Merge changes Ia2dd6bb1,Id1220b03 into nextgenv2
* changes:
transform tests: Avoid #if inside INSTANTIATE_TEST_CASE_P
variance_test: Avoid #if inside INSTANTIATE_TEST_CASE_P
diff --git a/vp10/common/blockd.h b/vp10/common/blockd.h
index b6eeee5..f7eb141 100644
--- a/vp10/common/blockd.h
+++ b/vp10/common/blockd.h
@@ -400,15 +400,44 @@
1, 16, 12, 2
};
static const int num_ext_tx_set_intra[EXT_TX_SETS_INTRA] = {
- 1, 12, 5
+ 1, 7, 5
};
#if EXT_TX_SIZES == 4
static INLINE int get_ext_tx_set(TX_SIZE tx_size, BLOCK_SIZE bs,
int is_inter) {
if (tx_size > TX_32X32 || bs < BLOCK_8X8) return 0;
+#if USE_REDUCED_TXSET_FOR_16X16
+ if (tx_size == TX_32X32)
+ return is_inter ? 3 - USE_MSKTX_FOR_32X32 : 0;
+ return (tx_size == TX_16X16 ? 2 : 1);
+#else
if (tx_size == TX_32X32)
return is_inter ? 3 - 2 * USE_MSKTX_FOR_32X32 : 0;
+ return (tx_size == TX_16X16 && !is_inter ? 2 : 1);
+#endif // USE_REDUCED_TXSET_FOR_16X16
+}
+
+static const int use_intra_ext_tx_for_txsize[EXT_TX_SETS_INTRA][TX_SIZES] = {
+ { 0, 0, 0, 0, }, // unused
+ { 1, 1, 0, 0, },
+ { 0, 0, 1, 0, },
+};
+
+static const int use_inter_ext_tx_for_txsize[EXT_TX_SETS_INTER][TX_SIZES] = {
+ { 0, 0, 0, 0, }, // unused
+ { 1, 1, (!USE_REDUCED_TXSET_FOR_16X16), USE_MSKTX_FOR_32X32, },
+ { 0, 0, USE_REDUCED_TXSET_FOR_16X16, 0, },
+ { 0, 0, 0, (!USE_MSKTX_FOR_32X32), },
+};
+
+#else // EXT_TX_SIZES == 4
+
+static INLINE int get_ext_tx_set(TX_SIZE tx_size, BLOCK_SIZE bs,
+ int is_inter) {
+ (void) is_inter;
+ if (tx_size > TX_32X32 || bs < BLOCK_8X8) return 0;
+ if (tx_size == TX_32X32) return 0;
#if USE_REDUCED_TXSET_FOR_16X16
return (tx_size == TX_16X16 ? 2 : 1);
#else
@@ -424,31 +453,8 @@
static const int use_inter_ext_tx_for_txsize[EXT_TX_SETS_INTER][TX_SIZES] = {
{ 0, 0, 0, 0, }, // unused
- { 1, 1, 1, USE_MSKTX_FOR_32X32, },
- { 0, 0, 0, 0, },
- { 0, 0, 0, (!USE_MSKTX_FOR_32X32), },
-};
-
-#else // EXT_TX_SIZES == 4
-
-static INLINE int get_ext_tx_set(TX_SIZE tx_size, BLOCK_SIZE bs,
- int is_inter) {
- (void) is_inter;
- if (tx_size > TX_32X32 || bs < BLOCK_8X8) return 0;
- if (tx_size == TX_32X32) return 0;
- return tx_size == TX_16X16 ? 2 : 1;
-}
-
-static const int use_intra_ext_tx_for_txsize[EXT_TX_SETS_INTRA][TX_SIZES] = {
- { 0, 0, 0, 0, }, // unused
- { 1, 1, 0, 0, },
- { 0, 0, 1, 0, },
-};
-
-static const int use_inter_ext_tx_for_txsize[EXT_TX_SETS_INTER][TX_SIZES] = {
- { 0, 0, 0, 0, }, // unused
- { 1, 1, 0, 0, },
- { 0, 0, 1, 0, },
+ { 1, 1, (!USE_REDUCED_TXSET_FOR_16X16), 0, },
+ { 0, 0, USE_REDUCED_TXSET_FOR_16X16, 0, },
{ 0, 0, 0, 1, },
};
#endif // EXT_TX_SIZES == 4
@@ -456,7 +462,7 @@
// Transform types used in each intra set
static const int ext_tx_used_intra[EXT_TX_SETS_INTRA][TX_TYPES] = {
{1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
- {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0},
+ {1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0},
{1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
};
@@ -479,13 +485,10 @@
#define ALLOW_FILTER_INTRA_MODES 1
#define ANGLE_STEP 3
#define MAX_ANGLE_DELTAS 3
-#define ANGLE_FAST_SEARCH 1
-#define ANGLE_SKIP_THRESH 0.10
-#define FILTER_FAST_SEARCH 1
extern const int16_t dr_intra_derivative[270][2];
-static uint8_t mode_to_angle_map[INTRA_MODES] = {
+static const uint8_t mode_to_angle_map[INTRA_MODES] = {
0, 90, 180, 45, 135, 111, 157, 203, 67, 0,
};
@@ -715,14 +718,7 @@
}
static INLINE int is_neighbor_overlappable(const MB_MODE_INFO *mbmi) {
-#if CONFIG_EXT_INTER
- return (is_inter_block(mbmi) &&
- !(has_second_ref(mbmi) && get_wedge_bits(mbmi->sb_type) &&
- mbmi->use_wedge_interinter) &&
- !(is_interintra_pred(mbmi)));
-#else
return (is_inter_block(mbmi));
-#endif // CONFIG_EXT_INTER
}
#endif // CONFIG_OBMC
diff --git a/vp10/common/entropymode.c b/vp10/common/entropymode.c
index f1c8e30..e622ebc 100644
--- a/vp10/common/entropymode.c
+++ b/vp10/common/entropymode.c
@@ -976,13 +976,8 @@
-DCT_DCT, 4,
6, 8,
-V_DCT, -H_DCT,
- 10, 16,
- 12, 14,
+ -ADST_ADST, 10,
-ADST_DCT, -DCT_ADST,
- -FLIPADST_DCT, -DCT_FLIPADST,
- 18, 20,
- -ADST_ADST, -FLIPADST_FLIPADST,
- -ADST_FLIPADST, -FLIPADST_ADST,
}, {
-IDTX, 2,
-DCT_DCT, 4,
@@ -1041,50 +1036,50 @@
},
}, {
{
- { 8, 176, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 10, 28, 32, 128, 176, 192, 208, 128, 128, 128, 128, },
- { 10, 28, 32, 128, 176, 192, 48, 128, 128, 128, 128, },
- { 9, 160, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 8, 28, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
- { 7, 28, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 7, 20, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 10, 23, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 8, 29, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 3, 20, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
+ { 8, 224, 32, 128, 64, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 9, 200, 32, 128, 64, 128, },
+ { 8, 8, 32, 128, 224, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 32, 32, 128, 16, 64, },
}, {
- { 2, 176, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 4, 28, 32, 128, 176, 192, 208, 128, 128, 128, 128, },
- { 4, 28, 32, 128, 176, 192, 48, 128, 128, 128, 128, },
- { 8, 160, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 2, 28, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
- { 3, 28, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 3, 26, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 9, 24, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 5, 24, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 2, 25, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
+ { 8, 224, 32, 128, 64, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 9, 200, 32, 128, 64, 128, },
+ { 8, 8, 32, 128, 224, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 32, 32, 128, 16, 64, },
}, {
- { 2, 176, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 1, 28, 32, 128, 176, 192, 208, 128, 128, 128, 128, },
- { 1, 28, 32, 128, 176, 192, 48, 128, 128, 128, 128, },
- { 4, 160, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 2, 28, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
- { 2, 28, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 3, 29, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 4, 27, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 2, 34, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 1, 25, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
+ { 8, 224, 32, 128, 64, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 9, 200, 32, 128, 64, 128, },
+ { 8, 8, 32, 128, 224, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 32, 32, 128, 16, 64, },
#if EXT_TX_SIZES == 4
}, {
- { 2, 176, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 1, 12, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 1, 17, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 4, 41, 32, 128, 128, 128, 128, 128, 128, 128, 128, },
- { 2, 17, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
- { 2, 14, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 3, 19, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 4, 27, 32, 128, 160, 176, 64, 128, 128, 128, 128, },
- { 2, 34, 32, 128, 160, 176, 192, 128, 128, 128, 128, },
- { 1, 15, 32, 128, 96, 128, 128, 128, 160, 192, 128, },
+ { 8, 224, 32, 128, 64, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 9, 200, 32, 128, 64, 128, },
+ { 8, 8, 32, 128, 224, 128, },
+ { 10, 32, 32, 128, 16, 192, },
+ { 10, 32, 32, 128, 16, 64, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 23, 32, 128, 80, 176, },
+ { 10, 32, 32, 128, 16, 64, },
#endif
},
}, {
diff --git a/vp10/common/reconinter.c b/vp10/common/reconinter.c
index 2d2563e..e5e2442 100644
--- a/vp10/common/reconinter.c
+++ b/vp10/common/reconinter.c
@@ -1402,6 +1402,19 @@
} // each mi in the left column
}
+#if CONFIG_EXT_INTER
+void modify_neighbor_predictor_for_obmc(MB_MODE_INFO *mbmi) {
+ if (is_interintra_pred(mbmi)) {
+ mbmi->ref_frame[1] = NONE;
+ } else if (has_second_ref(mbmi) && get_wedge_bits(mbmi->sb_type) &&
+ mbmi->use_wedge_interinter) {
+ mbmi->use_wedge_interinter = 0;
+ mbmi->ref_frame[1] = NONE;
+ }
+ return;
+}
+#endif // CONFIG_EXT_INTER
+
void vp10_build_prediction_by_above_preds(VP10_COMMON *cm,
MACROBLOCKD *xd,
int mi_row, int mi_col,
@@ -1420,6 +1433,9 @@
MODE_INFO *above_mi = xd->mi[mi_col_offset +
mi_row_offset * xd->mi_stride];
MB_MODE_INFO *above_mbmi = &above_mi->mbmi;
+#if CONFIG_EXT_INTER
+ MB_MODE_INFO backup_mbmi;
+#endif // CONFIG_EXT_INTER
mi_step = VPXMIN(xd->n8_w,
num_8x8_blocks_wide_lookup[above_mbmi->sb_type]);
@@ -1427,6 +1443,11 @@
if (!is_neighbor_overlappable(above_mbmi))
continue;
+#if CONFIG_EXT_INTER
+ backup_mbmi = *above_mbmi;
+ modify_neighbor_predictor_for_obmc(above_mbmi);
+#endif // CONFIG_EXT_INTER
+
for (j = 0; j < MAX_MB_PLANE; ++j) {
struct macroblockd_plane *const pd = &xd->plane[j];
setup_pred_plane(&pd->dst,
@@ -1488,6 +1509,9 @@
mi_x, mi_y);
}
}
+#if CONFIG_EXT_INTER
+ *above_mbmi = backup_mbmi;
+#endif // CONFIG_EXT_INTER
}
xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
}
@@ -1513,6 +1537,9 @@
mi_row_offset * xd->mi_stride];
MB_MODE_INFO *left_mbmi = &left_mi->mbmi;
const int is_compound = has_second_ref(left_mbmi);
+#if CONFIG_EXT_INTER
+ MB_MODE_INFO backup_mbmi;
+#endif // CONFIG_EXT_INTER
mi_step = VPXMIN(xd->n8_h,
num_8x8_blocks_high_lookup[left_mbmi->sb_type]);
@@ -1520,6 +1547,11 @@
if (!is_neighbor_overlappable(left_mbmi))
continue;
+#if CONFIG_EXT_INTER
+ backup_mbmi = *left_mbmi;
+ modify_neighbor_predictor_for_obmc(left_mbmi);
+#endif // CONFIG_EXT_INTER
+
for (j = 0; j < MAX_MB_PLANE; ++j) {
struct macroblockd_plane *const pd = &xd->plane[j];
setup_pred_plane(&pd->dst,
@@ -1581,6 +1613,9 @@
mi_x, mi_y);
}
}
+#if CONFIG_EXT_INTER
+ *left_mbmi = backup_mbmi;
+#endif // CONFIG_EXT_INTER
}
xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
}
diff --git a/vp10/common/x86/vp10_txfm1d_sse4.h b/vp10/common/x86/vp10_txfm1d_sse4.h
index 803b86d..86ab660 100644
--- a/vp10/common/x86/vp10_txfm1d_sse4.h
+++ b/vp10/common/x86/vp10_txfm1d_sse4.h
@@ -81,33 +81,32 @@
}
}
-#define round_shift_32_sse4_1(vec, bit) \
- ({ \
- __m128i tmp, round; \
- round = _mm_set1_epi32(1 << (bit - 1)); \
- tmp = _mm_add_epi32(vec, round); \
- _mm_srai_epi32(tmp, bit); \
- })
+static INLINE __m128i round_shift_32_sse4_1(__m128i vec, int bit) {
+ __m128i tmp, round;
+ round = _mm_set1_epi32(1 << (bit - 1));
+ tmp = _mm_add_epi32(vec, round);
+ return _mm_srai_epi32(tmp, bit);
+}
-#define round_shift_array_32_sse4_1(input, output, size, bit) \
- ({ \
- if (bit > 0) { \
- int i; \
- for (i = 0; i < size; i++) { \
- output[i] = round_shift_32_sse4_1(input[i], bit); \
- } \
- } else { \
- int i; \
- for (i = 0; i < size; i++) { \
- output[i] = _mm_slli_epi32(input[i], -bit); \
- } \
- } \
- })
+static INLINE void round_shift_array_32_sse4_1(__m128i *input, __m128i *output,
+ const int size, const int bit) {
+ if (bit > 0) {
+ int i;
+ for (i = 0; i < size; i++) {
+ output[i] = round_shift_32_sse4_1(input[i], bit);
+ }
+ } else {
+ int i;
+ for (i = 0; i < size; i++) {
+ output[i] = _mm_slli_epi32(input[i], -bit);
+ }
+ }
+}
// out0 = in0*w0 + in1*w1
// out1 = -in1*w0 + in0*w1
#define btf_32_sse4_1_type0(w0, w1, in0, in1, out0, out1, bit) \
- ({ \
+ do { \
__m128i ww0, ww1, in0_w0, in1_w1, in0_w1, in1_w0; \
ww0 = _mm_set1_epi32(w0); \
ww1 = _mm_set1_epi32(w1); \
@@ -119,12 +118,12 @@
in1_w0 = _mm_mullo_epi32(in1, ww0); \
out1 = _mm_sub_epi32(in0_w1, in1_w0); \
out1 = round_shift_32_sse4_1(out1, bit); \
- })
+ } while (0)
// out0 = in0*w0 + in1*w1
// out1 = in1*w0 - in0*w1
#define btf_32_sse4_1_type1(w0, w1, in0, in1, out0, out1, bit) \
- ({ \
+ do { \
__m128i ww0, ww1, in0_w0, in1_w1, in0_w1, in1_w0; \
ww0 = _mm_set1_epi32(w0); \
ww1 = _mm_set1_epi32(w1); \
@@ -136,7 +135,7 @@
in1_w0 = _mm_mullo_epi32(in1, ww0); \
out1 = _mm_sub_epi32(in1_w0, in0_w1); \
out1 = round_shift_32_sse4_1(out1, bit); \
- })
+ } while (0)
#ifdef __cplusplus
}
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index 49ad376..004f49c 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -28,7 +28,9 @@
#include "vp10/common/seg_common.h"
#include "vp10/common/tile_common.h"
+#if CONFIG_ANS
#include "vp10/encoder/buf_ans.h"
+#endif // CONFIG_ANS
#include "vp10/encoder/cost.h"
#include "vp10/encoder/bitstream.h"
#include "vp10/encoder/encodemv.h"
diff --git a/vp10/encoder/buf_ans.h b/vp10/encoder/buf_ans.h
index 11055d9..8697ee4 100644
--- a/vp10/encoder/buf_ans.h
+++ b/vp10/encoder/buf_ans.h
@@ -52,7 +52,7 @@
}
static INLINE void buf_uabs_write(struct BufAnsCoder *const c,
- uint8_t val, AnsP8 prob) {
+ uint8_t val, AnsP8 prob) {
assert(c->offset <= c->size);
if (c->offset == c->size) {
vp10_buf_ans_grow(c);
@@ -85,7 +85,8 @@
sym.cum_prob = c->buf[offset].val_start;
rans_write(ans, &sym);
} else {
- uabs_write(ans, c->buf[offset].val_start, c->buf[offset].prob);
+ uabs_write(ans, (uint8_t)c->buf[offset].val_start,
+ (AnsP8)c->buf[offset].prob);
}
}
}
diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index 437b366..21ed965 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -833,14 +833,18 @@
cm->tile_height = clamp(cpi->oxcf.tile_rows, 1, 32);
cm->tile_width <<= MAX_MIB_SIZE_LOG2;
cm->tile_height <<= MAX_MIB_SIZE_LOG2;
- } else
-#endif // CONFIG_EXT_PARTITION
- {
+ } else {
cm->tile_width = clamp(cpi->oxcf.tile_columns, 1, 64);
cm->tile_height = clamp(cpi->oxcf.tile_rows, 1, 64);
cm->tile_width <<= MAX_MIB_SIZE_LOG2 - 1;
cm->tile_height <<= MAX_MIB_SIZE_LOG2 - 1;
}
+#else
+ cm->tile_width = clamp(cpi->oxcf.tile_columns, 1, 64);
+ cm->tile_height = clamp(cpi->oxcf.tile_rows, 1, 64);
+ cm->tile_width <<= MAX_MIB_SIZE_LOG2;
+ cm->tile_height <<= MAX_MIB_SIZE_LOG2;
+#endif // CONFIG_EXT_PARTITION
cm->tile_width = VPXMIN(cm->tile_width, cm->mi_cols);
cm->tile_height = VPXMIN(cm->tile_height, cm->mi_rows);
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 0d2dada..d7a0f1a 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -81,6 +81,12 @@
#define MIN_EARLY_TERM_INDEX 3
#define NEW_MV_DISCOUNT_FACTOR 8
+#if CONFIG_EXT_INTRA
+#define ANGLE_FAST_SEARCH 1
+#define ANGLE_SKIP_THRESH 10
+#define FILTER_FAST_SEARCH 1
+#endif // CONFIG_EXT_INTRA
+
const double ADST_FLIP_SVM[8] = {-6.6623, -2.8062, -3.2531, 3.1671, // vert
-7.7051, -3.2234, -3.6193, 3.4533}; // horz
@@ -2531,99 +2537,134 @@
return best_rd;
}
-static INLINE int get_angle_index(double angle) {
- const double step = 22.5, base = 45;
- return (int)lround((angle - base) / step);
-}
+// Indices are sign, integer, and fractional part of the gradient value
+static const uint8_t gradient_to_angle_bin[2][7][16] = {
+ {
+ {6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 0, 0, 0, 0, },
+ {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, },
+ {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, },
+ {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, },
+ {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, },
+ {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, },
+ {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, },
+ },
+ {
+ {6, 6, 6, 6, 5, 5, 5, 5, 5, 5, 5, 5, 4, 4, 4, 4, },
+ {4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 3, 3, },
+ {3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, },
+ {3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, },
+ {3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, },
+ {3, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, },
+ {2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, },
+ },
+};
+
+static const uint8_t mode_to_angle_bin[INTRA_MODES] = {
+ 0, 2, 6, 0, 4, 3, 5, 7, 1, 0,
+};
static void angle_estimation(const uint8_t *src, int src_stride,
- int rows, int cols, double *hist) {
- int r, c, i, index;
- double angle, dx, dy;
- double temp, divisor;
+ int rows, int cols,
+ uint8_t *directional_mode_skip_mask) {
+ int i, r, c, index, dx, dy, temp, sn, remd, quot;
+ uint64_t hist[DIRECTIONAL_MODES];
+ uint64_t hist_sum = 0;
- vpx_clear_system_state();
- divisor = 0;
- for (i = 0; i < DIRECTIONAL_MODES; ++i)
- hist[i] = 0;
-
+ memset(hist, 0, DIRECTIONAL_MODES * sizeof(hist[0]));
src += src_stride;
for (r = 1; r < rows; ++r) {
for (c = 1; c < cols; ++c) {
dx = src[c] - src[c - 1];
dy = src[c] - src[c - src_stride];
temp = dx * dx + dy * dy;
- if (dy == 0)
- angle = 90;
- else
- angle = (atan((double)dx / (double)dy)) * 180 / PI;
- assert(angle >= -90 && angle <= 90);
- index = get_angle_index(angle + 180);
- if (index < DIRECTIONAL_MODES) {
- hist[index] += temp;
- divisor += temp;
+ if (dy == 0) {
+ index = 2;
+ } else {
+ sn = (dx > 0) ^ (dy > 0);
+ dx = abs(dx);
+ dy = abs(dy);
+ remd = dx % dy;
+ quot = dx / dy;
+ remd = remd * 16 / dy;
+ index = gradient_to_angle_bin[sn][VPXMIN(quot, 6)][VPXMIN(remd, 15)];
}
- if (angle > 0) {
- index = get_angle_index(angle);
- if (index >= 0) {
- hist[index] += temp;
- divisor += temp;
- }
- }
+ hist[index] += temp;
}
src += src_stride;
}
- if (divisor < 1)
- divisor = 1;
for (i = 0; i < DIRECTIONAL_MODES; ++i)
- hist[i] /= divisor;
+ hist_sum += hist[i];
+ for (i = 0; i < INTRA_MODES; ++i) {
+ if (i != DC_PRED && i != TM_PRED) {
+ int index = mode_to_angle_bin[i];
+ uint64_t score = 2 * hist[index];
+ int weight = 2;
+ if (index > 0) {
+ score += hist[index - 1];
+ weight += 1;
+ }
+ if (index < DIRECTIONAL_MODES - 1) {
+ score += hist[index + 1];
+ weight += 1;
+ }
+ if (score * ANGLE_SKIP_THRESH < hist_sum * weight)
+ directional_mode_skip_mask[i] = 1;
+ }
+ }
}
#if CONFIG_VP9_HIGHBITDEPTH
static void highbd_angle_estimation(const uint8_t *src8, int src_stride,
- int rows, int cols, double *hist) {
- int r, c, i, index;
- double angle, dx, dy;
- double temp, divisor;
+ int rows, int cols,
+ uint8_t *directional_mode_skip_mask) {
+ int i, r, c, index, dx, dy, temp, sn, remd, quot;
+ uint64_t hist[DIRECTIONAL_MODES];
+ uint64_t hist_sum = 0;
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- vpx_clear_system_state();
- divisor = 0;
- for (i = 0; i < DIRECTIONAL_MODES; ++i)
- hist[i] = 0;
-
+ memset(hist, 0, DIRECTIONAL_MODES * sizeof(hist[0]));
src += src_stride;
for (r = 1; r < rows; ++r) {
for (c = 1; c < cols; ++c) {
dx = src[c] - src[c - 1];
dy = src[c] - src[c - src_stride];
temp = dx * dx + dy * dy;
- if (dy == 0)
- angle = 90;
- else
- angle = (atan((double)dx / (double)dy)) * 180 / PI;
- assert(angle >= -90 && angle <= 90);
- index = get_angle_index(angle + 180);
- if (index < DIRECTIONAL_MODES) {
- hist[index] += temp;
- divisor += temp;
+ if (dy == 0) {
+ index = 2;
+ } else {
+ sn = (dx > 0) ^ (dy > 0);
+ dx = abs(dx);
+ dy = abs(dy);
+ remd = dx % dy;
+ quot = dx / dy;
+ remd = remd * 16 / dy;
+ index = gradient_to_angle_bin[sn][VPXMIN(quot, 6)][VPXMIN(remd, 15)];
}
- if (angle > 0) {
- index = get_angle_index(angle);
- if (index >= 0) {
- hist[index] += temp;
- divisor += temp;
- }
- }
+ hist[index] += temp;
}
src += src_stride;
}
- if (divisor < 1)
- divisor = 1;
for (i = 0; i < DIRECTIONAL_MODES; ++i)
- hist[i] /= divisor;
+ hist_sum += hist[i];
+ for (i = 0; i < INTRA_MODES; ++i) {
+ if (i != DC_PRED && i != TM_PRED) {
+ int index = mode_to_angle_bin[i];
+ uint64_t score = 2 * hist[index];
+ int weight = 2;
+ if (index > 0) {
+ score += hist[index - 1];
+ weight += 1;
+ }
+ if (index < DIRECTIONAL_MODES - 1) {
+ score += hist[index + 1];
+ weight += 1;
+ }
+ if (score * ANGLE_SKIP_THRESH < hist_sum * weight)
+ directional_mode_skip_mask[i] = 1;
+ }
+ }
}
#endif // CONFIG_VP9_HIGHBITDEPTH
#endif // CONFIG_EXT_INTRA
@@ -2649,7 +2690,6 @@
uint8_t directional_mode_skip_mask[INTRA_MODES];
const int src_stride = x->plane[0].src.stride;
const uint8_t *src = x->plane[0].src.buf;
- double hist[DIRECTIONAL_MODES];
#endif // CONFIG_EXT_INTRA
TX_TYPE best_tx_type = DCT_DCT;
int *bmode_costs;
@@ -2675,29 +2715,11 @@
sizeof(directional_mode_skip_mask[0]) * INTRA_MODES);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
- highbd_angle_estimation(src, src_stride, rows, cols, hist);
+ highbd_angle_estimation(src, src_stride, rows, cols,
+ directional_mode_skip_mask);
else
#endif
- angle_estimation(src, src_stride, rows, cols, hist);
-
- for (mode = 0; mode < INTRA_MODES; ++mode) {
- if (mode != DC_PRED && mode != TM_PRED) {
- int index = get_angle_index((double)mode_to_angle_map[mode]);
- double score, weight = 1.0;
- score = hist[index];
- if (index > 0) {
- score += hist[index - 1] * 0.5;
- weight += 0.5;
- }
- if (index < DIRECTIONAL_MODES - 1) {
- score += hist[index + 1] * 0.5;
- weight += 0.5;
- }
- score /= weight;
- if (score < ANGLE_SKIP_THRESH)
- directional_mode_skip_mask[mode] = 1;
- }
- }
+ angle_estimation(src, src_stride, rows, cols, directional_mode_skip_mask);
#endif // CONFIG_EXT_INTRA
memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
palette_mode_info.palette_size[0] = 0;
@@ -6387,10 +6409,16 @@
joint_motion_search(cpi, x, bsize, frame_mv,
mi_row, mi_col, NULL, single_newmv, &rate_mv, 0);
} else {
+#if CONFIG_REF_MV
+ vp10_set_mvcost(x, mbmi->ref_frame[0]);
+#endif // CONFIG_REF_MV
rate_mv = vp10_mv_bit_cost(&frame_mv[refs[0]].as_mv,
&x->mbmi_ext->ref_mvs[refs[0]][0].as_mv,
x->nmvjointcost, x->mvcost,
MV_COST_WEIGHT);
+#if CONFIG_REF_MV
+ vp10_set_mvcost(x, mbmi->ref_frame[1]);
+#endif // CONFIG_REF_MV
rate_mv += vp10_mv_bit_cost(&frame_mv[refs[1]].as_mv,
&x->mbmi_ext->ref_mvs[refs[1]][0].as_mv,
x->nmvjointcost, x->mvcost,
@@ -6417,9 +6445,15 @@
mi_row, mi_col,
single_newmv, &rate_mv, 0);
} else {
+#if CONFIG_REF_MV
+ vp10_set_mvcost(x, mbmi->ref_frame[0]);
+#endif // CONFIG_REF_MV
rate_mv = vp10_mv_bit_cost(&frame_mv[refs[0]].as_mv,
&x->mbmi_ext->ref_mvs[refs[0]][0].as_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
+#if CONFIG_REF_MV
+ vp10_set_mvcost(x, mbmi->ref_frame[1]);
+#endif // CONFIG_REF_MV
rate_mv += vp10_mv_bit_cost(&frame_mv[refs[1]].as_mv,
&x->mbmi_ext->ref_mvs[refs[1]][0].as_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
@@ -8138,33 +8172,14 @@
const uint8_t *src = x->plane[0].src.buf;
const int rows = 4 * num_4x4_blocks_high_lookup[bsize];
const int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
- double hist[DIRECTIONAL_MODES];
- PREDICTION_MODE mode;
-
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
- highbd_angle_estimation(src, src_stride, rows, cols, hist);
+ highbd_angle_estimation(src, src_stride, rows, cols,
+ directional_mode_skip_mask);
else
#endif
- angle_estimation(src, src_stride, rows, cols, hist);
- for (mode = 0; mode < INTRA_MODES; ++mode) {
- if (mode != DC_PRED && mode != TM_PRED) {
- int index = get_angle_index((double)mode_to_angle_map[mode]);
- double score, weight = 1.0;
- score = hist[index];
- if (index > 0) {
- score += hist[index - 1] * 0.5;
- weight += 0.5;
- }
- if (index < DIRECTIONAL_MODES - 1) {
- score += hist[index + 1] * 0.5;
- weight += 0.5;
- }
- score /= weight;
- if (score < ANGLE_SKIP_THRESH)
- directional_mode_skip_mask[mode] = 1;
- }
- }
+ angle_estimation(src, src_stride, rows, cols,
+ directional_mode_skip_mask);
angle_stats_ready = 1;
}
if (directional_mode_skip_mask[mbmi->mode])
@@ -8433,8 +8448,10 @@
if (!mv_check_bounds(x, &cur_mv.as_mv)) {
INTERP_FILTER dummy_single_inter_filter[MB_MODE_COUNT]
- [MAX_REF_FRAMES];
- int dummy_single_skippable[MB_MODE_COUNT][MAX_REF_FRAMES];
+ [MAX_REF_FRAMES] =
+ { { 0 } };
+ int dummy_single_skippable[MB_MODE_COUNT][MAX_REF_FRAMES] =
+ { { 0 } };
int dummy_disable_skip = 0;
#if CONFIG_EXT_INTER
int_mv dummy_single_newmvs[2][MAX_REF_FRAMES] =
diff --git a/vp10/encoder/speed_features.c b/vp10/encoder/speed_features.c
index cd1c91a..b3304a7 100644
--- a/vp10/encoder/speed_features.c
+++ b/vp10/encoder/speed_features.c
@@ -551,6 +551,12 @@
else if (oxcf->mode == GOOD)
set_good_speed_feature(cpi, cm, sf, oxcf->speed);
+#if CONFIG_REF_MV
+ // TODO(geza): Temporarily turn this off for ref-mv to fix tests.
+ // Investigate/reimplement skip_recode better to enable this.
+ sf->allow_skip_recode = 0;
+#endif // CONFIG_REF_MV
+
// sf->partition_search_breakout_dist_thr is set assuming max 64x64
// blocks. Normalise this if the blocks are bigger.
if (MAX_SB_SIZE_LOG2 > 6) {
