Change skip_txfm in RD_STATS to a uint8_t
x86 requires specifying the entire immediate value most operations. On a
full x86_64 executable, binary size is reduced by 0.38 KiB.
1) Reorder zero_rate in RD_STATS, so it can be initialized with rate
instead of with skip_txfm (on an instruction level). This is done to
prevent 2) from growing binary size.
2) Change RD_STATS skip_txfm to uint8_t.
3) Change code that depends on skip_txfm's type.
4) Change MB_MODE_INFO's skip_flag from int8_t to uint8_t in an attempt
to unify the type used.
No measurable impact on performance.
Change-Id: I3886f5f48b33d99348fc350a1acdf6e0c9394c32
diff --git a/av1/common/blockd.h b/av1/common/blockd.h
index 79849c4..cfcf1ac 100644
--- a/av1/common/blockd.h
+++ b/av1/common/blockd.h
@@ -188,6 +188,7 @@
typedef struct RD_STATS {
int rate;
+ int zero_rate;
int64_t dist;
// Please be careful of using rdcost, it's not guaranteed to be set all the
// time.
@@ -196,8 +197,7 @@
// rate/dist.
int64_t rdcost;
int64_t sse;
- int skip_txfm; // sse should equal to dist when skip_txfm == 1
- int zero_rate;
+ uint8_t skip_txfm; // sse should equal to dist when skip_txfm == 1
#if CONFIG_RD_DEBUG
int txb_coeff_cost[MAX_MB_PLANE];
#endif // CONFIG_RD_DEBUG
@@ -285,7 +285,7 @@
****************************************************************************/
/**@{*/
/*! \brief Whether to skip transforming and sending. */
- int8_t skip_txfm;
+ uint8_t skip_txfm;
/*! \brief Transform size when fixed size txfm is used (e.g. intra modes). */
TX_SIZE tx_size;
/*! \brief Transform size when recursive txfm tree is on. */
diff --git a/av1/encoder/block.h b/av1/encoder/block.h
index 76b1883..fe46aec 100644
--- a/av1/encoder/block.h
+++ b/av1/encoder/block.h
@@ -496,7 +496,7 @@
*/
typedef struct {
//! Whether to skip transform and quantization on a partition block level.
- int skip_txfm;
+ uint8_t skip_txfm;
/*! \brief Whether to skip transform and quantization on a txfm block level.
*
diff --git a/av1/encoder/compound_type.c b/av1/encoder/compound_type.c
index 4f762b9..39c505d 100644
--- a/av1/encoder/compound_type.c
+++ b/av1/encoder/compound_type.c
@@ -465,7 +465,8 @@
INTERINTRA_MODE *best_interintra_mode, int64_t *best_interintra_rd,
INTERINTRA_MODE interintra_mode, BLOCK_SIZE bsize) {
const AV1_COMMON *const cm = &cpi->common;
- int rate, skip_txfm_sb;
+ int rate;
+ uint8_t skip_txfm_sb;
int64_t dist, skip_sse_sb;
const int bw = block_size_wide[bsize];
mbmi->interintra_mode = interintra_mode;
@@ -688,7 +689,8 @@
const int_mv mv0 = mbmi->mv[0];
// Refine motion vector for NEWMV case.
if (have_newmv_in_inter_mode(mbmi->mode)) {
- int rate_sum, skip_txfm_sb;
+ int rate_sum;
+ uint8_t skip_txfm_sb;
int64_t dist_sum, skip_sse_sb;
// get negative of mask
const uint8_t *mask =
@@ -1048,7 +1050,8 @@
const COMPOUND_TYPE compound_type = mbmi->interinter_comp.type;
// This function will be called only for COMPOUND_WEDGE and COMPOUND_DIFFWTD
assert(compound_type == COMPOUND_WEDGE || compound_type == COMPOUND_DIFFWTD);
- int rate_sum, tmp_skip_txfm_sb;
+ int rate_sum;
+ uint8_t tmp_skip_txfm_sb;
int64_t dist_sum, tmp_skip_sse_sb;
pick_interinter_mask_type pick_interinter_mask[2] = { pick_interinter_wedge,
pick_interinter_seg };
@@ -1300,7 +1303,8 @@
// Case COMPOUND_AVERAGE and COMPOUND_DISTWTD
if (cur_type < COMPOUND_WEDGE) {
if (cpi->sf.inter_sf.enable_fast_compound_mode_search == 2) {
- int rate_sum, tmp_skip_txfm_sb;
+ int rate_sum;
+ uint8_t tmp_skip_txfm_sb;
int64_t dist_sum, tmp_skip_sse_sb;
// Reuse data if matching record is found
diff --git a/av1/encoder/encodemb.h b/av1/encoder/encodemb.h
index b58d13d..b819e82 100644
--- a/av1/encoder/encodemb.h
+++ b/av1/encoder/encodemb.h
@@ -56,7 +56,7 @@
const struct AV1_COMP *cpi;
MACROBLOCK *x;
struct optimize_ctx *ctx;
- int8_t *skip;
+ uint8_t *skip;
ENTROPY_CONTEXT *ta;
ENTROPY_CONTEXT *tl;
RUN_TYPE dry_run;
diff --git a/av1/encoder/intra_mode_search.c b/av1/encoder/intra_mode_search.c
index 113144c..d863910 100644
--- a/av1/encoder/intra_mode_search.c
+++ b/av1/encoder/intra_mode_search.c
@@ -228,7 +228,7 @@
*/
static int rd_pick_filter_intra_sby(const AV1_COMP *const cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
- int64_t *distortion, int *skippable,
+ int64_t *distortion, uint8_t *skippable,
BLOCK_SIZE bsize, int mode_cost,
PREDICTION_MODE best_mode_so_far,
int64_t *best_rd, int64_t *best_model_rd,
@@ -812,7 +812,7 @@
int64_t av1_rd_pick_intra_sbuv_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
- int64_t *distortion, int *skippable,
+ int64_t *distortion, uint8_t *skippable,
BLOCK_SIZE bsize, TX_SIZE max_tx_size) {
const AV1_COMMON *const cm = &cpi->common;
MACROBLOCKD *xd = &x->e_mbd;
@@ -1137,7 +1137,8 @@
BLOCK_SIZE bsize, const int *bmode_costs,
int64_t *best_rd, int *rate,
int *rate_tokenonly, int64_t *distortion,
- int *skippable, MB_MODE_INFO *best_mbmi,
+ uint8_t *skippable,
+ MB_MODE_INFO *best_mbmi,
PICK_MODE_CONTEXT *ctx) {
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = xd->mi[0];
@@ -1431,7 +1432,7 @@
// Finds the best non-intrabc mode on an intra frame.
int64_t av1_rd_pick_intra_sby_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
- int64_t *distortion, int *skippable,
+ int64_t *distortion, uint8_t *skippable,
BLOCK_SIZE bsize, int64_t best_rd,
PICK_MODE_CONTEXT *ctx) {
MACROBLOCKD *const xd = &x->e_mbd;
diff --git a/av1/encoder/intra_mode_search.h b/av1/encoder/intra_mode_search.h
index 7ee61d7..75289c4 100644
--- a/av1/encoder/intra_mode_search.h
+++ b/av1/encoder/intra_mode_search.h
@@ -62,7 +62,7 @@
int rate_uv_intra; /*!< \brief Total rate to transmit uv_mode */
int rate_uv_tokenonly; /*!< \brief Rate transmit txfm tokens */
int64_t dist_uvs; /*!< \brief Distortion of the uv_mode's recon */
- int skip_uvs; /*!< \brief Whether the uv txfm is skippable */
+ uint8_t skip_uvs; /*!< \brief Whether the uv txfm is skippable */
UV_PREDICTION_MODE mode_uv; /*!< \brief The best uv mode */
PALETTE_MODE_INFO pmi_uv; /*!< \brief Color map if mode_uv is palette */
int8_t uv_angle_delta; /*!< \brief Angle delta if mode_uv directional */
@@ -234,7 +234,7 @@
*/
int64_t av1_rd_pick_intra_sby_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
- int64_t *distortion, int *skippable,
+ int64_t *distortion, uint8_t *skippable,
BLOCK_SIZE bsize, int64_t best_rd,
PICK_MODE_CONTEXT *ctx);
@@ -269,7 +269,7 @@
*/
int64_t av1_rd_pick_intra_sbuv_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
int *rate, int *rate_tokenonly,
- int64_t *distortion, int *skippable,
+ int64_t *distortion, uint8_t *skippable,
BLOCK_SIZE bsize, TX_SIZE max_tx_size);
/*! \brief Return the number of colors in src. Used by palette mode.
diff --git a/av1/encoder/model_rd.h b/av1/encoder/model_rd.h
index db5ede4..f7e8b96 100644
--- a/av1/encoder/model_rd.h
+++ b/av1/encoder/model_rd.h
@@ -35,13 +35,11 @@
#define MODELRD_TYPE_INTRA 1
#define MODELRD_TYPE_MOTION_MODE_RD 1
-typedef void (*model_rd_for_sb_type)(const AV1_COMP *const cpi,
- BLOCK_SIZE bsize, MACROBLOCK *x,
- MACROBLOCKD *xd, int plane_from,
- int plane_to, int *out_rate_sum,
- int64_t *out_dist_sum, int *skip_txfm_sb,
- int64_t *skip_sse_sb, int *plane_rate,
- int64_t *plane_sse, int64_t *plane_dist);
+typedef void (*model_rd_for_sb_type)(
+ const AV1_COMP *const cpi, BLOCK_SIZE bsize, MACROBLOCK *x, MACROBLOCKD *xd,
+ int plane_from, int plane_to, int *out_rate_sum, int64_t *out_dist_sum,
+ uint8_t *skip_txfm_sb, int64_t *skip_sse_sb, int *plane_rate,
+ int64_t *plane_sse, int64_t *plane_dist);
typedef void (*model_rd_from_sse_type)(const AV1_COMP *const cpi,
const MACROBLOCK *const x,
BLOCK_SIZE plane_bsize, int plane,
@@ -160,7 +158,7 @@
static AOM_INLINE void model_rd_for_sb(
const AV1_COMP *const cpi, BLOCK_SIZE bsize, MACROBLOCK *x, MACROBLOCKD *xd,
int plane_from, int plane_to, int *out_rate_sum, int64_t *out_dist_sum,
- int *skip_txfm_sb, int64_t *skip_sse_sb, int *plane_rate,
+ uint8_t *skip_txfm_sb, int64_t *skip_sse_sb, int *plane_rate,
int64_t *plane_sse, int64_t *plane_dist) {
// Note our transform coeffs are 8 times an orthogonal transform.
// Hence quantizer step is also 8 times. To get effective quantizer
@@ -212,7 +210,7 @@
static AOM_INLINE void model_rd_for_sb_with_curvfit(
const AV1_COMP *const cpi, BLOCK_SIZE bsize, MACROBLOCK *x, MACROBLOCKD *xd,
int plane_from, int plane_to, int *out_rate_sum, int64_t *out_dist_sum,
- int *skip_txfm_sb, int64_t *skip_sse_sb, int *plane_rate,
+ uint8_t *skip_txfm_sb, int64_t *skip_sse_sb, int *plane_rate,
int64_t *plane_sse, int64_t *plane_dist) {
// Note our transform coeffs are 8 times an orthogonal transform.
// Hence quantizer step is also 8 times. To get effective quantizer
diff --git a/av1/encoder/palette.c b/av1/encoder/palette.c
index 3363336..4375175 100644
--- a/av1/encoder/palette.c
+++ b/av1/encoder/palette.c
@@ -219,7 +219,7 @@
BLOCK_SIZE bsize, int dc_mode_cost, const int *data, int *centroids, int n,
uint16_t *color_cache, int n_cache, bool do_header_rd_based_gating,
MB_MODE_INFO *best_mbmi, uint8_t *best_palette_color_map, int64_t *best_rd,
- int *rate, int *rate_tokenonly, int64_t *distortion, int *skippable,
+ int *rate, int *rate_tokenonly, int64_t *distortion, uint8_t *skippable,
int *beat_best_rd, PICK_MODE_CONTEXT *ctx, uint8_t *blk_skip,
uint8_t *tx_type_map, int *beat_best_palette_rd,
bool *do_header_rd_based_breakout, int discount_color_cost) {
@@ -328,7 +328,7 @@
int start_n, int end_n, int step_size, bool do_header_rd_based_gating,
int *last_n_searched, uint16_t *color_cache, int n_cache,
MB_MODE_INFO *best_mbmi, uint8_t *best_palette_color_map, int64_t *best_rd,
- int *rate, int *rate_tokenonly, int64_t *distortion, int *skippable,
+ int *rate, int *rate_tokenonly, int64_t *distortion, uint8_t *skippable,
int *beat_best_rd, PICK_MODE_CONTEXT *ctx, uint8_t *best_blk_skip,
uint8_t *tx_type_map, int discount_color_cost) {
int centroids[PALETTE_MAX_SIZE];
@@ -376,7 +376,7 @@
bool do_header_rd_based_gating, int *last_n_searched, uint16_t *color_cache,
int n_cache, MB_MODE_INFO *best_mbmi, uint8_t *best_palette_color_map,
int64_t *best_rd, int *rate, int *rate_tokenonly, int64_t *distortion,
- int *skippable, int *beat_best_rd, PICK_MODE_CONTEXT *ctx,
+ uint8_t *skippable, int *beat_best_rd, PICK_MODE_CONTEXT *ctx,
uint8_t *best_blk_skip, uint8_t *tx_type_map, uint8_t *color_map,
int data_points, int discount_color_cost) {
int centroids[PALETTE_MAX_SIZE];
@@ -527,7 +527,7 @@
void av1_rd_pick_palette_intra_sby(
const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize, int dc_mode_cost,
MB_MODE_INFO *best_mbmi, uint8_t *best_palette_color_map, int64_t *best_rd,
- int *rate, int *rate_tokenonly, int64_t *distortion, int *skippable,
+ int *rate, int *rate_tokenonly, int64_t *distortion, uint8_t *skippable,
int *beat_best_rd, PICK_MODE_CONTEXT *ctx, uint8_t *best_blk_skip,
uint8_t *tx_type_map) {
MACROBLOCKD *const xd = &x->e_mbd;
@@ -740,7 +740,7 @@
MB_MODE_INFO *const best_mbmi,
int64_t *best_rd, int *rate,
int *rate_tokenonly, int64_t *distortion,
- int *skippable) {
+ uint8_t *skippable) {
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = xd->mi[0];
assert(!is_inter_block(mbmi));
diff --git a/av1/encoder/palette.h b/av1/encoder/palette.h
index e219208..34d2ddc 100644
--- a/av1/encoder/palette.h
+++ b/av1/encoder/palette.h
@@ -186,7 +186,7 @@
const struct AV1_COMP *cpi, struct macroblock *x, BLOCK_SIZE bsize,
int dc_mode_cost, MB_MODE_INFO *best_mbmi, uint8_t *best_palette_color_map,
int64_t *best_rd, int *rate, int *rate_tokenonly, int64_t *distortion,
- int *skippable, int *beat_best_rd, struct PICK_MODE_CONTEXT *ctx,
+ uint8_t *skippable, int *beat_best_rd, struct PICK_MODE_CONTEXT *ctx,
uint8_t *best_blk_skip, uint8_t *tx_type_map);
/*!\brief Search for the best palette in the chroma plane.
@@ -201,7 +201,7 @@
MB_MODE_INFO *const best_mbmi,
int64_t *best_rd, int *rate,
int *rate_tokenonly, int64_t *distortion,
- int *skippable);
+ uint8_t *skippable);
/*!\brief Resets palette color map for chroma channels.
*/
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index 867ccf1..a1d5778 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -3295,7 +3295,7 @@
const int num_planes = av1_num_planes(cm);
TxfmSearchInfo *txfm_info = &x->txfm_search_info;
int rate_y = 0, rate_uv = 0, rate_y_tokenonly = 0, rate_uv_tokenonly = 0;
- int y_skip_txfm = 0, uv_skip_txfm = 0;
+ uint8_t y_skip_txfm = 0, uv_skip_txfm = 0;
int64_t dist_y = 0, dist_uv = 0;
ctx->rd_stats.skip_txfm = 0;
diff --git a/av1/encoder/tx_search.c b/av1/encoder/tx_search.c
index ea0288b..74c9de2 100644
--- a/av1/encoder/tx_search.c
+++ b/av1/encoder/tx_search.c
@@ -3473,7 +3473,7 @@
assert(level >= 0 && level <= 2);
int model_rate;
int64_t model_dist;
- int model_skip;
+ uint8_t model_skip;
MACROBLOCKD *const xd = &x->e_mbd;
model_rd_sb_fn[MODELRD_TYPE_TX_SEARCH_PRUNE](
cpi, bsize, x, xd, 0, 0, &model_rate, &model_dist, &model_skip, NULL,
