Merge "Vidyo patch: Optimization for 1-to-2 downsampling and upsampling."
diff --git a/build/make/configure.sh b/build/make/configure.sh
index 98248b0..d888268 100644
--- a/build/make/configure.sh
+++ b/build/make/configure.sh
@@ -647,7 +647,7 @@
armv6*)
tgt_isa=armv6
;;
- armv7*-hardfloat*)
+ armv7*-hardfloat* | armv7*-gnueabihf | arm-*-gnueabihf)
tgt_isa=armv7
float_abi=hard
;;
@@ -877,7 +877,6 @@
case ${tgt_cc} in
gcc)
- CROSS=${CROSS:-arm-none-linux-gnueabi-}
link_with_cc=gcc
setup_gnu_toolchain
arch_int=${tgt_isa##armv}
@@ -1135,7 +1134,7 @@
CC=${CC:-${CROSS}gcc}
CXX=${CXX:-${CROSS}g++}
LD=${LD:-${CROSS}gcc}
- CROSS=${CROSS:-g}
+ CROSS=${CROSS-g}
;;
os2)
disable_feature pic
diff --git a/test/byte_alignment_test.cc b/test/byte_alignment_test.cc
index aa4b78b..3a808b0 100644
--- a/test/byte_alignment_test.cc
+++ b/test/byte_alignment_test.cc
@@ -21,14 +21,14 @@
namespace {
+#if CONFIG_WEBM_IO
+
const int kLegacyByteAlignment = 0;
const int kLegacyYPlaneByteAlignment = 32;
const int kNumPlanesToCheck = 3;
const char kVP9TestFile[] = "vp90-2-02-size-lf-1920x1080.webm";
const char kVP9Md5File[] = "vp90-2-02-size-lf-1920x1080.webm.md5";
-#if CONFIG_WEBM_IO
-
struct ByteAlignmentTestParam {
int byte_alignment;
vpx_codec_err_t expected_value;
diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index 332210d..0449b52 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -293,6 +293,7 @@
vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 12);
}
+#if HAVE_SSE2
void idct16x16_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
vpx_highbd_idct16x16_10_add_c(in, out, stride, 10);
}
@@ -301,7 +302,6 @@
vpx_highbd_idct16x16_10_add_c(in, out, stride, 12);
}
-#if HAVE_SSE2
void idct16x16_256_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
vpx_highbd_idct16x16_256_add_sse2(in, out, stride, 10);
}
diff --git a/test/dct32x32_test.cc b/test/dct32x32_test.cc
index f7327b1..2dac10b 100644
--- a/test/dct32x32_test.cc
+++ b/test/dct32x32_test.cc
@@ -81,10 +81,6 @@
Trans32x32Param;
#if CONFIG_VP9_HIGHBITDEPTH
-void idct32x32_8(const tran_low_t *in, uint8_t *out, int stride) {
- vpx_highbd_idct32x32_1024_add_c(in, out, stride, 8);
-}
-
void idct32x32_10(const tran_low_t *in, uint8_t *out, int stride) {
vpx_highbd_idct32x32_1024_add_c(in, out, stride, 10);
}
diff --git a/test/encode_api_test.cc b/test/encode_api_test.cc
new file mode 100644
index 0000000..a7200e6
--- /dev/null
+++ b/test/encode_api_test.cc
@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_config.h"
+#include "vpx/vp8cx.h"
+#include "vpx/vpx_encoder.h"
+
+namespace {
+
+#define NELEMENTS(x) static_cast<int>(sizeof(x) / sizeof(x[0]))
+
+TEST(EncodeAPI, InvalidParams) {
+ static const vpx_codec_iface_t *kCodecs[] = {
+#if CONFIG_VP8_ENCODER
+ &vpx_codec_vp8_cx_algo,
+#endif
+#if CONFIG_VP9_ENCODER
+ &vpx_codec_vp9_cx_algo,
+#endif
+#if CONFIG_VP10_ENCODER
+ &vpx_codec_vp10_cx_algo,
+#endif
+ };
+ uint8_t buf[1] = {0};
+ vpx_image_t img;
+ vpx_codec_ctx_t enc;
+ vpx_codec_enc_cfg_t cfg;
+
+ EXPECT_EQ(&img, vpx_img_wrap(&img, VPX_IMG_FMT_I420, 1, 1, 1, buf));
+
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM, vpx_codec_enc_init(NULL, NULL, NULL, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM, vpx_codec_enc_init(&enc, NULL, NULL, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM, vpx_codec_encode(NULL, NULL, 0, 0, 0, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM, vpx_codec_encode(NULL, &img, 0, 0, 0, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM, vpx_codec_destroy(NULL));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM,
+ vpx_codec_enc_config_default(NULL, NULL, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM,
+ vpx_codec_enc_config_default(NULL, &cfg, 0));
+ EXPECT_TRUE(vpx_codec_error(NULL) != NULL);
+
+ for (int i = 0; i < NELEMENTS(kCodecs); ++i) {
+ SCOPED_TRACE(vpx_codec_iface_name(kCodecs[i]));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM,
+ vpx_codec_enc_init(NULL, kCodecs[i], NULL, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM,
+ vpx_codec_enc_init(&enc, kCodecs[i], NULL, 0));
+ EXPECT_EQ(VPX_CODEC_INVALID_PARAM,
+ vpx_codec_enc_config_default(kCodecs[i], &cfg, 1));
+
+ EXPECT_EQ(VPX_CODEC_OK, vpx_codec_enc_config_default(kCodecs[i], &cfg, 0));
+ EXPECT_EQ(VPX_CODEC_OK, vpx_codec_enc_init(&enc, kCodecs[i], &cfg, 0));
+ EXPECT_EQ(VPX_CODEC_OK, vpx_codec_encode(&enc, NULL, 0, 0, 0, 0));
+
+ EXPECT_EQ(VPX_CODEC_OK, vpx_codec_destroy(&enc));
+ }
+}
+
+} // namespace
diff --git a/test/external_frame_buffer_test.cc b/test/external_frame_buffer_test.cc
index d02dca2..2570f44 100644
--- a/test/external_frame_buffer_test.cc
+++ b/test/external_frame_buffer_test.cc
@@ -24,7 +24,6 @@
namespace {
const int kVideoNameParam = 1;
-const char kVP9TestFile[] = "vp90-2-02-size-lf-1920x1080.webm";
struct ExternalFrameBuffer {
uint8_t *data;
@@ -155,6 +154,8 @@
ExternalFrameBuffer *ext_fb_list_;
};
+#if CONFIG_WEBM_IO
+
// Callback used by libvpx to request the application to return a frame
// buffer of at least |min_size| in bytes.
int get_vp9_frame_buffer(void *user_priv, size_t min_size,
@@ -197,6 +198,8 @@
return 0;
}
+#endif // CONFIG_WEBM_IO
+
// Class for testing passing in external frame buffers to libvpx.
class ExternalFrameBufferMD5Test
: public ::libvpx_test::DecoderTest,
@@ -278,6 +281,8 @@
};
#if CONFIG_WEBM_IO
+const char kVP9TestFile[] = "vp90-2-02-size-lf-1920x1080.webm";
+
// Class for testing passing in external frame buffers to libvpx.
class ExternalFrameBufferTest : public ::testing::Test {
protected:
diff --git a/test/fdct8x8_test.cc b/test/fdct8x8_test.cc
index c0deaf4..72d2aed 100644
--- a/test/fdct8x8_test.cc
+++ b/test/fdct8x8_test.cc
@@ -107,6 +107,8 @@
vp9_highbd_iht8x8_64_add_c(in, out, stride, tx_type, 12);
}
+#if HAVE_SSE2
+
void idct8x8_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
vpx_highbd_idct8x8_10_add_c(in, out, stride, 10);
}
@@ -115,7 +117,6 @@
vpx_highbd_idct8x8_10_add_c(in, out, stride, 12);
}
-#if HAVE_SSE2
void idct8x8_10_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
vpx_highbd_idct8x8_10_add_sse2(in, out, stride, 10);
}
diff --git a/test/sad_test.cc b/test/sad_test.cc
index a144cfc..3f0f74c 100644
--- a/test/sad_test.cc
+++ b/test/sad_test.cc
@@ -700,16 +700,6 @@
INSTANTIATE_TEST_CASE_P(MMX, SADTest, ::testing::ValuesIn(mmx_tests));
#endif // HAVE_MMX
-#if HAVE_SSE
-#if CONFIG_USE_X86INC
-const SadMxNx4Param x4d_sse_tests[] = {
- make_tuple(4, 8, &vpx_sad4x8x4d_sse, -1),
- make_tuple(4, 4, &vpx_sad4x4x4d_sse, -1),
-};
-INSTANTIATE_TEST_CASE_P(SSE, SADx4Test, ::testing::ValuesIn(x4d_sse_tests));
-#endif // CONFIG_USE_X86INC
-#endif // HAVE_SSE
-
#if HAVE_SSE2
#if CONFIG_USE_X86INC
const SadMxNParam sse2_tests[] = {
@@ -828,6 +818,8 @@
make_tuple(8, 16, &vpx_sad8x16x4d_sse2, -1),
make_tuple(8, 8, &vpx_sad8x8x4d_sse2, -1),
make_tuple(8, 4, &vpx_sad8x4x4d_sse2, -1),
+ make_tuple(4, 8, &vpx_sad4x8x4d_sse2, -1),
+ make_tuple(4, 4, &vpx_sad4x4x4d_sse2, -1),
#if CONFIG_VP9_HIGHBITDEPTH
make_tuple(64, 64, &vpx_highbd_sad64x64x4d_sse2, 8),
make_tuple(64, 32, &vpx_highbd_sad64x32x4d_sse2, 8),
diff --git a/test/test-data.mk b/test/test-data.mk
index 43c98bf..05a0885 100644
--- a/test/test-data.mk
+++ b/test/test-data.mk
@@ -562,6 +562,8 @@
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-03-size-226x224.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-03-size-226x226.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-03-size-226x226.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-03-size-352x288.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-03-size-352x288.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-05-resize.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-05-resize.ivf.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-06-bilinear.webm
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 357f377..a4ed174 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -550,6 +550,8 @@
83c6d8f2969b759e10e5c6542baca1265c874c29 *vp90-2-03-size-226x224.webm.md5
fe0af2ee47b1e5f6a66db369e2d7e9d870b38dce *vp90-2-03-size-226x226.webm
94ad19b8b699cea105e2ff18f0df2afd7242bcf7 *vp90-2-03-size-226x226.webm.md5
+52bc1dfd3a97b24d922eb8a31d07527891561f2a *vp90-2-03-size-352x288.webm
+3084d6d0a1eec22e85a394422fbc8faae58930a5 *vp90-2-03-size-352x288.webm.md5
b6524e4084d15b5d0caaa3d3d1368db30cbee69c *vp90-2-03-deltaq.webm
65f45ec9a55537aac76104818278e0978f94a678 *vp90-2-03-deltaq.webm.md5
4dbb87494c7f565ffc266c98d17d0d8c7a5c5aba *vp90-2-05-resize.ivf
diff --git a/test/test.mk b/test/test.mk
index 80b57e5..e8e8304 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -20,6 +20,7 @@
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += ../y4minput.h ../y4minput.c
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += aq_segment_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += datarate_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += encode_api_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += error_resilience_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += i420_video_source.h
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += resize_test.cc
diff --git a/test/test_vectors.cc b/test/test_vectors.cc
index 634ea44..c822479 100644
--- a/test/test_vectors.cc
+++ b/test/test_vectors.cc
@@ -180,7 +180,8 @@
"vp90-2-03-size-226x198.webm", "vp90-2-03-size-226x200.webm",
"vp90-2-03-size-226x202.webm", "vp90-2-03-size-226x208.webm",
"vp90-2-03-size-226x210.webm", "vp90-2-03-size-226x224.webm",
- "vp90-2-03-size-226x226.webm", "vp90-2-03-deltaq.webm",
+ "vp90-2-03-size-226x226.webm", "vp90-2-03-size-352x288.webm",
+ "vp90-2-03-deltaq.webm",
"vp90-2-05-resize.ivf", "vp90-2-06-bilinear.webm",
"vp90-2-07-frame_parallel.webm", "vp90-2-08-tile_1x2_frame_parallel.webm",
"vp90-2-08-tile_1x2.webm", "vp90-2-08-tile_1x4_frame_parallel.webm",
diff --git a/vp10/common/pred_common.c b/vp10/common/pred_common.c
index 2e79e0d..236ae54 100644
--- a/vp10/common/pred_common.c
+++ b/vp10/common/pred_common.c
@@ -17,8 +17,8 @@
int vp10_get_pred_context_switchable_interp(const MACROBLOCKD *xd) {
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int left_type = xd->left_available && is_inter_block(left_mbmi) ?
left_mbmi->interp_filter : SWITCHABLE_FILTERS;
@@ -70,8 +70,8 @@
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
if (has_above && has_left) { // both edges available
if (!has_second_ref(above_mbmi) && !has_second_ref(left_mbmi))
// neither edge uses comp pred (0/1)
@@ -114,8 +114,8 @@
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
const int fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
const int var_ref_idx = !fix_ref_idx;
@@ -194,8 +194,8 @@
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mbmi);
const int left_intra = !is_inter_block(left_mbmi);
@@ -261,8 +261,8 @@
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mbmi);
const int left_intra = !is_inter_block(left_mbmi);
diff --git a/vp10/decoder/decodemv.c b/vp10/decoder/decodemv.c
index a28ae55..01b796c 100644
--- a/vp10/decoder/decodemv.c
+++ b/vp10/decoder/decodemv.c
@@ -24,19 +24,6 @@
#include "vpx_dsp/vpx_dsp_common.h"
-static INLINE int read_uniform(vpx_reader *r, int n) {
- int l = get_unsigned_bits(n);
- int m = (1 << l) - n;
- int v = vpx_read_literal(r, l-1);
-
- assert(l != 0);
-
- if (v < m)
- return v;
- else
- return (v << 1) - m + vpx_read_literal(r, 1);
-}
-
static PREDICTION_MODE read_intra_mode(vpx_reader *r, const vpx_prob *p) {
return (PREDICTION_MODE)vpx_read_tree(r, vp10_intra_mode_tree, p);
}
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index ede8bb3..04ce61d 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -45,19 +45,6 @@
static const struct vp10_token inter_mode_encodings[INTER_MODES] =
{{2, 2}, {6, 3}, {0, 1}, {7, 3}};
-static INLINE void write_uniform(vpx_writer *w, int n, int v) {
- int l = get_unsigned_bits(n);
- int m = (1 << l) - n;
- if (l == 0)
- return;
- if (v < m) {
- vpx_write_literal(w, v, l - 1);
- } else {
- vpx_write_literal(w, m + ((v - m) >> 1), l - 1);
- vpx_write_literal(w, (v - m) & 1, 1);
- }
-}
-
static struct vp10_token ext_tx_encodings[TX_TYPES];
void vp10_encode_token_init() {
diff --git a/vp10/encoder/firstpass.c b/vp10/encoder/firstpass.c
index c41fa3e..bc1ce00 100644
--- a/vp10/encoder/firstpass.c
+++ b/vp10/encoder/firstpass.c
@@ -1613,7 +1613,6 @@
int mid_boost_bits = 0;
int mid_frame_idx;
unsigned char arf_buffer_indices[MAX_ACTIVE_ARFS];
- int alt_frame_index = frame_index;
key_frame = cpi->common.frame_type == KEY_FRAME;
@@ -1626,15 +1625,13 @@
gf_group->update_type[0] = OVERLAY_UPDATE;
gf_group->rf_level[0] = INTER_NORMAL;
gf_group->bit_allocation[0] = 0;
- gf_group->arf_update_idx[0] = arf_buffer_indices[0];
- gf_group->arf_ref_idx[0] = arf_buffer_indices[0];
} else {
gf_group->update_type[0] = GF_UPDATE;
gf_group->rf_level[0] = GF_ARF_STD;
gf_group->bit_allocation[0] = gf_arf_bits;
- gf_group->arf_update_idx[0] = arf_buffer_indices[0];
- gf_group->arf_ref_idx[0] = arf_buffer_indices[0];
}
+ gf_group->arf_update_idx[0] = arf_buffer_indices[0];
+ gf_group->arf_ref_idx[0] = arf_buffer_indices[0];
// Step over the golden frame / overlay frame
if (EOF == input_stats(twopass, &frame_stats))
@@ -1648,15 +1645,15 @@
// Store the bits to spend on the ARF if there is one.
if (rc->source_alt_ref_pending) {
- gf_group->update_type[alt_frame_index] = ARF_UPDATE;
- gf_group->rf_level[alt_frame_index] = GF_ARF_STD;
- gf_group->bit_allocation[alt_frame_index] = gf_arf_bits;
+ gf_group->update_type[frame_index] = ARF_UPDATE;
+ gf_group->rf_level[frame_index] = GF_ARF_STD;
+ gf_group->bit_allocation[frame_index] = gf_arf_bits;
- gf_group->arf_src_offset[alt_frame_index] =
+ gf_group->arf_src_offset[frame_index] =
(unsigned char)(rc->baseline_gf_interval - 1);
- gf_group->arf_update_idx[alt_frame_index] = arf_buffer_indices[0];
- gf_group->arf_ref_idx[alt_frame_index] =
+ gf_group->arf_update_idx[frame_index] = arf_buffer_indices[0];
+ gf_group->arf_ref_idx[frame_index] =
arf_buffer_indices[cpi->multi_arf_last_grp_enabled &&
rc->source_alt_ref_active];
++frame_index;
@@ -1819,10 +1816,10 @@
{
int int_max_q =
(int)(vp10_convert_qindex_to_q(twopass->active_worst_quality,
- cpi->common.bit_depth));
+ cpi->common.bit_depth));
int int_lbq =
(int)(vp10_convert_qindex_to_q(rc->last_boosted_qindex,
- cpi->common.bit_depth));
+ cpi->common.bit_depth));
active_min_gf_interval = rc->min_gf_interval + VPXMIN(2, int_max_q / 200);
if (active_min_gf_interval > rc->max_gf_interval)
active_min_gf_interval = rc->max_gf_interval;
@@ -1835,13 +1832,12 @@
// At high Q when there are few bits to spare we are better with a longer
// interval to spread the cost of the GF.
active_max_gf_interval = 12 + VPXMIN(4, (int_lbq / 6));
- if (active_max_gf_interval < active_min_gf_interval)
- active_max_gf_interval = active_min_gf_interval;
- if (active_max_gf_interval > rc->max_gf_interval)
- active_max_gf_interval = rc->max_gf_interval;
+ // We have: active_min_gf_interval <= rc->max_gf_interval
if (active_max_gf_interval < active_min_gf_interval)
active_max_gf_interval = active_min_gf_interval;
+ else if (active_max_gf_interval > rc->max_gf_interval)
+ active_max_gf_interval = rc->max_gf_interval;
}
}
@@ -2139,6 +2135,8 @@
return is_viable_kf;
}
+#define FRAMES_TO_CHECK_DECAY 8
+
static void find_next_key_frame(VP10_COMP *cpi, FIRSTPASS_STATS *this_frame) {
int i, j;
RATE_CONTROL *const rc = &cpi->rc;
@@ -2157,7 +2155,7 @@
double boost_score = 0.0;
double kf_mod_err = 0.0;
double kf_group_err = 0.0;
- double recent_loop_decay[8] = {1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0};
+ double recent_loop_decay[FRAMES_TO_CHECK_DECAY];
vp10_zero(next_frame);
@@ -2184,6 +2182,10 @@
kf_mod_err = calculate_modified_err(cpi, twopass, oxcf, this_frame);
+ // Initialize the decay rates for the recent frames to check
+ for (j = 0; j < FRAMES_TO_CHECK_DECAY; ++j)
+ recent_loop_decay[j] = 1.0;
+
// Find the next keyframe.
i = 0;
while (twopass->stats_in < twopass->stats_in_end &&
@@ -2210,9 +2212,9 @@
// We want to know something about the recent past... rather than
// as used elsewhere where we are concerned with decay in prediction
// quality since the last GF or KF.
- recent_loop_decay[i % 8] = loop_decay_rate;
+ recent_loop_decay[i % FRAMES_TO_CHECK_DECAY] = loop_decay_rate;
decay_accumulator = 1.0;
- for (j = 0; j < 8; ++j)
+ for (j = 0; j < FRAMES_TO_CHECK_DECAY; ++j)
decay_accumulator *= recent_loop_decay[j];
// Special check for transition or high motion followed by a
diff --git a/vp10/encoder/lookahead.c b/vp10/encoder/lookahead.c
index 431c631..dce0139 100644
--- a/vp10/encoder/lookahead.c
+++ b/vp10/encoder/lookahead.c
@@ -207,7 +207,7 @@
int drain) {
struct lookahead_entry *buf = NULL;
- if (ctx->sz && (drain || ctx->sz == ctx->max_sz - MAX_PRE_FRAMES)) {
+ if (ctx && ctx->sz && (drain || ctx->sz == ctx->max_sz - MAX_PRE_FRAMES)) {
buf = pop(ctx, &ctx->read_idx);
ctx->sz--;
}
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 90a716d..b1077cb 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -132,16 +132,6 @@
{{INTRA_FRAME, NONE}},
};
-static INLINE int write_uniform_cost(int n, int v) {
- int l = get_unsigned_bits(n), m = (1 << l) - n;
- if (l == 0)
- return 0;
- if (v < m)
- return (l - 1) * vp10_cost_bit(128, 0);
- else
- return l * vp10_cost_bit(128, 0);
-}
-
static void swap_block_ptr(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
int m, int n, int min_plane, int max_plane) {
int i;
diff --git a/vp10/encoder/subexp.c b/vp10/encoder/subexp.c
index d407477..eccee8e 100644
--- a/vp10/encoder/subexp.c
+++ b/vp10/encoder/subexp.c
@@ -143,8 +143,10 @@
vpx_prob *bestp,
vpx_prob upd,
int stepsize) {
- int i, old_b, new_b, update_b, savings, bestsavings, step;
+ int i, old_b, new_b, update_b, savings, bestsavings;
int newp;
+ const int step_sign = *bestp > oldp[PIVOT_NODE] ? -1 : 1;
+ const int step = stepsize * step_sign;
vpx_prob bestnewp, newplist[ENTROPY_NODES], oldplist[ENTROPY_NODES];
vp10_model_to_full_probs(oldp, oldplist);
memcpy(newplist, oldp, sizeof(vpx_prob) * UNCONSTRAINED_NODES);
@@ -155,41 +157,23 @@
bestsavings = 0;
bestnewp = oldp[PIVOT_NODE];
- if (*bestp > oldp[PIVOT_NODE]) {
- step = -stepsize;
- for (newp = *bestp; newp > oldp[PIVOT_NODE]; newp += step) {
- if (newp < 1 || newp > 255)
- continue;
- newplist[PIVOT_NODE] = newp;
- vp10_model_to_full_probs(newplist, newplist);
- for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
- new_b += cost_branch256(ct + 2 * i, newplist[i]);
- new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
- update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
- vp10_cost_upd256;
- savings = old_b - new_b - update_b;
- if (savings > bestsavings) {
- bestsavings = savings;
- bestnewp = newp;
- }
- }
- } else {
- step = stepsize;
- for (newp = *bestp; newp < oldp[PIVOT_NODE]; newp += step) {
- if (newp < 1 || newp > 255)
- continue;
- newplist[PIVOT_NODE] = newp;
- vp10_model_to_full_probs(newplist, newplist);
- for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
- new_b += cost_branch256(ct + 2 * i, newplist[i]);
- new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
- update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
- vp10_cost_upd256;
- savings = old_b - new_b - update_b;
- if (savings > bestsavings) {
- bestsavings = savings;
- bestnewp = newp;
- }
+ assert(stepsize > 0);
+
+ for (newp = *bestp; (newp - oldp[PIVOT_NODE]) * step_sign < 0;
+ newp += step) {
+ if (newp < 1 || newp > 255)
+ continue;
+ newplist[PIVOT_NODE] = newp;
+ vp10_model_to_full_probs(newplist, newplist);
+ for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
+ new_b += cost_branch256(ct + 2 * i, newplist[i]);
+ new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
+ update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
+ vp10_cost_upd256;
+ savings = old_b - new_b - update_b;
+ if (savings > bestsavings) {
+ bestsavings = savings;
+ bestnewp = newp;
}
}
diff --git a/vp8/common/reconintra4x4.h b/vp8/common/reconintra4x4.h
index 869841e..5dc5d13 100644
--- a/vp8/common/reconintra4x4.h
+++ b/vp8/common/reconintra4x4.h
@@ -17,8 +17,8 @@
extern "C" {
#endif
-static void intra_prediction_down_copy(MACROBLOCKD *xd,
- unsigned char *above_right_src)
+static INLINE void intra_prediction_down_copy(MACROBLOCKD *xd,
+ unsigned char *above_right_src)
{
int dst_stride = xd->dst.y_stride;
unsigned char *above_right_dst = xd->dst.y_buffer - dst_stride + 16;
diff --git a/vp8/common/threading.h b/vp8/common/threading.h
index bc99dbb..a433d03 100644
--- a/vp8/common/threading.h
+++ b/vp8/common/threading.h
@@ -24,7 +24,7 @@
/* Win32 */
#include <process.h>
#include <windows.h>
-#define THREAD_FUNCTION DWORD WINAPI
+#define THREAD_FUNCTION unsigned int __stdcall
#define THREAD_FUNCTION_RETURN DWORD
#define THREAD_SPECIFIC_INDEX DWORD
#define pthread_t HANDLE
diff --git a/vp8/encoder/lookahead.c b/vp8/encoder/lookahead.c
index ce2ce08..6623385 100644
--- a/vp8/encoder/lookahead.c
+++ b/vp8/encoder/lookahead.c
@@ -181,6 +181,7 @@
{
struct lookahead_entry* buf = NULL;
+ assert(ctx != NULL);
if(ctx->sz && (drain || ctx->sz == ctx->max_sz - 1))
{
buf = pop(ctx, &ctx->read_idx);
diff --git a/vp8/encoder/pickinter.c b/vp8/encoder/pickinter.c
index d0fff3f..0ea0632 100644
--- a/vp8/encoder/pickinter.c
+++ b/vp8/encoder/pickinter.c
@@ -36,6 +36,8 @@
extern unsigned int cnt_pm;
#endif
+#define MODEL_MODE 0
+
extern const int vp8_ref_frame_order[MAX_MODES];
extern const MB_PREDICTION_MODE vp8_mode_order[MAX_MODES];
@@ -45,18 +47,21 @@
// skin color classifier is defined.
// Fixed-point skin color model parameters.
-static const int skin_mean[2] = {7463, 9614}; // q6
+static const int skin_mean[5][2] =
+ {{7463, 9614}, {6400, 10240}, {7040, 10240}, {8320, 9280}, {6800, 9614}};
static const int skin_inv_cov[4] = {4107, 1663, 1663, 2157}; // q16
-static const int skin_threshold = 1570636; // q18
+static const int skin_threshold[2] = {1570636, 800000}; // q18
// Evaluates the Mahalanobis distance measure for the input CbCr values.
-static int evaluate_skin_color_difference(int cb, int cr)
-{
+static int evaluate_skin_color_difference(int cb, int cr, int idx) {
const int cb_q6 = cb << 6;
const int cr_q6 = cr << 6;
- const int cb_diff_q12 = (cb_q6 - skin_mean[0]) * (cb_q6 - skin_mean[0]);
- const int cbcr_diff_q12 = (cb_q6 - skin_mean[0]) * (cr_q6 - skin_mean[1]);
- const int cr_diff_q12 = (cr_q6 - skin_mean[1]) * (cr_q6 - skin_mean[1]);
+ const int cb_diff_q12 =
+ (cb_q6 - skin_mean[idx][0]) * (cb_q6 - skin_mean[idx][0]);
+ const int cbcr_diff_q12 =
+ (cb_q6 - skin_mean[idx][0]) * (cr_q6 - skin_mean[idx][1]);
+ const int cr_diff_q12 =
+ (cr_q6 - skin_mean[idx][1]) * (cr_q6 - skin_mean[idx][1]);
const int cb_diff_q2 = (cb_diff_q12 + (1 << 9)) >> 10;
const int cbcr_diff_q2 = (cbcr_diff_q12 + (1 << 9)) >> 10;
const int cr_diff_q2 = (cr_diff_q12 + (1 << 9)) >> 10;
@@ -67,6 +72,34 @@
return skin_diff;
}
+// Checks if the input yCbCr values corresponds to skin color.
+static int is_skin_color(int y, int cb, int cr)
+{
+ if (y < 40 || y > 220)
+ {
+ return 0;
+ }
+ else
+ {
+ if (MODEL_MODE == 0)
+ {
+ return (evaluate_skin_color_difference(cb, cr, 0) < skin_threshold[0]);
+ }
+ else
+ {
+ int i = 0;
+ for (; i < 5; i++)
+ {
+ if (evaluate_skin_color_difference(cb, cr, i) < skin_threshold[1])
+ {
+ return 1;
+ }
+ }
+ return 0;
+ }
+ }
+}
+
static int macroblock_corner_grad(unsigned char* signal, int stride,
int offsetx, int offsety, int sgnx, int sgny)
{
@@ -157,16 +190,6 @@
return 0;
}
-// Checks if the input yCbCr values corresponds to skin color.
-static int is_skin_color(int y, int cb, int cr)
-{
- if (y < 40 || y > 220)
- {
- return 0;
- }
- return (evaluate_skin_color_difference(cb, cr) < skin_threshold);
-}
-
int vp8_skip_fractional_mv_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
int_mv *bestmv, int_mv *ref_mv,
int error_per_bit,
diff --git a/vp8/vp8_cx_iface.c b/vp8/vp8_cx_iface.c
index c125ae8..257d2a0 100644
--- a/vp8/vp8_cx_iface.c
+++ b/vp8/vp8_cx_iface.c
@@ -1162,31 +1162,6 @@
return NULL;
}
-static vpx_codec_err_t vp8e_update_entropy(vpx_codec_alg_priv_t *ctx,
- va_list args)
-{
- int update = va_arg(args, int);
- vp8_update_entropy(ctx->cpi, update);
- return VPX_CODEC_OK;
-
-}
-
-static vpx_codec_err_t vp8e_update_reference(vpx_codec_alg_priv_t *ctx,
- va_list args)
-{
- int update = va_arg(args, int);
- vp8_update_reference(ctx->cpi, update);
- return VPX_CODEC_OK;
-}
-
-static vpx_codec_err_t vp8e_use_reference(vpx_codec_alg_priv_t *ctx,
- va_list args)
-{
- int reference_flag = va_arg(args, int);
- vp8_use_as_reference(ctx->cpi, reference_flag);
- return VPX_CODEC_OK;
-}
-
static vpx_codec_err_t vp8e_set_frame_flags(vpx_codec_alg_priv_t *ctx,
va_list args)
{
diff --git a/vp9/common/vp9_loopfilter.h b/vp9/common/vp9_loopfilter.h
index b2893a4..fca8830 100644
--- a/vp9/common/vp9_loopfilter.h
+++ b/vp9/common/vp9_loopfilter.h
@@ -69,6 +69,7 @@
struct loopfilter {
int filter_level;
+ int last_filt_level;
int sharpness_level;
int last_sharpness_level;
diff --git a/vp9/common/vp9_pred_common.c b/vp9/common/vp9_pred_common.c
index 1bb11ef..c201890 100644
--- a/vp9/common/vp9_pred_common.c
+++ b/vp9/common/vp9_pred_common.c
@@ -17,8 +17,8 @@
int vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd) {
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
const MODE_INFO *const left_mi = xd->left_mi;
const int left_type = xd->left_available && is_inter_block(left_mi) ?
left_mi->interp_filter : SWITCHABLE_FILTERS;
@@ -70,8 +70,8 @@
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
if (has_above && has_left) { // both edges available
if (!has_second_ref(above_mi) && !has_second_ref(left_mi))
// neither edge uses comp pred (0/1)
@@ -114,8 +114,8 @@
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
const int fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
const int var_ref_idx = !fix_ref_idx;
@@ -194,8 +194,8 @@
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mi);
const int left_intra = !is_inter_block(left_mi);
@@ -261,8 +261,8 @@
// Note:
// The mode info data structure has a one element border above and to the
- // left of the entries correpsonding to real macroblocks.
- // The prediction flags in these dummy entries are initialised to 0.
+ // left of the entries corresponding to real macroblocks.
+ // The prediction flags in these dummy entries are initialized to 0.
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mi);
const int left_intra = !is_inter_block(left_mi);
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 7d58cc6..39710c8 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -552,7 +552,7 @@
}
#endif // CONFIG_VP9_HIGHBITDEPTH
-static void dec_build_inter_predictors(VP9Decoder *const pbi, MACROBLOCKD *xd,
+static void dec_build_inter_predictors(VPxWorker *const worker, MACROBLOCKD *xd,
int plane, int bw, int bh, int x,
int y, int w, int h, int mi_x, int mi_y,
const InterpKernel *kernel,
@@ -662,8 +662,8 @@
// Wait until reference block is ready. Pad 7 more pixels as last 7
// pixels of each superblock row can be changed by next superblock row.
- if (pbi->frame_parallel_decode)
- vp9_frameworker_wait(pbi->frame_worker_owner, ref_frame_buf,
+ if (worker != NULL)
+ vp9_frameworker_wait(worker, ref_frame_buf,
VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
// Skip border extension if block is inside the frame.
@@ -689,11 +689,11 @@
} else {
// Wait until reference block is ready. Pad 7 more pixels as last 7
// pixels of each superblock row can be changed by next superblock row.
- if (pbi->frame_parallel_decode) {
- const int y1 = (y0_16 + (h - 1) * ys) >> SUBPEL_BITS;
- vp9_frameworker_wait(pbi->frame_worker_owner, ref_frame_buf,
- VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
- }
+ if (worker != NULL) {
+ const int y1 = (y0_16 + (h - 1) * ys) >> SUBPEL_BITS;
+ vp9_frameworker_wait(worker, ref_frame_buf,
+ VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
+ }
}
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
@@ -720,53 +720,60 @@
const BLOCK_SIZE sb_type = mi->sb_type;
const int is_compound = has_second_ref(mi);
int ref;
+ int is_scaled;
+ VPxWorker *const fwo = pbi->frame_parallel_decode ?
+ pbi->frame_worker_owner : NULL;
for (ref = 0; ref < 1 + is_compound; ++ref) {
const MV_REFERENCE_FRAME frame = mi->ref_frame[ref];
RefBuffer *ref_buf = &pbi->common.frame_refs[frame - LAST_FRAME];
+ const struct scale_factors *const sf = &ref_buf->sf;
+ const int idx = ref_buf->idx;
+ BufferPool *const pool = pbi->common.buffer_pool;
+ RefCntBuffer *const ref_frame_buf = &pool->frame_bufs[idx];
- xd->block_refs[ref] = ref_buf;
- if (!vp9_is_valid_scale(&ref_buf->sf))
+ if (!vp9_is_valid_scale(sf))
vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
"Reference frame has invalid dimensions");
- vp9_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, &ref_buf->sf);
- }
- for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
- struct macroblockd_plane *const pd = &xd->plane[plane];
- struct buf_2d *const dst_buf = &pd->dst;
- const int num_4x4_w = pd->n4_w;
- const int num_4x4_h = pd->n4_h;
+ is_scaled = vp9_is_scaled(sf);
+ vp9_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, sf);
+ xd->block_refs[ref] = ref_buf;
- const int n4w_x4 = 4 * num_4x4_w;
- const int n4h_x4 = 4 * num_4x4_h;
- int ref;
-
- for (ref = 0; ref < 1 + is_compound; ++ref) {
- const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
- struct buf_2d *const pre_buf = &pd->pre[ref];
- const int idx = xd->block_refs[ref]->idx;
- BufferPool *const pool = pbi->common.buffer_pool;
- RefCntBuffer *const ref_frame_buf = &pool->frame_bufs[idx];
- const int is_scaled = vp9_is_scaled(sf);
-
- if (sb_type < BLOCK_8X8) {
+ if (sb_type < BLOCK_8X8) {
+ for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
+ struct macroblockd_plane *const pd = &xd->plane[plane];
+ struct buf_2d *const dst_buf = &pd->dst;
+ const int num_4x4_w = pd->n4_w;
+ const int num_4x4_h = pd->n4_h;
+ const int n4w_x4 = 4 * num_4x4_w;
+ const int n4h_x4 = 4 * num_4x4_h;
+ struct buf_2d *const pre_buf = &pd->pre[ref];
int i = 0, x, y;
for (y = 0; y < num_4x4_h; ++y) {
for (x = 0; x < num_4x4_w; ++x) {
const MV mv = average_split_mvs(pd, mi, ref, i++);
- dec_build_inter_predictors(pbi, xd, plane, n4w_x4, n4h_x4,
+ dec_build_inter_predictors(fwo, xd, plane, n4w_x4, n4h_x4,
4 * x, 4 * y, 4, 4, mi_x, mi_y, kernel,
sf, pre_buf, dst_buf, &mv,
ref_frame_buf, is_scaled, ref);
}
}
- } else {
- const MV mv = mi->mv[ref].as_mv;
- dec_build_inter_predictors(pbi, xd, plane, n4w_x4, n4h_x4,
+ }
+ } else {
+ const MV mv = mi->mv[ref].as_mv;
+ for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
+ struct macroblockd_plane *const pd = &xd->plane[plane];
+ struct buf_2d *const dst_buf = &pd->dst;
+ const int num_4x4_w = pd->n4_w;
+ const int num_4x4_h = pd->n4_h;
+ const int n4w_x4 = 4 * num_4x4_w;
+ const int n4h_x4 = 4 * num_4x4_h;
+ struct buf_2d *const pre_buf = &pd->pre[ref];
+ dec_build_inter_predictors(fwo, xd, plane, n4w_x4, n4h_x4,
0, 0, n4w_x4, n4h_x4, mi_x, mi_y, kernel,
- sf, pre_buf, dst_buf, &mv, ref_frame_buf,
- is_scaled, ref);
+ sf, pre_buf, dst_buf, &mv,
+ ref_frame_buf, is_scaled, ref);
}
}
}
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index c2fd178..743ab39 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -419,7 +419,7 @@
static INLINE int assign_mv(VP9_COMMON *cm, MACROBLOCKD *xd,
PREDICTION_MODE mode,
int_mv mv[2], int_mv ref_mv[2],
- int_mv nearest_mv[2], int_mv near_mv[2],
+ int_mv near_nearest_mv[2],
int is_compound, int allow_hp, vpx_reader *r) {
int i;
int ret = 1;
@@ -435,22 +435,15 @@
}
break;
}
+ case NEARMV:
case NEARESTMV: {
- mv[0].as_int = nearest_mv[0].as_int;
- if (is_compound)
- mv[1].as_int = nearest_mv[1].as_int;
- break;
- }
- case NEARMV: {
- mv[0].as_int = near_mv[0].as_int;
- if (is_compound)
- mv[1].as_int = near_mv[1].as_int;
+ mv[0].as_int = near_nearest_mv[0].as_int;
+ mv[1].as_int = near_nearest_mv[1].as_int;
break;
}
case ZEROMV: {
mv[0].as_int = 0;
- if (is_compound)
- mv[1].as_int = 0;
+ mv[1].as_int = 0;
break;
}
default: {
@@ -475,17 +468,15 @@
}
static void dec_find_best_ref_mvs(MACROBLOCKD *xd, int allow_hp, int_mv *mvlist,
- int_mv *nearest_mv, int_mv *near_mv,
- int refmv_count) {
+ int_mv *best_mv, int refmv_count) {
int i;
// Make sure all the candidates are properly clamped etc
for (i = 0; i < refmv_count; ++i) {
lower_mv_precision(&mvlist[i].as_mv, allow_hp);
clamp_mv2(&mvlist[i].as_mv, xd);
+ *best_mv = mvlist[i];
}
- *nearest_mv = mvlist[0];
- *near_mv = mvlist[1];
}
static void fpm_sync(void *const data, int mi_row) {
@@ -532,10 +523,10 @@
// This function searches the neighborhood of a given MB/SB
// to try and find candidate reference vectors.
static int dec_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd,
- MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
+ PREDICTION_MODE mode, MV_REFERENCE_FRAME ref_frame,
const POSITION *const mv_ref_search,
int_mv *mv_ref_list,
- int mi_row, int mi_col,
+ int mi_row, int mi_col, int block, int is_sub8x8,
find_mv_refs_sync sync, void *const data) {
const int *ref_sign_bias = cm->ref_frame_sign_bias;
int i, refmv_count = 0;
@@ -545,16 +536,38 @@
const TileInfo *const tile = &xd->tile;
// If mode is nearestmv or newmv (uses nearestmv as a reference) then stop
// searching after the first mv is found.
- const int early_break = (mi->mode == NEARESTMV) ||
- (mi->mode == NEWMV);
+ const int early_break = (mode != NEARMV);
// Blank the reference vector list
memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES);
+ i = 0;
+ if (is_sub8x8) {
+ // If the size < 8x8 we get the mv from the bmi substructure for the
+ // nearest two blocks.
+ for (i = 0; i < 2; ++i) {
+ const POSITION *const mv_ref = &mv_ref_search[i];
+ if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
+ const MODE_INFO *const candidate_mi =
+ xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride];
+ different_ref_found = 1;
+
+ if (candidate_mi->ref_frame[0] == ref_frame)
+ ADD_MV_REF_LIST_EB(
+ get_sub_block_mv(candidate_mi, 0, mv_ref->col, block),
+ refmv_count, mv_ref_list, Done);
+ else if (candidate_mi->ref_frame[1] == ref_frame)
+ ADD_MV_REF_LIST_EB(
+ get_sub_block_mv(candidate_mi, 1, mv_ref->col, block),
+ refmv_count, mv_ref_list, Done);
+ }
+ }
+ }
+
// Check the rest of the neighbors in much the same way
// as before except we don't need to keep track of sub blocks or
// mode counts.
- for (i = 0; i < MVREF_NEIGHBOURS; ++i) {
+ for (; i < MVREF_NEIGHBOURS; ++i) {
const POSITION *const mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MODE_INFO *const candidate =
@@ -635,7 +648,7 @@
}
}
- if (mi->mode == NEARMV)
+ if (mode == NEARMV)
refmv_count = MAX_MV_REF_CANDIDATES;
else
// we only care about the nearestmv for the remaining modes
@@ -649,6 +662,62 @@
return refmv_count;
}
+static void append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
+ const POSITION *const mv_ref_search,
+ PREDICTION_MODE b_mode, int block,
+ int ref, int mi_row, int mi_col,
+ int_mv *best_sub8x8) {
+ int_mv mv_list[MAX_MV_REF_CANDIDATES];
+ MODE_INFO *const mi = xd->mi[0];
+ b_mode_info *bmi = mi->bmi;
+ int n;
+ int refmv_count;
+
+ assert(MAX_MV_REF_CANDIDATES == 2);
+
+ refmv_count = dec_find_mv_refs(cm, xd, b_mode, mi->ref_frame[ref],
+ mv_ref_search, mv_list, mi_row, mi_col, block,
+ 1, NULL, NULL);
+
+ switch (block) {
+ case 0:
+ best_sub8x8->as_int = mv_list[refmv_count - 1].as_int;
+ break;
+ case 1:
+ case 2:
+ if (b_mode == NEARESTMV) {
+ best_sub8x8->as_int = bmi[0].as_mv[ref].as_int;
+ } else {
+ best_sub8x8->as_int = 0;
+ for (n = 0; n < refmv_count; ++n)
+ if (bmi[0].as_mv[ref].as_int != mv_list[n].as_int) {
+ best_sub8x8->as_int = mv_list[n].as_int;
+ break;
+ }
+ }
+ break;
+ case 3:
+ if (b_mode == NEARESTMV) {
+ best_sub8x8->as_int = bmi[2].as_mv[ref].as_int;
+ } else {
+ int_mv candidates[2 + MAX_MV_REF_CANDIDATES];
+ candidates[0] = bmi[1].as_mv[ref];
+ candidates[1] = bmi[0].as_mv[ref];
+ candidates[2] = mv_list[0];
+ candidates[3] = mv_list[1];
+ best_sub8x8->as_int = 0;
+ for (n = 0; n < 2 + MAX_MV_REF_CANDIDATES; ++n)
+ if (bmi[2].as_mv[ref].as_int != candidates[n].as_int) {
+ best_sub8x8->as_int = candidates[n].as_int;
+ break;
+ }
+ }
+ break;
+ default:
+ assert(0 && "Invalid block index.");
+ }
+}
+
static uint8_t get_mode_context(const VP9_COMMON *cm, const MACROBLOCKD *xd,
const POSITION *const mv_ref_search,
int mi_row, int mi_col) {
@@ -677,7 +746,7 @@
VP9_COMMON *const cm = &pbi->common;
const BLOCK_SIZE bsize = mi->sb_type;
const int allow_hp = cm->allow_high_precision_mv;
- int_mv nearestmv[2], nearmv[2];
+ int_mv best_ref_mvs[2];
int ref, is_compound;
uint8_t inter_mode_ctx;
const POSITION *const mv_ref_search = mv_ref_blocks[bsize];
@@ -705,16 +774,16 @@
if (mi->mode != ZEROMV) {
for (ref = 0; ref < 1 + is_compound; ++ref) {
- int_mv ref_mvs[MAX_MV_REF_CANDIDATES];
+ int_mv tmp_mvs[MAX_MV_REF_CANDIDATES];
const MV_REFERENCE_FRAME frame = mi->ref_frame[ref];
int refmv_count;
- refmv_count = dec_find_mv_refs(cm, xd, mi, frame, mv_ref_search,
- ref_mvs, mi_row, mi_col, fpm_sync,
- (void *)pbi);
+ refmv_count = dec_find_mv_refs(cm, xd, mi->mode, frame, mv_ref_search,
+ tmp_mvs, mi_row, mi_col, -1, 0,
+ fpm_sync, (void *)pbi);
- dec_find_best_ref_mvs(xd, allow_hp, ref_mvs, &nearestmv[ref],
- &nearmv[ref], refmv_count);
+ dec_find_best_ref_mvs(xd, allow_hp, tmp_mvs, &best_ref_mvs[ref],
+ refmv_count);
}
}
}
@@ -728,33 +797,24 @@
const int num_4x4_h = 1 << xd->bmode_blocks_hl;
int idx, idy;
PREDICTION_MODE b_mode;
- int_mv nearest_sub8x8[2], near_sub8x8[2];
+ int_mv best_sub8x8[2];
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
- int_mv block[2];
const int j = idy * 2 + idx;
b_mode = read_inter_mode(cm, xd, r, inter_mode_ctx);
if (b_mode == NEARESTMV || b_mode == NEARMV) {
- uint8_t dummy_mode_ctx[MAX_REF_FRAMES];
for (ref = 0; ref < 1 + is_compound; ++ref)
- vp9_append_sub8x8_mvs_for_idx(cm, xd, j, ref, mi_row, mi_col,
- &nearest_sub8x8[ref],
- &near_sub8x8[ref],
- dummy_mode_ctx);
+ append_sub8x8_mvs_for_idx(cm, xd, mv_ref_search, b_mode, j, ref,
+ mi_row, mi_col, &best_sub8x8[ref]);
}
- if (!assign_mv(cm, xd, b_mode, block, nearestmv,
- nearest_sub8x8, near_sub8x8,
- is_compound, allow_hp, r)) {
+ if (!assign_mv(cm, xd, b_mode, mi->bmi[j].as_mv, best_ref_mvs,
+ best_sub8x8, is_compound, allow_hp, r)) {
xd->corrupted |= 1;
break;
}
- mi->bmi[j].as_mv[0].as_int = block[0].as_int;
- if (is_compound)
- mi->bmi[j].as_mv[1].as_int = block[1].as_int;
-
if (num_4x4_h == 2)
mi->bmi[j + 2] = mi->bmi[j];
if (num_4x4_w == 2)
@@ -767,8 +827,8 @@
mi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
mi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
} else {
- xd->corrupted |= !assign_mv(cm, xd, mi->mode, mi->mv, nearestmv,
- nearestmv, nearmv, is_compound, allow_hp, r);
+ xd->corrupted |= !assign_mv(cm, xd, mi->mode, mi->mv, best_ref_mvs,
+ best_ref_mvs, is_compound, allow_hp, r);
}
}
diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index dbcf2be..dcc75b9 100644
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -23,14 +23,6 @@
#define EOB_CONTEXT_NODE 0
#define ZERO_CONTEXT_NODE 1
#define ONE_CONTEXT_NODE 2
-#define LOW_VAL_CONTEXT_NODE 0
-#define TWO_CONTEXT_NODE 1
-#define THREE_CONTEXT_NODE 2
-#define HIGH_LOW_CONTEXT_NODE 3
-#define CAT_ONE_CONTEXT_NODE 4
-#define CAT_THREEFOUR_CONTEXT_NODE 5
-#define CAT_THREE_CONTEXT_NODE 6
-#define CAT_FIVE_CONTEXT_NODE 7
#define INCREMENT_COUNT(token) \
do { \
diff --git a/vp9/encoder/vp9_aq_360.c b/vp9/encoder/vp9_aq_360.c
index 7f93734..f8c187c 100644
--- a/vp9/encoder/vp9_aq_360.c
+++ b/vp9/encoder/vp9_aq_360.c
@@ -21,12 +21,6 @@
#include "vp9/encoder/vp9_rd.h"
#include "vp9/encoder/vp9_segmentation.h"
-#define ENERGY_MIN (-4)
-#define ENERGY_MAX (1)
-#define ENERGY_SPAN (ENERGY_MAX - ENERGY_MIN + 1)
-#define ENERGY_IN_BOUNDS(energy)\
- assert((energy) >= ENERGY_MIN && (energy) <= ENERGY_MAX)
-
static const double rate_ratio[MAX_SEGMENTS] =
{1.0, 0.75, 0.6, 0.5, 0.4, 0.3, 0.25};
diff --git a/vp9/encoder/vp9_aq_complexity.c b/vp9/encoder/vp9_aq_complexity.c
index e597322..2d979ec 100644
--- a/vp9/encoder/vp9_aq_complexity.c
+++ b/vp9/encoder/vp9_aq_complexity.c
@@ -35,9 +35,6 @@
{-3.5, -2.5, -1.5, 100.00, 100.0},
{-3.0, -2.0, -1.0, 100.00, 100.0} };
-#define DEFAULT_COMPLEXITY 64
-
-
static int get_aq_c_strength(int q_index, vpx_bit_depth_t bit_depth) {
// Approximate base quatizer (truncated to int)
const int base_quant = vp9_ac_quant(q_index, 0, bit_depth) / 4;
@@ -107,7 +104,6 @@
#define DEFAULT_LV_THRESH 10.0
#define MIN_DEFAULT_LV_THRESH 8.0
-#define VAR_STRENGTH_STEP 0.25
// Select a segment for the current block.
// The choice of segment for a block depends on the ratio of the projected
// bits for the block vs a target average and its spatial complexity.
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index de09519..0982fd6 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -210,7 +210,7 @@
int is_skin = 0;
if (refresh_this_block == 0 &&
bsize <= BLOCK_16X16 &&
- cpi->oxcf.content != VP9E_CONTENT_SCREEN) {
+ cpi->use_skin_detection) {
is_skin = vp9_compute_skin_block(p[0].src.buf,
p[1].src.buf,
p[2].src.buf,
@@ -321,11 +321,11 @@
void vp9_cyclic_refresh_set_golden_update(VP9_COMP *const cpi) {
RATE_CONTROL *const rc = &cpi->rc;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
- // Set minimum gf_interval for GF update to a multiple (== 2) of refresh
- // period. Depending on past encoding stats, GF flag may be reset and update
- // may not occur until next baseline_gf_interval.
+ // Set minimum gf_interval for GF update to a multiple of the refresh period,
+ // with some max limit. Depending on past encoding stats, GF flag may be
+ // reset and update may not occur until next baseline_gf_interval.
if (cr->percent_refresh > 0)
- rc->baseline_gf_interval = 4 * (100 / cr->percent_refresh);
+ rc->baseline_gf_interval = VPXMIN(4 * (100 / cr->percent_refresh), 40);
else
rc->baseline_gf_interval = 40;
}
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index ca5d746..3eaa9de 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -137,6 +137,8 @@
// the visual quality at the boundary of moving color objects.
uint8_t color_sensitivity[2];
+ uint8_t sb_is_skin;
+
void (*fwd_txm4x4)(const int16_t *input, tran_low_t *output, int stride);
void (*itxm_add)(const tran_low_t *input, uint8_t *dest, int stride, int eob);
#if CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_cost.c b/vp9/encoder/vp9_cost.c
index e2fbb34..c85f763 100644
--- a/vp9/encoder/vp9_cost.c
+++ b/vp9/encoder/vp9_cost.c
@@ -11,29 +11,32 @@
#include "vp9/encoder/vp9_cost.h"
-const unsigned int vp9_prob_cost[256] = {
- 2047, 2047, 1791, 1641, 1535, 1452, 1385, 1328, 1279, 1235, 1196, 1161,
- 1129, 1099, 1072, 1046, 1023, 1000, 979, 959, 940, 922, 905, 889,
- 873, 858, 843, 829, 816, 803, 790, 778, 767, 755, 744, 733,
- 723, 713, 703, 693, 684, 675, 666, 657, 649, 641, 633, 625,
- 617, 609, 602, 594, 587, 580, 573, 567, 560, 553, 547, 541,
- 534, 528, 522, 516, 511, 505, 499, 494, 488, 483, 477, 472,
- 467, 462, 457, 452, 447, 442, 437, 433, 428, 424, 419, 415,
- 410, 406, 401, 397, 393, 389, 385, 381, 377, 373, 369, 365,
- 361, 357, 353, 349, 346, 342, 338, 335, 331, 328, 324, 321,
- 317, 314, 311, 307, 304, 301, 297, 294, 291, 288, 285, 281,
- 278, 275, 272, 269, 266, 263, 260, 257, 255, 252, 249, 246,
- 243, 240, 238, 235, 232, 229, 227, 224, 221, 219, 216, 214,
- 211, 208, 206, 203, 201, 198, 196, 194, 191, 189, 186, 184,
- 181, 179, 177, 174, 172, 170, 168, 165, 163, 161, 159, 156,
- 154, 152, 150, 148, 145, 143, 141, 139, 137, 135, 133, 131,
- 129, 127, 125, 123, 121, 119, 117, 115, 113, 111, 109, 107,
- 105, 103, 101, 99, 97, 95, 93, 92, 90, 88, 86, 84,
- 82, 81, 79, 77, 75, 73, 72, 70, 68, 66, 65, 63,
- 61, 60, 58, 56, 55, 53, 51, 50, 48, 46, 45, 43,
- 41, 40, 38, 37, 35, 33, 32, 30, 29, 27, 25, 24,
- 22, 21, 19, 18, 16, 15, 13, 12, 10, 9, 7, 6,
- 4, 3, 1, 1};
+/* round(-log2(i/256.) * (1 << VP9_PROB_COST_SHIFT))
+ Begins and ends with a bogus entry to satisfy use of prob=0 in the firstpass.
+ https://code.google.com/p/webm/issues/detail?id=1089 */
+const uint16_t vp9_prob_cost[257] = {
+ 4096, 4096, 3584, 3284, 3072, 2907, 2772, 2659, 2560, 2473, 2395, 2325,
+ 2260, 2201, 2147, 2096, 2048, 2003, 1961, 1921, 1883, 1847, 1813, 1780,
+ 1748, 1718, 1689, 1661, 1635, 1609, 1584, 1559, 1536, 1513, 1491, 1470,
+ 1449, 1429, 1409, 1390, 1371, 1353, 1335, 1318, 1301, 1284, 1268, 1252,
+ 1236, 1221, 1206, 1192, 1177, 1163, 1149, 1136, 1123, 1110, 1097, 1084,
+ 1072, 1059, 1047, 1036, 1024, 1013, 1001, 990, 979, 968, 958, 947,
+ 937, 927, 917, 907, 897, 887, 878, 868, 859, 850, 841, 832,
+ 823, 814, 806, 797, 789, 780, 772, 764, 756, 748, 740, 732,
+ 724, 717, 709, 702, 694, 687, 680, 673, 665, 658, 651, 644,
+ 637, 631, 624, 617, 611, 604, 598, 591, 585, 578, 572, 566,
+ 560, 554, 547, 541, 535, 530, 524, 518, 512, 506, 501, 495,
+ 489, 484, 478, 473, 467, 462, 456, 451, 446, 441, 435, 430,
+ 425, 420, 415, 410, 405, 400, 395, 390, 385, 380, 375, 371,
+ 366, 361, 356, 352, 347, 343, 338, 333, 329, 324, 320, 316,
+ 311, 307, 302, 298, 294, 289, 285, 281, 277, 273, 268, 264,
+ 260, 256, 252, 248, 244, 240, 236, 232, 228, 224, 220, 216,
+ 212, 209, 205, 201, 197, 194, 190, 186, 182, 179, 175, 171,
+ 168, 164, 161, 157, 153, 150, 146, 143, 139, 136, 132, 129,
+ 125, 122, 119, 115, 112, 109, 105, 102, 99, 95, 92, 89,
+ 86, 82, 79, 76, 73, 70, 66, 63, 60, 57, 54, 51,
+ 48, 45, 42, 38, 35, 32, 29, 26, 23, 20, 18, 15,
+ 12, 9, 6, 3, 3};
static void cost(int *costs, vpx_tree tree, const vpx_prob *probs,
int i, int c) {
diff --git a/vp9/encoder/vp9_cost.h b/vp9/encoder/vp9_cost.h
index d8bf23f..9831013 100644
--- a/vp9/encoder/vp9_cost.h
+++ b/vp9/encoder/vp9_cost.h
@@ -12,21 +12,22 @@
#define VP9_ENCODER_VP9_COST_H_
#include "vpx_dsp/prob.h"
+#include "vpx/vpx_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
-extern const unsigned int vp9_prob_cost[256];
+extern const uint16_t vp9_prob_cost[257];
// The factor to scale from cost in bits to cost in vp9_prob_cost units.
-#define VP9_PROB_COST_SHIFT 8
+#define VP9_PROB_COST_SHIFT 9
#define vp9_cost_zero(prob) (vp9_prob_cost[prob])
-#define vp9_cost_one(prob) vp9_cost_zero(vpx_complement(prob))
+#define vp9_cost_one(prob) vp9_cost_zero(256 - (prob))
-#define vp9_cost_bit(prob, bit) vp9_cost_zero((bit) ? vpx_complement(prob) \
+#define vp9_cost_bit(prob, bit) vp9_cost_zero((bit) ? 256 - (prob) \
: (prob))
static INLINE unsigned int cost_branch256(const unsigned int ct[2],
diff --git a/vp9/encoder/vp9_denoiser.c b/vp9/encoder/vp9_denoiser.c
index 3280e73..99118f5 100644
--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -344,7 +344,9 @@
mv_col = ctx->best_sse_mv.as_mv.col;
mv_row = ctx->best_sse_mv.as_mv.row;
motion_magnitude = mv_row * mv_row + mv_col * mv_col;
- if (denoiser->denoising_level == kDenHigh && motion_magnitude < 16) {
+ if (!is_skin &&
+ denoiser->denoising_level == kDenHigh &&
+ motion_magnitude < 16) {
denoiser->increase_denoising = 1;
} else {
denoiser->increase_denoising = 0;
diff --git a/vp9/encoder/vp9_denoiser.h b/vp9/encoder/vp9_denoiser.h
index 12dd5b5..9f13bd5 100644
--- a/vp9/encoder/vp9_denoiser.h
+++ b/vp9/encoder/vp9_denoiser.h
@@ -75,7 +75,8 @@
// This function is used by both c and sse2 denoiser implementations.
// Define it as a static function within the scope where vp9_denoiser.h
// is referenced.
-static int total_adj_strong_thresh(BLOCK_SIZE bs, int increase_denoising) {
+static INLINE int total_adj_strong_thresh(BLOCK_SIZE bs,
+ int increase_denoising) {
return (1 << num_pels_log2_lookup[bs]) * (increase_denoising ? 3 : 2);
}
#endif
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index e5a80c3..4159fe5 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -714,6 +714,10 @@
s = x->plane[0].src.buf;
sp = x->plane[0].src.stride;
+ // Index for force_split: 0 for 64x64, 1-4 for 32x32 blocks,
+ // 5-20 for the 16x16 blocks.
+ force_split[0] = 0;
+
if (!is_key_frame) {
// In the case of spatial/temporal scalable coding, the assumption here is
// that the temporal reference frame will always be of type LAST_FRAME.
@@ -768,6 +772,50 @@
vp9_build_inter_predictors_sb(xd, mi_row, mi_col, BLOCK_64X64);
+ // Check if most of the superblock is skin content, and if so, force split
+ // to 32x32. Avoid checking superblocks on/near boundary and avoid low
+ // resolutons for now.
+ // Note superblock may still pick 64X64 if y_sad is very small
+ // (i.e., y_sad < cpi->vbp_threshold_sad) below. For now leave this as is.
+ x->sb_is_skin = 0;
+#if !CONFIG_VP9_HIGHBITDEPTH
+ if (cpi->use_skin_detection && !low_res && (mi_col >= 8 &&
+ mi_col + 8 < cm->mi_cols && mi_row >= 8 && mi_row + 8 < cm->mi_rows)) {
+ int num_16x16_skin = 0;
+ int num_16x16_nonskin = 0;
+ uint8_t *ysignal = x->plane[0].src.buf;
+ uint8_t *usignal = x->plane[1].src.buf;
+ uint8_t *vsignal = x->plane[2].src.buf;
+ int spuv = x->plane[1].src.stride;
+ for (i = 0; i < 4; i++) {
+ for (j = 0; j < 4; j++) {
+ int is_skin = vp9_compute_skin_block(ysignal,
+ usignal,
+ vsignal,
+ sp,
+ spuv,
+ BLOCK_16X16);
+ num_16x16_skin += is_skin;
+ num_16x16_nonskin += (1 - is_skin);
+ if (num_16x16_nonskin > 3) {
+ // Exit loop if at least 4 of the 16x16 blocks are not skin.
+ i = 4;
+ j = 4;
+ }
+ ysignal += 16;
+ usignal += 8;
+ vsignal += 8;
+ }
+ ysignal += (sp << 4) - 64;
+ usignal += (spuv << 3) - 32;
+ vsignal += (spuv << 3) - 32;
+ }
+ if (num_16x16_skin > 12) {
+ x->sb_is_skin = 1;
+ force_split[0] = 1;
+ }
+ }
+#endif
for (i = 1; i <= 2; ++i) {
struct macroblock_plane *p = &x->plane[i];
struct macroblockd_plane *pd = &xd->plane[i];
@@ -779,7 +827,9 @@
uv_sad = cpi->fn_ptr[bs].sdf(p->src.buf, p->src.stride,
pd->dst.buf, pd->dst.stride);
- x->color_sensitivity[i - 1] = uv_sad > (y_sad >> 2);
+ // TODO(marpan): Investigate if we should lower this threshold if
+ // superblock is detected as skin.
+ x->color_sensitivity[i - 1] = uv_sad > (y_sad >> 2);
}
d = xd->plane[0].dst.buf;
@@ -818,9 +868,6 @@
#endif // CONFIG_VP9_HIGHBITDEPTH
}
- // Index for force_split: 0 for 64x64, 1-4 for 32x32 blocks,
- // 5-20 for the 16x16 blocks.
- force_split[0] = 0;
// Fill in the entire tree of 8x8 (or 4x4 under some conditions) variances
// for splits.
for (i = 0; i < 4; i++) {
@@ -3629,6 +3676,7 @@
vp9_rd_cost_init(&dummy_rdc);
x->color_sensitivity[0] = 0;
x->color_sensitivity[1] = 0;
+ x->sb_is_skin = 0;
if (seg->enabled) {
const uint8_t *const map = seg->update_map ? cpi->segmentation_map
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 028d1da..689e8c0 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -117,9 +117,9 @@
EXTRABIT e0;
int best, band, pt, i, final_eob;
#if CONFIG_VP9_HIGHBITDEPTH
- const int16_t *cat6_high_cost = vp9_get_high_cost_table(xd->bd);
+ const int *cat6_high_cost = vp9_get_high_cost_table(xd->bd);
#else
- const int16_t *cat6_high_cost = vp9_get_high_cost_table(8);
+ const int *cat6_high_cost = vp9_get_high_cost_table(8);
#endif
assert((!type && !plane) || (type && plane));
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 3ac344d..8a46738 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -62,8 +62,6 @@
#define AM_SEGMENT_ID_INACTIVE 7
#define AM_SEGMENT_ID_ACTIVE 0
-#define SHARP_FILTER_QTHRESH 0 /* Q threshold for 8-tap sharp filter */
-
#define ALTREF_HIGH_PRECISION_MV 1 // Whether to use high precision mv
// for altref computation.
#define HIGH_PRECISION_MV_QTHRESH 200 // Q threshold for high precision
@@ -1671,6 +1669,7 @@
cpi->resize_state = 0;
cpi->resize_avg_qp = 0;
cpi->resize_buffer_underflow = 0;
+ cpi->use_skin_detection = 0;
cpi->common.buffer_pool = pool;
cpi->rc.high_source_sad = 0;
@@ -1966,11 +1965,14 @@
return cpi;
}
+
+#if CONFIG_INTERNAL_STATS
#define SNPRINT(H, T) \
snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T))
#define SNPRINT2(H, T, V) \
snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V))
+#endif // CONFIG_INTERNAL_STATS
void vp9_remove_compressor(VP9_COMP *cpi) {
VP9_COMMON *cm;
@@ -2848,6 +2850,7 @@
if (xd->lossless) {
lf->filter_level = 0;
+ lf->last_filt_level = 0;
} else {
struct vpx_usec_timer timer;
@@ -2855,7 +2858,16 @@
vpx_usec_timer_start(&timer);
- vp9_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
+ if (!cpi->rc.is_src_frame_alt_ref) {
+ if ((cpi->common.frame_type == KEY_FRAME) &&
+ (!cpi->rc.this_key_frame_forced)) {
+ lf->last_filt_level = 0;
+ }
+ vp9_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
+ lf->last_filt_level = lf->filter_level;
+ } else {
+ lf->filter_level = 0;
+ }
vpx_usec_timer_mark(&timer);
cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
@@ -3048,7 +3060,7 @@
"%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
"%6d %6d %5d %5d %5d "
"%10"PRId64" %10.3lf"
- "%10lf %8u %10"PRId64" %10d %10d %10d\n",
+ "%10lf %8u %10"PRId64" %10d %10d %10d %10d\n",
cpi->common.current_video_frame,
cm->width, cm->height,
cpi->td.rd_counts.m_search_count,
@@ -3080,7 +3092,8 @@
(1 + cpi->twopass.total_left_stats.coded_error),
cpi->tot_recode_hits, recon_err, cpi->rc.kf_boost,
cpi->twopass.kf_zeromotion_pct,
- cpi->twopass.fr_content_type);
+ cpi->twopass.fr_content_type,
+ cm->lf.filter_level);
fclose(f);
@@ -3156,7 +3169,7 @@
if (oxcf->pass == 2 && cpi->sf.static_segmentation)
configure_static_seg_features(cpi);
-#if CONFIG_VP9_POSTPROC
+#if CONFIG_VP9_POSTPROC && !(CONFIG_VP9_TEMPORAL_DENOISING)
if (oxcf->noise_sensitivity > 0) {
int l = 0;
switch (oxcf->noise_sensitivity) {
@@ -3354,6 +3367,14 @@
set_size_independent_vars(cpi);
set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
+ if (cpi->oxcf.speed >= 5 &&
+ cpi->oxcf.pass == 0 &&
+ cpi->oxcf.rc_mode == VPX_CBR &&
+ cpi->oxcf.content != VP9E_CONTENT_SCREEN &&
+ cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
+ cpi->use_skin_detection = 1;
+ }
+
vp9_set_quantizer(cm, q);
vp9_set_variance_partition_thresholds(cpi, q);
@@ -4341,7 +4362,7 @@
cpi->svc.layer_context[cpi->svc.spatial_layer_id].has_alt_frame = 1;
#endif
- if (oxcf->arnr_max_frames > 0) {
+ if ((oxcf->arnr_max_frames > 0) && (oxcf->arnr_strength > 0)) {
// Produce the filtered ARF frame.
vp9_temporal_filter(cpi, arf_src_index);
vpx_extend_frame_borders(&cpi->alt_ref_buffer);
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 76c36f4..838ceac 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -494,6 +494,8 @@
int resize_buffer_underflow;
int resize_count;
+ int use_skin_detection;
+
NOISE_ESTIMATE noise_estimate;
// VAR_BASED_PARTITION thresholds
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 43fcdbd..9d3b154 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -1720,15 +1720,13 @@
gf_group->update_type[0] = OVERLAY_UPDATE;
gf_group->rf_level[0] = INTER_NORMAL;
gf_group->bit_allocation[0] = 0;
- gf_group->arf_update_idx[0] = arf_buffer_indices[0];
- gf_group->arf_ref_idx[0] = arf_buffer_indices[0];
} else {
gf_group->update_type[0] = GF_UPDATE;
gf_group->rf_level[0] = GF_ARF_STD;
gf_group->bit_allocation[0] = gf_arf_bits;
- gf_group->arf_update_idx[0] = arf_buffer_indices[0];
- gf_group->arf_ref_idx[0] = arf_buffer_indices[0];
}
+ gf_group->arf_update_idx[0] = arf_buffer_indices[0];
+ gf_group->arf_ref_idx[0] = arf_buffer_indices[0];
// Step over the golden frame / overlay frame
if (EOF == input_stats(twopass, &frame_stats))
@@ -1939,13 +1937,12 @@
// At high Q when there are few bits to spare we are better with a longer
// interval to spread the cost of the GF.
active_max_gf_interval = 12 + VPXMIN(4, (int_lbq / 6));
- if (active_max_gf_interval < active_min_gf_interval)
- active_max_gf_interval = active_min_gf_interval;
- if (active_max_gf_interval > rc->max_gf_interval)
- active_max_gf_interval = rc->max_gf_interval;
+ // We have: active_min_gf_interval <= rc->max_gf_interval
if (active_max_gf_interval < active_min_gf_interval)
active_max_gf_interval = active_min_gf_interval;
+ else if (active_max_gf_interval > rc->max_gf_interval)
+ active_max_gf_interval = rc->max_gf_interval;
}
}
@@ -2263,6 +2260,8 @@
return is_viable_kf;
}
+#define FRAMES_TO_CHECK_DECAY 8
+
static void find_next_key_frame(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
int i, j;
RATE_CONTROL *const rc = &cpi->rc;
@@ -2281,7 +2280,7 @@
double boost_score = 0.0;
double kf_mod_err = 0.0;
double kf_group_err = 0.0;
- double recent_loop_decay[8] = {1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0};
+ double recent_loop_decay[FRAMES_TO_CHECK_DECAY];
vp9_zero(next_frame);
@@ -2308,6 +2307,10 @@
kf_mod_err = calculate_modified_err(cpi, twopass, oxcf, this_frame);
+ // Initialize the decay rates for the recent frames to check
+ for (j = 0; j < FRAMES_TO_CHECK_DECAY; ++j)
+ recent_loop_decay[j] = 1.0;
+
// Find the next keyframe.
i = 0;
while (twopass->stats_in < twopass->stats_in_end &&
@@ -2334,9 +2337,9 @@
// We want to know something about the recent past... rather than
// as used elsewhere where we are concerned with decay in prediction
// quality since the last GF or KF.
- recent_loop_decay[i % 8] = loop_decay_rate;
+ recent_loop_decay[i % FRAMES_TO_CHECK_DECAY] = loop_decay_rate;
decay_accumulator = 1.0;
- for (j = 0; j < 8; ++j)
+ for (j = 0; j < FRAMES_TO_CHECK_DECAY; ++j)
decay_accumulator *= recent_loop_decay[j];
// Special check for transition or high motion followed by a
diff --git a/vp9/encoder/vp9_lookahead.c b/vp9/encoder/vp9_lookahead.c
index 8787be8..def9b8c 100644
--- a/vp9/encoder/vp9_lookahead.c
+++ b/vp9/encoder/vp9_lookahead.c
@@ -207,7 +207,7 @@
int drain) {
struct lookahead_entry *buf = NULL;
- if (ctx->sz && (drain || ctx->sz == ctx->max_sz - MAX_PRE_FRAMES)) {
+ if (ctx && ctx->sz && (drain || ctx->sz == ctx->max_sz - MAX_PRE_FRAMES)) {
buf = pop(ctx, &ctx->read_idx);
ctx->sz--;
}
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index 84ef1b4..607941c 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -86,7 +86,9 @@
if (mvcost) {
const MV diff = { mv->row - ref->row,
mv->col - ref->col };
- return ROUND_POWER_OF_TWO(mv_cost(&diff, mvjcost, mvcost) *
+ // TODO(aconverse): See if this shift needs to be tied to
+ // VP9_PROB_COST_SHIFT.
+ return ROUND_POWER_OF_TWO((unsigned)mv_cost(&diff, mvjcost, mvcost) *
error_per_bit, 13);
}
return 0;
@@ -96,8 +98,9 @@
int error_per_bit) {
const MV diff = { mv->row - ref->row,
mv->col - ref->col };
- return ROUND_POWER_OF_TWO(mv_cost(&diff, x->nmvjointsadcost,
- x->nmvsadcost) * error_per_bit, 8);
+ // TODO(aconverse): See if this shift needs to be tied to VP9_PROB_COST_SHIFT.
+ return ROUND_POWER_OF_TWO((unsigned)mv_cost(&diff, x->nmvjointsadcost,
+ x->nmvsadcost) * error_per_bit, 8);
}
void vp9_init_dsmotion_compensation(search_site_config *cfg, int stride) {
@@ -150,10 +153,10 @@
*/
/* estimated cost of a motion vector (r,c) */
-#define MVC(r, c) \
- (mvcost ? \
- ((mvjcost[((r) != rr) * 2 + ((c) != rc)] + \
- mvcost[0][((r) - rr)] + mvcost[1][((c) - rc)]) * \
+#define MVC(r, c) \
+ (mvcost ? \
+ ((unsigned)(mvjcost[((r) != rr) * 2 + ((c) != rc)] + \
+ mvcost[0][((r) - rr)] + mvcost[1][((c) - rc)]) * \
error_per_bit + 4096) >> 13 : 0)
@@ -846,9 +849,9 @@
cost_list[i + 1] = fn_ptr->vf(what->buf, what->stride,
get_buf_from_mv(in_what, &this_mv),
in_what->stride, &sse) +
- // mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
- mv_err_cost(&this_mv, &fcenter_mv, x->nmvjointcost, x->mvcost,
- x->errorperbit);
+ mv_err_cost(&this_mv, &fcenter_mv,
+ x->nmvjointcost, x->mvcost,
+ x->errorperbit);
}
} else {
for (i = 0; i < 4; i++) {
@@ -860,9 +863,9 @@
cost_list[i + 1] = fn_ptr->vf(what->buf, what->stride,
get_buf_from_mv(in_what, &this_mv),
in_what->stride, &sse) +
- // mvsad_err_cost(x, &this_mv, &fcenter_mv, sadpb);
- mv_err_cost(&this_mv, &fcenter_mv, x->nmvjointcost, x->mvcost,
- x->errorperbit);
+ mv_err_cost(&this_mv, &fcenter_mv,
+ x->nmvjointcost, x->mvcost,
+ x->errorperbit);
}
}
}
diff --git a/vp9/encoder/vp9_noise_estimate.c b/vp9/encoder/vp9_noise_estimate.c
index 008a40a..e56cc9b 100644
--- a/vp9/encoder/vp9_noise_estimate.c
+++ b/vp9/encoder/vp9_noise_estimate.c
@@ -66,6 +66,7 @@
return 0;
}
+#if CONFIG_VP9_TEMPORAL_DENOISING
static void copy_frame(YV12_BUFFER_CONFIG * const dest,
const YV12_BUFFER_CONFIG * const src) {
int r;
@@ -81,6 +82,7 @@
srcbuf += src->y_stride;
}
}
+#endif // CONFIG_VP9_TEMPORAL_DENOISING
NOISE_LEVEL vp9_noise_estimate_extract_level(NOISE_ESTIMATE *const ne) {
int noise_level = kLowLow;
@@ -160,7 +162,9 @@
for (mi_row = 0; mi_row < cm->mi_rows; mi_row++) {
for (mi_col = 0; mi_col < cm->mi_cols; mi_col++) {
// 16x16 blocks, 1/4 sample of frame.
- if (mi_row % 4 == 0 && mi_col % 4 == 0) {
+ if (mi_row % 4 == 0 && mi_col % 4 == 0 &&
+ mi_row < cm->mi_rows - 1 &&
+ mi_col < cm->mi_cols - 1) {
int bl_index = mi_row * cm->mi_cols + mi_col;
int bl_index1 = bl_index + 1;
int bl_index2 = bl_index + cm->mi_cols;
diff --git a/vp9/encoder/vp9_picklpf.c b/vp9/encoder/vp9_picklpf.c
index 5444bc8..f6b1dfc 100644
--- a/vp9/encoder/vp9_picklpf.c
+++ b/vp9/encoder/vp9_picklpf.c
@@ -78,7 +78,8 @@
// Start the search at the previous frame filter level unless it is now out of
// range.
- int filt_mid = clamp(lf->filter_level, min_filter_level, max_filter_level);
+ int filt_mid =
+ clamp(lf->last_filt_level, min_filter_level, max_filter_level);
int filter_step = filt_mid < 16 ? 4 : filt_mid / 4;
// Sum squared error at each filter level
int64_t ss_err[MAX_LOOP_FILTER + 1];
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 8f6f2a3..829066c 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -852,6 +852,12 @@
if (var <= thresh_ac && (sse - var) <= thresh_dc) {
unsigned int sse_u, sse_v;
unsigned int var_u, var_v;
+ unsigned int thresh_ac_uv = thresh_ac;
+ unsigned int thresh_dc_uv = thresh_dc;
+ if (x->sb_is_skin) {
+ thresh_ac_uv = 0;
+ thresh_dc_uv = 0;
+ }
// Skip UV prediction unless breakout is zero (lossless) to save
// computation with low impact on the result
@@ -867,14 +873,14 @@
xd->plane[1].dst.stride, &sse_u);
// U skipping condition checking
- if (((var_u << 2) <= thresh_ac) && (sse_u - var_u <= thresh_dc)) {
+ if (((var_u << 2) <= thresh_ac_uv) && (sse_u - var_u <= thresh_dc_uv)) {
var_v = cpi->fn_ptr[uv_size].vf(x->plane[2].src.buf,
x->plane[2].src.stride,
xd->plane[2].dst.buf,
xd->plane[2].dst.stride, &sse_v);
// V skipping condition checking
- if (((var_v << 2) <= thresh_ac) && (sse_v - var_v <= thresh_dc)) {
+ if (((var_v << 2) <= thresh_ac_uv) && (sse_v - var_v <= thresh_dc_uv)) {
x->skip = 1;
// The cost of skip bit needs to be added.
@@ -1585,7 +1591,8 @@
this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv, this_rdc.rate, this_rdc.dist);
if (cpi->oxcf.speed >= 5 &&
- cpi->oxcf.content != VP9E_CONTENT_SCREEN) {
+ cpi->oxcf.content != VP9E_CONTENT_SCREEN &&
+ !x->sb_is_skin) {
// Bias against non-zero (above some threshold) motion for large blocks.
// This is temporary fix to avoid selection of large mv for big blocks.
if (frame_mv[this_mode][ref_frame].as_mv.row > 64 ||
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index 2579c60..5df2909 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -500,6 +500,12 @@
else
cpi->rc.rc_1_frame = 0;
+ // Turn off oscilation detection in the case of massive overshoot.
+ if (cpi->rc.rc_1_frame == -1 && cpi->rc.rc_2_frame == 1 &&
+ correction_factor > 1000) {
+ cpi->rc.rc_2_frame = 0;
+ }
+
if (correction_factor > 102) {
// We are not already at the worst allowable quality
correction_factor = (int)(100 + ((correction_factor - 100) *
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index ff3112c..f00a58c 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -373,9 +373,9 @@
int pt = combine_entropy_contexts(*A, *L);
int c, cost;
#if CONFIG_VP9_HIGHBITDEPTH
- const int16_t *cat6_high_cost = vp9_get_high_cost_table(xd->bd);
+ const int *cat6_high_cost = vp9_get_high_cost_table(xd->bd);
#else
- const int16_t *cat6_high_cost = vp9_get_high_cost_table(8);
+ const int *cat6_high_cost = vp9_get_high_cost_table(8);
#endif
// Check for consistency of tx_size with mode info
@@ -3355,15 +3355,19 @@
}
if (!disable_skip) {
+ vpx_prob skip_prob = vp9_get_skip_prob(cm, xd);
if (skippable) {
// Back out the coefficient coding costs
rate2 -= (rate_y + rate_uv);
// Cost the skip mb case
- rate2 += vp9_cost_bit(vp9_get_skip_prob(cm, xd), 1);
+ rate2 += vp9_cost_bit(skip_prob, 1);
} else if (ref_frame != INTRA_FRAME && !xd->lossless) {
- if (RDCOST(x->rdmult, x->rddiv, rate_y + rate_uv, distortion2) <
- RDCOST(x->rdmult, x->rddiv, 0, total_sse)) {
+ if (RDCOST(x->rdmult, x->rddiv,
+ rate_y + rate_uv + vp9_cost_bit(skip_prob, 0),
+ distortion2) <
+ RDCOST(x->rdmult, x->rddiv,
+ vp9_cost_bit(skip_prob, 1), total_sse)) {
// Add in the cost of the no skip flag.
rate2 += vp9_cost_bit(vp9_get_skip_prob(cm, xd), 0);
} else {
diff --git a/vp9/encoder/vp9_resize.c b/vp9/encoder/vp9_resize.c
index 59c7478..f4d0db4d 100644
--- a/vp9/encoder/vp9_resize.c
+++ b/vp9/encoder/vp9_resize.c
@@ -15,6 +15,7 @@
#include <stdlib.h>
#include <string.h>
+#include "./vpx_config.h"
#if CONFIG_VP9_HIGHBITDEPTH
#include "vpx_dsp/vpx_dsp_common.h"
#endif // CONFIG_VP9_HIGHBITDEPTH
diff --git a/vp9/encoder/vp9_skin_detection.c b/vp9/encoder/vp9_skin_detection.c
index 0ca1665..8e117eb 100644
--- a/vp9/encoder/vp9_skin_detection.c
+++ b/vp9/encoder/vp9_skin_detection.c
@@ -15,22 +15,29 @@
#include "vp9/encoder/vp9_encoder.h"
#include "vp9/encoder/vp9_skin_detection.h"
+#define MODEL_MODE 0
+
// Fixed-point skin color model parameters.
-static const int skin_mean[2] = {7463, 9614}; // q6
+static const int skin_mean[5][2] = {
+ {7463, 9614}, {6400, 10240}, {7040, 10240}, {8320, 9280}, {6800, 9614}};
static const int skin_inv_cov[4] = {4107, 1663, 1663, 2157}; // q16
-static const int skin_threshold = 1570636; // q18
+static const int skin_threshold[6] = {1570636, 1400000, 800000, 800000, 800000,
+ 800000}; // q18
// Thresholds on luminance.
-static const int y_low = 20;
+static const int y_low = 40;
static const int y_high = 220;
// Evaluates the Mahalanobis distance measure for the input CbCr values.
-static int evaluate_skin_color_difference(int cb, int cr) {
+static int evaluate_skin_color_difference(int cb, int cr, int idx) {
const int cb_q6 = cb << 6;
const int cr_q6 = cr << 6;
- const int cb_diff_q12 = (cb_q6 - skin_mean[0]) * (cb_q6 - skin_mean[0]);
- const int cbcr_diff_q12 = (cb_q6 - skin_mean[0]) * (cr_q6 - skin_mean[1]);
- const int cr_diff_q12 = (cr_q6 - skin_mean[1]) * (cr_q6 - skin_mean[1]);
+ const int cb_diff_q12 =
+ (cb_q6 - skin_mean[idx][0]) * (cb_q6 - skin_mean[idx][0]);
+ const int cbcr_diff_q12 =
+ (cb_q6 - skin_mean[idx][0]) * (cr_q6 - skin_mean[idx][1]);
+ const int cr_diff_q12 =
+ (cr_q6 - skin_mean[idx][1]) * (cr_q6 - skin_mean[idx][1]);
const int cb_diff_q2 = (cb_diff_q12 + (1 << 9)) >> 10;
const int cbcr_diff_q2 = (cbcr_diff_q12 + (1 << 9)) >> 10;
const int cr_diff_q2 = (cr_diff_q12 + (1 << 9)) >> 10;
@@ -42,10 +49,35 @@
}
int vp9_skin_pixel(const uint8_t y, const uint8_t cb, const uint8_t cr) {
- if (y < y_low || y > y_high)
+ if (y < y_low || y > y_high) {
return 0;
- else
- return (evaluate_skin_color_difference(cb, cr) < skin_threshold);
+ } else {
+ if (MODEL_MODE == 0) {
+ return (evaluate_skin_color_difference(cb, cr, 0) < skin_threshold[0]);
+ } else {
+ int i = 0;
+ // Exit on grey.
+ if (cb == 128 && cr == 128)
+ return 0;
+ // Exit on very strong cb.
+ if (cb > 150 && cr < 110)
+ return 0;
+ // Exit on (another) low luminance threshold if either color is high.
+ if (y < 50 && (cb > 140 || cr > 140))
+ return 0;
+ for (; i < 5; i++) {
+ if (evaluate_skin_color_difference(cb, cr, i) < skin_threshold[i + 1]) {
+ return 1;
+ }
+ // Exit if difference is much large than the threshold.
+ if (evaluate_skin_color_difference(cb, cr, i) >
+ (skin_threshold[i + 1] << 3)) {
+ return 0;
+ }
+ }
+ return 0;
+ }
+ }
}
int vp9_compute_skin_block(const uint8_t *y, const uint8_t *u, const uint8_t *v,
diff --git a/vp9/encoder/vp9_subexp.c b/vp9/encoder/vp9_subexp.c
index 1a87199..cbd3c49 100644
--- a/vp9/encoder/vp9_subexp.c
+++ b/vp9/encoder/vp9_subexp.c
@@ -142,8 +142,10 @@
vpx_prob *bestp,
vpx_prob upd,
int stepsize) {
- int i, old_b, new_b, update_b, savings, bestsavings, step;
+ int i, old_b, new_b, update_b, savings, bestsavings;
int newp;
+ const int step_sign = *bestp > oldp[PIVOT_NODE] ? -1 : 1;
+ const int step = stepsize * step_sign;
vpx_prob bestnewp, newplist[ENTROPY_NODES], oldplist[ENTROPY_NODES];
vp9_model_to_full_probs(oldp, oldplist);
memcpy(newplist, oldp, sizeof(vpx_prob) * UNCONSTRAINED_NODES);
@@ -154,41 +156,23 @@
bestsavings = 0;
bestnewp = oldp[PIVOT_NODE];
- if (*bestp > oldp[PIVOT_NODE]) {
- step = -stepsize;
- for (newp = *bestp; newp > oldp[PIVOT_NODE]; newp += step) {
- if (newp < 1 || newp > 255)
- continue;
- newplist[PIVOT_NODE] = newp;
- vp9_model_to_full_probs(newplist, newplist);
- for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
- new_b += cost_branch256(ct + 2 * i, newplist[i]);
- new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
- update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
- vp9_cost_upd256;
- savings = old_b - new_b - update_b;
- if (savings > bestsavings) {
- bestsavings = savings;
- bestnewp = newp;
- }
- }
- } else {
- step = stepsize;
- for (newp = *bestp; newp < oldp[PIVOT_NODE]; newp += step) {
- if (newp < 1 || newp > 255)
- continue;
- newplist[PIVOT_NODE] = newp;
- vp9_model_to_full_probs(newplist, newplist);
- for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
- new_b += cost_branch256(ct + 2 * i, newplist[i]);
- new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
- update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
- vp9_cost_upd256;
- savings = old_b - new_b - update_b;
- if (savings > bestsavings) {
- bestsavings = savings;
- bestnewp = newp;
- }
+ assert(stepsize > 0);
+
+ for (newp = *bestp; (newp - oldp[PIVOT_NODE]) * step_sign < 0;
+ newp += step) {
+ if (newp < 1 || newp > 255)
+ continue;
+ newplist[PIVOT_NODE] = newp;
+ vp9_model_to_full_probs(newplist, newplist);
+ for (i = UNCONSTRAINED_NODES, new_b = 0; i < ENTROPY_NODES; ++i)
+ new_b += cost_branch256(ct + 2 * i, newplist[i]);
+ new_b += cost_branch256(ct + 2 * PIVOT_NODE, newplist[PIVOT_NODE]);
+ update_b = prob_diff_update_cost(newp, oldp[PIVOT_NODE]) +
+ vp9_cost_upd256;
+ savings = old_b - new_b - update_b;
+ if (savings > bestsavings) {
+ bestsavings = savings;
+ bestnewp = newp;
}
}
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index e0236aa..9724df4 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -16,7 +16,6 @@
#include "vp9/encoder/vp9_extend.h"
#include "vpx_dsp/vpx_dsp_common.h"
-#define SMALL_FRAME_FB_IDX 7
#define SMALL_FRAME_WIDTH 32
#define SMALL_FRAME_HEIGHT 16
@@ -644,6 +643,8 @@
}
#if CONFIG_SPATIAL_SVC
+#define SMALL_FRAME_FB_IDX 7
+
int vp9_svc_start_frame(VP9_COMP *const cpi) {
int width = 0, height = 0;
LAYER_CONTEXT *lc;
@@ -754,7 +755,8 @@
return 0;
}
-#endif
+#undef SMALL_FRAME_FB_IDX
+#endif // CONFIG_SPATIAL_SVC
struct lookahead_entry *vp9_svc_lookahead_pop(VP9_COMP *const cpi,
struct lookahead_ctx *ctx,
diff --git a/vp9/encoder/vp9_tokenize.c b/vp9/encoder/vp9_tokenize.c
index a6dd3c5..93be6d7 100644
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -67,292 +67,170 @@
};
static const int16_t zero_cost[] = {0};
-static const int16_t sign_cost[] = {255, 257};
-static const int16_t cat1_cost[] = {429, 431, 616, 618};
-static const int16_t cat2_cost[] = {624, 626, 727, 729, 848, 850, 951, 953};
-static const int16_t cat3_cost[] = {
- 820, 822, 893, 895, 940, 942, 1013, 1015, 1096, 1098, 1169, 1171, 1216, 1218,
- 1289, 1291
-};
-static const int16_t cat4_cost[] = {
- 1032, 1034, 1075, 1077, 1105, 1107, 1148, 1150, 1194, 1196, 1237, 1239,
- 1267, 1269, 1310, 1312, 1328, 1330, 1371, 1373, 1401, 1403, 1444, 1446,
- 1490, 1492, 1533, 1535, 1563, 1565, 1606, 1608
-};
-static const int16_t cat5_cost[] = {
- 1269, 1271, 1283, 1285, 1306, 1308, 1320,
- 1322, 1347, 1349, 1361, 1363, 1384, 1386, 1398, 1400, 1443, 1445, 1457,
- 1459, 1480, 1482, 1494, 1496, 1521, 1523, 1535, 1537, 1558, 1560, 1572,
- 1574, 1592, 1594, 1606, 1608, 1629, 1631, 1643, 1645, 1670, 1672, 1684,
- 1686, 1707, 1709, 1721, 1723, 1766, 1768, 1780, 1782, 1803, 1805, 1817,
- 1819, 1844, 1846, 1858, 1860, 1881, 1883, 1895, 1897
-};
+static const int16_t sign_cost[1] = {512};
+static const int16_t cat1_cost[1 << 1] = {864, 1229};
+static const int16_t cat2_cost[1 << 2] = {1256, 1453, 1696, 1893};
+static const int16_t cat3_cost[1 << 3] = {1652, 1791, 1884, 2023,
+ 2195, 2334, 2427, 2566};
+static const int16_t cat4_cost[1 << 4] = {2079, 2160, 2218, 2299, 2395, 2476,
+ 2534, 2615, 2661, 2742, 2800, 2881,
+ 2977, 3058, 3116, 3197};
+static const int16_t cat5_cost[1 << 5] = {
+ 2553, 2576, 2622, 2645, 2703, 2726, 2772, 2795, 2894, 2917, 2963,
+ 2986, 3044, 3067, 3113, 3136, 3190, 3213, 3259, 3282, 3340, 3363,
+ 3409, 3432, 3531, 3554, 3600, 3623, 3681, 3704, 3750, 3773};
const int16_t vp9_cat6_low_cost[256] = {
- 1638, 1640, 1646, 1648, 1652, 1654, 1660, 1662,
- 1670, 1672, 1678, 1680, 1684, 1686, 1692, 1694, 1711, 1713, 1719, 1721,
- 1725, 1727, 1733, 1735, 1743, 1745, 1751, 1753, 1757, 1759, 1765, 1767,
- 1787, 1789, 1795, 1797, 1801, 1803, 1809, 1811, 1819, 1821, 1827, 1829,
- 1833, 1835, 1841, 1843, 1860, 1862, 1868, 1870, 1874, 1876, 1882, 1884,
- 1892, 1894, 1900, 1902, 1906, 1908, 1914, 1916, 1940, 1942, 1948, 1950,
- 1954, 1956, 1962, 1964, 1972, 1974, 1980, 1982, 1986, 1988, 1994, 1996,
- 2013, 2015, 2021, 2023, 2027, 2029, 2035, 2037, 2045, 2047, 2053, 2055,
- 2059, 2061, 2067, 2069, 2089, 2091, 2097, 2099, 2103, 2105, 2111, 2113,
- 2121, 2123, 2129, 2131, 2135, 2137, 2143, 2145, 2162, 2164, 2170, 2172,
- 2176, 2178, 2184, 2186, 2194, 2196, 2202, 2204, 2208, 2210, 2216, 2218,
- 2082, 2084, 2090, 2092, 2096, 2098, 2104, 2106, 2114, 2116, 2122, 2124,
- 2128, 2130, 2136, 2138, 2155, 2157, 2163, 2165, 2169, 2171, 2177, 2179,
- 2187, 2189, 2195, 2197, 2201, 2203, 2209, 2211, 2231, 2233, 2239, 2241,
- 2245, 2247, 2253, 2255, 2263, 2265, 2271, 2273, 2277, 2279, 2285, 2287,
- 2304, 2306, 2312, 2314, 2318, 2320, 2326, 2328, 2336, 2338, 2344, 2346,
- 2350, 2352, 2358, 2360, 2384, 2386, 2392, 2394, 2398, 2400, 2406, 2408,
- 2416, 2418, 2424, 2426, 2430, 2432, 2438, 2440, 2457, 2459, 2465, 2467,
- 2471, 2473, 2479, 2481, 2489, 2491, 2497, 2499, 2503, 2505, 2511, 2513,
- 2533, 2535, 2541, 2543, 2547, 2549, 2555, 2557, 2565, 2567, 2573, 2575,
- 2579, 2581, 2587, 2589, 2606, 2608, 2614, 2616, 2620, 2622, 2628, 2630,
- 2638, 2640, 2646, 2648, 2652, 2654, 2660, 2662
-};
-const int16_t vp9_cat6_high_cost[128] = {
- 72, 892, 1183, 2003, 1448, 2268, 2559, 3379,
- 1709, 2529, 2820, 3640, 3085, 3905, 4196, 5016, 2118, 2938, 3229, 4049,
- 3494, 4314, 4605, 5425, 3755, 4575, 4866, 5686, 5131, 5951, 6242, 7062,
- 2118, 2938, 3229, 4049, 3494, 4314, 4605, 5425, 3755, 4575, 4866, 5686,
- 5131, 5951, 6242, 7062, 4164, 4984, 5275, 6095, 5540, 6360, 6651, 7471,
- 5801, 6621, 6912, 7732, 7177, 7997, 8288, 9108, 2118, 2938, 3229, 4049,
- 3494, 4314, 4605, 5425, 3755, 4575, 4866, 5686, 5131, 5951, 6242, 7062,
- 4164, 4984, 5275, 6095, 5540, 6360, 6651, 7471, 5801, 6621, 6912, 7732,
- 7177, 7997, 8288, 9108, 4164, 4984, 5275, 6095, 5540, 6360, 6651, 7471,
- 5801, 6621, 6912, 7732, 7177, 7997, 8288, 9108, 6210, 7030, 7321, 8141,
- 7586, 8406, 8697, 9517, 7847, 8667, 8958, 9778, 9223, 10043, 10334, 11154
-};
+ 3378, 3390, 3401, 3413, 3435, 3447, 3458, 3470, 3517, 3529, 3540, 3552,
+ 3574, 3586, 3597, 3609, 3671, 3683, 3694, 3706, 3728, 3740, 3751, 3763,
+ 3810, 3822, 3833, 3845, 3867, 3879, 3890, 3902, 3973, 3985, 3996, 4008,
+ 4030, 4042, 4053, 4065, 4112, 4124, 4135, 4147, 4169, 4181, 4192, 4204,
+ 4266, 4278, 4289, 4301, 4323, 4335, 4346, 4358, 4405, 4417, 4428, 4440,
+ 4462, 4474, 4485, 4497, 4253, 4265, 4276, 4288, 4310, 4322, 4333, 4345,
+ 4392, 4404, 4415, 4427, 4449, 4461, 4472, 4484, 4546, 4558, 4569, 4581,
+ 4603, 4615, 4626, 4638, 4685, 4697, 4708, 4720, 4742, 4754, 4765, 4777,
+ 4848, 4860, 4871, 4883, 4905, 4917, 4928, 4940, 4987, 4999, 5010, 5022,
+ 5044, 5056, 5067, 5079, 5141, 5153, 5164, 5176, 5198, 5210, 5221, 5233,
+ 5280, 5292, 5303, 5315, 5337, 5349, 5360, 5372, 4988, 5000, 5011, 5023,
+ 5045, 5057, 5068, 5080, 5127, 5139, 5150, 5162, 5184, 5196, 5207, 5219,
+ 5281, 5293, 5304, 5316, 5338, 5350, 5361, 5373, 5420, 5432, 5443, 5455,
+ 5477, 5489, 5500, 5512, 5583, 5595, 5606, 5618, 5640, 5652, 5663, 5675,
+ 5722, 5734, 5745, 5757, 5779, 5791, 5802, 5814, 5876, 5888, 5899, 5911,
+ 5933, 5945, 5956, 5968, 6015, 6027, 6038, 6050, 6072, 6084, 6095, 6107,
+ 5863, 5875, 5886, 5898, 5920, 5932, 5943, 5955, 6002, 6014, 6025, 6037,
+ 6059, 6071, 6082, 6094, 6156, 6168, 6179, 6191, 6213, 6225, 6236, 6248,
+ 6295, 6307, 6318, 6330, 6352, 6364, 6375, 6387, 6458, 6470, 6481, 6493,
+ 6515, 6527, 6538, 6550, 6597, 6609, 6620, 6632, 6654, 6666, 6677, 6689,
+ 6751, 6763, 6774, 6786, 6808, 6820, 6831, 6843, 6890, 6902, 6913, 6925,
+ 6947, 6959, 6970, 6982};
+const int vp9_cat6_high_cost[64] = {
+ 88, 2251, 2727, 4890, 3148, 5311, 5787, 7950, 3666, 5829, 6305,
+ 8468, 6726, 8889, 9365, 11528, 3666, 5829, 6305, 8468, 6726, 8889,
+ 9365, 11528, 7244, 9407, 9883, 12046, 10304, 12467, 12943, 15106, 3666,
+ 5829, 6305, 8468, 6726, 8889, 9365, 11528, 7244, 9407, 9883, 12046,
+ 10304, 12467, 12943, 15106, 7244, 9407, 9883, 12046, 10304, 12467, 12943,
+ 15106, 10822, 12985, 13461, 15624, 13882, 16045, 16521, 18684};
#if CONFIG_VP9_HIGHBITDEPTH
-const int16_t vp9_cat6_high10_high_cost[512] = {
- 74, 894, 1185, 2005, 1450, 2270, 2561,
- 3381, 1711, 2531, 2822, 3642, 3087, 3907, 4198, 5018, 2120, 2940, 3231,
- 4051, 3496, 4316, 4607, 5427, 3757, 4577, 4868, 5688, 5133, 5953, 6244,
- 7064, 2120, 2940, 3231, 4051, 3496, 4316, 4607, 5427, 3757, 4577, 4868,
- 5688, 5133, 5953, 6244, 7064, 4166, 4986, 5277, 6097, 5542, 6362, 6653,
- 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290, 9110, 2120, 2940, 3231,
- 4051, 3496, 4316, 4607, 5427, 3757, 4577, 4868, 5688, 5133, 5953, 6244,
- 7064, 4166, 4986, 5277, 6097, 5542, 6362, 6653, 7473, 5803, 6623, 6914,
- 7734, 7179, 7999, 8290, 9110, 4166, 4986, 5277, 6097, 5542, 6362, 6653,
- 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290, 9110, 6212, 7032, 7323,
- 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960, 9780, 9225, 10045, 10336,
- 11156, 2120, 2940, 3231, 4051, 3496, 4316, 4607, 5427, 3757, 4577, 4868,
- 5688, 5133, 5953, 6244, 7064, 4166, 4986, 5277, 6097, 5542, 6362, 6653,
- 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290, 9110, 4166, 4986, 5277,
- 6097, 5542, 6362, 6653, 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290,
- 9110, 6212, 7032, 7323, 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960,
- 9780, 9225, 10045, 10336, 11156, 4166, 4986, 5277, 6097, 5542, 6362, 6653,
- 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290, 9110, 6212, 7032, 7323,
- 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960, 9780, 9225, 10045, 10336,
- 11156, 6212, 7032, 7323, 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960,
- 9780, 9225, 10045, 10336, 11156, 8258, 9078, 9369, 10189, 9634, 10454,
- 10745, 11565, 9895, 10715, 11006, 11826, 11271, 12091, 12382, 13202, 2120,
- 2940, 3231, 4051, 3496, 4316, 4607, 5427, 3757, 4577, 4868, 5688, 5133,
- 5953, 6244, 7064, 4166, 4986, 5277, 6097, 5542, 6362, 6653, 7473, 5803,
- 6623, 6914, 7734, 7179, 7999, 8290, 9110, 4166, 4986, 5277, 6097, 5542,
- 6362, 6653, 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290, 9110, 6212,
- 7032, 7323, 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960, 9780, 9225,
- 10045, 10336, 11156, 4166, 4986, 5277, 6097, 5542, 6362, 6653, 7473, 5803,
- 6623, 6914, 7734, 7179, 7999, 8290, 9110, 6212, 7032, 7323, 8143, 7588,
- 8408, 8699, 9519, 7849, 8669, 8960, 9780, 9225, 10045, 10336, 11156, 6212,
- 7032, 7323, 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960, 9780, 9225,
- 10045, 10336, 11156, 8258, 9078, 9369, 10189, 9634, 10454, 10745, 11565,
- 9895, 10715, 11006, 11826, 11271, 12091, 12382, 13202, 4166, 4986, 5277,
- 6097, 5542, 6362, 6653, 7473, 5803, 6623, 6914, 7734, 7179, 7999, 8290,
- 9110, 6212, 7032, 7323, 8143, 7588, 8408, 8699, 9519, 7849, 8669, 8960,
- 9780, 9225, 10045, 10336, 11156, 6212, 7032, 7323, 8143, 7588, 8408, 8699,
- 9519, 7849, 8669, 8960, 9780, 9225, 10045, 10336, 11156, 8258, 9078, 9369,
- 10189, 9634, 10454, 10745, 11565, 9895, 10715, 11006, 11826, 11271, 12091,
- 12382, 13202, 6212, 7032, 7323, 8143, 7588, 8408, 8699, 9519, 7849, 8669,
- 8960, 9780, 9225, 10045, 10336, 11156, 8258, 9078, 9369, 10189, 9634, 10454,
- 10745, 11565, 9895, 10715, 11006, 11826, 11271, 12091, 12382, 13202, 8258,
- 9078, 9369, 10189, 9634, 10454, 10745, 11565, 9895, 10715, 11006, 11826,
- 11271, 12091, 12382, 13202, 10304, 11124, 11415, 12235, 11680, 12500, 12791,
- 13611, 11941, 12761, 13052, 13872, 13317, 14137, 14428, 15248,
-};
-const int16_t vp9_cat6_high12_high_cost[2048] = {
- 76, 896, 1187, 2007, 1452, 2272, 2563,
- 3383, 1713, 2533, 2824, 3644, 3089, 3909, 4200, 5020, 2122, 2942, 3233,
- 4053, 3498, 4318, 4609, 5429, 3759, 4579, 4870, 5690, 5135, 5955, 6246,
- 7066, 2122, 2942, 3233, 4053, 3498, 4318, 4609, 5429, 3759, 4579, 4870,
- 5690, 5135, 5955, 6246, 7066, 4168, 4988, 5279, 6099, 5544, 6364, 6655,
- 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 2122, 2942, 3233,
- 4053, 3498, 4318, 4609, 5429, 3759, 4579, 4870, 5690, 5135, 5955, 6246,
- 7066, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916,
- 7736, 7181, 8001, 8292, 9112, 4168, 4988, 5279, 6099, 5544, 6364, 6655,
- 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325,
- 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338,
- 11158, 2122, 2942, 3233, 4053, 3498, 4318, 4609, 5429, 3759, 4579, 4870,
- 5690, 5135, 5955, 6246, 7066, 4168, 4988, 5279, 6099, 5544, 6364, 6655,
- 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 4168, 4988, 5279,
- 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292,
- 9112, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962,
- 9782, 9227, 10047, 10338, 11158, 4168, 4988, 5279, 6099, 5544, 6364, 6655,
- 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325,
- 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338,
- 11158, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962,
- 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456,
- 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 2122,
- 2942, 3233, 4053, 3498, 4318, 4609, 5429, 3759, 4579, 4870, 5690, 5135,
- 5955, 6246, 7066, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805,
- 6625, 6916, 7736, 7181, 8001, 8292, 9112, 4168, 4988, 5279, 6099, 5544,
- 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 6214,
- 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227,
- 10047, 10338, 11158, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805,
- 6625, 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325, 8145, 7590,
- 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 6214,
- 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227,
- 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567,
- 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 4168, 4988, 5279,
- 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292,
- 9112, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962,
- 9782, 9227, 10047, 10338, 11158, 6214, 7034, 7325, 8145, 7590, 8410, 8701,
- 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371,
- 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093,
- 12384, 13204, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671,
- 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456,
- 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 8260,
- 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828,
- 11273, 12093, 12384, 13204, 10306, 11126, 11417, 12237, 11682, 12502, 12793,
- 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250, 2122, 2942,
- 3233, 4053, 3498, 4318, 4609, 5429, 3759, 4579, 4870, 5690, 5135, 5955,
- 6246, 7066, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625,
- 6916, 7736, 7181, 8001, 8292, 9112, 4168, 4988, 5279, 6099, 5544, 6364,
- 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034,
- 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047,
- 10338, 11158, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625,
- 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325, 8145, 7590, 8410,
- 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 6214, 7034,
- 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047,
- 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897,
- 10717, 11008, 11828, 11273, 12093, 12384, 13204, 4168, 4988, 5279, 6099,
- 5544, 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112,
- 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782,
- 9227, 10047, 10338, 11158, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521,
- 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191,
- 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384,
- 13204, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962,
- 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456,
- 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 8260,
- 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828,
- 11273, 12093, 12384, 13204, 10306, 11126, 11417, 12237, 11682, 12502, 12793,
- 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250, 4168, 4988,
- 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001,
- 8292, 9112, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671,
- 8962, 9782, 9227, 10047, 10338, 11158, 6214, 7034, 7325, 8145, 7590, 8410,
- 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080,
- 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273,
- 12093, 12384, 13204, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851,
- 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636,
- 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204,
- 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008,
- 11828, 11273, 12093, 12384, 13204, 10306, 11126, 11417, 12237, 11682, 12502,
- 12793, 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250, 6214,
- 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227,
- 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567,
- 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 8260, 9080, 9371,
- 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093,
- 12384, 13204, 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943,
- 12763, 13054, 13874, 13319, 14139, 14430, 15250, 8260, 9080, 9371, 10191,
- 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384,
- 13204, 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763,
- 13054, 13874, 13319, 14139, 14430, 15250, 10306, 11126, 11417, 12237, 11682,
- 12502, 12793, 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250,
- 12352, 13172, 13463, 14283, 13728, 14548, 14839, 15659, 13989, 14809, 15100,
- 15920, 15365, 16185, 16476, 17296, 2122, 2942, 3233, 4053, 3498, 4318, 4609,
- 5429, 3759, 4579, 4870, 5690, 5135, 5955, 6246, 7066, 4168, 4988, 5279,
- 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292,
- 9112, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916,
- 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325, 8145, 7590, 8410, 8701,
- 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 4168, 4988, 5279,
- 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916, 7736, 7181, 8001, 8292,
- 9112, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962,
- 9782, 9227, 10047, 10338, 11158, 6214, 7034, 7325, 8145, 7590, 8410, 8701,
- 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371,
- 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093,
- 12384, 13204, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625,
- 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325, 8145, 7590, 8410,
- 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 6214, 7034,
- 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047,
- 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897,
- 10717, 11008, 11828, 11273, 12093, 12384, 13204, 6214, 7034, 7325, 8145,
- 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158,
- 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008,
- 11828, 11273, 12093, 12384, 13204, 8260, 9080, 9371, 10191, 9636, 10456,
- 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 10306,
- 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874,
- 13319, 14139, 14430, 15250, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475,
- 5805, 6625, 6916, 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325, 8145,
- 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158,
- 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782,
- 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747,
- 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 6214, 7034,
- 7325, 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047,
- 10338, 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897,
- 10717, 11008, 11828, 11273, 12093, 12384, 13204, 8260, 9080, 9371, 10191,
- 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384,
- 13204, 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763,
- 13054, 13874, 13319, 14139, 14430, 15250, 6214, 7034, 7325, 8145, 7590,
- 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260,
- 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828,
- 11273, 12093, 12384, 13204, 8260, 9080, 9371, 10191, 9636, 10456, 10747,
- 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 10306, 11126,
- 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874, 13319,
- 14139, 14430, 15250, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567,
- 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 10306, 11126, 11417,
- 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874, 13319, 14139,
- 14430, 15250, 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943,
- 12763, 13054, 13874, 13319, 14139, 14430, 15250, 12352, 13172, 13463, 14283,
- 13728, 14548, 14839, 15659, 13989, 14809, 15100, 15920, 15365, 16185, 16476,
- 17296, 4168, 4988, 5279, 6099, 5544, 6364, 6655, 7475, 5805, 6625, 6916,
- 7736, 7181, 8001, 8292, 9112, 6214, 7034, 7325, 8145, 7590, 8410, 8701,
- 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 6214, 7034, 7325,
- 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338,
- 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717,
- 11008, 11828, 11273, 12093, 12384, 13204, 6214, 7034, 7325, 8145, 7590,
- 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260,
- 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828,
- 11273, 12093, 12384, 13204, 8260, 9080, 9371, 10191, 9636, 10456, 10747,
- 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 10306, 11126,
- 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874, 13319,
- 14139, 14430, 15250, 6214, 7034, 7325, 8145, 7590, 8410, 8701, 9521, 7851,
- 8671, 8962, 9782, 9227, 10047, 10338, 11158, 8260, 9080, 9371, 10191, 9636,
- 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204,
- 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008,
- 11828, 11273, 12093, 12384, 13204, 10306, 11126, 11417, 12237, 11682, 12502,
- 12793, 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250, 8260,
- 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717, 11008, 11828,
- 11273, 12093, 12384, 13204, 10306, 11126, 11417, 12237, 11682, 12502, 12793,
- 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250, 10306, 11126,
- 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874, 13319,
- 14139, 14430, 15250, 12352, 13172, 13463, 14283, 13728, 14548, 14839, 15659,
- 13989, 14809, 15100, 15920, 15365, 16185, 16476, 17296, 6214, 7034, 7325,
- 8145, 7590, 8410, 8701, 9521, 7851, 8671, 8962, 9782, 9227, 10047, 10338,
- 11158, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567, 9897, 10717,
- 11008, 11828, 11273, 12093, 12384, 13204, 8260, 9080, 9371, 10191, 9636,
- 10456, 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204,
- 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054,
- 13874, 13319, 14139, 14430, 15250, 8260, 9080, 9371, 10191, 9636, 10456,
- 10747, 11567, 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 10306,
- 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874,
- 13319, 14139, 14430, 15250, 10306, 11126, 11417, 12237, 11682, 12502, 12793,
- 13613, 11943, 12763, 13054, 13874, 13319, 14139, 14430, 15250, 12352, 13172,
- 13463, 14283, 13728, 14548, 14839, 15659, 13989, 14809, 15100, 15920, 15365,
- 16185, 16476, 17296, 8260, 9080, 9371, 10191, 9636, 10456, 10747, 11567,
- 9897, 10717, 11008, 11828, 11273, 12093, 12384, 13204, 10306, 11126, 11417,
- 12237, 11682, 12502, 12793, 13613, 11943, 12763, 13054, 13874, 13319, 14139,
- 14430, 15250, 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943,
- 12763, 13054, 13874, 13319, 14139, 14430, 15250, 12352, 13172, 13463, 14283,
- 13728, 14548, 14839, 15659, 13989, 14809, 15100, 15920, 15365, 16185, 16476,
- 17296, 10306, 11126, 11417, 12237, 11682, 12502, 12793, 13613, 11943, 12763,
- 13054, 13874, 13319, 14139, 14430, 15250, 12352, 13172, 13463, 14283, 13728,
- 14548, 14839, 15659, 13989, 14809, 15100, 15920, 15365, 16185, 16476, 17296,
- 12352, 13172, 13463, 14283, 13728, 14548, 14839, 15659, 13989, 14809, 15100,
- 15920, 15365, 16185, 16476, 17296, 14398, 15218, 15509, 16329, 15774, 16594,
- 16885, 17705, 16035, 16855, 17146, 17966, 17411, 18231, 18522, 19342
-};
+const int vp9_cat6_high10_high_cost[256] = {
+ 94, 2257, 2733, 4896, 3154, 5317, 5793, 7956, 3672, 5835, 6311,
+ 8474, 6732, 8895, 9371, 11534, 3672, 5835, 6311, 8474, 6732, 8895,
+ 9371, 11534, 7250, 9413, 9889, 12052, 10310, 12473, 12949, 15112, 3672,
+ 5835, 6311, 8474, 6732, 8895, 9371, 11534, 7250, 9413, 9889, 12052,
+ 10310, 12473, 12949, 15112, 7250, 9413, 9889, 12052, 10310, 12473, 12949,
+ 15112, 10828, 12991, 13467, 15630, 13888, 16051, 16527, 18690, 4187, 6350,
+ 6826, 8989, 7247, 9410, 9886, 12049, 7765, 9928, 10404, 12567, 10825,
+ 12988, 13464, 15627, 7765, 9928, 10404, 12567, 10825, 12988, 13464, 15627,
+ 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 7765, 9928, 10404,
+ 12567, 10825, 12988, 13464, 15627, 11343, 13506, 13982, 16145, 14403, 16566,
+ 17042, 19205, 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 14921,
+ 17084, 17560, 19723, 17981, 20144, 20620, 22783, 4187, 6350, 6826, 8989,
+ 7247, 9410, 9886, 12049, 7765, 9928, 10404, 12567, 10825, 12988, 13464,
+ 15627, 7765, 9928, 10404, 12567, 10825, 12988, 13464, 15627, 11343, 13506,
+ 13982, 16145, 14403, 16566, 17042, 19205, 7765, 9928, 10404, 12567, 10825,
+ 12988, 13464, 15627, 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205,
+ 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 14921, 17084, 17560,
+ 19723, 17981, 20144, 20620, 22783, 8280, 10443, 10919, 13082, 11340, 13503,
+ 13979, 16142, 11858, 14021, 14497, 16660, 14918, 17081, 17557, 19720, 11858,
+ 14021, 14497, 16660, 14918, 17081, 17557, 19720, 15436, 17599, 18075, 20238,
+ 18496, 20659, 21135, 23298, 11858, 14021, 14497, 16660, 14918, 17081, 17557,
+ 19720, 15436, 17599, 18075, 20238, 18496, 20659, 21135, 23298, 15436, 17599,
+ 18075, 20238, 18496, 20659, 21135, 23298, 19014, 21177, 21653, 23816, 22074,
+ 24237, 24713, 26876};
+const int vp9_cat6_high12_high_cost[1024] = {
+ 100, 2263, 2739, 4902, 3160, 5323, 5799, 7962, 3678, 5841, 6317,
+ 8480, 6738, 8901, 9377, 11540, 3678, 5841, 6317, 8480, 6738, 8901,
+ 9377, 11540, 7256, 9419, 9895, 12058, 10316, 12479, 12955, 15118, 3678,
+ 5841, 6317, 8480, 6738, 8901, 9377, 11540, 7256, 9419, 9895, 12058,
+ 10316, 12479, 12955, 15118, 7256, 9419, 9895, 12058, 10316, 12479, 12955,
+ 15118, 10834, 12997, 13473, 15636, 13894, 16057, 16533, 18696, 4193, 6356,
+ 6832, 8995, 7253, 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831,
+ 12994, 13470, 15633, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633,
+ 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410,
+ 12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572,
+ 17048, 19211, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927,
+ 17090, 17566, 19729, 17987, 20150, 20626, 22789, 4193, 6356, 6832, 8995,
+ 7253, 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831, 12994, 13470,
+ 15633, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349, 13512,
+ 13988, 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410, 12573, 10831,
+ 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211,
+ 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927, 17090, 17566,
+ 19729, 17987, 20150, 20626, 22789, 8286, 10449, 10925, 13088, 11346, 13509,
+ 13985, 16148, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864,
+ 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244,
+ 18502, 20665, 21141, 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563,
+ 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605,
+ 18081, 20244, 18502, 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080,
+ 24243, 24719, 26882, 4193, 6356, 6832, 8995, 7253, 9416, 9892, 12055,
+ 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 7771, 9934, 10410,
+ 12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572,
+ 17048, 19211, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349,
+ 13512, 13988, 16151, 14409, 16572, 17048, 19211, 11349, 13512, 13988, 16151,
+ 14409, 16572, 17048, 19211, 14927, 17090, 17566, 19729, 17987, 20150, 20626,
+ 22789, 8286, 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027,
+ 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924,
+ 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304,
+ 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081,
+ 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665,
+ 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882, 8286,
+ 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027, 14503, 16666,
+ 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924, 17087, 17563,
+ 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 11864, 14027,
+ 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502,
+ 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304,
+ 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882, 12379, 14542, 15018,
+ 17181, 15439, 17602, 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180,
+ 21656, 23819, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535,
+ 21698, 22174, 24337, 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759,
+ 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234,
+ 27397, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276,
+ 25752, 27915, 26173, 28336, 28812, 30975, 4193, 6356, 6832, 8995, 7253,
+ 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633,
+ 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988,
+ 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410, 12573, 10831, 12994,
+ 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 11349,
+ 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927, 17090, 17566, 19729,
+ 17987, 20150, 20626, 22789, 8286, 10449, 10925, 13088, 11346, 13509, 13985,
+ 16148, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027,
+ 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502,
+ 20665, 21141, 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726,
+ 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081,
+ 20244, 18502, 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243,
+ 24719, 26882, 8286, 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864,
+ 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666,
+ 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141,
+ 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605,
+ 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502,
+ 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882,
+ 12379, 14542, 15018, 17181, 15439, 17602, 18078, 20241, 15957, 18120, 18596,
+ 20759, 19017, 21180, 21656, 23819, 15957, 18120, 18596, 20759, 19017, 21180,
+ 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 15957,
+ 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337,
+ 22595, 24758, 25234, 27397, 19535, 21698, 22174, 24337, 22595, 24758, 25234,
+ 27397, 23113, 25276, 25752, 27915, 26173, 28336, 28812, 30975, 8286, 10449,
+ 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027, 14503, 16666, 14924,
+ 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726,
+ 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 11864, 14027, 14503,
+ 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665,
+ 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 19020,
+ 21183, 21659, 23822, 22080, 24243, 24719, 26882, 12379, 14542, 15018, 17181,
+ 15439, 17602, 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180, 21656,
+ 23819, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698,
+ 22174, 24337, 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759, 19017,
+ 21180, 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397,
+ 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276, 25752,
+ 27915, 26173, 28336, 28812, 30975, 12379, 14542, 15018, 17181, 15439, 17602,
+ 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 15957,
+ 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337,
+ 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759, 19017, 21180, 21656,
+ 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 19535, 21698,
+ 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276, 25752, 27915, 26173,
+ 28336, 28812, 30975, 16472, 18635, 19111, 21274, 19532, 21695, 22171, 24334,
+ 20050, 22213, 22689, 24852, 23110, 25273, 25749, 27912, 20050, 22213, 22689,
+ 24852, 23110, 25273, 25749, 27912, 23628, 25791, 26267, 28430, 26688, 28851,
+ 29327, 31490, 20050, 22213, 22689, 24852, 23110, 25273, 25749, 27912, 23628,
+ 25791, 26267, 28430, 26688, 28851, 29327, 31490, 23628, 25791, 26267, 28430,
+ 26688, 28851, 29327, 31490, 27206, 29369, 29845, 32008, 30266, 32429, 32905,
+ 35068};
#endif
const vp9_extra_bit vp9_extra_bits[ENTROPY_TOKENS] = {
diff --git a/vp9/encoder/vp9_tokenize.h b/vp9/encoder/vp9_tokenize.h
index 26da6fe..df979b2 100644
--- a/vp9/encoder/vp9_tokenize.h
+++ b/vp9/encoder/vp9_tokenize.h
@@ -75,25 +75,25 @@
extern const TOKENVALUE *vp9_dct_value_tokens_ptr;
extern const TOKENVALUE *vp9_dct_cat_lt_10_value_tokens;
extern const int16_t vp9_cat6_low_cost[256];
-extern const int16_t vp9_cat6_high_cost[128];
-extern const int16_t vp9_cat6_high10_high_cost[512];
-extern const int16_t vp9_cat6_high12_high_cost[2048];
-static INLINE int16_t vp9_get_cost(int16_t token, EXTRABIT extrabits,
- const int16_t *cat6_high_table) {
+extern const int vp9_cat6_high_cost[64];
+extern const int vp9_cat6_high10_high_cost[256];
+extern const int vp9_cat6_high12_high_cost[1024];
+static INLINE int vp9_get_cost(int16_t token, EXTRABIT extrabits,
+ const int *cat6_high_table) {
if (token != CATEGORY6_TOKEN)
- return vp9_extra_bits[token].cost[extrabits];
- return vp9_cat6_low_cost[extrabits & 0xff]
- + cat6_high_table[extrabits >> 8];
+ return vp9_extra_bits[token].cost[extrabits >> 1];
+ return vp9_cat6_low_cost[(extrabits >> 1) & 0xff]
+ + cat6_high_table[extrabits >> 9];
}
#if CONFIG_VP9_HIGHBITDEPTH
-static INLINE const int16_t* vp9_get_high_cost_table(int bit_depth) {
+static INLINE const int* vp9_get_high_cost_table(int bit_depth) {
return bit_depth == 8 ? vp9_cat6_high_cost
: (bit_depth == 10 ? vp9_cat6_high10_high_cost :
vp9_cat6_high12_high_cost);
}
#else
-static INLINE const int16_t* vp9_get_high_cost_table(int bit_depth) {
+static INLINE const int* vp9_get_high_cost_table(int bit_depth) {
(void) bit_depth;
return vp9_cat6_high_cost;
}
diff --git a/vp9/encoder/x86/vp9_diamond_search_sad_avx.c b/vp9/encoder/x86/vp9_diamond_search_sad_avx.c
index 2ed3f1a..b475f8d 100644
--- a/vp9/encoder/x86/vp9_diamond_search_sad_avx.c
+++ b/vp9/encoder/x86/vp9_diamond_search_sad_avx.c
@@ -19,11 +19,11 @@
#include "vpx_ports/mem.h"
#ifdef __GNUC__
-# define __likely__(v) __builtin_expect(v, 1)
-# define __unlikely__(v) __builtin_expect(v, 0)
+# define LIKELY(v) __builtin_expect(v, 1)
+# define UNLIKELY(v) __builtin_expect(v, 0)
#else
-# define __likely__(v) (v)
-# define __unlikely__(v) (v)
+# define LIKELY(v) (v)
+# define UNLIKELY(v) (v)
#endif
static INLINE int_mv pack_int_mv(int16_t row, int16_t col) {
@@ -50,8 +50,9 @@
int error_per_bit) {
const int_mv diff = pack_int_mv(mv.as_mv.row - ref->row,
mv.as_mv.col - ref->col);
- return ROUND_POWER_OF_TWO(mv_cost(diff, x->nmvjointsadcost,
- x->nmvsadcost) * error_per_bit, 8);
+ return ROUND_POWER_OF_TWO((unsigned)mv_cost(diff, x->nmvjointsadcost,
+ x->nmvsadcost) *
+ error_per_bit, 8);
}
/*****************************************************************************
@@ -162,7 +163,7 @@
v_inside_d = _mm_cmpeq_epi32(v_these_mv_clamp_w, v_these_mv_w);
// If none of them are inside, then move on
- if (__likely__(_mm_test_all_zeros(v_inside_d, v_inside_d))) {
+ if (LIKELY(_mm_test_all_zeros(v_inside_d, v_inside_d))) {
continue;
}
@@ -268,7 +269,7 @@
// If the local best value is not saturated, just use it, otherwise
// find the horizontal minimum again the hard way on 32 bits.
// This is executed rarely.
- if (__unlikely__(local_best_sad == 0xffff)) {
+ if (UNLIKELY(local_best_sad == 0xffff)) {
__m128i v_loval_d, v_hival_d, v_loidx_d, v_hiidx_d, v_sel_d;
v_loval_d = v_sad_d;
@@ -293,7 +294,7 @@
}
// Update the global minimum if the local minimum is smaller
- if (__likely__(local_best_sad < best_sad)) {
+ if (LIKELY(local_best_sad < best_sad)) {
new_bmv = ((const int_mv *)&v_these_mv_w)[local_best_idx];
new_best_address = ((const uint8_t **)v_blocka)[local_best_idx];
@@ -312,7 +313,7 @@
v_ba_d = _mm_set1_epi32((intptr_t)best_address);
#endif
- if (__unlikely__(best_address == in_what)) {
+ if (UNLIKELY(best_address == in_what)) {
(*num00)++;
}
}
diff --git a/vpx_dsp/intrapred.c b/vpx_dsp/intrapred.c
index a9669e5..ba35b23 100644
--- a/vpx_dsp/intrapred.c
+++ b/vpx_dsp/intrapred.c
@@ -44,6 +44,7 @@
dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
}
+#if CONFIG_MISC_FIXES
static INLINE void d207e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
@@ -58,6 +59,7 @@
dst += stride;
}
}
+#endif // CONFIG_MISC_FIXES
static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
@@ -76,6 +78,7 @@
}
}
+#if CONFIG_MISC_FIXES
static INLINE void d63e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
@@ -89,6 +92,7 @@
dst += stride;
}
}
+#endif // CONFIG_MISC_FIXES
static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
@@ -109,6 +113,7 @@
}
}
+#if CONFIG_MISC_FIXES
static INLINE void d45e_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
int r, c;
@@ -121,6 +126,7 @@
dst += stride;
}
}
+#endif // CONFIG_MISC_FIXES
static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
@@ -528,6 +534,7 @@
}
}
+#if CONFIG_MISC_FIXES
static INLINE void highbd_d207e_predictor(uint16_t *dst, ptrdiff_t stride,
int bs, const uint16_t *above,
const uint16_t *left, int bd) {
@@ -544,6 +551,7 @@
dst += stride;
}
}
+#endif // CONFIG_MISC_FIXES
static INLINE void highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride,
int bs, const uint16_t *above,
@@ -579,6 +587,7 @@
}
}
+#if CONFIG_MISC_FIXES
static INLINE void highbd_d45e_predictor(uint16_t *dst, ptrdiff_t stride,
int bs, const uint16_t *above,
const uint16_t *left, int bd) {
@@ -593,6 +602,7 @@
dst += stride;
}
}
+#endif // CONFIG_MISC_FIXES
static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
int bs, const uint16_t *above,
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 6c6f15e..73726d2 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -699,7 +699,7 @@
specialize qw/vpx_iwht4x4_1_add/;
add_proto qw/void vpx_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_iwht4x4_16_add/;
+ specialize qw/vpx_iwht4x4_16_add/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
specialize qw/vpx_highbd_idct4x4_1_add/;
@@ -762,7 +762,7 @@
add_proto qw/void vpx_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_1_add/;
-
+
add_proto qw/void vpx_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
specialize qw/vpx_highbd_idct4x4_16_add/;
@@ -785,10 +785,10 @@
specialize qw/vpx_idct4x4_1_add sse2/;
add_proto qw/void vpx_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_idct8x8_64_add sse2/;
+ specialize qw/vpx_idct8x8_64_add sse2/, "$ssse3_x86_64_x86inc";
add_proto qw/void vpx_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_idct8x8_12_add sse2/;
+ specialize qw/vpx_idct8x8_12_add sse2/, "$ssse3_x86_64_x86inc";
add_proto qw/void vpx_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct8x8_1_add sse2/;
@@ -803,14 +803,15 @@
specialize qw/vpx_idct16x16_1_add sse2/;
add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_idct32x32_1024_add sse2/;
+ specialize qw/vpx_idct32x32_1024_add sse2/, "$ssse3_x86_64_x86inc";
add_proto qw/void vpx_idct32x32_135_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_idct32x32_135_add sse2/;
+ specialize qw/vpx_idct32x32_135_add sse2/, "$ssse3_x86_64_x86inc";
+ # Need to add 135 eob idct32x32 implementations.
$vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
add_proto qw/void vpx_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_idct32x32_34_add sse2/;
+ specialize qw/vpx_idct32x32_34_add sse2/, "$ssse3_x86_64_x86inc";
add_proto qw/void vpx_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_1_add sse2/;
@@ -1156,10 +1157,10 @@
specialize qw/vpx_sad8x4x4d msa/, "$sse2_x86inc";
add_proto qw/void vpx_sad4x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad4x8x4d msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x8x4d msa/, "$sse2_x86inc";
add_proto qw/void vpx_sad4x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
-specialize qw/vpx_sad4x4x4d msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x4x4d msa/, "$sse2_x86inc";
#
# Structured Similarity (SSIM)
diff --git a/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm b/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
index 80a330b..20baf82 100644
--- a/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
+++ b/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
@@ -220,7 +220,24 @@
mova m12, [pw_11585x2]
lea r3, [2 * strideq]
-
+%if CONFIG_VP9_HIGHBITDEPTH
+ mova m0, [inputq + 0]
+ packssdw m0, [inputq + 16]
+ mova m1, [inputq + 32]
+ packssdw m1, [inputq + 48]
+ mova m2, [inputq + 64]
+ packssdw m2, [inputq + 80]
+ mova m3, [inputq + 96]
+ packssdw m3, [inputq + 112]
+ mova m4, [inputq + 128]
+ packssdw m4, [inputq + 144]
+ mova m5, [inputq + 160]
+ packssdw m5, [inputq + 176]
+ mova m6, [inputq + 192]
+ packssdw m6, [inputq + 208]
+ mova m7, [inputq + 224]
+ packssdw m7, [inputq + 240]
+%else
mova m0, [inputq + 0]
mova m1, [inputq + 16]
mova m2, [inputq + 32]
@@ -229,7 +246,7 @@
mova m5, [inputq + 80]
mova m6, [inputq + 96]
mova m7, [inputq + 112]
-
+%endif
TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
IDCT8_1D
TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
@@ -254,10 +271,21 @@
lea r3, [2 * strideq]
+%if CONFIG_VP9_HIGHBITDEPTH
+ mova m0, [inputq + 0]
+ packssdw m0, [inputq + 16]
+ mova m1, [inputq + 32]
+ packssdw m1, [inputq + 48]
+ mova m2, [inputq + 64]
+ packssdw m2, [inputq + 80]
+ mova m3, [inputq + 96]
+ packssdw m3, [inputq + 112]
+%else
mova m0, [inputq + 0]
mova m1, [inputq + 16]
mova m2, [inputq + 32]
mova m3, [inputq + 48]
+%endif
punpcklwd m0, m1
punpcklwd m2, m3
@@ -765,6 +793,24 @@
lea r4, [rsp + transposed_in]
idct32x32_34_transpose:
+%if CONFIG_VP9_HIGHBITDEPTH
+ mova m0, [r3 + 0]
+ packssdw m0, [r3 + 16]
+ mova m1, [r3 + 32 * 4]
+ packssdw m1, [r3 + 32 * 4 + 16]
+ mova m2, [r3 + 32 * 8]
+ packssdw m2, [r3 + 32 * 8 + 16]
+ mova m3, [r3 + 32 * 12]
+ packssdw m3, [r3 + 32 * 12 + 16]
+ mova m4, [r3 + 32 * 16]
+ packssdw m4, [r3 + 32 * 16 + 16]
+ mova m5, [r3 + 32 * 20]
+ packssdw m5, [r3 + 32 * 20 + 16]
+ mova m6, [r3 + 32 * 24]
+ packssdw m6, [r3 + 32 * 24 + 16]
+ mova m7, [r3 + 32 * 28]
+ packssdw m7, [r3 + 32 * 28 + 16]
+%else
mova m0, [r3 + 0]
mova m1, [r3 + 16 * 4]
mova m2, [r3 + 16 * 8]
@@ -773,6 +819,7 @@
mova m5, [r3 + 16 * 20]
mova m6, [r3 + 16 * 24]
mova m7, [r3 + 16 * 28]
+%endif
TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
@@ -1176,6 +1223,24 @@
mov r7, 2
idct32x32_135_transpose:
+%if CONFIG_VP9_HIGHBITDEPTH
+ mova m0, [r3 + 0]
+ packssdw m0, [r3 + 16]
+ mova m1, [r3 + 32 * 4]
+ packssdw m1, [r3 + 32 * 4 + 16]
+ mova m2, [r3 + 32 * 8]
+ packssdw m2, [r3 + 32 * 8 + 16]
+ mova m3, [r3 + 32 * 12]
+ packssdw m3, [r3 + 32 * 12 + 16]
+ mova m4, [r3 + 32 * 16]
+ packssdw m4, [r3 + 32 * 16 + 16]
+ mova m5, [r3 + 32 * 20]
+ packssdw m5, [r3 + 32 * 20 + 16]
+ mova m6, [r3 + 32 * 24]
+ packssdw m6, [r3 + 32 * 24 + 16]
+ mova m7, [r3 + 32 * 28]
+ packssdw m7, [r3 + 32 * 28 + 16]
+%else
mova m0, [r3 + 0]
mova m1, [r3 + 16 * 4]
mova m2, [r3 + 16 * 8]
@@ -1184,7 +1249,7 @@
mova m5, [r3 + 16 * 20]
mova m6, [r3 + 16 * 24]
mova m7, [r3 + 16 * 28]
-
+%endif
TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
mova [r4 + 0], m0
@@ -1196,14 +1261,22 @@
mova [r4 + 16 * 6], m6
mova [r4 + 16 * 7], m7
+%if CONFIG_VP9_HIGHBITDEPTH
+ add r3, 32
+%else
add r3, 16
+%endif
add r4, 16 * 8
dec r7
jne idct32x32_135_transpose
IDCT32X32_135 16*0, 16*32, 16*64, 16*96
lea stp, [stp + 16 * 8]
+%if CONFIG_VP9_HIGHBITDEPTH
+ lea inputq, [inputq + 32 * 32]
+%else
lea inputq, [inputq + 16 * 32]
+%endif
dec r6
jnz idct32x32_135
@@ -1614,6 +1687,24 @@
mov r7, 4
idct32x32_1024_transpose:
+%if CONFIG_VP9_HIGHBITDEPTH
+ mova m0, [r3 + 0]
+ packssdw m0, [r3 + 16]
+ mova m1, [r3 + 32 * 4]
+ packssdw m1, [r3 + 32 * 4 + 16]
+ mova m2, [r3 + 32 * 8]
+ packssdw m2, [r3 + 32 * 8 + 16]
+ mova m3, [r3 + 32 * 12]
+ packssdw m3, [r3 + 32 * 12 + 16]
+ mova m4, [r3 + 32 * 16]
+ packssdw m4, [r3 + 32 * 16 + 16]
+ mova m5, [r3 + 32 * 20]
+ packssdw m5, [r3 + 32 * 20 + 16]
+ mova m6, [r3 + 32 * 24]
+ packssdw m6, [r3 + 32 * 24 + 16]
+ mova m7, [r3 + 32 * 28]
+ packssdw m7, [r3 + 32 * 28 + 16]
+%else
mova m0, [r3 + 0]
mova m1, [r3 + 16 * 4]
mova m2, [r3 + 16 * 8]
@@ -1622,6 +1713,7 @@
mova m5, [r3 + 16 * 20]
mova m6, [r3 + 16 * 24]
mova m7, [r3 + 16 * 28]
+%endif
TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
@@ -1633,8 +1725,11 @@
mova [r4 + 16 * 5], m5
mova [r4 + 16 * 6], m6
mova [r4 + 16 * 7], m7
-
+%if CONFIG_VP9_HIGHBITDEPTH
+ add r3, 32
+%else
add r3, 16
+%endif
add r4, 16 * 8
dec r7
jne idct32x32_1024_transpose
@@ -1642,7 +1737,11 @@
IDCT32X32_1024 16*0, 16*32, 16*64, 16*96
lea stp, [stp + 16 * 8]
+%if CONFIG_VP9_HIGHBITDEPTH
+ lea inputq, [inputq + 32 * 32]
+%else
lea inputq, [inputq + 16 * 32]
+%endif
dec r6
jnz idct32x32_1024
diff --git a/vpx_dsp/x86/inv_wht_sse2.asm b/vpx_dsp/x86/inv_wht_sse2.asm
index df6f469..fbbcd76 100644
--- a/vpx_dsp/x86/inv_wht_sse2.asm
+++ b/vpx_dsp/x86/inv_wht_sse2.asm
@@ -82,9 +82,15 @@
INIT_XMM sse2
cglobal iwht4x4_16_add, 3, 3, 7, input, output, stride
+%if CONFIG_VP9_HIGHBITDEPTH
+ mova m0, [inputq + 0]
+ packssdw m0, [inputq + 16]
+ mova m1, [inputq + 32]
+ packssdw m1, [inputq + 48]
+%else
mova m0, [inputq + 0]
mova m1, [inputq + 16]
-
+%endif
psraw m0, 2
psraw m1, 2
diff --git a/vpx_dsp/x86/sad4d_sse2.asm b/vpx_dsp/x86/sad4d_sse2.asm
index a2f0ae7..3f6e55c 100644
--- a/vpx_dsp/x86/sad4d_sse2.asm
+++ b/vpx_dsp/x86/sad4d_sse2.asm
@@ -20,33 +20,41 @@
movd m4, [ref2q+%3]
movd m7, [ref3q+%3]
movd m5, [ref4q+%3]
- punpckldq m0, [srcq +%4]
- punpckldq m6, [ref1q+%5]
- punpckldq m4, [ref2q+%5]
- punpckldq m7, [ref3q+%5]
- punpckldq m5, [ref4q+%5]
+ movd m1, [srcq +%4]
+ movd m2, [ref1q+%5]
+ punpckldq m0, m1
+ punpckldq m6, m2
+ movd m1, [ref2q+%5]
+ movd m2, [ref3q+%5]
+ movd m3, [ref4q+%5]
+ punpckldq m4, m1
+ punpckldq m7, m2
+ punpckldq m5, m3
+ movlhps m0, m0
+ movlhps m6, m4
+ movlhps m7, m5
psadbw m6, m0
- psadbw m4, m0
psadbw m7, m0
- psadbw m5, m0
- punpckldq m6, m4
- punpckldq m7, m5
%else
movd m1, [ref1q+%3]
+ movd m5, [ref1q+%5]
movd m2, [ref2q+%3]
+ movd m4, [ref2q+%5]
+ punpckldq m1, m5
+ punpckldq m2, m4
movd m3, [ref3q+%3]
+ movd m5, [ref3q+%5]
+ punpckldq m3, m5
movd m4, [ref4q+%3]
- punpckldq m0, [srcq +%4]
- punpckldq m1, [ref1q+%5]
- punpckldq m2, [ref2q+%5]
- punpckldq m3, [ref3q+%5]
- punpckldq m4, [ref4q+%5]
+ movd m5, [ref4q+%5]
+ punpckldq m4, m5
+ movd m5, [srcq +%4]
+ punpckldq m0, m5
+ movlhps m0, m0
+ movlhps m1, m2
+ movlhps m3, m4
psadbw m1, m0
- psadbw m2, m0
psadbw m3, m0
- psadbw m4, m0
- punpckldq m1, m2
- punpckldq m3, m4
paddd m6, m1
paddd m7, m3
%endif
@@ -170,7 +178,7 @@
; void vpx_sadNxNx4d_sse2(uint8_t *src, int src_stride,
; uint8_t *ref[4], int ref_stride,
; uint32_t res[4]);
-; where NxN = 64x64, 32x32, 16x16, 16x8, 8x16 or 8x8
+; where NxN = 64x64, 32x32, 16x16, 16x8, 8x16, 8x8, 8x4, 4x8 and 4x4
%macro SADNXN4D 2
%if UNIX64
cglobal sad%1x%2x4d, 5, 8, 8, src, src_stride, ref1, ref_stride, \
@@ -192,7 +200,7 @@
%endrep
PROCESS_%1x2x4 0, 0, 0, src_strideq, ref_strideq, 0
-%if mmsize == 16
+%if %1 > 4
pslldq m5, 4
pslldq m7, 4
por m4, m5
@@ -207,8 +215,10 @@
RET
%else
movifnidn r4, r4mp
- movq [r4+0], m6
- movq [r4+8], m7
+ pshufd m6, m6, 0x08
+ pshufd m7, m7, 0x08
+ movq [r4+0], m6
+ movq [r4+8], m7
RET
%endif
%endmacro
@@ -225,7 +235,5 @@
SADNXN4D 8, 16
SADNXN4D 8, 8
SADNXN4D 8, 4
-
-INIT_MMX sse
SADNXN4D 4, 8
SADNXN4D 4, 4
diff --git a/vpx_mem/vpx_mem.c b/vpx_mem/vpx_mem.c
index b98fe83..b261fc0 100644
--- a/vpx_mem/vpx_mem.c
+++ b/vpx_mem/vpx_mem.c
@@ -9,8 +9,6 @@
*/
-#define __VPX_MEM_C__
-
#include "vpx_mem.h"
#include <stdio.h>
#include <stdlib.h>
diff --git a/vpx_scale/generic/yv12config.c b/vpx_scale/generic/yv12config.c
index e1ee46e..e8fee52 100644
--- a/vpx_scale/generic/yv12config.c
+++ b/vpx_scale/generic/yv12config.c
@@ -160,29 +160,12 @@
const uint64_t uvplane_size = (uv_height + 2 * uv_border_h) *
(uint64_t)uv_stride + byte_alignment;
-#if CONFIG_ALPHA
- const int alpha_width = aligned_width;
- const int alpha_height = aligned_height;
- const int alpha_stride = y_stride;
- const int alpha_border_w = border;
- const int alpha_border_h = border;
- const uint64_t alpha_plane_size = (alpha_height + 2 * alpha_border_h) *
- (uint64_t)alpha_stride + byte_alignment;
-#if CONFIG_VP9_HIGHBITDEPTH
- const uint64_t frame_size = (1 + use_highbitdepth) *
- (yplane_size + 2 * uvplane_size + alpha_plane_size);
-#else
- const uint64_t frame_size = yplane_size + 2 * uvplane_size +
- alpha_plane_size;
-#endif // CONFIG_VP9_HIGHBITDEPTH
-#else
#if CONFIG_VP9_HIGHBITDEPTH
const uint64_t frame_size =
(1 + use_highbitdepth) * (yplane_size + 2 * uvplane_size);
#else
const uint64_t frame_size = yplane_size + 2 * uvplane_size;
#endif // CONFIG_VP9_HIGHBITDEPTH
-#endif // CONFIG_ALPHA
uint8_t *buf = NULL;
@@ -277,14 +260,6 @@
buf + yplane_size + uvplane_size + (uv_border_h * uv_stride) +
uv_border_w, vp9_byte_align);
-#if CONFIG_ALPHA
- ybf->alpha_width = alpha_width;
- ybf->alpha_height = alpha_height;
- ybf->alpha_stride = alpha_stride;
- ybf->alpha_buffer = (uint8_t *)yv12_align_addr(
- buf + yplane_size + 2 * uvplane_size +
- (alpha_border_h * alpha_stride) + alpha_border_w, vp9_byte_align);
-#endif
ybf->corrupted = 0; /* assume not corrupted by errors */
return 0;
}
diff --git a/vpxenc.c b/vpxenc.c
index 8798e69..c61d060 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -454,6 +454,9 @@
&frame_parallel_decoding, &aq_mode, &frame_periodic_boost,
&noise_sens, &tune_content, &input_color_space,
&min_gf_interval, &max_gf_interval,
+#if CONFIG_VP9_HIGHBITDEPTH
+ &bitdeptharg, &inbitdeptharg,
+#endif // CONFIG_VP9_HIGHBITDEPTH
NULL
};
static const int vp9_arg_ctrl_map[] = {
@@ -480,6 +483,9 @@
&frame_parallel_decoding, &aq_mode, &frame_periodic_boost,
&noise_sens, &tune_content, &input_color_space,
&min_gf_interval, &max_gf_interval,
+#if CONFIG_VP9_HIGHBITDEPTH
+ &bitdeptharg, &inbitdeptharg,
+#endif // CONFIG_VP9_HIGHBITDEPTH
NULL
};
static const int vp10_arg_ctrl_map[] = {
