Merge scale_factors and scale_factors_uv.

This prevents a duplicate memcpy of a 128-byte struct every time
set_scale_factors() is called (which is a lot), thus leading to a
decrease from 3.7 MB to 1.85 MB of struct copying per 64x64 block
RD/partition loop.

Overall, this decreases encoding time of the first 50 frames of bus
@ 1500kbps (speed 0) from 1min5.9 to 1min4.9, i.e. about a 1.5%
overall speedup. We can likely get more gains by removing the copy
of the other struct (and replacing it with an indexing) as well.

Change-Id: I3dceb7e79f71e6fe911b11cc994cf89a869dde7a

diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 0795975..7ad3e87 100644 (file)
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -219,7 +219,6 @@ typedef struct macroblockd {
   struct macroblockd_plane plane[MAX_MB_PLANE];
 
   struct scale_factors scale_factor[2];
-  struct scale_factors scale_factor_uv[2];
 
   MODE_INFO *prev_mode_info_context;
   MODE_INFO *mode_info_context;

diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index c1b64d8..63e5646 100644 (file)
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -296,8 +296,7 @@ static void build_inter_predictors(int plane, int block,
     const int pre_stride = arg->pre_stride[which_mv][plane];
     const uint8_t *const pre = base_pre +
         scaled_buffer_offset(x, y, pre_stride, &xd->scale_factor[which_mv]);
-    struct scale_factors * const scale =
-      plane == 0 ? &xd->scale_factor[which_mv] : &xd->scale_factor_uv[which_mv];
+    struct scale_factors * const scale = &xd->scale_factor[which_mv];
 
     // dest
     uint8_t *const dst = arg->dst[plane] + arg->dst_stride[plane] * y + x;

diff --git a/vp9/common/vp9_reconinter.h b/vp9/common/vp9_reconinter.h
index 5b939ed..e37750d 100644 (file)
--- a/vp9/common/vp9_reconinter.h
+++ b/vp9/common/vp9_reconinter.h
@@ -83,8 +83,7 @@ static void setup_dst_planes(MACROBLOCKD *xd,
 static void setup_pre_planes(MACROBLOCKD *xd, int i,
                              const YV12_BUFFER_CONFIG *src,
                              int mi_row, int mi_col,
-                             const struct scale_factors *scale,
-                             const struct scale_factors *scale_uv) {
+                             const struct scale_factors *sf) {
   if (src) {
     int j;
     uint8_t* buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
@@ -94,7 +93,6 @@ static void setup_pre_planes(MACROBLOCKD *xd, int i,
 
     for (j = 0; j < MAX_MB_PLANE; ++j) {
       struct macroblockd_plane *pd = &xd->plane[j];
-      const struct scale_factors *sf = j ? scale_uv : scale;
       setup_pred_plane(&pd->pre[i], buffers[j], strides[j],
                      mi_row, mi_col, sf, pd->subsampling_x, pd->subsampling_y);
     }
@@ -103,8 +101,8 @@ static void setup_pre_planes(MACROBLOCKD *xd, int i,
 
 static void set_scale_factors(MACROBLOCKD *xd, int ref0, int ref1,
                               struct scale_factors sf[MAX_REF_FRAMES]) {
-  xd->scale_factor[0] = xd->scale_factor_uv[0] = sf[ref0 >= 0 ? ref0 : 0];
-  xd->scale_factor[1] = xd->scale_factor_uv[1] = sf[ref1 >= 0 ? ref1 : 0];
+  xd->scale_factor[0] = sf[ref0 >= 0 ? ref0 : 0];
+  xd->scale_factor[1] = sf[ref1 >= 0 ? ref1 : 0];
 }
 
 void vp9_setup_scale_factors(VP9_COMMON *cm, int i);

diff --git a/vp9/decoder/vp9_decodframe.c b/vp9/decoder/vp9_decodframe.c
index 118843a..27e67d2 100644 (file)
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -215,9 +215,7 @@ static void set_ref(VP9D_COMP *pbi, int i, int mi_row, int mi_col) {
 
   const YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[cm->active_ref_idx[ref]];
   xd->scale_factor[i] = cm->active_ref_scale[ref];
-  xd->scale_factor_uv[i] = cm->active_ref_scale[ref];
-  setup_pre_planes(xd, i, cfg, mi_row, mi_col,
-                   &xd->scale_factor[i], &xd->scale_factor_uv[i]);
+  setup_pre_planes(xd, i, cfg, mi_row, mi_col, &xd->scale_factor[i]);
   xd->corrupted |= cfg->corrupted;
 }
 
@@ -969,8 +967,7 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
   update_frame_context(&pc->fc);
 
   // Initialize xd pointers. Any reference should do for xd->pre, so use 0.
-  setup_pre_planes(xd, 0, &pc->yv12_fb[pc->active_ref_idx[0]], 0, 0,
-                   NULL, NULL);
+  setup_pre_planes(xd, 0, &pc->yv12_fb[pc->active_ref_idx[0]], 0, 0, NULL);
   setup_dst_planes(xd, new_fb, 0, 0);
 
   new_fb->corrupted |= read_compressed_header(pbi, data, first_partition_size);

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 5d9471e..3c6ece3 100644 (file)
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1078,9 +1078,9 @@ static void choose_partitioning(VP9_COMP *cpi, MODE_INFO *m, int mi_row,
     YV12_BUFFER_CONFIG *second_ref_fb = NULL;
 
     setup_pre_planes(xd, 0, ref_fb, mi_row, mi_col,
-                     &xd->scale_factor[0], &xd->scale_factor_uv[0]);
+                     &xd->scale_factor[0]);
     setup_pre_planes(xd, 1, second_ref_fb, mi_row, mi_col,
-                     &xd->scale_factor[1], &xd->scale_factor_uv[1]);
+                     &xd->scale_factor[1]);
     xd->mode_info_context->mbmi.ref_frame[0] = LAST_FRAME;
     xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_SB64X64;
     vp9_find_best_ref_mvs(xd, m->mbmi.ref_mvs[m->mbmi.ref_frame[0]],
@@ -1917,7 +1917,7 @@ static void init_encode_frame_mb_context(VP9_COMP *cpi) {
 
   // TODO(jkoleszar): are these initializations required?
   setup_pre_planes(xd, 0, &cm->yv12_fb[cm->ref_frame_map[cpi->lst_fb_idx]],
-                   0, 0, NULL, NULL);
+                   0, 0, NULL);
   setup_dst_planes(xd, &cm->yv12_fb[cm->new_fb_idx], 0, 0);
 
   setup_block_dptrs(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
@@ -2551,9 +2551,9 @@ static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t, int output_enabled,
     assert(cm->frame_type != KEY_FRAME);
 
     setup_pre_planes(xd, 0, ref_fb, mi_row, mi_col,
-                     &xd->scale_factor[0], &xd->scale_factor_uv[0]);
+                     &xd->scale_factor[0]);
     setup_pre_planes(xd, 1, second_ref_fb, mi_row, mi_col,
-                     &xd->scale_factor[1], &xd->scale_factor_uv[1]);
+                     &xd->scale_factor[1]);
 
 
     vp9_build_inter_predictors_sb(

diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index b5932eb..ec2e361 100644 (file)
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -501,7 +501,7 @@ void vp9_first_pass(VP9_COMP *cpi) {
   vp9_clear_system_state();  // __asm emms;
 
   vp9_setup_src_planes(x, cpi->Source, 0, 0);
-  setup_pre_planes(xd, 0, lst_yv12, 0, 0, NULL, NULL);
+  setup_pre_planes(xd, 0, lst_yv12, 0, 0, NULL);
   setup_dst_planes(xd, new_yv12, 0, 0);
 
   x->partition_info = x->pi;

diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index d11d1f8..aea8cf0 100644 (file)
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -2420,7 +2420,7 @@ static void single_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
     for (i = 0; i < MAX_MB_PLANE; i++)
       backup_yv12[i] = xd->plane[i].pre[0];
 
-    setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL, NULL);
+    setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
   }
 
   vp9_clamp_mv_min_max(x, &ref_mv);
@@ -2528,7 +2528,7 @@ static void joint_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
     // motion search code to be used without additional modifications.
     for (i = 0; i < MAX_MB_PLANE; i++)
       backup_yv12[i] = xd->plane[i].pre[0];
-    setup_pre_planes(xd, 0, scaled_ref_frame[0], mi_row, mi_col, NULL, NULL);
+    setup_pre_planes(xd, 0, scaled_ref_frame[0], mi_row, mi_col, NULL);
   }
 
   if (scaled_ref_frame[1]) {
@@ -2536,7 +2536,7 @@ static void joint_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
     for (i = 0; i < MAX_MB_PLANE; i++)
       backup_second_yv12[i] = xd->plane[i].pre[1];
 
-    setup_pre_planes(xd, 0, scaled_ref_frame[1], mi_row, mi_col, NULL, NULL);
+    setup_pre_planes(xd, 0, scaled_ref_frame[1], mi_row, mi_col, NULL);
   }
 
   xd->scale_factor[0].set_scaled_offsets(&xd->scale_factor[0],

diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index c1bd93b..821b7c6 100644 (file)
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -58,7 +58,7 @@ static void temporal_filter_predictors_mb_c(MACROBLOCKD *xd,
   vp9_build_inter_predictor(u_mb_ptr, stride,
                             &pred[256], 8,
                             &mv,
-                            &xd->scale_factor_uv[which_mv],
+                            &xd->scale_factor[which_mv],
                             8, 8,
                             which_mv,
                             &xd->subpix, MV_PRECISION_Q4);
@@ -66,7 +66,7 @@ static void temporal_filter_predictors_mb_c(MACROBLOCKD *xd,
   vp9_build_inter_predictor(v_mb_ptr, stride,
                             &pred[320], 8,
                             &mv,
-                            &xd->scale_factor_uv[which_mv],
+                            &xd->scale_factor[which_mv],
                             8, 8,
                             which_mv,
                             &xd->subpix, MV_PRECISION_Q4);
@@ -443,7 +443,6 @@ void vp9_temporal_filter_prepare(VP9_COMP *cpi, int distance) {
       cm->yv12_fb[cm->new_fb_idx].y_crop_width,
       cm->yv12_fb[cm->new_fb_idx].y_crop_height,
       cm->width, cm->height);
-  cpi->mb.e_mbd.scale_factor_uv[0] = cpi->mb.e_mbd.scale_factor[0];
 
   // Setup frame pointers, NULL indicates frame not included in filter
   vpx_memset(cpi->frames, 0, max_frames * sizeof(YV12_BUFFER_CONFIG *));