Merge "Merge branch 'master' into nextgenv2" into nextgenv2
diff --git a/test/obmc_variance_test.cc b/test/obmc_variance_test.cc
index 40295f2..b7de9ed 100644
--- a/test/obmc_variance_test.cc
+++ b/test/obmc_variance_test.cc
@@ -122,7 +122,7 @@
////////////////////////////////////////////////////////////////////////////////
#if CONFIG_VP9_HIGHBITDEPTH
-class ObmcVarianceHBDTest : public FunctionEquivalenceTest<ObmcVarF> {}
+class ObmcVarianceHBDTest : public FunctionEquivalenceTest<ObmcVarF> {};
TEST_P(ObmcVarianceHBDTest, RandomValues) {
DECLARE_ALIGNED(32, uint16_t, pre[MAX_SB_SQUARE]);
@@ -133,19 +133,20 @@
const int pre_stride = this->rng_(MAX_SB_SIZE + 1);
for (int i = 0 ; i < MAX_SB_SQUARE ; ++i) {
- pre[i] = this->rng_(1 << this->bit_depth);
- wsrc[i] = this->rng_(1 << this->bit_depth) *
+ pre[i] = this->rng_(1 << params_.bit_depth);
+ wsrc[i] = this->rng_(1 << params_.bit_depth) *
this->rng_(kMaskMax * kMaskMax + 1);
mask[i] = this->rng_(kMaskMax * kMaskMax + 1);
}
unsigned int ref_sse, tst_sse;
- const unsigned int ref_res = ref_func_(CONVERT_TO_BYTEPTR(pre), pre_stride,
- wsrc, mask, &ref_sse);
+ const unsigned int ref_res = params_.ref_func(CONVERT_TO_BYTEPTR(pre),
+ pre_stride,
+ wsrc, mask, &ref_sse);
unsigned int tst_res;
ASM_REGISTER_STATE_CHECK(
- tst_res = tst_func_(CONVERT_TO_BYTEPTR(pre),
- pre_stride, wsrc, mask, &tst_sse));
+ tst_res = params_.tst_func(CONVERT_TO_BYTEPTR(pre),
+ pre_stride, wsrc, mask, &tst_sse));
ASSERT_EQ(ref_res, tst_res);
ASSERT_EQ(ref_sse, tst_sse);
@@ -161,18 +162,19 @@
const int pre_stride = iter;
for (int i = 0 ; i < MAX_SB_SQUARE ; ++i) {
- pre[i] = (1 << this->bit_depth) - 1;
- wsrc[i] = ((1 << this->bit_depth) - 1) * kMaskMax * kMaskMax;
+ pre[i] = (1 << params_.bit_depth) - 1;
+ wsrc[i] = ((1 << params_.bit_depth) - 1) * kMaskMax * kMaskMax;
mask[i] = kMaskMax * kMaskMax;
}
unsigned int ref_sse, tst_sse;
- const unsigned int ref_res = ref_func_(CONVERT_TO_BYTEPTR(pre), pre_stride,
- wsrc, mask, &ref_sse);
+ const unsigned int ref_res = params_.ref_func(CONVERT_TO_BYTEPTR(pre),
+ pre_stride,
+ wsrc, mask, &ref_sse);
unsigned int tst_res;
ASM_REGISTER_STATE_CHECK(
- tst_res = tst_func_(CONVERT_TO_BYTEPTR(pre), pre_stride,
- wsrc, mask, &tst_sse));
+ tst_res = params_.tst_func(CONVERT_TO_BYTEPTR(pre), pre_stride,
+ wsrc, mask, &tst_sse));
ASSERT_EQ(ref_res, tst_res);
ASSERT_EQ(ref_sse, tst_sse);
diff --git a/vp10/encoder/block.h b/vp10/encoder/block.h
index 4c9f8a5..c39aa21 100644
--- a/vp10/encoder/block.h
+++ b/vp10/encoder/block.h
@@ -140,9 +140,6 @@
int mv_row_min;
int mv_row_max;
- // Notes transform blocks where no coefficents are coded.
- // Set during mode selection. Read during block encoding.
- uint8_t zcoeff_blk[TX_SIZES][MAX_MIB_SIZE * MAX_MIB_SIZE * 4];
#if CONFIG_VAR_TX
uint8_t blk_skip[MAX_MB_PLANE][MAX_MIB_SIZE * MAX_MIB_SIZE * 4];
#if CONFIG_REF_MV
diff --git a/vp10/encoder/context_tree.c b/vp10/encoder/context_tree.c
index 41155c9..8555cdb 100644
--- a/vp10/encoder/context_tree.c
+++ b/vp10/encoder/context_tree.c
@@ -34,8 +34,6 @@
ctx->partition = partition;
#endif
- CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
- vpx_calloc(num_blk, sizeof(uint8_t)));
for (i = 0; i < MAX_MB_PLANE; ++i) {
#if CONFIG_VAR_TX
CHECK_MEM_ERROR(cm, ctx->blk_skip[i],
@@ -68,8 +66,6 @@
static void free_mode_context(PICK_MODE_CONTEXT *ctx) {
int i, k;
- vpx_free(ctx->zcoeff_blk);
- ctx->zcoeff_blk = 0;
for (i = 0; i < MAX_MB_PLANE; ++i) {
#if CONFIG_VAR_TX
vpx_free(ctx->blk_skip[i]);
diff --git a/vp10/encoder/context_tree.h b/vp10/encoder/context_tree.h
index 7b49354..a788788 100644
--- a/vp10/encoder/context_tree.h
+++ b/vp10/encoder/context_tree.h
@@ -26,7 +26,6 @@
typedef struct {
MODE_INFO mic;
MB_MODE_INFO_EXT mbmi_ext;
- uint8_t *zcoeff_blk;
uint8_t *color_index_map[2];
#if CONFIG_VAR_TX
uint8_t *blk_skip[MAX_MB_PLANE];
diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index ff1ee6b..287fb17 100644
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -1204,8 +1204,6 @@
memcpy(x->blk_skip[i], ctx->blk_skip[i],
sizeof(uint8_t) * ctx->num_4x4_blk);
#endif
- memcpy(x->zcoeff_blk[mbmi->tx_size], ctx->zcoeff_blk,
- sizeof(ctx->zcoeff_blk[0]) * ctx->num_4x4_blk);
if (!output_enabled)
return;
@@ -1362,8 +1360,6 @@
memcpy(x->blk_skip[i], ctx->blk_skip[i],
sizeof(uint8_t) * ctx->num_4x4_blk);
#endif // CONFIG_VAR_TX
- memcpy(x->zcoeff_blk[mbmi->tx_size], ctx->zcoeff_blk,
- sizeof(uint8_t) * ctx->num_4x4_blk);
#if CONFIG_VAR_TX
{
@@ -1569,8 +1565,6 @@
memcpy(ctx->blk_skip[i], x->blk_skip[i],
sizeof(uint8_t) * ctx->num_4x4_blk);
#endif // CONFIG_VAR_TX
- memcpy(ctx->zcoeff_blk, x->zcoeff_blk[supertx_size],
- sizeof(uint8_t) * ctx->num_4x4_blk);
ctx->mic.mbmi.tx_size = supertx_size;
ctx->skip = x->skip;
ctx->mic.mbmi.tx_type = best_tx;
@@ -6083,7 +6077,6 @@
#endif // CONFIG_EXT_TX
int tmp_rate_tx = 0, skip_tx = 0;
int64_t tmp_dist_tx = 0, rd_tx, bestrd_tx = INT64_MAX;
- uint8_t tmp_zcoeff_blk = 0;
set_skip_context(xd, mi_row, mi_col);
set_mode_info_offsets(cpi, x, xd, mi_row, mi_col);
@@ -6230,10 +6223,8 @@
tmp_rate_tx = *tmp_rate;
tmp_dist_tx = *tmp_dist;
skip_tx = x->skip;
- tmp_zcoeff_blk = x->zcoeff_blk[tx_size][0];
}
}
- x->zcoeff_blk[tx_size][0] = tmp_zcoeff_blk;
*tmp_rate = tmp_rate_tx;
*tmp_dist = tmp_dist_tx;
x->skip = skip_tx;
diff --git a/vp10/encoder/encodemb.c b/vp10/encoder/encodemb.c
index 3810be5..34428a7 100644
--- a/vp10/encoder/encodemb.c
+++ b/vp10/encoder/encodemb.c
@@ -982,16 +982,6 @@
ctx = combine_entropy_contexts(*a, *l);
#endif
- // TODO(jingning): per transformed block zero forcing only enabled for
- // luma component. will integrate chroma components as well.
- // Turn this back on when the rate-distortion loop is synchronized with
- // the recursive transform block coding.
-// if (x->zcoeff_blk[tx_size][block] && plane == 0) {
-// p->eobs[block] = 0;
-// *a = *l = 0;
-// return;
-// }
-
#if CONFIG_VAR_TX
// Assert not magic number (uninitialised).
assert(x->blk_skip[plane][(blk_row << bwl) + blk_col] != 234);
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 45c10cd..52dcf35 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -1166,22 +1166,13 @@
MACROBLOCK *const x = args->x;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-#if CONFIG_NEW_QUANT
- int ctx;
- struct macroblockd_plane *const pd = &xd->plane[plane];
-#endif // CONFIG_NEW_QUANT
int64_t rd1, rd2, rd;
int rate;
int64_t dist;
int64_t sse;
- ENTROPY_CONTEXT coeff_ctx = combine_entropy_contexts(
+ int coeff_ctx = combine_entropy_contexts(
*(args->t_above + blk_col), *(args->t_left + blk_row));
-#if CONFIG_NEW_QUANT
- ctx = get_entropy_context(tx_size, pd->above_context + blk_col,
- pd->left_context + blk_row);
-#endif // CONFIG_NEW_QUANT
-
if (args->exit_early)
return;
@@ -1228,7 +1219,7 @@
// full forward transform and quantization
#if CONFIG_NEW_QUANT
vp10_xform_quant_fp_nuq(x, plane, block, blk_row, blk_col, plane_bsize,
- tx_size, ctx);
+ tx_size, coeff_ctx);
#else
vp10_xform_quant(x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
VP10_XFORM_QUANT_FP);
@@ -1251,9 +1242,6 @@
// TODO(jingning): temporarily enabled only for luma component
rd = VPXMIN(rd1, rd2);
- if (plane == 0)
- x->zcoeff_blk[tx_size][block] = !x->plane[plane].eobs[block] ||
- (rd1 > rd2 && !xd->lossless[mbmi->segment_id]);
args->this_rate += rate;
args->this_dist += dist;
@@ -1453,7 +1441,6 @@
int64_t best_rd = INT64_MAX, last_rd = INT64_MAX;
const TX_SIZE max_tx_size = max_txsize_lookup[bs];
TX_SIZE best_tx = max_tx_size;
- uint8_t zcoeff_blk[TX_SIZES][MAX_MIB_SIZE * MAX_MIB_SIZE * 4];
const int tx_select = cm->tx_mode == TX_MODE_SELECT;
const int is_inter = is_inter_block(mbmi);
#if CONFIG_EXT_TX
@@ -1530,17 +1517,10 @@
*rate = r;
*skip = s;
*psse = sse;
- memcpy(zcoeff_blk[mbmi->tx_size], x->zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
}
}
mbmi->tx_size = best_tx;
- memcpy(x->zcoeff_blk[mbmi->tx_size], zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
-
return best_rd;
}
@@ -1730,7 +1710,6 @@
BLOCK_SIZE bs) {
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
- uint8_t zcoeff_blk[TX_SIZES][MAX_MIB_SIZE * MAX_MIB_SIZE * 4];
int r, s;
int64_t d, sse;
int64_t rd = INT64_MAX;
@@ -1765,9 +1744,6 @@
*psse = sse;
best_tx_type = tx_type;
best_tx = mbmi->tx_size;
- memcpy(zcoeff_blk[mbmi->tx_size], x->zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
}
}
@@ -1778,10 +1754,6 @@
if (mbmi->tx_size >= TX_32X32)
assert(mbmi->tx_type == DCT_DCT);
#endif
-
- memcpy(x->zcoeff_blk[mbmi->tx_size], zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
}
static void super_block_yrd(VP10_COMP *cpi, MACROBLOCK *x, int *rate,
@@ -2070,12 +2042,17 @@
if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type, 0);
-#if CONFIG_VAR_TX
+#if CONFIG_VAR_TX | CONFIG_NEW_QUANT
const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy));
-#endif // CONFIG_VAR_TX
+#endif // CONFIG_VAR_TX | CONFIG_NEW_QUANT
+#if CONFIG_NEW_QUANT
+ vp10_xform_quant_fp_nuq(x, 0, block, row + idy, col + idx,
+ BLOCK_8X8, TX_4X4, coeff_ctx);
+#else
vp10_xform_quant(x, 0, block, row + idy, col + idx, BLOCK_8X8,
TX_4X4, VP10_XFORM_QUANT_FP);
+#endif // CONFIG_NEW_QUANT
#if CONFIG_VAR_TX
ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
so->neighbors, cpi->sf.use_fast_coef_costing);
@@ -2099,8 +2076,13 @@
const scan_order *so = get_scan(TX_4X4, tx_type, 0);
const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy));
+#if CONFIG_NEW_QUANT
+ vp10_xform_quant_fp_nuq(x, 0, block, row + idy, col + idx,
+ BLOCK_8X8, TX_4X4, coeff_ctx);
+#else
vp10_xform_quant(x, 0, block, row + idy, col + idx, BLOCK_8X8,
TX_4X4, VP10_XFORM_QUANT_FP);
+#endif // CONFIG_NEW_QUANT
vp10_optimize_b(x, 0, block, TX_4X4, coeff_ctx);
#if CONFIG_VAR_TX
ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
@@ -2192,12 +2174,17 @@
if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block, TX_4X4);
const scan_order *so = get_scan(TX_4X4, tx_type, 0);
-#if CONFIG_VAR_TX
+#if CONFIG_VAR_TX | CONFIG_NEW_QUANT
const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy));
-#endif
+#endif // CONFIG_VAR_TX | CONFIG_NEW_QUANT
+#if CONFIG_NEW_QUANT
+ vp10_xform_quant_fp_nuq(x, 0, block, row + idy, col + idx, BLOCK_8X8,
+ TX_4X4, coeff_ctx);
+#else
vp10_xform_quant(x, 0, block, row + idy, col + idx, BLOCK_8X8,
TX_4X4, VP10_XFORM_QUANT_B);
+#endif // CONFIG_NEW_QUANT
#if CONFIG_VAR_TX
ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
so->neighbors, cpi->sf.use_fast_coef_costing);
@@ -2220,8 +2207,13 @@
const scan_order *so = get_scan(TX_4X4, tx_type, 0);
const int coeff_ctx = combine_entropy_contexts(*(tempa + idx),
*(templ + idy));
+#if CONFIG_NEW_QUANT
+ vp10_xform_quant_fp_nuq(x, 0, block, row + idy, col + idx, BLOCK_8X8,
+ TX_4X4, coeff_ctx);
+#else
vp10_xform_quant(x, 0, block, row + idy, col + idx, BLOCK_8X8,
TX_4X4, VP10_XFORM_QUANT_FP);
+#endif // CONFIG_NEW_QUANT
vp10_optimize_b(x, 0, block, TX_4X4, coeff_ctx);
#if CONFIG_VAR_TX
ratey += cost_coeffs(x, 0, block, coeff_ctx, TX_4X4, so->scan,
@@ -2444,9 +2436,7 @@
TX_TYPE *best_tx_type,
INTRA_FILTER *best_filter,
BLOCK_SIZE bsize, int rate_overhead,
- int64_t *best_rd,
- uint8_t zcoeff_blk[][MAX_MIB_SIZE *
- MAX_MIB_SIZE * 4]) {
+ int64_t *best_rd) {
int this_rate, this_rate_tokenonly, s;
int64_t this_distortion, this_rd;
MB_MODE_INFO *mbmi = &x->e_mbd.mi[0]->mbmi;
@@ -2468,9 +2458,6 @@
*rate_tokenonly = this_rate_tokenonly;
*distortion = this_distortion;
*skippable = s;
- memcpy(zcoeff_blk[mbmi->tx_size], x->zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
}
}
@@ -2490,7 +2477,6 @@
int64_t this_distortion, this_rd;
TX_SIZE best_tx_size = mic->mbmi.tx_size;
TX_TYPE best_tx_type = mbmi->tx_type;
- uint8_t zcoeff_blk[TX_SIZES][MAX_MIB_SIZE * MAX_MIB_SIZE * 4];
if (ANGLE_FAST_SEARCH) {
int deltas_level1[3] = {0, -2, 2};
@@ -2537,9 +2523,6 @@
*rate_tokenonly = this_rate_tokenonly;
*distortion = this_distortion;
*skippable = s;
- memcpy(zcoeff_blk[mbmi->tx_size], x->zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
}
}
}
@@ -2561,7 +2544,7 @@
rate_overhead +
cpi->intra_filter_cost
[intra_filter_ctx][filter],
- &best_rd, zcoeff_blk);
+ &best_rd);
}
}
}
@@ -2583,7 +2566,7 @@
rate_overhead +
cpi->intra_filter_cost
[intra_filter_ctx][filter],
- &best_rd, zcoeff_blk);
+ &best_rd);
}
}
}
@@ -2600,8 +2583,7 @@
&best_angle_delta, &best_tx_size,
&best_tx_type, &best_filter, bsize,
rate_overhead + cpi->intra_filter_cost
- [intra_filter_ctx][filter], &best_rd,
- zcoeff_blk);
+ [intra_filter_ctx][filter], &best_rd);
}
}
}
@@ -2610,11 +2592,6 @@
mbmi->angle_delta[0] = best_angle_delta;
mic->mbmi.intra_filter = best_filter;
mbmi->tx_type = best_tx_type;
- if (*rate_tokenonly < INT_MAX)
- memcpy(x->zcoeff_blk[mbmi->tx_size], zcoeff_blk[mbmi->tx_size],
- sizeof(zcoeff_blk[mbmi->tx_size][0]) *
- MAX_MIB_SIZE * MAX_MIB_SIZE * 4);
-
return best_rd;
}
@@ -4457,8 +4434,13 @@
k += (idy * 2 + idx);
coeff_ctx = combine_entropy_contexts(*(ta + (k & 1)),
*(tl + (k >> 1)));
+#if CONFIG_NEW_QUANT
+ vp10_xform_quant_fp_nuq(x, 0, k, idy + (i >> 1), idx + (i & 0x01),
+ BLOCK_8X8, TX_4X4, coeff_ctx);
+#else
vp10_xform_quant(x, 0, k, idy + (i >> 1), idx + (i & 0x01), BLOCK_8X8,
TX_4X4, VP10_XFORM_QUANT_FP);
+#endif // CONFIG_NEW_QUANT
if (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0)
vp10_optimize_b(x, 0, k, TX_4X4, coeff_ctx);
dist_block(cpi, x, 0, k, idy + (i >> 1), idx + (i & 0x1), TX_4X4,
@@ -8384,8 +8366,6 @@
*best_mbmode = *mbmi;
*best_skip2 = 0;
*best_mode_skippable = skippable;
- memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
- sizeof(ctx->zcoeff_blk[0]) * ctx->num_4x4_blk);
}
}
#endif // CONFIG_EXT_INTRA
@@ -9492,9 +9472,6 @@
for (i = 0; i < MAX_MB_PLANE; ++i)
memcpy(ctx->blk_skip[i], x->blk_skip[i],
sizeof(uint8_t) * ctx->num_4x4_blk);
-#else
- memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
- sizeof(ctx->zcoeff_blk[0]) * ctx->num_4x4_blk);
#endif
// TODO(debargha): enhance this test with a better distortion prediction
@@ -9756,8 +9733,6 @@
best_mbmode = *mbmi;
best_skip2 = 0;
best_mode_skippable = skippable;
- memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
- sizeof(ctx->zcoeff_blk[0]) * ctx->num_4x4_blk);
}
}
PALETTE_EXIT:
@@ -10220,7 +10195,6 @@
best_rd = best_rd_so_far;
best_yrd = best_rd_so_far;
#endif // CONFIG_SUPERTX
- memset(x->zcoeff_blk[TX_4X4], 0, 4);
vp10_zero(best_mbmode);
#if CONFIG_EXT_INTRA
@@ -10648,7 +10622,6 @@
tmp_best_mbmode = *mbmi;
for (i = 0; i < 4; i++) {
tmp_best_bmodes[i] = xd->mi[0]->bmi[i];
- x->zcoeff_blk[TX_4X4][i] = !x->plane[0].eobs[i];
}
pred_exists = 1;
}
@@ -10845,9 +10818,6 @@
#if CONFIG_VAR_TX
for (i = 0; i < MAX_MB_PLANE; ++i)
memset(ctx->blk_skip[i], 0, sizeof(uint8_t) * ctx->num_4x4_blk);
-#else
- memcpy(ctx->zcoeff_blk, x->zcoeff_blk[TX_4X4],
- sizeof(ctx->zcoeff_blk[0]) * ctx->num_4x4_blk);
#endif
for (i = 0; i < 4; i++)
