#define DELAYED_PIC_REF 4
static VLC coeff_token_vlc[4];
+static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
+static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
+
static VLC chroma_dc_coeff_token_vlc;
+static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
+static const int chroma_dc_coeff_token_vlc_table_size = 256;
static VLC total_zeros_vlc[15];
+static VLC_TYPE total_zeros_vlc_tables[15][512][2];
+static const int total_zeros_vlc_tables_size = 512;
+
static VLC chroma_dc_total_zeros_vlc[3];
+static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
+static const int chroma_dc_total_zeros_vlc_tables_size = 8;
static VLC run_vlc[6];
+static VLC_TYPE run_vlc_tables[6][8][2];
+static const int run_vlc_tables_size = 8;
+
static VLC run7_vlc;
+static VLC_TYPE run7_vlc_table[96][2];
+static const int run7_vlc_table_size = 96;
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
}
static inline void direct_dist_scale_factor(H264Context * const h){
- const int poc = h->s.current_picture_ptr->poc;
+ MpegEncContext * const s = &h->s;
+ const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
const int poc1 = h->ref_list[1][0].poc;
int i;
for(i=0; i<h->ref_count[0]; i++){
int poc0 = h->ref_list[0][i].poc;
int td = av_clip(poc1 - poc0, -128, 127);
- if(td == 0 /* FIXME || pic0 is a long-term ref */){
+ if(td == 0 || h->ref_list[0][i].long_ref){
h->dist_scale_factor[i] = 256;
}else{
int tb = av_clip(poc - poc0, -128, 127);
Picture * const ref1 = &h->ref_list[1][0];
Picture * const cur = s->current_picture_ptr;
int list, i, j;
- if(cur->pict_type == FF_I_TYPE)
- cur->ref_count[0] = 0;
- if(cur->pict_type != FF_B_TYPE)
- cur->ref_count[1] = 0;
+ int sidx= s->picture_structure&1;
+ int ref1sidx= ref1->reference&1;
for(list=0; list<2; list++){
- cur->ref_count[list] = h->ref_count[list];
+ cur->ref_count[sidx][list] = h->ref_count[list];
for(j=0; j<h->ref_count[list]; j++)
- cur->ref_poc[list][j] = h->ref_list[list][j].poc;
+ cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
+ }
+ if(s->picture_structure == PICT_FRAME){
+ memcpy(cur->ref_count[0], cur->ref_count[1], sizeof(cur->ref_count[0]));
+ memcpy(cur->ref_poc [0], cur->ref_poc [1], sizeof(cur->ref_poc [0]));
}
if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
return;
for(list=0; list<2; list++){
- for(i=0; i<ref1->ref_count[list]; i++){
- const int poc = ref1->ref_poc[list][i];
+ for(i=0; i<ref1->ref_count[ref1sidx][list]; i++){
+ int poc = ref1->ref_poc[ref1sidx][list][i];
+ if(((poc&3) == 3) != (s->picture_structure == PICT_FRAME))
+ poc= (poc&~3) + s->picture_structure;
h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
for(j=0; j<h->ref_count[list]; j++)
- if(h->ref_list[list][j].poc == poc){
+ if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
h->map_col_to_list0[list][i] = j;
break;
}
}
if(FRAME_MBAFF){
for(list=0; list<2; list++){
- for(i=0; i<ref1->ref_count[list]; i++){
+ for(i=0; i<ref1->ref_count[ref1sidx][list]; i++){
j = h->map_col_to_list0[list][i];
h->map_col_to_list0_field[list][2*i] = 2*j;
h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
MpegEncContext * const s = &h->s;
- const int mb_xy = h->mb_xy;
- const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
- const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
+ const int fieldoff= (s->picture_structure & h->ref_list[1][0].reference) ? 0 : (3-2*s->picture_structure);
+ const int mb_xy = h->mb_xy + s->mb_stride*fieldoff;
+ const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride + 2*h->b8_stride*fieldoff;
+ const int b4_xy = 4*s->mb_x + 4*s->mb_y*h-> b_stride + 4*h-> b_stride*fieldoff;
const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
int refa = h->ref_cache[list][scan8[0] - 1];
int refb = h->ref_cache[list][scan8[0] - 8];
int refc = h->ref_cache[list][scan8[0] - 8 + 4];
- if(refc == -2)
+ if(refc == PART_NOT_AVAILABLE)
refc = h->ref_cache[list][scan8[0] - 8 - 1];
ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
if(ref[list] < 0)
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;
- if(FRAME_MBAFF){
- if(IS_INTERLACED(*mb_type)){
+ if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
map_col_to_list0[0] = h->map_col_to_list0_field[0];
map_col_to_list0[1] = h->map_col_to_list0_field[1];
dist_scale_factor = h->dist_scale_factor_field;
/* FIXME assumes direct_8x8_inference == 1 */
const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
int mb_types_col[2];
+ int b8_stride = h->b8_stride;
+ int b4_stride = h->b_stride;
int y_shift;
+ int ref_shift;
*mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
| (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
l1mv1 -= 4*h->b_stride;
}
y_shift = 0;
+ ref_shift= FRAME_MBAFF ? 0 : 1;
if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
&& (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
*mb_type |= MB_TYPE_16x8;
else
*mb_type |= MB_TYPE_8x8;
+ b8_stride *= 3;
+ b4_stride *= 6;
}else{
- /* field to frame scaling */
- /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
- * but in MBAFF, top and bottom POC are equal */
- int dy = (s->mb_y&1) ? 1 : 2;
+ int cur_poc = s->current_picture_ptr->poc;
+ int *col_poc = h->ref_list[1]->field_poc;
+ int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
+ int dy = 2*col_parity - (s->mb_y&1);
mb_types_col[0] =
- mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
+ mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy + col_parity*s->mb_stride];
l1ref0 += dy*h->b8_stride;
l1ref1 += dy*h->b8_stride;
l1mv0 += 2*dy*h->b_stride;
l1mv1 += 2*dy*h->b_stride;
y_shift = 2;
+ ref_shift= FRAME_MBAFF ? 2 : 1;
if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
&& !is_b8x8)
*mb_type |= MB_TYPE_16x16;
else
*mb_type |= MB_TYPE_8x8;
+ b8_stride = 0;
}
for(i8=0; i8<4; i8++){
continue;
}
- ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
+ ref0 = l1ref0[x8 + y8*b8_stride];
if(ref0 >= 0)
- ref0 = map_col_to_list0[0][ref0*2>>y_shift];
+ ref0 = map_col_to_list0[0][ref0*2>>ref_shift];
else{
- ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
+ ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]*2>>ref_shift];
l1mv= l1mv1;
}
scale = dist_scale_factor[ref0];
fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
{
- const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
+ const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
int my_col = (mv_col[1]<<y_shift)/2;
int mx = (scale * mv_col[0] + 128) >> 8;
int my = (scale * my_col + 128) >> 8;
}
return;
}
- }
/* one-to-one mv scaling */
if (!done) {
int i;
+ int offset;
done = 1;
+ chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
+ chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
&chroma_dc_coeff_token_len [0], 1, 1,
- &chroma_dc_coeff_token_bits[0], 1, 1, 1);
+ &chroma_dc_coeff_token_bits[0], 1, 1,
+ INIT_VLC_USE_NEW_STATIC);
+ offset = 0;
for(i=0; i<4; i++){
+ coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
+ coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
&coeff_token_len [i][0], 1, 1,
- &coeff_token_bits[i][0], 1, 1, 1);
+ &coeff_token_bits[i][0], 1, 1,
+ INIT_VLC_USE_NEW_STATIC);
+ offset += coeff_token_vlc_tables_size[i];
}
+ /*
+ * This is a one time safety check to make sure that
+ * the packed static coeff_token_vlc table sizes
+ * were initialized correctly.
+ */
+ assert(offset == sizeof(coeff_token_vlc_tables)/(sizeof(VLC_TYPE)*2));
for(i=0; i<3; i++){
- init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
+ chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
+ chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
+ init_vlc(&chroma_dc_total_zeros_vlc[i],
+ CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
&chroma_dc_total_zeros_len [i][0], 1, 1,
- &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
+ &chroma_dc_total_zeros_bits[i][0], 1, 1,
+ INIT_VLC_USE_NEW_STATIC);
}
for(i=0; i<15; i++){
- init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
+ total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
+ total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
+ init_vlc(&total_zeros_vlc[i],
+ TOTAL_ZEROS_VLC_BITS, 16,
&total_zeros_len [i][0], 1, 1,
- &total_zeros_bits[i][0], 1, 1, 1);
+ &total_zeros_bits[i][0], 1, 1,
+ INIT_VLC_USE_NEW_STATIC);
}
for(i=0; i<6; i++){
- init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
+ run_vlc[i].table = run_vlc_tables[i];
+ run_vlc[i].table_allocated = run_vlc_tables_size;
+ init_vlc(&run_vlc[i],
+ RUN_VLC_BITS, 7,
&run_len [i][0], 1, 1,
- &run_bits[i][0], 1, 1, 1);
+ &run_bits[i][0], 1, 1,
+ INIT_VLC_USE_NEW_STATIC);
}
+ run7_vlc.table = run7_vlc_table,
+ run7_vlc.table_allocated = run7_vlc_table_size;
init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
&run_len [6][0], 1, 1,
- &run_bits[6][0], 1, 1, 1);
+ &run_bits[6][0], 1, 1,
+ INIT_VLC_USE_NEW_STATIC);
}
}
}
if (!simple && IS_INTRA_PCM(mb_type)) {
- unsigned int x, y;
-
- // The pixels are stored in h->mb array in the same order as levels,
- // copy them in output in the correct order.
- for(i=0; i<16; i++) {
- for (y=0; y<4; y++) {
- for (x=0; x<4; x++) {
- *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
- }
- }
+ for (i=0; i<16; i++) {
+ memcpy(dest_y + i* linesize, h->mb + i*8, 16);
}
- for(i=16; i<16+4; i++) {
- for (y=0; y<4; y++) {
- for (x=0; x<4; x++) {
- *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
- }
- }
- }
- for(i=20; i<20+4; i++) {
- for (y=0; y<4; y++) {
- for (x=0; x<4; x++) {
- *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
- }
- }
+ for (i=0; i<8; i++) {
+ memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
+ memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
}
} else {
if(IS_INTRA(mb_type)){
pic->reference = parity;
pic->linesize[i] *= 2;
}
+ pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
}
static int split_field_copy(Picture *dest, Picture *src,
h->mb_aff_frame = h->sps.mb_aff;
}
}
+ h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
if(h0->current_slice == 0){
while(h->frame_num != h->prev_frame_num &&
if(FRAME_MBAFF){
if( (s->mb_y&1) == 0 )
h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);
- }else
- h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
+ }
h->prev_mb_skipped= 0;
h->slice_table[ mb_xy ]= h->slice_num;
if(IS_INTRA_PCM(mb_type)){
- unsigned int x, y;
+ unsigned int x;
// We assume these blocks are very rare so we do not optimize it.
align_get_bits(&s->gb);
// The pixels are stored in the same order as levels in h->mb array.
- for(y=0; y<16; y++){
- const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
- for(x=0; x<16; x++){
- tprintf(s->avctx, "LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
- h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
- }
- }
- if(CHROMA){
- for(y=0; y<8; y++){
- const int index= 256 + 4*(y&3) + 32*(y>>2);
- for(x=0; x<8; x++){
- tprintf(s->avctx, "CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
- h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
- }
- }
- for(y=0; y<8; y++){
- const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
- for(x=0; x<8; x++){
- tprintf(s->avctx, "CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
- h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
- }
- }
+ for(x=0; x < (CHROMA ? 384 : 256); x++){
+ ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
}
// In deblocking, the quantizer is 0
}
}
- while( coeff_count-- ) {
+ do {
uint8_t *ctx = coeff_abs_level1_ctx[node_ctx] + abs_level_m1_ctx_base;
- int j= scantable[index[coeff_count]];
+ int j= scantable[index[--coeff_count]];
if( get_cabac( CC, ctx ) == 0 ) {
node_ctx = coeff_abs_level_transition[0][node_ctx];
block[j] = (get_cabac_bypass_sign( CC, -coeff_abs ) * qmul[j] + 32) >> 6;
}
}
- }
+ } while( coeff_count );
#ifdef CABAC_ON_STACK
h->cabac.range = cc.range ;
h->cabac.low = cc.low ;
if( (s->mb_y&1) == 0 )
h->mb_mbaff =
h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
- }else
- h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
+ }
h->prev_mb_skipped = 0;
if(IS_INTRA_PCM(mb_type)) {
const uint8_t *ptr;
- unsigned int x, y;
// We assume these blocks are very rare so we do not optimize it.
// FIXME The two following lines get the bitstream position in the cabac
}
// The pixels are stored in the same order as levels in h->mb array.
- for(y=0; y<16; y++){
- const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
- for(x=0; x<16; x++){
- tprintf(s->avctx, "LUMA ICPM LEVEL (%3d)\n", *ptr);
- h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;
- }
- }
+ memcpy(h->mb, ptr, 256); ptr+=256;
if(CHROMA){
- for(y=0; y<8; y++){
- const int index= 256 + 4*(y&3) + 32*(y>>2);
- for(x=0; x<8; x++){
- tprintf(s->avctx, "CHROMA U ICPM LEVEL (%3d)\n", *ptr);
- h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
- }
- }
- for(y=0; y<8; y++){
- const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
- for(x=0; x<8; x++){
- tprintf(s->avctx, "CHROMA V ICPM LEVEL (%3d)\n", *ptr);
- h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
- }
- }
+ memcpy(h->mb+128, ptr, 128); ptr+=128;
}
ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);