Merge "Prunes out full-rd computation based on modeled rd"
diff --git a/build/make/gen_msvs_proj.sh b/build/make/gen_msvs_proj.sh
index cff27c8..fc5011b 100755
--- a/build/make/gen_msvs_proj.sh
+++ b/build/make/gen_msvs_proj.sh
@@ -381,7 +381,7 @@
RuntimeLibrary="$debug_runtime" \
UsePrecompiledHeader="0" \
WarningLevel="3" \
- DebugInformationFormat="1" \
+ DebugInformationFormat="2" \
$warn_64bit \
$uses_asm && tag Tool Name="YASM" IncludePaths="$incs" Debug="true"
@@ -395,7 +395,7 @@
RuntimeLibrary="$debug_runtime" \
UsePrecompiledHeader="0" \
WarningLevel="3" \
- DebugInformationFormat="1" \
+ DebugInformationFormat="2" \
$warn_64bit \
$uses_asm && tag Tool Name="YASM" IncludePaths="$incs" Debug="true"
diff --git a/configure b/configure
index 28676fb..36fc08f 100755
--- a/configure
+++ b/configure
@@ -682,6 +682,14 @@
# iOS/ARM builds do not work with gtest. This does not match
# x86 targets.
;;
+ *-win*)
+ # Some mingw toolchains don't have pthread available by default.
+ # Treat these more like visual studio where threading in gtest
+ # would be disabled for the same reason.
+ check_cxx "$@" <<EOF && soft_enable unit_tests
+int z;
+EOF
+ ;;
*)
enabled pthread_h && check_cxx "$@" <<EOF && soft_enable unit_tests
int z;
diff --git a/libs.mk b/libs.mk
index f7d8395..4aa7dc4 100644
--- a/libs.mk
+++ b/libs.mk
@@ -448,6 +448,10 @@
include $(SRC_PATH_BARE)/third_party/googletest/gtest.mk
GTEST_SRCS := $(addprefix third_party/googletest/src/,$(call enabled,GTEST_SRCS))
GTEST_OBJS=$(call objs,$(GTEST_SRCS))
+ifeq ($(filter win%,$(TGT_OS)),$(TGT_OS))
+# Disabling pthreads globally will cause issues on darwin and possibly elsewhere
+$(GTEST_OBJS) $(GTEST_OBJS:.o=.d): CXXFLAGS += -DGTEST_HAS_PTHREAD=0
+endif
$(GTEST_OBJS) $(GTEST_OBJS:.o=.d): CXXFLAGS += -I$(SRC_PATH_BARE)/third_party/googletest/src
$(GTEST_OBJS) $(GTEST_OBJS:.o=.d): CXXFLAGS += -I$(SRC_PATH_BARE)/third_party/googletest/src/include
OBJS-$(BUILD_LIBVPX) += $(GTEST_OBJS)
@@ -472,7 +476,7 @@
lib$(CODEC_LIB)$(CODEC_LIB_SUF) libgtest.a ))\
$(if $(BUILD_LIBVPX),$(eval $(call linkerxx_template,$(bin),\
$(LIBVPX_TEST_OBJS) \
- -L. -lvpx -lgtest -lpthread -lm)\
+ -L. -lvpx -lgtest $(extralibs) -lm)\
)))\
$(if $(LIPO_LIBS),$(eval $(call lipo_bin_template,$(bin))))\
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 439d98c..385dcc1 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -391,10 +391,9 @@
static BLOCK_SIZE_TYPE get_subsize(BLOCK_SIZE_TYPE bsize,
PARTITION_TYPE partition) {
- BLOCK_SIZE_TYPE subsize;
+ BLOCK_SIZE_TYPE subsize = bsize;
switch (partition) {
case PARTITION_NONE:
- subsize = bsize;
break;
case PARTITION_HORZ:
if (bsize == BLOCK_SIZE_SB64X64)
diff --git a/vp9/common/vp9_entropymv.c b/vp9/common/vp9_entropymv.c
index a62fa47..50e8463 100644
--- a/vp9/common/vp9_entropymv.c
+++ b/vp9/common/vp9_entropymv.c
@@ -85,7 +85,7 @@
#define mv_class_base(c) ((c) ? (CLASS0_SIZE << (c + 2)) : 0)
MV_CLASS_TYPE vp9_get_mv_class(int z, int *offset) {
- MV_CLASS_TYPE c;
+ MV_CLASS_TYPE c = MV_CLASS_0;
if (z < CLASS0_SIZE * 8) c = MV_CLASS_0;
else if (z < CLASS0_SIZE * 16) c = MV_CLASS_1;
else if (z < CLASS0_SIZE * 32) c = MV_CLASS_2;
diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index f331a8f..265a19a 100644
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -365,17 +365,16 @@
MACROBLOCKD * const xd = arg->xd;
const int bwl = b_width_log2(bsize) - xd->plane[plane].subsampling_x;
const int bhl = b_height_log2(bsize) - xd->plane[plane].subsampling_y;
- const int bh = 4 << bhl, bw = 4 << bwl;
const int x = 4 * (block & ((1 << bwl) - 1)), y = 4 * (block >> bwl);
const int use_second_ref = xd->mode_info_context->mbmi.ref_frame[1] > 0;
int which_mv;
- assert(x < bw);
- assert(y < bh);
+ assert(x < (4 << bwl));
+ assert(y < (4 << bhl));
assert(xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8 ||
- 4 << pred_w == bw);
+ 4 << pred_w == (4 << bwl));
assert(xd->mode_info_context->mbmi.sb_type < BLOCK_SIZE_SB8X8 ||
- 4 << pred_h == bh);
+ 4 << pred_h == (4 << bhl));
for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
// source
diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh
index 313cfd6..d861a7a 100644
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -22,6 +22,8 @@
}
forward_decls vp9_common_forward_decls
+[ $arch = "x86_64" ] && mmx_x86_64=mmx && sse2_x86_64=sse2
+
#
# Dequant
#
@@ -62,7 +64,7 @@
specialize vp9_d63_predictor_4x4
prototype void vp9_h_predictor_4x4 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_h_predictor_4x4
+specialize vp9_h_predictor_4x4 ssse3
prototype void vp9_d117_predictor_4x4 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_d117_predictor_4x4
@@ -77,7 +79,7 @@
specialize vp9_v_predictor_4x4 sse
prototype void vp9_tm_predictor_4x4 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_tm_predictor_4x4
+specialize vp9_tm_predictor_4x4 sse
prototype void vp9_dc_predictor_4x4 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_dc_predictor_4x4 sse
@@ -101,7 +103,7 @@
specialize vp9_d63_predictor_8x8
prototype void vp9_h_predictor_8x8 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_h_predictor_8x8
+specialize vp9_h_predictor_8x8 ssse3
prototype void vp9_d117_predictor_8x8 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_d117_predictor_8x8
@@ -116,7 +118,7 @@
specialize vp9_v_predictor_8x8 sse
prototype void vp9_tm_predictor_8x8 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_tm_predictor_8x8
+specialize vp9_tm_predictor_8x8 sse2
prototype void vp9_dc_predictor_8x8 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_dc_predictor_8x8 sse
@@ -140,7 +142,7 @@
specialize vp9_d63_predictor_16x16
prototype void vp9_h_predictor_16x16 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_h_predictor_16x16
+specialize vp9_h_predictor_16x16 ssse3
prototype void vp9_d117_predictor_16x16 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_d117_predictor_16x16
@@ -155,7 +157,7 @@
specialize vp9_v_predictor_16x16 sse2
prototype void vp9_tm_predictor_16x16 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_tm_predictor_16x16
+specialize vp9_tm_predictor_16x16 sse2
prototype void vp9_dc_predictor_16x16 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_dc_predictor_16x16 sse2
@@ -179,7 +181,7 @@
specialize vp9_d63_predictor_32x32
prototype void vp9_h_predictor_32x32 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_h_predictor_32x32
+specialize vp9_h_predictor_32x32 ssse3
prototype void vp9_d117_predictor_32x32 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_d117_predictor_32x32
@@ -194,7 +196,7 @@
specialize vp9_v_predictor_32x32 sse2
prototype void vp9_tm_predictor_32x32 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
-specialize vp9_tm_predictor_32x32
+specialize vp9_tm_predictor_32x32 sse2_x86_64
prototype void vp9_dc_predictor_32x32 "uint8_t *ypred_ptr, ptrdiff_t y_stride, uint8_t *yabove_row, uint8_t *yleft_col"
specialize vp9_dc_predictor_32x32 sse2
@@ -362,8 +364,6 @@
# variance
-[ $arch = "x86_64" ] && mmx_x86_64=mmx && sse2_x86_64=sse2
-
prototype unsigned int vp9_variance32x16 "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse"
specialize vp9_variance32x16 sse2
@@ -744,7 +744,7 @@
specialize vp9_short_fht8x8 sse2
prototype void vp9_short_fht16x16 "int16_t *InputData, int16_t *OutputData, int pitch, int tx_type"
-specialize vp9_short_fht16x16
+specialize vp9_short_fht16x16 sse2
prototype void vp9_short_fdct8x8 "int16_t *InputData, int16_t *OutputData, int pitch"
specialize vp9_short_fdct8x8 sse2
diff --git a/vp9/common/x86/vp9_intrapred_sse2.asm b/vp9/common/x86/vp9_intrapred_sse2.asm
index 5aa3dba..980b8b9 100644
--- a/vp9/common/x86/vp9_intrapred_sse2.asm
+++ b/vp9/common/x86/vp9_intrapred_sse2.asm
@@ -188,3 +188,154 @@
dec nlines4d
jnz .loop
REP_RET
+
+INIT_MMX sse
+cglobal tm_predictor_4x4, 4, 4, 4, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ movd m0, [aboveq]
+ punpcklbw m2, m1
+ punpcklbw m0, m1
+ pshufw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -2
+ add leftq, 4
+ psubw m0, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m3, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m3, m1
+ pshufw m2, m2, 0x0
+ pshufw m3, m3, 0x0
+ paddw m2, m0
+ paddw m3, m0
+ packuswb m2, m2
+ packuswb m3, m3
+ movd [dstq ], m2
+ movd [dstq+strideq], m3
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal tm_predictor_8x8, 4, 4, 4, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ movq m0, [aboveq]
+ punpcklbw m2, m1
+ punpcklbw m0, m1
+ pshuflw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -4
+ punpcklqdq m2, m2
+ add leftq, 8
+ psubw m0, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m3, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m3, m1
+ pshuflw m2, m2, 0x0
+ pshuflw m3, m3, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m3, m3
+ paddw m2, m0
+ paddw m3, m0
+ packuswb m2, m3
+ movq [dstq ], m2
+ movhps [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal tm_predictor_16x16, 4, 4, 7, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ mova m0, [aboveq]
+ punpcklbw m2, m1
+ punpckhbw m4, m0, m1
+ punpcklbw m0, m1
+ pshuflw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -8
+ punpcklqdq m2, m2
+ add leftq, 16
+ psubw m0, m2
+ psubw m4, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m3, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m3, m1
+ pshuflw m2, m2, 0x0
+ pshuflw m3, m3, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m3, m3
+ paddw m5, m2, m0
+ paddw m6, m3, m0
+ paddw m2, m4
+ paddw m3, m4
+ packuswb m5, m2
+ packuswb m6, m3
+ mova [dstq ], m5
+ mova [dstq+strideq], m6
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal tm_predictor_32x32, 4, 4, 10, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ mova m0, [aboveq]
+ mova m4, [aboveq+16]
+ punpcklbw m2, m1
+ punpckhbw m3, m0, m1
+ punpckhbw m5, m4, m1
+ punpcklbw m0, m1
+ punpcklbw m4, m1
+ pshuflw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -16
+ punpcklqdq m2, m2
+ add leftq, 32
+ psubw m0, m2
+ psubw m3, m2
+ psubw m4, m2
+ psubw m5, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m6, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m6, m1
+ pshuflw m2, m2, 0x0
+ pshuflw m6, m6, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m6, m6
+ paddw m7, m2, m0
+ paddw m8, m2, m3
+ paddw m9, m2, m4
+ paddw m2, m5
+ packuswb m7, m8
+ packuswb m9, m2
+ paddw m2, m6, m0
+ paddw m8, m6, m3
+ mova [dstq ], m7
+ paddw m7, m6, m4
+ paddw m6, m5
+ mova [dstq +16], m9
+ packuswb m2, m8
+ packuswb m7, m6
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq+16], m7
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+%endif
diff --git a/vp9/common/x86/vp9_intrapred_ssse3.asm b/vp9/common/x86/vp9_intrapred_ssse3.asm
new file mode 100644
index 0000000..bc8ed5c
--- /dev/null
+++ b/vp9/common/x86/vp9_intrapred_ssse3.asm
@@ -0,0 +1,87 @@
+;
+; 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.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION .text
+
+INIT_MMX ssse3
+cglobal h_predictor_4x4, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 4
+ mov lineq, -2
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ movd [dstq ], m1
+ movd [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_MMX ssse3
+cglobal h_predictor_8x8, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 8
+ mov lineq, -4
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ movq [dstq ], m1
+ movq [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM ssse3
+cglobal h_predictor_16x16, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 16
+ mov lineq, -8
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ mova [dstq ], m1
+ mova [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM ssse3
+cglobal h_predictor_32x32, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 32
+ mov lineq, -16
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ mova [dstq ], m1
+ mova [dstq +16], m1
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq+16], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
diff --git a/vp9/decoder/vp9_decodframe.c b/vp9/decoder/vp9_decodframe.c
index 31cf69a..37bdad2 100644
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -55,32 +55,30 @@
return data > max ? max : data;
}
-static void setup_txfm_mode(VP9_COMMON *pc, int lossless, vp9_reader *r) {
- if (lossless) {
- pc->txfm_mode = ONLY_4X4;
- } else {
- pc->txfm_mode = vp9_read_literal(r, 2);
- if (pc->txfm_mode == ALLOW_32X32)
- pc->txfm_mode += vp9_read_bit(r);
+static TXFM_MODE read_tx_mode(vp9_reader *r) {
+ TXFM_MODE txfm_mode = vp9_read_literal(r, 2);
+ if (txfm_mode == ALLOW_32X32)
+ txfm_mode += vp9_read_bit(r);
+ return txfm_mode;
+}
- if (pc->txfm_mode == TX_MODE_SELECT) {
- int i, j;
- for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
- for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j)
- if (vp9_read(r, VP9_MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &pc->fc.tx_probs_8x8p[i][j]);
+static void read_tx_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
+ int i, j;
- for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
- for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j)
- if (vp9_read(r, VP9_MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &pc->fc.tx_probs_16x16p[i][j]);
+ for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
+ for (j = 0; j < TX_SIZE_MAX_SB - 3; ++j)
+ if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+ vp9_diff_update_prob(r, &fc->tx_probs_8x8p[i][j]);
- for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
- for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j)
- if (vp9_read(r, VP9_MODE_UPDATE_PROB))
- vp9_diff_update_prob(r, &pc->fc.tx_probs_32x32p[i][j]);
- }
- }
+ for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
+ for (j = 0; j < TX_SIZE_MAX_SB - 2; ++j)
+ if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+ vp9_diff_update_prob(r, &fc->tx_probs_16x16p[i][j]);
+
+ for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
+ for (j = 0; j < TX_SIZE_MAX_SB - 1; ++j)
+ if (vp9_read(r, VP9_MODE_UPDATE_PROB))
+ vp9_diff_update_prob(r, &fc->tx_probs_32x32p[i][j]);
}
static void mb_init_dequantizer(VP9_COMMON *pc, MACROBLOCKD *xd) {
@@ -370,33 +368,21 @@
static void read_coef_probs_common(FRAME_CONTEXT *fc, TX_SIZE tx_size,
vp9_reader *r) {
vp9_coeff_probs_model *coef_probs = fc->coef_probs[tx_size];
+ int i, j, k, l, m;
- if (vp9_read_bit(r)) {
- int i, j, k, l, m;
- for (i = 0; i < BLOCK_TYPES; i++) {
- for (j = 0; j < REF_TYPES; j++) {
- for (k = 0; k < COEF_BANDS; k++) {
- for (l = 0; l < PREV_COEF_CONTEXTS; l++) {
- if (l >= 3 && k == 0)
- continue;
-
- for (m = 0; m < UNCONSTRAINED_NODES; m++) {
- vp9_prob *const p = coef_probs[i][j][k][l] + m;
-
- if (vp9_read(r, VP9_COEF_UPDATE_PROB))
- vp9_diff_update_prob(r, p);
- }
- }
- }
- }
- }
- }
+ if (vp9_read_bit(r))
+ for (i = 0; i < BLOCK_TYPES; i++)
+ for (j = 0; j < REF_TYPES; j++)
+ for (k = 0; k < COEF_BANDS; k++)
+ for (l = 0; l < PREV_COEF_CONTEXTS; l++)
+ if (k > 0 || l < 3)
+ for (m = 0; m < UNCONSTRAINED_NODES; m++)
+ if (vp9_read(r, VP9_COEF_UPDATE_PROB))
+ vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]);
}
-static void read_coef_probs(VP9D_COMP *pbi, vp9_reader *r) {
- const TXFM_MODE txfm_mode = pbi->common.txfm_mode;
- FRAME_CONTEXT *const fc = &pbi->common.fc;
-
+static void read_coef_probs(FRAME_CONTEXT *fc, TXFM_MODE txfm_mode,
+ vp9_reader *r) {
read_coef_probs_common(fc, TX_4X4, r);
if (txfm_mode > ONLY_4X4)
@@ -923,6 +909,26 @@
return vp9_rb_read_literal(rb, 16);
}
+static int read_compressed_header(VP9D_COMP *pbi, const uint8_t *data,
+ size_t partition_size) {
+ VP9_COMMON *const cm = &pbi->common;
+ MACROBLOCKD *const xd = &pbi->mb;
+ vp9_reader r;
+
+ if (vp9_reader_init(&r, data, partition_size))
+ vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
+ "Failed to allocate bool decoder 0");
+
+ cm->txfm_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r);
+ if (cm->txfm_mode == TX_MODE_SELECT)
+ read_tx_probs(&cm->fc, &r);
+ read_coef_probs(&cm->fc, cm->txfm_mode, &r);
+
+ vp9_prepare_read_mode_info(pbi, &r);
+
+ return vp9_reader_has_error(&r);
+}
+
void vp9_init_dequantizer(VP9_COMMON *pc) {
int q;
@@ -939,7 +945,7 @@
int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
int i;
- vp9_reader header_bc, residual_bc;
+ vp9_reader residual_bc;
VP9_COMMON *const pc = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
@@ -973,10 +979,6 @@
xd->frame_type = pc->frame_type;
xd->mode_info_stride = pc->mode_info_stride;
- if (vp9_reader_init(&header_bc, data, first_partition_size))
- vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
- "Failed to allocate bool decoder 0");
-
mb_init_dequantizer(pc, &pbi->mb); // MB level dequantizer setup
if (!keyframe)
@@ -986,15 +988,13 @@
update_frame_context(&pc->fc);
- setup_txfm_mode(pc, xd->lossless, &header_bc);
-
- read_coef_probs(pbi, &header_bc);
-
// Initialize xd pointers. Any reference should do for xd->pre, so use 0.
setup_pre_planes(xd, 0, &pc->yv12_fb[pc->active_ref_idx[0]], 0, 0,
NULL, NULL);
setup_dst_planes(xd, new_fb, 0, 0);
+ new_fb->corrupted |= read_compressed_header(pbi, data, first_partition_size);
+
// Create the segmentation map structure and set to 0
if (!pc->last_frame_seg_map)
CHECK_MEM_ERROR(pc, pc->last_frame_seg_map,
@@ -1008,14 +1008,12 @@
set_prev_mi(pc);
- vp9_prepare_read_mode_info(pbi, &header_bc);
-
decode_tiles(pbi, data, first_partition_size, &residual_bc);
pc->last_width = pc->width;
pc->last_height = pc->height;
- new_fb->corrupted = vp9_reader_has_error(&header_bc) | xd->corrupted;
+ new_fb->corrupted |= xd->corrupted;
if (!pbi->decoded_key_frame) {
if (keyframe && !new_fb->corrupted)
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index ef29525..3a2990c 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -629,7 +629,7 @@
int bsl = b_width_log2(bsize);
int bs = (1 << bsl) / 4; // mode_info step for subsize
int n;
- PARTITION_TYPE partition;
+ PARTITION_TYPE partition = PARTITION_NONE;
BLOCK_SIZE_TYPE subsize;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index ce81e05..adaf667 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -324,9 +324,8 @@
int mb_mode_index = ctx->best_mode_index;
const int mis = cpi->common.mode_info_stride;
const int bh = 1 << mi_height_log2(bsize), bw = 1 << mi_width_log2(bsize);
- const MB_PREDICTION_MODE mb_mode = mi->mbmi.mode;
- assert(mb_mode < MB_MODE_COUNT);
+ assert(mi->mbmi.mode < MB_MODE_COUNT);
assert(mb_mode_index < MAX_MODES);
assert(mi->mbmi.ref_frame[0] < MAX_REF_FRAMES);
assert(mi->mbmi.ref_frame[1] < MAX_REF_FRAMES);
@@ -390,7 +389,7 @@
THR_TM /*TM_PRED*/,
THR_B_PRED /*I4X4_PRED*/,
};
- cpi->mode_chosen_counts[kf_mode_index[mb_mode]]++;
+ cpi->mode_chosen_counts[kf_mode_index[mi->mbmi.mode]]++;
#endif
} else {
// Note how often each mode chosen as best
@@ -1170,7 +1169,7 @@
int mh = bh / 2;
int bss = (1 << bsl) / 4;
int i, pl;
- PARTITION_TYPE partition;
+ PARTITION_TYPE partition = PARTITION_NONE;
BLOCK_SIZE_TYPE subsize;
ENTROPY_CONTEXT l[16 * MAX_MB_PLANE], a[16 * MAX_MB_PLANE];
PARTITION_CONTEXT sl[8], sa[8];
@@ -1421,6 +1420,8 @@
int srate = INT_MAX;
int64_t sdist = INT_MAX;
+ (void) *tp_orig;
+
if (bsize < BLOCK_SIZE_SB8X8)
if (xd->ab_index != 0) {
*rate = 0;
@@ -1928,13 +1929,13 @@
if (mbmi->txfm_size > txfm_max) {
MACROBLOCK * const x = &cpi->mb;
MACROBLOCKD * const xd = &x->e_mbd;
- const int segment_id = mbmi->segment_id;
const int ymbs = MIN(bh, cm->mi_rows - mi_row);
const int xmbs = MIN(bw, cm->mi_cols - mi_col);
xd->mode_info_context = mi;
assert(
- vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP) || get_skip_flag(mi, mis, ymbs, xmbs));
+ vp9_segfeature_active(xd, mbmi->segment_id, SEG_LVL_SKIP) ||
+ get_skip_flag(mi, mis, ymbs, xmbs));
set_txfm_flag(mi, mis, ymbs, xmbs, txfm_max);
}
}
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 859bca9..4cb38f7 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -503,8 +503,8 @@
MACROBLOCK *x, MACROBLOCKD *xd,
int *out_rate_sum, int64_t *out_dist_sum,
int *out_skip) {
- int t, j, k;
- enum BlockSize bs;
+ int t = 4, j, k;
+ enum BlockSize bs = BLOCK_4X4;
struct macroblock_plane *const p = &x->plane[0];
struct macroblockd_plane *const pd = &xd->plane[0];
const int bw = plane_block_width(bsize, pd);
@@ -576,26 +576,25 @@
int pt;
int c = 0;
int cost = 0;
- const int16_t *scan, *nb;
+ const int16_t *scan = NULL, *nb;
const int eob = xd->plane[plane].eobs[block];
const int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[plane].qcoeff, block, 16);
const int ref = mbmi->ref_frame[0] != INTRA_FRAME;
unsigned int (*token_costs)[COEF_BANDS][PREV_COEF_CONTEXTS]
[MAX_ENTROPY_TOKENS] = mb->token_costs[tx_size][type][ref];
- ENTROPY_CONTEXT above_ec, left_ec;
+ ENTROPY_CONTEXT above_ec = 0, left_ec = 0;
TX_TYPE tx_type = DCT_DCT;
const int segment_id = xd->mode_info_context->mbmi.segment_id;
- int seg_eob;
+ int seg_eob = 0;
uint8_t token_cache[1024];
- const uint8_t * band_translate;
+ const uint8_t *band_translate = NULL;
// Check for consistency of tx_size with mode info
assert((!type && !plane) || (type && plane));
if (type == PLANE_TYPE_Y_WITH_DC) {
assert(xd->mode_info_context->mbmi.txfm_size == tx_size);
} else {
- TX_SIZE tx_size_uv = get_uv_tx_size(mbmi);
- assert(tx_size == tx_size_uv);
+ assert(tx_size == get_uv_tx_size(mbmi));
}
switch (tx_size) {
diff --git a/vp9/encoder/x86/vp9_dct_sse2.c b/vp9/encoder/x86/vp9_dct_sse2.c
index 87a774c..bf09c7a 100644
--- a/vp9/encoder/x86/vp9_dct_sse2.c
+++ b/vp9/encoder/x86/vp9_dct_sse2.c
@@ -557,6 +557,8 @@
// right shift and rounding
static INLINE void right_shift_8x8(__m128i *res, int const bit) {
+ const __m128i kOne = _mm_set1_epi16(1);
+ const int bit_m02 = bit - 2;
__m128i sign0 = _mm_srai_epi16(res[0], 15);
__m128i sign1 = _mm_srai_epi16(res[1], 15);
__m128i sign2 = _mm_srai_epi16(res[2], 15);
@@ -566,6 +568,18 @@
__m128i sign6 = _mm_srai_epi16(res[6], 15);
__m128i sign7 = _mm_srai_epi16(res[7], 15);
+ if (bit_m02 >= 0) {
+ __m128i k_const_rounding = _mm_slli_epi16(kOne, bit_m02);
+ res[0] = _mm_add_epi16(res[0], k_const_rounding);
+ res[1] = _mm_add_epi16(res[1], k_const_rounding);
+ res[2] = _mm_add_epi16(res[2], k_const_rounding);
+ res[3] = _mm_add_epi16(res[3], k_const_rounding);
+ res[4] = _mm_add_epi16(res[4], k_const_rounding);
+ res[5] = _mm_add_epi16(res[5], k_const_rounding);
+ res[6] = _mm_add_epi16(res[6], k_const_rounding);
+ res[7] = _mm_add_epi16(res[7], k_const_rounding);
+ }
+
res[0] = _mm_sub_epi16(res[0], sign0);
res[1] = _mm_sub_epi16(res[1], sign1);
res[2] = _mm_sub_epi16(res[2], sign2);
@@ -587,8 +601,6 @@
// write 8x8 array
static INLINE void write_buffer_8x8(int16_t *output, __m128i *res, int stride) {
- right_shift_8x8(res, 1);
-
_mm_store_si128((__m128i *)(output + 0 * stride), res[0]);
_mm_store_si128((__m128i *)(output + 1 * stride), res[1]);
_mm_store_si128((__m128i *)(output + 2 * stride), res[2]);
@@ -1046,6 +1058,7 @@
assert(0);
break;
}
+ right_shift_8x8(in, 1);
write_buffer_8x8(output, in, 8);
}
@@ -1675,6 +1688,890 @@
}
}
+static INLINE void load_buffer_16x16(int16_t* input, __m128i *in0,
+ __m128i *in1, int stride) {
+ // load first 8 columns
+ load_buffer_8x8(input, in0, stride);
+ load_buffer_8x8(input + 8 * stride, in0 + 8, stride);
+
+ input += 8;
+ // load second 8 columns
+ load_buffer_8x8(input, in1, stride);
+ load_buffer_8x8(input + 8 * stride, in1 + 8, stride);
+}
+
+static INLINE void write_buffer_16x16(int16_t *output, __m128i *in0,
+ __m128i *in1, int stride) {
+ // write first 8 columns
+ write_buffer_8x8(output, in0, stride);
+ write_buffer_8x8(output + 8 * stride, in0 + 8, stride);
+ // write second 8 columns
+ output += 8;
+ write_buffer_8x8(output, in1, stride);
+ write_buffer_8x8(output + 8 * stride, in1 + 8, stride);
+}
+
+static INLINE void array_transpose_16x16(__m128i *res0, __m128i *res1) {
+ __m128i tbuf[8];
+ array_transpose_8x8(res0, res0);
+ array_transpose_8x8(res1, tbuf);
+ array_transpose_8x8(res0 + 8, res1);
+ array_transpose_8x8(res1 + 8, res1 + 8);
+
+ res0[8] = tbuf[0];
+ res0[9] = tbuf[1];
+ res0[10] = tbuf[2];
+ res0[11] = tbuf[3];
+ res0[12] = tbuf[4];
+ res0[13] = tbuf[5];
+ res0[14] = tbuf[6];
+ res0[15] = tbuf[7];
+}
+
+static INLINE void right_shift_16x16(__m128i *res0, __m128i *res1) {
+ // perform rounding operations
+ right_shift_8x8(res0, 2);
+ right_shift_8x8(res0 + 8, 2);
+ right_shift_8x8(res1, 2);
+ right_shift_8x8(res1 + 8, 2);
+}
+
+void fdct16_1d_8col(__m128i *in) {
+ // perform 16x16 1-D DCT for 8 columns
+ __m128i i[8], s[8], p[8], t[8], u[16], v[16];
+ const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
+ const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+ const __m128i k__cospi_m16_p16 = pair_set_epi16(-cospi_16_64, cospi_16_64);
+ const __m128i k__cospi_p24_p08 = pair_set_epi16(cospi_24_64, cospi_8_64);
+ const __m128i k__cospi_m24_m08 = pair_set_epi16(-cospi_24_64, -cospi_8_64);
+ const __m128i k__cospi_m08_p24 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+ const __m128i k__cospi_p28_p04 = pair_set_epi16(cospi_28_64, cospi_4_64);
+ const __m128i k__cospi_m04_p28 = pair_set_epi16(-cospi_4_64, cospi_28_64);
+ const __m128i k__cospi_p12_p20 = pair_set_epi16(cospi_12_64, cospi_20_64);
+ const __m128i k__cospi_m20_p12 = pair_set_epi16(-cospi_20_64, cospi_12_64);
+ const __m128i k__cospi_p30_p02 = pair_set_epi16(cospi_30_64, cospi_2_64);
+ const __m128i k__cospi_p14_p18 = pair_set_epi16(cospi_14_64, cospi_18_64);
+ const __m128i k__cospi_m02_p30 = pair_set_epi16(-cospi_2_64, cospi_30_64);
+ const __m128i k__cospi_m18_p14 = pair_set_epi16(-cospi_18_64, cospi_14_64);
+ const __m128i k__cospi_p22_p10 = pair_set_epi16(cospi_22_64, cospi_10_64);
+ const __m128i k__cospi_p06_p26 = pair_set_epi16(cospi_6_64, cospi_26_64);
+ const __m128i k__cospi_m10_p22 = pair_set_epi16(-cospi_10_64, cospi_22_64);
+ const __m128i k__cospi_m26_p06 = pair_set_epi16(-cospi_26_64, cospi_6_64);
+ const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
+
+ // stage 1
+ i[0] = _mm_add_epi16(in[0], in[15]);
+ i[1] = _mm_add_epi16(in[1], in[14]);
+ i[2] = _mm_add_epi16(in[2], in[13]);
+ i[3] = _mm_add_epi16(in[3], in[12]);
+ i[4] = _mm_add_epi16(in[4], in[11]);
+ i[5] = _mm_add_epi16(in[5], in[10]);
+ i[6] = _mm_add_epi16(in[6], in[9]);
+ i[7] = _mm_add_epi16(in[7], in[8]);
+
+ s[0] = _mm_sub_epi16(in[7], in[8]);
+ s[1] = _mm_sub_epi16(in[6], in[9]);
+ s[2] = _mm_sub_epi16(in[5], in[10]);
+ s[3] = _mm_sub_epi16(in[4], in[11]);
+ s[4] = _mm_sub_epi16(in[3], in[12]);
+ s[5] = _mm_sub_epi16(in[2], in[13]);
+ s[6] = _mm_sub_epi16(in[1], in[14]);
+ s[7] = _mm_sub_epi16(in[0], in[15]);
+
+ p[0] = _mm_add_epi16(i[0], i[7]);
+ p[1] = _mm_add_epi16(i[1], i[6]);
+ p[2] = _mm_add_epi16(i[2], i[5]);
+ p[3] = _mm_add_epi16(i[3], i[4]);
+ p[4] = _mm_sub_epi16(i[3], i[4]);
+ p[5] = _mm_sub_epi16(i[2], i[5]);
+ p[6] = _mm_sub_epi16(i[1], i[6]);
+ p[7] = _mm_sub_epi16(i[0], i[7]);
+
+ u[0] = _mm_add_epi16(p[0], p[3]);
+ u[1] = _mm_add_epi16(p[1], p[2]);
+ u[2] = _mm_sub_epi16(p[1], p[2]);
+ u[3] = _mm_sub_epi16(p[0], p[3]);
+
+ v[0] = _mm_unpacklo_epi16(u[0], u[1]);
+ v[1] = _mm_unpackhi_epi16(u[0], u[1]);
+ v[2] = _mm_unpacklo_epi16(u[2], u[3]);
+ v[3] = _mm_unpackhi_epi16(u[2], u[3]);
+
+ u[0] = _mm_madd_epi16(v[0], k__cospi_p16_p16);
+ u[1] = _mm_madd_epi16(v[1], k__cospi_p16_p16);
+ u[2] = _mm_madd_epi16(v[0], k__cospi_p16_m16);
+ u[3] = _mm_madd_epi16(v[1], k__cospi_p16_m16);
+ u[4] = _mm_madd_epi16(v[2], k__cospi_p24_p08);
+ u[5] = _mm_madd_epi16(v[3], k__cospi_p24_p08);
+ u[6] = _mm_madd_epi16(v[2], k__cospi_m08_p24);
+ u[7] = _mm_madd_epi16(v[3], k__cospi_m08_p24);
+
+ v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
+ v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
+ v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING);
+ v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING);
+ v[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING);
+ v[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING);
+ v[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING);
+ v[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING);
+
+ u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS);
+ u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS);
+ u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
+ u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
+ u[4] = _mm_srai_epi32(v[4], DCT_CONST_BITS);
+ u[5] = _mm_srai_epi32(v[5], DCT_CONST_BITS);
+ u[6] = _mm_srai_epi32(v[6], DCT_CONST_BITS);
+ u[7] = _mm_srai_epi32(v[7], DCT_CONST_BITS);
+
+ in[0] = _mm_packs_epi32(u[0], u[1]);
+ in[4] = _mm_packs_epi32(u[4], u[5]);
+ in[8] = _mm_packs_epi32(u[2], u[3]);
+ in[12] = _mm_packs_epi32(u[6], u[7]);
+
+ u[0] = _mm_unpacklo_epi16(p[5], p[6]);
+ u[1] = _mm_unpackhi_epi16(p[5], p[6]);
+ v[0] = _mm_madd_epi16(u[0], k__cospi_m16_p16);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_m16_p16);
+ v[2] = _mm_madd_epi16(u[0], k__cospi_p16_p16);
+ v[3] = _mm_madd_epi16(u[1], k__cospi_p16_p16);
+
+ u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+
+ u[0] = _mm_packs_epi32(v[0], v[1]);
+ u[1] = _mm_packs_epi32(v[2], v[3]);
+
+ t[0] = _mm_add_epi16(p[4], u[0]);
+ t[1] = _mm_sub_epi16(p[4], u[0]);
+ t[2] = _mm_sub_epi16(p[7], u[1]);
+ t[3] = _mm_add_epi16(p[7], u[1]);
+
+ u[0] = _mm_unpacklo_epi16(t[0], t[3]);
+ u[1] = _mm_unpackhi_epi16(t[0], t[3]);
+ u[2] = _mm_unpacklo_epi16(t[1], t[2]);
+ u[3] = _mm_unpackhi_epi16(t[1], t[2]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_p28_p04);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_p28_p04);
+ v[2] = _mm_madd_epi16(u[2], k__cospi_p12_p20);
+ v[3] = _mm_madd_epi16(u[3], k__cospi_p12_p20);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_m20_p12);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_m20_p12);
+ v[6] = _mm_madd_epi16(u[0], k__cospi_m04_p28);
+ v[7] = _mm_madd_epi16(u[1], k__cospi_m04_p28);
+
+ u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING);
+ u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING);
+ u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING);
+ u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING);
+ u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+ v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS);
+ v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS);
+ v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS);
+ v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS);
+
+ in[2] = _mm_packs_epi32(v[0], v[1]);
+ in[6] = _mm_packs_epi32(v[4], v[5]);
+ in[10] = _mm_packs_epi32(v[2], v[3]);
+ in[14] = _mm_packs_epi32(v[6], v[7]);
+
+ // stage 2
+ u[0] = _mm_unpacklo_epi16(s[2], s[5]);
+ u[1] = _mm_unpackhi_epi16(s[2], s[5]);
+ u[2] = _mm_unpacklo_epi16(s[3], s[4]);
+ u[3] = _mm_unpackhi_epi16(s[3], s[4]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_m16_p16);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_m16_p16);
+ v[2] = _mm_madd_epi16(u[2], k__cospi_m16_p16);
+ v[3] = _mm_madd_epi16(u[3], k__cospi_m16_p16);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_p16_p16);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_p16_p16);
+ v[6] = _mm_madd_epi16(u[0], k__cospi_p16_p16);
+ v[7] = _mm_madd_epi16(u[1], k__cospi_p16_p16);
+
+ u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING);
+ u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING);
+ u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING);
+ u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING);
+ u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+ v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS);
+ v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS);
+ v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS);
+ v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS);
+
+ t[2] = _mm_packs_epi32(v[0], v[1]);
+ t[3] = _mm_packs_epi32(v[2], v[3]);
+ t[4] = _mm_packs_epi32(v[4], v[5]);
+ t[5] = _mm_packs_epi32(v[6], v[7]);
+
+ // stage 3
+ p[0] = _mm_add_epi16(s[0], t[3]);
+ p[1] = _mm_add_epi16(s[1], t[2]);
+ p[2] = _mm_sub_epi16(s[1], t[2]);
+ p[3] = _mm_sub_epi16(s[0], t[3]);
+ p[4] = _mm_sub_epi16(s[7], t[4]);
+ p[5] = _mm_sub_epi16(s[6], t[5]);
+ p[6] = _mm_add_epi16(s[6], t[5]);
+ p[7] = _mm_add_epi16(s[7], t[4]);
+
+ // stage 4
+ u[0] = _mm_unpacklo_epi16(p[1], p[6]);
+ u[1] = _mm_unpackhi_epi16(p[1], p[6]);
+ u[2] = _mm_unpacklo_epi16(p[2], p[5]);
+ u[3] = _mm_unpackhi_epi16(p[2], p[5]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_m08_p24);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_m08_p24);
+ v[2] = _mm_madd_epi16(u[2], k__cospi_m24_m08);
+ v[3] = _mm_madd_epi16(u[3], k__cospi_m24_m08);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_m08_p24);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_m08_p24);
+ v[6] = _mm_madd_epi16(u[0], k__cospi_p24_p08);
+ v[7] = _mm_madd_epi16(u[1], k__cospi_p24_p08);
+
+ u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING);
+ u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING);
+ u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING);
+ u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING);
+ u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+ v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS);
+ v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS);
+ v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS);
+ v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS);
+
+ t[1] = _mm_packs_epi32(v[0], v[1]);
+ t[2] = _mm_packs_epi32(v[2], v[3]);
+ t[5] = _mm_packs_epi32(v[4], v[5]);
+ t[6] = _mm_packs_epi32(v[6], v[7]);
+
+ // stage 5
+ s[0] = _mm_add_epi16(p[0], t[1]);
+ s[1] = _mm_sub_epi16(p[0], t[1]);
+ s[2] = _mm_sub_epi16(p[3], t[2]);
+ s[3] = _mm_add_epi16(p[3], t[2]);
+ s[4] = _mm_add_epi16(p[4], t[5]);
+ s[5] = _mm_sub_epi16(p[4], t[5]);
+ s[6] = _mm_sub_epi16(p[7], t[6]);
+ s[7] = _mm_add_epi16(p[7], t[6]);
+
+ // stage 6
+ u[0] = _mm_unpacklo_epi16(s[0], s[7]);
+ u[1] = _mm_unpackhi_epi16(s[0], s[7]);
+ u[2] = _mm_unpacklo_epi16(s[1], s[6]);
+ u[3] = _mm_unpackhi_epi16(s[1], s[6]);
+ u[4] = _mm_unpacklo_epi16(s[2], s[5]);
+ u[5] = _mm_unpackhi_epi16(s[2], s[5]);
+ u[6] = _mm_unpacklo_epi16(s[3], s[4]);
+ u[7] = _mm_unpackhi_epi16(s[3], s[4]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_p30_p02);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_p30_p02);
+ v[2] = _mm_madd_epi16(u[2], k__cospi_p14_p18);
+ v[3] = _mm_madd_epi16(u[3], k__cospi_p14_p18);
+ v[4] = _mm_madd_epi16(u[4], k__cospi_p22_p10);
+ v[5] = _mm_madd_epi16(u[5], k__cospi_p22_p10);
+ v[6] = _mm_madd_epi16(u[6], k__cospi_p06_p26);
+ v[7] = _mm_madd_epi16(u[7], k__cospi_p06_p26);
+ v[8] = _mm_madd_epi16(u[6], k__cospi_m26_p06);
+ v[9] = _mm_madd_epi16(u[7], k__cospi_m26_p06);
+ v[10] = _mm_madd_epi16(u[4], k__cospi_m10_p22);
+ v[11] = _mm_madd_epi16(u[5], k__cospi_m10_p22);
+ v[12] = _mm_madd_epi16(u[2], k__cospi_m18_p14);
+ v[13] = _mm_madd_epi16(u[3], k__cospi_m18_p14);
+ v[14] = _mm_madd_epi16(u[0], k__cospi_m02_p30);
+ v[15] = _mm_madd_epi16(u[1], k__cospi_m02_p30);
+
+ u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING);
+ u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING);
+ u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING);
+ u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING);
+ u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING);
+ u[8] = _mm_add_epi32(v[8], k__DCT_CONST_ROUNDING);
+ u[9] = _mm_add_epi32(v[9], k__DCT_CONST_ROUNDING);
+ u[10] = _mm_add_epi32(v[10], k__DCT_CONST_ROUNDING);
+ u[11] = _mm_add_epi32(v[11], k__DCT_CONST_ROUNDING);
+ u[12] = _mm_add_epi32(v[12], k__DCT_CONST_ROUNDING);
+ u[13] = _mm_add_epi32(v[13], k__DCT_CONST_ROUNDING);
+ u[14] = _mm_add_epi32(v[14], k__DCT_CONST_ROUNDING);
+ u[15] = _mm_add_epi32(v[15], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+ v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS);
+ v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS);
+ v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS);
+ v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS);
+ v[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS);
+ v[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS);
+ v[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS);
+ v[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS);
+ v[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS);
+ v[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS);
+ v[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS);
+ v[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS);
+
+ in[1] = _mm_packs_epi32(v[0], v[1]);
+ in[9] = _mm_packs_epi32(v[2], v[3]);
+ in[5] = _mm_packs_epi32(v[4], v[5]);
+ in[13] = _mm_packs_epi32(v[6], v[7]);
+ in[3] = _mm_packs_epi32(v[8], v[9]);
+ in[11] = _mm_packs_epi32(v[10], v[11]);
+ in[7] = _mm_packs_epi32(v[12], v[13]);
+ in[15] = _mm_packs_epi32(v[14], v[15]);
+}
+
+void fadst16_1d_8col(__m128i *in) {
+ // perform 16x16 1-D ADST for 8 columns
+ __m128i s[16], x[16], u[32], v[32];
+ const __m128i k__cospi_p01_p31 = pair_set_epi16(cospi_1_64, cospi_31_64);
+ const __m128i k__cospi_p31_m01 = pair_set_epi16(cospi_31_64, -cospi_1_64);
+ const __m128i k__cospi_p05_p27 = pair_set_epi16(cospi_5_64, cospi_27_64);
+ const __m128i k__cospi_p27_m05 = pair_set_epi16(cospi_27_64, -cospi_5_64);
+ const __m128i k__cospi_p09_p23 = pair_set_epi16(cospi_9_64, cospi_23_64);
+ const __m128i k__cospi_p23_m09 = pair_set_epi16(cospi_23_64, -cospi_9_64);
+ const __m128i k__cospi_p13_p19 = pair_set_epi16(cospi_13_64, cospi_19_64);
+ const __m128i k__cospi_p19_m13 = pair_set_epi16(cospi_19_64, -cospi_13_64);
+ const __m128i k__cospi_p17_p15 = pair_set_epi16(cospi_17_64, cospi_15_64);
+ const __m128i k__cospi_p15_m17 = pair_set_epi16(cospi_15_64, -cospi_17_64);
+ const __m128i k__cospi_p21_p11 = pair_set_epi16(cospi_21_64, cospi_11_64);
+ const __m128i k__cospi_p11_m21 = pair_set_epi16(cospi_11_64, -cospi_21_64);
+ const __m128i k__cospi_p25_p07 = pair_set_epi16(cospi_25_64, cospi_7_64);
+ const __m128i k__cospi_p07_m25 = pair_set_epi16(cospi_7_64, -cospi_25_64);
+ const __m128i k__cospi_p29_p03 = pair_set_epi16(cospi_29_64, cospi_3_64);
+ const __m128i k__cospi_p03_m29 = pair_set_epi16(cospi_3_64, -cospi_29_64);
+ const __m128i k__cospi_p04_p28 = pair_set_epi16(cospi_4_64, cospi_28_64);
+ const __m128i k__cospi_p28_m04 = pair_set_epi16(cospi_28_64, -cospi_4_64);
+ const __m128i k__cospi_p20_p12 = pair_set_epi16(cospi_20_64, cospi_12_64);
+ const __m128i k__cospi_p12_m20 = pair_set_epi16(cospi_12_64, -cospi_20_64);
+ const __m128i k__cospi_m28_p04 = pair_set_epi16(-cospi_28_64, cospi_4_64);
+ const __m128i k__cospi_m12_p20 = pair_set_epi16(-cospi_12_64, cospi_20_64);
+ const __m128i k__cospi_p08_p24 = pair_set_epi16(cospi_8_64, cospi_24_64);
+ const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64);
+ const __m128i k__cospi_m24_p08 = pair_set_epi16(-cospi_24_64, cospi_8_64);
+ const __m128i k__cospi_m16_m16 = _mm_set1_epi16(-cospi_16_64);
+ const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
+ const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
+ const __m128i k__cospi_m16_p16 = pair_set_epi16(-cospi_16_64, cospi_16_64);
+ const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
+ const __m128i kZero = _mm_set1_epi16(0);
+
+ u[0] = _mm_unpacklo_epi16(in[15], in[0]);
+ u[1] = _mm_unpackhi_epi16(in[15], in[0]);
+ u[2] = _mm_unpacklo_epi16(in[13], in[2]);
+ u[3] = _mm_unpackhi_epi16(in[13], in[2]);
+ u[4] = _mm_unpacklo_epi16(in[11], in[4]);
+ u[5] = _mm_unpackhi_epi16(in[11], in[4]);
+ u[6] = _mm_unpacklo_epi16(in[9], in[6]);
+ u[7] = _mm_unpackhi_epi16(in[9], in[6]);
+ u[8] = _mm_unpacklo_epi16(in[7], in[8]);
+ u[9] = _mm_unpackhi_epi16(in[7], in[8]);
+ u[10] = _mm_unpacklo_epi16(in[5], in[10]);
+ u[11] = _mm_unpackhi_epi16(in[5], in[10]);
+ u[12] = _mm_unpacklo_epi16(in[3], in[12]);
+ u[13] = _mm_unpackhi_epi16(in[3], in[12]);
+ u[14] = _mm_unpacklo_epi16(in[1], in[14]);
+ u[15] = _mm_unpackhi_epi16(in[1], in[14]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_p01_p31);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_p01_p31);
+ v[2] = _mm_madd_epi16(u[0], k__cospi_p31_m01);
+ v[3] = _mm_madd_epi16(u[1], k__cospi_p31_m01);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_p05_p27);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_p05_p27);
+ v[6] = _mm_madd_epi16(u[2], k__cospi_p27_m05);
+ v[7] = _mm_madd_epi16(u[3], k__cospi_p27_m05);
+ v[8] = _mm_madd_epi16(u[4], k__cospi_p09_p23);
+ v[9] = _mm_madd_epi16(u[5], k__cospi_p09_p23);
+ v[10] = _mm_madd_epi16(u[4], k__cospi_p23_m09);
+ v[11] = _mm_madd_epi16(u[5], k__cospi_p23_m09);
+ v[12] = _mm_madd_epi16(u[6], k__cospi_p13_p19);
+ v[13] = _mm_madd_epi16(u[7], k__cospi_p13_p19);
+ v[14] = _mm_madd_epi16(u[6], k__cospi_p19_m13);
+ v[15] = _mm_madd_epi16(u[7], k__cospi_p19_m13);
+ v[16] = _mm_madd_epi16(u[8], k__cospi_p17_p15);
+ v[17] = _mm_madd_epi16(u[9], k__cospi_p17_p15);
+ v[18] = _mm_madd_epi16(u[8], k__cospi_p15_m17);
+ v[19] = _mm_madd_epi16(u[9], k__cospi_p15_m17);
+ v[20] = _mm_madd_epi16(u[10], k__cospi_p21_p11);
+ v[21] = _mm_madd_epi16(u[11], k__cospi_p21_p11);
+ v[22] = _mm_madd_epi16(u[10], k__cospi_p11_m21);
+ v[23] = _mm_madd_epi16(u[11], k__cospi_p11_m21);
+ v[24] = _mm_madd_epi16(u[12], k__cospi_p25_p07);
+ v[25] = _mm_madd_epi16(u[13], k__cospi_p25_p07);
+ v[26] = _mm_madd_epi16(u[12], k__cospi_p07_m25);
+ v[27] = _mm_madd_epi16(u[13], k__cospi_p07_m25);
+ v[28] = _mm_madd_epi16(u[14], k__cospi_p29_p03);
+ v[29] = _mm_madd_epi16(u[15], k__cospi_p29_p03);
+ v[30] = _mm_madd_epi16(u[14], k__cospi_p03_m29);
+ v[31] = _mm_madd_epi16(u[15], k__cospi_p03_m29);
+
+ u[0] = _mm_add_epi32(v[0], v[16]);
+ u[1] = _mm_add_epi32(v[1], v[17]);
+ u[2] = _mm_add_epi32(v[2], v[18]);
+ u[3] = _mm_add_epi32(v[3], v[19]);
+ u[4] = _mm_add_epi32(v[4], v[20]);
+ u[5] = _mm_add_epi32(v[5], v[21]);
+ u[6] = _mm_add_epi32(v[6], v[22]);
+ u[7] = _mm_add_epi32(v[7], v[23]);
+ u[8] = _mm_add_epi32(v[8], v[24]);
+ u[9] = _mm_add_epi32(v[9], v[25]);
+ u[10] = _mm_add_epi32(v[10], v[26]);
+ u[11] = _mm_add_epi32(v[11], v[27]);
+ u[12] = _mm_add_epi32(v[12], v[28]);
+ u[13] = _mm_add_epi32(v[13], v[29]);
+ u[14] = _mm_add_epi32(v[14], v[30]);
+ u[15] = _mm_add_epi32(v[15], v[31]);
+ u[16] = _mm_sub_epi32(v[0], v[16]);
+ u[17] = _mm_sub_epi32(v[1], v[17]);
+ u[18] = _mm_sub_epi32(v[2], v[18]);
+ u[19] = _mm_sub_epi32(v[3], v[19]);
+ u[20] = _mm_sub_epi32(v[4], v[20]);
+ u[21] = _mm_sub_epi32(v[5], v[21]);
+ u[22] = _mm_sub_epi32(v[6], v[22]);
+ u[23] = _mm_sub_epi32(v[7], v[23]);
+ u[24] = _mm_sub_epi32(v[8], v[24]);
+ u[25] = _mm_sub_epi32(v[9], v[25]);
+ u[26] = _mm_sub_epi32(v[10], v[26]);
+ u[27] = _mm_sub_epi32(v[11], v[27]);
+ u[28] = _mm_sub_epi32(v[12], v[28]);
+ u[29] = _mm_sub_epi32(v[13], v[29]);
+ u[30] = _mm_sub_epi32(v[14], v[30]);
+ u[31] = _mm_sub_epi32(v[15], v[31]);
+
+ v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
+ v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
+ v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING);
+ v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING);
+ v[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING);
+ v[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING);
+ v[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING);
+ v[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING);
+ v[8] = _mm_add_epi32(u[8], k__DCT_CONST_ROUNDING);
+ v[9] = _mm_add_epi32(u[9], k__DCT_CONST_ROUNDING);
+ v[10] = _mm_add_epi32(u[10], k__DCT_CONST_ROUNDING);
+ v[11] = _mm_add_epi32(u[11], k__DCT_CONST_ROUNDING);
+ v[12] = _mm_add_epi32(u[12], k__DCT_CONST_ROUNDING);
+ v[13] = _mm_add_epi32(u[13], k__DCT_CONST_ROUNDING);
+ v[14] = _mm_add_epi32(u[14], k__DCT_CONST_ROUNDING);
+ v[15] = _mm_add_epi32(u[15], k__DCT_CONST_ROUNDING);
+ v[16] = _mm_add_epi32(u[16], k__DCT_CONST_ROUNDING);
+ v[17] = _mm_add_epi32(u[17], k__DCT_CONST_ROUNDING);
+ v[18] = _mm_add_epi32(u[18], k__DCT_CONST_ROUNDING);
+ v[19] = _mm_add_epi32(u[19], k__DCT_CONST_ROUNDING);
+ v[20] = _mm_add_epi32(u[20], k__DCT_CONST_ROUNDING);
+ v[21] = _mm_add_epi32(u[21], k__DCT_CONST_ROUNDING);
+ v[22] = _mm_add_epi32(u[22], k__DCT_CONST_ROUNDING);
+ v[23] = _mm_add_epi32(u[23], k__DCT_CONST_ROUNDING);
+ v[24] = _mm_add_epi32(u[24], k__DCT_CONST_ROUNDING);
+ v[25] = _mm_add_epi32(u[25], k__DCT_CONST_ROUNDING);
+ v[26] = _mm_add_epi32(u[26], k__DCT_CONST_ROUNDING);
+ v[27] = _mm_add_epi32(u[27], k__DCT_CONST_ROUNDING);
+ v[28] = _mm_add_epi32(u[28], k__DCT_CONST_ROUNDING);
+ v[29] = _mm_add_epi32(u[29], k__DCT_CONST_ROUNDING);
+ v[30] = _mm_add_epi32(u[30], k__DCT_CONST_ROUNDING);
+ v[31] = _mm_add_epi32(u[31], k__DCT_CONST_ROUNDING);
+
+ u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS);
+ u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS);
+ u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
+ u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
+ u[4] = _mm_srai_epi32(v[4], DCT_CONST_BITS);
+ u[5] = _mm_srai_epi32(v[5], DCT_CONST_BITS);
+ u[6] = _mm_srai_epi32(v[6], DCT_CONST_BITS);
+ u[7] = _mm_srai_epi32(v[7], DCT_CONST_BITS);
+ u[8] = _mm_srai_epi32(v[8], DCT_CONST_BITS);
+ u[9] = _mm_srai_epi32(v[9], DCT_CONST_BITS);
+ u[10] = _mm_srai_epi32(v[10], DCT_CONST_BITS);
+ u[11] = _mm_srai_epi32(v[11], DCT_CONST_BITS);
+ u[12] = _mm_srai_epi32(v[12], DCT_CONST_BITS);
+ u[13] = _mm_srai_epi32(v[13], DCT_CONST_BITS);
+ u[14] = _mm_srai_epi32(v[14], DCT_CONST_BITS);
+ u[15] = _mm_srai_epi32(v[15], DCT_CONST_BITS);
+ u[16] = _mm_srai_epi32(v[16], DCT_CONST_BITS);
+ u[17] = _mm_srai_epi32(v[17], DCT_CONST_BITS);
+ u[18] = _mm_srai_epi32(v[18], DCT_CONST_BITS);
+ u[19] = _mm_srai_epi32(v[19], DCT_CONST_BITS);
+ u[20] = _mm_srai_epi32(v[20], DCT_CONST_BITS);
+ u[21] = _mm_srai_epi32(v[21], DCT_CONST_BITS);
+ u[22] = _mm_srai_epi32(v[22], DCT_CONST_BITS);
+ u[23] = _mm_srai_epi32(v[23], DCT_CONST_BITS);
+ u[24] = _mm_srai_epi32(v[24], DCT_CONST_BITS);
+ u[25] = _mm_srai_epi32(v[25], DCT_CONST_BITS);
+ u[26] = _mm_srai_epi32(v[26], DCT_CONST_BITS);
+ u[27] = _mm_srai_epi32(v[27], DCT_CONST_BITS);
+ u[28] = _mm_srai_epi32(v[28], DCT_CONST_BITS);
+ u[29] = _mm_srai_epi32(v[29], DCT_CONST_BITS);
+ u[30] = _mm_srai_epi32(v[30], DCT_CONST_BITS);
+ u[31] = _mm_srai_epi32(v[31], DCT_CONST_BITS);
+
+ s[0] = _mm_packs_epi32(u[0], u[1]);
+ s[1] = _mm_packs_epi32(u[2], u[3]);
+ s[2] = _mm_packs_epi32(u[4], u[5]);
+ s[3] = _mm_packs_epi32(u[6], u[7]);
+ s[4] = _mm_packs_epi32(u[8], u[9]);
+ s[5] = _mm_packs_epi32(u[10], u[11]);
+ s[6] = _mm_packs_epi32(u[12], u[13]);
+ s[7] = _mm_packs_epi32(u[14], u[15]);
+ s[8] = _mm_packs_epi32(u[16], u[17]);
+ s[9] = _mm_packs_epi32(u[18], u[19]);
+ s[10] = _mm_packs_epi32(u[20], u[21]);
+ s[11] = _mm_packs_epi32(u[22], u[23]);
+ s[12] = _mm_packs_epi32(u[24], u[25]);
+ s[13] = _mm_packs_epi32(u[26], u[27]);
+ s[14] = _mm_packs_epi32(u[28], u[29]);
+ s[15] = _mm_packs_epi32(u[30], u[31]);
+
+ // stage 2
+ u[0] = _mm_unpacklo_epi16(s[8], s[9]);
+ u[1] = _mm_unpackhi_epi16(s[8], s[9]);
+ u[2] = _mm_unpacklo_epi16(s[10], s[11]);
+ u[3] = _mm_unpackhi_epi16(s[10], s[11]);
+ u[4] = _mm_unpacklo_epi16(s[12], s[13]);
+ u[5] = _mm_unpackhi_epi16(s[12], s[13]);
+ u[6] = _mm_unpacklo_epi16(s[14], s[15]);
+ u[7] = _mm_unpackhi_epi16(s[14], s[15]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_p04_p28);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_p04_p28);
+ v[2] = _mm_madd_epi16(u[0], k__cospi_p28_m04);
+ v[3] = _mm_madd_epi16(u[1], k__cospi_p28_m04);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_p20_p12);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_p20_p12);
+ v[6] = _mm_madd_epi16(u[2], k__cospi_p12_m20);
+ v[7] = _mm_madd_epi16(u[3], k__cospi_p12_m20);
+ v[8] = _mm_madd_epi16(u[4], k__cospi_m28_p04);
+ v[9] = _mm_madd_epi16(u[5], k__cospi_m28_p04);
+ v[10] = _mm_madd_epi16(u[4], k__cospi_p04_p28);
+ v[11] = _mm_madd_epi16(u[5], k__cospi_p04_p28);
+ v[12] = _mm_madd_epi16(u[6], k__cospi_m12_p20);
+ v[13] = _mm_madd_epi16(u[7], k__cospi_m12_p20);
+ v[14] = _mm_madd_epi16(u[6], k__cospi_p20_p12);
+ v[15] = _mm_madd_epi16(u[7], k__cospi_p20_p12);
+
+ u[0] = _mm_add_epi32(v[0], v[8]);
+ u[1] = _mm_add_epi32(v[1], v[9]);
+ u[2] = _mm_add_epi32(v[2], v[10]);
+ u[3] = _mm_add_epi32(v[3], v[11]);
+ u[4] = _mm_add_epi32(v[4], v[12]);
+ u[5] = _mm_add_epi32(v[5], v[13]);
+ u[6] = _mm_add_epi32(v[6], v[14]);
+ u[7] = _mm_add_epi32(v[7], v[15]);
+ u[8] = _mm_sub_epi32(v[0], v[8]);
+ u[9] = _mm_sub_epi32(v[1], v[9]);
+ u[10] = _mm_sub_epi32(v[2], v[10]);
+ u[11] = _mm_sub_epi32(v[3], v[11]);
+ u[12] = _mm_sub_epi32(v[4], v[12]);
+ u[13] = _mm_sub_epi32(v[5], v[13]);
+ u[14] = _mm_sub_epi32(v[6], v[14]);
+ u[15] = _mm_sub_epi32(v[7], v[15]);
+
+ v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
+ v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
+ v[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING);
+ v[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING);
+ v[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING);
+ v[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING);
+ v[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING);
+ v[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING);
+ v[8] = _mm_add_epi32(u[8], k__DCT_CONST_ROUNDING);
+ v[9] = _mm_add_epi32(u[9], k__DCT_CONST_ROUNDING);
+ v[10] = _mm_add_epi32(u[10], k__DCT_CONST_ROUNDING);
+ v[11] = _mm_add_epi32(u[11], k__DCT_CONST_ROUNDING);
+ v[12] = _mm_add_epi32(u[12], k__DCT_CONST_ROUNDING);
+ v[13] = _mm_add_epi32(u[13], k__DCT_CONST_ROUNDING);
+ v[14] = _mm_add_epi32(u[14], k__DCT_CONST_ROUNDING);
+ v[15] = _mm_add_epi32(u[15], k__DCT_CONST_ROUNDING);
+
+ u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS);
+ u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS);
+ u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
+ u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
+ u[4] = _mm_srai_epi32(v[4], DCT_CONST_BITS);
+ u[5] = _mm_srai_epi32(v[5], DCT_CONST_BITS);
+ u[6] = _mm_srai_epi32(v[6], DCT_CONST_BITS);
+ u[7] = _mm_srai_epi32(v[7], DCT_CONST_BITS);
+ u[8] = _mm_srai_epi32(v[8], DCT_CONST_BITS);
+ u[9] = _mm_srai_epi32(v[9], DCT_CONST_BITS);
+ u[10] = _mm_srai_epi32(v[10], DCT_CONST_BITS);
+ u[11] = _mm_srai_epi32(v[11], DCT_CONST_BITS);
+ u[12] = _mm_srai_epi32(v[12], DCT_CONST_BITS);
+ u[13] = _mm_srai_epi32(v[13], DCT_CONST_BITS);
+ u[14] = _mm_srai_epi32(v[14], DCT_CONST_BITS);
+ u[15] = _mm_srai_epi32(v[15], DCT_CONST_BITS);
+
+ x[0] = _mm_add_epi16(s[0], s[4]);
+ x[1] = _mm_add_epi16(s[1], s[5]);
+ x[2] = _mm_add_epi16(s[2], s[6]);
+ x[3] = _mm_add_epi16(s[3], s[7]);
+ x[4] = _mm_sub_epi16(s[0], s[4]);
+ x[5] = _mm_sub_epi16(s[1], s[5]);
+ x[6] = _mm_sub_epi16(s[2], s[6]);
+ x[7] = _mm_sub_epi16(s[3], s[7]);
+ x[8] = _mm_packs_epi32(u[0], u[1]);
+ x[9] = _mm_packs_epi32(u[2], u[3]);
+ x[10] = _mm_packs_epi32(u[4], u[5]);
+ x[11] = _mm_packs_epi32(u[6], u[7]);
+ x[12] = _mm_packs_epi32(u[8], u[9]);
+ x[13] = _mm_packs_epi32(u[10], u[11]);
+ x[14] = _mm_packs_epi32(u[12], u[13]);
+ x[15] = _mm_packs_epi32(u[14], u[15]);
+
+ // stage 3
+ u[0] = _mm_unpacklo_epi16(x[4], x[5]);
+ u[1] = _mm_unpackhi_epi16(x[4], x[5]);
+ u[2] = _mm_unpacklo_epi16(x[6], x[7]);
+ u[3] = _mm_unpackhi_epi16(x[6], x[7]);
+ u[4] = _mm_unpacklo_epi16(x[12], x[13]);
+ u[5] = _mm_unpackhi_epi16(x[12], x[13]);
+ u[6] = _mm_unpacklo_epi16(x[14], x[15]);
+ u[7] = _mm_unpackhi_epi16(x[14], x[15]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_p08_p24);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_p08_p24);
+ v[2] = _mm_madd_epi16(u[0], k__cospi_p24_m08);
+ v[3] = _mm_madd_epi16(u[1], k__cospi_p24_m08);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_m24_p08);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_m24_p08);
+ v[6] = _mm_madd_epi16(u[2], k__cospi_p08_p24);
+ v[7] = _mm_madd_epi16(u[3], k__cospi_p08_p24);
+ v[8] = _mm_madd_epi16(u[4], k__cospi_p08_p24);
+ v[9] = _mm_madd_epi16(u[5], k__cospi_p08_p24);
+ v[10] = _mm_madd_epi16(u[4], k__cospi_p24_m08);
+ v[11] = _mm_madd_epi16(u[5], k__cospi_p24_m08);
+ v[12] = _mm_madd_epi16(u[6], k__cospi_m24_p08);
+ v[13] = _mm_madd_epi16(u[7], k__cospi_m24_p08);
+ v[14] = _mm_madd_epi16(u[6], k__cospi_p08_p24);
+ v[15] = _mm_madd_epi16(u[7], k__cospi_p08_p24);
+
+ u[0] = _mm_add_epi32(v[0], v[4]);
+ u[1] = _mm_add_epi32(v[1], v[5]);
+ u[2] = _mm_add_epi32(v[2], v[6]);
+ u[3] = _mm_add_epi32(v[3], v[7]);
+ u[4] = _mm_sub_epi32(v[0], v[4]);
+ u[5] = _mm_sub_epi32(v[1], v[5]);
+ u[6] = _mm_sub_epi32(v[2], v[6]);
+ u[7] = _mm_sub_epi32(v[3], v[7]);
+ u[8] = _mm_add_epi32(v[8], v[12]);
+ u[9] = _mm_add_epi32(v[9], v[13]);
+ u[10] = _mm_add_epi32(v[10], v[14]);
+ u[11] = _mm_add_epi32(v[11], v[15]);
+ u[12] = _mm_sub_epi32(v[8], v[12]);
+ u[13] = _mm_sub_epi32(v[9], v[13]);
+ u[14] = _mm_sub_epi32(v[10], v[14]);
+ u[15] = _mm_sub_epi32(v[11], v[15]);
+
+ u[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(u[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(u[3], k__DCT_CONST_ROUNDING);
+ u[4] = _mm_add_epi32(u[4], k__DCT_CONST_ROUNDING);
+ u[5] = _mm_add_epi32(u[5], k__DCT_CONST_ROUNDING);
+ u[6] = _mm_add_epi32(u[6], k__DCT_CONST_ROUNDING);
+ u[7] = _mm_add_epi32(u[7], k__DCT_CONST_ROUNDING);
+ u[8] = _mm_add_epi32(u[8], k__DCT_CONST_ROUNDING);
+ u[9] = _mm_add_epi32(u[9], k__DCT_CONST_ROUNDING);
+ u[10] = _mm_add_epi32(u[10], k__DCT_CONST_ROUNDING);
+ u[11] = _mm_add_epi32(u[11], k__DCT_CONST_ROUNDING);
+ u[12] = _mm_add_epi32(u[12], k__DCT_CONST_ROUNDING);
+ u[13] = _mm_add_epi32(u[13], k__DCT_CONST_ROUNDING);
+ u[14] = _mm_add_epi32(u[14], k__DCT_CONST_ROUNDING);
+ u[15] = _mm_add_epi32(u[15], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+ v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS);
+ v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS);
+ v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS);
+ v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS);
+ v[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS);
+ v[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS);
+ v[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS);
+ v[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS);
+ v[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS);
+ v[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS);
+ v[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS);
+ v[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS);
+
+ s[0] = _mm_add_epi16(x[0], x[2]);
+ s[1] = _mm_add_epi16(x[1], x[3]);
+ s[2] = _mm_sub_epi16(x[0], x[2]);
+ s[3] = _mm_sub_epi16(x[1], x[3]);
+ s[4] = _mm_packs_epi32(v[0], v[1]);
+ s[5] = _mm_packs_epi32(v[2], v[3]);
+ s[6] = _mm_packs_epi32(v[4], v[5]);
+ s[7] = _mm_packs_epi32(v[6], v[7]);
+ s[8] = _mm_add_epi16(x[8], x[10]);
+ s[9] = _mm_add_epi16(x[9], x[11]);
+ s[10] = _mm_sub_epi16(x[8], x[10]);
+ s[11] = _mm_sub_epi16(x[9], x[11]);
+ s[12] = _mm_packs_epi32(v[8], v[9]);
+ s[13] = _mm_packs_epi32(v[10], v[11]);
+ s[14] = _mm_packs_epi32(v[12], v[13]);
+ s[15] = _mm_packs_epi32(v[14], v[15]);
+
+ // stage 4
+ u[0] = _mm_unpacklo_epi16(s[2], s[3]);
+ u[1] = _mm_unpackhi_epi16(s[2], s[3]);
+ u[2] = _mm_unpacklo_epi16(s[6], s[7]);
+ u[3] = _mm_unpackhi_epi16(s[6], s[7]);
+ u[4] = _mm_unpacklo_epi16(s[10], s[11]);
+ u[5] = _mm_unpackhi_epi16(s[10], s[11]);
+ u[6] = _mm_unpacklo_epi16(s[14], s[15]);
+ u[7] = _mm_unpackhi_epi16(s[14], s[15]);
+
+ v[0] = _mm_madd_epi16(u[0], k__cospi_m16_m16);
+ v[1] = _mm_madd_epi16(u[1], k__cospi_m16_m16);
+ v[2] = _mm_madd_epi16(u[0], k__cospi_p16_m16);
+ v[3] = _mm_madd_epi16(u[1], k__cospi_p16_m16);
+ v[4] = _mm_madd_epi16(u[2], k__cospi_p16_p16);
+ v[5] = _mm_madd_epi16(u[3], k__cospi_p16_p16);
+ v[6] = _mm_madd_epi16(u[2], k__cospi_m16_p16);
+ v[7] = _mm_madd_epi16(u[3], k__cospi_m16_p16);
+ v[8] = _mm_madd_epi16(u[4], k__cospi_p16_p16);
+ v[9] = _mm_madd_epi16(u[5], k__cospi_p16_p16);
+ v[10] = _mm_madd_epi16(u[4], k__cospi_m16_p16);
+ v[11] = _mm_madd_epi16(u[5], k__cospi_m16_p16);
+ v[12] = _mm_madd_epi16(u[6], k__cospi_m16_m16);
+ v[13] = _mm_madd_epi16(u[7], k__cospi_m16_m16);
+ v[14] = _mm_madd_epi16(u[6], k__cospi_p16_m16);
+ v[15] = _mm_madd_epi16(u[7], k__cospi_p16_m16);
+
+ u[0] = _mm_add_epi32(v[0], k__DCT_CONST_ROUNDING);
+ u[1] = _mm_add_epi32(v[1], k__DCT_CONST_ROUNDING);
+ u[2] = _mm_add_epi32(v[2], k__DCT_CONST_ROUNDING);
+ u[3] = _mm_add_epi32(v[3], k__DCT_CONST_ROUNDING);
+ u[4] = _mm_add_epi32(v[4], k__DCT_CONST_ROUNDING);
+ u[5] = _mm_add_epi32(v[5], k__DCT_CONST_ROUNDING);
+ u[6] = _mm_add_epi32(v[6], k__DCT_CONST_ROUNDING);
+ u[7] = _mm_add_epi32(v[7], k__DCT_CONST_ROUNDING);
+ u[8] = _mm_add_epi32(v[8], k__DCT_CONST_ROUNDING);
+ u[9] = _mm_add_epi32(v[9], k__DCT_CONST_ROUNDING);
+ u[10] = _mm_add_epi32(v[10], k__DCT_CONST_ROUNDING);
+ u[11] = _mm_add_epi32(v[11], k__DCT_CONST_ROUNDING);
+ u[12] = _mm_add_epi32(v[12], k__DCT_CONST_ROUNDING);
+ u[13] = _mm_add_epi32(v[13], k__DCT_CONST_ROUNDING);
+ u[14] = _mm_add_epi32(v[14], k__DCT_CONST_ROUNDING);
+ u[15] = _mm_add_epi32(v[15], k__DCT_CONST_ROUNDING);
+
+ v[0] = _mm_srai_epi32(u[0], DCT_CONST_BITS);
+ v[1] = _mm_srai_epi32(u[1], DCT_CONST_BITS);
+ v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
+ v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
+ v[4] = _mm_srai_epi32(u[4], DCT_CONST_BITS);
+ v[5] = _mm_srai_epi32(u[5], DCT_CONST_BITS);
+ v[6] = _mm_srai_epi32(u[6], DCT_CONST_BITS);
+ v[7] = _mm_srai_epi32(u[7], DCT_CONST_BITS);
+ v[8] = _mm_srai_epi32(u[8], DCT_CONST_BITS);
+ v[9] = _mm_srai_epi32(u[9], DCT_CONST_BITS);
+ v[10] = _mm_srai_epi32(u[10], DCT_CONST_BITS);
+ v[11] = _mm_srai_epi32(u[11], DCT_CONST_BITS);
+ v[12] = _mm_srai_epi32(u[12], DCT_CONST_BITS);
+ v[13] = _mm_srai_epi32(u[13], DCT_CONST_BITS);
+ v[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS);
+ v[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS);
+
+ in[0] = s[0];
+ in[1] = _mm_sub_epi16(kZero, s[8]);
+ in[2] = s[12];
+ in[3] = _mm_sub_epi16(kZero, s[4]);
+ in[4] = _mm_packs_epi32(v[4], v[5]);
+ in[5] = _mm_packs_epi32(v[12], v[13]);
+ in[6] = _mm_packs_epi32(v[8], v[9]);
+ in[7] = _mm_packs_epi32(v[0], v[1]);
+ in[8] = _mm_packs_epi32(v[2], v[3]);
+ in[9] = _mm_packs_epi32(v[10], v[11]);
+ in[10] = _mm_packs_epi32(v[14], v[15]);
+ in[11] = _mm_packs_epi32(v[6], v[7]);
+ in[12] = s[5];
+ in[13] = _mm_sub_epi16(kZero, s[13]);
+ in[14] = s[9];
+ in[15] = _mm_sub_epi16(kZero, s[1]);
+}
+
+void fdct16_1d_sse2(__m128i *in0, __m128i *in1) {
+ fdct16_1d_8col(in0);
+ fdct16_1d_8col(in1);
+ array_transpose_16x16(in0, in1);
+}
+
+void fadst16_1d_sse2(__m128i *in0, __m128i *in1) {
+ fadst16_1d_8col(in0);
+ fadst16_1d_8col(in1);
+ array_transpose_16x16(in0, in1);
+}
+
+void vp9_short_fht16x16_sse2(int16_t *input, int16_t *output,
+ int stride, int tx_type) {
+ __m128i in0[16], in1[16];
+ load_buffer_16x16(input, in0, in1, stride);
+ switch (tx_type) {
+ case 0: // DCT_DCT
+ fdct16_1d_sse2(in0, in1);
+ right_shift_16x16(in0, in1);
+ fdct16_1d_sse2(in0, in1);
+ break;
+ case 1: // ADST_DCT
+ fadst16_1d_sse2(in0, in1);
+ right_shift_16x16(in0, in1);
+ fdct16_1d_sse2(in0, in1);
+ break;
+ case 2: // DCT_ADST
+ fdct16_1d_sse2(in0, in1);
+ right_shift_16x16(in0, in1);
+ fadst16_1d_sse2(in0, in1);
+ break;
+ case 3: // ADST_ADST
+ fadst16_1d_sse2(in0, in1);
+ right_shift_16x16(in0, in1);
+ fadst16_1d_sse2(in0, in1);
+ break;
+ default:
+ assert(0);
+ break;
+ }
+ write_buffer_16x16(output, in0, in1, 16);
+}
+
void vp9_short_fdct32x32_rd_sse2(int16_t *input,
int16_t *output_org, int pitch) {
// Calculate pre-multiplied strides
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index d08d04f..95ea60b 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -82,6 +82,7 @@
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_loopfilter_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_recon_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_intrapred_sse2.asm
+VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_intrapred_ssse3.asm
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
ifeq ($(CONFIG_POSTPROC),yes)
VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.asm
