Mercurial > libavcodec.hg
comparison motion_est_template.c @ 1799:95612d423fde libavcodec
multithreaded/SMP motion estimation
multithreaded/SMP encoding for MPEG1/MPEG2/MPEG4/H263
all pthread specific code is in pthread.c
to try it, run configure --enable-pthreads and ffmpeg ... -threads <num>
the internal thread API is a simple AVCodecContext.execute() callback which executes a given function pointer with different arguments and returns after finishing all, that way no mutexes or other thread-mess is needed outside pthread.c
| author | michael |
|---|---|
| date | Fri, 13 Feb 2004 17:54:10 +0000 |
| parents | 07a484280a82 |
| children | e039d79185c2 |
comparison
equal
deleted
inserted
replaced
| 1798:a3da4b429984 | 1799:95612d423fde |
|---|---|
| 555 }\ | 555 }\ |
| 556 } | 556 } |
| 557 | 557 |
| 558 #define CHECK_CLIPED_MV(ax,ay)\ | 558 #define CHECK_CLIPED_MV(ax,ay)\ |
| 559 {\ | 559 {\ |
| 560 const int x= FFMAX(xmin, FFMIN(ax, xmax));\ | 560 const int x= ax;\ |
| 561 const int y= FFMAX(ymin, FFMIN(ay, ymax));\ | 561 const int y= ay;\ |
| 562 CHECK_MV(x, y)\ | 562 const int x2= FFMAX(xmin, FFMIN(x, xmax));\ |
| 563 const int y2= FFMAX(ymin, FFMIN(y, ymax));\ | |
| 564 CHECK_MV(x2, y2)\ | |
| 563 } | 565 } |
| 564 | 566 |
| 565 #define CHECK_MV_DIR(x,y,new_dir)\ | 567 #define CHECK_MV_DIR(x,y,new_dir)\ |
| 566 {\ | 568 {\ |
| 567 const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\ | 569 const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\ |
| 910 CMP(dmin, 0, 0, size); | 912 CMP(dmin, 0, 0, size); |
| 911 map[0]= map_generation; | 913 map[0]= map_generation; |
| 912 score_map[0]= dmin; | 914 score_map[0]= dmin; |
| 913 | 915 |
| 914 /* first line */ | 916 /* first line */ |
| 915 if (s->mb_y == 0) { | 917 if (s->first_slice_line) { |
| 916 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) | 918 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) |
| 917 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, | 919 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, |
| 918 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) | 920 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) |
| 919 }else{ | 921 }else{ |
| 920 if(dmin<256 && ( P_LEFT[0] |P_LEFT[1] | 922 if(dmin<256 && ( P_LEFT[0] |P_LEFT[1] |
| 936 } | 938 } |
| 937 if(dmin>256*4){ | 939 if(dmin>256*4){ |
| 938 if(s->me.pre_pass){ | 940 if(s->me.pre_pass){ |
| 939 CHECK_CLIPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16, | 941 CHECK_CLIPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16, |
| 940 (last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16) | 942 (last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16) |
| 941 CHECK_CLIPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, | 943 if(!s->first_slice_line) |
| 942 (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | 944 CHECK_CLIPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, |
| 945 (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | |
| 943 }else{ | 946 }else{ |
| 944 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, | 947 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, |
| 945 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) | 948 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) |
| 946 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, | 949 if(s->end_mb_y == s->mb_height || s->mb_y+1<s->end_mb_y) //FIXME replace at least with last_slice_line |
| 947 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | 950 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, |
| 951 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | |
| 948 } | 952 } |
| 949 } | 953 } |
| 950 | 954 |
| 951 if(s->avctx->last_predictor_count){ | 955 if(s->avctx->last_predictor_count){ |
| 952 const int count= s->avctx->last_predictor_count; | 956 const int count= s->avctx->last_predictor_count; |
| 1022 map_generation= update_map_generation(s); | 1026 map_generation= update_map_generation(s); |
| 1023 | 1027 |
| 1024 dmin = 1000000; | 1028 dmin = 1000000; |
| 1025 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax); | 1029 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax); |
| 1026 /* first line */ | 1030 /* first line */ |
| 1027 if (s->mb_y == 0/* && block<2*/) { | 1031 if (s->first_slice_line) { |
| 1028 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) | 1032 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) |
| 1029 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, | 1033 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, |
| 1030 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) | 1034 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) |
| 1031 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift) | 1035 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift) |
| 1032 }else{ | 1036 }else{ |
| 1042 } | 1046 } |
| 1043 } | 1047 } |
| 1044 if(dmin>64*4){ | 1048 if(dmin>64*4){ |
| 1045 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, | 1049 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, |
| 1046 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) | 1050 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) |
| 1047 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, | 1051 if(s->end_mb_y == s->mb_height || s->mb_y+1<s->end_mb_y) //FIXME replace at least with last_slice_line |
| 1048 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | 1052 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, |
| 1053 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | |
| 1049 } | 1054 } |
| 1050 | 1055 |
| 1051 if(s->me.dia_size==-1) | 1056 if(s->me.dia_size==-1) |
| 1052 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride, | 1057 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride, |
| 1053 pred_x, pred_y, penalty_factor, | 1058 pred_x, pred_y, penalty_factor, |
| 1100 map_generation= update_map_generation(s); | 1105 map_generation= update_map_generation(s); |
| 1101 | 1106 |
| 1102 dmin = 1000000; | 1107 dmin = 1000000; |
| 1103 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax); | 1108 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax); |
| 1104 /* first line */ | 1109 /* first line */ |
| 1105 if (s->mb_y == 0) { | 1110 if (s->first_slice_line) { |
| 1106 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) | 1111 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) |
| 1107 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, | 1112 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, |
| 1108 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) | 1113 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) |
| 1109 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift) | 1114 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift) |
| 1110 }else{ | 1115 }else{ |
| 1120 } | 1125 } |
| 1121 } | 1126 } |
| 1122 if(dmin>64*4){ | 1127 if(dmin>64*4){ |
| 1123 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, | 1128 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, |
| 1124 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) | 1129 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) |
| 1125 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, | 1130 if(s->end_mb_y == s->mb_height || s->mb_y+1<s->end_mb_y) //FIXME replace at least with last_slice_line |
| 1126 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | 1131 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, |
| 1132 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | |
| 1127 } | 1133 } |
| 1128 | 1134 |
| 1129 if(s->me.dia_size==-1) | 1135 if(s->me.dia_size==-1) |
| 1130 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride, | 1136 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride, |
| 1131 pred_x, pred_y, penalty_factor, | 1137 pred_x, pred_y, penalty_factor, |