diff options
Diffstat (limited to 'libavcodec')
| -rw-r--r-- | libavcodec/aacps.c | 24 | ||||
| -rw-r--r-- | libavcodec/h264.h | 418 | ||||
| -rw-r--r-- | libavcodec/h264_mvpred.h | 418 | ||||
| -rw-r--r-- | libavcodec/lpc.c | 6 | ||||
| -rw-r--r-- | libavcodec/msmpeg4.c | 2 | ||||
| -rw-r--r-- | libavcodec/version.h | 2 | ||||
| -rw-r--r-- | libavcodec/x86/h264_intrapred_10bit.asm | 231 | ||||
| -rw-r--r-- | libavcodec/x86/h264_intrapred_init.c | 23 |
8 files changed, 687 insertions, 437 deletions
diff --git a/libavcodec/aacps.c b/libavcodec/aacps.c index f2caedb..e7341b0 100644 --- a/libavcodec/aacps.c +++ b/libavcodec/aacps.c @@ -69,19 +69,19 @@ static const int huff_iid[] = { static VLC vlc_ps[10]; -/** - * Read Inter-channel Intensity Difference/Inter-Channel Coherence/ - * Inter-channel Phase Difference/Overall Phase Difference parameters from the - * bitstream. - * - * @param avctx contains the current codec context - * @param gb pointer to the input bitstream - * @param ps pointer to the Parametric Stereo context - * @param PAR pointer to the parameter to be read - * @param e envelope to decode - * @param dt 1: time delta-coded, 0: frequency delta-coded - */ #define READ_PAR_DATA(PAR, OFFSET, MASK, ERR_CONDITION) \ +/** \ + * Read Inter-channel Intensity Difference/Inter-Channel Coherence/ \ + * Inter-channel Phase Difference/Overall Phase Difference parameters from the \ + * bitstream. \ + * \ + * @param avctx contains the current codec context \ + * @param gb pointer to the input bitstream \ + * @param ps pointer to the Parametric Stereo context \ + * @param PAR pointer to the parameter to be read \ + * @param e envelope to decode \ + * @param dt 1: time delta-coded, 0: frequency delta-coded \ + */ \ static int read_ ## PAR ## _data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, \ int8_t (*PAR)[PS_MAX_NR_IIDICC], int table_idx, int e, int dt) \ { \ diff --git a/libavcodec/h264.h b/libavcodec/h264.h index ee52dee..d448fc3 100644 --- a/libavcodec/h264.h +++ b/libavcodec/h264.h @@ -770,424 +770,6 @@ static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){ return h->pps.chroma_qp_table[t][qscale]; } -static void fill_decode_neighbors(H264Context *h, int mb_type){ - MpegEncContext * const s = &h->s; - const int mb_xy= h->mb_xy; - int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; - static const uint8_t left_block_options[4][32]={ - {0,1,2,3,7,10,8,11,3+0*4, 3+1*4, 3+2*4, 3+3*4, 1+4*4, 1+8*4, 1+5*4, 1+9*4}, - {2,2,3,3,8,11,8,11,3+2*4, 3+2*4, 3+3*4, 3+3*4, 1+5*4, 1+9*4, 1+5*4, 1+9*4}, - {0,0,1,1,7,10,7,10,3+0*4, 3+0*4, 3+1*4, 3+1*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4}, - {0,2,0,2,7,10,7,10,3+0*4, 3+2*4, 3+0*4, 3+2*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4} - }; - - h->topleft_partition= -1; - - top_xy = mb_xy - (s->mb_stride << MB_FIELD); - - /* Wow, what a mess, why didn't they simplify the interlacing & intra - * stuff, I can't imagine that these complex rules are worth it. */ - - topleft_xy = top_xy - 1; - topright_xy= top_xy + 1; - left_xy[LBOT] = left_xy[LTOP] = mb_xy-1; - h->left_block = left_block_options[0]; - if(FRAME_MBAFF){ - const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]); - const int curr_mb_field_flag = IS_INTERLACED(mb_type); - if(s->mb_y&1){ - if (left_mb_field_flag != curr_mb_field_flag) { - left_xy[LBOT] = left_xy[LTOP] = mb_xy - s->mb_stride - 1; - if (curr_mb_field_flag) { - left_xy[LBOT] += s->mb_stride; - h->left_block = left_block_options[3]; - } else { - topleft_xy += s->mb_stride; - // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition - h->topleft_partition = 0; - h->left_block = left_block_options[1]; - } - } - }else{ - if(curr_mb_field_flag){ - topleft_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy - 1] >> 7) & 1) - 1); - topright_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy + 1] >> 7) & 1) - 1); - top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy ] >> 7) & 1) - 1); - } - if (left_mb_field_flag != curr_mb_field_flag) { - if (curr_mb_field_flag) { - left_xy[LBOT] += s->mb_stride; - h->left_block = left_block_options[3]; - } else { - h->left_block = left_block_options[2]; - } - } - } - } - - h->topleft_mb_xy = topleft_xy; - h->top_mb_xy = top_xy; - h->topright_mb_xy= topright_xy; - h->left_mb_xy[LTOP] = left_xy[LTOP]; - h->left_mb_xy[LBOT] = left_xy[LBOT]; - //FIXME do we need all in the context? - - h->topleft_type = s->current_picture.f.mb_type[topleft_xy]; - h->top_type = s->current_picture.f.mb_type[top_xy]; - h->topright_type = s->current_picture.f.mb_type[topright_xy]; - h->left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]]; - h->left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]]; - - if(FMO){ - if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0; - if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0; - if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0; - }else{ - if(h->slice_table[topleft_xy ] != h->slice_num){ - h->topleft_type = 0; - if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0; - if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0; - } - } - if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0; -} - -static void fill_decode_caches(H264Context *h, int mb_type){ - MpegEncContext * const s = &h->s; - int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; - int topleft_type, top_type, topright_type, left_type[LEFT_MBS]; - const uint8_t * left_block= h->left_block; - int i; - uint8_t *nnz; - uint8_t *nnz_cache; - - topleft_xy = h->topleft_mb_xy; - top_xy = h->top_mb_xy; - topright_xy = h->topright_mb_xy; - left_xy[LTOP] = h->left_mb_xy[LTOP]; - left_xy[LBOT] = h->left_mb_xy[LBOT]; - topleft_type = h->topleft_type; - top_type = h->top_type; - topright_type = h->topright_type; - left_type[LTOP]= h->left_type[LTOP]; - left_type[LBOT]= h->left_type[LBOT]; - - if(!IS_SKIP(mb_type)){ - if(IS_INTRA(mb_type)){ - int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1; - h->topleft_samples_available= - h->top_samples_available= - h->left_samples_available= 0xFFFF; - h->topright_samples_available= 0xEEEA; - - if(!(top_type & type_mask)){ - h->topleft_samples_available= 0xB3FF; - h->top_samples_available= 0x33FF; - h->topright_samples_available= 0x26EA; - } - if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])){ - if(IS_INTERLACED(mb_type)){ - if(!(left_type[LTOP] & type_mask)){ - h->topleft_samples_available&= 0xDFFF; - h->left_samples_available&= 0x5FFF; - } - if(!(left_type[LBOT] & type_mask)){ - h->topleft_samples_available&= 0xFF5F; - h->left_samples_available&= 0xFF5F; - } - }else{ - int left_typei = s->current_picture.f.mb_type[left_xy[LTOP] + s->mb_stride]; - - assert(left_xy[LTOP] == left_xy[LBOT]); - if(!((left_typei & type_mask) && (left_type[LTOP] & type_mask))){ - h->topleft_samples_available&= 0xDF5F; - h->left_samples_available&= 0x5F5F; - } - } - }else{ - if(!(left_type[LTOP] & type_mask)){ - h->topleft_samples_available&= 0xDF5F; - h->left_samples_available&= 0x5F5F; - } - } - - if(!(topleft_type & type_mask)) - h->topleft_samples_available&= 0x7FFF; - - if(!(topright_type & type_mask)) - h->topright_samples_available&= 0xFBFF; - - if(IS_INTRA4x4(mb_type)){ - if(IS_INTRA4x4(top_type)){ - AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]); - }else{ - h->intra4x4_pred_mode_cache[4+8*0]= - h->intra4x4_pred_mode_cache[5+8*0]= - h->intra4x4_pred_mode_cache[6+8*0]= - h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask); - } - for(i=0; i<2; i++){ - if(IS_INTRA4x4(left_type[LEFT(i)])){ - int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]]; - h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]]; - h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]]; - }else{ - h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= - h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[LEFT(i)] & type_mask); - } - } - } - } - - -/* -0 . T T. T T T T -1 L . .L . . . . -2 L . .L . . . . -3 . T TL . . . . -4 L . .L . . . . -5 L . .. . . . . -*/ -//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec) - nnz_cache = h->non_zero_count_cache; - if(top_type){ - nnz = h->non_zero_count[top_xy]; - AV_COPY32(&nnz_cache[4+8* 0], &nnz[4*3]); - if(CHROMA444){ - AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 7]); - AV_COPY32(&nnz_cache[4+8*10], &nnz[4*11]); - }else{ - AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 5]); - AV_COPY32(&nnz_cache[4+8*10], &nnz[4* 9]); - } - }else{ - uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040; - AV_WN32A(&nnz_cache[4+8* 0], top_empty); - AV_WN32A(&nnz_cache[4+8* 5], top_empty); - AV_WN32A(&nnz_cache[4+8*10], top_empty); - } - - for (i=0; i<2; i++) { - if(left_type[LEFT(i)]){ - nnz = h->non_zero_count[left_xy[LEFT(i)]]; - nnz_cache[3+8* 1 + 2*8*i]= nnz[left_block[8+0+2*i]]; - nnz_cache[3+8* 2 + 2*8*i]= nnz[left_block[8+1+2*i]]; - if(CHROMA444){ - nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]+4*4]; - nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]+4*4]; - nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]+8*4]; - nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]+8*4]; - }else{ - nnz_cache[3+8* 6 + 8*i]= nnz[left_block[8+4+2*i]]; - nnz_cache[3+8*11 + 8*i]= nnz[left_block[8+5+2*i]]; - } - }else{ - nnz_cache[3+8* 1 + 2*8*i]= - nnz_cache[3+8* 2 + 2*8*i]= - nnz_cache[3+8* 6 + 2*8*i]= - nnz_cache[3+8* 7 + 2*8*i]= - nnz_cache[3+8*11 + 2*8*i]= - nnz_cache[3+8*12 + 2*8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64; - } - } - - if( CABAC ) { - // top_cbp - if(top_type) { - h->top_cbp = h->cbp_table[top_xy]; - } else { - h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; - } - // left_cbp - if (left_type[LTOP]) { - h->left_cbp = (h->cbp_table[left_xy[LTOP]] & 0x7F0) - | ((h->cbp_table[left_xy[LTOP]]>>(left_block[0]&(~1)))&2) - | (((h->cbp_table[left_xy[LBOT]]>>(left_block[2]&(~1)))&2) << 2); - } else { - h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; - } - } - } - - if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){ - int list; - int b_stride = h->b_stride; - for(list=0; list<h->list_count; list++){ - int8_t *ref_cache = &h->ref_cache[list][scan8[0]]; - int8_t *ref = s->current_picture.f.ref_index[list]; - int16_t (*mv_cache)[2] = &h->mv_cache[list][scan8[0]]; - int16_t (*mv)[2] = s->current_picture.f.motion_val[list]; - if(!USES_LIST(mb_type, list)){ - continue; - } - assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)); - - if(USES_LIST(top_type, list)){ - const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride; - AV_COPY128(mv_cache[0 - 1*8], mv[b_xy + 0]); - ref_cache[0 - 1*8]= - ref_cache[1 - 1*8]= ref[4*top_xy + 2]; - ref_cache[2 - 1*8]= - ref_cache[3 - 1*8]= ref[4*top_xy + 3]; - }else{ - AV_ZERO128(mv_cache[0 - 1*8]); - AV_WN32A(&ref_cache[0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101); - } - - if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){ - for(i=0; i<2; i++){ - int cache_idx = -1 + i*2*8; - if(USES_LIST(left_type[LEFT(i)], list)){ - const int b_xy= h->mb2b_xy[left_xy[LEFT(i)]] + 3; - const int b8_xy= 4*left_xy[LEFT(i)] + 1; - AV_COPY32(mv_cache[cache_idx ], mv[b_xy + b_stride*left_block[0+i*2]]); - AV_COPY32(mv_cache[cache_idx+8], mv[b_xy + b_stride*left_block[1+i*2]]); - ref_cache[cache_idx ]= ref[b8_xy + (left_block[0+i*2]&~1)]; - ref_cache[cache_idx+8]= ref[b8_xy + (left_block[1+i*2]&~1)]; - }else{ - AV_ZERO32(mv_cache[cache_idx ]); - AV_ZERO32(mv_cache[cache_idx+8]); - ref_cache[cache_idx ]= - ref_cache[cache_idx+8]= (left_type[LEFT(i)]) ? LIST_NOT_USED : PART_NOT_AVAILABLE; - } - } - }else{ - if(USES_LIST(left_type[LTOP], list)){ - const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3; - const int b8_xy= 4*left_xy[LTOP] + 1; - AV_COPY32(mv_cache[-1], mv[b_xy + b_stride*left_block[0]]); - ref_cache[-1]= ref[b8_xy + (left_block[0]&~1)]; - }else{ - AV_ZERO32(mv_cache[-1]); - ref_cache[-1]= left_type[LTOP] ? LIST_NOT_USED : PART_NOT_AVAILABLE; - } - } - - if(USES_LIST(topright_type, list)){ - const int b_xy= h->mb2b_xy[topright_xy] + 3*b_stride; - AV_COPY32(mv_cache[4 - 1*8], mv[b_xy]); - ref_cache[4 - 1*8]= ref[4*topright_xy + 2]; - }else{ - AV_ZERO32(mv_cache[4 - 1*8]); - ref_cache[4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; - } - if(ref_cache[4 - 1*8] < 0){ - if(USES_LIST(topleft_type, list)){ - const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride); - const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2); - AV_COPY32(mv_cache[-1 - 1*8], mv[b_xy]); - ref_cache[-1 - 1*8]= ref[b8_xy]; - }else{ - AV_ZERO32(mv_cache[-1 - 1*8]); - ref_cache[-1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; - } - } - - if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF) - continue; - - if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))){ - uint8_t (*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]]; - uint8_t (*mvd)[2] = h->mvd_table[list]; - ref_cache[2+8*0] = - ref_cache[2+8*2] = PART_NOT_AVAILABLE; - AV_ZERO32(mv_cache[2+8*0]); - AV_ZERO32(mv_cache[2+8*2]); - - if( CABAC ) { - if(USES_LIST(top_type, list)){ - const int b_xy= h->mb2br_xy[top_xy]; - AV_COPY64(mvd_cache[0 - 1*8], mvd[b_xy + 0]); - }else{ - AV_ZERO64(mvd_cache[0 - 1*8]); - } - if(USES_LIST(left_type[LTOP], list)){ - const int b_xy= h->mb2br_xy[left_xy[LTOP]] + 6; - AV_COPY16(mvd_cache[-1 + 0*8], mvd[b_xy - left_block[0]]); - AV_COPY16(mvd_cache[-1 + 1*8], mvd[b_xy - left_block[1]]); - }else{ - AV_ZERO16(mvd_cache[-1 + 0*8]); - AV_ZERO16(mvd_cache[-1 + 1*8]); - } - if(USES_LIST(left_type[LBOT], list)){ - const int b_xy= h->mb2br_xy[left_xy[LBOT]] + 6; - AV_COPY16(mvd_cache[-1 + 2*8], mvd[b_xy - left_block[2]]); - AV_COPY16(mvd_cache[-1 + 3*8], mvd[b_xy - left_block[3]]); - }else{ - AV_ZERO16(mvd_cache[-1 + 2*8]); - AV_ZERO16(mvd_cache[-1 + 3*8]); - } - AV_ZERO16(mvd_cache[2+8*0]); - AV_ZERO16(mvd_cache[2+8*2]); - if(h->slice_type_nos == AV_PICTURE_TYPE_B){ - uint8_t *direct_cache = &h->direct_cache[scan8[0]]; - uint8_t *direct_table = h->direct_table; - fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16>>1, 1); - - if(IS_DIRECT(top_type)){ - AV_WN32A(&direct_cache[-1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1)); - }else if(IS_8X8(top_type)){ - int b8_xy = 4*top_xy; - direct_cache[0 - 1*8]= direct_table[b8_xy + 2]; - direct_cache[2 - 1*8]= direct_table[b8_xy + 3]; - }else{ - AV_WN32A(&direct_cache[-1*8], 0x01010101*(MB_TYPE_16x16>>1)); - } - - if(IS_DIRECT(left_type[LTOP])) - direct_cache[-1 + 0*8]= MB_TYPE_DIRECT2>>1; - else if(IS_8X8(left_type[LTOP])) - direct_cache[-1 + 0*8]= direct_table[4*left_xy[LTOP] + 1 + (left_block[0]&~1)]; - else - direct_cache[-1 + 0*8]= MB_TYPE_16x16>>1; - - if(IS_DIRECT(left_type[LBOT])) - direct_cache[-1 + 2*8]= MB_TYPE_DIRECT2>>1; - else if(IS_8X8(left_type[LBOT])) - direct_cache[-1 + 2*8]= direct_table[4*left_xy[LBOT] + 1 + (left_block[2]&~1)]; - else - direct_cache[-1 + 2*8]= MB_TYPE_16x16>>1; - } - } - } - if(FRAME_MBAFF){ -#define MAP_MVS\ - MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\ - MAP_F2F(scan8[0] + 0 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 1 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 2 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 3 - 1*8, top_type)\ - MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\ - MAP_F2F(scan8[0] - 1 + 0*8, left_type[LTOP])\ - MAP_F2F(scan8[0] - 1 + 1*8, left_type[LTOP])\ - MAP_F2F(scan8[0] - 1 + 2*8, left_type[LBOT])\ - MAP_F2F(scan8[0] - 1 + 3*8, left_type[LBOT]) - if(MB_FIELD){ -#define MAP_F2F(idx, mb_type)\ - if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ - h->ref_cache[list][idx] <<= 1;\ - h->mv_cache[list][idx][1] /= 2;\ - h->mvd_cache[list][idx][1] >>=1;\ - } - MAP_MVS -#undef MAP_F2F - }else{ -#define MAP_F2F(idx, mb_type)\ - if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ - h->ref_cache[list][idx] >>= 1;\ - h->mv_cache[list][idx][1] <<= 1;\ - h->mvd_cache[list][idx][1] <<= 1;\ - } - MAP_MVS -#undef MAP_F2F - } - } - } - } - - h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]); -} - /** * gets the predicted intra4x4 prediction mode. */ diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h index 227dc4a..d2677ea 100644 --- a/libavcodec/h264_mvpred.h +++ b/libavcodec/h264_mvpred.h @@ -327,6 +327,424 @@ zeromv: return; } +static void fill_decode_neighbors(H264Context *h, int mb_type){ + MpegEncContext * const s = &h->s; + const int mb_xy= h->mb_xy; + int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; + static const uint8_t left_block_options[4][32]={ + {0,1,2,3,7,10,8,11,3+0*4, 3+1*4, 3+2*4, 3+3*4, 1+4*4, 1+8*4, 1+5*4, 1+9*4}, + {2,2,3,3,8,11,8,11,3+2*4, 3+2*4, 3+3*4, 3+3*4, 1+5*4, 1+9*4, 1+5*4, 1+9*4}, + {0,0,1,1,7,10,7,10,3+0*4, 3+0*4, 3+1*4, 3+1*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4}, + {0,2,0,2,7,10,7,10,3+0*4, 3+2*4, 3+0*4, 3+2*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4} + }; + + h->topleft_partition= -1; + + top_xy = mb_xy - (s->mb_stride << MB_FIELD); + + /* Wow, what a mess, why didn't they simplify the interlacing & intra + * stuff, I can't imagine that these complex rules are worth it. */ + + topleft_xy = top_xy - 1; + topright_xy= top_xy + 1; + left_xy[LBOT] = left_xy[LTOP] = mb_xy-1; + h->left_block = left_block_options[0]; + if(FRAME_MBAFF){ + const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]); + const int curr_mb_field_flag = IS_INTERLACED(mb_type); + if(s->mb_y&1){ + if (left_mb_field_flag != curr_mb_field_flag) { + left_xy[LBOT] = left_xy[LTOP] = mb_xy - s->mb_stride - 1; + if (curr_mb_field_flag) { + left_xy[LBOT] += s->mb_stride; + h->left_block = left_block_options[3]; + } else { + topleft_xy += s->mb_stride; + // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition + h->topleft_partition = 0; + h->left_block = left_block_options[1]; + } + } + }else{ + if(curr_mb_field_flag){ + topleft_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy - 1] >> 7) & 1) - 1); + topright_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy + 1] >> 7) & 1) - 1); + top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy ] >> 7) & 1) - 1); + } + if (left_mb_field_flag != curr_mb_field_flag) { + if (curr_mb_field_flag) { + left_xy[LBOT] += s->mb_stride; + h->left_block = left_block_options[3]; + } else { + h->left_block = left_block_options[2]; + } + } + } + } + + h->topleft_mb_xy = topleft_xy; + h->top_mb_xy = top_xy; + h->topright_mb_xy= topright_xy; + h->left_mb_xy[LTOP] = left_xy[LTOP]; + h->left_mb_xy[LBOT] = left_xy[LBOT]; + //FIXME do we need all in the context? + + h->topleft_type = s->current_picture.f.mb_type[topleft_xy]; + h->top_type = s->current_picture.f.mb_type[top_xy]; + h->topright_type = s->current_picture.f.mb_type[topright_xy]; + h->left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]]; + h->left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]]; + + if(FMO){ + if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0; + if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0; + if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0; + }else{ + if(h->slice_table[topleft_xy ] != h->slice_num){ + h->topleft_type = 0; + if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0; + if(h->slice_table[left_xy[LTOP] ] != h->slice_num) h->left_type[LTOP] = h->left_type[LBOT] = 0; + } + } + if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0; +} + +static void fill_decode_caches(H264Context *h, int mb_type){ + MpegEncContext * const s = &h->s; + int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; + int topleft_type, top_type, topright_type, left_type[LEFT_MBS]; + const uint8_t * left_block= h->left_block; + int i; + uint8_t *nnz; + uint8_t *nnz_cache; + + topleft_xy = h->topleft_mb_xy; + top_xy = h->top_mb_xy; + topright_xy = h->topright_mb_xy; + left_xy[LTOP] = h->left_mb_xy[LTOP]; + left_xy[LBOT] = h->left_mb_xy[LBOT]; + topleft_type = h->topleft_type; + top_type = h->top_type; + topright_type = h->topright_type; + left_type[LTOP]= h->left_type[LTOP]; + left_type[LBOT]= h->left_type[LBOT]; + + if(!IS_SKIP(mb_type)){ + if(IS_INTRA(mb_type)){ + int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1; + h->topleft_samples_available= + h->top_samples_available= + h->left_samples_available= 0xFFFF; + h->topright_samples_available= 0xEEEA; + + if(!(top_type & type_mask)){ + h->topleft_samples_available= 0xB3FF; + h->top_samples_available= 0x33FF; + h->topright_samples_available= 0x26EA; + } + if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[LTOP])){ + if(IS_INTERLACED(mb_type)){ + if(!(left_type[LTOP] & type_mask)){ + h->topleft_samples_available&= 0xDFFF; + h->left_samples_available&= 0x5FFF; + } + if(!(left_type[LBOT] & type_mask)){ + h->topleft_samples_available&= 0xFF5F; + h->left_samples_available&= 0xFF5F; + } + }else{ + int left_typei = s->current_picture.f.mb_type[left_xy[LTOP] + s->mb_stride]; + + assert(left_xy[LTOP] == left_xy[LBOT]); + if(!((left_typei & type_mask) && (left_type[LTOP] & type_mask))){ + h->topleft_samples_available&= 0xDF5F; + h->left_samples_available&= 0x5F5F; + } + } + }else{ + if(!(left_type[LTOP] & type_mask)){ + h->topleft_samples_available&= 0xDF5F; + h->left_samples_available&= 0x5F5F; + } + } + + if(!(topleft_type & type_mask)) + h->topleft_samples_available&= 0x7FFF; + + if(!(topright_type & type_mask)) + h->topright_samples_available&= 0xFBFF; + + if(IS_INTRA4x4(mb_type)){ + if(IS_INTRA4x4(top_type)){ + AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]); + }else{ + h->intra4x4_pred_mode_cache[4+8*0]= + h->intra4x4_pred_mode_cache[5+8*0]= + h->intra4x4_pred_mode_cache[6+8*0]= + h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask); + } + for(i=0; i<2; i++){ + if(IS_INTRA4x4(left_type[LEFT(i)])){ + int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[LEFT(i)]]; + h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]]; + h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]]; + }else{ + h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= + h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[LEFT(i)] & type_mask); + } + } + } + } + + +/* +0 . T T. T T T T +1 L . .L . . . . +2 L . .L . . . . +3 . T TL . . . . +4 L . .L . . . . +5 L . .. . . . . +*/ +//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec) + nnz_cache = h->non_zero_count_cache; + if(top_type){ + nnz = h->non_zero_count[top_xy]; + AV_COPY32(&nnz_cache[4+8* 0], &nnz[4*3]); + if(CHROMA444){ + AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 7]); + AV_COPY32(&nnz_cache[4+8*10], &nnz[4*11]); + }else{ + AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 5]); + AV_COPY32(&nnz_cache[4+8*10], &nnz[4* 9]); + } + }else{ + uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040; + AV_WN32A(&nnz_cache[4+8* 0], top_empty); + AV_WN32A(&nnz_cache[4+8* 5], top_empty); + AV_WN32A(&nnz_cache[4+8*10], top_empty); + } + + for (i=0; i<2; i++) { + if(left_type[LEFT(i)]){ + nnz = h->non_zero_count[left_xy[LEFT(i)]]; + nnz_cache[3+8* 1 + 2*8*i]= nnz[left_block[8+0+2*i]]; + nnz_cache[3+8* 2 + 2*8*i]= nnz[left_block[8+1+2*i]]; + if(CHROMA444){ + nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]+4*4]; + nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]+4*4]; + nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]+8*4]; + nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]+8*4]; + }else{ + nnz_cache[3+8* 6 + 8*i]= nnz[left_block[8+4+2*i]]; + nnz_cache[3+8*11 + 8*i]= nnz[left_block[8+5+2*i]]; + } + }else{ + nnz_cache[3+8* 1 + 2*8*i]= + nnz_cache[3+8* 2 + 2*8*i]= + nnz_cache[3+8* 6 + 2*8*i]= + nnz_cache[3+8* 7 + 2*8*i]= + nnz_cache[3+8*11 + 2*8*i]= + nnz_cache[3+8*12 + 2*8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64; + } + } + + if( CABAC ) { + // top_cbp + if(top_type) { + h->top_cbp = h->cbp_table[top_xy]; + } else { + h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; + } + // left_cbp + if (left_type[LTOP]) { + h->left_cbp = (h->cbp_table[left_xy[LTOP]] & 0x7F0) + | ((h->cbp_table[left_xy[LTOP]]>>(left_block[0]&(~1)))&2) + | (((h->cbp_table[left_xy[LBOT]]>>(left_block[2]&(~1)))&2) << 2); + } else { + h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F; + } + } + } + + if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){ + int list; + int b_stride = h->b_stride; + for(list=0; list<h->list_count; list++){ + int8_t *ref_cache = &h->ref_cache[list][scan8[0]]; + int8_t *ref = s->current_picture.f.ref_index[list]; + int16_t (*mv_cache)[2] = &h->mv_cache[list][scan8[0]]; + int16_t (*mv)[2] = s->current_picture.f.motion_val[list]; + if(!USES_LIST(mb_type, list)){ + continue; + } + assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)); + + if(USES_LIST(top_type, list)){ + const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride; + AV_COPY128(mv_cache[0 - 1*8], mv[b_xy + 0]); + ref_cache[0 - 1*8]= + ref_cache[1 - 1*8]= ref[4*top_xy + 2]; + ref_cache[2 - 1*8]= + ref_cache[3 - 1*8]= ref[4*top_xy + 3]; + }else{ + AV_ZERO128(mv_cache[0 - 1*8]); + AV_WN32A(&ref_cache[0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101); + } + + if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){ + for(i=0; i<2; i++){ + int cache_idx = -1 + i*2*8; + if(USES_LIST(left_type[LEFT(i)], list)){ + const int b_xy= h->mb2b_xy[left_xy[LEFT(i)]] + 3; + const int b8_xy= 4*left_xy[LEFT(i)] + 1; + AV_COPY32(mv_cache[cache_idx ], mv[b_xy + b_stride*left_block[0+i*2]]); + AV_COPY32(mv_cache[cache_idx+8], mv[b_xy + b_stride*left_block[1+i*2]]); + ref_cache[cache_idx ]= ref[b8_xy + (left_block[0+i*2]&~1)]; + ref_cache[cache_idx+8]= ref[b8_xy + (left_block[1+i*2]&~1)]; + }else{ + AV_ZERO32(mv_cache[cache_idx ]); + AV_ZERO32(mv_cache[cache_idx+8]); + ref_cache[cache_idx ]= + ref_cache[cache_idx+8]= (left_type[LEFT(i)]) ? LIST_NOT_USED : PART_NOT_AVAILABLE; + } + } + }else{ + if(USES_LIST(left_type[LTOP], list)){ + const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3; + const int b8_xy= 4*left_xy[LTOP] + 1; + AV_COPY32(mv_cache[-1], mv[b_xy + b_stride*left_block[0]]); + ref_cache[-1]= ref[b8_xy + (left_block[0]&~1)]; + }else{ + AV_ZERO32(mv_cache[-1]); + ref_cache[-1]= left_type[LTOP] ? LIST_NOT_USED : PART_NOT_AVAILABLE; + } + } + + if(USES_LIST(topright_type, list)){ + const int b_xy= h->mb2b_xy[topright_xy] + 3*b_stride; + AV_COPY32(mv_cache[4 - 1*8], mv[b_xy]); + ref_cache[4 - 1*8]= ref[4*topright_xy + 2]; + }else{ + AV_ZERO32(mv_cache[4 - 1*8]); + ref_cache[4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; + } + if(ref_cache[4 - 1*8] < 0){ + if(USES_LIST(topleft_type, list)){ + const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride); + const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2); + AV_COPY32(mv_cache[-1 - 1*8], mv[b_xy]); + ref_cache[-1 - 1*8]= ref[b8_xy]; + }else{ + AV_ZERO32(mv_cache[-1 - 1*8]); + ref_cache[-1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; + } + } + + if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF) + continue; + + if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))){ + uint8_t (*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]]; + uint8_t (*mvd)[2] = h->mvd_table[list]; + ref_cache[2+8*0] = + ref_cache[2+8*2] = PART_NOT_AVAILABLE; + AV_ZERO32(mv_cache[2+8*0]); + AV_ZERO32(mv_cache[2+8*2]); + + if( CABAC ) { + if(USES_LIST(top_type, list)){ + const int b_xy= h->mb2br_xy[top_xy]; + AV_COPY64(mvd_cache[0 - 1*8], mvd[b_xy + 0]); + }else{ + AV_ZERO64(mvd_cache[0 - 1*8]); + } + if(USES_LIST(left_type[LTOP], list)){ + const int b_xy= h->mb2br_xy[left_xy[LTOP]] + 6; + AV_COPY16(mvd_cache[-1 + 0*8], mvd[b_xy - left_block[0]]); + AV_COPY16(mvd_cache[-1 + 1*8], mvd[b_xy - left_block[1]]); + }else{ + AV_ZERO16(mvd_cache[-1 + 0*8]); + AV_ZERO16(mvd_cache[-1 + 1*8]); + } + if(USES_LIST(left_type[LBOT], list)){ + const int b_xy= h->mb2br_xy[left_xy[LBOT]] + 6; + AV_COPY16(mvd_cache[-1 + 2*8], mvd[b_xy - left_block[2]]); + AV_COPY16(mvd_cache[-1 + 3*8], mvd[b_xy - left_block[3]]); + }else{ + AV_ZERO16(mvd_cache[-1 + 2*8]); + AV_ZERO16(mvd_cache[-1 + 3*8]); + } + AV_ZERO16(mvd_cache[2+8*0]); + AV_ZERO16(mvd_cache[2+8*2]); + if(h->slice_type_nos == AV_PICTURE_TYPE_B){ + uint8_t *direct_cache = &h->direct_cache[scan8[0]]; + uint8_t *direct_table = h->direct_table; + fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16>>1, 1); + + if(IS_DIRECT(top_type)){ + AV_WN32A(&direct_cache[-1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1)); + }else if(IS_8X8(top_type)){ + int b8_xy = 4*top_xy; + direct_cache[0 - 1*8]= direct_table[b8_xy + 2]; + direct_cache[2 - 1*8]= direct_table[b8_xy + 3]; + }else{ + AV_WN32A(&direct_cache[-1*8], 0x01010101*(MB_TYPE_16x16>>1)); + } + + if(IS_DIRECT(left_type[LTOP])) + direct_cache[-1 + 0*8]= MB_TYPE_DIRECT2>>1; + else if(IS_8X8(left_type[LTOP])) + direct_cache[-1 + 0*8]= direct_table[4*left_xy[LTOP] + 1 + (left_block[0]&~1)]; + else + direct_cache[-1 + 0*8]= MB_TYPE_16x16>>1; + + if(IS_DIRECT(left_type[LBOT])) + direct_cache[-1 + 2*8]= MB_TYPE_DIRECT2>>1; + else if(IS_8X8(left_type[LBOT])) + direct_cache[-1 + 2*8]= direct_table[4*left_xy[LBOT] + 1 + (left_block[2]&~1)]; + else + direct_cache[-1 + 2*8]= MB_TYPE_16x16>>1; + } + } + } + if(FRAME_MBAFF){ +#define MAP_MVS\ + MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\ + MAP_F2F(scan8[0] + 0 - 1*8, top_type)\ + MAP_F2F(scan8[0] + 1 - 1*8, top_type)\ + MAP_F2F(scan8[0] + 2 - 1*8, top_type)\ + MAP_F2F(scan8[0] + 3 - 1*8, top_type)\ + MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\ + MAP_F2F(scan8[0] - 1 + 0*8, left_type[LTOP])\ + MAP_F2F(scan8[0] - 1 + 1*8, left_type[LTOP])\ + MAP_F2F(scan8[0] - 1 + 2*8, left_type[LBOT])\ + MAP_F2F(scan8[0] - 1 + 3*8, left_type[LBOT]) + if(MB_FIELD){ +#define MAP_F2F(idx, mb_type)\ + if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ + h->ref_cache[list][idx] <<= 1;\ + h->mv_cache[list][idx][1] /= 2;\ + h->mvd_cache[list][idx][1] >>=1;\ + } + MAP_MVS +#undef MAP_F2F + }else{ +#define MAP_F2F(idx, mb_type)\ + if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ + h->ref_cache[list][idx] >>= 1;\ + h->mv_cache[list][idx][1] <<= 1;\ + h->mvd_cache[list][idx][1] <<= 1;\ + } + MAP_MVS +#undef MAP_F2F + } + } + } + } + + h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[LTOP]); +} + /** * decodes a P_SKIP or B_SKIP macroblock */ diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c index 9488a0f..c272088 100644 --- a/libavcodec/lpc.c +++ b/libavcodec/lpc.c @@ -149,10 +149,8 @@ static int estimate_best_order(double *ref, int min_order, int max_order) /** * Calculate LPC coefficients for multiple orders * - * @param lpc_type LPC method for determining coefficients - * 0 = LPC with fixed pre-defined coeffs - * 1 = LPC with coeffs determined by Levinson-Durbin recursion - * 2+ = LPC with coeffs determined by Cholesky factorization using (use_lpc-1) passes. + * @param lpc_type LPC method for determining coefficients, + * see #FFLPCType for details */ int ff_lpc_calc_coefs(LPCContext *s, const int32_t *samples, int blocksize, int min_order, diff --git a/libavcodec/msmpeg4.c b/libavcodec/msmpeg4.c index 07b27bc..62927ac 100644 --- a/libavcodec/msmpeg4.c +++ b/libavcodec/msmpeg4.c @@ -1892,7 +1892,7 @@ AVCodec ff_msmpeg4v1_decoder = { NULL, ff_h263_decode_end, ff_h263_decode_frame, - CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | CODEC_CAP_EXPERIMENTAL, + CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1, .max_lowres= 3, .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"), .pix_fmts= ff_pixfmt_list_420, diff --git a/libavcodec/version.h b/libavcodec/version.h index de67bd4..475e748 100644 --- a/libavcodec/version.h +++ b/libavcodec/version.h @@ -21,7 +21,7 @@ #define AVCODEC_VERSION_H #define LIBAVCODEC_VERSION_MAJOR 53 -#define LIBAVCODEC_VERSION_MINOR 7 +#define LIBAVCODEC_VERSION_MINOR 8 #define LIBAVCODEC_VERSION_MICRO 0 #define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \ diff --git a/libavcodec/x86/h264_intrapred_10bit.asm b/libavcodec/x86/h264_intrapred_10bit.asm index d57fc79..24a7bfa 100644 --- a/libavcodec/x86/h264_intrapred_10bit.asm +++ b/libavcodec/x86/h264_intrapred_10bit.asm @@ -29,11 +29,19 @@ SECTION_RODATA SECTION .text +cextern pw_16 cextern pw_8 cextern pw_4 cextern pw_2 cextern pw_1 +pw_m32101234: dw -3, -2, -1, 0, 1, 2, 3, 4 +pw_m3: times 8 dw -3 +pw_pixel_max: times 8 dw ((1 << 10)-1) +pw_512: times 8 dw 512 +pd_17: times 4 dd 17 +pd_16: times 4 dd 16 + ; dest, left, right, src ; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2 %macro PRED4x4_LOWPASS 4 @@ -464,7 +472,92 @@ PRED8x8_TOP_DC mmxext, pshufw INIT_XMM PRED8x8_TOP_DC sse2 , pshuflw +;----------------------------------------------------------------------------- +; void pred8x8_plane(pixel *src, int stride) +;----------------------------------------------------------------------------- +INIT_XMM +cglobal pred8x8_plane_10_sse2, 2,7,7 + sub r0, r1 + lea r2, [r1+r1*2] + lea r3, [r0+r1*4] + mova m2, [r0] + pmaddwd m2, [pw_m32101234] + HADDD m2, m1 + movd m0, [r0-4] + psrld m0, 14 + psubw m2, m0 ; H + movd m0, [r3+r1*4-4] + movd m1, [r0+12] + paddw m0, m1 + psllw m0, 4 ; 16*(src[7*stride-1] + src[-stride+7]) + movzx r4d, word [r3+r1*1-2] ; src[4*stride-1] + movzx r5d, word [r0+r2*1-2] ; src[2*stride-1] + sub r4d, r5d + movzx r6d, word [r3+r1*2-2] ; src[5*stride-1] + movzx r5d, word [r0+r1*2-2] ; src[1*stride-1] + sub r6d, r5d + lea r4d, [r4+r6*2] + movzx r5d, word [r3+r2*1-2] ; src[6*stride-1] + movzx r6d, word [r0+r1*1-2] ; src[0*stride-1] + sub r5d, r6d + lea r5d, [r5+r5*2] + add r4d, r5d + movzx r6d, word [r3+r1*4-2] ; src[7*stride-1] + movzx r5d, word [r0+r1*0-2] ; src[ -stride-1] + sub r6d, r5d + lea r4d, [r4+r6*4] + movd m3, r4d ; V + punpckldq m2, m3 + pmaddwd m2, [pd_17] + paddd m2, [pd_16] + psrad m2, 5 ; b, c + mova m3, [pw_pixel_max] + pxor m1, m1 + SPLATW m0, m0, 1 + SPLATW m4, m2, 2 + SPLATW m2, m2, 0 + pmullw m2, [pw_m32101234] ; b + pmullw m5, m4, [pw_m3] ; c + paddw m5, [pw_16] + mov r2d, 8 + add r0, r1 +.loop: + paddsw m6, m2, m5 + paddsw m6, m0 + psraw m6, 5 + CLIPW m6, m1, m3 + mova [r0], m6 + paddw m5, m4 + add r0, r1 + dec r2d + jg .loop + REP_RET + + +;----------------------------------------------------------------------------- +; void pred8x8l_128_dc(pixel *src, int has_topleft, int has_topright, int stride) +;----------------------------------------------------------------------------- +%macro PRED8x8L_128_DC 1 +cglobal pred8x8l_128_dc_10_%1, 4,4 + mova m0, [pw_512] + lea r1, [r3+r3*2] + lea r2, [r0+r3*4] + MOV8 r0+r3*0, m0, m0 + MOV8 r0+r3*1, m0, m0 + MOV8 r0+r3*2, m0, m0 + MOV8 r0+r1*1, m0, m0 + MOV8 r2+r3*0, m0, m0 + MOV8 r2+r3*1, m0, m0 + MOV8 r2+r3*2, m0, m0 + MOV8 r2+r1*1, m0, m0 + RET +%endmacro + +INIT_MMX +PRED8x8L_128_DC mmxext +INIT_XMM +PRED8x8L_128_DC sse2 ;----------------------------------------------------------------------------- ; void pred8x8l_top_dc(pixel *src, int has_topleft, int has_topright, int stride) @@ -1258,7 +1351,7 @@ cglobal pred16x16_horizontal_10_%1, 2,3 MOV16 r0+r1*1, m1, m1, m1, m1 lea r0, [r0+r1*2] dec r2 - jge .vloop + jg .vloop REP_RET %endmacro @@ -1266,3 +1359,139 @@ INIT_MMX PRED16x16_HORIZONTAL mmxext INIT_XMM PRED16x16_HORIZONTAL sse2 + +;----------------------------------------------------------------------------- +; void pred16x16_dc(pixel *src, int stride) +;----------------------------------------------------------------------------- +%macro PRED16x16_DC 1 +cglobal pred16x16_dc_10_%1, 2,7 + mov r4, r0 + sub r0, r1 + mova m0, [r0+0] + paddw m0, [r0+mmsize] +%if mmsize==8 + paddw m0, [r0+16] + paddw m0, [r0+24] +%endif + HADDW m0, m2 + + sub r0, 2 + movzx r3d, word [r0+r1*1] + movzx r5d, word [r0+r1*2] +%rep 7 + lea r0, [r0+r1*2] + movzx r2d, word [r0+r1*1] + add r3d, r2d + movzx r2d, word [r0+r1*2] + add r5d, r2d +%endrep + lea r3d, [r3+r5+16] + + movd m1, r3d + paddw m0, m1 + psrlw m0, 5 + SPLATW m0, m0 + mov r3d, 8 +.loop: + MOV16 r4+r1*0, m0, m0, m0, m0 + MOV16 r4+r1*1, m0, m0, m0, m0 + lea r4, [r4+r1*2] + dec r3d + jg .loop + REP_RET +%endmacro + +INIT_MMX +PRED16x16_DC mmxext +INIT_XMM +PRED16x16_DC sse2 + +;----------------------------------------------------------------------------- +; void pred16x16_top_dc(pixel *src, int stride) +;----------------------------------------------------------------------------- +%macro PRED16x16_TOP_DC 1 +cglobal pred16x16_top_dc_10_%1, 2,3 + sub r0, r1 + mova m0, [r0+0] + paddw m0, [r0+mmsize] +%if mmsize==8 + paddw m0, [r0+16] + paddw m0, [r0+24] +%endif + HADDW m0, m2 + + SPLATW m0, m0 + paddw m0, [pw_8] + psrlw m0, 4 + mov r2d, 8 +.loop: + MOV16 r0+r1*1, m0, m0, m0, m0 + MOV16 r0+r1*2, m0, m0, m0, m0 + lea r0, [r0+r1*2] + dec r2d + jg .loop + REP_RET +%endmacro + +INIT_MMX +PRED16x16_TOP_DC mmxext +INIT_XMM +PRED16x16_TOP_DC sse2 + +;----------------------------------------------------------------------------- +; void pred16x16_left_dc(pixel *src, int stride) +;----------------------------------------------------------------------------- +%macro PRED16x16_LEFT_DC 1 +cglobal pred16x16_left_dc_10_%1, 2,7 + mov r4, r0 + + sub r0, 2 + movzx r5d, word [r0+r1*0] + movzx r6d, word [r0+r1*1] +%rep 7 + lea r0, [r0+r1*2] + movzx r2d, word [r0+r1*0] + movzx r3d, word [r0+r1*1] + add r5d, r2d + add r6d, r3d +%endrep + lea r2d, [r5+r6+8] + shr r2d, 4 + + movd m0, r2d + SPLATW m0, m0 + mov r3d, 8 +.loop: + MOV16 r4+r1*0, m0, m0, m0, m0 + MOV16 r4+r1*1, m0, m0, m0, m0 + lea r4, [r4+r1*2] + dec r3d + jg .loop + REP_RET +%endmacro + +INIT_MMX +PRED16x16_LEFT_DC mmxext +INIT_XMM +PRED16x16_LEFT_DC sse2 + +;----------------------------------------------------------------------------- +; void pred16x16_128_dc(pixel *src, int stride) +;----------------------------------------------------------------------------- +%macro PRED16x16_128_DC 1 +cglobal pred16x16_128_dc_10_%1, 2,3 + mova m0, [pw_512] + mov r2d, 8 +.loop: + MOV16 r0+r1*0, m0, m0, m0, m0 + MOV16 r0+r1*1, m0, m0, m0, m0 + lea r0, [r0+r1*2] + dec r2d + jg .loop + REP_RET +%endmacro + +INIT_MMX +PRED16x16_128_DC mmxext +INIT_XMM +PRED16x16_128_DC sse2 diff --git a/libavcodec/x86/h264_intrapred_init.c b/libavcodec/x86/h264_intrapred_init.c index 2ccc584..5153c4e 100644 --- a/libavcodec/x86/h264_intrapred_init.c +++ b/libavcodec/x86/h264_intrapred_init.c @@ -47,6 +47,7 @@ PRED8x8(dc, 10, mmxext) PRED8x8(dc, 10, sse2) PRED8x8(top_dc, 10, mmxext) PRED8x8(top_dc, 10, sse2) +PRED8x8(plane, 10, sse2) PRED8x8(vertical, 10, sse2) PRED8x8(horizontal, 10, sse2) @@ -55,6 +56,8 @@ void ff_pred8x8l_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int has_tople PRED8x8L(dc, 10, sse2) PRED8x8L(dc, 10, ssse3) +PRED8x8L(128_dc, 10, mmxext) +PRED8x8L(128_dc, 10, sse2) PRED8x8L(top_dc, 10, sse2) PRED8x8L(top_dc, 10, ssse3) PRED8x8L(vertical, 10, sse2) @@ -73,6 +76,14 @@ PRED8x8L(horizontal_up, 10, ssse3) #define PRED16x16(TYPE, DEPTH, OPT)\ void ff_pred16x16_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride); +PRED16x16(dc, 10, mmxext) +PRED16x16(dc, 10, sse2) +PRED16x16(top_dc, 10, mmxext) +PRED16x16(top_dc, 10, sse2) +PRED16x16(128_dc, 10, mmxext) +PRED16x16(128_dc, 10, sse2) +PRED16x16(left_dc, 10, mmxext) +PRED16x16(left_dc, 10, sse2) PRED16x16(vertical, 10, mmxext) PRED16x16(vertical, 10, sse2) PRED16x16(horizontal, 10, mmxext) @@ -289,6 +300,12 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_10_mmxext; h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_10_mmxext; + h->pred8x8l[DC_128_PRED ] = ff_pred8x8l_128_dc_10_mmxext; + + h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_10_mmxext; + h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_10_mmxext; + h->pred16x16[DC_128_PRED8x8 ] = ff_pred16x16_128_dc_10_mmxext; + h->pred16x16[LEFT_DC_PRED8x8 ] = ff_pred16x16_left_dc_10_mmxext; h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_10_mmxext; h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_10_mmxext; } @@ -301,18 +318,24 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_10_sse2; h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_10_sse2; + h->pred8x8[PLANE_PRED8x8 ] = ff_pred8x8_plane_10_sse2; h->pred8x8[VERT_PRED8x8 ] = ff_pred8x8_vertical_10_sse2; h->pred8x8[HOR_PRED8x8 ] = ff_pred8x8_horizontal_10_sse2; h->pred8x8l[VERT_PRED ] = ff_pred8x8l_vertical_10_sse2; h->pred8x8l[HOR_PRED ] = ff_pred8x8l_horizontal_10_sse2; h->pred8x8l[DC_PRED ] = ff_pred8x8l_dc_10_sse2; + h->pred8x8l[DC_128_PRED ] = ff_pred8x8l_128_dc_10_sse2; h->pred8x8l[TOP_DC_PRED ] = ff_pred8x8l_top_dc_10_sse2; h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_sse2; h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_sse2; h->pred8x8l[VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_10_sse2; h->pred8x8l[HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_10_sse2; + h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_10_sse2; + h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_10_sse2; + h->pred16x16[DC_128_PRED8x8 ] = ff_pred16x16_128_dc_10_sse2; + h->pred16x16[LEFT_DC_PRED8x8 ] = ff_pred16x16_left_dc_10_sse2; h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_10_sse2; h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_10_sse2; } |
