Merge "[spatial svc]Add golden frame to first pass rate control"
diff --git a/examples/decode_to_md5.c b/examples/decode_to_md5.c
index 1c56303..fbc0f4a 100644
--- a/examples/decode_to_md5.c
+++ b/examples/decode_to_md5.c
@@ -33,8 +33,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
-
#include "vpx/vp8dx.h"
#include "vpx/vpx_decoder.h"
diff --git a/examples/decode_with_drops.c b/examples/decode_with_drops.c
index a20fdac..9423e38 100644
--- a/examples/decode_with_drops.c
+++ b/examples/decode_with_drops.c
@@ -56,8 +56,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
-
#include "vpx/vp8dx.h"
#include "vpx/vpx_decoder.h"
diff --git a/examples/postproc.c b/examples/postproc.c
index 59c50b1..c74347c 100644
--- a/examples/postproc.c
+++ b/examples/postproc.c
@@ -43,8 +43,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
-
#include "vpx/vp8dx.h"
#include "vpx/vpx_decoder.h"
diff --git a/examples/set_maps.c b/examples/set_maps.c
index 2ee5bca..851adc4 100644
--- a/examples/set_maps.c
+++ b/examples/set_maps.c
@@ -47,7 +47,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vp8cx.h"
#include "vpx/vpx_encoder.h"
diff --git a/examples/simple_decoder.c b/examples/simple_decoder.c
index 3318758..3f7d6aa 100644
--- a/examples/simple_decoder.c
+++ b/examples/simple_decoder.c
@@ -29,9 +29,7 @@
// -----------------
// For decoders, you only have to include `vpx_decoder.h` and then any
// header files for the specific codecs you use. In this case, we're using
-// vp8. The `VPX_CODEC_DISABLE_COMPAT` macro can be defined to ensure
-// strict compliance with the latest SDK by disabling some backwards
-// compatibility features. Defining this macro is encouraged.
+// vp8.
//
// Initializing The Codec
// ----------------------
@@ -81,8 +79,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
-
#include "vpx/vp8dx.h"
#include "vpx/vpx_decoder.h"
diff --git a/examples/simple_encoder.c b/examples/simple_encoder.c
index 30bb73a..f20c246 100644
--- a/examples/simple_encoder.c
+++ b/examples/simple_encoder.c
@@ -28,9 +28,7 @@
// -----------------
// For encoders, you only have to include `vpx_encoder.h` and then any
// header files for the specific codecs you use. In this case, we're using
-// vp8. The `VPX_CODEC_DISABLE_COMPAT` macro can be defined to ensure
-// strict compliance with the latest SDK by disabling some backwards
-// compatibility features. Defining this macro is encouraged.
+// vp8.
//
// Getting The Default Configuration
// ---------------------------------
@@ -101,7 +99,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vpx_encoder.h"
#include "./tools_common.h"
diff --git a/examples/twopass_encoder.c b/examples/twopass_encoder.c
index 76d5a28..653ae94 100644
--- a/examples/twopass_encoder.c
+++ b/examples/twopass_encoder.c
@@ -28,9 +28,8 @@
// Encoding A Frame
// ----------------
// Encoding a frame in two pass mode is identical to the simple encoder
-// example, except the deadline is set to VPX_DL_BEST_QUALITY to get the
-// best quality possible. VPX_DL_GOOD_QUALITY could also be used.
-//
+// example. To increase the quality while sacrificing encoding speed,
+// VPX_DL_BEST_QUALITY can be used in place of VPX_DL_GOOD_QUALITY.
//
// Processing Statistics Packets
// -----------------------------
@@ -52,7 +51,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vpx_encoder.h"
#include "./tools_common.h"
@@ -142,13 +140,13 @@
// Calculate frame statistics.
while (vpx_img_read(raw, infile)) {
++frame_count;
- get_frame_stats(&codec, raw, frame_count, 1, 0, VPX_DL_BEST_QUALITY,
+ get_frame_stats(&codec, raw, frame_count, 1, 0, VPX_DL_GOOD_QUALITY,
&stats);
}
// Flush encoder.
while (get_frame_stats(&codec, NULL, frame_count, 1, 0,
- VPX_DL_BEST_QUALITY, &stats)) {}
+ VPX_DL_GOOD_QUALITY, &stats)) {}
printf("Pass 0 complete. Processed %d frames.\n", frame_count);
if (vpx_codec_destroy(&codec))
@@ -182,11 +180,11 @@
// Encode frames.
while (vpx_img_read(raw, infile)) {
++frame_count;
- encode_frame(&codec, raw, frame_count, 1, 0, VPX_DL_BEST_QUALITY, writer);
+ encode_frame(&codec, raw, frame_count, 1, 0, VPX_DL_GOOD_QUALITY, writer);
}
// Flush encoder.
- while (encode_frame(&codec, NULL, -1, 1, 0, VPX_DL_BEST_QUALITY, writer)) {}
+ while (encode_frame(&codec, NULL, -1, 1, 0, VPX_DL_GOOD_QUALITY, writer)) {}
printf("\n");
diff --git a/examples/vp8_multi_resolution_encoder.c b/examples/vp8_multi_resolution_encoder.c
index 7c050fa..9f50dc7 100644
--- a/examples/vp8_multi_resolution_encoder.c
+++ b/examples/vp8_multi_resolution_encoder.c
@@ -24,7 +24,6 @@
#include "third_party/libyuv/include/libyuv/scale.h"
#include "third_party/libyuv/include/libyuv/cpu_id.h"
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vpx_encoder.h"
#include "vpx/vp8cx.h"
diff --git a/examples/vp8cx_set_ref.c b/examples/vp8cx_set_ref.c
index 5f3f0a3..b0961a2 100644
--- a/examples/vp8cx_set_ref.c
+++ b/examples/vp8cx_set_ref.c
@@ -50,7 +50,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vp8cx.h"
#include "vpx/vpx_encoder.h"
diff --git a/examples/vpx_temporal_svc_encoder.c b/examples/vpx_temporal_svc_encoder.c
index 5eac92c..3de983e 100644
--- a/examples/vpx_temporal_svc_encoder.c
+++ b/examples/vpx_temporal_svc_encoder.c
@@ -18,7 +18,6 @@
#include <stdlib.h>
#include <string.h>
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "./vpx_config.h"
#include "vpx_ports/vpx_timer.h"
#include "vpx/vp8cx.h"
diff --git a/libs.mk b/libs.mk
index c7c2748..f9f2d80 100644
--- a/libs.mk
+++ b/libs.mk
@@ -531,7 +531,6 @@
@echo " [CREATE] $@"
@rm -f $@
@echo "INPUT += $^" >> $@
- @echo "PREDEFINED = VPX_CODEC_DISABLE_COMPAT" >> $@
@echo "INCLUDE_PATH += ." >> $@;
@echo "ENABLED_SECTIONS += $(sort $(CODEC_DOC_SECTIONS))" >> $@
diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 50e7c23..b61d490 100644
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -118,7 +118,9 @@
{1, "invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf"},
{1, "invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf"},
{1, "invalid-vp90-2-05-resize.ivf.s59293_r01-05_b6-.ivf"},
- {1, "invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf"},
+ {1, "invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf"},
+ {1, "invalid-vp91-2-mixedrefcsp-444to420.ivf"},
+ {1, "invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf"},
};
VP9_INSTANTIATE_TEST_CASE(InvalidFileTest,
@@ -149,8 +151,8 @@
{4, "invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm"},
{4, "invalid-"
"vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf"},
- {2, "invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf"},
- {4, "invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf"},
+ {2, "invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf"},
+ {4, "invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf"},
};
INSTANTIATE_TEST_CASE_P(
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 84b13f9..e6114ab 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -683,10 +683,10 @@
717da707afcaa1f692ff1946f291054eb75a4f06 screendata.y4m
b7c1296630cdf1a7ef493d15ff4f9eb2999202f6 invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf
0a3884edb3fd8f9d9b500223e650f7de257b67d8 invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
-fac89b5735be8a86b0dc05159f996a5c3208ae32 invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
-22e0ee8babe574722baf4ef6d7ff5d7cf80d386c invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
-4506dfdcdf8ee4250924b075a0dcf1f070f72e5a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
-d3ea592c8d7b05d14c7ed48befc0a3aaf7709b7a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
+fac89b5735be8a86b0dc05159f996a5c3208ae32 invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf
+0a3884edb3fd8f9d9b500223e650f7de257b67d8 invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf.res
+4506dfdcdf8ee4250924b075a0dcf1f070f72e5a invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf
+bcdedaf168ac225575468fda77502d2dc9fd5baa invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf.res
65e93f9653bcf65b022f7d225268d1a90a76e7bb vp90-2-19-skip.webm
368dccdde5288c13c25695d2eacdc7402cadf613 vp90-2-19-skip.webm.md5
ffe460282df2b0e7d4603c2158653ad96f574b02 vp90-2-19-skip-01.webm
@@ -695,5 +695,9 @@
0a3884edb3fd8f9d9b500223e650f7de257b67d8 invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf.res
5e67e24e7f53fd189e565513cef8519b1bd6c712 invalid-vp90-2-05-resize.ivf.s59293_r01-05_b6-.ivf
741158f67c0d9d23726624d06bdc482ad368afc9 invalid-vp90-2-05-resize.ivf.s59293_r01-05_b6-.ivf.res
-8b1f7bf7e86c0976d277f60e8fcd9539e75a079a invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf
-fb79dcbbbb8c82d5a750e339acce66e39a32f15f invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf.res
+8b1f7bf7e86c0976d277f60e8fcd9539e75a079a invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf
+9c6bdf048fb2e66f07d4b4db5b32e6f303bd6109 invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf.res
+552e372e9b78127389fb06b34545df2cec15ba6d invalid-vp91-2-mixedrefcsp-444to420.ivf
+a61774cf03fc584bd9f0904fc145253bb8ea6c4c invalid-vp91-2-mixedrefcsp-444to420.ivf.res
+812d05a64a0d83c1b504d0519927ddc5a2cdb273 invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf
+1e472baaf5f6113459f0399a38a5a5e68d17799d invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf.res
diff --git a/test/test.mk b/test/test.mk
index c839c92..b92b6da 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -811,14 +811,18 @@
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x2_frame_parallel.webm.ivf.s47039_r01-05_b6-.ivf.res
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-08-tile_1x4_frame_parallel_all_key.webm.res
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.ivf.res
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.ivf.res
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf
-LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-aq2.webm.ivf.s3984_r01-05_b6-.v2.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s19552_r01-05_b6-.v2.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-09-subpixel-00.ivf.s20492_r01-05_b6-.v2.ivf.res
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf.res
ifeq ($(CONFIG_DECODE_PERF_TESTS),yes)
# BBB VP9 streams
diff --git a/test/variance_test.cc b/test/variance_test.cc
index b135470..f76402e 100644
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
@@ -35,6 +35,14 @@
using ::std::tr1::tuple;
using libvpx_test::ACMRandom;
+static unsigned int mb_ss_ref(const int16_t *src) {
+ unsigned int res = 0;
+ for (int i = 0; i < 256; ++i) {
+ res += src[i] * src[i];
+ }
+ return res;
+}
+
static unsigned int variance_ref(const uint8_t *ref, const uint8_t *src,
int l2w, int l2h, unsigned int *sse_ptr) {
int se = 0;
@@ -76,6 +84,50 @@
return sse - (((int64_t) se * se) >> (l2w + l2h));
}
+typedef unsigned int (*SumOfSquaresFunction)(const int16_t *src);
+
+class SumOfSquaresTest : public ::testing::TestWithParam<SumOfSquaresFunction> {
+ public:
+ SumOfSquaresTest() : func_(GetParam()) {}
+
+ virtual ~SumOfSquaresTest() {
+ libvpx_test::ClearSystemState();
+ }
+
+ protected:
+ void ConstTest();
+ void RefTest();
+
+ SumOfSquaresFunction func_;
+ ACMRandom rnd_;
+};
+
+void SumOfSquaresTest::ConstTest() {
+ int16_t mem[256];
+ unsigned int res;
+ for (int v = 0; v < 256; ++v) {
+ for (int i = 0; i < 256; ++i) {
+ mem[i] = v;
+ }
+ ASM_REGISTER_STATE_CHECK(res = func_(mem));
+ EXPECT_EQ(256u * (v * v), res);
+ }
+}
+
+void SumOfSquaresTest::RefTest() {
+ int16_t mem[256];
+ for (int i = 0; i < 100; ++i) {
+ for (int j = 0; j < 256; ++j) {
+ mem[j] = rnd_.Rand8() - rnd_.Rand8();
+ }
+
+ const unsigned int expected = mb_ss_ref(mem);
+ unsigned int res;
+ ASM_REGISTER_STATE_CHECK(res = func_(mem));
+ EXPECT_EQ(expected, res);
+ }
+}
+
template<typename VarianceFunctionType>
class VarianceTest
: public ::testing::TestWithParam<tuple<int, int, VarianceFunctionType> > {
@@ -362,6 +414,13 @@
namespace vp9 {
#if CONFIG_VP9_ENCODER
+
+TEST_P(SumOfSquaresTest, Const) { ConstTest(); }
+TEST_P(SumOfSquaresTest, Ref) { RefTest(); }
+
+INSTANTIATE_TEST_CASE_P(C, SumOfSquaresTest,
+ ::testing::Values(vp9_get_mb_ss_c));
+
typedef VarianceTest<vp9_variance_fn_t> VP9VarianceTest;
typedef SubpelVarianceTest<vp9_subpixvariance_fn_t> VP9SubpelVarianceTest;
typedef SubpelVarianceTest<vp9_subp_avg_variance_fn_t> VP9SubpelAvgVarianceTest;
@@ -487,6 +546,10 @@
#if HAVE_SSE2
#if CONFIG_USE_X86INC
+
+INSTANTIATE_TEST_CASE_P(SSE2, SumOfSquaresTest,
+ ::testing::Values(vp9_get_mb_ss_sse2));
+
const vp9_variance_fn_t variance4x4_sse2 = vp9_variance4x4_sse2;
const vp9_variance_fn_t variance4x8_sse2 = vp9_variance4x8_sse2;
const vp9_variance_fn_t variance8x4_sse2 = vp9_variance8x4_sse2;
diff --git a/third_party/libyuv/README.libvpx b/third_party/libyuv/README.libvpx
index fb2b9d1..3869d25 100644
--- a/third_party/libyuv/README.libvpx
+++ b/third_party/libyuv/README.libvpx
@@ -13,4 +13,4 @@
in order to encode multiple resolution bit streams.
Local Modifications:
-None.
+cherry-pick 'Issue 24479004: Fix building with MSVC for arm'
diff --git a/third_party/libyuv/source/cpu_id.cc b/third_party/libyuv/source/cpu_id.cc
index deb4c44..8f8a403 100644
--- a/third_party/libyuv/source/cpu_id.cc
+++ b/third_party/libyuv/source/cpu_id.cc
@@ -15,7 +15,8 @@
#endif
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
!defined(__native_client__) && \
- defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219)
+ defined(_MSC_VER) && (_MSC_FULL_VER >= 160040219) && \
+ (defined(_M_IX86) || defined(_M_X64))
#include <immintrin.h> // For _xgetbv()
#endif
diff --git a/third_party/libyuv/source/row_win.cc b/third_party/libyuv/source/row_win.cc
index f58fc51..d79c353 100644
--- a/third_party/libyuv/source/row_win.cc
+++ b/third_party/libyuv/source/row_win.cc
@@ -21,7 +21,8 @@
#endif
// This module is for Visual C.
-#if !defined(LIBYUV_DISABLE_X86) && defined(_MSC_VER)
+#if !defined(LIBYUV_DISABLE_X86) && defined(_MSC_VER) && \
+ (defined(_M_IX86) || defined(_M_X64))
#define YG 74 /* (int8)(1.164 * 64 + 0.5) */
diff --git a/vp8/common/rtcd_defs.pl b/vp8/common/rtcd_defs.pl
index 651809d..0070c28 100644
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -289,22 +289,19 @@
$vp8_sub_pixel_variance16x16_neon_asm=vp8_sub_pixel_variance16x16_neon;
add_proto qw/unsigned int vp8_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp8_variance_halfpixvar16x16_h mmx sse2 media neon_asm/;
+specialize qw/vp8_variance_halfpixvar16x16_h mmx sse2 media neon/;
$vp8_variance_halfpixvar16x16_h_sse2=vp8_variance_halfpixvar16x16_h_wmt;
$vp8_variance_halfpixvar16x16_h_media=vp8_variance_halfpixvar16x16_h_armv6;
-$vp8_variance_halfpixvar16x16_h_neon_asm=vp8_variance_halfpixvar16x16_h_neon;
add_proto qw/unsigned int vp8_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp8_variance_halfpixvar16x16_v mmx sse2 media neon_asm/;
+specialize qw/vp8_variance_halfpixvar16x16_v mmx sse2 media neon/;
$vp8_variance_halfpixvar16x16_v_sse2=vp8_variance_halfpixvar16x16_v_wmt;
$vp8_variance_halfpixvar16x16_v_media=vp8_variance_halfpixvar16x16_v_armv6;
-$vp8_variance_halfpixvar16x16_v_neon_asm=vp8_variance_halfpixvar16x16_v_neon;
add_proto qw/unsigned int vp8_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int *sse";
-specialize qw/vp8_variance_halfpixvar16x16_hv mmx sse2 media neon_asm/;
+specialize qw/vp8_variance_halfpixvar16x16_hv mmx sse2 media neon/;
$vp8_variance_halfpixvar16x16_hv_sse2=vp8_variance_halfpixvar16x16_hv_wmt;
$vp8_variance_halfpixvar16x16_hv_media=vp8_variance_halfpixvar16x16_hv_armv6;
-$vp8_variance_halfpixvar16x16_hv_neon_asm=vp8_variance_halfpixvar16x16_hv_neon;
#
# Single block SAD
@@ -530,13 +527,6 @@
}
#
-# Pick Loopfilter
-#
-add_proto qw/void vp8_yv12_copy_partial_frame/, "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc";
-specialize qw/vp8_yv12_copy_partial_frame neon_asm/;
-$vp8_yv12_copy_partial_frame_neon_asm=vp8_yv12_copy_partial_frame_neon;
-
-#
# Denoiser filter
#
if (vpx_config("CONFIG_TEMPORAL_DENOISING") eq "yes") {
@@ -544,7 +534,6 @@
specialize qw/vp8_denoiser_filter sse2 neon/;
add_proto qw/int vp8_denoiser_filter_uv/, "unsigned char *mc_running_avg, int mc_avg_stride, unsigned char *running_avg, int avg_stride, unsigned char *sig, int sig_stride, unsigned int motion_magnitude, int increase_denoising";
specialize qw/vp8_denoiser_filter_uv sse2 neon/;
-
}
# End of encoder only functions
diff --git a/vp8/encoder/arm/neon/picklpf_arm.c b/vp8/encoder/arm/neon/picklpf_arm.c
deleted file mode 100644
index ec8071e..0000000
--- a/vp8/encoder/arm/neon/picklpf_arm.c
+++ /dev/null
@@ -1,46 +0,0 @@
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "vp8/common/loopfilter.h"
-#include "vpx_scale/yv12config.h"
-
-extern void vp8_memcpy_partial_neon(unsigned char *dst_ptr,
- unsigned char *src_ptr,
- int sz);
-
-
-void vp8_yv12_copy_partial_frame_neon(YV12_BUFFER_CONFIG *src_ybc,
- YV12_BUFFER_CONFIG *dst_ybc)
-{
- unsigned char *src_y, *dst_y;
- int yheight;
- int ystride;
- int yoffset;
- int linestocopy;
-
- yheight = src_ybc->y_height;
- ystride = src_ybc->y_stride;
-
- /* number of MB rows to use in partial filtering */
- linestocopy = (yheight >> 4) / PARTIAL_FRAME_FRACTION;
- linestocopy = linestocopy ? linestocopy << 4 : 16; /* 16 lines per MB */
-
- /* Copy extra 4 so that full filter context is available if filtering done
- * on the copied partial frame and not original. Partial filter does mb
- * filtering for top row also, which can modify3 pixels above.
- */
- linestocopy += 4;
- /* partial image starts at ~middle of frame (macroblock border) */
- yoffset = ystride * (((yheight >> 5) * 16) - 4);
- src_y = src_ybc->y_buffer + yoffset;
- dst_y = dst_ybc->y_buffer + yoffset;
-
- vp8_memcpy_partial_neon(dst_y, src_y, ystride * linestocopy);
-}
diff --git a/vp8/encoder/arm/neon/vp8_memcpy_neon.asm b/vp8/encoder/arm/neon/vp8_memcpy_neon.asm
deleted file mode 100644
index d219e2d..0000000
--- a/vp8/encoder/arm/neon/vp8_memcpy_neon.asm
+++ /dev/null
@@ -1,72 +0,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
- EXPORT |vp8_memcpy_partial_neon|
-
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-;=========================================
-;this is not a full memcpy function!!!
-;void vp8_memcpy_partial_neon(unsigned char *dst_ptr, unsigned char *src_ptr,
-; int sz);
-|vp8_memcpy_partial_neon| PROC
- vpush {d8-d15}
- ;pld [r1] ;preload pred data
- ;pld [r1, #128]
- ;pld [r1, #256]
- ;pld [r1, #384]
-
- mov r12, r2, lsr #8 ;copy 256 bytes data at one time
-
-memcpy_neon_loop
- vld1.8 {q0, q1}, [r1]! ;load src data
- subs r12, r12, #1
- vld1.8 {q2, q3}, [r1]!
- vst1.8 {q0, q1}, [r0]! ;copy to dst_ptr
- vld1.8 {q4, q5}, [r1]!
- vst1.8 {q2, q3}, [r0]!
- vld1.8 {q6, q7}, [r1]!
- vst1.8 {q4, q5}, [r0]!
- vld1.8 {q8, q9}, [r1]!
- vst1.8 {q6, q7}, [r0]!
- vld1.8 {q10, q11}, [r1]!
- vst1.8 {q8, q9}, [r0]!
- vld1.8 {q12, q13}, [r1]!
- vst1.8 {q10, q11}, [r0]!
- vld1.8 {q14, q15}, [r1]!
- vst1.8 {q12, q13}, [r0]!
- vst1.8 {q14, q15}, [r0]!
-
- ;pld [r1] ;preload pred data -- need to adjust for real device
- ;pld [r1, #128]
- ;pld [r1, #256]
- ;pld [r1, #384]
-
- bne memcpy_neon_loop
-
- ands r3, r2, #0xff ;extra copy
- beq done_copy_neon_loop
-
-extra_copy_neon_loop
- vld1.8 {q0}, [r1]! ;load src data
- subs r3, r3, #16
- vst1.8 {q0}, [r0]!
- bne extra_copy_neon_loop
-
-done_copy_neon_loop
- vpop {d8-d15}
- bx lr
- ENDP
-
- END
diff --git a/vp8/encoder/denoising.c b/vp8/encoder/denoising.c
index 0c98eb1..2f33d4a 100644
--- a/vp8/encoder/denoising.c
+++ b/vp8/encoder/denoising.c
@@ -68,6 +68,10 @@
int adj_val[3] = {3, 4, 6};
int shift_inc1 = 0;
int shift_inc2 = 1;
+ int col_sum[16] = {0, 0, 0, 0,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0};
/* If motion_magnitude is small, making the denoiser more aggressive by
* increasing the adjustment for each level. Add another increment for
* blocks that are labeled for increase denoising. */
@@ -98,11 +102,11 @@
if (absdiff <= 3 + shift_inc1)
{
running_avg_y[c] = mc_running_avg_y[c];
- sum_diff += diff;
+ col_sum[c] += diff;
}
else
{
- if (absdiff >= 4 && absdiff <= 7)
+ if (absdiff >= 4 + shift_inc1 && absdiff <= 7)
adjustment = adj_val[0];
else if (absdiff >= 8 && absdiff <= 15)
adjustment = adj_val[1];
@@ -116,7 +120,7 @@
else
running_avg_y[c] = sig[c] + adjustment;
- sum_diff += adjustment;
+ col_sum[c] += adjustment;
}
else
{
@@ -125,7 +129,7 @@
else
running_avg_y[c] = sig[c] - adjustment;
- sum_diff -= adjustment;
+ col_sum[c] -= adjustment;
}
}
}
@@ -136,6 +140,23 @@
running_avg_y += avg_y_stride;
}
+ for (c = 0; c < 16; ++c) {
+ // Below we clip the value in the same way which SSE code use.
+ // When adopting aggressive denoiser, the adj_val for each pixel
+ // could be at most 8 (this is current max adjustment of the map).
+ // In SSE code, we calculate the sum of adj_val for
+ // the columns, so the sum could be upto 128(16 rows). However,
+ // the range of the value is -128 ~ 127 in SSE code, that's why
+ // we do this change in C code.
+ // We don't do this for UV denoiser, since there are only 8 rows,
+ // and max adjustments <= 8, so the sum of the columns will not
+ // exceed 64.
+ if (col_sum[c] >= 128) {
+ col_sum[c] = 127;
+ }
+ sum_diff += col_sum[c];
+ }
+
sum_diff_thresh= SUM_DIFF_THRESHOLD;
if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH;
if (abs(sum_diff) > sum_diff_thresh) {
@@ -166,14 +187,14 @@
running_avg_y[c] = 0;
else
running_avg_y[c] = running_avg_y[c] - adjustment;
- sum_diff -= adjustment;
+ col_sum[c] -= adjustment;
} else if (diff < 0) {
// Bring denoised signal up.
if (running_avg_y[c] + adjustment > 255)
running_avg_y[c] = 255;
else
running_avg_y[c] = running_avg_y[c] + adjustment;
- sum_diff += adjustment;
+ col_sum[c] += adjustment;
}
}
// TODO(marpan): Check here if abs(sum_diff) has gone below the
@@ -182,6 +203,15 @@
mc_running_avg_y += mc_avg_y_stride;
running_avg_y += avg_y_stride;
}
+
+ sum_diff = 0;
+ for (c = 0; c < 16; ++c) {
+ if (col_sum[c] >= 128) {
+ col_sum[c] = 127;
+ }
+ sum_diff += col_sum[c];
+ }
+
if (abs(sum_diff) > sum_diff_thresh)
return COPY_BLOCK;
} else {
@@ -412,12 +442,27 @@
vp8_denoiser_set_parameters(denoiser, mode);
denoiser->nmse_source_diff = 0;
denoiser->nmse_source_diff_count = 0;
+ denoiser->qp_avg = 0;
+ // QP threshold below which we can go up to aggressive mode.
+ denoiser->qp_threshold_up = 80;
+ // QP threshold above which we can go back down to normal mode.
+ // For now keep this second threshold high, so not used currently.
+ denoiser->qp_threshold_down = 128;
+ // Bitrate thresholds and noise metric (nmse) thresholds for switching to
+ // aggressive mode.
// TODO(marpan): Adjust thresholds, including effect on resolution.
+ denoiser->bitrate_threshold = 200000; // (bits/sec).
denoiser->threshold_aggressive_mode = 35;
- if (width * height > 640 * 480)
+ if (width * height > 640 * 480) {
+ denoiser->bitrate_threshold = 500000;
+ denoiser->threshold_aggressive_mode = 100;
+ } else if (width * height > 960 * 540) {
+ denoiser->bitrate_threshold = 800000;
denoiser->threshold_aggressive_mode = 150;
- else if (width * height > 1280 * 720)
+ } else if (width * height > 1280 * 720) {
+ denoiser->bitrate_threshold = 2000000;
denoiser->threshold_aggressive_mode = 1400;
+ }
return 0;
}
diff --git a/vp8/encoder/denoising.h b/vp8/encoder/denoising.h
index e1844f1..bb0c2ce 100644
--- a/vp8/encoder/denoising.h
+++ b/vp8/encoder/denoising.h
@@ -19,7 +19,7 @@
#endif
#define SUM_DIFF_THRESHOLD (16 * 16 * 2)
-#define SUM_DIFF_THRESHOLD_HIGH (16 * 16 * 3)
+#define SUM_DIFF_THRESHOLD_HIGH (600)
#define MOTION_MAGNITUDE_THRESHOLD (8*3)
#define SUM_DIFF_THRESHOLD_UV (96) // (8 * 8 * 1.5)
@@ -81,6 +81,10 @@
int threshold_aggressive_mode;
int nmse_source_diff;
int nmse_source_diff_count;
+ int qp_avg;
+ int qp_threshold_up;
+ int qp_threshold_down;
+ int bitrate_threshold;
denoise_params denoise_pars;
} VP8_DENOISER;
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 74e75c4..791a485 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -3370,33 +3370,43 @@
if (total > 0 &&
(num_blocks > (tot_num_blocks >> 4))) {
// Update the recursive mean square source_diff.
- if (cpi->denoiser.nmse_source_diff_count == 0)
+ if (cpi->denoiser.nmse_source_diff_count == 0) {
// First sample in new interval.
cpi->denoiser.nmse_source_diff = total;
- else
+ cpi->denoiser.qp_avg = cm->base_qindex;
+ } else {
// For subsequent samples, use average with weight ~1/4 for new sample.
cpi->denoiser.nmse_source_diff = (int)((total >> 2) +
3 * (cpi->denoiser.nmse_source_diff >> 2));
+ cpi->denoiser.qp_avg = (int)((cm->base_qindex >> 2) +
+ 3 * (cpi->denoiser.qp_avg >> 2));
+ }
cpi->denoiser.nmse_source_diff_count++;
}
// Check for changing the denoiser mode, when we have obtained #samples =
- // num_mode_change.
+ // num_mode_change. Condition the change also on the bitrate and QP.
if (cpi->denoiser.nmse_source_diff_count == num_mode_change) {
// Check for going up: from normal to aggressive mode.
if ((cpi->denoiser.denoiser_mode == kDenoiserOnYUV) &&
(cpi->denoiser.nmse_source_diff >
- cpi->denoiser.threshold_aggressive_mode)) {
+ cpi->denoiser.threshold_aggressive_mode) &&
+ (cpi->denoiser.qp_avg < cpi->denoiser.qp_threshold_up &&
+ cpi->target_bandwidth > cpi->denoiser.bitrate_threshold)) {
vp8_denoiser_set_parameters(&cpi->denoiser, kDenoiserOnYUVAggressive);
} else {
// Check for going down: from aggressive to normal mode.
- if ((cpi->denoiser.denoiser_mode == kDenoiserOnYUVAggressive) &&
+ if (((cpi->denoiser.denoiser_mode == kDenoiserOnYUVAggressive) &&
(cpi->denoiser.nmse_source_diff <
- cpi->denoiser.threshold_aggressive_mode)) {
+ cpi->denoiser.threshold_aggressive_mode)) ||
+ ((cpi->denoiser.denoiser_mode == kDenoiserOnYUVAggressive) &&
+ (cpi->denoiser.qp_avg > cpi->denoiser.qp_threshold_down ||
+ cpi->target_bandwidth < cpi->denoiser.bitrate_threshold))) {
vp8_denoiser_set_parameters(&cpi->denoiser, kDenoiserOnYUV);
}
}
// Reset metric and counter for next interval.
cpi->denoiser.nmse_source_diff = 0;
+ cpi->denoiser.qp_avg = 0;
cpi->denoiser.nmse_source_diff_count = 0;
}
}
diff --git a/vp8/encoder/picklpf.c b/vp8/encoder/picklpf.c
index 250d04c..f0c8f28 100644
--- a/vp8/encoder/picklpf.c
+++ b/vp8/encoder/picklpf.c
@@ -23,8 +23,8 @@
extern int vp8_calc_ss_err(YV12_BUFFER_CONFIG *source, YV12_BUFFER_CONFIG *dest);
-void vp8_yv12_copy_partial_frame_c(YV12_BUFFER_CONFIG *src_ybc,
- YV12_BUFFER_CONFIG *dst_ybc)
+static void yv12_copy_partial_frame(YV12_BUFFER_CONFIG *src_ybc,
+ YV12_BUFFER_CONFIG *dst_ybc)
{
unsigned char *src_y, *dst_y;
int yheight;
@@ -173,7 +173,7 @@
/* Get the err using the previous frame's filter value. */
/* Copy the unfiltered / processed recon buffer to the new buffer */
- vp8_yv12_copy_partial_frame(saved_frame, cm->frame_to_show);
+ yv12_copy_partial_frame(saved_frame, cm->frame_to_show);
vp8_loop_filter_partial_frame(cm, &cpi->mb.e_mbd, filt_val);
best_err = calc_partial_ssl_err(sd, cm->frame_to_show);
@@ -184,7 +184,7 @@
while (filt_val >= min_filter_level)
{
/* Apply the loop filter */
- vp8_yv12_copy_partial_frame(saved_frame, cm->frame_to_show);
+ yv12_copy_partial_frame(saved_frame, cm->frame_to_show);
vp8_loop_filter_partial_frame(cm, &cpi->mb.e_mbd, filt_val);
/* Get the err for filtered frame */
@@ -214,7 +214,7 @@
while (filt_val < max_filter_level)
{
/* Apply the loop filter */
- vp8_yv12_copy_partial_frame(saved_frame, cm->frame_to_show);
+ yv12_copy_partial_frame(saved_frame, cm->frame_to_show);
vp8_loop_filter_partial_frame(cm, &cpi->mb.e_mbd, filt_val);
diff --git a/vp8/vp8cx_arm.mk b/vp8/vp8cx_arm.mk
index eb94202..551271e 100644
--- a/vp8/vp8cx_arm.mk
+++ b/vp8/vp8cx_arm.mk
@@ -36,9 +36,7 @@
#File list for neon
# encoder
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/fastquantizeb_neon$(ASM)
-VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/picklpf_arm.c
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_mse16x16_neon$(ASM)
-VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_memcpy_neon$(ASM)
VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/denoising_neon.c
VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp8_shortwalsh4x4_neon.c
diff --git a/vp9/common/vp9_loopfilter.c b/vp9/common/vp9_loopfilter.c
index 3b39d42..4d03c4d 100644
--- a/vp9/common/vp9_loopfilter.c
+++ b/vp9/common/vp9_loopfilter.c
@@ -1193,7 +1193,7 @@
}
}
-void vp9_loop_filter_rows(const YV12_BUFFER_CONFIG *frame_buffer,
+void vp9_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
VP9_COMMON *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only) {
@@ -1247,9 +1247,8 @@
y_only);
}
-int vp9_loop_filter_worker(void *arg1, void *arg2) {
- LFWorkerData *const lf_data = (LFWorkerData*)arg1;
- (void)arg2;
+int vp9_loop_filter_worker(LFWorkerData *const lf_data, void *unused) {
+ (void)unused;
vp9_loop_filter_rows(lf_data->frame_buffer, lf_data->cm, lf_data->planes,
lf_data->start, lf_data->stop, lf_data->y_only);
return 1;
diff --git a/vp9/common/vp9_loopfilter.h b/vp9/common/vp9_loopfilter.h
index 6fa2773..69e7dd0 100644
--- a/vp9/common/vp9_loopfilter.h
+++ b/vp9/common/vp9_loopfilter.h
@@ -111,13 +111,13 @@
int y_only, int partial_frame);
// Apply the loop filter to [start, stop) macro block rows in frame_buffer.
-void vp9_loop_filter_rows(const YV12_BUFFER_CONFIG *frame_buffer,
+void vp9_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
struct VP9Common *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only);
typedef struct LoopFilterWorkerData {
- const YV12_BUFFER_CONFIG *frame_buffer;
+ YV12_BUFFER_CONFIG *frame_buffer;
struct VP9Common *cm;
struct macroblockd_plane planes[MAX_MB_PLANE];
@@ -129,8 +129,8 @@
int num_lf_workers;
} LFWorkerData;
-// Operates on the rows described by LFWorkerData passed as 'arg1'.
-int vp9_loop_filter_worker(void *arg1, void *arg2);
+// Operates on the rows described by 'lf_data'.
+int vp9_loop_filter_worker(LFWorkerData *const lf_data, void *unused);
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index fa49568..667e057 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -268,7 +268,7 @@
#
if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
add_proto qw/void vp9_mbpost_proc_down/, "uint8_t *dst, int pitch, int rows, int cols, int flimit";
-specialize qw/vp9_mbpost_proc_down mmx sse2/;
+specialize qw/vp9_mbpost_proc_down sse2/;
$vp9_mbpost_proc_down_sse2=vp9_mbpost_proc_down_xmm;
add_proto qw/void vp9_mbpost_proc_across_ip/, "uint8_t *src, int pitch, int rows, int cols, int flimit";
@@ -276,11 +276,11 @@
$vp9_mbpost_proc_across_ip_sse2=vp9_mbpost_proc_across_ip_xmm;
add_proto qw/void vp9_post_proc_down_and_across/, "const uint8_t *src_ptr, uint8_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
-specialize qw/vp9_post_proc_down_and_across mmx sse2/;
+specialize qw/vp9_post_proc_down_and_across sse2/;
$vp9_post_proc_down_and_across_sse2=vp9_post_proc_down_and_across_xmm;
add_proto qw/void vp9_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
-specialize qw/vp9_plane_add_noise mmx sse2/;
+specialize qw/vp9_plane_add_noise sse2/;
$vp9_plane_add_noise_sse2=vp9_plane_add_noise_wmt;
}
@@ -696,7 +696,7 @@
specialize qw/vp9_mse8x8/, "$sse2_x86inc";
add_proto qw/unsigned int vp9_get_mb_ss/, "const int16_t *";
-specialize qw/vp9_get_mb_ss sse2/;
+specialize qw/vp9_get_mb_ss/, "$sse2_x86inc";
# ENCODEMB INVOKE
add_proto qw/int64_t vp9_block_error/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size, int64_t *ssz";
diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c
index b60f8a0..df60987 100644
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -3573,6 +3573,7 @@
int stride) {
const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
const __m128i final_rounding = _mm_set1_epi16(1<<5);
+ const __m128i zero = _mm_setzero_si128();
// idct constants for each stage
const __m128i stg1_0 = pair_set_epi16(cospi_31_64, -cospi_1_64);
@@ -3635,7 +3636,6 @@
stp2_30, stp2_31;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
int i, j, i32;
- int zero_flag[2];
for (i = 0; i < 4; i++) {
i32 = (i << 5);
@@ -3710,13 +3710,7 @@
zero_idx[13] = _mm_or_si128(zero_idx[10], zero_idx[11]);
zero_idx[14] = _mm_or_si128(zero_idx[12], zero_idx[13]);
- zero_idx[0] = _mm_unpackhi_epi64(zero_idx[14], zero_idx[14]);
- zero_idx[1] = _mm_or_si128(zero_idx[0], zero_idx[14]);
- zero_idx[2] = _mm_srli_epi64(zero_idx[1], 32);
- zero_flag[0] = _mm_cvtsi128_si32(zero_idx[1]);
- zero_flag[1] = _mm_cvtsi128_si32(zero_idx[2]);
-
- if (!zero_flag[0] && !zero_flag[1]) {
+ if (_mm_movemask_epi8(_mm_cmpeq_epi32(zero_idx[14], zero)) == 0xFFFF) {
col[i32 + 0] = _mm_setzero_si128();
col[i32 + 1] = _mm_setzero_si128();
col[i32 + 2] = _mm_setzero_si128();
@@ -3795,7 +3789,6 @@
col[i32 + 31] = _mm_sub_epi16(stp1_0, stp1_31);
}
for (i = 0; i < 4; i++) {
- const __m128i zero = _mm_setzero_si128();
// Second 1-D idct
j = i << 3;
diff --git a/vp9/common/x86/vp9_postproc_mmx.asm b/vp9/common/x86/vp9_postproc_mmx.asm
deleted file mode 100644
index 5b8deef..0000000
--- a/vp9/common/x86/vp9_postproc_mmx.asm
+++ /dev/null
@@ -1,533 +0,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-%define VP9_FILTER_WEIGHT 128
-%define VP9_FILTER_SHIFT 7
-
-;void vp9_post_proc_down_and_across_mmx
-;(
-; unsigned char *src_ptr,
-; unsigned char *dst_ptr,
-; int src_pixels_per_line,
-; int dst_pixels_per_line,
-; int rows,
-; int cols,
-; int flimit
-;)
-global sym(vp9_post_proc_down_and_across_mmx) PRIVATE
-sym(vp9_post_proc_down_and_across_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-%if ABI_IS_32BIT=1 && CONFIG_PIC=1
- ; move the global rd onto the stack, since we don't have enough registers
- ; to do PIC addressing
- movq mm0, [GLOBAL(rd)]
- sub rsp, 8
- movq [rsp], mm0
-%define RD [rsp]
-%else
-%define RD [GLOBAL(rd)]
-%endif
-
- push rbx
- lea rbx, [GLOBAL(Blur)]
- movd mm2, dword ptr arg(6) ;flimit
- punpcklwd mm2, mm2
- punpckldq mm2, mm2
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(1) ;dst_ptr
-
- movsxd rcx, DWORD PTR arg(4) ;rows
- movsxd rax, DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
- pxor mm0, mm0 ; mm0 = 00000000
-
-.nextrow:
-
- xor rdx, rdx ; clear out rdx for use as loop counter
-.nextcol:
-
- pxor mm7, mm7 ; mm7 = 00000000
- movq mm6, [rbx + 32 ] ; mm6 = kernel 2 taps
- movq mm3, [rsi] ; mm4 = r0 p0..p7
- punpcklbw mm3, mm0 ; mm3 = p0..p3
- movq mm1, mm3 ; mm1 = p0..p3
- pmullw mm3, mm6 ; mm3 *= kernel 2 modifiers
-
- movq mm6, [rbx + 48] ; mm6 = kernel 3 taps
- movq mm5, [rsi + rax] ; mm4 = r1 p0..p7
- punpcklbw mm5, mm0 ; mm5 = r1 p0..p3
- pmullw mm6, mm5 ; mm6 *= p0..p3 * kernel 3 modifiers
- paddusw mm3, mm6 ; mm3 += mm6
-
- ; thresholding
- movq mm7, mm1 ; mm7 = r0 p0..p3
- psubusw mm7, mm5 ; mm7 = r0 p0..p3 - r1 p0..p3
- psubusw mm5, mm1 ; mm5 = r1 p0..p3 - r0 p0..p3
- paddusw mm7, mm5 ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
- pcmpgtw mm7, mm2
-
- movq mm6, [rbx + 64 ] ; mm6 = kernel 4 modifiers
- movq mm5, [rsi + 2*rax] ; mm4 = r2 p0..p7
- punpcklbw mm5, mm0 ; mm5 = r2 p0..p3
- pmullw mm6, mm5 ; mm5 *= kernel 4 modifiers
- paddusw mm3, mm6 ; mm3 += mm5
-
- ; thresholding
- movq mm6, mm1 ; mm6 = r0 p0..p3
- psubusw mm6, mm5 ; mm6 = r0 p0..p3 - r2 p0..p3
- psubusw mm5, mm1 ; mm5 = r2 p0..p3 - r2 p0..p3
- paddusw mm6, mm5 ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
- pcmpgtw mm6, mm2
- por mm7, mm6 ; accumulate thresholds
-
-
- neg rax
- movq mm6, [rbx ] ; kernel 0 taps
- movq mm5, [rsi+2*rax] ; mm4 = r-2 p0..p7
- punpcklbw mm5, mm0 ; mm5 = r-2 p0..p3
- pmullw mm6, mm5 ; mm5 *= kernel 0 modifiers
- paddusw mm3, mm6 ; mm3 += mm5
-
- ; thresholding
- movq mm6, mm1 ; mm6 = r0 p0..p3
- psubusw mm6, mm5 ; mm6 = p0..p3 - r-2 p0..p3
- psubusw mm5, mm1 ; mm5 = r-2 p0..p3 - p0..p3
- paddusw mm6, mm5 ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
- pcmpgtw mm6, mm2
- por mm7, mm6 ; accumulate thresholds
-
- movq mm6, [rbx + 16] ; kernel 1 taps
- movq mm4, [rsi+rax] ; mm4 = r-1 p0..p7
- punpcklbw mm4, mm0 ; mm4 = r-1 p0..p3
- pmullw mm6, mm4 ; mm4 *= kernel 1 modifiers.
- paddusw mm3, mm6 ; mm3 += mm5
-
- ; thresholding
- movq mm6, mm1 ; mm6 = r0 p0..p3
- psubusw mm6, mm4 ; mm6 = p0..p3 - r-2 p0..p3
- psubusw mm4, mm1 ; mm5 = r-1 p0..p3 - p0..p3
- paddusw mm6, mm4 ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
- pcmpgtw mm6, mm2
- por mm7, mm6 ; accumulate thresholds
-
-
- paddusw mm3, RD ; mm3 += round value
- psraw mm3, VP9_FILTER_SHIFT ; mm3 /= 128
-
- pand mm1, mm7 ; mm1 select vals > thresh from source
- pandn mm7, mm3 ; mm7 select vals < thresh from blurred result
- paddusw mm1, mm7 ; combination
-
- packuswb mm1, mm0 ; pack to bytes
-
- movd [rdi], mm1 ;
- neg rax ; pitch is positive
-
-
- add rsi, 4
- add rdi, 4
- add rdx, 4
-
- cmp edx, dword ptr arg(5) ;cols
- jl .nextcol
- ; done with the all cols, start the across filtering in place
- sub rsi, rdx
- sub rdi, rdx
-
-
- push rax
- xor rdx, rdx
- mov rax, [rdi-4];
-
-.acrossnextcol:
- pxor mm7, mm7 ; mm7 = 00000000
- movq mm6, [rbx + 32 ] ;
- movq mm4, [rdi+rdx] ; mm4 = p0..p7
- movq mm3, mm4 ; mm3 = p0..p7
- punpcklbw mm3, mm0 ; mm3 = p0..p3
- movq mm1, mm3 ; mm1 = p0..p3
- pmullw mm3, mm6 ; mm3 *= kernel 2 modifiers
-
- movq mm6, [rbx + 48]
- psrlq mm4, 8 ; mm4 = p1..p7
- movq mm5, mm4 ; mm5 = p1..p7
- punpcklbw mm5, mm0 ; mm5 = p1..p4
- pmullw mm6, mm5 ; mm6 *= p1..p4 * kernel 3 modifiers
- paddusw mm3, mm6 ; mm3 += mm6
-
- ; thresholding
- movq mm7, mm1 ; mm7 = p0..p3
- psubusw mm7, mm5 ; mm7 = p0..p3 - p1..p4
- psubusw mm5, mm1 ; mm5 = p1..p4 - p0..p3
- paddusw mm7, mm5 ; mm7 = abs(p0..p3 - p1..p4)
- pcmpgtw mm7, mm2
-
- movq mm6, [rbx + 64 ]
- psrlq mm4, 8 ; mm4 = p2..p7
- movq mm5, mm4 ; mm5 = p2..p7
- punpcklbw mm5, mm0 ; mm5 = p2..p5
- pmullw mm6, mm5 ; mm5 *= kernel 4 modifiers
- paddusw mm3, mm6 ; mm3 += mm5
-
- ; thresholding
- movq mm6, mm1 ; mm6 = p0..p3
- psubusw mm6, mm5 ; mm6 = p0..p3 - p1..p4
- psubusw mm5, mm1 ; mm5 = p1..p4 - p0..p3
- paddusw mm6, mm5 ; mm6 = abs(p0..p3 - p1..p4)
- pcmpgtw mm6, mm2
- por mm7, mm6 ; accumulate thresholds
-
-
- movq mm6, [rbx ]
- movq mm4, [rdi+rdx-2] ; mm4 = p-2..p5
- movq mm5, mm4 ; mm5 = p-2..p5
- punpcklbw mm5, mm0 ; mm5 = p-2..p1
- pmullw mm6, mm5 ; mm5 *= kernel 0 modifiers
- paddusw mm3, mm6 ; mm3 += mm5
-
- ; thresholding
- movq mm6, mm1 ; mm6 = p0..p3
- psubusw mm6, mm5 ; mm6 = p0..p3 - p1..p4
- psubusw mm5, mm1 ; mm5 = p1..p4 - p0..p3
- paddusw mm6, mm5 ; mm6 = abs(p0..p3 - p1..p4)
- pcmpgtw mm6, mm2
- por mm7, mm6 ; accumulate thresholds
-
- movq mm6, [rbx + 16]
- psrlq mm4, 8 ; mm4 = p-1..p5
- punpcklbw mm4, mm0 ; mm4 = p-1..p2
- pmullw mm6, mm4 ; mm4 *= kernel 1 modifiers.
- paddusw mm3, mm6 ; mm3 += mm5
-
- ; thresholding
- movq mm6, mm1 ; mm6 = p0..p3
- psubusw mm6, mm4 ; mm6 = p0..p3 - p1..p4
- psubusw mm4, mm1 ; mm5 = p1..p4 - p0..p3
- paddusw mm6, mm4 ; mm6 = abs(p0..p3 - p1..p4)
- pcmpgtw mm6, mm2
- por mm7, mm6 ; accumulate thresholds
-
- paddusw mm3, RD ; mm3 += round value
- psraw mm3, VP9_FILTER_SHIFT ; mm3 /= 128
-
- pand mm1, mm7 ; mm1 select vals > thresh from source
- pandn mm7, mm3 ; mm7 select vals < thresh from blurred result
- paddusw mm1, mm7 ; combination
-
- packuswb mm1, mm0 ; pack to bytes
- mov DWORD PTR [rdi+rdx-4], eax ; store previous four bytes
- movd eax, mm1
-
- add rdx, 4
- cmp edx, dword ptr arg(5) ;cols
- jl .acrossnextcol;
-
- mov DWORD PTR [rdi+rdx-4], eax
- pop rax
-
- ; done with this rwo
- add rsi,rax ; next line
- movsxd rax, dword ptr arg(3) ;dst_pixels_per_line ; destination pitch?
- add rdi,rax ; next destination
- movsxd rax, dword ptr arg(2) ;src_pixels_per_line ; destination pitch?
-
- dec rcx ; decrement count
- jnz .nextrow ; next row
- pop rbx
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-%undef RD
-
-
-;void vp9_mbpost_proc_down_mmx(unsigned char *dst,
-; int pitch, int rows, int cols,int flimit)
-extern sym(vp9_rv)
-global sym(vp9_mbpost_proc_down_mmx) PRIVATE
-sym(vp9_mbpost_proc_down_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 5
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 136
-
- ; unsigned char d[16][8] at [rsp]
- ; create flimit2 at [rsp+128]
- mov eax, dword ptr arg(4) ;flimit
- mov [rsp+128], eax
- mov [rsp+128+4], eax
-%define flimit2 [rsp+128]
-
-%if ABI_IS_32BIT=0
- lea r8, [GLOBAL(sym(vp9_rv))]
-%endif
-
- ;rows +=8;
- add dword ptr arg(2), 8
-
- ;for(c=0; c<cols; c+=4)
-.loop_col:
- mov rsi, arg(0) ;s
- pxor mm0, mm0 ;
-
- movsxd rax, dword ptr arg(1) ;pitch ;
- neg rax ; rax = -pitch
-
- lea rsi, [rsi + rax*8]; ; rdi = s[-pitch*8]
- neg rax
-
-
- pxor mm5, mm5
- pxor mm6, mm6 ;
-
- pxor mm7, mm7 ;
- mov rdi, rsi
-
- mov rcx, 15 ;
-
-.loop_initvar:
- movd mm1, DWORD PTR [rdi];
- punpcklbw mm1, mm0 ;
-
- paddw mm5, mm1 ;
- pmullw mm1, mm1 ;
-
- movq mm2, mm1 ;
- punpcklwd mm1, mm0 ;
-
- punpckhwd mm2, mm0 ;
- paddd mm6, mm1 ;
-
- paddd mm7, mm2 ;
- lea rdi, [rdi+rax] ;
-
- dec rcx
- jne .loop_initvar
- ;save the var and sum
- xor rdx, rdx
-.loop_row:
- movd mm1, DWORD PTR [rsi] ; [s-pitch*8]
- movd mm2, DWORD PTR [rdi] ; [s+pitch*7]
-
- punpcklbw mm1, mm0
- punpcklbw mm2, mm0
-
- paddw mm5, mm2
- psubw mm5, mm1
-
- pmullw mm2, mm2
- movq mm4, mm2
-
- punpcklwd mm2, mm0
- punpckhwd mm4, mm0
-
- paddd mm6, mm2
- paddd mm7, mm4
-
- pmullw mm1, mm1
- movq mm2, mm1
-
- punpcklwd mm1, mm0
- psubd mm6, mm1
-
- punpckhwd mm2, mm0
- psubd mm7, mm2
-
-
- movq mm3, mm6
- pslld mm3, 4
-
- psubd mm3, mm6
- movq mm1, mm5
-
- movq mm4, mm5
- pmullw mm1, mm1
-
- pmulhw mm4, mm4
- movq mm2, mm1
-
- punpcklwd mm1, mm4
- punpckhwd mm2, mm4
-
- movq mm4, mm7
- pslld mm4, 4
-
- psubd mm4, mm7
-
- psubd mm3, mm1
- psubd mm4, mm2
-
- psubd mm3, flimit2
- psubd mm4, flimit2
-
- psrad mm3, 31
- psrad mm4, 31
-
- packssdw mm3, mm4
- packsswb mm3, mm0
-
- movd mm1, DWORD PTR [rsi+rax*8]
-
- movq mm2, mm1
- punpcklbw mm1, mm0
-
- paddw mm1, mm5
- mov rcx, rdx
-
- and rcx, 127
-%if ABI_IS_32BIT=1 && CONFIG_PIC=1
- push rax
- lea rax, [GLOBAL(sym(vp9_rv))]
- movq mm4, [rax + rcx*2] ;vp9_rv[rcx*2]
- pop rax
-%elif ABI_IS_32BIT=0
- movq mm4, [r8 + rcx*2] ;vp9_rv[rcx*2]
-%else
- movq mm4, [sym(vp9_rv) + rcx*2]
-%endif
- paddw mm1, mm4
- ;paddw xmm1, eight8s
- psraw mm1, 4
-
- packuswb mm1, mm0
- pand mm1, mm3
-
- pandn mm3, mm2
- por mm1, mm3
-
- and rcx, 15
- movd DWORD PTR [rsp+rcx*4], mm1 ;d[rcx*4]
-
- mov rcx, rdx
- sub rcx, 8
-
- and rcx, 15
- movd mm1, DWORD PTR [rsp+rcx*4] ;d[rcx*4]
-
- movd [rsi], mm1
- lea rsi, [rsi+rax]
-
- lea rdi, [rdi+rax]
- add rdx, 1
-
- cmp edx, dword arg(2) ;rows
- jl .loop_row
-
-
- add dword arg(0), 4 ; s += 4
- sub dword arg(3), 4 ; cols -= 4
- cmp dword arg(3), 0
- jg .loop_col
-
- add rsp, 136
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-%undef flimit2
-
-
-;void vp9_plane_add_noise_mmx (unsigned char *start, unsigned char *noise,
-; unsigned char blackclamp[16],
-; unsigned char whiteclamp[16],
-; unsigned char bothclamp[16],
-; unsigned int width, unsigned int height, int pitch)
-global sym(vp9_plane_add_noise_mmx) PRIVATE
-sym(vp9_plane_add_noise_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 8
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- ; we rely on the fact that the clamping vectors are stored contiguously
- ; in black/white/both order. Note that we have to reload this here because
- ; rdx could be trashed by rand()
- mov rdx, arg(2) ; blackclamp
-
-
- mov rdi, rcx
- movsxd rcx, dword arg(5) ;[Width]
- mov rsi, arg(0) ;Pos
- xor rax,rax
-
-.addnoise_nextset:
- movq mm1,[rsi+rax] ; get the source
-
- psubusb mm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
- paddusb mm1, [rdx+32] ;bothclamp
- psubusb mm1, [rdx+16] ;whiteclamp
-
- movq mm2,[rdi+rax] ; get the noise for this line
- paddb mm1,mm2 ; add it in
- movq [rsi+rax],mm1 ; store the result
-
- add rax,8 ; move to the next line
-
- cmp rax, rcx
- jl .addnoise_nextset
-
- movsxd rax, dword arg(7) ; Pitch
- add arg(0), rax ; Start += Pitch
- sub dword arg(6), 1 ; Height -= 1
- jg .addnoise_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-SECTION_RODATA
-align 16
-Blur:
- times 16 dw 16
- times 8 dw 64
- times 16 dw 16
- times 8 dw 0
-
-rd:
- times 4 dw 0x40
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index a9c03f0..38e35ca 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -668,6 +668,15 @@
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffer");
}
+ cm->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
+}
+
+static INLINE int valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth,
+ int ref_xss, int ref_yss,
+ vpx_bit_depth_t this_bit_depth,
+ int this_xss, int this_yss) {
+ return ref_bit_depth == this_bit_depth && ref_xss == this_xss &&
+ ref_yss == this_yss;
}
static void setup_frame_size_with_refs(VP9_COMMON *cm,
@@ -707,6 +716,18 @@
if (!has_valid_ref_frame)
vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
"Referenced frame has invalid size");
+ for (i = 0; i < REFS_PER_FRAME; ++i) {
+ RefBuffer *const ref_frame = &cm->frame_refs[i];
+ if (!valid_ref_frame_img_fmt(
+ ref_frame->buf->bit_depth,
+ ref_frame->buf->uv_crop_width < ref_frame->buf->y_crop_width,
+ ref_frame->buf->uv_crop_height < ref_frame->buf->y_crop_height,
+ cm->bit_depth,
+ cm->subsampling_x,
+ cm->subsampling_y))
+ vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+ "Referenced frame has incompatible color space");
+ }
resize_context_buffers(cm, width, height);
setup_display_size(cm, rb);
@@ -723,6 +744,7 @@
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffer");
}
+ cm->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
}
static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
@@ -938,9 +960,8 @@
return vp9_reader_find_end(&tile_data->bit_reader);
}
-static int tile_worker_hook(void *arg1, void *arg2) {
- TileWorkerData *const tile_data = (TileWorkerData*)arg1;
- const TileInfo *const tile = (TileInfo*)arg2;
+static int tile_worker_hook(TileWorkerData *const tile_data,
+ const TileInfo *const tile) {
int mi_row, mi_col;
for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
@@ -1201,6 +1222,7 @@
}
setup_frame_size(cm, rb);
+ pbi->need_resync = 0;
} else {
cm->intra_only = cm->show_frame ? 0 : vp9_rb_read_bit(rb);
@@ -1224,6 +1246,7 @@
pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
setup_frame_size(cm, rb);
+ pbi->need_resync = 0;
} else {
pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
for (i = 0; i < REFS_PER_FRAME; ++i) {
@@ -1252,6 +1275,12 @@
}
}
+ if (pbi->need_resync) {
+ vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
+ "Keyframe / intra-only frame required to reset decoder"
+ " state");
+ }
+
if (!cm->error_resilient_mode) {
cm->refresh_frame_context = vp9_rb_read_bit(rb);
cm->frame_parallel_decoding_mode = vp9_rb_read_bit(rb);
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index 9106b0d..6ee3d70 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -60,6 +60,7 @@
}
cm->error.setjmp = 1;
+ pbi->need_resync = 1;
initialize_dec();
// Initialize the references to not point to any frame buffers.
@@ -238,6 +239,7 @@
cm->new_fb_idx = get_free_fb(cm);
if (setjmp(cm->error.jmp)) {
+ pbi->need_resync = 1;
cm->error.setjmp = 0;
vp9_clear_system_state();
diff --git a/vp9/decoder/vp9_decoder.h b/vp9/decoder/vp9_decoder.h
index 848d212..4f52bb9 100644
--- a/vp9/decoder/vp9_decoder.h
+++ b/vp9/decoder/vp9_decoder.h
@@ -58,6 +58,7 @@
int max_threads;
int inv_tile_order;
+ int need_resync; // wait for key/intra-only frame
} VP9Decoder;
int vp9_receive_compressed_data(struct VP9Decoder *pbi,
diff --git a/vp9/decoder/vp9_dthread.c b/vp9/decoder/vp9_dthread.c
index b82ea6a..6635880 100644
--- a/vp9/decoder/vp9_dthread.c
+++ b/vp9/decoder/vp9_dthread.c
@@ -121,10 +121,10 @@
}
// Row-based multi-threaded loopfilter hook
-static int loop_filter_row_worker(void *arg1, void *arg2) {
- TileWorkerData *const tile_data = (TileWorkerData*)arg1;
+static int loop_filter_row_worker(TileWorkerData *const tile_data,
+ void *unused) {
LFWorkerData *const lf_data = &tile_data->lfdata;
- (void) arg2;
+ (void)unused;
loop_filter_rows_mt(lf_data->frame_buffer, lf_data->cm, lf_data->planes,
lf_data->start, lf_data->stop, lf_data->y_only,
lf_data->lf_sync, lf_data->num_lf_workers);
@@ -145,15 +145,13 @@
const int num_workers = MIN(pbi->max_threads & ~1, tile_cols);
int i;
- // Allocate memory used in thread synchronization.
- // This always needs to be done even if frame_filter_level is 0.
+ if (!frame_filter_level) return;
+
if (!lf_sync->sync_range || cm->last_height != cm->height) {
vp9_loop_filter_dealloc(lf_sync);
- vp9_loop_filter_alloc(cm, lf_sync, sb_rows, cm->width);
+ vp9_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width);
}
- if (!frame_filter_level) return;
-
vp9_loop_filter_frame_init(cm, frame_filter_level);
// Initialize cur_sb_col to -1 for all SB rows.
@@ -216,7 +214,7 @@
}
// Allocate memory for lf row synchronization
-void vp9_loop_filter_alloc(VP9_COMMON *cm, VP9LfSync *lf_sync, int rows,
+void vp9_loop_filter_alloc(VP9LfSync *lf_sync, VP9_COMMON *cm, int rows,
int width) {
lf_sync->rows = rows;
#if CONFIG_MULTITHREAD
diff --git a/vp9/decoder/vp9_dthread.h b/vp9/decoder/vp9_dthread.h
index 8b02ef7..b1fbdeb 100644
--- a/vp9/decoder/vp9_dthread.h
+++ b/vp9/decoder/vp9_dthread.h
@@ -42,8 +42,8 @@
} VP9LfSync;
// Allocate memory for loopfilter row synchronization.
-void vp9_loop_filter_alloc(struct VP9Common *cm, VP9LfSync *lf_sync,
- int rows, int width);
+void vp9_loop_filter_alloc(VP9LfSync *lf_sync, VP9_COMMON *cm, int rows,
+ int width);
// Deallocate loopfilter synchronization related mutex and data.
void vp9_loop_filter_dealloc(VP9LfSync *lf_sync);
diff --git a/vp9/encoder/vp9_denoiser.c b/vp9/encoder/vp9_denoiser.c
index c4cf5ee..75b9449 100644
--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -89,9 +89,9 @@
int total_adj = 0;
int shift_inc = 1;
- /* If motion_magnitude is small, making the denoiser more aggressive by
- * increasing the adjustment for each level. Add another increment for
- * blocks that are labeled for increase denoising. */
+ // If motion_magnitude is small, making the denoiser more aggressive by
+ // increasing the adjustment for each level. Add another increment for
+ // blocks that are labeled for increase denoising.
if (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) {
if (increase_denoising) {
shift_inc = 2;
diff --git a/vp9/encoder/vp9_denoiser.h b/vp9/encoder/vp9_denoiser.h
index a913add..fa714b1 100644
--- a/vp9/encoder/vp9_denoiser.h
+++ b/vp9/encoder/vp9_denoiser.h
@@ -18,7 +18,7 @@
extern "C" {
#endif
-#define MOTION_MAGNITUDE_THRESHOLD (8*3)
+#define MOTION_MAGNITUDE_THRESHOLD (8 * 3)
typedef enum vp9_denoiser_decision {
COPY_BLOCK,
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 72ced05..c62b52f 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -2466,6 +2466,9 @@
vp9_zero(cpi->mb.pred_mv);
cpi->pc_root->index = 0;
+ // TODO(yunqingwang): use_lastframe_partitioning is no longer used in good-
+ // quality encoding. Need to evaluate it in real-time encoding later to
+ // decide if it can be removed too. And then, do the code cleanup.
if ((sf->partition_search_type == SEARCH_PARTITION &&
sf->use_lastframe_partitioning) ||
sf->partition_search_type == FIXED_PARTITION ||
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 2ca91b9..b6e606d 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2233,9 +2233,11 @@
#endif
#if CONFIG_INTERNAL_STATS
- int i;
- for (i = 0; i < MAX_MODES; ++i)
- cpi->mode_chosen_counts[i] = 0;
+ {
+ int i;
+ for (i = 0; i < MAX_MODES; ++i)
+ cpi->mode_chosen_counts[i] = 0;
+ }
#endif
vp9_set_speed_features(cpi);
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index eee6ffe..a97d778 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -28,11 +28,17 @@
#include "vp9/encoder/vp9_ratectrl.h"
#include "vp9/encoder/vp9_rd.h"
+typedef struct {
+ uint8_t *data;
+ int stride;
+ int in_use;
+} PRED_BUFFER;
+
static int mv_refs_rt(const VP9_COMMON *cm, const MACROBLOCKD *xd,
- const TileInfo *const tile,
- MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
- int_mv *mv_ref_list,
- int mi_row, int mi_col) {
+ const TileInfo *const tile,
+ MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
+ int_mv *mv_ref_list,
+ int mi_row, int mi_col) {
const int *ref_sign_bias = cm->ref_frame_sign_bias;
int i, refmv_count = 0;
@@ -224,13 +230,15 @@
if (cpi->common.tx_mode == TX_MODE_SELECT) {
if (sse > (var << 2))
- xd->mi[0]->mbmi.tx_size = MIN(max_txsize_lookup[bsize],
- tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+ xd->mi[0]->mbmi.tx_size =
+ MIN(max_txsize_lookup[bsize],
+ tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
else
xd->mi[0]->mbmi.tx_size = TX_8X8;
} else {
- xd->mi[0]->mbmi.tx_size = MIN(max_txsize_lookup[bsize],
- tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+ xd->mi[0]->mbmi.tx_size =
+ MIN(max_txsize_lookup[bsize],
+ tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
}
vp9_model_rd_from_var_lapndz(sse - var, 1 << num_pels_log2_lookup[bsize],
@@ -331,7 +339,7 @@
// The cost of skip bit needs to be added.
*rate = cpi->inter_mode_cost[mbmi->mode_context[ref_frame]]
- [INTER_OFFSET(this_mode)];
+ [INTER_OFFSET(this_mode)];
// More on this part of rate
// rate += vp9_cost_bit(vp9_get_skip_prob(cm, xd), 1);
@@ -402,20 +410,21 @@
// TODO(jingning) placeholder for inter-frame non-RD mode decision.
// this needs various further optimizations. to be continued..
-int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
- const TileInfo *const tile,
- int mi_row, int mi_col,
- int *returnrate,
- int64_t *returndistortion,
- BLOCK_SIZE bsize,
- PICK_MODE_CONTEXT *ctx) {
- MACROBLOCKD *xd = &x->e_mbd;
- MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
+ int mi_row, int mi_col,
+ int *returnrate,
+ int64_t *returndistortion,
+ BLOCK_SIZE bsize,
+ PICK_MODE_CONTEXT *ctx) {
+ VP9_COMMON *const cm = &cpi->common;
+ MACROBLOCKD *const xd = &x->e_mbd;
+ MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
struct macroblockd_plane *const pd = &xd->plane[0];
- PREDICTION_MODE this_mode, best_mode = ZEROMV;
+ PREDICTION_MODE best_mode = ZEROMV;
MV_REFERENCE_FRAME ref_frame, best_ref_frame = LAST_FRAME;
TX_SIZE best_tx_size = MIN(max_txsize_lookup[bsize],
- tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+ tx_mode_to_biggest_tx_size[cm->tx_mode]);
INTERP_FILTER best_pred_filter = EIGHTTAP;
int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
struct buf_2d yv12_mb[4][MAX_MB_PLANE];
@@ -430,26 +439,23 @@
unsigned int var_y = UINT_MAX;
unsigned int sse_y = UINT_MAX;
- VP9_COMMON *cm = &cpi->common;
- int intra_cost_penalty = 20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q);
-
+ const int intra_cost_penalty =
+ 20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q);
const int64_t inter_mode_thresh = RDCOST(x->rdmult, x->rddiv,
intra_cost_penalty, 0);
- const int64_t intra_mode_cost = 50;
+ const int intra_mode_cost = 50;
- unsigned char segment_id = mbmi->segment_id;
+ const int8_t segment_id = mbmi->segment_id;
const int *const rd_threshes = cpi->rd.threshes[segment_id][bsize];
const int *const rd_thresh_freq_fact = cpi->rd.thresh_freq_fact[bsize];
- // Mode index conversion form THR_MODES to PREDICTION_MODE for a ref frame.
INTERP_FILTER filter_ref = cm->interp_filter;
- int bsl = mi_width_log2(bsize);
+ const int bsl = mi_width_log2(bsize);
const int pred_filter_search = cm->interp_filter == SWITCHABLE ?
(((mi_row + mi_col) >> bsl) +
get_chessboard_index(cm->current_video_frame)) & 0x1 : 0;
int const_motion[MAX_REF_FRAMES] = { 0 };
const int bh = num_4x4_blocks_high_lookup[bsize] << 2;
const int bw = num_4x4_blocks_wide_lookup[bsize] << 2;
- const int pixels_in_block = bh * bw;
// For speed 6, the result of interp filter is reused later in actual encoding
// process.
// tmp[3] points to dst buffer, and the other 3 point to allocated buffers.
@@ -458,15 +464,11 @@
struct buf_2d orig_dst = pd->dst;
PRED_BUFFER *best_pred = NULL;
PRED_BUFFER *this_mode_pred = NULL;
- int i;
- // CTX is used by the temporal denoiser which is currently being developed.
- // TODO(jbb): when temporal denoiser is finished and in the default build
- // remove the following line;
- (void) ctx;
if (cpi->sf.reuse_inter_pred_sby) {
+ int i;
for (i = 0; i < 3; i++) {
- tmp[i].data = &pred_buf[pixels_in_block * i];
+ tmp[i].data = &pred_buf[bw * bh * i];
tmp[i].stride = bw;
tmp[i].in_use = 0;
}
@@ -492,6 +494,7 @@
mbmi->segment_id = segment_id;
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
+ PREDICTION_MODE this_mode;
x->pred_mv_sad[ref_frame] = INT_MAX;
frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
frame_mv[ZEROMV][ref_frame].as_int = 0;
@@ -546,8 +549,9 @@
if (!(cpi->sf.inter_mode_mask[bsize] & (1 << this_mode)))
continue;
- mode_rd_thresh = rd_threshes[mode_idx[ref_frame - LAST_FRAME]
- [INTER_OFFSET(this_mode)]];
+ mode_rd_thresh =
+ rd_threshes[mode_idx[ref_frame -
+ LAST_FRAME][INTER_OFFSET(this_mode)]];
if (rd_less_than_thresh(best_rd, mode_rd_thresh,
rd_thresh_freq_fact[this_mode]))
continue;
@@ -564,7 +568,7 @@
if (this_mode != NEARESTMV &&
frame_mv[this_mode][ref_frame].as_int ==
frame_mv[NEARESTMV][ref_frame].as_int)
- continue;
+ continue;
mbmi->mode = this_mode;
mbmi->mv[0].as_int = frame_mv[this_mode][ref_frame].as_int;
@@ -643,7 +647,7 @@
rate += rate_mv;
rate += cpi->inter_mode_cost[mbmi->mode_context[ref_frame]]
- [INTER_OFFSET(this_mode)];
+ [INTER_OFFSET(this_mode)];
this_rd = RDCOST(x->rdmult, x->rddiv, rate, dist);
// Skipping checking: test to see if this block can be reconstructed by
@@ -661,6 +665,8 @@
if (cpi->oxcf.noise_sensitivity > 0) {
vp9_denoiser_update_frame_stats(mbmi, sse_y, this_mode, ctx);
}
+#else
+ (void)ctx;
#endif
if (this_rd < best_rd || x->skip) {
@@ -696,15 +702,9 @@
// temp buf to dst buf.
if (best_pred != NULL && cpi->sf.reuse_inter_pred_sby &&
best_pred->data != orig_dst.buf) {
- uint8_t *copy_from, *copy_to;
-
pd->dst = orig_dst;
- copy_to = pd->dst.buf;
-
- copy_from = best_pred->data;
-
- vp9_convolve_copy(copy_from, bw, copy_to, pd->dst.stride, NULL, 0, NULL, 0,
- bw, bh);
+ vp9_convolve_copy(best_pred->data, bw, pd->dst.buf, pd->dst.stride, NULL, 0,
+ NULL, 0, bw, bh);
}
mbmi->mode = best_mode;
@@ -719,8 +719,8 @@
// threshold.
if (!x->skip && best_rd > inter_mode_thresh &&
bsize <= cpi->sf.max_intra_bsize) {
+ PREDICTION_MODE this_mode;
struct estimate_block_intra_args args = { cpi, x, DC_PRED, 0, 0 };
-
const TX_SIZE intra_tx_size =
MIN(max_txsize_lookup[bsize],
tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
@@ -761,6 +761,4 @@
if (cpi->sf.reuse_inter_pred_sby)
pd->dst = orig_dst;
}
-
- return INT64_MAX;
}
diff --git a/vp9/encoder/vp9_pickmode.h b/vp9/encoder/vp9_pickmode.h
index 49c6feb..97aeca7 100644
--- a/vp9/encoder/vp9_pickmode.h
+++ b/vp9/encoder/vp9_pickmode.h
@@ -17,19 +17,13 @@
extern "C" {
#endif
-typedef struct {
- uint8_t *data;
- int stride;
- int in_use;
-} PRED_BUFFER;
-
-int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
- const struct TileInfo *const tile,
- int mi_row, int mi_col,
- int *returnrate,
- int64_t *returndistortion,
- BLOCK_SIZE bsize,
- PICK_MODE_CONTEXT *ctx);
+void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
+ const struct TileInfo *const tile,
+ int mi_row, int mi_col,
+ int *returnrate,
+ int64_t *returndistortion,
+ BLOCK_SIZE bsize,
+ PICK_MODE_CONTEXT *ctx);
#ifdef __cplusplus
} // extern "C"
diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index b9e4408..b826ff4 100644
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -374,7 +374,6 @@
uint8_t *ref_y_ptr;
const int num_mv_refs = MAX_MV_REF_CANDIDATES +
(cpi->sf.adaptive_motion_search &&
- cpi->common.show_frame &&
block_size < cpi->sf.max_partition_size);
MV pred_mv[3];
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 9f9bed6..9096677 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1398,7 +1398,7 @@
mvp_full.row = bsi->mvp.as_mv.row >> 3;
mvp_full.col = bsi->mvp.as_mv.col >> 3;
- if (cpi->sf.adaptive_motion_search && cm->show_frame) {
+ if (cpi->sf.adaptive_motion_search) {
mvp_full.row = x->pred_mv[mbmi->ref_frame[0]].row >> 3;
mvp_full.col = x->pred_mv[mbmi->ref_frame[0]].col >> 3;
step_param = MAX(step_param, 8);
@@ -1815,8 +1815,7 @@
step_param = cpi->mv_step_param;
}
- if (cpi->sf.adaptive_motion_search && bsize < BLOCK_64X64 &&
- cm->show_frame) {
+ if (cpi->sf.adaptive_motion_search && bsize < BLOCK_64X64) {
int boffset = 2 * (b_width_log2(BLOCK_64X64) - MIN(b_height_log2(bsize),
b_width_log2(bsize)));
step_param = MAX(step_param, boffset);
@@ -1876,7 +1875,7 @@
*rate_mv = vp9_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
- if (cpi->sf.adaptive_motion_search && cm->show_frame)
+ if (cpi->sf.adaptive_motion_search)
x->pred_mv[ref] = tmp_mv->as_mv;
if (scaled_ref_frame) {
@@ -2576,6 +2575,7 @@
vp9_prob comp_mode_p;
int64_t best_intra_rd = INT64_MAX;
int64_t best_inter_rd = INT64_MAX;
+ unsigned int best_pred_sse = UINT_MAX;
PREDICTION_MODE best_intra_mode = DC_PRED;
MV_REFERENCE_FRAME best_inter_ref_frame = LAST_FRAME;
int rate_uv_intra[TX_SIZES], rate_uv_tokenonly[TX_SIZES];
@@ -2675,14 +2675,6 @@
}
}
- if (bsize > cpi->sf.max_intra_bsize) {
- const int all_intra_modes = (1 << THR_DC) | (1 << THR_TM) |
- (1 << THR_H_PRED) | (1 << THR_V_PRED) | (1 << THR_D135_PRED) |
- (1 << THR_D207_PRED) | (1 << THR_D153_PRED) | (1 << THR_D63_PRED) |
- (1 << THR_D117_PRED) | (1 << THR_D45_PRED);
- mode_skip_mask |= all_intra_modes;
- }
-
for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
int mode_excluded = 0;
int64_t this_rd = INT64_MAX;
@@ -2706,7 +2698,7 @@
// Look at the reference frame of the best mode so far and set the
// skip mask to look at a subset of the remaining modes.
if (mode_index == mode_skip_start && best_mode_index >= 0) {
- switch (vp9_mode_order[best_mode_index].ref_frame[0]) {
+ switch (best_mbmode.ref_frame[0]) {
case INTRA_FRAME:
break;
case LAST_FRAME:
@@ -2731,6 +2723,10 @@
continue;
}
+ if (bsize > cpi->sf.max_intra_bsize)
+ if (ref_frame == INTRA_FRAME)
+ continue;
+
if (mode_skip_mask & (1 << mode_index))
continue;
@@ -2795,7 +2791,7 @@
if ((mode_search_skip_flags & FLAG_SKIP_COMP_BESTINTRA) &&
best_mode_index >=0 &&
- vp9_mode_order[best_mode_index].ref_frame[0] == INTRA_FRAME)
+ best_mbmode.ref_frame[0] == INTRA_FRAME)
continue;
if ((mode_search_skip_flags & FLAG_SKIP_COMP_REFMISMATCH) &&
ref_frame != best_inter_ref_frame &&
@@ -2809,7 +2805,7 @@
if (ref_frame == INTRA_FRAME) {
if (cpi->sf.adaptive_mode_search)
- if ((x->source_variance << num_pels_log2_lookup[bsize]) > best_intra_rd)
+ if ((x->source_variance << num_pels_log2_lookup[bsize]) > best_pred_sse)
continue;
if (!(intra_y_mode_mask & (1 << this_mode)))
@@ -2827,7 +2823,7 @@
if ((mode_search_skip_flags & FLAG_SKIP_INTRA_BESTINTER) &&
(this_mode >= D45_PRED && this_mode <= TM_PRED)) {
if (best_mode_index >= 0 &&
- vp9_mode_order[best_mode_index].ref_frame[0] > INTRA_FRAME)
+ best_mbmode.ref_frame[0] > INTRA_FRAME)
continue;
}
if (mode_search_skip_flags & FLAG_SKIP_INTRA_DIRMISMATCH) {
@@ -2988,7 +2984,7 @@
mbmi->mv[0].as_int = 0;
max_plane = 1;
} else {
- best_intra_rd = x->pred_sse[ref_frame];
+ best_pred_sse = x->pred_sse[ref_frame];
}
*returnrate = rate2;
@@ -3125,7 +3121,7 @@
// If we used an estimate for the uv intra rd in the loop above...
if (cpi->sf.use_uv_intra_rd_estimate) {
// Do Intra UV best rd mode selection if best mode choice above was intra.
- if (vp9_mode_order[best_mode_index].ref_frame[0] == INTRA_FRAME) {
+ if (best_mbmode.ref_frame[0] == INTRA_FRAME) {
TX_SIZE uv_tx_size;
*mbmi = best_mbmode;
uv_tx_size = get_uv_tx_size(mbmi, &xd->plane[1]);
@@ -3175,7 +3171,6 @@
vp9_zero(best_tx_diff);
}
- set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
store_coding_context(x, ctx, best_mode_index, best_pred_diff,
best_tx_diff, best_filter_diff, best_mode_skippable);
@@ -3888,7 +3883,6 @@
vp9_zero(best_filter_diff);
}
- set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
store_coding_context(x, ctx, best_ref_index,
best_pred_diff, best_tx_diff, best_filter_diff, 0);
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index dbf4ae9..04f3cc2 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -104,15 +104,11 @@
: USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720) {
- sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
- sf->last_partitioning_redo_frequency = 3;
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
sf->adaptive_pred_interp_filter = 0;
} else {
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
- sf->last_partitioning_redo_frequency = 2;
- sf->lf_motion_threshold = NO_MOTION_THRESHOLD;
}
sf->reference_masking = 1;
@@ -123,8 +119,6 @@
sf->disable_filter_search_var_thresh = 100;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = CONSTRAIN_NEIGHBORING_MIN_MAX;
- sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
- sf->adjust_partitioning_from_last_frame = 1;
if (MIN(cm->width, cm->height) >= 720)
sf->partition_search_breakout_dist_thr = (1 << 24);
@@ -148,8 +142,6 @@
sf->cb_pred_filter_search = 1;
sf->alt_ref_search_fp = 1;
sf->motion_field_mode_search = !boosted;
- sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
- sf->last_partitioning_redo_frequency = 2;
sf->recode_loop = ALLOW_RECODE_KFMAXBW;
sf->adaptive_rd_thresh = 3;
sf->mode_skip_start = 6;
@@ -172,7 +164,6 @@
sf->mode_search_skip_flags |= FLAG_SKIP_COMP_REFMISMATCH |
FLAG_EARLY_TERMINATE;
sf->disable_filter_search_var_thresh = 200;
- sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->use_lp32x32fdct = 1;
sf->use_fast_coef_updates = ONE_LOOP_REDUCED;
sf->use_fast_coef_costing = 1;
diff --git a/vp9/encoder/vp9_variance.c b/vp9/encoder/vp9_variance.c
index eb5ae2e..afbb191 100644
--- a/vp9/encoder/vp9_variance.c
+++ b/vp9/encoder/vp9_variance.c
@@ -103,8 +103,9 @@
unsigned int vp9_get_mb_ss_c(const int16_t *src_ptr) {
unsigned int i, sum = 0;
- for (i = 0; i < 256; i++)
+ for (i = 0; i < 256; ++i) {
sum += src_ptr[i] * src_ptr[i];
+ }
return sum;
}
diff --git a/vp9/encoder/x86/vp9_variance_impl_sse2.asm b/vp9/encoder/x86/vp9_variance_impl_sse2.asm
deleted file mode 100644
index 6278f2a..0000000
--- a/vp9/encoder/x86/vp9_variance_impl_sse2.asm
+++ /dev/null
@@ -1,69 +0,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-;unsigned int vp9_get_mb_ss_sse2
-;(
-; short *src_ptr
-;)
-global sym(vp9_get_mb_ss_sse2) PRIVATE
-sym(vp9_get_mb_ss_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 1
- GET_GOT rbx
- push rsi
- push rdi
- sub rsp, 16
- ; end prolog
-
-
- mov rax, arg(0) ;[src_ptr]
- mov rcx, 8
- pxor xmm4, xmm4
-
-.NEXTROW:
- movdqa xmm0, [rax]
- movdqa xmm1, [rax+16]
- movdqa xmm2, [rax+32]
- movdqa xmm3, [rax+48]
- pmaddwd xmm0, xmm0
- pmaddwd xmm1, xmm1
- pmaddwd xmm2, xmm2
- pmaddwd xmm3, xmm3
-
- paddd xmm0, xmm1
- paddd xmm2, xmm3
- paddd xmm4, xmm0
- paddd xmm4, xmm2
-
- add rax, 0x40
- dec rcx
- ja .NEXTROW
-
- movdqa xmm3,xmm4
- psrldq xmm4,8
- paddd xmm4,xmm3
- movdqa xmm3,xmm4
- psrldq xmm4,4
- paddd xmm4,xmm3
- movq rax,xmm4
-
-
- ; begin epilog
- add rsp, 16
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
diff --git a/vp9/encoder/x86/vp9_variance_sse2.c b/vp9/encoder/x86/vp9_variance_sse2.c
index 1cc4bbc..b4d2b0a 100644
--- a/vp9/encoder/x86/vp9_variance_sse2.c
+++ b/vp9/encoder/x86/vp9_variance_sse2.c
@@ -19,6 +19,21 @@
const unsigned char *ref, int ref_stride,
unsigned int *sse, int *sum);
+unsigned int vp9_get_mb_ss_sse2(const int16_t *src) {
+ __m128i vsum = _mm_setzero_si128();
+ int i;
+
+ for (i = 0; i < 32; ++i) {
+ const __m128i v = _mm_loadu_si128((const __m128i *)src);
+ vsum = _mm_add_epi32(vsum, _mm_madd_epi16(v, v));
+ src += 8;
+ }
+
+ vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
+ vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
+ return _mm_cvtsi128_si32(vsum);
+}
+
#define READ64(p, stride, i) \
_mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const uint32_t *)(p + i * stride)), \
_mm_cvtsi32_si128(*(const uint32_t *)(p + (i + 1) * stride)))
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 8e3e885..90f0342 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -80,7 +80,6 @@
VP9_COMMON_SRCS-$(HAVE_AVX2) += common/x86/vp9_subpixel_8t_intrin_avx2.c
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_intrin_ssse3.c
ifeq ($(CONFIG_VP9_POSTPROC),yes)
-VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_postproc_sse2.asm
endif
diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk
index 312f717..e450f7b 100644
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -93,7 +93,6 @@
VP9_CX_SRCS-yes += encoder/vp9_mbgraph.c
VP9_CX_SRCS-yes += encoder/vp9_mbgraph.h
-VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_impl_sse2.asm
VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_variance_impl_intrin_avx2.c
VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad4d_sse2.asm
VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_sad4d_intrin_avx2.c
diff --git a/vpx/src/svc_encodeframe.c b/vpx/src/svc_encodeframe.c
index 8911e83..ea30b44 100644
--- a/vpx/src/svc_encodeframe.c
+++ b/vpx/src/svc_encodeframe.c
@@ -15,12 +15,12 @@
#include <assert.h>
#include <math.h>
+#include <limits.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#define VPX_DISABLE_CTRL_TYPECHECKS 1
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "./vpx_config.h"
#include "vpx/svc_context.h"
#include "vpx/vp8cx.h"
@@ -47,8 +47,33 @@
#define OPTION_BUFFER_SIZE 256
#define COMPONENTS 4 // psnr & sse statistics maintained for total, y, u, v
-static const char *DEFAULT_QUANTIZER_VALUES = "60,53,39,33,27";
-static const char *DEFAULT_SCALE_FACTORS = "4/16,5/16,7/16,11/16,16/16";
+static const int DEFAULT_QUANTIZER_VALUES[VPX_SS_MAX_LAYERS] = {
+ 60, 53, 39, 33, 27
+};
+
+static const int DEFAULT_SCALE_FACTORS_NUM[VPX_SS_MAX_LAYERS] = {
+ 4, 5, 7, 11, 16
+};
+
+static const int DEFAULT_SCALE_FACTORS_DEN[VPX_SS_MAX_LAYERS] = {
+ 16, 16, 16, 16, 16
+};
+
+typedef enum {
+ QUANTIZER = 0,
+ BITRATE,
+ SCALE_FACTOR,
+ AUTO_ALT_REF,
+ ALL_OPTION_TYPES
+} LAYER_OPTION_TYPE;
+
+static const int option_max_values[ALL_OPTION_TYPES] = {
+ 63, INT_MAX, INT_MAX, 1
+};
+
+static const int option_min_values[ALL_OPTION_TYPES] = {
+ 0, 0, 1, 0
+};
// One encoded frame
typedef struct FrameData {
@@ -68,6 +93,7 @@
int scaling_factor_den[VPX_SS_MAX_LAYERS];
int quantizer[VPX_SS_MAX_LAYERS];
int enable_auto_alt_ref[VPX_SS_MAX_LAYERS];
+ int bitrates[VPX_SS_MAX_LAYERS];
// accumulated statistics
double psnr_sum[VPX_SS_MAX_LAYERS][COMPONENTS]; // total/Y/U/V
@@ -197,158 +223,62 @@
return retval;
}
-static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
- const char *quantizer_values) {
- char *input_string;
- char *token;
- const char *delim = ",";
- char *save_ptr;
- int found = 0;
- int i, q;
- vpx_codec_err_t res = VPX_CODEC_OK;
- SvcInternal *const si = get_svc_internal(svc_ctx);
+static vpx_codec_err_t extract_option(LAYER_OPTION_TYPE type,
+ char *input,
+ int *value0,
+ int *value1) {
+ if (type == SCALE_FACTOR) {
+ *value0 = strtol(input, &input, 10);
+ if (*input++ != '/')
+ return VPX_CODEC_INVALID_PARAM;
+ *value1 = strtol(input, &input, 10);
- if (quantizer_values == NULL || strlen(quantizer_values) == 0) {
- input_string = strdup(DEFAULT_QUANTIZER_VALUES);
+ if (*value0 < option_min_values[SCALE_FACTOR] ||
+ *value1 < option_min_values[SCALE_FACTOR] ||
+ *value0 > option_max_values[SCALE_FACTOR] ||
+ *value1 > option_max_values[SCALE_FACTOR])
+ return VPX_CODEC_INVALID_PARAM;
} else {
- input_string = strdup(quantizer_values);
+ *value0 = atoi(input);
+ if (*value0 < option_min_values[type] ||
+ *value0 > option_max_values[type])
+ return VPX_CODEC_INVALID_PARAM;
}
-
- token = strtok_r(input_string, delim, &save_ptr);
- for (i = 0; i < svc_ctx->spatial_layers; ++i) {
- if (token != NULL) {
- q = atoi(token);
- if (q <= 0 || q > 100) {
- svc_log(svc_ctx, SVC_LOG_ERROR,
- "svc-quantizer-values: invalid value %s\n", token);
- res = VPX_CODEC_INVALID_PARAM;
- break;
- }
- token = strtok_r(NULL, delim, &save_ptr);
- found = i + 1;
- } else {
- q = 0;
- }
- si->quantizer[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] = q;
- }
- if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
- svc_log(svc_ctx, SVC_LOG_ERROR,
- "svc: quantizers: %d values required, but only %d specified\n",
- svc_ctx->spatial_layers, found);
- res = VPX_CODEC_INVALID_PARAM;
- }
- free(input_string);
- return res;
+ return VPX_CODEC_OK;
}
-static vpx_codec_err_t parse_auto_alt_ref(SvcContext *svc_ctx,
- const char *alt_ref_options) {
- char *input_string;
- char *token;
- const char *delim = ",";
- char *save_ptr;
- int found = 0, enabled = 0;
- int i, value;
- vpx_codec_err_t res = VPX_CODEC_OK;
- SvcInternal *const si = get_svc_internal(svc_ctx);
-
- if (alt_ref_options == NULL || strlen(alt_ref_options) == 0) {
- return VPX_CODEC_INVALID_PARAM;
- } else {
- input_string = strdup(alt_ref_options);
- }
-
- token = strtok_r(input_string, delim, &save_ptr);
- for (i = 0; i < svc_ctx->spatial_layers; ++i) {
- if (token != NULL) {
- value = atoi(token);
- if (value < 0 || value > 1) {
- svc_log(svc_ctx, SVC_LOG_ERROR,
- "enable auto alt ref values: invalid value %s\n", token);
- res = VPX_CODEC_INVALID_PARAM;
- break;
- }
- token = strtok_r(NULL, delim, &save_ptr);
- found = i + 1;
- } else {
- value = 0;
- }
- si->enable_auto_alt_ref[i] = value;
- if (value > 0)
- ++enabled;
- }
- if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
- svc_log(svc_ctx, SVC_LOG_ERROR,
- "svc: quantizers: %d values required, but only %d specified\n",
- svc_ctx->spatial_layers, found);
- res = VPX_CODEC_INVALID_PARAM;
- }
- if (enabled > REF_FRAMES - svc_ctx->spatial_layers) {
- svc_log(svc_ctx, SVC_LOG_ERROR,
- "svc: auto alt ref: Maxinum %d(REF_FRAMES - layers) layers could"
- "enabled auto alt reference frame, but % layers are enabled\n",
- REF_FRAMES - svc_ctx->spatial_layers, enabled);
- res = VPX_CODEC_INVALID_PARAM;
- }
- free(input_string);
- return res;
-}
-
-static void log_invalid_scale_factor(SvcContext *svc_ctx, const char *value) {
- svc_log(svc_ctx, SVC_LOG_ERROR, "svc scale-factors: invalid value %s\n",
- value);
-}
-
-static vpx_codec_err_t parse_scale_factors(SvcContext *svc_ctx,
- const char *scale_factors) {
- char *input_string;
- char *token;
- const char *delim = ",";
- char *save_ptr;
- int found = 0;
+static vpx_codec_err_t parse_layer_options_from_string(SvcContext *svc_ctx,
+ LAYER_OPTION_TYPE type,
+ const char *input,
+ int *option0,
+ int *option1) {
int i;
- int64_t num, den;
vpx_codec_err_t res = VPX_CODEC_OK;
- SvcInternal *const si = get_svc_internal(svc_ctx);
+ char *input_string;
+ char *token;
+ const char *delim = ",";
+ char *save_ptr;
- if (scale_factors == NULL || strlen(scale_factors) == 0) {
- input_string = strdup(DEFAULT_SCALE_FACTORS);
- } else {
- input_string = strdup(scale_factors);
- }
+ if (input == NULL || option0 == NULL ||
+ (option1 == NULL && type == SCALE_FACTOR))
+ return VPX_CODEC_INVALID_PARAM;
+
+ input_string = strdup(input);
token = strtok_r(input_string, delim, &save_ptr);
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
- num = den = 0;
if (token != NULL) {
- num = strtol(token, &token, 10);
- if (num <= 0) {
- log_invalid_scale_factor(svc_ctx, token);
- res = VPX_CODEC_INVALID_PARAM;
+ res = extract_option(type, token, option0 + i, option1 + i);
+ if (res != VPX_CODEC_OK)
break;
- }
- if (*token++ != '/') {
- log_invalid_scale_factor(svc_ctx, token);
- res = VPX_CODEC_INVALID_PARAM;
- break;
- }
- den = strtol(token, &token, 10);
- if (den <= 0) {
- log_invalid_scale_factor(svc_ctx, token);
- res = VPX_CODEC_INVALID_PARAM;
- break;
- }
token = strtok_r(NULL, delim, &save_ptr);
- found = i + 1;
+ } else {
+ break;
}
- si->scaling_factor_num[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
- (int)num;
- si->scaling_factor_den[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] =
- (int)den;
}
- if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
+ if (res == VPX_CODEC_OK && i != svc_ctx->spatial_layers) {
svc_log(svc_ctx, SVC_LOG_ERROR,
- "svc: scale-factors: %d values required, but only %d specified\n",
- svc_ctx->spatial_layers, found);
+ "svc: layer params type: %d %d values required, "
+ "but only %d specified\n", type, svc_ctx->spatial_layers, i);
res = VPX_CODEC_INVALID_PARAM;
}
free(input_string);
@@ -369,6 +299,7 @@
char *input_ptr;
SvcInternal *const si = get_svc_internal(svc_ctx);
vpx_codec_err_t res = VPX_CODEC_OK;
+ int i, alt_ref_enabled = 0;
if (options == NULL) return VPX_CODEC_OK;
input_string = strdup(options);
@@ -389,13 +320,21 @@
} else if (strcmp("temporal-layers", option_name) == 0) {
svc_ctx->temporal_layers = atoi(option_value);
} else if (strcmp("scale-factors", option_name) == 0) {
- res = parse_scale_factors(svc_ctx, option_value);
+ res = parse_layer_options_from_string(svc_ctx, SCALE_FACTOR, option_value,
+ si->scaling_factor_num,
+ si->scaling_factor_den);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("quantizers", option_name) == 0) {
- res = parse_quantizer_values(svc_ctx, option_value);
+ res = parse_layer_options_from_string(svc_ctx, QUANTIZER, option_value,
+ si->quantizer, NULL);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("auto-alt-refs", option_name) == 0) {
- res = parse_auto_alt_ref(svc_ctx, option_value);
+ res = parse_layer_options_from_string(svc_ctx, AUTO_ALT_REF, option_value,
+ si->enable_auto_alt_ref, NULL);
+ if (res != VPX_CODEC_OK) break;
+ } else if (strcmp("bitrates", option_name) == 0) {
+ res = parse_layer_options_from_string(svc_ctx, BITRATE, option_value,
+ si->bitrates, NULL);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("multi-frame-contexts", option_name) == 0) {
si->use_multiple_frame_contexts = atoi(option_value);
@@ -413,6 +352,16 @@
svc_ctx->spatial_layers * svc_ctx->temporal_layers > 4))
res = VPX_CODEC_INVALID_PARAM;
+ for (i = 0; i < svc_ctx->spatial_layers; ++i)
+ alt_ref_enabled += si->enable_auto_alt_ref[i];
+ if (alt_ref_enabled > REF_FRAMES - svc_ctx->spatial_layers) {
+ svc_log(svc_ctx, SVC_LOG_ERROR,
+ "svc: auto alt ref: Maxinum %d(REF_FRAMES - layers) layers could"
+ "enabled auto alt reference frame, but % layers are enabled\n",
+ REF_FRAMES - svc_ctx->spatial_layers, alt_ref_enabled);
+ res = VPX_CODEC_INVALID_PARAM;
+ }
+
return res;
}
@@ -448,6 +397,39 @@
return VPX_CODEC_OK;
}
+void assign_layer_bitrates(const SvcInternal *const si,
+ vpx_codec_enc_cfg_t *const enc_cfg) {
+ int i;
+
+ if (si->bitrates[0] != 0) {
+ enc_cfg->rc_target_bitrate = 0;
+ for (i = 0; i < si->layers; ++i) {
+ enc_cfg->ss_target_bitrate[i] = (unsigned int)si->bitrates[i];
+ enc_cfg->rc_target_bitrate += si->bitrates[i];
+ }
+ } else {
+ float total = 0;
+ float alloc_ratio[VPX_SS_MAX_LAYERS] = {0};
+
+ for (i = 0; i < si->layers; ++i) {
+ if (si->scaling_factor_den[i] > 0) {
+ alloc_ratio[i] = (float)(si->scaling_factor_num[i] * 1.0 /
+ si->scaling_factor_den[i]);
+
+ alloc_ratio[i] *= alloc_ratio[i];
+ total += alloc_ratio[i];
+ }
+ }
+
+ for (i = 0; i < si->layers; ++i) {
+ if (total > 0) {
+ enc_cfg->ss_target_bitrate[i] = (unsigned int)
+ (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total);
+ }
+ }
+ }
+}
+
vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
vpx_codec_iface_t *iface,
vpx_codec_enc_cfg_t *enc_cfg) {
@@ -481,11 +463,27 @@
return VPX_CODEC_INVALID_PARAM;
}
- res = parse_quantizer_values(svc_ctx, si->quantizers);
- if (res != VPX_CODEC_OK) return res;
+ for (i = 0; i < VPX_SS_MAX_LAYERS; ++i) {
+ si->quantizer[i] = DEFAULT_QUANTIZER_VALUES[i];
+ si->scaling_factor_num[i] = DEFAULT_SCALE_FACTORS_NUM[i];
+ si->scaling_factor_den[i] = DEFAULT_SCALE_FACTORS_DEN[i];
+ }
- res = parse_scale_factors(svc_ctx, si->scale_factors);
- if (res != VPX_CODEC_OK) return res;
+ if (strlen(si->quantizers) > 0) {
+ res = parse_layer_options_from_string(svc_ctx, QUANTIZER, si->quantizers,
+ si->quantizer, NULL);
+ if (res != VPX_CODEC_OK)
+ return res;
+ }
+
+ if (strlen(si->scale_factors) > 0) {
+ res = parse_layer_options_from_string(svc_ctx, SCALE_FACTOR,
+ si->scale_factors,
+ si->scaling_factor_num,
+ si->scaling_factor_den);
+ if (res != VPX_CODEC_OK)
+ return res;
+ }
// Parse aggregate command line options. Options must start with
// "layers=xx" then followed by other options
@@ -504,33 +502,7 @@
si->layers = svc_ctx->spatial_layers;
- // Assign target bitrate for each layer. We calculate the ratio
- // from the resolution for now.
- // TODO(Minghai): Optimize the mechanism of allocating bits after
- // implementing svc two pass rate control.
- if (si->layers > 1) {
- float total = 0;
- float alloc_ratio[VPX_SS_MAX_LAYERS] = {0};
-
- assert(si->layers <= VPX_SS_MAX_LAYERS);
- for (i = 0; i < si->layers; ++i) {
- int pos = i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers;
- if (pos < VPX_SS_MAX_LAYERS && si->scaling_factor_den[pos] > 0) {
- alloc_ratio[i] = (float)(si->scaling_factor_num[pos] * 1.0 /
- si->scaling_factor_den[pos]);
-
- alloc_ratio[i] *= alloc_ratio[i];
- total += alloc_ratio[i];
- }
- }
-
- for (i = 0; i < si->layers; ++i) {
- if (total > 0) {
- enc_cfg->ss_target_bitrate[i] = (unsigned int)
- (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total);
- }
- }
- }
+ assign_layer_bitrates(si, enc_cfg);
#if CONFIG_SPATIAL_SVC
for (i = 0; i < si->layers; ++i)
@@ -585,7 +557,7 @@
int layer,
unsigned int *width,
unsigned int *height) {
- int w, h, index, num, den;
+ int w, h, num, den;
const SvcInternal *const si = get_const_svc_internal(svc_ctx);
if (svc_ctx == NULL || si == NULL || width == NULL || height == NULL) {
@@ -593,9 +565,8 @@
}
if (layer < 0 || layer >= si->layers) return VPX_CODEC_INVALID_PARAM;
- index = layer + VPX_SS_MAX_LAYERS - si->layers;
- num = si->scaling_factor_num[index];
- den = si->scaling_factor_den[index];
+ num = si->scaling_factor_num[layer];
+ den = si->scaling_factor_den[layer];
if (num == 0 || den == 0) return VPX_CODEC_INVALID_PARAM;
w = si->width * num / den;
@@ -613,7 +584,7 @@
static void set_svc_parameters(SvcContext *svc_ctx,
vpx_codec_ctx_t *codec_ctx) {
- int layer, layer_index;
+ int layer;
vpx_svc_parameters_t svc_params;
SvcInternal *const si = get_svc_internal(svc_ctx);
@@ -627,11 +598,10 @@
&svc_params.height)) {
svc_log(svc_ctx, SVC_LOG_ERROR, "vpx_svc_get_layer_resolution failed\n");
}
- layer_index = layer + VPX_SS_MAX_LAYERS - si->layers;
if (codec_ctx->config.enc->g_pass == VPX_RC_ONE_PASS) {
- svc_params.min_quantizer = si->quantizer[layer_index];
- svc_params.max_quantizer = si->quantizer[layer_index];
+ svc_params.min_quantizer = si->quantizer[layer];
+ svc_params.max_quantizer = si->quantizer[layer];
} else {
svc_params.min_quantizer = codec_ctx->config.enc->rc_min_quantizer;
svc_params.max_quantizer = codec_ctx->config.enc->rc_max_quantizer;
diff --git a/vpx/vpx_image.h b/vpx/vpx_image.h
index 0b7bb90..ef6d1dd 100644
--- a/vpx/vpx_image.h
+++ b/vpx/vpx_image.h
@@ -64,40 +64,6 @@
VPX_IMG_FMT_I44416 = VPX_IMG_FMT_I444 | VPX_IMG_FMT_HIGHBITDEPTH
} vpx_img_fmt_t; /**< alias for enum vpx_img_fmt */
-#if !defined(VPX_CODEC_DISABLE_COMPAT) || !VPX_CODEC_DISABLE_COMPAT
-#define IMG_FMT_PLANAR VPX_IMG_FMT_PLANAR /**< \deprecated Use #VPX_IMG_FMT_PLANAR */
-#define IMG_FMT_UV_FLIP VPX_IMG_FMT_UV_FLIP /**< \deprecated Use #VPX_IMG_FMT_UV_FLIP */
-#define IMG_FMT_HAS_ALPHA VPX_IMG_FMT_HAS_ALPHA /**< \deprecated Use #VPX_IMG_FMT_HAS_ALPHA */
-
- /*!\brief Deprecated list of supported image formats
- * \deprecated New code should use #vpx_img_fmt
- */
-#define img_fmt vpx_img_fmt
- /*!\brief alias for enum img_fmt.
- * \deprecated New code should use #vpx_img_fmt_t
- */
-#define img_fmt_t vpx_img_fmt_t
-
-#define IMG_FMT_NONE VPX_IMG_FMT_NONE /**< \deprecated Use #VPX_IMG_FMT_NONE */
-#define IMG_FMT_RGB24 VPX_IMG_FMT_RGB24 /**< \deprecated Use #VPX_IMG_FMT_RGB24 */
-#define IMG_FMT_RGB32 VPX_IMG_FMT_RGB32 /**< \deprecated Use #VPX_IMG_FMT_RGB32 */
-#define IMG_FMT_RGB565 VPX_IMG_FMT_RGB565 /**< \deprecated Use #VPX_IMG_FMT_RGB565 */
-#define IMG_FMT_RGB555 VPX_IMG_FMT_RGB555 /**< \deprecated Use #VPX_IMG_FMT_RGB555 */
-#define IMG_FMT_UYVY VPX_IMG_FMT_UYVY /**< \deprecated Use #VPX_IMG_FMT_UYVY */
-#define IMG_FMT_YUY2 VPX_IMG_FMT_YUY2 /**< \deprecated Use #VPX_IMG_FMT_YUY2 */
-#define IMG_FMT_YVYU VPX_IMG_FMT_YVYU /**< \deprecated Use #VPX_IMG_FMT_YVYU */
-#define IMG_FMT_BGR24 VPX_IMG_FMT_BGR24 /**< \deprecated Use #VPX_IMG_FMT_BGR24 */
-#define IMG_FMT_RGB32_LE VPX_IMG_FMT_RGB32_LE /**< \deprecated Use #VPX_IMG_FMT_RGB32_LE */
-#define IMG_FMT_ARGB VPX_IMG_FMT_ARGB /**< \deprecated Use #VPX_IMG_FMT_ARGB */
-#define IMG_FMT_ARGB_LE VPX_IMG_FMT_ARGB_LE /**< \deprecated Use #VPX_IMG_FMT_ARGB_LE */
-#define IMG_FMT_RGB565_LE VPX_IMG_FMT_RGB565_LE /**< \deprecated Use #VPX_IMG_FMT_RGB565_LE */
-#define IMG_FMT_RGB555_LE VPX_IMG_FMT_RGB555_LE /**< \deprecated Use #VPX_IMG_FMT_RGB555_LE */
-#define IMG_FMT_YV12 VPX_IMG_FMT_YV12 /**< \deprecated Use #VPX_IMG_FMT_YV12 */
-#define IMG_FMT_I420 VPX_IMG_FMT_I420 /**< \deprecated Use #VPX_IMG_FMT_I420 */
-#define IMG_FMT_VPXYV12 VPX_IMG_FMT_VPXYV12 /**< \deprecated Use #VPX_IMG_FMT_VPXYV12 */
-#define IMG_FMT_VPXI420 VPX_IMG_FMT_VPXI420 /**< \deprecated Use #VPX_IMG_FMT_VPXI420 */
-#endif /* VPX_CODEC_DISABLE_COMPAT */
-
/**\brief Image Descriptor */
typedef struct vpx_image {
vpx_img_fmt_t fmt; /**< Image Format */
@@ -121,13 +87,6 @@
#define VPX_PLANE_U 1 /**< U (Chroma) plane */
#define VPX_PLANE_V 2 /**< V (Chroma) plane */
#define VPX_PLANE_ALPHA 3 /**< A (Transparency) plane */
-#if !defined(VPX_CODEC_DISABLE_COMPAT) || !VPX_CODEC_DISABLE_COMPAT
-#define PLANE_PACKED VPX_PLANE_PACKED
-#define PLANE_Y VPX_PLANE_Y
-#define PLANE_U VPX_PLANE_U
-#define PLANE_V VPX_PLANE_V
-#define PLANE_ALPHA VPX_PLANE_ALPHA
-#endif
unsigned char *planes[4]; /**< pointer to the top left pixel for each plane */
int stride[4]; /**< stride between rows for each plane */
diff --git a/vpx_ports/arm_cpudetect.c b/vpx_ports/arm_cpudetect.c
index fa0e030..f03feff 100644
--- a/vpx_ports/arm_cpudetect.c
+++ b/vpx_ports/arm_cpudetect.c
@@ -10,7 +10,8 @@
#include <stdlib.h>
#include <string.h>
-#include "arm.h"
+#include "vpx_ports/arm.h"
+#include "./vpx_config.h"
#ifdef WINAPI_FAMILY
#include <winapifamily.h>
@@ -54,9 +55,9 @@
#if HAVE_MEDIA
flags |= HAS_MEDIA;
#endif /* HAVE_MEDIA */
-#if HAVE_NEON
+#if HAVE_NEON || HAVE_NEON_ASM
flags |= HAS_NEON;
-#endif /* HAVE_NEON */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
return flags & mask;
}
@@ -87,6 +88,7 @@
/*Ignore exception.*/
}
}
+#endif /* HAVE_EDSP */
#if HAVE_MEDIA
if (mask & HAS_MEDIA)
__try {
@@ -97,7 +99,8 @@
/*Ignore exception.*/
}
}
-#if HAVE_NEON
+#endif /* HAVE_MEDIA */
+#if HAVE_NEON || HAVE_NEON_ASM
if (mask &HAS_NEON) {
__try {
/*VORR q0,q0,q0*/
@@ -107,9 +110,7 @@
/*Ignore exception.*/
}
}
-#endif /* HAVE_NEON */
-#endif /* HAVE_MEDIA */
-#endif /* HAVE_EDSP */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
return flags & mask;
}
@@ -132,10 +133,10 @@
#if HAVE_MEDIA
flags |= HAS_MEDIA;
#endif /* HAVE_MEDIA */
-#if HAVE_NEON
+#if HAVE_NEON || HAVE_NEON_ASM
if (features & ANDROID_CPU_ARM_FEATURE_NEON)
flags |= HAS_NEON;
-#endif /* HAVE_NEON */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
return flags & mask;
}
@@ -162,7 +163,7 @@
*/
char buf[512];
while (fgets(buf, 511, fin) != NULL) {
-#if HAVE_EDSP || HAVE_NEON
+#if HAVE_EDSP || HAVE_NEON || HAVE_NEON_ASM
if (memcmp(buf, "Features", 8) == 0) {
char *p;
#if HAVE_EDSP
@@ -170,15 +171,15 @@
if (p != NULL && (p[5] == ' ' || p[5] == '\n')) {
flags |= HAS_EDSP;
}
-#if HAVE_NEON
+#endif /* HAVE_EDSP */
+#if HAVE_NEON || HAVE_NEON_ASM
p = strstr(buf, " neon");
if (p != NULL && (p[5] == ' ' || p[5] == '\n')) {
flags |= HAS_NEON;
}
-#endif /* HAVE_NEON */
-#endif /* HAVE_EDSP */
+#endif /* HAVE_NEON || HAVE_NEON_ASM */
}
-#endif /* HAVE_EDSP || HAVE_NEON */
+#endif /* HAVE_EDSP || HAVE_NEON || HAVE_NEON_ASM */
#if HAVE_MEDIA
if (memcmp(buf, "CPU architecture:", 17) == 0) {
int version;
diff --git a/vpx_scale/yv12config.h b/vpx_scale/yv12config.h
index eb0a8d6..9ff764c 100644
--- a/vpx_scale/yv12config.h
+++ b/vpx_scale/yv12config.h
@@ -15,6 +15,7 @@
extern "C" {
#endif
+#include "vpx/vpx_codec.h"
#include "vpx/vpx_frame_buffer.h"
#include "vpx/vpx_integer.h"
@@ -50,6 +51,7 @@
int buffer_alloc_sz;
int border;
int frame_size;
+ unsigned int bit_depth;
int corrupted;
int flags;
diff --git a/vpxdec.c b/vpxdec.c
index 6470081..091522f 100644
--- a/vpxdec.c
+++ b/vpxdec.c
@@ -24,7 +24,6 @@
#include "./args.h"
#include "./ivfdec.h"
-#define VPX_CODEC_DISABLE_COMPAT 1
#include "vpx/vpx_decoder.h"
#include "vpx_ports/mem_ops.h"
#include "vpx_ports/vpx_timer.h"
diff --git a/vpxenc.c b/vpxenc.c
index b1156e1..5afca24 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -1202,7 +1202,7 @@
// Check that the codec bit depth is greater than the input bit depth.
if (stream->config.cfg.g_input_bit_depth >
- (int)stream->config.cfg.g_bit_depth) {
+ (unsigned int)stream->config.cfg.g_bit_depth) {
fatal("Stream %d: codec bit depth (%d) less than input bit depth (%d)",
stream->index, (int)stream->config.cfg.g_bit_depth,
stream->config.cfg.g_input_bit_depth);
diff --git a/y4minput.c b/y4minput.c
index bcc742a..34ea96d 100644
--- a/y4minput.c
+++ b/y4minput.c
@@ -1041,12 +1041,12 @@
c_w *= bytes_per_sample;
c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
c_sz = c_w * c_h;
- _img->stride[PLANE_Y] = _img->stride[PLANE_ALPHA] =
+ _img->stride[VPX_PLANE_Y] = _img->stride[VPX_PLANE_ALPHA] =
_y4m->pic_w * bytes_per_sample;
- _img->stride[PLANE_U] = _img->stride[PLANE_V] = c_w;
- _img->planes[PLANE_Y] = _y4m->dst_buf;
- _img->planes[PLANE_U] = _y4m->dst_buf + pic_sz;
- _img->planes[PLANE_V] = _y4m->dst_buf + pic_sz + c_sz;
- _img->planes[PLANE_ALPHA] = _y4m->dst_buf + pic_sz + 2 * c_sz;
+ _img->stride[VPX_PLANE_U] = _img->stride[VPX_PLANE_V] = c_w;
+ _img->planes[VPX_PLANE_Y] = _y4m->dst_buf;
+ _img->planes[VPX_PLANE_U] = _y4m->dst_buf + pic_sz;
+ _img->planes[VPX_PLANE_V] = _y4m->dst_buf + pic_sz + c_sz;
+ _img->planes[VPX_PLANE_ALPHA] = _y4m->dst_buf + pic_sz + 2 * c_sz;
return 1;
}
