Merge "Reducing size of MODE_INFO struct"
diff --git a/CHANGELOG b/CHANGELOG
index 4ada43a..b0d3064 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,5 +1,7 @@
xxxx-yy-zz v1.4.0 "Changes for next release"
+ vpxenc is changed to use VP9 by default.
Encoder controls added for 1 pass SVC.
+ Decoder control to toggle on/off loopfilter.
2015-04-03 v1.4.0 "Indian Runner Duck"
This release includes significant improvements to the VP9 codec.
diff --git a/build/make/gen_msvs_vcxproj.sh b/build/make/gen_msvs_vcxproj.sh
index 643ebd6..b653651 100755
--- a/build/make/gen_msvs_vcxproj.sh
+++ b/build/make/gen_msvs_vcxproj.sh
@@ -263,8 +263,8 @@
;;
arm*)
platforms[0]="ARM"
- asm_Debug_cmdline="armasm -nologo "%(FullPath)""
- asm_Release_cmdline="armasm -nologo "%(FullPath)""
+ asm_Debug_cmdline="armasm -nologo -oldit "%(FullPath)""
+ asm_Release_cmdline="armasm -nologo -oldit "%(FullPath)""
;;
*) die "Unsupported target $target!"
;;
diff --git a/rate_hist.c b/rate_hist.c
index 1cef19b..a77222b 100644
--- a/rate_hist.c
+++ b/rate_hist.c
@@ -88,6 +88,9 @@
if (now < cfg->rc_buf_initial_sz)
return;
+ if (!cfg->rc_target_bitrate)
+ return;
+
then = now;
/* Sum the size over the past rc_buf_sz ms */
diff --git a/test/lpf_8_test.cc b/test/lpf_8_test.cc
index 3d21e6c..ba51309 100644
--- a/test/lpf_8_test.cc
+++ b/test/lpf_8_test.cc
@@ -137,6 +137,20 @@
#endif // CONFIG_VP9_HIGHBITDEPTH
#endif // HAVE_NEON_ASM
+#if HAVE_MSA && (!CONFIG_VP9_HIGHBITDEPTH)
+void wrapper_vertical_16_msa(uint8_t *s, int p, const uint8_t *blimit,
+ const uint8_t *limit, const uint8_t *thresh,
+ int count) {
+ vp9_lpf_vertical_16_msa(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);
+}
+#endif // HAVE_MSA && (!CONFIG_VP9_HIGHBITDEPTH)
+
class Loop8Test6Param : public ::testing::TestWithParam<loop8_param_t> {
public:
virtual ~Loop8Test6Param() {}
@@ -676,4 +690,27 @@
#endif // CONFIG_VP9_HIGHBITDEPTH
#endif // HAVE_NEON
+#if HAVE_MSA && (!CONFIG_VP9_HIGHBITDEPTH)
+INSTANTIATE_TEST_CASE_P(
+ MSA, Loop8Test6Param,
+ ::testing::Values(
+ make_tuple(&vp9_lpf_horizontal_8_msa, &vp9_lpf_horizontal_8_c, 8, 1),
+ make_tuple(&vp9_lpf_horizontal_16_msa, &vp9_lpf_horizontal_16_c, 8, 1),
+ make_tuple(&vp9_lpf_horizontal_16_msa, &vp9_lpf_horizontal_16_c, 8, 2),
+ make_tuple(&vp9_lpf_vertical_8_msa, &vp9_lpf_vertical_8_c, 8, 1),
+ make_tuple(&wrapper_vertical_16_msa, &wrapper_vertical_16_c, 8, 1)));
+
+INSTANTIATE_TEST_CASE_P(
+ MSA, Loop8Test9Param,
+ ::testing::Values(
+ make_tuple(&vp9_lpf_horizontal_4_dual_msa,
+ &vp9_lpf_horizontal_4_dual_c, 8),
+ make_tuple(&vp9_lpf_horizontal_8_dual_msa,
+ &vp9_lpf_horizontal_8_dual_c, 8),
+ make_tuple(&vp9_lpf_vertical_4_dual_msa,
+ &vp9_lpf_vertical_4_dual_c, 8),
+ make_tuple(&vp9_lpf_vertical_8_dual_msa,
+ &vp9_lpf_vertical_8_dual_c, 8)));
+#endif // HAVE_MSA && (!CONFIG_VP9_HIGHBITDEPTH)
+
} // namespace
diff --git a/test/test-data.mk b/test/test-data.mk
index 1055e12..dda1c18 100644
--- a/test/test-data.mk
+++ b/test/test-data.mk
@@ -12,6 +12,7 @@
LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_12_422.y4m
LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_12_444.y4m
LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_12_440.yuv
+LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_8_420_a10-1.y4m
LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_8_420.y4m
LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_8_422.y4m
LIBVPX_TEST_DATA-$(CONFIG_ENCODERS) += park_joy_90p_8_444.y4m
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 0bb4481..3590f4e 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -22,6 +22,7 @@
c92825f1ea25c5c37855083a69faac6ac4641a9e *park_joy_90p_12_422.y4m
b592189b885b6cc85db55cc98512a197d73d3b34 *park_joy_90p_12_444.y4m
82c1bfcca368c2f22bad7d693d690d5499ecdd11 *park_joy_90p_12_440.yuv
+b9e1e90aece2be6e2c90d89e6ab2372d5f8c792d *park_joy_90p_8_420_a10-1.y4m
4e0eb61e76f0684188d9bc9f3ce61f6b6b77bb2c *park_joy_90p_8_420.y4m
7a193ff7dfeb96ba5f82b2afd7afa9e1fe83d947 *park_joy_90p_8_422.y4m
bdb7856e6bc93599bdda05c2e773a9f22b6c6d03 *park_joy_90p_8_444.y4m
diff --git a/test/test.mk b/test/test.mk
index 0dad062..4d63a63 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -66,6 +66,7 @@
LIBVPX_TEST_SRCS-$(CONFIG_DECODERS) += ../webmdec.cc
LIBVPX_TEST_SRCS-$(CONFIG_DECODERS) += ../webmdec.h
LIBVPX_TEST_SRCS-$(CONFIG_DECODERS) += webm_video_source.h
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_skip_loopfilter_test.cc
endif
LIBVPX_TEST_SRCS-$(CONFIG_DECODERS) += decode_api_test.cc
diff --git a/test/tools_common.sh b/test/tools_common.sh
index 0ae011e..60424ed 100755
--- a/test/tools_common.sh
+++ b/test/tools_common.sh
@@ -408,6 +408,8 @@
YUV_RAW_INPUT_WIDTH=352
YUV_RAW_INPUT_HEIGHT=288
+Y4M_NOSQ_PAR_INPUT="${LIBVPX_TEST_DATA_PATH}/park_joy_90p_8_420_a10-1.y4m"
+
# Setup a trap function to clean up after tests complete.
trap cleanup EXIT
@@ -429,6 +431,7 @@
VPX_TEST_VERBOSE_OUTPUT=${VPX_TEST_VERBOSE_OUTPUT}
YUV_RAW_INPUT=${YUV_RAW_INPUT}
YUV_RAW_INPUT_WIDTH=${YUV_RAW_INPUT_WIDTH}
- YUV_RAW_INPUT_HEIGHT=${YUV_RAW_INPUT_HEIGHT}"
+ YUV_RAW_INPUT_HEIGHT=${YUV_RAW_INPUT_HEIGHT}
+ Y4M_NOSQ_PAR_INPUT=${Y4M_NOSQ_PAR_INPUT}"
fi # End $VPX_TEST_TOOLS_COMMON_SH pseudo include guard.
diff --git a/test/variance_test.cc b/test/variance_test.cc
index 08a247d..2d17119 100644
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
@@ -1976,12 +1976,9 @@
#if HAVE_NEON_ASM
const SubpixVarMxNFunc vp8_subpel_variance16x16_neon =
vp8_sub_pixel_variance16x16_neon;
-const SubpixVarMxNFunc vp8_subpel_variance8x8_neon =
- vp8_sub_pixel_variance8x8_neon;
INSTANTIATE_TEST_CASE_P(
NEON, VP8SubpelVarianceTest,
- ::testing::Values(make_tuple(3, 3, vp8_subpel_variance8x8_neon, 0),
- make_tuple(4, 4, vp8_subpel_variance16x16_neon, 0)));
+ ::testing::Values(make_tuple(4, 4, vp8_subpel_variance16x16_neon, 0)));
#endif // HAVE_NEON_ASM
#endif // CONFIG_VP8
diff --git a/test/vp9_skip_loopfilter_test.cc b/test/vp9_skip_loopfilter_test.cc
new file mode 100644
index 0000000..b0cc7ba
--- /dev/null
+++ b/test/vp9_skip_loopfilter_test.cc
@@ -0,0 +1,180 @@
+/*
+ * Copyright (c) 2015 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/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#include "test/webm_video_source.h"
+
+namespace {
+
+const char kVp9TestFile[] = "vp90-2-08-tile_1x8_frame_parallel.webm";
+const char kVp9Md5File[] = "vp90-2-08-tile_1x8_frame_parallel.webm.md5";
+
+// Class for testing shutting off the loop filter.
+class SkipLoopFilterTest {
+ public:
+ SkipLoopFilterTest()
+ : video_(NULL),
+ decoder_(NULL),
+ md5_file_(NULL) {}
+
+ ~SkipLoopFilterTest() {
+ if (md5_file_ != NULL)
+ fclose(md5_file_);
+ delete decoder_;
+ delete video_;
+ }
+
+ // If |threads| > 0 then set the decoder with that number of threads.
+ void Init(int num_threads) {
+ expected_md5_[0] = '\0';
+ junk_[0] = '\0';
+ video_ = new libvpx_test::WebMVideoSource(kVp9TestFile);
+ ASSERT_TRUE(video_ != NULL);
+ video_->Init();
+ video_->Begin();
+
+ vpx_codec_dec_cfg_t cfg = vpx_codec_dec_cfg_t();
+ if (num_threads > 0)
+ cfg.threads = num_threads;
+ decoder_ = new libvpx_test::VP9Decoder(cfg, 0);
+ ASSERT_TRUE(decoder_ != NULL);
+
+ OpenMd5File(kVp9Md5File);
+ }
+
+ // Set the VP9 skipLoopFilter control value.
+ void SetSkipLoopFilter(int value, vpx_codec_err_t expected_value) {
+ decoder_->Control(VP9_SET_SKIP_LOOP_FILTER, value, expected_value);
+ }
+
+ vpx_codec_err_t DecodeOneFrame() {
+ const vpx_codec_err_t res =
+ decoder_->DecodeFrame(video_->cxdata(), video_->frame_size());
+ if (res == VPX_CODEC_OK) {
+ ReadMd5();
+ video_->Next();
+ }
+ return res;
+ }
+
+ vpx_codec_err_t DecodeRemainingFrames() {
+ for (; video_->cxdata() != NULL; video_->Next()) {
+ const vpx_codec_err_t res =
+ decoder_->DecodeFrame(video_->cxdata(), video_->frame_size());
+ if (res != VPX_CODEC_OK)
+ return res;
+ ReadMd5();
+ }
+ return VPX_CODEC_OK;
+ }
+
+ // Checks if MD5 matches or doesn't.
+ void CheckMd5(bool matches) {
+ libvpx_test::DxDataIterator dec_iter = decoder_->GetDxData();
+ const vpx_image_t *img = dec_iter.Next();
+ CheckMd5Vpx(*img, matches);
+ }
+
+ private:
+ // TODO(fgalligan): Move the MD5 testing code into another class.
+ void OpenMd5File(const std::string &md5_file_name) {
+ md5_file_ = libvpx_test::OpenTestDataFile(md5_file_name);
+ ASSERT_TRUE(md5_file_ != NULL) << "MD5 file open failed. Filename: "
+ << md5_file_name;
+ }
+
+ // Reads the next line of the MD5 file.
+ void ReadMd5() {
+ ASSERT_TRUE(md5_file_ != NULL);
+ const int res = fscanf(md5_file_, "%s %s", expected_md5_, junk_);
+ ASSERT_NE(EOF, res) << "Read md5 data failed";
+ expected_md5_[32] = '\0';
+ }
+
+ // Checks if the last read MD5 matches |img| or doesn't.
+ void CheckMd5Vpx(const vpx_image_t &img, bool matches) {
+ ::libvpx_test::MD5 md5_res;
+ md5_res.Add(&img);
+ const char *const actual_md5 = md5_res.Get();
+
+ // Check MD5.
+ if (matches)
+ ASSERT_STREQ(expected_md5_, actual_md5) << "MD5 checksums don't match";
+ else
+ ASSERT_STRNE(expected_md5_, actual_md5) << "MD5 checksums match";
+ }
+
+ libvpx_test::WebMVideoSource *video_;
+ libvpx_test::VP9Decoder *decoder_;
+ FILE *md5_file_;
+ char expected_md5_[33];
+ char junk_[128];
+};
+
+TEST(SkipLoopFilterTest, ShutOffLoopFilter) {
+ const int non_zero_value = 1;
+ const int num_threads = 0;
+ SkipLoopFilterTest skip_loop_filter;
+ skip_loop_filter.Init(num_threads);
+ skip_loop_filter.SetSkipLoopFilter(non_zero_value, VPX_CODEC_OK);
+ ASSERT_EQ(VPX_CODEC_OK, skip_loop_filter.DecodeRemainingFrames());
+ skip_loop_filter.CheckMd5(false);
+}
+
+TEST(SkipLoopFilterTest, ShutOffLoopFilterSingleThread) {
+ const int non_zero_value = 1;
+ const int num_threads = 1;
+ SkipLoopFilterTest skip_loop_filter;
+ skip_loop_filter.Init(num_threads);
+ skip_loop_filter.SetSkipLoopFilter(non_zero_value, VPX_CODEC_OK);
+ ASSERT_EQ(VPX_CODEC_OK, skip_loop_filter.DecodeRemainingFrames());
+ skip_loop_filter.CheckMd5(false);
+}
+
+TEST(SkipLoopFilterTest, ShutOffLoopFilter8Threads) {
+ const int non_zero_value = 1;
+ const int num_threads = 8;
+ SkipLoopFilterTest skip_loop_filter;
+ skip_loop_filter.Init(num_threads);
+ skip_loop_filter.SetSkipLoopFilter(non_zero_value, VPX_CODEC_OK);
+ ASSERT_EQ(VPX_CODEC_OK, skip_loop_filter.DecodeRemainingFrames());
+ skip_loop_filter.CheckMd5(false);
+}
+
+TEST(SkipLoopFilterTest, WithLoopFilter) {
+ const int non_zero_value = 1;
+ const int num_threads = 0;
+ SkipLoopFilterTest skip_loop_filter;
+ skip_loop_filter.Init(num_threads);
+ skip_loop_filter.SetSkipLoopFilter(non_zero_value, VPX_CODEC_OK);
+ skip_loop_filter.SetSkipLoopFilter(0, VPX_CODEC_OK);
+ ASSERT_EQ(VPX_CODEC_OK, skip_loop_filter.DecodeRemainingFrames());
+ skip_loop_filter.CheckMd5(true);
+}
+
+TEST(SkipLoopFilterTest, ToggleLoopFilter) {
+ const int num_threads = 0;
+ SkipLoopFilterTest skip_loop_filter;
+ skip_loop_filter.Init(num_threads);
+
+ for (int i = 0; i < 10; ++i) {
+ skip_loop_filter.SetSkipLoopFilter(i % 2, VPX_CODEC_OK);
+ ASSERT_EQ(VPX_CODEC_OK, skip_loop_filter.DecodeOneFrame());
+ }
+ ASSERT_EQ(VPX_CODEC_OK, skip_loop_filter.DecodeRemainingFrames());
+ skip_loop_filter.CheckMd5(false);
+}
+
+} // namespace
diff --git a/test/vpxenc.sh b/test/vpxenc.sh
index 9674bdc..1faa145 100755
--- a/test/vpxenc.sh
+++ b/test/vpxenc.sh
@@ -23,6 +23,13 @@
elog "The file ${YUV_RAW_INPUT##*/} must exist in LIBVPX_TEST_DATA_PATH."
return 1
fi
+ if [ "$(vpxenc_can_encode_vp9)" = "yes" ]; then
+ if [ ! -e "${Y4M_NOSQ_PAR_INPUT}" ]; then
+ elog "The file ${Y4M_NOSQ_PAR_INPUT##*/} must exist in"
+ elog "LIBVPX_TEST_DATA_PATH."
+ return 1
+ fi
+ fi
if [ -z "$(vpx_tool_path vpxenc)" ]; then
elog "vpxenc not found. It must exist in LIBVPX_BIN_PATH or its parent."
return 1
@@ -49,6 +56,10 @@
--height="${YUV_RAW_INPUT_HEIGHT}""
}
+y4m_input_non_square_par() {
+ echo ""${Y4M_NOSQ_PAR_INPUT}""
+}
+
# Echo default vpxenc real time encoding params. $1 is the codec, which defaults
# to vp8 if unspecified.
vpxenc_rt_params() {
@@ -320,6 +331,23 @@
fi
}
+# TODO(fgalligan): Test that DisplayWidth is different than video width.
+vpxenc_vp9_webm_non_square_par() {
+ if [ "$(vpxenc_can_encode_vp9)" = "yes" ] && \
+ [ "$(webm_io_available)" = "yes" ]; then
+ local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9_non_square_par.webm"
+ vpxenc $(y4m_input_non_square_par) \
+ --codec=vp9 \
+ --limit="${TEST_FRAMES}" \
+ --output="${output}"
+
+ if [ ! -e "${output}" ]; then
+ elog "Output file does not exist."
+ return 1
+ fi
+ fi
+}
+
vpxenc_tests="vpxenc_vp8_ivf
vpxenc_vp8_webm
vpxenc_vp8_webm_rt
@@ -332,6 +360,7 @@
vpxenc_vp9_webm_2pass
vpxenc_vp9_ivf_lossless
vpxenc_vp9_ivf_minq0_maxq0
- vpxenc_vp9_webm_lag10_frames20"
+ vpxenc_vp9_webm_lag10_frames20
+ vpxenc_vp9_webm_non_square_par"
run_tests vpxenc_verify_environment "${vpxenc_tests}"
diff --git a/tools_common.h b/tools_common.h
index aa7f025..adccec8 100644
--- a/tools_common.h
+++ b/tools_common.h
@@ -89,6 +89,7 @@
enum VideoFileType file_type;
uint32_t width;
uint32_t height;
+ struct VpxRational pixel_aspect_ratio;
vpx_img_fmt_t fmt;
vpx_bit_depth_t bit_depth;
int only_i420;
diff --git a/vp8/common/rtcd_defs.pl b/vp8/common/rtcd_defs.pl
index 245677f..fed2088 100644
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -245,10 +245,9 @@
$vp8_sub_pixel_variance4x4_sse2=vp8_sub_pixel_variance4x4_wmt;
add_proto qw/unsigned int vp8_sub_pixel_variance8x8/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
-specialize qw/vp8_sub_pixel_variance8x8 mmx sse2 media neon_asm/;
+specialize qw/vp8_sub_pixel_variance8x8 mmx sse2 media/;
$vp8_sub_pixel_variance8x8_sse2=vp8_sub_pixel_variance8x8_wmt;
$vp8_sub_pixel_variance8x8_media=vp8_sub_pixel_variance8x8_armv6;
-$vp8_sub_pixel_variance8x8_neon_asm=vp8_sub_pixel_variance8x8_neon;
add_proto qw/unsigned int vp8_sub_pixel_variance8x16/, "const unsigned char *src_ptr, int source_stride, int xoffset, int yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse";
specialize qw/vp8_sub_pixel_variance8x16 mmx sse2/;
diff --git a/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm b/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
index d4f6d9b..14f574a 100644
--- a/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
+++ b/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
@@ -298,8 +298,7 @@
|vp9_tm_predictor_4x4_neon| PROC
; Load ytop_left = above[-1];
sub r12, r2, #1
- ldrb r12, [r12]
- vdup.u8 d0, r12
+ vld1.u8 {d0[]}, [r12]
; Load above 4 pixels
vld1.32 {d2[0]}, [r2]
@@ -309,10 +308,10 @@
; Load left row by row and compute left + (above - ytop_left)
; 1st row and 2nd row
- ldrb r12, [r3], #1
- ldrb r2, [r3], #1
- vdup.u16 q1, r12
- vdup.u16 q2, r2
+ vld1.u8 {d2[]}, [r3]!
+ vld1.u8 {d4[]}, [r3]!
+ vmovl.u8 q1, d2
+ vmovl.u8 q2, d4
vadd.s16 q1, q1, q3
vadd.s16 q2, q2, q3
vqmovun.s16 d0, q1
@@ -321,10 +320,10 @@
vst1.32 {d1[0]}, [r0], r1
; 3rd row and 4th row
- ldrb r12, [r3], #1
- ldrb r2, [r3], #1
- vdup.u16 q1, r12
- vdup.u16 q2, r2
+ vld1.u8 {d2[]}, [r3]!
+ vld1.u8 {d4[]}, [r3]
+ vmovl.u8 q1, d2
+ vmovl.u8 q2, d4
vadd.s16 q1, q1, q3
vadd.s16 q2, q2, q3
vqmovun.s16 d0, q1
diff --git a/vp9/common/mips/msa/vp9_loopfilter_16_msa.c b/vp9/common/mips/msa/vp9_loopfilter_16_msa.c
new file mode 100644
index 0000000..aeaa48e
--- /dev/null
+++ b/vp9/common/mips/msa/vp9_loopfilter_16_msa.c
@@ -0,0 +1,1480 @@
+/*
+ * Copyright (c) 2015 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/mem.h"
+#include "vp9/common/mips/msa/vp9_loopfilter_msa.h"
+
+int32_t vp9_hz_lpf_t4_and_t8_16w(uint8_t *src, int32_t pitch,
+ uint8_t *filter48,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
+ v16u8 flat, mask, hev, thresh, b_limit, limit;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
+ v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
+ v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
+ v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
+ v16u8 zero = { 0 };
+
+ /* load vector elements */
+ LD_UB8(src - (4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ /* mask and hev */
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ if (__msa_test_bz_v(flat)) {
+ ST_UB4(p1_out, p0_out, q0_out, q1_out, (src - 2 * pitch), pitch);
+
+ return 1;
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r,
+ q2_r, q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
+ p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
+
+ ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
+ ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
+ VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
+ p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
+
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
+ p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
+ p0_filt8_r, q0_filt8_r);
+ PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
+ q2_filt8_r);
+
+ /* store pixel values */
+ p2_out = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
+ p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
+ p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
+ q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
+ q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
+ q2_out = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
+
+ ST_UB4(p2_out, p1_out, p0_out, q0_out, filter48, 16);
+ filter48 += (4 * 16);
+ ST_UB2(q1_out, q2_out, filter48, 16);
+ filter48 += (2 * 16);
+ ST_UB(flat, filter48);
+
+ return 0;
+ }
+}
+
+void vp9_hz_lpf_t16_16w(uint8_t *src, int32_t pitch, uint8_t *filter48) {
+ v16u8 flat, flat2, filter8;
+ v16i8 zero = { 0 };
+ v16u8 p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5, q6, q7;
+ v8u16 p7_r_in, p6_r_in, p5_r_in, p4_r_in, p3_r_in, p2_r_in, p1_r_in, p0_r_in;
+ v8u16 q7_r_in, q6_r_in, q5_r_in, q4_r_in, q3_r_in, q2_r_in, q1_r_in, q0_r_in;
+ v8u16 p7_l_in, p6_l_in, p5_l_in, p4_l_in, p3_l_in, p2_l_in, p1_l_in, p0_l_in;
+ v8u16 q7_l_in, q6_l_in, q5_l_in, q4_l_in, q3_l_in, q2_l_in, q1_l_in, q0_l_in;
+ v8u16 tmp0_r, tmp1_r, tmp0_l, tmp1_l;
+ v8i16 l_out, r_out;
+
+ flat = LD_UB(filter48 + 96);
+
+ LD_UB8((src - 8 * pitch), pitch, p7, p6, p5, p4, p3, p2, p1, p0);
+ LD_UB8(src, pitch, q0, q1, q2, q3, q4, q5, q6, q7);
+ VP9_FLAT5(p7, p6, p5, p4, p0, q0, q4, q5, q6, q7, flat, flat2);
+
+ if (__msa_test_bz_v(flat2)) {
+ LD_UB4(filter48, 16, p2, p1, p0, q0);
+ LD_UB2(filter48 + 4 * 16, 16, q1, q2);
+
+ src -= 3 * pitch;
+ ST_UB4(p2, p1, p0, q0, src, pitch);
+ src += (4 * pitch);
+ ST_UB2(q1, q2, src, pitch);
+ } else {
+ src -= 7 * pitch;
+
+ ILVR_B8_UH(zero, p7, zero, p6, zero, p5, zero, p4, zero, p3, zero, p2,
+ zero, p1, zero, p0, p7_r_in, p6_r_in, p5_r_in, p4_r_in, p3_r_in,
+ p2_r_in, p1_r_in, p0_r_in);
+
+ q0_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q0);
+
+ tmp0_r = p7_r_in << 3;
+ tmp0_r -= p7_r_in;
+ tmp0_r += p6_r_in;
+ tmp0_r += q0_r_in;
+ tmp1_r = p6_r_in + p5_r_in;
+ tmp1_r += p4_r_in;
+ tmp1_r += p3_r_in;
+ tmp1_r += p2_r_in;
+ tmp1_r += p1_r_in;
+ tmp1_r += p0_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ ILVL_B4_UH(zero, p7, zero, p6, zero, p5, zero, p4, p7_l_in, p6_l_in,
+ p5_l_in, p4_l_in);
+ ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l_in, p2_l_in,
+ p1_l_in, p0_l_in);
+ q0_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q0);
+
+ tmp0_l = p7_l_in << 3;
+ tmp0_l -= p7_l_in;
+ tmp0_l += p6_l_in;
+ tmp0_l += q0_l_in;
+ tmp1_l = p6_l_in + p5_l_in;
+ tmp1_l += p4_l_in;
+ tmp1_l += p3_l_in;
+ tmp1_l += p2_l_in;
+ tmp1_l += p1_l_in;
+ tmp1_l += p0_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p6 = __msa_bmnz_v(p6, (v16u8)r_out, flat2);
+ ST_UB(p6, src);
+ src += pitch;
+
+ /* p5 */
+ q1_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q1);
+ tmp0_r = p5_r_in - p6_r_in;
+ tmp0_r += q1_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ q1_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q1);
+ tmp0_l = p5_l_in - p6_l_in;
+ tmp0_l += q1_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p5 = __msa_bmnz_v(p5, (v16u8)r_out, flat2);
+ ST_UB(p5, src);
+ src += pitch;
+
+ /* p4 */
+ q2_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q2);
+ tmp0_r = p4_r_in - p5_r_in;
+ tmp0_r += q2_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = (v8i16)__msa_srari_h((v8i16)tmp1_r, 4);
+
+ q2_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q2);
+ tmp0_l = p4_l_in - p5_l_in;
+ tmp0_l += q2_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p4 = __msa_bmnz_v(p4, (v16u8)r_out, flat2);
+ ST_UB(p4, src);
+ src += pitch;
+
+ /* p3 */
+ q3_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q3);
+ tmp0_r = p3_r_in - p4_r_in;
+ tmp0_r += q3_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ q3_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q3);
+ tmp0_l = p3_l_in - p4_l_in;
+ tmp0_l += q3_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p3 = __msa_bmnz_v(p3, (v16u8)r_out, flat2);
+ ST_UB(p3, src);
+ src += pitch;
+
+ /* p2 */
+ q4_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q4);
+ filter8 = LD_UB(filter48);
+ tmp0_r = p2_r_in - p3_r_in;
+ tmp0_r += q4_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ q4_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q4);
+ tmp0_l = p2_l_in - p3_l_in;
+ tmp0_l += q4_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += pitch;
+
+ /* p1 */
+ q5_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q5);
+ filter8 = LD_UB(filter48 + 16);
+ tmp0_r = p1_r_in - p2_r_in;
+ tmp0_r += q5_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ q5_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q5);
+ tmp0_l = p1_l_in - p2_l_in;
+ tmp0_l += q5_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += pitch;
+
+ /* p0 */
+ q6_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q6);
+ filter8 = LD_UB(filter48 + 32);
+ tmp0_r = p0_r_in - p1_r_in;
+ tmp0_r += q6_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ q6_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q6);
+ tmp0_l = p0_l_in - p1_l_in;
+ tmp0_l += q6_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += pitch;
+
+ /* q0 */
+ q7_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q7);
+ filter8 = LD_UB(filter48 + 48);
+ tmp0_r = q7_r_in - p0_r_in;
+ tmp0_r += q0_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ q7_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q7);
+ tmp0_l = q7_l_in - p0_l_in;
+ tmp0_l += q0_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += pitch;
+
+ /* q1 */
+ filter8 = LD_UB(filter48 + 64);
+ tmp0_r = q7_r_in - q0_r_in;
+ tmp0_r += q1_r_in;
+ tmp0_r -= p6_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ tmp0_l = q7_l_in - q0_l_in;
+ tmp0_l += q1_l_in;
+ tmp0_l -= p6_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += pitch;
+
+ /* q2 */
+ filter8 = LD_UB(filter48 + 80);
+ tmp0_r = q7_r_in - q1_r_in;
+ tmp0_r += q2_r_in;
+ tmp0_r -= p5_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ tmp0_l = q7_l_in - q1_l_in;
+ tmp0_l += q2_l_in;
+ tmp0_l -= p5_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += pitch;
+
+ /* q3 */
+ tmp0_r = q7_r_in - q2_r_in;
+ tmp0_r += q3_r_in;
+ tmp0_r -= p4_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ tmp0_l = q7_l_in - q2_l_in;
+ tmp0_l += q3_l_in;
+ tmp0_l -= p4_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q3 = __msa_bmnz_v(q3, (v16u8)r_out, flat2);
+ ST_UB(q3, src);
+ src += pitch;
+
+ /* q4 */
+ tmp0_r = q7_r_in - q3_r_in;
+ tmp0_r += q4_r_in;
+ tmp0_r -= p3_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ tmp0_l = q7_l_in - q3_l_in;
+ tmp0_l += q4_l_in;
+ tmp0_l -= p3_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q4 = __msa_bmnz_v(q4, (v16u8)r_out, flat2);
+ ST_UB(q4, src);
+ src += pitch;
+
+ /* q5 */
+ tmp0_r = q7_r_in - q4_r_in;
+ tmp0_r += q5_r_in;
+ tmp0_r -= p2_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ tmp0_l = q7_l_in - q4_l_in;
+ tmp0_l += q5_l_in;
+ tmp0_l -= p2_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q5 = __msa_bmnz_v(q5, (v16u8)r_out, flat2);
+ ST_UB(q5, src);
+ src += pitch;
+
+ /* q6 */
+ tmp0_r = q7_r_in - q5_r_in;
+ tmp0_r += q6_r_in;
+ tmp0_r -= p1_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ tmp0_l = q7_l_in - q5_l_in;
+ tmp0_l += q6_l_in;
+ tmp0_l -= p1_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q6 = __msa_bmnz_v(q6, (v16u8)r_out, flat2);
+ ST_UB(q6, src);
+ }
+}
+
+void vp9_lpf_horizontal_16_dual_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr,
+ int32_t count) {
+ DECLARE_ALIGNED(32, uint8_t, filter48[16 * 8]);
+ uint8_t early_exit = 0;
+
+ (void)count;
+
+ early_exit = vp9_hz_lpf_t4_and_t8_16w(src, pitch, &filter48[0], b_limit_ptr,
+ limit_ptr, thresh_ptr);
+
+ if (0 == early_exit) {
+ vp9_hz_lpf_t16_16w(src, pitch, filter48);
+ }
+}
+
+void vp9_lpf_horizontal_16_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr,
+ int32_t count) {
+ if (1 == count) {
+ uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
+ uint64_t dword0, dword1;
+ v16u8 flat2, mask, hev, flat, thresh, b_limit, limit;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0, p7, p6, p5, p4, q4, q5, q6, q7;
+ v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
+ v16u8 p0_filter16, p1_filter16;
+ v8i16 p2_filter8, p1_filter8, p0_filter8;
+ v8i16 q0_filter8, q1_filter8, q2_filter8;
+ v8u16 p7_r, p6_r, p5_r, p4_r, q7_r, q6_r, q5_r, q4_r;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q3_r, q2_r, q1_r, q0_r;
+ v16i8 zero = { 0 };
+ v8u16 tmp0, tmp1, tmp2;
+
+ /* load vector elements */
+ LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out,
+ q1_out);
+
+ flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
+
+ if (__msa_test_bz_v(flat)) {
+ p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
+ p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
+ q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
+ q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
+ SD4(p1_d, p0_d, q0_d, q1_d, src - 2 * pitch, pitch);
+ } else {
+ /* convert 8 bit input data into 16 bit */
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
+ q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filter8,
+ p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8);
+
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(zero, p2_filter8, zero, p1_filter8, zero, p0_filter8,
+ zero, q0_filter8, p2_filter8, p1_filter8, p0_filter8,
+ q0_filter8);
+ PCKEV_B2_SH(zero, q1_filter8, zero, q2_filter8, q1_filter8, q2_filter8);
+
+ /* store pixel values */
+ p2_out = __msa_bmnz_v(p2, (v16u8)p2_filter8, flat);
+ p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filter8, flat);
+ p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filter8, flat);
+ q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filter8, flat);
+ q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filter8, flat);
+ q2_out = __msa_bmnz_v(q2, (v16u8)q2_filter8, flat);
+
+ /* load 16 vector elements */
+ LD_UB4((src - 8 * pitch), pitch, p7, p6, p5, p4);
+ LD_UB4(src + (4 * pitch), pitch, q4, q5, q6, q7);
+
+ VP9_FLAT5(p7, p6, p5, p4, p0, q0, q4, q5, q6, q7, flat, flat2);
+
+ if (__msa_test_bz_v(flat2)) {
+ p2_d = __msa_copy_u_d((v2i64)p2_out, 0);
+ p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
+ p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
+ q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
+ q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
+ q2_d = __msa_copy_u_d((v2i64)q2_out, 0);
+
+ SD4(p2_d, p1_d, p0_d, q0_d, src - 3 * pitch, pitch);
+ SD(q1_d, src + pitch);
+ SD(q2_d, src + 2 * pitch);
+ } else {
+ /* LSB(right) 8 pixel operation */
+ ILVR_B8_UH(zero, p7, zero, p6, zero, p5, zero, p4, zero, q4, zero, q5,
+ zero, q6, zero, q7, p7_r, p6_r, p5_r, p4_r, q4_r, q5_r, q6_r,
+ q7_r);
+
+ tmp0 = p7_r << 3;
+ tmp0 -= p7_r;
+ tmp0 += p6_r;
+ tmp0 += q0_r;
+
+ src -= 7 * pitch;
+
+ /* calculation of p6 and p5 */
+ tmp1 = p6_r + p5_r + p4_r + p3_r;
+ tmp1 += (p2_r + p1_r + p0_r);
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp0 = p5_r - p6_r + q1_r - p7_r;
+ tmp1 += tmp0;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(p6, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(p5, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ src += pitch;
+
+ /* calculation of p4 and p3 */
+ tmp0 = p4_r - p5_r + q2_r - p7_r;
+ tmp2 = p3_r - p4_r + q3_r - p7_r;
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp1 += tmp2;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(p4, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(p3, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ src += pitch;
+
+ /* calculation of p2 and p1 */
+ tmp0 = p2_r - p3_r + q4_r - p7_r;
+ tmp2 = p1_r - p2_r + q5_r - p7_r;
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp1 += tmp2;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(p2_out, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(p1_out, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ src += pitch;
+
+ /* calculation of p0 and q0 */
+ tmp0 = (p0_r - p1_r) + (q6_r - p7_r);
+ tmp2 = (q7_r - p0_r) + (q0_r - p7_r);
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp1 += tmp2;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(p0_out, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(q0_out, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ src += pitch;
+
+ /* calculation of q1 and q2 */
+ tmp0 = q7_r - q0_r + q1_r - p6_r;
+ tmp2 = q7_r - q1_r + q2_r - p5_r;
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp1 += tmp2;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(q1_out, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(q2_out, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ src += pitch;
+
+ /* calculation of q3 and q4 */
+ tmp0 = (q7_r - q2_r) + (q3_r - p4_r);
+ tmp2 = (q7_r - q3_r) + (q4_r - p3_r);
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp1 += tmp2;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(q3, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(q4, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ src += pitch;
+
+ /* calculation of q5 and q6 */
+ tmp0 = (q7_r - q4_r) + (q5_r - p2_r);
+ tmp2 = (q7_r - q5_r) + (q6_r - p1_r);
+ tmp1 += tmp0;
+ p0_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ tmp1 += tmp2;
+ p1_filter16 = (v16u8)__msa_srari_h((v8i16)tmp1, 4);
+ PCKEV_B2_UB(zero, p0_filter16, zero, p1_filter16, p0_filter16,
+ p1_filter16);
+ p0_filter16 = __msa_bmnz_v(q5, p0_filter16, flat2);
+ p1_filter16 = __msa_bmnz_v(q6, p1_filter16, flat2);
+ dword0 = __msa_copy_u_d((v2i64)p0_filter16, 0);
+ dword1 = __msa_copy_u_d((v2i64)p1_filter16, 0);
+ SD(dword0, src);
+ src += pitch;
+ SD(dword1, src);
+ }
+ }
+ } else {
+ vp9_lpf_horizontal_16_dual_msa(src, pitch, b_limit_ptr, limit_ptr,
+ thresh_ptr, count);
+ }
+}
+
+static void vp9_transpose_16x8_to_8x16(uint8_t *input, int32_t in_pitch,
+ uint8_t *output, int32_t out_pitch) {
+ v16u8 p7_org, p6_org, p5_org, p4_org, p3_org, p2_org, p1_org, p0_org;
+ v16i8 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ v16u8 p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5, q6, q7;
+
+ LD_UB8(input, in_pitch,
+ p7_org, p6_org, p5_org, p4_org, p3_org, p2_org, p1_org, p0_org);
+ /* 8x8 transpose */
+ TRANSPOSE8x8_UB_UB(p7_org, p6_org, p5_org, p4_org, p3_org, p2_org, p1_org,
+ p0_org, p7, p6, p5, p4, p3, p2, p1, p0);
+ /* 8x8 transpose */
+ ILVL_B4_SB(p5_org, p7_org, p4_org, p6_org, p1_org, p3_org, p0_org, p2_org,
+ tmp0, tmp1, tmp2, tmp3);
+ ILVR_B2_SB(tmp1, tmp0, tmp3, tmp2, tmp4, tmp6);
+ ILVL_B2_SB(tmp1, tmp0, tmp3, tmp2, tmp5, tmp7);
+ ILVR_W2_UB(tmp6, tmp4, tmp7, tmp5, q0, q4);
+ ILVL_W2_UB(tmp6, tmp4, tmp7, tmp5, q2, q6);
+ SLDI_B4_0_UB(q0, q2, q4, q6, q1, q3, q5, q7, 8);
+
+ ST_UB8(p7, p6, p5, p4, p3, p2, p1, p0, output, out_pitch);
+ output += (8 * out_pitch);
+ ST_UB8(q0, q1, q2, q3, q4, q5, q6, q7, output, out_pitch);
+}
+
+static void vp9_transpose_8x16_to_16x8(uint8_t *input, int32_t in_pitch,
+ uint8_t *output, int32_t out_pitch) {
+ v16u8 p7_o, p6_o, p5_o, p4_o, p3_o, p2_o, p1_o, p0_o;
+ v16u8 p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5, q6, q7;
+
+ LD_UB8(input, in_pitch, p7, p6, p5, p4, p3, p2, p1, p0);
+ LD_UB8(input + (8 * in_pitch), in_pitch, q0, q1, q2, q3, q4, q5, q6, q7);
+ TRANSPOSE16x8_UB_UB(p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5,
+ q6, q7, p7_o, p6_o, p5_o, p4_o, p3_o, p2_o, p1_o, p0_o);
+ ST_UB8(p7_o, p6_o, p5_o, p4_o, p3_o, p2_o, p1_o, p0_o, output, out_pitch);
+}
+
+static void vp9_transpose_16x16(uint8_t *input, int32_t in_pitch,
+ uint8_t *output, int32_t out_pitch) {
+ v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
+ v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
+ v16u8 p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5, q6, q7;
+ v8i16 tmp0, tmp1, tmp4, tmp5, tmp6, tmp7;
+ v4i32 tmp2, tmp3;
+
+ LD_UB8(input, in_pitch, row0, row1, row2, row3, row4, row5, row6, row7);
+ input += (8 * in_pitch);
+ LD_UB8(input, in_pitch,
+ row8, row9, row10, row11, row12, row13, row14, row15);
+
+ TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
+ row8, row9, row10, row11, row12, row13, row14, row15,
+ p7, p6, p5, p4, p3, p2, p1, p0);
+
+ /* transpose 16x8 matrix into 8x16 */
+ /* total 8 intermediate register and 32 instructions */
+ q7 = (v16u8)__msa_ilvod_d((v2i64)row8, (v2i64)row0);
+ q6 = (v16u8)__msa_ilvod_d((v2i64)row9, (v2i64)row1);
+ q5 = (v16u8)__msa_ilvod_d((v2i64)row10, (v2i64)row2);
+ q4 = (v16u8)__msa_ilvod_d((v2i64)row11, (v2i64)row3);
+ q3 = (v16u8)__msa_ilvod_d((v2i64)row12, (v2i64)row4);
+ q2 = (v16u8)__msa_ilvod_d((v2i64)row13, (v2i64)row5);
+ q1 = (v16u8)__msa_ilvod_d((v2i64)row14, (v2i64)row6);
+ q0 = (v16u8)__msa_ilvod_d((v2i64)row15, (v2i64)row7);
+
+ ILVEV_B2_SH(q7, q6, q5, q4, tmp0, tmp1);
+ tmp4 = (v8i16)__msa_ilvod_b((v16i8)q6, (v16i8)q7);
+ tmp5 = (v8i16)__msa_ilvod_b((v16i8)q4, (v16i8)q5);
+
+ ILVEV_B2_UB(q3, q2, q1, q0, q5, q7);
+ tmp6 = (v8i16)__msa_ilvod_b((v16i8)q2, (v16i8)q3);
+ tmp7 = (v8i16)__msa_ilvod_b((v16i8)q0, (v16i8)q1);
+
+ ILVEV_H2_SW(tmp0, tmp1, q5, q7, tmp2, tmp3);
+ q0 = (v16u8)__msa_ilvev_w(tmp3, tmp2);
+ q4 = (v16u8)__msa_ilvod_w(tmp3, tmp2);
+
+ tmp2 = (v4i32)__msa_ilvod_h(tmp1, tmp0);
+ tmp3 = (v4i32)__msa_ilvod_h((v8i16)q7, (v8i16)q5);
+ q2 = (v16u8)__msa_ilvev_w(tmp3, tmp2);
+ q6 = (v16u8)__msa_ilvod_w(tmp3, tmp2);
+
+ ILVEV_H2_SW(tmp4, tmp5, tmp6, tmp7, tmp2, tmp3);
+ q1 = (v16u8)__msa_ilvev_w(tmp3, tmp2);
+ q5 = (v16u8)__msa_ilvod_w(tmp3, tmp2);
+
+ tmp2 = (v4i32)__msa_ilvod_h(tmp5, tmp4);
+ tmp3 = (v4i32)__msa_ilvod_h(tmp7, tmp6);
+ q3 = (v16u8)__msa_ilvev_w(tmp3, tmp2);
+ q7 = (v16u8)__msa_ilvod_w(tmp3, tmp2);
+
+ ST_UB8(p7, p6, p5, p4, p3, p2, p1, p0, output, out_pitch);
+ output += (8 * out_pitch);
+ ST_UB8(q0, q1, q2, q3, q4, q5, q6, q7, output, out_pitch);
+}
+
+int32_t vp9_vt_lpf_t4_and_t8_8w(uint8_t *src, uint8_t *filter48,
+ uint8_t *src_org, int32_t pitch_org,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
+ v16u8 flat, mask, hev, thresh, b_limit, limit;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
+ v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
+ v16i8 zero = { 0 };
+ v8i16 vec0, vec1, vec2, vec3;
+
+ /* load vector elements */
+ LD_UB8(src - (4 * 16), 16, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ /* mask and hev */
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ /* flat4 */
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ /* filter4 */
+ VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
+
+ if (__msa_test_bz_v(flat)) {
+ ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec2, vec3);
+ ST4x8_UB(vec2, vec3, (src_org - 2), pitch_org);
+ return 1;
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
+ q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
+ p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
+
+ /* convert 16 bit output data into 8 bit */
+ p2_r = (v8u16)__msa_pckev_b((v16i8)p2_filt8_r, (v16i8)p2_filt8_r);
+ p1_r = (v8u16)__msa_pckev_b((v16i8)p1_filt8_r, (v16i8)p1_filt8_r);
+ p0_r = (v8u16)__msa_pckev_b((v16i8)p0_filt8_r, (v16i8)p0_filt8_r);
+ q0_r = (v8u16)__msa_pckev_b((v16i8)q0_filt8_r, (v16i8)q0_filt8_r);
+ q1_r = (v8u16)__msa_pckev_b((v16i8)q1_filt8_r, (v16i8)q1_filt8_r);
+ q2_r = (v8u16)__msa_pckev_b((v16i8)q2_filt8_r, (v16i8)q2_filt8_r);
+
+ /* store pixel values */
+ p2_out = __msa_bmnz_v(p2, (v16u8)p2_r, flat);
+ p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_r, flat);
+ p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_r, flat);
+ q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_r, flat);
+ q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_r, flat);
+ q2_out = __msa_bmnz_v(q2, (v16u8)q2_r, flat);
+
+ ST_UB4(p2_out, p1_out, p0_out, q0_out, filter48, 16);
+ filter48 += (4 * 16);
+ ST_UB2(q1_out, q2_out, filter48, 16);
+ filter48 += (2 * 16);
+ ST_UB(flat, filter48);
+
+ return 0;
+ }
+}
+
+int32_t vp9_vt_lpf_t16_8w(uint8_t *src, uint8_t *src_org, int32_t pitch,
+ uint8_t *filter48) {
+ v16i8 zero = { 0 };
+ v16u8 filter8, flat, flat2;
+ v16u8 p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5, q6, q7;
+ v8u16 p7_r_in, p6_r_in, p5_r_in, p4_r_in, p3_r_in, p2_r_in, p1_r_in, p0_r_in;
+ v8u16 q7_r_in, q6_r_in, q5_r_in, q4_r_in, q3_r_in, q2_r_in, q1_r_in, q0_r_in;
+ v8u16 tmp0_r, tmp1_r;
+ v8i16 r_out;
+
+ flat = LD_UB(filter48 + 6 * 16);
+
+ LD_UB8((src - 8 * 16), 16, p7, p6, p5, p4, p3, p2, p1, p0);
+ LD_UB8(src, 16, q0, q1, q2, q3, q4, q5, q6, q7);
+
+ VP9_FLAT5(p7, p6, p5, p4, p0, q0, q4, q5, q6, q7, flat, flat2);
+
+ if (__msa_test_bz_v(flat2)) {
+ v8i16 vec0, vec1, vec2, vec3, vec4;
+
+ LD_UB4(filter48, 16, p2, p1, p0, q0);
+ LD_UB2(filter48 + 4 * 16, 16, q1, q2);
+
+ ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec3, vec4);
+ vec2 = (v8i16)__msa_ilvr_b((v16i8)q2, (v16i8)q1);
+
+ src_org -= 3;
+ ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src_org, pitch);
+ ST2x4_UB(vec2, 0, (src_org + 4), pitch);
+ src_org += (4 * pitch);
+ ST4x4_UB(vec4, vec4, 0, 1, 2, 3, src_org, pitch);
+ ST2x4_UB(vec2, 4, (src_org + 4), pitch);
+
+ return 1;
+ } else {
+ src -= 7 * 16;
+
+ ILVR_B8_UH(zero, p7, zero, p6, zero, p5, zero, p4, zero, p3, zero, p2,
+ zero, p1, zero, p0, p7_r_in, p6_r_in, p5_r_in, p4_r_in,
+ p3_r_in, p2_r_in, p1_r_in, p0_r_in);
+ q0_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q0);
+
+ tmp0_r = p7_r_in << 3;
+ tmp0_r -= p7_r_in;
+ tmp0_r += p6_r_in;
+ tmp0_r += q0_r_in;
+ tmp1_r = p6_r_in + p5_r_in;
+ tmp1_r += p4_r_in;
+ tmp1_r += p3_r_in;
+ tmp1_r += p2_r_in;
+ tmp1_r += p1_r_in;
+ tmp1_r += p0_r_in;
+ tmp1_r += tmp0_r;
+
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ p6 = __msa_bmnz_v(p6, (v16u8)r_out, flat2);
+ ST8x1_UB(p6, src);
+ src += 16;
+
+ /* p5 */
+ q1_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q1);
+ tmp0_r = p5_r_in - p6_r_in;
+ tmp0_r += q1_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ p5 = __msa_bmnz_v(p5, (v16u8)r_out, flat2);
+ ST8x1_UB(p5, src);
+ src += 16;
+
+ /* p4 */
+ q2_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q2);
+ tmp0_r = p4_r_in - p5_r_in;
+ tmp0_r += q2_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ p4 = __msa_bmnz_v(p4, (v16u8)r_out, flat2);
+ ST8x1_UB(p4, src);
+ src += 16;
+
+ /* p3 */
+ q3_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q3);
+ tmp0_r = p3_r_in - p4_r_in;
+ tmp0_r += q3_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ p3 = __msa_bmnz_v(p3, (v16u8)r_out, flat2);
+ ST8x1_UB(p3, src);
+ src += 16;
+
+ /* p2 */
+ q4_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q4);
+ filter8 = LD_UB(filter48);
+ tmp0_r = p2_r_in - p3_r_in;
+ tmp0_r += q4_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST8x1_UB(filter8, src);
+ src += 16;
+
+ /* p1 */
+ q5_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q5);
+ filter8 = LD_UB(filter48 + 16);
+ tmp0_r = p1_r_in - p2_r_in;
+ tmp0_r += q5_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST8x1_UB(filter8, src);
+ src += 16;
+
+ /* p0 */
+ q6_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q6);
+ filter8 = LD_UB(filter48 + 32);
+ tmp0_r = p0_r_in - p1_r_in;
+ tmp0_r += q6_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST8x1_UB(filter8, src);
+ src += 16;
+
+ /* q0 */
+ q7_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q7);
+ filter8 = LD_UB(filter48 + 48);
+ tmp0_r = q7_r_in - p0_r_in;
+ tmp0_r += q0_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST8x1_UB(filter8, src);
+ src += 16;
+
+ /* q1 */
+ filter8 = LD_UB(filter48 + 64);
+ tmp0_r = q7_r_in - q0_r_in;
+ tmp0_r += q1_r_in;
+ tmp0_r -= p6_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST8x1_UB(filter8, src);
+ src += 16;
+
+ /* q2 */
+ filter8 = LD_UB(filter48 + 80);
+ tmp0_r = q7_r_in - q1_r_in;
+ tmp0_r += q2_r_in;
+ tmp0_r -= p5_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST8x1_UB(filter8, src);
+ src += 16;
+
+ /* q3 */
+ tmp0_r = q7_r_in - q2_r_in;
+ tmp0_r += q3_r_in;
+ tmp0_r -= p4_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ q3 = __msa_bmnz_v(q3, (v16u8)r_out, flat2);
+ ST8x1_UB(q3, src);
+ src += 16;
+
+ /* q4 */
+ tmp0_r = q7_r_in - q3_r_in;
+ tmp0_r += q4_r_in;
+ tmp0_r -= p3_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ q4 = __msa_bmnz_v(q4, (v16u8)r_out, flat2);
+ ST8x1_UB(q4, src);
+ src += 16;
+
+ /* q5 */
+ tmp0_r = q7_r_in - q4_r_in;
+ tmp0_r += q5_r_in;
+ tmp0_r -= p2_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ q5 = __msa_bmnz_v(q5, (v16u8)r_out, flat2);
+ ST8x1_UB(q5, src);
+ src += 16;
+
+ /* q6 */
+ tmp0_r = q7_r_in - q5_r_in;
+ tmp0_r += q6_r_in;
+ tmp0_r -= p1_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)r_out, (v16i8)r_out);
+ q6 = __msa_bmnz_v(q6, (v16u8)r_out, flat2);
+ ST8x1_UB(q6, src);
+
+ return 0;
+ }
+}
+
+void vp9_lpf_vertical_16_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ uint8_t early_exit = 0;
+ DECLARE_ALIGNED(32, uint8_t, transposed_input[16 * 24]);
+ uint8_t *filter48 = &transposed_input[16 * 16];
+
+ vp9_transpose_16x8_to_8x16(src - 8, pitch, transposed_input, 16);
+
+ early_exit = vp9_vt_lpf_t4_and_t8_8w((transposed_input + 16 * 8),
+ &filter48[0], src, pitch, b_limit_ptr,
+ limit_ptr, thresh_ptr);
+
+ if (0 == early_exit) {
+ early_exit = vp9_vt_lpf_t16_8w((transposed_input + 16 * 8), src, pitch,
+ &filter48[0]);
+
+ if (0 == early_exit) {
+ vp9_transpose_8x16_to_16x8(transposed_input, 16, src - 8, pitch);
+ }
+ }
+}
+
+int32_t vp9_vt_lpf_t4_and_t8_16w(uint8_t *src, uint8_t *filter48,
+ uint8_t *src_org, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
+ v16u8 flat, mask, hev, thresh, b_limit, limit;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
+ v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
+ v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
+ v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
+ v16i8 zero = { 0 };
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5;
+
+ /* load vector elements */
+ LD_UB8(src - (4 * 16), 16, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ /* mask and hev */
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ /* flat4 */
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ /* filter4 */
+ VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ if (__msa_test_bz_v(flat)) {
+ ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec2, vec3);
+ ILVL_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec4, vec5);
+
+ src_org -= 2;
+ ST4x8_UB(vec2, vec3, src_org, pitch);
+ src_org += 8 * pitch;
+ ST4x8_UB(vec4, vec5, src_org, pitch);
+
+ return 1;
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
+ q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
+ p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
+ ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
+ ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
+ VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
+ p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
+
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
+ p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
+ p0_filt8_r, q0_filt8_r);
+ PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
+ q2_filt8_r);
+
+ /* store pixel values */
+ p2_out = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
+ p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
+ p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
+ q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
+ q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
+ q2_out = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
+
+ ST_UB4(p2_out, p1_out, p0_out, q0_out, filter48, 16);
+ filter48 += (4 * 16);
+ ST_UB2(q1_out, q2_out, filter48, 16);
+ filter48 += (2 * 16);
+ ST_UB(flat, filter48);
+
+ return 0;
+ }
+}
+
+int32_t vp9_vt_lpf_t16_16w(uint8_t *src, uint8_t *src_org, int32_t pitch,
+ uint8_t *filter48) {
+ v16u8 flat, flat2, filter8;
+ v16i8 zero = { 0 };
+ v16u8 p7, p6, p5, p4, p3, p2, p1, p0, q0, q1, q2, q3, q4, q5, q6, q7;
+ v8u16 p7_r_in, p6_r_in, p5_r_in, p4_r_in, p3_r_in, p2_r_in, p1_r_in, p0_r_in;
+ v8u16 q7_r_in, q6_r_in, q5_r_in, q4_r_in, q3_r_in, q2_r_in, q1_r_in, q0_r_in;
+ v8u16 p7_l_in, p6_l_in, p5_l_in, p4_l_in, p3_l_in, p2_l_in, p1_l_in, p0_l_in;
+ v8u16 q7_l_in, q6_l_in, q5_l_in, q4_l_in, q3_l_in, q2_l_in, q1_l_in, q0_l_in;
+ v8u16 tmp0_r, tmp1_r, tmp0_l, tmp1_l;
+ v8i16 l_out, r_out;
+
+ flat = LD_UB(filter48 + 6 * 16);
+
+ LD_UB8((src - 8 * 16), 16, p7, p6, p5, p4, p3, p2, p1, p0);
+ LD_UB8(src, 16, q0, q1, q2, q3, q4, q5, q6, q7);
+
+ VP9_FLAT5(p7, p6, p5, p4, p0, q0, q4, q5, q6, q7, flat, flat2);
+
+ if (__msa_test_bz_v(flat2)) {
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+
+ LD_UB4(filter48, 16, p2, p1, p0, q0);
+ LD_UB2(filter48 + 4 * 16, 16, q1, q2);
+
+ ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec3, vec4);
+ ILVL_B2_SH(p1, p2, q0, p0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec6, vec7);
+ ILVRL_B2_SH(q2, q1, vec2, vec5);
+
+ src_org -= 3;
+ ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src_org, pitch);
+ ST2x4_UB(vec2, 0, (src_org + 4), pitch);
+ src_org += (4 * pitch);
+ ST4x4_UB(vec4, vec4, 0, 1, 2, 3, src_org, pitch);
+ ST2x4_UB(vec2, 4, (src_org + 4), pitch);
+ src_org += (4 * pitch);
+ ST4x4_UB(vec6, vec6, 0, 1, 2, 3, src_org, pitch);
+ ST2x4_UB(vec5, 0, (src_org + 4), pitch);
+ src_org += (4 * pitch);
+ ST4x4_UB(vec7, vec7, 0, 1, 2, 3, src_org, pitch);
+ ST2x4_UB(vec5, 4, (src_org + 4), pitch);
+
+ return 1;
+ } else {
+ src -= 7 * 16;
+
+ ILVR_B8_UH(zero, p7, zero, p6, zero, p5, zero, p4, zero, p3, zero, p2,
+ zero, p1, zero, p0, p7_r_in, p6_r_in, p5_r_in, p4_r_in,
+ p3_r_in, p2_r_in, p1_r_in, p0_r_in);
+ q0_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q0);
+
+ tmp0_r = p7_r_in << 3;
+ tmp0_r -= p7_r_in;
+ tmp0_r += p6_r_in;
+ tmp0_r += q0_r_in;
+ tmp1_r = p6_r_in + p5_r_in;
+ tmp1_r += p4_r_in;
+ tmp1_r += p3_r_in;
+ tmp1_r += p2_r_in;
+ tmp1_r += p1_r_in;
+ tmp1_r += p0_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+
+ ILVL_B4_UH(zero, p7, zero, p6, zero, p5, zero, p4, p7_l_in, p6_l_in,
+ p5_l_in, p4_l_in);
+ ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l_in, p2_l_in,
+ p1_l_in, p0_l_in);
+ q0_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q0);
+
+ tmp0_l = p7_l_in << 3;
+ tmp0_l -= p7_l_in;
+ tmp0_l += p6_l_in;
+ tmp0_l += q0_l_in;
+ tmp1_l = p6_l_in + p5_l_in;
+ tmp1_l += p4_l_in;
+ tmp1_l += p3_l_in;
+ tmp1_l += p2_l_in;
+ tmp1_l += p1_l_in;
+ tmp1_l += p0_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p6 = __msa_bmnz_v(p6, (v16u8)r_out, flat2);
+ ST_UB(p6, src);
+ src += 16;
+
+ /* p5 */
+ q1_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q1);
+ tmp0_r = p5_r_in - p6_r_in;
+ tmp0_r += q1_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q1_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q1);
+ tmp0_l = p5_l_in - p6_l_in;
+ tmp0_l += q1_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p5 = __msa_bmnz_v(p5, (v16u8)r_out, flat2);
+ ST_UB(p5, src);
+ src += 16;
+
+ /* p4 */
+ q2_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q2);
+ tmp0_r = p4_r_in - p5_r_in;
+ tmp0_r += q2_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q2_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q2);
+ tmp0_l = p4_l_in - p5_l_in;
+ tmp0_l += q2_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p4 = __msa_bmnz_v(p4, (v16u8)r_out, flat2);
+ ST_UB(p4, src);
+ src += 16;
+
+ /* p3 */
+ q3_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q3);
+ tmp0_r = p3_r_in - p4_r_in;
+ tmp0_r += q3_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q3_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q3);
+ tmp0_l = p3_l_in - p4_l_in;
+ tmp0_l += q3_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ p3 = __msa_bmnz_v(p3, (v16u8)r_out, flat2);
+ ST_UB(p3, src);
+ src += 16;
+
+ /* p2 */
+ q4_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q4);
+ filter8 = LD_UB(filter48);
+ tmp0_r = p2_r_in - p3_r_in;
+ tmp0_r += q4_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q4_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q4);
+ tmp0_l = p2_l_in - p3_l_in;
+ tmp0_l += q4_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += 16;
+
+ /* p1 */
+ q5_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q5);
+ filter8 = LD_UB(filter48 + 16);
+ tmp0_r = p1_r_in - p2_r_in;
+ tmp0_r += q5_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q5_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q5);
+ tmp0_l = p1_l_in - p2_l_in;
+ tmp0_l += q5_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)(tmp1_l), 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += 16;
+
+ /* p0 */
+ q6_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q6);
+ filter8 = LD_UB(filter48 + 32);
+ tmp0_r = p0_r_in - p1_r_in;
+ tmp0_r += q6_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q6_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q6);
+ tmp0_l = p0_l_in - p1_l_in;
+ tmp0_l += q6_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += 16;
+
+ /* q0 */
+ q7_r_in = (v8u16)__msa_ilvr_b(zero, (v16i8)q7);
+ filter8 = LD_UB(filter48 + 48);
+ tmp0_r = q7_r_in - p0_r_in;
+ tmp0_r += q0_r_in;
+ tmp0_r -= p7_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ q7_l_in = (v8u16)__msa_ilvl_b(zero, (v16i8)q7);
+ tmp0_l = q7_l_in - p0_l_in;
+ tmp0_l += q0_l_in;
+ tmp0_l -= p7_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += 16;
+
+ /* q1 */
+ filter8 = LD_UB(filter48 + 64);
+ tmp0_r = q7_r_in - q0_r_in;
+ tmp0_r += q1_r_in;
+ tmp0_r -= p6_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ tmp0_l = q7_l_in - q0_l_in;
+ tmp0_l += q1_l_in;
+ tmp0_l -= p6_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += 16;
+
+ /* q2 */
+ filter8 = LD_UB(filter48 + 80);
+ tmp0_r = q7_r_in - q1_r_in;
+ tmp0_r += q2_r_in;
+ tmp0_r -= p5_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ tmp0_l = q7_l_in - q1_l_in;
+ tmp0_l += q2_l_in;
+ tmp0_l -= p5_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ filter8 = __msa_bmnz_v(filter8, (v16u8)r_out, flat2);
+ ST_UB(filter8, src);
+ src += 16;
+
+ /* q3 */
+ tmp0_r = q7_r_in - q2_r_in;
+ tmp0_r += q3_r_in;
+ tmp0_r -= p4_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ tmp0_l = q7_l_in - q2_l_in;
+ tmp0_l += q3_l_in;
+ tmp0_l -= p4_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q3 = __msa_bmnz_v(q3, (v16u8)r_out, flat2);
+ ST_UB(q3, src);
+ src += 16;
+
+ /* q4 */
+ tmp0_r = q7_r_in - q3_r_in;
+ tmp0_r += q4_r_in;
+ tmp0_r -= p3_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ tmp0_l = q7_l_in - q3_l_in;
+ tmp0_l += q4_l_in;
+ tmp0_l -= p3_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q4 = __msa_bmnz_v(q4, (v16u8)r_out, flat2);
+ ST_UB(q4, src);
+ src += 16;
+
+ /* q5 */
+ tmp0_r = q7_r_in - q4_r_in;
+ tmp0_r += q5_r_in;
+ tmp0_r -= p2_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ tmp0_l = q7_l_in - q4_l_in;
+ tmp0_l += q5_l_in;
+ tmp0_l -= p2_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q5 = __msa_bmnz_v(q5, (v16u8)r_out, flat2);
+ ST_UB(q5, src);
+ src += 16;
+
+ /* q6 */
+ tmp0_r = q7_r_in - q5_r_in;
+ tmp0_r += q6_r_in;
+ tmp0_r -= p1_r_in;
+ tmp1_r += tmp0_r;
+ r_out = __msa_srari_h((v8i16)tmp1_r, 4);
+ tmp0_l = q7_l_in - q5_l_in;
+ tmp0_l += q6_l_in;
+ tmp0_l -= p1_l_in;
+ tmp1_l += tmp0_l;
+ l_out = __msa_srari_h((v8i16)tmp1_l, 4);
+ r_out = (v8i16)__msa_pckev_b((v16i8)l_out, (v16i8)r_out);
+ q6 = __msa_bmnz_v(q6, (v16u8)r_out, flat2);
+ ST_UB(q6, src);
+
+ return 0;
+ }
+}
+
+void vp9_lpf_vertical_16_dual_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr) {
+ uint8_t early_exit = 0;
+ DECLARE_ALIGNED(32, uint8_t, transposed_input[16 * 24]);
+ uint8_t *filter48 = &transposed_input[16 * 16];
+
+ vp9_transpose_16x16((src - 8), pitch, &transposed_input[0], 16);
+
+ early_exit = vp9_vt_lpf_t4_and_t8_16w((transposed_input + 16 * 8),
+ &filter48[0], src, pitch, b_limit_ptr,
+ limit_ptr, thresh_ptr);
+
+ if (0 == early_exit) {
+ early_exit = vp9_vt_lpf_t16_16w((transposed_input + 16 * 8), src, pitch,
+ &filter48[0]);
+
+ if (0 == early_exit) {
+ vp9_transpose_16x16(transposed_input, 16, (src - 8), pitch);
+ }
+ }
+}
diff --git a/vp9/common/mips/msa/vp9_loopfilter_4_msa.c b/vp9/common/mips/msa/vp9_loopfilter_4_msa.c
new file mode 100644
index 0000000..7f69135
--- /dev/null
+++ b/vp9/common/mips/msa/vp9_loopfilter_4_msa.c
@@ -0,0 +1,152 @@
+/*
+ * Copyright (c) 2015 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 "vp9/common/mips/msa/vp9_loopfilter_msa.h"
+
+void vp9_lpf_horizontal_4_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr,
+ int32_t count) {
+ uint64_t p1_d, p0_d, q0_d, q1_d;
+ v16u8 mask, hev, flat, thresh, b_limit, limit;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0, p1_out, p0_out, q0_out, q1_out;
+
+ (void)count;
+
+ /* load vector elements */
+ LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
+ p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
+ q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
+ q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
+ SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), pitch);
+}
+
+void vp9_lpf_horizontal_4_dual_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit0_ptr,
+ const uint8_t *limit0_ptr,
+ const uint8_t *thresh0_ptr,
+ const uint8_t *b_limit1_ptr,
+ const uint8_t *limit1_ptr,
+ const uint8_t *thresh1_ptr) {
+ v16u8 mask, hev, flat, thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+
+ /* load vector elements */
+ LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
+ thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
+ thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
+
+ b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
+ b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
+ b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
+
+ limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
+ limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
+ limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
+
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0,
+ hev, mask, flat);
+ VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
+
+ ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch);
+}
+
+void vp9_lpf_vertical_4_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr,
+ int32_t count) {
+ v16u8 mask, hev, flat, limit, thresh, b_limit;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v8i16 vec0, vec1, vec2, vec3;
+
+ (void)count;
+
+ LD_UB8((src - 4), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3,
+ p3, p2, p1, p0, q0, q1, q2, q3);
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
+ ILVR_B2_SH(p0, p1, q1, q0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec2, vec3);
+
+ src -= 2;
+ ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
+ src += 4 * pitch;
+ ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
+}
+
+void vp9_lpf_vertical_4_dual_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit0_ptr,
+ const uint8_t *limit0_ptr,
+ const uint8_t *thresh0_ptr,
+ const uint8_t *b_limit1_ptr,
+ const uint8_t *limit1_ptr,
+ const uint8_t *thresh1_ptr) {
+ v16u8 mask, hev, flat;
+ v16u8 thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
+ v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
+ v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+ LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
+ LD_UB8(src - 4 + (8 * pitch), pitch,
+ row8, row9, row10, row11, row12, row13, row14, row15);
+
+ TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
+ row8, row9, row10, row11, row12, row13, row14, row15,
+ p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
+ thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
+ thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
+
+ b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
+ b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
+ b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
+
+ limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
+ limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
+ limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
+
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0,
+ hev, mask, flat);
+ VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
+ ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
+ ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3);
+ ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
+ ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5);
+
+ src -= 2;
+
+ ST4x8_UB(tmp2, tmp3, src, pitch);
+ src += (8 * pitch);
+ ST4x8_UB(tmp4, tmp5, src, pitch);
+}
diff --git a/vp9/common/mips/msa/vp9_loopfilter_8_msa.c b/vp9/common/mips/msa/vp9_loopfilter_8_msa.c
new file mode 100644
index 0000000..26a858d
--- /dev/null
+++ b/vp9/common/mips/msa/vp9_loopfilter_8_msa.c
@@ -0,0 +1,348 @@
+/*
+ * Copyright (c) 2015 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 "vp9/common/mips/msa/vp9_loopfilter_msa.h"
+
+void vp9_lpf_horizontal_8_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr,
+ int32_t count) {
+ uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
+ v16u8 mask, hev, flat, thresh, b_limit, limit;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
+ v8i16 p2_filter8, p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q3_r, q2_r, q1_r, q0_r;
+ v16i8 zero = { 0 };
+
+ (void)count;
+
+ /* load vector elements */
+ LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
+
+ if (__msa_test_bz_v(flat)) {
+ p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
+ p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
+ q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
+ q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
+ SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), pitch);
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r,
+ q2_r, q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filter8,
+ p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8);
+
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(zero, p2_filter8, zero, p1_filter8, zero, p0_filter8,
+ zero, q0_filter8, p2_filter8, p1_filter8, p0_filter8,
+ q0_filter8);
+ PCKEV_B2_SH(zero, q1_filter8, zero, q2_filter8, q1_filter8, q2_filter8);
+
+ /* store pixel values */
+ p2_out = __msa_bmnz_v(p2, (v16u8)p2_filter8, flat);
+ p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filter8, flat);
+ p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filter8, flat);
+ q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filter8, flat);
+ q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filter8, flat);
+ q2_out = __msa_bmnz_v(q2, (v16u8)q2_filter8, flat);
+
+ p2_d = __msa_copy_u_d((v2i64)p2_out, 0);
+ p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
+ p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
+ q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
+ q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
+ q2_d = __msa_copy_u_d((v2i64)q2_out, 0);
+
+ src -= 3 * pitch;
+
+ SD4(p2_d, p1_d, p0_d, q0_d, src, pitch);
+ src += (4 * pitch);
+ SD(q1_d, src);
+ src += pitch;
+ SD(q2_d, src);
+ }
+}
+
+void vp9_lpf_horizontal_8_dual_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit0,
+ const uint8_t *limit0,
+ const uint8_t *thresh0,
+ const uint8_t *b_limit1,
+ const uint8_t *limit1,
+ const uint8_t *thresh1) {
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
+ v16u8 flat, mask, hev, tmp, thresh, b_limit, limit;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
+ v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
+ v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
+ v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
+ v16u8 zero = { 0 };
+
+ /* load vector elements */
+ LD_UB8(src - (4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh0);
+ tmp = (v16u8)__msa_fill_b(*thresh1);
+ thresh = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)thresh);
+
+ b_limit = (v16u8)__msa_fill_b(*b_limit0);
+ tmp = (v16u8)__msa_fill_b(*b_limit1);
+ b_limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)b_limit);
+
+ limit = (v16u8)__msa_fill_b(*limit0);
+ tmp = (v16u8)__msa_fill_b(*limit1);
+ limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)limit);
+
+ /* mask and hev */
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ if (__msa_test_bz_v(flat)) {
+ ST_UB4(p1_out, p0_out, q0_out, q1_out, (src - 2 * pitch), pitch);
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r,
+ q2_r, q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
+ p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
+
+ ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
+ ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
+ VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
+ p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
+
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
+ p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
+ p0_filt8_r, q0_filt8_r);
+ PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
+ q2_filt8_r);
+
+ /* store pixel values */
+ p2_out = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
+ p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
+ p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
+ q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
+ q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
+ q2_out = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
+
+ src -= 3 * pitch;
+
+ ST_UB4(p2_out, p1_out, p0_out, q0_out, src, pitch);
+ src += (4 * pitch);
+ ST_UB2(q1_out, q2_out, src, pitch);
+ src += (2 * pitch);
+ }
+}
+
+void vp9_lpf_vertical_8_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit_ptr,
+ const uint8_t *limit_ptr,
+ const uint8_t *thresh_ptr,
+ int32_t count) {
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p1_out, p0_out, q0_out, q1_out;
+ v16u8 flat, mask, hev, thresh, b_limit, limit;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
+ v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
+ v16u8 zero = { 0 };
+ v8i16 vec0, vec1, vec2, vec3, vec4;
+
+ (void)count;
+
+ /* load vector elements */
+ LD_UB8(src - 4, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
+
+ TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3,
+ p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh_ptr);
+ b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
+ limit = (v16u8)__msa_fill_b(*limit_ptr);
+
+ /* mask and hev */
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ /* flat4 */
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ /* filter4 */
+ VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
+
+ if (__msa_test_bz_v(flat)) {
+ /* Store 4 pixels p1-_q1 */
+ ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec2, vec3);
+
+ src -= 2;
+ ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
+ src += 4 * pitch;
+ ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
+ q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
+ p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(p2_filt8_r, p2_filt8_r, p1_filt8_r, p1_filt8_r, p0_filt8_r,
+ p0_filt8_r, q0_filt8_r, q0_filt8_r, p2_filt8_r, p1_filt8_r,
+ p0_filt8_r, q0_filt8_r);
+ PCKEV_B2_SH(q1_filt8_r, q1_filt8_r, q2_filt8_r, q2_filt8_r, q1_filt8_r,
+ q2_filt8_r);
+
+ /* store pixel values */
+ p2 = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
+ p1 = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
+ p0 = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
+ q0 = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
+ q1 = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
+ q2 = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
+
+ /* Store 6 pixels p2-_q2 */
+ ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec2, vec3);
+ vec4 = (v8i16)__msa_ilvr_b((v16i8)q2, (v16i8)q1);
+
+ src -= 3;
+ ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
+ ST2x4_UB(vec4, 0, src + 4, pitch);
+ src += (4 * pitch);
+ ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
+ ST2x4_UB(vec4, 4, src + 4, pitch);
+ }
+}
+
+void vp9_lpf_vertical_8_dual_msa(uint8_t *src, int32_t pitch,
+ const uint8_t *b_limit0,
+ const uint8_t *limit0,
+ const uint8_t *thresh0,
+ const uint8_t *b_limit1,
+ const uint8_t *limit1,
+ const uint8_t *thresh1) {
+ uint8_t *temp_src;
+ v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
+ v16u8 p1_out, p0_out, q0_out, q1_out;
+ v16u8 flat, mask, hev, thresh, b_limit, limit;
+ v16u8 row4, row5, row6, row7, row12, row13, row14, row15;
+ v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
+ v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
+ v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
+ v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
+ v16u8 zero = { 0 };
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+
+ temp_src = src - 4;
+
+ LD_UB8(temp_src, pitch, p0, p1, p2, p3, row4, row5, row6, row7);
+ temp_src += (8 * pitch);
+ LD_UB8(temp_src, pitch, q3, q2, q1, q0, row12, row13, row14, row15);
+
+ /* transpose 16x8 matrix into 8x16 */
+ TRANSPOSE16x8_UB_UB(p0, p1, p2, p3, row4, row5, row6, row7,
+ q3, q2, q1, q0, row12, row13, row14, row15,
+ p3, p2, p1, p0, q0, q1, q2, q3);
+
+ thresh = (v16u8)__msa_fill_b(*thresh0);
+ vec0 = (v8i16)__msa_fill_b(*thresh1);
+ thresh = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)thresh);
+
+ b_limit = (v16u8)__msa_fill_b(*b_limit0);
+ vec0 = (v8i16)__msa_fill_b(*b_limit1);
+ b_limit = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)b_limit);
+
+ limit = (v16u8)__msa_fill_b(*limit0);
+ vec0 = (v8i16)__msa_fill_b(*limit1);
+ limit = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)limit);
+
+ /* mask and hev */
+ LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
+ hev, mask, flat);
+ /* flat4 */
+ VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
+ /* filter4 */
+ VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
+
+ if (__msa_test_bz_v(flat)) {
+ ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec2, vec3);
+ ILVL_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec4, vec5);
+
+ src -= 2;
+ ST4x8_UB(vec2, vec3, src, pitch);
+ src += 8 * pitch;
+ ST4x8_UB(vec4, vec5, src, pitch);
+ } else {
+ ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
+ zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
+ q3_r);
+ VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
+ p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
+
+ ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
+ ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
+
+ /* filter8 */
+ VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
+ p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
+
+ /* convert 16 bit output data into 8 bit */
+ PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
+ p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
+ p0_filt8_r, q0_filt8_r);
+ PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
+ q2_filt8_r);
+
+ /* store pixel values */
+ p2 = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
+ p1 = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
+ p0 = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
+ q0 = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
+ q1 = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
+ q2 = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
+
+ ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec3, vec4);
+ ILVL_B2_SH(p1, p2, q0, p0, vec0, vec1);
+ ILVRL_H2_SH(vec1, vec0, vec6, vec7);
+ ILVRL_B2_SH(q2, q1, vec2, vec5);
+
+ src -= 3;
+ ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
+ ST2x4_UB(vec2, 0, src + 4, pitch);
+ src += (4 * pitch);
+ ST4x4_UB(vec4, vec4, 0, 1, 2, 3, src, pitch);
+ ST2x4_UB(vec2, 4, src + 4, pitch);
+ src += (4 * pitch);
+ ST4x4_UB(vec6, vec6, 0, 1, 2, 3, src, pitch);
+ ST2x4_UB(vec5, 0, src + 4, pitch);
+ src += (4 * pitch);
+ ST4x4_UB(vec7, vec7, 0, 1, 2, 3, src, pitch);
+ ST2x4_UB(vec5, 4, src + 4, pitch);
+ }
+}
diff --git a/vp9/common/mips/msa/vp9_loopfilter_msa.h b/vp9/common/mips/msa/vp9_loopfilter_msa.h
new file mode 100644
index 0000000..0643e41
--- /dev/null
+++ b/vp9/common/mips/msa/vp9_loopfilter_msa.h
@@ -0,0 +1,246 @@
+/*
+ * Copyright (c) 2015 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.
+ */
+
+#ifndef VP9_COMMON_MIPS_MSA_VP9_LOOPFILTER_MSA_H_
+#define VP9_COMMON_MIPS_MSA_VP9_LOOPFILTER_MSA_H_
+
+#include "vp9/common/mips/msa/vp9_macros_msa.h"
+
+#define VP9_LPF_FILTER4_8W(p1_in, p0_in, q0_in, q1_in, mask_in, hev_in, \
+ p1_out, p0_out, q0_out, q1_out) { \
+ v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt_sign; \
+ v16i8 filt, filt1, filt2, cnst4b, cnst3b; \
+ v8i16 q0_sub_p0_r, filt_r, cnst3h; \
+ \
+ p1_m = (v16i8)__msa_xori_b(p1_in, 0x80); \
+ p0_m = (v16i8)__msa_xori_b(p0_in, 0x80); \
+ q0_m = (v16i8)__msa_xori_b(q0_in, 0x80); \
+ q1_m = (v16i8)__msa_xori_b(q1_in, 0x80); \
+ \
+ filt = __msa_subs_s_b(p1_m, q1_m); \
+ filt = filt & (v16i8)hev_in; \
+ q0_sub_p0 = q0_m - p0_m; \
+ filt_sign = __msa_clti_s_b(filt, 0); \
+ \
+ cnst3h = __msa_ldi_h(3); \
+ q0_sub_p0_r = (v8i16)__msa_ilvr_b(q0_sub_p0, q0_sub_p0); \
+ q0_sub_p0_r = __msa_dotp_s_h((v16i8)q0_sub_p0_r, (v16i8)cnst3h); \
+ filt_r = (v8i16)__msa_ilvr_b(filt_sign, filt); \
+ filt_r += q0_sub_p0_r; \
+ filt_r = __msa_sat_s_h(filt_r, 7); \
+ \
+ /* combine left and right part */ \
+ filt = __msa_pckev_b((v16i8)filt_r, (v16i8)filt_r); \
+ \
+ filt = filt & (v16i8)mask_in; \
+ cnst4b = __msa_ldi_b(4); \
+ filt1 = __msa_adds_s_b(filt, cnst4b); \
+ filt1 >>= 3; \
+ \
+ cnst3b = __msa_ldi_b(3); \
+ filt2 = __msa_adds_s_b(filt, cnst3b); \
+ filt2 >>= 3; \
+ \
+ q0_m = __msa_subs_s_b(q0_m, filt1); \
+ q0_out = __msa_xori_b((v16u8)q0_m, 0x80); \
+ p0_m = __msa_adds_s_b(p0_m, filt2); \
+ p0_out = __msa_xori_b((v16u8)p0_m, 0x80); \
+ \
+ filt = __msa_srari_b(filt1, 1); \
+ hev_in = __msa_xori_b((v16u8)hev_in, 0xff); \
+ filt = filt & (v16i8)hev_in; \
+ \
+ q1_m = __msa_subs_s_b(q1_m, filt); \
+ q1_out = __msa_xori_b((v16u8)q1_m, 0x80); \
+ p1_m = __msa_adds_s_b(p1_m, filt); \
+ p1_out = __msa_xori_b((v16u8)p1_m, 0x80); \
+}
+
+#define VP9_LPF_FILTER4_4W(p1_in, p0_in, q0_in, q1_in, mask_in, hev_in, \
+ p1_out, p0_out, q0_out, q1_out) { \
+ v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt_sign; \
+ v16i8 filt, filt1, filt2, cnst4b, cnst3b; \
+ v8i16 q0_sub_p0_r, q0_sub_p0_l, filt_l, filt_r, cnst3h; \
+ \
+ p1_m = (v16i8)__msa_xori_b(p1_in, 0x80); \
+ p0_m = (v16i8)__msa_xori_b(p0_in, 0x80); \
+ q0_m = (v16i8)__msa_xori_b(q0_in, 0x80); \
+ q1_m = (v16i8)__msa_xori_b(q1_in, 0x80); \
+ \
+ filt = __msa_subs_s_b(p1_m, q1_m); \
+ \
+ filt = filt & (v16i8)hev_in; \
+ \
+ q0_sub_p0 = q0_m - p0_m; \
+ filt_sign = __msa_clti_s_b(filt, 0); \
+ \
+ cnst3h = __msa_ldi_h(3); \
+ q0_sub_p0_r = (v8i16)__msa_ilvr_b(q0_sub_p0, q0_sub_p0); \
+ q0_sub_p0_r = __msa_dotp_s_h((v16i8)q0_sub_p0_r, (v16i8)cnst3h); \
+ filt_r = (v8i16)__msa_ilvr_b(filt_sign, filt); \
+ filt_r += q0_sub_p0_r; \
+ filt_r = __msa_sat_s_h(filt_r, 7); \
+ \
+ q0_sub_p0_l = (v8i16)__msa_ilvl_b(q0_sub_p0, q0_sub_p0); \
+ q0_sub_p0_l = __msa_dotp_s_h((v16i8)q0_sub_p0_l, (v16i8)cnst3h); \
+ filt_l = (v8i16)__msa_ilvl_b(filt_sign, filt); \
+ filt_l += q0_sub_p0_l; \
+ filt_l = __msa_sat_s_h(filt_l, 7); \
+ \
+ filt = __msa_pckev_b((v16i8)filt_l, (v16i8)filt_r); \
+ filt = filt & (v16i8)mask_in; \
+ \
+ cnst4b = __msa_ldi_b(4); \
+ filt1 = __msa_adds_s_b(filt, cnst4b); \
+ filt1 >>= 3; \
+ \
+ cnst3b = __msa_ldi_b(3); \
+ filt2 = __msa_adds_s_b(filt, cnst3b); \
+ filt2 >>= 3; \
+ \
+ q0_m = __msa_subs_s_b(q0_m, filt1); \
+ q0_out = __msa_xori_b((v16u8)q0_m, 0x80); \
+ p0_m = __msa_adds_s_b(p0_m, filt2); \
+ p0_out = __msa_xori_b((v16u8)p0_m, 0x80); \
+ \
+ filt = __msa_srari_b(filt1, 1); \
+ hev_in = __msa_xori_b((v16u8)hev_in, 0xff); \
+ filt = filt & (v16i8)hev_in; \
+ \
+ q1_m = __msa_subs_s_b(q1_m, filt); \
+ q1_out = __msa_xori_b((v16u8)q1_m, 0x80); \
+ p1_m = __msa_adds_s_b(p1_m, filt); \
+ p1_out = __msa_xori_b((v16u8)p1_m, 0x80); \
+}
+
+#define VP9_FLAT4(p3_in, p2_in, p0_in, q0_in, q2_in, q3_in, flat_out) { \
+ v16u8 tmp, p2_a_sub_p0, q2_a_sub_q0, p3_a_sub_p0, q3_a_sub_q0; \
+ v16u8 zero_in = { 0 }; \
+ \
+ tmp = __msa_ori_b(zero_in, 1); \
+ p2_a_sub_p0 = __msa_asub_u_b(p2_in, p0_in); \
+ q2_a_sub_q0 = __msa_asub_u_b(q2_in, q0_in); \
+ p3_a_sub_p0 = __msa_asub_u_b(p3_in, p0_in); \
+ q3_a_sub_q0 = __msa_asub_u_b(q3_in, q0_in); \
+ \
+ p2_a_sub_p0 = __msa_max_u_b(p2_a_sub_p0, q2_a_sub_q0); \
+ flat_out = __msa_max_u_b(p2_a_sub_p0, flat_out); \
+ p3_a_sub_p0 = __msa_max_u_b(p3_a_sub_p0, q3_a_sub_q0); \
+ flat_out = __msa_max_u_b(p3_a_sub_p0, flat_out); \
+ \
+ flat_out = (tmp < (v16u8)flat_out); \
+ flat_out = __msa_xori_b(flat_out, 0xff); \
+ flat_out = flat_out & (mask); \
+}
+
+#define VP9_FLAT5(p7_in, p6_in, p5_in, p4_in, p0_in, q0_in, q4_in, \
+ q5_in, q6_in, q7_in, flat_in, flat2_out) { \
+ v16u8 tmp, zero_in = { 0 }; \
+ v16u8 p4_a_sub_p0, q4_a_sub_q0, p5_a_sub_p0, q5_a_sub_q0; \
+ v16u8 p6_a_sub_p0, q6_a_sub_q0, p7_a_sub_p0, q7_a_sub_q0; \
+ \
+ tmp = __msa_ori_b(zero_in, 1); \
+ p4_a_sub_p0 = __msa_asub_u_b(p4_in, p0_in); \
+ q4_a_sub_q0 = __msa_asub_u_b(q4_in, q0_in); \
+ p5_a_sub_p0 = __msa_asub_u_b(p5_in, p0_in); \
+ q5_a_sub_q0 = __msa_asub_u_b(q5_in, q0_in); \
+ p6_a_sub_p0 = __msa_asub_u_b(p6_in, p0_in); \
+ q6_a_sub_q0 = __msa_asub_u_b(q6_in, q0_in); \
+ p7_a_sub_p0 = __msa_asub_u_b(p7_in, p0_in); \
+ q7_a_sub_q0 = __msa_asub_u_b(q7_in, q0_in); \
+ \
+ p4_a_sub_p0 = __msa_max_u_b(p4_a_sub_p0, q4_a_sub_q0); \
+ flat2_out = __msa_max_u_b(p5_a_sub_p0, q5_a_sub_q0); \
+ flat2_out = __msa_max_u_b(p4_a_sub_p0, flat2_out); \
+ p6_a_sub_p0 = __msa_max_u_b(p6_a_sub_p0, q6_a_sub_q0); \
+ flat2_out = __msa_max_u_b(p6_a_sub_p0, flat2_out); \
+ p7_a_sub_p0 = __msa_max_u_b(p7_a_sub_p0, q7_a_sub_q0); \
+ flat2_out = __msa_max_u_b(p7_a_sub_p0, flat2_out); \
+ \
+ flat2_out = (tmp < (v16u8)flat2_out); \
+ flat2_out = __msa_xori_b(flat2_out, 0xff); \
+ flat2_out = flat2_out & flat_in; \
+}
+
+#define VP9_FILTER8(p3_in, p2_in, p1_in, p0_in, \
+ q0_in, q1_in, q2_in, q3_in, \
+ p2_filt8_out, p1_filt8_out, p0_filt8_out, \
+ q0_filt8_out, q1_filt8_out, q2_filt8_out) { \
+ v8u16 tmp0, tmp1, tmp2; \
+ \
+ tmp2 = p2_in + p1_in + p0_in; \
+ tmp0 = p3_in << 1; \
+ \
+ tmp0 = tmp0 + tmp2 + q0_in; \
+ tmp1 = tmp0 + p3_in + p2_in; \
+ p2_filt8_out = (v8i16)__msa_srari_h((v8i16)tmp1, 3); \
+ \
+ tmp1 = tmp0 + p1_in + q1_in; \
+ p1_filt8_out = (v8i16)__msa_srari_h((v8i16)tmp1, 3); \
+ \
+ tmp1 = q2_in + q1_in + q0_in; \
+ tmp2 = tmp2 + tmp1; \
+ tmp0 = tmp2 + (p0_in); \
+ tmp0 = tmp0 + (p3_in); \
+ p0_filt8_out = (v8i16)__msa_srari_h((v8i16)tmp0, 3); \
+ \
+ tmp0 = q2_in + q3_in; \
+ tmp0 = p0_in + tmp1 + tmp0; \
+ tmp1 = q3_in + q3_in; \
+ tmp1 = tmp1 + tmp0; \
+ q2_filt8_out = (v8i16)__msa_srari_h((v8i16)tmp1, 3); \
+ \
+ tmp0 = tmp2 + q3_in; \
+ tmp1 = tmp0 + q0_in; \
+ q0_filt8_out = (v8i16)__msa_srari_h((v8i16)tmp1, 3); \
+ \
+ tmp1 = tmp0 - p2_in; \
+ tmp0 = q1_in + q3_in; \
+ tmp1 = tmp0 + tmp1; \
+ q1_filt8_out = (v8i16)__msa_srari_h((v8i16)tmp1, 3); \
+}
+
+#define LPF_MASK_HEV(p3_in, p2_in, p1_in, p0_in, \
+ q0_in, q1_in, q2_in, q3_in, \
+ limit_in, b_limit_in, thresh_in, \
+ hev_out, mask_out, flat_out) { \
+ v16u8 p3_asub_p2_m, p2_asub_p1_m, p1_asub_p0_m, q1_asub_q0_m; \
+ v16u8 p1_asub_q1_m, p0_asub_q0_m, q3_asub_q2_m, q2_asub_q1_m; \
+ \
+ /* absolute subtraction of pixel values */ \
+ p3_asub_p2_m = __msa_asub_u_b(p3_in, p2_in); \
+ p2_asub_p1_m = __msa_asub_u_b(p2_in, p1_in); \
+ p1_asub_p0_m = __msa_asub_u_b(p1_in, p0_in); \
+ q1_asub_q0_m = __msa_asub_u_b(q1_in, q0_in); \
+ q2_asub_q1_m = __msa_asub_u_b(q2_in, q1_in); \
+ q3_asub_q2_m = __msa_asub_u_b(q3_in, q2_in); \
+ p0_asub_q0_m = __msa_asub_u_b(p0_in, q0_in); \
+ p1_asub_q1_m = __msa_asub_u_b(p1_in, q1_in); \
+ \
+ /* calculation of hev */ \
+ flat_out = __msa_max_u_b(p1_asub_p0_m, q1_asub_q0_m); \
+ hev_out = thresh_in < (v16u8)flat_out; \
+ \
+ /* calculation of mask */ \
+ p0_asub_q0_m = __msa_adds_u_b(p0_asub_q0_m, p0_asub_q0_m); \
+ p1_asub_q1_m >>= 1; \
+ p0_asub_q0_m = __msa_adds_u_b(p0_asub_q0_m, p1_asub_q1_m); \
+ \
+ mask_out = b_limit_in < p0_asub_q0_m; \
+ mask_out = __msa_max_u_b(flat_out, mask_out); \
+ p3_asub_p2_m = __msa_max_u_b(p3_asub_p2_m, p2_asub_p1_m); \
+ mask_out = __msa_max_u_b(p3_asub_p2_m, mask_out); \
+ q2_asub_q1_m = __msa_max_u_b(q2_asub_q1_m, q3_asub_q2_m); \
+ mask_out = __msa_max_u_b(q2_asub_q1_m, mask_out); \
+ \
+ mask_out = limit_in < (v16u8)mask_out; \
+ mask_out = __msa_xori_b(mask_out, 0xff); \
+}
+#endif /* VP9_COMMON_MIPS_MSA_VP9_LOOPFILTER_MSA_H_ */
diff --git a/vp9/common/mips/msa/vp9_macros_msa.h b/vp9/common/mips/msa/vp9_macros_msa.h
index 11dbd49..3751e35 100644
--- a/vp9/common/mips/msa/vp9_macros_msa.h
+++ b/vp9/common/mips/msa/vp9_macros_msa.h
@@ -413,6 +413,33 @@
}
#define ST_SH8(...) ST_H8(v8i16, __VA_ARGS__)
+/* Description : Store as 2x4 byte block to destination memory from input vector
+ Arguments : Inputs - in, stidx, pdst, stride
+ Return Type - unsigned byte
+ Details : Index stidx halfword element from 'in' vector is copied and
+ stored on first line
+ Index stidx+1 halfword element from 'in' vector is copied and
+ stored on second line
+ Index stidx+2 halfword element from 'in' vector is copied and
+ stored on third line
+ Index stidx+3 halfword element from 'in' vector is copied and
+ stored on fourth line
+*/
+#define ST2x4_UB(in, stidx, pdst, stride) { \
+ uint16_t out0_m, out1_m, out2_m, out3_m; \
+ uint8_t *pblk_2x4_m = (uint8_t *)(pdst); \
+ \
+ out0_m = __msa_copy_u_h((v8i16)in, (stidx)); \
+ out1_m = __msa_copy_u_h((v8i16)in, (stidx + 1)); \
+ out2_m = __msa_copy_u_h((v8i16)in, (stidx + 2)); \
+ out3_m = __msa_copy_u_h((v8i16)in, (stidx + 3)); \
+ \
+ SH(out0_m, pblk_2x4_m); \
+ SH(out1_m, pblk_2x4_m + stride); \
+ SH(out2_m, pblk_2x4_m + 2 * stride); \
+ SH(out3_m, pblk_2x4_m + 3 * stride); \
+}
+
/* Description : Store as 4x4 byte block to destination memory from input vector
Arguments : Inputs - in0, in1, pdst, stride
Return Type - unsigned byte
@@ -518,6 +545,13 @@
}
#define SLDI_B2_0_SW(...) SLDI_B2_0(v4i32, __VA_ARGS__)
+#define SLDI_B4_0(RTYPE, in0, in1, in2, in3, \
+ out0, out1, out2, out3, slide_val) { \
+ SLDI_B2_0(RTYPE, in0, in1, out0, out1, slide_val); \
+ SLDI_B2_0(RTYPE, in2, in3, out2, out3, slide_val); \
+}
+#define SLDI_B4_0_UB(...) SLDI_B4_0(v16u8, __VA_ARGS__)
+
/* Description : Immediate number of columns to slide
Arguments : Inputs - in0_0, in0_1, in1_0, in1_1, slide_val
Outputs - out0, out1
@@ -754,6 +788,21 @@
#define ILVEV_H2_SH(...) ILVEV_H2(v8i16, __VA_ARGS__)
#define ILVEV_H2_SW(...) ILVEV_H2(v4i32, __VA_ARGS__)
+/* Description : Interleave even double word elements from vectors
+ Arguments : Inputs - in0, in1, in2, in3
+ Outputs - out0, out1
+ Return Type - as per RTYPE
+ Details : Even double word elements of 'in0' and even double word
+ elements of 'in1' are interleaved and copied to 'out0'
+ Even double word elements of 'in2' and even double word
+ elements of 'in3' are interleaved and copied to 'out1'
+*/
+#define ILVEV_D2(RTYPE, in0, in1, in2, in3, out0, out1) { \
+ out0 = (RTYPE)__msa_ilvev_d((v2i64)in1, (v2i64)in0); \
+ out1 = (RTYPE)__msa_ilvev_d((v2i64)in3, (v2i64)in2); \
+}
+#define ILVEV_D2_UB(...) ILVEV_D2(v16u8, __VA_ARGS__)
+
/* Description : Interleave left half of byte elements from vectors
Arguments : Inputs - in0, in1, in2, in3
Outputs - out0, out1
@@ -840,6 +889,16 @@
#define ILVR_B4_UH(...) ILVR_B4(v8u16, __VA_ARGS__)
#define ILVR_B4_SH(...) ILVR_B4(v8i16, __VA_ARGS__)
+#define ILVR_B8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
+ in8, in9, in10, in11, in12, in13, in14, in15, \
+ out0, out1, out2, out3, out4, out5, out6, out7) { \
+ ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
+ out0, out1, out2, out3); \
+ ILVR_B4(RTYPE, in8, in9, in10, in11, in12, in13, in14, in15, \
+ out4, out5, out6, out7); \
+}
+#define ILVR_B8_UH(...) ILVR_B8(v8u16, __VA_ARGS__)
+
/* Description : Interleave right half of halfword elements from vectors
Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
Outputs - out0, out1, out2, out3
@@ -1301,6 +1360,78 @@
out7 = in0 - in7; \
}
+/* Description : Transposes input 8x8 byte block
+ Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
+ (input 8x8 byte block)
+ Outputs - out0, out1, out2, out3, out4, out5, out6, out7
+ (output 8x8 byte block)
+ Return Type - unsigned byte
+ Details :
+*/
+#define TRANSPOSE8x8_UB(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
+ out0, out1, out2, out3, out4, out5, out6, out7) { \
+ v16i8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
+ v16i8 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \
+ \
+ ILVR_B4_SB(in2, in0, in3, in1, in6, in4, in7, in5, \
+ tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
+ ILVRL_B2_SB(tmp1_m, tmp0_m, tmp4_m, tmp5_m); \
+ ILVRL_B2_SB(tmp3_m, tmp2_m, tmp6_m, tmp7_m); \
+ ILVRL_W2(RTYPE, tmp6_m, tmp4_m, out0, out2); \
+ ILVRL_W2(RTYPE, tmp7_m, tmp5_m, out4, out6); \
+ SLDI_B2_0(RTYPE, out0, out2, out1, out3, 8); \
+ SLDI_B2_0(RTYPE, out4, out6, out5, out7, 8); \
+}
+#define TRANSPOSE8x8_UB_UB(...) TRANSPOSE8x8_UB(v16u8, __VA_ARGS__)
+
+/* Description : Transposes 16x8 block into 8x16 with byte elements in vectors
+ Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15
+ Outputs - out0, out1, out2, out3, out4, out5, out6, out7
+ Return Type - unsigned byte
+ Details :
+*/
+#define TRANSPOSE16x8_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
+ in8, in9, in10, in11, in12, in13, in14, in15, \
+ out0, out1, out2, out3, out4, out5, out6, out7) { \
+ v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
+ v16u8 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \
+ \
+ ILVEV_D2_UB(in0, in8, in1, in9, out7, out6); \
+ ILVEV_D2_UB(in2, in10, in3, in11, out5, out4); \
+ ILVEV_D2_UB(in4, in12, in5, in13, out3, out2); \
+ ILVEV_D2_UB(in6, in14, in7, in15, out1, out0); \
+ \
+ tmp0_m = (v16u8)__msa_ilvev_b((v16i8)out6, (v16i8)out7); \
+ tmp4_m = (v16u8)__msa_ilvod_b((v16i8)out6, (v16i8)out7); \
+ tmp1_m = (v16u8)__msa_ilvev_b((v16i8)out4, (v16i8)out5); \
+ tmp5_m = (v16u8)__msa_ilvod_b((v16i8)out4, (v16i8)out5); \
+ out5 = (v16u8)__msa_ilvev_b((v16i8)out2, (v16i8)out3); \
+ tmp6_m = (v16u8)__msa_ilvod_b((v16i8)out2, (v16i8)out3); \
+ out7 = (v16u8)__msa_ilvev_b((v16i8)out0, (v16i8)out1); \
+ tmp7_m = (v16u8)__msa_ilvod_b((v16i8)out0, (v16i8)out1); \
+ \
+ ILVEV_H2_UB(tmp0_m, tmp1_m, out5, out7, tmp2_m, tmp3_m); \
+ out0 = (v16u8)__msa_ilvev_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ out4 = (v16u8)__msa_ilvod_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ \
+ tmp2_m = (v16u8)__msa_ilvod_h((v8i16)tmp1_m, (v8i16)tmp0_m); \
+ tmp3_m = (v16u8)__msa_ilvod_h((v8i16)out7, (v8i16)out5); \
+ out2 = (v16u8)__msa_ilvev_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ out6 = (v16u8)__msa_ilvod_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ \
+ ILVEV_H2_UB(tmp4_m, tmp5_m, tmp6_m, tmp7_m, tmp2_m, tmp3_m); \
+ out1 = (v16u8)__msa_ilvev_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ out5 = (v16u8)__msa_ilvod_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ \
+ tmp2_m = (v16u8)__msa_ilvod_h((v8i16)tmp5_m, (v8i16)tmp4_m); \
+ tmp2_m = (v16u8)__msa_ilvod_h((v8i16)tmp5_m, (v8i16)tmp4_m); \
+ tmp3_m = (v16u8)__msa_ilvod_h((v8i16)tmp7_m, (v8i16)tmp6_m); \
+ tmp3_m = (v16u8)__msa_ilvod_h((v8i16)tmp7_m, (v8i16)tmp6_m); \
+ out3 = (v16u8)__msa_ilvev_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+ out7 = (v16u8)__msa_ilvod_w((v4i32)tmp3_m, (v4i32)tmp2_m); \
+}
+
/* Description : Transposes 4x4 block with half word elements in vectors
Arguments : Inputs - in0, in1, in2, in3
Outputs - out0, out1, out2, out3
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index c7d5ee1..53ae921 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -129,9 +129,12 @@
int subsampling_y;
struct buf_2d dst;
struct buf_2d pre[2];
- const int16_t *dequant;
ENTROPY_CONTEXT *above_context;
ENTROPY_CONTEXT *left_context;
+ int16_t seg_dequant[MAX_SEGMENTS][2];
+
+ // encoder
+ const int16_t *dequant;
};
#define BLOCK_OFFSET(x, i) ((x) + (i) * 16)
@@ -164,6 +167,9 @@
int mb_to_top_edge;
int mb_to_bottom_edge;
+ FRAME_CONTEXT *fc;
+ int frame_parallel_decoding_mode;
+
/* pointers to reference frames */
RefBuffer *block_refs[2];
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index 188b03d..3af2a41 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -264,6 +264,7 @@
int log2_tile_cols, log2_tile_rows;
int byte_alignment;
+ int skip_loop_filter;
// Private data associated with the frame buffer callbacks.
void *cb_priv;
@@ -341,6 +342,14 @@
xd->plane[i].dqcoeff = xd->dqcoeff;
xd->above_context[i] = cm->above_context +
i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);
+
+ if (xd->plane[i].plane_type == PLANE_TYPE_Y) {
+ memcpy(xd->plane[i].seg_dequant, cm->y_dequant, sizeof(cm->y_dequant));
+ } else {
+ memcpy(xd->plane[i].seg_dequant, cm->uv_dequant, sizeof(cm->uv_dequant));
+ }
+ xd->fc = cm->fc;
+ xd->frame_parallel_decoding_mode = cm->frame_parallel_decoding_mode;
}
xd->above_seg_context = cm->above_seg_context;
diff --git a/vp9/common/vp9_reconintra.c b/vp9/common/vp9_reconintra.c
index f7c01fd..3312f29 100644
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -102,6 +102,10 @@
intra_pred_sized(type, 32)
#endif // CONFIG_VP9_HIGHBITDEPTH
+#define DST(x, y) dst[(x) + (y) * stride]
+#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
+#define AVG2(a, b) (((a) + (b) + 1) >> 1)
+
#if CONFIG_VP9_HIGHBITDEPTH
static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride,
int bs, const uint16_t *above,
@@ -112,18 +116,16 @@
// First column.
for (r = 0; r < bs - 1; ++r) {
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+ dst[r * stride] = AVG2(left[r], left[r + 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[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
}
- dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
- left[bs - 1] * 3, 2);
+ dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
dst[(bs - 1) * stride] = left[bs - 1];
dst++;
@@ -145,11 +147,9 @@
(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[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
+ above[(r >> 1) + c + 2])
+ : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
}
dst += stride;
}
@@ -163,9 +163,8 @@
(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)
+ dst[c] = r + c + 2 < bs * 2 ? AVG3(above[r + c], above[r + c + 1],
+ above[r + c + 2])
: above[bs * 2 - 1];
}
dst += stride;
@@ -180,20 +179,19 @@
// first row
for (c = 0; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+ dst[c] = AVG2(above[c - 1], above[c]);
dst += stride;
// second row
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
dst += stride;
// the rest of first col
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[0] = AVG3(above[-1], left[0], left[1]);
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);
+ dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
// the rest of the block
for (r = 2; r < bs; ++r) {
@@ -208,14 +206,13 @@
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);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
- dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
dst += stride;
for (r = 1; r < bs; ++r) {
@@ -230,20 +227,19 @@
const uint16_t *left, int bd) {
int r, c;
(void) bd;
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+ dst[0] = AVG2(above[-1], left[0]);
for (r = 1; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+ dst[r * stride] = AVG2(left[r - 1], left[r]);
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);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
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[c] = AVG3(above[c - 1], above[c], above[c + 1]);
dst += stride;
for (r = 1; r < bs; ++r) {
@@ -359,10 +355,6 @@
}
#endif // CONFIG_VP9_HIGHBITDEPTH
-#define DST(x, y) dst[(x) + (y) * stride]
-#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
-#define AVG2(a, b) (((a) + (b) + 1) >> 1)
-
void vp9_d207_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left) {
const int I = left[0];
@@ -386,16 +378,14 @@
(void) above;
// first column
for (r = 0; r < bs - 1; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+ dst[r * stride] = AVG2(left[r], left[r + 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[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
+ dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
dst[(bs - 1) * stride] = left[bs - 1];
dst++;
@@ -418,33 +408,34 @@
const int E = above[4];
const int F = above[5];
const int G = above[6];
- const int H = above[7];
(void)left;
DST(0, 0) = AVG2(A, B);
DST(1, 0) = DST(0, 2) = AVG2(B, C);
DST(2, 0) = DST(1, 2) = AVG2(C, D);
DST(3, 0) = DST(2, 2) = AVG2(D, E);
+ DST(3, 2) = AVG2(E, F); // differs from vp8
DST(0, 1) = AVG3(A, B, C);
DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
- DST(3, 2) = AVG3(E, F, G);
- DST(3, 3) = AVG3(F, G, H);
+ DST(3, 3) = AVG3(E, F, G); // differs from vp8
}
static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
- (void) left;
- 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;
+ int size;
+ (void)left;
+ for (c = 0; c < bs; ++c) {
+ dst[c] = AVG2(above[c], above[c + 1]);
+ dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]);
+ }
+ for (r = 2, size = bs - 2; r < bs; r += 2, --size) {
+ memcpy(dst + (r + 0) * stride, dst + (r >> 1), size);
+ memset(dst + (r + 0) * stride + size, above[bs - 1], bs - size);
+ memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1), size);
+ memset(dst + (r + 1) * stride + size, above[bs - 1], bs - size);
}
}
intra_pred_no_4x4(d63)
@@ -467,7 +458,7 @@
DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
- DST(3, 3) = AVG3(G, H, H);
+ DST(3, 3) = H; // differs from vp8
}
static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
@@ -517,20 +508,19 @@
// first row
for (c = 0; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+ dst[c] = AVG2(above[c - 1], above[c]);
dst += stride;
// second row
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
dst += stride;
// the rest of first col
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[0] = AVG3(above[-1], left[0], left[1]);
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);
+ dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
// the rest of the block
for (r = 2; r < bs; ++r) {
@@ -565,14 +555,13 @@
static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
- dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
for (c = 1; c < bs; c++)
- dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
- dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; ++r)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
dst += stride;
for (r = 1; r < bs; ++r) {
@@ -610,20 +599,19 @@
static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
- dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+ dst[0] = AVG2(above[-1], left[0]);
for (r = 1; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+ dst[r * stride] = AVG2(left[r - 1], left[r]);
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);
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
for (r = 2; r < bs; r++)
- dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
- left[r], 2);
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
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[c] = AVG3(above[c - 1], above[c], above[c + 1]);
dst += stride;
for (r = 1; r < bs; ++r) {
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 65de9d2..bbe200d 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -216,44 +216,44 @@
# Loopfilter
#
add_proto qw/void vp9_lpf_vertical_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
-specialize qw/vp9_lpf_vertical_16 sse2 neon_asm dspr2/;
+specialize qw/vp9_lpf_vertical_16 sse2 neon_asm dspr2 msa/;
$vp9_lpf_vertical_16_neon_asm=vp9_lpf_vertical_16_neon;
add_proto qw/void vp9_lpf_vertical_16_dual/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
-specialize qw/vp9_lpf_vertical_16_dual sse2 neon_asm dspr2/;
+specialize qw/vp9_lpf_vertical_16_dual sse2 neon_asm dspr2 msa/;
$vp9_lpf_vertical_16_dual_neon_asm=vp9_lpf_vertical_16_dual_neon;
add_proto qw/void vp9_lpf_vertical_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/vp9_lpf_vertical_8 sse2 neon_asm dspr2/;
+specialize qw/vp9_lpf_vertical_8 sse2 neon_asm dspr2 msa/;
$vp9_lpf_vertical_8_neon_asm=vp9_lpf_vertical_8_neon;
add_proto qw/void vp9_lpf_vertical_8_dual/, "uint8_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";
-specialize qw/vp9_lpf_vertical_8_dual sse2 neon_asm dspr2/;
+specialize qw/vp9_lpf_vertical_8_dual sse2 neon_asm dspr2 msa/;
$vp9_lpf_vertical_8_dual_neon_asm=vp9_lpf_vertical_8_dual_neon;
add_proto qw/void vp9_lpf_vertical_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/vp9_lpf_vertical_4 mmx neon dspr2/;
+specialize qw/vp9_lpf_vertical_4 mmx neon dspr2 msa/;
add_proto qw/void vp9_lpf_vertical_4_dual/, "uint8_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";
-specialize qw/vp9_lpf_vertical_4_dual sse2 neon dspr2/;
+specialize qw/vp9_lpf_vertical_4_dual sse2 neon dspr2 msa/;
add_proto qw/void vp9_lpf_horizontal_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/vp9_lpf_horizontal_16 sse2 avx2 neon_asm dspr2/;
+specialize qw/vp9_lpf_horizontal_16 sse2 avx2 neon_asm dspr2 msa/;
$vp9_lpf_horizontal_16_neon_asm=vp9_lpf_horizontal_16_neon;
add_proto qw/void vp9_lpf_horizontal_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/vp9_lpf_horizontal_8 sse2 neon_asm dspr2/;
+specialize qw/vp9_lpf_horizontal_8 sse2 neon_asm dspr2 msa/;
$vp9_lpf_horizontal_8_neon_asm=vp9_lpf_horizontal_8_neon;
add_proto qw/void vp9_lpf_horizontal_8_dual/, "uint8_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";
-specialize qw/vp9_lpf_horizontal_8_dual sse2 neon_asm dspr2/;
+specialize qw/vp9_lpf_horizontal_8_dual sse2 neon_asm dspr2 msa/;
$vp9_lpf_horizontal_8_dual_neon_asm=vp9_lpf_horizontal_8_dual_neon;
add_proto qw/void vp9_lpf_horizontal_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
-specialize qw/vp9_lpf_horizontal_4 mmx neon dspr2/;
+specialize qw/vp9_lpf_horizontal_4 mmx neon dspr2 msa/;
add_proto qw/void vp9_lpf_horizontal_4_dual/, "uint8_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";
-specialize qw/vp9_lpf_horizontal_4_dual sse2 neon dspr2/;
+specialize qw/vp9_lpf_horizontal_4_dual sse2 neon dspr2 msa/;
#
# post proc
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index d7f8b6f..30ca2d0 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -290,7 +290,6 @@
}
struct intra_args {
- VP9_COMMON *cm;
MACROBLOCKD *xd;
vp9_reader *r;
int seg_id;
@@ -300,7 +299,6 @@
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct intra_args *const args = (struct intra_args *)arg;
- 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];
@@ -317,7 +315,7 @@
x, y, plane);
if (!mi->mbmi.skip) {
- const int eob = vp9_decode_block_tokens(cm, xd, plane, block,
+ const int eob = vp9_decode_block_tokens(xd, plane, block,
plane_bsize, x, y, tx_size,
args->r, args->seg_id);
inverse_transform_block(xd, plane, block, tx_size, dst, pd->dst.stride,
@@ -326,7 +324,6 @@
}
struct inter_args {
- VP9_COMMON *cm;
MACROBLOCKD *xd;
vp9_reader *r;
int *eobtotal;
@@ -337,12 +334,11 @@
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct inter_args *args = (struct inter_args *)arg;
- VP9_COMMON *const cm = args->cm;
MACROBLOCKD *const xd = args->xd;
struct macroblockd_plane *const pd = &xd->plane[plane];
int x, y, eob;
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
- eob = vp9_decode_block_tokens(cm, xd, plane, block, plane_bsize,
+ eob = vp9_decode_block_tokens(xd, plane, block, plane_bsize,
x, y, tx_size, args->r, args->seg_id);
inverse_transform_block(xd, plane, block, tx_size,
&pd->dst.buf[4 * y * pd->dst.stride + 4 * x],
@@ -385,6 +381,15 @@
VP9_COMMON *const cm = &pbi->common;
const int less8x8 = bsize < BLOCK_8X8;
MB_MODE_INFO *mbmi = set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
+
+ if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) {
+ const BLOCK_SIZE uv_subsize =
+ ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y];
+ if (uv_subsize == BLOCK_INVALID)
+ vpx_internal_error(xd->error_info,
+ VPX_CODEC_CORRUPT_FRAME, "Invalid block size.");
+ }
+
vp9_read_mode_info(pbi, xd, tile, mi_row, mi_col, r);
if (less8x8)
@@ -395,7 +400,7 @@
}
if (!is_inter_block(mbmi)) {
- struct intra_args arg = {cm, xd, r, mbmi->segment_id};
+ struct intra_args arg = {xd, r, mbmi->segment_id};
vp9_foreach_transformed_block(xd, bsize,
predict_and_reconstruct_intra_block, &arg);
} else {
@@ -405,7 +410,7 @@
// Reconstruction
if (!mbmi->skip) {
int eobtotal = 0;
- struct inter_args arg = {cm, xd, r, &eobtotal, mbmi->segment_id};
+ struct inter_args arg = {xd, r, &eobtotal, mbmi->segment_id};
vp9_foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
if (!less8x8 && eobtotal == 0)
mbmi->skip = 1; // skip loopfilter
@@ -448,18 +453,14 @@
VP9_COMMON *const cm = &pbi->common;
const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
PARTITION_TYPE partition;
- BLOCK_SIZE subsize, uv_subsize;
+ BLOCK_SIZE subsize;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
return;
partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r);
subsize = get_subsize(bsize, partition);
- uv_subsize = ss_size_lookup[subsize][cm->subsampling_x][cm->subsampling_y];
- if (subsize >= BLOCK_8X8 && uv_subsize == BLOCK_INVALID)
- vpx_internal_error(xd->error_info,
- VPX_CODEC_CORRUPT_FRAME, "Invalid block size.");
- if (subsize < BLOCK_8X8) {
+ if (bsize == BLOCK_8X8) {
decode_block(pbi, xd, tile, mi_row, mi_col, r, subsize);
} else {
switch (partition) {
@@ -925,7 +926,8 @@
int mi_row, mi_col;
TileData *tile_data = NULL;
- if (cm->lf.filter_level && pbi->lf_worker.data1 == NULL) {
+ if (cm->lf.filter_level && !cm->skip_loop_filter &&
+ pbi->lf_worker.data1 == NULL) {
CHECK_MEM_ERROR(cm, pbi->lf_worker.data1,
vpx_memalign(32, sizeof(LFWorkerData)));
pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker;
@@ -935,7 +937,7 @@
}
}
- if (cm->lf.filter_level) {
+ if (cm->lf.filter_level && !cm->skip_loop_filter) {
LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
// Be sure to sync as we might be resuming after a failed frame decode.
winterface->sync(&pbi->lf_worker);
@@ -1008,7 +1010,7 @@
"Failed to decode tile data");
}
// Loopfilter one row.
- if (cm->lf.filter_level) {
+ if (cm->lf.filter_level && !cm->skip_loop_filter) {
const int lf_start = mi_row - MI_BLOCK_SIZE;
LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
@@ -1037,7 +1039,7 @@
}
// Loopfilter remaining rows in the frame.
- if (cm->lf.filter_level) {
+ if (cm->lf.filter_level && !cm->skip_loop_filter) {
LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
winterface->sync(&pbi->lf_worker);
lf_data->start = lf_data->stop;
@@ -1661,7 +1663,7 @@
vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
"Decode failed. Frame data header is corrupted.");
- if (cm->lf.filter_level) {
+ if (cm->lf.filter_level && !cm->skip_loop_filter) {
vp9_loop_filter_frame_init(cm, cm->lf.filter_level);
}
@@ -1687,11 +1689,13 @@
// Multi-threaded tile decoder
*p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end);
if (!xd->corrupted) {
- // If multiple threads are used to decode tiles, then we use those threads
- // to do parallel loopfiltering.
- vp9_loop_filter_frame_mt(new_fb, cm, pbi->mb.plane, cm->lf.filter_level,
- 0, 0, pbi->tile_workers, pbi->num_tile_workers,
- &pbi->lf_row_sync);
+ if (!cm->skip_loop_filter) {
+ // If multiple threads are used to decode tiles, then we use those
+ // threads to do parallel loopfiltering.
+ vp9_loop_filter_frame_mt(new_fb, cm, pbi->mb.plane,
+ cm->lf.filter_level, 0, 0, pbi->tile_workers,
+ pbi->num_tile_workers, &pbi->lf_row_sync);
+ }
} else {
vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
"Decode failed. Frame data is corrupted.");
diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index 9395a8b..3304e64 100644
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -46,23 +46,21 @@
return val;
}
-static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd,
+static int decode_coefs(const MACROBLOCKD *xd,
PLANE_TYPE type,
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) {
FRAME_COUNTS *counts = xd->counts;
const int max_eob = 16 << (tx_size << 1);
- const FRAME_CONTEXT *const fc = cm->fc;
+ const FRAME_CONTEXT *const fc = xd->fc;
const int ref = is_inter_block(&xd->mi[0]->mbmi);
int band, c = 0;
const vp9_prob (*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] =
fc->coef_probs[tx_size][type][ref];
const vp9_prob *prob;
- unsigned int (*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1] =
- counts->coef[tx_size][type][ref];
- unsigned int (*eob_branch_count)[COEFF_CONTEXTS] =
- counts->eob_branch[tx_size][type][ref];
+ unsigned int (*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1];
+ unsigned int (*eob_branch_count)[COEFF_CONTEXTS];
uint8_t token_cache[32 * 32];
const uint8_t *band_translate = get_band_translate(tx_size);
const int dq_shift = (tx_size == TX_32X32);
@@ -75,9 +73,14 @@
const uint8_t *cat5_prob;
const uint8_t *cat6_prob;
+ if (counts) {
+ coef_counts = counts->coef[tx_size][type][ref];
+ eob_branch_count = counts->eob_branch[tx_size][type][ref];
+ }
+
#if CONFIG_VP9_HIGHBITDEPTH
- if (cm->use_highbitdepth) {
- if (cm->bit_depth == VPX_BITS_10) {
+ if (xd->bd > VPX_BITS_8) {
+ if (xd->bd == VPX_BITS_10) {
cat1_prob = vp9_cat1_prob_high10;
cat2_prob = vp9_cat2_prob_high10;
cat3_prob = vp9_cat3_prob_high10;
@@ -163,7 +166,7 @@
break;
case CATEGORY6_TOKEN:
#if CONFIG_VP9_HIGHBITDEPTH
- switch (cm->bit_depth) {
+ switch (xd->bd) {
case VPX_BITS_8:
val = CAT6_MIN_VAL + read_coeff(cat6_prob, 14, r);
break;
@@ -187,7 +190,7 @@
#if CONFIG_COEFFICIENT_RANGE_CHECKING
#if CONFIG_VP9_HIGHBITDEPTH
dqcoeff[scan[c]] = highbd_check_range((vp9_read_bit(r) ? -v : v),
- cm->bit_depth);
+ xd->bd);
#else
dqcoeff[scan[c]] = check_range(vp9_read_bit(r) ? -v : v);
#endif // CONFIG_VP9_HIGHBITDEPTH
@@ -203,18 +206,17 @@
return c;
}
-int vp9_decode_block_tokens(VP9_COMMON *cm, MACROBLOCKD *xd,
+int vp9_decode_block_tokens(MACROBLOCKD *xd,
int plane, int block,
BLOCK_SIZE plane_bsize, int x, int y,
TX_SIZE tx_size, vp9_reader *r,
int seg_id) {
struct macroblockd_plane *const pd = &xd->plane[plane];
- const int16_t *const dequant = (plane == 0) ? cm->y_dequant[seg_id]
- : cm->uv_dequant[seg_id];
+ const int16_t *const dequant = pd->seg_dequant[seg_id];
const int ctx = get_entropy_context(tx_size, pd->above_context + x,
pd->left_context + y);
const scan_order *so = get_scan(xd, tx_size, pd->plane_type, block);
- const int eob = decode_coefs(cm, xd, pd->plane_type,
+ const int eob = decode_coefs(xd, pd->plane_type,
BLOCK_OFFSET(pd->dqcoeff, block), tx_size,
dequant, ctx, so->scan, so->neighbors, r);
vp9_set_contexts(xd, pd, plane_bsize, tx_size, eob > 0, x, y);
diff --git a/vp9/decoder/vp9_detokenize.h b/vp9/decoder/vp9_detokenize.h
index 25687c2..df17606 100644
--- a/vp9/decoder/vp9_detokenize.h
+++ b/vp9/decoder/vp9_detokenize.h
@@ -19,7 +19,7 @@
extern "C" {
#endif
-int vp9_decode_block_tokens(VP9_COMMON *cm, MACROBLOCKD *xd,
+int vp9_decode_block_tokens(MACROBLOCKD *xd,
int plane, int block,
BLOCK_SIZE plane_bsize, int x, int y,
TX_SIZE tx_size, vp9_reader *r,
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index 4b1c959..71c1e0b 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -48,6 +48,8 @@
int16_t motion_thresh;
// Rate target ratio to set q delta.
double rate_ratio_qdelta;
+ // Boost factor for rate target ratio, for segment CR_SEGMENT_ID_BOOST2.
+ double rate_boost_fac;
double low_content_avg;
int qindex_delta_seg1;
int qindex_delta_seg2;
@@ -93,6 +95,19 @@
return 1;
}
+static void adjust_cyclic_refresh_parameters(VP9_COMP *const cpi) {
+ const VP9_COMMON *const cm = &cpi->common;
+ const RATE_CONTROL *const rc = &cpi->rc;
+ CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
+ // Adjust some parameters, currently only for low resolutions at low bitrates.
+ if (cm->width <= 352 &&
+ cm->height <= 288 &&
+ rc->avg_frame_bandwidth < 3400) {
+ cr->motion_thresh = 4;
+ cr->rate_boost_fac = 1.25;
+ }
+}
+
// Check if this coding block, of size bsize, should be considered for refresh
// (lower-qp coding). Decision can be based on various factors, such as
// size of the coding block (i.e., below min_block size rejected), coding
@@ -462,6 +477,7 @@
vp9_clear_system_state();
cr->max_qdelta_perc = 50;
cr->time_for_refresh = 0;
+ cr->rate_boost_fac = 1.7;
// Set rate threshold to some multiple (set to 2 for now) of the target
// rate (target is given by sb64_target_rate and scaled by 256).
cr->thresh_rate_sb = ((int64_t)(rc->sb64_target_rate) << 8) << 2;
@@ -470,6 +486,9 @@
// vp9_convert_qindex_to_q(), vp9_ac_quant(), ac_qlookup*[].
cr->thresh_dist_sb = ((int64_t)(q * q)) << 2;
cr->motion_thresh = 32;
+
+ adjust_cyclic_refresh_parameters(cpi);
+
// Set up segmentation.
// Clear down the segment map.
vp9_enable_segmentation(&cm->seg);
@@ -505,7 +524,7 @@
// Set a more aggressive (higher) q delta for segment BOOST2.
qindex_delta = compute_deltaq(cpi, cm->base_qindex,
MIN(CR_MAX_RATE_TARGET_RATIO,
- CR_BOOST2_FAC * cr->rate_ratio_qdelta));
+ cr->rate_boost_fac * cr->rate_ratio_qdelta));
cr->qindex_delta_seg2 = qindex_delta;
vp9_set_segdata(seg, CR_SEGMENT_ID_BOOST2, SEG_LVL_ALT_Q, qindex_delta);
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.h b/vp9/encoder/vp9_aq_cyclicrefresh.h
index 21f114b..99bb98e 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.h
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.h
@@ -27,9 +27,6 @@
// Maximum rate target ratio for setting segment delta-qp.
#define CR_MAX_RATE_TARGET_RATIO 4.0
-// Boost factor for rate target ratio, for segment CR_SEGMENT_ID_BOOST2.
-#define CR_BOOST2_FAC 1.7
-
struct VP9_COMP;
struct CYCLIC_REFRESH;
diff --git a/vp9/encoder/vp9_aq_variance.c b/vp9/encoder/vp9_aq_variance.c
index 9e5d9ee..f072717 100644
--- a/vp9/encoder/vp9_aq_variance.c
+++ b/vp9/encoder/vp9_aq_variance.c
@@ -82,6 +82,61 @@
}
}
+/* TODO(agrange, paulwilkins): The block_variance calls the unoptimized versions
+ * of variance() and highbd_8_variance(). It should not.
+ */
+static void aq_variance(const uint8_t *a, int a_stride,
+ const uint8_t *b, int b_stride,
+ int w, int h, unsigned int *sse, int *sum) {
+ int i, j;
+
+ *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;
+ }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void aq_highbd_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;
+ }
+}
+
+static void aq_highbd_8_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;
+ aq_highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
+ *sse = (unsigned int)sse_long;
+ *sum = (int)sum_long;
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
static unsigned int block_variance(VP9_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bs) {
@@ -98,18 +153,18 @@
int avg;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- highbd_8_variance(x->plane[0].src.buf, x->plane[0].src.stride,
- CONVERT_TO_BYTEPTR(vp9_highbd_64_zeros), 0, bw, bh,
- &sse, &avg);
+ aq_highbd_8_variance(x->plane[0].src.buf, x->plane[0].src.stride,
+ CONVERT_TO_BYTEPTR(vp9_highbd_64_zeros), 0, bw, bh,
+ &sse, &avg);
sse >>= 2 * (xd->bd - 8);
avg >>= (xd->bd - 8);
} else {
- variance(x->plane[0].src.buf, x->plane[0].src.stride,
- vp9_64_zeros, 0, bw, bh, &sse, &avg);
+ aq_variance(x->plane[0].src.buf, x->plane[0].src.stride,
+ vp9_64_zeros, 0, bw, bh, &sse, &avg);
}
#else
- variance(x->plane[0].src.buf, x->plane[0].src.stride,
- vp9_64_zeros, 0, bw, bh, &sse, &avg);
+ aq_variance(x->plane[0].src.buf, x->plane[0].src.stride,
+ vp9_64_zeros, 0, bw, bh, &sse, &avg);
#endif // CONFIG_VP9_HIGHBITDEPTH
var = sse - (((int64_t)avg * avg) / (bw * bh));
return (256 * var) / (bw * bh);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 95f9bc4..ba38d64 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2056,6 +2056,65 @@
#endif
}
+/* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
+ * and highbd_8_variance(). It should not.
+ */
+static void encoder_variance(const uint8_t *a, int a_stride,
+ const uint8_t *b, int b_stride,
+ int w, int h, unsigned int *sse, int *sum) {
+ int i, j;
+
+ *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;
+ }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void encoder_highbd_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;
+ }
+}
+
+static void encoder_highbd_8_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;
+ encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
+ &sse_long, &sum_long);
+ *sse = (unsigned int)sse_long;
+ *sum = (int)sum_long;
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
static int64_t get_sse(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
int width, int height) {
@@ -2067,15 +2126,15 @@
int x, y;
if (dw > 0) {
- variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
- dw, height, &sse, &sum);
+ encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
+ dw, height, &sse, &sum);
total_sse += sse;
}
if (dh > 0) {
- variance(&a[(height - dh) * a_stride], a_stride,
- &b[(height - dh) * b_stride], b_stride,
- width - dw, dh, &sse, &sum);
+ encoder_variance(&a[(height - dh) * a_stride], a_stride,
+ &b[(height - dh) * b_stride], b_stride,
+ width - dw, dh, &sse, &sum);
total_sse += sse;
}
@@ -2128,14 +2187,15 @@
unsigned int sse = 0;
int sum = 0;
if (dw > 0) {
- highbd_8_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
- dw, height, &sse, &sum);
+ encoder_highbd_8_variance(&a[width - dw], a_stride,
+ &b[width - dw], b_stride,
+ dw, height, &sse, &sum);
total_sse += sse;
}
if (dh > 0) {
- highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
- &b[(height - dh) * b_stride], b_stride,
- width - dw, dh, &sse, &sum);
+ encoder_highbd_8_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) {
@@ -2812,7 +2872,7 @@
"%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
"%6d %6d %5d %5d %5d "
"%10"PRId64" %10.3lf"
- "%10lf %8u %10"PRId64" %10d %10d\n",
+ "%10lf %8u %10"PRId64" %10d %10d %10d\n",
cpi->common.current_video_frame,
cm->width, cm->height,
cpi->rc.source_alt_ref_pending,
@@ -2841,7 +2901,8 @@
cpi->twopass.bits_left /
(1 + cpi->twopass.total_left_stats.coded_error),
cpi->tot_recode_hits, recon_err, cpi->rc.kf_boost,
- cpi->twopass.kf_zeromotion_pct);
+ cpi->twopass.kf_zeromotion_pct,
+ cpi->twopass.fr_content_type);
fclose(f);
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 856a665..f801851 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -113,8 +113,8 @@
fpfile = fopen("firstpass.stt", "a");
fprintf(fpfile, "%12.0lf %12.4lf %12.0lf %12.0lf %12.0lf %12.4lf %12.4lf"
- "%12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf"
- "%12.4lf %12.0lf %12.0lf %12.0lf %12.4lf\n",
+ "%12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf %12.4lf"
+ "%12.4lf %12.4lf %12.0lf %12.0lf %12.0lf %12.4lf\n",
stats->frame,
stats->weight,
stats->intra_error,
@@ -124,6 +124,8 @@
stats->pcnt_motion,
stats->pcnt_second_ref,
stats->pcnt_neutral,
+ stats->ul_intra_pct,
+ stats->image_start_row,
stats->MVr,
stats->mvr_abs,
stats->MVc,
@@ -160,7 +162,9 @@
section->pcnt_motion = 0.0;
section->pcnt_second_ref = 0.0;
section->pcnt_neutral = 0.0;
- section->MVr = 0.0;
+ section->ul_intra_pct = 0.0;
+ section->image_start_row = 0.0;
+ section->MVr = 0.0;
section->mvr_abs = 0.0;
section->MVc = 0.0;
section->mvc_abs = 0.0;
@@ -185,7 +189,9 @@
section->pcnt_motion += frame->pcnt_motion;
section->pcnt_second_ref += frame->pcnt_second_ref;
section->pcnt_neutral += frame->pcnt_neutral;
- section->MVr += frame->MVr;
+ section->ul_intra_pct += frame->ul_intra_pct;
+ section->image_start_row += frame->image_start_row;
+ section->MVr += frame->MVr;
section->mvr_abs += frame->mvr_abs;
section->MVc += frame->MVc;
section->mvc_abs += frame->mvc_abs;
@@ -208,7 +214,9 @@
section->pcnt_motion -= frame->pcnt_motion;
section->pcnt_second_ref -= frame->pcnt_second_ref;
section->pcnt_neutral -= frame->pcnt_neutral;
- section->MVr -= frame->MVr;
+ section->ul_intra_pct -= frame->ul_intra_pct;
+ section->image_start_row -= frame->image_start_row;
+ section->MVr -= frame->MVr;
section->mvr_abs -= frame->mvr_abs;
section->MVc -= frame->MVc;
section->mvc_abs -= frame->mvc_abs;
@@ -453,6 +461,8 @@
cpi->rc.frames_to_key = INT_MAX;
}
+#define UL_INTRA_THRESH 50
+#define INVALID_ROW -1
void vp9_first_pass(VP9_COMP *cpi, const struct lookahead_entry *source) {
int mb_row, mb_col;
MACROBLOCK *const x = &cpi->td.mb;
@@ -477,6 +487,8 @@
int second_ref_count = 0;
const int intrapenalty = INTRA_MODE_PENALTY;
double neutral_count;
+ int ul_intra_count = 0;
+ int image_data_start_row = INVALID_ROW;
int new_mv_count = 0;
int sum_in_vectors = 0;
MV lastmv = {0, 0};
@@ -636,6 +648,18 @@
(bsize >= BLOCK_16X16 ? TX_16X16 : TX_8X8) : TX_4X4;
vp9_encode_intra_block_plane(x, bsize, 0);
this_error = vpx_get_mb_ss(x->plane[0].src_diff);
+
+ // Keep a record of blocks that have almost no intra error residual
+ // (i.e. are in effect completely flat and untextured in the intra
+ // domain). In natural videos this is uncommon, but it is much more
+ // common in animations, graphics and screen content, so may be used
+ // as a signal to detect these types of content.
+ if (this_error < UL_INTRA_THRESH) {
+ ++ul_intra_count;
+ } else if ((mb_col > 0) && (image_data_start_row == INVALID_ROW)) {
+ image_data_start_row = mb_row;
+ }
+
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth) {
switch (cm->bit_depth) {
@@ -963,6 +987,18 @@
vp9_clear_system_state();
}
+ // Clamp the image start to rows/2. This number of rows is discarded top
+ // and bottom as dead data so rows / 2 means the frame is blank.
+ if ((image_data_start_row > cm->mb_rows / 2) ||
+ (image_data_start_row == INVALID_ROW)) {
+ image_data_start_row = cm->mb_rows / 2;
+ }
+ // Exclude any image dead zone
+ if (image_data_start_row > 0) {
+ ul_intra_count =
+ MAX(0, ul_intra_count - (image_data_start_row * cm->mb_cols * 2));
+ }
+
{
FIRSTPASS_STATS fps;
// The minimum error here insures some bit allocation to frames even
@@ -987,6 +1023,8 @@
fps.pcnt_inter = (double)intercount / num_mbs;
fps.pcnt_second_ref = (double)second_ref_count / num_mbs;
fps.pcnt_neutral = (double)neutral_count / num_mbs;
+ fps.ul_intra_pct = (double)ul_intra_count / num_mbs;
+ fps.image_start_row = (double)image_data_start_row;
if (mvcount > 0) {
fps.MVr = (double)sum_mvr / mvcount;
@@ -2565,6 +2603,12 @@
if (EOF == input_stats(twopass, &this_frame))
return;
+ // Set the frame content type flag.
+ if (this_frame.ul_intra_pct >= FC_ANIMATION_THRESH)
+ twopass->fr_content_type = FC_GRAPHICS_ANIMATION;
+ else
+ twopass->fr_content_type = FC_NORMAL;
+
// Keyframe and section processing.
if (rc->frames_to_key == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY)) {
FIRSTPASS_STATS this_frame_copy;
diff --git a/vp9/encoder/vp9_firstpass.h b/vp9/encoder/vp9_firstpass.h
index 4a03855..45f1132 100644
--- a/vp9/encoder/vp9_firstpass.h
+++ b/vp9/encoder/vp9_firstpass.h
@@ -51,6 +51,8 @@
double pcnt_motion;
double pcnt_second_ref;
double pcnt_neutral;
+ double ul_intra_pct;
+ double image_start_row;
double MVr;
double mvr_abs;
double MVc;
@@ -73,6 +75,13 @@
FRAME_UPDATE_TYPES = 5
} FRAME_UPDATE_TYPE;
+#define FC_ANIMATION_THRESH 0.15
+typedef enum {
+ FC_NORMAL = 0,
+ FC_GRAPHICS_ANIMATION = 1,
+ FRAME_CONTENT_TYPES = 2
+} FRAME_CONTENT_TYPE;
+
typedef struct {
unsigned char index;
RATE_FACTOR_LEVEL rf_level[(MAX_LAG_BUFFERS * 2) + 1];
@@ -103,6 +112,8 @@
uint8_t *this_frame_mb_stats;
FIRSTPASS_MB_STATS firstpass_mb_stats;
#endif
+ // An indication of the content type of the current frame
+ FRAME_CONTENT_TYPE fr_content_type;
// Projected total bits available for a key frame group of frames
int64_t kf_group_bits;
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 2a3a3fa..60bff57 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -1058,6 +1058,7 @@
int mi_row, int mi_col, RD_COST *rd_cost,
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx) {
VP9_COMMON *const cm = &cpi->common;
+ SPEED_FEATURES *const sf = &cpi->sf;
TileInfo *const tile_info = &tile_data->tile_info;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
@@ -1218,7 +1219,7 @@
continue;
i = (ref_frame == LAST_FRAME) ? GOLDEN_FRAME : LAST_FRAME;
- if (cpi->ref_frame_flags & flag_list[i])
+ if ((cpi->ref_frame_flags & flag_list[i]) && sf->reference_masking)
if (x->pred_mv_sad[ref_frame] > (x->pred_mv_sad[i] << 1))
ref_frame_skip_mask |= (1 << ref_frame);
if (ref_frame_skip_mask & (1 << ref_frame))
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 61c066e..6c6c4ed 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -19,9 +19,33 @@
return frame_is_kf_gf_arf(cpi) || vp9_is_upper_layer_key_frame(cpi);
}
-static void set_good_speed_feature_framesize_dependent(VP9_COMMON *cm,
+// Sets a partition size down to which the auto partition code will always
+// search (can go lower), based on the image dimensions. The logic here
+// is that the extent to which ringing artefacts are offensive, depends
+// partly on the screen area that over which they propogate. Propogation is
+// limited by transform block size but the screen area take up by a given block
+// size will be larger for a small image format stretched to full screen.
+static BLOCK_SIZE set_partition_min_limit(VP9_COMMON *const cm) {
+ unsigned int screen_area = (cm->width * cm->height);
+
+ // Select block size based on image format size.
+ if (screen_area < 1280 * 720) {
+ // Formats smaller in area than 720P
+ return BLOCK_4X4;
+ } else if (screen_area < 1920 * 1080) {
+ // Format >= 720P and < 1080P
+ return BLOCK_8X8;
+ } else {
+ // Formats 1080P and up
+ return BLOCK_16X16;
+ }
+}
+
+static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
SPEED_FEATURES *sf,
int speed) {
+ VP9_COMMON *const cm = &cpi->common;
+
if (speed >= 1) {
if (MIN(cm->width, cm->height) >= 720) {
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
@@ -45,6 +69,7 @@
sf->partition_search_breakout_dist_thr = (1 << 22);
sf->partition_search_breakout_rate_thr = 100;
}
+ sf->rd_auto_partition_min_limit = set_partition_min_limit(cm);
}
if (speed >= 3) {
@@ -62,6 +87,13 @@
}
}
+ // If this is a two pass clip that fits the criteria for animated or
+ // graphics content then reset disable_split_mask for speeds 1-4.
+ if ((speed >= 1) && (cpi->oxcf.pass == 2) &&
+ (cpi->twopass.fr_content_type == FC_GRAPHICS_ANIMATION)) {
+ sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
+ }
+
if (speed >= 4) {
if (MIN(cm->width, cm->height) >= 720) {
sf->partition_search_breakout_dist_thr = (1 << 26);
@@ -72,29 +104,6 @@
}
}
-// Sets a partition size down to which the auto partition code will always
-// search (can go lower), based on the image dimensions. The logic here
-// is that the extent to which ringing artefacts are offensive, depends
-// partly on the screen area that over which they propogate. Propogation is
-// limited by transform block size but the screen area take up by a given block
-// size will be larger for a small image format stretched to full screen.
-static BLOCK_SIZE set_partition_min_limit(VP9_COMP *cpi) {
- VP9_COMMON *const cm = &cpi->common;
- unsigned int screen_area = (cm->width * cm->height);
-
- // Select block size based on image format size.
- if (screen_area < 1280 * 720) {
- // Formats smaller in area than 720P
- return BLOCK_4X4;
- } else if (screen_area < 1920 * 1080) {
- // Format >= 720P and < 1080P
- return BLOCK_8X8;
- } else {
- // Formats 1080P and up
- return BLOCK_16X16;
- }
-}
-
static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
SPEED_FEATURES *sf, int speed) {
const int boosted = frame_is_boosted(cpi);
@@ -139,7 +148,6 @@
sf->disable_filter_search_var_thresh = 100;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
- sf->rd_auto_partition_min_limit = set_partition_min_limit(cpi);
sf->allow_partition_search_skip = 1;
}
@@ -260,8 +268,12 @@
FLAG_SKIP_INTRA_LOWVAR;
sf->adaptive_pred_interp_filter = 2;
- // Reference masking is not supported in dynamic scaling mode.
- sf->reference_masking = cpi->oxcf.resize_mode != RESIZE_DYNAMIC ? 1 : 0;
+ // Disable reference masking if using spatial scaling since
+ // pred_mv_sad will not be set (since vp9_mv_pred will not
+ // be called).
+ // TODO(marpan/agrange): Fix this condition.
+ sf->reference_masking = (cpi->oxcf.resize_mode != RESIZE_DYNAMIC &&
+ cpi->svc.number_spatial_layers == 1) ? 1 : 0;
sf->disable_filter_search_var_thresh = 50;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
@@ -379,7 +391,6 @@
void vp9_set_speed_features_framesize_dependent(VP9_COMP *cpi) {
SPEED_FEATURES *const sf = &cpi->sf;
- VP9_COMMON *const cm = &cpi->common;
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
RD_OPT *const rd = &cpi->rd;
int i;
@@ -387,7 +398,7 @@
if (oxcf->mode == REALTIME) {
set_rt_speed_feature_framesize_dependent(cpi, sf, oxcf->speed);
} else if (oxcf->mode == GOOD) {
- set_good_speed_feature_framesize_dependent(cm, sf, oxcf->speed);
+ set_good_speed_feature_framesize_dependent(cpi, sf, oxcf->speed);
}
if (sf->disable_split_mask == DISABLE_ALL_SPLIT) {
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index 9197048..cb1b0df 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -131,7 +131,6 @@
LAYER_CONTEXT *const lc =
&svc->layer_context[sl * oxcf->ts_number_layers + tl];
RATE_CONTROL *const lrc = &lc->rc;
- layer = LAYER_IDS_TO_IDX(sl, tl, oxcf->ts_number_layers);
lc->spatial_layer_target_bandwidth = spatial_layer_target;
bitrate_alloc = (float)lc->target_bandwidth / spatial_layer_target;
@@ -144,7 +143,7 @@
lrc->bits_off_target =
MIN(lrc->bits_off_target, lrc->maximum_buffer_size);
lrc->buffer_level = MIN(lrc->buffer_level, lrc->maximum_buffer_size);
- lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[layer];
+ lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[tl];
lrc->avg_frame_bandwidth = (int)(lc->target_bandwidth / lc->framerate);
lrc->max_frame_bandwidth = rc->max_frame_bandwidth;
lrc->worst_quality = rc->worst_quality;
diff --git a/vp9/encoder/vp9_variance.h b/vp9/encoder/vp9_variance.h
index 8fc47a8..0a87395 100644
--- a/vp9/encoder/vp9_variance.h
+++ b/vp9/encoder/vp9_variance.h
@@ -18,60 +18,6 @@
extern "C" {
#endif
-// TODO(johannkoenig): All functions which depend on
-// [highbd_][8|10|12_]variance should be refactored or moved to vpx_dsp.
-static void variance(const uint8_t *a, int a_stride,
- const uint8_t *b, int b_stride,
- int w, int h, unsigned int *sse, int *sum) {
- int i, j;
-
- *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;
- }
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void highbd_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;
- }
-}
-static void highbd_8_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;
- highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
- *sse = (unsigned int)sse_long;
- *sum = (int)sum_long;
-}
-#endif
-
typedef unsigned int(*vp9_sad_fn_t)(const uint8_t *src_ptr,
int source_stride,
const uint8_t *ref_ptr,
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 166926b..d018699 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -146,6 +146,10 @@
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct16x16_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct32x32_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct_msa.h
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_loopfilter_4_msa.c
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_loopfilter_8_msa.c
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_loopfilter_16_msa.c
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_loopfilter_msa.h
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.c
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.h
diff --git a/vp9/vp9_dx_iface.c b/vp9/vp9_dx_iface.c
index 8cff965..4080d64 100644
--- a/vp9/vp9_dx_iface.c
+++ b/vp9/vp9_dx_iface.c
@@ -55,6 +55,7 @@
int invert_tile_order;
int last_show_frame; // Index of last output frame.
int byte_alignment;
+ int skip_loop_filter;
// Frame parallel related.
int frame_parallel_decode; // frame-based threading.
@@ -285,6 +286,7 @@
cm->new_fb_idx = INVALID_IDX;
cm->byte_alignment = ctx->byte_alignment;
+ cm->skip_loop_filter = ctx->skip_loop_filter;
if (ctx->get_ext_fb_cb != NULL && ctx->release_ext_fb_cb != NULL) {
pool->get_fb_cb = ctx->get_ext_fb_cb;
@@ -1059,6 +1061,19 @@
return VPX_CODEC_OK;
}
+static vpx_codec_err_t ctrl_set_skip_loop_filter(vpx_codec_alg_priv_t *ctx,
+ va_list args) {
+ ctx->skip_loop_filter = va_arg(args, int);
+
+ if (ctx->frame_workers) {
+ VP9Worker *const worker = ctx->frame_workers;
+ FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1;
+ frame_worker_data->pbi->common.skip_loop_filter = ctx->skip_loop_filter;
+ }
+
+ return VPX_CODEC_OK;
+}
+
static vpx_codec_ctrl_fn_map_t decoder_ctrl_maps[] = {
{VP8_COPY_REFERENCE, ctrl_copy_reference},
@@ -1072,6 +1087,7 @@
{VP9_INVERT_TILE_DECODE_ORDER, ctrl_set_invert_tile_order},
{VPXD_SET_DECRYPTOR, ctrl_set_decryptor},
{VP9_SET_BYTE_ALIGNMENT, ctrl_set_byte_alignment},
+ {VP9_SET_SKIP_LOOP_FILTER, ctrl_set_skip_loop_filter},
// Getters
{VP8D_GET_LAST_REF_UPDATES, ctrl_get_last_ref_updates},
diff --git a/vpx/vp8dx.h b/vpx/vp8dx.h
index 83898bf..bc9cb1a 100644
--- a/vpx/vp8dx.h
+++ b/vpx/vp8dx.h
@@ -106,6 +106,13 @@
*/
VP9_INVERT_TILE_DECODE_ORDER,
+ /** control function to set the skip loop filter flag. Valid values are
+ * integers. The decoder will skip the loop filter when its value is set to
+ * nonzero. If the loop filter is skipped the decoder may accumulate decode
+ * artifacts. The default value is 0.
+ */
+ VP9_SET_SKIP_LOOP_FILTER,
+
VP8_DECODER_CTRL_ID_MAX
};
diff --git a/vpxdec.c b/vpxdec.c
index f219c2e..baa6115 100644
--- a/vpxdec.c
+++ b/vpxdec.c
@@ -106,24 +106,25 @@
};
#if CONFIG_VP8_DECODER
-static const arg_def_t addnoise_level = ARG_DEF(NULL, "noise-level", 1,
- "Enable VP8 postproc add noise");
-static const arg_def_t deblock = ARG_DEF(NULL, "deblock", 0,
- "Enable VP8 deblocking");
-static const arg_def_t demacroblock_level = ARG_DEF(NULL, "demacroblock-level", 1,
- "Enable VP8 demacroblocking, w/ level");
-static const arg_def_t pp_debug_info = ARG_DEF(NULL, "pp-debug-info", 1,
- "Enable VP8 visible debug info");
-static const arg_def_t pp_disp_ref_frame = ARG_DEF(NULL, "pp-dbg-ref-frame", 1,
- "Display only selected reference frame per macro block");
-static const arg_def_t pp_disp_mb_modes = ARG_DEF(NULL, "pp-dbg-mb-modes", 1,
- "Display only selected macro block modes");
-static const arg_def_t pp_disp_b_modes = ARG_DEF(NULL, "pp-dbg-b-modes", 1,
- "Display only selected block modes");
-static const arg_def_t pp_disp_mvs = ARG_DEF(NULL, "pp-dbg-mvs", 1,
- "Draw only selected motion vectors");
-static const arg_def_t mfqe = ARG_DEF(NULL, "mfqe", 0,
- "Enable multiframe quality enhancement");
+static const arg_def_t addnoise_level = ARG_DEF(
+ NULL, "noise-level", 1, "Enable VP8 postproc add noise");
+static const arg_def_t deblock = ARG_DEF(
+ NULL, "deblock", 0, "Enable VP8 deblocking");
+static const arg_def_t demacroblock_level = ARG_DEF(
+ NULL, "demacroblock-level", 1, "Enable VP8 demacroblocking, w/ level");
+static const arg_def_t pp_debug_info = ARG_DEF(
+ NULL, "pp-debug-info", 1, "Enable VP8 visible debug info");
+static const arg_def_t pp_disp_ref_frame = ARG_DEF(
+ NULL, "pp-dbg-ref-frame", 1,
+ "Display only selected reference frame per macro block");
+static const arg_def_t pp_disp_mb_modes = ARG_DEF(
+ NULL, "pp-dbg-mb-modes", 1, "Display only selected macro block modes");
+static const arg_def_t pp_disp_b_modes = ARG_DEF(
+ NULL, "pp-dbg-b-modes", 1, "Display only selected block modes");
+static const arg_def_t pp_disp_mvs = ARG_DEF(
+ NULL, "pp-dbg-mvs", 1, "Draw only selected motion vectors");
+static const arg_def_t mfqe = ARG_DEF(
+ NULL, "mfqe", 0, "Enable multiframe quality enhancement");
static const arg_def_t *vp8_pp_args[] = {
&addnoise_level, &deblock, &demacroblock_level, &pp_debug_info,
@@ -813,34 +814,42 @@
fprintf(stderr, "%s\n", decoder.name);
#if CONFIG_VP8_DECODER
-
if (vp8_pp_cfg.post_proc_flag
&& vpx_codec_control(&decoder, VP8_SET_POSTPROC, &vp8_pp_cfg)) {
- fprintf(stderr, "Failed to configure postproc: %s\n", vpx_codec_error(&decoder));
+ fprintf(stderr, "Failed to configure postproc: %s\n",
+ vpx_codec_error(&decoder));
return EXIT_FAILURE;
}
if (vp8_dbg_color_ref_frame
- && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_REF_FRAME, vp8_dbg_color_ref_frame)) {
- fprintf(stderr, "Failed to configure reference block visualizer: %s\n", vpx_codec_error(&decoder));
+ && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_REF_FRAME,
+ vp8_dbg_color_ref_frame)) {
+ fprintf(stderr, "Failed to configure reference block visualizer: %s\n",
+ vpx_codec_error(&decoder));
return EXIT_FAILURE;
}
if (vp8_dbg_color_mb_modes
- && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_MB_MODES, vp8_dbg_color_mb_modes)) {
- fprintf(stderr, "Failed to configure macro block visualizer: %s\n", vpx_codec_error(&decoder));
+ && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_MB_MODES,
+ vp8_dbg_color_mb_modes)) {
+ fprintf(stderr, "Failed to configure macro block visualizer: %s\n",
+ vpx_codec_error(&decoder));
return EXIT_FAILURE;
}
if (vp8_dbg_color_b_modes
- && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_B_MODES, vp8_dbg_color_b_modes)) {
- fprintf(stderr, "Failed to configure block visualizer: %s\n", vpx_codec_error(&decoder));
+ && vpx_codec_control(&decoder, VP8_SET_DBG_COLOR_B_MODES,
+ vp8_dbg_color_b_modes)) {
+ fprintf(stderr, "Failed to configure block visualizer: %s\n",
+ vpx_codec_error(&decoder));
return EXIT_FAILURE;
}
if (vp8_dbg_display_mv
- && vpx_codec_control(&decoder, VP8_SET_DBG_DISPLAY_MV, vp8_dbg_display_mv)) {
- fprintf(stderr, "Failed to configure motion vector visualizer: %s\n", vpx_codec_error(&decoder));
+ && vpx_codec_control(&decoder, VP8_SET_DBG_DISPLAY_MV,
+ vp8_dbg_display_mv)) {
+ fprintf(stderr, "Failed to configure motion vector visualizer: %s\n",
+ vpx_codec_error(&decoder));
return EXIT_FAILURE;
}
#endif
diff --git a/vpxenc.c b/vpxenc.c
index 4519acb..8bbb9fc 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -330,10 +330,6 @@
NULL, "sharpness", 1, "Loop filter sharpness (0..7)");
static const arg_def_t static_thresh = ARG_DEF(
NULL, "static-thresh", 1, "Motion detection threshold");
-static const arg_def_t cpu_used_vp8 = ARG_DEF(
- NULL, "cpu-used", 1, "CPU Used (-16..16)");
-static const arg_def_t cpu_used_vp9 = ARG_DEF(
- NULL, "cpu-used", 1, "CPU Used (-8..8)");
static const arg_def_t auto_altref = ARG_DEF(
NULL, "auto-alt-ref", 1, "Enable automatic alt reference frames");
static const arg_def_t arnr_maxframes = ARG_DEF(
@@ -353,17 +349,14 @@
NULL, "cq-level", 1, "Constant/Constrained Quality level");
static const arg_def_t max_intra_rate_pct = ARG_DEF(
NULL, "max-intra-rate", 1, "Max I-frame bitrate (pct)");
-static const arg_def_t max_inter_rate_pct = ARG_DEF(
- NULL, "max-inter-rate", 1, "Max P-frame bitrate (pct)");
-static const arg_def_t gf_cbr_boost_pct = ARG_DEF(
- NULL, "gf-cbr-boost", 1, "Boost for Golden Frame in CBR mode (pct)");
-
-static const arg_def_t screen_content_mode = ARG_DEF(NULL, "screen-content-mode", 1,
- "Screen content mode");
#if CONFIG_VP8_ENCODER
+static const arg_def_t cpu_used_vp8 = ARG_DEF(
+ NULL, "cpu-used", 1, "CPU Used (-16..16)");
static const arg_def_t token_parts = ARG_DEF(
NULL, "token-parts", 1, "Number of token partitions to use, log2");
+static const arg_def_t screen_content_mode = ARG_DEF(
+ NULL, "screen-content-mode", 1, "Screen content mode");
static const arg_def_t *vp8_args[] = {
&cpu_used_vp8, &auto_altref, &noise_sens, &sharpness, &static_thresh,
&token_parts, &arnr_maxframes, &arnr_strength, &arnr_type,
@@ -382,6 +375,8 @@
#endif
#if CONFIG_VP9_ENCODER
+static const arg_def_t cpu_used_vp9 = ARG_DEF(
+ NULL, "cpu-used", 1, "CPU Used (-8..8)");
static const arg_def_t tile_cols = ARG_DEF(
NULL, "tile-columns", 1, "Number of tile columns to use, log2");
static const arg_def_t tile_rows = ARG_DEF(
@@ -397,6 +392,10 @@
static const arg_def_t frame_periodic_boost = ARG_DEF(
NULL, "frame-boost", 1,
"Enable frame periodic boost (0: off (default), 1: on)");
+static const arg_def_t gf_cbr_boost_pct = ARG_DEF(
+ NULL, "gf-cbr-boost", 1, "Boost for Golden Frame in CBR mode (pct)");
+static const arg_def_t max_inter_rate_pct = ARG_DEF(
+ NULL, "max-inter-rate", 1, "Max P-frame bitrate (pct)");
static const struct arg_enum_list color_space_enum[] = {
{ "unknown", VPX_CS_UNKNOWN },
@@ -469,6 +468,7 @@
void usage_exit(void) {
int i;
+ const int num_encoder = get_vpx_encoder_count();
fprintf(stderr, "Usage: %s <options> -o dst_filename src_filename \n",
exec_name);
@@ -496,11 +496,15 @@
" in fractional seconds. Default is 1/1000.\n");
fprintf(stderr, "\nIncluded encoders:\n\n");
- for (i = 0; i < get_vpx_encoder_count(); ++i) {
+ for (i = 0; i < num_encoder; ++i) {
const VpxInterface *const encoder = get_vpx_encoder_by_index(i);
- fprintf(stderr, " %-6s - %s\n",
- encoder->name, vpx_codec_iface_name(encoder->codec_interface()));
+ const char* defstr = (i == (num_encoder - 1)) ? "(default)" : "";
+ fprintf(stderr, " %-6s - %s %s\n",
+ encoder->name, vpx_codec_iface_name(encoder->codec_interface()),
+ defstr);
}
+ fprintf(stderr, "\n ");
+ fprintf(stderr, "Use --codec to switch to a non-default encoder.\n\n");
exit(EXIT_FAILURE);
}
@@ -811,10 +815,14 @@
static void parse_global_config(struct VpxEncoderConfig *global, char **argv) {
char **argi, **argj;
struct arg arg;
+ const int num_encoder = get_vpx_encoder_count();
+
+ if (num_encoder < 1)
+ die("Error: no valid encoder available\n");
/* Initialize default parameters */
memset(global, 0, sizeof(*global));
- global->codec = get_vpx_encoder_by_index(0);
+ global->codec = get_vpx_encoder_by_index(num_encoder - 1);
global->passes = 0;
global->color_type = I420;
/* Assign default deadline to good quality */
@@ -935,6 +943,10 @@
rewind(input->file);
}
+ /* Default to 1:1 pixel aspect ratio. */
+ input->pixel_aspect_ratio.numerator = 1;
+ input->pixel_aspect_ratio.denominator = 1;
+
/* For RAW input sources, these bytes will applied on the first frame
* in read_frame().
*/
@@ -948,6 +960,8 @@
input->file_type = FILE_TYPE_Y4M;
input->width = input->y4m.pic_w;
input->height = input->y4m.pic_h;
+ input->pixel_aspect_ratio.numerator = input->y4m.par_n;
+ input->pixel_aspect_ratio.denominator = input->y4m.par_d;
input->framerate.numerator = input->y4m.fps_n;
input->framerate.denominator = input->y4m.fps_d;
input->fmt = input->y4m.vpx_fmt;
@@ -1381,7 +1395,8 @@
static void open_output_file(struct stream_state *stream,
- struct VpxEncoderConfig *global) {
+ struct VpxEncoderConfig *global,
+ const struct VpxRational *pixel_aspect_ratio) {
const char *fn = stream->config.out_fn;
const struct vpx_codec_enc_cfg *const cfg = &stream->config.cfg;
@@ -1402,7 +1417,8 @@
write_webm_file_header(&stream->ebml, cfg,
&global->framerate,
stream->config.stereo_fmt,
- global->codec->fourcc);
+ global->codec->fourcc,
+ pixel_aspect_ratio);
}
#endif
@@ -2035,7 +2051,8 @@
}
FOREACH_STREAM(setup_pass(stream, &global, pass));
- FOREACH_STREAM(open_output_file(stream, &global));
+ FOREACH_STREAM(open_output_file(stream, &global,
+ &input.pixel_aspect_ratio));
FOREACH_STREAM(initialize_encoder(stream, &global));
#if CONFIG_VP9 && CONFIG_VP9_HIGHBITDEPTH
diff --git a/webmenc.cc b/webmenc.cc
index a0e542b..8212ee3 100644
--- a/webmenc.cc
+++ b/webmenc.cc
@@ -24,7 +24,8 @@
const vpx_codec_enc_cfg_t *cfg,
const struct vpx_rational *fps,
stereo_format_t stereo_fmt,
- unsigned int fourcc) {
+ unsigned int fourcc,
+ const struct VpxRational *par) {
mkvmuxer::MkvWriter *const writer = new mkvmuxer::MkvWriter(glob->stream);
mkvmuxer::Segment *const segment = new mkvmuxer::Segment();
segment->Init(writer);
@@ -49,6 +50,15 @@
segment->GetTrackByNumber(video_track_id));
video_track->SetStereoMode(stereo_fmt);
video_track->set_codec_id(fourcc == VP8_FOURCC ? "V_VP8" : "V_VP9");
+ if (par->numerator > 1 || par->denominator > 1) {
+ // TODO(fgalligan): Add support of DisplayUnit, Display Aspect Ratio type
+ // to WebM format.
+ const uint64_t display_width =
+ static_cast<uint64_t>(((cfg->g_w * par->numerator * 1.0) /
+ par->denominator) + .5);
+ video_track->set_display_width(display_width);
+ video_track->set_display_height(cfg->g_h);
+ }
if (glob->debug) {
video_track->set_uid(kDebugTrackUid);
}
diff --git a/webmenc.h b/webmenc.h
index 0ac606b..c255d3d 100644
--- a/webmenc.h
+++ b/webmenc.h
@@ -42,7 +42,8 @@
const vpx_codec_enc_cfg_t *cfg,
const struct vpx_rational *fps,
stereo_format_t stereo_fmt,
- unsigned int fourcc);
+ unsigned int fourcc,
+ const struct VpxRational *par);
void write_webm_block(struct EbmlGlobal *glob,
const vpx_codec_enc_cfg_t *cfg,
