h264: K&R formatting cosmetics

Signed-off-by: Diego Biurrun <diego@biurrun.de>
Signed-off-by: Vittorio Giovara <vittorio.giovara@gmail.com>
pull/77/merge
Luca Barbato 11 years ago committed by Diego Biurrun
parent a11ef610ee
commit 2db953f846
  1. 23
      libavcodec/h264.c
  2. 758
      libavcodec/h264_direct.c
  3. 20
      libavcodec/h264_slice.c

@ -501,8 +501,10 @@ int ff_h264_context_init(H264Context *h)
er->mb_stride = h->mb_stride; er->mb_stride = h->mb_stride;
er->b8_stride = h->mb_width * 2 + 1; er->b8_stride = h->mb_width * 2 + 1;
FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int), // error resilience code looks cleaner with this
fail); // error ressilience code looks cleaner with this FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
(h->mb_num + 1) * sizeof(int), fail);
for (y = 0; y < h->mb_height; y++) for (y = 0; y < h->mb_height; y++)
for (x = 0; x < h->mb_width; x++) for (x = 0; x < h->mb_width; x++)
er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride; er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
@ -518,10 +520,11 @@ int ff_h264_context_init(H264Context *h)
FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail); FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail);
FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride, FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
fail); h->mb_height * h->mb_stride, fail);
FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail); FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base,
yc_size * sizeof(int16_t), fail);
er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2; er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2;
er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1; er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1;
er->dc_val[2] = er->dc_val[1] + c_size; er->dc_val[2] = er->dc_val[1] + c_size;
@ -586,7 +589,7 @@ int ff_h264_decode_extradata(H264Context *h)
} }
p += nalsize; p += nalsize;
} }
// Now store right nal length size, that will be used to parse all other nals // Store right nal length size that will be used to parse all other nals
h->nal_length_size = (avctx->extradata[4] & 0x03) + 1; h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
} else { } else {
h->is_avc = 0; h->is_avc = 0;
@ -1037,8 +1040,8 @@ int ff_pred_weight_table(H264Context *h)
static void idr(H264Context *h) static void idr(H264Context *h)
{ {
ff_h264_remove_all_refs(h); ff_h264_remove_all_refs(h);
h->prev_frame_num = 0; h->prev_frame_num =
h->prev_frame_num_offset = 0; h->prev_frame_num_offset =
h->prev_poc_msb = h->prev_poc_msb =
h->prev_poc_lsb = 0; h->prev_poc_lsb = 0;
} }
@ -1692,9 +1695,9 @@ end:
static int get_consumed_bytes(int pos, int buf_size) static int get_consumed_bytes(int pos, int buf_size)
{ {
if (pos == 0) if (pos == 0)
pos = 1; // avoid infinite loops (i doubt that is needed but ...) pos = 1; // avoid infinite loops (I doubt that is needed but...)
if (pos + 10 > buf_size) if (pos + 10 > buf_size)
pos = buf_size; // oops ;) pos = buf_size; // oops ;)
return pos; return pos;
} }

@ -34,63 +34,67 @@
#include <assert.h> #include <assert.h>
static int get_scale_factor(H264Context *const h, int poc, int poc1, int i)
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){ {
int poc0 = h->ref_list[0][i].poc; int poc0 = h->ref_list[0][i].poc;
int td = av_clip(poc1 - poc0, -128, 127); int td = av_clip(poc1 - poc0, -128, 127);
if(td == 0 || h->ref_list[0][i].long_ref){ if (td == 0 || h->ref_list[0][i].long_ref) {
return 256; return 256;
}else{ } else {
int tb = av_clip(poc - poc0, -128, 127); int tb = av_clip(poc - poc0, -128, 127);
int tx = (16384 + (FFABS(td) >> 1)) / td; int tx = (16384 + (FFABS(td) >> 1)) / td;
return av_clip((tb*tx + 32) >> 6, -1024, 1023); return av_clip((tb * tx + 32) >> 6, -1024, 1023);
} }
} }
void ff_h264_direct_dist_scale_factor(H264Context * const h){ void ff_h264_direct_dist_scale_factor(H264Context *const h)
const int poc = h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD]; {
const int poc = h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD];
const int poc1 = h->ref_list[1][0].poc; const int poc1 = h->ref_list[1][0].poc;
int i, field; int i, field;
if (FRAME_MBAFF(h)) if (FRAME_MBAFF(h))
for (field = 0; field < 2; field++){ for (field = 0; field < 2; field++) {
const int poc = h->cur_pic_ptr->field_poc[field]; const int poc = h->cur_pic_ptr->field_poc[field];
const int poc1 = h->ref_list[1][0].field_poc[field]; const int poc1 = h->ref_list[1][0].field_poc[field];
for (i = 0; i < 2 * h->ref_count[0]; i++) for (i = 0; i < 2 * h->ref_count[0]; i++)
h->dist_scale_factor_field[field][i^field] = h->dist_scale_factor_field[field][i ^ field] =
get_scale_factor(h, poc, poc1, i+16); get_scale_factor(h, poc, poc1, i + 16);
} }
for (i = 0; i < h->ref_count[0]; i++){ for (i = 0; i < h->ref_count[0]; i++)
h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i); h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
}
} }
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){ static void fill_colmap(H264Context *h, int map[2][16 + 32], int list,
H264Picture * const ref1 = &h->ref_list[1][0]; int field, int colfield, int mbafi)
{
H264Picture *const ref1 = &h->ref_list[1][0];
int j, old_ref, rfield; int j, old_ref, rfield;
int start= mbafi ? 16 : 0; int start = mbafi ? 16 : 0;
int end = mbafi ? 16+2*h->ref_count[0] : h->ref_count[0]; int end = mbafi ? 16 + 2 * h->ref_count[0] : h->ref_count[0];
int interl= mbafi || h->picture_structure != PICT_FRAME; int interl = mbafi || h->picture_structure != PICT_FRAME;
/* bogus; fills in for missing frames */ /* bogus; fills in for missing frames */
memset(map[list], 0, sizeof(map[list])); memset(map[list], 0, sizeof(map[list]));
for(rfield=0; rfield<2; rfield++){ for (rfield = 0; rfield < 2; rfield++) {
for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){ for (old_ref = 0; old_ref < ref1->ref_count[colfield][list]; old_ref++) {
int poc = ref1->ref_poc[colfield][list][old_ref]; int poc = ref1->ref_poc[colfield][list][old_ref];
if (!interl) if (!interl)
poc |= 3; poc |= 3;
else if( interl && (poc&3) == 3) // FIXME: store all MBAFF references so this is not needed // FIXME: store all MBAFF references so this is not needed
poc= (poc&~3) + rfield + 1; else if (interl && (poc & 3) == 3)
poc = (poc & ~3) + rfield + 1;
for(j=start; j<end; j++){
if (4 * h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference & 3) == poc) { for (j = start; j < end; j++) {
int cur_ref= mbafi ? (j-16)^field : j; if (4 * h->ref_list[0][j].frame_num +
(h->ref_list[0][j].reference & 3) == poc) {
int cur_ref = mbafi ? (j - 16) ^ field : j;
if (ref1->mbaff) if (ref1->mbaff)
map[list][2 * old_ref + (rfield^field) + 16] = cur_ref; map[list][2 * old_ref + (rfield ^ field) + 16] = cur_ref;
if(rfield == field || !interl) if (rfield == field || !interl)
map[list][old_ref] = cur_ref; map[list][old_ref] = cur_ref;
break; break;
} }
@ -99,65 +103,75 @@ static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field,
} }
} }
void ff_h264_direct_ref_list_init(H264Context * const h){ void ff_h264_direct_ref_list_init(H264Context *const h)
H264Picture * const ref1 = &h->ref_list[1][0]; {
H264Picture * const cur = h->cur_pic_ptr; H264Picture *const ref1 = &h->ref_list[1][0];
H264Picture *const cur = h->cur_pic_ptr;
int list, j, field; int list, j, field;
int sidx= (h->picture_structure&1)^1; int sidx = (h->picture_structure & 1) ^ 1;
int ref1sidx = (ref1->reference&1)^1; int ref1sidx = (ref1->reference & 1) ^ 1;
for(list=0; list<2; list++){ for (list = 0; list < 2; list++) {
cur->ref_count[sidx][list] = h->ref_count[list]; cur->ref_count[sidx][list] = h->ref_count[list];
for(j=0; j<h->ref_count[list]; j++) for (j = 0; j < h->ref_count[list]; j++)
cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference & 3); cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num +
(h->ref_list[list][j].reference & 3);
} }
if(h->picture_structure == PICT_FRAME){ if (h->picture_structure == PICT_FRAME) {
memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0])); memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0])); memcpy(cur->ref_poc[1], cur->ref_poc[0], sizeof(cur->ref_poc[0]));
} }
cur->mbaff = FRAME_MBAFF(h); cur->mbaff = FRAME_MBAFF(h);
h->col_fieldoff= 0; h->col_fieldoff = 0;
if(h->picture_structure == PICT_FRAME){ if (h->picture_structure == PICT_FRAME) {
int cur_poc = h->cur_pic_ptr->poc; int cur_poc = h->cur_pic_ptr->poc;
int *col_poc = h->ref_list[1]->field_poc; int *col_poc = h->ref_list[1]->field_poc;
h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc)); h->col_parity = (FFABS(col_poc[0] - cur_poc) >=
ref1sidx=sidx= h->col_parity; FFABS(col_poc[1] - cur_poc));
} else if (!(h->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity ref1sidx =
sidx = h->col_parity;
// FL -> FL & differ parity
} else if (!(h->picture_structure & h->ref_list[1][0].reference) &&
!h->ref_list[1][0].mbaff) {
h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3; h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3;
} }
if (h->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred) if (h->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)
return; return;
for(list=0; list<2; list++){ for (list = 0; list < 2; list++) {
fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0); fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
if (FRAME_MBAFF(h)) if (FRAME_MBAFF(h))
for(field=0; field<2; field++) for (field = 0; field < 2; field++)
fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1); fill_colmap(h, h->map_col_to_list0_field[field], list, field,
field, 1);
} }
} }
static void await_reference_mb_row(H264Context * const h, H264Picture *ref, int mb_y) static void await_reference_mb_row(H264Context *const h, H264Picture *ref,
int mb_y)
{ {
int ref_field = ref->reference - 1; int ref_field = ref->reference - 1;
int ref_field_picture = ref->field_picture; int ref_field_picture = ref->field_picture;
int ref_height = 16*h->mb_height >> ref_field_picture; int ref_height = 16 * h->mb_height >> ref_field_picture;
if(!HAVE_THREADS || !(h->avctx->active_thread_type&FF_THREAD_FRAME)) if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_FRAME))
return; return;
//FIXME it can be safe to access mb stuff /* FIXME: It can be safe to access mb stuff
//even if pixels aren't deblocked yet * even if pixels aren't deblocked yet. */
ff_thread_await_progress(&ref->tf, ff_thread_await_progress(&ref->tf,
FFMIN(16 * mb_y >> ref_field_picture, ref_height - 1), FFMIN(16 * mb_y >> ref_field_picture,
ref_height - 1),
ref_field_picture && ref_field); ref_field_picture && ref_field);
} }
static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){ static void pred_spatial_direct_motion(H264Context *const h, int *mb_type)
{
int b8_stride = 2; int b8_stride = 2;
int b4_stride = h->b_stride; int b4_stride = h->b_stride;
int mb_xy = h->mb_xy, mb_y = h->mb_y; int mb_xy = h->mb_xy, mb_y = h->mb_y;
@ -165,7 +179,7 @@ static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
const int16_t (*l1mv0)[2], (*l1mv1)[2]; const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1; const int8_t *l1ref0, *l1ref1;
const int is_b8x8 = IS_8X8(*mb_type); const int is_b8x8 = IS_8X8(*mb_type);
unsigned int sub_mb_type= MB_TYPE_L0L1; unsigned int sub_mb_type = MB_TYPE_L0L1;
int i8, i4; int i8, i4;
int ref[2]; int ref[2];
int mv[2]; int mv[2];
@ -173,241 +187,276 @@ static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
assert(h->ref_list[1][0].reference & 3); assert(h->ref_list[1][0].reference & 3);
await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type)); await_reference_mb_row(h, &h->ref_list[1][0],
h->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)
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16 | MB_TYPE_INTRA4x4 | \
MB_TYPE_INTRA16x16 | MB_TYPE_INTRA_PCM)
/* ref = min(neighbors) */ /* ref = min(neighbors) */
for(list=0; list<2; list++){ for (list = 0; list < 2; list++) {
int left_ref = h->ref_cache[list][scan8[0] - 1]; int left_ref = h->ref_cache[list][scan8[0] - 1];
int top_ref = h->ref_cache[list][scan8[0] - 8]; int top_ref = h->ref_cache[list][scan8[0] - 8];
int refc = h->ref_cache[list][scan8[0] - 8 + 4]; int refc = h->ref_cache[list][scan8[0] - 8 + 4];
const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4]; const int16_t *C = h->mv_cache[list][scan8[0] - 8 + 4];
if(refc == PART_NOT_AVAILABLE){ if (refc == PART_NOT_AVAILABLE) {
refc = h->ref_cache[list][scan8[0] - 8 - 1]; refc = h->ref_cache[list][scan8[0] - 8 - 1];
C = h-> mv_cache[list][scan8[0] - 8 - 1]; C = h->mv_cache[list][scan8[0] - 8 - 1];
} }
ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc); ref[list] = FFMIN3((unsigned)left_ref,
if(ref[list] >= 0){ (unsigned)top_ref,
//this is just pred_motion() but with the cases removed that cannot happen for direct blocks (unsigned)refc);
const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; if (ref[list] >= 0) {
const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; /* This is just pred_motion() but with the cases removed that
* cannot happen for direct blocks. */
int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]); const int16_t *const A = h->mv_cache[list][scan8[0] - 1];
if(match_count > 1){ //most common const int16_t *const B = h->mv_cache[list][scan8[0] - 8];
mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
mid_pred(A[1], B[1], C[1]) ); int match_count = (left_ref == ref[list]) +
}else { (top_ref == ref[list]) +
assert(match_count==1); (refc == ref[list]);
if(left_ref==ref[list]){
mv[list]= AV_RN32A(A); if (match_count > 1) { // most common
}else if(top_ref==ref[list]){ mv[list] = pack16to32(mid_pred(A[0], B[0], C[0]),
mv[list]= AV_RN32A(B); mid_pred(A[1], B[1], C[1]));
}else{ } else {
mv[list]= AV_RN32A(C); assert(match_count == 1);
} if (left_ref == ref[list])
mv[list] = AV_RN32A(A);
else if (top_ref == ref[list])
mv[list] = AV_RN32A(B);
else
mv[list] = AV_RN32A(C);
} }
}else{ } else {
int mask= ~(MB_TYPE_L0 << (2*list)); int mask = ~(MB_TYPE_L0 << (2 * list));
mv[list] = 0; mv[list] = 0;
ref[list] = -1; ref[list] = -1;
if(!is_b8x8) if (!is_b8x8)
*mb_type &= mask; *mb_type &= mask;
sub_mb_type &= mask; sub_mb_type &= mask;
} }
} }
if(ref[0] < 0 && ref[1] < 0){ if (ref[0] < 0 && ref[1] < 0) {
ref[0] = ref[1] = 0; ref[0] = ref[1] = 0;
if(!is_b8x8) if (!is_b8x8)
*mb_type |= MB_TYPE_L0L1; *mb_type |= MB_TYPE_L0L1;
sub_mb_type |= MB_TYPE_L0L1; sub_mb_type |= MB_TYPE_L0L1;
} }
if(!(is_b8x8|mv[0]|mv[1])){ if (!(is_b8x8 | mv[0] | mv[1])) {
fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1); fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4); fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
*mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2; *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
MB_TYPE_16x16 | MB_TYPE_DIRECT2;
return; return;
} }
if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
mb_y = (h->mb_y&~1) + h->col_parity; mb_y = (h->mb_y & ~1) + h->col_parity;
mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride; mb_xy = h->mb_x +
((h->mb_y & ~1) + h->col_parity) * h->mb_stride;
b8_stride = 0; b8_stride = 0;
}else{ } else {
mb_y += h->col_fieldoff; mb_y += h->col_fieldoff;
mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity mb_xy += h->mb_stride * h->col_fieldoff; // non-zero for FL -> FL & differ parity
} }
goto single_col; goto single_col;
}else{ // AFL/AFR/FR/FL -> AFR/FR } else { // AFL/AFR/FR/FL -> AFR/FR
if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR if (IS_INTERLACED(*mb_type)) { // AFL /FL -> AFR/FR
mb_y = h->mb_y&~1; mb_y = h->mb_y & ~1;
mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride; mb_xy = (h->mb_y & ~1) * h->mb_stride + h->mb_x;
mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride]; mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
b8_stride = 2+4*h->mb_stride; b8_stride = 2 + 4 * h->mb_stride;
b4_stride *= 6; b4_stride *= 6;
if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) { if (IS_INTERLACED(mb_type_col[0]) !=
IS_INTERLACED(mb_type_col[1])) {
mb_type_col[0] &= ~MB_TYPE_INTERLACED; mb_type_col[0] &= ~MB_TYPE_INTERLACED;
mb_type_col[1] &= ~MB_TYPE_INTERLACED; mb_type_col[1] &= ~MB_TYPE_INTERLACED;
} }
sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) if ((mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) &&
&& (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) &&
&& !is_b8x8){ !is_b8x8) {
*mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */ *mb_type |= MB_TYPE_16x8 | MB_TYPE_DIRECT2; /* B_16x8 */
}else{ } else {
*mb_type |= MB_TYPE_8x8; *mb_type |= MB_TYPE_8x8;
} }
}else{ // AFR/FR -> AFR/FR } else { // AFR/FR -> AFR/FR
single_col: single_col:
mb_type_col[0] = mb_type_col[0] =
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){ if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {
*mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */ *mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_16x16 */
}else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){ } else if (!is_b8x8 &&
*mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16)); (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {
}else{ *mb_type |= MB_TYPE_DIRECT2 |
if(!h->sps.direct_8x8_inference_flag){ (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));
/* FIXME save sub mb types from previous frames (or derive from MVs) } else {
* so we know exactly what block size to use */ if (!h->sps.direct_8x8_inference_flag) {
sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */ /* FIXME: Save sub mb types from previous frames (or derive
* from MVs) so we know exactly what block size to use. */
sub_mb_type += (MB_TYPE_8x8 - MB_TYPE_16x16); /* B_SUB_4x4 */
} }
*mb_type |= MB_TYPE_8x8; *mb_type |= MB_TYPE_8x8;
} }
} }
} }
await_reference_mb_row(h, &h->ref_list[1][0], mb_y); await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy[mb_xy]];
l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy[mb_xy]];
l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy]; l1ref0 = &h->ref_list[1][0].ref_index[0][4 * mb_xy];
l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy]; l1ref1 = &h->ref_list[1][0].ref_index[1][4 * mb_xy];
if(!b8_stride){ if (!b8_stride) {
if(h->mb_y&1){ if (h->mb_y & 1) {
l1ref0 += 2; l1ref0 += 2;
l1ref1 += 2; l1ref1 += 2;
l1mv0 += 2*b4_stride; l1mv0 += 2 * b4_stride;
l1mv1 += 2*b4_stride; l1mv1 += 2 * b4_stride;
} }
} }
if (IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])) {
if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ int n = 0;
int n=0; for (i8 = 0; i8 < 4; i8++) {
for(i8=0; i8<4; i8++){ int x8 = i8 & 1;
int x8 = i8&1; int y8 = i8 >> 1;
int y8 = i8>>1; int xy8 = x8 + y8 * b8_stride;
int xy8 = x8+y8*b8_stride; int xy4 = x8 * 3 + y8 * b4_stride;
int xy4 = 3*x8+y8*b4_stride; int a, b;
int a,b;
if (is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) continue;
continue; h->sub_mb_type[i8] = sub_mb_type;
h->sub_mb_type[i8] = sub_mb_type;
fill_rectangle(&h->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); (uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); fill_rectangle(&h->ref_cache[1][scan8[i8 * 4]], 2, 2, 8,
if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref (uint8_t)ref[1], 1);
&& ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1) if (!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref &&
|| (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){ ((l1ref0[xy8] == 0 &&
a=b=0; FFABS(l1mv0[xy4][0]) <= 1 &&
if(ref[0] > 0) FFABS(l1mv0[xy4][1]) <= 1) ||
a= mv[0]; (l1ref0[xy8] < 0 &&
if(ref[1] > 0) l1ref1[xy8] == 0 &&
b= mv[1]; FFABS(l1mv1[xy4][0]) <= 1 &&
n++; FFABS(l1mv1[xy4][1]) <= 1))) {
}else{ a =
a= mv[0]; b = 0;
b= mv[1]; if (ref[0] > 0)
} a = mv[0];
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4); if (ref[1] > 0)
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4); b = mv[1];
} n++;
if(!is_b8x8 && !(n&3)) } else {
*mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2; a = mv[0];
}else if(IS_16X16(*mb_type)){ b = mv[1];
int a,b;
fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
&& ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
|| (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
&& h->x264_build>33U))){
a=b=0;
if(ref[0] > 0)
a= mv[0];
if(ref[1] > 0)
b= mv[1];
}else{
a= mv[0];
b= mv[1];
} }
fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4); fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, a, 4);
fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4); fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, b, 4);
}else{ }
int n=0; if (!is_b8x8 && !(n & 3))
for(i8=0; i8<4; i8++){ *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
const int x8 = i8&1; MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
const int y8 = i8>>1; MB_TYPE_16x16 | MB_TYPE_DIRECT2;
} else if (IS_16X16(*mb_type)) {
if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) int a, b;
continue;
h->sub_mb_type[i8] = sub_mb_type;
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4); fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4); fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); if (!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref &&
fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); ((l1ref0[0] == 0 &&
FFABS(l1mv0[0][0]) <= 1 &&
assert(b8_stride==2); FFABS(l1mv0[0][1]) <= 1) ||
/* col_zero_flag */ (l1ref0[0] < 0 && !l1ref1[0] &&
if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0 FFABS(l1mv1[0][0]) <= 1 &&
|| (l1ref0[i8] < 0 && l1ref1[i8] == 0 FFABS(l1mv1[0][1]) <= 1 &&
&& h->x264_build>33U))){ h->x264_build > 33U))) {
const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1; a = b = 0;
if(IS_SUB_8X8(sub_mb_type)){ if (ref[0] > 0)
const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; a = mv[0];
if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ if (ref[1] > 0)
if(ref[0] == 0) b = mv[1];
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); } else {
if(ref[1] == 0) a = mv[0];
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); b = mv[1];
n+=4; }
} fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
}else{ fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
int m=0; } else {
for(i4=0; i4<4; i4++){ int n = 0;
const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; for (i8 = 0; i8 < 4; i8++) {
if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ const int x8 = i8 & 1;
if(ref[0] == 0) const int y8 = i8 >> 1;
AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
if(ref[1] == 0) if (is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]); continue;
h->sub_mb_type[i8] = sub_mb_type;
fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, mv[0], 4);
fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, mv[1], 4);
fill_rectangle(&h->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
(uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[i8 * 4]], 2, 2, 8,
(uint8_t)ref[1], 1);
assert(b8_stride == 2);
/* col_zero_flag */
if (!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref &&
(l1ref0[i8] == 0 ||
(l1ref0[i8] < 0 &&
l1ref1[i8] == 0 &&
h->x264_build > 33U))) {
const int16_t (*l1mv)[2] = l1ref0[i8] == 0 ? l1mv0 : l1mv1;
if (IS_SUB_8X8(sub_mb_type)) {
const int16_t *mv_col = l1mv[x8 * 3 + y8 * 3 * b4_stride];
if (FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1) {
if (ref[0] == 0)
fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2,
8, 0, 4);
if (ref[1] == 0)
fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2,
8, 0, 4);
n += 4;
}
} else {
int m = 0;
for (i4 = 0; i4 < 4; i4++) {
const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +
(y8 * 2 + (i4 >> 1)) * b4_stride];
if (FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1) {
if (ref[0] == 0)
AV_ZERO32(h->mv_cache[0][scan8[i8 * 4 + i4]]);
if (ref[1] == 0)
AV_ZERO32(h->mv_cache[1][scan8[i8 * 4 + i4]]);
m++; m++;
} }
} }
if(!(m&3)) if (!(m & 3))
h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8; h->sub_mb_type[i8] += MB_TYPE_16x16 - MB_TYPE_8x8;
n+=m; n += m;
}
} }
} }
if(!is_b8x8 && !(n&15))
*mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
} }
if (!is_b8x8 && !(n & 15))
*mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
MB_TYPE_16x16 | MB_TYPE_DIRECT2;
}
} }
static void pred_temp_direct_motion(H264Context * const h, int *mb_type){ static void pred_temp_direct_motion(H264Context *const h, int *mb_type)
{
int b8_stride = 2; int b8_stride = 2;
int b4_stride = h->b_stride; int b4_stride = h->b_stride;
int mb_xy = h->mb_xy, mb_y = h->mb_y; int mb_xy = h->mb_xy, mb_y = h->mb_y;
@ -420,127 +469,142 @@ static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
assert(h->ref_list[1][0].reference & 3); assert(h->ref_list[1][0].reference & 3);
await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type)); await_reference_mb_row(h, &h->ref_list[1][0],
h->mb_y + !!IS_INTERLACED(*mb_type));
if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
mb_y = (h->mb_y&~1) + h->col_parity; mb_y = (h->mb_y & ~1) + h->col_parity;
mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride; mb_xy = h->mb_x +
((h->mb_y & ~1) + h->col_parity) * h->mb_stride;
b8_stride = 0; b8_stride = 0;
}else{ } else {
mb_y += h->col_fieldoff; mb_y += h->col_fieldoff;
mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity mb_xy += h->mb_stride * h->col_fieldoff; // non-zero for FL -> FL & differ parity
} }
goto single_col; goto single_col;
}else{ // AFL/AFR/FR/FL -> AFR/FR } else { // AFL/AFR/FR/FL -> AFR/FR
if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR if (IS_INTERLACED(*mb_type)) { // AFL /FL -> AFR/FR
mb_y = h->mb_y&~1; mb_y = h->mb_y & ~1;
mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride; mb_xy = h->mb_x + (h->mb_y & ~1) * h->mb_stride;
mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride]; mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
b8_stride = 2+4*h->mb_stride; b8_stride = 2 + 4 * h->mb_stride;
b4_stride *= 6; b4_stride *= 6;
if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) { if (IS_INTERLACED(mb_type_col[0]) !=
IS_INTERLACED(mb_type_col[1])) {
mb_type_col[0] &= ~MB_TYPE_INTERLACED; mb_type_col[0] &= ~MB_TYPE_INTERLACED;
mb_type_col[1] &= ~MB_TYPE_INTERLACED; mb_type_col[1] &= ~MB_TYPE_INTERLACED;
} }
sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) if ((mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) &&
&& (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) &&
&& !is_b8x8){ !is_b8x8) {
*mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */ *mb_type |= MB_TYPE_16x8 | MB_TYPE_L0L1 |
}else{ MB_TYPE_DIRECT2; /* B_16x8 */
*mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; } else {
*mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1;
} }
}else{ // AFR/FR -> AFR/FR } else { // AFR/FR -> AFR/FR
single_col: single_col:
mb_type_col[0] = mb_type_col[0] =
mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){ MB_TYPE_DIRECT2; /* B_SUB_8x8 */
*mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */ if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {
}else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){ *mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
*mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16)); MB_TYPE_DIRECT2; /* B_16x16 */
}else{ } else if (!is_b8x8 &&
if(!h->sps.direct_8x8_inference_flag){ (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {
/* FIXME save sub mb types from previous frames (or derive from MVs) *mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 |
* so we know exactly what block size to use */ (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));
sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */ } else {
if (!h->sps.direct_8x8_inference_flag) {
/* FIXME: save sub mb types from previous frames (or derive
* from MVs) so we know exactly what block size to use */
sub_mb_type = MB_TYPE_8x8 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
MB_TYPE_DIRECT2; /* B_SUB_4x4 */
} }
*mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; *mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1;
} }
} }
} }
await_reference_mb_row(h, &h->ref_list[1][0], mb_y); await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy[mb_xy]];
l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy[mb_xy]];
l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy]; l1ref0 = &h->ref_list[1][0].ref_index[0][4 * mb_xy];
l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy]; l1ref1 = &h->ref_list[1][0].ref_index[1][4 * mb_xy];
if(!b8_stride){ if (!b8_stride) {
if(h->mb_y&1){ if (h->mb_y & 1) {
l1ref0 += 2; l1ref0 += 2;
l1ref1 += 2; l1ref1 += 2;
l1mv0 += 2*b4_stride; l1mv0 += 2 * b4_stride;
l1mv1 += 2*b4_stride; l1mv1 += 2 * b4_stride;
} }
} }
{ {
const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]}; const int *map_col_to_list0[2] = { h->map_col_to_list0[0],
h->map_col_to_list0[1] };
const int *dist_scale_factor = h->dist_scale_factor; const int *dist_scale_factor = h->dist_scale_factor;
int ref_offset; int ref_offset;
if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) { if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {
map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y&1][0]; map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y & 1][0];
map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y&1][1]; map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y & 1][1];
dist_scale_factor =h->dist_scale_factor_field[h->mb_y&1]; dist_scale_factor = h->dist_scale_factor_field[h->mb_y & 1];
} }
ref_offset = (h->ref_list[1][0].mbaff<<4) & (mb_type_col[0]>>3); ref_offset = (h->ref_list[1][0].mbaff << 4) & (mb_type_col[0] >> 3);
if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ if (IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])) {
int y_shift = 2*!IS_INTERLACED(*mb_type); int y_shift = 2 * !IS_INTERLACED(*mb_type);
assert(h->sps.direct_8x8_inference_flag); assert(h->sps.direct_8x8_inference_flag);
for(i8=0; i8<4; i8++){ for (i8 = 0; i8 < 4; i8++) {
const int x8 = i8&1; const int x8 = i8 & 1;
const int y8 = i8>>1; const int y8 = i8 >> 1;
int ref0, scale; int ref0, scale;
const int16_t (*l1mv)[2]= l1mv0; const int16_t (*l1mv)[2] = l1mv0;
if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) if (is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
continue; continue;
h->sub_mb_type[i8] = sub_mb_type; h->sub_mb_type[i8] = sub_mb_type;
fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); fill_rectangle(&h->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 1);
if(IS_INTRA(mb_type_col[y8])){ if (IS_INTRA(mb_type_col[y8])) {
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); fill_rectangle(&h->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 1);
fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 4);
fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 4);
continue; continue;
} }
ref0 = l1ref0[x8 + y8*b8_stride]; ref0 = l1ref0[x8 + y8 * b8_stride];
if(ref0 >= 0) if (ref0 >= 0)
ref0 = map_col_to_list0[0][ref0 + ref_offset]; ref0 = map_col_to_list0[0][ref0 + ref_offset];
else{ else {
ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset]; ref0 = map_col_to_list0[1][l1ref1[x8 + y8 * b8_stride] +
l1mv= l1mv1; ref_offset];
l1mv = l1mv1;
} }
scale = dist_scale_factor[ref0]; scale = dist_scale_factor[ref0];
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); fill_rectangle(&h->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
ref0, 1);
{ {
const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride]; const int16_t *mv_col = l1mv[x8 * 3 + y8 * b4_stride];
int my_col = (mv_col[1]<<y_shift)/2; int my_col = (mv_col[1] << y_shift) / 2;
int mx = (scale * mv_col[0] + 128) >> 8; int mx = (scale * mv_col[0] + 128) >> 8;
int my = (scale * my_col + 128) >> 8; int my = (scale * my_col + 128) >> 8;
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2, 8,
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4); pack16to32(mx, my), 4);
fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2, 8,
pack16to32(mx - mv_col[0], my - my_col), 4);
} }
} }
return; return;
@ -548,13 +612,13 @@ single_col:
/* one-to-one mv scaling */ /* one-to-one mv scaling */
if(IS_16X16(*mb_type)){ if (IS_16X16(*mb_type)) {
int ref, mv0, mv1; int ref, mv0, mv1;
fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
if(IS_INTRA(mb_type_col[0])){ if (IS_INTRA(mb_type_col[0])) {
ref=mv0=mv1=0; ref = mv0 = mv1 = 0;
}else{ } else {
const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset] const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
: map_col_to_list0[1][l1ref1[0] + ref_offset]; : map_col_to_list0[1][l1ref1[0] + ref_offset];
const int scale = dist_scale_factor[ref0]; const int scale = dist_scale_factor[ref0];
@ -562,66 +626,72 @@ single_col:
int mv_l0[2]; int mv_l0[2];
mv_l0[0] = (scale * mv_col[0] + 128) >> 8; mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
mv_l0[1] = (scale * mv_col[1] + 128) >> 8; mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
ref= ref0; ref = ref0;
mv0= pack16to32(mv_l0[0],mv_l0[1]); mv0 = pack16to32(mv_l0[0], mv_l0[1]);
mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); mv1 = pack16to32(mv_l0[0] - mv_col[0], mv_l0[1] - mv_col[1]);
} }
fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1); fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4); fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4); fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
}else{ } else {
for(i8=0; i8<4; i8++){ for (i8 = 0; i8 < 4; i8++) {
const int x8 = i8&1; const int x8 = i8 & 1;
const int y8 = i8>>1; const int y8 = i8 >> 1;
int ref0, scale; int ref0, scale;
const int16_t (*l1mv)[2]= l1mv0; const int16_t (*l1mv)[2] = l1mv0;
if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) if (is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
continue; continue;
h->sub_mb_type[i8] = sub_mb_type; h->sub_mb_type[i8] = sub_mb_type;
fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); fill_rectangle(&h->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 1);
if(IS_INTRA(mb_type_col[0])){ if (IS_INTRA(mb_type_col[0])) {
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); fill_rectangle(&h->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 1);
fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 4);
fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 4);
continue; continue;
} }
assert(b8_stride == 2); assert(b8_stride == 2);
ref0 = l1ref0[i8]; ref0 = l1ref0[i8];
if(ref0 >= 0) if (ref0 >= 0)
ref0 = map_col_to_list0[0][ref0 + ref_offset]; ref0 = map_col_to_list0[0][ref0 + ref_offset];
else{ else {
ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset]; ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
l1mv= l1mv1; l1mv = l1mv1;
} }
scale = dist_scale_factor[ref0]; scale = dist_scale_factor[ref0];
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); fill_rectangle(&h->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,
if(IS_SUB_8X8(sub_mb_type)){ ref0, 1);
const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; if (IS_SUB_8X8(sub_mb_type)) {
int mx = (scale * mv_col[0] + 128) >> 8; const int16_t *mv_col = l1mv[x8 * 3 + y8 * 3 * b4_stride];
int my = (scale * mv_col[1] + 128) >> 8; int mx = (scale * mv_col[0] + 128) >> 8;
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); int my = (scale * mv_col[1] + 128) >> 8;
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4); fill_rectangle(&h->mv_cache[0][scan8[i8 * 4]], 2, 2, 8,
}else pack16to32(mx, my), 4);
for(i4=0; i4<4; i4++){ fill_rectangle(&h->mv_cache[1][scan8[i8 * 4]], 2, 2, 8,
const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; pack16to32(mx - mv_col[0], my - mv_col[1]), 4);
int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; } else {
mv_l0[0] = (scale * mv_col[0] + 128) >> 8; for (i4 = 0; i4 < 4; i4++) {
mv_l0[1] = (scale * mv_col[1] + 128) >> 8; const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +
AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]], (y8 * 2 + (i4 >> 1)) * b4_stride];
pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1])); int16_t *mv_l0 = h->mv_cache[0][scan8[i8 * 4 + i4]];
mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
AV_WN32A(h->mv_cache[1][scan8[i8 * 4 + i4]],
pack16to32(mv_l0[0] - mv_col[0],
mv_l0[1] - mv_col[1]));
}
} }
} }
} }
} }
} }
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){ void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type)
if(h->direct_spatial_mv_pred){ {
if (h->direct_spatial_mv_pred)
pred_spatial_direct_motion(h, mb_type); pred_spatial_direct_motion(h, mb_type);
}else{ else
pred_temp_direct_motion(h, mb_type); pred_temp_direct_motion(h, mb_type);
}
} }

@ -226,9 +226,9 @@ static int init_table_pools(H264Context *h)
av_buffer_allocz); av_buffer_allocz);
h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) * h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
sizeof(uint32_t), av_buffer_allocz); sizeof(uint32_t), av_buffer_allocz);
h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) * h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
sizeof(int16_t), av_buffer_allocz); sizeof(int16_t), av_buffer_allocz);
h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz); h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool || if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
!h->ref_index_pool) { !h->ref_index_pool) {
@ -1045,7 +1045,10 @@ static int init_dimensions(H264Context *h)
return AVERROR_INVALIDDATA; return AVERROR_INVALIDDATA;
av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n"); av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0; h->sps.crop_bottom =
h->sps.crop_top =
h->sps.crop_right =
h->sps.crop_left =
h->sps.crop = 0; h->sps.crop = 0;
width = h->width; width = h->width;
@ -2295,14 +2298,12 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
if (get_bits_left(&h->gb) == 0) { if (get_bits_left(&h->gb) == 0) {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, h->mb_x - 1, h->mb_y, ER_MB_END);
ER_MB_END);
return 0; return 0;
} else { } else {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, h->mb_x - 1, h->mb_y, ER_MB_END);
ER_MB_END);
return AVERROR_INVALIDDATA; return AVERROR_INVALIDDATA;
} }
@ -2315,8 +2316,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
if (get_bits_left(&h->gb) == 0) { if (get_bits_left(&h->gb) == 0) {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, h->mb_x - 1, h->mb_y, ER_MB_END);
ER_MB_END);
if (h->mb_x > lf_x_start) if (h->mb_x > lf_x_start)
loop_filter(h, lf_x_start, h->mb_x); loop_filter(h, lf_x_start, h->mb_x);

Loading…
Cancel
Save