Optimize memory in palette tile token buffer
This CL optimizes the memory usage in palette tile token buffer
by replacing color_map_cdf pointer in TokenExtra with the index
to color_ctx.
This CL is bit-exact with no impact on encode time.
Resolution Tile Avg Memory reduction
Single Multi
thread thread
640x360 2x1 7.11% 6.65%
832x480 2x1 8.99% 8.16%
1280x720 2x2 9.96% 9.75%
1920x1080 4x2 11.36% 10.72%
Memory measuring command:
$ command time -v ./aomenc ...
Change-Id: Ife166a9da2cfb2d7f9bed483109d87cc4c8de0ea
diff --git a/av1/encoder/bitstream.c b/av1/encoder/bitstream.c
index 6c29bd9..e7044fb 100644
--- a/av1/encoder/bitstream.c
+++ b/av1/encoder/bitstream.c
@@ -340,13 +340,16 @@
}
static AOM_INLINE void pack_map_tokens(aom_writer *w, const TokenExtra **tp,
- int n, int num) {
+ int n, int num, MapCdf map_pb_cdf) {
const TokenExtra *p = *tp;
+ const int palette_size_idx = n - PALETTE_MIN_SIZE;
write_uniform(w, n, p->token); // The first color index.
++p;
--num;
for (int i = 0; i < num; ++i) {
- aom_write_symbol(w, p->token, p->color_map_cdf, n);
+ assert((p->color_ctx >= 0) && (p->color_ctx < CDF_SIZE(PALETTE_COLORS)));
+ aom_cdf_prob *color_map_cdf = map_pb_cdf[palette_size_idx][p->color_ctx];
+ aom_write_symbol(w, p->token, color_map_cdf, n);
++p;
}
*tp = p;
@@ -1488,6 +1491,7 @@
const AV1_COMMON *cm = &cpi->common;
const CommonModeInfoParams *const mi_params = &cm->mi_params;
MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
+ FRAME_CONTEXT *tile_ctx = xd->tile_ctx;
const int grid_idx = mi_row * mi_params->mi_stride + mi_col;
xd->mi = mi_params->mi_grid_base + grid_idx;
cpi->td.mb.mbmi_ext_frame =
@@ -1526,7 +1530,9 @@
av1_get_block_dimensions(mbmi->bsize, plane, xd, NULL, NULL, &rows,
&cols);
assert(*tok < tok_end);
- pack_map_tokens(w, tok, palette_size_plane, rows * cols);
+ MapCdf map_pb_cdf = plane ? tile_ctx->palette_uv_color_index_cdf
+ : tile_ctx->palette_y_color_index_cdf;
+ pack_map_tokens(w, tok, palette_size_plane, rows * cols, map_pb_cdf);
}
}
diff --git a/av1/encoder/tokenize.c b/av1/encoder/tokenize.c
index aa8f7fd..43ac270 100644
--- a/av1/encoder/tokenize.c
+++ b/av1/encoder/tokenize.c
@@ -29,7 +29,7 @@
static int cost_and_tokenize_map(Av1ColorMapParam *param, TokenExtra **t,
int plane, int calc_rate, int allow_update_cdf,
- FRAME_COUNTS *counts, MapCdf map_pb_cdf) {
+ FRAME_COUNTS *counts) {
const uint8_t *const color_map = param->color_map;
MapCdf map_cdf = param->map_cdf;
ColorCost color_cost = param->color_cost;
@@ -54,7 +54,7 @@
this_rate += (*color_cost)[palette_size_idx][color_ctx][color_new_idx];
} else {
(*t)->token = color_new_idx;
- (*t)->color_map_cdf = map_pb_cdf[palette_size_idx][color_ctx];
+ (*t)->color_ctx = color_ctx;
++(*t);
if (allow_update_cdf)
update_cdf(map_cdf[palette_size_idx][color_ctx], color_new_idx, n);
@@ -107,10 +107,7 @@
assert(plane == 0 || plane == 1);
Av1ColorMapParam color_map_params;
get_color_map_params(x, plane, bsize, tx_size, type, &color_map_params);
- MapCdf map_pb_cdf = plane ? x->tile_pb_ctx->palette_uv_color_index_cdf
- : x->tile_pb_ctx->palette_y_color_index_cdf;
- return cost_and_tokenize_map(&color_map_params, NULL, plane, 1, 0, NULL,
- map_pb_cdf);
+ return cost_and_tokenize_map(&color_map_params, NULL, plane, 1, 0, NULL);
}
void av1_tokenize_color_map(const MACROBLOCK *const x, int plane,
@@ -122,12 +119,10 @@
get_color_map_params(x, plane, bsize, tx_size, type, &color_map_params);
// The first color index does not use context or entropy.
(*t)->token = color_map_params.color_map[0];
- (*t)->color_map_cdf = NULL;
+ (*t)->color_ctx = -1;
++(*t);
- MapCdf map_pb_cdf = plane ? x->tile_pb_ctx->palette_uv_color_index_cdf
- : x->tile_pb_ctx->palette_y_color_index_cdf;
cost_and_tokenize_map(&color_map_params, t, plane, 0, allow_update_cdf,
- counts, map_pb_cdf);
+ counts);
}
static void tokenize_vartx(ThreadData *td, TX_SIZE tx_size,
diff --git a/av1/encoder/tokenize.h b/av1/encoder/tokenize.h
index 85bec53..d5e1dfd 100644
--- a/av1/encoder/tokenize.h
+++ b/av1/encoder/tokenize.h
@@ -21,7 +21,7 @@
#endif
typedef struct {
- aom_cdf_prob *color_map_cdf;
+ int8_t color_ctx;
uint8_t token;
} TokenExtra;