Masking intra mode choice adaptively

The commit changes to mask available intra prediction modes for test
based on prediction block size.

With this patch, encoding time of CpuUsed 2 reduces from 10% to 20% for
HD clips with a compression drop of 0.2%

Change-Id: I65f320f1237c0f5ae3a355bf7caf447f55625455

diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index a979a4b5eaf056d116967f769effef2585506b51..0833b4ac8b2bab8678c038af26d0c46248abffd4 100644 (file)
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -737,8 +737,10 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
   sf->mode_search_skip_flags = 0;
   sf->disable_split_var_thresh = 0;
   sf->disable_filter_search_var_thresh = 0;
-  sf->intra_y_mode_mask = ALL_INTRA_MODES;
-  sf->intra_uv_mode_mask = ALL_INTRA_MODES;
+  for (i = 0; i < TX_SIZES; i++) {
+    sf->intra_y_mode_mask[i] = ALL_INTRA_MODES;
+    sf->intra_uv_mode_mask[i] = ALL_INTRA_MODES;
+  }
   sf->use_rd_breakout = 0;
   sf->skip_encode_sb = 0;
   sf->use_uv_intra_rd_estimate = 0;
@@ -788,6 +790,9 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
         sf->auto_mv_step_size = 1;
         sf->adaptive_rd_thresh = 2;
         sf->recode_loop = 2;
+        sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
+        sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
+        sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
       }
       if (speed == 2) {
         sf->use_square_partition_only = !(cpi->common.frame_type == KEY_FRAME ||
@@ -823,6 +828,10 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
 
         sf->adaptive_rd_thresh = 2;
         sf->mode_skip_start = 11;
+        sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
+        sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V;
+        sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
+        sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
       }
       if (speed == 3) {
         sf->use_square_partition_only = 1;
@@ -923,8 +932,10 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
         sf->subpel_iters_per_step = 1;
         sf->disable_split_var_thresh = 64;
         sf->disable_filter_search_var_thresh = 96;
-        sf->intra_y_mode_mask = INTRA_DC_ONLY;
-        sf->intra_uv_mode_mask = INTRA_DC_ONLY;
+        for (i = 0; i < TX_SIZES; i++) {
+          sf->intra_y_mode_mask[i] = INTRA_DC_ONLY;
+          sf->intra_uv_mode_mask[i] = INTRA_DC_ONLY;
+        }
         sf->use_fast_coef_updates = 2;
         sf->adaptive_rd_thresh = 4;
         sf->mode_skip_start = 6;

diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h
index d706ec71bf48457883603627c6551900f3f4b2ea..f88ae8ad0322a9f7216d79fad844cf32526d4451 100644 (file)
--- a/vp9/encoder/vp9_onyx_int.h
+++ b/vp9/encoder/vp9_onyx_int.h
@@ -230,6 +230,7 @@ typedef enum {
 #define ALL_INTRA_MODES 0x3FF
 #define INTRA_DC_ONLY 0x01
 #define INTRA_DC_TM ((1 << TM_PRED) | (1 << DC_PRED))
+#define INTRA_DC_H_V ((1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED))
 #define INTRA_DC_TM_H_V (INTRA_DC_TM | (1 << V_PRED) | (1 << H_PRED))
 
 typedef enum {
@@ -285,8 +286,8 @@ typedef struct {
   // A source variance threshold below which filter search is disabled
   // Choose a very large value (UINT_MAX) to use 8-tap always
   unsigned int disable_filter_search_var_thresh;
-  int intra_y_mode_mask;
-  int intra_uv_mode_mask;
+  int intra_y_mode_mask[TX_SIZES];
+  int intra_uv_mode_mask[TX_SIZES];
   int use_rd_breakout;
   int use_uv_intra_rd_estimate;
   int use_fast_lpf_pick;

diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 60fec389d41a50c3259ad28ac9eaee4c2d590a29..a1be626233665516ca3f651e335d90427643fb9f 100644 (file)
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1044,7 +1044,7 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
     int64_t this_rd;
     int ratey = 0;
 
-    if (!(cpi->sf.intra_y_mode_mask & (1 << mode)))
+    if (!(cpi->sf.intra_y_mode_mask[TX_4X4] & (1 << mode)))
       continue;
 
     // Only do the oblique modes if the best so far is
@@ -1236,7 +1236,7 @@ static int64_t rd_pick_intra_sby_mode(VP9_COMP *cpi, MACROBLOCK *x,
     MODE_INFO *above_mi = xd->mi_8x8[-xd->mode_info_stride];
     MODE_INFO *left_mi = xd->mi_8x8[-1];
 
-    if (!(cpi->sf.intra_y_mode_mask & (1 << mode)))
+    if (!(cpi->sf.intra_y_mode_mask[max_txsize_lookup[bsize]] & (1 << mode)))
       continue;
 
     if (cpi->common.frame_type == KEY_FRAME) {
@@ -1339,9 +1339,10 @@ static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
   // int mode_mask = (bsize <= BLOCK_8X8)
   //                ? ALL_INTRA_MODES : cpi->sf.intra_uv_mode_mask;
 
-  for (mode = DC_PRED; mode <= TM_PRED; mode++) {
+  for (mode = DC_PRED; mode <= TM_PRED; mode ++) {
     // if (!(mode_mask & (1 << mode)))
-    if (!(cpi->sf.intra_uv_mode_mask & (1 << mode)))
+    if (!(cpi->sf.intra_uv_mode_mask[max_uv_txsize_lookup[bsize]]
+          & (1 << mode)))
       continue;
 
     x->e_mbd.mi_8x8[0]->mbmi.uv_mode = mode;