4 |
* - Decoder Module - |
* - Decoder Module - |
5 |
* |
* |
6 |
* Copyright(C) 2002 MinChen <chenm001@163.com> |
* Copyright(C) 2002 MinChen <chenm001@163.com> |
7 |
* 2002-2004 Peter Ross <pross@xvid.org> |
* 2002-2010 Peter Ross <pross@xvid.org> |
8 |
* |
* |
9 |
* This program is free software ; you can redistribute it and/or modify |
* This program is free software ; you can redistribute it and/or modify |
10 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
61 |
#include "image/postprocessing.h" |
#include "image/postprocessing.h" |
62 |
#include "utils/mem_align.h" |
#include "utils/mem_align.h" |
63 |
|
|
64 |
|
#define DIV2ROUND(n) (((n)>>1)|((n)&1)) |
65 |
|
#define DIV2(n) ((n)>>1) |
66 |
|
#define DIVUVMOV(n) (((n) >> 1) + roundtab_79[(n) & 0x3]) // |
67 |
|
|
68 |
static int |
static int |
69 |
decoder_resize(DECODER * dec) |
decoder_resize(DECODER * dec) |
70 |
{ |
{ |
172 |
dec->width = create->width; |
dec->width = create->width; |
173 |
dec->height = create->height; |
dec->height = create->height; |
174 |
|
|
175 |
|
dec->num_threads = MAX(0, create->num_threads); |
176 |
|
|
177 |
image_null(&dec->cur); |
image_null(&dec->cur); |
178 |
image_null(&dec->refn[0]); |
image_null(&dec->refn[0]); |
179 |
image_null(&dec->refn[1]); |
image_null(&dec->refn[1]); |
197 |
dec->low_delay = 0; |
dec->low_delay = 0; |
198 |
dec->packed_mode = 0; |
dec->packed_mode = 0; |
199 |
dec->time_inc_resolution = 1; /* until VOL header says otherwise */ |
dec->time_inc_resolution = 1; /* until VOL header says otherwise */ |
200 |
|
dec->ver_id = 1; |
201 |
|
|
202 |
|
if (create->fourcc == ((int)('X')|((int)('V')<<8)| |
203 |
|
((int)('I')<<16)|((int)('D')<<24))) { /* XVID */ |
204 |
|
dec->bs_version = 0; /* Initially assume oldest xvid version */ |
205 |
|
} |
206 |
|
else { |
207 |
|
dec->bs_version = 0xffff; /* Initialize to very high value -> assume bugfree stream */ |
208 |
|
} |
209 |
|
|
210 |
dec->fixed_dimensions = (dec->width > 0 && dec->height > 0); |
dec->fixed_dimensions = (dec->width > 0 && dec->height > 0); |
211 |
|
|
212 |
if (dec->fixed_dimensions) |
if (dec->fixed_dimensions) { |
213 |
return decoder_resize(dec); |
int ret = decoder_resize(dec); |
214 |
|
if (ret == XVID_ERR_MEMORY) create->handle = NULL; |
215 |
|
return ret; |
216 |
|
} |
217 |
else |
else |
218 |
return 0; |
return 0; |
219 |
} |
} |
330 |
stop_iquant_timer(); |
stop_iquant_timer(); |
331 |
|
|
332 |
start_timer(); |
start_timer(); |
333 |
idct(&data[i * 64]); |
idct((short * const)&data[i * 64]); |
334 |
stop_idct_timer(); |
stop_idct_timer(); |
335 |
|
|
336 |
} |
} |
362 |
DECLARE_ALIGNED_MATRIX(data, 1, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(data, 1, 64, int16_t, CACHE_LINE); |
363 |
|
|
364 |
int stride = dec->edged_width; |
int stride = dec->edged_width; |
|
int next_block = stride * 8; |
|
365 |
int i; |
int i; |
366 |
const uint32_t iQuant = pMB->quant; |
const uint32_t iQuant = pMB->quant; |
367 |
const int direction = dec->alternate_vertical_scan ? 2 : 0; |
const int direction = dec->alternate_vertical_scan ? 2 : 0; |
385 |
|
|
386 |
|
|
387 |
if (dec->interlacing && pMB->field_dct) { |
if (dec->interlacing && pMB->field_dct) { |
|
next_block = stride; |
|
|
stride *= 2; |
|
|
} |
|
|
|
|
388 |
dst[0] = pY_Cur; |
dst[0] = pY_Cur; |
389 |
dst[2] = pY_Cur + next_block; |
dst[1] = pY_Cur + 8; |
390 |
dst[1] = dst[0] + 8; |
dst[2] = pY_Cur + stride; |
391 |
|
dst[3] = dst[2] + 8; |
392 |
|
dst[4] = pU_Cur; |
393 |
|
dst[5] = pV_Cur; |
394 |
|
strides[0] = strides[1] = strides[2] = strides[3] = stride*2; |
395 |
|
strides[4] = stride/2; |
396 |
|
strides[5] = stride/2; |
397 |
|
} else { |
398 |
|
dst[0] = pY_Cur; |
399 |
|
dst[1] = pY_Cur + 8; |
400 |
|
dst[2] = pY_Cur + 8*stride; |
401 |
dst[3] = dst[2] + 8; |
dst[3] = dst[2] + 8; |
402 |
dst[4] = pU_Cur; |
dst[4] = pU_Cur; |
403 |
dst[5] = pV_Cur; |
dst[5] = pV_Cur; |
404 |
strides[0] = strides[1] = strides[2] = strides[3] = stride; |
strides[0] = strides[1] = strides[2] = strides[3] = stride; |
405 |
strides[4] = stride/2; |
strides[4] = stride/2; |
406 |
strides[5] = stride/2; |
strides[5] = stride/2; |
407 |
|
} |
408 |
|
|
409 |
for (i = 0; i < 6; i++) { |
for (i = 0; i < 6; i++) { |
410 |
/* Process only coded blocks */ |
/* Process only coded blocks */ |
420 |
|
|
421 |
/* iDCT */ |
/* iDCT */ |
422 |
start_timer(); |
start_timer(); |
423 |
idct(&data[0]); |
idct((short * const)&data[0]); |
424 |
stop_idct_timer(); |
stop_idct_timer(); |
425 |
|
|
426 |
/* Add this residual to the predicted block */ |
/* Add this residual to the predicted block */ |
467 |
CHECK_MV(mv[3]); |
CHECK_MV(mv[3]); |
468 |
} |
} |
469 |
|
|
470 |
|
/* Up to this version, chroma rounding was wrong with qpel. |
471 |
|
* So we try to be backward compatible to avoid artifacts */ |
472 |
|
#define BS_VERSION_BUGGY_CHROMA_ROUNDING 1 |
473 |
|
|
474 |
/* decode an inter macroblock */ |
/* decode an inter macroblock */ |
475 |
static void |
static void |
476 |
decoder_mbinter(DECODER * dec, |
decoder_mbinter(DECODER * dec, |
480 |
const uint32_t cbp, |
const uint32_t cbp, |
481 |
Bitstream * bs, |
Bitstream * bs, |
482 |
const uint32_t rounding, |
const uint32_t rounding, |
483 |
const int ref) |
const int ref, |
484 |
|
const int bvop) |
485 |
{ |
{ |
486 |
uint32_t stride = dec->edged_width; |
uint32_t stride = dec->edged_width; |
487 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
502 |
|
|
503 |
start_timer(); |
start_timer(); |
504 |
|
|
505 |
if (pMB->mode != MODE_INTER4V) { /* INTER, INTER_Q, NOT_CODED, FORWARD, BACKWARD */ |
if ((pMB->mode != MODE_INTER4V) || (bvop)) { /* INTER, INTER_Q, NOT_CODED, FORWARD, BACKWARD */ |
506 |
|
|
507 |
uv_dx = mv[0].x; |
uv_dx = mv[0].x; |
508 |
uv_dy = mv[0].y; |
uv_dy = mv[0].y; |
509 |
if (dec->quarterpel) { |
if (dec->quarterpel) { |
510 |
|
if (dec->bs_version <= BS_VERSION_BUGGY_CHROMA_ROUNDING) { |
511 |
|
uv_dx = (uv_dx>>1) | (uv_dx&1); |
512 |
|
uv_dy = (uv_dy>>1) | (uv_dy&1); |
513 |
|
} |
514 |
|
else { |
515 |
uv_dx /= 2; |
uv_dx /= 2; |
516 |
uv_dy /= 2; |
uv_dy /= 2; |
517 |
} |
} |
518 |
|
} |
519 |
uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; |
uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; |
520 |
uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; |
uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; |
521 |
|
|
530 |
} else { /* MODE_INTER4V */ |
} else { /* MODE_INTER4V */ |
531 |
|
|
532 |
if(dec->quarterpel) { |
if(dec->quarterpel) { |
533 |
|
if (dec->bs_version <= BS_VERSION_BUGGY_CHROMA_ROUNDING) { |
534 |
|
int z; |
535 |
|
uv_dx = 0; uv_dy = 0; |
536 |
|
for (z = 0; z < 4; z++) { |
537 |
|
uv_dx += ((mv[z].x>>1) | (mv[z].x&1)); |
538 |
|
uv_dy += ((mv[z].y>>1) | (mv[z].y&1)); |
539 |
|
} |
540 |
|
} |
541 |
|
else { |
542 |
uv_dx = (mv[0].x / 2) + (mv[1].x / 2) + (mv[2].x / 2) + (mv[3].x / 2); |
uv_dx = (mv[0].x / 2) + (mv[1].x / 2) + (mv[2].x / 2) + (mv[3].x / 2); |
543 |
uv_dy = (mv[0].y / 2) + (mv[1].y / 2) + (mv[2].y / 2) + (mv[3].y / 2); |
uv_dy = (mv[0].y / 2) + (mv[1].y / 2) + (mv[2].y / 2) + (mv[3].y / 2); |
544 |
|
} |
545 |
} else { |
} else { |
546 |
uv_dx = mv[0].x + mv[1].x + mv[2].x + mv[3].x; |
uv_dx = mv[0].x + mv[1].x + mv[2].x + mv[3].x; |
547 |
uv_dy = mv[0].y + mv[1].y + mv[2].y + mv[3].y; |
uv_dy = mv[0].y + mv[1].y + mv[2].y + mv[3].y; |
587 |
decoder_mb_decode(dec, cbp, bs, pY_Cur, pU_Cur, pV_Cur, pMB); |
decoder_mb_decode(dec, cbp, bs, pY_Cur, pU_Cur, pV_Cur, pMB); |
588 |
} |
} |
589 |
|
|
590 |
|
/* decode an inter macroblock in field mode */ |
591 |
|
static void |
592 |
|
decoder_mbinter_field(DECODER * dec, |
593 |
|
const MACROBLOCK * pMB, |
594 |
|
const uint32_t x_pos, |
595 |
|
const uint32_t y_pos, |
596 |
|
const uint32_t cbp, |
597 |
|
Bitstream * bs, |
598 |
|
const uint32_t rounding, |
599 |
|
const int ref, |
600 |
|
const int bvop) |
601 |
|
{ |
602 |
|
uint32_t stride = dec->edged_width; |
603 |
|
uint32_t stride2 = stride / 2; |
604 |
|
|
605 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
606 |
|
|
607 |
|
int uvtop_dx, uvtop_dy; |
608 |
|
int uvbot_dx, uvbot_dy; |
609 |
|
VECTOR mv[4]; /* local copy of mvs */ |
610 |
|
|
611 |
|
/* Get pointer to memory areas */ |
612 |
|
pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); |
613 |
|
pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3); |
614 |
|
pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); |
615 |
|
|
616 |
|
mv[0] = pMB->mvs[0]; |
617 |
|
mv[1] = pMB->mvs[1]; |
618 |
|
memset(&mv[2],0,2*sizeof(VECTOR)); |
619 |
|
|
620 |
|
validate_vector(mv, x_pos, y_pos, dec); |
621 |
|
|
622 |
|
start_timer(); |
623 |
|
|
624 |
|
if((pMB->mode!=MODE_INTER4V) || (bvop)) /* INTER, INTER_Q, NOT_CODED, FORWARD, BACKWARD */ |
625 |
|
{ |
626 |
|
/* Prepare top field vector */ |
627 |
|
uvtop_dx = DIV2ROUND(mv[0].x); |
628 |
|
uvtop_dy = DIV2ROUND(mv[0].y); |
629 |
|
|
630 |
|
/* Prepare bottom field vector */ |
631 |
|
uvbot_dx = DIV2ROUND(mv[1].x); |
632 |
|
uvbot_dy = DIV2ROUND(mv[1].y); |
633 |
|
|
634 |
|
if(dec->quarterpel) |
635 |
|
{ |
636 |
|
/* NOT supported */ |
637 |
|
} |
638 |
|
else |
639 |
|
{ |
640 |
|
/* Interpolate top field left part(we use double stride for every 2nd line) */ |
641 |
|
interpolate8x8_switch(dec->cur.y,dec->refn[ref].y+pMB->field_for_top*stride, |
642 |
|
16*x_pos,8*y_pos,mv[0].x, mv[0].y>>1,2*stride, rounding); |
643 |
|
/* top field right part */ |
644 |
|
interpolate8x8_switch(dec->cur.y,dec->refn[ref].y+pMB->field_for_top*stride, |
645 |
|
16*x_pos+8,8*y_pos,mv[0].x, mv[0].y>>1,2*stride, rounding); |
646 |
|
|
647 |
|
/* Interpolate bottom field left part(we use double stride for every 2nd line) */ |
648 |
|
interpolate8x8_switch(dec->cur.y+stride,dec->refn[ref].y+pMB->field_for_bot*stride, |
649 |
|
16*x_pos,8*y_pos,mv[1].x, mv[1].y>>1,2*stride, rounding); |
650 |
|
/* Bottom field right part */ |
651 |
|
interpolate8x8_switch(dec->cur.y+stride,dec->refn[ref].y+pMB->field_for_bot*stride, |
652 |
|
16*x_pos+8,8*y_pos,mv[1].x, mv[1].y>>1,2*stride, rounding); |
653 |
|
|
654 |
|
/* Interpolate field1 U */ |
655 |
|
interpolate8x4_switch(dec->cur.u,dec->refn[ref].u+pMB->field_for_top*stride2, |
656 |
|
8*x_pos,4*y_pos,uvtop_dx,DIV2ROUND(uvtop_dy),stride,rounding); |
657 |
|
|
658 |
|
/* Interpolate field1 V */ |
659 |
|
interpolate8x4_switch(dec->cur.v,dec->refn[ref].v+pMB->field_for_top*stride2, |
660 |
|
8*x_pos,4*y_pos,uvtop_dx,DIV2ROUND(uvtop_dy),stride,rounding); |
661 |
|
|
662 |
|
/* Interpolate field2 U */ |
663 |
|
interpolate8x4_switch(dec->cur.u+stride2,dec->refn[ref].u+pMB->field_for_bot*stride2, |
664 |
|
8*x_pos,4*y_pos,uvbot_dx,DIV2ROUND(uvbot_dy),stride,rounding); |
665 |
|
|
666 |
|
/* Interpolate field2 V */ |
667 |
|
interpolate8x4_switch(dec->cur.v+stride2,dec->refn[ref].v+pMB->field_for_bot*stride2, |
668 |
|
8*x_pos,4*y_pos,uvbot_dx,DIV2ROUND(uvbot_dy),stride,rounding); |
669 |
|
} |
670 |
|
} |
671 |
|
else |
672 |
|
{ |
673 |
|
/* We don't expect 4 motion vectors in interlaced mode */ |
674 |
|
} |
675 |
|
|
676 |
|
stop_comp_timer(); |
677 |
|
|
678 |
|
/* Must add error correction? */ |
679 |
|
if(cbp) |
680 |
|
decoder_mb_decode(dec, cbp, bs, pY_Cur, pU_Cur, pV_Cur, pMB); |
681 |
|
} |
682 |
|
|
683 |
static void |
static void |
684 |
decoder_mbgmc(DECODER * dec, |
decoder_mbgmc(DECODER * dec, |
685 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
759 |
bound = read_video_packet_header(bs, dec, 0, |
bound = read_video_packet_header(bs, dec, 0, |
760 |
&quant, NULL, NULL, &intra_dc_threshold); |
&quant, NULL, NULL, &intra_dc_threshold); |
761 |
x = bound % mb_width; |
x = bound % mb_width; |
762 |
y = bound / mb_width; |
y = MIN((bound / mb_width), (mb_height-1)); |
763 |
} |
} |
764 |
mb = &dec->mbs[y * dec->mb_width + x]; |
mb = &dec->mbs[y * dec->mb_width + x]; |
765 |
|
|
847 |
ret_mv->y = mv.y; |
ret_mv->y = mv.y; |
848 |
} |
} |
849 |
|
|
850 |
|
/* We use this when decoder runs interlaced -> different prediction */ |
851 |
|
|
852 |
|
static void get_motion_vector_interlaced(DECODER * dec, |
853 |
|
Bitstream * bs, |
854 |
|
int x, |
855 |
|
int y, |
856 |
|
int k, |
857 |
|
MACROBLOCK *pMB, |
858 |
|
int fcode, |
859 |
|
const int bound) |
860 |
|
{ |
861 |
|
const int scale_fac = 1 << (fcode - 1); |
862 |
|
const int high = (32 * scale_fac) - 1; |
863 |
|
const int low = ((-32) * scale_fac); |
864 |
|
const int range = (64 * scale_fac); |
865 |
|
|
866 |
|
/* Get interlaced prediction */ |
867 |
|
const VECTOR pmv=get_pmv2_interlaced(dec->mbs,dec->mb_width,bound,x,y,k); |
868 |
|
VECTOR mv,mvf1,mvf2; |
869 |
|
|
870 |
|
if(!pMB->field_pred) |
871 |
|
{ |
872 |
|
mv.x = get_mv(bs,fcode); |
873 |
|
mv.y = get_mv(bs,fcode); |
874 |
|
|
875 |
|
mv.x += pmv.x; |
876 |
|
mv.y += pmv.y; |
877 |
|
|
878 |
|
if(mv.x<low) { |
879 |
|
mv.x += range; |
880 |
|
} else if (mv.x>high) { |
881 |
|
mv.x-=range; |
882 |
|
} |
883 |
|
|
884 |
|
if (mv.y < low) { |
885 |
|
mv.y += range; |
886 |
|
} else if (mv.y > high) { |
887 |
|
mv.y -= range; |
888 |
|
} |
889 |
|
|
890 |
|
pMB->mvs[0]=pMB->mvs[1]=pMB->mvs[2]=pMB->mvs[3]=mv; |
891 |
|
} |
892 |
|
else |
893 |
|
{ |
894 |
|
mvf1.x = get_mv(bs, fcode); |
895 |
|
mvf1.y = get_mv(bs, fcode); |
896 |
|
|
897 |
|
mvf1.x += pmv.x; |
898 |
|
mvf1.y = 2*(mvf1.y+pmv.y/2); /* It's multiple of 2 */ |
899 |
|
|
900 |
|
if (mvf1.x < low) { |
901 |
|
mvf1.x += range; |
902 |
|
} else if (mvf1.x > high) { |
903 |
|
mvf1.x -= range; |
904 |
|
} |
905 |
|
|
906 |
|
if (mvf1.y < low) { |
907 |
|
mvf1.y += range; |
908 |
|
} else if (mvf1.y > high) { |
909 |
|
mvf1.y -= range; |
910 |
|
} |
911 |
|
|
912 |
|
mvf2.x = get_mv(bs, fcode); |
913 |
|
mvf2.y = get_mv(bs, fcode); |
914 |
|
|
915 |
|
mvf2.x += pmv.x; |
916 |
|
mvf2.y = 2*(mvf2.y+pmv.y/2); /* It's multiple of 2 */ |
917 |
|
|
918 |
|
if (mvf2.x < low) { |
919 |
|
mvf2.x += range; |
920 |
|
} else if (mvf2.x > high) { |
921 |
|
mvf2.x -= range; |
922 |
|
} |
923 |
|
|
924 |
|
if (mvf2.y < low) { |
925 |
|
mvf2.y += range; |
926 |
|
} else if (mvf2.y > high) { |
927 |
|
mvf2.y -= range; |
928 |
|
} |
929 |
|
|
930 |
|
pMB->mvs[0]=mvf1; |
931 |
|
pMB->mvs[1]=mvf2; |
932 |
|
pMB->mvs[2].x=pMB->mvs[3].x=0; |
933 |
|
pMB->mvs[2].y=pMB->mvs[3].y=0; |
934 |
|
|
935 |
|
/* Calculate average for as it is field predicted */ |
936 |
|
pMB->mvs_avg.x=DIV2ROUND(pMB->mvs[0].x+pMB->mvs[1].x); |
937 |
|
pMB->mvs_avg.y=DIV2ROUND(pMB->mvs[0].y+pMB->mvs[1].y); |
938 |
|
} |
939 |
|
} |
940 |
|
|
941 |
/* for P_VOP set gmc_warp to NULL */ |
/* for P_VOP set gmc_warp to NULL */ |
942 |
static void |
static void |
943 |
decoder_pframe(DECODER * dec, |
decoder_pframe(DECODER * dec, |
986 |
bound = read_video_packet_header(bs, dec, fcode - 1, |
bound = read_video_packet_header(bs, dec, fcode - 1, |
987 |
&quant, &fcode, NULL, &intra_dc_threshold); |
&quant, &fcode, NULL, &intra_dc_threshold); |
988 |
x = bound % mb_width; |
x = bound % mb_width; |
989 |
y = bound / mb_width; |
y = MIN((bound / mb_width), (mb_height-1)); |
990 |
} |
} |
991 |
mb = &dec->mbs[y * dec->mb_width + x]; |
mb = &dec->mbs[y * dec->mb_width + x]; |
992 |
|
|
1030 |
} |
} |
1031 |
mb->quant = quant; |
mb->quant = quant; |
1032 |
|
|
1033 |
|
mb->field_pred=0; |
1034 |
if (dec->interlacing) { |
if (dec->interlacing) { |
1035 |
if (cbp || intra) { |
if (cbp || intra) { |
1036 |
mb->field_dct = BitstreamGetBit(bs); |
mb->field_dct = BitstreamGetBit(bs); |
1056 |
|
|
1057 |
} else if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
} else if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
1058 |
|
|
1059 |
if (dec->interlacing && mb->field_pred) { |
if(dec->interlacing) { |
1060 |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); |
/* Get motion vectors interlaced, field_pred is handled there */ |
1061 |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[1], fcode, bound); |
get_motion_vector_interlaced(dec, bs, x, y, 0, mb, fcode, bound); |
1062 |
} else { |
} else { |
1063 |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); |
1064 |
mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; |
mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; |
1065 |
} |
} |
1066 |
} else if (mb->mode == MODE_INTER4V ) { |
} else if (mb->mode == MODE_INTER4V ) { |
1067 |
|
/* interlaced missing here */ |
1068 |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); |
1069 |
get_motion_vector(dec, bs, x, y, 1, &mb->mvs[1], fcode, bound); |
get_motion_vector(dec, bs, x, y, 1, &mb->mvs[1], fcode, bound); |
1070 |
get_motion_vector(dec, bs, x, y, 2, &mb->mvs[2], fcode, bound); |
get_motion_vector(dec, bs, x, y, 2, &mb->mvs[2], fcode, bound); |
1077 |
continue; |
continue; |
1078 |
} |
} |
1079 |
|
|
1080 |
decoder_mbinter(dec, mb, x, y, cbp, bs, rounding, 0); |
/* See how to decode */ |
1081 |
|
if(!mb->field_pred) |
1082 |
|
decoder_mbinter(dec, mb, x, y, cbp, bs, rounding, 0, 0); |
1083 |
|
else |
1084 |
|
decoder_mbinter_field(dec, mb, x, y, cbp, bs, rounding, 0, 0); |
1085 |
|
|
1086 |
} else if (gmc_warp) { /* a not coded S(GMC)-VOP macroblock */ |
} else if (gmc_warp) { /* a not coded S(GMC)-VOP macroblock */ |
1087 |
mb->mode = MODE_NOT_CODED_GMC; |
mb->mode = MODE_NOT_CODED_GMC; |
1099 |
|
|
1100 |
mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = 0; |
mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = 0; |
1101 |
mb->mvs[0].y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = 0; |
mb->mvs[0].y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = 0; |
1102 |
|
mb->field_pred=0; /* (!) */ |
1103 |
|
|
1104 |
decoder_mbinter(dec, mb, x, y, 0, bs, |
decoder_mbinter(dec, mb, x, y, 0, bs, |
1105 |
rounding, 0); |
rounding, 0, 0); |
1106 |
|
|
1107 |
if(dec->out_frm && cp_mb > 0) { |
if(dec->out_frm && cp_mb > 0) { |
1108 |
output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,st_mb,y,cp_mb); |
output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,st_mb,y,cp_mb); |
1184 |
b_uv_dy = pMB->b_mvs[0].y; |
b_uv_dy = pMB->b_mvs[0].y; |
1185 |
|
|
1186 |
if (dec->quarterpel) { |
if (dec->quarterpel) { |
1187 |
|
if (dec->bs_version <= BS_VERSION_BUGGY_CHROMA_ROUNDING) { |
1188 |
|
uv_dx = (uv_dx>>1) | (uv_dx&1); |
1189 |
|
uv_dy = (uv_dy>>1) | (uv_dy&1); |
1190 |
|
b_uv_dx = (b_uv_dx>>1) | (b_uv_dx&1); |
1191 |
|
b_uv_dy = (b_uv_dy>>1) | (b_uv_dy&1); |
1192 |
|
} |
1193 |
|
else { |
1194 |
uv_dx /= 2; |
uv_dx /= 2; |
1195 |
uv_dy /= 2; |
uv_dy /= 2; |
1196 |
b_uv_dx /= 2; |
b_uv_dx /= 2; |
1197 |
b_uv_dy /= 2; |
b_uv_dy /= 2; |
1198 |
} |
} |
1199 |
|
} |
1200 |
|
|
1201 |
uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; |
uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; |
1202 |
uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; |
uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; |
1204 |
b_uv_dy = (b_uv_dy >> 1) + roundtab_79[b_uv_dy & 0x3]; |
b_uv_dy = (b_uv_dy >> 1) + roundtab_79[b_uv_dy & 0x3]; |
1205 |
|
|
1206 |
} else { |
} else { |
1207 |
|
if (dec->quarterpel) { /* for qpel the /2 shall be done before summation. We've done it right in the encoder in the past. */ |
1208 |
|
/* TODO: figure out if we ever did it wrong on the encoder side. If yes, add some workaround */ |
1209 |
|
if (dec->bs_version <= BS_VERSION_BUGGY_CHROMA_ROUNDING) { |
1210 |
|
int z; |
1211 |
|
uv_dx = 0; uv_dy = 0; |
1212 |
|
b_uv_dx = 0; b_uv_dy = 0; |
1213 |
|
for (z = 0; z < 4; z++) { |
1214 |
|
uv_dx += ((pMB->mvs[z].x>>1) | (pMB->mvs[z].x&1)); |
1215 |
|
uv_dy += ((pMB->mvs[z].y>>1) | (pMB->mvs[z].y&1)); |
1216 |
|
b_uv_dx += ((pMB->b_mvs[z].x>>1) | (pMB->b_mvs[z].x&1)); |
1217 |
|
b_uv_dy += ((pMB->b_mvs[z].y>>1) | (pMB->b_mvs[z].y&1)); |
1218 |
|
} |
1219 |
|
} |
1220 |
|
else { |
1221 |
|
uv_dx = (pMB->mvs[0].x / 2) + (pMB->mvs[1].x / 2) + (pMB->mvs[2].x / 2) + (pMB->mvs[3].x / 2); |
1222 |
|
uv_dy = (pMB->mvs[0].y / 2) + (pMB->mvs[1].y / 2) + (pMB->mvs[2].y / 2) + (pMB->mvs[3].y / 2); |
1223 |
|
b_uv_dx = (pMB->b_mvs[0].x / 2) + (pMB->b_mvs[1].x / 2) + (pMB->b_mvs[2].x / 2) + (pMB->b_mvs[3].x / 2); |
1224 |
|
b_uv_dy = (pMB->b_mvs[0].y / 2) + (pMB->b_mvs[1].y / 2) + (pMB->b_mvs[2].y / 2) + (pMB->b_mvs[3].y / 2); |
1225 |
|
} |
1226 |
|
} else { |
1227 |
uv_dx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
uv_dx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1228 |
uv_dy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
uv_dy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
1229 |
b_uv_dx = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
b_uv_dx = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
1230 |
b_uv_dy = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
b_uv_dy = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
|
|
|
|
if (dec->quarterpel) { |
|
|
uv_dx /= 2; |
|
|
uv_dy /= 2; |
|
|
b_uv_dx /= 2; |
|
|
b_uv_dy /= 2; |
|
1231 |
} |
} |
1232 |
|
|
1233 |
uv_dx = (uv_dx >> 3) + roundtab_76[uv_dx & 0xf]; |
uv_dx = (uv_dx >> 3) + roundtab_76[uv_dx & 0xf]; |
1346 |
return -1; |
return -1; |
1347 |
} |
} |
1348 |
|
|
1349 |
|
static int __inline get_resync_len_b(const int fcode_backward, |
1350 |
|
const int fcode_forward) { |
1351 |
|
int resync_len = ((fcode_forward>fcode_backward) ? fcode_forward : fcode_backward) - 1; |
1352 |
|
if (resync_len < 1) resync_len = 1; |
1353 |
|
return resync_len; |
1354 |
|
} |
1355 |
|
|
1356 |
static void |
static void |
1357 |
decoder_bframe(DECODER * dec, |
decoder_bframe(DECODER * dec, |
1358 |
Bitstream * bs, |
Bitstream * bs, |
1364 |
VECTOR mv; |
VECTOR mv; |
1365 |
const VECTOR zeromv = {0,0}; |
const VECTOR zeromv = {0,0}; |
1366 |
int i; |
int i; |
1367 |
|
int resync_len; |
1368 |
|
|
1369 |
if (!dec->is_edged[0]) { |
if (!dec->is_edged[0]) { |
1370 |
start_timer(); |
start_timer(); |
1382 |
stop_edges_timer(); |
stop_edges_timer(); |
1383 |
} |
} |
1384 |
|
|
1385 |
|
resync_len = get_resync_len_b(fcode_backward, fcode_forward); |
1386 |
for (y = 0; y < dec->mb_height; y++) { |
for (y = 0; y < dec->mb_height; y++) { |
1387 |
/* Initialize Pred Motion Vector */ |
/* Initialize Pred Motion Vector */ |
1388 |
dec->p_fmv = dec->p_bmv = zeromv; |
dec->p_fmv = dec->p_bmv = zeromv; |
1389 |
for (x = 0; x < dec->mb_width; x++) { |
for (x = 0; x < dec->mb_width; x++) { |
1390 |
MACROBLOCK *mb = &dec->mbs[y * dec->mb_width + x]; |
MACROBLOCK *mb = &dec->mbs[y * dec->mb_width + x]; |
1391 |
MACROBLOCK *last_mb = &dec->last_mbs[y * dec->mb_width + x]; |
MACROBLOCK *last_mb = &dec->last_mbs[y * dec->mb_width + x]; |
1392 |
const int fcode_max = (fcode_forward>fcode_backward) ? fcode_forward : fcode_backward; |
int intra_dc_threshold; /* fake variable */ |
|
int32_t intra_dc_threshold; /* fake variable */ |
|
1393 |
|
|
1394 |
if (check_resync_marker(bs, fcode_max - 1)) { |
if (check_resync_marker(bs, resync_len)) { |
1395 |
int bound = read_video_packet_header(bs, dec, fcode_max - 1, &quant, |
int bound = read_video_packet_header(bs, dec, resync_len, &quant, |
1396 |
&fcode_forward, &fcode_backward, &intra_dc_threshold); |
&fcode_forward, &fcode_backward, &intra_dc_threshold); |
1397 |
x = bound % dec->mb_width; |
x = bound % dec->mb_width; |
1398 |
y = bound / dec->mb_width; |
y = MIN((bound / dec->mb_width), (dec->mb_height-1)); |
1399 |
/* reset predicted macroblocks */ |
/* reset predicted macroblocks */ |
1400 |
dec->p_fmv = dec->p_bmv = zeromv; |
dec->p_fmv = dec->p_bmv = zeromv; |
1401 |
|
/* update resync len with new fcodes */ |
1402 |
|
resync_len = get_resync_len_b(fcode_backward, fcode_forward); |
1403 |
} |
} |
1404 |
|
|
1405 |
mv = |
mv = |
1416 |
if (last_mb->mode == MODE_NOT_CODED) { |
if (last_mb->mode == MODE_NOT_CODED) { |
1417 |
mb->cbp = 0; |
mb->cbp = 0; |
1418 |
mb->mode = MODE_FORWARD; |
mb->mode = MODE_FORWARD; |
1419 |
decoder_mbinter(dec, mb, x, y, mb->cbp, bs, 0, 1); |
decoder_mbinter(dec, mb, x, y, mb->cbp, bs, 0, 1, 1); |
1420 |
continue; |
continue; |
1421 |
} |
} |
1422 |
|
|
1499 |
get_b_motion_vector(bs, &mb->mvs[0], fcode_backward, dec->p_bmv, dec, x, y); |
get_b_motion_vector(bs, &mb->mvs[0], fcode_backward, dec->p_bmv, dec, x, y); |
1500 |
dec->p_bmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; |
dec->p_bmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; |
1501 |
|
|
1502 |
decoder_mbinter(dec, mb, x, y, mb->cbp, bs, 0, 0); |
decoder_mbinter(dec, mb, x, y, mb->cbp, bs, 0, 0, 1); |
1503 |
break; |
break; |
1504 |
|
|
1505 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1506 |
get_b_motion_vector(bs, &mb->mvs[0], fcode_forward, dec->p_fmv, dec, x, y); |
get_b_motion_vector(bs, &mb->mvs[0], fcode_forward, dec->p_fmv, dec, x, y); |
1507 |
dec->p_fmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; |
dec->p_fmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; |
1508 |
|
|
1509 |
decoder_mbinter(dec, mb, x, y, mb->cbp, bs, 0, 1); |
decoder_mbinter(dec, mb, x, y, mb->cbp, bs, 0, 1, 1); |
1510 |
break; |
break; |
1511 |
|
|
1512 |
default: |
default: |
1517 |
} |
} |
1518 |
|
|
1519 |
/* perform post processing if necessary, and output the image */ |
/* perform post processing if necessary, and output the image */ |
1520 |
void decoder_output(DECODER * dec, IMAGE * img, MACROBLOCK * mbs, |
static void decoder_output(DECODER * dec, IMAGE * img, MACROBLOCK * mbs, |
1521 |
xvid_dec_frame_t * frame, xvid_dec_stats_t * stats, |
xvid_dec_frame_t * frame, xvid_dec_stats_t * stats, |
1522 |
int coding_type, int quant) |
int coding_type, int quant) |
1523 |
{ |
{ |
1533 |
image_copy(&dec->tmp, img, dec->edged_width, dec->height); |
image_copy(&dec->tmp, img, dec->edged_width, dec->height); |
1534 |
image_postproc(&dec->postproc, &dec->tmp, dec->edged_width, |
image_postproc(&dec->postproc, &dec->tmp, dec->edged_width, |
1535 |
mbs, dec->mb_width, dec->mb_height, dec->mb_width, |
mbs, dec->mb_width, dec->mb_height, dec->mb_width, |
1536 |
frame->general, brightness, dec->frames, (coding_type == B_VOP)); |
frame->general, brightness, dec->frames, (coding_type == B_VOP), dec->num_threads); |
1537 |
img = &dec->tmp; |
img = &dec->tmp; |
1538 |
} |
} |
1539 |
|
|
1548 |
stats->data.vop.qscale_stride = dec->mb_width; |
stats->data.vop.qscale_stride = dec->mb_width; |
1549 |
stats->data.vop.qscale = dec->qscale; |
stats->data.vop.qscale = dec->qscale; |
1550 |
if (stats->data.vop.qscale != NULL && mbs != NULL) { |
if (stats->data.vop.qscale != NULL && mbs != NULL) { |
1551 |
int i; |
unsigned int i; |
1552 |
for (i = 0; i < dec->mb_width*dec->mb_height; i++) |
for (i = 0; i < dec->mb_width*dec->mb_height; i++) |
1553 |
stats->data.vop.qscale[i] = mbs[i].quant; |
stats->data.vop.qscale[i] = mbs[i].quant; |
1554 |
} else |
} else |
1639 |
if (coding_type == -2 || coding_type == -3) { /* vol and/or resize */ |
if (coding_type == -2 || coding_type == -3) { /* vol and/or resize */ |
1640 |
|
|
1641 |
if (coding_type == -3) |
if (coding_type == -3) |
1642 |
decoder_resize(dec); |
if (decoder_resize(dec)) return XVID_ERR_MEMORY; |
1643 |
|
|
1644 |
if (stats) { |
if (stats) { |
1645 |
stats->type = XVID_TYPE_VOL; |
stats->type = XVID_TYPE_VOL; |
1662 |
goto repeat; |
goto repeat; |
1663 |
} |
} |
1664 |
|
|
1665 |
dec->p_bmv.x = dec->p_bmv.y = dec->p_fmv.y = dec->p_fmv.y = 0; /* init pred vector to 0 */ |
dec->p_bmv.x = dec->p_bmv.y = dec->p_fmv.x = dec->p_fmv.y = 0; /* init pred vector to 0 */ |
1666 |
|
|
1667 |
/* packed_mode: special-N_VOP treament */ |
/* packed_mode: special-N_VOP treament */ |
1668 |
if (dec->packed_mode && coding_type == N_VOP) { |
if (dec->packed_mode && coding_type == N_VOP) { |