3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Encoder main module - |
* - Encoder main module - |
5 |
* |
* |
6 |
* Copyright(C) 2002 Michael Militzer <isibaar@xvid.org> |
* Copyright(C) 2002-2010 Michael Militzer <isibaar@xvid.org> |
7 |
* 2002-2003 Peter Ross <pross@xvid.org> |
* 2002-2003 Peter Ross <pross@xvid.org> |
8 |
* 2002 Daniel Smith <danielsmith@astroboymail.com> |
* 2002 Daniel Smith <danielsmith@astroboymail.com> |
9 |
* |
* |
49 |
#include "quant/quant_matrix.h" |
#include "quant/quant_matrix.h" |
50 |
#include "utils/mem_align.h" |
#include "utils/mem_align.h" |
51 |
|
|
52 |
|
# include "motion/motion_smp.h" |
53 |
|
|
54 |
|
|
55 |
/***************************************************************************** |
/***************************************************************************** |
56 |
* Local function prototypes |
* Local function prototypes |
57 |
****************************************************************************/ |
****************************************************************************/ |
88 |
/* |
/* |
89 |
* Simplify the "fincr/fbase" fraction |
* Simplify the "fincr/fbase" fraction |
90 |
*/ |
*/ |
91 |
|
static int |
92 |
|
gcd(int a, int b) |
93 |
|
{ |
94 |
|
int r ; |
95 |
|
|
96 |
|
if (b > a) { |
97 |
|
r = a; |
98 |
|
a = b; |
99 |
|
b = r; |
100 |
|
} |
101 |
|
|
102 |
|
while ((r = a % b)) { |
103 |
|
a = b; |
104 |
|
b = r; |
105 |
|
} |
106 |
|
return b; |
107 |
|
} |
108 |
|
|
109 |
static void |
static void |
110 |
simplify_time(int *inc, int *base) |
simplify_time(int *inc, int *base) |
111 |
{ |
{ |
112 |
/* common factor */ |
/* common factor */ |
113 |
int i = *inc; |
const int s = gcd(*inc, *base); |
114 |
while (i > 1) { |
*inc /= s; |
115 |
if (*inc % i == 0 && *base % i == 0) { |
*base /= s; |
|
*inc /= i; |
|
|
*base /= i; |
|
|
i = *inc; |
|
|
continue; |
|
|
} |
|
|
i--; |
|
|
} |
|
116 |
|
|
117 |
if (*base > 65535 || *inc > 65535) { |
if (*base > 65535 || *inc > 65535) { |
118 |
int *biggest; |
int *biggest; |
128 |
} |
} |
129 |
|
|
130 |
div = ((float)*biggest)/((float)65535); |
div = ((float)*biggest)/((float)65535); |
131 |
*biggest = (int)(((float)*biggest)/div); |
*biggest = (unsigned int)(((float)*biggest)/div); |
132 |
*other = (int)(((float)*other)/div); |
*other = (unsigned int)(((float)*other)/div); |
133 |
} |
} |
134 |
} |
} |
135 |
|
|
160 |
|
|
161 |
/* global flags */ |
/* global flags */ |
162 |
pEnc->mbParam.global_flags = create->global; |
pEnc->mbParam.global_flags = create->global; |
163 |
|
if ((pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED)) |
164 |
|
pEnc->mbParam.global_flags |= XVID_GLOBAL_DIVX5_USERDATA; |
165 |
|
|
166 |
/* width, height */ |
/* width, height */ |
167 |
pEnc->mbParam.width = create->width; |
pEnc->mbParam.width = create->width; |
175 |
pEnc->mbParam.fincr = MAX(create->fincr, 0); |
pEnc->mbParam.fincr = MAX(create->fincr, 0); |
176 |
pEnc->mbParam.fbase = create->fincr <= 0 ? 25 : create->fbase; |
pEnc->mbParam.fbase = create->fincr <= 0 ? 25 : create->fbase; |
177 |
if (pEnc->mbParam.fincr>0) |
if (pEnc->mbParam.fincr>0) |
178 |
simplify_time(&pEnc->mbParam.fincr, &pEnc->mbParam.fbase); |
simplify_time((int*)&pEnc->mbParam.fincr, (int*)&pEnc->mbParam.fbase); |
179 |
|
|
180 |
/* zones */ |
/* zones */ |
181 |
if(create->num_zones > 0) { |
if(create->num_zones > 0) { |
206 |
|
|
207 |
memset(&pinfo, 0, sizeof(xvid_plg_info_t)); |
memset(&pinfo, 0, sizeof(xvid_plg_info_t)); |
208 |
pinfo.version = XVID_VERSION; |
pinfo.version = XVID_VERSION; |
209 |
if (create->plugins[n].func(0, XVID_PLG_INFO, &pinfo, 0) >= 0) { |
if (create->plugins[n].func(NULL, XVID_PLG_INFO, &pinfo, NULL) >= 0) { |
210 |
pEnc->mbParam.plugin_flags |= pinfo.flags; |
pEnc->mbParam.plugin_flags |= pinfo.flags; |
211 |
} |
} |
212 |
|
|
223 |
pcreate.param = create->plugins[n].param; |
pcreate.param = create->plugins[n].param; |
224 |
|
|
225 |
pEnc->plugins[n].func = NULL; /* disable plugins that fail */ |
pEnc->plugins[n].func = NULL; /* disable plugins that fail */ |
226 |
if (create->plugins[n].func(0, XVID_PLG_CREATE, &pcreate, &pEnc->plugins[n].param) >= 0) { |
if (create->plugins[n].func(NULL, XVID_PLG_CREATE, &pcreate, &pEnc->plugins[n].param) >= 0) { |
227 |
pEnc->plugins[n].func = create->plugins[n].func; |
pEnc->plugins[n].func = create->plugins[n].func; |
228 |
} |
} |
229 |
} |
} |
241 |
goto xvid_err_memory1a; |
goto xvid_err_memory1a; |
242 |
} |
} |
243 |
|
|
244 |
|
/* temp lambdas */ |
245 |
|
if (pEnc->mbParam.plugin_flags & XVID_REQLAMBDA) { |
246 |
|
pEnc->temp_lambda = (float *) xvid_malloc(pEnc->mbParam.mb_width * |
247 |
|
pEnc->mbParam.mb_height * 6 * sizeof(float), CACHE_LINE); |
248 |
|
if (pEnc->temp_lambda == NULL) |
249 |
|
goto xvid_err_memory1a; |
250 |
|
} |
251 |
|
|
252 |
/* bframes */ |
/* bframes */ |
253 |
pEnc->mbParam.max_bframes = MAX(create->max_bframes, 0); |
pEnc->mbParam.max_bframes = MAX(create->max_bframes, 0); |
254 |
pEnc->mbParam.bquant_ratio = MAX(create->bquant_ratio, 0); |
pEnc->mbParam.bquant_ratio = MAX(create->bquant_ratio, 0); |
435 |
/* timestamp stuff */ |
/* timestamp stuff */ |
436 |
|
|
437 |
pEnc->mbParam.m_stamp = 0; |
pEnc->mbParam.m_stamp = 0; |
438 |
pEnc->m_framenum = 0; |
pEnc->m_framenum = create->start_frame_num; |
439 |
pEnc->current->stamp = 0; |
pEnc->current->stamp = 0; |
440 |
pEnc->reference->stamp = 0; |
pEnc->reference->stamp = 0; |
441 |
|
|
444 |
pEnc->iFrameNum = 0; |
pEnc->iFrameNum = 0; |
445 |
pEnc->fMvPrevSigma = -1; |
pEnc->fMvPrevSigma = -1; |
446 |
|
|
447 |
|
/* slices */ |
448 |
|
pEnc->num_slices = MIN(MAX(1, create->num_slices), (int) pEnc->mbParam.mb_height); |
449 |
|
|
450 |
|
/* multithreaded stuff */ |
451 |
|
if (create->num_threads > 0) { |
452 |
|
#ifndef HAVE_PTHREAD |
453 |
|
int t = MAX(1, create->num_threads); |
454 |
|
#else |
455 |
|
int t = MIN(create->num_threads, (int) (pEnc->mbParam.mb_height>>1)); /* at least two rows per thread */ |
456 |
|
#endif |
457 |
|
int threads_per_slice = MAX(1, (t / pEnc->num_slices)); |
458 |
|
int rows_per_thread = (pEnc->mbParam.mb_height + threads_per_slice - 1) / threads_per_slice; |
459 |
|
|
460 |
|
pEnc->num_threads = t; |
461 |
|
pEnc->smpData = xvid_malloc(t*sizeof(SMPData), CACHE_LINE); |
462 |
|
if (!pEnc->smpData) |
463 |
|
goto xvid_err_nosmp; |
464 |
|
|
465 |
|
/* tmp bitstream buffer for slice coding */ |
466 |
|
pEnc->smpData[0].tmp_buffer = xvid_malloc(16*pEnc->mbParam.edged_width*pEnc->mbParam.mb_height*sizeof(uint8_t), CACHE_LINE); |
467 |
|
if (! pEnc->smpData[0].tmp_buffer) goto xvid_err_nosmp; |
468 |
|
|
469 |
|
for (n = 0; n < t; n++) { |
470 |
|
int s = MIN(pEnc->num_threads, pEnc->num_slices); |
471 |
|
|
472 |
|
pEnc->smpData[n].complete_count_self = |
473 |
|
xvid_malloc(rows_per_thread * sizeof(int), CACHE_LINE); |
474 |
|
|
475 |
|
if (!pEnc->smpData[n].complete_count_self) |
476 |
|
goto xvid_err_nosmp; |
477 |
|
|
478 |
|
if (n > 0 && n < s) { |
479 |
|
pEnc->smpData[n].bs = (Bitstream *) xvid_malloc(sizeof(Bitstream), CACHE_LINE); |
480 |
|
if (!pEnc->smpData[n].bs) |
481 |
|
goto xvid_err_nosmp; |
482 |
|
|
483 |
|
pEnc->smpData[n].sStat = (Statistics *) xvid_malloc(sizeof(Statistics), CACHE_LINE); |
484 |
|
if (!pEnc->smpData[n].sStat) |
485 |
|
goto xvid_err_nosmp; |
486 |
|
|
487 |
|
pEnc->smpData[n].tmp_buffer = pEnc->smpData[0].tmp_buffer + 16*(((n-1)*pEnc->mbParam.edged_width*pEnc->mbParam.mb_height)/s); |
488 |
|
BitstreamInit(pEnc->smpData[n].bs, pEnc->smpData[n].tmp_buffer, 0); |
489 |
|
} |
490 |
|
|
491 |
|
if (n != 0) |
492 |
|
pEnc->smpData[n].complete_count_above = |
493 |
|
pEnc->smpData[n-1].complete_count_self; |
494 |
|
} |
495 |
|
pEnc->smpData[0].complete_count_above = |
496 |
|
pEnc->smpData[t-1].complete_count_self - 1; |
497 |
|
|
498 |
|
} else { |
499 |
|
xvid_err_nosmp: |
500 |
|
/* no SMP */ |
501 |
|
if (pEnc->smpData) { |
502 |
|
if (pEnc->smpData[0].tmp_buffer) |
503 |
|
xvid_free(pEnc->smpData[0].tmp_buffer); |
504 |
|
} |
505 |
|
else { |
506 |
|
pEnc->smpData = xvid_malloc(1*sizeof(SMPData), CACHE_LINE); |
507 |
|
if (pEnc->smpData == NULL) |
508 |
|
goto xvid_err_memory5; |
509 |
|
} |
510 |
|
|
511 |
|
create->num_threads = 0; |
512 |
|
} |
513 |
|
|
514 |
create->handle = (void *) pEnc; |
create->handle = (void *) pEnc; |
515 |
|
|
516 |
init_timer(); |
init_timer(); |
597 |
xvid_free(pEnc->temp_dquants); |
xvid_free(pEnc->temp_dquants); |
598 |
} |
} |
599 |
|
|
600 |
|
if(pEnc->mbParam.plugin_flags & XVID_REQLAMBDA) { |
601 |
|
xvid_free(pEnc->temp_lambda); |
602 |
|
} |
603 |
|
|
604 |
xvid_err_memory0: |
xvid_err_memory0: |
605 |
for (n=0; n<pEnc->num_plugins;n++) { |
for (n=0; n<pEnc->num_plugins;n++) { |
606 |
if (pEnc->plugins[n].func) { |
if (pEnc->plugins[n].func) { |
607 |
pEnc->plugins[n].func(pEnc->plugins[n].param, XVID_PLG_DESTROY, 0, 0); |
pEnc->plugins[n].func(pEnc->plugins[n].param, XVID_PLG_DESTROY, NULL, NULL); |
608 |
} |
} |
609 |
} |
} |
610 |
xvid_free(pEnc->plugins); |
xvid_free(pEnc->plugins); |
699 |
xvid_free(pEnc->temp_dquants); |
xvid_free(pEnc->temp_dquants); |
700 |
} |
} |
701 |
|
|
702 |
|
if ((pEnc->mbParam.plugin_flags & XVID_REQLAMBDA)) { |
703 |
|
xvid_free(pEnc->temp_lambda); |
704 |
|
} |
705 |
|
|
706 |
if (pEnc->num_plugins>0) { |
if (pEnc->num_plugins>0) { |
707 |
xvid_plg_destroy_t pdestroy; |
xvid_plg_destroy_t pdestroy; |
712 |
|
|
713 |
for (i=0; i<pEnc->num_plugins;i++) { |
for (i=0; i<pEnc->num_plugins;i++) { |
714 |
if (pEnc->plugins[i].func) { |
if (pEnc->plugins[i].func) { |
715 |
pEnc->plugins[i].func(pEnc->plugins[i].param, XVID_PLG_DESTROY, &pdestroy, 0); |
pEnc->plugins[i].func(pEnc->plugins[i].param, XVID_PLG_DESTROY, &pdestroy, NULL); |
716 |
} |
} |
717 |
} |
} |
718 |
xvid_free(pEnc->plugins); |
xvid_free(pEnc->plugins); |
720 |
|
|
721 |
xvid_free(pEnc->mbParam.mpeg_quant_matrices); |
xvid_free(pEnc->mbParam.mpeg_quant_matrices); |
722 |
|
|
723 |
if (pEnc->num_plugins>0) |
if (pEnc->num_zones > 0) |
724 |
xvid_free(pEnc->zones); |
xvid_free(pEnc->zones); |
725 |
|
|
726 |
|
if (pEnc->num_threads > 0) { |
727 |
|
for (i = 1; i < MAX(1, MIN(pEnc->num_threads, pEnc->num_slices)); i++) { |
728 |
|
xvid_free(pEnc->smpData[i].bs); |
729 |
|
xvid_free(pEnc->smpData[i].sStat); |
730 |
|
} |
731 |
|
if (pEnc->smpData[0].tmp_buffer) xvid_free(pEnc->smpData[0].tmp_buffer); |
732 |
|
|
733 |
|
for (i = 0; i < pEnc->num_threads; i++) |
734 |
|
xvid_free(pEnc->smpData[i].complete_count_self); |
735 |
|
} |
736 |
|
xvid_free(pEnc->smpData); |
737 |
|
|
738 |
xvid_free(pEnc); |
xvid_free(pEnc); |
739 |
|
|
740 |
return 0; /* ok */ |
return 0; /* ok */ |
748 |
static void call_plugins(Encoder * pEnc, FRAMEINFO * frame, IMAGE * original, |
static void call_plugins(Encoder * pEnc, FRAMEINFO * frame, IMAGE * original, |
749 |
int opt, int * type, int * quant, xvid_enc_stats_t * stats) |
int opt, int * type, int * quant, xvid_enc_stats_t * stats) |
750 |
{ |
{ |
751 |
unsigned int i, j; |
unsigned int i, j, k; |
752 |
xvid_plg_data_t data; |
xvid_plg_data_t data; |
753 |
|
|
754 |
/* set data struct */ |
/* set data struct */ |
807 |
if ((pEnc->mbParam.plugin_flags & XVID_REQDQUANTS)) { |
if ((pEnc->mbParam.plugin_flags & XVID_REQDQUANTS)) { |
808 |
data.dquant = pEnc->temp_dquants; |
data.dquant = pEnc->temp_dquants; |
809 |
data.dquant_stride = pEnc->mbParam.mb_width; |
data.dquant_stride = pEnc->mbParam.mb_width; |
810 |
memset(data.dquant, 0, data.mb_width*data.mb_height); |
memset(data.dquant, 0, data.mb_width*data.mb_height*sizeof(int)); |
811 |
|
} |
812 |
|
|
813 |
|
if(pEnc->mbParam.plugin_flags & XVID_REQLAMBDA) { |
814 |
|
int block = 0; |
815 |
|
emms(); |
816 |
|
data.lambda = pEnc->temp_lambda; |
817 |
|
for(i = 0;i < pEnc->mbParam.mb_height; i++) |
818 |
|
for(j = 0;j < pEnc->mbParam.mb_width; j++) |
819 |
|
for (k = 0; k < 6; k++) |
820 |
|
data.lambda[block++] = 1.0f; |
821 |
} |
} |
822 |
|
|
823 |
} else { /* XVID_PLG_AFTER */ |
} else { /* XVID_PLG_AFTER */ |
894 |
for (i=0; i<(unsigned int)pEnc->num_plugins;i++) { |
for (i=0; i<(unsigned int)pEnc->num_plugins;i++) { |
895 |
emms(); |
emms(); |
896 |
if (pEnc->plugins[i].func) { |
if (pEnc->plugins[i].func) { |
897 |
if (pEnc->plugins[i].func(pEnc->plugins[i].param, opt, &data, 0) < 0) { |
if (pEnc->plugins[i].func(pEnc->plugins[i].param, opt, &data, NULL) < 0) { |
898 |
continue; |
continue; |
899 |
} |
} |
900 |
} |
} |
923 |
frame->mbs[j*pEnc->mbParam.mb_width + i].dquant = 0; |
frame->mbs[j*pEnc->mbParam.mb_width + i].dquant = 0; |
924 |
} |
} |
925 |
} |
} |
926 |
|
|
927 |
|
if (pEnc->mbParam.plugin_flags & XVID_REQLAMBDA) { |
928 |
|
for (j = 0; j < pEnc->mbParam.mb_height; j++) |
929 |
|
for (i = 0; i < pEnc->mbParam.mb_width; i++) |
930 |
|
for (k = 0; k < 6; k++) { |
931 |
|
frame->mbs[j*pEnc->mbParam.mb_width + i].lambda[k] = |
932 |
|
(int) ((float)(1<<LAMBDA_EXP) * data.lambda[6 * (j * data.mb_width + i) + k]); |
933 |
|
} |
934 |
|
} else { |
935 |
|
for (j = 0; j<pEnc->mbParam.mb_height; j++) |
936 |
|
for (i = 0; i<pEnc->mbParam.mb_width; i++) |
937 |
|
for (k = 0; k < 6; k++) { |
938 |
|
frame->mbs[j*pEnc->mbParam.mb_width + i].lambda[k] = 1<<LAMBDA_EXP; |
939 |
|
} |
940 |
|
} |
941 |
|
|
942 |
|
|
943 |
frame->mbs[0].quant = data.quant; /* FRAME will not affect the quant in stats */ |
frame->mbs[0].quant = data.quant; /* FRAME will not affect the quant in stats */ |
944 |
} |
} |
945 |
|
|
1009 |
#endif |
#endif |
1010 |
} |
} |
1011 |
|
|
|
static int |
|
|
gcd(int a, int b) |
|
|
{ |
|
|
int r ; |
|
|
|
|
|
if (b > a) { |
|
|
r = a; |
|
|
a = b; |
|
|
b = r; |
|
|
} |
|
|
|
|
|
while ((r = a % b)) { |
|
|
a = b; |
|
|
b = r; |
|
|
} |
|
|
return b; |
|
|
} |
|
|
|
|
1012 |
static void |
static void |
1013 |
simplify_par(int *par_width, int *par_height) |
simplify_par(int *par_width, int *par_height) |
1014 |
{ |
{ |
1039 |
return; |
return; |
1040 |
} |
} |
1041 |
|
|
|
|
|
1042 |
/***************************************************************************** |
/***************************************************************************** |
1043 |
* IPB frame encoder entry point |
* IPB frame encoder entry point |
1044 |
* |
* |
1133 |
} |
} |
1134 |
|
|
1135 |
FrameCodeB(pEnc, pEnc->bframes[pEnc->bframenum_head], &bs); |
FrameCodeB(pEnc, pEnc->bframes[pEnc->bframenum_head], &bs); |
1136 |
call_plugins(pEnc, pEnc->bframes[pEnc->bframenum_head], &pEnc->sOriginal2, XVID_PLG_AFTER, 0, 0, stats); |
call_plugins(pEnc, pEnc->bframes[pEnc->bframenum_head], &pEnc->sOriginal2, XVID_PLG_AFTER, NULL, NULL, stats); |
1137 |
pEnc->bframenum_head++; |
pEnc->bframenum_head++; |
1138 |
|
|
1139 |
goto done; |
goto done; |
1165 |
|
|
1166 |
/* add the not-coded length to the reference frame size */ |
/* add the not-coded length to the reference frame size */ |
1167 |
pEnc->current->length += (BitstreamPos(&bs) - bits) / 8; |
pEnc->current->length += (BitstreamPos(&bs) - bits) / 8; |
1168 |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, 0, 0, stats); |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, NULL, NULL, stats); |
1169 |
|
|
1170 |
/* flush complete: reset counters */ |
/* flush complete: reset counters */ |
1171 |
pEnc->flush_bframes = 0; |
pEnc->flush_bframes = 0; |
1193 |
pEnc->queue_head, pEnc->queue_tail, pEnc->queue_size); |
pEnc->queue_head, pEnc->queue_tail, pEnc->queue_size); |
1194 |
|
|
1195 |
if (!(pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) && pEnc->mbParam.max_bframes > 0) { |
if (!(pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) && pEnc->mbParam.max_bframes > 0) { |
1196 |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, 0, 0, stats); |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, NULL, NULL, stats); |
1197 |
} |
} |
1198 |
|
|
1199 |
/* if the very last frame is to be b-vop, we must change it to a p-vop */ |
/* if the very last frame is to be b-vop, we must change it to a p-vop */ |
1222 |
|
|
1223 |
|
|
1224 |
if ((pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) && pEnc->bframenum_tail==0) { |
if ((pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) && pEnc->bframenum_tail==0) { |
1225 |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, 0, 0, stats); |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, NULL, NULL, stats); |
1226 |
}else{ |
}else{ |
1227 |
pEnc->flush_bframes = 1; |
pEnc->flush_bframes = 1; |
1228 |
goto done; |
goto done; |
1271 |
type = frame->type; |
type = frame->type; |
1272 |
pEnc->current->quant = frame->quant; |
pEnc->current->quant = frame->quant; |
1273 |
|
|
1274 |
call_plugins(pEnc, pEnc->current, NULL, XVID_PLG_BEFORE, &type, &pEnc->current->quant, stats); |
call_plugins(pEnc, pEnc->current, NULL, XVID_PLG_BEFORE, &type, (int*)&pEnc->current->quant, stats); |
1275 |
|
|
1276 |
if (type > 0){ /* XVID_TYPE_?VOP */ |
if (type > 0){ /* XVID_TYPE_?VOP */ |
1277 |
type = type2coding(type); /* convert XVID_TYPE_?VOP to bitstream coding type */ |
type = type2coding(type); /* convert XVID_TYPE_?VOP to bitstream coding type */ |
1346 |
if (!(pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) && pEnc->mbParam.max_bframes > 0) |
if (!(pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) && pEnc->mbParam.max_bframes > 0) |
1347 |
{ |
{ |
1348 |
if (pEnc->current->stamp > 0) { |
if (pEnc->current->stamp > 0) { |
1349 |
call_plugins(pEnc, pEnc->reference, &pEnc->sOriginal, XVID_PLG_AFTER, 0, 0, stats); |
call_plugins(pEnc, pEnc->reference, &pEnc->sOriginal, XVID_PLG_AFTER, NULL, NULL, stats); |
1350 |
} |
} |
1351 |
else |
else if (stats) { |
1352 |
stats->type = XVID_TYPE_NOTHING; |
stats->type = XVID_TYPE_NOTHING; |
1353 |
} |
} |
1354 |
|
} |
1355 |
|
|
1356 |
/* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
/* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
1357 |
* closed-gop |
* closed-gop |
1375 |
SWAP(FRAMEINFO*, pEnc->current, pEnc->bframes[pEnc->bframenum_tail]); |
SWAP(FRAMEINFO*, pEnc->current, pEnc->bframes[pEnc->bframenum_tail]); |
1376 |
|
|
1377 |
if ((pEnc->current->vop_flags & XVID_VOP_DEBUG)) { |
if ((pEnc->current->vop_flags & XVID_VOP_DEBUG)) { |
1378 |
image_printf(&pEnc->current->image, pEnc->mbParam.edged_width, pEnc->mbParam.height, 5, 100, "DX50 BVOP->PVOP"); |
image_printf(&pEnc->current->image, pEnc->mbParam.edged_width, pEnc->mbParam.height, 5, 100, "CLOSED GOP BVOP->PVOP"); |
1379 |
} |
} |
1380 |
|
|
1381 |
/* convert B-VOP quant to P-VOP */ |
/* convert B-VOP quant to P-VOP */ |
1436 |
|
|
1437 |
/* prevent vol/vop misuse */ |
/* prevent vol/vop misuse */ |
1438 |
|
|
|
if (!(pEnc->current->vol_flags & XVID_VOL_REDUCED_ENABLE)) |
|
|
pEnc->current->vop_flags &= ~XVID_VOP_REDUCED; |
|
|
|
|
1439 |
if (!(pEnc->current->vol_flags & XVID_VOL_INTERLACING)) |
if (!(pEnc->current->vol_flags & XVID_VOL_INTERLACING)) |
1440 |
pEnc->current->vop_flags &= ~(XVID_VOP_TOPFIELDFIRST|XVID_VOP_ALTERNATESCAN); |
pEnc->current->vop_flags &= ~(XVID_VOP_TOPFIELDFIRST|XVID_VOP_ALTERNATESCAN); |
1441 |
|
|
1470 |
|
|
1471 |
if ( FrameCodeP(pEnc, &bs) == 0 ) { |
if ( FrameCodeP(pEnc, &bs) == 0 ) { |
1472 |
/* N-VOP, we mustn't code b-frames yet */ |
/* N-VOP, we mustn't code b-frames yet */ |
1473 |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, 0, 0, stats); |
if ((pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) || |
1474 |
|
pEnc->mbParam.max_bframes == 0) |
1475 |
|
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, NULL, NULL, stats); |
1476 |
goto done; |
goto done; |
1477 |
} |
} |
1478 |
} |
} |
1493 |
|
|
1494 |
/* packed or no-bframes or no-bframes-queued: output stats */ |
/* packed or no-bframes or no-bframes-queued: output stats */ |
1495 |
if ((pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) || pEnc->mbParam.max_bframes == 0 ) { |
if ((pEnc->mbParam.global_flags & XVID_GLOBAL_PACKED) || pEnc->mbParam.max_bframes == 0 ) { |
1496 |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, 0, 0, stats); |
call_plugins(pEnc, pEnc->current, &pEnc->sOriginal, XVID_PLG_AFTER, NULL, NULL, stats); |
1497 |
} |
} |
1498 |
|
|
1499 |
/* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
/* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
1533 |
|
|
1534 |
|
|
1535 |
static __inline void |
static __inline void |
1536 |
CodeIntraMB(Encoder * pEnc, |
CodeIntraMB(MACROBLOCK * pMB) |
|
MACROBLOCK * pMB) |
|
1537 |
{ |
{ |
|
|
|
1538 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
1539 |
|
|
1540 |
/* zero mv statistics */ |
/* zero mv statistics */ |
1548 |
} |
} |
1549 |
} |
} |
1550 |
|
|
1551 |
|
static void |
1552 |
|
SliceCodeI(SMPData *data) |
|
static int |
|
|
FrameCodeI(Encoder * pEnc, |
|
|
Bitstream * bs) |
|
1553 |
{ |
{ |
1554 |
int bits = BitstreamPos(bs); |
Encoder *pEnc = (Encoder *) data->pEnc; |
1555 |
|
Bitstream *bs = (Bitstream *) data->bs; |
1556 |
|
|
1557 |
|
uint16_t x, y; |
1558 |
int mb_width = pEnc->mbParam.mb_width; |
int mb_width = pEnc->mbParam.mb_width; |
1559 |
int mb_height = pEnc->mbParam.mb_height; |
int mb_height = pEnc->mbParam.mb_height; |
1560 |
|
|
1561 |
|
int bound = 0, num_slices = pEnc->num_slices; |
1562 |
|
FRAMEINFO *const current = pEnc->current; |
1563 |
|
|
1564 |
DECLARE_ALIGNED_MATRIX(dct_codes, 6, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(dct_codes, 6, 64, int16_t, CACHE_LINE); |
1565 |
DECLARE_ALIGNED_MATRIX(qcoeff, 6, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(qcoeff, 6, 64, int16_t, CACHE_LINE); |
1566 |
|
|
1567 |
uint16_t x, y; |
if (data->start_y > 0) { /* write resync marker */ |
1568 |
|
bound = data->start_y*mb_width; |
1569 |
|
write_video_packet_header(bs, &pEnc->mbParam, current, bound); |
1570 |
|
} |
1571 |
|
|
1572 |
if ((pEnc->current->vol_flags & XVID_VOL_REDUCED_ENABLE)) |
for (y = data->start_y; y < data->stop_y; y++) { |
1573 |
{ |
int new_bound = mb_width * ((((y*num_slices) / mb_height) * mb_height + (num_slices-1)) / num_slices); |
1574 |
mb_width = (pEnc->mbParam.width + 31) / 32; |
|
1575 |
mb_height = (pEnc->mbParam.height + 31) / 32; |
if (new_bound > bound) { |
1576 |
|
bound = new_bound; |
1577 |
|
BitstreamPadAlways(bs); |
1578 |
|
write_video_packet_header(bs, &pEnc->mbParam, current, bound); |
1579 |
|
} |
1580 |
|
|
1581 |
|
for (x = 0; x < mb_width; x++) { |
1582 |
|
MACROBLOCK *pMB = ¤t->mbs[x + y * mb_width]; |
1583 |
|
|
1584 |
|
CodeIntraMB(pMB); |
1585 |
|
|
1586 |
|
MBTransQuantIntra(&pEnc->mbParam, current, pMB, x, y, |
1587 |
|
dct_codes, qcoeff); |
1588 |
|
|
|
/* 16x16->8x8 downsample requires 1 additional edge pixel*/ |
|
|
/* XXX: setedges is overkill */ |
|
1589 |
start_timer(); |
start_timer(); |
1590 |
image_setedges(&pEnc->current->image, |
MBPrediction(current, x, y, mb_width, qcoeff, bound); |
1591 |
pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, |
stop_prediction_timer(); |
1592 |
pEnc->mbParam.width, pEnc->mbParam.height, 0); |
|
1593 |
stop_edges_timer(); |
start_timer(); |
1594 |
|
MBCoding(current, pMB, qcoeff, bs, data->sStat); |
1595 |
|
stop_coding_timer(); |
1596 |
|
|
1597 |
|
} |
1598 |
|
} |
1599 |
|
|
1600 |
|
emms(); |
1601 |
|
BitstreamPadAlways(bs); |
1602 |
|
} |
1603 |
|
|
1604 |
|
static __inline void |
1605 |
|
SerializeBitstreams(Encoder *pEnc, FRAMEINFO *current, Bitstream *bs, int num_threads) |
1606 |
|
{ |
1607 |
|
int k; |
1608 |
|
uint32_t pos = BitstreamLength(bs); |
1609 |
|
|
1610 |
|
for (k = 1; k < num_threads; k++) { |
1611 |
|
uint32_t len = BitstreamLength(pEnc->smpData[k].bs); |
1612 |
|
|
1613 |
|
memcpy((void *)((ptr_t)bs->start + pos), |
1614 |
|
(void *)((ptr_t)pEnc->smpData[k].bs->start), len); |
1615 |
|
|
1616 |
|
current->length += len; |
1617 |
|
pos += len; |
1618 |
|
|
1619 |
|
/* collect stats */ |
1620 |
|
current->sStat.iTextBits += pEnc->smpData[k].sStat->iTextBits; |
1621 |
|
current->sStat.kblks += pEnc->smpData[k].sStat->kblks; |
1622 |
|
current->sStat.mblks += pEnc->smpData[k].sStat->mblks; |
1623 |
|
current->sStat.ublks += pEnc->smpData[k].sStat->ublks; |
1624 |
|
current->sStat.iMVBits += pEnc->smpData[k].sStat->iMVBits; |
1625 |
|
} |
1626 |
|
|
1627 |
|
if (num_threads > 1) { |
1628 |
|
uint32_t pos32 = pos>>2; |
1629 |
|
bs->tail = bs->start + pos32; |
1630 |
|
bs->pos = 8*(pos - (pos32<<2)); |
1631 |
|
bs->buf = 0; |
1632 |
|
|
1633 |
|
if (bs->pos > 0) { |
1634 |
|
uint32_t pos8 = bs->pos/8; |
1635 |
|
memset((void *)((ptr_t)bs->tail+pos8), 0, (4-pos8)); |
1636 |
|
pos = *bs->tail; |
1637 |
|
#ifndef ARCH_IS_BIG_ENDIAN |
1638 |
|
BSWAP(pos); |
1639 |
|
#endif |
1640 |
|
bs->buf = pos; |
1641 |
|
} |
1642 |
|
} |
1643 |
} |
} |
1644 |
|
|
1645 |
|
static int |
1646 |
|
FrameCodeI(Encoder * pEnc, |
1647 |
|
Bitstream * bs) |
1648 |
|
{ |
1649 |
|
int bits = BitstreamPos(bs); |
1650 |
|
int bound = 0, num_slices = pEnc->num_slices; |
1651 |
|
int num_threads = MAX(1, MIN(pEnc->num_threads, num_slices)); |
1652 |
|
int slices_per_thread = (num_slices*1024 / num_threads); |
1653 |
|
int mb_height = pEnc->mbParam.mb_height; |
1654 |
|
#ifdef HAVE_PTHREAD |
1655 |
|
void * status = NULL; |
1656 |
|
#endif |
1657 |
|
uint16_t k; |
1658 |
|
|
1659 |
pEnc->mbParam.m_rounding_type = 1; |
pEnc->mbParam.m_rounding_type = 1; |
1660 |
pEnc->current->rounding_type = pEnc->mbParam.m_rounding_type; |
pEnc->current->rounding_type = pEnc->mbParam.m_rounding_type; |
1661 |
pEnc->current->coding_type = I_VOP; |
pEnc->current->coding_type = I_VOP; |
1664 |
|
|
1665 |
SetMacroblockQuants(&pEnc->mbParam, pEnc->current); |
SetMacroblockQuants(&pEnc->mbParam, pEnc->current); |
1666 |
|
|
1667 |
BitstreamWriteVolHeader(bs, &pEnc->mbParam, pEnc->current); |
BitstreamWriteVolHeader(bs, &pEnc->mbParam, pEnc->current, num_slices); |
1668 |
|
|
1669 |
set_timecodes(pEnc->current,pEnc->reference,pEnc->mbParam.fbase); |
set_timecodes(pEnc->current,pEnc->reference,pEnc->mbParam.fbase); |
1670 |
|
|
1673 |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, pEnc->current, 1, pEnc->current->mbs[0].quant); |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, pEnc->current, 1, pEnc->current->mbs[0].quant); |
1674 |
|
|
1675 |
pEnc->current->sStat.iTextBits = 0; |
pEnc->current->sStat.iTextBits = 0; |
|
pEnc->current->sStat.kblks = mb_width * mb_height; |
|
|
pEnc->current->sStat.mblks = pEnc->current->sStat.ublks = 0; |
|
1676 |
|
|
1677 |
for (y = 0; y < mb_height; y++) |
/* multithreaded intra coding - dispatch threads */ |
1678 |
for (x = 0; x < mb_width; x++) { |
for (k = 0; k < num_threads; k++) { |
1679 |
MACROBLOCK *pMB = |
int add = ((slices_per_thread + 512) >> 10); |
|
&pEnc->current->mbs[x + y * pEnc->mbParam.mb_width]; |
|
1680 |
|
|
1681 |
CodeIntraMB(pEnc, pMB); |
slices_per_thread += ((num_slices*1024 / num_threads) - add*1024); |
1682 |
|
|
1683 |
MBTransQuantIntra(&pEnc->mbParam, pEnc->current, pMB, x, y, |
pEnc->smpData[k].pEnc = (void *) pEnc; |
1684 |
dct_codes, qcoeff); |
pEnc->smpData[k].stop_y = (((bound+add) * mb_height + (num_slices-1)) / num_slices); |
1685 |
|
pEnc->smpData[k].start_y = ((bound * mb_height + (num_slices-1)) / num_slices); |
1686 |
|
|
1687 |
start_timer(); |
bound += add; |
|
MBPrediction(pEnc->current, x, y, pEnc->mbParam.mb_width, qcoeff); |
|
|
stop_prediction_timer(); |
|
1688 |
|
|
1689 |
start_timer(); |
if (k > 0) { |
1690 |
if (pEnc->current->vop_flags & XVID_VOP_GREYSCALE) |
BitstreamReset(pEnc->smpData[k].bs); |
1691 |
{ pMB->cbp &= 0x3C; /* keep only bits 5-2 */ |
pEnc->smpData[k].sStat->iTextBits = 0; |
|
qcoeff[4*64+0]=0; /* zero, because for INTRA MBs DC value is saved */ |
|
|
qcoeff[5*64+0]=0; |
|
1692 |
} |
} |
|
MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->current->sStat); |
|
|
stop_coding_timer(); |
|
1693 |
} |
} |
1694 |
|
pEnc->smpData[0].bs = bs; |
1695 |
|
pEnc->smpData[0].sStat = &pEnc->current->sStat; |
1696 |
|
|
1697 |
if ((pEnc->current->vop_flags & XVID_VOP_REDUCED)) |
#ifdef HAVE_PTHREAD |
1698 |
{ |
/* create threads */ |
1699 |
image_deblock_rrv(&pEnc->current->image, pEnc->mbParam.edged_width, |
for (k = 1; k < num_threads; k++) { |
1700 |
pEnc->current->mbs, mb_width, mb_height, pEnc->mbParam.mb_width, |
pthread_create(&pEnc->smpData[k].handle, NULL, |
1701 |
16, 0); |
(void*)SliceCodeI, (void*)&pEnc->smpData[k]); |
1702 |
} |
} |
1703 |
emms(); |
#endif |
1704 |
|
|
1705 |
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
SliceCodeI(&pEnc->smpData[0]); |
1706 |
|
|
1707 |
|
#ifdef HAVE_PTHREAD |
1708 |
|
/* wait until all threads are finished */ |
1709 |
|
for (k = 1; k < num_threads; k++) { |
1710 |
|
pthread_join(pEnc->smpData[k].handle, &status); |
1711 |
|
} |
1712 |
|
#endif |
1713 |
|
|
1714 |
|
pEnc->current->length = BitstreamLength(bs) - (bits/8); |
1715 |
|
|
1716 |
|
/* reassemble the pieces together */ |
1717 |
|
SerializeBitstreams(pEnc, pEnc->current, bs, num_threads); |
1718 |
|
|
1719 |
pEnc->current->length = (BitstreamPos(bs) - bits) / 8; |
pEnc->current->sStat.iMVBits = 0; |
1720 |
|
pEnc->current->sStat.mblks = pEnc->current->sStat.ublks = 0; |
1721 |
|
pEnc->current->sStat.kblks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; |
1722 |
|
|
1723 |
pEnc->fMvPrevSigma = -1; |
pEnc->fMvPrevSigma = -1; |
1724 |
pEnc->mbParam.m_fcode = 2; |
pEnc->mbParam.m_fcode = 2; |
1729 |
return 1; /* intra */ |
return 1; /* intra */ |
1730 |
} |
} |
1731 |
|
|
1732 |
|
static __inline void |
1733 |
|
updateFcode(Statistics * sStat, Encoder * pEnc) |
1734 |
|
{ |
1735 |
|
float fSigma; |
1736 |
|
int iSearchRange; |
1737 |
|
|
1738 |
|
if (sStat->iMvCount == 0) |
1739 |
|
sStat->iMvCount = 1; |
1740 |
|
|
1741 |
#define INTRA_THRESHOLD 0.5 |
fSigma = (float) sqrt((float) sStat->iMvSum / sStat->iMvCount); |
|
#define BFRAME_SKIP_THRESHHOLD 30 |
|
1742 |
|
|
1743 |
|
iSearchRange = 16 << pEnc->mbParam.m_fcode; |
1744 |
|
|
1745 |
/* FrameCodeP also handles S(GMC)-VOPs */ |
if ((3.0 * fSigma > iSearchRange) && (pEnc->mbParam.m_fcode <= 5) ) |
1746 |
static int |
pEnc->mbParam.m_fcode++; |
1747 |
FrameCodeP(Encoder * pEnc, |
|
1748 |
Bitstream * bs) |
else if ((5.0 * fSigma < iSearchRange) |
1749 |
|
&& (4.0 * pEnc->fMvPrevSigma < iSearchRange) |
1750 |
|
&& (pEnc->mbParam.m_fcode >= 2) ) |
1751 |
|
pEnc->mbParam.m_fcode--; |
1752 |
|
|
1753 |
|
pEnc->fMvPrevSigma = fSigma; |
1754 |
|
} |
1755 |
|
|
1756 |
|
#define BFRAME_SKIP_THRESHHOLD 30 |
1757 |
|
|
1758 |
|
static void |
1759 |
|
SliceCodeP(SMPData *data) |
1760 |
{ |
{ |
1761 |
float fSigma; |
Encoder *pEnc = (Encoder *) data->pEnc; |
1762 |
int bits = BitstreamPos(bs); |
Bitstream *bs = (Bitstream *) data->bs; |
1763 |
|
|
1764 |
|
int x, y, k; |
1765 |
|
FRAMEINFO *const current = pEnc->current; |
1766 |
|
FRAMEINFO *const reference = pEnc->reference; |
1767 |
|
MBParam * const pParam = &pEnc->mbParam; |
1768 |
|
int mb_width = pParam->mb_width; |
1769 |
|
int mb_height = pParam->mb_height; |
1770 |
|
|
1771 |
DECLARE_ALIGNED_MATRIX(dct_codes, 6, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(dct_codes, 6, 64, int16_t, CACHE_LINE); |
1772 |
DECLARE_ALIGNED_MATRIX(qcoeff, 6, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(qcoeff, 6, 64, int16_t, CACHE_LINE); |
1773 |
|
|
1774 |
int x, y, k; |
int bound = 0, num_slices = pEnc->num_slices; |
1775 |
int iSearchRange; |
|
1776 |
|
if (data->start_y > 0) { /* write resync marker */ |
1777 |
|
bound = data->start_y*mb_width; |
1778 |
|
write_video_packet_header(bs, pParam, current, bound); |
1779 |
|
} |
1780 |
|
|
1781 |
|
for (y = data->start_y; y < data->stop_y; y++) { |
1782 |
|
int new_bound = mb_width * ((((y*num_slices) / mb_height) * mb_height + (num_slices-1)) / num_slices); |
1783 |
|
|
1784 |
|
if (new_bound > bound) { |
1785 |
|
bound = new_bound; |
1786 |
|
BitstreamPadAlways(bs); |
1787 |
|
write_video_packet_header(bs, pParam, current, bound); |
1788 |
|
} |
1789 |
|
|
1790 |
|
for (x = 0; x < mb_width; x++) { |
1791 |
|
MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; |
1792 |
int skip_possible; |
int skip_possible; |
1793 |
|
|
1794 |
|
if (pMB->mode == MODE_INTRA || pMB->mode == MODE_INTRA_Q) { |
1795 |
|
CodeIntraMB(pMB); |
1796 |
|
MBTransQuantIntra(pParam, current, pMB, x, y, |
1797 |
|
dct_codes, qcoeff); |
1798 |
|
|
1799 |
|
start_timer(); |
1800 |
|
MBPrediction(current, x, y, pParam->mb_width, qcoeff, bound); |
1801 |
|
stop_prediction_timer(); |
1802 |
|
|
1803 |
|
data->sStat->kblks++; |
1804 |
|
|
1805 |
|
MBCoding(current, pMB, qcoeff, bs, data->sStat); |
1806 |
|
stop_coding_timer(); |
1807 |
|
continue; |
1808 |
|
} |
1809 |
|
|
1810 |
|
start_timer(); |
1811 |
|
MBMotionCompensation(pMB, x, y, &reference->image, |
1812 |
|
&pEnc->vInterH, &pEnc->vInterV, |
1813 |
|
&pEnc->vInterHV, &pEnc->vGMC, |
1814 |
|
¤t->image, |
1815 |
|
dct_codes, pParam->width, |
1816 |
|
pParam->height, |
1817 |
|
pParam->edged_width, |
1818 |
|
(current->vol_flags & XVID_VOL_QUARTERPEL), |
1819 |
|
current->rounding_type, |
1820 |
|
data->RefQ); |
1821 |
|
|
1822 |
|
stop_comp_timer(); |
1823 |
|
|
1824 |
|
pMB->field_pred = 0; |
1825 |
|
|
1826 |
|
if (pMB->cbp != 0) { |
1827 |
|
pMB->cbp = MBTransQuantInter(pParam, current, pMB, x, y, |
1828 |
|
dct_codes, qcoeff); |
1829 |
|
} |
1830 |
|
|
1831 |
|
if (pMB->dquant != 0) |
1832 |
|
MBSetDquant(pMB, x, y, pParam); |
1833 |
|
|
1834 |
|
|
1835 |
|
if (pMB->cbp || pMB->mvs[0].x || pMB->mvs[0].y || |
1836 |
|
pMB->mvs[1].x || pMB->mvs[1].y || pMB->mvs[2].x || |
1837 |
|
pMB->mvs[2].y || pMB->mvs[3].x || pMB->mvs[3].y) { |
1838 |
|
data->sStat->mblks++; |
1839 |
|
} else { |
1840 |
|
data->sStat->ublks++; |
1841 |
|
} |
1842 |
|
|
1843 |
|
start_timer(); |
1844 |
|
|
1845 |
|
/* Finished processing the MB, now check if to CODE or SKIP */ |
1846 |
|
|
1847 |
|
skip_possible = (pMB->cbp == 0) && (pMB->mode == MODE_INTER); |
1848 |
|
|
1849 |
|
if (current->coding_type == S_VOP) |
1850 |
|
skip_possible &= (pMB->mcsel == 1); |
1851 |
|
else { /* PVOP */ |
1852 |
|
const VECTOR * const mv = (pParam->vol_flags & XVID_VOL_QUARTERPEL) ? |
1853 |
|
pMB->qmvs : pMB->mvs; |
1854 |
|
skip_possible &= ((mv->x|mv->y) == 0); |
1855 |
|
} |
1856 |
|
|
1857 |
|
if ((pMB->mode == MODE_NOT_CODED) || (skip_possible)) { |
1858 |
|
/* This is a candidate for SKIPping, but for P-VOPs check intermediate B-frames first */ |
1859 |
|
int bSkip = 1; |
1860 |
|
|
1861 |
|
if (current->coding_type == P_VOP) { /* special rule for P-VOP's SKIP */ |
1862 |
|
for (k = pEnc->bframenum_head; k < pEnc->bframenum_tail; k++) { |
1863 |
|
int iSAD; |
1864 |
|
iSAD = sad16(reference->image.y + 16*y*pParam->edged_width + 16*x, |
1865 |
|
pEnc->bframes[k]->image.y + 16*y*pParam->edged_width + 16*x, |
1866 |
|
pParam->edged_width, BFRAME_SKIP_THRESHHOLD * pMB->quant); |
1867 |
|
if (iSAD >= BFRAME_SKIP_THRESHHOLD * pMB->quant || ((bound > 1) && |
1868 |
|
((y*mb_width+x == bound) || (y*mb_width+x == bound+1)))) { /* Some third-party decoders have problems with coloc skip MB before or after |
1869 |
|
resync marker in BVOP. We avoid any ambiguity and force no skip at slice boundary */ |
1870 |
|
bSkip = 0; /* could not SKIP */ |
1871 |
|
if (pParam->vol_flags & XVID_VOL_QUARTERPEL) { |
1872 |
|
VECTOR predMV = get_qpmv2(current->mbs, pParam->mb_width, bound, x, y, 0); |
1873 |
|
pMB->pmvs[0].x = - predMV.x; |
1874 |
|
pMB->pmvs[0].y = - predMV.y; |
1875 |
|
} else { |
1876 |
|
VECTOR predMV = get_pmv2(current->mbs, pParam->mb_width, bound, x, y, 0); |
1877 |
|
pMB->pmvs[0].x = - predMV.x; |
1878 |
|
pMB->pmvs[0].y = - predMV.y; |
1879 |
|
} |
1880 |
|
pMB->mode = MODE_INTER; |
1881 |
|
pMB->cbp = 0; |
1882 |
|
break; |
1883 |
|
} |
1884 |
|
} |
1885 |
|
} |
1886 |
|
|
1887 |
|
if (bSkip) { |
1888 |
|
/* do SKIP */ |
1889 |
|
pMB->mode = MODE_NOT_CODED; |
1890 |
|
MBSkip(bs); |
1891 |
|
stop_coding_timer(); |
1892 |
|
continue; /* next MB */ |
1893 |
|
} |
1894 |
|
} |
1895 |
|
|
1896 |
|
/* ordinary case: normal coded INTER/INTER4V block */ |
1897 |
|
MBCoding(current, pMB, qcoeff, bs, data->sStat); |
1898 |
|
stop_coding_timer(); |
1899 |
|
} |
1900 |
|
} |
1901 |
|
|
1902 |
|
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
1903 |
|
emms(); |
1904 |
|
} |
1905 |
|
|
1906 |
|
/* FrameCodeP also handles S(GMC)-VOPs */ |
1907 |
|
static int |
1908 |
|
FrameCodeP(Encoder * pEnc, Bitstream * bs) |
1909 |
|
{ |
1910 |
|
int bits = BitstreamPos(bs); |
1911 |
|
|
1912 |
FRAMEINFO *const current = pEnc->current; |
FRAMEINFO *const current = pEnc->current; |
1913 |
FRAMEINFO *const reference = pEnc->reference; |
FRAMEINFO *const reference = pEnc->reference; |
1914 |
MBParam * const pParam = &pEnc->mbParam; |
MBParam * const pParam = &pEnc->mbParam; |
1916 |
int mb_height = pParam->mb_height; |
int mb_height = pParam->mb_height; |
1917 |
int coded = 1; |
int coded = 1; |
1918 |
|
|
1919 |
|
int k = 0, bound = 0, num_slices = pEnc->num_slices; |
1920 |
|
int num_threads = MAX(1, MIN(pEnc->num_threads, num_slices)); |
1921 |
|
#ifdef HAVE_PTHREAD |
1922 |
|
void * status = NULL; |
1923 |
|
int threads_per_slice = (pEnc->num_threads*1024 / num_threads); |
1924 |
|
#endif |
1925 |
|
int slices_per_thread = (num_slices*1024 / num_threads); |
1926 |
|
|
|
/* IMAGE *pCurrent = ¤t->image; */ |
|
1927 |
IMAGE *pRef = &reference->image; |
IMAGE *pRef = &reference->image; |
1928 |
|
|
|
if ((current->vop_flags & XVID_VOP_REDUCED)) |
|
|
{ |
|
|
mb_width = (pParam->width + 31) / 32; |
|
|
mb_height = (pParam->height + 31) / 32; |
|
|
} |
|
|
|
|
|
|
|
1929 |
if (!reference->is_edged) { |
if (!reference->is_edged) { |
1930 |
start_timer(); |
start_timer(); |
1931 |
image_setedges(pRef, pParam->edged_width, pParam->edged_height, |
image_setedges(pRef, pParam->edged_width, pParam->edged_height, |
1932 |
pParam->width, pParam->height, 0); |
pParam->width, pParam->height, XVID_BS_VERSION); |
1933 |
stop_edges_timer(); |
stop_edges_timer(); |
1934 |
reference->is_edged = 1; |
reference->is_edged = 1; |
1935 |
} |
} |
1941 |
if ((current->vop_flags & XVID_VOP_HALFPEL)) { |
if ((current->vop_flags & XVID_VOP_HALFPEL)) { |
1942 |
if (reference->is_interpolated != current->rounding_type) { |
if (reference->is_interpolated != current->rounding_type) { |
1943 |
start_timer(); |
start_timer(); |
1944 |
image_interpolate(pRef, &pEnc->vInterH, &pEnc->vInterV, |
image_interpolate(pRef->y, pEnc->vInterH.y, pEnc->vInterV.y, |
1945 |
&pEnc->vInterHV, pParam->edged_width, |
pEnc->vInterHV.y, pParam->edged_width, |
1946 |
pParam->edged_height, |
pParam->edged_height, |
1947 |
(pParam->vol_flags & XVID_VOL_QUARTERPEL), |
(pParam->vol_flags & XVID_VOL_QUARTERPEL), |
1948 |
current->rounding_type); |
current->rounding_type); |
1951 |
} |
} |
1952 |
} |
} |
1953 |
|
|
1954 |
|
current->sStat.iTextBits = current->sStat.iMvSum = current->sStat.iMvCount = |
1955 |
|
current->sStat.kblks = current->sStat.mblks = current->sStat.ublks = |
1956 |
|
current->sStat.iMVBits = 0; |
1957 |
|
|
1958 |
current->coding_type = P_VOP; |
current->coding_type = P_VOP; |
1959 |
|
|
1960 |
|
if (current->vop_flags & XVID_VOP_RD_PSNRHVSM) { |
1961 |
|
image_block_variance(¤t->image, pParam->edged_width, current->mbs, |
1962 |
|
pParam->mb_width, pParam->mb_height); |
1963 |
|
} |
1964 |
|
|
1965 |
call_plugins(pEnc, pEnc->current, NULL, XVID_PLG_FRAME, NULL, NULL, NULL); |
call_plugins(pEnc, pEnc->current, NULL, XVID_PLG_FRAME, NULL, NULL, NULL); |
1966 |
|
|
1967 |
SetMacroblockQuants(&pEnc->mbParam, current); |
SetMacroblockQuants(&pEnc->mbParam, current); |
1970 |
if (current->vol_flags & XVID_VOL_GMC ) /* GMC only for S(GMC)-VOPs */ |
if (current->vol_flags & XVID_VOL_GMC ) /* GMC only for S(GMC)-VOPs */ |
1971 |
{ int gmcval; |
{ int gmcval; |
1972 |
current->warp = GlobalMotionEst( current->mbs, pParam, current, reference, |
current->warp = GlobalMotionEst( current->mbs, pParam, current, reference, |
1973 |
&pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV); |
&pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, num_slices); |
1974 |
|
|
1975 |
if (current->motion_flags & XVID_ME_GME_REFINE) { |
if (current->motion_flags & XVID_ME_GME_REFINE) { |
1976 |
gmcval = GlobalMotionEstRefine(¤t->warp, |
gmcval = GlobalMotionEstRefine(¤t->warp, |
2017 |
} |
} |
2018 |
} |
} |
2019 |
|
|
2020 |
MotionEstimation(&pEnc->mbParam, current, reference, |
#ifdef HAVE_PTHREAD |
2021 |
&pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, |
if (pEnc->num_threads > 0) { |
|
&pEnc->vGMC, 256*4096); |
|
|
|
|
|
|
|
|
stop_motion_timer(); |
|
|
|
|
|
set_timecodes(current,reference,pParam->fbase); |
|
|
|
|
|
BitstreamWriteVopHeader(bs, &pEnc->mbParam, current, 1, current->mbs[0].quant); |
|
|
|
|
|
current->sStat.iTextBits = current->sStat.iMvSum = current->sStat.iMvCount = |
|
|
current->sStat.kblks = current->sStat.mblks = current->sStat.ublks = 0; |
|
|
|
|
2022 |
|
|
2023 |
for (y = 0; y < mb_height; y++) { |
/* multithreaded motion estimation - dispatch threads */ |
2024 |
for (x = 0; x < mb_width; x++) { |
while (k < pEnc->num_threads) { |
2025 |
MACROBLOCK *pMB = |
int i, add_s = (slices_per_thread + 512) >> 10; |
2026 |
¤t->mbs[x + y * pParam->mb_width]; |
int add_t = (threads_per_slice + 512) >> 10; |
2027 |
|
|
2028 |
int bIntra = (pMB->mode == MODE_INTRA) || (pMB->mode == MODE_INTRA_Q); |
int start_y = (bound * mb_height + (num_slices-1)) / num_slices; |
2029 |
|
int stop_y = ((bound+add_s) * mb_height + (num_slices-1)) / num_slices; |
2030 |
|
int rows_per_thread = (stop_y - start_y + add_t - 1) / add_t; |
2031 |
|
|
2032 |
if (bIntra) { |
slices_per_thread += ((num_slices*1024 / num_threads) - add_s*1024); |
2033 |
CodeIntraMB(pEnc, pMB); |
threads_per_slice += ((pEnc->num_threads*1024 / num_threads) - add_t*1024); |
|
MBTransQuantIntra(&pEnc->mbParam, current, pMB, x, y, |
|
|
dct_codes, qcoeff); |
|
2034 |
|
|
2035 |
start_timer(); |
for (i = 0; i < add_t; i++) { |
2036 |
MBPrediction(current, x, y, pParam->mb_width, qcoeff); |
memset(pEnc->smpData[k+i].complete_count_self, 0, rows_per_thread * sizeof(int)); |
|
stop_prediction_timer(); |
|
2037 |
|
|
2038 |
current->sStat.kblks++; |
pEnc->smpData[k+i].pEnc = (void *) pEnc; |
2039 |
|
pEnc->smpData[k+i].y_row = i; |
2040 |
|
pEnc->smpData[k+i].y_step = add_t; |
2041 |
|
pEnc->smpData[k+i].stop_y = stop_y; |
2042 |
|
pEnc->smpData[k+i].start_y = start_y; |
2043 |
|
|
2044 |
if (pEnc->current->vop_flags & XVID_VOP_GREYSCALE) |
/* todo: sort out temp space once and for all */ |
2045 |
{ pMB->cbp &= 0x3C; /* keep only bits 5-2 */ |
pEnc->smpData[k+i].RefQ = (((k+i)&1) ? pEnc->vInterV.u : pEnc->vInterV.v) + |
2046 |
qcoeff[4*64+0]=0; /* zero, because for INTRA MBs DC value is saved */ |
16*((k+i)>>1)*pParam->edged_width; |
|
qcoeff[5*64+0]=0; |
|
|
} |
|
|
MBCoding(current, pMB, qcoeff, bs, ¤t->sStat); |
|
|
stop_coding_timer(); |
|
|
continue; |
|
2047 |
} |
} |
2048 |
|
|
2049 |
start_timer(); |
pEnc->smpData[k].complete_count_above = |
2050 |
MBMotionCompensation(pMB, x, y, &reference->image, |
pEnc->smpData[k+add_t-1].complete_count_self - 1; |
|
&pEnc->vInterH, &pEnc->vInterV, |
|
|
&pEnc->vInterHV, &pEnc->vGMC, |
|
|
¤t->image, |
|
|
dct_codes, pParam->width, |
|
|
pParam->height, |
|
|
pParam->edged_width, |
|
|
(current->vol_flags & XVID_VOL_QUARTERPEL), |
|
|
(current->vop_flags & XVID_VOP_REDUCED), |
|
|
current->rounding_type); |
|
2051 |
|
|
2052 |
stop_comp_timer(); |
bound += add_s; |
2053 |
|
k += add_t; |
|
pMB->field_pred = 0; |
|
|
|
|
|
if (pMB->mode != MODE_NOT_CODED) |
|
|
{ pMB->cbp = |
|
|
MBTransQuantInter(&pEnc->mbParam, current, pMB, x, y, |
|
|
dct_codes, qcoeff); |
|
2054 |
} |
} |
2055 |
|
|
2056 |
if (pMB->dquant != 0) |
for (k = 1; k < pEnc->num_threads; k++) { |
2057 |
MBSetDquant(pMB, x, y, &pEnc->mbParam); |
pthread_create(&pEnc->smpData[k].handle, NULL, |
2058 |
|
(void*)MotionEstimateSMP, (void*)&pEnc->smpData[k]); |
|
|
|
|
if (pMB->cbp || pMB->mvs[0].x || pMB->mvs[0].y || |
|
|
pMB->mvs[1].x || pMB->mvs[1].y || pMB->mvs[2].x || |
|
|
pMB->mvs[2].y || pMB->mvs[3].x || pMB->mvs[3].y) { |
|
|
current->sStat.mblks++; |
|
|
} else { |
|
|
current->sStat.ublks++; |
|
2059 |
} |
} |
2060 |
|
|
2061 |
start_timer(); |
MotionEstimateSMP(&pEnc->smpData[0]); |
|
|
|
|
/* Finished processing the MB, now check if to CODE or SKIP */ |
|
2062 |
|
|
2063 |
skip_possible = (pMB->cbp == 0) && (pMB->mode == MODE_INTER) && |
for (k = 1; k < pEnc->num_threads; k++) { |
2064 |
(pMB->dquant == 0); |
pthread_join(pEnc->smpData[k].handle, &status); |
2065 |
|
} |
2066 |
|
|
2067 |
if (current->coding_type == S_VOP) |
current->fcode = 0; |
2068 |
skip_possible &= (pMB->mcsel == 1); |
for (k = 0; k < pEnc->num_threads; k++) { |
2069 |
else if (current->coding_type == P_VOP) { |
current->sStat.iMvSum += pEnc->smpData[k].mvSum; |
2070 |
if ((pParam->vol_flags & XVID_VOL_QUARTERPEL)) |
current->sStat.iMvCount += pEnc->smpData[k].mvCount; |
2071 |
skip_possible &= ( (pMB->qmvs[0].x == 0) && (pMB->qmvs[0].y == 0) ); |
if (pEnc->smpData[k].minfcode > current->fcode) |
2072 |
else |
current->fcode = pEnc->smpData[k].minfcode; |
|
skip_possible &= ( (pMB->mvs[0].x == 0) && (pMB->mvs[0].y == 0) ); |
|
2073 |
} |
} |
2074 |
|
|
2075 |
if ( (pMB->mode == MODE_NOT_CODED) || (skip_possible)) { |
} else |
2076 |
|
#endif |
2077 |
|
{ |
2078 |
|
|
2079 |
/* This is a candidate for SKIPping, but for P-VOPs check intermediate B-frames first */ |
/* regular ME */ |
2080 |
|
|
2081 |
if (current->coding_type == P_VOP) /* special rule for P-VOP's SKIP */ |
MotionEstimation(&pEnc->mbParam, current, reference, |
2082 |
{ |
&pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, |
2083 |
int bSkip = 1; |
&pEnc->vGMC, 256*4096, num_slices); |
2084 |
|
|
|
for (k=pEnc->bframenum_head; k< pEnc->bframenum_tail; k++) |
|
|
{ |
|
|
int iSAD; |
|
|
iSAD = sad16(reference->image.y + 16*y*pParam->edged_width + 16*x, |
|
|
pEnc->bframes[k]->image.y + 16*y*pParam->edged_width + 16*x, |
|
|
pParam->edged_width,BFRAME_SKIP_THRESHHOLD); |
|
|
if (iSAD >= BFRAME_SKIP_THRESHHOLD * pMB->quant) |
|
|
{ bSkip = 0; |
|
|
break; |
|
|
} |
|
2085 |
} |
} |
2086 |
|
|
2087 |
if (!bSkip) { /* no SKIP, but trivial block */ |
stop_motion_timer(); |
|
if((pParam->vol_flags & XVID_VOL_QUARTERPEL)) { |
|
|
VECTOR predMV = get_qpmv2(current->mbs, pParam->mb_width, 0, x, y, 0); |
|
|
pMB->pmvs[0].x = - predMV.x; |
|
|
pMB->pmvs[0].y = - predMV.y; |
|
|
} |
|
|
else { |
|
|
VECTOR predMV = get_pmv2(current->mbs, pParam->mb_width, 0, x, y, 0); |
|
|
pMB->pmvs[0].x = - predMV.x; |
|
|
pMB->pmvs[0].y = - predMV.y; |
|
|
} |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->cbp = 0; |
|
|
MBCoding(current, pMB, qcoeff, bs, ¤t->sStat); |
|
|
stop_coding_timer(); |
|
2088 |
|
|
2089 |
continue; /* next MB */ |
set_timecodes(current,reference,pParam->fbase); |
|
} |
|
|
} |
|
|
/* do SKIP */ |
|
2090 |
|
|
2091 |
pMB->mode = MODE_NOT_CODED; |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, current, 1, current->mbs[0].quant); |
|
MBSkip(bs); |
|
|
stop_coding_timer(); |
|
|
continue; /* next MB */ |
|
|
} |
|
|
/* ordinary case: normal coded INTER/INTER4V block */ |
|
2092 |
|
|
2093 |
if ((current->vop_flags & XVID_VOP_GREYSCALE)) |
/* multithreaded inter coding - dispatch threads */ |
|
{ pMB->cbp &= 0x3C; /* keep only bits 5-2 */ |
|
|
qcoeff[4*64+0]=0; /* zero, because DC for INTRA MBs DC value is saved */ |
|
|
qcoeff[5*64+0]=0; |
|
|
} |
|
2094 |
|
|
2095 |
if((pParam->vol_flags & XVID_VOL_QUARTERPEL)) { |
bound = 0; |
2096 |
VECTOR predMV = get_qpmv2(current->mbs, pParam->mb_width, 0, x, y, 0); |
slices_per_thread = (num_slices*1024 / num_threads); |
|
pMB->pmvs[0].x = pMB->qmvs[0].x - predMV.x; |
|
|
pMB->pmvs[0].y = pMB->qmvs[0].y - predMV.y; |
|
|
DPRINTF(XVID_DEBUG_MV,"mv_diff (%i,%i) pred (%i,%i) result (%i,%i)\n", pMB->pmvs[0].x, pMB->pmvs[0].y, predMV.x, predMV.y, pMB->mvs[0].x, pMB->mvs[0].y); |
|
|
} else { |
|
|
VECTOR predMV = get_pmv2(current->mbs, pParam->mb_width, 0, x, y, 0); |
|
|
pMB->pmvs[0].x = pMB->mvs[0].x - predMV.x; |
|
|
pMB->pmvs[0].y = pMB->mvs[0].y - predMV.y; |
|
|
DPRINTF(XVID_DEBUG_MV,"mv_diff (%i,%i) pred (%i,%i) result (%i,%i)\n", pMB->pmvs[0].x, pMB->pmvs[0].y, predMV.x, predMV.y, pMB->mvs[0].x, pMB->mvs[0].y); |
|
|
} |
|
2097 |
|
|
2098 |
|
for (k = 0; k < num_threads; k++) { |
2099 |
|
int add = ((slices_per_thread + 512) >> 10); |
2100 |
|
|
2101 |
if (pMB->mode == MODE_INTER4V) |
slices_per_thread += ((num_slices*1024 / num_threads) - add*1024); |
|
{ int k; |
|
|
for (k=1;k<4;k++) |
|
|
{ |
|
|
if((pParam->vol_flags & XVID_VOL_QUARTERPEL)) { |
|
|
VECTOR predMV = get_qpmv2(current->mbs, pParam->mb_width, 0, x, y, k); |
|
|
pMB->pmvs[k].x = pMB->qmvs[k].x - predMV.x; |
|
|
pMB->pmvs[k].y = pMB->qmvs[k].y - predMV.y; |
|
|
DPRINTF(XVID_DEBUG_MV,"mv_diff (%i,%i) pred (%i,%i) result (%i,%i)\n", pMB->pmvs[k].x, pMB->pmvs[k].y, predMV.x, predMV.y, pMB->mvs[k].x, pMB->mvs[k].y); |
|
|
} else { |
|
|
VECTOR predMV = get_pmv2(current->mbs, pParam->mb_width, 0, x, y, k); |
|
|
pMB->pmvs[k].x = pMB->mvs[k].x - predMV.x; |
|
|
pMB->pmvs[k].y = pMB->mvs[k].y - predMV.y; |
|
|
DPRINTF(XVID_DEBUG_MV,"mv_diff (%i,%i) pred (%i,%i) result (%i,%i)\n", pMB->pmvs[k].x, pMB->pmvs[k].y, predMV.x, predMV.y, pMB->mvs[k].x, pMB->mvs[k].y); |
|
|
} |
|
2102 |
|
|
2103 |
} |
pEnc->smpData[k].pEnc = (void *) pEnc; |
2104 |
} |
pEnc->smpData[k].stop_y = (((bound+add) * mb_height + (num_slices-1)) / num_slices); |
2105 |
|
pEnc->smpData[k].start_y = ((bound * mb_height + (num_slices-1)) / num_slices); |
2106 |
|
pEnc->smpData[k].RefQ = ((k&1) ? pEnc->vInterV.u : pEnc->vInterV.v) + 16*(k>>1)*pParam->edged_width; |
2107 |
|
|
2108 |
MBCoding(current, pMB, qcoeff, bs, &pEnc->current->sStat); |
bound += add; |
2109 |
stop_coding_timer(); |
|
2110 |
|
if (k > 0) { |
2111 |
|
pEnc->smpData[k].sStat->iTextBits = pEnc->smpData[k].sStat->kblks = |
2112 |
|
pEnc->smpData[k].sStat->mblks = pEnc->smpData[k].sStat->ublks = |
2113 |
|
pEnc->smpData[k].sStat->iMVBits = 0; |
2114 |
|
|
2115 |
|
BitstreamReset(pEnc->smpData[k].bs); |
2116 |
} |
} |
2117 |
} |
} |
2118 |
|
pEnc->smpData[0].bs = bs; |
2119 |
|
pEnc->smpData[0].sStat = ¤t->sStat; |
2120 |
|
|
2121 |
if ((current->vop_flags & XVID_VOP_REDUCED)) |
#ifdef HAVE_PTHREAD |
2122 |
{ |
/* create threads */ |
2123 |
image_deblock_rrv(¤t->image, pParam->edged_width, |
for (k = 1; k < num_threads; k++) { |
2124 |
current->mbs, mb_width, mb_height, pParam->mb_width, |
pthread_create(&pEnc->smpData[k].handle, NULL, |
2125 |
16, 0); |
(void*)SliceCodeP, (void*)&pEnc->smpData[k]); |
2126 |
} |
} |
2127 |
|
#endif |
2128 |
|
|
2129 |
emms(); |
SliceCodeP(&pEnc->smpData[0]); |
|
|
|
|
if (current->sStat.iMvCount == 0) |
|
|
current->sStat.iMvCount = 1; |
|
2130 |
|
|
2131 |
fSigma = (float) sqrt((float) current->sStat.iMvSum / current->sStat.iMvCount); |
#ifdef HAVE_PTHREAD |
2132 |
|
/* wait until all threads are finished */ |
2133 |
|
for (k = 1; k < num_threads; k++) { |
2134 |
|
pthread_join(pEnc->smpData[k].handle, &status); |
2135 |
|
} |
2136 |
|
#endif |
2137 |
|
|
2138 |
iSearchRange = 1 << (3 + pParam->m_fcode); |
current->length = BitstreamLength(bs) - (bits/8); |
2139 |
|
|
2140 |
if ((fSigma > iSearchRange / 3) |
/* reassemble the pieces together */ |
2141 |
&& (pParam->m_fcode <= (3 + (pParam->vol_flags & XVID_VOL_QUARTERPEL?1:0) ))) /* maximum search range 128 */ |
SerializeBitstreams(pEnc, pEnc->current, bs, num_threads); |
|
{ |
|
|
pParam->m_fcode++; |
|
|
iSearchRange *= 2; |
|
|
} else if ((fSigma < iSearchRange / 6) |
|
|
&& (pEnc->fMvPrevSigma >= 0) |
|
|
&& (pEnc->fMvPrevSigma < iSearchRange / 6) |
|
|
&& (pParam->m_fcode >= (2 + (pParam->vol_flags & XVID_VOL_QUARTERPEL?1:0) ))) /* minimum search range 16 */ |
|
|
{ |
|
|
pParam->m_fcode--; |
|
|
iSearchRange /= 2; |
|
|
} |
|
2142 |
|
|
2143 |
pEnc->fMvPrevSigma = fSigma; |
updateFcode(¤t->sStat, pEnc); |
2144 |
|
|
2145 |
/* frame drop code */ |
/* frame drop code */ |
2146 |
#if 0 |
#if 0 |
2147 |
DPRINTF(XVID_DEBUG_DEBUG, "kmu %i %i %i\n", current->sStat.kblks, current->sStat.mblks, current->sStat.ublks); |
DPRINTF(XVID_DEBUG_DEBUG, "kmu %i %i %i\n", current->sStat.kblks, current->sStat.mblks, current->sStat.ublks); |
2148 |
#endif |
#endif |
2149 |
if (current->sStat.kblks + current->sStat.mblks <= |
|
2150 |
|
if (current->sStat.kblks + current->sStat.mblks < |
2151 |
(pParam->frame_drop_ratio * mb_width * mb_height) / 100 && |
(pParam->frame_drop_ratio * mb_width * mb_height) / 100 && |
2152 |
( (pEnc->bframenum_head >= pEnc->bframenum_tail) || !(pEnc->mbParam.global_flags & XVID_GLOBAL_CLOSED_GOP)) ) |
( (pEnc->bframenum_head >= pEnc->bframenum_tail) || !(pEnc->mbParam.global_flags & XVID_GLOBAL_CLOSED_GOP)) && |
2153 |
|
(current->coding_type == P_VOP) ) |
2154 |
{ |
{ |
2155 |
current->sStat.kblks = current->sStat.mblks = 0; |
current->sStat.kblks = current->sStat.mblks = current->sStat.iTextBits = 0; |
2156 |
current->sStat.ublks = mb_width * mb_height; |
current->sStat.ublks = mb_width * mb_height; |
2157 |
|
|
2158 |
BitstreamReset(bs); |
BitstreamReset(bs); |
2171 |
memcpy(current->mbs, reference->mbs, sizeof(MACROBLOCK) * mb_width * mb_height); |
memcpy(current->mbs, reference->mbs, sizeof(MACROBLOCK) * mb_width * mb_height); |
2172 |
coded = 0; |
coded = 0; |
2173 |
|
|
2174 |
|
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
2175 |
|
|
2176 |
|
current->length = (BitstreamPos(bs) - bits) / 8; |
2177 |
|
|
2178 |
} else { |
} else { |
2179 |
|
|
2180 |
pEnc->current->is_edged = 0; /* not edged */ |
pEnc->current->is_edged = 0; /* not edged */ |
2203 |
} |
} |
2204 |
*/ |
*/ |
2205 |
|
|
2206 |
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
return coded; |
2207 |
|
} |
2208 |
|
|
2209 |
current->length = (BitstreamPos(bs) - bits) / 8; |
static void |
2210 |
|
SliceCodeB(SMPData *data) |
2211 |
|
{ |
2212 |
|
Encoder *pEnc = (Encoder *) data->pEnc; |
2213 |
|
Bitstream *bs = (Bitstream *) data->bs; |
2214 |
|
|
2215 |
return coded; |
DECLARE_ALIGNED_MATRIX(dct_codes, 6, 64, int16_t, CACHE_LINE); |
2216 |
|
DECLARE_ALIGNED_MATRIX(qcoeff, 6, 64, int16_t, CACHE_LINE); |
2217 |
|
|
2218 |
|
int x, y; |
2219 |
|
FRAMEINFO * const frame = (FRAMEINFO * const) data->current; |
2220 |
|
MBParam * const pParam = &pEnc->mbParam; |
2221 |
|
int mb_width = pParam->mb_width; |
2222 |
|
int mb_height = pParam->mb_height; |
2223 |
|
IMAGE *f_ref = &pEnc->reference->image; |
2224 |
|
IMAGE *b_ref = &pEnc->current->image; |
2225 |
|
|
2226 |
|
int bound = data->start_y*mb_width; |
2227 |
|
int num_slices = pEnc->num_slices; |
2228 |
|
|
2229 |
|
if (data->start_y > 0) { /* write resync marker */ |
2230 |
|
write_video_packet_header(bs, pParam, frame, bound+1); |
2231 |
|
} |
2232 |
|
|
2233 |
|
for (y = data->start_y; y < MIN(data->stop_y+1, mb_height); y++) { |
2234 |
|
int new_bound = mb_width * ((((y*num_slices) / mb_height) * mb_height + (num_slices-1)) / num_slices); |
2235 |
|
int stop_x = (y == data->stop_y) ? 1 : mb_width; |
2236 |
|
int start_x = (y == data->start_y && y > 0) ? 1 : 0; |
2237 |
|
|
2238 |
|
for (x = start_x; x < stop_x; x++) { |
2239 |
|
MACROBLOCK * const mb = &frame->mbs[x + y * pEnc->mbParam.mb_width]; |
2240 |
|
|
2241 |
|
/* decoder ignores mb when refence block is INTER(0,0), CBP=0 */ |
2242 |
|
if (mb->mode == MODE_NOT_CODED) { |
2243 |
|
if (pParam->plugin_flags & XVID_REQORIGINAL) { |
2244 |
|
MBMotionCompensation(mb, x, y, f_ref, NULL, f_ref, NULL, NULL, &frame->image, |
2245 |
|
NULL, 0, 0, pParam->edged_width, 0, 0, data->RefQ); |
2246 |
|
} |
2247 |
|
continue; |
2248 |
|
} |
2249 |
|
|
2250 |
|
if (new_bound > bound && x > 0) { |
2251 |
|
bound = new_bound; |
2252 |
|
BitstreamPadAlways(bs); |
2253 |
|
write_video_packet_header(bs, pParam, frame, y*mb_width+x); |
2254 |
|
} |
2255 |
|
|
2256 |
|
mb->quant = frame->quant; |
2257 |
|
|
2258 |
|
if (mb->cbp != 0 || pParam->plugin_flags & XVID_REQORIGINAL) { |
2259 |
|
/* we have to motion-compensate, transfer etc, |
2260 |
|
because there might be blocks to code */ |
2261 |
|
|
2262 |
|
MBMotionCompensationBVOP(pParam, mb, x, y, &frame->image, |
2263 |
|
f_ref, &pEnc->f_refh, &pEnc->f_refv, |
2264 |
|
&pEnc->f_refhv, b_ref, &pEnc->vInterH, |
2265 |
|
&pEnc->vInterV, &pEnc->vInterHV, dct_codes, |
2266 |
|
data->RefQ); |
2267 |
|
|
2268 |
|
mb->cbp = MBTransQuantInterBVOP(pParam, frame, mb, x, y, dct_codes, qcoeff); |
2269 |
} |
} |
2270 |
|
|
2271 |
|
if (mb->mode == MODE_DIRECT_NO4V) |
2272 |
|
mb->mode = MODE_DIRECT; |
2273 |
|
|
2274 |
|
if (mb->mode == MODE_DIRECT && (mb->cbp | mb->pmvs[3].x | mb->pmvs[3].y) == 0) |
2275 |
|
mb->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
2276 |
|
else |
2277 |
|
if (frame->vop_flags & XVID_VOP_GREYSCALE) |
2278 |
|
/* keep only bits 5-2 -- Chroma blocks will just be skipped by MBCodingBVOP */ |
2279 |
|
mb->cbp &= 0x3C; |
2280 |
|
|
2281 |
|
start_timer(); |
2282 |
|
MBCodingBVOP(frame, mb, qcoeff, frame->fcode, frame->bcode, bs, data->sStat); |
2283 |
|
stop_coding_timer(); |
2284 |
|
} |
2285 |
|
} |
2286 |
|
|
2287 |
|
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
2288 |
|
emms(); |
2289 |
|
} |
2290 |
|
|
2291 |
static void |
static void |
2292 |
FrameCodeB(Encoder * pEnc, |
FrameCodeB(Encoder * pEnc, |
2294 |
Bitstream * bs) |
Bitstream * bs) |
2295 |
{ |
{ |
2296 |
int bits = BitstreamPos(bs); |
int bits = BitstreamPos(bs); |
2297 |
DECLARE_ALIGNED_MATRIX(dct_codes, 6, 64, int16_t, CACHE_LINE); |
int k = 0, bound = 0, num_slices = pEnc->num_slices; |
2298 |
DECLARE_ALIGNED_MATRIX(qcoeff, 6, 64, int16_t, CACHE_LINE); |
int num_threads = MAX(1, MIN(pEnc->num_threads, num_slices)); |
2299 |
uint32_t x, y; |
#ifdef HAVE_PTHREAD |
2300 |
|
void * status = NULL; |
2301 |
|
int threads_per_slice = (pEnc->num_threads*1024 / num_threads); |
2302 |
|
#endif |
2303 |
|
int slices_per_thread = (num_slices*1024 / num_threads); |
2304 |
|
|
2305 |
IMAGE *f_ref = &pEnc->reference->image; |
IMAGE *f_ref = &pEnc->reference->image; |
2306 |
IMAGE *b_ref = &pEnc->current->image; |
IMAGE *b_ref = &pEnc->current->image; |
2307 |
|
|
2308 |
|
MBParam * const pParam = &pEnc->mbParam; |
2309 |
|
int mb_height = pParam->mb_height; |
2310 |
|
|
2311 |
#ifdef BFRAMES_DEC_DEBUG |
#ifdef BFRAMES_DEC_DEBUG |
2312 |
FILE *fp; |
FILE *fp; |
2313 |
static char first=0; |
static char first=0; |
2315 |
fprintf(fp,"Y=%3d X=%3d MB=%2d CBP=%02X\n",y,x,mb->mode,mb->cbp); \ |
fprintf(fp,"Y=%3d X=%3d MB=%2d CBP=%02X\n",y,x,mb->mode,mb->cbp); \ |
2316 |
} |
} |
2317 |
|
|
|
/* XXX: pEnc->current->global_flags &= ~XVID_VOP_REDUCED; reduced resoltion not yet supported */ |
|
|
|
|
2318 |
if (!first){ |
if (!first){ |
2319 |
fp=fopen("C:\\XVIDDBGE.TXT","w"); |
fp=fopen("C:\\XVIDDBGE.TXT","w"); |
2320 |
} |
} |
2324 |
if (!pEnc->reference->is_edged) { |
if (!pEnc->reference->is_edged) { |
2325 |
image_setedges(f_ref, pEnc->mbParam.edged_width, |
image_setedges(f_ref, pEnc->mbParam.edged_width, |
2326 |
pEnc->mbParam.edged_height, pEnc->mbParam.width, |
pEnc->mbParam.edged_height, pEnc->mbParam.width, |
2327 |
pEnc->mbParam.height, 0); |
pEnc->mbParam.height, XVID_BS_VERSION); |
2328 |
pEnc->current->is_edged = 1; |
pEnc->reference->is_edged = 1; |
2329 |
} |
} |
2330 |
|
|
2331 |
if (pEnc->reference->is_interpolated != 0) { |
if (pEnc->reference->is_interpolated != 0) { |
2332 |
start_timer(); |
start_timer(); |
2333 |
image_interpolate(f_ref, &pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, |
image_interpolate(f_ref->y, pEnc->f_refh.y, pEnc->f_refv.y, pEnc->f_refhv.y, |
2334 |
pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, |
pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, |
2335 |
(pEnc->mbParam.vol_flags & XVID_VOL_QUARTERPEL), 0); |
(pEnc->mbParam.vol_flags & XVID_VOL_QUARTERPEL), 0); |
2336 |
stop_inter_timer(); |
stop_inter_timer(); |
2341 |
if (!pEnc->current->is_edged) { |
if (!pEnc->current->is_edged) { |
2342 |
image_setedges(b_ref, pEnc->mbParam.edged_width, |
image_setedges(b_ref, pEnc->mbParam.edged_width, |
2343 |
pEnc->mbParam.edged_height, pEnc->mbParam.width, |
pEnc->mbParam.edged_height, pEnc->mbParam.width, |
2344 |
pEnc->mbParam.height, 0); |
pEnc->mbParam.height, XVID_BS_VERSION); |
2345 |
pEnc->current->is_edged = 1; |
pEnc->current->is_edged = 1; |
2346 |
} |
} |
2347 |
|
|
2348 |
if (pEnc->current->is_interpolated != 0) { |
if (pEnc->current->is_interpolated != 0) { |
2349 |
start_timer(); |
start_timer(); |
2350 |
image_interpolate(b_ref, &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, |
image_interpolate(b_ref->y, pEnc->vInterH.y, pEnc->vInterV.y, pEnc->vInterHV.y, |
2351 |
pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, |
pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, |
2352 |
(pEnc->mbParam.vol_flags & XVID_VOL_QUARTERPEL), 0); |
(pEnc->mbParam.vol_flags & XVID_VOL_QUARTERPEL), 0); |
2353 |
stop_inter_timer(); |
stop_inter_timer(); |
2355 |
} |
} |
2356 |
|
|
2357 |
frame->coding_type = B_VOP; |
frame->coding_type = B_VOP; |
2358 |
call_plugins(pEnc, pEnc->current, NULL, XVID_PLG_FRAME, NULL, NULL, NULL); |
|
2359 |
|
if ((frame->vop_flags & XVID_VOP_RD_PSNRHVSM) && (frame->vop_flags & XVID_VOP_RD_BVOP)) { |
2360 |
|
image_block_variance(&frame->image, pEnc->mbParam.edged_width, frame->mbs, |
2361 |
|
pEnc->mbParam.mb_width, pEnc->mbParam.mb_height); |
2362 |
|
} |
2363 |
|
|
2364 |
|
call_plugins(pEnc, frame, NULL, XVID_PLG_FRAME, NULL, NULL, NULL); |
2365 |
|
|
2366 |
|
frame->fcode = frame->bcode = pEnc->current->fcode; |
2367 |
|
|
2368 |
start_timer(); |
start_timer(); |
2369 |
|
|
2370 |
|
#ifdef HAVE_PTHREAD |
2371 |
|
if (pEnc->num_threads > 0) { |
2372 |
|
|
2373 |
|
/* multithreaded motion estimation - dispatch threads */ |
2374 |
|
while (k < pEnc->num_threads) { |
2375 |
|
int i, add_s = (slices_per_thread + 512) >> 10; |
2376 |
|
int add_t = (threads_per_slice + 512) >> 10; |
2377 |
|
|
2378 |
|
int start_y = (bound * mb_height + (num_slices-1)) / num_slices; |
2379 |
|
int stop_y = ((bound+add_s) * mb_height + (num_slices-1)) / num_slices; |
2380 |
|
int rows_per_thread = (stop_y - start_y + add_t - 1) / add_t; |
2381 |
|
|
2382 |
|
slices_per_thread += ((num_slices*1024 / num_threads) - add_s*1024); |
2383 |
|
threads_per_slice += ((pEnc->num_threads*1024 / num_threads) - add_t*1024); |
2384 |
|
|
2385 |
|
for (i = 0; i < add_t; i++) { |
2386 |
|
memset(pEnc->smpData[k+i].complete_count_self, 0, rows_per_thread * sizeof(int)); |
2387 |
|
|
2388 |
|
pEnc->smpData[k+i].pEnc = (void *) pEnc; |
2389 |
|
pEnc->smpData[k+i].current = frame; |
2390 |
|
|
2391 |
|
pEnc->smpData[k+i].y_row = i; |
2392 |
|
pEnc->smpData[k+i].y_step = add_t; |
2393 |
|
pEnc->smpData[k+i].stop_y = stop_y; |
2394 |
|
pEnc->smpData[k+i].start_y = start_y; |
2395 |
|
|
2396 |
|
/* todo: sort out temp space once and for all */ |
2397 |
|
pEnc->smpData[k+i].RefQ = (((k+i)&1) ? pEnc->vInterV.u : pEnc->vInterV.v) + |
2398 |
|
16*((k+i)>>1)*pParam->edged_width; |
2399 |
|
} |
2400 |
|
|
2401 |
|
pEnc->smpData[k].complete_count_above = |
2402 |
|
pEnc->smpData[k+add_t-1].complete_count_self - 1; |
2403 |
|
|
2404 |
|
bound += add_s; |
2405 |
|
k += add_t; |
2406 |
|
} |
2407 |
|
|
2408 |
|
for (k = 1; k < pEnc->num_threads; k++) { |
2409 |
|
pthread_create(&pEnc->smpData[k].handle, NULL, |
2410 |
|
(void*)SMPMotionEstimationBVOP, (void*)&pEnc->smpData[k]); |
2411 |
|
} |
2412 |
|
|
2413 |
|
SMPMotionEstimationBVOP(&pEnc->smpData[0]); |
2414 |
|
|
2415 |
|
for (k = 1; k < pEnc->num_threads; k++) { |
2416 |
|
pthread_join(pEnc->smpData[k].handle, &status); |
2417 |
|
} |
2418 |
|
|
2419 |
|
frame->fcode = frame->bcode = 0; |
2420 |
|
for (k = 0; k < pEnc->num_threads; k++) { |
2421 |
|
if (pEnc->smpData[k].minfcode > frame->fcode) |
2422 |
|
frame->fcode = pEnc->smpData[k].minfcode; |
2423 |
|
if (pEnc->smpData[k].minbcode > frame->bcode) |
2424 |
|
frame->bcode = pEnc->smpData[k].minbcode; |
2425 |
|
} |
2426 |
|
} else |
2427 |
|
#endif |
2428 |
|
{ |
2429 |
|
|
2430 |
MotionEstimationBVOP(&pEnc->mbParam, frame, |
MotionEstimationBVOP(&pEnc->mbParam, frame, |
2431 |
((int32_t)(pEnc->current->stamp - frame->stamp)), /* time_bp */ |
((int32_t)(pEnc->current->stamp - frame->stamp)), /* time_bp */ |
2432 |
((int32_t)(pEnc->current->stamp - pEnc->reference->stamp)), /* time_pp */ |
((int32_t)(pEnc->current->stamp - pEnc->reference->stamp)), /* time_pp */ |
2433 |
pEnc->reference->mbs, f_ref, |
pEnc->reference->mbs, f_ref, |
2434 |
&pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, |
&pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, |
2435 |
pEnc->current, b_ref, &pEnc->vInterH, |
pEnc->current, b_ref, &pEnc->vInterH, |
2436 |
&pEnc->vInterV, &pEnc->vInterHV); |
&pEnc->vInterV, &pEnc->vInterHV, |
2437 |
|
pEnc->num_slices); |
2438 |
|
} |
2439 |
stop_motion_timer(); |
stop_motion_timer(); |
2440 |
|
|
2441 |
set_timecodes(frame, pEnc->reference,pEnc->mbParam.fbase); |
set_timecodes(frame, pEnc->reference,pEnc->mbParam.fbase); |
2442 |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, frame, 1, frame->quant); |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, frame, 1, frame->quant); |
2443 |
|
|
2444 |
|
/* reset stats */ |
2445 |
frame->sStat.iTextBits = 0; |
frame->sStat.iTextBits = 0; |
2446 |
|
frame->sStat.iMVBits = 0; |
2447 |
frame->sStat.iMvSum = 0; |
frame->sStat.iMvSum = 0; |
2448 |
frame->sStat.iMvCount = 0; |
frame->sStat.iMvCount = 0; |
2449 |
frame->sStat.kblks = frame->sStat.mblks = frame->sStat.ublks = 0; |
frame->sStat.kblks = frame->sStat.mblks = frame->sStat.ublks = 0; |
2450 |
frame->sStat.mblks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; |
frame->sStat.mblks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; |
2451 |
frame->sStat.kblks = frame->sStat.ublks = 0; |
frame->sStat.kblks = frame->sStat.ublks = 0; |
2452 |
|
|
2453 |
for (y = 0; y < pEnc->mbParam.mb_height; y++) { |
/* multithreaded inter coding - dispatch threads */ |
2454 |
for (x = 0; x < pEnc->mbParam.mb_width; x++) { |
bound = 0; |
2455 |
MACROBLOCK * const mb = &frame->mbs[x + y * pEnc->mbParam.mb_width]; |
slices_per_thread = (num_slices*1024 / num_threads); |
2456 |
|
|
2457 |
/* decoder ignores mb when refence block is INTER(0,0), CBP=0 */ |
for (k = 0; k < num_threads; k++) { |
2458 |
if (mb->mode == MODE_NOT_CODED) { |
int add = ((slices_per_thread + 512) >> 10); |
|
if (pEnc->mbParam.plugin_flags & XVID_REQORIGINAL) { |
|
|
MBMotionCompensation(mb, x, y, f_ref, NULL, f_ref, NULL, NULL, &frame->image, |
|
|
NULL, 0, 0, pEnc->mbParam.edged_width, 0, 0, 0); |
|
|
} |
|
2459 |
|
|
2460 |
continue; |
slices_per_thread += ((num_slices*1024 / num_threads) - add*1024); |
|
} |
|
2461 |
|
|
2462 |
if (mb->mode != MODE_DIRECT_NONE_MV || pEnc->mbParam.plugin_flags & XVID_REQORIGINAL) { |
pEnc->smpData[k].pEnc = (void *) pEnc; |
2463 |
MBMotionCompensationBVOP(&pEnc->mbParam, mb, x, y, &frame->image, |
pEnc->smpData[k].current = frame; |
2464 |
f_ref, &pEnc->f_refh, &pEnc->f_refv, |
pEnc->smpData[k].stop_y = (((bound+add) * mb_height + (num_slices-1)) / num_slices); |
2465 |
&pEnc->f_refhv, b_ref, &pEnc->vInterH, |
pEnc->smpData[k].start_y = ((bound * mb_height + (num_slices-1)) / num_slices); |
2466 |
&pEnc->vInterV, &pEnc->vInterHV, |
bound += add; |
|
dct_codes); |
|
|
|
|
|
if (mb->mode == MODE_DIRECT_NO4V) mb->mode = MODE_DIRECT; |
|
|
mb->quant = frame->quant; |
|
2467 |
|
|
2468 |
if (mb->mode != MODE_DIRECT_NONE_MV) |
/* todo: sort out temp space once and for all */ |
2469 |
mb->cbp = MBTransQuantInterBVOP(&pEnc->mbParam, frame, mb, x, y, dct_codes, qcoeff); |
pEnc->smpData[k].RefQ = ((k&1) ? pEnc->vInterV.u : pEnc->vInterV.v) + 16*(k>>1)*pParam->edged_width; |
2470 |
|
|
2471 |
if ( (mb->mode == MODE_DIRECT) && (mb->cbp == 0) |
if (k > 0) { |
2472 |
&& (mb->pmvs[3].x == 0) && (mb->pmvs[3].y == 0) ) { |
BitstreamReset(pEnc->smpData[k].bs); |
2473 |
mb->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
pEnc->smpData[k].sStat->iTextBits = pEnc->smpData[k].sStat->kblks = |
2474 |
|
pEnc->smpData[k].sStat->mblks = pEnc->smpData[k].sStat->ublks = pEnc->smpData[k].sStat->iMVBits = 0; |
2475 |
} |
} |
2476 |
} |
} |
2477 |
|
|
2478 |
/* keep only bits 5-2 -- Chroma blocks will just be skipped by the |
#ifdef HAVE_PTHREAD |
2479 |
* coding function for BFrames, that's why we don't zero teh DC |
for (k = 1; k < num_threads; k++) { |
2480 |
* coeffs */ |
pthread_create(&pEnc->smpData[k].handle, NULL, |
2481 |
if ((frame->vop_flags & XVID_VOP_GREYSCALE)) |
(void*)SliceCodeB, (void*)&pEnc->smpData[k]); |
|
mb->cbp &= 0x3C; |
|
|
|
|
|
start_timer(); |
|
|
MBCodingBVOP(frame, mb, qcoeff, frame->fcode, frame->bcode, bs, |
|
|
&frame->sStat); |
|
|
stop_coding_timer(); |
|
|
} |
|
2482 |
} |
} |
2483 |
|
#endif |
2484 |
|
|
2485 |
emms(); |
pEnc->smpData[0].bs = bs; |
2486 |
|
pEnc->smpData[0].sStat = &frame->sStat; |
2487 |
|
SliceCodeB(&pEnc->smpData[0]); |
2488 |
|
|
2489 |
|
#ifdef HAVE_PTHREAD |
2490 |
|
for (k = 1; k < num_threads; k++) { |
2491 |
|
pthread_join(pEnc->smpData[k].handle, &status); |
2492 |
|
} |
2493 |
|
#endif |
2494 |
|
|
2495 |
/* TODO: dynamic fcode/bcode ??? */ |
frame->length = BitstreamLength(bs) - (bits/8); |
2496 |
|
|
2497 |
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
/* reassemble the pieces together */ |
2498 |
frame->length = (BitstreamPos(bs) - bits) / 8; |
SerializeBitstreams(pEnc, frame, bs, num_threads); |
2499 |
|
|
2500 |
#ifdef BFRAMES_DEC_DEBUG |
#ifdef BFRAMES_DEC_DEBUG |
2501 |
if (!first){ |
if (!first){ |