Merge "Speed up rd selection in OBMC experiment" into nextgenv2
diff --git a/configure b/configure
index e7eb152..97366e4 100755
--- a/configure
+++ b/configure
@@ -284,7 +284,6 @@
ext_partition
ext_tile
obmc
- affine_motion
"
CONFIG_LIST="
dependency_tracking
diff --git a/test/vp10_convolve_test.cc b/test/vp10_convolve_test.cc
index 07b0dda..122a8e5 100644
--- a/test/vp10_convolve_test.cc
+++ b/test/vp10_convolve_test.cc
@@ -5,6 +5,7 @@
#include "vp10/common/filter.h"
#include "vp10/common/vp10_convolve.h"
#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_ports/mem.h"
using libvpx_test::ACMRandom;
@@ -270,4 +271,132 @@
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
+
+#define CONVOLVE_SPEED_TEST 0
+#if CONVOLVE_SPEED_TEST
+#define highbd_convolve_speed(func, block_size, frame_size) \
+ TEST(VP10ConvolveTest, func##_speed_##block_size##_##frame_size) { \
+ ACMRandom rnd(ACMRandom::DeterministicSeed()); \
+ INTERP_FILTER interp_filter = EIGHTTAP; \
+ InterpFilterParams filter_params = \
+ vp10_get_interp_filter_params(interp_filter); \
+ ptrdiff_t filter_size = filter_params.tap; \
+ int filter_center = filter_size / 2 - 1; \
+ DECLARE_ALIGNED(16, uint16_t, \
+ src[(frame_size + 7) * (frame_size + 7)]) = {0}; \
+ int src_stride = frame_size + 7; \
+ DECLARE_ALIGNED(16, uint16_t, dst[frame_size * frame_size]) = {0}; \
+ int dst_stride = frame_size; \
+ int x_step_q4 = 16; \
+ int y_step_q4 = 16; \
+ int subpel_x_q4 = 8; \
+ int subpel_y_q4 = 6; \
+ int bd = 10; \
+ \
+ int w = block_size; \
+ int h = block_size; \
+ \
+ const int16_t* filter_x = \
+ vp10_get_interp_filter_kernel(filter_params, subpel_x_q4); \
+ const int16_t* filter_y = \
+ vp10_get_interp_filter_kernel(filter_params, subpel_y_q4); \
+ \
+ for (int i = 0; i < src_stride * src_stride; i++) { \
+ src[i] = rnd.Rand16() % (1 << bd); \
+ } \
+ \
+ int offset = filter_center * src_stride + filter_center; \
+ int row_offset = 0; \
+ int col_offset = 0; \
+ for (int i = 0; i < 100000; i++) { \
+ int src_total_offset = offset + col_offset * src_stride + row_offset; \
+ int dst_total_offset = col_offset * dst_stride + row_offset; \
+ func(CONVERT_TO_BYTEPTR(src + src_total_offset), src_stride, \
+ CONVERT_TO_BYTEPTR(dst + dst_total_offset), dst_stride, filter_x, \
+ x_step_q4, filter_y, y_step_q4, w, h, bd); \
+ if (offset + w + w < frame_size) { \
+ row_offset += w; \
+ } else { \
+ row_offset = 0; \
+ col_offset += h; \
+ } \
+ if (col_offset + h >= frame_size) { \
+ col_offset = 0; \
+ } \
+ } \
+ }
+
+#define lowbd_convolve_speed(func, block_size, frame_size) \
+ TEST(VP10ConvolveTest, func##_speed_l_##block_size##_##frame_size) { \
+ ACMRandom rnd(ACMRandom::DeterministicSeed()); \
+ INTERP_FILTER interp_filter = EIGHTTAP; \
+ InterpFilterParams filter_params = \
+ vp10_get_interp_filter_params(interp_filter); \
+ ptrdiff_t filter_size = filter_params.tap; \
+ int filter_center = filter_size / 2 - 1; \
+ DECLARE_ALIGNED(16, uint8_t, src[(frame_size + 7) * (frame_size + 7)]); \
+ int src_stride = frame_size + 7; \
+ DECLARE_ALIGNED(16, uint8_t, dst[frame_size * frame_size]); \
+ int dst_stride = frame_size; \
+ int x_step_q4 = 16; \
+ int y_step_q4 = 16; \
+ int subpel_x_q4 = 8; \
+ int subpel_y_q4 = 6; \
+ int bd = 8; \
+ \
+ int w = block_size; \
+ int h = block_size; \
+ \
+ const int16_t* filter_x = \
+ vp10_get_interp_filter_kernel(filter_params, subpel_x_q4); \
+ const int16_t* filter_y = \
+ vp10_get_interp_filter_kernel(filter_params, subpel_y_q4); \
+ \
+ for (int i = 0; i < src_stride * src_stride; i++) { \
+ src[i] = rnd.Rand16() % (1 << bd); \
+ } \
+ \
+ int offset = filter_center * src_stride + filter_center; \
+ int row_offset = 0; \
+ int col_offset = 0; \
+ for (int i = 0; i < 100000; i++) { \
+ func(src + offset, src_stride, dst, dst_stride, filter_x, x_step_q4, \
+ filter_y, y_step_q4, w, h); \
+ if (offset + w + w < frame_size) { \
+ row_offset += w; \
+ } else { \
+ row_offset = 0; \
+ col_offset += h; \
+ } \
+ if (col_offset + h >= frame_size) { \
+ col_offset = 0; \
+ } \
+ } \
+ }
+
+// This experiment shows that when frame size is 64x64
+// vpx_highbd_convolve8_sse2 and vpx_convolve8_sse2's speed are similar.
+// However when frame size becomes 1024x1024
+// vpx_highbd_convolve8_sse2 is around 50% slower than vpx_convolve8_sse2
+// we think the bottleneck is from memory IO
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 8, 64);
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 16, 64);
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 32, 64);
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 64, 64);
+
+lowbd_convolve_speed(vpx_convolve8_sse2, 8, 64);
+lowbd_convolve_speed(vpx_convolve8_sse2, 16, 64);
+lowbd_convolve_speed(vpx_convolve8_sse2, 32, 64);
+lowbd_convolve_speed(vpx_convolve8_sse2, 64, 64);
+
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 8, 1024);
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 16, 1024);
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 32, 1024);
+highbd_convolve_speed(vpx_highbd_convolve8_sse2, 64, 1024);
+
+lowbd_convolve_speed(vpx_convolve8_sse2, 8, 1024);
+lowbd_convolve_speed(vpx_convolve8_sse2, 16, 1024);
+lowbd_convolve_speed(vpx_convolve8_sse2, 32, 1024);
+lowbd_convolve_speed(vpx_convolve8_sse2, 64, 1024);
+#endif // CONVOLVE_SPEED_TEST
} // namespace
diff --git a/vp10/common/entropymode.c b/vp10/common/entropymode.c
index b939424..d48679e 100644
--- a/vp10/common/entropymode.c
+++ b/vp10/common/entropymode.c
@@ -383,7 +383,7 @@
};
-const vpx_prob default_uv_palette_mode_prob[2] = {
+const vpx_prob vp10_default_palette_uv_mode_prob[2] = {
253, 229
};
diff --git a/vp10/common/entropymode.h b/vp10/common/entropymode.h
index ba36ddb..2443d60 100644
--- a/vp10/common/entropymode.h
+++ b/vp10/common/entropymode.h
@@ -176,6 +176,7 @@
[INTRA_MODES - 1];
extern const vpx_prob
vp10_default_palette_y_mode_prob[PALETTE_BLOCK_SIZES][PALETTE_Y_MODE_CONTEXTS];
+extern const vpx_prob vp10_default_palette_uv_mode_prob[2];
extern const vpx_prob
vp10_default_palette_y_size_prob[PALETTE_BLOCK_SIZES][PALETTE_SIZES - 1];
extern const vpx_prob
diff --git a/vp10/common/restoration.h b/vp10/common/restoration.h
index 43b140e..980fe72 100644
--- a/vp10/common/restoration.h
+++ b/vp10/common/restoration.h
@@ -42,6 +42,9 @@
#define WIENER_FILT_TAP1_BITS 4
#define WIENER_FILT_TAP2_BITS 5
+#define WIENER_FILT_BITS \
+ ((WIENER_FILT_TAP0_BITS + WIENER_FILT_TAP1_BITS + WIENER_FILT_TAP2_BITS) * 2)
+
#define WIENER_FILT_TAP0_MAXV \
(WIENER_FILT_TAP0_MINV -1 + (1 << WIENER_FILT_TAP0_BITS))
#define WIENER_FILT_TAP1_MAXV \
diff --git a/vp10/common/scan.c b/vp10/common/scan.c
index e26b40d..6dc5604 100644
--- a/vp10/common/scan.c
+++ b/vp10/common/scan.c
@@ -19,6 +19,7 @@
7, 14, 11, 15,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, mcol_scan_4x4[16]) = {
0, 4, 8, 12,
1, 5, 9, 13,
@@ -32,6 +33,7 @@
8, 9, 10, 11,
12, 13, 14, 15,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, col_scan_4x4[16]) = {
0, 4, 8, 1,
@@ -58,6 +60,7 @@
46, 39, 61, 54, 47, 62, 55, 63,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, mcol_scan_8x8[64]) = {
0, 8, 16, 24, 32, 40, 48, 56,
1, 9, 17, 25, 33, 41, 49, 57,
@@ -79,6 +82,7 @@
48, 49, 50, 51, 52, 53, 54, 55,
56, 57, 58, 59, 60, 61, 62, 63,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, col_scan_8x8[64]) = {
0, 8, 16, 1, 24, 9, 32, 17,
@@ -123,6 +127,7 @@
255,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, mcol_scan_16x16[256]) = {
0, 16, 32, 48, 64, 80, 96, 112, 128, 144, 160, 176, 192, 208, 224, 240,
1, 17, 33, 49, 65, 81, 97, 113, 129, 145, 161, 177, 193, 209, 225, 241,
@@ -169,6 +174,7 @@
240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252,
253, 254, 255,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, col_scan_16x16[256]) = {
0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81,
@@ -213,6 +219,7 @@
255,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, mcol_scan_32x32[1024]) = {
0, 32, 64, 96, 128, 160, 192, 224, 256, 288, 320, 352, 384, 416, 448, 480,
512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832, 864, 896, 928, 960,
@@ -407,6 +414,7 @@
1004, 1005, 1006, 1007, 1008, 1009, 1010, 1011, 1012, 1013, 1014,
1015, 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, default_scan_32x32[1024]) = {
0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160,
@@ -789,6 +797,7 @@
3, 6, 10, 13, 7, 10, 11, 14, 0, 0,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
mcol_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 4, 4, 8, 8, 0, 0, 1, 4, 5, 8, 9, 12, 1, 1, 2, 5, 6, 9, 10, 13,
@@ -800,6 +809,7 @@
0, 0, 0, 0, 1, 1, 2, 2, 0, 0, 1, 4, 2, 5, 3, 6, 4, 4, 5, 8, 6, 9, 7, 10, 8,
8, 9, 12, 10, 13, 11, 14, 0, 0,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = {
@@ -824,6 +834,7 @@
47, 47, 55, 55, 0, 0,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
mcol_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 8, 8, 16, 16, 24, 24, 32, 32, 40, 40, 48, 48, 0, 0, 1, 8, 9,
@@ -845,6 +856,7 @@
41, 48, 42, 49, 43, 50, 44, 51, 45, 52, 46, 53, 47, 54, 48, 48, 49, 56,
50, 57, 51, 58, 52, 59, 53, 60, 54, 61, 55, 62, 0, 0,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = {
@@ -868,6 +880,7 @@
46, 54, 61, 47, 54, 55, 62, 0, 0,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
mcol_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 16, 16, 32, 32, 48, 48, 64, 64, 80, 80, 96, 96,
@@ -975,6 +988,7 @@
253, 239, 254,
0, 0,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = {
@@ -1087,6 +1101,7 @@
238, 239, 254, 0, 0,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
mcol_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 32, 32, 64, 64, 96, 96, 128, 128, 160, 160, 192, 192,
@@ -1474,6 +1489,7 @@
987, 1018, 988, 1019, 989, 1020, 990, 1021, 991, 1022,
0, 0,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]) = {
@@ -2143,6 +2159,7 @@
0, 2, 5, 8, 1, 3, 9, 12, 4, 7, 11, 14, 6, 10, 13, 15,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_mcol_iscan_4x4[16]) = {
0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15,
};
@@ -2150,6 +2167,7 @@
DECLARE_ALIGNED(16, static const int16_t, vp10_mrow_iscan_4x4[16]) = {
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_col_iscan_4x4[16]) = {
0, 3, 7, 11, 1, 5, 9, 12, 2, 6, 10, 14, 4, 8, 13, 15,
@@ -2159,6 +2177,7 @@
0, 1, 3, 5, 2, 4, 6, 9, 7, 8, 11, 13, 10, 12, 14, 15,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_mcol_iscan_8x8[64]) = {
0, 8, 16, 24, 32, 40, 48, 56, 1, 9, 17, 25, 33, 41, 49, 57, 2, 10,
18, 26, 34, 42, 50, 58, 3, 11, 19, 27, 35, 43, 51, 59, 4, 12, 20,
@@ -2172,6 +2191,7 @@
37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53,
54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_col_iscan_8x8[64]) = {
0, 3, 8, 15, 22, 32, 40, 47, 1, 5, 11, 18, 26, 34, 44, 51,
@@ -2194,6 +2214,7 @@
25, 32, 39, 45, 50, 55, 59, 62, 33, 40, 46, 51, 54, 58, 61, 63,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_mcol_iscan_16x16[256]) = {
0, 16, 32, 48, 64, 80, 96, 112, 128, 144, 160, 176, 192, 208, 224, 240,
1, 17, 33, 49, 65, 81, 97, 113, 129, 145, 161, 177, 193, 209, 225, 241,
@@ -2240,6 +2261,7 @@
240, 241, 242, 243, 244, 245, 246, 247,
248, 249, 250, 251, 252, 253, 254, 255,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_col_iscan_16x16[256]) = {
0, 4, 11, 20, 31, 43, 59, 75, 85, 109, 130, 150, 165, 181, 195, 198,
@@ -2300,6 +2322,7 @@
249, 253, 255,
};
+#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_mcol_iscan_32x32[1024]) = {
0, 32, 64, 96, 128, 160, 192, 224, 256, 288, 320, 352, 384, 416,
448, 480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832,
@@ -2497,6 +2520,7 @@
1011, 1012, 1013, 1014, 1015, 1016, 1017, 1018, 1019,
1020, 1021, 1022, 1023,
};
+#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t, vp10_default_iscan_32x32[1024]) = {
0, 2, 5, 10, 17, 25, 38, 47, 62, 83, 101, 121, 145, 170, 193, 204,
diff --git a/vp10/decoder/decodeframe.c b/vp10/decoder/decodeframe.c
index 320e66e..ce6317c 100644
--- a/vp10/decoder/decodeframe.c
+++ b/vp10/decoder/decodeframe.c
@@ -1978,6 +1978,10 @@
}
if (!is_inter_block(mbmi)) {
int plane;
+ for (plane = 0; plane <= 1; ++plane) {
+ if (mbmi->palette_mode_info.palette_size[plane])
+ vp10_decode_palette_tokens(xd, plane, r);
+ }
for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const TX_SIZE tx_size =
@@ -1994,9 +1998,6 @@
(xd->mb_to_bottom_edge >= 0 ?
0 : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
- if (plane <= 1 && mbmi->palette_mode_info.palette_size[plane])
- vp10_decode_palette_tokens(xd, plane, r);
-
for (row = 0; row < max_blocks_high; row += step)
for (col = 0; col < max_blocks_wide; col += step)
predict_and_reconstruct_intra_block(xd,
diff --git a/vp10/decoder/decodemv.c b/vp10/decoder/decodemv.c
index 4cd6d1d..eb336be 100644
--- a/vp10/decoder/decodemv.c
+++ b/vp10/decoder/decodemv.c
@@ -419,30 +419,48 @@
vpx_reader *r) {
MODE_INFO *const mi = xd->mi[0];
MB_MODE_INFO *const mbmi = &mi->mbmi;
- const MODE_INFO *above_mi = xd->above_mi;
- const MODE_INFO *left_mi = xd->left_mi;
+ const MODE_INFO *const above_mi = xd->above_mi;
+ const MODE_INFO *const left_mi = xd->left_mi;
const BLOCK_SIZE bsize = mbmi->sb_type;
- int i, palette_ctx = 0;
+ int i, n, palette_ctx = 0;
+ PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
- if (above_mi)
- palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (left_mi)
- palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (vpx_read(r, vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
- [palette_ctx])) {
- int n;
- PALETTE_MODE_INFO *pmi = &mbmi->palette_mode_info;
-
- pmi->palette_size[0] =
+ if (mbmi->mode == DC_PRED) {
+ if (above_mi)
+ palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
+ if (left_mi)
+ palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
+ if (vpx_read(r, vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
+ [palette_ctx])) {
+ pmi->palette_size[0] =
vpx_read_tree(r, vp10_palette_size_tree,
vp10_default_palette_y_size_prob[bsize - BLOCK_8X8]) + 2;
- n = pmi->palette_size[0];
+ n = pmi->palette_size[0];
+ for (i = 0; i < n; ++i)
+ pmi->palette_colors[i] = vpx_read_literal(r, cm->bit_depth);
- for (i = 0; i < n; ++i)
- pmi->palette_colors[i] = vpx_read_literal(r, cm->bit_depth);
+ xd->plane[0].color_index_map[0] = read_uniform(r, n);
+ assert(xd->plane[0].color_index_map[0] < n);
+ }
+ }
- xd->plane[0].color_index_map[0] = read_uniform(r, n);
- assert(xd->plane[0].color_index_map[0] < n);
+ if (mbmi->uv_mode == DC_PRED) {
+ if (vpx_read(r,
+ vp10_default_palette_uv_mode_prob[pmi->palette_size[0] > 0])) {
+ pmi->palette_size[1] =
+ vpx_read_tree(r, vp10_palette_size_tree,
+ vp10_default_palette_uv_size_prob[bsize - BLOCK_8X8])
+ + 2;
+ n = pmi->palette_size[1];
+ for (i = 0; i < n; ++i) {
+ pmi->palette_colors[PALETTE_MAX_SIZE + i] =
+ vpx_read_literal(r, cm->bit_depth);
+ pmi->palette_colors[2 * PALETTE_MAX_SIZE + i] =
+ vpx_read_literal(r, cm->bit_depth);
+ }
+ xd->plane[1].color_index_map[0] = read_uniform(r, n);
+ assert(xd->plane[1].color_index_map[0] < n);
+ }
}
}
@@ -456,7 +474,8 @@
#if !ALLOW_FILTER_INTRA_MODES
return;
#endif
- if (mbmi->mode == DC_PRED) {
+ if (mbmi->mode == DC_PRED &&
+ mbmi->palette_mode_info.palette_size[0] == 0) {
mbmi->ext_intra_mode_info.use_ext_intra_mode[0] =
vpx_read(r, cm->fc->ext_intra_probs[0]);
if (mbmi->ext_intra_mode_info.use_ext_intra_mode[0]) {
@@ -466,7 +485,8 @@
if (counts)
++counts->ext_intra[0][mbmi->ext_intra_mode_info.use_ext_intra_mode[0]];
}
- if (mbmi->uv_mode == DC_PRED) {
+ if (mbmi->uv_mode == DC_PRED &&
+ mbmi->palette_mode_info.palette_size[1] == 0) {
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] =
vpx_read(r, cm->fc->ext_intra_probs[1]);
if (mbmi->ext_intra_mode_info.use_ext_intra_mode[1]) {
@@ -555,8 +575,7 @@
mbmi->palette_mode_info.palette_size[0] = 0;
mbmi->palette_mode_info.palette_size[1] = 0;
- if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools &&
- mbmi->mode == DC_PRED)
+ if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools)
read_palette_mode_info(cm, xd, r);
if (!FIXED_TX_TYPE) {
@@ -868,9 +887,10 @@
mbmi->angle_delta[1] =
read_uniform(r, 2 * MAX_ANGLE_DELTAS + 1) - MAX_ANGLE_DELTAS;
#endif // CONFIG_EXT_INTRA
-
mbmi->palette_mode_info.palette_size[0] = 0;
mbmi->palette_mode_info.palette_size[1] = 0;
+ if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools)
+ read_palette_mode_info(cm, xd, r);
#if CONFIG_EXT_INTRA
mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 0;
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0;
@@ -1151,6 +1171,9 @@
int16_t mode_ctx = 0;
MV_REFERENCE_FRAME ref_frame;
+ mbmi->palette_mode_info.palette_size[0] = 0;
+ mbmi->palette_mode_info.palette_size[1] = 0;
+
read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame);
is_compound = has_second_ref(mbmi);
diff --git a/vp10/decoder/detokenize.c b/vp10/decoder/detokenize.c
index c5dec87..c1153d2 100644
--- a/vp10/decoder/detokenize.c
+++ b/vp10/decoder/detokenize.c
@@ -429,13 +429,15 @@
MODE_INFO *const mi = xd->mi[0];
MB_MODE_INFO *const mbmi = &mi->mbmi;
const BLOCK_SIZE bsize = mbmi->sb_type;
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
+ const int rows = (4 * num_4x4_blocks_high_lookup[bsize]) >>
+ (xd->plane[plane != 0].subsampling_y);
+ const int cols = (4 * num_4x4_blocks_wide_lookup[bsize]) >>
+ (xd->plane[plane != 0].subsampling_x);
int color_idx, color_ctx, color_order[PALETTE_MAX_SIZE];
int n = mbmi->palette_mode_info.palette_size[plane != 0];
int i, j;
- uint8_t *color_map = xd->plane[plane].color_index_map;
- const vpx_prob (* prob)[PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] =
+ uint8_t *color_map = xd->plane[plane != 0].color_index_map;
+ const vpx_prob (* const prob)[PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] =
plane ? vp10_default_palette_uv_color_prob :
vp10_default_palette_y_color_prob;
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index 24a9366..1ef2ea5 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -509,13 +509,11 @@
#endif // CONFIG_EXT_TX
static void pack_palette_tokens(vpx_writer *w, TOKENEXTRA **tp,
- BLOCK_SIZE bsize, int n) {
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
+ int n, int num) {
int i;
TOKENEXTRA *p = *tp;
- for (i = 0; i < rows * cols -1; ++i) {
+ for (i = 0; i < num; ++i) {
vp10_write_token(w, vp10_palette_color_tree[n - 2], p->context_tree,
&palette_color_encodings[n - 2][p->token]);
++p;
@@ -842,7 +840,8 @@
#if !ALLOW_FILTER_INTRA_MODES
return;
#endif
- if (mbmi->mode == DC_PRED) {
+ if (mbmi->mode == DC_PRED &&
+ mbmi->palette_mode_info.palette_size[0] == 0) {
vpx_write(w, mbmi->ext_intra_mode_info.use_ext_intra_mode[0],
cm->fc->ext_intra_probs[0]);
if (mbmi->ext_intra_mode_info.use_ext_intra_mode[0]) {
@@ -850,7 +849,8 @@
write_uniform(w, FILTER_INTRA_MODES, mode);
}
}
- if (mbmi->uv_mode == DC_PRED) {
+ if (mbmi->uv_mode == DC_PRED &&
+ mbmi->palette_mode_info.palette_size[1] == 0) {
vpx_write(w, mbmi->ext_intra_mode_info.use_ext_intra_mode[1],
cm->fc->ext_intra_probs[1]);
if (mbmi->ext_intra_mode_info.use_ext_intra_mode[1]) {
@@ -881,6 +881,55 @@
}
}
+static void write_palette_mode_info(const VP10_COMMON *cm,
+ const MACROBLOCKD *xd,
+ const MODE_INFO *const mi,
+ vpx_writer *w) {
+ const MB_MODE_INFO *const mbmi = &mi->mbmi;
+ const MODE_INFO *const above_mi = xd->above_mi;
+ const MODE_INFO *const left_mi = xd->left_mi;
+ const BLOCK_SIZE bsize = mbmi->sb_type;
+ const PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
+ int palette_ctx = 0;
+ int n, i;
+
+ if (mbmi->mode == DC_PRED) {
+ n = pmi->palette_size[0];
+ if (above_mi)
+ palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
+ if (left_mi)
+ palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
+ vpx_write(w, n > 0,
+ vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8][palette_ctx]);
+ if (n > 0) {
+ vp10_write_token(w, vp10_palette_size_tree,
+ vp10_default_palette_y_size_prob[bsize - BLOCK_8X8],
+ &palette_size_encodings[n - 2]);
+ for (i = 0; i < n; ++i)
+ vpx_write_literal(w, pmi->palette_colors[i], cm->bit_depth);
+ write_uniform(w, n, pmi->palette_first_color_idx[0]);
+ }
+ }
+
+ if (mbmi->uv_mode == DC_PRED) {
+ n = pmi->palette_size[1];
+ vpx_write(w, n > 0,
+ vp10_default_palette_uv_mode_prob[pmi->palette_size[0] > 0]);
+ if (n > 0) {
+ vp10_write_token(w, vp10_palette_size_tree,
+ vp10_default_palette_uv_size_prob[bsize - BLOCK_8X8],
+ &palette_size_encodings[n - 2]);
+ for (i = 0; i < n; ++i) {
+ vpx_write_literal(w, pmi->palette_colors[PALETTE_MAX_SIZE + i],
+ cm->bit_depth);
+ vpx_write_literal(w, pmi->palette_colors[2 * PALETTE_MAX_SIZE + i],
+ cm->bit_depth);
+ }
+ write_uniform(w, n, pmi->palette_first_color_idx[1]);
+ }
+ }
+}
+
static void pack_inter_mode_mvs(VP10_COMP *cpi, const MODE_INFO *mi,
#if CONFIG_SUPERTX
int supertx_enabled,
@@ -995,7 +1044,10 @@
bsize >= BLOCK_8X8)
write_uniform(w, 2 * MAX_ANGLE_DELTAS + 1,
MAX_ANGLE_DELTAS + mbmi->angle_delta[1]);
-
+#endif // CONFIG_EXT_INTRA
+ if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools)
+ write_palette_mode_info(cm, xd, mi, w);
+#if CONFIG_EXT_INTRA
if (bsize >= BLOCK_8X8)
write_ext_intra_mode_info(cm, mbmi, w);
#endif // CONFIG_EXT_INTRA
@@ -1263,36 +1315,6 @@
}
}
-static void write_palette_mode_info(const VP10_COMMON *cm,
- const MACROBLOCKD *xd,
- const MODE_INFO *const mi,
- vpx_writer *w) {
- const MB_MODE_INFO *const mbmi = &mi->mbmi;
- const MODE_INFO *const above_mi = xd->above_mi;
- const MODE_INFO *const left_mi = xd->left_mi;
- const BLOCK_SIZE bsize = mbmi->sb_type;
- const PALETTE_MODE_INFO *pmi = &mbmi->palette_mode_info;
- int palette_ctx = 0;
- int n, i;
-
- n = pmi->palette_size[0];
- if (above_mi)
- palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (left_mi)
- palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- vpx_write(w, n > 0,
- vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8][palette_ctx]);
- if (n > 0) {
- vp10_write_token(w, vp10_palette_size_tree,
- vp10_default_palette_y_size_prob[bsize - BLOCK_8X8],
- &palette_size_encodings[n - 2]);
- for (i = 0; i < n; ++i)
- vpx_write_literal(w, pmi->palette_colors[i],
- cm->bit_depth);
- write_uniform(w, n, pmi->palette_first_color_idx[0]);
- }
-}
-
static void write_mb_modes_kf(const VP10_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO **mi_8x8, vpx_writer *w) {
const struct segmentation *const seg = &cm->seg;
@@ -1351,9 +1373,7 @@
write_uniform(w, 2 * MAX_ANGLE_DELTAS + 1,
MAX_ANGLE_DELTAS + mbmi->angle_delta[1]);
#endif // CONFIG_EXT_INTRA
-
- if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools &&
- mbmi->mode == DC_PRED)
+ if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools)
write_palette_mode_info(cm, xd, mi, w);
if (!FIXED_TX_TYPE) {
@@ -1428,11 +1448,17 @@
w);
}
- if (m->mbmi.palette_mode_info.palette_size[0] > 0) {
- assert(*tok < tok_end);
- pack_palette_tokens(w, tok, m->mbmi.sb_type,
- m->mbmi.palette_mode_info.palette_size[0]);
- assert(*tok < tok_end);
+ for (plane = 0; plane <= 1; ++plane) {
+ if (m->mbmi.palette_mode_info.palette_size[plane] > 0) {
+ const int rows = (4 * num_4x4_blocks_high_lookup[m->mbmi.sb_type]) >>
+ (xd->plane[plane].subsampling_y);
+ const int cols = (4 * num_4x4_blocks_wide_lookup[m->mbmi.sb_type]) >>
+ (xd->plane[plane].subsampling_x);
+ assert(*tok < tok_end);
+ pack_palette_tokens(w, tok, m->mbmi.palette_mode_info.palette_size[plane],
+ rows * cols - 1);
+ assert(*tok < tok_end);
+ }
}
#if CONFIG_SUPERTX
diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index 7f55295..c5a68a9 100644
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -4432,9 +4432,11 @@
#if CONFIG_EXT_INTRA
if (output_enabled && bsize >= BLOCK_8X8) {
FRAME_COUNTS *counts = td->counts;
- if (mbmi->mode == DC_PRED)
+ if (mbmi->mode == DC_PRED &&
+ mbmi->palette_mode_info.palette_size[0] == 0)
++counts->ext_intra[0][mbmi->ext_intra_mode_info.use_ext_intra_mode[0]];
- if (mbmi->uv_mode == DC_PRED)
+ if (mbmi->uv_mode == DC_PRED &&
+ mbmi->palette_mode_info.palette_size[1] == 0)
++counts->ext_intra[1][mbmi->ext_intra_mode_info.use_ext_intra_mode[1]];
if (mbmi->mode != DC_PRED && mbmi->mode != TM_PRED) {
int p_angle;
@@ -4448,12 +4450,14 @@
#endif // CONFIG_EXT_INTRA
if (bsize >= BLOCK_8X8 && output_enabled) {
- if (mbmi->palette_mode_info.palette_size[0] > 0) {
- mbmi->palette_mode_info.palette_first_color_idx[0] =
- xd->plane[0].color_index_map[0];
- // TODO(huisu): this increases the use of token buffer. Needs stretch
- // test to verify.
- vp10_tokenize_palette_sb(td, bsize, 0, t);
+ for (plane = 0; plane <= 1; ++plane) {
+ if (mbmi->palette_mode_info.palette_size[plane] > 0) {
+ mbmi->palette_mode_info.palette_first_color_idx[plane] =
+ xd->plane[plane].color_index_map[0];
+ // TODO(huisu): this increases the use of token buffer. Needs stretch
+ // test to verify.
+ vp10_tokenize_palette_sb(td, bsize, plane, t);
+ }
}
}
vp10_tokenize_sb(cpi, td, t, !output_enabled, VPXMAX(bsize, BLOCK_8X8));
diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index 34dd8d5..fc65e72 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -358,9 +358,7 @@
static void dealloc_compressor_data(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
-#if CONFIG_REF_MV
int i;
-#endif
vpx_free(cpi->mbmi_ext_base);
cpi->mbmi_ext_base = NULL;
@@ -413,14 +411,9 @@
vpx_free(cpi->active_map.map);
cpi->active_map.map = NULL;
-#if CONFIG_AFFINE_MOTION
- {
- // Free up-sampled reference buffers.
- int i;
- for (i = 0; i < MAX_REF_FRAMES; i++)
- vpx_free_frame_buffer(&cpi->upsampled_ref_bufs[i].buf);
- }
-#endif
+ // Free up-sampled reference buffers.
+ for (i = 0; i < MAX_REF_FRAMES; i++)
+ vpx_free_frame_buffer(&cpi->upsampled_ref_bufs[i].buf);
vp10_free_ref_frame_buffers(cm->buffer_pool);
#if CONFIG_VP9_POSTPROC
@@ -756,26 +749,6 @@
NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate scaled last source buffer");
-
-#if CONFIG_AFFINE_MOTION
- {
- // Allocate up-sampled reference buffers.
- int i;
-
- for (i = 0; i < MAX_REF_FRAMES; i++)
- if (vpx_realloc_frame_buffer(&cpi->upsampled_ref_bufs[i].buf,
- (cm->width << 3), (cm->height << 3),
- cm->subsampling_x, cm->subsampling_y,
-#if CONFIG_VP9_HIGHBITDEPTH
- cm->use_highbitdepth,
-#endif
- (VP9_ENC_BORDER_IN_PIXELS << 3),
- cm->byte_alignment,
- NULL, NULL, NULL))
- vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
- "Failed to allocate up-sampled reference frame buffer");
- }
-#endif
}
@@ -2069,6 +2042,14 @@
} while (++i <= MV_MAX);
}
+static INLINE void init_upsampled_ref_frame_bufs(VP10_COMP *cpi) {
+ int i;
+
+ for (i = 0; i < MAX_REF_FRAMES; ++i) {
+ cpi->upsampled_ref_bufs[i].ref_count = 0;
+ cpi->upsampled_ref_idx[i] = INVALID_IDX;
+ }
+}
VP10_COMP *vp10_create_compressor(VP10EncoderConfig *oxcf,
BufferPool *const pool) {
@@ -2267,6 +2248,8 @@
vp10_init_second_pass(cpi);
}
+ init_upsampled_ref_frame_bufs(cpi);
+
vp10_set_speed_features_framesize_independent(cpi);
vp10_set_speed_features_framesize_dependent(cpi);
@@ -2929,7 +2912,6 @@
return force_recode;
}
-#if CONFIG_AFFINE_MOTION
static INLINE int get_free_upsampled_ref_buf(EncRefCntBuffer *ubufs) {
int i;
@@ -2941,50 +2923,59 @@
return INVALID_IDX;
}
-// Up-sample reference frames.
-static INLINE int upsample_ref_frame(RefCntBuffer *bufs,
-#if CONFIG_VP9_HIGHBITDEPTH
- EncRefCntBuffer *ubufs, int new_idx,
- int bit_depth) {
-#else
- EncRefCntBuffer *ubufs, int new_idx) {
-#endif
+// Up-sample 1 reference frame.
+static INLINE int upsample_ref_frame(VP10_COMP *cpi,
+ const YV12_BUFFER_CONFIG *const ref) {
+ VP10_COMMON * const cm = &cpi->common;
+ EncRefCntBuffer *ubufs = cpi->upsampled_ref_bufs;
int new_uidx = get_free_upsampled_ref_buf(ubufs);
if (new_uidx == INVALID_IDX) {
return INVALID_IDX;
} else {
- const YV12_BUFFER_CONFIG *const ref = &bufs[new_idx].buf;
YV12_BUFFER_CONFIG *upsampled_ref = &ubufs[new_uidx].buf;
+ // Can allocate buffer for Y plane only.
+ if (upsampled_ref->buffer_alloc_sz < (ref->buffer_alloc_sz << 6))
+ if (vpx_realloc_frame_buffer(upsampled_ref,
+ (cm->width << 3), (cm->height << 3),
+ cm->subsampling_x, cm->subsampling_y,
+#if CONFIG_VP9_HIGHBITDEPTH
+ cm->use_highbitdepth,
+#endif
+ (VP9_ENC_BORDER_IN_PIXELS << 3),
+ cm->byte_alignment,
+ NULL, NULL, NULL))
+ vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
+ "Failed to allocate up-sampled frame buffer");
+
// Currently, only Y plane is up-sampled, U, V are not used.
#if CONFIG_VP9_HIGHBITDEPTH
- scale_and_extend_frame(ref, upsampled_ref, 1, bit_depth);
+ scale_and_extend_frame(ref, upsampled_ref, 1, (int)cm->bit_depth);
#else
scale_and_extend_frame(ref, upsampled_ref, 1);
#endif
return new_uidx;
}
}
-#endif
void vp10_update_reference_frames(VP10_COMP *cpi) {
VP10_COMMON * const cm = &cpi->common;
BufferPool *const pool = cm->buffer_pool;
+ const int use_upsampled_ref = cpi->sf.use_upsampled_references;
+ int new_uidx = 0;
+
#if CONFIG_EXT_REFS
int ref_frame;
#endif // CONFIG_EXT_REFS
-#if CONFIG_AFFINE_MOTION
- // Always up-sample the current encoded frame.
-#if CONFIG_VP9_HIGHBITDEPTH
- int new_uidx = upsample_ref_frame(pool->frame_bufs, cpi->upsampled_ref_bufs,
- cm->new_fb_idx, (int)cm->bit_depth);
-#else
- int new_uidx = upsample_ref_frame(pool->frame_bufs, cpi->upsampled_ref_bufs,
- cm->new_fb_idx);
-#endif
-#endif
+ if (use_upsampled_ref) {
+ // Up-sample the current encoded frame.
+ RefCntBuffer *bufs = pool->frame_bufs;
+ const YV12_BUFFER_CONFIG *const ref = &bufs[cm->new_fb_idx].buf;
+
+ new_uidx = upsample_ref_frame(cpi, ref);
+ }
// At this point the new frame has been encoded.
// If any buffer copy / swapping is signaled it should be done here.
@@ -2994,12 +2985,12 @@
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
-#if CONFIG_AFFINE_MOTION
- uref_cnt_fb(cpi->upsampled_ref_bufs,
- &cpi->upsampled_ref_idx[cpi->gld_fb_idx], new_uidx);
- uref_cnt_fb(cpi->upsampled_ref_bufs,
- &cpi->upsampled_ref_idx[cpi->alt_fb_idx], new_uidx);
-#endif
+ if (use_upsampled_ref) {
+ uref_cnt_fb(cpi->upsampled_ref_bufs,
+ &cpi->upsampled_ref_idx[cpi->gld_fb_idx], new_uidx);
+ uref_cnt_fb(cpi->upsampled_ref_bufs,
+ &cpi->upsampled_ref_idx[cpi->alt_fb_idx], new_uidx);
+ }
} else if (vp10_preserve_existing_gf(cpi)) {
// We have decided to preserve the previously existing golden frame as our
// new ARF frame. However, in the short term in function
@@ -3013,10 +3004,10 @@
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
-#if CONFIG_AFFINE_MOTION
- uref_cnt_fb(cpi->upsampled_ref_bufs,
- &cpi->upsampled_ref_idx[cpi->alt_fb_idx], new_uidx);
-#endif
+ if (use_upsampled_ref)
+ uref_cnt_fb(cpi->upsampled_ref_bufs,
+ &cpi->upsampled_ref_idx[cpi->alt_fb_idx], new_uidx);
+
tmp = cpi->alt_fb_idx;
cpi->alt_fb_idx = cpi->gld_fb_idx;
cpi->gld_fb_idx = tmp;
@@ -3030,10 +3021,10 @@
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[arf_idx], cm->new_fb_idx);
-#if CONFIG_AFFINE_MOTION
- uref_cnt_fb(cpi->upsampled_ref_bufs,
- &cpi->upsampled_ref_idx[cpi->alt_fb_idx], new_uidx);
-#endif
+ if (use_upsampled_ref)
+ uref_cnt_fb(cpi->upsampled_ref_bufs,
+ &cpi->upsampled_ref_idx[cpi->alt_fb_idx], new_uidx);
+
memcpy(cpi->interp_filter_selected[ALTREF_FRAME],
cpi->interp_filter_selected[0],
sizeof(cpi->interp_filter_selected[0]));
@@ -3042,10 +3033,10 @@
if (cpi->refresh_golden_frame) {
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
-#if CONFIG_AFFINE_MOTION
- uref_cnt_fb(cpi->upsampled_ref_bufs,
- &cpi->upsampled_ref_idx[cpi->gld_fb_idx], new_uidx);
-#endif
+ if (use_upsampled_ref)
+ uref_cnt_fb(cpi->upsampled_ref_bufs,
+ &cpi->upsampled_ref_idx[cpi->gld_fb_idx], new_uidx);
+
if (!cpi->rc.is_src_frame_alt_ref)
memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
cpi->interp_filter_selected[0],
@@ -3080,10 +3071,10 @@
if (cpi->refresh_last_frame) {
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
-#if CONFIG_AFFINE_MOTION
- uref_cnt_fb(cpi->upsampled_ref_bufs,
- &cpi->upsampled_ref_idx[cpi->lst_fb_idx], new_uidx);
-#endif
+ if (use_upsampled_ref)
+ uref_cnt_fb(cpi->upsampled_ref_bufs,
+ &cpi->upsampled_ref_idx[cpi->lst_fb_idx], new_uidx);
+
if (!cpi->rc.is_src_frame_alt_ref) {
memcpy(cpi->interp_filter_selected[LAST_FRAME],
cpi->interp_filter_selected[0],
@@ -3249,8 +3240,9 @@
}
#endif // CONFIG_VP9_HIGHBITDEPTH
-#if CONFIG_AFFINE_MOTION
- {
+ if (cpi->sf.use_upsampled_references && (force_scaling ||
+ new_fb_ptr->buf.y_crop_width != cm->width ||
+ new_fb_ptr->buf.y_crop_height != cm->height)) {
const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
EncRefCntBuffer *ubuf =
&cpi->upsampled_ref_bufs[cpi->upsampled_ref_idx[map_idx]];
@@ -3267,15 +3259,12 @@
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate up-sampled frame buffer");
#if CONFIG_VP9_HIGHBITDEPTH
- scale_and_extend_frame(&new_fb_ptr->buf, &ubuf->buf, MAX_MB_PLANE,
+ scale_and_extend_frame(&new_fb_ptr->buf, &ubuf->buf, 1,
(int)cm->bit_depth);
#else
- scale_and_extend_frame(&new_fb_ptr->buf, &ubuf->buf, MAX_MB_PLANE);
+ scale_and_extend_frame(&new_fb_ptr->buf, &ubuf->buf, 1);
#endif
- cpi->scaled_ref_idx[ref_frame - LAST_FRAME] = new_fb;
- alloc_frame_mvs(cm, new_fb);
}
-#endif
} else {
const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
RefCntBuffer *const buf = &pool->frame_bufs[buf_idx];
@@ -3610,9 +3599,28 @@
set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME);
}
+static void reset_use_upsampled_references(VP10_COMP *cpi) {
+ MV_REFERENCE_FRAME ref_frame;
+
+ // reset up-sampled reference buffer structure.
+ init_upsampled_ref_frame_bufs(cpi);
+
+ for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
+ const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi,
+ ref_frame);
+ int new_uidx = upsample_ref_frame(cpi, ref);
+
+ // Update the up-sampled reference index.
+ cpi->upsampled_ref_idx[get_ref_frame_map_idx(cpi, ref_frame)] =
+ new_uidx;
+ cpi->upsampled_ref_bufs[new_uidx].ref_count++;
+ }
+}
+
static void encode_without_recode_loop(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
int q = 0, bottom_index = 0, top_index = 0; // Dummy variables.
+ const int use_upsampled_ref = cpi->sf.use_upsampled_references;
vpx_clear_system_state();
@@ -3647,6 +3655,12 @@
set_size_independent_vars(cpi);
set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
+ // cpi->sf.use_upsampled_references can be different from frame to frame.
+ // Every time when cpi->sf.use_upsampled_references is changed from 0 to 1.
+ // The reference frames for this frame have to be up-sampled before encoding.
+ if (!use_upsampled_ref && cpi->sf.use_upsampled_references)
+ reset_use_upsampled_references(cpi);
+
vp10_set_quantizer(cm, q);
vp10_set_variance_partition_thresholds(cpi, q);
@@ -3694,9 +3708,16 @@
int frame_over_shoot_limit;
int frame_under_shoot_limit;
int q = 0, q_low = 0, q_high = 0;
+ const int use_upsampled_ref = cpi->sf.use_upsampled_references;
set_size_independent_vars(cpi);
+ // cpi->sf.use_upsampled_references can be different from frame to frame.
+ // Every time when cpi->sf.use_upsampled_references is changed from 0 to 1.
+ // The reference frames for this frame have to be up-sampled before encoding.
+ if (!use_upsampled_ref && cpi->sf.use_upsampled_references)
+ reset_use_upsampled_references(cpi);
+
do {
vpx_clear_system_state();
@@ -4355,17 +4376,6 @@
}
}
-#if CONFIG_AFFINE_MOTION
-static INLINE void init_upsampled_ref_frame_bufs(VP10_COMP *cpi) {
- int i;
-
- for (i = 0; i < MAX_REF_FRAMES; ++i) {
- cpi->upsampled_ref_bufs[i].ref_count = 0;
- cpi->upsampled_ref_idx[i] = INVALID_IDX;
- }
-}
-#endif
-
static void check_initial_width(VP10_COMP *cpi,
#if CONFIG_VP9_HIGHBITDEPTH
int use_highbitdepth,
@@ -4388,9 +4398,7 @@
alloc_raw_frame_buffers(cpi);
init_ref_frame_bufs(cm);
alloc_util_frame_buffers(cpi);
-#if CONFIG_AFFINE_MOTION
- init_upsampled_ref_frame_bufs(cpi);
-#endif
+
init_motion_estimation(cpi); // TODO(agrange) This can be removed.
cpi->initial_width = cm->width;
diff --git a/vp10/encoder/encoder.h b/vp10/encoder/encoder.h
index b2c242c..afe3292 100644
--- a/vp10/encoder/encoder.h
+++ b/vp10/encoder/encoder.h
@@ -286,12 +286,10 @@
double worst;
} ImageStat;
-#if CONFIG_AFFINE_MOTION
typedef struct {
int ref_count;
YV12_BUFFER_CONFIG buf;
} EncRefCntBuffer;
-#endif
typedef struct VP10_COMP {
QUANTS quants;
@@ -311,11 +309,9 @@
YV12_BUFFER_CONFIG *unscaled_last_source;
YV12_BUFFER_CONFIG scaled_last_source;
-#if CONFIG_AFFINE_MOTION
// Up-sampled reference buffers
EncRefCntBuffer upsampled_ref_bufs[MAX_REF_FRAMES];
int upsampled_ref_idx[MAX_REF_FRAMES];
-#endif
TileDataEnc *tile_data;
int allocated_tiles; // Keep track of memory allocated for tiles.
@@ -702,7 +698,6 @@
#define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
-#if CONFIG_AFFINE_MOTION
// Update up-sampled reference frame index.
static INLINE void uref_cnt_fb(EncRefCntBuffer *ubufs, int *uidx,
int new_uidx) {
@@ -714,7 +709,6 @@
*uidx = new_uidx;
ubufs[new_uidx].ref_count++;
}
-#endif
#ifdef __cplusplus
} // extern "C"
diff --git a/vp10/encoder/mbgraph.c b/vp10/encoder/mbgraph.c
index 32ff0fa..5e66ce5 100644
--- a/vp10/encoder/mbgraph.c
+++ b/vp10/encoder/mbgraph.c
@@ -64,11 +64,7 @@
&v_fn_ptr, 0, mv_sf->subpel_iters_per_step,
cond_cost_list(cpi, cost_list),
NULL, NULL,
-#if CONFIG_AFFINE_MOTION
&distortion, &sse, NULL, 0, 0, 0);
-#else
- &distortion, &sse, NULL, 0, 0);
-#endif
}
#if CONFIG_EXT_INTER
diff --git a/vp10/encoder/mcomp.c b/vp10/encoder/mcomp.c
index 2c93976..1f147d7 100644
--- a/vp10/encoder/mcomp.c
+++ b/vp10/encoder/mcomp.c
@@ -210,7 +210,6 @@
#define CHECK_BETTER0(v, r, c) CHECK_BETTER(v, r, c)
-#if CONFIG_AFFINE_MOTION
static INLINE const uint8_t *upre(const uint8_t *buf, int stride,
int r, int c) {
return &buf[(r) * stride + (c)];
@@ -232,7 +231,6 @@
} else { \
v = INT_MAX; \
}
-#endif
#define FIRST_LEVEL_CHECKS \
{ \
@@ -438,11 +436,7 @@
int *distortion,
unsigned int *sse1,
const uint8_t *second_pred,
-#if CONFIG_AFFINE_MOTION
int w, int h, int use_upsampled_ref) {
-#else
- int w, int h) {
-#endif
SETUP_SUBPEL_SEARCH;
besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
z, src_stride, y, y_stride, second_pred,
@@ -455,9 +449,7 @@
(void) allow_hp;
(void) forced_stop;
(void) hstep;
-#if CONFIG_AFFINE_MOTION
(void) use_upsampled_ref;
-#endif
if (cost_list &&
cost_list[0] != INT_MAX && cost_list[1] != INT_MAX &&
@@ -524,16 +516,10 @@
int *distortion,
unsigned int *sse1,
const uint8_t *second_pred,
-#if CONFIG_AFFINE_MOTION
int w, int h,
int use_upsampled_ref) {
-#else
- int w, int h) {
-#endif
SETUP_SUBPEL_SEARCH;
-#if CONFIG_AFFINE_MOTION
(void) use_upsampled_ref;
-#endif
besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
z, src_stride, y, y_stride, second_pred,
@@ -607,15 +593,9 @@
int *distortion,
unsigned int *sse1,
const uint8_t *second_pred,
-#if CONFIG_AFFINE_MOTION
int w, int h, int use_upsampled_ref) {
-#else
- int w, int h) {
-#endif
SETUP_SUBPEL_SEARCH;
-#if CONFIG_AFFINE_MOTION
(void) use_upsampled_ref;
-#endif
besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
z, src_stride, y, y_stride, second_pred,
@@ -705,9 +685,8 @@
{0, -1}, {0, 1}, {-1, 0}, {1, 0}
};
-
-#if CONFIG_AFFINE_MOTION
#if CONFIG_VP9_HIGHBITDEPTH
+// TODO(yunqing): Optimize the following 2 functions.
static void highbd_comp_avg_upsampled_pred(uint16_t *comp_pred,
const uint8_t *pred8,
int width, int height,
@@ -798,7 +777,6 @@
besterr += mv_err_cost(bestmv, ref_mv, mvjcost, mvcost, error_per_bit);
return besterr;
}
-#endif
int vp10_find_best_sub_pixel_tree(const MACROBLOCK *x,
MV *bestmv, const MV *ref_mv,
@@ -812,11 +790,7 @@
int *distortion,
unsigned int *sse1,
const uint8_t *second_pred,
-#if CONFIG_AFFINE_MOTION
int w, int h, int use_upsampled_ref) {
-#else
- int w, int h) {
-#endif
const uint8_t *const z = x->plane[0].src.buf;
const uint8_t *const src_address = z;
const int src_stride = x->plane[0].src.stride;
@@ -852,7 +826,6 @@
bestmv->row *= 8;
bestmv->col *= 8;
-#if CONFIG_AFFINE_MOTION
// use_upsampled_ref can be 0 or 1
if (use_upsampled_ref)
besterr = upsampled_setup_center_error(xd, bestmv, ref_mv, error_per_bit,
@@ -860,7 +833,6 @@
second_pred, w, h, (offset << 3),
mvjcost, mvcost, sse1, distortion);
else
-#endif
besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
z, src_stride, y, y_stride, second_pred,
w, h, offset, mvjcost, mvcost,
@@ -876,7 +848,6 @@
if (tc >= minc && tc <= maxc && tr >= minr && tr <= maxr) {
MV this_mv = {tr, tc};
-#if CONFIG_AFFINE_MOTION
if (use_upsampled_ref) {
const uint8_t *const pre_address = y + tr * y_stride + tc;
@@ -884,7 +855,6 @@
pre_address, y_stride, second_pred,
w, h, &sse);
} else {
-#endif
const uint8_t *const pre_address = y + (tr >> 3) * y_stride +
(tc >> 3);
if (second_pred == NULL)
@@ -893,9 +863,7 @@
else
thismse = vfp->svaf(pre_address, y_stride, sp(tc), sp(tr),
src_address, src_stride, &sse, second_pred);
-#if CONFIG_AFFINE_MOTION
}
-#endif
cost_array[idx] = thismse +
mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);
@@ -920,7 +888,6 @@
if (tc >= minc && tc <= maxc && tr >= minr && tr <= maxr) {
MV this_mv = {tr, tc};
-#if CONFIG_AFFINE_MOTION
if (use_upsampled_ref) {
const uint8_t *const pre_address = y + tr * y_stride + tc;
@@ -928,7 +895,6 @@
pre_address, y_stride, second_pred,
w, h, &sse);
} else {
-#endif
const uint8_t *const pre_address = y + (tr >> 3) * y_stride + (tc >> 3);
if (second_pred == NULL)
@@ -937,9 +903,7 @@
else
thismse = vfp->svaf(pre_address, y_stride, sp(tc), sp(tr),
src_address, src_stride, &sse, second_pred);
-#if CONFIG_AFFINE_MOTION
}
-#endif
cost_array[4] = thismse +
mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);
@@ -963,15 +927,11 @@
}
if (iters_per_step > 1 && best_idx != -1) {
-#if CONFIG_AFFINE_MOTION
if (use_upsampled_ref) {
SECOND_LEVEL_CHECKS_BEST(1);
} else {
-#endif
SECOND_LEVEL_CHECKS_BEST(0);
-#if CONFIG_AFFINE_MOTION
}
-#endif
}
tr = br;
diff --git a/vp10/encoder/mcomp.h b/vp10/encoder/mcomp.h
index a430c76..f99cd8b 100644
--- a/vp10/encoder/mcomp.h
+++ b/vp10/encoder/mcomp.h
@@ -116,11 +116,7 @@
int *mvjcost, int *mvcost[2],
int *distortion, unsigned int *sse1,
const uint8_t *second_pred,
-#if CONFIG_AFFINE_MOTION
int w, int h, int use_upsampled_ref);
-#else
- int w, int h);
-#endif
extern fractional_mv_step_fp vp10_find_best_sub_pixel_tree;
extern fractional_mv_step_fp vp10_find_best_sub_pixel_tree_pruned;
diff --git a/vp10/encoder/palette.c b/vp10/encoder/palette.c
index 522e185..d413935 100644
--- a/vp10/encoder/palette.c
+++ b/vp10/encoder/palette.c
@@ -93,7 +93,7 @@
uint8_t *pre_indices, int n, int k, int dim, int max_itr) {
int i = 0;
double pre_dist, this_dist;
- double pre_centroids[PALETTE_MAX_SIZE];
+ double pre_centroids[2 * PALETTE_MAX_SIZE];
vp10_calc_indices(data, centroids, indices, n, k, dim);
pre_dist = calc_total_dist(data, centroids, indices, n, k, dim);
diff --git a/vp10/encoder/pickrst.c b/vp10/encoder/pickrst.c
index 9982836..13f955d 100644
--- a/vp10/encoder/pickrst.c
+++ b/vp10/encoder/pickrst.c
@@ -9,6 +9,7 @@
*/
#include <assert.h>
+#include <float.h>
#include <limits.h>
#include <math.h>
@@ -27,12 +28,12 @@
#include "vp10/encoder/picklpf.h"
#include "vp10/encoder/pickrst.h"
-static int try_restoration_frame(const YV12_BUFFER_CONFIG *sd,
- VP10_COMP *const cpi,
- RestorationInfo *rsi,
- int partial_frame) {
+static int64_t try_restoration_frame(const YV12_BUFFER_CONFIG *sd,
+ VP10_COMP *const cpi,
+ RestorationInfo *rsi,
+ int partial_frame) {
VP10_COMMON *const cm = &cpi->common;
- int filt_err;
+ int64_t filt_err;
vp10_loop_restoration_frame(cm->frame_to_show, cm,
rsi, 1, partial_frame);
#if CONFIG_VP9_HIGHBITDEPTH
@@ -55,7 +56,8 @@
int filter_level, int partial_frame,
double *best_cost_ret) {
VP10_COMMON *const cm = &cpi->common;
- int i, restoration_best, err;
+ int i, restoration_best;
+ int64_t err;
double best_cost;
double cost;
const int restoration_level_bits = vp10_restoration_level_bits(&cpi->common);
@@ -400,8 +402,8 @@
}
}
-static void wiener_decompose_sep_sym(double *M, double *H,
- double *a, double *b) {
+static int wiener_decompose_sep_sym(double *M, double *H,
+ double *a, double *b) {
static const double init_filt[RESTORATION_WIN] = {
0.035623, -0.127154, 0.211436, 0.760190, 0.211436, -0.127154, 0.035623,
};
@@ -424,6 +426,7 @@
update_b_sep_sym(Mc, Hc, a, b);
iter++;
}
+ return 1;
}
#define CLIP(x, lo, hi) ((x) < (lo) ? (lo) : (x) > (hi) ? (hi) : (x))
@@ -448,7 +451,8 @@
double *best_cost_ret) {
VP10_COMMON *const cm = &cpi->common;
RestorationInfo rsi;
- int err, bits;
+ int64_t err;
+ int bits;
double cost_wiener, cost_norestore;
MACROBLOCK *x = &cpi->td.mb;
double M[RESTORATION_WIN2];
@@ -485,7 +489,10 @@
compute_stats(dgd->y_buffer, src->y_buffer, width, height,
dgd_stride, src_stride, M, H);
- wiener_decompose_sep_sym(M, H, vfilterd, hfilterd);
+ if (!wiener_decompose_sep_sym(M, H, vfilterd, hfilterd)) {
+ *best_cost_ret = DBL_MAX;
+ return 0;
+ }
quantize_sym_filter(vfilterd, vfilter);
quantize_sym_filter(hfilterd, hfilter);
@@ -493,7 +500,7 @@
memcpy(rsi.vfilter, vfilter, sizeof(rsi.vfilter));
memcpy(rsi.hfilter, hfilter, sizeof(rsi.hfilter));
err = try_restoration_frame(src, cpi, &rsi, partial_frame);
- bits = 22;
+ bits = WIENER_FILT_BITS;
cost_wiener = RDCOST_DBL(x->rdmult, x->rddiv, (bits << 2), err);
vpx_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show);
@@ -511,10 +518,10 @@
const YV12_BUFFER_CONFIG *sd, VP10_COMP *cpi, LPF_PICK_METHOD method) {
VP10_COMMON *const cm = &cpi->common;
struct loopfilter *const lf = &cm->lf;
- int wiener_success;
- double cost_bilateral = 1e12;
- double cost_wiener = 1e12;
- double cost_norestore = 1e12;
+ int wiener_success = 0;
+ double cost_bilateral = DBL_MAX;
+ double cost_wiener = DBL_MAX;
+ double cost_norestore = DBL_MAX;
lf->sharpness_level =
cm->frame_type == KEY_FRAME ? 0 : cpi->oxcf.sharpness;
@@ -577,8 +584,6 @@
wiener_success = search_wiener_filter(
sd, cpi, lf->filter_level, method == LPF_PICK_FROM_SUBIMAGE,
cm->rst_info.vfilter, cm->rst_info.hfilter, &cost_wiener);
- // printf("Costs %g %g (%d) %g\n",
- // cost_norestore, cost_bilateral, lf->filter_level, cost_wiener);
if (cost_bilateral < cost_wiener) {
lf->filter_level = blf_filter_level;
if (cm->rst_info.restoration_level != -1)
@@ -591,5 +596,8 @@
else
cm->rst_info.restoration_type = RESTORE_NONE;
}
+ // printf("[%d] Costs %g %g (%d) %g (%d)\n", cm->rst_info.restoration_type,
+ // cost_norestore, cost_bilateral, lf->filter_level, cost_wiener,
+ // wiener_success);
}
}
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index cd4c209..6543b33 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -1691,21 +1691,23 @@
return 0;
}
-void rd_pick_palette_intra_sby(VP10_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
- int palette_ctx, int dc_mode_cost,
- PALETTE_MODE_INFO *palette_mode_info,
- uint8_t *best_palette_color_map,
- TX_SIZE *best_tx, PREDICTION_MODE *mode_selected,
- int64_t *best_rd) {
+static int rd_pick_palette_intra_sby(VP10_COMP *cpi, MACROBLOCK *x,
+ BLOCK_SIZE bsize,
+ int palette_ctx, int dc_mode_cost,
+ PALETTE_MODE_INFO *palette_mode_info,
+ uint8_t *best_palette_color_map,
+ TX_SIZE *best_tx,
+ PREDICTION_MODE *mode_selected,
+ int64_t *best_rd) {
MACROBLOCKD *const xd = &x->e_mbd;
MODE_INFO *const mic = xd->mi[0];
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
- int this_rate, this_rate_tokenonly, s;
+ const int rows = 4 * num_4x4_blocks_high_lookup[bsize];
+ const int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
+ int this_rate, this_rate_tokenonly, s, colors, n;
+ int rate_overhead = 0;
int64_t this_distortion, this_rd;
- int colors, n;
- int src_stride = x->plane[0].src.stride;
- uint8_t *src = x->plane[0].src.buf;
+ const int src_stride = x->plane[0].src.stride;
+ const uint8_t *const src = x->plane[0].src.buf;
#if CONFIG_VP9_HIGHBITDEPTH
if (cpi->common.use_highbitdepth)
@@ -1715,19 +1717,23 @@
#endif // CONFIG_VP9_HIGHBITDEPTH
colors = vp10_count_colors(src, src_stride, rows, cols);
palette_mode_info->palette_size[0] = 0;
+#if CONFIG_EXT_INTRA
+ mic->mbmi.ext_intra_mode_info.use_ext_intra_mode[0] = 0;
+#endif // CONFIG_EXT_INTRA
if (colors > 1 && colors <= 64 && cpi->common.allow_screen_content_tools) {
int r, c, i, j, k;
- int max_itr = 50;
+ const int max_itr = 50;
int color_ctx, color_idx = 0;
int color_order[PALETTE_MAX_SIZE];
- double *data = x->palette_buffer->kmeans_data_buf;
- uint8_t *indices = x->palette_buffer->kmeans_indices_buf;
- uint8_t *pre_indices = x->palette_buffer->kmeans_pre_indices_buf;
+ double *const data = x->palette_buffer->kmeans_data_buf;
+ uint8_t *const indices = x->palette_buffer->kmeans_indices_buf;
+ uint8_t *const pre_indices = x->palette_buffer->kmeans_pre_indices_buf;
double centroids[PALETTE_MAX_SIZE];
- uint8_t *color_map;
+ uint8_t *const color_map = xd->plane[0].color_index_map;
double lb, ub, val;
- PALETTE_MODE_INFO *pmi = &mic->mbmi.palette_mode_info;
+ MB_MODE_INFO *const mbmi = &mic->mbmi;
+ PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
#if CONFIG_VP9_HIGHBITDEPTH
uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
if (cpi->common.use_highbitdepth)
@@ -1764,7 +1770,10 @@
}
#endif // CONFIG_VP9_HIGHBITDEPTH
- mic->mbmi.mode = DC_PRED;
+ mbmi->mode = DC_PRED;
+#if CONFIG_EXT_INTRA
+ mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 0;
+#endif // CONFIG_EXT_INTRA
for (n = colors > PALETTE_MAX_SIZE ? PALETTE_MAX_SIZE : colors;
n >= 2; --n) {
@@ -1805,7 +1814,7 @@
vp10_calc_indices(data, centroids, indices, rows * cols, k, 1);
for (r = 0; r < rows; ++r)
for (c = 0; c < cols; ++c)
- xd->plane[0].color_index_map[r * cols + c] = indices[r * cols + c];
+ color_map[r * cols + c] = indices[r * cols + c];
super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion,
&s, NULL, bsize, *best_rd);
@@ -1814,12 +1823,10 @@
this_rate = this_rate_tokenonly + dc_mode_cost +
cpi->common.bit_depth * k * vp10_cost_bit(128, 0) +
- cpi->palette_y_size_cost[bsize - BLOCK_8X8][k - 2];
- this_rate +=
+ cpi->palette_y_size_cost[bsize - BLOCK_8X8][k - 2] +
+ write_uniform_cost(k, color_map[0]) +
vp10_cost_bit(vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
- [palette_ctx], 1);
- color_map = xd->plane[0].color_index_map;
- this_rate += write_uniform_cost(k, xd->plane[0].color_index_map[0]);
+ [palette_ctx], 1);
for (i = 0; i < rows; ++i) {
for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
color_ctx = vp10_get_palette_color_context(color_map, cols, i, j,
@@ -1829,7 +1836,7 @@
color_idx = r;
break;
}
- assert(color_idx < k);
+ assert(color_idx >= 0 && color_idx < k);
this_rate +=
cpi->palette_y_color_cost[k - 2][color_ctx][color_idx];
}
@@ -1838,14 +1845,17 @@
if (this_rd < *best_rd) {
*best_rd = this_rd;
- *palette_mode_info = mic->mbmi.palette_mode_info;
- memcpy(best_palette_color_map, xd->plane[0].color_index_map,
- rows * cols * sizeof(xd->plane[0].color_index_map[0]));
+ *palette_mode_info = *pmi;
+ memcpy(best_palette_color_map, color_map,
+ rows * cols * sizeof(color_map[0]));
*mode_selected = DC_PRED;
- *best_tx = mic->mbmi.tx_size;
+ *best_tx = mbmi->tx_size;
+ rate_overhead = this_rate - this_rate_tokenonly;
}
}
}
+
+ return rate_overhead;
}
static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
@@ -2247,6 +2257,7 @@
vp10_zero(ext_intra_mode_info);
mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 1;
mbmi->mode = DC_PRED;
+ mbmi->palette_mode_info.palette_size[0] = 0;
for (mode = 0; mode < FILTER_INTRA_MODES; ++mode) {
mbmi->ext_intra_mode_info.ext_intra_mode[0] = mode;
@@ -2584,6 +2595,7 @@
TX_TYPE best_tx_type = DCT_DCT;
int *bmode_costs;
PALETTE_MODE_INFO palette_mode_info;
+ PALETTE_MODE_INFO *const pmi = &mic->mbmi.palette_mode_info;
uint8_t *best_palette_color_map = cpi->common.allow_screen_content_tools ?
x->palette_buffer->best_palette_color_map : NULL;
const int rows = 4 * num_4x4_blocks_high_lookup[bsize];
@@ -2630,7 +2642,7 @@
#endif // CONFIG_EXT_INTRA
memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
palette_mode_info.palette_size[0] = 0;
- mic->mbmi.palette_mode_info.palette_size[0] = 0;
+ pmi->palette_size[0] = 0;
if (above_mi)
palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
if (left_mi)
@@ -2717,7 +2729,7 @@
&best_tx, &mode_selected, &best_rd);
#if CONFIG_EXT_INTRA
- if (!palette_mode_info.palette_size[0] > 0 && ALLOW_FILTER_INTRA_MODES) {
+ if (ALLOW_FILTER_INTRA_MODES) {
if (rd_pick_ext_intra_sby(cpi, x, rate, rate_tokenonly, distortion,
skippable, bsize, bmode_costs[DC_PRED],
&best_rd)) {
@@ -2733,6 +2745,7 @@
if (ext_intra_mode_info.use_ext_intra_mode[0]) {
mic->mbmi.ext_intra_mode_info.ext_intra_mode[0] =
ext_intra_mode_info.ext_intra_mode[0];
+ palette_mode_info.palette_size[0] = 0;
}
#endif // CONFIG_EXT_INTRA
@@ -2743,11 +2756,9 @@
mic->mbmi.intra_filter = best_filter;
#endif // CONFIG_EXT_INTRA
mic->mbmi.tx_type = best_tx_type;
- mic->mbmi.palette_mode_info.palette_size[0] =
- palette_mode_info.palette_size[0];
+ pmi->palette_size[0] = palette_mode_info.palette_size[0];
if (palette_mode_info.palette_size[0] > 0) {
- memcpy(mic->mbmi.palette_mode_info.palette_colors,
- palette_mode_info.palette_colors,
+ memcpy(pmi->palette_colors, palette_mode_info.palette_colors,
PALETTE_MAX_SIZE * sizeof(palette_mode_info.palette_colors[0]));
memcpy(xd->plane[0].color_index_map, best_palette_color_map,
rows * cols * sizeof(best_palette_color_map[0]));
@@ -3498,6 +3509,183 @@
return is_cost_valid;
}
+static void rd_pick_palette_intra_sbuv(VP10_COMP *cpi, MACROBLOCK *x,
+ PICK_MODE_CONTEXT *ctx, int dc_mode_cost,
+ PALETTE_MODE_INFO *palette_mode_info,
+ uint8_t *best_palette_color_map,
+ PREDICTION_MODE *mode_selected,
+ int64_t *best_rd, int *rate,
+ int *rate_tokenonly,
+ int64_t *distortion, int *skippable) {
+ MACROBLOCKD *const xd = &x->e_mbd;
+ MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+ const BLOCK_SIZE bsize = mbmi->sb_type;
+ const int rows = (4 * num_4x4_blocks_high_lookup[bsize]) >>
+ (xd->plane[1].subsampling_y);
+ const int cols = (4 * num_4x4_blocks_wide_lookup[bsize]) >>
+ (xd->plane[1].subsampling_x);
+ int this_rate, this_rate_tokenonly, s;
+ int64_t this_distortion, this_rd;
+ int colors_u, colors_v, colors;
+ const int src_stride = x->plane[1].src.stride;
+ const uint8_t *const src_u = x->plane[1].src.buf;
+ const uint8_t *const src_v = x->plane[2].src.buf;
+
+#if CONFIG_EXT_INTRA
+ mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0;
+#endif // CONFIG_EXT_INTRA
+
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (cpi->common.use_highbitdepth) {
+ colors_u = vp10_count_colors_highbd(src_u, src_stride, rows, cols,
+ cpi->common.bit_depth);
+ colors_v = vp10_count_colors_highbd(src_v, src_stride, rows, cols,
+ cpi->common.bit_depth);
+ } else {
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ colors_u = vp10_count_colors(src_u, src_stride, rows, cols);
+ colors_v = vp10_count_colors(src_v, src_stride, rows, cols);
+#if CONFIG_VP9_HIGHBITDEPTH
+ }
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+ colors = colors_u > colors_v ? colors_u : colors_v;
+ if (colors > 1 && colors <= 64) {
+ int r, c, n, i, j;
+ const int max_itr = 50;
+ int color_ctx, color_idx = 0;
+ int color_order[PALETTE_MAX_SIZE];
+ int64_t this_sse;
+ double lb_u, ub_u, val_u;
+ double lb_v, ub_v, val_v;
+ double *const data = x->palette_buffer->kmeans_data_buf;
+ uint8_t *const indices = x->palette_buffer->kmeans_indices_buf;
+ uint8_t *const pre_indices = x->palette_buffer->kmeans_pre_indices_buf;
+ double centroids[2 * PALETTE_MAX_SIZE];
+ uint8_t *const color_map = xd->plane[1].color_index_map;
+ PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
+
+#if CONFIG_VP9_HIGHBITDEPTH
+ uint16_t *src_u16 = CONVERT_TO_SHORTPTR(src_u);
+ uint16_t *src_v16 = CONVERT_TO_SHORTPTR(src_v);
+ if (cpi->common.use_highbitdepth) {
+ lb_u = src_u16[0];
+ ub_u = src_u16[0];
+ lb_v = src_v16[0];
+ ub_v = src_v16[0];
+ } else {
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ lb_u = src_u[0];
+ ub_u = src_u[0];
+ lb_v = src_v[0];
+ ub_v = src_v[0];
+#if CONFIG_VP9_HIGHBITDEPTH
+ }
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+ mbmi->uv_mode = DC_PRED;
+#if CONFIG_EXT_INTRA
+ mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0;
+#endif // CONFIG_EXT_INTRA
+ for (r = 0; r < rows; ++r) {
+ for (c = 0; c < cols; ++c) {
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (cpi->common.use_highbitdepth) {
+ val_u = src_u16[r * src_stride + c];
+ val_v = src_v16[r * src_stride + c];
+ data[(r * cols + c) * 2 ] = val_u;
+ data[(r * cols + c) * 2 + 1] = val_v;
+ } else {
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ val_u = src_u[r * src_stride + c];
+ val_v = src_v[r * src_stride + c];
+ data[(r * cols + c) * 2 ] = val_u;
+ data[(r * cols + c) * 2 + 1] = val_v;
+#if CONFIG_VP9_HIGHBITDEPTH
+ }
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ if (val_u < lb_u)
+ lb_u = val_u;
+ else if (val_u > ub_u)
+ ub_u = val_u;
+ if (val_v < lb_v)
+ lb_v = val_v;
+ else if (val_v > ub_v)
+ ub_v = val_v;
+ }
+ }
+
+ for (n = colors > PALETTE_MAX_SIZE ? PALETTE_MAX_SIZE : colors;
+ n >= 2; --n) {
+ for (i = 0; i < n; ++i) {
+ centroids[i * 2] = lb_u + (2 * i + 1) * (ub_u - lb_u) / n / 2;
+ centroids[i * 2 + 1] =
+ lb_v + (2 * i + 1) * (ub_v - lb_v) / n / 2;;
+ }
+ r = vp10_k_means(data, centroids, indices, pre_indices, rows * cols, n,
+ 2, max_itr);
+ pmi->palette_size[1] = n;
+ for (i = 1; i < 3; ++i) {
+ for (j = 0; j < n; ++j) {
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (cpi->common.use_highbitdepth)
+ pmi->palette_colors[i * PALETTE_MAX_SIZE + j] =
+ clip_pixel_highbd(round(centroids[j * 2 + i - 1]),
+ cpi->common.bit_depth);
+ else
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ pmi->palette_colors[i * PALETTE_MAX_SIZE + j] =
+ clip_pixel(round(centroids[j * 2 + i - 1]));
+ }
+ }
+ for (r = 0; r < rows; ++r)
+ for (c = 0; c < cols; ++c)
+ color_map[r * cols + c] = indices[r * cols + c];
+
+ super_block_uvrd(cpi, x, &this_rate_tokenonly,
+ &this_distortion, &s, &this_sse, bsize, *best_rd);
+ if (this_rate_tokenonly == INT_MAX)
+ continue;
+ this_rate = this_rate_tokenonly + dc_mode_cost +
+ 2 * cpi->common.bit_depth * n * vp10_cost_bit(128, 0) +
+ cpi->palette_uv_size_cost[bsize - BLOCK_8X8][n - 2] +
+ write_uniform_cost(n, color_map[0]) +
+ vp10_cost_bit(vp10_default_palette_uv_mode_prob
+ [pmi->palette_size[0] > 0], 1);
+
+ for (i = 0; i < rows; ++i) {
+ for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
+ color_ctx = vp10_get_palette_color_context(color_map, cols, i, j, n,
+ color_order);
+ for (r = 0; r < n; ++r)
+ if (color_map[i * cols + j] == color_order[r]) {
+ color_idx = r;
+ break;
+ }
+ assert(color_idx >= 0 && color_idx < n);
+ this_rate +=
+ cpi->palette_uv_color_cost[n - 2][color_ctx][color_idx];
+ }
+ }
+
+ this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
+ if (this_rd < *best_rd) {
+ *best_rd = this_rd;
+ *palette_mode_info = *pmi;
+ memcpy(best_palette_color_map, xd->plane[1].color_index_map,
+ rows * cols * sizeof(best_palette_color_map[0]));
+ *mode_selected = DC_PRED;
+ *rate = this_rate;
+ *distortion = this_distortion;
+ *rate_tokenonly = this_rate_tokenonly;
+ *skippable = s;
+ if (!x->select_tx_size)
+ swap_block_ptr(x, ctx, 2, 0, 1, MAX_MB_PLANE);
+ }
+ }
+ }
+}
+
#if CONFIG_EXT_INTRA
// Return 1 if an ext intra mode is selected; return 0 otherwise.
static int rd_pick_ext_intra_sbuv(VP10_COMP *cpi, MACROBLOCK *x,
@@ -3516,6 +3704,7 @@
vp10_zero(ext_intra_mode_info);
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 1;
mbmi->uv_mode = DC_PRED;
+ mbmi->palette_mode_info.palette_size[1] = 0;
for (mode = 0; mode < FILTER_INTRA_MODES; ++mode) {
mbmi->ext_intra_mode_info.ext_intra_mode[1] = mode;
@@ -3670,6 +3859,13 @@
int64_t best_rd = INT64_MAX, this_rd;
int this_rate_tokenonly, this_rate, s;
int64_t this_distortion, this_sse;
+ const int rows = (4 * num_4x4_blocks_high_lookup[bsize]) >>
+ (xd->plane[1].subsampling_y);
+ const int cols = (4 * num_4x4_blocks_wide_lookup[bsize]) >>
+ (xd->plane[1].subsampling_x);
+ PALETTE_MODE_INFO palette_mode_info;
+ PALETTE_MODE_INFO *const pmi = &xd->mi[0]->mbmi.palette_mode_info;
+ uint8_t *best_palette_color_map = NULL;
#if CONFIG_EXT_INTRA
int is_directional_mode, rate_overhead, best_angle_delta = 0;
EXT_INTRA_MODE_INFO ext_intra_mode_info;
@@ -3678,7 +3874,8 @@
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0;
#endif // CONFIG_EXT_INTRA
memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
- xd->mi[0]->mbmi.palette_mode_info.palette_size[1] = 0;
+ palette_mode_info.palette_size[1] = 0;
+ pmi->palette_size[1] = 0;
for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
if (!(cpi->sf.intra_uv_mode_mask[max_tx_size] & (1 << mode)))
continue;
@@ -3705,15 +3902,19 @@
this_rate += write_uniform_cost(2 * MAX_ANGLE_DELTAS + 1,
MAX_ANGLE_DELTAS +
mbmi->angle_delta[1]);
- if (mode == DC_PRED && 0)
+ if (mode == DC_PRED)
this_rate += vp10_cost_bit(cpi->common.fc->ext_intra_probs[1], 0);
#else
if (!super_block_uvrd(cpi, x, &this_rate_tokenonly,
&this_distortion, &s, &this_sse, bsize, best_rd))
continue;
this_rate = this_rate_tokenonly +
- cpi->intra_uv_mode_cost[xd->mi[0]->mbmi.mode][mode];
+ cpi->intra_uv_mode_cost[mbmi->mode][mode];
#endif // CONFIG_EXT_INTRA
+ if (cpi->common.allow_screen_content_tools && mbmi->sb_type >= BLOCK_8X8 &&
+ mode == DC_PRED)
+ this_rate += vp10_cost_bit(vp10_default_palette_uv_mode_prob
+ [pmi->palette_size[0] > 0], 0);
this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
@@ -3732,6 +3933,15 @@
}
}
+ if (cpi->common.allow_screen_content_tools && mbmi->sb_type >= BLOCK_8X8) {
+ best_palette_color_map = x->palette_buffer->best_palette_color_map;
+ rd_pick_palette_intra_sbuv(cpi, x, ctx,
+ cpi->intra_uv_mode_cost[mbmi->mode][DC_PRED],
+ &palette_mode_info, best_palette_color_map,
+ &mode_selected, &best_rd, rate, rate_tokenonly,
+ distortion, skippable);
+ }
+
#if CONFIG_EXT_INTRA
if (mbmi->sb_type >= BLOCK_8X8 && ALLOW_FILTER_INTRA_MODES) {
if (rd_pick_ext_intra_sbuv(cpi, x, ctx, rate, rate_tokenonly, distortion,
@@ -3743,12 +3953,23 @@
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] =
ext_intra_mode_info.use_ext_intra_mode[1];
- if (ext_intra_mode_info.use_ext_intra_mode[1])
+ if (ext_intra_mode_info.use_ext_intra_mode[1]) {
mbmi->ext_intra_mode_info.ext_intra_mode[1] =
ext_intra_mode_info.ext_intra_mode[1];
+ palette_mode_info.palette_size[1] = 0;
+ }
mbmi->angle_delta[1] = best_angle_delta;
#endif // CONFIG_EXT_INTRA
mbmi->uv_mode = mode_selected;
+ pmi->palette_size[1] = palette_mode_info.palette_size[1];
+ if (palette_mode_info.palette_size[1] > 0) {
+ memcpy(pmi->palette_colors + PALETTE_MAX_SIZE,
+ palette_mode_info.palette_colors + PALETTE_MAX_SIZE,
+ 2 * PALETTE_MAX_SIZE * sizeof(palette_mode_info.palette_colors[0]));
+ memcpy(xd->plane[1].color_index_map, best_palette_color_map,
+ rows * cols * sizeof(best_palette_color_map[0]));
+ }
+
return best_rd;
}
@@ -4443,52 +4664,52 @@
if (bestsme < INT_MAX) {
int dis; /* TODO: use dis in distortion calculation later. */
unsigned int sse;
-#if CONFIG_AFFINE_MOTION
- // Use up-sampled reference frames.
- struct macroblockd_plane *const pd = &xd->plane[0];
- struct buf_2d backup_pred = pd->pre[0];
- const YV12_BUFFER_CONFIG *upsampled_ref =
- get_upsampled_ref(cpi, refs[id]);
+ if (cpi->sf.use_upsampled_references) {
+ // Use up-sampled reference frames.
+ struct macroblockd_plane *const pd = &xd->plane[0];
+ struct buf_2d backup_pred = pd->pre[0];
+ const YV12_BUFFER_CONFIG *upsampled_ref =
+ get_upsampled_ref(cpi, refs[id]);
- // Set pred for Y plane
- setup_pred_plane(&pd->pre[0], upsampled_ref->y_buffer,
- upsampled_ref->y_stride, (mi_row << 3), (mi_col << 3),
- NULL, pd->subsampling_x, pd->subsampling_y);
+ // Set pred for Y plane
+ setup_pred_plane(&pd->pre[0], upsampled_ref->y_buffer,
+ upsampled_ref->y_stride, (mi_row << 3), (mi_col << 3),
+ NULL, pd->subsampling_x, pd->subsampling_y);
- // If bsize < BLOCK_8X8, adjust pred pointer for this block
- if (bsize < BLOCK_8X8)
- pd->pre[0].buf =
- &pd->pre[0].buf[(vp10_raster_block_offset(BLOCK_8X8, block,
- pd->pre[0].stride)) << 3];
+ // If bsize < BLOCK_8X8, adjust pred pointer for this block
+ if (bsize < BLOCK_8X8)
+ pd->pre[0].buf =
+ &pd->pre[0].buf[(vp10_raster_block_offset(BLOCK_8X8, block,
+ pd->pre[0].stride)) << 3];
- bestsme = cpi->find_fractional_mv_step(
- x, &tmp_mv,
- &ref_mv[id].as_mv,
- cpi->common.allow_high_precision_mv,
- x->errorperbit,
- &cpi->fn_ptr[bsize],
- 0, cpi->sf.mv.subpel_iters_per_step,
- NULL,
- x->nmvjointcost, x->mvcost,
- &dis, &sse, second_pred,
- pw, ph, 1);
+ bestsme = cpi->find_fractional_mv_step(
+ x, &tmp_mv,
+ &ref_mv[id].as_mv,
+ cpi->common.allow_high_precision_mv,
+ x->errorperbit,
+ &cpi->fn_ptr[bsize],
+ 0, cpi->sf.mv.subpel_iters_per_step,
+ NULL,
+ x->nmvjointcost, x->mvcost,
+ &dis, &sse, second_pred,
+ pw, ph, 1);
- // Restore the reference frames.
- pd->pre[0] = backup_pred;
-#else
- (void) block;
- bestsme = cpi->find_fractional_mv_step(
- x, &tmp_mv,
- &ref_mv[id].as_mv,
- cpi->common.allow_high_precision_mv,
- x->errorperbit,
- &cpi->fn_ptr[bsize],
- 0, cpi->sf.mv.subpel_iters_per_step,
- NULL,
- x->nmvjointcost, x->mvcost,
- &dis, &sse, second_pred,
- pw, ph);
-#endif
+ // Restore the reference frames.
+ pd->pre[0] = backup_pred;
+ } else {
+ (void) block;
+ bestsme = cpi->find_fractional_mv_step(
+ x, &tmp_mv,
+ &ref_mv[id].as_mv,
+ cpi->common.allow_high_precision_mv,
+ x->errorperbit,
+ &cpi->fn_ptr[bsize],
+ 0, cpi->sf.mv.subpel_iters_per_step,
+ NULL,
+ x->nmvjointcost, x->mvcost,
+ &dis, &sse, second_pred,
+ pw, ph, 0);
+ }
}
// Restore the pointer to the first (possibly scaled) prediction buffer.
@@ -4769,57 +4990,57 @@
if (bestsme < INT_MAX) {
int distortion;
-#if CONFIG_AFFINE_MOTION
- const int pw = 4 * num_4x4_blocks_wide_lookup[bsize];
- const int ph = 4 * num_4x4_blocks_high_lookup[bsize];
- // Use up-sampled reference frames.
- struct macroblockd_plane *const pd = &xd->plane[0];
- struct buf_2d backup_pred = pd->pre[0];
- const YV12_BUFFER_CONFIG *upsampled_ref =
- get_upsampled_ref(cpi, mbmi->ref_frame[0]);
+ if (cpi->sf.use_upsampled_references) {
+ const int pw = 4 * num_4x4_blocks_wide_lookup[bsize];
+ const int ph = 4 * num_4x4_blocks_high_lookup[bsize];
+ // Use up-sampled reference frames.
+ struct macroblockd_plane *const pd = &xd->plane[0];
+ struct buf_2d backup_pred = pd->pre[0];
+ const YV12_BUFFER_CONFIG *upsampled_ref =
+ get_upsampled_ref(cpi, mbmi->ref_frame[0]);
- // Set pred for Y plane
- setup_pred_plane(&pd->pre[0], upsampled_ref->y_buffer,
- upsampled_ref->y_stride,
- (mi_row << 3), (mi_col << 3),
- NULL, pd->subsampling_x, pd->subsampling_y);
+ // Set pred for Y plane
+ setup_pred_plane(&pd->pre[0], upsampled_ref->y_buffer,
+ upsampled_ref->y_stride,
+ (mi_row << 3), (mi_col << 3),
+ NULL, pd->subsampling_x, pd->subsampling_y);
- // adjust pred pointer for this block
- pd->pre[0].buf =
- &pd->pre[0].buf[(vp10_raster_block_offset(BLOCK_8X8, i,
- pd->pre[0].stride)) << 3];
+ // adjust pred pointer for this block
+ pd->pre[0].buf =
+ &pd->pre[0].buf[(vp10_raster_block_offset(BLOCK_8X8, i,
+ pd->pre[0].stride)) << 3];
- cpi->find_fractional_mv_step(
- x,
- new_mv,
- &bsi->ref_mv[0]->as_mv,
- cm->allow_high_precision_mv,
- x->errorperbit, &cpi->fn_ptr[bsize],
- cpi->sf.mv.subpel_force_stop,
- cpi->sf.mv.subpel_iters_per_step,
- cond_cost_list(cpi, cost_list),
- x->nmvjointcost, x->mvcost,
- &distortion,
- &x->pred_sse[mbmi->ref_frame[0]],
- NULL, pw, ph, 1);
+ cpi->find_fractional_mv_step(
+ x,
+ new_mv,
+ &bsi->ref_mv[0]->as_mv,
+ cm->allow_high_precision_mv,
+ x->errorperbit, &cpi->fn_ptr[bsize],
+ cpi->sf.mv.subpel_force_stop,
+ cpi->sf.mv.subpel_iters_per_step,
+ cond_cost_list(cpi, cost_list),
+ x->nmvjointcost, x->mvcost,
+ &distortion,
+ &x->pred_sse[mbmi->ref_frame[0]],
+ NULL, pw, ph, 1);
- // Restore the reference frames.
- pd->pre[0] = backup_pred;
-#else
- cpi->find_fractional_mv_step(
- x,
- new_mv,
- &bsi->ref_mv[0]->as_mv,
- cm->allow_high_precision_mv,
- x->errorperbit, &cpi->fn_ptr[bsize],
- cpi->sf.mv.subpel_force_stop,
- cpi->sf.mv.subpel_iters_per_step,
- cond_cost_list(cpi, cost_list),
- x->nmvjointcost, x->mvcost,
- &distortion,
- &x->pred_sse[mbmi->ref_frame[0]],
- NULL, 0, 0);
-#endif
+ // Restore the reference frames.
+ pd->pre[0] = backup_pred;
+ } else {
+ cpi->find_fractional_mv_step(
+ x,
+ new_mv,
+ &bsi->ref_mv[0]->as_mv,
+ cm->allow_high_precision_mv,
+ x->errorperbit, &cpi->fn_ptr[bsize],
+ cpi->sf.mv.subpel_force_stop,
+ cpi->sf.mv.subpel_iters_per_step,
+ cond_cost_list(cpi, cost_list),
+ x->nmvjointcost, x->mvcost,
+ &distortion,
+ &x->pred_sse[mbmi->ref_frame[0]],
+ NULL, 0, 0, 0);
+ }
// save motion search result for use in compound prediction
#if CONFIG_EXT_INTER
@@ -5416,43 +5637,43 @@
if (bestsme < INT_MAX) {
int dis; /* TODO: use dis in distortion calculation later. */
-#if CONFIG_AFFINE_MOTION
- const int pw = 4 * num_4x4_blocks_wide_lookup[bsize];
- const int ph = 4 * num_4x4_blocks_high_lookup[bsize];
- // Use up-sampled reference frames.
- struct macroblockd_plane *const pd = &xd->plane[0];
- struct buf_2d backup_pred = pd->pre[ref_idx];
- const YV12_BUFFER_CONFIG *upsampled_ref = get_upsampled_ref(cpi, ref);
+ if (cpi->sf.use_upsampled_references) {
+ const int pw = 4 * num_4x4_blocks_wide_lookup[bsize];
+ const int ph = 4 * num_4x4_blocks_high_lookup[bsize];
+ // Use up-sampled reference frames.
+ struct macroblockd_plane *const pd = &xd->plane[0];
+ struct buf_2d backup_pred = pd->pre[ref_idx];
+ const YV12_BUFFER_CONFIG *upsampled_ref = get_upsampled_ref(cpi, ref);
- // Set pred for Y plane
- setup_pred_plane(&pd->pre[ref_idx], upsampled_ref->y_buffer,
- upsampled_ref->y_stride, (mi_row << 3), (mi_col << 3),
- NULL, pd->subsampling_x, pd->subsampling_y);
+ // Set pred for Y plane
+ setup_pred_plane(&pd->pre[ref_idx], upsampled_ref->y_buffer,
+ upsampled_ref->y_stride, (mi_row << 3), (mi_col << 3),
+ NULL, pd->subsampling_x, pd->subsampling_y);
- bestsme = cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
- cm->allow_high_precision_mv,
- x->errorperbit,
- &cpi->fn_ptr[bsize],
- cpi->sf.mv.subpel_force_stop,
- cpi->sf.mv.subpel_iters_per_step,
- cond_cost_list(cpi, cost_list),
- x->nmvjointcost, x->mvcost,
- &dis, &x->pred_sse[ref], NULL,
- pw, ph, 1);
+ bestsme = cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
+ cm->allow_high_precision_mv,
+ x->errorperbit,
+ &cpi->fn_ptr[bsize],
+ cpi->sf.mv.subpel_force_stop,
+ cpi->sf.mv.subpel_iters_per_step,
+ cond_cost_list(cpi, cost_list),
+ x->nmvjointcost, x->mvcost,
+ &dis, &x->pred_sse[ref], NULL,
+ pw, ph, 1);
- // Restore the reference frames.
- pd->pre[ref_idx] = backup_pred;
-#else
- cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
- cm->allow_high_precision_mv,
- x->errorperbit,
- &cpi->fn_ptr[bsize],
- cpi->sf.mv.subpel_force_stop,
- cpi->sf.mv.subpel_iters_per_step,
- cond_cost_list(cpi, cost_list),
- x->nmvjointcost, x->mvcost,
- &dis, &x->pred_sse[ref], NULL, 0, 0);
-#endif
+ // Restore the reference frames.
+ pd->pre[ref_idx] = backup_pred;
+ } else {
+ cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
+ cm->allow_high_precision_mv,
+ x->errorperbit,
+ &cpi->fn_ptr[bsize],
+ cpi->sf.mv.subpel_force_stop,
+ cpi->sf.mv.subpel_iters_per_step,
+ cond_cost_list(cpi, cost_list),
+ x->nmvjointcost, x->mvcost,
+ &dis, &x->pred_sse[ref], NULL, 0, 0, 0);
+ }
}
*rate_mv = vp10_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
@@ -7066,6 +7287,63 @@
vp10_active_v_edge(cpi, mi_col, MI_BLOCK_SIZE);
}
+static void restore_uv_color_map(VP10_COMP *cpi, MACROBLOCK *x) {
+ MACROBLOCKD *const xd = &x->e_mbd;
+ MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+ PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
+ const BLOCK_SIZE bsize = mbmi->sb_type;
+ const int rows = (4 * num_4x4_blocks_high_lookup[bsize]) >>
+ (xd->plane[1].subsampling_y);
+ const int cols = (4 * num_4x4_blocks_wide_lookup[bsize]) >>
+ (xd->plane[1].subsampling_x);
+ int src_stride = x->plane[1].src.stride;
+ const uint8_t *const src_u = x->plane[1].src.buf;
+ const uint8_t *const src_v = x->plane[2].src.buf;
+ double *const data = x->palette_buffer->kmeans_data_buf;
+ uint8_t *const indices = x->palette_buffer->kmeans_indices_buf;
+ double centroids[2 * PALETTE_MAX_SIZE];
+ uint8_t *const color_map = xd->plane[1].color_index_map;
+ int r, c;
+#if CONFIG_VP9_HIGHBITDEPTH
+ const uint16_t *const src_u16 = CONVERT_TO_SHORTPTR(src_u);
+ const uint16_t *const src_v16 = CONVERT_TO_SHORTPTR(src_v);
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ (void)cpi;
+
+ for (r = 0; r < rows; ++r) {
+ for (c = 0; c < cols; ++c) {
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (cpi->common.use_highbitdepth) {
+ data[(r * cols + c) * 2 ] =
+ src_u16[r * src_stride + c];
+ data[(r * cols + c) * 2 + 1] =
+ src_v16[r * src_stride + c];
+ } else {
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ data[(r * cols + c) * 2 ] =
+ src_u[r * src_stride + c];
+ data[(r * cols + c) * 2 + 1] =
+ src_v[r * src_stride + c];
+#if CONFIG_VP9_HIGHBITDEPTH
+ }
+#endif // CONFIG_VP9_HIGHBITDEPTH
+ }
+ }
+
+ for (r = 1; r < 3; ++r) {
+ for (c = 0; c < pmi->palette_size[1]; ++c) {
+ centroids[c * 2 + r - 1] = pmi->palette_colors[r * PALETTE_MAX_SIZE + c];
+ }
+ }
+
+ vp10_calc_indices(data, centroids, indices, rows * cols,
+ pmi->palette_size[1], 2);
+
+ for (r = 0; r < rows; ++r)
+ for (c = 0; c < cols; ++c)
+ color_map[r * cols + c] = indices[r * cols + c];
+}
+
void vp10_rd_pick_inter_mode_sb(VP10_COMP *cpi,
TileDataEnc *tile_data,
MACROBLOCK *x,
@@ -7082,6 +7360,7 @@
SPEED_FEATURES *const sf = &cpi->sf;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+ PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
const struct segmentation *const seg = &cm->seg;
PREDICTION_MODE this_mode;
@@ -7126,6 +7405,7 @@
int64_t dist_uv[TX_SIZES];
int skip_uv[TX_SIZES];
PREDICTION_MODE mode_uv[TX_SIZES];
+ PALETTE_MODE_INFO pmi_uv[TX_SIZES];
#if CONFIG_EXT_INTRA
EXT_INTRA_MODE_INFO ext_intra_mode_info_uv[TX_SIZES];
int8_t uv_angle_delta[TX_SIZES];
@@ -7153,6 +7433,11 @@
int64_t mask_filter = 0;
int64_t filter_cache[SWITCHABLE_FILTER_CONTEXTS];
const TX_SIZE max_tx_size = max_txsize_lookup[bsize];
+ int palette_ctx = 0;
+ const int rows = 4 * num_4x4_blocks_high_lookup[bsize];
+ const int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
+ const MODE_INFO *above_mi = xd->above_mi;
+ const MODE_INFO *left_mi = xd->left_mi;
#if CONFIG_OBMC
#if CONFIG_VP9_HIGHBITDEPTH
DECLARE_ALIGNED(16, uint8_t, tmp_buf1[2 * MAX_MB_PLANE * 64 * 64]);
@@ -7188,6 +7473,14 @@
#endif // CONFIG_OBMC
vp10_zero(best_mbmode);
+ vp10_zero(pmi_uv);
+
+ if (cm->allow_screen_content_tools) {
+ if (above_mi)
+ palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
+ if (left_mi)
+ palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
+ }
#if CONFIG_EXT_INTRA
memset(directional_mode_skip_mask, 0,
@@ -7370,8 +7663,6 @@
midx = end_pos;
}
- mbmi->palette_mode_info.palette_size[0] = 0;
- mbmi->palette_mode_info.palette_size[1] = 0;
for (midx = 0; midx < MAX_MODES; ++midx) {
int mode_index = mode_map[midx];
int mode_excluded = 0;
@@ -7565,6 +7856,8 @@
mbmi->uv_mode = DC_PRED;
mbmi->ref_frame[0] = ref_frame;
mbmi->ref_frame[1] = second_ref_frame;
+ pmi->palette_size[0] = 0;
+ pmi->palette_size[1] = 0;
#if CONFIG_EXT_INTRA
mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 0;
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0;
@@ -7597,7 +7890,6 @@
TX_SIZE uv_tx;
struct macroblockd_plane *const pd = &xd->plane[1];
memset(x->skip_txfm, 0, sizeof(x->skip_txfm));
-
#if CONFIG_EXT_INTRA
is_directional_mode = (mbmi->mode != DC_PRED && mbmi->mode != TM_PRED);
if (is_directional_mode) {
@@ -7681,6 +7973,8 @@
choose_intra_uv_mode(cpi, x, ctx, bsize, uv_tx,
&rate_uv_intra[uv_tx], &rate_uv_tokenonly[uv_tx],
&dist_uv[uv_tx], &skip_uv[uv_tx], &mode_uv[uv_tx]);
+ if (cm->allow_screen_content_tools)
+ pmi_uv[uv_tx] = *pmi;
#if CONFIG_EXT_INTRA
ext_intra_mode_info_uv[uv_tx] = mbmi->ext_intra_mode_info;
uv_angle_delta[uv_tx] = mbmi->angle_delta[1];
@@ -7691,6 +7985,12 @@
distortion_uv = dist_uv[uv_tx];
skippable = skippable && skip_uv[uv_tx];
mbmi->uv_mode = mode_uv[uv_tx];
+ if (cm->allow_screen_content_tools) {
+ pmi->palette_size[1] = pmi_uv[uv_tx].palette_size[1];
+ memcpy(pmi->palette_colors + PALETTE_MAX_SIZE,
+ pmi_uv[uv_tx].palette_colors + PALETTE_MAX_SIZE,
+ 2 * PALETTE_MAX_SIZE * sizeof(pmi->palette_colors[0]));
+ }
#if CONFIG_EXT_INTRA
mbmi->angle_delta[1] = uv_angle_delta[uv_tx];
mbmi->ext_intra_mode_info.use_ext_intra_mode[1] =
@@ -7702,6 +8002,10 @@
#endif // CONFIG_EXT_INTRA
rate2 = rate_y + intra_mode_cost[mbmi->mode] + rate_uv_intra[uv_tx];
+ if (cpi->common.allow_screen_content_tools && mbmi->mode == DC_PRED)
+ rate2 +=
+ vp10_cost_bit(vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
+ [palette_ctx], 0);
if (!xd->lossless[mbmi->segment_id]) {
// super_block_yrd above includes the cost of the tx_size in the
@@ -8140,6 +8444,104 @@
break;
}
+ // Only try palette mode when the best mode so far is an intra mode.
+ if (cm->allow_screen_content_tools && !is_inter_mode(best_mbmode.mode)) {
+ PREDICTION_MODE mode_selected;
+ int rate2 = 0, rate_y = 0;
+ int64_t distortion2 = 0, distortion_y = 0, dummy_rd = best_rd, this_rd;
+ int skippable = 0, rate_overhead = 0;
+ TX_SIZE best_tx_size, uv_tx;
+ PALETTE_MODE_INFO palette_mode_info;
+ uint8_t *const best_palette_color_map =
+ x->palette_buffer->best_palette_color_map;
+ uint8_t *const color_map = xd->plane[0].color_index_map;
+
+ mbmi->mode = DC_PRED;
+ mbmi->uv_mode = DC_PRED;
+ mbmi->ref_frame[0] = INTRA_FRAME;
+ mbmi->ref_frame[1] = NONE;
+ memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
+ palette_mode_info.palette_size[0] = 0;
+ rate_overhead =
+ rd_pick_palette_intra_sby(cpi, x, bsize, palette_ctx,
+ intra_mode_cost[DC_PRED],
+ &palette_mode_info, best_palette_color_map,
+ &best_tx_size, &mode_selected, &dummy_rd);
+ if (palette_mode_info.palette_size[0] == 0)
+ goto PALETTE_EXIT;
+
+ pmi->palette_size[0] =
+ palette_mode_info.palette_size[0];
+ if (palette_mode_info.palette_size[0] > 0) {
+ memcpy(pmi->palette_colors, palette_mode_info.palette_colors,
+ PALETTE_MAX_SIZE * sizeof(palette_mode_info.palette_colors[0]));
+ memcpy(color_map, best_palette_color_map,
+ rows * cols * sizeof(best_palette_color_map[0]));
+ }
+ super_block_yrd(cpi, x, &rate_y, &distortion_y, &skippable,
+ NULL, bsize, best_rd);
+ if (rate_y == INT_MAX)
+ goto PALETTE_EXIT;
+ uv_tx = get_uv_tx_size_impl(mbmi->tx_size, bsize,
+ xd->plane[1].subsampling_x,
+ xd->plane[1].subsampling_y);
+ if (rate_uv_intra[uv_tx] == INT_MAX) {
+ choose_intra_uv_mode(cpi, x, ctx, bsize, uv_tx,
+ &rate_uv_intra[uv_tx], &rate_uv_tokenonly[uv_tx],
+ &dist_uv[uv_tx], &skip_uv[uv_tx], &mode_uv[uv_tx]);
+ pmi_uv[uv_tx] = *pmi;
+#if CONFIG_EXT_INTRA
+ ext_intra_mode_info_uv[uv_tx] = mbmi->ext_intra_mode_info;
+ uv_angle_delta[uv_tx] = mbmi->angle_delta[1];
+#endif // CONFIG_EXT_INTRA
+ }
+ mbmi->uv_mode = mode_uv[uv_tx];
+ pmi->palette_size[1] = pmi_uv[uv_tx].palette_size[1];
+ if (pmi->palette_size[1] > 0)
+ memcpy(pmi->palette_colors + PALETTE_MAX_SIZE,
+ pmi_uv[uv_tx].palette_colors + PALETTE_MAX_SIZE,
+ 2 * PALETTE_MAX_SIZE * sizeof(pmi->palette_colors[0]));
+#if CONFIG_EXT_INTRA
+ mbmi->angle_delta[1] = uv_angle_delta[uv_tx];
+ mbmi->ext_intra_mode_info.use_ext_intra_mode[1] =
+ ext_intra_mode_info_uv[uv_tx].use_ext_intra_mode[1];
+ if (ext_intra_mode_info_uv[uv_tx].use_ext_intra_mode[1]) {
+ mbmi->ext_intra_mode_info.ext_intra_mode[1] =
+ ext_intra_mode_info_uv[uv_tx].ext_intra_mode[1];
+ }
+#endif // CONFIG_EXT_INTRA
+ skippable = skippable && skip_uv[uv_tx];
+ distortion2 = distortion_y + dist_uv[uv_tx];
+ rate2 = rate_y + rate_overhead + rate_uv_intra[uv_tx];
+ rate2 += ref_costs_single[INTRA_FRAME];
+
+ if (skippable) {
+ rate2 -= (rate_y + rate_uv_tokenonly[uv_tx]);
+ rate2 += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 1);
+ } else {
+ rate2 += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 0);
+ }
+ this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
+ if (this_rd < best_rd) {
+ int max_plane = MAX_MB_PLANE;
+ best_mode_index = 3;
+ mbmi->mv[0].as_int = 0;
+ max_plane = 1;
+ rd_cost->rate = rate2;
+ rd_cost->dist = distortion2;
+ rd_cost->rdcost = this_rd;
+ best_rd = this_rd;
+ best_mbmode = *mbmi;
+ best_skip2 = 0;
+ best_mode_skippable = skippable;
+ if (!x->select_tx_size)
+ swap_block_ptr(x, ctx, 1, 0, 0, max_plane);
+ memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
+ sizeof(ctx->zcoeff_blk[0]) * ctx->num_4x4_blk);
+ }
+ }
+ PALETTE_EXIT:
+
// The inter modes' rate costs are not calculated precisely in some cases.
// Therefore, sometimes, NEWMV is chosen instead of NEARESTMV, NEARMV, and
// ZEROMV. Here, checks are added for those cases, and the mode decisions
@@ -8366,6 +8768,10 @@
store_coding_context(x, ctx, best_mode_index, best_pred_diff,
best_filter_diff, best_mode_skippable);
+
+ if (cm->allow_screen_content_tools && pmi->palette_size[1] > 0) {
+ restore_uv_color_map(cpi, x);
+ }
}
void vp10_rd_pick_inter_mode_sb_seg_skip(VP10_COMP *cpi,
@@ -8951,16 +9357,6 @@
if (!vp10_is_interp_needed(xd) && cm->interp_filter == SWITCHABLE &&
mbmi->interp_filter != EIGHTTAP_REGULAR) {
mbmi->interp_filter = EIGHTTAP_REGULAR;
- tmp_rd = rd_pick_best_sub8x8_mode(cpi, x,
- &x->mbmi_ext->ref_mvs[ref_frame][0],
- second_ref, best_yrd, &rate, &rate_y,
- &distortion, &skippable, &total_sse,
- (int) this_rd_thresh, seg_mvs,
-#if CONFIG_EXT_INTER
- compound_seg_newmvs,
-#endif // CONFIG_EXT_INTER
- bsi, 0,
- mi_row, mi_col);
}
#endif // CONFIG_EXT_INTERP
if (tmp_rd == INT64_MAX)
diff --git a/vp10/encoder/rdopt.h b/vp10/encoder/rdopt.h
index f4d9b95..174ad4d 100644
--- a/vp10/encoder/rdopt.h
+++ b/vp10/encoder/rdopt.h
@@ -103,13 +103,16 @@
int tmp_stride[MAX_MB_PLANE]);
#endif // CONFIG_OBMC
-#if CONFIG_AFFINE_MOTION
static INLINE const YV12_BUFFER_CONFIG *get_upsampled_ref(VP10_COMP *cpi,
const int ref) {
// Use up-sampled reference frames.
int ref_idx = 0;
if (ref == LAST_FRAME)
+#if CONFIG_EXT_REFS
+ ref_idx = cpi->lst_fb_idxes[ref - LAST_FRAME];
+#else
ref_idx = cpi->lst_fb_idx;
+#endif
else if (ref == GOLDEN_FRAME)
ref_idx = cpi->gld_fb_idx;
else if (ref == ALTREF_FRAME)
@@ -117,7 +120,6 @@
return &cpi->upsampled_ref_bufs[cpi->upsampled_ref_idx[ref_idx]].buf;
}
-#endif
#ifdef __cplusplus
} // extern "C"
diff --git a/vp10/encoder/speed_features.c b/vp10/encoder/speed_features.c
index ec8acda..169ae2c 100644
--- a/vp10/encoder/speed_features.c
+++ b/vp10/encoder/speed_features.c
@@ -177,6 +177,7 @@
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->allow_partition_search_skip = 1;
+ sf->use_upsampled_references = 0;
#if CONFIG_EXT_TX
sf->tx_type_search = PRUNE_TWO;
#endif
@@ -279,6 +280,7 @@
sf->use_fast_coef_costing = 1;
sf->allow_exhaustive_searches = 0;
sf->exhaustive_searches_thresh = INT_MAX;
+ sf->use_upsampled_references = 0;
// Use transform domain distortion computation
// Note var-tx expt always uses pixel domain distortion.
@@ -495,6 +497,11 @@
sf->disable_filter_search_var_thresh = 0;
sf->adaptive_interp_filter_search = 0;
sf->allow_partition_search_skip = 0;
+#if CONFIG_EXT_REFS
+ sf->use_upsampled_references = 0;
+#else
+ sf->use_upsampled_references = 1;
+#endif
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = INTRA_ALL;
diff --git a/vp10/encoder/speed_features.h b/vp10/encoder/speed_features.h
index fbb6988..02ee204 100644
--- a/vp10/encoder/speed_features.h
+++ b/vp10/encoder/speed_features.h
@@ -479,6 +479,9 @@
// Fast approximation of vp10_model_rd_from_var_lapndz
int simple_model_rd_from_var;
+ // Do sub-pixel search in up-sampled reference frames
+ int use_upsampled_references;
+
// Whether to compute distortion in the image domain (slower but
// more accurate), or in the transform domain (faster but less acurate).
int use_transform_domain_distortion;
diff --git a/vp10/encoder/temporal_filter.c b/vp10/encoder/temporal_filter.c
index 3e1246a..b3cf899 100644
--- a/vp10/encoder/temporal_filter.c
+++ b/vp10/encoder/temporal_filter.c
@@ -320,11 +320,7 @@
0, mv_sf->subpel_iters_per_step,
cond_cost_list(cpi, cost_list),
NULL, NULL,
-#if CONFIG_AFFINE_MOTION
&distortion, &sse, NULL, 0, 0, 0);
-#else
- &distortion, &sse, NULL, 0, 0);
-#endif
// Restore input state
x->plane[0].src = src;
diff --git a/vp10/encoder/tokenize.c b/vp10/encoder/tokenize.c
index 5cae8e3..c71c985 100644
--- a/vp10/encoder/tokenize.c
+++ b/vp10/encoder/tokenize.c
@@ -398,13 +398,18 @@
MACROBLOCK *const x = &td->mb;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
- uint8_t *color_map = xd->plane[0].color_index_map;
+ uint8_t *color_map = xd->plane[plane != 0].color_index_map;
PALETTE_MODE_INFO *pmi = &mbmi->palette_mode_info;
int n = pmi->palette_size[plane != 0];
int i, j, k;
int color_new_idx = -1, color_ctx, color_order[PALETTE_MAX_SIZE];
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
+ const int rows = (4 * num_4x4_blocks_high_lookup[bsize]) >>
+ (xd->plane[plane != 0].subsampling_y);
+ const int cols = (4 * num_4x4_blocks_wide_lookup[bsize]) >>
+ (xd->plane[plane != 0].subsampling_x);
+ const vpx_prob (* const probs)[PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] =
+ plane == 0 ? vp10_default_palette_y_color_prob :
+ vp10_default_palette_uv_color_prob;
for (i = 0; i < rows; ++i) {
for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
@@ -416,9 +421,8 @@
break;
}
assert(color_new_idx >= 0 && color_new_idx < n);
-
(*t)->token = color_new_idx;
- (*t)->context_tree = vp10_default_palette_y_color_prob[n - 2][color_ctx];
+ (*t)->context_tree = probs[n - 2][color_ctx];
(*t)->skip_eob_node = 0;
++(*t);
}
diff --git a/vpx_dsp/variance.c b/vpx_dsp/variance.c
index 169769a..ee1e305 100644
--- a/vpx_dsp/variance.c
+++ b/vpx_dsp/variance.c
@@ -267,7 +267,6 @@
}
}
-#if CONFIG_AFFINE_MOTION
// Get pred block from up-sampled reference.
void vpx_upsampled_pred_c(uint8_t *comp_pred,
int width, int height,
@@ -300,7 +299,6 @@
ref += stride;
}
}
-#endif
#if CONFIG_VP9_HIGHBITDEPTH
static void highbd_variance64(const uint8_t *a8, int a_stride,
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 583d9fa..e5c002a 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -1233,12 +1233,10 @@
#
# ...
#
-if (vpx_config("CONFIG_AFFINE_MOTION") eq "yes") {
- add_proto qw/void vpx_upsampled_pred/, "uint8_t *comp_pred, int width, int height, const uint8_t *ref, int ref_stride";
- specialize qw/vpx_upsampled_pred sse2/;
- add_proto qw/void vpx_comp_avg_upsampled_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride";
- specialize qw/vpx_comp_avg_upsampled_pred sse2/;
-}
+add_proto qw/void vpx_upsampled_pred/, "uint8_t *comp_pred, int width, int height, const uint8_t *ref, int ref_stride";
+specialize qw/vpx_upsampled_pred sse2/;
+add_proto qw/void vpx_comp_avg_upsampled_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride";
+specialize qw/vpx_comp_avg_upsampled_pred sse2/;
#
# ...
diff --git a/vpx_dsp/x86/variance_sse2.c b/vpx_dsp/x86/variance_sse2.c
index 7943c84..63fc1e6 100644
--- a/vpx_dsp/x86/variance_sse2.c
+++ b/vpx_dsp/x86/variance_sse2.c
@@ -476,7 +476,6 @@
#undef FN
#endif // CONFIG_USE_X86INC
-#if CONFIG_AFFINE_MOTION
void vpx_upsampled_pred_sse2(uint8_t *comp_pred,
int width, int height,
const uint8_t *ref, int ref_stride) {
@@ -703,4 +702,3 @@
}
}
}
-#endif
