left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
- if(FRAME_MBAFF && !IS_INTRA(mb_type)){
+ if(MB_MBAFF && !IS_INTRA(mb_type)){
int list;
- int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
- for(i=0; i<16; i++)
- h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
for(list=0; list<h->list_count; list++){
+ //These values where changed for ease of performing MC, we need to change them back
+ //FIXME maybe we can make MC and loop filter use the same values or prevent
+ //the MC code from changing ref_cache and rather use a temporary array.
if(USES_LIST(mb_type,list)){
- uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
- uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
- for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
- dst[0] = src[0];
- dst[1] = src[1];
- dst[2] = src[2];
- dst[3] = src[3];
- }
*(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
- *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
+ *(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])*0x0101;
- }else{
- fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
- fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
+ *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
}
}
}
}
}
- if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
+ if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
continue;
if(USES_LIST(topleft_type, list)){
h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
-
- if(FRAME_MBAFF){
- // store all luma nnzs, for deblocking
- int v = 0, i;
- for(i=0; i<16; i++)
- v += (!!h->non_zero_count_cache[scan8[i]]) << i;
- *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
- }
}
/**
int *col_poc = h->ref_list[1]->field_poc;
int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
- l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
- l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
- l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
- l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
- if(s->mb_y&1){
- l1ref0 += b8_stride;
- l1ref1 += b8_stride;
- l1mv0 += 2*b4_stride;
- l1mv1 += 2*b4_stride;
- }
b8_stride = 0;
}
}else if(!(s->picture_structure & h->ref_list[1][0].reference)){// FL -> FL & differ parity
single_col:
mb_type_col[0] =
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
- if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
- /* FIXME save sub mb types from previous frames (or derive from MVs)
- * so we know exactly what block size to use */
- sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
- *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
- }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
- sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
- *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
- }else{
- sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
- *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
- }
+ if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
+ /* FIXME save sub mb types from previous frames (or derive from MVs)
+ * so we know exactly what block size to use */
+ sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
+ *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
+ }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
+ sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
+ *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
+ }else{
+ sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
+ *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
+ }
}
}
- if(b8_stride){
- l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
- l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
- l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
- l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
+ l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
+ l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
+ l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
+ l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
+ if(!b8_stride){
+ if(s->mb_y&1){
+ l1ref0 += h->b8_stride;
+ l1ref1 += h->b8_stride;
+ l1mv0 += 2*b4_stride;
+ l1mv1 += 2*b4_stride;
+ }
}
if(h->direct_spatial_mv_pred){
fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
/* col_zero_flag */
- if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*h->b8_stride] == 0
- || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
+ if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*b8_stride] == 0
+ || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
&& (h->x264_build>33 || !h->x264_build)))){
- const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
+ const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
if(IS_SUB_8X8(sub_mb_type)){
- const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
+ const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
if(ref[0] == 0)
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
}
}else
for(i4=0; i4<4; i4++){
- const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
+ const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
if(ref[0] == 0)
*(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
const int *dist_scale_factor = h->dist_scale_factor;
- if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
- map_col_to_list0[0] = h->map_col_to_list0_field[0];
- map_col_to_list0[1] = h->map_col_to_list0_field[1];
- dist_scale_factor = h->dist_scale_factor_field;
- }
- if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
- /* FIXME assumes direct_8x8_inference == 1 */
- int y_shift;
- int ref_shift;
+ if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
+ map_col_to_list0[0] = h->map_col_to_list0_field[0];
+ map_col_to_list0[1] = h->map_col_to_list0_field[1];
+ dist_scale_factor = h->dist_scale_factor_field;
+ }
+ if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
+ /* FIXME assumes direct_8x8_inference == 1 */
+ int y_shift = 2*!IS_INTERLACED(*mb_type);
+ int ref_shift= FRAME_MBAFF ? y_shift : 1;
- if(IS_INTERLACED(*mb_type)){
- /* frame to field scaling */
- y_shift = 0;
- ref_shift= FRAME_MBAFF ? 0 : 1;
- }else{
- y_shift = 2;
- ref_shift= FRAME_MBAFF ? 2 : 1;
- }
+ for(i8=0; i8<4; i8++){
+ const int x8 = i8&1;
+ const int y8 = i8>>1;
+ int ref0, scale;
+ const int16_t (*l1mv)[2]= l1mv0;
- for(i8=0; i8<4; i8++){
- const int x8 = i8&1;
- const int y8 = i8>>1;
- int ref0, scale;
- const int16_t (*l1mv)[2]= l1mv0;
-
- if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
- continue;
- h->sub_mb_type[i8] = sub_mb_type;
-
- fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
- if(IS_INTRA(mb_type_col[y8])){
- fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
- fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
- fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
- continue;
- }
+ if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
+ continue;
+ h->sub_mb_type[i8] = sub_mb_type;
- ref0 = l1ref0[x8 + y8*b8_stride];
- if(ref0 >= 0)
- ref0 = map_col_to_list0[0][ref0*2>>ref_shift];
- else{
- ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]*2>>ref_shift];
- l1mv= l1mv1;
- }
- scale = dist_scale_factor[ref0];
- fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
+ fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
+ if(IS_INTRA(mb_type_col[y8])){
+ fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
+ fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
+ fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
+ continue;
+ }
- {
- const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
- int my_col = (mv_col[1]<<y_shift)/2;
- int mx = (scale * mv_col[0] + 128) >> 8;
- int my = (scale * my_col + 128) >> 8;
- fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
- fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
- }
+ ref0 = l1ref0[x8 + y8*b8_stride];
+ if(ref0 >= 0)
+ ref0 = map_col_to_list0[0][ref0*2>>ref_shift];
+ else{
+ ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]*2>>ref_shift];
+ l1mv= l1mv1;
+ }
+ scale = dist_scale_factor[ref0];
+ fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
+
+ {
+ const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
+ int my_col = (mv_col[1]<<y_shift)/2;
+ int mx = (scale * mv_col[0] + 128) >> 8;
+ int my = (scale * my_col + 128) >> 8;
+ fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
+ fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
}
- return;
}
+ return;
+ }
/* one-to-one mv scaling */
continue;
}
- ref0 = l1ref0[x8 + y8*h->b8_stride];
+ ref0 = l1ref0[x8 + y8*b8_stride];
if(ref0 >= 0)
ref0 = map_col_to_list0[0][ref0];
else{
- ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
+ ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]];
l1mv= l1mv1;
}
scale = dist_scale_factor[ref0];
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
if(IS_SUB_8X8(sub_mb_type)){
- const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
+ const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
int mx = (scale * mv_col[0] + 128) >> 8;
int my = (scale * mv_col[1] + 128) >> 8;
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
}else
for(i4=0; i4<4; i4++){
- const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
+ const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
MpegEncContext * const s = &h->s;
int i;
+ int step = 1;
+ int offset = 1;
+ int uvoffset= 1;
+ int top_idx = 1;
+ int skiplast= 0;
src_y -= linesize;
src_cb -= uvlinesize;
src_cr -= uvlinesize;
+ if(!simple && FRAME_MBAFF){
+ if(s->mb_y&1){
+ offset = MB_MBAFF ? 1 : 17;
+ uvoffset= MB_MBAFF ? 1 : 9;
+ if(!MB_MBAFF){
+ *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y + 15*linesize);
+ *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
+ if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
+ *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
+ *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
+ }
+ }
+ }else{
+ if(!MB_MBAFF){
+ h->left_border[0]= h->top_borders[0][s->mb_x][15];
+ if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
+ h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7 ];
+ h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
+ }
+ skiplast= 1;
+ }
+ offset =
+ uvoffset=
+ top_idx = MB_MBAFF ? 0 : 1;
+ }
+ step= MB_MBAFF ? 2 : 1;
+ }
+
// There are two lines saved, the line above the the top macroblock of a pair,
// and the line above the bottom macroblock
- h->left_border[0]= h->top_borders[0][s->mb_x][15];
- for(i=1; i<17; i++){
- h->left_border[i]= src_y[15+i* linesize];
+ h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
+ for(i=1; i<17 - skiplast; i++){
+ h->left_border[offset+i*step]= src_y[15+i* linesize];
}
- *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
- *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
+ *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
+ *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
- h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
- h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
- for(i=1; i<9; i++){
- h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
- h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
+ h->left_border[uvoffset+34 ]= h->top_borders[top_idx][s->mb_x][16+7];
+ h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
+ for(i=1; i<9 - skiplast; i++){
+ h->left_border[uvoffset+34 +i*step]= src_cb[7+i*uvlinesize];
+ h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
}
- *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
- *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
+ *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
+ *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
}
}
int deblock_left;
int deblock_top;
int mb_xy;
+ int step = 1;
+ int offset = 1;
+ int uvoffset= 1;
+ int top_idx = 1;
+
+ if(!simple && FRAME_MBAFF){
+ if(s->mb_y&1){
+ offset = MB_MBAFF ? 1 : 17;
+ uvoffset= MB_MBAFF ? 1 : 9;
+ }else{
+ offset =
+ uvoffset=
+ top_idx = MB_MBAFF ? 0 : 1;
+ }
+ step= MB_MBAFF ? 2 : 1;
+ }
if(h->deblocking_filter == 2) {
mb_xy = h->mb_xy;
b= t;
if(deblock_left){
- for(i = !deblock_top; i<17; i++){
- XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
+ for(i = !deblock_top; i<16; i++){
+ XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, xchg);
}
+ XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, 1);
}
if(deblock_top){
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
+ XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
+ XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
if(s->mb_x+1 < s->mb_width){
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
+ XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
}
}
if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
if(deblock_left){
- for(i = !deblock_top; i<9; i++){
- XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
- XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
- }
- }
- if(deblock_top){
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
- }
- }
-}
-
-static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
- MpegEncContext * const s = &h->s;
- int i;
-
- src_y -= 2 * linesize;
- src_cb -= 2 * uvlinesize;
- src_cr -= 2 * uvlinesize;
-
- // There are two lines saved, the line above the the top macroblock of a pair,
- // and the line above the bottom macroblock
- h->left_border[0]= h->top_borders[0][s->mb_x][15];
- h->left_border[1]= h->top_borders[1][s->mb_x][15];
- for(i=2; i<34; i++){
- h->left_border[i]= src_y[15+i* linesize];
- }
-
- *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
- *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
- *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
- *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
-
- if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
- h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
- h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
- h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
- h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
- for(i=2; i<18; i++){
- h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
- h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
- }
- *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
- *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
- *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
- *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
- }
-}
-
-static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
- MpegEncContext * const s = &h->s;
- int temp8, i;
- uint64_t temp64;
- int deblock_left = (s->mb_x > 0);
- int deblock_top = (s->mb_y > 1);
-
- tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
-
- src_y -= 2 * linesize + 1;
- src_cb -= 2 * uvlinesize + 1;
- src_cr -= 2 * uvlinesize + 1;
-
-#define XCHG(a,b,t,xchg)\
-t= a;\
-if(xchg)\
- a= b;\
-b= t;
-
- if(deblock_left){
- for(i = (!deblock_top)<<1; i<34; i++){
- XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
- }
- }
-
- if(deblock_top){
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
- XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
- XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
- if(s->mb_x+1 < s->mb_width){
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
- XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
- }
- }
-
- if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
- if(deblock_left){
- for(i = (!deblock_top) << 1; i<18; i++){
- XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
- XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
+ for(i = !deblock_top; i<8; i++){
+ XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, xchg);
+ XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
}
+ XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, 1);
+ XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
}
if(deblock_top){
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
- XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
- XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
- XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
+ XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
+ XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
}
}
}
idct_add = s->dsp.h264_idct_add;
}
- if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
- && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
- int mbt_y = mb_y&~1;
- uint8_t *top_y = s->current_picture.data[0] + (mbt_y * 16* s->linesize ) + mb_x * 16;
- uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
- uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
- xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
- }
-
if (!simple && IS_INTRA_PCM(mb_type)) {
for (i=0; i<16; i++) {
memcpy(dest_y + i* linesize, h->mb + i*8, 16);
}
} else {
if(IS_INTRA(mb_type)){
- if(h->deblocking_filter && (simple || !FRAME_MBAFF))
+ if(h->deblocking_filter)
xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
}else
svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
}
- if(h->deblocking_filter && (simple || !FRAME_MBAFF))
+ if(h->deblocking_filter)
xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
}else if(is_h264){
hl_motion(h, dest_y, dest_cb, dest_cr,
}
}
if(h->deblocking_filter) {
+ backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
+ fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
+ h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
+ h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
if (!simple && FRAME_MBAFF) {
- //FIXME try deblocking one mb at a time?
- // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
- const int mb_y = s->mb_y - 1;
- uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
- const int mb_xy= mb_x + mb_y*s->mb_stride;
- const int mb_type_top = s->current_picture.mb_type[mb_xy];
- const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
- if (!bottom) return;
- pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
- pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
- pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
-
- if(IS_INTRA(mb_type_top | mb_type_bottom))
- xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
-
- backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
- // deblock a pair
- // top
- s->mb_y--; h->mb_xy -= s->mb_stride;
- tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
- fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
- h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
- h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
- filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
- // bottom
- s->mb_y++; h->mb_xy += s->mb_stride;
- tprintf(h->s.avctx, "call mbaff filter_mb\n");
- fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
- h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
- h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
- filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
+ filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
} else {
- tprintf(h->s.avctx, "call filter_mb\n");
- backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
- fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
- h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
- h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
}
}
int *ref2frm= h->ref2frm[h->slice_num&15][j];
ref2frm[0]=
ref2frm[1]= -1;
- for(i=0; i<48; i++)
+ for(i=0; i<16; i++)
ref2frm[i+2]= 4*h->ref_list[j][i].frame_num
+(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
+ +(h->ref_list[j][i].reference&3);
}
h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
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) [48+2] = h->ref2frm[ h->slice_num &15 ];
- int (*ref2frmm)[48+2] = h->ref2frm[ h->slice_table[mbm_xy]&15 ];
+ int (*ref2frm) [64] = h->ref2frm[ h->slice_num &15 ][0] + (MB_MBAFF ? 20 : 2);
+ int (*ref2frmm)[64] = h->ref2frm[ h->slice_table[mbm_xy]&15 ][0] + (MB_MBAFF ? 20 : 2);
int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))
/* 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)[48+2] = edge > 0 ? ref2frm : ref2frmm;
+ int (*ref2frmn)[64] = edge > 0 ? ref2frm : ref2frmm;
int16_t bS[4];
int qp;
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]+2] != ref2frmn[l][h->ref_cache[l][bn_idx]+2] ||
+ v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] ||
FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
}
v=0;
for( l = 0; !v && l < 2; l++ ) {
int ln= 1-l;
- v |= ref2frm[l][h->ref_cache[l][b_idx]+2] != ref2frmn[ln][h->ref_cache[ln][bn_idx]+2] ||
+ v |= ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] ||
FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] ) >= 4 ||
FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit;
}
{
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]+2] != ref2frmn[l][h->ref_cache[l][bn_idx]+2] ||
+ if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[l][h->ref_cache[l][bn_idx]] ||
FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
bS[i] = 1;
bS[i] = 0;
for( l = 0; l < 2; l++ ) {
int ln= 1-l;
- if( ref2frm[l][h->ref_cache[l][b_idx]+2] != ref2frmn[ln][h->ref_cache[ln][bn_idx]+2] ||
+ if( ref2frm[l][h->ref_cache[l][b_idx]] != ref2frmn[ln][h->ref_cache[ln][bn_idx]] ||
FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] ) >= 4 ||
FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) {
bS[i] = 1;