[CFL] DC_PRED on Partition Unit Neighboring Pixels
To avoid a cascade of encodes when performing CfL RDO,
we compute DC_PRED on the partition unit. To do so, we
change the tx_size of CfL to match the size of the
partition unit (i.e. CfL partitions only contain 1
transform block).
This change requires disabling CfL when a chroma
partition-unit-sized DC_PRED is unavailable
(i.e. 4:1, 1:4 partitions and chroma partitions > 32X32).
Results on Subset1 (compared to disabling 4:1 and 1:4 partU):
PSNR | PSNR Cb | PSNR Cr | PSNR HVS | SSIM | MS SSIM | CIEDE 2000
-0.1243 | -1.9286 | -2.0140 | -0.1514 | -0.1512 | -0.1947 | -0.8066
https://two.arewecompressedyet.com/?job=master%402017-12-12T14%3A53%3A01.451Z&job=CfL-PartU%402017-12-12T15%3A39%3A36.794Z
Change-Id: I2a4adde79c10089130775b8e0df5f9c198855cad
diff --git a/av1/common/blockd.h b/av1/common/blockd.h
index 75297e2..24be7d9 100644
--- a/av1/common/blockd.h
+++ b/av1/common/blockd.h
@@ -1158,6 +1158,14 @@
static INLINE TX_SIZE av1_get_uv_tx_size(const MB_MODE_INFO *mbmi,
const struct macroblockd_plane *pd) {
+#if CONFIG_CFL
+ if (!is_inter_block(mbmi) && mbmi->uv_mode == UV_CFL_PRED) {
+ const BLOCK_SIZE plane_bsize =
+ AOMMAX(BLOCK_4X4, get_plane_block_size(mbmi->sb_type, pd));
+ assert(plane_bsize < BLOCK_SIZES_ALL);
+ return max_txsize_rect_lookup[0][plane_bsize];
+ }
+#endif
const TX_SIZE uv_txsize =
uv_txsize_lookup[mbmi->sb_type][mbmi->tx_size][pd->subsampling_x]
[pd->subsampling_y];
diff --git a/av1/common/cfl.c b/av1/common/cfl.c
index 95c61d9..29a7d0d 100644
--- a/av1/common/cfl.c
+++ b/av1/common/cfl.c
@@ -186,27 +186,27 @@
}
void cfl_predict_block(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
- int row, int col, TX_SIZE tx_size, int plane) {
+ TX_SIZE tx_size, int plane) {
CFL_CTX *const cfl = &xd->cfl;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+ assert(is_cfl_allowed(xd));
if (!cfl->are_parameters_computed) cfl_compute_parameters(xd, tx_size);
- const int16_t *pred_buf_q3 =
- cfl->pred_buf_q3 + ((row * MAX_SB_SIZE + col) << tx_size_wide_log2[0]);
const int alpha_q3 =
cfl_idx_to_alpha(mbmi->cfl_alpha_idx, mbmi->cfl_alpha_signs, plane - 1);
#if CONFIG_HIGHBITDEPTH
if (get_bitdepth_data_path_index(xd)) {
uint16_t *dst_16 = CONVERT_TO_SHORTPTR(dst);
- cfl_build_prediction_hbd(pred_buf_q3, dst_16, dst_stride,
+ cfl_build_prediction_hbd(cfl->pred_buf_q3, dst_16, dst_stride,
tx_size_wide[tx_size], tx_size_high[tx_size],
alpha_q3, xd->bd);
return;
}
#endif // CONFIG_HIGHBITDEPTH
- cfl_build_prediction_lbd(pred_buf_q3, dst, dst_stride, tx_size_wide[tx_size],
- tx_size_high[tx_size], alpha_q3);
+ cfl_build_prediction_lbd(cfl->pred_buf_q3, dst, dst_stride,
+ tx_size_wide[tx_size], tx_size_high[tx_size],
+ alpha_q3);
}
static void cfl_luma_subsampling_420_lbd(const uint8_t *input, int input_stride,
@@ -466,7 +466,7 @@
uint8_t *dst =
&pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
- assert(is_cfl_allowed(&xd->mi[0]->mbmi));
+ assert(is_cfl_allowed(xd));
if (block_size_high[bsize] == 4 || block_size_wide[bsize] == 4) {
// Only dimensions of size 4 can have an odd offset.
assert(!((col & 1) && tx_size_wide[tx_size] != 4));
@@ -487,7 +487,7 @@
int col = 0;
bsize = AOMMAX(BLOCK_4X4, bsize);
- assert(is_cfl_allowed(&xd->mi[0]->mbmi));
+ assert(is_cfl_allowed(xd));
if (block_size_high[bsize] == 4 || block_size_wide[bsize] == 4) {
sub8x8_adjust_offset(cfl, &row, &col);
#if CONFIG_DEBUG
diff --git a/av1/common/cfl.h b/av1/common/cfl.h
index 84e6b32..92fbbd0 100644
--- a/av1/common/cfl.h
+++ b/av1/common/cfl.h
@@ -14,19 +14,21 @@
#include "av1/common/blockd.h"
-static INLINE int is_cfl_allowed(const MB_MODE_INFO *mbmi) {
- const BLOCK_SIZE bsize = mbmi->sb_type;
- assert(bsize >= BLOCK_4X4); // Intra luma partitions can't be < 4X4
- assert(bsize < BLOCK_SIZES_ALL);
- return (bsize >= BLOCK_4X4) && (bsize < BLOCK_SIZES);
+static INLINE int is_cfl_allowed(const MACROBLOCKD *const xd) {
+ const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+ const BLOCK_SIZE plane_bsize = AOMMAX(
+ BLOCK_4X4, get_plane_block_size(mbmi->sb_type, &xd->plane[AOM_PLANE_U]));
+ assert(plane_bsize < BLOCK_SIZES_ALL);
+ return plane_bsize <= BLOCK_32X32;
}
+
static INLINE int get_scaled_luma_q0(int alpha_q3, int16_t pred_buf_q3) {
int scaled_luma_q6 = alpha_q3 * pred_buf_q3;
return ROUND_POWER_OF_TWO_SIGNED(scaled_luma_q6, 6);
}
void cfl_predict_block(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
- int row, int col, TX_SIZE tx_size, int plane);
+ TX_SIZE tx_size, int plane);
void cfl_store_block(MACROBLOCKD *const xd, BLOCK_SIZE bsize, TX_SIZE tx_size);
diff --git a/av1/common/reconintra.c b/av1/common/reconintra.c
index 0b41123..6ce57a8 100644
--- a/av1/common/reconintra.c
+++ b/av1/common/reconintra.c
@@ -2779,7 +2779,17 @@
#if CONFIG_CFL
if (plane != AOM_PLANE_Y && mbmi->uv_mode == UV_CFL_PRED) {
- cfl_predict_block(xd, dst, dst_stride, blk_row, blk_col, tx_size, plane);
+#if CONFIG_DEBUG
+ assert(blk_col == 0);
+ assert(blk_row == 0);
+ assert(is_cfl_allowed(xd));
+ const BLOCK_SIZE plane_bsize =
+ AOMMAX(BLOCK_4X4, get_plane_block_size(mbmi->sb_type, pd));
+ assert(plane_bsize < BLOCK_SIZES_ALL);
+ assert(block_size_wide[plane_bsize] == tx_size_wide[tx_size]);
+ assert(block_size_high[plane_bsize] == tx_size_high[tx_size]);
+#endif
+ cfl_predict_block(xd, dst, dst_stride, tx_size, plane);
}
#endif
}
diff --git a/av1/decoder/decodeframe.c b/av1/decoder/decodeframe.c
index 4e6ce01..47b3647 100644
--- a/av1/decoder/decodeframe.c
+++ b/av1/decoder/decodeframe.c
@@ -213,7 +213,7 @@
}
}
#if CONFIG_CFL
- if (plane == AOM_PLANE_Y && xd->cfl.store_y && is_cfl_allowed(mbmi)) {
+ if (plane == AOM_PLANE_Y && xd->cfl.store_y && is_cfl_allowed(xd)) {
cfl_store_tx(xd, row, col, tx_size, mbmi->sb_type);
}
#endif // CONFIG_CFL
@@ -584,7 +584,7 @@
#if CONFIG_CFL
if (mbmi->uv_mode != UV_CFL_PRED) {
if (!cfl->is_chroma_reference && is_inter_block(mbmi) &&
- is_cfl_allowed(mbmi)) {
+ is_cfl_allowed(xd)) {
cfl_store_block(xd, mbmi->sb_type, mbmi->tx_size);
}
}
diff --git a/av1/decoder/decodemv.c b/av1/decoder/decodemv.c
index 0928fc0..c1f60f0 100644
--- a/av1/decoder/decodemv.c
+++ b/av1/decoder/decodemv.c
@@ -1200,7 +1200,7 @@
#if CONFIG_CFL
if (mbmi->uv_mode == UV_CFL_PRED) {
- if (!is_cfl_allowed(mbmi)) {
+ if (!is_cfl_allowed(xd)) {
aom_internal_error(
&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
"Chroma from Luma (CfL) cannot be signaled for a %dx%d block.",
@@ -1560,7 +1560,7 @@
#if CONFIG_CFL
if (mbmi->uv_mode == UV_CFL_PRED) {
- if (!is_cfl_allowed(mbmi)) {
+ if (!is_cfl_allowed(xd)) {
aom_internal_error(
&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
"Chroma from Luma (CfL) cannot be signaled for a %dx%d block.",
diff --git a/av1/encoder/bitstream.c b/av1/encoder/bitstream.c
index 1c04554..75abce6 100644
--- a/av1/encoder/bitstream.c
+++ b/av1/encoder/bitstream.c
@@ -1424,7 +1424,7 @@
#if CONFIG_CFL
if (mbmi->uv_mode == UV_CFL_PRED) {
- if (!is_cfl_allowed(mbmi)) {
+ if (!is_cfl_allowed(xd)) {
aom_internal_error(
&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
"Chroma from Luma (CfL) cannot be signaled for a %dx%d block.",
@@ -1735,7 +1735,7 @@
#if CONFIG_CFL
if (mbmi->uv_mode == UV_CFL_PRED) {
- if (!is_cfl_allowed(mbmi)) {
+ if (!is_cfl_allowed(xd)) {
aom_internal_error(
&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
"Chroma from Luma (CfL) cannot be signaled for a %dx%d block.",
diff --git a/av1/encoder/encodeframe.c b/av1/encoder/encodeframe.c
index f1d516e..c0c7f31 100644
--- a/av1/encoder/encodeframe.c
+++ b/av1/encoder/encodeframe.c
@@ -4827,7 +4827,7 @@
if (is_inter_block(mbmi) &&
!is_chroma_reference(mi_row, mi_col, bsize, cfl->subsampling_x,
cfl->subsampling_y) &&
- is_cfl_allowed(mbmi)) {
+ is_cfl_allowed(xd)) {
cfl_store_block(xd, mbmi->sb_type, mbmi->tx_size);
}
#endif // CONFIG_CFL
diff --git a/av1/encoder/encodemb.c b/av1/encoder/encodemb.c
index 65bc6be..354dca8 100644
--- a/av1/encoder/encodemb.c
+++ b/av1/encoder/encodemb.c
@@ -972,8 +972,7 @@
if (*eob) *(args->skip) = 0;
#if CONFIG_CFL
- if (plane == AOM_PLANE_Y && xd->cfl.store_y &&
- is_cfl_allowed(&xd->mi[0]->mbmi)) {
+ if (plane == AOM_PLANE_Y && xd->cfl.store_y && is_cfl_allowed(xd)) {
cfl_store_tx(xd, blk_row, blk_col, tx_size, plane_bsize);
}
#endif // CONFIG_CFL
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index f4dad69..2917271 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -2142,7 +2142,7 @@
return;
}
#if CONFIG_CFL
- if (plane == AOM_PLANE_Y && xd->cfl.store_y && is_cfl_allowed(mbmi)) {
+ if (plane == AOM_PLANE_Y && xd->cfl.store_y && is_cfl_allowed(xd)) {
assert(!is_inter_block(mbmi) || plane_bsize < BLOCK_8X8);
cfl_store_tx(xd, blk_row, blk_col, tx_size, plane_bsize);
}
@@ -5427,12 +5427,19 @@
}
static int cfl_rd_pick_alpha(MACROBLOCK *const x, const AV1_COMP *const cpi,
- BLOCK_SIZE bsize, TX_SIZE tx_size,
- int64_t best_rd) {
+ TX_SIZE tx_size, int64_t best_rd) {
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
- bsize = scale_chroma_bsize(bsize, xd->plane[AOM_PLANE_U].subsampling_x,
- xd->plane[AOM_PLANE_U].subsampling_y);
+
+ const BLOCK_SIZE bsize = mbmi->sb_type;
+#if CONFIG_DEBUG
+ assert(is_cfl_allowed(xd));
+ const BLOCK_SIZE plane_bsize = AOMMAX(
+ BLOCK_4X4, get_plane_block_size(mbmi->sb_type, &xd->plane[AOM_PLANE_U]));
+ assert(plane_bsize < BLOCK_SIZES_ALL);
+ assert(block_size_wide[plane_bsize] == tx_size_wide[tx_size]);
+ assert(block_size_high[plane_bsize] == tx_size_high[tx_size]);
+#endif
int rates[CFL_PRED_PLANES][CFL_MAGS_SIZE];
int64_t dists[CFL_PRED_PLANES][CFL_MAGS_SIZE];
@@ -5546,10 +5553,11 @@
#if CONFIG_CFL
int cfl_alpha_rate = 0;
if (mode == UV_CFL_PRED) {
- if (!is_cfl_allowed(mbmi)) continue;
+ if (!is_cfl_allowed(xd)) continue;
assert(!is_directional_mode);
- const TX_SIZE uv_tx_size = av1_get_uv_tx_size(mbmi, &xd->plane[1]);
- cfl_alpha_rate = cfl_rd_pick_alpha(x, cpi, bsize, uv_tx_size, best_rd);
+ const TX_SIZE uv_tx_size =
+ av1_get_uv_tx_size(mbmi, &xd->plane[AOM_PLANE_U]);
+ cfl_alpha_rate = cfl_rd_pick_alpha(x, cpi, uv_tx_size, best_rd);
if (cfl_alpha_rate == INT_MAX) continue;
}
#endif
@@ -5578,7 +5586,7 @@
#if CONFIG_CFL
if (mode == UV_CFL_PRED) {
- assert(is_cfl_allowed(mbmi));
+ assert(is_cfl_allowed(xd));
this_rate += cfl_alpha_rate;
#if CONFIG_DEBUG
assert(xd->cfl.rate == this_rate);
