h264: move some neighbour information into the per-slice context

pull/125/head
Anton Khirnov 10 years ago
parent 4bd5ac200d
commit 8b00f4df20
  1. 26
      libavcodec/h264.h
  2. 47
      libavcodec/h264_cabac.c
  3. 12
      libavcodec/h264_cavlc.c
  4. 29
      libavcodec/h264_loopfilter.c
  5. 5
      libavcodec/h264_mb.c
  6. 8
      libavcodec/h264_mb_template.c
  7. 158
      libavcodec/h264_mvpred.h
  8. 12
      libavcodec/h264_slice.c
  9. 3
      libavcodec/svq3.c

@ -320,6 +320,19 @@ typedef struct H264SliceContext {
int chroma_pred_mode; int chroma_pred_mode;
int intra16x16_pred_mode; int intra16x16_pred_mode;
int topleft_mb_xy;
int top_mb_xy;
int topright_mb_xy;
int left_mb_xy[LEFT_MBS];
int topleft_type;
int top_type;
int topright_type;
int left_type[LEFT_MBS];
const uint8_t *left_block;
int topleft_partition;
} H264SliceContext; } H264SliceContext;
/** /**
@ -357,19 +370,6 @@ typedef struct H264Context {
int workaround_bugs; int workaround_bugs;
// prediction stuff // prediction stuff
int topleft_mb_xy;
int top_mb_xy;
int topright_mb_xy;
int left_mb_xy[LEFT_MBS];
int topleft_type;
int top_type;
int topright_type;
int left_type[LEFT_MBS];
const uint8_t *left_block;
int topleft_partition;
int8_t intra4x4_pred_mode_cache[5 * 8]; int8_t intra4x4_pred_mode_cache[5 * 8];
int8_t(*intra4x4_pred_mode); int8_t(*intra4x4_pred_mode);
H264PredContext hpc; H264PredContext hpc;

@ -1293,15 +1293,17 @@ static int decode_cabac_field_decoding_flag(H264Context *h) {
return get_cabac_noinline( &h->cabac, &(h->cabac_state+70)[ctx] ); return get_cabac_noinline( &h->cabac, &(h->cabac_state+70)[ctx] );
} }
static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) { static int decode_cabac_intra_mb_type(H264Context *h, H264SliceContext *sl,
int ctx_base, int intra_slice)
{
uint8_t *state= &h->cabac_state[ctx_base]; uint8_t *state= &h->cabac_state[ctx_base];
int mb_type; int mb_type;
if(intra_slice){ if(intra_slice){
int ctx=0; int ctx=0;
if( h->left_type[LTOP] & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)) if (sl->left_type[LTOP] & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM))
ctx++; ctx++;
if( h->top_type & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)) if (sl->top_type & (MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM))
ctx++; ctx++;
if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 ) if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 )
return 0; /* I4x4 */ return 0; /* I4x4 */
@ -1371,17 +1373,18 @@ static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
return mode + ( mode >= pred_mode ); return mode + ( mode >= pred_mode );
} }
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) { static int decode_cabac_mb_chroma_pre_mode(H264Context *h, H264SliceContext *sl)
const int mba_xy = h->left_mb_xy[0]; {
const int mbb_xy = h->top_mb_xy; const int mba_xy = sl->left_mb_xy[0];
const int mbb_xy = sl->top_mb_xy;
int ctx = 0; int ctx = 0;
/* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */ /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
if( h->left_type[LTOP] && h->chroma_pred_mode_table[mba_xy] != 0 ) if (sl->left_type[LTOP] && h->chroma_pred_mode_table[mba_xy] != 0)
ctx++; ctx++;
if( h->top_type && h->chroma_pred_mode_table[mbb_xy] != 0 ) if (sl->top_type && h->chroma_pred_mode_table[mbb_xy] != 0)
ctx++; ctx++;
if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 ) if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
@ -1928,15 +1931,15 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl)
sl->prev_mb_skipped = 0; sl->prev_mb_skipped = 0;
fill_decode_neighbors(h, -(MB_FIELD(h))); fill_decode_neighbors(h, sl, -(MB_FIELD(h)));
if( h->slice_type_nos == AV_PICTURE_TYPE_B ) { if( h->slice_type_nos == AV_PICTURE_TYPE_B ) {
int ctx = 0; int ctx = 0;
assert(h->slice_type_nos == AV_PICTURE_TYPE_B); assert(h->slice_type_nos == AV_PICTURE_TYPE_B);
if( !IS_DIRECT( h->left_type[LTOP]-1 ) ) if (!IS_DIRECT(sl->left_type[LTOP] - 1))
ctx++; ctx++;
if( !IS_DIRECT( h->top_type-1 ) ) if (!IS_DIRECT(sl->top_type - 1))
ctx++; ctx++;
if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) ){ if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) ){
@ -1952,7 +1955,7 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl)
if( bits < 8 ){ if( bits < 8 ){
mb_type= bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */ mb_type= bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
}else if( bits == 13 ){ }else if( bits == 13 ){
mb_type= decode_cabac_intra_mb_type(h, 32, 0); mb_type = decode_cabac_intra_mb_type(h, sl, 32, 0);
goto decode_intra_mb; goto decode_intra_mb;
}else if( bits == 14 ){ }else if( bits == 14 ){
mb_type= 11; /* B_L1_L0_8x16 */ mb_type= 11; /* B_L1_L0_8x16 */
@ -1978,11 +1981,11 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl)
partition_count= p_mb_type_info[mb_type].partition_count; partition_count= p_mb_type_info[mb_type].partition_count;
mb_type= p_mb_type_info[mb_type].type; mb_type= p_mb_type_info[mb_type].type;
} else { } else {
mb_type= decode_cabac_intra_mb_type(h, 17, 0); mb_type = decode_cabac_intra_mb_type(h, sl, 17, 0);
goto decode_intra_mb; goto decode_intra_mb;
} }
} else { } else {
mb_type= decode_cabac_intra_mb_type(h, 3, 1); mb_type = decode_cabac_intra_mb_type(h, sl, 3, 1);
if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type) if(h->slice_type == AV_PICTURE_TYPE_SI && mb_type)
mb_type--; mb_type--;
assert(h->slice_type_nos == AV_PICTURE_TYPE_I); assert(h->slice_type_nos == AV_PICTURE_TYPE_I);
@ -2031,7 +2034,7 @@ decode_intra_mb:
return 0; return 0;
} }
fill_decode_caches(h, mb_type); fill_decode_caches(h, sl, mb_type);
if( IS_INTRA( mb_type ) ) { if( IS_INTRA( mb_type ) ) {
int i, pred_mode; int i, pred_mode;
@ -2060,7 +2063,7 @@ decode_intra_mb:
} }
if(decode_chroma){ if(decode_chroma){
h->chroma_pred_mode_table[mb_xy] = h->chroma_pred_mode_table[mb_xy] =
pred_mode = decode_cabac_mb_chroma_pre_mode( h ); pred_mode = decode_cabac_mb_chroma_pre_mode(h, sl);
pred_mode= ff_h264_check_intra_pred_mode( h, pred_mode, 1 ); pred_mode= ff_h264_check_intra_pred_mode( h, pred_mode, 1 );
if( pred_mode < 0 ) return -1; if( pred_mode < 0 ) return -1;
@ -2136,7 +2139,7 @@ decode_intra_mb:
const int index= 4*i + block_width*j; const int index= 4*i + block_width*j;
int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
uint8_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ]; uint8_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); pred_motion(h, sl, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
DECODE_CABAC_MB_MVD( h, list, index) DECODE_CABAC_MB_MVD( h, list, index)
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -2200,7 +2203,7 @@ decode_intra_mb:
for(list=0; list<h->list_count; list++){ for(list=0; list<h->list_count; list++){
if(IS_DIR(mb_type, 0, list)){ if(IS_DIR(mb_type, 0, list)){
int mx,my,mpx,mpy; int mx,my,mpx,mpy;
pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); pred_motion(h, sl, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
DECODE_CABAC_MB_MVD( h, list, 0) DECODE_CABAC_MB_MVD( h, list, 0)
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -2231,7 +2234,7 @@ decode_intra_mb:
for(i=0; i<2; i++){ for(i=0; i<2; i++){
if(IS_DIR(mb_type, i, list)){ if(IS_DIR(mb_type, i, list)){
int mx,my,mpx,mpy; int mx,my,mpx,mpy;
pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); pred_16x8_motion(h, sl, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
DECODE_CABAC_MB_MVD( h, list, 8*i) DECODE_CABAC_MB_MVD( h, list, 8*i)
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -2266,7 +2269,7 @@ decode_intra_mb:
for(i=0; i<2; i++){ for(i=0; i<2; i++){
if(IS_DIR(mb_type, i, list)){ if(IS_DIR(mb_type, i, list)){
int mx,my,mpx,mpy; int mx,my,mpx,mpy;
pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); pred_8x16_motion(h, sl, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
DECODE_CABAC_MB_MVD( h, list, 4*i) DECODE_CABAC_MB_MVD( h, list, 4*i)
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -2304,7 +2307,7 @@ decode_intra_mb:
int i; int i;
uint8_t *nnz_cache = h->non_zero_count_cache; uint8_t *nnz_cache = h->non_zero_count_cache;
for (i = 0; i < 2; i++){ for (i = 0; i < 2; i++){
if (h->left_type[LEFT(i)] && !IS_8x8DCT(h->left_type[LEFT(i)])){ if (sl->left_type[LEFT(i)] && !IS_8x8DCT(sl->left_type[LEFT(i)])) {
nnz_cache[3+8* 1 + 2*8*i]= nnz_cache[3+8* 1 + 2*8*i]=
nnz_cache[3+8* 2 + 2*8*i]= nnz_cache[3+8* 2 + 2*8*i]=
nnz_cache[3+8* 6 + 2*8*i]= nnz_cache[3+8* 6 + 2*8*i]=
@ -2313,7 +2316,7 @@ decode_intra_mb:
nnz_cache[3+8*12 + 2*8*i]= IS_INTRA(mb_type) ? 64 : 0; nnz_cache[3+8*12 + 2*8*i]= IS_INTRA(mb_type) ? 64 : 0;
} }
} }
if (h->top_type && !IS_8x8DCT(h->top_type)){ if (sl->top_type && !IS_8x8DCT(sl->top_type)){
uint32_t top_empty = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 0x40404040; uint32_t top_empty = CABAC(h) && !IS_INTRA(mb_type) ? 0 : 0x40404040;
AV_WN32A(&nnz_cache[4+8* 0], top_empty); AV_WN32A(&nnz_cache[4+8* 0], top_empty);
AV_WN32A(&nnz_cache[4+8* 5], top_empty); AV_WN32A(&nnz_cache[4+8* 5], top_empty);

@ -781,8 +781,8 @@ decode_intra_mb:
return 0; return 0;
} }
fill_decode_neighbors(h, mb_type); fill_decode_neighbors(h, sl, mb_type);
fill_decode_caches(h, mb_type); fill_decode_caches(h, sl, mb_type);
//mb_pred //mb_pred
if(IS_INTRA(mb_type)){ if(IS_INTRA(mb_type)){
@ -903,7 +903,7 @@ decode_intra_mb:
int mx, my; int mx, my;
const int index= 4*i + block_width*j; const int index= 4*i + block_width*j;
int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); pred_motion(h, sl, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
mx += get_se_golomb(&h->gb); mx += get_se_golomb(&h->gb);
my += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb);
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -957,7 +957,7 @@ decode_intra_mb:
} }
for(list=0; list<h->list_count; list++){ for(list=0; list<h->list_count; list++){
if(IS_DIR(mb_type, 0, list)){ if(IS_DIR(mb_type, 0, list)){
pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); pred_motion(h, sl, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
mx += get_se_golomb(&h->gb); mx += get_se_golomb(&h->gb);
my += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb);
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -992,7 +992,7 @@ decode_intra_mb:
for(i=0; i<2; i++){ for(i=0; i<2; i++){
unsigned int val; unsigned int val;
if(IS_DIR(mb_type, i, list)){ if(IS_DIR(mb_type, i, list)){
pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); pred_16x8_motion(h, sl, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
mx += get_se_golomb(&h->gb); mx += get_se_golomb(&h->gb);
my += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb);
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);
@ -1030,7 +1030,7 @@ decode_intra_mb:
for(i=0; i<2; i++){ for(i=0; i<2; i++){
unsigned int val; unsigned int val;
if(IS_DIR(mb_type, i, list)){ if(IS_DIR(mb_type, i, list)){
pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); pred_8x16_motion(h, sl, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
mx += get_se_golomb(&h->gb); mx += get_se_golomb(&h->gb);
my += get_se_golomb(&h->gb); my += get_se_golomb(&h->gb);
tprintf(h->avctx, "final mv:%d %d\n", mx, my); tprintf(h->avctx, "final mv:%d %d\n", mx, my);

@ -235,6 +235,7 @@ static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride,
} }
static av_always_inline void h264_filter_mb_fast_internal(H264Context *h, static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,
H264SliceContext *sl,
int mb_x, int mb_y, int mb_x, int mb_y,
uint8_t *img_y, uint8_t *img_y,
uint8_t *img_cb, uint8_t *img_cb,
@ -248,8 +249,8 @@ static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,
int chroma422 = CHROMA422(h); int chroma422 = CHROMA422(h);
int mb_xy = h->mb_xy; int mb_xy = h->mb_xy;
int left_type= h->left_type[LTOP]; int left_type = sl->left_type[LTOP];
int top_type= h->top_type; int top_type = sl->top_type;
int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8); int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset; int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset;
@ -258,7 +259,7 @@ static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,
int mb_type = h->cur_pic.mb_type[mb_xy]; int mb_type = h->cur_pic.mb_type[mb_xy];
int qp = h->cur_pic.qscale_table[mb_xy]; int qp = h->cur_pic.qscale_table[mb_xy];
int qp0 = h->cur_pic.qscale_table[mb_xy - 1]; int qp0 = h->cur_pic.qscale_table[mb_xy - 1];
int qp1 = h->cur_pic.qscale_table[h->top_mb_xy]; int qp1 = h->cur_pic.qscale_table[sl->top_mb_xy];
int qpc = get_chroma_qp( h, 0, qp ); int qpc = get_chroma_qp( h, 0, qp );
int qpc0 = get_chroma_qp( h, 0, qp0 ); int qpc0 = get_chroma_qp( h, 0, qp0 );
int qpc1 = get_chroma_qp( h, 0, qp1 ); int qpc1 = get_chroma_qp( h, 0, qp1 );
@ -427,12 +428,12 @@ void ff_h264_filter_mb_fast(H264Context *h, H264SliceContext *sl,
} }
#if CONFIG_SMALL #if CONFIG_SMALL
h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift); h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);
#else #else
if(h->pixel_shift){ if(h->pixel_shift){
h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1); h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);
}else{ }else{
h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0); h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);
} }
#endif #endif
} }
@ -478,8 +479,8 @@ static av_always_inline void filter_mb_dir(H264Context *h, H264SliceContext *sl,
int chroma_qp_avg[2]; int chroma_qp_avg[2];
int chroma444 = CHROMA444(h); int chroma444 = CHROMA444(h);
int chroma422 = CHROMA422(h); int chroma422 = CHROMA422(h);
const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; const int mbm_xy = dir == 0 ? mb_xy -1 : sl->top_mb_xy;
const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type; const int mbm_type = dir == 0 ? sl->left_type[LTOP] : sl->top_type;
// how often to recheck mv-based bS when iterating between edges // how often to recheck mv-based bS when iterating between edges
static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1}, static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
@ -726,9 +727,9 @@ void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl,
if (FRAME_MBAFF(h) if (FRAME_MBAFF(h)
// and current and left pair do not have the same interlaced type // and current and left pair do not have the same interlaced type
&& IS_INTERLACED(mb_type^h->left_type[LTOP]) && IS_INTERLACED(mb_type ^ sl->left_type[LTOP])
// and left mb is in available to us // and left mb is in available to us
&& h->left_type[LTOP]) { && sl->left_type[LTOP]) {
/* First vertical edge is different in MBAFF frames /* First vertical edge is different in MBAFF frames
* There are 8 different bS to compute and 2 different Qp * There are 8 different bS to compute and 2 different Qp
*/ */
@ -756,8 +757,8 @@ void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl,
const uint8_t *off= offset[MB_FIELD(h)][mb_y&1]; const uint8_t *off= offset[MB_FIELD(h)][mb_y&1];
for( i = 0; i < 8; i++ ) { for( i = 0; i < 8; i++ ) {
int j= MB_FIELD(h) ? i>>2 : i&1; int j= MB_FIELD(h) ? i>>2 : i&1;
int mbn_xy = h->left_mb_xy[LEFT(j)]; int mbn_xy = sl->left_mb_xy[LEFT(j)];
int mbn_type= h->left_type[LEFT(j)]; int mbn_type = sl->left_type[LEFT(j)];
if( IS_INTRA( mbn_type ) ) if( IS_INTRA( mbn_type ) )
bS[i] = 4; bS[i] = 4;
@ -772,8 +773,8 @@ void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl,
} }
mb_qp = h->cur_pic.qscale_table[mb_xy]; mb_qp = h->cur_pic.qscale_table[mb_xy];
mbn0_qp = h->cur_pic.qscale_table[h->left_mb_xy[0]]; mbn0_qp = h->cur_pic.qscale_table[sl->left_mb_xy[0]];
mbn1_qp = h->cur_pic.qscale_table[h->left_mb_xy[1]]; mbn1_qp = h->cur_pic.qscale_table[sl->left_mb_xy[1]];
qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1; qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) + bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1; get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;

@ -496,7 +496,8 @@ static av_always_inline void prefetch_motion(H264Context *h, int list,
} }
} }
static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y, static av_always_inline void xchg_mb_border(H264Context *h, H264SliceContext *sl,
uint8_t *src_y,
uint8_t *src_cb, uint8_t *src_cr, uint8_t *src_cb, uint8_t *src_cr,
int linesize, int uvlinesize, int linesize, int uvlinesize,
int xchg, int chroma444, int xchg, int chroma444,
@ -519,7 +520,7 @@ static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
if (h->deblocking_filter == 2) { if (h->deblocking_filter == 2) {
deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num; deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num;
deblock_top = h->top_type; deblock_top = sl->top_type;
} else { } else {
deblock_topleft = (h->mb_x > 0); deblock_topleft = (h->mb_x > 0);
deblock_top = (h->mb_y > !!MB_FIELD(h)); deblock_top = (h->mb_y > !!MB_FIELD(h));

@ -159,7 +159,7 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl)
} else { } else {
if (IS_INTRA(mb_type)) { if (IS_INTRA(mb_type)) {
if (h->deblocking_filter) if (h->deblocking_filter)
xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT); uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT);
if (SIMPLE || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { if (SIMPLE || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
@ -172,7 +172,7 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl)
block_offset, linesize, dest_y, 0); block_offset, linesize, dest_y, 0);
if (h->deblocking_filter) if (h->deblocking_filter)
xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT); uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT);
} else if (is_h264) { } else if (is_h264) {
if (chroma422) { if (chroma422) {
@ -343,7 +343,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext
} else { } else {
if (IS_INTRA(mb_type)) { if (IS_INTRA(mb_type)) {
if (h->deblocking_filter) if (h->deblocking_filter)
xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize,
linesize, 1, 1, SIMPLE, PIXEL_SHIFT); linesize, 1, 1, SIMPLE, PIXEL_SHIFT);
for (p = 0; p < plane_count; p++) for (p = 0; p < plane_count; p++)
@ -352,7 +352,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext
block_offset, linesize, dest[p], p); block_offset, linesize, dest[p], p);
if (h->deblocking_filter) if (h->deblocking_filter)
xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize,
linesize, 0, 1, SIMPLE, PIXEL_SHIFT); linesize, 0, 1, SIMPLE, PIXEL_SHIFT);
} else { } else {
FUNC(hl_motion_444)(h, sl, dest[0], dest[1], dest[2], FUNC(hl_motion_444)(h, sl, dest[0], dest[1], dest[2],

@ -35,7 +35,8 @@
#include <assert.h> #include <assert.h>
static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, static av_always_inline int fetch_diagonal_mv(H264Context *h, H264SliceContext *sl,
const int16_t **C,
int i, int list, int part_width) int i, int list, int part_width)
{ {
const int topright_ref = h->ref_cache[list][i - 8 + part_width]; const int topright_ref = h->ref_cache[list][i - 8 + part_width];
@ -61,13 +62,13 @@ static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C,
AV_ZERO32(h->mv_cache[list][scan8[0] - 2]); AV_ZERO32(h->mv_cache[list][scan8[0] - 2]);
*C = h->mv_cache[list][scan8[0] - 2]; *C = h->mv_cache[list][scan8[0] - 2];
if (!MB_FIELD(h) && IS_INTERLACED(h->left_type[0])) { if (!MB_FIELD(h) && IS_INTERLACED(sl->left_type[0])) {
SET_DIAG_MV(* 2, >> 1, h->left_mb_xy[0] + h->mb_stride, SET_DIAG_MV(* 2, >> 1, sl->left_mb_xy[0] + h->mb_stride,
(h->mb_y & 1) * 2 + (i >> 5)); (h->mb_y & 1) * 2 + (i >> 5));
} }
if (MB_FIELD(h) && !IS_INTERLACED(h->left_type[0])) { if (MB_FIELD(h) && !IS_INTERLACED(sl->left_type[0])) {
// left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK. // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
SET_DIAG_MV(/ 2, << 1, h->left_mb_xy[i >= 36], ((i >> 2)) & 3); SET_DIAG_MV(/ 2, << 1, sl->left_mb_xy[i >= 36], ((i >> 2)) & 3);
} }
} }
#undef SET_DIAG_MV #undef SET_DIAG_MV
@ -91,7 +92,9 @@ static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C,
* @param mx the x component of the predicted motion vector * @param mx the x component of the predicted motion vector
* @param my the y component of the predicted motion vector * @param my the y component of the predicted motion vector
*/ */
static av_always_inline void pred_motion(H264Context *const h, int n, static av_always_inline void pred_motion(H264Context *const h,
H264SliceContext *sl,
int n,
int part_width, int list, int ref, int part_width, int list, int ref,
int *const mx, int *const my) int *const mx, int *const my)
{ {
@ -113,7 +116,7 @@ static av_always_inline void pred_motion(H264Context *const h, int n,
* . . . L . . . . * . . . L . . . .
*/ */
diagonal_ref = fetch_diagonal_mv(h, &C, index8, list, part_width); diagonal_ref = fetch_diagonal_mv(h, sl, &C, index8, list, part_width);
match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref); match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref);
tprintf(h->avctx, "pred_motion match_count=%d\n", match_count); tprintf(h->avctx, "pred_motion match_count=%d\n", match_count);
if (match_count > 1) { //most common if (match_count > 1) { //most common
@ -155,6 +158,7 @@ static av_always_inline void pred_motion(H264Context *const h, int n,
* @param my the y component of the predicted motion vector * @param my the y component of the predicted motion vector
*/ */
static av_always_inline void pred_16x8_motion(H264Context *const h, static av_always_inline void pred_16x8_motion(H264Context *const h,
H264SliceContext *sl,
int n, int list, int ref, int n, int list, int ref,
int *const mx, int *const my) int *const mx, int *const my)
{ {
@ -185,7 +189,7 @@ static av_always_inline void pred_16x8_motion(H264Context *const h,
} }
//RARE //RARE
pred_motion(h, n, 4, list, ref, mx, my); pred_motion(h, sl, n, 4, list, ref, mx, my);
} }
/** /**
@ -195,6 +199,7 @@ static av_always_inline void pred_16x8_motion(H264Context *const h,
* @param my the y component of the predicted motion vector * @param my the y component of the predicted motion vector
*/ */
static av_always_inline void pred_8x16_motion(H264Context *const h, static av_always_inline void pred_8x16_motion(H264Context *const h,
H264SliceContext *sl,
int n, int list, int ref, int n, int list, int ref,
int *const mx, int *const my) int *const mx, int *const my)
{ {
@ -214,7 +219,7 @@ static av_always_inline void pred_8x16_motion(H264Context *const h,
const int16_t *C; const int16_t *C;
int diagonal_ref; int diagonal_ref;
diagonal_ref = fetch_diagonal_mv(h, &C, scan8[4], list, 2); diagonal_ref = fetch_diagonal_mv(h, sl, &C, scan8[4], list, 2);
tprintf(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", tprintf(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n",
diagonal_ref, C[0], C[1], h->mb_x, h->mb_y, n, list); diagonal_ref, C[0], C[1], h->mb_x, h->mb_y, n, list);
@ -227,7 +232,7 @@ static av_always_inline void pred_8x16_motion(H264Context *const h,
} }
//RARE //RARE
pred_motion(h, n, 2, list, ref, mx, my); pred_motion(h, sl, n, 2, list, ref, mx, my);
} }
#define FIX_MV_MBAFF(type, refn, mvn, idx) \ #define FIX_MV_MBAFF(type, refn, mvn, idx) \
@ -249,7 +254,8 @@ static av_always_inline void pred_8x16_motion(H264Context *const h,
} \ } \
} }
static av_always_inline void pred_pskip_motion(H264Context *const h) static av_always_inline void pred_pskip_motion(H264Context *const h,
H264SliceContext *sl)
{ {
DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = { 0 }; DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = { 0 };
DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2]; DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2];
@ -266,26 +272,26 @@ static av_always_inline void pred_pskip_motion(H264Context *const h)
* FIXME: this is a partial duplicate of the logic in fill_decode_caches, * FIXME: this is a partial duplicate of the logic in fill_decode_caches,
* but it's faster this way. Is there a way to avoid this duplication? * but it's faster this way. Is there a way to avoid this duplication?
*/ */
if (USES_LIST(h->left_type[LTOP], 0)) { if (USES_LIST(sl->left_type[LTOP], 0)) {
left_ref = ref[4 * h->left_mb_xy[LTOP] + 1 + (h->left_block[0] & ~1)]; left_ref = ref[4 * sl->left_mb_xy[LTOP] + 1 + (sl->left_block[0] & ~1)];
A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride * h->left_block[0]]; A = mv[h->mb2b_xy[sl->left_mb_xy[LTOP]] + 3 + b_stride * sl->left_block[0]];
FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0); FIX_MV_MBAFF(sl->left_type[LTOP], left_ref, A, 0);
if (!(left_ref | AV_RN32A(A))) if (!(left_ref | AV_RN32A(A)))
goto zeromv; goto zeromv;
} else if (h->left_type[LTOP]) { } else if (sl->left_type[LTOP]) {
left_ref = LIST_NOT_USED; left_ref = LIST_NOT_USED;
A = zeromv; A = zeromv;
} else { } else {
goto zeromv; goto zeromv;
} }
if (USES_LIST(h->top_type, 0)) { if (USES_LIST(sl->top_type, 0)) {
top_ref = ref[4 * h->top_mb_xy + 2]; top_ref = ref[4 * sl->top_mb_xy + 2];
B = mv[h->mb2b_xy[h->top_mb_xy] + 3 * b_stride]; B = mv[h->mb2b_xy[sl->top_mb_xy] + 3 * b_stride];
FIX_MV_MBAFF(h->top_type, top_ref, B, 1); FIX_MV_MBAFF(sl->top_type, top_ref, B, 1);
if (!(top_ref | AV_RN32A(B))) if (!(top_ref | AV_RN32A(B)))
goto zeromv; goto zeromv;
} else if (h->top_type) { } else if (sl->top_type) {
top_ref = LIST_NOT_USED; top_ref = LIST_NOT_USED;
B = zeromv; B = zeromv;
} else { } else {
@ -295,21 +301,21 @@ static av_always_inline void pred_pskip_motion(H264Context *const h)
tprintf(h->avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", tprintf(h->avctx, "pred_pskip: (%d) (%d) at %2d %2d\n",
top_ref, left_ref, h->mb_x, h->mb_y); top_ref, left_ref, h->mb_x, h->mb_y);
if (USES_LIST(h->topright_type, 0)) { if (USES_LIST(sl->topright_type, 0)) {
diagonal_ref = ref[4 * h->topright_mb_xy + 2]; diagonal_ref = ref[4 * sl->topright_mb_xy + 2];
C = mv[h->mb2b_xy[h->topright_mb_xy] + 3 * b_stride]; C = mv[h->mb2b_xy[sl->topright_mb_xy] + 3 * b_stride];
FIX_MV_MBAFF(h->topright_type, diagonal_ref, C, 2); FIX_MV_MBAFF(sl->topright_type, diagonal_ref, C, 2);
} else if (h->topright_type) { } else if (sl->topright_type) {
diagonal_ref = LIST_NOT_USED; diagonal_ref = LIST_NOT_USED;
C = zeromv; C = zeromv;
} else { } else {
if (USES_LIST(h->topleft_type, 0)) { if (USES_LIST(sl->topleft_type, 0)) {
diagonal_ref = ref[4 * h->topleft_mb_xy + 1 + diagonal_ref = ref[4 * sl->topleft_mb_xy + 1 +
(h->topleft_partition & 2)]; (sl->topleft_partition & 2)];
C = mv[h->mb2b_xy[h->topleft_mb_xy] + 3 + b_stride + C = mv[h->mb2b_xy[sl->topleft_mb_xy] + 3 + b_stride +
(h->topleft_partition & 2 * b_stride)]; (sl->topleft_partition & 2 * b_stride)];
FIX_MV_MBAFF(h->topleft_type, diagonal_ref, C, 2); FIX_MV_MBAFF(sl->topleft_type, diagonal_ref, C, 2);
} else if (h->topleft_type) { } else if (sl->topleft_type) {
diagonal_ref = LIST_NOT_USED; diagonal_ref = LIST_NOT_USED;
C = zeromv; C = zeromv;
} else { } else {
@ -347,7 +353,7 @@ zeromv:
return; return;
} }
static void fill_decode_neighbors(H264Context *h, int mb_type) static void fill_decode_neighbors(H264Context *h, H264SliceContext *sl, int mb_type)
{ {
const int mb_xy = h->mb_xy; const int mb_xy = h->mb_xy;
int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS];
@ -358,7 +364,7 @@ static void fill_decode_neighbors(H264Context *h, int mb_type)
{ 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 } { 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; sl->topleft_partition = -1;
top_xy = mb_xy - (h->mb_stride << MB_FIELD(h)); top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
@ -368,7 +374,7 @@ static void fill_decode_neighbors(H264Context *h, int mb_type)
topleft_xy = top_xy - 1; topleft_xy = top_xy - 1;
topright_xy = top_xy + 1; topright_xy = top_xy + 1;
left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1; left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
h->left_block = left_block_options[0]; sl->left_block = left_block_options[0];
if (FRAME_MBAFF(h)) { if (FRAME_MBAFF(h)) {
const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]); const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
const int curr_mb_field_flag = IS_INTERLACED(mb_type); const int curr_mb_field_flag = IS_INTERLACED(mb_type);
@ -377,13 +383,13 @@ static void fill_decode_neighbors(H264Context *h, int mb_type)
left_xy[LBOT] = left_xy[LTOP] = mb_xy - h->mb_stride - 1; left_xy[LBOT] = left_xy[LTOP] = mb_xy - h->mb_stride - 1;
if (curr_mb_field_flag) { if (curr_mb_field_flag) {
left_xy[LBOT] += h->mb_stride; left_xy[LBOT] += h->mb_stride;
h->left_block = left_block_options[3]; sl->left_block = left_block_options[3];
} else { } else {
topleft_xy += h->mb_stride; topleft_xy += h->mb_stride;
/* take top left mv from the middle of the mb, as opposed /* take top left mv from the middle of the mb, as opposed
* to all other modes which use the bottom right partition */ * to all other modes which use the bottom right partition */
h->topleft_partition = 0; sl->topleft_partition = 0;
h->left_block = left_block_options[1]; sl->left_block = left_block_options[1];
} }
} }
} else { } else {
@ -395,66 +401,66 @@ static void fill_decode_neighbors(H264Context *h, int mb_type)
if (left_mb_field_flag != curr_mb_field_flag) { if (left_mb_field_flag != curr_mb_field_flag) {
if (curr_mb_field_flag) { if (curr_mb_field_flag) {
left_xy[LBOT] += h->mb_stride; left_xy[LBOT] += h->mb_stride;
h->left_block = left_block_options[3]; sl->left_block = left_block_options[3];
} else { } else {
h->left_block = left_block_options[2]; sl->left_block = left_block_options[2];
} }
} }
} }
} }
h->topleft_mb_xy = topleft_xy; sl->topleft_mb_xy = topleft_xy;
h->top_mb_xy = top_xy; sl->top_mb_xy = top_xy;
h->topright_mb_xy = topright_xy; sl->topright_mb_xy = topright_xy;
h->left_mb_xy[LTOP] = left_xy[LTOP]; sl->left_mb_xy[LTOP] = left_xy[LTOP];
h->left_mb_xy[LBOT] = left_xy[LBOT]; sl->left_mb_xy[LBOT] = left_xy[LBOT];
//FIXME do we need all in the context? //FIXME do we need all in the context?
h->topleft_type = h->cur_pic.mb_type[topleft_xy]; sl->topleft_type = h->cur_pic.mb_type[topleft_xy];
h->top_type = h->cur_pic.mb_type[top_xy]; sl->top_type = h->cur_pic.mb_type[top_xy];
h->topright_type = h->cur_pic.mb_type[topright_xy]; sl->topright_type = h->cur_pic.mb_type[topright_xy];
h->left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; sl->left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
h->left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; sl->left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
if (FMO) { if (FMO) {
if (h->slice_table[topleft_xy] != h->slice_num) if (h->slice_table[topleft_xy] != h->slice_num)
h->topleft_type = 0; sl->topleft_type = 0;
if (h->slice_table[top_xy] != h->slice_num) if (h->slice_table[top_xy] != h->slice_num)
h->top_type = 0; sl->top_type = 0;
if (h->slice_table[left_xy[LTOP]] != h->slice_num) if (h->slice_table[left_xy[LTOP]] != h->slice_num)
h->left_type[LTOP] = h->left_type[LBOT] = 0; sl->left_type[LTOP] = sl->left_type[LBOT] = 0;
} else { } else {
if (h->slice_table[topleft_xy] != h->slice_num) { if (h->slice_table[topleft_xy] != h->slice_num) {
h->topleft_type = 0; sl->topleft_type = 0;
if (h->slice_table[top_xy] != h->slice_num) if (h->slice_table[top_xy] != h->slice_num)
h->top_type = 0; sl->top_type = 0;
if (h->slice_table[left_xy[LTOP]] != h->slice_num) if (h->slice_table[left_xy[LTOP]] != h->slice_num)
h->left_type[LTOP] = h->left_type[LBOT] = 0; sl->left_type[LTOP] = sl->left_type[LBOT] = 0;
} }
} }
if (h->slice_table[topright_xy] != h->slice_num) if (h->slice_table[topright_xy] != h->slice_num)
h->topright_type = 0; sl->topright_type = 0;
} }
static void fill_decode_caches(H264Context *h, int mb_type) static void fill_decode_caches(H264Context *h, H264SliceContext *sl, int mb_type)
{ {
int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS]; int topleft_xy, top_xy, topright_xy, left_xy[LEFT_MBS];
int topleft_type, top_type, topright_type, left_type[LEFT_MBS]; int topleft_type, top_type, topright_type, left_type[LEFT_MBS];
const uint8_t *left_block = h->left_block; const uint8_t *left_block = sl->left_block;
int i; int i;
uint8_t *nnz; uint8_t *nnz;
uint8_t *nnz_cache; uint8_t *nnz_cache;
topleft_xy = h->topleft_mb_xy; topleft_xy = sl->topleft_mb_xy;
top_xy = h->top_mb_xy; top_xy = sl->top_mb_xy;
topright_xy = h->topright_mb_xy; topright_xy = sl->topright_mb_xy;
left_xy[LTOP] = h->left_mb_xy[LTOP]; left_xy[LTOP] = sl->left_mb_xy[LTOP];
left_xy[LBOT] = h->left_mb_xy[LBOT]; left_xy[LBOT] = sl->left_mb_xy[LBOT];
topleft_type = h->topleft_type; topleft_type = sl->topleft_type;
top_type = h->top_type; top_type = sl->top_type;
topright_type = h->topright_type; topright_type = sl->topright_type;
left_type[LTOP] = h->left_type[LTOP]; left_type[LTOP] = sl->left_type[LTOP];
left_type[LBOT] = h->left_type[LBOT]; left_type[LBOT] = sl->left_type[LBOT];
if (!IS_SKIP(mb_type)) { if (!IS_SKIP(mb_type)) {
if (IS_INTRA(mb_type)) { if (IS_INTRA(mb_type)) {
@ -667,8 +673,8 @@ static void fill_decode_caches(H264Context *h, int mb_type)
if (ref_cache[4 - 1 * 8] < 0) { if (ref_cache[4 - 1 * 8] < 0) {
if (USES_LIST(topleft_type, list)) { if (USES_LIST(topleft_type, list)) {
const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride +
(h->topleft_partition & 2 * b_stride); (sl->topleft_partition & 2 * b_stride);
const int b8_xy = 4 * topleft_xy + 1 + (h->topleft_partition & 2); const int b8_xy = 4 * topleft_xy + 1 + (sl->topleft_partition & 2);
AV_COPY32(mv_cache[-1 - 1 * 8], mv[b_xy]); AV_COPY32(mv_cache[-1 - 1 * 8], mv[b_xy]);
ref_cache[-1 - 1 * 8] = ref[b8_xy]; ref_cache[-1 - 1 * 8] = ref[b8_xy];
} else { } else {
@ -808,16 +814,16 @@ static void av_unused decode_mb_skip(H264Context *h, H264SliceContext *sl)
// just for fill_caches. pred_direct_motion will set the real mb_type // just for fill_caches. pred_direct_motion will set the real mb_type
mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 | MB_TYPE_SKIP; mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 | MB_TYPE_SKIP;
if (h->direct_spatial_mv_pred) { if (h->direct_spatial_mv_pred) {
fill_decode_neighbors(h, mb_type); fill_decode_neighbors(h, sl, mb_type);
fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ... fill_decode_caches(h, sl, mb_type); //FIXME check what is needed and what not ...
} }
ff_h264_pred_direct_motion(h, &mb_type); ff_h264_pred_direct_motion(h, &mb_type);
mb_type |= MB_TYPE_SKIP; mb_type |= MB_TYPE_SKIP;
} else { } else {
mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P1L0 | MB_TYPE_SKIP; mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P1L0 | MB_TYPE_SKIP;
fill_decode_neighbors(h, mb_type); fill_decode_neighbors(h, sl, mb_type);
pred_pskip_motion(h); pred_pskip_motion(h, sl);
} }
write_back_motion(h, mb_type); write_back_motion(h, mb_type);

@ -1932,9 +1932,9 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
} }
} }
h->top_mb_xy = top_xy; sl->top_mb_xy = top_xy;
h->left_mb_xy[LTOP] = left_xy[LTOP]; sl->left_mb_xy[LTOP] = left_xy[LTOP];
h->left_mb_xy[LBOT] = left_xy[LBOT]; sl->left_mb_xy[LBOT] = left_xy[LBOT];
{ {
/* For sufficiently low qp, filtering wouldn't do anything. /* For sufficiently low qp, filtering wouldn't do anything.
* This is a conservative estimate: could also check beta_offset * This is a conservative estimate: could also check beta_offset
@ -1970,9 +1970,9 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
if (h->slice_table[left_xy[LBOT]] == 0xFFFF) if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
left_type[LTOP] = left_type[LBOT] = 0; left_type[LTOP] = left_type[LBOT] = 0;
} }
h->top_type = top_type; sl->top_type = top_type;
h->left_type[LTOP] = left_type[LTOP]; sl->left_type[LTOP] = left_type[LTOP];
h->left_type[LBOT] = left_type[LBOT]; sl->left_type[LBOT] = left_type[LBOT];
if (IS_INTRA(mb_type)) if (IS_INTRA(mb_type))
return 0; return 0;

@ -369,6 +369,7 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode,
{ {
int i, j, k, mx, my, dx, dy, x, y; int i, j, k, mx, my, dx, dy, x, y;
H264Context *h = &s->h; H264Context *h = &s->h;
H264SliceContext *sl = &h->slice_ctx[0];
const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1); const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1);
const int part_height = 16 >> ((unsigned)(size + 1) / 3); const int part_height = 16 >> ((unsigned)(size + 1) / 3);
const int extra_width = (mode == PREDICT_MODE) ? -16 * 6 : 0; const int extra_width = (mode == PREDICT_MODE) ? -16 * 6 : 0;
@ -386,7 +387,7 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode,
(j >> 1 & 4) + (i & 8); (j >> 1 & 4) + (i & 8);
if (mode != PREDICT_MODE) { if (mode != PREDICT_MODE) {
pred_motion(h, k, part_width >> 2, dir, 1, &mx, &my); pred_motion(h, sl, k, part_width >> 2, dir, 1, &mx, &my);
} else { } else {
mx = s->next_pic->motion_val[0][b_xy][0] << 1; mx = s->next_pic->motion_val[0][b_xy][0] << 1;
my = s->next_pic->motion_val[0][b_xy][1] << 1; my = s->next_pic->motion_val[0][b_xy][1] << 1;

Loading…
Cancel
Save