summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rwxr-xr-xtest/vpxenc.sh25
-rw-r--r--vp9/encoder/vp9_encodeframe.c23
-rw-r--r--vp9/encoder/vp9_encoder.c13
-rw-r--r--vp9/encoder/vp9_encoder.h9
-rw-r--r--vp9/encoder/vp9_firstpass.c10
-rw-r--r--vp9/encoder/vp9_firstpass.h1
6 files changed, 54 insertions, 27 deletions
diff --git a/test/vpxenc.sh b/test/vpxenc.sh
index e24c10672..f94e2e094 100755
--- a/test/vpxenc.sh
+++ b/test/vpxenc.sh
@@ -291,15 +291,14 @@ vpxenc_vp9_webm_rt_multithread_tiled() {
--threads=${threads} \
--tile-columns=${tile_cols} \
--output="${output}"
+
+ if [ ! -e "${output}" ]; then
+ elog "Output file does not exist."
+ return 1
+ fi
+ rm "${output}"
done
done
-
- if [ ! -e "${output}" ]; then
- elog "Output file does not exist."
- return 1
- fi
-
- rm "${output}"
fi
}
@@ -320,15 +319,13 @@ vpxenc_vp9_webm_rt_multithread_tiled_frameparallel() {
--tile-columns=${tile_cols} \
--frame-parallel=1 \
--output="${output}"
+ if [ ! -e "${output}" ]; then
+ elog "Output file does not exist."
+ return 1
+ fi
+ rm "${output}"
done
done
-
- if [ ! -e "${output}" ]; then
- elog "Output file does not exist."
- return 1
- fi
-
- rm "${output}"
fi
}
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index ba1d71bf7..0d727ab00 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -3183,7 +3183,7 @@ static int ml_pruning_partition(VP9_COMMON *const cm, MACROBLOCKD *const xd,
#define FEATURES 4
// ML-based partition search breakout.
-static int ml_predict_breakout(const VP9_COMP *const cpi, BLOCK_SIZE bsize,
+static int ml_predict_breakout(VP9_COMP *const cpi, BLOCK_SIZE bsize,
const MACROBLOCK *const x,
const RD_COST *const rd_cost) {
DECLARE_ALIGNED(16, static const uint8_t, vp9_64_zeros[64]) = { 0 };
@@ -3214,14 +3214,29 @@ static int ml_predict_breakout(const VP9_COMP *const cpi, BLOCK_SIZE bsize,
if (!linear_weights) return 0;
{ // Generate feature values.
+#if CONFIG_VP9_HIGHBITDEPTH
+ const int ac_q =
+ vp9_ac_quant(cm->base_qindex, 0, cm->bit_depth) >> (x->e_mbd.bd - 8);
+#else
const int ac_q = vp9_ac_quant(qindex, 0, cm->bit_depth);
+#endif // CONFIG_VP9_HIGHBITDEPTH
const int num_pels_log2 = num_pels_log2_lookup[bsize];
int feature_index = 0;
unsigned int var, sse;
float rate_f, dist_f;
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+ var =
+ vp9_high_get_sby_variance(cpi, &x->plane[0].src, bsize, x->e_mbd.bd);
+ } else {
+ var = cpi->fn_ptr[bsize].vf(x->plane[0].src.buf, x->plane[0].src.stride,
+ vp9_64_zeros, 0, &sse);
+ }
+#else
var = cpi->fn_ptr[bsize].vf(x->plane[0].src.buf, x->plane[0].src.stride,
vp9_64_zeros, 0, &sse);
+#endif
var = var >> num_pels_log2;
vpx_clear_system_state();
@@ -3822,13 +3837,9 @@ static void rd_pick_partition(VP9_COMP *cpi, ThreadData *td,
}
if ((do_split || do_rect) && !x->e_mbd.lossless && ctx->skippable) {
- int use_ml_based_breakout =
+ const int use_ml_based_breakout =
cpi->sf.use_ml_partition_search_breakout &&
cm->base_qindex >= 100;
-#if CONFIG_VP9_HIGHBITDEPTH
- if (x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
- use_ml_based_breakout = 0;
-#endif // CONFIG_VP9_HIGHBITDEPTH
if (use_ml_based_breakout) {
if (ml_predict_breakout(cpi, bsize, x, &this_rdc)) {
do_split = 0;
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 7b126ba1f..d9299f39e 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2359,11 +2359,10 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf,
vp9_set_speed_features_framesize_dependent(cpi);
if (cpi->sf.enable_tpl_model) {
+ const int mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
+ const int mi_rows = mi_cols_aligned_to_sb(cm->mi_rows);
// TODO(jingning): Reduce the actual memory use for tpl model build up.
for (frame = 0; frame < MAX_ARF_GOP_SIZE; ++frame) {
- int mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
- int mi_rows = mi_cols_aligned_to_sb(cm->mi_rows);
-
CHECK_MEM_ERROR(cm, cpi->tpl_stats[frame].tpl_stats_ptr,
vpx_calloc(mi_rows * mi_cols,
sizeof(*cpi->tpl_stats[frame].tpl_stats_ptr)));
@@ -2374,6 +2373,11 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf,
cpi->tpl_stats[frame].mi_rows = cm->mi_rows;
cpi->tpl_stats[frame].mi_cols = cm->mi_cols;
}
+
+ for (frame = 0; frame < REF_FRAMES; ++frame) {
+ cpi->enc_frame_buf[frame].mem_valid = 0;
+ cpi->enc_frame_buf[frame].released = 1;
+ }
}
// Allocate memory to store variances for a frame.
@@ -5412,8 +5416,9 @@ void init_gop_frames(VP9_COMP *cpi, GF_PICTURE *gf_picture,
// Initialize P frames
for (frame_idx = 2; frame_idx < MAX_LAG_BUFFERS; ++frame_idx) {
+ const int frame_gop_offset = gf_group->frame_gop_index[frame_idx];
struct lookahead_entry *buf =
- vp9_lookahead_peek(cpi->lookahead, frame_idx - 2);
+ vp9_lookahead_peek(cpi->lookahead, frame_gop_offset - 1);
if (buf == NULL) break;
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 45cc97315..79346ed09 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -495,6 +495,12 @@ typedef struct ARNRFilterData {
struct scale_factors sf;
} ARNRFilterData;
+typedef struct EncFrameBuf {
+ int mem_valid;
+ int released;
+ YV12_BUFFER_CONFIG frame;
+} EncFrameBuf;
+
// Maximum operating frame buffer size needed for a GOP using ARF reference.
#define MAX_ARF_GOP_SIZE (2 * MAX_LAG_BUFFERS)
@@ -522,7 +528,8 @@ typedef struct VP9_COMP {
YV12_BUFFER_CONFIG *raw_source_frame;
TplDepFrame tpl_stats[MAX_ARF_GOP_SIZE];
- YV12_BUFFER_CONFIG *tpl_recon_frames[REFS_PER_FRAME + 1];
+ YV12_BUFFER_CONFIG *tpl_recon_frames[REF_FRAMES];
+ EncFrameBuf enc_frame_buf[REF_FRAMES];
TileDataEnc *tile_data;
int allocated_tiles; // Keep track of memory allocated for tiles.
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 58c3a435d..318dd21b7 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -2124,6 +2124,7 @@ static void find_arf_order(VP9_COMP *cpi, GF_GROUP *gf_group,
for (idx = start; idx < end; ++idx) {
gf_group->update_type[*index_counter] = LF_UPDATE;
gf_group->arf_src_offset[*index_counter] = 0;
+ gf_group->frame_gop_index[*index_counter] = idx;
gf_group->rf_level[*index_counter] = INTER_NORMAL;
gf_group->layer_depth[*index_counter] = depth;
++(*index_counter);
@@ -2137,6 +2138,7 @@ static void find_arf_order(VP9_COMP *cpi, GF_GROUP *gf_group,
gf_group->layer_depth[*index_counter] = depth;
gf_group->update_type[*index_counter] = ARF_UPDATE;
gf_group->arf_src_offset[*index_counter] = mid - start;
+ gf_group->frame_gop_index[*index_counter] = mid;
gf_group->rf_level[*index_counter] = GF_ARF_LOW;
for (idx = 0; idx <= mid; ++idx)
@@ -2153,6 +2155,7 @@ static void find_arf_order(VP9_COMP *cpi, GF_GROUP *gf_group,
gf_group->update_type[*index_counter] = USE_BUF_FRAME;
gf_group->arf_src_offset[*index_counter] = 0;
+ gf_group->frame_gop_index[*index_counter] = mid;
gf_group->rf_level[*index_counter] = INTER_NORMAL;
gf_group->layer_depth[*index_counter] = depth;
++(*index_counter);
@@ -2203,17 +2206,18 @@ static int define_gf_group_structure(VP9_COMP *cpi) {
gf_group->layer_depth[frame_index] = 1;
gf_group->arf_src_offset[frame_index] =
(unsigned char)(rc->baseline_gf_interval - 1);
+ gf_group->frame_gop_index[frame_index] = rc->baseline_gf_interval;
++frame_index;
}
if (rc->source_alt_ref_pending && cpi->multi_layer_arf) {
- find_arf_order(cpi, gf_group, &frame_index, 2, 0,
- rc->baseline_gf_interval - 1);
+ find_arf_order(cpi, gf_group, &frame_index, 2, 1, rc->baseline_gf_interval);
set_gf_overlay_frame_type(gf_group, frame_index,
rc->source_alt_ref_pending);
gf_group->arf_src_offset[frame_index] = 0;
+ gf_group->frame_gop_index[frame_index] = rc->baseline_gf_interval;
return frame_index;
}
@@ -2227,6 +2231,7 @@ static int define_gf_group_structure(VP9_COMP *cpi) {
gf_group->update_type[frame_index] = LF_UPDATE;
gf_group->rf_level[frame_index] = INTER_NORMAL;
gf_group->arf_src_offset[frame_index] = 0;
+ gf_group->frame_gop_index[frame_index] = i + 1;
gf_group->layer_depth[frame_index] = MAX_ARF_LAYERS - 1;
++frame_index;
@@ -2247,6 +2252,7 @@ static int define_gf_group_structure(VP9_COMP *cpi) {
gf_group->rf_level[frame_index] = GF_ARF_STD;
}
gf_group->arf_src_offset[frame_index] = 0;
+ gf_group->frame_gop_index[frame_index] = rc->baseline_gf_interval;
return frame_index;
}
diff --git a/vp9/encoder/vp9_firstpass.h b/vp9/encoder/vp9_firstpass.h
index 9bd0a9e04..9d1e9355a 100644
--- a/vp9/encoder/vp9_firstpass.h
+++ b/vp9/encoder/vp9_firstpass.h
@@ -129,6 +129,7 @@ typedef struct {
FRAME_UPDATE_TYPE update_type[MAX_STATIC_GF_GROUP_LENGTH + 2];
unsigned char arf_src_offset[MAX_STATIC_GF_GROUP_LENGTH + 2];
unsigned char layer_depth[MAX_STATIC_GF_GROUP_LENGTH + 2];
+ unsigned char frame_gop_index[MAX_STATIC_GF_GROUP_LENGTH + 2];
int bit_allocation[MAX_STATIC_GF_GROUP_LENGTH + 2];
int gfu_boost[MAX_STATIC_GF_GROUP_LENGTH + 2];
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-12 11:23:09 +0000