ffmpeg roll of source to mar 9 version...

patched-ffmpeg-mt/libavcodec/h264_loopfilter.c

 /*
  * H.26L/H.264/AVC/JVT/14496-10/... loop filter
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  *
  * This file is part of FFmpeg.
  *
  * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
  * FFmpeg is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
  * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 /**
  * @file libavcodec/h264_loopfilter.c
  * H.264 / AVC / MPEG4 part10 loop filter.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
+#include "libavutil/intreadwrite.h"
 #include "internal.h"
 #include "dsputil.h"
 #include "avcodec.h"
 #include "mpegvideo.h"
 #include "h264.h"
 #include "mathops.h"
 #include "rectangle.h"
 
 //#undef NDEBUG
 #include <assert.h>
@@ skipping 322 matching lines @@
             filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
         }
         filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
         filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
         filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
         filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
         filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
         filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
         return;
     } else {
-        DECLARE_ALIGNED_8(int16_t, bS)[2][4][4];
-        uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
+        LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
         int edges;
         if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
             edges = 4;
-            bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
+            AV_WN64A(bS[0][0], 0x0002000200020002ULL);
+            AV_WN64A(bS[0][2], 0x0002000200020002ULL);
+            AV_WN64A(bS[1][0], 0x0002000200020002ULL);
+            AV_WN64A(bS[1][2], 0x0002000200020002ULL);
         } else {
             int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0;
             int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0;
             int step =  1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
             edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
             s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
                                               h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
         }
         if( IS_INTRA(left_type) )
-            bSv[0][0] = 0x0004000400040004ULL;
+            AV_WN64A(bS[0][0], 0x0004000400040004ULL);
         if( IS_INTRA(h->top_type) )
-            bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;
+            AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
 
 #define FILTER(hv,dir,edge)\
-        if(bSv[dir][edge]) {\
+        if(AV_RN64A(bS[dir][edge])) {                                   \
             filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
             if(!(edge&1)) {\
                 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
                 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
             }\
         }
         if(left_type)
             FILTER(v,0,0);
         if( edges == 1 ) {
             FILTER(h,1,0);
@@ skipping 66 matching lines @@
             // This is a special case in the norm where the filtering must
             // be done twice (one each of the field) even if we are in a
             // frame macroblock.
             //
             unsigned int tmp_linesize   = 2 *   linesize;
             unsigned int tmp_uvlinesize = 2 * uvlinesize;
             int mbn_xy = mb_xy - 2 * s->mb_stride;
             int j;
 
             for(j=0; j<2; j++, mbn_xy += s->mb_stride){
-                DECLARE_ALIGNED_8(int16_t, bS)[4];
+                DECLARE_ALIGNED(8, int16_t, bS)[4];
                 int qp;
                 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
-                    *(uint64_t*)bS= 0x0003000300030003ULL;
+                    AV_WN64A(bS, 0x0003000300030003ULL);
                 } else {
-                    const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8; //FIXME  8x8dct?
+                    if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
+                        bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]);
+                        bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]);
+                        bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]);
+                        bS[3]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+3]);
+                    }else{
+                    const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8;
                     int i;
                     for( i = 0; i < 4; i++ ) {
                         bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
                     }
+                    }
                 }
                 // Do not use s->qscale as luma quantizer because it has not the same
                 // value in IPCM macroblocks.
                 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
                 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
                 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
                 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
                 filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
                                 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
                 filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
                                 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
             }
         }else{
-            DECLARE_ALIGNED_8(int16_t, bS)[4];
+            DECLARE_ALIGNED(8, int16_t, bS)[4];
             int qp;
 
             if( IS_INTRA(mb_type|mbm_type)) {
-                *(uint64_t*)bS= 0x0003000300030003ULL;
+                AV_WN64A(bS, 0x0003000300030003ULL);
                 if (   (!IS_INTERLACED(mb_type|mbm_type))
                     || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
                 )
-                    *(uint64_t*)bS= 0x0004000400040004ULL;
+                    AV_WN64A(bS, 0x0004000400040004ULL);
             } else {
-                int i, l;
+                int i;
                 int mv_done;
 
                 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
-                    *(uint64_t*)bS= 0x0001000100010001ULL;
+                    AV_WN64A(bS, 0x0001000100010001ULL);
                     mv_done = 1;
                 }
                 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
                     int b_idx= 8 + 4;
                     int bn_idx= b_idx - (dir ? 8:1);
 
                     bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
                     mv_done = 1;
                 }
                 else
@@ skipping 42 matching lines @@
                             qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
                         filter_mb_edgech( &img_cr[0], uvlinesize, bS, qp, h);
                     }
                 }
             }
         }
     }
 
     /* Calculate bS */
     for( edge = 1; edge < edges; edge++ ) {
-        DECLARE_ALIGNED_8(int16_t, bS)[4];
+        DECLARE_ALIGNED(8, int16_t, bS)[4];
         int qp;
 
         if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
             continue;
 
         if( IS_INTRA(mb_type)) {
-            *(uint64_t*)bS= 0x0003000300030003ULL;
+            AV_WN64A(bS, 0x0003000300030003ULL);
         } else {
-            int i, l;
+            int i;
             int mv_done;
 
             if( edge & mask_edge ) {
-                *(uint64_t*)bS= 0;
+                AV_ZERO64(bS);
                 mv_done = 1;
             }
             else if( mask_par0 ) {
                 int b_idx= 8 + 4 + edge * (dir ? 8:1);
                 int bn_idx= b_idx - (dir ? 8:1);
 
                 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
                 mv_done = 1;
             }
             else
@@ skipping 42 matching lines @@
     }
 }
 
 void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
     MpegEncContext * const s = &h->s;
     const int mb_xy= mb_x + mb_y*s->mb_stride;
     const int mb_type = s->current_picture.mb_type[mb_xy];
     const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
     int first_vertical_edge_done = 0;
     av_unused int dir;
-    int list;
 
     if (FRAME_MBAFF
             // and current and left pair do not have the same interlaced type
             && IS_INTERLACED(mb_type^h->left_type[0])
             // and left mb is in available to us
             && h->left_type[0]) {
         /* First vertical edge is different in MBAFF frames
          * There are 8 different bS to compute and 2 different Qp
          */
-        DECLARE_ALIGNED_8(int16_t, bS)[8];
+        DECLARE_ALIGNED(8, int16_t, bS)[8];
         int qp[2];
         int bqp[2];
         int rqp[2];
         int mb_qp, mbn0_qp, mbn1_qp;
         int i;
         first_vertical_edge_done = 1;
 
-        if( IS_INTRA(mb_type) )
-            *(uint64_t*)&bS[0]=
-            *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
-        else {
+        if( IS_INTRA(mb_type) ) {
+            AV_WN64A(&bS[0], 0x0004000400040004ULL);
+            AV_WN64A(&bS[4], 0x0004000400040004ULL);
+        } else {
             static const uint8_t offset[2][2][8]={
                 {
                     {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
                     {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
                 },{
                     {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
                     {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
                 }
             };
             const uint8_t *off= offset[MB_FIELD][mb_y&1];
@@ skipping 49 matching lines @@
     }
 
 #if CONFIG_SMALL
     for( dir = 0; dir < 2; dir++ )
         filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, dir);
 #else
     filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, 0);
     filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
 #endif
 }