Perform reference remapping at fill_cache() time instead of in the

loop filter. This removes one obstacle of getting ff_h264_filter_mb_fast()
bitexact. code is maybe 0.1% faster

Originally committed as revision 21280 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/h264.c

@@ -2087,16 +2087,35 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     }
 
     for(j=0; j<2; j++){
+        int id_list[16];
         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
+        for(i=0; i<16; i++){
+            id_list[i]= 60;
+            if(h->ref_list[j][i].data[0]){
+                int k;
+                uint8_t *base= h->ref_list[j][i].base[0];
+                for(k=0; k<h->short_ref_count; k++)
+                    if(h->short_ref[k]->base[0] == base){
+                        id_list[i]= k;
+                        break;
+                    }
+                for(k=0; k<h->long_ref_count; k++)
+                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
+                        id_list[i]= h->short_ref_count + k;
+                        break;
+                    }
+            }
+        }
+
         ref2frm[0]=
         ref2frm[1]= -1;
         for(i=0; i<16; i++)
-            ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
+            ref2frm[i+2]= 4*id_list[i]
                           +(h->ref_list[j][i].reference&3);
         ref2frm[18+0]=
         ref2frm[18+1]= -1;
         for(i=16; i<48; i++)
-            ref2frm[i+4]= 4*h->ref_list[j][i].frame_num
+            ref2frm[i+4]= 4*id_list[(i-16)>>1]
                           +(h->ref_list[j][i].reference&3);
     }
 

libavcodec/h264.h

@@ -834,11 +834,20 @@ static av_always_inline int fill_caches(H264Context *h, int mb_type, int for_deb
                 }
 
                 ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
+                if(for_deblock){
+                    int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                    *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
+                    *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
+                    ref += h->b8_stride;
+                    *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
+                    *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
+                }else{
                 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
                 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
                 ref += h->b8_stride;
                 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
                 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
+                }
 
                 b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
                 for(y=0; y<4; y++){
@@ -1029,16 +1038,24 @@ static av_always_inline int fill_caches(H264Context *h, int mb_type, int for_deb
                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
+                if(for_deblock){
+                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
+                    h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
+                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
+                    h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
+                }else{
                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
+                }
             }else{
                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
-                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
+                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= (((for_deblock||top_type) ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
             }
 
             for(i=0; i<2; i++){
@@ -1048,13 +1065,19 @@ static av_always_inline int fill_caches(H264Context *h, int mb_type, int for_deb
                     const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
                     *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
                     *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
+                    if(for_deblock){
+                        int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[i]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                        h->ref_cache[list][cache_idx  ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)]];
+                        h->ref_cache[list][cache_idx+8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)]];
+                    }else{
                     h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
                     h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
+                    }
                 }else{
                     *(uint32_t*)h->mv_cache [list][cache_idx  ]=
                     *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
                     h->ref_cache[list][cache_idx  ]=
-                    h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
+                    h->ref_cache[list][cache_idx+8]= (for_deblock||left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
                 }
             }
 

libavcodec/h264_loopfilter.c

@@ -445,8 +445,6 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
     int edge;
     const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
     const int mbm_type = s->current_picture.mb_type[mbm_xy];
-    int (*ref2frm) [64] = h->ref2frm[ h->slice_num          &(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
-    int (*ref2frmm)[64] = h->ref2frm[ h->slice_table[mbm_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
     int start = h->slice_table[mbm_xy] == 0xFFFF ? 1 : 0;
 
     const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))
@@ -513,7 +511,6 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
         /* mbn_xy: neighbor macroblock */
         const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
         const int mbn_type = s->current_picture.mb_type[mbn_xy];
-        int (*ref2frmn)[64] = edge > 0 ? ref2frm : ref2frmm;
         int16_t bS[4];
         int qp;
 
@@ -553,7 +550,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
                 int v = 0;
 
                 for( l = 0; !v && l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
-                    v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] |
+                    v |= h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
                          h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
                          FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
                 }
@@ -562,7 +559,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
                     v=0;
                     for( l = 0; !v && l < 2; l++ ) {
                         int ln= 1-l;
-                        v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] |
+                        v |= h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
                             h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
                             FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit;
                     }
@@ -588,7 +585,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
                 {
                     bS[i] = 0;
                     for( l = 0; l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
-                        if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] |
+                        if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
                             h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
                             FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
                             bS[i] = 1;
@@ -600,7 +597,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
                         bS[i] = 0;
                         for( l = 0; l < 2; l++ ) {
                             int ln= 1-l;
-                            if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] |
+                            if( h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
                                 h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
                                 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) {
                                 bS[i] = 1;