Mercurial > libavcodec.hg
changeset 10852:86d7ab878805 libavcodec
Get rid of #include "svq3.c"
functions called more than per mb are moved into the header, scan8 is also
as it must be known at compiletime.
The code after this patch duplicates h264data.h, this has been done to minimize
the changes in this step and allow more fine grained benchmarking.
Speedwise this is 1% faster on my pentium dual core with diegos cursed cathedral
sample.
| author | michael |
|---|---|
| date | Tue, 12 Jan 2010 05:30:31 +0000 |
| parents | 07dbba7c16ec |
| children | 7ba2dc9ed4ac |
| files | Makefile h264.c h264.h h264data.h svq3.c |
| diffstat | 5 files changed, 250 insertions(+), 224 deletions(-) [+] |
line wrap: on
line diff
--- a/Makefile Tue Jan 12 04:49:18 2010 +0000 +++ b/Makefile Tue Jan 12 05:30:31 2010 +0000 @@ -131,7 +131,7 @@ OBJS-$(CONFIG_H263_ENCODER) += mpegvideo_enc.o mpeg4video.o mpeg4videoenc.o motion_est.o \ ratecontrol.o h263.o ituh263enc.o flvenc.o mpeg12data.o \ mpegvideo.o error_resilience.o -OBJS-$(CONFIG_H264_DECODER) += h264.o h264idct.o h264pred.o cabac.o \ +OBJS-$(CONFIG_H264_DECODER) += h264.o svq3.o h264idct.o h264pred.o cabac.o \ mpegvideo.o error_resilience.o OBJS-$(CONFIG_H264_ENCODER) += h264enc.o h264dspenc.o OBJS-$(CONFIG_H264_VAAPI_HWACCEL) += vaapi_h264.o
--- a/h264.c Tue Jan 12 04:49:18 2010 +0000 +++ b/h264.c Tue Jan 12 05:30:31 2010 +0000 @@ -81,14 +81,6 @@ static void filter_mb_fast( 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); static Picture * remove_long(H264Context *h, int i, int ref_mask); -static av_always_inline uint32_t pack16to32(int a, int b){ -#if HAVE_BIGENDIAN - return (b&0xFFFF) + (a<<16); -#else - return (a&0xFFFF) + (b<<16); -#endif -} - static const uint8_t rem6[52]={ 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, }; @@ -546,7 +538,7 @@ h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]); } -static inline void write_back_intra_pred_mode(H264Context *h){ +void ff_h264_write_back_intra_pred_mode(H264Context *h){ const int mb_xy= h->mb_xy; h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1]; @@ -561,46 +553,7 @@ /** * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. */ -static inline int check_intra4x4_pred_mode(H264Context *h){ - MpegEncContext * const s = &h->s; - static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; - static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; - int i; - - if(!(h->top_samples_available&0x8000)){ - for(i=0; i<4; i++){ - int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; - if(status<0){ - av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); - return -1; - } else if(status){ - h->intra4x4_pred_mode_cache[scan8[0] + i]= status; - } - } - } - - if((h->left_samples_available&0x8888)!=0x8888){ - static const int mask[4]={0x8000,0x2000,0x80,0x20}; - for(i=0; i<4; i++){ - if(!(h->left_samples_available&mask[i])){ - int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; - if(status<0){ - av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); - return -1; - } else if(status){ - h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; - } - } - } - } - - return 0; -} //FIXME cleanup like next - -/** - * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. - */ -static inline int check_intra_pred_mode(H264Context *h, int mode){ +int ff_h264_check_intra_pred_mode(H264Context *h, int mode){ MpegEncContext * const s = &h->s; static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1}; static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8}; @@ -684,119 +637,6 @@ return i&31; } -static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ - const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; - MpegEncContext *s = &h->s; - - /* there is no consistent mapping of mvs to neighboring locations that will - * make mbaff happy, so we can't move all this logic to fill_caches */ - if(FRAME_MBAFF){ - const uint32_t *mb_types = s->current_picture_ptr->mb_type; - const int16_t *mv; - *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0; - *C = h->mv_cache[list][scan8[0]-2]; - - if(!MB_FIELD - && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){ - int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3); - if(IS_INTERLACED(mb_types[topright_xy])){ -#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\ - const int x4 = X4, y4 = Y4;\ - const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\ - if(!USES_LIST(mb_type,list))\ - return LIST_NOT_USED;\ - mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\ - h->mv_cache[list][scan8[0]-2][0] = mv[0];\ - h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\ - return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP; - - SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1); - } - } - if(topright_ref == PART_NOT_AVAILABLE - && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4 - && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){ - if(!MB_FIELD - && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){ - SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1); - } - if(MB_FIELD - && !IS_INTERLACED(mb_types[h->left_mb_xy[0]]) - && i >= scan8[0]+8){ - // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK. - SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2); - } - } -#undef SET_DIAG_MV - } - - if(topright_ref != PART_NOT_AVAILABLE){ - *C= h->mv_cache[list][ i - 8 + part_width ]; - return topright_ref; - }else{ - tprintf(s->avctx, "topright MV not available\n"); - - *C= h->mv_cache[list][ i - 8 - 1 ]; - return h->ref_cache[list][ i - 8 - 1 ]; - } -} - -/** - * gets the predicted MV. - * @param n the block index - * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) - * @param mx the x component of the predicted motion vector - * @param my the y component of the predicted motion vector - */ -static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ - const int index8= scan8[n]; - const int top_ref= h->ref_cache[list][ index8 - 8 ]; - const int left_ref= h->ref_cache[list][ index8 - 1 ]; - const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; - const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; - const int16_t * C; - int diagonal_ref, match_count; - - assert(part_width==1 || part_width==2 || part_width==4); - -/* mv_cache - B . . A T T T T - U . . L . . , . - U . . L . . . . - U . . L . . , . - . . . L . . . . -*/ - - diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); - match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); - tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count); - if(match_count > 1){ //most common - *mx= mid_pred(A[0], B[0], C[0]); - *my= mid_pred(A[1], B[1], C[1]); - }else if(match_count==1){ - if(left_ref==ref){ - *mx= A[0]; - *my= A[1]; - }else if(top_ref==ref){ - *mx= B[0]; - *my= B[1]; - }else{ - *mx= C[0]; - *my= C[1]; - } - }else{ - if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ - *mx= A[0]; - *my= A[1]; - }else{ - *mx= mid_pred(A[0], B[0], C[0]); - *my= mid_pred(A[1], B[1], C[1]); - } - } - - tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list); -} - /** * gets the directionally predicted 16x8 MV. * @param n the block index @@ -2070,11 +1910,7 @@ } -/** - * allocates tables. - * needs width/height - */ -static int alloc_tables(H264Context *h){ +int ff_h264_alloc_tables(H264Context *h){ MpegEncContext * const s = &h->s; const int big_mb_num= s->mb_stride * (s->mb_height+1); int x,y; @@ -2180,7 +2016,7 @@ h->sei_buffering_period_present = 0; } -static av_cold int decode_init(AVCodecContext *avctx){ +av_cold int ff_h264_decode_init(AVCodecContext *avctx){ H264Context *h= avctx->priv_data; MpegEncContext * const s = &h->s; @@ -2223,7 +2059,7 @@ return 0; } -static int frame_start(H264Context *h){ +int ff_h264_frame_start(H264Context *h){ MpegEncContext * const s = &h->s; int i; @@ -2686,7 +2522,7 @@ hl_decode_mb_internal(h, 0); } -static void hl_decode_mb(H264Context *h){ +void ff_h264_hl_decode_mb(H264Context *h){ MpegEncContext * const s = &h->s; const int mb_xy= h->mb_xy; const int mb_type= s->current_picture.mb_type[mb_xy]; @@ -3838,7 +3674,7 @@ h->prev_interlaced_frame = 1; init_scan_tables(h); - alloc_tables(h); + ff_h264_alloc_tables(h); for(i = 1; i < s->avctx->thread_count; i++) { H264Context *c; @@ -3877,7 +3713,7 @@ while(h->frame_num != h->prev_frame_num && h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){ av_log(NULL, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num); - if (frame_start(h) < 0) + if (ff_h264_frame_start(h) < 0) return -1; h->prev_frame_num++; h->prev_frame_num %= 1<<h->sps.log2_max_frame_num; @@ -3925,7 +3761,7 @@ s0->first_field = FIELD_PICTURE; } - if((!FIELD_PICTURE || s0->first_field) && frame_start(h) < 0) { + if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) { s0->first_field = 0; return -1; } @@ -4561,16 +4397,16 @@ else h->intra4x4_pred_mode_cache[ scan8[i] ] = mode; } - write_back_intra_pred_mode(h); + ff_h264_write_back_intra_pred_mode(h); if( check_intra4x4_pred_mode(h) < 0) return -1; }else{ - h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode); + h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode(h, h->intra16x16_pred_mode); if(h->intra16x16_pred_mode < 0) return -1; } if(CHROMA){ - pred_mode= check_intra_pred_mode(h, get_ue_golomb_31(&s->gb)); + pred_mode= ff_h264_check_intra_pred_mode(h, get_ue_golomb_31(&s->gb)); if(pred_mode < 0) return -1; h->chroma_pred_mode= pred_mode; @@ -5662,17 +5498,17 @@ //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] ); } } - write_back_intra_pred_mode(h); + ff_h264_write_back_intra_pred_mode(h); if( check_intra4x4_pred_mode(h) < 0 ) return -1; } else { - h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode ); + h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode( h, h->intra16x16_pred_mode ); if( h->intra16x16_pred_mode < 0 ) return -1; } if(CHROMA){ h->chroma_pred_mode_table[mb_xy] = pred_mode = decode_cabac_mb_chroma_pre_mode( h ); - pred_mode= check_intra_pred_mode( h, pred_mode ); + pred_mode= ff_h264_check_intra_pred_mode( h, pred_mode ); if( pred_mode < 0 ) return -1; h->chroma_pred_mode= pred_mode; } @@ -6734,14 +6570,14 @@ int eos; //STOP_TIMER("decode_mb_cabac") - if(ret>=0) hl_decode_mb(h); + if(ret>=0) ff_h264_hl_decode_mb(h); if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ? s->mb_y++; ret = decode_mb_cabac(h); - if(ret>=0) hl_decode_mb(h); + if(ret>=0) ff_h264_hl_decode_mb(h); s->mb_y--; } eos = get_cabac_terminate( &h->cabac ); @@ -6772,13 +6608,13 @@ for(;;){ int ret = decode_mb_cavlc(h); - if(ret>=0) hl_decode_mb(h); + if(ret>=0) ff_h264_hl_decode_mb(h); if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ? s->mb_y++; ret = decode_mb_cavlc(h); - if(ret>=0) hl_decode_mb(h); + if(ret>=0) ff_h264_hl_decode_mb(h); s->mb_y--; } @@ -6831,7 +6667,7 @@ for(;s->mb_x < s->mb_width; s->mb_x++){ int ret= decode_mb(h); - hl_decode_mb(h); + ff_h264_hl_decode_mb(h); if(ret<0){ av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); @@ -8176,7 +8012,7 @@ av_freep(h->pps_buffers + i); } -static av_cold int decode_end(AVCodecContext *avctx) +av_cold int ff_h264_decode_end(AVCodecContext *avctx) { H264Context *h = avctx->priv_data; MpegEncContext *s = &h->s; @@ -8196,9 +8032,9 @@ CODEC_TYPE_VIDEO, CODEC_ID_H264, sizeof(H264Context), - decode_init, + ff_h264_decode_init, NULL, - decode_end, + ff_h264_decode_end, decode_frame, /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY, .flush= flush_dpb, @@ -8212,9 +8048,9 @@ CODEC_TYPE_VIDEO, CODEC_ID_H264, sizeof(H264Context), - decode_init, + ff_h264_decode_init, NULL, - decode_end, + ff_h264_decode_end, decode_frame, CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU, .flush= flush_dpb, @@ -8222,7 +8058,3 @@ .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE}, }; #endif - -#if CONFIG_SVQ3_DECODER -#include "svq3.c" -#endif
--- a/h264.h Tue Jan 12 04:49:18 2010 +0000 +++ b/h264.h Tue Jan 12 05:30:31 2010 +0000 @@ -593,4 +593,203 @@ */ int ff_h264_get_slice_type(H264Context *h); +/** + * allocates tables. + * needs width/height + */ +int ff_h264_alloc_tables(H264Context *h); + +/** + * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. + */ +int ff_h264_check_intra_pred_mode(H264Context *h, int mode); + +void ff_h264_write_back_intra_pred_mode(H264Context *h); +void ff_h264_hl_decode_mb(H264Context *h); +int ff_h264_frame_start(H264Context *h); +av_cold int ff_h264_decode_init(AVCodecContext *avctx); +av_cold int ff_h264_decode_end(AVCodecContext *avctx); + +/* +o-o o-o + / / / +o-o o-o + ,---' +o-o o-o + / / / +o-o o-o +*/ +//This table must be here because scan8[constant] must be known at compiletime +static const uint8_t scan8[16 + 2*4]={ + 4+1*8, 5+1*8, 4+2*8, 5+2*8, + 6+1*8, 7+1*8, 6+2*8, 7+2*8, + 4+3*8, 5+3*8, 4+4*8, 5+4*8, + 6+3*8, 7+3*8, 6+4*8, 7+4*8, + 1+1*8, 2+1*8, + 1+2*8, 2+2*8, + 1+4*8, 2+4*8, + 1+5*8, 2+5*8, +}; + +static av_always_inline uint32_t pack16to32(int a, int b){ +#if HAVE_BIGENDIAN + return (b&0xFFFF) + (a<<16); +#else + return (a&0xFFFF) + (b<<16); +#endif +} + +/** + * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. + */ +static inline int check_intra4x4_pred_mode(H264Context *h){ + MpegEncContext * const s = &h->s; + static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; + static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; + int i; + + if(!(h->top_samples_available&0x8000)){ + for(i=0; i<4; i++){ + int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; + if(status<0){ + av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); + return -1; + } else if(status){ + h->intra4x4_pred_mode_cache[scan8[0] + i]= status; + } + } + } + + if((h->left_samples_available&0x8888)!=0x8888){ + static const int mask[4]={0x8000,0x2000,0x80,0x20}; + for(i=0; i<4; i++){ + if(!(h->left_samples_available&mask[i])){ + int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; + if(status<0){ + av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); + return -1; + } else if(status){ + h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; + } + } + } + } + + return 0; +} //FIXME cleanup like ff_h264_check_intra_pred_mode + +static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ + const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; + MpegEncContext *s = &h->s; + + /* there is no consistent mapping of mvs to neighboring locations that will + * make mbaff happy, so we can't move all this logic to fill_caches */ + if(FRAME_MBAFF){ + const uint32_t *mb_types = s->current_picture_ptr->mb_type; + const int16_t *mv; + *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0; + *C = h->mv_cache[list][scan8[0]-2]; + + if(!MB_FIELD + && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){ + int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3); + if(IS_INTERLACED(mb_types[topright_xy])){ +#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\ + const int x4 = X4, y4 = Y4;\ + const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\ + if(!USES_LIST(mb_type,list))\ + return LIST_NOT_USED;\ + mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\ + h->mv_cache[list][scan8[0]-2][0] = mv[0];\ + h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\ + return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP; + + SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1); + } + } + if(topright_ref == PART_NOT_AVAILABLE + && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4 + && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){ + if(!MB_FIELD + && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){ + SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1); + } + if(MB_FIELD + && !IS_INTERLACED(mb_types[h->left_mb_xy[0]]) + && i >= scan8[0]+8){ + // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK. + SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2); + } + } +#undef SET_DIAG_MV + } + + if(topright_ref != PART_NOT_AVAILABLE){ + *C= h->mv_cache[list][ i - 8 + part_width ]; + return topright_ref; + }else{ + tprintf(s->avctx, "topright MV not available\n"); + + *C= h->mv_cache[list][ i - 8 - 1 ]; + return h->ref_cache[list][ i - 8 - 1 ]; + } +} + +/** + * gets the predicted MV. + * @param n the block index + * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) + * @param mx the x component of the predicted motion vector + * @param my the y component of the predicted motion vector + */ +static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ + const int index8= scan8[n]; + const int top_ref= h->ref_cache[list][ index8 - 8 ]; + const int left_ref= h->ref_cache[list][ index8 - 1 ]; + const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; + const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; + const int16_t * C; + int diagonal_ref, match_count; + + assert(part_width==1 || part_width==2 || part_width==4); + +/* mv_cache + B . . A T T T T + U . . L . . , . + U . . L . . . . + U . . L . . , . + . . . L . . . . +*/ + + diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); + match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); + tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count); + if(match_count > 1){ //most common + *mx= mid_pred(A[0], B[0], C[0]); + *my= mid_pred(A[1], B[1], C[1]); + }else if(match_count==1){ + if(left_ref==ref){ + *mx= A[0]; + *my= A[1]; + }else if(top_ref==ref){ + *mx= B[0]; + *my= B[1]; + }else{ + *mx= C[0]; + *my= C[1]; + } + }else{ + if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ + *mx= A[0]; + *my= A[1]; + }else{ + *mx= mid_pred(A[0], B[0], C[0]); + *my= mid_pred(A[1], B[1], C[1]); + } + } + + tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list); +} + + #endif /* AVCODEC_H264_H */
--- a/h264data.h Tue Jan 12 04:49:18 2010 +0000 +++ b/h264data.h Tue Jan 12 05:30:31 2010 +0000 @@ -232,26 +232,6 @@ {7,6,5,4,3,2,1,1,1,1,1,1,1,1,1}, }; -/* -o-o o-o - / / / -o-o o-o - ,---' -o-o o-o - / / / -o-o o-o -*/ - -static const uint8_t scan8[16 + 2*4]={ - 4+1*8, 5+1*8, 4+2*8, 5+2*8, - 6+1*8, 7+1*8, 6+2*8, 7+2*8, - 4+3*8, 5+3*8, 4+4*8, 5+4*8, - 6+3*8, 7+3*8, 6+4*8, 7+4*8, - 1+1*8, 2+1*8, - 1+2*8, 2+2*8, - 1+4*8, 2+4*8, - 1+5*8, 2+5*8, -}; static const uint8_t zigzag_scan[16]={ 0+0*4, 1+0*4, 0+1*4, 0+2*4,
--- a/svq3.c Tue Jan 12 04:49:18 2010 +0000 +++ b/svq3.c Tue Jan 12 05:30:31 2010 +0000 @@ -39,6 +39,21 @@ * correctly decodes this file: * http://samples.mplayerhq.hu/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov */ +#include "internal.h" +#include "dsputil.h" +#include "avcodec.h" +#include "mpegvideo.h" +#include "h264.h" + +#include "h264data.h" //FIXME FIXME FIXME + +#include "golomb.h" +#include "rectangle.h" +#include "vdpau_internal.h" + +#if ARCH_X86 +#include "x86/h264_i386.h" +#endif #if CONFIG_ZLIB #include <zlib.h> @@ -114,7 +129,7 @@ }; -static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp) +void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp) { const int qmul = svq3_dequant_coeff[qp]; #define stride 16 @@ -151,7 +166,7 @@ } #undef stride -static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, +void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc) { const int qmul = svq3_dequant_coeff[qp]; @@ -571,7 +586,7 @@ } } - write_back_intra_pred_mode(h); + ff_h264_write_back_intra_pred_mode(h); if (mb_type == 8) { check_intra4x4_pred_mode(h); @@ -592,7 +607,7 @@ dir = i_mb_type_info[mb_type - 8].pred_mode; dir = (dir >> 1) ^ 3*(dir & 1) ^ 1; - if ((h->intra16x16_pred_mode = check_intra_pred_mode(h, dir)) == -1){ + if ((h->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, dir)) == -1){ av_log(h->s.avctx, AV_LOG_ERROR, "check_intra_pred_mode = -1\n"); return -1; } @@ -685,7 +700,7 @@ s->current_picture.mb_type[mb_xy] = mb_type; if (IS_INTRA(mb_type)) { - h->chroma_pred_mode = check_intra_pred_mode(h, DC_PRED8x8); + h->chroma_pred_mode = ff_h264_check_intra_pred_mode(h, DC_PRED8x8); } return 0; @@ -789,7 +804,7 @@ return -1; } - if (decode_init(avctx) < 0) + if (ff_h264_decode_init(avctx) < 0) return -1; s->flags = avctx->flags; @@ -811,7 +826,7 @@ h->b_stride = 4*s->mb_width; - alloc_tables(h); + ff_h264_alloc_tables(h); /* prowl for the "SEQH" marker in the extradata */ extradata = (unsigned char *)avctx->extradata; @@ -965,7 +980,7 @@ s->next_p_frame_damaged = 0; } - if (frame_start(h) < 0) + if (ff_h264_frame_start(h) < 0) return -1; if (s->pict_type == FF_B_TYPE) { @@ -1028,7 +1043,7 @@ } if (mb_type != 0) { - hl_decode_mb (h); + ff_h264_hl_decode_mb (h); } if (s->pict_type != FF_B_TYPE && !s->low_delay) { @@ -1064,7 +1079,7 @@ sizeof(H264Context), svq3_decode_init, NULL, - decode_end, + ff_h264_decode_end, svq3_decode_frame, CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | CODEC_CAP_DELAY, .long_name = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),