static void dca_qmf_32_subbands(float samples_in[32][8], int sb_act,
SynthFilterContext *synth, FFTContext *imdct,
float synth_buf_ptr[512],
int *synth_buf_offset, float synth_buf2[32],
const float window[512], float *samples_out,
float raXin[32], float scale)
{
int i;
int subindex;
for (i = sb_act; i < 32; i++)
raXin[i] = 0.0;
/* Reconstructed channel sample index */
for (subindex = 0; subindex < 8; subindex++) {
/* Load in one sample from each subband and clear inactive subbands */
for (i = 0; i < sb_act; i++) {
unsigned sign = (i - 1) & 2;
uint32_t v = AV_RN32A(&samples_in[i][subindex]) ^ sign << 30;
AV_WN32A(&raXin[i], v);
}
synth->synth_filter_float(imdct, synth_buf_ptr, synth_buf_offset,
synth_buf2, window, samples_out, raXin,
scale);
samples_out += 32;
}
}
/* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
* between these functions would be nice. */
int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
HEVCNAL *nal)
{
int i, si, di;
uint8_t *dst;
if (s)
s->skipped_bytes = 0;
#define STARTCODE_TEST \
if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
if (src[i + 2] != 3) { \
/* startcode, so we must be past the end */ \
length = i; \
} \
break; \
}
#if HAVE_FAST_UNALIGNED
#define FIND_FIRST_ZERO \
if (i > 0 && !src[i]) \
i--; \
while (src[i]) \
i++
#if HAVE_FAST_64BIT
for (i = 0; i + 1 < length; i += 9) {
if (!((~AV_RN64A(src + i) &
(AV_RN64A(src + i) - 0x0100010001000101ULL)) &
0x8000800080008080ULL))
continue;
FIND_FIRST_ZERO;
STARTCODE_TEST;
i -= 7;
}
#else
for (i = 0; i + 1 < length; i += 5) {
if (!((~AV_RN32A(src + i) &
(AV_RN32A(src + i) - 0x01000101U)) &
0x80008080U))
continue;
FIND_FIRST_ZERO;
STARTCODE_TEST;
i -= 3;
}
#endif /* HAVE_FAST_64BIT */
#else
for (i = 0; i + 1 < length; i += 2) {
if (src[i])
continue;
if (i > 0 && src[i - 1] == 0)
i--;
STARTCODE_TEST;
}
#endif /* HAVE_FAST_UNALIGNED */
if (i >= length - 1) { // no escaped 0
nal->data =
nal->raw_data = src;
nal->size =
nal->raw_size = length;
return length;
}
av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
length + FF_INPUT_BUFFER_PADDING_SIZE);
if (!nal->rbsp_buffer)
return AVERROR(ENOMEM);
dst = nal->rbsp_buffer;
memcpy(dst, src, i);
si = di = i;
while (si + 2 < length) {
// remove escapes (very rare 1:2^22)
if (src[si + 2] > 3) {
dst[di++] = src[si++];
dst[di++] = src[si++];
} else if (src[si] == 0 && src[si + 1] == 0) {
if (src[si + 2] == 3) { // escape
dst[di++] = 0;
dst[di++] = 0;
si += 3;
if (s) {
s->skipped_bytes++;
if (s->skipped_bytes_pos_size < s->skipped_bytes) {
s->skipped_bytes_pos_size *= 2;
av_reallocp_array(&s->skipped_bytes_pos,
s->skipped_bytes_pos_size,
sizeof(*s->skipped_bytes_pos));
if (!s->skipped_bytes_pos)
return AVERROR(ENOMEM);
}
if (s->skipped_bytes_pos)
s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
}
continue;
} else // next start code
goto nsc;
}
dst[di++] = src[si++];
}
while (si < length)
dst[di++] = src[si++];
nsc:
memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
nal->data = dst;
nal->size = di;
nal->raw_data = src;
nal->raw_size = si;
return si;
}
static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
MpegEncContext * const s = &h->s;
int b8_stride = 2;
int b4_stride = h->b_stride;
int mb_xy = h->mb_xy, mb_y = s->mb_y;
int mb_type_col[2];
const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1;
const int is_b8x8 = IS_8X8(*mb_type);
unsigned int sub_mb_type= MB_TYPE_L0L1;
int i8, i4;
int ref[2];
int mv[2];
int list;
assert(h->ref_list[1][0].f.reference & 3);
await_reference_mb_row(h, &h->ref_list[1][0], s->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)
/* ref = min(neighbors) */
for(list=0; list<2; list++){
int left_ref = h->ref_cache[list][scan8[0] - 1];
int top_ref = h->ref_cache[list][scan8[0] - 8];
int refc = h->ref_cache[list][scan8[0] - 8 + 4];
const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
if(refc == PART_NOT_AVAILABLE){
refc = h->ref_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);
if(ref[list] >= 0){
//this is just pred_motion() but with the cases removed that cannot happen for direct blocks
const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
if(match_count > 1){ //most common
mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
mid_pred(A[1], B[1], C[1]) );
}else {
assert(match_count==1);
if(left_ref==ref[list]){
mv[list]= AV_RN32A(A);
}else if(top_ref==ref[list]){
mv[list]= AV_RN32A(B);
}else{
mv[list]= AV_RN32A(C);
}
}
}else{
int mask= ~(MB_TYPE_L0 << (2*list));
mv[list] = 0;
ref[list] = -1;
if(!is_b8x8)
*mb_type &= mask;
sub_mb_type &= mask;
}
}
if(ref[0] < 0 && ref[1] < 0){
ref[0] = ref[1] = 0;
if(!is_b8x8)
*mb_type |= MB_TYPE_L0L1;
sub_mb_type |= MB_TYPE_L0L1;
}
if(!(is_b8x8|mv[0]|mv[1])){
fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
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;
return;
}
if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
mb_y = (s->mb_y&~1) + h->col_parity;
mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
b8_stride = 0;
}else{
mb_y += h->col_fieldoff;
mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
}
goto single_col;
}else{ // AFL/AFR/FR/FL -> AFR/FR
if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
mb_y = s->mb_y&~1;
mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
b8_stride = 2+4*s->mb_stride;
b4_stride *= 6;
if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
mb_type_col[0] &= ~MB_TYPE_INTERLACED;
mb_type_col[1] &= ~MB_TYPE_INTERLACED;
}
sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
&& (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
&& !is_b8x8){
*mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
}else{
*mb_type |= MB_TYPE_8x8;
}
}else{ // AFR/FR -> AFR/FR
single_col:
mb_type_col[0] =
mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
*mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
}else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
*mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
}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_16x16); /* B_SUB_4x4 */
}
*mb_type |= MB_TYPE_8x8;
}
}
}
await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
if(!b8_stride){
if(s->mb_y&1){
l1ref0 += 2;
l1ref1 += 2;
l1mv0 += 2*b4_stride;
l1mv1 += 2*b4_stride;
}
}
if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
int n=0;
for(i8=0; i8<4; i8++){
int x8 = i8&1;
int y8 = i8>>1;
int xy8 = x8+y8*b8_stride;
int xy4 = 3*x8+y8*b4_stride;
int a,b;
if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
continue;
h->sub_mb_type[i8] = sub_mb_type;
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);
if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
&& ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
|| (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
a=b=0;
if(ref[0] > 0)
a= mv[0];
if(ref[1] > 0)
b= mv[1];
n++;
}else{
a= mv[0];
b= mv[1];
}
fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
}
if(!is_b8x8 && !(n&3))
*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;
}else if(IS_16X16(*mb_type)){