*/
/**
- * @file libavcodec/h264_direct.c
+ * @file
* H.264 / AVC / MPEG4 part10 direct mb/block decoding.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
MpegEncContext * const s = &h->s;
- int b8_stride = h->b8_stride;
+ int b8_stride = 2;
int b4_stride = h->b_stride;
int mb_xy = h->mb_xy;
int mb_type_col[2];
const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1;
const int is_b8x8 = IS_8X8(*mb_type);
- unsigned int sub_mb_type= MB_TYPE_L0L1;;
+ unsigned int sub_mb_type= MB_TYPE_L0L1;
int i8, i4;
int ref[2];
int mv[2];
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
- *mb_type |= MB_TYPE_L0L1;
/* ref = min(neighbors) */
for(list=0; list<2; list++){
int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
if(match_count > 1){ //most common
- mv[list]= (mid_pred(A[0], B[0], C[0])&0xFFFF)
- +(mid_pred(A[1], B[1], C[1])<<16);
+ mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
+ mid_pred(A[1], B[1], C[1]) );
}else {
assert(match_count==1);
if(left_ref==ref[list]){
- mv[list]= *(uint32_t*)A;
+ mv[list]= AV_RN32A(A);
}else if(top_ref==ref[list]){
- mv[list]= *(uint32_t*)B;
+ mv[list]= AV_RN32A(B);
}else{
- mv[list]= *(uint32_t*)C;
+ mv[list]= AV_RN32A(C);
}
}
}else{
sub_mb_type |= MB_TYPE_L0L1;
}
- if(!is_b8x8 && (mv[0]|mv[1]) == 0){
+ if(!(is_b8x8|mv[0]|mv[1])){
fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
- b8_stride *= 3;
+ b8_stride = 2+4*s->mb_stride;
b4_stride *= 6;
sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
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]];
+ l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy];
+ l1ref1 = &h->ref_list[1][0].ref_index [1][4*mb_xy];
if(!b8_stride){
if(s->mb_y&1){
- l1ref0 += h->b8_stride;
- l1ref1 += h->b8_stride;
+ l1ref0 += 2;
+ l1ref1 += 2;
l1mv0 += 2*b4_stride;
l1mv1 += 2*b4_stride;
}
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
+ assert(b8_stride==2);
/* col_zero_flag */
- if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[x8 + y8*b8_stride] == 0
- || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
+ if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0
+ || (l1ref0[i8] < 0 && l1ref1[i8] == 0
&& h->x264_build>33U))){
- const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
+ const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
if(IS_SUB_8X8(sub_mb_type)){
const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
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;
+ AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
if(ref[1] == 0)
- *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
+ AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
m++;
}
}
static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
MpegEncContext * const s = &h->s;
- int b8_stride = h->b8_stride;
+ int b8_stride = 2;
int b4_stride = h->b_stride;
int mb_xy = h->mb_xy;
int mb_type_col[2];
mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
- b8_stride *= 3;
+ b8_stride = 2+4*s->mb_stride;
b4_stride *= 6;
sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
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]];
+ l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy];
+ l1ref1 = &h->ref_list[1][0].ref_index [1][4*mb_xy];
if(!b8_stride){
if(s->mb_y&1){
- l1ref0 += h->b8_stride;
- l1ref1 += h->b8_stride;
+ l1ref0 += 2;
+ l1ref1 += 2;
l1mv0 += 2*b4_stride;
l1mv1 += 2*b4_stride;
}
continue;
}
- ref0 = l1ref0[x8 + y8*b8_stride];
+ assert(b8_stride == 2);
+ ref0 = l1ref0[i8];
if(ref0 >= 0)
ref0 = map_col_to_list0[0][ref0 + ref_offset];
else{
- ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
+ ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
l1mv= l1mv1;
}
scale = dist_scale_factor[ref0];
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;
- *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
- pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
+ AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
+ pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
}
}
}