1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00

H.264: fix CODEC_FLAG_GRAY

It was broken in 4:4:4, and still did chroma deblocking for no reason in 4:2:0.
This commit is contained in:
Jason Garrett-Glaser 2011-06-09 16:17:41 -07:00
parent c9c493872c
commit c177cfb4fb
2 changed files with 84 additions and 65 deletions

View File

@ -484,6 +484,8 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
}
if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
if(chroma444){
src_cb = pic->data[1] + offset;
if(emu){
@ -509,8 +511,6 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
return;
}
if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
if(MB_FIELD){
// chroma offset when predicting from a field of opposite parity
my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
@ -1847,24 +1847,28 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
for (j = 0; j < 16; j++)
tmp_y[j] = get_bits(&gb, bit_depth);
}
for (i = 0; i < 8; i++) {
uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
for (j = 0; j < 8; j++)
tmp_cb[j] = get_bits(&gb, bit_depth);
}
for (i = 0; i < 8; i++) {
uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
for (j = 0; j < 8; j++)
tmp_cr[j] = get_bits(&gb, bit_depth);
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
for (i = 0; i < 8; i++) {
uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
for (j = 0; j < 8; j++)
tmp_cb[j] = get_bits(&gb, bit_depth);
}
for (i = 0; i < 8; i++) {
uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
for (j = 0; j < 8; j++)
tmp_cr[j] = get_bits(&gb, bit_depth);
}
}
} else {
for (i=0; i<16; i++) {
memcpy(dest_y + i* linesize, h->mb + i*8, 16);
}
for (i=0; i<8; i++) {
memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
}
for (i=0; i<16; i++) {
memcpy(dest_y + i* linesize, h->mb + i*8, 16);
}
if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
for (i=0; i<8; i++) {
memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
}
}
}
} else {
if(IS_INTRA(mb_type)){
@ -1954,8 +1958,9 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
int i, j, p;
int *block_offset = &h->block_offset[0];
const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
for (p = 0; p < 3; p++)
for (p = 0; p < plane_count; p++)
{
dest[p] = s->current_picture.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
@ -1996,7 +2001,7 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
GetBitContext gb;
init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
for (p = 0; p < 3; p++) {
for (p = 0; p < plane_count; p++) {
for (i = 0; i < 16; i++) {
uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
for (j = 0; j < 16; j++)
@ -2004,7 +2009,7 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
}
}
} else {
for (p = 0; p < 3; p++) {
for (p = 0; p < plane_count; p++) {
for (i = 0; i < 16; i++) {
memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
}
@ -2015,7 +2020,7 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
if(h->deblocking_filter)
xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
for (p = 0; p < 3; p++)
for (p = 0; p < plane_count; p++)
hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
if(h->deblocking_filter)
@ -2035,7 +2040,7 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
h->h264dsp.biweight_h264_pixels_tab, 1);
}
for (p = 0; p < 3; p++)
for (p = 0; p < plane_count; p++)
hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
}
if(h->cbp || IS_INTRA(mb_type))

View File

@ -217,6 +217,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
int mb_xy;
int mb_type, left_type;
int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
mb_xy = h->mb_xy;
@ -262,16 +263,18 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
}
if(left_type){
filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
if(chroma){
if(left_type){
filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
}
filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
}
filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
return;
} else {
LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
@ -298,7 +301,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
#define FILTER(hv,dir,edge)\
if(AV_RN64A(bS[dir][edge])) { \
filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
if(!(edge&1)) {\
if(chroma && !(edge&1)) {\
filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
}\
@ -353,7 +356,7 @@ static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
return v;
}
static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int chroma444, int dir) {
static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int chroma, int chroma444, int dir) {
MpegEncContext * const s = &h->s;
int edge;
int chroma_qp_avg[2];
@ -410,12 +413,14 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
if (chroma444) {
filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
} else {
filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
if (chroma) {
if (chroma444) {
filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
} else {
filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
}
}
}
}else{
@ -475,7 +480,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
if( dir == 0 ) {
filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
{
if (chroma) {
if (chroma444) {
filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
@ -486,7 +491,7 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
}
} else {
filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
{
if (chroma) {
if (chroma444) {
filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
@ -557,21 +562,25 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
//{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
if( dir == 0 ) {
filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, h );
if (chroma444) {
filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
} else if( (edge&1) == 0 ) {
filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
if (chroma) {
if (chroma444) {
filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
} else if( (edge&1) == 0 ) {
filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
}
}
} else {
filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
if (chroma444) {
filter_mb_edgeh ( &img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgeh ( &img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
} else if( (edge&1) == 0 ) {
filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
if (chroma) {
if (chroma444) {
filter_mb_edgeh ( &img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgeh ( &img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
} else if( (edge&1) == 0 ) {
filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
}
}
}
}
@ -584,6 +593,7 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint
const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
int first_vertical_edge_done = 0;
av_unused int dir;
int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
if (FRAME_MBAFF
// and current and left pair do not have the same interlaced type
@ -652,25 +662,29 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint
if(MB_FIELD){
filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
if (chroma){
filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
}
}else{
filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
if (chroma){
filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
}
}
}
#if CONFIG_SMALL
for( dir = 0; dir < 2; dir++ )
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, CHROMA444, dir);
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, chroma, CHROMA444, dir);
#else
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, CHROMA444, 0);
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, CHROMA444, 1);
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, chroma, CHROMA444, 0);
filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, chroma, CHROMA444, 1);
#endif
}