falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

h264: move mb_{x,y} into the per-slice context

tags/n2.7
Anton Khirnov 11 years ago
parent
0edbe6faa7
commit
d4d9068cdf
13 changed files with 193 additions and 188 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +1
    -1
      libavcodec/dxva2_h264.c
  2. +8
    -8
      libavcodec/h264.c
  3. +3
    -2
      libavcodec/h264.h
  4. +9
    -9
      libavcodec/h264_cabac.c
  5. +12
    -12
      libavcodec/h264_cavlc.c
  6. +19
    -19
      libavcodec/h264_direct.c
  7. +22
    -22
      libavcodec/h264_mb.c
  8. +11
    -11
      libavcodec/h264_mb_template.c
  9. +1
    -1
      libavcodec/h264_mc_template.c
  10. +8
    -8
      libavcodec/h264_mvpred.h
  11. +67
    -63
      libavcodec/h264_slice.c
  12. +31
    -31
      libavcodec/svq3.c
  13. +1
    -1
      libavcodec/vaapi_h264.c

+ 1
- 1
libavcodec/dxva2_h264.c View File

@@ -220,7 +220,7 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice,
slice->SliceBytesInBuffer = size;
slice->wBadSliceChopping = 0;

slice->first_mb_in_slice = (h->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + h->mb_x;
slice->first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x;
slice->NumMbsForSlice = 0; /* XXX it is set once we have all slices */
slice->BitOffsetToSliceData = get_bits_count(&h->gb);
slice->slice_type = ff_h264_get_slice_type(sl);


+ 8
- 8
libavcodec/h264.c View File

@@ -58,8 +58,8 @@ static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
H264Context *h = opaque;
H264SliceContext *sl = &h->slice_ctx[0];

h->mb_x = mb_x;
h->mb_y = mb_y;
sl->mb_x = mb_x;
sl->mb_y = mb_y;
sl->mb_xy = mb_x + mb_y * h->mb_stride;
memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
assert(ref >= 0);
@@ -143,7 +143,7 @@ int ff_h264_check_intra4x4_pred_mode(H264Context *h, H264SliceContext *sl)
if (status < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"top block unavailable for requested intra4x4 mode %d at %d %d\n",
status, h->mb_x, h->mb_y);
status, sl->mb_x, sl->mb_y);
return AVERROR_INVALIDDATA;
} else if (status) {
sl->intra4x4_pred_mode_cache[scan8[0] + i] = status;
@@ -159,7 +159,7 @@ int ff_h264_check_intra4x4_pred_mode(H264Context *h, H264SliceContext *sl)
if (status < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"left block unavailable for requested intra4x4 mode %d at %d %d\n",
status, h->mb_x, h->mb_y);
status, sl->mb_x, sl->mb_y);
return AVERROR_INVALIDDATA;
} else if (status) {
sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
@@ -183,7 +183,7 @@ int ff_h264_check_intra_pred_mode(H264Context *h, H264SliceContext *sl,
if (mode > 3U) {
av_log(h->avctx, AV_LOG_ERROR,
"out of range intra chroma pred mode at %d %d\n",
h->mb_x, h->mb_y);
sl->mb_x, sl->mb_y);
return AVERROR_INVALIDDATA;
}

@@ -192,7 +192,7 @@ int ff_h264_check_intra_pred_mode(H264Context *h, H264SliceContext *sl,
if (mode < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"top block unavailable for requested intra mode at %d %d\n",
h->mb_x, h->mb_y);
sl->mb_x, sl->mb_y);
return AVERROR_INVALIDDATA;
}
}
@@ -208,7 +208,7 @@ int ff_h264_check_intra_pred_mode(H264Context *h, H264SliceContext *sl,
if (mode < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"left block unavailable for requested intra mode at %d %d\n",
h->mb_x, h->mb_y);
sl->mb_x, sl->mb_y);
return AVERROR_INVALIDDATA;
}
}
@@ -1114,7 +1114,7 @@ static void flush_dpb(AVCodecContext *avctx)
h->cur_pic_ptr = NULL;
ff_h264_unref_picture(h, &h->cur_pic);

h->mb_x = h->mb_y = 0;
h->mb_y = 0;

ff_h264_free_tables(h, 1);
h->context_initialized = 0;


+ 3
- 2
libavcodec/h264.h View File

@@ -356,6 +356,7 @@ typedef struct H264SliceContext {
ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
ptrdiff_t mb_uvlinesize;

int mb_x, mb_y;
int mb_xy;
int mb_skip_run;
int is_complex;
@@ -526,7 +527,7 @@ typedef struct H264Context {

int x264_build;

int mb_x, mb_y;
int mb_y;
int resync_mb_x;
int resync_mb_y;
int mb_height, mb_width;
@@ -1020,7 +1021,7 @@ static av_always_inline void write_back_motion(H264Context *h,
int mb_type)
{
const int b_stride = h->b_stride;
const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
const int b8_xy = 4 * sl->mb_xy;

if (USES_LIST(mb_type, 0)) {


+ 9
- 9
libavcodec/h264_cabac.c View File

@@ -1288,7 +1288,7 @@ static int decode_cabac_field_decoding_flag(H264Context *h, H264SliceContext *sl

unsigned long ctx = 0;

ctx += h->mb_field_decoding_flag & !!h->mb_x; //for FMO:(s->current_picture.mb_type[mba_xy] >> 7) & (h->slice_table[mba_xy] == h->slice_num);
ctx += h->mb_field_decoding_flag & !!sl->mb_x; //for FMO:(s->current_picture.mb_type[mba_xy] >> 7) & (h->slice_table[mba_xy] == h->slice_num);
ctx += (h->cur_pic.mb_type[mbb_xy] >> 7) & (h->slice_table[mbb_xy] == sl->slice_num);

return get_cabac_noinline( &sl->cabac, &(sl->cabac_state+70)[ctx] );
@@ -1914,21 +1914,21 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl)
int decode_chroma = h->sps.chroma_format_idc == 1 || h->sps.chroma_format_idc == 2;
const int pixel_shift = h->pixel_shift;

mb_xy = sl->mb_xy = h->mb_x + h->mb_y*h->mb_stride;
mb_xy = sl->mb_xy = sl->mb_x + sl->mb_y*h->mb_stride;

tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, h->mb_x, h->mb_y);
tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, sl->mb_x, sl->mb_y);
if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
int skip;
/* a skipped mb needs the aff flag from the following mb */
if (FRAME_MBAFF(h) && (h->mb_y & 1) == 1 && sl->prev_mb_skipped)
if (FRAME_MBAFF(h) && (sl->mb_y & 1) == 1 && sl->prev_mb_skipped)
skip = sl->next_mb_skipped;
else
skip = decode_cabac_mb_skip(h, sl, h->mb_x, h->mb_y );
skip = decode_cabac_mb_skip(h, sl, sl->mb_x, sl->mb_y );
/* read skip flags */
if( skip ) {
if (FRAME_MBAFF(h) && (h->mb_y & 1) == 0) {
if (FRAME_MBAFF(h) && (sl->mb_y & 1) == 0) {
h->cur_pic.mb_type[mb_xy] = MB_TYPE_SKIP;
sl->next_mb_skipped = decode_cabac_mb_skip(h, sl, h->mb_x, h->mb_y+1 );
sl->next_mb_skipped = decode_cabac_mb_skip(h, sl, sl->mb_x, sl->mb_y+1 );
if(!sl->next_mb_skipped)
h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h, sl);
}
@@ -1944,7 +1944,7 @@ int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl)
}
}
if (FRAME_MBAFF(h)) {
if( (h->mb_y&1) == 0 )
if ((sl->mb_y & 1) == 0)
h->mb_mbaff =
h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h, sl);
}
@@ -2367,7 +2367,7 @@ decode_intra_mb:
ctx= 3;
val++;
if(val > 2*max_qp){ //prevent infinite loop
av_log(h->avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", sl->mb_x, sl->mb_y);
return -1;
}
}


+ 12
- 12
libavcodec/h264_cavlc.c View File

@@ -477,7 +477,7 @@ static int decode_residual(H264Context *h, H264SliceContext *sl,
if(total_coeff==0)
return 0;
if(total_coeff > (unsigned)max_coeff) {
av_log(h->avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", h->mb_x, h->mb_y, total_coeff);
av_log(h->avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff=%d)\n", sl->mb_x, sl->mb_y, total_coeff);
return -1;
}

@@ -615,7 +615,7 @@ static int decode_residual(H264Context *h, H264SliceContext *sl,

if (zeros_left < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"negative number of zero coeffs at %d %d\n", h->mb_x, h->mb_y);
"negative number of zero coeffs at %d %d\n", sl->mb_x, sl->mb_y);
return AVERROR_INVALIDDATA;
}

@@ -703,9 +703,9 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
int decode_chroma = h->sps.chroma_format_idc == 1 || h->sps.chroma_format_idc == 2;
const int pixel_shift = h->pixel_shift;

mb_xy = sl->mb_xy = h->mb_x + h->mb_y*h->mb_stride;
mb_xy = sl->mb_xy = sl->mb_x + sl->mb_y*h->mb_stride;

tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, h->mb_x, h->mb_y);
tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, sl->mb_x, sl->mb_y);
cbp = 0; /* avoid warning. FIXME: find a solution without slowing
down the code */
if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
@@ -713,7 +713,7 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
sl->mb_skip_run = get_ue_golomb(&h->gb);

if (sl->mb_skip_run--) {
if(FRAME_MBAFF(h) && (h->mb_y&1) == 0){
if (FRAME_MBAFF(h) && (sl->mb_y & 1) == 0) {
if (sl->mb_skip_run == 0)
h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&h->gb);
}
@@ -722,7 +722,7 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
}
}
if (FRAME_MBAFF(h)) {
if( (h->mb_y&1) == 0 )
if ((sl->mb_y & 1) == 0)
h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&h->gb);
}

@@ -751,7 +751,7 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
mb_type--;
decode_intra_mb:
if(mb_type > 25){
av_log(h->avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_picture_type_char(sl->slice_type), h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_picture_type_char(sl->slice_type), sl->mb_x, sl->mb_y);
return -1;
}
partition_count=0;
@@ -838,7 +838,7 @@ decode_intra_mb:
for(i=0; i<4; i++){
sl->sub_mb_type[i]= get_ue_golomb_31(&h->gb);
if(sl->sub_mb_type[i] >=13){
av_log(h->avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], sl->mb_x, sl->mb_y);
return -1;
}
sub_partition_count[i]= b_sub_mb_type_info[ sl->sub_mb_type[i] ].partition_count;
@@ -856,7 +856,7 @@ decode_intra_mb:
for(i=0; i<4; i++){
sl->sub_mb_type[i]= get_ue_golomb_31(&h->gb);
if(sl->sub_mb_type[i] >=4){
av_log(h->avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], sl->mb_x, sl->mb_y);
return -1;
}
sub_partition_count[i]= p_sub_mb_type_info[ sl->sub_mb_type[i] ].partition_count;
@@ -1057,14 +1057,14 @@ decode_intra_mb:

if(decode_chroma){
if(cbp > 47){
av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, sl->mb_x, sl->mb_y);
return -1;
}
if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
else cbp= golomb_to_inter_cbp [cbp];
}else{
if(cbp > 15){
av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, sl->mb_x, sl->mb_y);
return -1;
}
if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp_gray[cbp];
@@ -1103,7 +1103,7 @@ decode_intra_mb:
if (sl->qscale < 0) sl->qscale += max_qp + 1;
else sl->qscale -= max_qp+1;
if (((unsigned)sl->qscale) > max_qp){
av_log(h->avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, h->mb_x, h->mb_y);
av_log(h->avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, sl->mb_x, sl->mb_y);
return -1;
}
}


+ 19
- 19
libavcodec/h264_direct.c View File

@@ -179,7 +179,7 @@ static void pred_spatial_direct_motion(H264Context *const h, H264SliceContext *s
{
int b8_stride = 2;
int b4_stride = h->b_stride;
int mb_xy = sl->mb_xy, mb_y = h->mb_y;
int mb_xy = sl->mb_xy, mb_y = sl->mb_y;
int mb_type_col[2];
const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1;
@@ -193,7 +193,7 @@ static void pred_spatial_direct_motion(H264Context *const h, H264SliceContext *s
assert(sl->ref_list[1][0].reference & 3);

await_reference_mb_row(h, &sl->ref_list[1][0],
h->mb_y + !!IS_INTERLACED(*mb_type));
sl->mb_y + !!IS_INTERLACED(*mb_type));

#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16 | MB_TYPE_INTRA4x4 | \
MB_TYPE_INTRA16x16 | MB_TYPE_INTRA_PCM)
@@ -262,9 +262,9 @@ static void pred_spatial_direct_motion(H264Context *const h, H264SliceContext *s

if (IS_INTERLACED(sl->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
mb_y = (h->mb_y & ~1) + sl->col_parity;
mb_xy = h->mb_x +
((h->mb_y & ~1) + sl->col_parity) * h->mb_stride;
mb_y = (sl->mb_y & ~1) + sl->col_parity;
mb_xy = sl->mb_x +
((sl->mb_y & ~1) + sl->col_parity) * h->mb_stride;
b8_stride = 0;
} else {
mb_y += sl->col_fieldoff;
@@ -273,8 +273,8 @@ static void pred_spatial_direct_motion(H264Context *const h, H264SliceContext *s
goto single_col;
} else { // AFL/AFR/FR/FL -> AFR/FR
if (IS_INTERLACED(*mb_type)) { // AFL /FL -> AFR/FR
mb_y = h->mb_y & ~1;
mb_xy = (h->mb_y & ~1) * h->mb_stride + h->mb_x;
mb_y = sl->mb_y & ~1;
mb_xy = (sl->mb_y & ~1) * h->mb_stride + sl->mb_x;
mb_type_col[0] = sl->ref_list[1][0].mb_type[mb_xy];
mb_type_col[1] = sl->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
b8_stride = 2 + 4 * h->mb_stride;
@@ -323,7 +323,7 @@ single_col:
l1ref0 = &sl->ref_list[1][0].ref_index[0][4 * mb_xy];
l1ref1 = &sl->ref_list[1][0].ref_index[1][4 * mb_xy];
if (!b8_stride) {
if (h->mb_y & 1) {
if (sl->mb_y & 1) {
l1ref0 += 2;
l1ref1 += 2;
l1mv0 += 2 * b4_stride;
@@ -465,7 +465,7 @@ static void pred_temp_direct_motion(H264Context *const h, H264SliceContext *sl,
{
int b8_stride = 2;
int b4_stride = h->b_stride;
int mb_xy = sl->mb_xy, mb_y = h->mb_y;
int mb_xy = sl->mb_xy, mb_y = sl->mb_y;
int mb_type_col[2];
const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1;
@@ -476,13 +476,13 @@ static void pred_temp_direct_motion(H264Context *const h, H264SliceContext *sl,
assert(sl->ref_list[1][0].reference & 3);

await_reference_mb_row(h, &sl->ref_list[1][0],
h->mb_y + !!IS_INTERLACED(*mb_type));
sl->mb_y + !!IS_INTERLACED(*mb_type));

if (IS_INTERLACED(sl->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
mb_y = (h->mb_y & ~1) + sl->col_parity;
mb_xy = h->mb_x +
((h->mb_y & ~1) + sl->col_parity) * h->mb_stride;
mb_y = (sl->mb_y & ~1) + sl->col_parity;
mb_xy = sl->mb_x +
((sl->mb_y & ~1) + sl->col_parity) * h->mb_stride;
b8_stride = 0;
} else {
mb_y += sl->col_fieldoff;
@@ -491,8 +491,8 @@ static void pred_temp_direct_motion(H264Context *const h, H264SliceContext *sl,
goto single_col;
} else { // AFL/AFR/FR/FL -> AFR/FR
if (IS_INTERLACED(*mb_type)) { // AFL /FL -> AFR/FR
mb_y = h->mb_y & ~1;
mb_xy = h->mb_x + (h->mb_y & ~1) * h->mb_stride;
mb_y = sl->mb_y & ~1;
mb_xy = sl->mb_x + (sl->mb_y & ~1) * h->mb_stride;
mb_type_col[0] = sl->ref_list[1][0].mb_type[mb_xy];
mb_type_col[1] = sl->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
b8_stride = 2 + 4 * h->mb_stride;
@@ -547,7 +547,7 @@ single_col:
l1ref0 = &sl->ref_list[1][0].ref_index[0][4 * mb_xy];
l1ref1 = &sl->ref_list[1][0].ref_index[1][4 * mb_xy];
if (!b8_stride) {
if (h->mb_y & 1) {
if (sl->mb_y & 1) {
l1ref0 += 2;
l1ref1 += 2;
l1mv0 += 2 * b4_stride;
@@ -562,9 +562,9 @@ single_col:
int ref_offset;

if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {
map_col_to_list0[0] = sl->map_col_to_list0_field[h->mb_y & 1][0];
map_col_to_list0[1] = sl->map_col_to_list0_field[h->mb_y & 1][1];
dist_scale_factor = sl->dist_scale_factor_field[h->mb_y & 1];
map_col_to_list0[0] = sl->map_col_to_list0_field[sl->mb_y & 1][0];
map_col_to_list0[1] = sl->map_col_to_list0_field[sl->mb_y & 1][1];
dist_scale_factor = sl->dist_scale_factor_field[sl->mb_y & 1];
}
ref_offset = (sl->ref_list[1][0].mbaff << 4) & (mb_type_col[0] >> 3);



+ 22
- 22
libavcodec/h264_mb.c View File

@@ -57,7 +57,7 @@ static inline void get_lowest_part_y(H264Context *h, H264SliceContext *sl,
{
int my;

y_offset += 16 * (h->mb_y >> MB_FIELD(h));
y_offset += 16 * (sl->mb_y >> MB_FIELD(h));

if (list0) {
int ref_n = sl->ref_cache[0][scan8[n]];
@@ -287,7 +287,7 @@ static av_always_inline void mc_dir_part(H264Context *h, H264SliceContext *sl,
ysh = 3 - (chroma_idc == 2 /* yuv422 */);
if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
// chroma offset when predicting from a field of opposite parity
my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
my += 2 * ((sl->mb_y & 1) - (pic->reference - 1));
emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
}

@@ -345,8 +345,8 @@ static av_always_inline void mc_part_std(H264Context *h, H264SliceContext *sl,
dest_cb += (x_offset << pixel_shift) + y_offset * sl->mb_uvlinesize;
dest_cr += (x_offset << pixel_shift) + y_offset * sl->mb_uvlinesize;
}
x_offset += 8 * h->mb_x;
y_offset += 8 * (h->mb_y >> MB_FIELD(h));
x_offset += 8 * sl->mb_x;
y_offset += 8 * (sl->mb_y >> MB_FIELD(h));

if (list0) {
H264Picture *ref = &sl->ref_list[0][sl->ref_cache[0][scan8[n]]];
@@ -399,8 +399,8 @@ static av_always_inline void mc_part_weighted(H264Context *h, H264SliceContext *
dest_cb += (x_offset << pixel_shift) + y_offset * sl->mb_uvlinesize;
dest_cr += (x_offset << pixel_shift) + y_offset * sl->mb_uvlinesize;
}
x_offset += 8 * h->mb_x;
y_offset += 8 * (h->mb_y >> MB_FIELD(h));
x_offset += 8 * sl->mb_x;
y_offset += 8 * (sl->mb_y >> MB_FIELD(h));

if (list0 && list1) {
/* don't optimize for luma-only case, since B-frames usually
@@ -421,7 +421,7 @@ static av_always_inline void mc_part_weighted(H264Context *h, H264SliceContext *
pixel_shift, chroma_idc);

if (sl->use_weight == 2) {
int weight0 = sl->implicit_weight[refn0][refn1][h->mb_y & 1];
int weight0 = sl->implicit_weight[refn0][refn1][sl->mb_y & 1];
int weight1 = 64 - weight0;
luma_weight_avg(dest_y, tmp_y, sl->mb_linesize,
height, 5, weight0, weight1, 0);
@@ -482,11 +482,11 @@ static av_always_inline void prefetch_motion(H264Context *h, H264SliceContext *s
* optimized for 64byte cache lines */
const int refn = sl->ref_cache[list][scan8[0]];
if (refn >= 0) {
const int mx = (sl->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
const int my = (sl->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
const int mx = (sl->mv_cache[list][scan8[0]][0] >> 2) + 16 * sl->mb_x + 8;
const int my = (sl->mv_cache[list][scan8[0]][1] >> 2) + 16 * sl->mb_y;
uint8_t **src = sl->ref_list[list][refn].f.data;
int off = (mx << pixel_shift) +
(my + (h->mb_x & 3) * 4) * sl->mb_linesize +
(my + (sl->mb_x & 3) * 4) * sl->mb_linesize +
(64 << pixel_shift);
h->vdsp.prefetch(src[0] + off, h->linesize, 4);
if (chroma_idc == 3 /* yuv444 */) {
@@ -494,7 +494,7 @@ static av_always_inline void prefetch_motion(H264Context *h, H264SliceContext *s
h->vdsp.prefetch(src[2] + off, h->linesize, 4);
} else {
off = ((mx >> 1) << pixel_shift) +
((my >> 1) + (h->mb_x & 7)) * h->uvlinesize +
((my >> 1) + (sl->mb_x & 7)) * h->uvlinesize +
(64 << pixel_shift);
h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
}
@@ -515,7 +515,7 @@ static av_always_inline void xchg_mb_border(H264Context *h, H264SliceContext *sl
uint8_t *top_border;

if (!simple && FRAME_MBAFF(h)) {
if (h->mb_y & 1) {
if (sl->mb_y & 1) {
if (!MB_MBAFF(h))
return;
} else {
@@ -527,16 +527,16 @@ static av_always_inline void xchg_mb_border(H264Context *h, H264SliceContext *sl
deblock_topleft = h->slice_table[sl->mb_xy - 1 - h->mb_stride] == sl->slice_num;
deblock_top = sl->top_type;
} else {
deblock_topleft = (h->mb_x > 0);
deblock_top = (h->mb_y > !!MB_FIELD(h));
deblock_topleft = (sl->mb_x > 0);
deblock_top = (sl->mb_y > !!MB_FIELD(h));
}

src_y -= linesize + 1 + pixel_shift;
src_cb -= uvlinesize + 1 + pixel_shift;
src_cr -= uvlinesize + 1 + pixel_shift;

top_border_m1 = h->top_borders[top_idx][h->mb_x - 1];
top_border = h->top_borders[top_idx][h->mb_x];
top_border_m1 = h->top_borders[top_idx][sl->mb_x - 1];
top_border = h->top_borders[top_idx][sl->mb_x];

#define XCHG(a, b, xchg) \
if (pixel_shift) { \
@@ -558,8 +558,8 @@ static av_always_inline void xchg_mb_border(H264Context *h, H264SliceContext *sl
}
XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
if (h->mb_x + 1 < h->mb_width) {
XCHG(h->top_borders[top_idx][h->mb_x + 1],
if (sl->mb_x + 1 < h->mb_width) {
XCHG(h->top_borders[top_idx][sl->mb_x + 1],
src_y + (17 << pixel_shift), 1);
}
}
@@ -574,9 +574,9 @@ static av_always_inline void xchg_mb_border(H264Context *h, H264SliceContext *sl
XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
if (h->mb_x + 1 < h->mb_width) {
XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
if (sl->mb_x + 1 < h->mb_width) {
XCHG(h->top_borders[top_idx][sl->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
XCHG(h->top_borders[top_idx][sl->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
}
}
} else {
@@ -671,7 +671,7 @@ static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
uint64_t tr_high;
if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
const int topright_avail = (sl->topright_samples_available << i) & 0x8000;
assert(h->mb_y || linesize <= block_offset[i]);
assert(sl->mb_y || linesize <= block_offset[i]);
if (!topright_avail) {
if (pixel_shift) {
tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;


+ 11
- 11
libavcodec/h264_mb_template.c View File

@@ -42,8 +42,8 @@

static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl)
{
const int mb_x = h->mb_x;
const int mb_y = h->mb_y;
const int mb_x = sl->mb_x;
const int mb_y = sl->mb_y;
const int mb_xy = sl->mb_xy;
const int mb_type = h->cur_pic.mb_type[mb_xy];
uint8_t *dest_y, *dest_cb, *dest_cr;
@@ -61,8 +61,8 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl)
dest_cb = h->cur_pic.f.data[1] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * h->uvlinesize * block_h;
dest_cr = h->cur_pic.f.data[2] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * h->uvlinesize * block_h;

h->vdsp.prefetch(dest_y + (h->mb_x & 3) * 4 * h->linesize + (64 << PIXEL_SHIFT), h->linesize, 4);
h->vdsp.prefetch(dest_cb + (h->mb_x & 7) * h->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2);
h->vdsp.prefetch(dest_y + (sl->mb_x & 3) * 4 * h->linesize + (64 << PIXEL_SHIFT), h->linesize, 4);
h->vdsp.prefetch(dest_cb + (sl->mb_x & 7) * h->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2);

h->list_counts[mb_xy] = sl->list_count;

@@ -82,13 +82,13 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl)
continue;
if (IS_16X16(mb_type)) {
int8_t *ref = &sl->ref_cache[list][scan8[0]];
fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (h->mb_y & 1), 1);
fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (sl->mb_y & 1), 1);
} else {
for (i = 0; i < 16; i += 4) {
int ref = sl->ref_cache[list][scan8[i]];
if (ref >= 0)
fill_rectangle(&sl->ref_cache[list][scan8[i]], 2, 2,
8, (16 + ref) ^ (h->mb_y & 1), 1);
8, (16 + ref) ^ (sl->mb_y & 1), 1);
}
}
}
@@ -274,8 +274,8 @@ static av_noinline void FUNC(hl_decode_mb)(H264Context *h, H264SliceContext *sl)

static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext *sl)
{
const int mb_x = h->mb_x;
const int mb_y = h->mb_y;
const int mb_x = sl->mb_x;
const int mb_y = sl->mb_y;
const int mb_xy = sl->mb_xy;
const int mb_type = h->cur_pic.mb_type[mb_xy];
uint8_t *dest[3];
@@ -288,7 +288,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext
for (p = 0; p < plane_count; p++) {
dest[p] = h->cur_pic.f.data[p] +
((mb_x << PIXEL_SHIFT) + mb_y * h->linesize) * 16;
h->vdsp.prefetch(dest[p] + (h->mb_x & 3) * 4 * h->linesize + (64 << PIXEL_SHIFT),
h->vdsp.prefetch(dest[p] + (sl->mb_x & 3) * 4 * h->linesize + (64 << PIXEL_SHIFT),
h->linesize, 4);
}

@@ -307,13 +307,13 @@ static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h, H264SliceContext
continue;
if (IS_16X16(mb_type)) {
int8_t *ref = &sl->ref_cache[list][scan8[0]];
fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (h->mb_y & 1), 1);
fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (sl->mb_y & 1), 1);
} else {
for (i = 0; i < 16; i += 4) {
int ref = sl->ref_cache[list][scan8[i]];
if (ref >= 0)
fill_rectangle(&sl->ref_cache[list][scan8[i]], 2, 2,
8, (16 + ref) ^ (h->mb_y & 1), 1);
8, (16 + ref) ^ (sl->mb_y & 1), 1);
}
}
}


+ 1
- 1
libavcodec/h264_mc_template.c View File

@@ -49,7 +49,7 @@ static void mc_part(H264Context *h, H264SliceContext *sl,
int list0, int list1)
{
if ((sl->use_weight == 2 && list0 && list1 &&
(sl->implicit_weight[sl->ref_cache[0][scan8[n]]][sl->ref_cache[1][scan8[n]]][h->mb_y & 1] != 32)) ||
(sl->implicit_weight[sl->ref_cache[0][scan8[n]]][sl->ref_cache[1][scan8[n]]][sl->mb_y & 1] != 32)) ||
sl->use_weight == 1)
mc_part_weighted(h, sl, n, square, height, delta, dest_y, dest_cb, dest_cr,
x_offset, y_offset, qpix_put, chroma_put,


+ 8
- 8
libavcodec/h264_mvpred.h View File

@@ -64,7 +64,7 @@ static av_always_inline int fetch_diagonal_mv(H264Context *h, H264SliceContext *

if (!MB_FIELD(h) && IS_INTERLACED(sl->left_type[0])) {
SET_DIAG_MV(* 2, >> 1, sl->left_mb_xy[0] + h->mb_stride,
(h->mb_y & 1) * 2 + (i >> 5));
(sl->mb_y & 1) * 2 + (i >> 5));
}
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.
@@ -148,7 +148,7 @@ static av_always_inline void pred_motion(H264Context *const h,
tprintf(h->avctx,
"pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n",
top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref,
A[0], A[1], ref, *mx, *my, h->mb_x, h->mb_y, n, list);
A[0], A[1], ref, *mx, *my, sl->mb_x, sl->mb_y, n, list);
}

/**
@@ -167,7 +167,7 @@ static av_always_inline void pred_16x8_motion(H264Context *const h,
const int16_t *const B = sl->mv_cache[list][scan8[0] - 8];

tprintf(h->avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n",
top_ref, B[0], B[1], h->mb_x, h->mb_y, n, list);
top_ref, B[0], B[1], sl->mb_x, sl->mb_y, n, list);

if (top_ref == ref) {
*mx = B[0];
@@ -179,7 +179,7 @@ static av_always_inline void pred_16x8_motion(H264Context *const h,
const int16_t *const A = sl->mv_cache[list][scan8[8] - 1];

tprintf(h->avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n",
left_ref, A[0], A[1], h->mb_x, h->mb_y, n, list);
left_ref, A[0], A[1], sl->mb_x, sl->mb_y, n, list);

if (left_ref == ref) {
*mx = A[0];
@@ -208,7 +208,7 @@ static av_always_inline void pred_8x16_motion(H264Context *const h,
const int16_t *const A = sl->mv_cache[list][scan8[0] - 1];

tprintf(h->avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n",
left_ref, A[0], A[1], h->mb_x, h->mb_y, n, list);
left_ref, A[0], A[1], sl->mb_x, sl->mb_y, n, list);

if (left_ref == ref) {
*mx = A[0];
@@ -222,7 +222,7 @@ static av_always_inline void pred_8x16_motion(H264Context *const h,
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",
diagonal_ref, C[0], C[1], h->mb_x, h->mb_y, n, list);
diagonal_ref, C[0], C[1], sl->mb_x, sl->mb_y, n, list);

if (diagonal_ref == ref) {
*mx = C[0];
@@ -299,7 +299,7 @@ static av_always_inline void pred_pskip_motion(H264Context *const h,
}

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, sl->mb_x, sl->mb_y);

if (USES_LIST(sl->topright_type, 0)) {
diagonal_ref = ref[4 * sl->topright_mb_xy + 2];
@@ -378,7 +378,7 @@ static void fill_decode_neighbors(H264Context *h, H264SliceContext *sl, int mb_t
if (FRAME_MBAFF(h)) {
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);
if (h->mb_y & 1) {
if (sl->mb_y & 1) {
if (left_mb_field_flag != curr_mb_field_flag) {
left_xy[LBOT] = left_xy[LTOP] = mb_xy - h->mb_stride - 1;
if (curr_mb_field_flag) {


+ 67
- 63
libavcodec/h264_slice.c View File

@@ -719,7 +719,8 @@ static int h264_frame_start(H264Context *h)
return 0;
}

static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *sl,
uint8_t *src_y,
uint8_t *src_cb, uint8_t *src_cr,
int linesize, int uvlinesize,
int simple)
@@ -735,9 +736,9 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
src_cr -= uvlinesize;

if (!simple && FRAME_MBAFF(h)) {
if (h->mb_y & 1) {
if (sl->mb_y & 1) {
if (!MB_MBAFF(h)) {
top_border = h->top_borders[0][h->mb_x];
top_border = h->top_borders[0][sl->mb_x];
AV_COPY128(top_border, src_y + 15 * linesize);
if (pixel_shift)
AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
@@ -777,7 +778,7 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
return;
}

top_border = h->top_borders[top_idx][h->mb_x];
top_border = h->top_borders[top_idx][sl->mb_x];
/* There are two lines saved, the line above the top macroblock
* of a pair, and the line above the bottom macroblock. */
AV_COPY128(top_border, src_y + 16 * linesize);
@@ -1207,8 +1208,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
slice_type = get_ue_golomb_31(&h->gb);
if (slice_type > 9) {
av_log(h->avctx, AV_LOG_ERROR,
"slice type %d too large at %d %d\n",
slice_type, h->mb_x, h->mb_y);
"slice type %d too large at %d\n",
slice_type, first_mb_in_slice);
return AVERROR_INVALIDDATA;
}
if (slice_type > 4) {
@@ -1556,12 +1557,12 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
return AVERROR_INVALIDDATA;
}
h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
h->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
h->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
FIELD_OR_MBAFF_PICTURE(h);
if (h->picture_structure == PICT_BOTTOM_FIELD)
h->resync_mb_y = h->mb_y = h->mb_y + 1;
assert(h->mb_y < h->mb_height);
h->resync_mb_y = sl->mb_y = sl->mb_y + 1;
assert(sl->mb_y < h->mb_height);

if (h->picture_structure == PICT_FRAME) {
h->curr_pic_num = h->frame_num;
@@ -1895,7 +1896,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h,
}

{
int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
@@ -1924,7 +1925,7 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
if (FRAME_MBAFF(h)) {
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);
if (h->mb_y & 1) {
if (sl->mb_y & 1) {
if (left_mb_field_flag != curr_mb_field_flag)
left_xy[LTOP] -= h->mb_stride;
} else {
@@ -2056,7 +2057,7 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
{
uint8_t *dest_y, *dest_cb, *dest_cr;
int linesize, uvlinesize, mb_x, mb_y;
const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
const int old_slice_type = sl->slice_type;
const int pixel_shift = h->pixel_shift;
const int block_h = 16 >> h->chroma_y_shift;
@@ -2074,8 +2075,8 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
h->mb_mbaff =
h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);

h->mb_x = mb_x;
h->mb_y = mb_y;
sl->mb_x = mb_x;
sl->mb_y = mb_y;
dest_y = h->cur_pic.f.data[0] +
((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
dest_cb = h->cur_pic.f.data[1] +
@@ -2098,7 +2099,7 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
linesize = sl->mb_linesize = h->linesize;
uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
}
backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
uvlinesize, 0);
if (fill_filter_caches(h, sl, mb_type))
continue;
@@ -2115,15 +2116,15 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
}
}
sl->slice_type = old_slice_type;
h->mb_x = end_x;
h->mb_y = end_mb_y - FRAME_MBAFF(h);
sl->mb_x = end_x;
sl->mb_y = end_mb_y - FRAME_MBAFF(h);
sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
}

static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
{
const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
h->cur_pic.mb_type[mb_xy - 1] :
(h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
@@ -2136,7 +2137,7 @@ static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
*/
static void decode_finish_row(H264Context *h, H264SliceContext *sl)
{
int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
int height = 16 << FRAME_MBAFF(h);
int deblock_border = (16 + 4) << FRAME_MBAFF(h);
@@ -2181,7 +2182,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
{
H264SliceContext *sl = arg;
H264Context *h = sl->h264;
int lf_x_start = h->mb_x;
int lf_x_start = sl->mb_x;

sl->mb_skip_run = -1;

@@ -2211,53 +2212,53 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)

// FIXME optimal? or let mb_decode decode 16x32 ?
if (ret >= 0 && FRAME_MBAFF(h)) {
h->mb_y++;
sl->mb_y++;

ret = ff_h264_decode_mb_cabac(h, sl);

if (ret >= 0)
ff_h264_hl_decode_mb(h, sl);
h->mb_y--;
sl->mb_y--;
}
eos = get_cabac_terminate(&sl->cabac);

if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
h->mb_y, ER_MB_END);
if (h->mb_x >= lf_x_start)
loop_filter(h, sl, lf_x_start, h->mb_x + 1);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, sl->mb_x - 1,
sl->mb_y, ER_MB_END);
if (sl->mb_x >= lf_x_start)
loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
return 0;
}
if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
av_log(h->avctx, AV_LOG_ERROR,
"error while decoding MB %d %d, bytestream %td\n",
h->mb_x, h->mb_y,
sl->mb_x, sl->mb_y,
sl->cabac.bytestream_end - sl->cabac.bytestream);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
h->mb_y, ER_MB_ERROR);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, sl->mb_x,
sl->mb_y, ER_MB_ERROR);
return AVERROR_INVALIDDATA;
}

if (++h->mb_x >= h->mb_width) {
loop_filter(h, sl, lf_x_start, h->mb_x);
h->mb_x = lf_x_start = 0;
if (++sl->mb_x >= h->mb_width) {
loop_filter(h, sl, lf_x_start, sl->mb_x);
sl->mb_x = lf_x_start = 0;
decode_finish_row(h, sl);
++h->mb_y;
++sl->mb_y;
if (FIELD_OR_MBAFF_PICTURE(h)) {
++h->mb_y;
if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
++sl->mb_y;
if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
predict_field_decoding_flag(h, sl);
}
}

if (eos || h->mb_y >= h->mb_height) {
if (eos || sl->mb_y >= h->mb_height) {
tprintf(h->avctx, "slice end %d %d\n",
get_bits_count(&h->gb), h->gb.size_in_bits);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
h->mb_y, ER_MB_END);
if (h->mb_x > lf_x_start)
loop_filter(h, sl, lf_x_start, h->mb_x);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, sl->mb_x - 1,
sl->mb_y, ER_MB_END);
if (sl->mb_x > lf_x_start)
loop_filter(h, sl, lf_x_start, sl->mb_x);
return 0;
}
}
@@ -2270,44 +2271,44 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)

// FIXME optimal? or let mb_decode decode 16x32 ?
if (ret >= 0 && FRAME_MBAFF(h)) {
h->mb_y++;
sl->mb_y++;
ret = ff_h264_decode_mb_cavlc(h, sl);

if (ret >= 0)
ff_h264_hl_decode_mb(h, sl);
h->mb_y--;
sl->mb_y--;
}

if (ret < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"error while decoding MB %d %d\n", h->mb_x, h->mb_y);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
h->mb_y, ER_MB_ERROR);
"error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, sl->mb_x,
sl->mb_y, ER_MB_ERROR);
return ret;
}

if (++h->mb_x >= h->mb_width) {
loop_filter(h, sl, lf_x_start, h->mb_x);
h->mb_x = lf_x_start = 0;
if (++sl->mb_x >= h->mb_width) {
loop_filter(h, sl, lf_x_start, sl->mb_x);
sl->mb_x = lf_x_start = 0;
decode_finish_row(h, sl);
++h->mb_y;
++sl->mb_y;
if (FIELD_OR_MBAFF_PICTURE(h)) {
++h->mb_y;
if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
++sl->mb_y;
if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
predict_field_decoding_flag(h, sl);
}
if (h->mb_y >= h->mb_height) {
if (sl->mb_y >= h->mb_height) {
tprintf(h->avctx, "slice end %d %d\n",
get_bits_count(&h->gb), h->gb.size_in_bits);

if (get_bits_left(&h->gb) == 0) {
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, ER_MB_END);
sl->mb_x - 1, sl->mb_y, ER_MB_END);

return 0;
} else {
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, ER_MB_END);
sl->mb_x - 1, sl->mb_y, ER_MB_END);

return AVERROR_INVALIDDATA;
}
@@ -2320,14 +2321,14 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)

if (get_bits_left(&h->gb) == 0) {
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, ER_MB_END);
if (h->mb_x > lf_x_start)
loop_filter(h, sl, lf_x_start, h->mb_x);
sl->mb_x - 1, sl->mb_y, ER_MB_END);
if (sl->mb_x > lf_x_start)
loop_filter(h, sl, lf_x_start, sl->mb_x);

return 0;
} else {
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
h->mb_y, ER_MB_ERROR);
er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, sl->mb_x,
sl->mb_y, ER_MB_ERROR);

return AVERROR_INVALIDDATA;
}
@@ -2346,12 +2347,15 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
{
AVCodecContext *const avctx = h->avctx;
H264Context *hx;
H264SliceContext *sl;
int i;

if (h->avctx->hwaccel)
return 0;
if (context_count == 1) {
return decode_slice(avctx, &h->slice_ctx[0]);
int ret = decode_slice(avctx, &h->slice_ctx[0]);
h->mb_y = h->slice_ctx[0].mb_y;
return ret;
} else {
for (i = 1; i < context_count; i++) {
hx = h->thread_context[i];
@@ -2363,8 +2367,8 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)

/* pull back stuff from slices to master context */
hx = h->thread_context[context_count - 1];
h->mb_x = hx->mb_x;
h->mb_y = hx->mb_y;
sl = &h->slice_ctx[context_count - 1];
h->mb_y = sl->mb_y;
h->droppable = hx->droppable;
h->picture_structure = hx->picture_structure;
for (i = 1; i < context_count; i++)


+ 31
- 31
libavcodec/svq3.c View File

@@ -378,11 +378,11 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode,

for (i = 0; i < 16; i += part_height)
for (j = 0; j < 16; j += part_width) {
const int b_xy = (4 * h->mb_x + (j >> 2)) +
(4 * h->mb_y + (i >> 2)) * h->b_stride;
const int b_xy = (4 * sl->mb_x + (j >> 2)) +
(4 * sl->mb_y + (i >> 2)) * h->b_stride;
int dxy;
x = 16 * h->mb_x + j;
y = 16 * h->mb_y + i;
x = 16 * sl->mb_x + j;
y = 16 * sl->mb_y + i;
k = (j >> 2 & 1) + (i >> 1 & 2) +
(j >> 1 & 4) + (i & 8);

@@ -488,20 +488,20 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)
uint32_t vlc;
int8_t *top, *left;
const int mb_xy = sl->mb_xy;
const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride;
const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride;

sl->top_samples_available = (h->mb_y == 0) ? 0x33FF : 0xFFFF;
sl->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF;
sl->top_samples_available = (sl->mb_y == 0) ? 0x33FF : 0xFFFF;
sl->left_samples_available = (sl->mb_x == 0) ? 0x5F5F : 0xFFFF;
sl->topright_samples_available = 0xFFFF;

if (mb_type == 0) { /* SKIP */
if (h->pict_type == AV_PICTURE_TYPE_P ||
s->next_pic->mb_type[mb_xy] == -1) {
svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16,
svq3_mc_dir_part(s, 16 * sl->mb_x, 16 * sl->mb_y, 16, 16,
0, 0, 0, 0, 0, 0);

if (h->pict_type == AV_PICTURE_TYPE_B)
svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16,
svq3_mc_dir_part(s, 16 * sl->mb_x, 16 * sl->mb_y, 16, 16,
0, 0, 0, 0, 1, 1);

mb_type = MB_TYPE_SKIP;
@@ -533,7 +533,7 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)
*/

for (m = 0; m < 2; m++) {
if (h->mb_x > 0 && sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6] != -1) {
if (sl->mb_x > 0 && sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6] != -1) {
for (i = 0; i < 4; i++)
AV_COPY32(sl->mv_cache[m][scan8[0] - 1 + i * 8],
h->cur_pic.motion_val[m][b_xy - 1 + i * h->b_stride]);
@@ -541,14 +541,14 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)
for (i = 0; i < 4; i++)
AV_ZERO32(sl->mv_cache[m][scan8[0] - 1 + i * 8]);
}
if (h->mb_y > 0) {
if (sl->mb_y > 0) {
memcpy(sl->mv_cache[m][scan8[0] - 1 * 8],
h->cur_pic.motion_val[m][b_xy - h->b_stride],
4 * 2 * sizeof(int16_t));
memset(&sl->ref_cache[m][scan8[0] - 1 * 8],
(sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1, 4);

if (h->mb_x < h->mb_width - 1) {
if (sl->mb_x < h->mb_width - 1) {
AV_COPY32(sl->mv_cache[m][scan8[0] + 4 - 1 * 8],
h->cur_pic.motion_val[m][b_xy - h->b_stride + 4]);
sl->ref_cache[m][scan8[0] + 4 - 1 * 8] =
@@ -556,7 +556,7 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)
sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1;
} else
sl->ref_cache[m][scan8[0] + 4 - 1 * 8] = PART_NOT_AVAILABLE;
if (h->mb_x > 0) {
if (sl->mb_x > 0) {
AV_COPY32(sl->mv_cache[m][scan8[0] - 1 - 1 * 8],
h->cur_pic.motion_val[m][b_xy - h->b_stride - 1]);
sl->ref_cache[m][scan8[0] - 1 - 1 * 8] =
@@ -599,13 +599,13 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)
memset(sl->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t));

if (mb_type == 8) {
if (h->mb_x > 0) {
if (sl->mb_x > 0) {
for (i = 0; i < 4; i++)
sl->intra4x4_pred_mode_cache[scan8[0] - 1 + i * 8] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6 - i];
if (sl->intra4x4_pred_mode_cache[scan8[0] - 1] == -1)
sl->left_samples_available = 0x5F5F;
}
if (h->mb_y > 0) {
if (sl->mb_y > 0) {
sl->intra4x4_pred_mode_cache[4 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 0];
sl->intra4x4_pred_mode_cache[5 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 1];
sl->intra4x4_pred_mode_cache[6 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 2];
@@ -646,8 +646,8 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)
if (mb_type == 8) {
ff_h264_check_intra4x4_pred_mode(h, sl);

sl->top_samples_available = (h->mb_y == 0) ? 0x33FF : 0xFFFF;
sl->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF;
sl->top_samples_available = (sl->mb_y == 0) ? 0x33FF : 0xFFFF;
sl->left_samples_available = (sl->mb_x == 0) ? 0x5F5F : 0xFFFF;
} else {
for (i = 0; i < 4; i++)
memset(&sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_128_PRED, 4);
@@ -822,7 +822,7 @@ static int svq3_decode_slice_header(AVCodecContext *avctx)
if ((header & 0x9F) == 2) {
i = (h->mb_num < 64) ? 6 : (1 + av_log2(h->mb_num - 1));
sl->mb_skip_run = get_bits(&h->gb, i) -
(h->mb_y * h->mb_width + h->mb_x);
(sl->mb_y * h->mb_width + sl->mb_x);
} else {
skip_bits1(&h->gb);
sl->mb_skip_run = 0;
@@ -845,17 +845,17 @@ static int svq3_decode_slice_header(AVCodecContext *avctx)
skip_bits(&h->gb, 8);

/* reset intra predictors and invalidate motion vector references */
if (h->mb_x > 0) {
if (sl->mb_x > 0) {
memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - 1] + 3,
-1, 4 * sizeof(int8_t));
memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_x],
-1, 8 * sizeof(int8_t) * h->mb_x);
memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - sl->mb_x],
-1, 8 * sizeof(int8_t) * sl->mb_x);
}
if (h->mb_y > 0) {
if (sl->mb_y > 0) {
memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_stride],
-1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));
-1, 8 * sizeof(int8_t) * (h->mb_width - sl->mb_x));

if (h->mb_x > 0)
if (sl->mb_x > 0)
sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] = -1;
}

@@ -1132,7 +1132,7 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data,

init_get_bits(&h->gb, buf, 8 * buf_size);

h->mb_x = h->mb_y = sl->mb_xy = 0;
sl->mb_x = sl->mb_y = sl->mb_xy = 0;

if (svq3_decode_slice_header(avctx))
return -1;
@@ -1245,10 +1245,10 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data,
}
}

for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {
for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {
for (sl->mb_y = 0; sl->mb_y < h->mb_height; sl->mb_y++) {
for (sl->mb_x = 0; sl->mb_x < h->mb_width; sl->mb_x++) {
unsigned mb_type;
sl->mb_xy = h->mb_x + h->mb_y * h->mb_stride;
sl->mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;

if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&
((get_bits_count(&h->gb) & 7) == 0 ||
@@ -1270,7 +1270,7 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data,
mb_type += 4;
if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {
av_log(h->avctx, AV_LOG_ERROR,
"error while decoding MB %d %d\n", h->mb_x, h->mb_y);
"error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
return -1;
}

@@ -1278,13 +1278,13 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data,
ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);

if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)
h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =
h->cur_pic.mb_type[sl->mb_x + sl->mb_y * h->mb_stride] =
(h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;
}

ff_draw_horiz_band(avctx, &s->cur_pic->f,
s->last_pic->f.data[0] ? &s->last_pic->f : NULL,
16 * h->mb_y, 16, h->picture_structure, 0,
16 * sl->mb_y, 16, h->picture_structure, 0,
h->low_delay);
}



+ 1
- 1
libavcodec/vaapi_h264.c View File

@@ -329,7 +329,7 @@ static int vaapi_h264_decode_slice(AVCodecContext *avctx,
if (!slice_param)
return -1;
slice_param->slice_data_bit_offset = get_bits_count(&h->gb) + 8; /* bit buffer started beyond nal_unit_type */
slice_param->first_mb_in_slice = (h->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + h->mb_x;
slice_param->first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x;
slice_param->slice_type = ff_h264_get_slice_type(sl);
slice_param->direct_spatial_mv_pred_flag = sl->slice_type == AV_PICTURE_TYPE_B ? sl->direct_spatial_mv_pred : 0;
slice_param->num_ref_idx_l0_active_minus1 = sl->list_count > 0 ? sl->ref_count[0] - 1 : 0;


Write Preview
Loading…
Cancel
Save