candidates[s][next_chkpts_indices[i]].col};
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        } else {
+          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
+            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
+                                bc + candidates[s][next_chkpts_indices[i]].col};
+            if (!is_mv_in(x, &this_mv))
+              continue;
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        }
+
+        if (best_site != -1) {
+          k = next_chkpts_indices[best_site];
+          br += candidates[s][k].row;
+          bc += candidates[s][k].col;
+        }
+      } while (best_site != -1);
+    }
+
+    // Note: If we enter the if below, then sad_list must be non-NULL.
+    if (s == 0) {
+      sad_list[0] = bestsad;
+      if (!do_init_search || s != best_init_s) {
+        if (check_bounds(x, br, bc, 1 << s)) {
+          for (i = 0; i < num_candidates[s]; i++) {
+            const MV this_mv = {br + candidates[s][i].row,
+                                bc + candidates[s][i].col};
+            sad_list[i + 1] =
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        } else {
+          for (i = 0; i < num_candidates[s]; i++) {
+            const MV this_mv = {br + candidates[s][i].row,
+                                bc + candidates[s][i].col};
+            if (!is_mv_in(x, &this_mv))
+              continue;
+            sad_list[i + 1] =
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        }
+
+        if (best_site != -1) {
+          br += candidates[s][best_site].row;
+          bc += candidates[s][best_site].col;
+          k = best_site;
+        }
+      }
+      while (best_site != -1) {
+        int next_chkpts_indices[PATTERN_CANDIDATES_REF];
+        best_site = -1;
+        next_chkpts_indices[0] = (k == 0) ? num_candidates[s] - 1 : k - 1;
+        next_chkpts_indices[1] = k;
+        next_chkpts_indices[2] = (k == num_candidates[s] - 1) ? 0 : k + 1;
+        sad_list[1] = sad_list[2] = sad_list[3] = sad_list[4] = INT_MAX;
+        sad_list[((k + 2) % 4) + 1] = sad_list[0];
+        sad_list[0] = bestsad;
+
+        if (check_bounds(x, br, bc, 1 << s)) {
+          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
+            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
+                                bc + candidates[s][next_chkpts_indices[i]].col};
+            sad_list[next_chkpts_indices[i] + 1] =
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        } else {
+          for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
+            const MV this_mv = {br + candidates[s][next_chkpts_indices[i]].row,
+                                bc + candidates[s][next_chkpts_indices[i]].col};
+            if (!is_mv_in(x, &this_mv)) {
+              sad_list[next_chkpts_indices[i] + 1] = INT_MAX;
+              continue;
+            }
+            sad_list[next_chkpts_indices[i] + 1] =
+            thissad = vfp->sdf(what->buf, what->stride,
+                               get_buf_from_mv(in_what, &this_mv),
+                               in_what->stride);
+            CHECK_BETTER
+          }
+        }
+
+        if (best_site != -1) {
+          k = next_chkpts_indices[best_site];
+          br += candidates[s][k].row;
+          bc += candidates[s][k].col;
+        }
+      }
+    }
+  }
+
+  // Returns the one-away integer pel sad values around the best as follows:
+  // sad_list[0]: sad at the best integer pel
+  // sad_list[1]: sad at delta {0, -1} (left)   from the best integer pel
+  // sad_list[2]: sad at delta { 1, 0} (bottom) from the best integer pel
+  // sad_list[3]: sad at delta { 0, 1} (right)  from the best integer pel
+  // sad_list[4]: sad at delta {-1, 0} (top)    from the best integer pel
+  if (sad_list) {
+    static const MV neighbors[4] = {{0, -1}, {1, 0}, {0, 1}, {-1, 0}};
+    if (sad_list[0] == INT_MAX) {
+      sad_list[0] = bestsad;
+      if (check_bounds(x, br, bc, 1)) {
+        for (i = 0; i < 4; i++) {
+          const MV this_mv = {br + neighbors[i].row,
+            bc + neighbors[i].col};
+          sad_list[i + 1] = vfp->sdf(what->buf, what->stride,
+                                     get_buf_from_mv(in_what, &this_mv),
                                      in_what->stride);
+        }
+      } else {
+        for (i = 0; i < 4; i++) {
+          const MV this_mv = {br + neighbors[i].row,
+            bc + neighbors[i].col};
+          if (!is_mv_in(x, &this_mv))
+            sad_list[i + 1] = INT_MAX;
+          else
+            sad_list[i + 1] = vfp->sdf(what->buf, what->stride,
+                                       get_buf_from_mv(in_what, &this_mv),
+                                       in_what->stride);
+        }
+      }
+    } else {
+      if (use_mvcost) {
+        for (i = 0; i < 4; i++) {
+          const MV this_mv = {br + neighbors[i].row,
+            bc + neighbors[i].col};
+          if (sad_list[i + 1] != INT_MAX) {
+            sad_list[i + 1] +=
+                mvsad_err_cost(x, &this_mv, &fcenter_mv, sad_per_bit);
+          }
+        }
       }
     }
   }
@@ -784,10 +1126,10 @@
     {{-512, -512}, {0, -1024}, {512, -512}, {1024, 0}, {512, 512}, {0, 1024},
       {-512, 512}, {-1024, 0}},
   };
-  return vp9_pattern_search(x, ref_mv, search_param, sad_per_bit,
-                            do_init_search, sad_list, vfp, use_mvcost,
-                            center_mv, best_mv,
-                            bigdia_num_candidates, bigdia_candidates);
+  return vp9_pattern_search_sad(x, ref_mv, search_param, sad_per_bit,
+                                do_init_search, sad_list, vfp, use_mvcost,
+                                center_mv, best_mv,
+                                bigdia_num_candidates, bigdia_candidates);
 }
 
 int vp9_square_search(const MACROBLOCK *x,
@@ -1082,10 +1424,10 @@
 /* do_refine: If last step (1-away) of n-step search doesn't pick the center
               point as the best match, we will do a final 1-away diamond
               refining search  */
-
 int vp9_full_pixel_diamond(const VP9_COMP *cpi, MACROBLOCK *x,
                            MV *mvp_full, int step_param,
                            int sadpb, int further_steps, int do_refine,
+                           int *cost_list,
                            const vp9_variance_fn_ptr_t *fn_ptr,
                            const MV *ref_mv, MV *dst_mv) {
   MV temp_mv;
@@ -1138,6 +1480,11 @@
       *dst_mv = best_mv;
     }
   }
+
+  // Return cost list.
+  if (cost_list) {
+    calc_int_sad_cost_list(x, ref_mv, sadpb, fn_ptr, dst_mv, cost_list);
+  }
   return bestsme;
 }
 
@@ -1496,7 +1843,7 @@
     case NSTEP:
       var = vp9_full_pixel_diamond(cpi, x, mvp_full, step_param, error_per_bit,
                                    MAX_MVSEARCH_STEPS - 1 - step_param,
-                                   1, fn_ptr, ref_mv, tmp_mv);
+                                   1, sad_list, fn_ptr, ref_mv, tmp_mv);
       break;
     default:
       assert(!"Invalid search method.");
diff --git a/vp9/encoder/vp9_mcomp.h b/vp9/encoder/vp9_mcomp.h
index 9b4734a..eb3f3eb 100644
--- a/vp9/encoder/vp9_mcomp.h
+++ b/vp9/encoder/vp9_mcomp.h
@@ -70,6 +70,7 @@
 int vp9_full_pixel_diamond(const struct VP9_COMP *cpi, MACROBLOCK *x,
                            MV *mvp_full, int step_param,
                            int sadpb, int further_steps, int do_refine,
+                           int *cost_list,
                            const vp9_variance_fn_ptr_t *fn_ptr,
                            const MV *ref_mv, MV *dst_mv);
 
diff --git a/vp9/encoder/vp9_picklpf.c b/vp9/encoder/vp9_picklpf.c
index d365489..2fc05e7 100644
--- a/vp9/encoder/vp9_picklpf.c
+++ b/vp9/encoder/vp9_picklpf.c
@@ -77,7 +77,6 @@
   while (filter_step > 0) {
     const int filt_high = MIN(filt_mid + filter_step, max_filter_level);
     const int filt_low = MAX(filt_mid - filter_step, min_filter_level);
-    int filt_err;
 
     // Bias against raising loop filter in favor of lowering it.
     int bias = (best_err >> (15 - (filt_mid / 8))) * filter_step;
@@ -92,17 +91,14 @@
     if (filt_direction <= 0 && filt_low != filt_mid) {
       // Get Low filter error score
       if (ss_err[filt_low] < 0) {
-        filt_err = try_filter_frame(sd, cpi, filt_low, partial_frame);
-        ss_err[filt_low] = filt_err;
-      } else {
-        filt_err = ss_err[filt_low];
+        ss_err[filt_low] = try_filter_frame(sd, cpi, filt_low, partial_frame);
       }
       // If value is close to the best so far then bias towards a lower loop
       // filter value.
-      if ((filt_err - bias) < best_err) {
+      if ((ss_err[filt_low] - bias) < best_err) {
         // Was it actually better than the previous best?
-        if (filt_err < best_err)
-          best_err = filt_err;
+        if (ss_err[filt_low] < best_err)
+          best_err = ss_err[filt_low];
 
         filt_best = filt_low;
       }
@@ -111,14 +107,11 @@
     // Now look at filt_high
     if (filt_direction >= 0 && filt_high != filt_mid) {
       if (ss_err[filt_high] < 0) {
-        filt_err = try_filter_frame(sd, cpi, filt_high, partial_frame);
-        ss_err[filt_high] = filt_err;
-      } else {
-        filt_err = ss_err[filt_high];
+        ss_err[filt_high] = try_filter_frame(sd, cpi, filt_high, partial_frame);
       }
       // Was it better than the previous best?
-      if (filt_err < (best_err - bias)) {
-        best_err = filt_err;
+      if (ss_err[filt_high] < (best_err - bias)) {
+        best_err = ss_err[filt_high];
         filt_best = filt_high;
       }
     }
@@ -149,7 +142,7 @@
   } else if (method >= LPF_PICK_FROM_Q) {
     const int min_filter_level = 0;
     const int max_filter_level = get_max_filter_level(cpi);
-    const int q = vp9_ac_quant(cm->base_qindex, 0);
+    const int q = vp9_ac_quant(cm->base_qindex, 0, cm->bit_depth);
     // These values were determined by linear fitting the result of the
     // searched level, filt_guess = q * 0.316206 + 3.87252
     int filt_guess = ROUND_POWER_OF_TWO(q * 20723 + 1015158, 18);
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index eee6ffe..a788c1d 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -28,11 +28,17 @@
 #include "vp9/encoder/vp9_ratectrl.h"
 #include "vp9/encoder/vp9_rd.h"
 
+typedef struct {
+  uint8_t *data;
+  int stride;
+  int in_use;
+} PRED_BUFFER;
+
 static int mv_refs_rt(const VP9_COMMON *cm, const MACROBLOCKD *xd,
-                       const TileInfo *const tile,
-                       MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
-                       int_mv *mv_ref_list,
-                       int mi_row, int mi_col) {
+                      const TileInfo *const tile,
+                      MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
+                      int_mv *mv_ref_list,
+                      int mi_row, int mi_col) {
   const int *ref_sign_bias = cm->ref_frame_sign_bias;
   int i, refmv_count = 0;
 
@@ -52,7 +58,7 @@
     const POSITION *const mv_ref = &mv_ref_search[i];
     if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
       const MODE_INFO *const candidate_mi = xd->mi[mv_ref->col + mv_ref->row *
-                                                   xd->mi_stride];
+                                                   xd->mi_stride].src_mi;
       const MB_MODE_INFO *const candidate = &candidate_mi->mbmi;
       // Keep counts for entropy encoding.
       context_counter += mode_2_counter[candidate->mode];
@@ -72,7 +78,7 @@
     const POSITION *const mv_ref = &mv_ref_search[i];
     if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
       const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row *
-                                                    xd->mi_stride]->mbmi;
+                                                    xd->mi_stride].src_mi->mbmi;
       different_ref_found = 1;
 
       if (candidate->ref_frame[0] == ref_frame)
@@ -88,7 +94,7 @@
       const POSITION *mv_ref = &mv_ref_search[i];
       if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
         const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row
-                                              * xd->mi_stride]->mbmi;
+                                              * xd->mi_stride].src_mi->mbmi;
 
         // If the candidate is INTRA we don't want to consider its mv.
         IF_DIFF_REF_FRAME_ADD_MV(candidate);
@@ -112,7 +118,7 @@
                                   int_mv *tmp_mv, int *rate_mv,
                                   int64_t best_rd_sofar) {
   MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0, 0}};
   const int step_param = cpi->sf.mv.fullpel_search_step_param;
   const int sadpb = x->sadperbit16;
@@ -224,13 +230,15 @@
 
   if (cpi->common.tx_mode == TX_MODE_SELECT) {
     if (sse > (var << 2))
-      xd->mi[0]->mbmi.tx_size = MIN(max_txsize_lookup[bsize],
-                          tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+      xd->mi[0].src_mi->mbmi.tx_size =
+          MIN(max_txsize_lookup[bsize],
+              tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
     else
-      xd->mi[0]->mbmi.tx_size = TX_8X8;
+      xd->mi[0].src_mi->mbmi.tx_size = TX_8X8;
   } else {
-    xd->mi[0]->mbmi.tx_size = MIN(max_txsize_lookup[bsize],
-                         tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+    xd->mi[0].src_mi->mbmi.tx_size =
+        MIN(max_txsize_lookup[bsize],
+            tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
   }
 
   vp9_model_rd_from_var_lapndz(sse - var, 1 << num_pels_log2_lookup[bsize],
@@ -269,7 +277,7 @@
                                  struct buf_2d yv12_mb[][MAX_MB_PLANE],
                                  int *rate, int64_t *dist) {
   MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
 
   const BLOCK_SIZE uv_size = get_plane_block_size(bsize, &xd->plane[1]);
   unsigned int var = var_y, sse = sse_y;
@@ -331,7 +339,7 @@
 
         // The cost of skip bit needs to be added.
         *rate = cpi->inter_mode_cost[mbmi->mode_context[ref_frame]]
-                                     [INTER_OFFSET(this_mode)];
+                                    [INTER_OFFSET(this_mode)];
 
         // More on this part of rate
         // rate += vp9_cost_bit(vp9_get_skip_prob(cm, xd), 1);
@@ -402,20 +410,21 @@
 
 // TODO(jingning) placeholder for inter-frame non-RD mode decision.
 // this needs various further optimizations. to be continued..
-int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
-                            const TileInfo *const tile,
-                            int mi_row, int mi_col,
-                            int *returnrate,
-                            int64_t *returndistortion,
-                            BLOCK_SIZE bsize,
-                            PICK_MODE_CONTEXT *ctx) {
-  MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
+                         const TileInfo *const tile,
+                         int mi_row, int mi_col,
+                         int *returnrate,
+                         int64_t *returndistortion,
+                         BLOCK_SIZE bsize,
+                         PICK_MODE_CONTEXT *ctx) {
+  VP9_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   struct macroblockd_plane *const pd = &xd->plane[0];
-  PREDICTION_MODE this_mode, best_mode = ZEROMV;
+  PREDICTION_MODE best_mode = ZEROMV;
   MV_REFERENCE_FRAME ref_frame, best_ref_frame = LAST_FRAME;
   TX_SIZE best_tx_size = MIN(max_txsize_lookup[bsize],
-                             tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+                             tx_mode_to_biggest_tx_size[cm->tx_mode]);
   INTERP_FILTER best_pred_filter = EIGHTTAP;
   int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
   struct buf_2d yv12_mb[4][MAX_MB_PLANE];
@@ -430,26 +439,23 @@
   unsigned int var_y = UINT_MAX;
   unsigned int sse_y = UINT_MAX;
 
-  VP9_COMMON *cm = &cpi->common;
-  int intra_cost_penalty = 20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q);
-
+  const int intra_cost_penalty =
+      20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q, cm->bit_depth);
   const int64_t inter_mode_thresh = RDCOST(x->rdmult, x->rddiv,
                                            intra_cost_penalty, 0);
-  const int64_t intra_mode_cost = 50;
+  const int intra_mode_cost = 50;
 
-  unsigned char segment_id = mbmi->segment_id;
+  const int8_t segment_id = mbmi->segment_id;
   const int *const rd_threshes = cpi->rd.threshes[segment_id][bsize];
   const int *const rd_thresh_freq_fact = cpi->rd.thresh_freq_fact[bsize];
-  // Mode index conversion form THR_MODES to PREDICTION_MODE for a ref frame.
   INTERP_FILTER filter_ref = cm->interp_filter;
-  int bsl = mi_width_log2(bsize);
+  const int bsl = mi_width_log2(bsize);
   const int pred_filter_search = cm->interp_filter == SWITCHABLE ?
       (((mi_row + mi_col) >> bsl) +
        get_chessboard_index(cm->current_video_frame)) & 0x1 : 0;
   int const_motion[MAX_REF_FRAMES] = { 0 };
   const int bh = num_4x4_blocks_high_lookup[bsize] << 2;
   const int bw = num_4x4_blocks_wide_lookup[bsize] << 2;
-  const int pixels_in_block = bh * bw;
   // For speed 6, the result of interp filter is reused later in actual encoding
   // process.
   // tmp[3] points to dst buffer, and the other 3 point to allocated buffers.
@@ -458,15 +464,11 @@
   struct buf_2d orig_dst = pd->dst;
   PRED_BUFFER *best_pred = NULL;
   PRED_BUFFER *this_mode_pred = NULL;
-  int i;
 
-  // CTX is used by the temporal denoiser which is currently being developed.
-  // TODO(jbb): when temporal denoiser is finished and in the default build
-  // remove the following line;
-  (void) ctx;
   if (cpi->sf.reuse_inter_pred_sby) {
+    int i;
     for (i = 0; i < 3; i++) {
-      tmp[i].data = &pred_buf[pixels_in_block * i];
+      tmp[i].data = &pred_buf[bw * bh * i];
       tmp[i].stride = bw;
       tmp[i].in_use = 0;
     }
@@ -492,14 +494,15 @@
   mbmi->segment_id = segment_id;
 
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
+    PREDICTION_MODE this_mode;
     x->pred_mv_sad[ref_frame] = INT_MAX;
     frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
     frame_mv[ZEROMV][ref_frame].as_int = 0;
 
     if (xd->up_available)
-      filter_ref = xd->mi[-xd->mi_stride]->mbmi.interp_filter;
+      filter_ref = xd->mi[-xd->mi_stride].src_mi->mbmi.interp_filter;
     else if (xd->left_available)
-      filter_ref = xd->mi[-1]->mbmi.interp_filter;
+      filter_ref = xd->mi[-1].src_mi->mbmi.interp_filter;
 
     if (cpi->ref_frame_flags & flag_list[ref_frame]) {
       const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
@@ -509,10 +512,10 @@
                            sf, sf);
 
       if (!cm->error_resilient_mode)
-        vp9_find_mv_refs(cm, xd, tile, xd->mi[0], ref_frame,
+        vp9_find_mv_refs(cm, xd, tile, xd->mi[0].src_mi, ref_frame,
                          candidates, mi_row, mi_col);
       else
-        const_motion[ref_frame] = mv_refs_rt(cm, xd, tile, xd->mi[0],
+        const_motion[ref_frame] = mv_refs_rt(cm, xd, tile, xd->mi[0].src_mi,
                                              ref_frame, candidates,
                                              mi_row, mi_col);
 
@@ -546,8 +549,9 @@
       if (!(cpi->sf.inter_mode_mask[bsize] & (1 << this_mode)))
         continue;
 
-      mode_rd_thresh =  rd_threshes[mode_idx[ref_frame - LAST_FRAME]
-                                    [INTER_OFFSET(this_mode)]];
+      mode_rd_thresh =
+          rd_threshes[mode_idx[ref_frame -
+                               LAST_FRAME][INTER_OFFSET(this_mode)]];
       if (rd_less_than_thresh(best_rd, mode_rd_thresh,
                               rd_thresh_freq_fact[this_mode]))
         continue;
@@ -564,7 +568,7 @@
       if (this_mode != NEARESTMV &&
           frame_mv[this_mode][ref_frame].as_int ==
               frame_mv[NEARESTMV][ref_frame].as_int)
-          continue;
+        continue;
 
       mbmi->mode = this_mode;
       mbmi->mv[0].as_int = frame_mv[this_mode][ref_frame].as_int;
@@ -643,7 +647,7 @@
 
       rate += rate_mv;
       rate += cpi->inter_mode_cost[mbmi->mode_context[ref_frame]]
-                                [INTER_OFFSET(this_mode)];
+                                  [INTER_OFFSET(this_mode)];
       this_rd = RDCOST(x->rdmult, x->rddiv, rate, dist);
 
       // Skipping checking: test to see if this block can be reconstructed by
@@ -661,6 +665,8 @@
       if (cpi->oxcf.noise_sensitivity > 0) {
         vp9_denoiser_update_frame_stats(mbmi, sse_y, this_mode, ctx);
       }
+#else
+      (void)ctx;
 #endif
 
       if (this_rd < best_rd || x->skip) {
@@ -696,15 +702,9 @@
   // temp buf to dst buf.
   if (best_pred != NULL && cpi->sf.reuse_inter_pred_sby &&
       best_pred->data != orig_dst.buf) {
-    uint8_t *copy_from, *copy_to;
-
     pd->dst = orig_dst;
-    copy_to = pd->dst.buf;
-
-    copy_from = best_pred->data;
-
-    vp9_convolve_copy(copy_from, bw, copy_to, pd->dst.stride, NULL, 0, NULL, 0,
-                      bw, bh);
+    vp9_convolve_copy(best_pred->data, bw, pd->dst.buf, pd->dst.stride, NULL, 0,
+                      NULL, 0, bw, bh);
   }
 
   mbmi->mode          = best_mode;
@@ -712,15 +712,15 @@
   mbmi->tx_size       = best_tx_size;
   mbmi->ref_frame[0]  = best_ref_frame;
   mbmi->mv[0].as_int  = frame_mv[best_mode][best_ref_frame].as_int;
-  xd->mi[0]->bmi[0].as_mv[0].as_int = mbmi->mv[0].as_int;
+  xd->mi[0].src_mi->bmi[0].as_mv[0].as_int = mbmi->mv[0].as_int;
   x->skip_txfm[0] = skip_txfm;
 
   // Perform intra prediction search, if the best SAD is above a certain
   // threshold.
   if (!x->skip && best_rd > inter_mode_thresh &&
       bsize <= cpi->sf.max_intra_bsize) {
+    PREDICTION_MODE this_mode;
     struct estimate_block_intra_args args = { cpi, x, DC_PRED, 0, 0 };
-
     const TX_SIZE intra_tx_size =
         MIN(max_txsize_lookup[bsize],
             tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
@@ -761,6 +761,4 @@
     if (cpi->sf.reuse_inter_pred_sby)
       pd->dst = orig_dst;
   }
-
-  return INT64_MAX;
 }
diff --git a/vp9/encoder/vp9_pickmode.h b/vp9/encoder/vp9_pickmode.h
index 49c6feb..97aeca7 100644
--- a/vp9/encoder/vp9_pickmode.h
+++ b/vp9/encoder/vp9_pickmode.h
@@ -17,19 +17,13 @@
 extern "C" {
 #endif
 
-typedef struct {
-  uint8_t *data;
-  int stride;
-  int in_use;
-} PRED_BUFFER;
-
-int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
-                            const struct TileInfo *const tile,
-                            int mi_row, int mi_col,
-                            int *returnrate,
-                            int64_t *returndistortion,
-                            BLOCK_SIZE bsize,
-                            PICK_MODE_CONTEXT *ctx);
+void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
+                         const struct TileInfo *const tile,
+                         int mi_row, int mi_col,
+                         int *returnrate,
+                         int64_t *returndistortion,
+                         BLOCK_SIZE bsize,
+                         PICK_MODE_CONTEXT *ctx);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/vp9/encoder/vp9_quantize.c b/vp9/encoder/vp9_quantize.c
index eababdb..3d2c409 100644
--- a/vp9/encoder/vp9_quantize.c
+++ b/vp9/encoder/vp9_quantize.c
@@ -19,9 +19,9 @@
 #include "vp9/encoder/vp9_quantize.h"
 #include "vp9/encoder/vp9_rd.h"
 
-void vp9_quantize_dc(const int16_t *coeff_ptr, int skip_block,
+void vp9_quantize_dc(const tran_low_t *coeff_ptr, int skip_block,
                      const int16_t *round_ptr, const int16_t quant,
-                     int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                     tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                      const int16_t dequant_ptr, uint16_t *eob_ptr) {
   const int rc = 0;
   const int coeff = coeff_ptr[rc];
@@ -40,9 +40,34 @@
   *eob_ptr = eob + 1;
 }
 
-void vp9_quantize_dc_32x32(const int16_t *coeff_ptr, int skip_block,
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_dc(const tran_low_t *coeff_ptr, int skip_block,
+                          const int16_t *round_ptr, const int16_t quant,
+                          tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                          const int16_t dequant_ptr, uint16_t *eob_ptr) {
+  int eob = -1;
+
+  if (!skip_block) {
+    const int rc = 0;
+    const int coeff = coeff_ptr[rc];
+    const int coeff_sign = (coeff >> 31);
+    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+    const int64_t tmp =
+        (clamp(abs_coeff + round_ptr[rc != 0], INT32_MIN, INT32_MAX) *
+         quant) >> 16;
+    qcoeff_ptr[rc]  = (tmp ^ coeff_sign) - coeff_sign;
+    dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr;
+    if (tmp)
+      eob = 0;
+  }
+  *eob_ptr = eob + 1;
+}
+#endif
+
+void vp9_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
                            const int16_t *round_ptr, const int16_t quant,
-                           int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                           tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                            const int16_t dequant_ptr, uint16_t *eob_ptr) {
   const int rc = 0;
   const int coeff = coeff_ptr[rc];
@@ -62,11 +87,36 @@
   *eob_ptr = eob + 1;
 }
 
-void vp9_quantize_fp_c(const int16_t *coeff_ptr, intptr_t count,
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
+                                const int16_t *round_ptr, const int16_t quant,
+                                tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                                const int16_t dequant_ptr, uint16_t *eob_ptr) {
+  int eob = -1;
+
+  if (!skip_block) {
+    const int rc = 0;
+    const int coeff = coeff_ptr[rc];
+    const int coeff_sign = (coeff >> 31);
+    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+    const int64_t tmp =
+        (clamp(abs_coeff + round_ptr[rc != 0], INT32_MIN, INT32_MAX) *
+         quant) >> 15;
+    qcoeff_ptr[rc]  = (tmp ^ coeff_sign) - coeff_sign;
+    dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr / 2;
+    if (tmp)
+      eob = 0;
+  }
+  *eob_ptr = eob + 1;
+}
+#endif
+
+void vp9_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                        int skip_block,
                        const int16_t *zbin_ptr, const int16_t *round_ptr,
                        const int16_t *quant_ptr, const int16_t *quant_shift_ptr,
-                       int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                       tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                        const int16_t *dequant_ptr,
                        int zbin_oq_value, uint16_t *eob_ptr,
                        const int16_t *scan, const int16_t *iscan) {
@@ -78,13 +128,13 @@
   (void)zbin_oq_value;
   (void)iscan;
 
-  vpx_memset(qcoeff_ptr, 0, count * sizeof(int16_t));
-  vpx_memset(dqcoeff_ptr, 0, count * sizeof(int16_t));
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
 
   if (!skip_block) {
     // Quantization pass: All coefficients with index >= zero_flag are
     // skippable. Note: zero_flag can be zero.
-    for (i = 0; i < count; i++) {
+    for (i = 0; i < n_coeffs; i++) {
       const int rc = scan[i];
       const int coeff = coeff_ptr[rc];
       const int coeff_sign = (coeff >> 31);
@@ -103,14 +153,59 @@
   *eob_ptr = eob + 1;
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t count,
+                            int skip_block, const int16_t *zbin_ptr,
+                            const int16_t *round_ptr, const int16_t *quant_ptr,
+                            const int16_t *quant_shift_ptr,
+                            tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                            const int16_t *dequant_ptr,
+                            int zbin_oq_value, uint16_t *eob_ptr,
+                            const int16_t *scan, const int16_t *iscan) {
+  int i;
+  int eob = -1;
+  // TODO(jingning) Decide the need of these arguments after the
+  // quantization process is completed.
+  (void)zbin_ptr;
+  (void)quant_shift_ptr;
+  (void)zbin_oq_value;
+  (void)iscan;
+
+  vpx_memset(qcoeff_ptr, 0, count * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, count * sizeof(*dqcoeff_ptr));
+
+  if (!skip_block) {
+    // Quantization pass: All coefficients with index >= zero_flag are
+    // skippable. Note: zero_flag can be zero.
+    for (i = 0; i < count; i++) {
+      const int rc = scan[i];
+      const int coeff = coeff_ptr[rc];
+      const int coeff_sign = (coeff >> 31);
+      const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+      const int64_t tmp =
+          (clamp(abs_coeff + round_ptr[rc != 0], INT32_MIN, INT32_MAX) *
+           quant_ptr[rc != 0]) >> 16;
+
+      qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
+      dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
+
+      if (tmp)
+        eob = i;
+    }
+  }
+  *eob_ptr = eob + 1;
+}
+#endif
+
 // TODO(jingning) Refactor this file and combine functions with similar
 // operations.
-void vp9_quantize_fp_32x32_c(const int16_t *coeff_ptr, intptr_t n_coeffs,
+void vp9_quantize_fp_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                              int skip_block,
                              const int16_t *zbin_ptr, const int16_t *round_ptr,
                              const int16_t *quant_ptr,
                              const int16_t *quant_shift_ptr,
-                             int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                             tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                              const int16_t *dequant_ptr,
                              int zbin_oq_value, uint16_t *eob_ptr,
                              const int16_t *scan, const int16_t *iscan) {
@@ -120,8 +215,8 @@
   (void)zbin_oq_value;
   (void)iscan;
 
-  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(int16_t));
-  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(int16_t));
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
 
   if (!skip_block) {
     for (i = 0; i < n_coeffs; i++) {
@@ -146,27 +241,72 @@
   *eob_ptr = eob + 1;
 }
 
-void vp9_quantize_b_c(const int16_t *coeff_ptr, intptr_t count,
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_fp_32x32_c(const tran_low_t *coeff_ptr,
+                                  intptr_t n_coeffs, int skip_block,
+                                  const int16_t *zbin_ptr,
+                                  const int16_t *round_ptr,
+                                  const int16_t *quant_ptr,
+                                  const int16_t *quant_shift_ptr,
+                                  tran_low_t *qcoeff_ptr,
+                                  tran_low_t *dqcoeff_ptr,
+                                  const int16_t *dequant_ptr,
+                                  int zbin_oq_value, uint16_t *eob_ptr,
+                                  const int16_t *scan, const int16_t *iscan) {
+  int i, eob = -1;
+  (void)zbin_ptr;
+  (void)quant_shift_ptr;
+  (void)zbin_oq_value;
+  (void)iscan;
+
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
+
+  if (!skip_block) {
+    for (i = 0; i < n_coeffs; i++) {
+      const int rc = scan[i];
+      const int coeff = coeff_ptr[rc];
+      const int coeff_sign = (coeff >> 31);
+      int64_t tmp = 0;
+      const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+      if (abs_coeff >= (dequant_ptr[rc != 0] >> 2)) {
+        tmp = clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
+                    INT32_MIN, INT32_MAX);
+        tmp = (tmp * quant_ptr[rc != 0]) >> 15;
+        qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
+        dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
+      }
+
+      if (tmp)
+        eob = i;
+    }
+  }
+  *eob_ptr = eob + 1;
+}
+#endif
+
+void vp9_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                       int skip_block,
                       const int16_t *zbin_ptr, const int16_t *round_ptr,
                       const int16_t *quant_ptr, const int16_t *quant_shift_ptr,
-                      int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                      tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                       const int16_t *dequant_ptr,
                       int zbin_oq_value, uint16_t *eob_ptr,
                       const int16_t *scan, const int16_t *iscan) {
-  int i, non_zero_count = (int)count, eob = -1;
+  int i, non_zero_count = (int)n_coeffs, eob = -1;
   const int zbins[2] = { zbin_ptr[0] + zbin_oq_value,
                          zbin_ptr[1] + zbin_oq_value };
   const int nzbins[2] = { zbins[0] * -1,
                           zbins[1] * -1 };
   (void)iscan;
 
-  vpx_memset(qcoeff_ptr, 0, count * sizeof(int16_t));
-  vpx_memset(dqcoeff_ptr, 0, count * sizeof(int16_t));
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
 
   if (!skip_block) {
     // Pre-scan pass
-    for (i = (int)count - 1; i >= 0; i--) {
+    for (i = (int)n_coeffs - 1; i >= 0; i--) {
       const int rc = scan[i];
       const int coeff = coeff_ptr[rc];
 
@@ -199,12 +339,68 @@
   *eob_ptr = eob + 1;
 }
 
-void vp9_quantize_b_32x32_c(const int16_t *coeff_ptr, intptr_t n_coeffs,
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
+                           int skip_block, const int16_t *zbin_ptr,
+                           const int16_t *round_ptr, const int16_t *quant_ptr,
+                           const int16_t *quant_shift_ptr,
+                           tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                           const int16_t *dequant_ptr, int zbin_oq_value,
+                           uint16_t *eob_ptr, const int16_t *scan,
+                           const int16_t *iscan) {
+  int i, non_zero_count = (int)n_coeffs, eob = -1;
+  const int zbins[2] = { zbin_ptr[0] + zbin_oq_value,
+                         zbin_ptr[1] + zbin_oq_value };
+  const int nzbins[2] = { zbins[0] * -1,
+                          zbins[1] * -1 };
+  (void)iscan;
+
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
+
+  if (!skip_block) {
+    // Pre-scan pass
+    for (i = (int)n_coeffs - 1; i >= 0; i--) {
+      const int rc = scan[i];
+      const int coeff = coeff_ptr[rc];
+
+      if (coeff < zbins[rc != 0] && coeff > nzbins[rc != 0])
+        non_zero_count--;
+      else
+        break;
+    }
+
+    // Quantization pass: All coefficients with index >= zero_flag are
+    // skippable. Note: zero_flag can be zero.
+    for (i = 0; i < non_zero_count; i++) {
+      const int rc = scan[i];
+      const int coeff = coeff_ptr[rc];
+      const int coeff_sign = (coeff >> 31);
+      const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+      if (abs_coeff >= zbins[rc != 0]) {
+        int64_t tmp = clamp(abs_coeff + round_ptr[rc != 0],
+                            INT32_MIN, INT32_MAX);
+        tmp = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+                  quant_shift_ptr[rc != 0]) >> 16;  // quantization
+        qcoeff_ptr[rc]  = (tmp ^ coeff_sign) - coeff_sign;
+        dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
+
+        if (tmp)
+          eob = i;
+      }
+    }
+  }
+  *eob_ptr = eob + 1;
+}
+#endif
+
+void vp9_quantize_b_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                             int skip_block,
                             const int16_t *zbin_ptr, const int16_t *round_ptr,
                             const int16_t *quant_ptr,
                             const int16_t *quant_shift_ptr,
-                            int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                            tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                             const int16_t *dequant_ptr,
                             int zbin_oq_value, uint16_t *eob_ptr,
                             const int16_t *scan, const int16_t *iscan) {
@@ -217,8 +413,8 @@
   int i, eob = -1;
   (void)iscan;
 
-  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(int16_t));
-  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(int16_t));
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
 
   if (!skip_block) {
     // Pre-scan pass
@@ -255,12 +451,84 @@
   *eob_ptr = eob + 1;
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_b_32x32_c(const tran_low_t *coeff_ptr,
+                                 intptr_t n_coeffs, int skip_block,
+                                 const int16_t *zbin_ptr,
+                                 const int16_t *round_ptr,
+                                 const int16_t *quant_ptr,
+                                 const int16_t *quant_shift_ptr,
+                                 tran_low_t *qcoeff_ptr,
+                                 tran_low_t *dqcoeff_ptr,
+                                 const int16_t *dequant_ptr,
+                                 int zbin_oq_value, uint16_t *eob_ptr,
+                                 const int16_t *scan, const int16_t *iscan) {
+  const int zbins[2] = { ROUND_POWER_OF_TWO(zbin_ptr[0] + zbin_oq_value, 1),
+                         ROUND_POWER_OF_TWO(zbin_ptr[1] + zbin_oq_value, 1) };
+  const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
+
+  int idx = 0;
+  int idx_arr[1024];
+  int i, eob = -1;
+  (void)iscan;
+
+  vpx_memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  vpx_memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
+
+  if (!skip_block) {
+    // Pre-scan pass
+    for (i = 0; i < n_coeffs; i++) {
+      const int rc = scan[i];
+      const int coeff = coeff_ptr[rc];
+
+      // If the coefficient is out of the base ZBIN range, keep it for
+      // quantization.
+      if (coeff >= zbins[rc != 0] || coeff <= nzbins[rc != 0])
+        idx_arr[idx++] = i;
+    }
+
+    // Quantization pass: only process the coefficients selected in
+    // pre-scan pass. Note: idx can be zero.
+    for (i = 0; i < idx; i++) {
+      const int rc = scan[idx_arr[i]];
+      const int coeff = coeff_ptr[rc];
+      const int coeff_sign = (coeff >> 31);
+      const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+      int64_t tmp = clamp(abs_coeff +
+                          ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
+                          INT32_MIN, INT32_MAX);
+      tmp = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+               quant_shift_ptr[rc != 0]) >> 15;
+
+      qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
+      dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
+
+      if (tmp)
+        eob = idx_arr[i];
+    }
+  }
+  *eob_ptr = eob + 1;
+}
+#endif
+
 void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int plane, int block,
                                 const int16_t *scan, const int16_t *iscan) {
   MACROBLOCKD *const xd = &x->e_mbd;
   struct macroblock_plane *p = &x->plane[plane];
   struct macroblockd_plane *pd = &xd->plane[plane];
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    vp9_high_quantize_b(BLOCK_OFFSET(p->coeff, block),
+                        16, x->skip_block,
+                        p->zbin, p->round, p->quant, p->quant_shift,
+                        BLOCK_OFFSET(p->qcoeff, block),
+                        BLOCK_OFFSET(pd->dqcoeff, block),
+                        pd->dequant, p->zbin_extra, &p->eobs[block],
+                        scan, iscan);
+    return;
+  }
+#endif
   vp9_quantize_b(BLOCK_OFFSET(p->coeff, block),
            16, x->skip_block,
            p->zbin, p->round, p->quant, p->quant_shift,
@@ -280,13 +548,33 @@
   *shift = 1 << (16 - l);
 }
 
+static int get_qzbin_factor(int q, vpx_bit_depth_t bit_depth) {
+  const int quant = vp9_dc_quant(q, 0, bit_depth);
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      return q == 0 ? 64 : (quant < 148 ? 84 : 80);
+    case VPX_BITS_10:
+      return q == 0 ? 64 : (quant < 592 ? 84 : 80);
+    case VPX_BITS_12:
+      return q == 0 ? 64 : (quant < 2368 ? 84 : 80);
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
+  (void) bit_depth;
+  return q == 0 ? 64 : (quant < 148 ? 84 : 80);
+#endif
+}
+
 void vp9_init_quantizer(VP9_COMP *cpi) {
   VP9_COMMON *const cm = &cpi->common;
   QUANTS *const quants = &cpi->quants;
   int i, q, quant;
 
   for (q = 0; q < QINDEX_RANGE; q++) {
-    const int qzbin_factor = q == 0 ? 64 : (vp9_dc_quant(q, 0) < 148 ? 84 : 80);
+    const int qzbin_factor = get_qzbin_factor(q, cm->bit_depth);
     const int qrounding_factor = q == 0 ? 64 : 48;
 
     for (i = 0; i < 2; ++i) {
@@ -295,8 +583,8 @@
         qrounding_factor_fp = 64;
 
       // y
-      quant = i == 0 ? vp9_dc_quant(q, cm->y_dc_delta_q)
-                     : vp9_ac_quant(q, 0);
+      quant = i == 0 ? vp9_dc_quant(q, cm->y_dc_delta_q, cm->bit_depth)
+                     : vp9_ac_quant(q, 0, cm->bit_depth);
       invert_quant(&quants->y_quant[q][i], &quants->y_quant_shift[q][i], quant);
       quants->y_quant_fp[q][i] = (1 << 16) / quant;
       quants->y_round_fp[q][i] = (qrounding_factor_fp * quant) >> 7;
@@ -305,8 +593,8 @@
       cm->y_dequant[q][i] = quant;
 
       // uv
-      quant = i == 0 ? vp9_dc_quant(q, cm->uv_dc_delta_q)
-                     : vp9_ac_quant(q, cm->uv_ac_delta_q);
+      quant = i == 0 ? vp9_dc_quant(q, cm->uv_dc_delta_q, cm->bit_depth)
+                     : vp9_ac_quant(q, cm->uv_ac_delta_q, cm->bit_depth);
       invert_quant(&quants->uv_quant[q][i],
                    &quants->uv_quant_shift[q][i], quant);
       quants->uv_quant_fp[q][i] = (1 << 16) / quant;
@@ -340,7 +628,7 @@
   const VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &x->e_mbd;
   QUANTS *const quants = &cpi->quants;
-  const int segment_id = xd->mi[0]->mbmi.segment_id;
+  const int segment_id = xd->mi[0].src_mi->mbmi.segment_id;
   const int qindex = vp9_get_qindex(&cm->seg, segment_id, cm->base_qindex);
   const int rdmult = vp9_compute_rd_mult(cpi, qindex + cm->y_dc_delta_q);
   const int zbin = cpi->zbin_mode_boost;
diff --git a/vp9/encoder/vp9_quantize.h b/vp9/encoder/vp9_quantize.h
index 262529b..d7edb0b 100644
--- a/vp9/encoder/vp9_quantize.h
+++ b/vp9/encoder/vp9_quantize.h
@@ -37,17 +37,29 @@
   DECLARE_ALIGNED(16, int16_t, uv_round[QINDEX_RANGE][8]);
 } QUANTS;
 
-void vp9_quantize_dc(const int16_t *coeff_ptr, int skip_block,
+void vp9_quantize_dc(const tran_low_t *coeff_ptr, int skip_block,
                      const int16_t *round_ptr, const int16_t quant_ptr,
-                     int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                     tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                      const int16_t dequant_ptr, uint16_t *eob_ptr);
-void vp9_quantize_dc_32x32(const int16_t *coeff_ptr, int skip_block,
+void vp9_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
                            const int16_t *round_ptr, const int16_t quant_ptr,
-                           int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
+                           tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                            const int16_t dequant_ptr, uint16_t *eob_ptr);
 void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int plane, int block,
                                 const int16_t *scan, const int16_t *iscan);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_quantize_dc(const tran_low_t *coeff_ptr, int skip_block,
+                          const int16_t *round_ptr, const int16_t quant_ptr,
+                          tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                          const int16_t dequant_ptr, uint16_t *eob_ptr);
+void vp9_high_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
+                                const int16_t *round_ptr,
+                                const int16_t quant_ptr, tran_low_t *qcoeff_ptr,
+                                tran_low_t *dqcoeff_ptr,
+                                const int16_t dequant_ptr, uint16_t *eob_ptr);
+#endif
+
 struct VP9_COMP;
 struct VP9Common;
 
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index b607c85..9b6c773 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -42,13 +42,56 @@
 
 #define FRAME_OVERHEAD_BITS 200
 
+#if CONFIG_VP9_HIGHBITDEPTH
+#define ASSIGN_MINQ_TABLE(bit_depth, name) \
+  do { \
+    switch (bit_depth) { \
+      case VPX_BITS_8: \
+        name = name##_8; \
+        break; \
+      case VPX_BITS_10: \
+        name = name##_10; \
+        break; \
+      case VPX_BITS_12: \
+        name = name##_12; \
+        break; \
+      default: \
+        assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10" \
+                    " or VPX_BITS_12"); \
+        name = NULL; \
+    } \
+  } while (0)
+#else
+#define ASSIGN_MINQ_TABLE(bit_depth, name) \
+  do { \
+    (void) bit_depth; \
+    name = name##_8; \
+  } while (0)
+#endif
+
 // Tables relating active max Q to active min Q
-static int kf_low_motion_minq[QINDEX_RANGE];
-static int kf_high_motion_minq[QINDEX_RANGE];
-static int arfgf_low_motion_minq[QINDEX_RANGE];
-static int arfgf_high_motion_minq[QINDEX_RANGE];
-static int inter_minq[QINDEX_RANGE];
-static int rtc_minq[QINDEX_RANGE];
+static int kf_low_motion_minq_8[QINDEX_RANGE];
+static int kf_high_motion_minq_8[QINDEX_RANGE];
+static int arfgf_low_motion_minq_8[QINDEX_RANGE];
+static int arfgf_high_motion_minq_8[QINDEX_RANGE];
+static int inter_minq_8[QINDEX_RANGE];
+static int rtc_minq_8[QINDEX_RANGE];
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static int kf_low_motion_minq_10[QINDEX_RANGE];
+static int kf_high_motion_minq_10[QINDEX_RANGE];
+static int arfgf_low_motion_minq_10[QINDEX_RANGE];
+static int arfgf_high_motion_minq_10[QINDEX_RANGE];
+static int inter_minq_10[QINDEX_RANGE];
+static int rtc_minq_10[QINDEX_RANGE];
+static int kf_low_motion_minq_12[QINDEX_RANGE];
+static int kf_high_motion_minq_12[QINDEX_RANGE];
+static int arfgf_low_motion_minq_12[QINDEX_RANGE];
+static int arfgf_high_motion_minq_12[QINDEX_RANGE];
+static int inter_minq_12[QINDEX_RANGE];
+static int rtc_minq_12[QINDEX_RANGE];
+#endif
+
 static int gf_high = 2000;
 static int gf_low = 400;
 static int kf_high = 5000;
@@ -58,7 +101,8 @@
 // formulaic approach to facilitate easier adjustment of the Q tables.
 // The formulae were derived from computing a 3rd order polynomial best
 // fit to the original data (after plotting real maxq vs minq (not q index))
-static int get_minq_index(double maxq, double x3, double x2, double x1) {
+static int get_minq_index(double maxq, double x3, double x2, double x1,
+                          vpx_bit_depth_t bit_depth) {
   int i;
   const double minqtarget = MIN(((x3 * maxq + x2) * maxq + x1) * maxq,
                                 maxq);
@@ -68,39 +112,70 @@
   if (minqtarget <= 2.0)
     return 0;
 
-  for (i = 0; i < QINDEX_RANGE; i++)
-    if (minqtarget <= vp9_convert_qindex_to_q(i))
+  for (i = 0; i < QINDEX_RANGE; i++) {
+    if (minqtarget <= vp9_convert_qindex_to_q(i, bit_depth))
       return i;
+  }
 
   return QINDEX_RANGE - 1;
 }
 
-void vp9_rc_init_minq_luts() {
+static void init_minq_luts(int *kf_low_m, int *kf_high_m,
+                           int *arfgf_low, int *arfgf_high,
+                           int *inter, int *rtc, vpx_bit_depth_t bit_depth) {
   int i;
-
   for (i = 0; i < QINDEX_RANGE; i++) {
-    const double maxq = vp9_convert_qindex_to_q(i);
-    kf_low_motion_minq[i] = get_minq_index(maxq, 0.000001, -0.0004, 0.125);
-    kf_high_motion_minq[i] = get_minq_index(maxq, 0.000002, -0.0012, 0.50);
-    arfgf_low_motion_minq[i] = get_minq_index(maxq, 0.0000015, -0.0009, 0.30);
-    arfgf_high_motion_minq[i] = get_minq_index(maxq, 0.0000021, -0.00125, 0.50);
-    inter_minq[i] = get_minq_index(maxq, 0.00000271, -0.00113, 0.90);
-    rtc_minq[i] = get_minq_index(maxq, 0.00000271, -0.00113, 0.70);
+    const double maxq = vp9_convert_qindex_to_q(i, bit_depth);
+    kf_low_m[i] = get_minq_index(maxq, 0.000001, -0.0004, 0.150, bit_depth);
+    kf_high_m[i] = get_minq_index(maxq, 0.0000021, -0.00125, 0.55, bit_depth);
+    arfgf_low[i] = get_minq_index(maxq, 0.0000015, -0.0009, 0.30, bit_depth);
+    arfgf_high[i] = get_minq_index(maxq, 0.0000021, -0.00125, 0.55, bit_depth);
+    inter[i] = get_minq_index(maxq, 0.00000271, -0.00113, 0.90, bit_depth);
+    rtc[i] = get_minq_index(maxq, 0.00000271, -0.00113, 0.70, bit_depth);
   }
 }
 
+void vp9_rc_init_minq_luts() {
+  init_minq_luts(kf_low_motion_minq_8, kf_high_motion_minq_8,
+                 arfgf_low_motion_minq_8, arfgf_high_motion_minq_8,
+                 inter_minq_8, rtc_minq_8, VPX_BITS_8);
+#if CONFIG_VP9_HIGHBITDEPTH
+  init_minq_luts(kf_low_motion_minq_10, kf_high_motion_minq_10,
+                 arfgf_low_motion_minq_10, arfgf_high_motion_minq_10,
+                 inter_minq_10, rtc_minq_10, VPX_BITS_10);
+  init_minq_luts(kf_low_motion_minq_12, kf_high_motion_minq_12,
+                 arfgf_low_motion_minq_12, arfgf_high_motion_minq_12,
+                 inter_minq_12, rtc_minq_12, VPX_BITS_12);
+#endif
+}
+
 // These functions use formulaic calculations to make playing with the
 // quantizer tables easier. If necessary they can be replaced by lookup
 // tables if and when things settle down in the experimental bitstream
-double vp9_convert_qindex_to_q(int qindex) {
+double vp9_convert_qindex_to_q(int qindex, vpx_bit_depth_t bit_depth) {
   // Convert the index to a real Q value (scaled down to match old Q values)
-  return vp9_ac_quant(qindex, 0) / 4.0;
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      return vp9_ac_quant(qindex, 0, bit_depth) / 4.0;
+    case VPX_BITS_10:
+      return vp9_ac_quant(qindex, 0, bit_depth) / 16.0;
+    case VPX_BITS_12:
+      return vp9_ac_quant(qindex, 0, bit_depth) / 64.0;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1.0;
+  }
+#else
+  return vp9_ac_quant(qindex, 0, bit_depth) / 4.0;
+#endif
 }
 
 int vp9_rc_bits_per_mb(FRAME_TYPE frame_type, int qindex,
-                       double correction_factor) {
-  const double q = vp9_convert_qindex_to_q(qindex);
-  int enumerator = frame_type == KEY_FRAME ? 3300000 : 2250000;
+                       double correction_factor,
+                       vpx_bit_depth_t bit_depth) {
+  const double q = vp9_convert_qindex_to_q(qindex, bit_depth);
+  int enumerator = frame_type == KEY_FRAME ? 2700000 : 1800000;
 
   // q based adjustment to baseline enumerator
   enumerator += (int)(enumerator * q) >> 12;
@@ -108,8 +183,10 @@
 }
 
 static int estimate_bits_at_q(FRAME_TYPE frame_type, int q, int mbs,
-                              double correction_factor) {
-  const int bpm = (int)(vp9_rc_bits_per_mb(frame_type, q, correction_factor));
+                              double correction_factor,
+                              vpx_bit_depth_t bit_depth) {
+  const int bpm = (int)(vp9_rc_bits_per_mb(frame_type, q, correction_factor,
+                                           bit_depth));
   return ((uint64_t)bpm * mbs) >> BPER_MB_NORMBITS;
 }
 
@@ -227,7 +304,7 @@
   rc->ni_frames = 0;
 
   rc->tot_q = 0.0;
-  rc->avg_q = vp9_convert_qindex_to_q(oxcf->worst_allowed_q);
+  rc->avg_q = vp9_convert_qindex_to_q(oxcf->worst_allowed_q, oxcf->bit_depth);
 
   for (i = 0; i < RATE_FACTOR_LEVELS; ++i) {
     rc->rate_correction_factors[i] = 1.0;
@@ -330,7 +407,8 @@
   // Stay in double to avoid int overflow when values are large
   projected_size_based_on_q = estimate_bits_at_q(cm->frame_type,
                                                  cm->base_qindex, cm->MBs,
-                                                 rate_correction_factor);
+                                                 rate_correction_factor,
+                                                 cm->bit_depth);
   // Work out a size correction factor.
   if (projected_size_based_on_q > 0)
     correction_factor = (100 * cpi->rc.projected_frame_size) /
@@ -392,7 +470,8 @@
 
   do {
     const int bits_per_mb_at_this_q = (int)vp9_rc_bits_per_mb(cm->frame_type, i,
-                                                             correction_factor);
+                                                              correction_factor,
+                                                              cm->bit_depth);
 
     if (bits_per_mb_at_this_q <= target_bits_per_mb) {
       if ((target_bits_per_mb - bits_per_mb_at_this_q) <= last_error)
@@ -424,12 +503,22 @@
   }
 }
 
-static int get_kf_active_quality(const RATE_CONTROL *const rc, int q) {
+static int get_kf_active_quality(const RATE_CONTROL *const rc, int q,
+                                 vpx_bit_depth_t bit_depth) {
+  int *kf_low_motion_minq;
+  int *kf_high_motion_minq;
+  ASSIGN_MINQ_TABLE(bit_depth, kf_low_motion_minq);
+  ASSIGN_MINQ_TABLE(bit_depth, kf_high_motion_minq);
   return get_active_quality(q, rc->kf_boost, kf_low, kf_high,
                             kf_low_motion_minq, kf_high_motion_minq);
 }
 
-static int get_gf_active_quality(const RATE_CONTROL *const rc, int q) {
+static int get_gf_active_quality(const RATE_CONTROL *const rc, int q,
+                                 vpx_bit_depth_t bit_depth) {
+  int *arfgf_low_motion_minq;
+  int *arfgf_high_motion_minq;
+  ASSIGN_MINQ_TABLE(bit_depth, arfgf_low_motion_minq);
+  ASSIGN_MINQ_TABLE(bit_depth, arfgf_high_motion_minq);
   return get_active_quality(q, rc->gfu_boost, gf_low, gf_high,
                             arfgf_low_motion_minq, arfgf_high_motion_minq);
 }
@@ -516,17 +605,20 @@
   int active_best_quality;
   int active_worst_quality = calc_active_worst_quality_one_pass_cbr(cpi);
   int q;
+  int *rtc_minq;
+  ASSIGN_MINQ_TABLE(cm->bit_depth, rtc_minq);
 
   if (frame_is_intra_only(cm)) {
     active_best_quality = rc->best_quality;
-    // Handle the special case for key frames forced when we have75 reached
+    // Handle the special case for key frames forced when we have reached
     // the maximum key frame interval. Here force the Q to a range
     // based on the ambient Q to reduce the risk of popping.
     if (rc->this_key_frame_forced) {
       int qindex = rc->last_boosted_qindex;
-      double last_boosted_q = vp9_convert_qindex_to_q(qindex);
+      double last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
       int delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
-                                            (last_boosted_q * 0.75));
+                                            (last_boosted_q * 0.75),
+                                            cm->bit_depth);
       active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
     } else if (cm->current_video_frame > 0) {
       // not first frame of one pass and kf_boost is set
@@ -534,7 +626,8 @@
       double q_val;
 
       active_best_quality =
-          get_kf_active_quality(rc, rc->avg_frame_qindex[KEY_FRAME]);
+          get_kf_active_quality(rc, rc->avg_frame_qindex[KEY_FRAME],
+                                cm->bit_depth);
 
       // Allow somewhat lower kf minq with small image formats.
       if ((cm->width * cm->height) <= (352 * 288)) {
@@ -543,9 +636,10 @@
 
       // Convert the adjustment factor to a qindex delta
       // on active_best_quality.
-      q_val = vp9_convert_qindex_to_q(active_best_quality);
+      q_val = vp9_convert_qindex_to_q(active_best_quality, cm->bit_depth);
       active_best_quality += vp9_compute_qdelta(rc, q_val,
-                                                q_val * q_adj_factor);
+                                                q_val * q_adj_factor,
+                                                cm->bit_depth);
     }
   } else if (!rc->is_src_frame_alt_ref &&
              !cpi->use_svc &&
@@ -559,7 +653,7 @@
     } else {
       q = active_worst_quality;
     }
-    active_best_quality = get_gf_active_quality(rc, q);
+    active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
   } else {
     // Use the lower of active_worst_quality and recent/average Q.
     if (cm->current_video_frame > 1) {
@@ -592,7 +686,8 @@
     int qdelta = 0;
     vp9_clear_system_state();
     qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
-                                        active_worst_quality, 2.0);
+                                        active_worst_quality, 2.0,
+                                        cm->bit_depth);
     *top_index = active_worst_quality + qdelta;
     *top_index = (*top_index > *bottom_index) ? *top_index : *bottom_index;
   }
@@ -644,6 +739,8 @@
   int active_best_quality;
   int active_worst_quality = calc_active_worst_quality_one_pass_vbr(cpi);
   int q;
+  int *inter_minq;
+  ASSIGN_MINQ_TABLE(cm->bit_depth, inter_minq);
 
   if (frame_is_intra_only(cm)) {
 
@@ -652,9 +749,10 @@
     // based on the ambient Q to reduce the risk of popping.
     if (rc->this_key_frame_forced) {
       int qindex = rc->last_boosted_qindex;
-      double last_boosted_q = vp9_convert_qindex_to_q(qindex);
+      double last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
       int delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
-                                            last_boosted_q * 0.75);
+                                            last_boosted_q * 0.75,
+                                            cm->bit_depth);
       active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
     } else {
       // not first frame of one pass and kf_boost is set
@@ -662,7 +760,8 @@
       double q_val;
 
       active_best_quality =
-          get_kf_active_quality(rc, rc->avg_frame_qindex[KEY_FRAME]);
+          get_kf_active_quality(rc, rc->avg_frame_qindex[KEY_FRAME],
+                                cm->bit_depth);
 
       // Allow somewhat lower kf minq with small image formats.
       if ((cm->width * cm->height) <= (352 * 288)) {
@@ -671,9 +770,10 @@
 
       // Convert the adjustment factor to a qindex delta
       // on active_best_quality.
-      q_val = vp9_convert_qindex_to_q(active_best_quality);
+      q_val = vp9_convert_qindex_to_q(active_best_quality, cm->bit_depth);
       active_best_quality += vp9_compute_qdelta(rc, q_val,
-                                                q_val * q_adj_factor);
+                                                q_val * q_adj_factor,
+                                                cm->bit_depth);
     }
   } else if (!rc->is_src_frame_alt_ref &&
              (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
@@ -691,7 +791,7 @@
       if (q < cq_level)
         q = cq_level;
 
-      active_best_quality = get_gf_active_quality(rc, q);
+      active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
 
       // Constrained quality use slightly lower active best.
       active_best_quality = active_best_quality * 15 / 16;
@@ -700,10 +800,10 @@
       if (!cpi->refresh_alt_ref_frame) {
         active_best_quality = cq_level;
       } else {
-        active_best_quality = get_gf_active_quality(rc, q);
+        active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
       }
     } else {
-      active_best_quality = get_gf_active_quality(rc, q);
+      active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
     }
   } else {
     if (oxcf->rc_mode == VPX_Q) {
@@ -742,11 +842,13 @@
         !rc->this_key_frame_forced &&
         !(cm->current_video_frame == 0)) {
       qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
-                                          active_worst_quality, 2.0);
+                                          active_worst_quality, 2.0,
+                                          cm->bit_depth);
     } else if (!rc->is_src_frame_alt_ref &&
                (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
       qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
-                                          active_worst_quality, 1.75);
+                                          active_worst_quality, 1.75,
+                                          cm->bit_depth);
     }
     *top_index = active_worst_quality + qdelta;
     *top_index = (*top_index > *bottom_index) ? *top_index : *bottom_index;
@@ -778,6 +880,7 @@
   return q;
 }
 
+#define STATIC_MOTION_THRESH 95
 static int rc_pick_q_and_bounds_two_pass(const VP9_COMP *cpi,
                                          int *bottom_index,
                                          int *top_index) {
@@ -788,23 +891,42 @@
   int active_best_quality;
   int active_worst_quality = cpi->twopass.active_worst_quality;
   int q;
+  int *inter_minq;
+  ASSIGN_MINQ_TABLE(cm->bit_depth, inter_minq);
 
   if (frame_is_intra_only(cm) || vp9_is_upper_layer_key_frame(cpi)) {
-    // Handle the special case for key frames forced when we have75 reached
+    // Handle the special case for key frames forced when we have reached
     // the maximum key frame interval. Here force the Q to a range
     // based on the ambient Q to reduce the risk of popping.
     if (rc->this_key_frame_forced) {
-      int qindex = rc->last_boosted_qindex;
-      double last_boosted_q = vp9_convert_qindex_to_q(qindex);
-      int delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
-                                            last_boosted_q * 0.75);
-      active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+      double last_boosted_q;
+      int delta_qindex;
+      int qindex;
+
+      if (cpi->twopass.last_kfgroup_zeromotion_pct >= STATIC_MOTION_THRESH) {
+        qindex = MIN(rc->last_kf_qindex, rc->last_boosted_qindex);
+        active_best_quality = qindex;
+        last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
+        delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
+                                              last_boosted_q * 1.25,
+                                              cm->bit_depth);
+        active_worst_quality = MIN(qindex + delta_qindex, active_worst_quality);
+
+      } else {
+        qindex = rc->last_boosted_qindex;
+        last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
+        delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
+                                              last_boosted_q * 0.75,
+                                              cm->bit_depth);
+        active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+      }
     } else {
       // Not forced keyframe.
       double q_adj_factor = 1.0;
       double q_val;
       // Baseline value derived from cpi->active_worst_quality and kf boost.
-      active_best_quality = get_kf_active_quality(rc, active_worst_quality);
+      active_best_quality = get_kf_active_quality(rc, active_worst_quality,
+                                                  cm->bit_depth);
 
       // Allow somewhat lower kf minq with small image formats.
       if ((cm->width * cm->height) <= (352 * 288)) {
@@ -816,9 +938,10 @@
 
       // Convert the adjustment factor to a qindex delta
       // on active_best_quality.
-      q_val = vp9_convert_qindex_to_q(active_best_quality);
+      q_val = vp9_convert_qindex_to_q(active_best_quality, cm->bit_depth);
       active_best_quality += vp9_compute_qdelta(rc, q_val,
-                                                q_val * q_adj_factor);
+                                                q_val * q_adj_factor,
+                                                cm->bit_depth);
     }
   } else if (!rc->is_src_frame_alt_ref &&
              (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
@@ -836,7 +959,7 @@
       if (q < cq_level)
         q = cq_level;
 
-      active_best_quality = get_gf_active_quality(rc, q);
+      active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
 
       // Constrained quality use slightly lower active best.
       active_best_quality = active_best_quality * 15 / 16;
@@ -845,10 +968,10 @@
       if (!cpi->refresh_alt_ref_frame) {
         active_best_quality = cq_level;
       } else {
-        active_best_quality = get_gf_active_quality(rc, q);
+        active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
       }
     } else {
-      active_best_quality = get_gf_active_quality(rc, q);
+      active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
     }
   } else {
     if (oxcf->rc_mode == VPX_Q) {
@@ -865,18 +988,12 @@
     }
   }
 
-  // Clip the active best and worst quality values to limits.
-  active_best_quality = clamp(active_best_quality,
-                              rc->best_quality, rc->worst_quality);
-  active_worst_quality = clamp(active_worst_quality,
-                               active_best_quality, rc->worst_quality);
-
-  *top_index = active_worst_quality;
-  *bottom_index = active_best_quality;
-
 #if LIMIT_QRANGE_FOR_ALTREF_AND_KEY
   vp9_clear_system_state();
-  {
+  // Static forced key frames Q restrictions dealt with elsewhere.
+  if (!((frame_is_intra_only(cm) || vp9_is_upper_layer_key_frame(cpi))) ||
+      !rc->this_key_frame_forced ||
+      (cpi->twopass.last_kfgroup_zeromotion_pct < STATIC_MOTION_THRESH)) {
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     const double rate_factor_deltas[RATE_FACTOR_LEVELS] = {
       1.00,  // INTER_NORMAL
@@ -888,28 +1005,45 @@
     const double rate_factor =
       rate_factor_deltas[gf_group->rf_level[gf_group->index]];
     int qdelta = vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
-                                            active_worst_quality, rate_factor);
-    *top_index = active_worst_quality + qdelta;
-    *top_index = (*top_index > *bottom_index) ? *top_index : *bottom_index;
+                                            active_worst_quality, rate_factor,
+                                            cm->bit_depth);
+    active_worst_quality = active_worst_quality + qdelta;
+    active_worst_quality = MAX(active_worst_quality, active_best_quality);
   }
 #endif
 
+  // Clip the active best and worst quality values to limits.
+  active_best_quality = clamp(active_best_quality,
+                              rc->best_quality, rc->worst_quality);
+  active_worst_quality = clamp(active_worst_quality,
+                               active_best_quality, rc->worst_quality);
+
   if (oxcf->rc_mode == VPX_Q) {
     q = active_best_quality;
   // Special case code to try and match quality with forced key frames.
-  } else if ((cm->frame_type == KEY_FRAME) && rc->this_key_frame_forced) {
-    q = rc->last_boosted_qindex;
+  } else if ((frame_is_intra_only(cm) || vp9_is_upper_layer_key_frame(cpi)) &&
+             rc->this_key_frame_forced) {
+    // If static since last kf use better of last boosted and last kf q.
+    if (cpi->twopass.last_kfgroup_zeromotion_pct >= STATIC_MOTION_THRESH) {
+      q = MIN(rc->last_kf_qindex, rc->last_boosted_qindex);
+    } else {
+      q = rc->last_boosted_qindex;
+    }
   } else {
     q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
                           active_best_quality, active_worst_quality);
-    if (q > *top_index) {
+    if (q > active_worst_quality) {
       // Special case when we are targeting the max allowed rate.
       if (rc->this_frame_target >= rc->max_frame_bandwidth)
-        *top_index = q;
+        active_worst_quality = q;
       else
-        q = *top_index;
+        q = active_worst_quality;
     }
   }
+  clamp(q, active_best_quality, active_worst_quality);
+
+  *top_index = active_worst_quality;
+  *bottom_index = active_best_quality;
 
   assert(*top_index <= rc->worst_quality &&
          *top_index >= rc->best_quality);
@@ -1038,7 +1172,7 @@
       rc->avg_frame_qindex[INTER_FRAME] =
         ROUND_POWER_OF_TWO(3 * rc->avg_frame_qindex[INTER_FRAME] + qindex, 2);
       rc->ni_frames++;
-      rc->tot_q += vp9_convert_qindex_to_q(qindex);
+      rc->tot_q += vp9_convert_qindex_to_q(qindex, cm->bit_depth);
       rc->avg_q = rc->tot_q / rc->ni_frames;
       // Calculate the average Q for normal inter frames (not key or GFU
       // frames).
@@ -1053,11 +1187,12 @@
   // better than that already stored.
   // This is used to help set quality in forced key frames to reduce popping
   if ((qindex < rc->last_boosted_qindex) ||
-      ((cpi->static_mb_pct < 100) &&
-       ((cm->frame_type == KEY_FRAME) || cpi->refresh_alt_ref_frame ||
+      (((cm->frame_type == KEY_FRAME) || cpi->refresh_alt_ref_frame ||
         (cpi->refresh_golden_frame && !rc->is_src_frame_alt_ref)))) {
     rc->last_boosted_qindex = qindex;
   }
+  if (cm->frame_type == KEY_FRAME)
+    rc->last_kf_qindex = qindex;
 
   update_buffer_level(cpi, rc->projected_frame_size);
 
@@ -1294,7 +1429,8 @@
   rc->baseline_gf_interval = INT_MAX;
 }
 
-int vp9_compute_qdelta(const RATE_CONTROL *rc, double qstart, double qtarget) {
+int vp9_compute_qdelta(const RATE_CONTROL *rc, double qstart, double qtarget,
+                       vpx_bit_depth_t bit_depth) {
   int start_index = rc->worst_quality;
   int target_index = rc->worst_quality;
   int i;
@@ -1302,14 +1438,14 @@
   // Convert the average q value to an index.
   for (i = rc->best_quality; i < rc->worst_quality; ++i) {
     start_index = i;
-    if (vp9_convert_qindex_to_q(i) >= qstart)
+    if (vp9_convert_qindex_to_q(i, bit_depth) >= qstart)
       break;
   }
 
   // Convert the q target to an index
   for (i = rc->best_quality; i < rc->worst_quality; ++i) {
     target_index = i;
-    if (vp9_convert_qindex_to_q(i) >= qtarget)
+    if (vp9_convert_qindex_to_q(i, bit_depth) >= qtarget)
       break;
   }
 
@@ -1317,12 +1453,14 @@
 }
 
 int vp9_compute_qdelta_by_rate(const RATE_CONTROL *rc, FRAME_TYPE frame_type,
-                               int qindex, double rate_target_ratio) {
+                               int qindex, double rate_target_ratio,
+                               vpx_bit_depth_t bit_depth) {
   int target_index = rc->worst_quality;
   int i;
 
   // Look up the current projected bits per block for the base index
-  const int base_bits_per_mb = vp9_rc_bits_per_mb(frame_type, qindex, 1.0);
+  const int base_bits_per_mb = vp9_rc_bits_per_mb(frame_type, qindex, 1.0,
+                                                  bit_depth);
 
   // Find the target bits per mb based on the base value and given ratio.
   const int target_bits_per_mb = (int)(rate_target_ratio * base_bits_per_mb);
@@ -1330,7 +1468,7 @@
   // Convert the q target to an index
   for (i = rc->best_quality; i < rc->worst_quality; ++i) {
     target_index = i;
-    if (vp9_rc_bits_per_mb(frame_type, i, 1.0) <= target_bits_per_mb )
+    if (vp9_rc_bits_per_mb(frame_type, i, 1.0, bit_depth) <= target_bits_per_mb)
       break;
   }
 
diff --git a/vp9/encoder/vp9_ratectrl.h b/vp9/encoder/vp9_ratectrl.h
index 456daf4..edfb9fc 100644
--- a/vp9/encoder/vp9_ratectrl.h
+++ b/vp9/encoder/vp9_ratectrl.h
@@ -12,6 +12,7 @@
 #ifndef VP9_ENCODER_VP9_RATECTRL_H_
 #define VP9_ENCODER_VP9_RATECTRL_H_
 
+#include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
 
 #include "vp9/common/vp9_blockd.h"
@@ -41,6 +42,7 @@
   int sb64_target_rate;
   int last_q[FRAME_TYPES];         // Separate values for Intra/Inter
   int last_boosted_qindex;         // Last boosted GF/KF/ARF q
+  int last_kf_qindex;              // Q index of the last key frame coded.
 
   int gfu_boost;
   int last_boost;
@@ -104,7 +106,7 @@
 void vp9_rc_init(const struct VP9EncoderConfig *oxcf, int pass,
                  RATE_CONTROL *rc);
 
-double vp9_convert_qindex_to_q(int qindex);
+double vp9_convert_qindex_to_q(int qindex, vpx_bit_depth_t bit_depth);
 
 void vp9_rc_init_minq_luts();
 
@@ -167,7 +169,7 @@
 
 // Estimates bits per mb for a given qindex and correction factor.
 int vp9_rc_bits_per_mb(FRAME_TYPE frame_type, int qindex,
-                       double correction_factor);
+                       double correction_factor, vpx_bit_depth_t bit_depth);
 
 // Clamping utilities for bitrate targets for iframes and pframes.
 int vp9_rc_clamp_iframe_target_size(const struct VP9_COMP *const cpi,
@@ -180,12 +182,14 @@
 
 // Computes a q delta (in "q index" terms) to get from a starting q value
 // to a target q value
-int vp9_compute_qdelta(const RATE_CONTROL *rc, double qstart, double qtarget);
+int vp9_compute_qdelta(const RATE_CONTROL *rc, double qstart, double qtarget,
+                       vpx_bit_depth_t bit_depth);
 
 // Computes a q delta (in "q index" terms) to get from a starting q value
 // to a value that should equate to the given rate ratio.
 int vp9_compute_qdelta_by_rate(const RATE_CONTROL *rc, FRAME_TYPE frame_type,
-                               int qindex, double rate_target_ratio);
+                               int qindex, double rate_target_ratio,
+                               vpx_bit_depth_t bit_depth);
 
 void vp9_rc_update_framerate(struct VP9_COMP *cpi);
 
diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index b826ff4..17369d4 100644
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -93,34 +93,69 @@
 }
 
 // Values are now correlated to quantizer.
-static int sad_per_bit16lut[QINDEX_RANGE];
-static int sad_per_bit4lut[QINDEX_RANGE];
+static int sad_per_bit16lut_8[QINDEX_RANGE];
+static int sad_per_bit4lut_8[QINDEX_RANGE];
 
-void vp9_init_me_luts() {
+#if CONFIG_VP9_HIGHBITDEPTH
+static int sad_per_bit16lut_10[QINDEX_RANGE];
+static int sad_per_bit4lut_10[QINDEX_RANGE];
+static int sad_per_bit16lut_12[QINDEX_RANGE];
+static int sad_per_bit4lut_12[QINDEX_RANGE];
+#endif
+
+static void init_me_luts_bd(int *bit16lut, int *bit4lut, int range,
+                            vpx_bit_depth_t bit_depth) {
   int i;
-
   // Initialize the sad lut tables using a formulaic calculation for now.
   // This is to make it easier to resolve the impact of experimental changes
   // to the quantizer tables.
-  for (i = 0; i < QINDEX_RANGE; ++i) {
-    const double q = vp9_convert_qindex_to_q(i);
-    sad_per_bit16lut[i] = (int)(0.0418 * q + 2.4107);
-    sad_per_bit4lut[i] = (int)(0.063 * q + 2.742);
+  for (i = 0; i < range; i++) {
+    const double q = vp9_convert_qindex_to_q(i, bit_depth);
+    bit16lut[i] = (int)(0.0418 * q + 2.4107);
+    bit4lut[i] = (int)(0.063 * q + 2.742);
   }
 }
 
+void vp9_init_me_luts() {
+  init_me_luts_bd(sad_per_bit16lut_8, sad_per_bit4lut_8, QINDEX_RANGE,
+                  VPX_BITS_8);
+#if CONFIG_VP9_HIGHBITDEPTH
+  init_me_luts_bd(sad_per_bit16lut_10, sad_per_bit4lut_10, QINDEX_RANGE,
+                  VPX_BITS_10);
+  init_me_luts_bd(sad_per_bit16lut_12, sad_per_bit4lut_12, QINDEX_RANGE,
+                  VPX_BITS_12);
+#endif
+}
+
 static const int rd_boost_factor[16] = {
   64, 32, 32, 32, 24, 16, 12, 12,
   8, 8, 4, 4, 2, 2, 1, 0
 };
 static const int rd_frame_type_factor[FRAME_UPDATE_TYPES] = {
-128, 144, 128, 128, 144
+  128, 144, 128, 128, 144
 };
 
 int vp9_compute_rd_mult(const VP9_COMP *cpi, int qindex) {
-  const int q = vp9_dc_quant(qindex, 0);
+  const int q = vp9_dc_quant(qindex, 0, cpi->common.bit_depth);
+#if CONFIG_VP9_HIGHBITDEPTH
+  int rdmult = 0;
+  switch (cpi->common.bit_depth) {
+    case VPX_BITS_8:
+      rdmult = 88 * q * q / 24;
+      break;
+    case VPX_BITS_10:
+      rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 4);
+      break;
+    case VPX_BITS_12:
+      rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 8);
+      break;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
   int rdmult = 88 * q * q / 24;
-
+#endif
   if (cpi->oxcf.pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
@@ -132,15 +167,53 @@
   return rdmult;
 }
 
-static int compute_rd_thresh_factor(int qindex) {
+static int compute_rd_thresh_factor(int qindex, vpx_bit_depth_t bit_depth) {
+  double q;
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
+      break;
+    case VPX_BITS_10:
+      q = vp9_dc_quant(qindex, 0, VPX_BITS_10) / 16.0;
+      break;
+    case VPX_BITS_12:
+      q = vp9_dc_quant(qindex, 0, VPX_BITS_12) / 64.0;
+      break;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
+  (void) bit_depth;
+  q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
+#endif
   // TODO(debargha): Adjust the function below.
-  const int q = (int)(pow(vp9_dc_quant(qindex, 0) / 4.0, RD_THRESH_POW) * 5.12);
-  return MAX(q, 8);
+  return MAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
 }
 
 void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex) {
-  cpi->mb.sadperbit16 = sad_per_bit16lut[qindex];
-  cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (cpi->common.bit_depth) {
+    case VPX_BITS_8:
+      cpi->mb.sadperbit16 = sad_per_bit16lut_8[qindex];
+      cpi->mb.sadperbit4 = sad_per_bit4lut_8[qindex];
+      break;
+    case VPX_BITS_10:
+      cpi->mb.sadperbit16 = sad_per_bit16lut_10[qindex];
+      cpi->mb.sadperbit4 = sad_per_bit4lut_10[qindex];
+      break;
+    case VPX_BITS_12:
+      cpi->mb.sadperbit16 = sad_per_bit16lut_12[qindex];
+      cpi->mb.sadperbit4 = sad_per_bit4lut_12[qindex];
+      break;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+  }
+#else
+  cpi->mb.sadperbit16 = sad_per_bit16lut_8[qindex];
+  cpi->mb.sadperbit4 = sad_per_bit4lut_8[qindex];
+#endif
 }
 
 static void set_block_thresholds(const VP9_COMMON *cm, RD_OPT *rd) {
@@ -149,9 +222,8 @@
   for (segment_id = 0; segment_id < MAX_SEGMENTS; ++segment_id) {
     const int qindex =
         clamp(vp9_get_qindex(&cm->seg, segment_id, cm->base_qindex) +
-                  cm->y_dc_delta_q,
-              0, MAXQ);
-    const int q = compute_rd_thresh_factor(qindex);
+              cm->y_dc_delta_q, 0, MAXQ);
+    const int q = compute_rd_thresh_factor(qindex, cm->bit_depth);
 
     for (bsize = 0; bsize < BLOCK_SIZES; ++bsize) {
       // Threshold here seems unnecessarily harsh but fine given actual
@@ -363,7 +435,7 @@
                  uint8_t *ref_y_buffer, int ref_y_stride,
                  int ref_frame, BLOCK_SIZE block_size) {
   MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   int i;
   int zero_seen = 0;
   int best_index = 0;
@@ -442,7 +514,7 @@
 
 int vp9_get_switchable_rate(const VP9_COMP *cpi) {
   const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  const MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const int ctx = vp9_get_pred_context_switchable_interp(xd);
   return SWITCHABLE_INTERP_RATE_FACTOR *
              cpi->switchable_interp_costs[ctx][mbmi->interp_filter];
@@ -457,9 +529,15 @@
   for (i = 0; i < MAX_MODES; ++i)
     rd->thresh_mult[i] = cpi->oxcf.mode == BEST ? -500 : 0;
 
-  rd->thresh_mult[THR_NEARESTMV] = 0;
-  rd->thresh_mult[THR_NEARESTG] = 0;
-  rd->thresh_mult[THR_NEARESTA] = 0;
+  if (sf->adaptive_rd_thresh) {
+    rd->thresh_mult[THR_NEARESTMV] = 300;
+    rd->thresh_mult[THR_NEARESTG] = 300;
+    rd->thresh_mult[THR_NEARESTA] = 300;
+  } else {
+    rd->thresh_mult[THR_NEARESTMV] = 0;
+    rd->thresh_mult[THR_NEARESTG] = 0;
+    rd->thresh_mult[THR_NEARESTA] = 0;
+  }
 
   rd->thresh_mult[THR_DC] += 1000;
 
@@ -498,41 +576,6 @@
   rd->thresh_mult[THR_D153_PRED] += 2500;
   rd->thresh_mult[THR_D207_PRED] += 2500;
   rd->thresh_mult[THR_D63_PRED] += 2500;
-
-  // Disable frame modes if flags not set.
-  if (!(cpi->ref_frame_flags & VP9_LAST_FLAG)) {
-    rd->thresh_mult[THR_NEWMV    ] = INT_MAX;
-    rd->thresh_mult[THR_NEARESTMV] = INT_MAX;
-    rd->thresh_mult[THR_ZEROMV   ] = INT_MAX;
-    rd->thresh_mult[THR_NEARMV   ] = INT_MAX;
-  }
-  if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
-    rd->thresh_mult[THR_NEARESTG ] = INT_MAX;
-    rd->thresh_mult[THR_ZEROG    ] = INT_MAX;
-    rd->thresh_mult[THR_NEARG    ] = INT_MAX;
-    rd->thresh_mult[THR_NEWG     ] = INT_MAX;
-  }
-  if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
-    rd->thresh_mult[THR_NEARESTA ] = INT_MAX;
-    rd->thresh_mult[THR_ZEROA    ] = INT_MAX;
-    rd->thresh_mult[THR_NEARA    ] = INT_MAX;
-    rd->thresh_mult[THR_NEWA     ] = INT_MAX;
-  }
-
-  if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_LAST_FLAG | VP9_ALT_FLAG)) {
-    rd->thresh_mult[THR_COMP_ZEROLA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARLA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEWLA    ] = INT_MAX;
-  }
-  if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
-    rd->thresh_mult[THR_COMP_ZEROGA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARGA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEWGA    ] = INT_MAX;
-  }
 }
 
 void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
@@ -554,18 +597,4 @@
   for (i = 0; i < MAX_REFS; ++i)
     if (sf->disable_split_mask & (1 << i))
       rd->thresh_mult_sub8x8[i] = INT_MAX;
-
-  // Disable mode test if frame flag is not set.
-  if (!(cpi->ref_frame_flags & VP9_LAST_FLAG))
-    rd->thresh_mult_sub8x8[THR_LAST] = INT_MAX;
-  if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG))
-    rd->thresh_mult_sub8x8[THR_GOLD] = INT_MAX;
-  if (!(cpi->ref_frame_flags & VP9_ALT_FLAG))
-    rd->thresh_mult_sub8x8[THR_ALTR] = INT_MAX;
-  if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_LAST_FLAG | VP9_ALT_FLAG))
-    rd->thresh_mult_sub8x8[THR_COMP_LA] = INT_MAX;
-  if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_GOLD_FLAG | VP9_ALT_FLAG))
-    rd->thresh_mult_sub8x8[THR_COMP_GA] = INT_MAX;
 }
diff --git a/vp9/encoder/vp9_rd.h b/vp9/encoder/vp9_rd.h
index eeb5e0f..5dcb2f8 100644
--- a/vp9/encoder/vp9_rd.h
+++ b/vp9/encoder/vp9_rd.h
@@ -51,6 +51,12 @@
 
   THR_NEARMV,
   THR_NEARA,
+  THR_NEARG,
+
+  THR_ZEROMV,
+  THR_ZEROG,
+  THR_ZEROA,
+
   THR_COMP_NEARESTLA,
   THR_COMP_NEARESTGA,
 
@@ -58,13 +64,9 @@
 
   THR_COMP_NEARLA,
   THR_COMP_NEWLA,
-  THR_NEARG,
   THR_COMP_NEARGA,
   THR_COMP_NEWGA,
 
-  THR_ZEROMV,
-  THR_ZEROG,
-  THR_ZEROA,
   THR_COMP_ZEROLA,
   THR_COMP_ZEROGA,
 
@@ -98,6 +100,8 @@
   int threshes[MAX_SEGMENTS][BLOCK_SIZES][MAX_MODES];
   int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
 
+  int mode_map[BLOCK_SIZES][MAX_MODES];
+
   int64_t comp_pred_diff[REFERENCE_MODES];
   int64_t prediction_type_threshes[MAX_REF_FRAMES][REFERENCE_MODES];
   int64_t tx_select_diff[TX_MODES];
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 506c9bc..e7403c4 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -41,9 +41,14 @@
 #define RD_THRESH_MAX_FACT 64
 #define RD_THRESH_INC      1
 
-#define LAST_FRAME_MODE_MASK    0xFFEDCD60
-#define GOLDEN_FRAME_MODE_MASK  0xFFDA3BB0
-#define ALT_REF_MODE_MASK       0xFFC648D0
+#define LAST_FRAME_MODE_MASK    ((1 << GOLDEN_FRAME) | (1 << ALTREF_FRAME) | \
+                                 (1 << INTRA_FRAME))
+#define GOLDEN_FRAME_MODE_MASK  ((1 << LAST_FRAME) | (1 << ALTREF_FRAME) | \
+                                 (1 << INTRA_FRAME))
+#define ALT_REF_MODE_MASK       ((1 << LAST_FRAME) | (1 << GOLDEN_FRAME) | \
+                                 (1 << INTRA_FRAME))
+
+#define SECOND_REF_FRAME_MASK   ((1 << ALTREF_FRAME) | 0x01)
 
 #define MIN_EARLY_TERM_INDEX    3
 
@@ -86,6 +91,12 @@
 
   {NEARMV,    {LAST_FRAME,   NONE}},
   {NEARMV,    {ALTREF_FRAME, NONE}},
+  {NEARMV,    {GOLDEN_FRAME, NONE}},
+
+  {ZEROMV,    {LAST_FRAME,   NONE}},
+  {ZEROMV,    {GOLDEN_FRAME, NONE}},
+  {ZEROMV,    {ALTREF_FRAME, NONE}},
+
   {NEARESTMV, {LAST_FRAME,   ALTREF_FRAME}},
   {NEARESTMV, {GOLDEN_FRAME, ALTREF_FRAME}},
 
@@ -93,13 +104,9 @@
 
   {NEARMV,    {LAST_FRAME,   ALTREF_FRAME}},
   {NEWMV,     {LAST_FRAME,   ALTREF_FRAME}},
-  {NEARMV,    {GOLDEN_FRAME, NONE}},
   {NEARMV,    {GOLDEN_FRAME, ALTREF_FRAME}},
   {NEWMV,     {GOLDEN_FRAME, ALTREF_FRAME}},
 
-  {ZEROMV,    {LAST_FRAME,   NONE}},
-  {ZEROMV,    {GOLDEN_FRAME, NONE}},
-  {ZEROMV,    {ALTREF_FRAME, NONE}},
   {ZEROMV,    {LAST_FRAME,   ALTREF_FRAME}},
   {ZEROMV,    {GOLDEN_FRAME, ALTREF_FRAME}},
 
@@ -169,7 +176,7 @@
   int i;
   int64_t rate_sum = 0;
   int64_t dist_sum = 0;
-  const int ref = xd->mi[0]->mbmi.ref_frame[0];
+  const int ref = xd->mi[0].src_mi->mbmi.ref_frame[0];
   unsigned int sse;
   unsigned int var = 0;
   unsigned int sum_sse = 0;
@@ -244,7 +251,7 @@
   *out_dist_sum = dist_sum << 4;
 }
 
-int64_t vp9_block_error_c(const int16_t *coeff, const int16_t *dqcoeff,
+int64_t vp9_block_error_c(const tran_low_t *coeff, const tran_low_t *dqcoeff,
                           intptr_t block_size, int64_t *ssz) {
   int i;
   int64_t error = 0, sqcoeff = 0;
@@ -277,13 +284,13 @@
                               const int16_t *scan, const int16_t *nb,
                               int use_fast_coef_costing) {
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   const struct macroblock_plane *p = &x->plane[plane];
   const struct macroblockd_plane *pd = &xd->plane[plane];
   const PLANE_TYPE type = pd->plane_type;
   const int16_t *band_count = &band_counts[tx_size][1];
   const int eob = p->eobs[block];
-  const int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  const tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   unsigned int (*token_costs)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
                    x->token_costs[tx_size][type][is_inter_block(mbmi)];
   uint8_t token_cache[32 * 32];
@@ -353,13 +360,13 @@
   const struct macroblockd_plane *const pd = &xd->plane[plane];
   int64_t this_sse;
   int shift = tx_size == TX_32X32 ? 0 : 2;
-  int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
+  tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   args->dist = vp9_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
                                &this_sse) >> shift;
   args->sse  = this_sse >> shift;
 
-  if (x->skip_encode && !is_inter_block(&xd->mi[0]->mbmi)) {
+  if (x->skip_encode && !is_inter_block(&xd->mi[0].src_mi->mbmi)) {
     // TODO(jingning): tune the model to better capture the distortion.
     int64_t p = (pd->dequant[1] * pd->dequant[1] *
                     (1 << ss_txfrm_size)) >> (shift + 2);
@@ -384,7 +391,7 @@
   struct rdcost_block_args *args = arg;
   MACROBLOCK *const x = args->x;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   int64_t rd1, rd2, rd;
 
   if (args->skip)
@@ -400,8 +407,8 @@
       dist_block(plane, block, tx_size, args);
     } else if (x->skip_txfm[(plane << 2) + (block >> (tx_size << 1))] == 2) {
       // compute DC coefficient
-      int16_t *const coeff   = BLOCK_OFFSET(x->plane[plane].coeff, block);
-      int16_t *const dqcoeff = BLOCK_OFFSET(xd->plane[plane].dqcoeff, block);
+      tran_low_t *const coeff   = BLOCK_OFFSET(x->plane[plane].coeff, block);
+      tran_low_t *const dqcoeff = BLOCK_OFFSET(xd->plane[plane].dqcoeff, block);
       vp9_xform_quant_dc(x, plane, block, plane_bsize, tx_size);
       args->sse  = x->bsse[(plane << 2) + (block >> (tx_size << 1))] << 4;
       args->dist = args->sse;
@@ -456,7 +463,7 @@
   args.use_fast_coef_costing = use_fast_coef_casting;
 
   if (plane == 0)
-    xd->mi[0]->mbmi.tx_size = tx_size;
+    xd->mi[0].src_mi->mbmi.tx_size = tx_size;
 
   vp9_get_entropy_contexts(bsize, tx_size, pd, args.t_above, args.t_left);
 
@@ -486,7 +493,7 @@
   VP9_COMMON *const cm = &cpi->common;
   const TX_SIZE largest_tx_size = tx_mode_to_biggest_tx_size[cm->tx_mode];
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
 
   mbmi->tx_size = MIN(max_tx_size, largest_tx_size);
 
@@ -506,7 +513,7 @@
   const TX_SIZE max_tx_size = max_txsize_lookup[bs];
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   vp9_prob skip_prob = vp9_get_skip_prob(cm, xd);
   int r[TX_SIZES][2], s[TX_SIZES];
   int64_t d[TX_SIZES], sse[TX_SIZES];
@@ -593,7 +600,7 @@
   int64_t sse;
   int64_t *ret_sse = psse ? psse : &sse;
 
-  assert(bs == xd->mi[0]->mbmi.sb_type);
+  assert(bs == xd->mi[0].src_mi->mbmi.sb_type);
 
   if (cpi->sf.tx_size_search_method == USE_LARGESTALL || xd->lossless) {
     vpx_memset(txfm_cache, 0, TX_MODES * sizeof(int64_t));
@@ -657,7 +664,7 @@
 
   vpx_memcpy(ta, a, sizeof(ta));
   vpx_memcpy(tl, l, sizeof(tl));
-  xd->mi[0]->mbmi.tx_size = TX_4X4;
+  xd->mi[0].src_mi->mbmi.tx_size = TX_4X4;
 
   for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
     int64_t this_rd;
@@ -685,8 +692,8 @@
         uint8_t *const dst = &dst_init[idx * 4 + idy * 4 * dst_stride];
         int16_t *const src_diff = raster_block_offset_int16(BLOCK_8X8, block,
                                                             p->src_diff);
-        int16_t *const coeff = BLOCK_OFFSET(x->plane[0].coeff, block);
-        xd->mi[0]->bmi[block].as_mode = mode;
+        tran_low_t *const coeff = BLOCK_OFFSET(x->plane[0].coeff, block);
+        xd->mi[0].src_mi->bmi[block].as_mode = mode;
         vp9_predict_intra_block(xd, block, 1,
                                 TX_4X4, mode,
                                 x->skip_encode ? src : dst,
@@ -759,10 +766,10 @@
                                             int64_t best_rd) {
   int i, j;
   const MACROBLOCKD *const xd = &mb->e_mbd;
-  MODE_INFO *const mic = xd->mi[0];
-  const MODE_INFO *above_mi = xd->mi[-xd->mi_stride];
-  const MODE_INFO *left_mi = xd->left_available ? xd->mi[-1] : NULL;
-  const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
+  MODE_INFO *const mic = xd->mi[0].src_mi;
+  const MODE_INFO *above_mi = xd->mi[-xd->mi_stride].src_mi;
+  const MODE_INFO *left_mi = xd->left_available ? xd->mi[-1].src_mi : NULL;
+  const BLOCK_SIZE bsize = xd->mi[0].src_mi->mbmi.sb_type;
   const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
   const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
   int idx, idy;
@@ -820,6 +827,7 @@
   return RDCOST(mb->rdmult, mb->rddiv, cost, total_distortion);
 }
 
+// This function is used only for intra_only frames
 static int64_t rd_pick_intra_sby_mode(VP9_COMP *cpi, MACROBLOCK *x,
                                       int *rate, int *rate_tokenonly,
                                       int64_t *distortion, int *skippable,
@@ -829,12 +837,17 @@
   PREDICTION_MODE mode;
   PREDICTION_MODE mode_selected = DC_PRED;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MODE_INFO *const mic = xd->mi[0];
+  MODE_INFO *const mic = xd->mi[0].src_mi;
   int this_rate, this_rate_tokenonly, s;
   int64_t this_distortion, this_rd;
   TX_SIZE best_tx = TX_4X4;
   int i;
-  int *bmode_costs = cpi->mbmode_cost;
+  int *bmode_costs;
+  const MODE_INFO *above_mi = xd->mi[-xd->mi_stride].src_mi;
+  const MODE_INFO *left_mi = xd->left_available ? xd->mi[-1].src_mi : NULL;
+  const PREDICTION_MODE A = vp9_above_block_mode(mic, above_mi, 0);
+  const PREDICTION_MODE L = vp9_left_block_mode(mic, left_mi, 0);
+  bmode_costs = cpi->y_mode_costs[A][L];
 
   if (cpi->sf.tx_size_search_method == USE_FULL_RD)
     for (i = 0; i < TX_MODES; i++)
@@ -843,15 +856,6 @@
   /* Y Search for intra prediction mode */
   for (mode = DC_PRED; mode <= TM_PRED; mode++) {
     int64_t local_tx_cache[TX_MODES];
-    MODE_INFO *above_mi = xd->mi[-xd->mi_stride];
-    MODE_INFO *left_mi = xd->left_available ? xd->mi[-1] : NULL;
-
-    if (cpi->common.frame_type == KEY_FRAME) {
-      const PREDICTION_MODE A = vp9_above_block_mode(mic, above_mi, 0);
-      const PREDICTION_MODE L = vp9_left_block_mode(mic, left_mi, 0);
-
-      bmode_costs = cpi->y_mode_costs[A][L];
-    }
     mic->mbmi.mode = mode;
 
     super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion,
@@ -895,7 +899,7 @@
                              int64_t *sse, BLOCK_SIZE bsize,
                              int64_t ref_best_rd) {
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const TX_SIZE uv_tx_size = get_uv_tx_size(mbmi, &xd->plane[1]);
   int plane;
   int pnrate = 0, pnskip = 1;
@@ -952,7 +956,7 @@
     if (!(cpi->sf.intra_uv_mode_mask[max_tx_size] & (1 << mode)))
       continue;
 
-    xd->mi[0]->mbmi.uv_mode = mode;
+    xd->mi[0].src_mi->mbmi.uv_mode = mode;
 
     super_block_uvrd(cpi, x, &this_rate_tokenonly,
                      &this_distortion, &s, &this_sse, bsize, best_rd);
@@ -974,7 +978,7 @@
     }
   }
 
-  xd->mi[0]->mbmi.uv_mode = mode_selected;
+  xd->mi[0].src_mi->mbmi.uv_mode = mode_selected;
   return best_rd;
 }
 
@@ -985,7 +989,7 @@
   const VP9_COMMON *cm = &cpi->common;
   int64_t unused;
 
-  x->e_mbd.mi[0]->mbmi.uv_mode = DC_PRED;
+  x->e_mbd.mi[0].src_mi->mbmi.uv_mode = DC_PRED;
   super_block_uvrd(cpi, x, rate_tokenonly, distortion,
                    skippable, &unused, bsize, INT64_MAX);
   *rate = *rate_tokenonly + cpi->intra_uv_mode_cost[cm->frame_type][DC_PRED];
@@ -1011,7 +1015,7 @@
                             rate_uv, rate_uv_tokenonly, dist_uv, skip_uv,
                             bsize < BLOCK_8X8 ? BLOCK_8X8 : bsize, max_tx_size);
   }
-  *mode_uv = x->e_mbd.mi[0]->mbmi.uv_mode;
+  *mode_uv = x->e_mbd.mi[0].src_mi->mbmi.uv_mode;
 }
 
 static int cost_mv_ref(const VP9_COMP *cpi, PREDICTION_MODE mode,
@@ -1033,7 +1037,7 @@
                                 int_mv seg_mvs[MAX_REF_FRAMES],
                                 int_mv *best_ref_mv[2], const int *mvjcost,
                                 int *mvcost[2]) {
-  MODE_INFO *const mic = xd->mi[0];
+  MODE_INFO *const mic = xd->mi[0].src_mi;
   const MB_MODE_INFO *const mbmi = &mic->mbmi;
   int thismvcost = 0;
   int idx, idy;
@@ -1095,7 +1099,7 @@
   MACROBLOCKD *xd = &x->e_mbd;
   struct macroblockd_plane *const pd = &xd->plane[0];
   struct macroblock_plane *const p = &x->plane[0];
-  MODE_INFO *const mi = xd->mi[0];
+  MODE_INFO *const mi = xd->mi[0].src_mi;
   const BLOCK_SIZE plane_bsize = get_plane_block_size(mi->mbmi.sb_type, pd);
   const int width = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
   const int height = 4 * num_4x4_blocks_high_lookup[plane_bsize];
@@ -1132,7 +1136,7 @@
   for (idy = 0; idy < height / 4; ++idy) {
     for (idx = 0; idx < width / 4; ++idx) {
       int64_t ssz, rd, rd1, rd2;
-      int16_t* coeff;
+      tran_low_t* coeff;
 
       k += (idy * 2 + idx);
       coeff = BLOCK_OFFSET(p->coeff, k);
@@ -1194,7 +1198,7 @@
 }
 
 static INLINE void mi_buf_shift(MACROBLOCK *x, int i) {
-  MB_MODE_INFO *const mbmi = &x->e_mbd.mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &x->e_mbd.mi[0].src_mi->mbmi;
   struct macroblock_plane *const p = &x->plane[0];
   struct macroblockd_plane *const pd = &x->e_mbd.plane[0];
 
@@ -1209,7 +1213,7 @@
 
 static INLINE void mi_buf_restore(MACROBLOCK *x, struct buf_2d orig_src,
                                   struct buf_2d orig_pre[2]) {
-  MB_MODE_INFO *mbmi = &x->e_mbd.mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &x->e_mbd.mi[0].src_mi->mbmi;
   x->plane[0].src = orig_src;
   x->e_mbd.plane[0].pre[0] = orig_pre[0];
   if (has_second_ref(mbmi))
@@ -1224,11 +1228,9 @@
 // TODO(aconverse): Find out if this is still productive then clean up or remove
 static int check_best_zero_mv(
     const VP9_COMP *cpi, const uint8_t mode_context[MAX_REF_FRAMES],
-    int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES],
-    int inter_mode_mask, int this_mode,
+    int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES], int this_mode,
     const MV_REFERENCE_FRAME ref_frames[2]) {
-  if ((inter_mode_mask & (1 << ZEROMV)) &&
-      (this_mode == NEARMV || this_mode == NEARESTMV || this_mode == ZEROMV) &&
+  if ((this_mode == NEARMV || this_mode == NEARESTMV || this_mode == ZEROMV) &&
       frame_mv[this_mode][ref_frames[0]].as_int == 0 &&
       (ref_frames[1] == NONE ||
        frame_mv[this_mode][ref_frames[1]].as_int == 0)) {
@@ -1274,7 +1276,7 @@
   int i;
   BEST_SEG_INFO *bsi = bsi_buf + filter_idx;
   MACROBLOCKD *xd = &x->e_mbd;
-  MODE_INFO *mi = xd->mi[0];
+  MODE_INFO *mi = xd->mi[0].src_mi;
   MB_MODE_INFO *mbmi = &mi->mbmi;
   int mode_idx;
   int k, br = 0, idx, idy;
@@ -1346,7 +1348,6 @@
           continue;
 
         if (!check_best_zero_mv(cpi, mbmi->mode_context, frame_mv,
-                                inter_mode_mask,
                                 this_mode, mbmi->ref_frame))
           continue;
 
@@ -1716,7 +1717,7 @@
   ctx->skip = x->skip;
   ctx->skippable = skippable;
   ctx->best_mode_index = mode_index;
-  ctx->mic = *xd->mi[0];
+  ctx->mic = *xd->mi[0].src_mi;
   ctx->single_pred_diff = (int)comp_pred_diff[SINGLE_REFERENCE];
   ctx->comp_pred_diff   = (int)comp_pred_diff[COMPOUND_REFERENCE];
   ctx->hybrid_pred_diff = (int)comp_pred_diff[REFERENCE_MODE_SELECT];
@@ -1737,7 +1738,7 @@
   const VP9_COMMON *cm = &cpi->common;
   const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
   MACROBLOCKD *const xd = &x->e_mbd;
-  MODE_INFO *const mi = xd->mi[0];
+  MODE_INFO *const mi = xd->mi[0].src_mi;
   int_mv *const candidates = mi->mbmi.ref_mvs[ref_frame];
   const struct scale_factors *const sf = &cm->frame_refs[ref_frame - 1].sf;
 
@@ -1767,7 +1768,7 @@
                                  int_mv *tmp_mv, int *rate_mv) {
   MACROBLOCKD *xd = &x->e_mbd;
   const VP9_COMMON *cm = &cpi->common;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0, 0}};
   int bestsme = INT_MAX;
   int step_param;
@@ -1894,7 +1895,7 @@
   const int pw = 4 * num_4x4_blocks_wide_lookup[bsize];
   const int ph = 4 * num_4x4_blocks_high_lookup[bsize];
   MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   const int refs[2] = { mbmi->ref_frame[0],
                         mbmi->ref_frame[1] < 0 ? 0 : mbmi->ref_frame[1] };
   int_mv ref_mv[2];
@@ -2136,7 +2137,7 @@
   VP9_COMMON *cm = &cpi->common;
   RD_OPT *rd_opt = &cpi->rd;
   MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   const int is_comp_pred = has_second_ref(mbmi);
   const int this_mode = mbmi->mode;
   int_mv *frame_mv = mode_mv[this_mode];
@@ -2165,9 +2166,9 @@
   if (pred_filter_search) {
     INTERP_FILTER af = SWITCHABLE, lf = SWITCHABLE;
     if (xd->up_available)
-      af = xd->mi[-xd->mi_stride]->mbmi.interp_filter;
+      af = xd->mi[-xd->mi_stride].src_mi->mbmi.interp_filter;
     if (xd->left_available)
-      lf = xd->mi[-1]->mbmi.interp_filter;
+      lf = xd->mi[-1].src_mi->mbmi.interp_filter;
 
     if ((this_mode != NEWMV) || (af == lf))
       best_filter = af;
@@ -2212,7 +2213,7 @@
         return INT64_MAX;
       *rate2 += rate_mv;
       frame_mv[refs[0]].as_int =
-          xd->mi[0]->bmi[0].as_mv[0].as_int = tmp_mv.as_int;
+          xd->mi[0].src_mi->bmi[0].as_mv[0].as_int = tmp_mv.as_int;
       single_newmv[refs[0]].as_int = tmp_mv.as_int;
     }
   }
@@ -2464,7 +2465,7 @@
   TX_SIZE max_uv_tx_size;
   x->skip_encode = 0;
   ctx->skip = 0;
-  xd->mi[0]->mbmi.ref_frame[0] = INTRA_FRAME;
+  xd->mi[0].src_mi->mbmi.ref_frame[0] = INTRA_FRAME;
 
   if (bsize >= BLOCK_8X8) {
     if (rd_pick_intra_sby_mode(cpi, x, &rate_y, &rate_y_tokenonly,
@@ -2473,11 +2474,6 @@
       *returnrate = INT_MAX;
       return;
     }
-    max_uv_tx_size = get_uv_tx_size_impl(xd->mi[0]->mbmi.tx_size, bsize,
-                                         pd[1].subsampling_x,
-                                         pd[1].subsampling_y);
-    rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
-                            &dist_uv, &uv_skip, bsize, max_uv_tx_size);
   } else {
     y_skip = 0;
     if (rd_pick_intra_sub_8x8_y_mode(cpi, x, &rate_y, &rate_y_tokenonly,
@@ -2485,12 +2481,13 @@
       *returnrate = INT_MAX;
       return;
     }
-    max_uv_tx_size = get_uv_tx_size_impl(xd->mi[0]->mbmi.tx_size, bsize,
-                                         pd[1].subsampling_x,
-                                         pd[1].subsampling_y);
-    rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
-                            &dist_uv, &uv_skip, BLOCK_8X8, max_uv_tx_size);
   }
+  max_uv_tx_size = get_uv_tx_size_impl(xd->mi[0].src_mi->mbmi.tx_size, bsize,
+                                       pd[1].subsampling_x,
+                                       pd[1].subsampling_y);
+  rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
+                          &dist_uv, &uv_skip, MAX(BLOCK_8X8, bsize),
+                          max_uv_tx_size);
 
   if (y_skip && uv_skip) {
     *returnrate = rate_y + rate_uv - rate_y_tokenonly - rate_uv_tokenonly +
@@ -2510,27 +2507,26 @@
       }
   }
 
-  ctx->mic = *xd->mi[0];
+  ctx->mic = *xd->mi[0].src_mi;
 }
 
-// Updating rd_thresh_freq_fact[] here means that the different
-// partition/block sizes are handled independently based on the best
-// choice for the current partition. It may well be better to keep a scaled
-// best rd so far value and update rd_thresh_freq_fact based on the mode/size
-// combination that wins out.
 static void update_rd_thresh_fact(VP9_COMP *cpi, int bsize,
                                   int best_mode_index) {
   if (cpi->sf.adaptive_rd_thresh > 0) {
     const int top_mode = bsize < BLOCK_8X8 ? MAX_REFS : MAX_MODES;
     int mode;
     for (mode = 0; mode < top_mode; ++mode) {
-      int *const fact = &cpi->rd.thresh_freq_fact[bsize][mode];
-
-      if (mode == best_mode_index) {
-        *fact -= (*fact >> 3);
-      } else {
-        *fact = MIN(*fact + RD_THRESH_INC,
-                    cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
+      const BLOCK_SIZE min_size = MAX(bsize - 1, BLOCK_4X4);
+      const BLOCK_SIZE max_size = MIN(bsize + 2, BLOCK_64X64);
+      BLOCK_SIZE bs;
+      for (bs = min_size; bs <= max_size; ++bs) {
+        int *const fact = &cpi->rd.thresh_freq_fact[bs][mode];
+        if (mode == best_mode_index) {
+          *fact -= (*fact >> 4);
+        } else {
+          *fact = MIN(*fact + RD_THRESH_INC,
+                      cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
+        }
       }
     }
   }
@@ -2547,7 +2543,7 @@
   VP9_COMMON *const cm = &cpi->common;
   RD_OPT *const rd_opt = &cpi->rd;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const struct segmentation *const seg = &cm->seg;
   struct macroblockd_plane *const pd = xd->plane;
   PREDICTION_MODE this_mode;
@@ -2570,28 +2566,29 @@
   int64_t best_filter_diff[SWITCHABLE_FILTER_CONTEXTS];
   MB_MODE_INFO best_mbmode;
   int best_mode_skippable = 0;
-  int mode_index, best_mode_index = -1;
+  int midx, best_mode_index = -1;
   unsigned int ref_costs_single[MAX_REF_FRAMES], ref_costs_comp[MAX_REF_FRAMES];
   vp9_prob comp_mode_p;
   int64_t best_intra_rd = INT64_MAX;
-  int64_t best_inter_rd = INT64_MAX;
+  unsigned int best_pred_sse = UINT_MAX;
   PREDICTION_MODE best_intra_mode = DC_PRED;
-  MV_REFERENCE_FRAME best_inter_ref_frame = LAST_FRAME;
   int rate_uv_intra[TX_SIZES], rate_uv_tokenonly[TX_SIZES];
   int64_t dist_uv[TX_SIZES];
   int skip_uv[TX_SIZES];
   PREDICTION_MODE mode_uv[TX_SIZES];
-  int intra_cost_penalty = 20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q);
+  const int intra_cost_penalty =
+      20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q, cm->bit_depth);
   int best_skip2 = 0;
-  int mode_skip_mask = 0;
+  uint8_t ref_frame_skip_mask[2] = { 0 };
+  uint16_t mode_skip_mask[MAX_REF_FRAMES] = { 0 };
   int mode_skip_start = cpi->sf.mode_skip_start + 1;
   const int *const rd_threshes = rd_opt->threshes[segment_id][bsize];
   const int *const rd_thresh_freq_fact = rd_opt->thresh_freq_fact[bsize];
+  int64_t mode_threshold[MAX_MODES];
+  int *mode_map = rd_opt->mode_map[bsize];
   const int mode_search_skip_flags = cpi->sf.mode_search_skip_flags;
-  const int intra_y_mode_mask =
-      cpi->sf.intra_y_mode_mask[max_txsize_lookup[bsize]];
-  int inter_mode_mask = cpi->sf.inter_mode_mask[bsize];
   vp9_zero(best_mbmode);
+
   x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
 
   estimate_ref_frame_costs(cm, xd, segment_id, ref_costs_single, ref_costs_comp,
@@ -2627,23 +2624,17 @@
   }
 
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    // All modes from vp9_mode_order that use this frame as any ref
-    static const int ref_frame_mask_all[] = {
-        0x0, 0x123291, 0x25c444, 0x39b722
-    };
-    // Fixed mv modes (NEARESTMV, NEARMV, ZEROMV) from vp9_mode_order that use
-    // this frame as their primary ref
-    static const int ref_frame_mask_fixedmv[] = {
-        0x0, 0x121281, 0x24c404, 0x080102
-    };
     if (!(cpi->ref_frame_flags & flag_list[ref_frame])) {
-      // Skip modes for missing references
-      mode_skip_mask |= ref_frame_mask_all[ref_frame];
+      // Skip checking missing references in both single and compound reference
+      // modes. Note that a mode will be skipped iff both reference frames
+      // are masked out.
+      ref_frame_skip_mask[0] |= (1 << ref_frame);
+      ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
     } else if (cpi->sf.reference_masking) {
       for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
         // Skip fixed mv modes for poor references
         if ((x->pred_mv_sad[ref_frame] >> 2) > x->pred_mv_sad[i]) {
-          mode_skip_mask |= ref_frame_mask_fixedmv[ref_frame];
+          mode_skip_mask[ref_frame] |= INTER_NEAREST_NEAR_ZERO;
           break;
         }
       }
@@ -2652,7 +2643,8 @@
     // then do nothing if the current ref frame is not allowed..
     if (vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
         vp9_get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) != (int)ref_frame) {
-      mode_skip_mask |= ref_frame_mask_all[ref_frame];
+      ref_frame_skip_mask[0] |= (1 << ref_frame);
+      ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
     }
   }
 
@@ -2665,24 +2657,51 @@
     // an unfiltered alternative. We allow near/nearest as well
     // because they may result in zero-zero MVs but be cheaper.
     if (cpi->rc.is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
-      mode_skip_mask =
-          ~((1 << THR_NEARESTA) | (1 << THR_NEARA) | (1 << THR_ZEROA));
+      ref_frame_skip_mask[0] = (1 << LAST_FRAME) | (1 << GOLDEN_FRAME);
+      ref_frame_skip_mask[1] = SECOND_REF_FRAME_MASK;
+      mode_skip_mask[ALTREF_FRAME] = ~INTER_NEAREST_NEAR_ZERO;
       if (frame_mv[NEARMV][ALTREF_FRAME].as_int != 0)
-        mode_skip_mask |= (1 << THR_NEARA);
+        mode_skip_mask[ALTREF_FRAME] |= (1 << NEARMV);
       if (frame_mv[NEARESTMV][ALTREF_FRAME].as_int != 0)
-        mode_skip_mask |= (1 << THR_NEARESTA);
+        mode_skip_mask[ALTREF_FRAME] |= (1 << NEARESTMV);
+    }
+  }
+
+  if (cpi->rc.is_src_frame_alt_ref) {
+    if (cpi->sf.alt_ref_search_fp) {
+      mode_skip_mask[ALTREF_FRAME] = 0;
+      ref_frame_skip_mask[0] = ~(1 << ALTREF_FRAME);
+      ref_frame_skip_mask[1] = SECOND_REF_FRAME_MASK;
     }
   }
 
   if (bsize > cpi->sf.max_intra_bsize) {
-    const int all_intra_modes = (1 << THR_DC) | (1 << THR_TM) |
-        (1 << THR_H_PRED) | (1 << THR_V_PRED) | (1 << THR_D135_PRED) |
-        (1 << THR_D207_PRED) | (1 << THR_D153_PRED) | (1 << THR_D63_PRED) |
-        (1 << THR_D117_PRED) | (1 << THR_D45_PRED);
-    mode_skip_mask |= all_intra_modes;
+    ref_frame_skip_mask[0] |= (1 << INTRA_FRAME);
+    ref_frame_skip_mask[1] |= (1 << INTRA_FRAME);
   }
 
-  for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
+  mode_skip_mask[INTRA_FRAME] |=
+      ~(cpi->sf.intra_y_mode_mask[max_txsize_lookup[bsize]]);
+
+  for (i = 0; i < MAX_MODES; ++i)
+    mode_threshold[i] = ((int64_t)rd_threshes[i] * rd_thresh_freq_fact[i]) >> 5;
+
+  midx =  cpi->sf.schedule_mode_search ? mode_skip_start : 0;
+  while (midx > 4) {
+    uint8_t end_pos = 0;
+    for (i = 5; i < midx; ++i) {
+      if (mode_threshold[mode_map[i - 1]] > mode_threshold[mode_map[i]]) {
+        uint8_t tmp = mode_map[i];
+        mode_map[i] = mode_map[i - 1];
+        mode_map[i - 1] = tmp;
+        end_pos = i;
+      }
+    }
+    midx = end_pos;
+  }
+
+  for (midx = 0; midx < MAX_MODES; ++midx) {
+    int mode_index = mode_map[midx];
     int mode_excluded = 0;
     int64_t this_rd = INT64_MAX;
     int disable_skip = 0;
@@ -2691,31 +2710,30 @@
     int64_t distortion2 = 0, distortion_y = 0, distortion_uv = 0;
     int skippable = 0;
     int64_t tx_cache[TX_MODES];
-    int i;
     int this_skip2 = 0;
     int64_t total_sse = INT64_MAX;
     int early_term = 0;
 
     this_mode = vp9_mode_order[mode_index].mode;
     ref_frame = vp9_mode_order[mode_index].ref_frame[0];
-    if (ref_frame != INTRA_FRAME && !(inter_mode_mask & (1 << this_mode)))
-      continue;
     second_ref_frame = vp9_mode_order[mode_index].ref_frame[1];
 
     // Look at the reference frame of the best mode so far and set the
     // skip mask to look at a subset of the remaining modes.
-    if (mode_index == mode_skip_start && best_mode_index >= 0) {
-      switch (vp9_mode_order[best_mode_index].ref_frame[0]) {
+    if (midx == mode_skip_start && best_mode_index >= 0) {
+      switch (best_mbmode.ref_frame[0]) {
         case INTRA_FRAME:
           break;
         case LAST_FRAME:
-          mode_skip_mask |= LAST_FRAME_MODE_MASK;
+          ref_frame_skip_mask[0] |= LAST_FRAME_MODE_MASK;
+          ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
           break;
         case GOLDEN_FRAME:
-          mode_skip_mask |= GOLDEN_FRAME_MODE_MASK;
+          ref_frame_skip_mask[0] |= GOLDEN_FRAME_MODE_MASK;
+          ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
           break;
         case ALTREF_FRAME:
-          mode_skip_mask |= ALT_REF_MODE_MASK;
+          ref_frame_skip_mask[0] |= ALT_REF_MODE_MASK;
           break;
         case NONE:
         case MAX_REF_FRAMES:
@@ -2724,18 +2742,18 @@
       }
     }
 
-    if (cpi->sf.alt_ref_search_fp && cpi->rc.is_src_frame_alt_ref) {
-      mode_skip_mask = 0;
-      if (!(ref_frame == ALTREF_FRAME && second_ref_frame == NONE))
-        continue;
-    }
+    if (ref_frame_skip_mask[0] & (1 << ref_frame) &&
+        ref_frame_skip_mask[1] & (1 << MAX(0, second_ref_frame)))
+      continue;
 
-    if (mode_skip_mask & (1 << mode_index))
+    if (mode_skip_mask[ref_frame] & (1 << this_mode))
       continue;
 
     // Test best rd so far against threshold for trying this mode.
-    if (rd_less_than_thresh(best_rd, rd_threshes[mode_index],
-                            rd_thresh_freq_fact[mode_index]))
+    if (best_mode_skippable && cpi->sf.schedule_mode_search)
+      mode_threshold[mode_index] <<= 1;
+
+    if (best_rd < mode_threshold[mode_index])
       continue;
 
     if (cpi->sf.motion_field_mode_search) {
@@ -2754,10 +2772,10 @@
       ref_mv.as_int = INVALID_MV;
 
       if ((mi_row - 1) >= tile->mi_row_start) {
-        ref_mv = xd->mi[-xd->mi_stride]->mbmi.mv[0];
-        rf = xd->mi[-xd->mi_stride]->mbmi.ref_frame[0];
+        ref_mv = xd->mi[-xd->mi_stride].src_mi->mbmi.mv[0];
+        rf = xd->mi[-xd->mi_stride].src_mi->mbmi.ref_frame[0];
         for (i = 0; i < mi_width; ++i) {
-          ref_mbmi = &xd->mi[-xd->mi_stride + i]->mbmi;
+          ref_mbmi = &xd->mi[-xd->mi_stride + i].src_mi->mbmi;
           const_motion &= (ref_mv.as_int == ref_mbmi->mv[0].as_int) &&
                           (ref_frame == ref_mbmi->ref_frame[0]);
           skip_ref_frame &= (rf == ref_mbmi->ref_frame[0]);
@@ -2766,11 +2784,11 @@
 
       if ((mi_col - 1) >= tile->mi_col_start) {
         if (ref_mv.as_int == INVALID_MV)
-          ref_mv = xd->mi[-1]->mbmi.mv[0];
+          ref_mv = xd->mi[-1].src_mi->mbmi.mv[0];
         if (rf == NONE)
-          rf = xd->mi[-1]->mbmi.ref_frame[0];
+          rf = xd->mi[-1].src_mi->mbmi.ref_frame[0];
         for (i = 0; i < mi_height; ++i) {
-          ref_mbmi = &xd->mi[i * xd->mi_stride - 1]->mbmi;
+          ref_mbmi = &xd->mi[i * xd->mi_stride - 1].src_mi->mbmi;
           const_motion &= (ref_mv.as_int == ref_mbmi->mv[0].as_int) &&
                           (ref_frame == ref_mbmi->ref_frame[0]);
           skip_ref_frame &= (rf == ref_mbmi->ref_frame[0]);
@@ -2792,14 +2810,19 @@
       if (!cm->allow_comp_inter_inter)
         continue;
 
+      // Skip compound inter modes if ARF is not available.
+      if (!(cpi->ref_frame_flags & flag_list[second_ref_frame]))
+        continue;
+
+      // Do not allow compound prediction if the segment level reference frame
+      // feature is in use as in this case there can only be one reference.
+      if (vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
+        continue;
+
       if ((mode_search_skip_flags & FLAG_SKIP_COMP_BESTINTRA) &&
-          best_mode_index >=0 &&
-          vp9_mode_order[best_mode_index].ref_frame[0] == INTRA_FRAME)
+          best_mode_index >= 0 && best_mbmode.ref_frame[0] == INTRA_FRAME)
         continue;
-      if ((mode_search_skip_flags & FLAG_SKIP_COMP_REFMISMATCH) &&
-          ref_frame != best_inter_ref_frame &&
-          second_ref_frame != best_inter_ref_frame)
-        continue;
+
       mode_excluded = cm->reference_mode == SINGLE_REFERENCE;
     } else {
       if (ref_frame != INTRA_FRAME)
@@ -2808,11 +2831,9 @@
 
     if (ref_frame == INTRA_FRAME) {
       if (cpi->sf.adaptive_mode_search)
-        if ((x->source_variance << num_pels_log2_lookup[bsize]) > best_intra_rd)
+        if ((x->source_variance << num_pels_log2_lookup[bsize]) > best_pred_sse)
           continue;
 
-      if (!(intra_y_mode_mask & (1 << this_mode)))
-        continue;
       if (this_mode != DC_PRED) {
         // Disable intra modes other than DC_PRED for blocks with low variance
         // Threshold for intra skipping based on source variance
@@ -2826,7 +2847,7 @@
         if ((mode_search_skip_flags & FLAG_SKIP_INTRA_BESTINTER) &&
             (this_mode >= D45_PRED && this_mode <= TM_PRED)) {
           if (best_mode_index >= 0 &&
-              vp9_mode_order[best_mode_index].ref_frame[0] > INTRA_FRAME)
+              best_mbmode.ref_frame[0] > INTRA_FRAME)
             continue;
         }
         if (mode_search_skip_flags & FLAG_SKIP_INTRA_DIRMISMATCH) {
@@ -2837,7 +2858,7 @@
     } else {
       const MV_REFERENCE_FRAME ref_frames[2] = {ref_frame, second_ref_frame};
       if (!check_best_zero_mv(cpi, mbmi->mode_context, frame_mv,
-                              inter_mode_mask, this_mode, ref_frames))
+                              this_mode, ref_frames))
         continue;
     }
 
@@ -2918,18 +2939,11 @@
 
     if (!disable_skip) {
       if (skippable) {
-        vp9_prob skip_prob = vp9_get_skip_prob(cm, xd);
-
         // Back out the coefficient coding costs
         rate2 -= (rate_y + rate_uv);
-        // for best yrd calculation
-        rate_uv = 0;
 
         // Cost the skip mb case
-        if (skip_prob) {
-          int prob_skip_cost = vp9_cost_bit(skip_prob, 1);
-          rate2 += prob_skip_cost;
-        }
+        rate2 += vp9_cost_bit(vp9_get_skip_prob(cm, xd), 1);
       } else if (ref_frame != INTRA_FRAME && !xd->lossless) {
         if (RDCOST(x->rdmult, x->rddiv, rate_y + rate_uv, distortion2) <
             RDCOST(x->rdmult, x->rddiv, 0, total_sse)) {
@@ -2941,8 +2955,6 @@
           distortion2 = total_sse;
           assert(total_sse >= 0);
           rate2 -= (rate_y + rate_uv);
-          rate_y = 0;
-          rate_uv = 0;
           this_skip2 = 1;
         }
       } else {
@@ -2960,12 +2972,6 @@
         best_intra_rd = this_rd;
         best_intra_mode = mbmi->mode;
       }
-    } else {
-      // Keep record of best inter rd with single reference
-      if (!comp_pred && !mode_excluded && this_rd < best_inter_rd) {
-        best_inter_rd = this_rd;
-        best_inter_ref_frame = ref_frame;
-      }
     }
 
     if (!disable_skip && ref_frame == INTRA_FRAME) {
@@ -2987,7 +2993,7 @@
           mbmi->mv[0].as_int = 0;
           max_plane = 1;
         } else {
-          best_intra_rd = x->pred_sse[ref_frame];
+          best_pred_sse = x->pred_sse[ref_frame];
         }
 
         *returnrate = rate2;
@@ -3037,13 +3043,11 @@
       hybrid_rd = RDCOST(x->rdmult, x->rddiv, hybrid_rate, distortion2);
 
       if (!comp_pred) {
-        if (single_rd < best_pred_rd[SINGLE_REFERENCE]) {
+        if (single_rd < best_pred_rd[SINGLE_REFERENCE])
           best_pred_rd[SINGLE_REFERENCE] = single_rd;
-        }
       } else {
-        if (single_rd < best_pred_rd[COMPOUND_REFERENCE]) {
+        if (single_rd < best_pred_rd[COMPOUND_REFERENCE])
           best_pred_rd[COMPOUND_REFERENCE] = single_rd;
-        }
       }
       if (hybrid_rd < best_pred_rd[REFERENCE_MODE_SELECT])
         best_pred_rd[REFERENCE_MODE_SELECT] = hybrid_rd;
@@ -3124,7 +3128,7 @@
   // If we used an estimate for the uv intra rd in the loop above...
   if (cpi->sf.use_uv_intra_rd_estimate) {
     // Do Intra UV best rd mode selection if best mode choice above was intra.
-    if (vp9_mode_order[best_mode_index].ref_frame[0] == INTRA_FRAME) {
+    if (best_mbmode.ref_frame[0] == INTRA_FRAME) {
       TX_SIZE uv_tx_size;
       *mbmi = best_mbmode;
       uv_tx_size = get_uv_tx_size(mbmi, &xd->plane[1]);
@@ -3141,7 +3145,8 @@
          (cm->interp_filter == best_mbmode.interp_filter) ||
          !is_inter_block(&best_mbmode));
 
-  update_rd_thresh_fact(cpi, bsize, best_mode_index);
+  if (!cpi->rc.is_src_frame_alt_ref)
+    update_rd_thresh_fact(cpi, bsize, best_mode_index);
 
   // macroblock modes
   *mbmi = best_mbmode;
@@ -3174,7 +3179,10 @@
     vp9_zero(best_tx_diff);
   }
 
-  set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
+  // TODO(yunqingwang): Moving this line in front of the above best_filter_diff
+  // updating code causes PSNR loss. Need to figure out the confliction.
+  x->skip |= best_mode_skippable;
+
   store_coding_context(x, ctx, best_mode_index, best_pred_diff,
                        best_tx_diff, best_filter_diff, best_mode_skippable);
 
@@ -3190,7 +3198,7 @@
   VP9_COMMON *const cm = &cpi->common;
   RD_OPT *const rd_opt = &cpi->rd;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   unsigned char segment_id = mbmi->segment_id;
   const int comp_pred = 0;
   int i;
@@ -3297,7 +3305,7 @@
   VP9_COMMON *const cm = &cpi->common;
   RD_OPT *const rd_opt = &cpi->rd;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const struct segmentation *const seg = &cm->seg;
   MV_REFERENCE_FRAME ref_frame, second_ref_frame;
   unsigned char segment_id = mbmi->segment_id;
@@ -3317,18 +3325,17 @@
   int ref_index, best_ref_index = 0;
   unsigned int ref_costs_single[MAX_REF_FRAMES], ref_costs_comp[MAX_REF_FRAMES];
   vp9_prob comp_mode_p;
-  int64_t best_inter_rd = INT64_MAX;
-  MV_REFERENCE_FRAME best_inter_ref_frame = LAST_FRAME;
   INTERP_FILTER tmp_best_filter = SWITCHABLE;
   int rate_uv_intra, rate_uv_tokenonly;
   int64_t dist_uv;
   int skip_uv;
   PREDICTION_MODE mode_uv = DC_PRED;
-  int intra_cost_penalty = 20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q);
+  const int intra_cost_penalty =
+      20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q, cm->bit_depth);
   int_mv seg_mvs[4][MAX_REF_FRAMES];
   b_mode_info best_bmodes[4];
   int best_skip2 = 0;
-  int mode_skip_mask = 0;
+  int ref_frame_skip_mask[2] = { 0 };
 
   x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
   vpx_memset(x->zcoeff_blk[TX_4X4], 0, 4);
@@ -3357,6 +3364,9 @@
                              ref_frame, bsize, mi_row, mi_col,
                              frame_mv[NEARESTMV], frame_mv[NEARMV],
                              yv12_mb);
+    } else {
+      ref_frame_skip_mask[0] |= (1 << ref_frame);
+      ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
     }
     frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
     frame_mv[ZEROMV][ref_frame].as_int = 0;
@@ -3382,18 +3392,19 @@
     // skip mask to look at a subset of the remaining modes.
     if (ref_index > 2 && cpi->sf.mode_skip_start < MAX_MODES) {
       if (ref_index == 3) {
-        switch (vp9_ref_order[best_ref_index].ref_frame[0]) {
+        switch (best_mbmode.ref_frame[0]) {
           case INTRA_FRAME:
-            mode_skip_mask = 0;
             break;
           case LAST_FRAME:
-            mode_skip_mask = 0x0010;
+            ref_frame_skip_mask[0] |= (1 << GOLDEN_FRAME) | (1 << ALTREF_FRAME);
+            ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
             break;
           case GOLDEN_FRAME:
-            mode_skip_mask = 0x0008;
+            ref_frame_skip_mask[0] |= (1 << LAST_FRAME) | (1 << ALTREF_FRAME);
+            ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
             break;
           case ALTREF_FRAME:
-            mode_skip_mask = 0x0000;
+            ref_frame_skip_mask[0] |= (1 << GOLDEN_FRAME) | (1 << LAST_FRAME);
             break;
           case NONE:
           case MAX_REF_FRAMES:
@@ -3401,38 +3412,31 @@
             break;
         }
       }
-      if (mode_skip_mask & (1 << ref_index))
-        continue;
     }
 
+    if (ref_frame_skip_mask[0] & (1 << ref_frame) &&
+        ref_frame_skip_mask[1] & (1 << MAX(0, second_ref_frame)))
+      continue;
+
     // Test best rd so far against threshold for trying this mode.
     if (rd_less_than_thresh(best_rd,
                             rd_opt->threshes[segment_id][bsize][ref_index],
                             rd_opt->thresh_freq_fact[bsize][ref_index]))
       continue;
 
-    if (ref_frame > INTRA_FRAME &&
-        !(cpi->ref_frame_flags & flag_list[ref_frame])) {
-      continue;
-    }
-
     comp_pred = second_ref_frame > INTRA_FRAME;
     if (comp_pred) {
       if (!cm->allow_comp_inter_inter)
         continue;
-
       if (!(cpi->ref_frame_flags & flag_list[second_ref_frame]))
         continue;
       // Do not allow compound prediction if the segment level reference frame
       // feature is in use as in this case there can only be one reference.
       if (vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME))
         continue;
+
       if ((cpi->sf.mode_search_skip_flags & FLAG_SKIP_COMP_BESTINTRA) &&
-          vp9_ref_order[best_ref_index].ref_frame[0] == INTRA_FRAME)
-        continue;
-      if ((cpi->sf.mode_search_skip_flags & FLAG_SKIP_COMP_REFMISMATCH) &&
-          ref_frame != best_inter_ref_frame &&
-          second_ref_frame != best_inter_ref_frame)
+          best_mbmode.ref_frame[0] == INTRA_FRAME)
         continue;
     }
 
@@ -3454,14 +3458,12 @@
     // If the segment reference frame feature is enabled....
     // then do nothing if the current ref frame is not allowed..
     if (vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
-        vp9_get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) !=
-            (int)ref_frame) {
+        vp9_get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) != (int)ref_frame) {
       continue;
     // Disable this drop out case if the ref frame
     // segment level feature is enabled for this segment. This is to
     // prevent the possibility that we end up unable to pick any mode.
-    } else if (!vp9_segfeature_active(seg, segment_id,
-                                      SEG_LVL_REF_FRAME)) {
+    } else if (!vp9_segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME)) {
       // Only consider ZEROMV/ALTREF_FRAME for alt ref frame,
       // unless ARNR filtering is enabled in which case we want
       // an unfiltered alternative. We allow near/nearest as well
@@ -3590,7 +3592,7 @@
               tmp_best_skippable = skippable;
               tmp_best_mbmode = *mbmi;
               for (i = 0; i < 4; i++) {
-                tmp_best_bmodes[i] = xd->mi[0]->bmi[i];
+                tmp_best_bmodes[i] = xd->mi[0].src_mi->bmi[i];
                 x->zcoeff_blk[TX_4X4][i] = !x->plane[0].eobs[i];
               }
               pred_exists = 1;
@@ -3634,7 +3636,7 @@
         skippable = tmp_best_skippable;
         *mbmi = tmp_best_mbmode;
         for (i = 0; i < 4; i++)
-          xd->mi[0]->bmi[i] = tmp_best_bmodes[i];
+          xd->mi[0].src_mi->bmi[i] = tmp_best_bmodes[i];
       }
 
       rate2 += rate;
@@ -3708,15 +3710,6 @@
       this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
     }
 
-    // Keep record of best inter rd with single reference
-    if (is_inter_block(mbmi) &&
-        !has_second_ref(mbmi) &&
-        !mode_excluded &&
-        this_rd < best_inter_rd) {
-      best_inter_rd = this_rd;
-      best_inter_ref_frame = ref_frame;
-    }
-
     if (!disable_skip && ref_frame == INTRA_FRAME) {
       for (i = 0; i < REFERENCE_MODES; ++i)
         best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
@@ -3750,7 +3743,7 @@
                    sizeof(uint8_t) * ctx->num_4x4_blk);
 
         for (i = 0; i < 4; i++)
-          best_bmodes[i] = xd->mi[0]->bmi[i];
+          best_bmodes[i] = xd->mi[0].src_mi->bmi[i];
 
         // TODO(debargha): enhance this test with a better distortion prediction
         // based on qp, activity mask and history
@@ -3786,11 +3779,11 @@
       single_rd = RDCOST(x->rdmult, x->rddiv, single_rate, distortion2);
       hybrid_rd = RDCOST(x->rdmult, x->rddiv, hybrid_rate, distortion2);
 
-      if (!comp_pred && single_rd < best_pred_rd[SINGLE_REFERENCE]) {
+      if (!comp_pred && single_rd < best_pred_rd[SINGLE_REFERENCE])
         best_pred_rd[SINGLE_REFERENCE] = single_rd;
-      } else if (comp_pred && single_rd < best_pred_rd[COMPOUND_REFERENCE]) {
+      else if (comp_pred && single_rd < best_pred_rd[COMPOUND_REFERENCE])
         best_pred_rd[COMPOUND_REFERENCE] = single_rd;
-      }
+
       if (hybrid_rd < best_pred_rd[REFERENCE_MODE_SELECT])
         best_pred_rd[REFERENCE_MODE_SELECT] = hybrid_rd;
     }
@@ -3831,7 +3824,7 @@
   // If we used an estimate for the uv intra rd in the loop above...
   if (cpi->sf.use_uv_intra_rd_estimate) {
     // Do Intra UV best rd mode selection if best mode choice above was intra.
-    if (vp9_ref_order[best_ref_index].ref_frame[0] == INTRA_FRAME) {
+    if (best_mbmode.ref_frame[0] == INTRA_FRAME) {
       *mbmi = best_mbmode;
       rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv_intra,
                               &rate_uv_tokenonly,
@@ -3858,13 +3851,14 @@
   x->skip |= best_skip2;
   if (!is_inter_block(&best_mbmode)) {
     for (i = 0; i < 4; i++)
-      xd->mi[0]->bmi[i].as_mode = best_bmodes[i].as_mode;
+      xd->mi[0].src_mi->bmi[i].as_mode = best_bmodes[i].as_mode;
   } else {
     for (i = 0; i < 4; ++i)
-      vpx_memcpy(&xd->mi[0]->bmi[i], &best_bmodes[i], sizeof(b_mode_info));
+      vpx_memcpy(&xd->mi[0].src_mi->bmi[i], &best_bmodes[i],
+                 sizeof(b_mode_info));
 
-    mbmi->mv[0].as_int = xd->mi[0]->bmi[3].as_mv[0].as_int;
-    mbmi->mv[1].as_int = xd->mi[0]->bmi[3].as_mv[1].as_int;
+    mbmi->mv[0].as_int = xd->mi[0].src_mi->bmi[3].as_mv[0].as_int;
+    mbmi->mv[1].as_int = xd->mi[0].src_mi->bmi[3].as_mv[1].as_int;
   }
 
   for (i = 0; i < REFERENCE_MODES; ++i) {
@@ -3887,9 +3881,9 @@
     vp9_zero(best_filter_diff);
   }
 
-  set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
   store_coding_context(x, ctx, best_ref_index,
                        best_pred_diff, best_tx_diff, best_filter_diff, 0);
 
   return best_rd;
 }
+
diff --git a/vp9/encoder/vp9_resize.c b/vp9/encoder/vp9_resize.c
index 4e6efae..4a8a521 100644
--- a/vp9/encoder/vp9_resize.c
+++ b/vp9/encoder/vp9_resize.c
@@ -312,7 +312,7 @@
 static void down2_symeven(const uint8_t *const input, int length,
                           uint8_t *output) {
   // Actual filter len = 2 * filter_len_half.
-  static const int16_t *filter = vp9_down2_symeven_half_filter;
+  const int16_t *filter = vp9_down2_symeven_half_filter;
   const int filter_len_half = sizeof(vp9_down2_symeven_half_filter) / 2;
   int i, j;
   uint8_t *optr = output;
@@ -368,7 +368,7 @@
 static void down2_symodd(const uint8_t *const input, int length,
                          uint8_t *output) {
   // Actual filter len = 2 * filter_len_half - 1.
-  static const int16_t *filter = vp9_down2_symodd_half_filter;
+  const int16_t *filter = vp9_down2_symodd_half_filter;
   const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
   int i, j;
   uint8_t *optr = output;
@@ -529,6 +529,302 @@
   free(arrbuf);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static void highbd_interpolate(const uint16_t *const input, int inlength,
+                               uint16_t *output, int outlength, int bd) {
+  const int64_t delta =
+      (((uint64_t)inlength << 32) + outlength / 2) / outlength;
+  const int64_t offset = inlength > outlength ?
+      (((int64_t)(inlength - outlength) << 31) + outlength / 2) / outlength :
+      -(((int64_t)(outlength - inlength) << 31) + outlength / 2) / outlength;
+  uint16_t *optr = output;
+  int x, x1, x2, sum, k, int_pel, sub_pel;
+  int64_t y;
+
+  const interp_kernel *interp_filters =
+      choose_interp_filter(inlength, outlength);
+
+  x = 0;
+  y = offset;
+  while ((y >> INTERP_PRECISION_BITS) < (INTERP_TAPS / 2 - 1)) {
+    x++;
+    y += delta;
+  }
+  x1 = x;
+  x = outlength - 1;
+  y = delta * x + offset;
+  while ((y >> INTERP_PRECISION_BITS) +
+         (int64_t)(INTERP_TAPS / 2) >= inlength) {
+    x--;
+    y -= delta;
+  }
+  x2 = x;
+  if (x1 > x2) {
+    for (x = 0, y = offset; x < outlength; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k) {
+        const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
+        sum += filter[k] *
+            input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
+      }
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+  } else {
+    // Initial part.
+    for (x = 0, y = offset; x < x1; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k)
+        sum += filter[k] *
+            input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0 ?
+                   0 : int_pel - INTERP_TAPS / 2 + 1 + k)];
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+    // Middle part.
+    for (; x <= x2; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k)
+        sum += filter[k] * input[int_pel - INTERP_TAPS / 2 + 1 + k];
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+    // End part.
+    for (; x < outlength; ++x, y += delta) {
+      const int16_t *filter;
+      int_pel = y >> INTERP_PRECISION_BITS;
+      sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
+      filter = interp_filters[sub_pel];
+      sum = 0;
+      for (k = 0; k < INTERP_TAPS; ++k)
+        sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >=
+                                  inlength ?  inlength - 1 :
+                                  int_pel - INTERP_TAPS / 2 + 1 + k)];
+      *optr++ = clip_pixel_high(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+    }
+  }
+}
+
+static void highbd_down2_symeven(const uint16_t *const input, int length,
+                                 uint16_t *output, int bd) {
+  // Actual filter len = 2 * filter_len_half.
+  static const int16_t *filter = vp9_down2_symeven_half_filter;
+  const int filter_len_half = sizeof(vp9_down2_symeven_half_filter) / 2;
+  int i, j;
+  uint16_t *optr = output;
+  int l1 = filter_len_half;
+  int l2 = (length - filter_len_half);
+  l1 += (l1 & 1);
+  l2 += (l2 & 1);
+  if (l1 > l2) {
+    // Short input length.
+    for (i = 0; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] +
+                input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  } else {
+    // Initial part.
+    for (i = 0; i < l1; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // Middle part.
+    for (; i < l2; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[i - j] + input[i + 1 + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // End part.
+    for (; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1));
+      for (j = 0; j < filter_len_half; ++j) {
+        sum += (input[i - j] +
+                input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  }
+}
+
+static void highbd_down2_symodd(const uint16_t *const input, int length,
+                              uint16_t *output, int bd) {
+  // Actual filter len = 2 * filter_len_half - 1.
+  static const int16_t *filter = vp9_down2_symodd_half_filter;
+  const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
+  int i, j;
+  uint16_t *optr = output;
+  int l1 = filter_len_half - 1;
+  int l2 = (length - filter_len_half + 1);
+  l1 += (l1 & 1);
+  l2 += (l2 & 1);
+  if (l1 > l2) {
+    // Short input length.
+    for (i = 0; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] +
+                input[(i + j >= length ? length - 1 : i + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  } else {
+    // Initial part.
+    for (i = 0; i < l1; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // Middle part.
+    for (; i < l2; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[i - j] + input[i + j]) * filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+    // End part.
+    for (; i < length; i += 2) {
+      int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
+      for (j = 1; j < filter_len_half; ++j) {
+        sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
+            filter[j];
+      }
+      sum >>= FILTER_BITS;
+      *optr++ = clip_pixel_high(sum, bd);
+    }
+  }
+}
+
+static void highbd_resize_multistep(const uint16_t *const input,
+                                    int length,
+                                    uint16_t *output,
+                                    int olength,
+                                    uint16_t *buf,
+                                    int bd) {
+  int steps;
+  if (length == olength) {
+    memcpy(output, input, sizeof(uint16_t) * length);
+    return;
+  }
+  steps = get_down2_steps(length, olength);
+
+  if (steps > 0) {
+    int s;
+    uint16_t *out = NULL;
+    uint16_t *tmpbuf = NULL;
+    uint16_t *otmp, *otmp2;
+    int filteredlength = length;
+    if (!tmpbuf) {
+      tmpbuf = (uint16_t *)malloc(sizeof(uint16_t) * length);
+      otmp = tmpbuf;
+    } else {
+      otmp = buf;
+    }
+    otmp2 = otmp + get_down2_length(length, 1);
+    for (s = 0; s < steps; ++s) {
+      const int proj_filteredlength = get_down2_length(filteredlength, 1);
+      const uint16_t *const in = (s == 0 ? input : out);
+      if (s == steps - 1 && proj_filteredlength == olength)
+        out = output;
+      else
+        out = (s & 1 ? otmp2 : otmp);
+      if (filteredlength & 1)
+        highbd_down2_symodd(in, filteredlength, out, bd);
+      else
+        highbd_down2_symeven(in, filteredlength, out, bd);
+      filteredlength = proj_filteredlength;
+    }
+    if (filteredlength != olength) {
+      highbd_interpolate(out, filteredlength, output, olength, bd);
+    }
+    if (tmpbuf)
+      free(tmpbuf);
+  } else {
+    highbd_interpolate(input, length, output, olength, bd);
+  }
+}
+
+static void highbd_fill_col_to_arr(uint16_t *img, int stride, int len,
+                                   uint16_t *arr) {
+  int i;
+  uint16_t *iptr = img;
+  uint16_t *aptr = arr;
+  for (i = 0; i < len; ++i, iptr += stride) {
+    *aptr++ = *iptr;
+  }
+}
+
+static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len,
+                                   uint16_t *arr) {
+  int i;
+  uint16_t *iptr = img;
+  uint16_t *aptr = arr;
+  for (i = 0; i < len; ++i, iptr += stride) {
+    *iptr = *aptr++;
+  }
+}
+
+void vp9_highbd_resize_plane(const uint8_t *const input,
+                             int height,
+                             int width,
+                             int in_stride,
+                             uint8_t *output,
+                             int height2,
+                             int width2,
+                             int out_stride,
+                             int bd) {
+  int i;
+  uint16_t *intbuf = (uint16_t *)malloc(sizeof(uint16_t) * width2 * height);
+  uint16_t *tmpbuf = (uint16_t *)malloc(sizeof(uint16_t) *
+                                        (width < height ? height : width));
+  uint16_t *arrbuf = (uint16_t *)malloc(sizeof(uint16_t) * (height + height2));
+  for (i = 0; i < height; ++i) {
+    highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width,
+                            intbuf + width2 * i, width2, tmpbuf, bd);
+  }
+  for (i = 0; i < width2; ++i) {
+    highbd_fill_col_to_arr(intbuf + i, width2, height, arrbuf);
+    highbd_resize_multistep(arrbuf, height, arrbuf + height, height2, tmpbuf,
+                            bd);
+    highbd_fill_arr_to_col(CONVERT_TO_SHORTPTR(output + i), out_stride, height2,
+                           arrbuf + height);
+  }
+  free(intbuf);
+  free(tmpbuf);
+  free(arrbuf);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 void vp9_resize_frame420(const uint8_t *const y,
                          int y_stride,
                          const uint8_t *const u, const uint8_t *const v,
@@ -574,3 +870,51 @@
   vp9_resize_plane(v, height, width, uv_stride,
                    ov, oheight, owidth, ouv_stride);
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_resize_frame420(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u, const uint8_t *const v,
+                                int uv_stride,
+                                int height, int width,
+                                uint8_t *oy, int oy_stride,
+                                uint8_t *ou, uint8_t *ov, int ouv_stride,
+                                int oheight, int owidth, int bd) {
+  vp9_highbd_resize_plane(y, height, width, y_stride,
+                          oy, oheight, owidth, oy_stride, bd);
+  vp9_highbd_resize_plane(u, height / 2, width / 2, uv_stride,
+                          ou, oheight / 2, owidth / 2, ouv_stride, bd);
+  vp9_highbd_resize_plane(v, height / 2, width / 2, uv_stride,
+                          ov, oheight / 2, owidth / 2, ouv_stride, bd);
+}
+
+void vp9_highbd_resize_frame422(const uint8_t *const y, int y_stride,
+                                const uint8_t *const u, const uint8_t *const v,
+                                int uv_stride,
+                                int height, int width,
+                                uint8_t *oy, int oy_stride,
+                                uint8_t *ou, uint8_t *ov, int ouv_stride,
+                                int oheight, int owidth, int bd) {
+  vp9_highbd_resize_plane(y, height, width, y_stride,
+                          oy, oheight, owidth, oy_stride, bd);
+  vp9_highbd_resize_plane(u, height, width / 2, uv_stride,
+                          ou, oheight, owidth / 2, ouv_stride, bd);
+  vp9_highbd_resize_plane(v, height, width / 2, uv_stride,
+                          ov, oheight, owidth / 2, ouv_stride, bd);
+}
+
+void vp9_highbd_resize_frame444(const uint8_t *const y, int y_stride,
+                                const uint8_t *const u, const uint8_t *const v,
+                                int uv_stride,
+                                int height, int width,
+                                uint8_t *oy, int oy_stride,
+                                uint8_t *ou, uint8_t *ov, int ouv_stride,
+                                int oheight, int owidth, int bd) {
+  vp9_highbd_resize_plane(y, height, width, y_stride,
+                          oy, oheight, owidth, oy_stride, bd);
+  vp9_highbd_resize_plane(u, height, width, uv_stride,
+                          ou, oheight, owidth, ouv_stride, bd);
+  vp9_highbd_resize_plane(v, height, width, uv_stride,
+                          ov, oheight, owidth, ouv_stride, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_resize.h b/vp9/encoder/vp9_resize.h
index 1818cd4..067af53 100644
--- a/vp9/encoder/vp9_resize.h
+++ b/vp9/encoder/vp9_resize.h
@@ -65,4 +65,60 @@
                          int oheight,
                          int owidth);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_resize_plane(const uint8_t *const input,
+                             int height,
+                             int width,
+                             int in_stride,
+                             uint8_t *output,
+                             int height2,
+                             int width2,
+                             int out_stride,
+                             int bd);
+void vp9_highbd_resize_frame420(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u,
+                                const uint8_t *const v,
+                                int uv_stride,
+                                int height,
+                                int width,
+                                uint8_t *oy,
+                                int oy_stride,
+                                uint8_t *ou,
+                                uint8_t *ov,
+                                int ouv_stride,
+                                int oheight,
+                                int owidth,
+                                int bd);
+void vp9_highbd_resize_frame422(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u,
+                                const uint8_t *const v,
+                                int uv_stride,
+                                int height,
+                                int width,
+                                uint8_t *oy,
+                                int oy_stride,
+                                uint8_t *ou,
+                                uint8_t *ov,
+                                int ouv_stride,
+                                int oheight,
+                                int owidth,
+                                int bd);
+void vp9_highbd_resize_frame444(const uint8_t *const y,
+                                int y_stride,
+                                const uint8_t *const u,
+                                const uint8_t *const v,
+                                int uv_stride,
+                                int height,
+                                int width,
+                                uint8_t *oy,
+                                int oy_stride,
+                                uint8_t *ou,
+                                uint8_t *ov,
+                                int ouv_stride,
+                                int oheight,
+                                int owidth,
+                                int bd);
+#endif    // CONFIG_VP9_HIGHBITDEPTH
 #endif    // VP9_ENCODER_VP9_RESIZE_H_
diff --git a/vp9/encoder/vp9_sad.c b/vp9/encoder/vp9_sad.c
index d062636..cee6ce1 100644
--- a/vp9/encoder/vp9_sad.c
+++ b/vp9/encoder/vp9_sad.c
@@ -14,6 +14,9 @@
 #include "./vpx_config.h"
 
 #include "vpx/vpx_integer.h"
+#if CONFIG_VP9_HIGHBITDEPTH
+#include "vp9/common/vp9_common.h"
+#endif
 #include "vp9/encoder/vp9_variance.h"
 
 static INLINE unsigned int sad(const uint8_t *a, int a_stride,
@@ -131,3 +134,138 @@
 sadMxNxK(4, 4, 3)
 sadMxNxK(4, 4, 8)
 sadMxNx4D(4, 4)
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE unsigned int high_sad(const uint8_t *a8, int a_stride,
+                                    const uint8_t *b8, int b_stride,
+                                    int width, int height) {
+  int y, x;
+  unsigned int sad = 0;
+  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+  for (y = 0; y < height; y++) {
+    for (x = 0; x < width; x++)
+      sad += abs(a[x] - b[x]);
+
+    a += a_stride;
+    b += b_stride;
+  }
+  return sad;
+}
+
+static INLINE unsigned int high_sadb(const uint8_t *a8, int a_stride,
+                                     const uint16_t *b, int b_stride,
+                                     int width, int height) {
+  int y, x;
+  unsigned int sad = 0;
+  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  for (y = 0; y < height; y++) {
+    for (x = 0; x < width; x++)
+      sad += abs(a[x] - b[x]);
+
+    a += a_stride;
+    b += b_stride;
+  }
+  return sad;
+}
+
+#define high_sadMxN(m, n) \
+unsigned int vp9_high_sad##m##x##n##_c(const uint8_t *src, int src_stride, \
+                                       const uint8_t *ref, int ref_stride) { \
+  return high_sad(src, src_stride, ref, ref_stride, m, n); \
+} \
+unsigned int vp9_high_sad##m##x##n##_avg_c(const uint8_t *src, int src_stride, \
+                                           const uint8_t *ref, int ref_stride, \
+                                           const uint8_t *second_pred) { \
+  uint16_t comp_pred[m * n]; \
+  vp9_high_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride); \
+  return high_sadb(src, src_stride, comp_pred, m, m, n); \
+}
+
+#define high_sadMxNxK(m, n, k) \
+void vp9_high_sad##m##x##n##x##k##_c(const uint8_t *src, int src_stride, \
+                                     const uint8_t *ref, int ref_stride, \
+                                     unsigned int *sads) { \
+  int i; \
+  for (i = 0; i < k; ++i) \
+    sads[i] = vp9_high_sad##m##x##n##_c(src, src_stride, &ref[i], ref_stride); \
+}
+
+#define high_sadMxNx4D(m, n) \
+void vp9_high_sad##m##x##n##x4d_c(const uint8_t *src, int src_stride, \
+                                  const uint8_t *const refs[], \
+                                  int ref_stride, unsigned int *sads) { \
+  int i; \
+  for (i = 0; i < 4; ++i) \
+    sads[i] = vp9_high_sad##m##x##n##_c(src, src_stride, refs[i], ref_stride); \
+}
+
+// 64x64
+high_sadMxN(64, 64)
+high_sadMxNxK(64, 64, 3)
+high_sadMxNxK(64, 64, 8)
+high_sadMxNx4D(64, 64)
+
+// 64x32
+high_sadMxN(64, 32)
+high_sadMxNx4D(64, 32)
+
+// 32x64
+high_sadMxN(32, 64)
+high_sadMxNx4D(32, 64)
+
+// 32x32
+high_sadMxN(32, 32)
+high_sadMxNxK(32, 32, 3)
+high_sadMxNxK(32, 32, 8)
+high_sadMxNx4D(32, 32)
+
+// 32x16
+high_sadMxN(32, 16)
+high_sadMxNx4D(32, 16)
+
+// 16x32
+high_sadMxN(16, 32)
+high_sadMxNx4D(16, 32)
+
+// 16x16
+high_sadMxN(16, 16)
+high_sadMxNxK(16, 16, 3)
+high_sadMxNxK(16, 16, 8)
+high_sadMxNx4D(16, 16)
+
+// 16x8
+high_sadMxN(16, 8)
+high_sadMxNxK(16, 8, 3)
+high_sadMxNxK(16, 8, 8)
+high_sadMxNx4D(16, 8)
+
+// 8x16
+high_sadMxN(8, 16)
+high_sadMxNxK(8, 16, 3)
+high_sadMxNxK(8, 16, 8)
+high_sadMxNx4D(8, 16)
+
+// 8x8
+high_sadMxN(8, 8)
+high_sadMxNxK(8, 8, 3)
+high_sadMxNxK(8, 8, 8)
+high_sadMxNx4D(8, 8)
+
+// 8x4
+high_sadMxN(8, 4)
+high_sadMxNxK(8, 4, 8)
+high_sadMxNx4D(8, 4)
+
+// 4x8
+high_sadMxN(4, 8)
+high_sadMxNxK(4, 8, 8)
+high_sadMxNx4D(4, 8)
+
+// 4x4
+high_sadMxN(4, 4)
+high_sadMxNxK(4, 4, 3)
+high_sadMxNxK(4, 4, 8)
+high_sadMxNx4D(4, 4)
+
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_segmentation.c b/vp9/encoder/vp9_segmentation.c
index d5676c3..f1d5177 100644
--- a/vp9/encoder/vp9_segmentation.c
+++ b/vp9/encoder/vp9_segmentation.c
@@ -111,7 +111,7 @@
 }
 
 static void count_segs(const VP9_COMMON *cm, MACROBLOCKD *xd,
-                       const TileInfo *tile, MODE_INFO **mi,
+                       const TileInfo *tile, MODE_INFO *mi,
                        int *no_pred_segcounts,
                        int (*temporal_predictor_count)[2],
                        int *t_unpred_seg_counts,
@@ -122,7 +122,7 @@
     return;
 
   xd->mi = mi;
-  segment_id = xd->mi[0]->mbmi.segment_id;
+  segment_id = xd->mi[0].src_mi->mbmi.segment_id;
 
   set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
 
@@ -131,7 +131,7 @@
 
   // Temporal prediction not allowed on key frames
   if (cm->frame_type != KEY_FRAME) {
-    const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
+    const BLOCK_SIZE bsize = xd->mi[0].src_mi->mbmi.sb_type;
     // Test to see if the segment id matches the predicted value.
     const int pred_segment_id = vp9_get_segment_id(cm, cm->last_frame_seg_map,
                                                    bsize, mi_row, mi_col);
@@ -140,7 +140,7 @@
 
     // Store the prediction status for this mb and update counts
     // as appropriate
-    xd->mi[0]->mbmi.seg_id_predicted = pred_flag;
+    xd->mi[0].src_mi->mbmi.seg_id_predicted = pred_flag;
     temporal_predictor_count[pred_context][pred_flag]++;
 
     // Update the "unpredicted" segment count
@@ -150,7 +150,7 @@
 }
 
 static void count_segs_sb(const VP9_COMMON *cm, MACROBLOCKD *xd,
-                          const TileInfo *tile, MODE_INFO **mi,
+                          const TileInfo *tile, MODE_INFO *mi,
                           int *no_pred_segcounts,
                           int (*temporal_predictor_count)[2],
                           int *t_unpred_seg_counts,
@@ -163,8 +163,8 @@
   if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
     return;
 
-  bw = num_8x8_blocks_wide_lookup[mi[0]->mbmi.sb_type];
-  bh = num_8x8_blocks_high_lookup[mi[0]->mbmi.sb_type];
+  bw = num_8x8_blocks_wide_lookup[mi[0].src_mi->mbmi.sb_type];
+  bh = num_8x8_blocks_high_lookup[mi[0].src_mi->mbmi.sb_type];
 
   if (bw == bs && bh == bs) {
     count_segs(cm, xd, tile, mi, no_pred_segcounts, temporal_predictor_count,
@@ -224,13 +224,13 @@
   // predicts this one
   for (tile_col = 0; tile_col < 1 << cm->log2_tile_cols; tile_col++) {
     TileInfo tile;
-    MODE_INFO **mi_ptr;
+    MODE_INFO *mi_ptr;
     vp9_tile_init(&tile, cm, 0, tile_col);
 
-    mi_ptr = cm->mi_grid_visible + tile.mi_col_start;
+    mi_ptr = cm->mi + tile.mi_col_start;
     for (mi_row = 0; mi_row < cm->mi_rows;
          mi_row += 8, mi_ptr += 8 * cm->mi_stride) {
-      MODE_INFO **mi = mi_ptr;
+      MODE_INFO *mi = mi_ptr;
       for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end;
            mi_col += 8, mi += 8)
         count_segs_sb(cm, xd, &tile, mi, no_pred_segcounts,
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index dbf4ae9..92e3149 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -13,43 +13,6 @@
 #include "vp9/encoder/vp9_encoder.h"
 #include "vp9/encoder/vp9_speed_features.h"
 
-enum {
-  INTRA_ALL       = (1 << DC_PRED) |
-                    (1 << V_PRED) | (1 << H_PRED) |
-                    (1 << D45_PRED) | (1 << D135_PRED) |
-                    (1 << D117_PRED) | (1 << D153_PRED) |
-                    (1 << D207_PRED) | (1 << D63_PRED) |
-                    (1 << TM_PRED),
-  INTRA_DC        = (1 << DC_PRED),
-  INTRA_DC_TM     = (1 << DC_PRED) | (1 << TM_PRED),
-  INTRA_DC_H_V    = (1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED),
-  INTRA_DC_TM_H_V = (1 << DC_PRED) | (1 << TM_PRED) | (1 << V_PRED) |
-                    (1 << H_PRED)
-};
-
-enum {
-  INTER_ALL = (1 << NEARESTMV) | (1 << NEARMV) | (1 << ZEROMV) | (1 << NEWMV),
-  INTER_NEAREST = (1 << NEARESTMV),
-  INTER_NEAREST_NEAR_NEW = (1 << NEARESTMV) | (1 << NEARMV) | (1 << NEWMV)
-};
-
-enum {
-  DISABLE_ALL_INTER_SPLIT   = (1 << THR_COMP_GA) |
-                              (1 << THR_COMP_LA) |
-                              (1 << THR_ALTR) |
-                              (1 << THR_GOLD) |
-                              (1 << THR_LAST),
-
-  DISABLE_ALL_SPLIT         = (1 << THR_INTRA) | DISABLE_ALL_INTER_SPLIT,
-
-  DISABLE_COMPOUND_SPLIT    = (1 << THR_COMP_GA) | (1 << THR_COMP_LA),
-
-  LAST_AND_INTRA_SPLIT_ONLY = (1 << THR_COMP_GA) |
-                              (1 << THR_COMP_LA) |
-                              (1 << THR_ALTR) |
-                              (1 << THR_GOLD)
-};
-
 // Intra only frames, golden frames (except alt ref overlays) and
 // alt ref frames tend to be coded at a higher than ambient quality
 static int frame_is_boosted(const VP9_COMP *cpi) {
@@ -104,15 +67,11 @@
                                                       : USE_LARGESTALL;
 
     if (MIN(cm->width, cm->height) >= 720) {
-      sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
-      sf->last_partitioning_redo_frequency = 3;
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
       sf->adaptive_pred_interp_filter = 0;
     } else {
       sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
-      sf->last_partitioning_redo_frequency = 2;
-      sf->lf_motion_threshold = NO_MOTION_THRESHOLD;
     }
 
     sf->reference_masking = 1;
@@ -123,8 +82,6 @@
     sf->disable_filter_search_var_thresh = 100;
     sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
     sf->auto_min_max_partition_size = CONSTRAIN_NEIGHBORING_MIN_MAX;
-    sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
-    sf->adjust_partitioning_from_last_frame = 1;
 
     if (MIN(cm->width, cm->height) >= 720)
       sf->partition_search_breakout_dist_thr = (1 << 24);
@@ -138,29 +95,26 @@
                                                         : USE_LARGESTALL;
     if (MIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = DISABLE_ALL_SPLIT;
+      sf->schedule_mode_search = cm->base_qindex < 220 ? 1 : 0;
+      sf->partition_search_breakout_dist_thr = (1 << 25);
     } else {
       sf->max_intra_bsize = BLOCK_32X32;
       sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT;
+      sf->schedule_mode_search = cm->base_qindex < 175 ? 1 : 0;
+      sf->partition_search_breakout_dist_thr = (1 << 23);
     }
+    sf->mv.subpel_search_method = SUBPEL_TREE_PRUNED;
     sf->adaptive_pred_interp_filter = 0;
     sf->adaptive_mode_search = 1;
     sf->cb_partition_search = !boosted;
     sf->cb_pred_filter_search = 1;
     sf->alt_ref_search_fp = 1;
-    sf->motion_field_mode_search = !boosted;
-    sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
-    sf->last_partitioning_redo_frequency = 2;
     sf->recode_loop = ALLOW_RECODE_KFMAXBW;
     sf->adaptive_rd_thresh = 3;
     sf->mode_skip_start = 6;
     sf->intra_y_mode_mask[TX_32X32] = INTRA_DC;
     sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC;
     sf->adaptive_interp_filter_search = 1;
-
-    if (MIN(cm->width, cm->height) >= 720)
-      sf->partition_search_breakout_dist_thr = (1 << 25);
-    else
-      sf->partition_search_breakout_dist_thr = (1 << 23);
     sf->partition_search_breakout_rate_thr = 1000;
   }
 
@@ -168,14 +122,15 @@
     sf->use_square_partition_only = 1;
     sf->tx_size_search_method = USE_LARGESTALL;
     sf->disable_split_mask = DISABLE_ALL_SPLIT;
+    sf->mv.search_method = BIGDIA;
+    sf->mv.subpel_search_method = SUBPEL_TREE_PRUNED;
     sf->adaptive_rd_thresh = 4;
-    sf->mode_search_skip_flags |= FLAG_SKIP_COMP_REFMISMATCH |
-                                  FLAG_EARLY_TERMINATE;
+    sf->mode_search_skip_flags |= FLAG_EARLY_TERMINATE;
     sf->disable_filter_search_var_thresh = 200;
-    sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
     sf->use_lp32x32fdct = 1;
     sf->use_fast_coef_updates = ONE_LOOP_REDUCED;
     sf->use_fast_coef_costing = 1;
+    sf->motion_field_mode_search = !boosted;
 
     if (MIN(cm->width, cm->height) >= 720)
       sf->partition_search_breakout_dist_thr = (1 << 26);
@@ -421,6 +376,7 @@
   sf->use_fast_coef_updates = TWO_LOOP;
   sf->use_fast_coef_costing = 0;
   sf->mode_skip_start = MAX_MODES;  // Mode index at which mode skip mask set
+  sf->schedule_mode_search = 0;
   sf->use_nonrd_pick_mode = 0;
   for (i = 0; i < BLOCK_SIZES; ++i)
     sf->inter_mode_mask[i] = INTER_ALL;
diff --git a/vp9/encoder/vp9_speed_features.h b/vp9/encoder/vp9_speed_features.h
index 33c441f..ed84008 100644
--- a/vp9/encoder/vp9_speed_features.h
+++ b/vp9/encoder/vp9_speed_features.h
@@ -17,6 +17,44 @@
 extern "C" {
 #endif
 
+enum {
+  INTRA_ALL       = (1 << DC_PRED) |
+                    (1 << V_PRED) | (1 << H_PRED) |
+                    (1 << D45_PRED) | (1 << D135_PRED) |
+                    (1 << D117_PRED) | (1 << D153_PRED) |
+                    (1 << D207_PRED) | (1 << D63_PRED) |
+                    (1 << TM_PRED),
+  INTRA_DC        = (1 << DC_PRED),
+  INTRA_DC_TM     = (1 << DC_PRED) | (1 << TM_PRED),
+  INTRA_DC_H_V    = (1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED),
+  INTRA_DC_TM_H_V = (1 << DC_PRED) | (1 << TM_PRED) | (1 << V_PRED) |
+                    (1 << H_PRED)
+};
+
+enum {
+  INTER_ALL = (1 << NEARESTMV) | (1 << NEARMV) | (1 << ZEROMV) | (1 << NEWMV),
+  INTER_NEAREST = (1 << NEARESTMV),
+  INTER_NEAREST_NEAR_NEW = (1 << NEARESTMV) | (1 << NEARMV) | (1 << NEWMV),
+  INTER_NEAREST_NEAR_ZERO = (1 << NEARESTMV) | (1 << NEARMV) | (1 << ZEROMV),
+};
+
+enum {
+  DISABLE_ALL_INTER_SPLIT   = (1 << THR_COMP_GA) |
+                              (1 << THR_COMP_LA) |
+                              (1 << THR_ALTR) |
+                              (1 << THR_GOLD) |
+                              (1 << THR_LAST),
+
+  DISABLE_ALL_SPLIT         = (1 << THR_INTRA) | DISABLE_ALL_INTER_SPLIT,
+
+  DISABLE_COMPOUND_SPLIT    = (1 << THR_COMP_GA) | (1 << THR_COMP_LA),
+
+  LAST_AND_INTRA_SPLIT_ONLY = (1 << THR_COMP_GA) |
+                              (1 << THR_COMP_LA) |
+                              (1 << THR_ALTR) |
+                              (1 << THR_GOLD)
+};
+
 typedef enum {
   DIAMOND = 0,
   NSTEP = 1,
@@ -87,11 +125,6 @@
   // Skips comp inter modes if the best so far is an intra mode.
   FLAG_SKIP_COMP_BESTINTRA = 1 << 1,
 
-  // Skips comp inter modes if the best single intermode so far does
-  // not have the same reference as one of the two references being
-  // tested.
-  FLAG_SKIP_COMP_REFMISMATCH = 1 << 2,
-
   // Skips oblique intra modes if the best so far is an inter mode.
   FLAG_SKIP_INTRA_BESTINTER = 1 << 3,
 
@@ -285,6 +318,8 @@
   // point for this motion search and limits the search range around it.
   int adaptive_motion_search;
 
+  int schedule_mode_search;
+
   // Allows sub 8x8 modes to use the prediction filter that was determined
   // best for 8x8 mode. If set to 0 we always re check all the filters for
   // sizes less than 8x8, 1 means we check all filter modes if no 8x8 filter
diff --git a/vp9/encoder/vp9_ssim.c b/vp9/encoder/vp9_ssim.c
index 8435640..d02d0c6 100644
--- a/vp9/encoder/vp9_ssim.c
+++ b/vp9/encoder/vp9_ssim.c
@@ -43,6 +43,44 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_ssim_parms_8x8_c(uint16_t *s, int sp, uint16_t *r, int rp,
+                                 uint32_t *sum_s, uint32_t *sum_r,
+                                 uint32_t *sum_sq_s, uint32_t *sum_sq_r,
+                                 uint32_t *sum_sxr) {
+  int i, j;
+  for (i = 0; i < 8; i++, s += sp, r += rp) {
+    for (j = 0; j < 8; j++) {
+      *sum_s += s[j];
+      *sum_r += r[j];
+      *sum_sq_s += s[j] * s[j];
+      *sum_sq_r += r[j] * r[j];
+      *sum_sxr += s[j] * r[j];
+    }
+  }
+}
+
+void vp9_highbd_ssim_parms_8x8_shift_c(uint16_t *s, int sp, uint16_t *r, int rp,
+                                       uint32_t *sum_s, uint32_t *sum_r,
+                                       uint32_t *sum_sq_s, uint32_t *sum_sq_r,
+                                       uint32_t *sum_sxr, unsigned int bd,
+                                       unsigned int shift) {
+  int i, j;
+  const int max_val = (1 << bd) - 1;
+  for (i = 0; i < 8; i++, s += sp, r += rp) {
+    for (j = 0; j < 8; j++) {
+      int sj = s[j];
+      int rj = r[j];
+      *sum_s += sj;
+      *sum_r += rj;
+      *sum_sq_s += sj * sj;
+      *sum_sq_r += rj * rj;
+      *sum_sxr += sj * rj;
+    }
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static const int64_t cc1 =  26634;  // (64^2*(.01*255)^2
 static const int64_t cc2 = 239708;  // (64^2*(.03*255)^2
 
@@ -73,6 +111,37 @@
   return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 64);
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static double high_ssim_8x8_shift(uint16_t *s, int sp, uint16_t *r, int rp,
+                                  unsigned int bd, unsigned int shift) {
+  uint32_t sum_s = 0, sum_r = 0, sum_sq_s = 0, sum_sq_r = 0, sum_sxr = 0;
+  const int oshift = bd - 8;
+  vp9_highbd_ssim_parms_8x8_shift(s, sp, r, rp, &sum_s,
+                                  &sum_r, &sum_sq_s, &sum_sq_r,
+                                  &sum_sxr, bd, shift);
+  return similarity(sum_s >> oshift,
+                    sum_r >> oshift,
+                    sum_sq_s >> (2 * oshift),
+                    sum_sq_r >> (2 * oshift),
+                    sum_sxr >> (2 * oshift),
+                    64);
+}
+
+static double high_ssim_8x8(uint16_t *s, int sp, uint16_t *r, int rp,
+                            unsigned int bd) {
+  uint32_t sum_s = 0, sum_r = 0, sum_sq_s = 0, sum_sq_r = 0, sum_sxr = 0;
+  const int oshift = bd - 8;
+  vp9_highbd_ssim_parms_8x8(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r,
+                            &sum_sxr);
+  return similarity(sum_s >> oshift,
+                    sum_r >> oshift,
+                    sum_sq_s >> (2 * oshift),
+                    sum_sq_r >> (2 * oshift),
+                    sum_sxr >> (2 * oshift),
+                    64);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 // We are using a 8x8 moving window with starting location of each 8x8 window
 // on the 4x4 pixel grid. Such arrangement allows the windows to overlap
 // block boundaries to penalize blocking artifacts.
@@ -94,6 +163,47 @@
   ssim_total /= samples;
   return ssim_total;
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+double vp9_highbd_ssim2(uint8_t *img1, uint8_t *img2, int stride_img1,
+                        int stride_img2, int width, int height,
+                        unsigned int bd, unsigned int shift) {
+  int i, j;
+  int samples = 0;
+  double ssim_total = 0;
+
+  if (shift) {
+    // sample point start with each 4x4 location
+    for (i = 0; i <= height - 8;
+         i += 4, img1 += stride_img1 * 4, img2 += stride_img2 * 4) {
+      for (j = 0; j <= width - 8; j += 4) {
+        double v = high_ssim_8x8_shift(CONVERT_TO_SHORTPTR(img1 + j),
+                                       stride_img1,
+                                       CONVERT_TO_SHORTPTR(img2 + j),
+                                       stride_img2,
+                                       bd, shift);
+        ssim_total += v;
+        samples++;
+      }
+    }
+  } else {
+    // sample point start with each 4x4 location
+    for (i = 0; i <= height - 8;
+         i += 4, img1 += stride_img1 * 4, img2 += stride_img2 * 4) {
+      for (j = 0; j <= width - 8; j += 4) {
+        double v = high_ssim_8x8(CONVERT_TO_SHORTPTR(img1 + j), stride_img1,
+                                 CONVERT_TO_SHORTPTR(img2 + j), stride_img2,
+                                 bd);
+        ssim_total += v;
+        samples++;
+      }
+    }
+  }
+  ssim_total /= samples;
+  return ssim_total;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 double vp9_calc_ssim(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest,
                      double *weight) {
   double a, b, c;
@@ -141,3 +251,63 @@
 
   return ssim_all;
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+double vp9_highbd_calc_ssim(YV12_BUFFER_CONFIG *source,
+                            YV12_BUFFER_CONFIG *dest,
+                            double *weight, unsigned int bd,
+                            unsigned int shift) {
+  double a, b, c;
+  double ssimv;
+
+  a = vp9_highbd_ssim2(source->y_buffer, dest->y_buffer,
+                       source->y_stride, dest->y_stride,
+                       source->y_crop_width, source->y_crop_height,
+                       bd, shift);
+
+  b = vp9_highbd_ssim2(source->u_buffer, dest->u_buffer,
+                       source->uv_stride, dest->uv_stride,
+                       source->uv_crop_width, source->uv_crop_height,
+                       bd, shift);
+
+  c = vp9_highbd_ssim2(source->v_buffer, dest->v_buffer,
+                       source->uv_stride, dest->uv_stride,
+                       source->uv_crop_width, source->uv_crop_height,
+                       bd, shift);
+
+  ssimv = a * .8 + .1 * (b + c);
+
+  *weight = 1;
+
+  return ssimv;
+}
+
+double vp9_highbd_calc_ssimg(YV12_BUFFER_CONFIG *source,
+                             YV12_BUFFER_CONFIG *dest, double *ssim_y,
+                             double *ssim_u, double *ssim_v,
+                             unsigned int bd, unsigned int shift) {
+  double ssim_all = 0;
+  double a, b, c;
+
+  a = vp9_highbd_ssim2(source->y_buffer, dest->y_buffer,
+                       source->y_stride, dest->y_stride,
+                       source->y_crop_width, source->y_crop_height,
+                       bd, shift);
+
+  b = vp9_highbd_ssim2(source->u_buffer, dest->u_buffer,
+                       source->uv_stride, dest->uv_stride,
+                       source->uv_crop_width, source->uv_crop_height,
+                       bd, shift);
+
+  c = vp9_highbd_ssim2(source->v_buffer, dest->v_buffer,
+                       source->uv_stride, dest->uv_stride,
+                       source->uv_crop_width, source->uv_crop_height,
+                       bd, shift);
+  *ssim_y = a;
+  *ssim_u = b;
+  *ssim_v = c;
+  ssim_all = (a * 4 + b + c) / 6;
+
+  return ssim_all;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_ssim.h b/vp9/encoder/vp9_ssim.h
index d1dd1b7..28baa4b 100644
--- a/vp9/encoder/vp9_ssim.h
+++ b/vp9/encoder/vp9_ssim.h
@@ -23,6 +23,22 @@
 double vp9_calc_ssimg(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest,
                       double *ssim_y, double *ssim_u, double *ssim_v);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+double vp9_highbd_calc_ssim(YV12_BUFFER_CONFIG *source,
+                            YV12_BUFFER_CONFIG *dest,
+                            double *weight,
+                            unsigned int bd,
+                            unsigned int shift);
+
+double vp9_highbd_calc_ssimg(YV12_BUFFER_CONFIG *source,
+                             YV12_BUFFER_CONFIG *dest,
+                             double *ssim_y,
+                             double *ssim_u,
+                             double *ssim_v,
+                             unsigned int bps,
+                             unsigned int shift);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index 7545d87..eed681c 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -19,7 +19,7 @@
   const VP9EncoderConfig *const oxcf = &cpi->oxcf;
   int layer;
   int layer_end;
-  int alt_ref_idx = svc->number_spatial_layers * svc->number_temporal_layers;
+  int alt_ref_idx = svc->number_spatial_layers;
 
   svc->spatial_layer_id = 0;
   svc->temporal_layer_id = 0;
@@ -233,51 +233,31 @@
 }
 
 #if CONFIG_SPATIAL_SVC
-int vp9_svc_lookahead_push(const VP9_COMP *const cpi, struct lookahead_ctx *ctx,
-                           YV12_BUFFER_CONFIG *src, int64_t ts_start,
-                           int64_t ts_end, unsigned int flags) {
-  struct lookahead_entry *buf;
-  int i, index;
+static void get_layer_resolution(const int width_org, const int height_org,
+                                 const int num, const int den,
+                                 int *width_out, int *height_out) {
+  int w, h;
 
-  if (vp9_lookahead_push(ctx, src, ts_start, ts_end, flags))
-    return 1;
+  if (width_out == NULL || height_out == NULL || den == 0)
+    return;
 
-  index = ctx->write_idx - 1;
-  if (index < 0)
-    index += ctx->max_sz;
+  w = width_org * num / den;
+  h = height_org * num / den;
 
-  buf = ctx->buf + index;
+  // make height and width even to make chrome player happy
+  w += w % 2;
+  h += h % 2;
 
-  if (buf == NULL)
-    return 1;
-
-  // Store svc parameters for each layer
-  for (i = 0; i < cpi->svc.number_spatial_layers; ++i)
-    buf->svc_params[i] = cpi->svc.layer_context[i].svc_params_received;
-
-  return 0;
+  *width_out = w;
+  *height_out = h;
 }
 
-static int copy_svc_params(VP9_COMP *const cpi, struct lookahead_entry *buf) {
-  int layer_id;
-  vpx_svc_parameters_t *layer_param;
+int vp9_svc_start_frame(VP9_COMP *const cpi) {
+  int width = 0, height = 0;
   LAYER_CONTEXT *lc;
   int count = 1 << (cpi->svc.number_temporal_layers - 1);
 
-  // Find the next layer to be encoded
-  for (layer_id = 0; layer_id < cpi->svc.number_spatial_layers; ++layer_id) {
-    if (buf->svc_params[layer_id].spatial_layer >=0)
-      break;
-  }
-
-  if (layer_id == cpi->svc.number_spatial_layers)
-    return 1;
-
-  layer_param = &buf->svc_params[layer_id];
-  cpi->svc.spatial_layer_id = layer_param->spatial_layer;
-  cpi->svc.temporal_layer_id = layer_param->temporal_layer;
-  cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
-
+  cpi->svc.spatial_layer_id = cpi->svc.spatial_layer_to_encode;
   lc = &cpi->svc.layer_context[cpi->svc.spatial_layer_id];
 
   cpi->svc.temporal_layer_id = 0;
@@ -286,30 +266,19 @@
     count >>= 1;
   }
 
-  cpi->lst_fb_idx =
-      cpi->svc.spatial_layer_id * cpi->svc.number_temporal_layers +
-      cpi->svc.temporal_layer_id;
-  if (lc->frames_from_key_frame < cpi->svc.number_temporal_layers)
-    cpi->ref_frame_flags &= ~VP9_LAST_FLAG;
+  cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
 
-  if (cpi->svc.spatial_layer_id == 0) {
-    if (cpi->svc.temporal_layer_id == 0)
-      cpi->gld_fb_idx = lc->gold_ref_idx >= 0 ?
-                        lc->gold_ref_idx : cpi->lst_fb_idx;
-    else
-      cpi->gld_fb_idx = cpi->lst_fb_idx - 1;
-  } else {
-    if (cpi->svc.temporal_layer_id == 0)
-      cpi->gld_fb_idx = cpi->svc.spatial_layer_id -
-                        cpi->svc.number_temporal_layers;
-    else
-      cpi->gld_fb_idx = cpi->lst_fb_idx - 1;
-  }
+  cpi->lst_fb_idx = cpi->svc.spatial_layer_id;
+
+  if (cpi->svc.spatial_layer_id == 0)
+    cpi->gld_fb_idx = (lc->gold_ref_idx >= 0) ?
+                      lc->gold_ref_idx : cpi->lst_fb_idx;
+  else
+    cpi->gld_fb_idx = cpi->svc.spatial_layer_id - 1;
 
   if (lc->current_video_frame_in_layer == 0) {
     if (cpi->svc.spatial_layer_id >= 2) {
-      cpi->alt_fb_idx =
-          cpi->svc.spatial_layer_id - 2 * cpi->svc.number_temporal_layers;
+      cpi->alt_fb_idx = cpi->svc.spatial_layer_id - 2;
     } else {
       cpi->alt_fb_idx = cpi->lst_fb_idx;
       cpi->ref_frame_flags &= (~VP9_LAST_FLAG & ~VP9_ALT_FLAG);
@@ -331,21 +300,21 @@
             lc_lower->alt_ref_source != NULL)
           cpi->alt_fb_idx = lc_lower->alt_ref_idx;
         else if (cpi->svc.spatial_layer_id >= 2)
-          cpi->alt_fb_idx =
-              cpi->svc.spatial_layer_id - 2 * cpi->svc.number_temporal_layers;
+          cpi->alt_fb_idx = cpi->svc.spatial_layer_id - 2;
         else
           cpi->alt_fb_idx = cpi->lst_fb_idx;
       }
     }
   }
 
-  if (vp9_set_size_literal(cpi, layer_param->width, layer_param->height) != 0)
+  get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height,
+                       lc->scaling_factor_num, lc->scaling_factor_den,
+                       &width, &height);
+  if (vp9_set_size_literal(cpi, width, height) != 0)
     return VPX_CODEC_INVALID_PARAM;
 
-  cpi->oxcf.worst_allowed_q =
-      vp9_quantizer_to_qindex(layer_param->max_quantizer);
-  cpi->oxcf.best_allowed_q =
-      vp9_quantizer_to_qindex(layer_param->min_quantizer);
+  cpi->oxcf.worst_allowed_q = vp9_quantizer_to_qindex(lc->max_q);
+  cpi->oxcf.best_allowed_q = vp9_quantizer_to_qindex(lc->min_q);
 
   vp9_change_config(cpi, &cpi->oxcf);
 
@@ -356,29 +325,15 @@
   return 0;
 }
 
-struct lookahead_entry *vp9_svc_lookahead_peek(VP9_COMP *const cpi,
-                                               struct lookahead_ctx *ctx,
-                                               int index, int copy_params) {
-  struct lookahead_entry *buf = vp9_lookahead_peek(ctx, index);
-
-  if (buf != NULL && copy_params != 0) {
-    if (copy_svc_params(cpi, buf) != 0)
-      return NULL;
-  }
-  return buf;
-}
-
 struct lookahead_entry *vp9_svc_lookahead_pop(VP9_COMP *const cpi,
                                               struct lookahead_ctx *ctx,
                                               int drain) {
   struct lookahead_entry *buf = NULL;
 
   if (ctx->sz && (drain || ctx->sz == ctx->max_sz - MAX_PRE_FRAMES)) {
-    buf = vp9_svc_lookahead_peek(cpi, ctx, 0, 1);
+    buf = vp9_lookahead_peek(ctx, 0);
     if (buf != NULL) {
-      // Only remove the buffer when pop the highest layer. Simply set the
-      // spatial_layer to -1 for lower layers.
-      buf->svc_params[cpi->svc.spatial_layer_id].spatial_layer = -1;
+      // Only remove the buffer when pop the highest layer.
       if (cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1) {
         vp9_lookahead_pop(ctx, drain);
       }
diff --git a/vp9/encoder/vp9_svc_layercontext.h b/vp9/encoder/vp9_svc_layercontext.h
index 1fc43a4..47a5456 100644
--- a/vp9/encoder/vp9_svc_layercontext.h
+++ b/vp9/encoder/vp9_svc_layercontext.h
@@ -24,18 +24,22 @@
   int target_bandwidth;
   double framerate;
   int avg_frame_size;
+  int max_q;
+  int min_q;
+  int scaling_factor_num;
+  int scaling_factor_den;
   TWO_PASS twopass;
   vpx_fixed_buf_t rc_twopass_stats_in;
   unsigned int current_video_frame_in_layer;
   int is_key_frame;
   int frames_from_key_frame;
   FRAME_TYPE last_frame_type;
-  vpx_svc_parameters_t svc_params_received;
   struct lookahead_entry  *alt_ref_source;
   int alt_ref_idx;
   int gold_ref_idx;
   int has_alt_frame;
   size_t layer_size;
+  struct vpx_psnr_pkt psnr_pkt;
 } LAYER_CONTEXT;
 
 typedef struct {
@@ -44,6 +48,8 @@
   int number_spatial_layers;
   int number_temporal_layers;
 
+  int spatial_layer_to_encode;
+
   // Store scaled source frames to be used for temporal filter to generate
   // a alt ref frame.
   YV12_BUFFER_CONFIG scaled_frames[MAX_LAG_BUFFERS];
@@ -87,22 +93,13 @@
 // Check if current layer is key frame in spatial upper layer
 int vp9_is_upper_layer_key_frame(const struct VP9_COMP *const cpi);
 
-// Copy the source image, flags and svc parameters into a new framebuffer
-// with the expected stride/border
-int vp9_svc_lookahead_push(const struct VP9_COMP *const cpi,
-                           struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
-                           int64_t ts_start, int64_t ts_end,
-                           unsigned int flags);
-
 // Get the next source buffer to encode
 struct lookahead_entry *vp9_svc_lookahead_pop(struct VP9_COMP *const cpi,
                                               struct lookahead_ctx *ctx,
                                               int drain);
 
-// Get a future source buffer to encode
-struct lookahead_entry *vp9_svc_lookahead_peek(struct VP9_COMP *const cpi,
-                                               struct lookahead_ctx *ctx,
-                                               int index, int copy_params);
+// Start a frame and initialize svc parameters
+int vp9_svc_start_frame(struct VP9_COMP *const cpi);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 18a6a91..eeb1ce9 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -44,7 +44,7 @@
   const int which_mv = 0;
   const MV mv = { mv_row, mv_col };
   const InterpKernel *const kernel =
-    vp9_get_interp_kernel(xd->mi[0]->mbmi.interp_filter);
+    vp9_get_interp_kernel(xd->mi[0].src_mi->mbmi.interp_filter);
 
   enum mv_precision mv_precision_uv;
   int uv_stride;
@@ -56,6 +56,34 @@
     mv_precision_uv = MV_PRECISION_Q3;
   }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    vp9_high_build_inter_predictor(y_mb_ptr, stride,
+                                   &pred[0], 16,
+                                   &mv,
+                                   scale,
+                                   16, 16,
+                                   which_mv,
+                                   kernel, MV_PRECISION_Q3, x, y, xd->bd);
+
+    vp9_high_build_inter_predictor(u_mb_ptr, uv_stride,
+                                   &pred[256], uv_block_width,
+                                   &mv,
+                                   scale,
+                                   uv_block_width, uv_block_height,
+                                   which_mv,
+                                   kernel, mv_precision_uv, x, y, xd->bd);
+
+    vp9_high_build_inter_predictor(v_mb_ptr, uv_stride,
+                                   &pred[512], uv_block_width,
+                                   &mv,
+                                   scale,
+                                   uv_block_width, uv_block_height,
+                                   which_mv,
+                                   kernel, mv_precision_uv, x, y, xd->bd);
+    return;
+  }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   vp9_build_inter_predictor(y_mb_ptr, stride,
                             &pred[0], 16,
                             &mv,
@@ -133,6 +161,54 @@
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
+                                        unsigned int stride,
+                                        uint8_t *frame2_8,
+                                        unsigned int block_width,
+                                        unsigned int block_height,
+                                        int strength,
+                                        int filter_weight,
+                                        unsigned int *accumulator,
+                                        uint16_t *count) {
+  uint16_t *frame1 = CONVERT_TO_SHORTPTR(frame1_8);
+  uint16_t *frame2 = CONVERT_TO_SHORTPTR(frame2_8);
+  unsigned int i, j, k;
+  int modifier;
+  int byte = 0;
+  const int rounding = strength > 0 ? 1 << (strength - 1) : 0;
+
+  for (i = 0, k = 0; i < block_height; i++) {
+    for (j = 0; j < block_width; j++, k++) {
+      int src_byte = frame1[byte];
+      int pixel_value = *frame2++;
+
+      modifier   = src_byte - pixel_value;
+      // This is an integer approximation of:
+      // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
+      // modifier =  (int)roundf(coeff > 16 ? 0 : 16-coeff);
+      modifier *= modifier;
+      modifier *= 3;
+      modifier += rounding;
+      modifier >>= strength;
+
+      if (modifier > 16)
+        modifier = 16;
+
+      modifier = 16 - modifier;
+      modifier *= filter_weight;
+
+      count[k] += modifier;
+      accumulator[k] += modifier * pixel_value;
+
+      byte++;
+    }
+
+    byte += stride - block_width;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static int temporal_filter_find_matching_mb_c(VP9_COMP *cpi,
                                               uint8_t *arf_frame_buf,
                                               uint8_t *frame_ptr_buf,
@@ -149,7 +225,7 @@
 
   MV best_ref_mv1 = {0, 0};
   MV best_ref_mv1_full; /* full-pixel value of best_ref_mv1 */
-  MV *ref_mv = &x->e_mbd.mi[0]->bmi[0].as_mv[0].as_mv;
+  MV *ref_mv = &x->e_mbd.mi[0].src_mi->bmi[0].as_mv[0].as_mv;
 
   // Save input state
   struct buf_2d src = x->plane[0].src;
@@ -200,8 +276,8 @@
   int frame;
   int mb_col, mb_row;
   unsigned int filter_weight;
-  int mb_cols = cpi->common.mb_cols;
-  int mb_rows = cpi->common.mb_rows;
+  int mb_cols = (frames[alt_ref_index]->y_crop_width + 15) >> 4;
+  int mb_rows = (frames[alt_ref_index]->y_crop_height + 15) >> 4;
   int mb_y_offset = 0;
   int mb_uv_offset = 0;
   DECLARE_ALIGNED_ARRAY(16, unsigned int, accumulator, 16 * 16 * 3);
@@ -209,13 +285,26 @@
   MACROBLOCKD *mbd = &cpi->mb.e_mbd;
   YV12_BUFFER_CONFIG *f = frames[alt_ref_index];
   uint8_t *dst1, *dst2;
+#if CONFIG_VP9_HIGHBITDEPTH
+  DECLARE_ALIGNED_ARRAY(16, uint16_t,  predictor16, 16 * 16 * 3);
+  DECLARE_ALIGNED_ARRAY(16, uint8_t,  predictor8, 16 * 16 * 3);
+  uint8_t *predictor;
+#else
   DECLARE_ALIGNED_ARRAY(16, uint8_t,  predictor, 16 * 16 * 3);
+#endif
   const int mb_uv_height = 16 >> mbd->plane[1].subsampling_y;
   const int mb_uv_width  = 16 >> mbd->plane[1].subsampling_x;
 
   // Save input state
   uint8_t* input_buffer[MAX_MB_PLANE];
   int i;
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    predictor = CONVERT_TO_BYTEPTR(predictor16);
+  } else {
+    predictor = predictor8;
+  }
+#endif
 
   for (i = 0; i < MAX_MB_PLANE; i++)
     input_buffer[i] = mbd->plane[i].pre[0].buf;
@@ -233,7 +322,7 @@
     // To keep the mv in play for both Y and UV planes the max that it
     //  can be on a border is therefore 16 - (2*VP9_INTERP_EXTEND+1).
     cpi->mb.mv_row_min = -((mb_row * 16) + (17 - 2 * VP9_INTERP_EXTEND));
-    cpi->mb.mv_row_max = ((cpi->common.mb_rows - 1 - mb_row) * 16)
+    cpi->mb.mv_row_max = ((mb_rows - 1 - mb_row) * 16)
                          + (17 - 2 * VP9_INTERP_EXTEND);
 
     for (mb_col = 0; mb_col < mb_cols; mb_col++) {
@@ -244,7 +333,7 @@
       vpx_memset(count, 0, 16 * 16 * 3 * sizeof(count[0]));
 
       cpi->mb.mv_col_min = -((mb_col * 16) + (17 - 2 * VP9_INTERP_EXTEND));
-      cpi->mb.mv_col_max = ((cpi->common.mb_cols - 1 - mb_col) * 16)
+      cpi->mb.mv_col_max = ((mb_cols - 1 - mb_col) * 16)
                            + (17 - 2 * VP9_INTERP_EXTEND);
 
       for (frame = 0; frame < frame_count; frame++) {
@@ -254,8 +343,8 @@
         if (frames[frame] == NULL)
           continue;
 
-        mbd->mi[0]->bmi[0].as_mv[0].as_mv.row = 0;
-        mbd->mi[0]->bmi[0].as_mv[0].as_mv.col = 0;
+        mbd->mi[0].src_mi->bmi[0].as_mv[0].as_mv.row = 0;
+        mbd->mi[0].src_mi->bmi[0].as_mv[0].as_mv.col = 0;
 
         if (frame == alt_ref_index) {
           filter_weight = 2;
@@ -281,11 +370,49 @@
               frames[frame]->v_buffer + mb_uv_offset,
               frames[frame]->y_stride,
               mb_uv_width, mb_uv_height,
-              mbd->mi[0]->bmi[0].as_mv[0].as_mv.row,
-              mbd->mi[0]->bmi[0].as_mv[0].as_mv.col,
+              mbd->mi[0].src_mi->bmi[0].as_mv[0].as_mv.row,
+              mbd->mi[0].src_mi->bmi[0].as_mv[0].as_mv.col,
               predictor, scale,
               mb_col * 16, mb_row * 16);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+          if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+            int adj_strength = strength + 2 * (mbd->bd - 8);
+            // Apply the filter (YUV)
+            vp9_highbd_temporal_filter_apply(f->y_buffer + mb_y_offset,
+                                             f->y_stride,
+                                             predictor, 16, 16, adj_strength,
+                                             filter_weight,
+                                             accumulator, count);
+            vp9_highbd_temporal_filter_apply(f->u_buffer + mb_uv_offset,
+                                             f->uv_stride, predictor + 256,
+                                             mb_uv_width, mb_uv_height,
+                                             adj_strength,
+                                             filter_weight, accumulator + 256,
+                                             count + 256);
+            vp9_highbd_temporal_filter_apply(f->v_buffer + mb_uv_offset,
+                                             f->uv_stride, predictor + 512,
+                                             mb_uv_width, mb_uv_height,
+                                             adj_strength, filter_weight,
+                                             accumulator + 512, count + 512);
+          } else {
+            // Apply the filter (YUV)
+            vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
+                                      predictor, 16, 16,
+                                      strength, filter_weight,
+                                      accumulator, count);
+            vp9_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
+                                      predictor + 256,
+                                      mb_uv_width, mb_uv_height, strength,
+                                      filter_weight, accumulator + 256,
+                                      count + 256);
+            vp9_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
+                                      predictor + 512,
+                                      mb_uv_width, mb_uv_height, strength,
+                                      filter_weight, accumulator + 512,
+                                      count + 512);
+          }
+#else
           // Apply the filter (YUV)
           vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
                                     predictor, 16, 16,
@@ -301,9 +428,108 @@
                                     mb_uv_width, mb_uv_height, strength,
                                     filter_weight, accumulator + 512,
                                     count + 512);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
         }
       }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+      if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+        uint16_t *dst1_16;
+        uint16_t *dst2_16;
+        // Normalize filter output to produce AltRef frame
+        dst1 = cpi->alt_ref_buffer.y_buffer;
+        dst1_16 = CONVERT_TO_SHORTPTR(dst1);
+        stride = cpi->alt_ref_buffer.y_stride;
+        byte = mb_y_offset;
+        for (i = 0, k = 0; i < 16; i++) {
+          for (j = 0; j < 16; j++, k++) {
+            unsigned int pval = accumulator[k] + (count[k] >> 1);
+            pval *= fixed_divide[count[k]];
+            pval >>= 19;
+
+            dst1_16[byte] = (uint16_t)pval;
+
+            // move to next pixel
+            byte++;
+          }
+
+          byte += stride - 16;
+        }
+
+        dst1 = cpi->alt_ref_buffer.u_buffer;
+        dst2 = cpi->alt_ref_buffer.v_buffer;
+        dst1_16 = CONVERT_TO_SHORTPTR(dst1);
+        dst2_16 = CONVERT_TO_SHORTPTR(dst2);
+        stride = cpi->alt_ref_buffer.uv_stride;
+        byte = mb_uv_offset;
+        for (i = 0, k = 256; i < mb_uv_height; i++) {
+          for (j = 0; j < mb_uv_width; j++, k++) {
+            int m = k + 256;
+
+            // U
+            unsigned int pval = accumulator[k] + (count[k] >> 1);
+            pval *= fixed_divide[count[k]];
+            pval >>= 19;
+            dst1_16[byte] = (uint16_t)pval;
+
+            // V
+            pval = accumulator[m] + (count[m] >> 1);
+            pval *= fixed_divide[count[m]];
+            pval >>= 19;
+            dst2_16[byte] = (uint16_t)pval;
+
+            // move to next pixel
+            byte++;
+          }
+
+          byte += stride - mb_uv_width;
+        }
+      } else {
+        // Normalize filter output to produce AltRef frame
+        dst1 = cpi->alt_ref_buffer.y_buffer;
+        stride = cpi->alt_ref_buffer.y_stride;
+        byte = mb_y_offset;
+        for (i = 0, k = 0; i < 16; i++) {
+          for (j = 0; j < 16; j++, k++) {
+            unsigned int pval = accumulator[k] + (count[k] >> 1);
+            pval *= fixed_divide[count[k]];
+            pval >>= 19;
+
+            dst1[byte] = (uint8_t)pval;
+
+            // move to next pixel
+            byte++;
+          }
+          byte += stride - 16;
+        }
+
+        dst1 = cpi->alt_ref_buffer.u_buffer;
+        dst2 = cpi->alt_ref_buffer.v_buffer;
+        stride = cpi->alt_ref_buffer.uv_stride;
+        byte = mb_uv_offset;
+        for (i = 0, k = 256; i < mb_uv_height; i++) {
+          for (j = 0; j < mb_uv_width; j++, k++) {
+            int m = k + 256;
+
+            // U
+            unsigned int pval = accumulator[k] + (count[k] >> 1);
+            pval *= fixed_divide[count[k]];
+            pval >>= 19;
+            dst1[byte] = (uint8_t)pval;
+
+            // V
+            pval = accumulator[m] + (count[m] >> 1);
+            pval *= fixed_divide[count[m]];
+            pval >>= 19;
+            dst2[byte] = (uint8_t)pval;
+
+            // move to next pixel
+            byte++;
+          }
+          byte += stride - mb_uv_width;
+        }
+      }
+#else
       // Normalize filter output to produce AltRef frame
       dst1 = cpi->alt_ref_buffer.y_buffer;
       stride = cpi->alt_ref_buffer.y_stride;
@@ -347,6 +573,7 @@
         }
         byte += stride - mb_uv_width;
       }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
       mb_y_offset += 16;
       mb_uv_offset += mb_uv_width;
     }
@@ -389,10 +616,10 @@
   // Adjust the strength based on active max q.
   if (cpi->common.current_video_frame > 1)
     q = ((int)vp9_convert_qindex_to_q(
-        cpi->rc.avg_frame_qindex[INTER_FRAME]));
+        cpi->rc.avg_frame_qindex[INTER_FRAME], cpi->common.bit_depth));
   else
     q = ((int)vp9_convert_qindex_to_q(
-        cpi->rc.avg_frame_qindex[KEY_FRAME]));
+        cpi->rc.avg_frame_qindex[KEY_FRAME], cpi->common.bit_depth));
   if (q > 16) {
     strength = oxcf->arnr_strength;
   } else {
@@ -454,11 +681,20 @@
     // In spatial svc the scaling factors might be less then 1/2. So we will use
     // non-normative scaling.
     int frame_used = 0;
+#if CONFIG_VP9_HIGHBITDEPTH
+    vp9_setup_scale_factors_for_frame(&sf,
+                                      get_frame_new_buffer(cm)->y_crop_width,
+                                      get_frame_new_buffer(cm)->y_crop_height,
+                                      get_frame_new_buffer(cm)->y_crop_width,
+                                      get_frame_new_buffer(cm)->y_crop_height,
+                                      cm->use_highbitdepth);
+#else
     vp9_setup_scale_factors_for_frame(&sf,
                                       get_frame_new_buffer(cm)->y_crop_width,
                                       get_frame_new_buffer(cm)->y_crop_height,
                                       get_frame_new_buffer(cm)->y_crop_width,
                                       get_frame_new_buffer(cm)->y_crop_height);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
     for (frame = 0; frame < frames_to_blur; ++frame) {
       if (cm->mi_cols * MI_SIZE != frames[frame]->y_width ||
@@ -480,10 +716,21 @@
       }
     }
   } else {
+    // ARF is produced at the native frame size and resized when coded.
+#if CONFIG_VP9_HIGHBITDEPTH
     vp9_setup_scale_factors_for_frame(&sf,
-                                      get_frame_new_buffer(cm)->y_crop_width,
-                                      get_frame_new_buffer(cm)->y_crop_height,
-                                      cm->width, cm->height);
+                                      frames[0]->y_crop_width,
+                                      frames[0]->y_crop_height,
+                                      frames[0]->y_crop_width,
+                                      frames[0]->y_crop_height,
+                                      cm->use_highbitdepth);
+#else
+    vp9_setup_scale_factors_for_frame(&sf,
+                                      frames[0]->y_crop_width,
+                                      frames[0]->y_crop_height,
+                                      frames[0]->y_crop_width,
+                                      frames[0]->y_crop_height);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   }
 
   temporal_filter_iterate_c(cpi, frames, frames_to_blur,
diff --git a/vp9/encoder/vp9_tokenize.c b/vp9/encoder/vp9_tokenize.c
index 6068b85..8b9aa91 100644
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -28,6 +28,18 @@
 static int16_t dct_value_cost[DCT_MAX_VALUE * 2];
 const int16_t *vp9_dct_value_cost_ptr;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static TOKENVALUE dct_value_tokens_high10[DCT_MAX_VALUE_HIGH10 * 2];
+const TOKENVALUE *vp9_dct_value_tokens_high10_ptr;
+static int16_t dct_value_cost_high10[DCT_MAX_VALUE_HIGH10 * 2];
+const int16_t *vp9_dct_value_cost_high10_ptr;
+
+static TOKENVALUE dct_value_tokens_high12[DCT_MAX_VALUE_HIGH12 * 2];
+const TOKENVALUE *vp9_dct_value_tokens_high12_ptr;
+static int16_t dct_value_cost_high12[DCT_MAX_VALUE_HIGH12 * 2];
+const int16_t *vp9_dct_value_cost_high12_ptr;
+#endif
+
 // Array indices are identical to previously-existing CONTEXT_NODE indices
 const vp9_tree_index vp9_coef_tree[TREE_SIZE(ENTROPY_TOKENS)] = {
   -EOB_TOKEN, 2,                       // 0  = EOB
@@ -57,6 +69,21 @@
 
 static vp9_tree_index cat1[2], cat2[4], cat3[6], cat4[8], cat5[10], cat6[28];
 
+#if CONFIG_VP9_HIGHBITDEPTH
+static vp9_tree_index cat1_high10[2];
+static vp9_tree_index cat2_high10[4];
+static vp9_tree_index cat3_high10[6];
+static vp9_tree_index cat4_high10[8];
+static vp9_tree_index cat5_high10[10];
+static vp9_tree_index cat6_high10[32];
+static vp9_tree_index cat1_high12[2];
+static vp9_tree_index cat2_high12[4];
+static vp9_tree_index cat3_high12[6];
+static vp9_tree_index cat4_high12[8];
+static vp9_tree_index cat5_high12[10];
+static vp9_tree_index cat6_high12[36];
+#endif
+
 static void init_bit_tree(vp9_tree_index *p, int n) {
   int i = 0;
 
@@ -75,6 +102,20 @@
   init_bit_tree(cat4, 4);
   init_bit_tree(cat5, 5);
   init_bit_tree(cat6, 14);
+#if CONFIG_VP9_HIGHBITDEPTH
+  init_bit_tree(cat1_high10, 1);
+  init_bit_tree(cat2_high10, 2);
+  init_bit_tree(cat3_high10, 3);
+  init_bit_tree(cat4_high10, 4);
+  init_bit_tree(cat5_high10, 5);
+  init_bit_tree(cat6_high10, 16);
+  init_bit_tree(cat1_high12, 1);
+  init_bit_tree(cat2_high12, 2);
+  init_bit_tree(cat3_high12, 3);
+  init_bit_tree(cat4_high12, 4);
+  init_bit_tree(cat5_high12, 5);
+  init_bit_tree(cat6_high12, 18);
+#endif
 }
 
 const vp9_extra_bit vp9_extra_bits[ENTROPY_TOKENS] = {
@@ -92,6 +133,37 @@
   {0, 0, 0, 0}                               // EOB_TOKEN
 };
 
+#if CONFIG_VP9_HIGHBITDEPTH
+const vp9_extra_bit vp9_extra_bits_high10[ENTROPY_TOKENS] = {
+  {0, 0, 0, 0},                                            // ZERO_TOKEN
+  {0, 0, 0, 1},                                            // ONE_TOKEN
+  {0, 0, 0, 2},                                            // TWO_TOKEN
+  {0, 0, 0, 3},                                            // THREE_TOKEN
+  {0, 0, 0, 4},                                            // FOUR_TOKEN
+  {cat1_high10, vp9_cat1_prob_high10, 1,  CAT1_MIN_VAL},   // CATEGORY1_TOKEN
+  {cat2_high10, vp9_cat2_prob_high10, 2,  CAT2_MIN_VAL},   // CATEGORY2_TOKEN
+  {cat3_high10, vp9_cat3_prob_high10, 3,  CAT3_MIN_VAL},   // CATEGORY3_TOKEN
+  {cat4_high10, vp9_cat4_prob_high10, 4,  CAT4_MIN_VAL},   // CATEGORY4_TOKEN
+  {cat5_high10, vp9_cat5_prob_high10, 5,  CAT5_MIN_VAL},   // CATEGORY5_TOKEN
+  {cat6_high10, vp9_cat6_prob_high10, 16, CAT6_MIN_VAL},   // CATEGORY6_TOKEN
+  {0, 0, 0, 0}                                             // EOB_TOKEN
+};
+const vp9_extra_bit vp9_extra_bits_high12[ENTROPY_TOKENS] = {
+  {0, 0, 0, 0},                                            // ZERO_TOKEN
+  {0, 0, 0, 1},                                            // ONE_TOKEN
+  {0, 0, 0, 2},                                            // TWO_TOKEN
+  {0, 0, 0, 3},                                            // THREE_TOKEN
+  {0, 0, 0, 4},                                            // FOUR_TOKEN
+  {cat1_high12, vp9_cat1_prob_high12, 1,  CAT1_MIN_VAL},   // CATEGORY1_TOKEN
+  {cat2_high12, vp9_cat2_prob_high12, 2,  CAT2_MIN_VAL},   // CATEGORY2_TOKEN
+  {cat3_high12, vp9_cat3_prob_high12, 3,  CAT3_MIN_VAL},   // CATEGORY3_TOKEN
+  {cat4_high12, vp9_cat4_prob_high12, 4,  CAT4_MIN_VAL},   // CATEGORY4_TOKEN
+  {cat5_high12, vp9_cat5_prob_high12, 5,  CAT5_MIN_VAL},   // CATEGORY5_TOKEN
+  {cat6_high12, vp9_cat6_prob_high12, 18, CAT6_MIN_VAL},   // CATEGORY6_TOKEN
+  {0, 0, 0, 0}                                             // EOB_TOKEN
+};
+#endif
+
 struct vp9_token vp9_coef_encodings[ENTROPY_TOKENS];
 
 void vp9_coef_tree_initialize() {
@@ -99,11 +171,9 @@
   vp9_tokens_from_tree(vp9_coef_encodings, vp9_coef_tree);
 }
 
-void vp9_tokenize_initialize() {
-  TOKENVALUE *const t = dct_value_tokens + DCT_MAX_VALUE;
-  const vp9_extra_bit *const e = vp9_extra_bits;
-
-  int i = -DCT_MAX_VALUE;
+static void tokenize_init_one(TOKENVALUE *t, const vp9_extra_bit *const e,
+                              int16_t *value_cost, int max_value) {
+  int i = -max_value;
   int sign = 1;
 
   do {
@@ -130,7 +200,7 @@
     // initialize the cost for extra bits for all possible coefficient value.
     {
       int cost = 0;
-      const vp9_extra_bit *p = &vp9_extra_bits[t[i].token];
+      const vp9_extra_bit *p = &e[t[i].token];
 
       if (p->base_val) {
         const int extra = t[i].extra;
@@ -140,13 +210,36 @@
           cost += treed_cost(p->tree, p->prob, extra >> 1, length);
 
         cost += vp9_cost_bit(vp9_prob_half, extra & 1); /* sign */
-        dct_value_cost[i + DCT_MAX_VALUE] = cost;
+        value_cost[i] = cost;
       }
     }
-  } while (++i < DCT_MAX_VALUE);
+  } while (++i < max_value);
+}
 
+void vp9_tokenize_initialize() {
   vp9_dct_value_tokens_ptr = dct_value_tokens + DCT_MAX_VALUE;
   vp9_dct_value_cost_ptr = dct_value_cost + DCT_MAX_VALUE;
+
+  tokenize_init_one(dct_value_tokens + DCT_MAX_VALUE, vp9_extra_bits,
+                    dct_value_cost + DCT_MAX_VALUE, DCT_MAX_VALUE);
+#if CONFIG_VP9_HIGHBITDEPTH
+  vp9_dct_value_tokens_high10_ptr = dct_value_tokens_high10 +
+      DCT_MAX_VALUE_HIGH10;
+  vp9_dct_value_cost_high10_ptr = dct_value_cost_high10 + DCT_MAX_VALUE_HIGH10;
+
+  tokenize_init_one(dct_value_tokens_high10 + DCT_MAX_VALUE_HIGH10,
+                    vp9_extra_bits_high10,
+                    dct_value_cost_high10 + DCT_MAX_VALUE_HIGH10,
+                    DCT_MAX_VALUE_HIGH10);
+  vp9_dct_value_tokens_high12_ptr = dct_value_tokens_high12 +
+      DCT_MAX_VALUE_HIGH12;
+  vp9_dct_value_cost_high12_ptr = dct_value_cost_high12 + DCT_MAX_VALUE_HIGH12;
+
+  tokenize_init_one(dct_value_tokens_high12 + DCT_MAX_VALUE_HIGH12,
+                    vp9_extra_bits_high12,
+                    dct_value_cost_high12 + DCT_MAX_VALUE_HIGH12,
+                    DCT_MAX_VALUE_HIGH12);
+#endif
 }
 
 struct tokenize_b_args {
@@ -206,13 +299,13 @@
   uint8_t token_cache[32 * 32];
   struct macroblock_plane *p = &cpi->mb.plane[plane];
   struct macroblockd_plane *pd = &xd->plane[plane];
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   int pt; /* near block/prev token context index */
   int c;
   TOKENEXTRA *t = *tp;        /* store tokens starting here */
   int eob = p->eobs[block];
   const PLANE_TYPE type = pd->plane_type;
-  const int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  const tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   const int segment_id = mbmi->segment_id;
   const int16_t *scan, *nb;
   const scan_order *so;
@@ -225,6 +318,7 @@
       cpi->common.counts.eob_branch[tx_size][type][ref];
   const uint8_t *const band = get_band_translate(tx_size);
   const int seg_eob = get_tx_eob(&cpi->common.seg, segment_id, tx_size);
+  const TOKENVALUE *dct_value_tokens;
 
   int aoff, loff;
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &aoff, &loff);
@@ -235,6 +329,18 @@
   scan = so->scan;
   nb = so->neighbors;
   c = 0;
+#if CONFIG_VP9_HIGH && CONFIG_HIGH_QUANT
+  if (cpi->common.profile >= PROFILE_2) {
+    dct_value_tokens = (cpi->common.bit_depth == VPX_BITS_10 ?
+                        vp9_dct_value_tokens_high10_ptr :
+                        vp9_dct_value_tokens_high12_ptr);
+  } else {
+    dct_value_tokens = vp9_dct_value_tokens_ptr;
+  }
+#else
+  dct_value_tokens = vp9_dct_value_tokens_ptr;
+#endif
+
   while (c < eob) {
     int v = 0;
     int skip_eob = 0;
@@ -253,14 +359,13 @@
     }
 
     add_token(&t, coef_probs[band[c]][pt],
-              vp9_dct_value_tokens_ptr[v].extra,
-              (uint8_t)vp9_dct_value_tokens_ptr[v].token,
+              dct_value_tokens[v].extra,
+              (uint8_t)dct_value_tokens[v].token,
               (uint8_t)skip_eob,
               counts[band[c]][pt]);
     eob_branch[band[c]][pt] += !skip_eob;
 
-    token_cache[scan[c]] =
-        vp9_pt_energy_class[vp9_dct_value_tokens_ptr[v].token];
+    token_cache[scan[c]] = vp9_pt_energy_class[dct_value_tokens[v].token];
     ++c;
     pt = get_coef_context(nb, token_cache, c);
   }
@@ -302,7 +407,7 @@
                      BLOCK_SIZE bsize) {
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   TOKENEXTRA *t_backup = *t;
   const int ctx = vp9_get_skip_context(xd);
   const int skip_inc = !vp9_segfeature_active(&cm->seg, mbmi->segment_id,
diff --git a/vp9/encoder/vp9_variance.c b/vp9/encoder/vp9_variance.c
index afbb191..c97f93f 100644
--- a/vp9/encoder/vp9_variance.c
+++ b/vp9/encoder/vp9_variance.c
@@ -267,3 +267,375 @@
     ref += ref_stride;
   }
 }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void high_variance64(const uint8_t *a8, int  a_stride,
+                     const uint8_t *b8, int  b_stride,
+                     int w, int h, uint64_t *sse,
+                     uint64_t *sum) {
+  int i, j;
+
+  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = a[j] - b[j];
+      *sum += diff;
+      *sse += diff * diff;
+    }
+    a += a_stride;
+    b += b_stride;
+  }
+}
+
+void high_variance(const uint8_t *a8, int  a_stride,
+                   const uint8_t *b8, int  b_stride,
+                   int w, int h, unsigned int *sse,
+                   int *sum) {
+  uint64_t sse_long = 0;
+  uint64_t sum_long = 0;
+  high_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
+  *sse = sse_long;
+  *sum = sum_long;
+}
+
+void high_10_variance(const uint8_t *a8, int  a_stride,
+                      const uint8_t *b8, int  b_stride,
+                      int w, int h, unsigned int *sse,
+                      int *sum) {
+  uint64_t sse_long = 0;
+  uint64_t sum_long = 0;
+  high_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
+  *sum = ROUND_POWER_OF_TWO(sum_long, 2);
+  *sse = ROUND_POWER_OF_TWO(sse_long, 4);
+}
+
+void high_12_variance(const uint8_t *a8, int  a_stride,
+                      const uint8_t *b8, int  b_stride,
+                      int w, int h, unsigned int *sse,
+                      int *sum) {
+  uint64_t sse_long = 0;
+  uint64_t sum_long = 0;
+  high_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
+  *sum = ROUND_POWER_OF_TWO(sum_long, 4);
+  *sse = ROUND_POWER_OF_TWO(sse_long, 8);
+}
+
+static void high_var_filter_block2d_bil_first_pass(
+    const uint8_t *src_ptr8,
+    uint16_t *output_ptr,
+    unsigned int src_pixels_per_line,
+    int pixel_step,
+    unsigned int output_height,
+    unsigned int output_width,
+    const int16_t *vp9_filter) {
+  unsigned int i, j;
+  uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src_ptr8);
+  for (i = 0; i < output_height; i++) {
+    for (j = 0; j < output_width; j++) {
+      output_ptr[j] =
+          ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
+                             (int)src_ptr[pixel_step] * vp9_filter[1],
+                             FILTER_BITS);
+
+      src_ptr++;
+    }
+
+    // Next row...
+    src_ptr += src_pixels_per_line - output_width;
+    output_ptr += output_width;
+  }
+}
+
+static void high_var_filter_block2d_bil_second_pass(
+    const uint16_t *src_ptr,
+    uint16_t *output_ptr,
+    unsigned int src_pixels_per_line,
+    unsigned int pixel_step,
+    unsigned int output_height,
+    unsigned int output_width,
+    const int16_t *vp9_filter) {
+  unsigned int  i, j;
+
+  for (i = 0; i < output_height; i++) {
+    for (j = 0; j < output_width; j++) {
+      output_ptr[j] =
+          ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
+                             (int)src_ptr[pixel_step] * vp9_filter[1],
+                             FILTER_BITS);
+      src_ptr++;
+    }
+
+    src_ptr += src_pixels_per_line - output_width;
+    output_ptr += output_width;
+  }
+}
+
+#define HIGH_VAR(W, H) \
+unsigned int vp9_high_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
+                                            const uint8_t *b, int b_stride, \
+                                            unsigned int *sse) { \
+  int sum; \
+  high_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+} \
+\
+unsigned int vp9_high_10_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
+                                               const uint8_t *b, int b_stride, \
+                                                unsigned int *sse) { \
+  int sum; \
+  high_10_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+} \
+\
+unsigned int vp9_high_12_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
+                                               const uint8_t *b, int b_stride, \
+                                               unsigned int *sse) { \
+  int sum; \
+  high_12_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
+  return *sse - (((int64_t)sum * sum) / (W * H)); \
+}
+
+#define HIGH_SUBPIX_VAR(W, H) \
+unsigned int vp9_high_sub_pixel_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  unsigned int *sse) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint16_t temp2[H * W]; \
+\
+  high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
+                                         W, BILINEAR_FILTERS_2TAP(xoffset)); \
+  high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                          BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  return vp9_high_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
+                                        dst_stride, sse); \
+} \
+\
+unsigned int vp9_high_10_sub_pixel_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  unsigned int *sse) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint16_t temp2[H * W]; \
+\
+  high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
+                                         W, BILINEAR_FILTERS_2TAP(xoffset)); \
+  high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                          BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  return vp9_high_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
+                                           dst_stride, sse); \
+} \
+\
+unsigned int vp9_high_12_sub_pixel_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  unsigned int *sse) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint16_t temp2[H * W]; \
+\
+  high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
+                                         W, BILINEAR_FILTERS_2TAP(xoffset)); \
+  high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                          BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  return vp9_high_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
+                                           dst_stride, sse); \
+}
+
+#define HIGH_SUBPIX_AVG_VAR(W, H) \
+unsigned int vp9_high_sub_pixel_avg_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  unsigned int *sse, \
+  const uint8_t *second_pred) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint16_t temp2[H * W]; \
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, temp3, H * W); \
+\
+  high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
+                                         W, BILINEAR_FILTERS_2TAP(xoffset)); \
+  high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                          BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  vp9_high_comp_avg_pred(temp3, second_pred, W, H, CONVERT_TO_BYTEPTR(temp2), \
+                         W); \
+\
+  return vp9_high_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
+                                        dst_stride, sse); \
+} \
+\
+unsigned int vp9_high_10_sub_pixel_avg_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  unsigned int *sse, \
+  const uint8_t *second_pred) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint16_t temp2[H * W]; \
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, temp3, H * W); \
+\
+  high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
+                                         W, BILINEAR_FILTERS_2TAP(xoffset)); \
+  high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                          BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  vp9_high_comp_avg_pred(temp3, second_pred, W, H, CONVERT_TO_BYTEPTR(temp2), \
+                         W); \
+\
+  return vp9_high_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
+                                        dst_stride, sse); \
+} \
+\
+unsigned int vp9_high_12_sub_pixel_avg_variance##W##x##H##_c( \
+  const uint8_t *src, int  src_stride, \
+  int xoffset, int  yoffset, \
+  const uint8_t *dst, int dst_stride, \
+  unsigned int *sse, \
+  const uint8_t *second_pred) { \
+  uint16_t fdata3[(H + 1) * W]; \
+  uint16_t temp2[H * W]; \
+  DECLARE_ALIGNED_ARRAY(16, uint16_t, temp3, H * W); \
+\
+  high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
+                                         W, BILINEAR_FILTERS_2TAP(xoffset)); \
+  high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
+                                          BILINEAR_FILTERS_2TAP(yoffset)); \
+\
+  vp9_high_comp_avg_pred(temp3, second_pred, W, H, CONVERT_TO_BYTEPTR(temp2), \
+                         W); \
+\
+  return vp9_high_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
+                                        dst_stride, sse); \
+}
+
+#define HIGH_GET_VAR(S) \
+void vp9_high_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
+                                  const uint8_t *ref, int ref_stride, \
+                                  unsigned int *sse, int *sum) { \
+  high_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
+} \
+\
+void vp9_high_10_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
+                                     const uint8_t *ref, int ref_stride, \
+                                     unsigned int *sse, int *sum) { \
+  high_10_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
+} \
+\
+void vp9_high_12_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
+                                     const uint8_t *ref, int ref_stride, \
+                                     unsigned int *sse, int *sum) { \
+  high_12_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
+}
+
+#define HIGH_MSE(W, H) \
+unsigned int vp9_high_mse##W##x##H##_c(const uint8_t *src, int src_stride, \
+                                       const uint8_t *ref, int ref_stride, \
+                                       unsigned int *sse) { \
+  int sum; \
+  high_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
+  return *sse; \
+} \
+\
+unsigned int vp9_high_10_mse##W##x##H##_c(const uint8_t *src, int src_stride, \
+                                          const uint8_t *ref, int ref_stride, \
+                                          unsigned int *sse) { \
+  int sum; \
+  high_10_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
+  return *sse; \
+} \
+\
+unsigned int vp9_high_12_mse##W##x##H##_c(const uint8_t *src, int src_stride, \
+                                          const uint8_t *ref, int ref_stride, \
+                                          unsigned int *sse) { \
+  int sum; \
+  high_12_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
+  return *sse; \
+}
+
+HIGH_GET_VAR(8)
+HIGH_GET_VAR(16)
+
+HIGH_MSE(16, 16)
+HIGH_MSE(16, 8)
+HIGH_MSE(8, 16)
+HIGH_MSE(8, 8)
+
+HIGH_VAR(4, 4)
+HIGH_SUBPIX_VAR(4, 4)
+HIGH_SUBPIX_AVG_VAR(4, 4)
+
+HIGH_VAR(4, 8)
+HIGH_SUBPIX_VAR(4, 8)
+HIGH_SUBPIX_AVG_VAR(4, 8)
+
+HIGH_VAR(8, 4)
+HIGH_SUBPIX_VAR(8, 4)
+HIGH_SUBPIX_AVG_VAR(8, 4)
+
+HIGH_VAR(8, 8)
+HIGH_SUBPIX_VAR(8, 8)
+HIGH_SUBPIX_AVG_VAR(8, 8)
+
+HIGH_VAR(8, 16)
+HIGH_SUBPIX_VAR(8, 16)
+HIGH_SUBPIX_AVG_VAR(8, 16)
+
+HIGH_VAR(16, 8)
+HIGH_SUBPIX_VAR(16, 8)
+HIGH_SUBPIX_AVG_VAR(16, 8)
+
+HIGH_VAR(16, 16)
+HIGH_SUBPIX_VAR(16, 16)
+HIGH_SUBPIX_AVG_VAR(16, 16)
+
+HIGH_VAR(16, 32)
+HIGH_SUBPIX_VAR(16, 32)
+HIGH_SUBPIX_AVG_VAR(16, 32)
+
+HIGH_VAR(32, 16)
+HIGH_SUBPIX_VAR(32, 16)
+HIGH_SUBPIX_AVG_VAR(32, 16)
+
+HIGH_VAR(32, 32)
+HIGH_SUBPIX_VAR(32, 32)
+HIGH_SUBPIX_AVG_VAR(32, 32)
+
+HIGH_VAR(32, 64)
+HIGH_SUBPIX_VAR(32, 64)
+HIGH_SUBPIX_AVG_VAR(32, 64)
+
+HIGH_VAR(64, 32)
+HIGH_SUBPIX_VAR(64, 32)
+HIGH_SUBPIX_AVG_VAR(64, 32)
+
+HIGH_VAR(64, 64)
+HIGH_SUBPIX_VAR(64, 64)
+HIGH_SUBPIX_AVG_VAR(64, 64)
+
+void vp9_high_comp_avg_pred(uint16_t *comp_pred, const uint8_t *pred8,
+                            int width, int height, const uint8_t *ref8,
+                            int ref_stride) {
+  int i, j;
+  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+  for (i = 0; i < height; i++) {
+    for (j = 0; j < width; j++) {
+      const int tmp = pred[j] + ref[j];
+      comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
+    }
+    comp_pred += width;
+    pred += width;
+    ref += ref_stride;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_variance.h b/vp9/encoder/vp9_variance.h
index 4a194b7..c51d08d 100644
--- a/vp9/encoder/vp9_variance.h
+++ b/vp9/encoder/vp9_variance.h
@@ -22,6 +22,23 @@
               int  w, int  h,
               unsigned int *sse, int *sum);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void high_variance(const uint8_t *a8, int a_stride,
+                   const uint8_t *b8, int b_stride,
+                   int w, int h,
+                   unsigned int *sse, int *sum);
+
+void high_10_variance(const uint8_t *a8, int a_stride,
+                      const uint8_t *b8, int b_stride,
+                      int w, int h,
+                      unsigned int *sse, int *sum);
+
+void high_12_variance(const uint8_t *a8, int a_stride,
+                      const uint8_t *b8, int b_stride,
+                      int w, int h,
+                      unsigned int *sse, int *sum);
+#endif
+
 typedef unsigned int(*vp9_sad_fn_t)(const uint8_t *src_ptr,
                                     int source_stride,
                                     const uint8_t *ref_ptr,
@@ -81,6 +98,11 @@
 void vp9_comp_avg_pred(uint8_t *comp_pred, const uint8_t *pred, int width,
                        int height, const uint8_t *ref, int ref_stride);
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_high_comp_avg_pred(uint16_t *comp_pred, const uint8_t *pred, int width,
+                            int height, const uint8_t *ref, int ref_stride);
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 90f0342..9414120 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -89,6 +89,13 @@
 VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_intrapred_ssse3.asm
 endif
 
+ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_intrapred_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_8t_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_bilinear_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_loopfilter_intrin_sse2.c
+endif
+
 # common (c)
 VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_common_dspr2.h
 VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_convolve2_avg_dspr2.c
diff --git a/vp9/vp9_cx_iface.c b/vp9/vp9_cx_iface.c
index 0f0b7a5..fbf4aa2 100644
--- a/vp9/vp9_cx_iface.c
+++ b/vp9/vp9_cx_iface.c
@@ -178,8 +178,6 @@
   }
 
 #if CONFIG_SPATIAL_SVC
-  if (cfg->ss_number_layers * cfg->ts_number_layers > REF_FRAMES)
-    ERROR("Too many layers. Maximum 8 layers could be set");
 
   if ((cfg->ss_number_layers > 1 || cfg->ts_number_layers > 1) &&
       cfg->g_pass == VPX_RC_LAST_PASS) {
@@ -188,8 +186,7 @@
       if (cfg->ss_enable_auto_alt_ref[i])
         ++alt_ref_sum;
     }
-    if (alt_ref_sum >
-        REF_FRAMES - cfg->ss_number_layers * cfg->ts_number_layers)
+    if (alt_ref_sum > REF_FRAMES - cfg->ss_number_layers)
       ERROR("Not enough ref buffers for svc alt ref frames");
     if ((cfg->ss_number_layers > 3 ||
          cfg->ss_number_layers * cfg->ts_number_layers > 4) &&
@@ -555,7 +552,7 @@
 static vpx_codec_err_t ctrl_set_noise_sensitivity(vpx_codec_alg_priv_t *ctx,
                                                   va_list args) {
   struct vp9_extracfg extra_cfg = ctx->extra_cfg;
-  extra_cfg.noise_sensitivity = CAST(VP8E_SET_NOISE_SENSITIVITY, args);
+  extra_cfg.noise_sensitivity = CAST(VP9E_SET_NOISE_SENSITIVITY, args);
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
@@ -686,6 +683,10 @@
 
     if (res == VPX_CODEC_OK) {
       set_encoder_config(&priv->oxcf, &priv->cfg, &priv->extra_cfg);
+#if CONFIG_VP9_HIGHBITDEPTH
+      priv->oxcf.use_highbitdepth =
+          (ctx->init_flags & VPX_CODEC_USE_HIGHBITDEPTH) ? 1 : 0;
+#endif
       priv->cpi = vp9_create_compressor(&priv->oxcf);
       if (priv->cpi == NULL)
         res = VPX_CODEC_MEM_ERROR;
@@ -981,15 +982,20 @@
         cx_data_sz -= size;
 #if CONFIG_SPATIAL_SVC
         if (is_two_pass_svc(cpi)) {
-          vpx_codec_cx_pkt_t pkt;
+          vpx_codec_cx_pkt_t pkt_sizes, pkt_psnr;
           int i;
-          vp9_zero(pkt);
-          pkt.kind = VPX_CODEC_SPATIAL_SVC_LAYER_SIZES;
+          vp9_zero(pkt_sizes);
+          vp9_zero(pkt_psnr);
+          pkt_sizes.kind = VPX_CODEC_SPATIAL_SVC_LAYER_SIZES;
+          pkt_psnr.kind = VPX_CODEC_SPATIAL_SVC_LAYER_PSNR;
           for (i = 0; i < cpi->svc.number_spatial_layers; ++i) {
-            pkt.data.layer_sizes[i] = cpi->svc.layer_context[i].layer_size;
-            cpi->svc.layer_context[i].layer_size = 0;
+            LAYER_CONTEXT *lc = &cpi->svc.layer_context[i];
+            pkt_sizes.data.layer_sizes[i] = lc->layer_size;
+            pkt_psnr.data.layer_psnr[i] = lc->psnr_pkt;
+            lc->layer_size = 0;
           }
-          vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt);
+          vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt_sizes);
+          vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt_psnr);
         }
 #endif
       }
@@ -1192,22 +1198,18 @@
 static vpx_codec_err_t ctrl_set_svc_parameters(vpx_codec_alg_priv_t *ctx,
                                                va_list args) {
   VP9_COMP *const cpi = ctx->cpi;
-  vpx_svc_parameters_t *const params = va_arg(args, vpx_svc_parameters_t *);
+  vpx_svc_extra_cfg_t *const params = va_arg(args, vpx_svc_extra_cfg_t *);
+  int i;
 
-  if (params == NULL || params->spatial_layer < 0 ||
-      params->spatial_layer >= cpi->svc.number_spatial_layers)
-    return VPX_CODEC_INVALID_PARAM;
+  for (i = 0; i < cpi->svc.number_spatial_layers; ++i) {
+    LAYER_CONTEXT *lc = &cpi->svc.layer_context[i];
 
-  if (params->spatial_layer == 0) {
-    int i;
-    for (i = 0; i < cpi->svc.number_spatial_layers; ++i) {
-      cpi->svc.layer_context[i].svc_params_received.spatial_layer = -1;
-    }
+    lc->max_q = params->max_quantizers[i];
+    lc->min_q = params->min_quantizers[i];
+    lc->scaling_factor_num = params->scaling_factor_num[i];
+    lc->scaling_factor_den = params->scaling_factor_den[i];
   }
 
-  cpi->svc.layer_context[params->spatial_layer].svc_params_received =
-      *params;
-
   return VPX_CODEC_OK;
 }
 
@@ -1231,7 +1233,6 @@
   {VP8E_SET_ACTIVEMAP,                ctrl_set_active_map},
   {VP8E_SET_SCALEMODE,                ctrl_set_scale_mode},
   {VP8E_SET_CPUUSED,                  ctrl_set_cpuused},
-  {VP8E_SET_NOISE_SENSITIVITY,        ctrl_set_noise_sensitivity},
   {VP8E_SET_ENABLEAUTOALTREF,         ctrl_set_enable_auto_alt_ref},
   {VP8E_SET_SHARPNESS,                ctrl_set_sharpness},
   {VP8E_SET_STATIC_THRESHOLD,         ctrl_set_static_thresh},
@@ -1251,6 +1252,7 @@
   {VP9E_SET_SVC_PARAMETERS,           ctrl_set_svc_parameters},
   {VP9E_SET_SVC_LAYER_ID,             ctrl_set_svc_layer_id},
   {VP9E_SET_TUNE_CONTENT,             ctrl_set_tune_content},
+  {VP9E_SET_NOISE_SENSITIVITY,        ctrl_set_noise_sensitivity},
 
   // Getters
   {VP8E_GET_LAST_QUANTIZER,           ctrl_get_quantizer},
@@ -1333,6 +1335,9 @@
 CODEC_INTERFACE(vpx_codec_vp9_cx) = {
   "WebM Project VP9 Encoder" VERSION_STRING,
   VPX_CODEC_INTERNAL_ABI_VERSION,
+#if CONFIG_VP9_HIGHBITDEPTH
+  VPX_CODEC_CAP_HIGHBITDEPTH |
+#endif
   VPX_CODEC_CAP_ENCODER | VPX_CODEC_CAP_PSNR,  // vpx_codec_caps_t
   encoder_init,       // vpx_codec_init_fn_t
   encoder_destroy,    // vpx_codec_destroy_fn_t
diff --git a/vpx/exports_enc b/vpx/exports_enc
index 07f0280..8885920 100644
--- a/vpx/exports_enc
+++ b/vpx/exports_enc
@@ -8,17 +8,7 @@
 text vpx_codec_set_cx_data_buf
 text vpx_svc_dump_statistics
 text vpx_svc_encode
-text vpx_svc_get_buffer
-text vpx_svc_get_encode_frame_count
-text vpx_svc_get_frame_size
 text vpx_svc_get_message
 text vpx_svc_init
-text vpx_svc_is_keyframe
 text vpx_svc_release
-text vpx_svc_set_keyframe
 text vpx_svc_set_options
-text vpx_svc_set_quantizers
-text vpx_svc_set_scale_factors
-text vpx_svc_get_layer_resolution
-text vpx_svc_get_rc_stats_buffer_size
-text vpx_svc_get_rc_stats_buffer
diff --git a/vpx/src/svc_encodeframe.c b/vpx/src/svc_encodeframe.c
index 8911e83..773087d 100644
--- a/vpx/src/svc_encodeframe.c
+++ b/vpx/src/svc_encodeframe.c
@@ -15,12 +15,12 @@
 
 #include <assert.h>
 #include <math.h>
+#include <limits.h>
 #include <stdarg.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #define VPX_DISABLE_CTRL_TYPECHECKS 1
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "./vpx_config.h"
 #include "vpx/svc_context.h"
 #include "vpx/vp8cx.h"
@@ -44,11 +44,34 @@
 #define SVC_REFERENCE_FRAMES 8
 #define SUPERFRAME_SLOTS (8)
 #define SUPERFRAME_BUFFER_SIZE (SUPERFRAME_SLOTS * sizeof(uint32_t) + 2)
-#define OPTION_BUFFER_SIZE 256
+#define OPTION_BUFFER_SIZE 1024
 #define COMPONENTS 4  // psnr & sse statistics maintained for total, y, u, v
 
-static const char *DEFAULT_QUANTIZER_VALUES = "60,53,39,33,27";
-static const char *DEFAULT_SCALE_FACTORS = "4/16,5/16,7/16,11/16,16/16";
+#define MAX_QUANTIZER 63
+
+static const int DEFAULT_SCALE_FACTORS_NUM[VPX_SS_MAX_LAYERS] = {
+  4, 5, 7, 11, 16
+};
+
+static const int DEFAULT_SCALE_FACTORS_DEN[VPX_SS_MAX_LAYERS] = {
+  16, 16, 16, 16, 16
+};
+
+typedef enum {
+  QUANTIZER = 0,
+  BITRATE,
+  SCALE_FACTOR,
+  AUTO_ALT_REF,
+  ALL_OPTION_TYPES
+} LAYER_OPTION_TYPE;
+
+static const int option_max_values[ALL_OPTION_TYPES] = {
+  63, INT_MAX, INT_MAX, 1
+};
+
+static const int option_min_values[ALL_OPTION_TYPES] = {
+  0, 0, 1, 0
+};
 
 // One encoded frame
 typedef struct FrameData {
@@ -60,14 +83,11 @@
 
 typedef struct SvcInternal {
   char options[OPTION_BUFFER_SIZE];        // set by vpx_svc_set_options
-  char quantizers[OPTION_BUFFER_SIZE];     // set by vpx_svc_set_quantizers
-  char scale_factors[OPTION_BUFFER_SIZE];  // set by vpx_svc_set_scale_factors
 
   // values extracted from option, quantizers
-  int scaling_factor_num[VPX_SS_MAX_LAYERS];
-  int scaling_factor_den[VPX_SS_MAX_LAYERS];
-  int quantizer[VPX_SS_MAX_LAYERS];
+  vpx_svc_extra_cfg_t svc_params;
   int enable_auto_alt_ref[VPX_SS_MAX_LAYERS];
+  int bitrates[VPX_SS_MAX_LAYERS];
 
   // accumulated statistics
   double psnr_sum[VPX_SS_MAX_LAYERS][COMPONENTS];   // total/Y/U/V
@@ -80,73 +100,14 @@
   int kf_dist;  // distance between keyframes
 
   // state variables
-  int encode_frame_count;
-  int frame_received;
-  int frame_within_gop;
-  int layers;
+  int psnr_pkt_received;
   int layer;
-  int is_keyframe;
   int use_multiple_frame_contexts;
 
-  FrameData *frame_list;
-  FrameData *frame_temp;
-
-  char *rc_stats_buf;
-  size_t rc_stats_buf_size;
-  size_t rc_stats_buf_used;
-
   char message_buffer[2048];
   vpx_codec_ctx_t *codec_ctx;
 } SvcInternal;
 
-// create FrameData from encoder output
-static struct FrameData *fd_create(void *buf, size_t size,
-                                   vpx_codec_frame_flags_t flags) {
-  struct FrameData *const frame_data =
-      (struct FrameData *)vpx_malloc(sizeof(*frame_data));
-  if (frame_data == NULL) {
-    return NULL;
-  }
-  frame_data->buf = vpx_malloc(size);
-  if (frame_data->buf == NULL) {
-    vpx_free(frame_data);
-    return NULL;
-  }
-  vpx_memcpy(frame_data->buf, buf, size);
-  frame_data->size = size;
-  frame_data->flags = flags;
-  return frame_data;
-}
-
-// free FrameData
-static void fd_free(struct FrameData *p) {
-  if (p) {
-    if (p->buf)
-      vpx_free(p->buf);
-    vpx_free(p);
-  }
-}
-
-// add FrameData to list
-static void fd_list_add(struct FrameData **list, struct FrameData *layer_data) {
-  struct FrameData **p = list;
-
-  while (*p != NULL) p = &(*p)->next;
-  *p = layer_data;
-  layer_data->next = NULL;
-}
-
-// free FrameData list
-static void fd_free_list(struct FrameData *list) {
-  struct FrameData *p = list;
-
-  while (p) {
-    list = list->next;
-    fd_free(p);
-    p = list;
-  }
-}
-
 static SvcInternal *get_svc_internal(SvcContext *svc_ctx) {
   if (svc_ctx == NULL) return NULL;
   if (svc_ctx->internal == NULL) {
@@ -197,158 +158,63 @@
   return retval;
 }
 
-static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
-                                              const char *quantizer_values) {
-  char *input_string;
-  char *token;
-  const char *delim = ",";
-  char *save_ptr;
-  int found = 0;
-  int i, q;
-  vpx_codec_err_t res = VPX_CODEC_OK;
-  SvcInternal *const si = get_svc_internal(svc_ctx);
+static vpx_codec_err_t extract_option(LAYER_OPTION_TYPE type,
+                                      char *input,
+                                      int *value0,
+                                      int *value1) {
+  if (type == SCALE_FACTOR) {
+    *value0 = strtol(input, &input, 10);
+    if (*input++ != '/')
+      return VPX_CODEC_INVALID_PARAM;
+    *value1 = strtol(input, &input, 10);
 
-  if (quantizer_values == NULL || strlen(quantizer_values) == 0) {
-    input_string = strdup(DEFAULT_QUANTIZER_VALUES);
+    if (*value0 < option_min_values[SCALE_FACTOR] ||
+        *value1 < option_min_values[SCALE_FACTOR] ||
+        *value0 > option_max_values[SCALE_FACTOR] ||
+        *value1 > option_max_values[SCALE_FACTOR] ||
+        *value0 > *value1)  // num shouldn't be greater than den
+      return VPX_CODEC_INVALID_PARAM;
   } else {
-    input_string = strdup(quantizer_values);
+    *value0 = atoi(input);
+    if (*value0 < option_min_values[type] ||
+        *value0 > option_max_values[type])
+      return VPX_CODEC_INVALID_PARAM;
   }
-
-  token = strtok_r(input_string, delim, &save_ptr);
-  for (i = 0; i < svc_ctx->spatial_layers; ++i) {
-    if (token != NULL) {
-      q = atoi(token);
-      if (q <= 0 || q > 100) {
-        svc_log(svc_ctx, SVC_LOG_ERROR,
-                "svc-quantizer-values: invalid value %s\n", token);
-        res = VPX_CODEC_INVALID_PARAM;
-        break;
-      }
-      token = strtok_r(NULL, delim, &save_ptr);
-      found = i + 1;
-    } else {
-      q = 0;
-    }
-    si->quantizer[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] = q;
-  }
-  if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
-    svc_log(svc_ctx, SVC_LOG_ERROR,
-            "svc: quantizers: %d values required, but only %d specified\n",
-            svc_ctx->spatial_layers, found);
-    res = VPX_CODEC_INVALID_PARAM;
-  }
-  free(input_string);
-  return res;
+  return VPX_CODEC_OK;
 }
 
-static vpx_codec_err_t parse_auto_alt_ref(SvcContext *svc_ctx,
-                                          const char *alt_ref_options) {
-  char *input_string;
-  char *token;
-  const char *delim = ",";
-  char *save_ptr;
-  int found = 0, enabled = 0;
-  int i, value;
-  vpx_codec_err_t res = VPX_CODEC_OK;
-  SvcInternal *const si = get_svc_internal(svc_ctx);
-
-  if (alt_ref_options == NULL || strlen(alt_ref_options) == 0) {
-    return VPX_CODEC_INVALID_PARAM;
-  } else {
-    input_string = strdup(alt_ref_options);
-  }
-
-  token = strtok_r(input_string, delim, &save_ptr);
-  for (i = 0; i < svc_ctx->spatial_layers; ++i) {
-    if (token != NULL) {
-      value = atoi(token);
-      if (value < 0 || value > 1) {
-        svc_log(svc_ctx, SVC_LOG_ERROR,
-                "enable auto alt ref values: invalid value %s\n", token);
-        res = VPX_CODEC_INVALID_PARAM;
-        break;
-      }
-      token = strtok_r(NULL, delim, &save_ptr);
-      found = i + 1;
-    } else {
-      value = 0;
-    }
-    si->enable_auto_alt_ref[i] = value;
-    if (value > 0)
-      ++enabled;
-  }
-  if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
-    svc_log(svc_ctx, SVC_LOG_ERROR,
-            "svc: quantizers: %d values required, but only %d specified\n",
-            svc_ctx->spatial_layers, found);
-    res = VPX_CODEC_INVALID_PARAM;
-  }
-  if (enabled > REF_FRAMES - svc_ctx->spatial_layers) {
-    svc_log(svc_ctx, SVC_LOG_ERROR,
-            "svc: auto alt ref: Maxinum %d(REF_FRAMES - layers) layers could"
-            "enabled auto alt reference frame, but % layers are enabled\n",
-            REF_FRAMES - svc_ctx->spatial_layers, enabled);
-    res = VPX_CODEC_INVALID_PARAM;
-  }
-  free(input_string);
-  return res;
-}
-
-static void log_invalid_scale_factor(SvcContext *svc_ctx, const char *value) {
-  svc_log(svc_ctx, SVC_LOG_ERROR, "svc scale-factors: invalid value %s\n",
-          value);
-}
-
-static vpx_codec_err_t parse_scale_factors(SvcContext *svc_ctx,
-                                           const char *scale_factors) {
-  char *input_string;
-  char *token;
-  const char *delim = ",";
-  char *save_ptr;
-  int found = 0;
+static vpx_codec_err_t parse_layer_options_from_string(SvcContext *svc_ctx,
+                                                       LAYER_OPTION_TYPE type,
+                                                       const char *input,
+                                                       int *option0,
+                                                       int *option1) {
   int i;
-  int64_t num, den;
   vpx_codec_err_t res = VPX_CODEC_OK;
-  SvcInternal *const si = get_svc_internal(svc_ctx);
+  char *input_string;
+  char *token;
+  const char *delim = ",";
+  char *save_ptr;
 
-  if (scale_factors == NULL || strlen(scale_factors) == 0) {
-    input_string = strdup(DEFAULT_SCALE_FACTORS);
-  } else {
-    input_string = strdup(scale_factors);
-  }
+  if (input == NULL || option0 == NULL ||
+      (option1 == NULL && type == SCALE_FACTOR))
+    return VPX_CODEC_INVALID_PARAM;
+
+  input_string = strdup(input);
   token = strtok_r(input_string, delim, &save_ptr);
   for (i = 0; i < svc_ctx->spatial_layers; ++i) {
-    num = den = 0;
     if (token != NULL) {
-      num = strtol(token, &token, 10);
-      if (num <= 0) {
-        log_invalid_scale_factor(svc_ctx, token);
-        res = VPX_CODEC_INVALID_PARAM;
+      res = extract_option(type, token, option0 + i, option1 + i);
+      if (res != VPX_CODEC_OK)
         break;
-      }
-      if (*token++ != '/') {
-        log_invalid_scale_factor(svc_ctx, token);
-        res = VPX_CODEC_INVALID_PARAM;
-        break;
-      }
-      den = strtol(token, &token, 10);
-      if (den <= 0) {
-        log_invalid_scale_factor(svc_ctx, token);
-        res = VPX_CODEC_INVALID_PARAM;
-        break;
-      }
       token = strtok_r(NULL, delim, &save_ptr);
-      found = i + 1;
+    } else {
+      break;
     }
-    si->scaling_factor_num[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
-        (int)num;
-    si->scaling_factor_den[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
-        (int)den;
   }
-  if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
+  if (res == VPX_CODEC_OK && i != svc_ctx->spatial_layers) {
     svc_log(svc_ctx, SVC_LOG_ERROR,
-            "svc: scale-factors: %d values required, but only %d specified\n",
-            svc_ctx->spatial_layers, found);
+            "svc: layer params type: %d    %d values required, "
+            "but only %d specified\n", type, svc_ctx->spatial_layers, i);
     res = VPX_CODEC_INVALID_PARAM;
   }
   free(input_string);
@@ -369,6 +235,7 @@
   char *input_ptr;
   SvcInternal *const si = get_svc_internal(svc_ctx);
   vpx_codec_err_t res = VPX_CODEC_OK;
+  int i, alt_ref_enabled = 0;
 
   if (options == NULL) return VPX_CODEC_OK;
   input_string = strdup(options);
@@ -389,13 +256,27 @@
     } else if (strcmp("temporal-layers", option_name) == 0) {
       svc_ctx->temporal_layers = atoi(option_value);
     } else if (strcmp("scale-factors", option_name) == 0) {
-      res = parse_scale_factors(svc_ctx, option_value);
+      res = parse_layer_options_from_string(svc_ctx, SCALE_FACTOR, option_value,
+                                            si->svc_params.scaling_factor_num,
+                                            si->svc_params.scaling_factor_den);
       if (res != VPX_CODEC_OK) break;
-    } else if (strcmp("quantizers", option_name) == 0) {
-      res = parse_quantizer_values(svc_ctx, option_value);
+    } else if (strcmp("max-quantizers", option_name) == 0) {
+      res = parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value,
+                                            si->svc_params.max_quantizers,
+                                            NULL);
+      if (res != VPX_CODEC_OK) break;
+    } else if (strcmp("min-quantizers", option_name) == 0) {
+      res = parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value,
+                                            si->svc_params.min_quantizers,
+                                            NULL);
       if (res != VPX_CODEC_OK) break;
     } else if (strcmp("auto-alt-refs", option_name) == 0) {
-      res = parse_auto_alt_ref(svc_ctx, option_value);
+      res = parse_layer_options_from_string(svc_ctx, AUTO_ALT_REF, option_value,
+                                            si->enable_auto_alt_ref, NULL);
+      if (res != VPX_CODEC_OK) break;
+    } else if (strcmp("bitrates", option_name) == 0) {
+      res = parse_layer_options_from_string(svc_ctx, BITRATE, option_value,
+                                            si->bitrates, NULL);
       if (res != VPX_CODEC_OK) break;
     } else if (strcmp("multi-frame-contexts", option_name) == 0) {
       si->use_multiple_frame_contexts = atoi(option_value);
@@ -408,11 +289,29 @@
   }
   free(input_string);
 
+  for (i = 0; i < svc_ctx->spatial_layers; ++i) {
+    if (si->svc_params.max_quantizers[i] > MAX_QUANTIZER ||
+        si->svc_params.max_quantizers[i] < 0 ||
+        si->svc_params.min_quantizers[i] > si->svc_params.max_quantizers[i] ||
+        si->svc_params.min_quantizers[i] < 0)
+      res = VPX_CODEC_INVALID_PARAM;
+  }
+
   if (si->use_multiple_frame_contexts &&
       (svc_ctx->spatial_layers > 3 ||
        svc_ctx->spatial_layers * svc_ctx->temporal_layers > 4))
     res = VPX_CODEC_INVALID_PARAM;
 
+  for (i = 0; i < svc_ctx->spatial_layers; ++i)
+    alt_ref_enabled += si->enable_auto_alt_ref[i];
+  if (alt_ref_enabled > REF_FRAMES - svc_ctx->spatial_layers) {
+    svc_log(svc_ctx, SVC_LOG_ERROR,
+            "svc: auto alt ref: Maxinum %d(REF_FRAMES - layers) layers could"
+            "enabled auto alt reference frame, but % layers are enabled\n",
+            REF_FRAMES - svc_ctx->spatial_layers, alt_ref_enabled);
+    res = VPX_CODEC_INVALID_PARAM;
+  }
+
   return res;
 }
 
@@ -426,26 +325,38 @@
   return VPX_CODEC_OK;
 }
 
-vpx_codec_err_t vpx_svc_set_quantizers(SvcContext *svc_ctx,
-                                       const char *quantizers) {
-  SvcInternal *const si = get_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || quantizers == NULL || si == NULL) {
-    return VPX_CODEC_INVALID_PARAM;
-  }
-  strncpy(si->quantizers, quantizers, sizeof(si->quantizers));
-  si->quantizers[sizeof(si->quantizers) - 1] = '\0';
-  return VPX_CODEC_OK;
-}
+void assign_layer_bitrates(const SvcContext *svc_ctx,
+                           vpx_codec_enc_cfg_t *const enc_cfg) {
+  int i;
+  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
 
-vpx_codec_err_t vpx_svc_set_scale_factors(SvcContext *svc_ctx,
-                                          const char *scale_factors) {
-  SvcInternal *const si = get_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || scale_factors == NULL || si == NULL) {
-    return VPX_CODEC_INVALID_PARAM;
+  if (si->bitrates[0] != 0) {
+    enc_cfg->rc_target_bitrate = 0;
+    for (i = 0; i < svc_ctx->spatial_layers; ++i) {
+      enc_cfg->ss_target_bitrate[i] = (unsigned int)si->bitrates[i];
+      enc_cfg->rc_target_bitrate += si->bitrates[i];
+    }
+  } else {
+    float total = 0;
+    float alloc_ratio[VPX_SS_MAX_LAYERS] = {0};
+
+    for (i = 0; i < svc_ctx->spatial_layers; ++i) {
+      if (si->svc_params.scaling_factor_den[i] > 0) {
+        alloc_ratio[i] = (float)(si->svc_params.scaling_factor_num[i] * 1.0 /
+                                 si->svc_params.scaling_factor_den[i]);
+
+        alloc_ratio[i] *= alloc_ratio[i];
+        total += alloc_ratio[i];
+      }
+    }
+
+    for (i = 0; i < svc_ctx->spatial_layers; ++i) {
+      if (total > 0) {
+        enc_cfg->ss_target_bitrate[i] = (unsigned int)
+            (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total);
+      }
+    }
   }
-  strncpy(si->scale_factors, scale_factors, sizeof(si->scale_factors));
-  si->scale_factors[sizeof(si->scale_factors) - 1] = '\0';
-  return VPX_CODEC_OK;
 }
 
 vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
@@ -481,11 +392,12 @@
     return VPX_CODEC_INVALID_PARAM;
   }
 
-  res = parse_quantizer_values(svc_ctx, si->quantizers);
-  if (res != VPX_CODEC_OK) return res;
-
-  res = parse_scale_factors(svc_ctx, si->scale_factors);
-  if (res != VPX_CODEC_OK) return res;
+  for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) {
+    si->svc_params.max_quantizers[i] = MAX_QUANTIZER;
+    si->svc_params.min_quantizers[i] = 0;
+    si->svc_params.scaling_factor_num[i] = DEFAULT_SCALE_FACTORS_NUM[i];
+    si->svc_params.scaling_factor_den[i] = DEFAULT_SCALE_FACTORS_DEN[i];
+  }
 
   // Parse aggregate command line options. Options must start with
   // "layers=xx" then followed by other options
@@ -502,38 +414,10 @@
   if (svc_ctx->temporal_layers > VPX_TS_MAX_LAYERS)
     svc_ctx->temporal_layers = VPX_TS_MAX_LAYERS;
 
-  si->layers = svc_ctx->spatial_layers;
-
-  // Assign target bitrate for each layer. We calculate the ratio
-  // from the resolution for now.
-  // TODO(Minghai): Optimize the mechanism of allocating bits after
-  // implementing svc two pass rate control.
-  if (si->layers > 1) {
-    float total = 0;
-    float alloc_ratio[VPX_SS_MAX_LAYERS] = {0};
-
-    assert(si->layers <= VPX_SS_MAX_LAYERS);
-    for (i = 0; i < si->layers; ++i) {
-      int pos = i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers;
-      if (pos < VPX_SS_MAX_LAYERS && si->scaling_factor_den[pos] > 0) {
-        alloc_ratio[i] = (float)(si->scaling_factor_num[pos] * 1.0 /
-            si->scaling_factor_den[pos]);
-
-        alloc_ratio[i] *= alloc_ratio[i];
-        total += alloc_ratio[i];
-      }
-    }
-
-    for (i = 0; i < si->layers; ++i) {
-      if (total > 0) {
-        enc_cfg->ss_target_bitrate[i] = (unsigned int)
-            (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total);
-      }
-    }
-  }
+  assign_layer_bitrates(svc_ctx, enc_cfg);
 
 #if CONFIG_SPATIAL_SVC
-  for (i = 0; i < si->layers; ++i)
+  for (i = 0; i < svc_ctx->spatial_layers; ++i)
     enc_cfg->ss_enable_auto_alt_ref[i] = si->enable_auto_alt_ref[i];
 #endif
 
@@ -547,24 +431,9 @@
   }
 
   // modify encoder configuration
-  enc_cfg->ss_number_layers = si->layers;
+  enc_cfg->ss_number_layers = svc_ctx->spatial_layers;
   enc_cfg->ts_number_layers = svc_ctx->temporal_layers;
 
-  // TODO(ivanmaltz): determine if these values need to be set explicitly for
-  // svc, or if the normal default/override mechanism can be used
-  enc_cfg->rc_dropframe_thresh = 0;
-  enc_cfg->rc_resize_allowed = 0;
-
-  if (enc_cfg->g_pass == VPX_RC_ONE_PASS) {
-    enc_cfg->rc_min_quantizer = 33;
-    enc_cfg->rc_max_quantizer = 33;
-  }
-
-  enc_cfg->rc_undershoot_pct = 100;
-  enc_cfg->rc_overshoot_pct = 15;
-  enc_cfg->rc_buf_initial_sz = 500;
-  enc_cfg->rc_buf_optimal_sz = 600;
-  enc_cfg->rc_buf_sz = 1000;
   if (enc_cfg->g_error_resilient == 0 && si->use_multiple_frame_contexts == 0)
     enc_cfg->g_error_resilient = 1;
 
@@ -576,71 +445,11 @@
   }
 
   vpx_codec_control(codec_ctx, VP9E_SET_SVC, 1);
-  vpx_codec_control(codec_ctx, VP8E_SET_TOKEN_PARTITIONS, 1);
+  vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &si->svc_params);
 
   return VPX_CODEC_OK;
 }
 
-vpx_codec_err_t vpx_svc_get_layer_resolution(const SvcContext *svc_ctx,
-                                             int layer,
-                                             unsigned int *width,
-                                             unsigned int *height) {
-  int w, h, index, num, den;
-  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
-
-  if (svc_ctx == NULL || si == NULL || width == NULL || height == NULL) {
-    return VPX_CODEC_INVALID_PARAM;
-  }
-  if (layer < 0 || layer >= si->layers) return VPX_CODEC_INVALID_PARAM;
-
-  index = layer + VPX_SS_MAX_LAYERS - si->layers;
-  num = si->scaling_factor_num[index];
-  den = si->scaling_factor_den[index];
-  if (num == 0 || den == 0) return VPX_CODEC_INVALID_PARAM;
-
-  w = si->width * num / den;
-  h = si->height * num / den;
-
-  // make height and width even to make chrome player happy
-  w += w % 2;
-  h += h % 2;
-
-  *width = w;
-  *height = h;
-
-  return VPX_CODEC_OK;
-}
-
-static void set_svc_parameters(SvcContext *svc_ctx,
-                               vpx_codec_ctx_t *codec_ctx) {
-  int layer, layer_index;
-  vpx_svc_parameters_t svc_params;
-  SvcInternal *const si = get_svc_internal(svc_ctx);
-
-  memset(&svc_params, 0, sizeof(svc_params));
-  svc_params.temporal_layer = 0;
-  svc_params.spatial_layer = si->layer;
-
-  layer = si->layer;
-  if (VPX_CODEC_OK != vpx_svc_get_layer_resolution(svc_ctx, layer,
-                                                   &svc_params.width,
-                                                   &svc_params.height)) {
-    svc_log(svc_ctx, SVC_LOG_ERROR, "vpx_svc_get_layer_resolution failed\n");
-  }
-  layer_index = layer + VPX_SS_MAX_LAYERS - si->layers;
-
-  if (codec_ctx->config.enc->g_pass == VPX_RC_ONE_PASS) {
-    svc_params.min_quantizer = si->quantizer[layer_index];
-    svc_params.max_quantizer = si->quantizer[layer_index];
-  } else {
-    svc_params.min_quantizer = codec_ctx->config.enc->rc_min_quantizer;
-    svc_params.max_quantizer = codec_ctx->config.enc->rc_max_quantizer;
-  }
-
-  svc_params.distance_from_i_frame = si->frame_within_gop;
-  vpx_codec_control(codec_ctx, VP9E_SET_SVC_PARAMETERS, &svc_params);
-}
-
 /**
  * Encode a frame into multiple layers
  * Create a superframe containing the individual layers
@@ -651,34 +460,12 @@
   vpx_codec_err_t res;
   vpx_codec_iter_t iter;
   const vpx_codec_cx_pkt_t *cx_pkt;
-  int layer_for_psnr = 0;
   SvcInternal *const si = get_svc_internal(svc_ctx);
   if (svc_ctx == NULL || codec_ctx == NULL || si == NULL) {
     return VPX_CODEC_INVALID_PARAM;
   }
 
   svc_log_reset(svc_ctx);
-  si->rc_stats_buf_used = 0;
-
-  si->layers = svc_ctx->spatial_layers;
-  if (si->encode_frame_count == 0) {
-    si->frame_within_gop = 0;
-  }
-  si->is_keyframe = (si->frame_within_gop == 0);
-
-  if (rawimg != NULL) {
-    svc_log(svc_ctx, SVC_LOG_DEBUG,
-            "vpx_svc_encode  layers: %d, frame_count: %d, "
-            "frame_within_gop: %d\n", si->layers, si->encode_frame_count,
-            si->frame_within_gop);
-  }
-
-  if (rawimg != NULL) {
-    // encode each layer
-    for (si->layer = 0; si->layer < si->layers; ++si->layer) {
-      set_svc_parameters(svc_ctx, codec_ctx);
-    }
-  }
 
   res = vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0,
                          deadline);
@@ -689,64 +476,40 @@
   iter = NULL;
   while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
     switch (cx_pkt->kind) {
-      case VPX_CODEC_CX_FRAME_PKT: {
-        fd_list_add(&si->frame_list, fd_create(cx_pkt->data.frame.buf,
-                                               cx_pkt->data.frame.sz,
-                                               cx_pkt->data.frame.flags));
-
-        svc_log(svc_ctx, SVC_LOG_DEBUG, "SVC frame: %d, kf: %d, size: %d, "
-                "pts: %d\n", si->frame_received,
-                (cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY) ? 1 : 0,
-                (int)cx_pkt->data.frame.sz, (int)cx_pkt->data.frame.pts);
-
-        ++si->frame_received;
-        layer_for_psnr = 0;
-        break;
-      }
-      case VPX_CODEC_PSNR_PKT: {
-        int i;
-        svc_log(svc_ctx, SVC_LOG_DEBUG,
-                "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
-                "%2.3f  %2.3f  %2.3f  %2.3f \n",
-                si->frame_received, layer_for_psnr,
-                cx_pkt->data.psnr.psnr[0], cx_pkt->data.psnr.psnr[1],
-                cx_pkt->data.psnr.psnr[2], cx_pkt->data.psnr.psnr[3]);
-        svc_log(svc_ctx, SVC_LOG_DEBUG,
-                "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
-                "%2.3f  %2.3f  %2.3f  %2.3f \n",
-                si->frame_received, layer_for_psnr,
-                cx_pkt->data.psnr.sse[0], cx_pkt->data.psnr.sse[1],
-                cx_pkt->data.psnr.sse[2], cx_pkt->data.psnr.sse[3]);
-        for (i = 0; i < COMPONENTS; i++) {
-          si->psnr_sum[layer_for_psnr][i] += cx_pkt->data.psnr.psnr[i];
-          si->sse_sum[layer_for_psnr][i] += cx_pkt->data.psnr.sse[i];
-        }
-        ++layer_for_psnr;
-        break;
-      }
-      case VPX_CODEC_STATS_PKT: {
-        size_t new_size = si->rc_stats_buf_used +
-            cx_pkt->data.twopass_stats.sz;
-
-        if (new_size > si->rc_stats_buf_size) {
-          char *p = (char*)realloc(si->rc_stats_buf, new_size);
-          if (p == NULL) {
-            svc_log(svc_ctx, SVC_LOG_ERROR, "Error allocating stats buf\n");
-            return VPX_CODEC_MEM_ERROR;
-          }
-          si->rc_stats_buf = p;
-          si->rc_stats_buf_size = new_size;
-        }
-
-        memcpy(si->rc_stats_buf + si->rc_stats_buf_used,
-               cx_pkt->data.twopass_stats.buf, cx_pkt->data.twopass_stats.sz);
-        si->rc_stats_buf_used += cx_pkt->data.twopass_stats.sz;
-        break;
-      }
 #if CONFIG_SPATIAL_SVC
+      case VPX_CODEC_SPATIAL_SVC_LAYER_PSNR: {
+        int i;
+        for (i = 0; i < svc_ctx->spatial_layers; ++i) {
+          int j;
+          svc_log(svc_ctx, SVC_LOG_DEBUG,
+                  "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
+                  "%2.3f  %2.3f  %2.3f  %2.3f \n",
+                  si->psnr_pkt_received, i,
+                  cx_pkt->data.layer_psnr[i].psnr[0],
+                  cx_pkt->data.layer_psnr[i].psnr[1],
+                  cx_pkt->data.layer_psnr[i].psnr[2],
+                  cx_pkt->data.layer_psnr[i].psnr[3]);
+          svc_log(svc_ctx, SVC_LOG_DEBUG,
+                  "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
+                  "%2.3f  %2.3f  %2.3f  %2.3f \n",
+                  si->psnr_pkt_received, i,
+                  cx_pkt->data.layer_psnr[i].sse[0],
+                  cx_pkt->data.layer_psnr[i].sse[1],
+                  cx_pkt->data.layer_psnr[i].sse[2],
+                  cx_pkt->data.layer_psnr[i].sse[3]);
+
+          for (j = 0; j < COMPONENTS; ++j) {
+            si->psnr_sum[i][j] +=
+                cx_pkt->data.layer_psnr[i].psnr[j];
+            si->sse_sum[i][j] += cx_pkt->data.layer_psnr[i].sse[j];
+          }
+        }
+        ++si->psnr_pkt_received;
+        break;
+      }
       case VPX_CODEC_SPATIAL_SVC_LAYER_SIZES: {
         int i;
-        for (i = 0; i < si->layers; ++i)
+        for (i = 0; i < svc_ctx->spatial_layers; ++i)
           si->bytes_sum[i] += cx_pkt->data.layer_sizes[i];
         break;
       }
@@ -757,11 +520,6 @@
     }
   }
 
-  if (rawimg != NULL) {
-    ++si->frame_within_gop;
-    ++si->encode_frame_count;
-  }
-
   return VPX_CODEC_OK;
 }
 
@@ -771,47 +529,6 @@
   return si->message_buffer;
 }
 
-// We will maintain a list of output frame buffers since with lag_in_frame
-// we need to output all frame buffers at the end. vpx_svc_get_buffer() will
-// remove a frame buffer from the list the put it to a temporal pointer, which
-// will be removed at the next vpx_svc_get_buffer() or when closing encoder.
-void *vpx_svc_get_buffer(SvcContext *svc_ctx) {
-  SvcInternal *const si = get_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL || si->frame_list == NULL) return NULL;
-
-  if (si->frame_temp)
-    fd_free(si->frame_temp);
-
-  si->frame_temp = si->frame_list;
-  si->frame_list = si->frame_list->next;
-
-  return si->frame_temp->buf;
-}
-
-size_t vpx_svc_get_frame_size(const SvcContext *svc_ctx) {
-  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL || si->frame_list == NULL) return 0;
-  return si->frame_list->size;
-}
-
-int vpx_svc_get_encode_frame_count(const SvcContext *svc_ctx) {
-  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL) return 0;
-  return si->encode_frame_count;
-}
-
-int vpx_svc_is_keyframe(const SvcContext *svc_ctx) {
-  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL || si->frame_list == NULL) return 0;
-  return (si->frame_list->flags & VPX_FRAME_IS_KEY) != 0;
-}
-
-void vpx_svc_set_keyframe(SvcContext *svc_ctx) {
-  SvcInternal *const si = get_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL) return;
-  si->frame_within_gop = 0;
-}
-
 static double calc_psnr(double d) {
   if (d == 0) return 100;
   return -10.0 * log(d) / log(10.0);
@@ -819,7 +536,7 @@
 
 // dump accumulated statistics and reset accumulated values
 const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
-  int number_of_frames, encode_frame_count;
+  int number_of_frames;
   int i, j;
   uint32_t bytes_total = 0;
   double scale[COMPONENTS];
@@ -832,12 +549,11 @@
 
   svc_log_reset(svc_ctx);
 
-  encode_frame_count = si->encode_frame_count;
-  if (si->encode_frame_count <= 0) return vpx_svc_get_message(svc_ctx);
+  number_of_frames = si->psnr_pkt_received;
+  if (number_of_frames <= 0) return vpx_svc_get_message(svc_ctx);
 
   svc_log(svc_ctx, SVC_LOG_INFO, "\n");
-  for (i = 0; i < si->layers; ++i) {
-    number_of_frames = encode_frame_count;
+  for (i = 0; i < svc_ctx->spatial_layers; ++i) {
 
     svc_log(svc_ctx, SVC_LOG_INFO,
             "Layer %d Average PSNR=[%2.3f, %2.3f, %2.3f, %2.3f], Bytes=[%u]\n",
@@ -872,7 +588,7 @@
   }
 
   // only display statistics once
-  si->encode_frame_count = 0;
+  si->psnr_pkt_received = 0;
 
   svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total);
   return vpx_svc_get_message(svc_ctx);
@@ -885,26 +601,8 @@
   // SvcInternal if it was not already allocated
   si = (SvcInternal *)svc_ctx->internal;
   if (si != NULL) {
-    fd_free(si->frame_temp);
-    fd_free_list(si->frame_list);
-    if (si->rc_stats_buf) {
-      free(si->rc_stats_buf);
-    }
     free(si);
     svc_ctx->internal = NULL;
   }
 }
 
-size_t vpx_svc_get_rc_stats_buffer_size(const SvcContext *svc_ctx) {
-  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL) return 0;
-  return si->rc_stats_buf_used;
-}
-
-char *vpx_svc_get_rc_stats_buffer(const SvcContext *svc_ctx) {
-  const SvcInternal *const si = get_const_svc_internal(svc_ctx);
-  if (svc_ctx == NULL || si == NULL) return NULL;
-  return si->rc_stats_buf;
-}
-
-
diff --git a/vpx/src/vpx_encoder.c b/vpx/src/vpx_encoder.c
index 1903b55..cd10c41 100644
--- a/vpx/src/vpx_encoder.c
+++ b/vpx/src/vpx_encoder.c
@@ -15,8 +15,8 @@
  */
 #include <limits.h>
 #include <string.h>
-#include "vpx/internal/vpx_codec_internal.h"
 #include "vpx_config.h"
+#include "vpx/internal/vpx_codec_internal.h"
 
 #define SAVE_STATUS(ctx,var) (ctx?(ctx->err = var):var)
 
diff --git a/vpx/svc_context.h b/vpx/svc_context.h
index eea3b13..61b5f4b 100644
--- a/vpx/svc_context.h
+++ b/vpx/svc_context.h
@@ -52,22 +52,6 @@
 vpx_codec_err_t vpx_svc_set_options(SvcContext *svc_ctx, const char *options);
 
 /**
- * Set SVC quantizer values
- * values comma separated, ordered from lowest resolution to highest
- * e.g., "60,53,39,33,27"
- */
-vpx_codec_err_t vpx_svc_set_quantizers(SvcContext *svc_ctx,
-                                       const char *quantizer_values);
-
-/**
- * Set SVC scale factors
- * values comma separated, ordered from lowest resolution to highest
- * e.g.,  "4/16,5/16,7/16,11/16,16/16"
- */
-vpx_codec_err_t vpx_svc_set_scale_factors(SvcContext *svc_ctx,
-                                          const char *scale_factors);
-
-/**
  * initialize SVC encoding
  */
 vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
@@ -95,51 +79,6 @@
  */
 const char *vpx_svc_get_message(const SvcContext *svc_ctx);
 
-/**
- * return size of encoded data to be returned by vpx_svc_get_buffer.
- * it needs to be called before vpx_svc_get_buffer.
- */
-size_t vpx_svc_get_frame_size(const SvcContext *svc_ctx);
-
-/**
- * return buffer with encoded data. encoder will maintain a list of frame
- * buffers. each call of vpx_svc_get_buffer() will return one frame.
- */
-void *vpx_svc_get_buffer(SvcContext *svc_ctx);
-
-/**
- * return size of two pass rate control stats data to be returned by
- * vpx_svc_get_rc_stats_buffer
- */
-size_t vpx_svc_get_rc_stats_buffer_size(const SvcContext *svc_ctx);
-
-/**
- * return buffer two pass of rate control stats data
- */
-char *vpx_svc_get_rc_stats_buffer(const SvcContext *svc_ctx);
-
-/**
- * return spatial resolution of the specified layer
- */
-vpx_codec_err_t vpx_svc_get_layer_resolution(const SvcContext *svc_ctx,
-                                             int layer,
-                                             unsigned int *width,
-                                             unsigned int *height);
-/**
- * return number of frames that have been encoded
- */
-int vpx_svc_get_encode_frame_count(const SvcContext *svc_ctx);
-
-/**
- * return 1 if last encoded frame was a keyframe
- */
-int vpx_svc_is_keyframe(const SvcContext *svc_ctx);
-
-/**
- * force the next frame to be a keyframe
- */
-void vpx_svc_set_keyframe(SvcContext *svc_ctx);
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/vpx/vp8cx.h b/vpx/vp8cx.h
index 796a7a1..77d9d6a 100644
--- a/vpx/vp8cx.h
+++ b/vpx/vp8cx.h
@@ -148,7 +148,12 @@
    */
   VP8E_SET_CPUUSED           = 13,
   VP8E_SET_ENABLEAUTOALTREF,       /**< control function to enable vp8 to automatic set and use altref frame */
-  VP8E_SET_NOISE_SENSITIVITY,      /**< control function to set noise sensitivity */
+  /*!\brief control function to set noise sensitivity
+   *
+   * 0: off, 1: OnYOnly, 2: OnYUV,
+   * 3: OnYUVAggressive, 4: Adaptive
+   */
+  VP8E_SET_NOISE_SENSITIVITY,
   VP8E_SET_SHARPNESS,              /**< control function to set sharpness */
   VP8E_SET_STATIC_THRESHOLD,       /**< control function to set the threshold for macroblocks treated static */
   VP8E_SET_TOKEN_PARTITIONS,       /**< control function to set the number of token partitions  */
@@ -197,6 +202,11 @@
   VP9E_SET_FRAME_PARALLEL_DECODING,
   VP9E_SET_AQ_MODE,
   VP9E_SET_FRAME_PERIODIC_BOOST,
+  /*!\brief control function to set noise sensitivity
+   *
+   *  0: off, 1: OnYOnly
+   */
+  VP9E_SET_NOISE_SENSITIVITY,
 
   VP9E_SET_SVC,
   VP9E_SET_SVC_PARAMETERS,
@@ -295,24 +305,6 @@
   VP8_TUNE_SSIM
 } vp8e_tuning;
 
-/*!\brief  vp9 svc parameters
- *
- * This defines parameters for svc encoding.
- *
- */
-typedef struct vpx_svc_parameters {
-  unsigned int width;         /**< width of current spatial layer */
-  unsigned int height;        /**< height of current spatial layer */
-  int spatial_layer;          /**< current spatial layer number - 0 = base */
-  int temporal_layer;         /**< current temporal layer number - 0 = base */
-  int max_quantizer;          /**< max quantizer for current layer */
-  int min_quantizer;          /**< min quantizer for current layer */
-  int distance_from_i_frame;  /**< frame number within current gop */
-  int lst_fb_idx;             /**< last frame frame buffer index */
-  int gld_fb_idx;             /**< golden frame frame buffer index */
-  int alt_fb_idx;             /**< alt reference frame frame buffer index */
-} vpx_svc_parameters_t;
-
 /*!\brief  vp9 svc layer parameters
  *
  * This defines the spatial and temporal layer id numbers for svc encoding.
@@ -345,7 +337,7 @@
 VPX_CTRL_USE_TYPE(VP8E_SET_SCALEMODE,          vpx_scaling_mode_t *)
 
 VPX_CTRL_USE_TYPE(VP9E_SET_SVC,                int)
-VPX_CTRL_USE_TYPE(VP9E_SET_SVC_PARAMETERS,     vpx_svc_parameters_t *)
+VPX_CTRL_USE_TYPE(VP9E_SET_SVC_PARAMETERS,     void *)
 VPX_CTRL_USE_TYPE(VP9E_SET_SVC_LAYER_ID,       vpx_svc_layer_id_t *)
 
 VPX_CTRL_USE_TYPE(VP8E_SET_CPUUSED,            int)
@@ -377,6 +369,8 @@
 
 VPX_CTRL_USE_TYPE(VP9E_SET_FRAME_PERIODIC_BOOST, unsigned int)
 
+VPX_CTRL_USE_TYPE(VP9E_SET_NOISE_SENSITIVITY,  unsigned int)
+
 VPX_CTRL_USE_TYPE(VP9E_SET_TUNE_CONTENT, int) /* vp9e_tune_content */
 /*! @} - end defgroup vp8_encoder */
 #ifdef __cplusplus
diff --git a/vpx/vpx_encoder.h b/vpx/vpx_encoder.h
index fdabed1..c6c7d08 100644
--- a/vpx/vpx_encoder.h
+++ b/vpx/vpx_encoder.h
@@ -163,6 +163,7 @@
     VPX_CODEC_PSNR_PKT,        /**< PSNR statistics for this frame */
 #if CONFIG_SPATIAL_SVC
     VPX_CODEC_SPATIAL_SVC_LAYER_SIZES, /**< Sizes for each layer in this frame*/
+    VPX_CODEC_SPATIAL_SVC_LAYER_PSNR, /**< PSNR for each layer in this frame*/
 #endif
     VPX_CODEC_CUSTOM_PKT = 256 /**< Algorithm extensions  */
   };
@@ -202,6 +203,7 @@
       vpx_fixed_buf_t raw;     /**< data for arbitrary packets */
 #if CONFIG_SPATIAL_SVC
       size_t layer_sizes[VPX_SS_MAX_LAYERS];
+      struct vpx_psnr_pkt layer_psnr[VPX_SS_MAX_LAYERS];
 #endif
 
       /* This packet size is fixed to allow codecs to extend this
@@ -709,6 +711,18 @@
     unsigned int           ts_layer_id[VPX_TS_MAX_PERIODICITY];
   } vpx_codec_enc_cfg_t; /**< alias for struct vpx_codec_enc_cfg */
 
+  /*!\brief  vp9 svc extra configure parameters
+   *
+   * This defines max/min quantizers and scale factors for each layer
+   *
+   */
+  typedef struct vpx_svc_parameters {
+    int max_quantizers[VPX_SS_MAX_LAYERS];
+    int min_quantizers[VPX_SS_MAX_LAYERS];
+    int scaling_factor_num[VPX_SS_MAX_LAYERS];
+    int scaling_factor_den[VPX_SS_MAX_LAYERS];
+  } vpx_svc_extra_cfg_t;
+
 
   /*!\brief Initialize an encoder instance
    *
diff --git a/vpx/vpx_frame_buffer.h b/vpx/vpx_frame_buffer.h
index e69df4b..41038b1 100644
--- a/vpx/vpx_frame_buffer.h
+++ b/vpx/vpx_frame_buffer.h
@@ -43,15 +43,15 @@
  *
  * This callback is invoked by the decoder to retrieve data for the frame
  * buffer in order for the decode call to complete. The callback must
- * allocate at least min_size in bytes and assign it to fb->data. Then the
- * callback must set fb->size to the allocated size. The application does not
- * need to align the allocated data. The callback is triggered when the
- * decoder needs a frame buffer to decode a compressed image into. This
- * function may be called more than once for every call to vpx_codec_decode.
- * The application may set fb->priv to some data which will be passed
- * back in the ximage and the release function call. |fb| is guaranteed to
- * not be NULL. On success the callback must return 0. Any failure the
- * callback must return a value less than 0.
+ * allocate at least min_size in bytes and assign it to fb->data. The callback
+ * must zero out all the data allocated. Then the callback must set fb->size
+ * to the allocated size. The application does not need to align the allocated
+ * data. The callback is triggered when the decoder needs a frame buffer to
+ * decode a compressed image into. This function may be called more than once
+ * for every call to vpx_codec_decode. The application may set fb->priv to
+ * some data which will be passed back in the ximage and the release function
+ * call. |fb| is guaranteed to not be NULL. On success the callback must
+ * return 0. Any failure the callback must return a value less than 0.
  *
  * \param[in] priv         Callback's private data
  * \param[in] new_size     Size in bytes needed by the buffer
diff --git a/vpx/vpx_image.h b/vpx/vpx_image.h
index 0b7bb90..ef6d1dd 100644
--- a/vpx/vpx_image.h
+++ b/vpx/vpx_image.h
@@ -64,40 +64,6 @@
     VPX_IMG_FMT_I44416    = VPX_IMG_FMT_I444 | VPX_IMG_FMT_HIGHBITDEPTH
   } vpx_img_fmt_t; /**< alias for enum vpx_img_fmt */
 
-#if !defined(VPX_CODEC_DISABLE_COMPAT) || !VPX_CODEC_DISABLE_COMPAT
-#define IMG_FMT_PLANAR         VPX_IMG_FMT_PLANAR     /**< \deprecated Use #VPX_IMG_FMT_PLANAR */
-#define IMG_FMT_UV_FLIP        VPX_IMG_FMT_UV_FLIP    /**< \deprecated Use #VPX_IMG_FMT_UV_FLIP */
-#define IMG_FMT_HAS_ALPHA      VPX_IMG_FMT_HAS_ALPHA  /**< \deprecated Use #VPX_IMG_FMT_HAS_ALPHA */
-
-  /*!\brief Deprecated list of supported image formats
-   * \deprecated New code should use #vpx_img_fmt
-   */
-#define img_fmt   vpx_img_fmt
-  /*!\brief alias for enum img_fmt.
-   * \deprecated New code should use #vpx_img_fmt_t
-   */
-#define img_fmt_t vpx_img_fmt_t
-
-#define IMG_FMT_NONE       VPX_IMG_FMT_NONE       /**< \deprecated Use #VPX_IMG_FMT_NONE */
-#define IMG_FMT_RGB24      VPX_IMG_FMT_RGB24      /**< \deprecated Use #VPX_IMG_FMT_RGB24 */
-#define IMG_FMT_RGB32      VPX_IMG_FMT_RGB32      /**< \deprecated Use #VPX_IMG_FMT_RGB32 */
-#define IMG_FMT_RGB565     VPX_IMG_FMT_RGB565     /**< \deprecated Use #VPX_IMG_FMT_RGB565 */
-#define IMG_FMT_RGB555     VPX_IMG_FMT_RGB555     /**< \deprecated Use #VPX_IMG_FMT_RGB555 */
-#define IMG_FMT_UYVY       VPX_IMG_FMT_UYVY       /**< \deprecated Use #VPX_IMG_FMT_UYVY */
-#define IMG_FMT_YUY2       VPX_IMG_FMT_YUY2       /**< \deprecated Use #VPX_IMG_FMT_YUY2 */
-#define IMG_FMT_YVYU       VPX_IMG_FMT_YVYU       /**< \deprecated Use #VPX_IMG_FMT_YVYU */
-#define IMG_FMT_BGR24      VPX_IMG_FMT_BGR24      /**< \deprecated Use #VPX_IMG_FMT_BGR24 */
-#define IMG_FMT_RGB32_LE   VPX_IMG_FMT_RGB32_LE   /**< \deprecated Use #VPX_IMG_FMT_RGB32_LE */
-#define IMG_FMT_ARGB       VPX_IMG_FMT_ARGB       /**< \deprecated Use #VPX_IMG_FMT_ARGB */
-#define IMG_FMT_ARGB_LE    VPX_IMG_FMT_ARGB_LE    /**< \deprecated Use #VPX_IMG_FMT_ARGB_LE */
-#define IMG_FMT_RGB565_LE  VPX_IMG_FMT_RGB565_LE  /**< \deprecated Use #VPX_IMG_FMT_RGB565_LE */
-#define IMG_FMT_RGB555_LE  VPX_IMG_FMT_RGB555_LE  /**< \deprecated Use #VPX_IMG_FMT_RGB555_LE */
-#define IMG_FMT_YV12       VPX_IMG_FMT_YV12       /**< \deprecated Use #VPX_IMG_FMT_YV12 */
-#define IMG_FMT_I420       VPX_IMG_FMT_I420       /**< \deprecated Use #VPX_IMG_FMT_I420 */
-#define IMG_FMT_VPXYV12    VPX_IMG_FMT_VPXYV12    /**< \deprecated Use #VPX_IMG_FMT_VPXYV12 */
-#define IMG_FMT_VPXI420    VPX_IMG_FMT_VPXI420    /**< \deprecated Use #VPX_IMG_FMT_VPXI420 */
-#endif /* VPX_CODEC_DISABLE_COMPAT */
-
   /**\brief Image Descriptor */
   typedef struct vpx_image {
     vpx_img_fmt_t fmt; /**< Image Format */
@@ -121,13 +87,6 @@
 #define VPX_PLANE_U      1   /**< U (Chroma) plane */
 #define VPX_PLANE_V      2   /**< V (Chroma) plane */
 #define VPX_PLANE_ALPHA  3   /**< A (Transparency) plane */
-#if !defined(VPX_CODEC_DISABLE_COMPAT) || !VPX_CODEC_DISABLE_COMPAT
-#define PLANE_PACKED     VPX_PLANE_PACKED
-#define PLANE_Y          VPX_PLANE_Y
-#define PLANE_U          VPX_PLANE_U
-#define PLANE_V          VPX_PLANE_V
-#define PLANE_ALPHA      VPX_PLANE_ALPHA
-#endif
     unsigned char *planes[4];  /**< pointer to the top left pixel for each plane */
     int      stride[4];  /**< stride between rows for each plane */
 
diff --git a/vpx_ports/arm_cpudetect.c b/vpx_ports/arm_cpudetect.c
index fa0e030..f03feff 100644
--- a/vpx_ports/arm_cpudetect.c
+++ b/vpx_ports/arm_cpudetect.c
@@ -10,7 +10,8 @@
 
 #include <stdlib.h>
 #include <string.h>
-#include "arm.h"
+#include "vpx_ports/arm.h"
+#include "./vpx_config.h"
 
 #ifdef WINAPI_FAMILY
 #include <winapifamily.h>
@@ -54,9 +55,9 @@
 #if HAVE_MEDIA
   flags |= HAS_MEDIA;
 #endif /* HAVE_MEDIA */
-#if HAVE_NEON
+#if HAVE_NEON || HAVE_NEON_ASM
   flags |= HAS_NEON;
-#endif /* HAVE_NEON */
+#endif /* HAVE_NEON  || HAVE_NEON_ASM */
   return flags & mask;
 }
 
@@ -87,6 +88,7 @@
       /*Ignore exception.*/
     }
   }
+#endif /* HAVE_EDSP */
 #if HAVE_MEDIA
   if (mask & HAS_MEDIA)
     __try {
@@ -97,7 +99,8 @@
     /*Ignore exception.*/
   }
 }
-#if HAVE_NEON
+#endif /* HAVE_MEDIA */
+#if HAVE_NEON || HAVE_NEON_ASM
 if (mask &HAS_NEON) {
   __try {
     /*VORR q0,q0,q0*/
@@ -107,9 +110,7 @@
     /*Ignore exception.*/
   }
 }
-#endif /* HAVE_NEON */
-#endif /* HAVE_MEDIA */
-#endif /* HAVE_EDSP */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
 return flags & mask;
 }
 
@@ -132,10 +133,10 @@
 #if HAVE_MEDIA
   flags |= HAS_MEDIA;
 #endif /* HAVE_MEDIA */
-#if HAVE_NEON
+#if HAVE_NEON || HAVE_NEON_ASM
   if (features & ANDROID_CPU_ARM_FEATURE_NEON)
     flags |= HAS_NEON;
-#endif /* HAVE_NEON */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
   return flags & mask;
 }
 
@@ -162,7 +163,7 @@
      */
     char buf[512];
     while (fgets(buf, 511, fin) != NULL) {
-#if HAVE_EDSP || HAVE_NEON
+#if HAVE_EDSP || HAVE_NEON || HAVE_NEON_ASM
       if (memcmp(buf, "Features", 8) == 0) {
         char *p;
 #if HAVE_EDSP
@@ -170,15 +171,15 @@
         if (p != NULL && (p[5] == ' ' || p[5] == '\n')) {
           flags |= HAS_EDSP;
         }
-#if HAVE_NEON
+#endif /* HAVE_EDSP */
+#if HAVE_NEON || HAVE_NEON_ASM
         p = strstr(buf, " neon");
         if (p != NULL && (p[5] == ' ' || p[5] == '\n')) {
           flags |= HAS_NEON;
         }
-#endif /* HAVE_NEON */
-#endif /* HAVE_EDSP */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
       }
-#endif /* HAVE_EDSP || HAVE_NEON */
+#endif /* HAVE_EDSP || HAVE_NEON || HAVE_NEON_ASM */
 #if HAVE_MEDIA
       if (memcmp(buf, "CPU architecture:", 17) == 0) {
         int version;
diff --git a/vpx_ports/mem.h b/vpx_ports/mem.h
index e91d776..1cb8c8c 100644
--- a/vpx_ports/mem.h
+++ b/vpx_ports/mem.h
@@ -23,7 +23,6 @@
 #warning No alignment directives known for this compiler.
 #define DECLARE_ALIGNED(n,typ,val)  typ val
 #endif
-#endif
 
 
 /* Declare an aligned array on the stack, for situations where the stack
@@ -44,4 +43,10 @@
 #define UNINITIALIZED_IS_SAFE(x) x=x
 #else
 #define UNINITIALIZED_IS_SAFE(x) x
+#endif
+
+#if HAVE_NEON && defined(_MSC_VER)
+#define __builtin_prefetch(x)
+#endif
+
 #endif  // VPX_PORTS_MEM_H_
diff --git a/vpx_scale/generic/yv12config.c b/vpx_scale/generic/yv12config.c
index 70d7ac0..475d231 100644
--- a/vpx_scale/generic/yv12config.c
+++ b/vpx_scale/generic/yv12config.c
@@ -199,11 +199,6 @@
       if (fb->data == NULL || fb->size < external_frame_size)
         return -1;
 
-      // This memset is needed for fixing valgrind error from C loop filter
-      // due to access uninitialized memory in frame border. It could be
-      // removed if border is totally removed.
-      vpx_memset(fb->data, 0, fb->size);
-
       ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32);
     } else if (frame_size > (size_t)ybf->buffer_alloc_sz) {
       // Allocation to hold larger frame, or first allocation.
diff --git a/vpx_scale/yv12config.h b/vpx_scale/yv12config.h
index eb0a8d6..9ff764c 100644
--- a/vpx_scale/yv12config.h
+++ b/vpx_scale/yv12config.h
@@ -15,6 +15,7 @@
 extern "C" {
 #endif
 
+#include "vpx/vpx_codec.h"
 #include "vpx/vpx_frame_buffer.h"
 #include "vpx/vpx_integer.h"
 
@@ -50,6 +51,7 @@
   int buffer_alloc_sz;
   int border;
   int frame_size;
+  unsigned int bit_depth;
 
   int corrupted;
   int flags;
diff --git a/vpxdec.c b/vpxdec.c
index 6470081..cf23c29 100644
--- a/vpxdec.c
+++ b/vpxdec.c
@@ -24,7 +24,6 @@
 #include "./args.h"
 #include "./ivfdec.h"
 
-#define VPX_CODEC_DISABLE_COMPAT 1
 #include "vpx/vpx_decoder.h"
 #include "vpx_ports/mem_ops.h"
 #include "vpx_ports/vpx_timer.h"
@@ -385,7 +384,7 @@
 
   if (ext_fb_list->ext_fb[i].size < min_size) {
     free(ext_fb_list->ext_fb[i].data);
-    ext_fb_list->ext_fb[i].data = (uint8_t *)malloc(min_size);
+    ext_fb_list->ext_fb[i].data = (uint8_t *)calloc(min_size, sizeof(uint8_t));
     if (!ext_fb_list->ext_fb[i].data)
       return -1;
 
diff --git a/vpxenc.c b/vpxenc.c
index 5afca24..1b0b632 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -415,10 +415,11 @@
     NULL, "tune-content", 1, "Tune content type", tune_content_enum);
 
 static const arg_def_t *vp9_args[] = {
-  &cpu_used, &auto_altref, &noise_sens, &sharpness, &static_thresh,
+  &cpu_used, &auto_altref, &sharpness, &static_thresh,
   &tile_cols, &tile_rows, &arnr_maxframes, &arnr_strength, &arnr_type,
   &tune_ssim, &cq_level, &max_intra_rate_pct, &lossless,
-  &frame_parallel_decoding, &aq_mode, &frame_periodic_boost, &tune_content,
+  &frame_parallel_decoding, &aq_mode, &frame_periodic_boost,
+  &noise_sens, &tune_content,
 #if CONFIG_VP9 && CONFIG_VP9_HIGHBITDEPTH
   &bitdeptharg, &inbitdeptharg,
 #endif
@@ -426,12 +427,13 @@
 };
 static const int vp9_arg_ctrl_map[] = {
   VP8E_SET_CPUUSED, VP8E_SET_ENABLEAUTOALTREF,
-  VP8E_SET_NOISE_SENSITIVITY, VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
+  VP8E_SET_SHARPNESS, VP8E_SET_STATIC_THRESHOLD,
   VP9E_SET_TILE_COLUMNS, VP9E_SET_TILE_ROWS,
   VP8E_SET_ARNR_MAXFRAMES, VP8E_SET_ARNR_STRENGTH, VP8E_SET_ARNR_TYPE,
   VP8E_SET_TUNING, VP8E_SET_CQ_LEVEL, VP8E_SET_MAX_INTRA_BITRATE_PCT,
   VP9E_SET_LOSSLESS, VP9E_SET_FRAME_PARALLEL_DECODING, VP9E_SET_AQ_MODE,
-  VP9E_SET_FRAME_PERIODIC_BOOST, VP9E_SET_TUNE_CONTENT,
+  VP9E_SET_FRAME_PERIODIC_BOOST, VP9E_SET_NOISE_SENSITIVITY,
+  VP9E_SET_TUNE_CONTENT,
   0
 };
 #endif
diff --git a/y4minput.c b/y4minput.c
index bcc742a..34ea96d 100644
--- a/y4minput.c
+++ b/y4minput.c
@@ -1041,12 +1041,12 @@
   c_w *= bytes_per_sample;
   c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
   c_sz = c_w * c_h;
-  _img->stride[PLANE_Y] = _img->stride[PLANE_ALPHA] =
+  _img->stride[VPX_PLANE_Y] = _img->stride[VPX_PLANE_ALPHA] =
       _y4m->pic_w * bytes_per_sample;
-  _img->stride[PLANE_U] = _img->stride[PLANE_V] = c_w;
-  _img->planes[PLANE_Y] = _y4m->dst_buf;
-  _img->planes[PLANE_U] = _y4m->dst_buf + pic_sz;
-  _img->planes[PLANE_V] = _y4m->dst_buf + pic_sz + c_sz;
-  _img->planes[PLANE_ALPHA] = _y4m->dst_buf + pic_sz + 2 * c_sz;
+  _img->stride[VPX_PLANE_U] = _img->stride[VPX_PLANE_V] = c_w;
+  _img->planes[VPX_PLANE_Y] = _y4m->dst_buf;
+  _img->planes[VPX_PLANE_U] = _y4m->dst_buf + pic_sz;
+  _img->planes[VPX_PLANE_V] = _y4m->dst_buf + pic_sz + c_sz;
+  _img->planes[VPX_PLANE_ALPHA] = _y4m->dst_buf + pic_sz + 2 * c_sz;
   return 1;
 }