00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024
00030 #include "libavutil/intmath.h"
00031 #include "avcodec.h"
00032 #include "dsputil.h"
00033 #include "mpegvideo.h"
00034 #include "mpegvideo_common.h"
00035 #include "h263.h"
00036 #include "mjpegenc.h"
00037 #include "msmpeg4.h"
00038 #include "faandct.h"
00039 #include "thread.h"
00040 #include "aandcttab.h"
00041 #include "flv.h"
00042 #include "mpeg4video.h"
00043 #include "internal.h"
00044 #include <limits.h>
00045
00046
00047
00048
00049 static int encode_picture(MpegEncContext *s, int picture_number);
00050 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
00051 static int sse_mb(MpegEncContext *s);
00052 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
00053 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
00054
00055
00056
00057
00058
00059
00060 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
00061 static uint8_t default_fcode_tab[MAX_MV*2+1];
00062
00063 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
00064 const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
00065 {
00066 int qscale;
00067 int shift=0;
00068
00069 for(qscale=qmin; qscale<=qmax; qscale++){
00070 int i;
00071 if (dsp->fdct == ff_jpeg_fdct_islow
00072 #ifdef FAAN_POSTSCALE
00073 || dsp->fdct == ff_faandct
00074 #endif
00075 ) {
00076 for(i=0;i<64;i++) {
00077 const int j= dsp->idct_permutation[i];
00078
00079
00080
00081
00082
00083 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
00084 (qscale * quant_matrix[j]));
00085 }
00086 } else if (dsp->fdct == fdct_ifast
00087 #ifndef FAAN_POSTSCALE
00088 || dsp->fdct == ff_faandct
00089 #endif
00090 ) {
00091 for(i=0;i<64;i++) {
00092 const int j= dsp->idct_permutation[i];
00093
00094
00095
00096
00097
00098 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
00099 (ff_aanscales[i] * qscale * quant_matrix[j]));
00100 }
00101 } else {
00102 for(i=0;i<64;i++) {
00103 const int j= dsp->idct_permutation[i];
00104
00105
00106
00107
00108
00109 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
00110
00111 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
00112
00113 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
00114 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
00115 }
00116 }
00117
00118 for(i=intra; i<64; i++){
00119 int64_t max= 8191;
00120 if (dsp->fdct == fdct_ifast
00121 #ifndef FAAN_POSTSCALE
00122 || dsp->fdct == ff_faandct
00123 #endif
00124 ) {
00125 max = (8191LL*ff_aanscales[i]) >> 14;
00126 }
00127 while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
00128 shift++;
00129 }
00130 }
00131 }
00132 if(shift){
00133 av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
00134 }
00135 }
00136
00137 static inline void update_qscale(MpegEncContext *s){
00138 s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
00139 s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
00140
00141 s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
00142 }
00143
00144 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
00145 int i;
00146
00147 if(matrix){
00148 put_bits(pb, 1, 1);
00149 for(i=0;i<64;i++) {
00150 put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
00151 }
00152 }else
00153 put_bits(pb, 1, 0);
00154 }
00155
00159 void ff_init_qscale_tab(MpegEncContext *s){
00160 int8_t * const qscale_table= s->current_picture.qscale_table;
00161 int i;
00162
00163 for(i=0; i<s->mb_num; i++){
00164 unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
00165 int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
00166 qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
00167 }
00168 }
00169
00170 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
00171 int i;
00172
00173 dst->pict_type = src->pict_type;
00174 dst->quality = src->quality;
00175 dst->coded_picture_number = src->coded_picture_number;
00176 dst->display_picture_number = src->display_picture_number;
00177
00178 dst->pts = src->pts;
00179 dst->interlaced_frame = src->interlaced_frame;
00180 dst->top_field_first = src->top_field_first;
00181
00182 if(s->avctx->me_threshold){
00183 if(!src->motion_val[0])
00184 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
00185 if(!src->mb_type)
00186 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
00187 if(!src->ref_index[0])
00188 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
00189 if(src->motion_subsample_log2 != dst->motion_subsample_log2)
00190 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
00191 src->motion_subsample_log2, dst->motion_subsample_log2);
00192
00193 memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
00194
00195 for(i=0; i<2; i++){
00196 int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
00197 int height= ((16*s->mb_height)>>src->motion_subsample_log2);
00198
00199 if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
00200 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
00201 }
00202 if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
00203 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t));
00204 }
00205 }
00206 }
00207 }
00208
00209 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
00210 #define COPY(a) dst->a= src->a
00211 COPY(pict_type);
00212 COPY(current_picture);
00213 COPY(f_code);
00214 COPY(b_code);
00215 COPY(qscale);
00216 COPY(lambda);
00217 COPY(lambda2);
00218 COPY(picture_in_gop_number);
00219 COPY(gop_picture_number);
00220 COPY(frame_pred_frame_dct);
00221 COPY(progressive_frame);
00222 COPY(partitioned_frame);
00223 #undef COPY
00224 }
00225
00230 static void MPV_encode_defaults(MpegEncContext *s){
00231 int i;
00232 MPV_common_defaults(s);
00233
00234 for(i=-16; i<16; i++){
00235 default_fcode_tab[i + MAX_MV]= 1;
00236 }
00237 s->me.mv_penalty= default_mv_penalty;
00238 s->fcode_tab= default_fcode_tab;
00239 }
00240
00241
00242 av_cold int MPV_encode_init(AVCodecContext *avctx)
00243 {
00244 MpegEncContext *s = avctx->priv_data;
00245 int i;
00246 int chroma_h_shift, chroma_v_shift;
00247
00248 MPV_encode_defaults(s);
00249
00250 switch (avctx->codec_id) {
00251 case CODEC_ID_MPEG2VIDEO:
00252 if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
00253 av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
00254 return -1;
00255 }
00256 break;
00257 case CODEC_ID_LJPEG:
00258 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_YUVJ444P && avctx->pix_fmt != PIX_FMT_BGRA &&
00259 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P && avctx->pix_fmt != PIX_FMT_YUV444P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
00260 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
00261 return -1;
00262 }
00263 break;
00264 case CODEC_ID_MJPEG:
00265 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
00266 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
00267 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
00268 return -1;
00269 }
00270 break;
00271 default:
00272 if(avctx->pix_fmt != PIX_FMT_YUV420P){
00273 av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
00274 return -1;
00275 }
00276 }
00277
00278 switch (avctx->pix_fmt) {
00279 case PIX_FMT_YUVJ422P:
00280 case PIX_FMT_YUV422P:
00281 s->chroma_format = CHROMA_422;
00282 break;
00283 case PIX_FMT_YUVJ420P:
00284 case PIX_FMT_YUV420P:
00285 default:
00286 s->chroma_format = CHROMA_420;
00287 break;
00288 }
00289
00290 s->bit_rate = avctx->bit_rate;
00291 s->width = avctx->width;
00292 s->height = avctx->height;
00293 if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
00294 av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
00295 avctx->gop_size=600;
00296 }
00297 s->gop_size = avctx->gop_size;
00298 s->avctx = avctx;
00299 s->flags= avctx->flags;
00300 s->flags2= avctx->flags2;
00301 s->max_b_frames= avctx->max_b_frames;
00302 s->codec_id= avctx->codec->id;
00303 s->luma_elim_threshold = avctx->luma_elim_threshold;
00304 s->chroma_elim_threshold= avctx->chroma_elim_threshold;
00305 s->strict_std_compliance= avctx->strict_std_compliance;
00306 s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
00307 s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
00308 s->mpeg_quant= avctx->mpeg_quant;
00309 s->rtp_mode= !!avctx->rtp_payload_size;
00310 s->intra_dc_precision= avctx->intra_dc_precision;
00311 s->user_specified_pts = AV_NOPTS_VALUE;
00312
00313 if (s->gop_size <= 1) {
00314 s->intra_only = 1;
00315 s->gop_size = 12;
00316 } else {
00317 s->intra_only = 0;
00318 }
00319
00320 s->me_method = avctx->me_method;
00321
00322
00323 s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
00324
00325 s->adaptive_quant= ( s->avctx->lumi_masking
00326 || s->avctx->dark_masking
00327 || s->avctx->temporal_cplx_masking
00328 || s->avctx->spatial_cplx_masking
00329 || s->avctx->p_masking
00330 || s->avctx->border_masking
00331 || (s->flags&CODEC_FLAG_QP_RD))
00332 && !s->fixed_qscale;
00333
00334 s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
00335 s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
00336 s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
00337 s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
00338 s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
00339
00340 if(avctx->rc_max_rate && !avctx->rc_buffer_size){
00341 av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
00342 return -1;
00343 }
00344
00345 if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
00346 av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
00347 }
00348
00349 if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
00350 av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
00351 return -1;
00352 }
00353
00354 if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
00355 av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
00356 return -1;
00357 }
00358
00359 if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
00360 av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
00361 }
00362
00363 if(avctx->rc_buffer_size && avctx->bit_rate*(int64_t)avctx->time_base.num > avctx->rc_buffer_size * (int64_t)avctx->time_base.den){
00364 av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
00365 return -1;
00366 }
00367
00368 if(!s->fixed_qscale && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
00369 av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
00370 return -1;
00371 }
00372
00373 if( s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
00374 && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
00375 && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
00376
00377 av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n");
00378 }
00379
00380 if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
00381 && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
00382 av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
00383 return -1;
00384 }
00385
00386 if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
00387 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
00388 return -1;
00389 }
00390
00391 if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
00392 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
00393 return -1;
00394 }
00395
00396 if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
00397 av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
00398 return -1;
00399 }
00400
00401 if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
00402 av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
00403 return -1;
00404 }
00405
00406 if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
00407 av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
00408 return -1;
00409 }
00410
00411 if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
00412 s->codec_id == CODEC_ID_H263P) &&
00413 (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
00414 av_log(avctx, AV_LOG_WARNING, "Invalid pixel aspect ratio %i/%i, limit is 255/255 reducing\n",
00415 avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
00416 av_reduce(&avctx->sample_aspect_ratio.num, &avctx->sample_aspect_ratio.den,
00417 avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den, 255);
00418 }
00419
00420 if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
00421 && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
00422 av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
00423 return -1;
00424 }
00425
00426 if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){
00427 av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
00428 return -1;
00429 }
00430
00431 if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
00432 av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
00433 return -1;
00434 }
00435
00436 if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
00437 av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
00438 return -1;
00439 }
00440
00441 if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
00442 av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
00443 return -1;
00444 }
00445
00446 if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
00447 av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
00448 return -1;
00449 }
00450
00451 if(s->flags & CODEC_FLAG_LOW_DELAY){
00452 if (s->codec_id != CODEC_ID_MPEG2VIDEO){
00453 av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
00454 return -1;
00455 }
00456 if (s->max_b_frames != 0){
00457 av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
00458 return -1;
00459 }
00460 }
00461
00462 if(s->q_scale_type == 1){
00463 if(s->codec_id != CODEC_ID_MPEG2VIDEO){
00464 av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
00465 return -1;
00466 }
00467 if(avctx->qmax > 12){
00468 av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
00469 return -1;
00470 }
00471 }
00472
00473 if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
00474 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
00475 && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
00476 av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
00477 return -1;
00478 }
00479
00480 if(s->avctx->thread_count < 1){
00481 av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
00482 return -1;
00483 }
00484
00485 if(s->avctx->thread_count > 1)
00486 s->rtp_mode= 1;
00487
00488 if(!avctx->time_base.den || !avctx->time_base.num){
00489 av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
00490 return -1;
00491 }
00492
00493 i= (INT_MAX/2+128)>>8;
00494 if(avctx->me_threshold >= i){
00495 av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
00496 return -1;
00497 }
00498 if(avctx->mb_threshold >= i){
00499 av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
00500 return -1;
00501 }
00502
00503 if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
00504 av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
00505 avctx->b_frame_strategy = 0;
00506 }
00507
00508 i= av_gcd(avctx->time_base.den, avctx->time_base.num);
00509 if(i > 1){
00510 av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
00511 avctx->time_base.den /= i;
00512 avctx->time_base.num /= i;
00513
00514 }
00515
00516 if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
00517 s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3);
00518 s->inter_quant_bias= 0;
00519 }else{
00520 s->intra_quant_bias=0;
00521 s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2));
00522 }
00523
00524 if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
00525 s->intra_quant_bias= avctx->intra_quant_bias;
00526 if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
00527 s->inter_quant_bias= avctx->inter_quant_bias;
00528
00529 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
00530
00531 if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
00532 av_log(avctx, AV_LOG_ERROR, "timebase %d/%d not supported by MPEG 4 standard, "
00533 "the maximum admitted value for the timebase denominator is %d\n",
00534 s->avctx->time_base.num, s->avctx->time_base.den, (1<<16)-1);
00535 return -1;
00536 }
00537 s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
00538
00539 switch(avctx->codec->id) {
00540 case CODEC_ID_MPEG1VIDEO:
00541 s->out_format = FMT_MPEG1;
00542 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
00543 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
00544 break;
00545 case CODEC_ID_MPEG2VIDEO:
00546 s->out_format = FMT_MPEG1;
00547 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
00548 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
00549 s->rtp_mode= 1;
00550 break;
00551 case CODEC_ID_LJPEG:
00552 case CODEC_ID_MJPEG:
00553 s->out_format = FMT_MJPEG;
00554 s->intra_only = 1;
00555 if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
00556 s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
00557 s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
00558 s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
00559 }else{
00560 s->mjpeg_vsample[0] = 2;
00561 s->mjpeg_vsample[1] = 2>>chroma_v_shift;
00562 s->mjpeg_vsample[2] = 2>>chroma_v_shift;
00563 s->mjpeg_hsample[0] = 2;
00564 s->mjpeg_hsample[1] = 2>>chroma_h_shift;
00565 s->mjpeg_hsample[2] = 2>>chroma_h_shift;
00566 }
00567 if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
00568 || ff_mjpeg_encode_init(s) < 0)
00569 return -1;
00570 avctx->delay=0;
00571 s->low_delay=1;
00572 break;
00573 case CODEC_ID_H261:
00574 if (!CONFIG_H261_ENCODER) return -1;
00575 if (ff_h261_get_picture_format(s->width, s->height) < 0) {
00576 av_log(avctx, AV_LOG_ERROR, "The specified picture size of %dx%d is not valid for the H.261 codec.\nValid sizes are 176x144, 352x288\n", s->width, s->height);
00577 return -1;
00578 }
00579 s->out_format = FMT_H261;
00580 avctx->delay=0;
00581 s->low_delay=1;
00582 break;
00583 case CODEC_ID_H263:
00584 if (!CONFIG_H263_ENCODER) return -1;
00585 if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 8) {
00586 av_log(avctx, AV_LOG_ERROR, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height);
00587 return -1;
00588 }
00589 s->out_format = FMT_H263;
00590 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
00591 avctx->delay=0;
00592 s->low_delay=1;
00593 break;
00594 case CODEC_ID_H263P:
00595 s->out_format = FMT_H263;
00596 s->h263_plus = 1;
00597
00598 s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
00599 s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
00600 s->modified_quant= s->h263_aic;
00601 s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
00602 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
00603 s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
00604 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
00605 s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
00606
00607
00608
00609 avctx->delay=0;
00610 s->low_delay=1;
00611 break;
00612 case CODEC_ID_FLV1:
00613 s->out_format = FMT_H263;
00614 s->h263_flv = 2;
00615 s->unrestricted_mv = 1;
00616 s->rtp_mode=0;
00617 avctx->delay=0;
00618 s->low_delay=1;
00619 break;
00620 case CODEC_ID_RV10:
00621 s->out_format = FMT_H263;
00622 avctx->delay=0;
00623 s->low_delay=1;
00624 break;
00625 case CODEC_ID_RV20:
00626 s->out_format = FMT_H263;
00627 avctx->delay=0;
00628 s->low_delay=1;
00629 s->modified_quant=1;
00630 s->h263_aic=1;
00631 s->h263_plus=1;
00632 s->loop_filter=1;
00633 s->unrestricted_mv= 0;
00634 break;
00635 case CODEC_ID_MPEG4:
00636 s->out_format = FMT_H263;
00637 s->h263_pred = 1;
00638 s->unrestricted_mv = 1;
00639 s->low_delay= s->max_b_frames ? 0 : 1;
00640 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
00641 break;
00642 case CODEC_ID_MSMPEG4V2:
00643 s->out_format = FMT_H263;
00644 s->h263_pred = 1;
00645 s->unrestricted_mv = 1;
00646 s->msmpeg4_version= 2;
00647 avctx->delay=0;
00648 s->low_delay=1;
00649 break;
00650 case CODEC_ID_MSMPEG4V3:
00651 s->out_format = FMT_H263;
00652 s->h263_pred = 1;
00653 s->unrestricted_mv = 1;
00654 s->msmpeg4_version= 3;
00655 s->flipflop_rounding=1;
00656 avctx->delay=0;
00657 s->low_delay=1;
00658 break;
00659 case CODEC_ID_WMV1:
00660 s->out_format = FMT_H263;
00661 s->h263_pred = 1;
00662 s->unrestricted_mv = 1;
00663 s->msmpeg4_version= 4;
00664 s->flipflop_rounding=1;
00665 avctx->delay=0;
00666 s->low_delay=1;
00667 break;
00668 case CODEC_ID_WMV2:
00669 s->out_format = FMT_H263;
00670 s->h263_pred = 1;
00671 s->unrestricted_mv = 1;
00672 s->msmpeg4_version= 5;
00673 s->flipflop_rounding=1;
00674 avctx->delay=0;
00675 s->low_delay=1;
00676 break;
00677 default:
00678 return -1;
00679 }
00680
00681 avctx->has_b_frames= !s->low_delay;
00682
00683 s->encoding = 1;
00684
00685 s->progressive_frame=
00686 s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
00687
00688
00689 if (MPV_common_init(s) < 0)
00690 return -1;
00691
00692 if(!s->dct_quantize)
00693 s->dct_quantize = dct_quantize_c;
00694 if(!s->denoise_dct)
00695 s->denoise_dct = denoise_dct_c;
00696 s->fast_dct_quantize = s->dct_quantize;
00697 if(avctx->trellis)
00698 s->dct_quantize = dct_quantize_trellis_c;
00699
00700 if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
00701 s->chroma_qscale_table= ff_h263_chroma_qscale_table;
00702
00703 s->quant_precision=5;
00704
00705 ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
00706 ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
00707
00708 if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
00709 ff_h261_encode_init(s);
00710 if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
00711 h263_encode_init(s);
00712 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
00713 ff_msmpeg4_encode_init(s);
00714 if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
00715 && s->out_format == FMT_MPEG1)
00716 ff_mpeg1_encode_init(s);
00717
00718
00719 for(i=0;i<64;i++) {
00720 int j= s->dsp.idct_permutation[i];
00721 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
00722 s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
00723 s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
00724 }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
00725 s->intra_matrix[j] =
00726 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
00727 }else
00728 {
00729 s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
00730 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
00731 }
00732 if(s->avctx->intra_matrix)
00733 s->intra_matrix[j] = s->avctx->intra_matrix[i];
00734 if(s->avctx->inter_matrix)
00735 s->inter_matrix[j] = s->avctx->inter_matrix[i];
00736 }
00737
00738
00739
00740 if (s->out_format != FMT_MJPEG) {
00741 ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
00742 s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
00743 ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
00744 s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
00745 }
00746
00747 if(ff_rate_control_init(s) < 0)
00748 return -1;
00749
00750 return 0;
00751 }
00752
00753 av_cold int MPV_encode_end(AVCodecContext *avctx)
00754 {
00755 MpegEncContext *s = avctx->priv_data;
00756
00757 ff_rate_control_uninit(s);
00758
00759 MPV_common_end(s);
00760 if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
00761 ff_mjpeg_encode_close(s);
00762
00763 av_freep(&avctx->extradata);
00764
00765 return 0;
00766 }
00767
00768 static int get_sae(uint8_t *src, int ref, int stride){
00769 int x,y;
00770 int acc=0;
00771
00772 for(y=0; y<16; y++){
00773 for(x=0; x<16; x++){
00774 acc+= FFABS(src[x+y*stride] - ref);
00775 }
00776 }
00777
00778 return acc;
00779 }
00780
00781 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
00782 int x, y, w, h;
00783 int acc=0;
00784
00785 w= s->width &~15;
00786 h= s->height&~15;
00787
00788 for(y=0; y<h; y+=16){
00789 for(x=0; x<w; x+=16){
00790 int offset= x + y*stride;
00791 int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
00792 int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
00793 int sae = get_sae(src + offset, mean, stride);
00794
00795 acc+= sae + 500 < sad;
00796 }
00797 }
00798 return acc;
00799 }
00800
00801
00802 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
00803 AVFrame *pic=NULL;
00804 int64_t pts;
00805 int i;
00806 const int encoding_delay= s->max_b_frames;
00807 int direct=1;
00808
00809 if(pic_arg){
00810 pts= pic_arg->pts;
00811 pic_arg->display_picture_number= s->input_picture_number++;
00812
00813 if(pts != AV_NOPTS_VALUE){
00814 if(s->user_specified_pts != AV_NOPTS_VALUE){
00815 int64_t time= pts;
00816 int64_t last= s->user_specified_pts;
00817
00818 if(time <= last){
00819 av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
00820 return -1;
00821 }
00822 }
00823 s->user_specified_pts= pts;
00824 }else{
00825 if(s->user_specified_pts != AV_NOPTS_VALUE){
00826 s->user_specified_pts=
00827 pts= s->user_specified_pts + 1;
00828 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
00829 }else{
00830 pts= pic_arg->display_picture_number;
00831 }
00832 }
00833 }
00834
00835 if(pic_arg){
00836 if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
00837 if(pic_arg->linesize[0] != s->linesize) direct=0;
00838 if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
00839 if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
00840
00841
00842
00843 if(direct){
00844 i= ff_find_unused_picture(s, 1);
00845
00846 pic= (AVFrame*)&s->picture[i];
00847 pic->reference= 3;
00848
00849 for(i=0; i<4; i++){
00850 pic->data[i]= pic_arg->data[i];
00851 pic->linesize[i]= pic_arg->linesize[i];
00852 }
00853 if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){
00854 return -1;
00855 }
00856 }else{
00857 i= ff_find_unused_picture(s, 0);
00858
00859 pic= (AVFrame*)&s->picture[i];
00860 pic->reference= 3;
00861
00862 if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){
00863 return -1;
00864 }
00865
00866 if( pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
00867 && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
00868 && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
00869
00870 }else{
00871 int h_chroma_shift, v_chroma_shift;
00872 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
00873
00874 for(i=0; i<3; i++){
00875 int src_stride= pic_arg->linesize[i];
00876 int dst_stride= i ? s->uvlinesize : s->linesize;
00877 int h_shift= i ? h_chroma_shift : 0;
00878 int v_shift= i ? v_chroma_shift : 0;
00879 int w= s->width >>h_shift;
00880 int h= s->height>>v_shift;
00881 uint8_t *src= pic_arg->data[i];
00882 uint8_t *dst= pic->data[i];
00883
00884 if(!s->avctx->rc_buffer_size)
00885 dst +=INPLACE_OFFSET;
00886
00887 if(src_stride==dst_stride)
00888 memcpy(dst, src, src_stride*h);
00889 else{
00890 while(h--){
00891 memcpy(dst, src, w);
00892 dst += dst_stride;
00893 src += src_stride;
00894 }
00895 }
00896 }
00897 }
00898 }
00899 copy_picture_attributes(s, pic, pic_arg);
00900 pic->pts= pts;
00901 }
00902
00903
00904 for(i=1; i<MAX_PICTURE_COUNT ; i++)
00905 s->input_picture[i-1]= s->input_picture[i];
00906
00907 s->input_picture[encoding_delay]= (Picture*)pic;
00908
00909 return 0;
00910 }
00911
00912 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
00913 int x, y, plane;
00914 int score=0;
00915 int64_t score64=0;
00916
00917 for(plane=0; plane<3; plane++){
00918 const int stride= p->linesize[plane];
00919 const int bw= plane ? 1 : 2;
00920 for(y=0; y<s->mb_height*bw; y++){
00921 for(x=0; x<s->mb_width*bw; x++){
00922 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
00923 int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8);
00924
00925 switch(s->avctx->frame_skip_exp){
00926 case 0: score= FFMAX(score, v); break;
00927 case 1: score+= FFABS(v);break;
00928 case 2: score+= v*v;break;
00929 case 3: score64+= FFABS(v*v*(int64_t)v);break;
00930 case 4: score64+= v*v*(int64_t)(v*v);break;
00931 }
00932 }
00933 }
00934 }
00935
00936 if(score) score64= score;
00937
00938 if(score64 < s->avctx->frame_skip_threshold)
00939 return 1;
00940 if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
00941 return 1;
00942 return 0;
00943 }
00944
00945 static int estimate_best_b_count(MpegEncContext *s){
00946 AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
00947 AVCodecContext *c= avcodec_alloc_context();
00948 AVFrame input[FF_MAX_B_FRAMES+2];
00949 const int scale= s->avctx->brd_scale;
00950 int i, j, out_size, p_lambda, b_lambda, lambda2;
00951 int outbuf_size= s->width * s->height;
00952 uint8_t *outbuf= av_malloc(outbuf_size);
00953 int64_t best_rd= INT64_MAX;
00954 int best_b_count= -1;
00955
00956 assert(scale>=0 && scale <=3);
00957
00958
00959 p_lambda= s->last_lambda_for[AV_PICTURE_TYPE_P];
00960 b_lambda= s->last_lambda_for[AV_PICTURE_TYPE_B];
00961 if(!b_lambda) b_lambda= p_lambda;
00962 lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
00963
00964 c->width = s->width >> scale;
00965 c->height= s->height>> scale;
00966 c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED ;
00967 c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
00968 c->mb_decision= s->avctx->mb_decision;
00969 c->me_cmp= s->avctx->me_cmp;
00970 c->mb_cmp= s->avctx->mb_cmp;
00971 c->me_sub_cmp= s->avctx->me_sub_cmp;
00972 c->pix_fmt = PIX_FMT_YUV420P;
00973 c->time_base= s->avctx->time_base;
00974 c->max_b_frames= s->max_b_frames;
00975
00976 if (avcodec_open2(c, codec, NULL) < 0)
00977 return -1;
00978
00979 for(i=0; i<s->max_b_frames+2; i++){
00980 int ysize= c->width*c->height;
00981 int csize= (c->width/2)*(c->height/2);
00982 Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
00983
00984 avcodec_get_frame_defaults(&input[i]);
00985 input[i].data[0]= av_malloc(ysize + 2*csize);
00986 input[i].data[1]= input[i].data[0] + ysize;
00987 input[i].data[2]= input[i].data[1] + csize;
00988 input[i].linesize[0]= c->width;
00989 input[i].linesize[1]=
00990 input[i].linesize[2]= c->width/2;
00991
00992 if(pre_input_ptr && (!i || s->input_picture[i-1])) {
00993 pre_input= *pre_input_ptr;
00994
00995 if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
00996 pre_input.data[0]+=INPLACE_OFFSET;
00997 pre_input.data[1]+=INPLACE_OFFSET;
00998 pre_input.data[2]+=INPLACE_OFFSET;
00999 }
01000
01001 s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
01002 s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1);
01003 s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1);
01004 }
01005 }
01006
01007 for(j=0; j<s->max_b_frames+1; j++){
01008 int64_t rd=0;
01009
01010 if(!s->input_picture[j])
01011 break;
01012
01013 c->error[0]= c->error[1]= c->error[2]= 0;
01014
01015 input[0].pict_type= AV_PICTURE_TYPE_I;
01016 input[0].quality= 1 * FF_QP2LAMBDA;
01017 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
01018
01019
01020 for(i=0; i<s->max_b_frames+1; i++){
01021 int is_p= i % (j+1) == j || i==s->max_b_frames;
01022
01023 input[i+1].pict_type= is_p ? AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
01024 input[i+1].quality= is_p ? p_lambda : b_lambda;
01025 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
01026 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
01027 }
01028
01029
01030 while(out_size){
01031 out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
01032 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
01033 }
01034
01035 rd += c->error[0] + c->error[1] + c->error[2];
01036
01037 if(rd < best_rd){
01038 best_rd= rd;
01039 best_b_count= j;
01040 }
01041 }
01042
01043 av_freep(&outbuf);
01044 avcodec_close(c);
01045 av_freep(&c);
01046
01047 for(i=0; i<s->max_b_frames+2; i++){
01048 av_freep(&input[i].data[0]);
01049 }
01050
01051 return best_b_count;
01052 }
01053
01054 static int select_input_picture(MpegEncContext *s){
01055 int i;
01056
01057 for(i=1; i<MAX_PICTURE_COUNT; i++)
01058 s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
01059 s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
01060
01061
01062 if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
01063 if( s->next_picture_ptr==NULL || s->intra_only){
01064 s->reordered_input_picture[0]= s->input_picture[0];
01065 s->reordered_input_picture[0]->pict_type= AV_PICTURE_TYPE_I;
01066 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
01067 }else{
01068 int b_frames;
01069
01070 if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
01071 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
01072
01073
01074
01075 if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
01076 for(i=0; i<4; i++)
01077 s->input_picture[0]->data[i]= NULL;
01078 s->input_picture[0]->type= 0;
01079 }else{
01080 assert( s->input_picture[0]->type==FF_BUFFER_TYPE_USER
01081 || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
01082
01083 s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
01084 }
01085
01086 emms_c();
01087 ff_vbv_update(s, 0);
01088
01089 goto no_output_pic;
01090 }
01091 }
01092
01093 if(s->flags&CODEC_FLAG_PASS2){
01094 for(i=0; i<s->max_b_frames+1; i++){
01095 int pict_num= s->input_picture[0]->display_picture_number + i;
01096
01097 if(pict_num >= s->rc_context.num_entries)
01098 break;
01099 if(!s->input_picture[i]){
01100 s->rc_context.entry[pict_num-1].new_pict_type = AV_PICTURE_TYPE_P;
01101 break;
01102 }
01103
01104 s->input_picture[i]->pict_type=
01105 s->rc_context.entry[pict_num].new_pict_type;
01106 }
01107 }
01108
01109 if(s->avctx->b_frame_strategy==0){
01110 b_frames= s->max_b_frames;
01111 while(b_frames && !s->input_picture[b_frames]) b_frames--;
01112 }else if(s->avctx->b_frame_strategy==1){
01113 for(i=1; i<s->max_b_frames+1; i++){
01114 if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
01115 s->input_picture[i]->b_frame_score=
01116 get_intra_count(s, s->input_picture[i ]->data[0],
01117 s->input_picture[i-1]->data[0], s->linesize) + 1;
01118 }
01119 }
01120 for(i=0; i<s->max_b_frames+1; i++){
01121 if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
01122 }
01123
01124 b_frames= FFMAX(0, i-1);
01125
01126
01127 for(i=0; i<b_frames+1; i++){
01128 s->input_picture[i]->b_frame_score=0;
01129 }
01130 }else if(s->avctx->b_frame_strategy==2){
01131 b_frames= estimate_best_b_count(s);
01132 }else{
01133 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
01134 b_frames=0;
01135 }
01136
01137 emms_c();
01138
01139
01140
01141
01142 for(i= b_frames - 1; i>=0; i--){
01143 int type= s->input_picture[i]->pict_type;
01144 if(type && type != AV_PICTURE_TYPE_B)
01145 b_frames= i;
01146 }
01147 if(s->input_picture[b_frames]->pict_type == AV_PICTURE_TYPE_B && b_frames == s->max_b_frames){
01148 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
01149 }
01150
01151 if(s->picture_in_gop_number + b_frames >= s->gop_size){
01152 if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
01153 b_frames= s->gop_size - s->picture_in_gop_number - 1;
01154 }else{
01155 if(s->flags & CODEC_FLAG_CLOSED_GOP)
01156 b_frames=0;
01157 s->input_picture[b_frames]->pict_type= AV_PICTURE_TYPE_I;
01158 }
01159 }
01160
01161 if( (s->flags & CODEC_FLAG_CLOSED_GOP)
01162 && b_frames
01163 && s->input_picture[b_frames]->pict_type== AV_PICTURE_TYPE_I)
01164 b_frames--;
01165
01166 s->reordered_input_picture[0]= s->input_picture[b_frames];
01167 if(s->reordered_input_picture[0]->pict_type != AV_PICTURE_TYPE_I)
01168 s->reordered_input_picture[0]->pict_type= AV_PICTURE_TYPE_P;
01169 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
01170 for(i=0; i<b_frames; i++){
01171 s->reordered_input_picture[i+1]= s->input_picture[i];
01172 s->reordered_input_picture[i+1]->pict_type= AV_PICTURE_TYPE_B;
01173 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
01174 }
01175 }
01176 }
01177 no_output_pic:
01178 if(s->reordered_input_picture[0]){
01179 s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=AV_PICTURE_TYPE_B ? 3 : 0;
01180
01181 ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
01182
01183 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
01184
01185
01186 int i= ff_find_unused_picture(s, 0);
01187 Picture *pic= &s->picture[i];
01188
01189 pic->reference = s->reordered_input_picture[0]->reference;
01190 if(ff_alloc_picture(s, pic, 0) < 0){
01191 return -1;
01192 }
01193
01194
01195 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
01196 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
01197 for(i=0; i<4; i++)
01198 s->reordered_input_picture[0]->data[i]= NULL;
01199 s->reordered_input_picture[0]->type= 0;
01200
01201 copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
01202
01203 s->current_picture_ptr= pic;
01204 }else{
01205
01206
01207 assert( s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
01208 || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
01209
01210 s->current_picture_ptr= s->reordered_input_picture[0];
01211 for(i=0; i<4; i++){
01212 s->new_picture.data[i]+= INPLACE_OFFSET;
01213 }
01214 }
01215 ff_copy_picture(&s->current_picture, s->current_picture_ptr);
01216
01217 s->picture_number= s->new_picture.display_picture_number;
01218
01219 }else{
01220 memset(&s->new_picture, 0, sizeof(Picture));
01221 }
01222 return 0;
01223 }
01224
01225 int MPV_encode_picture(AVCodecContext *avctx,
01226 unsigned char *buf, int buf_size, void *data)
01227 {
01228 MpegEncContext *s = avctx->priv_data;
01229 AVFrame *pic_arg = data;
01230 int i, stuffing_count, context_count = avctx->thread_count;
01231
01232 for(i=0; i<context_count; i++){
01233 int start_y= s->thread_context[i]->start_mb_y;
01234 int end_y= s->thread_context[i]-> end_mb_y;
01235 int h= s->mb_height;
01236 uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
01237 uint8_t *end = buf + (size_t)(((int64_t) buf_size)* end_y/h);
01238
01239 init_put_bits(&s->thread_context[i]->pb, start, end - start);
01240 }
01241
01242 s->picture_in_gop_number++;
01243
01244 if(load_input_picture(s, pic_arg) < 0)
01245 return -1;
01246
01247 if(select_input_picture(s) < 0){
01248 return -1;
01249 }
01250
01251
01252 if(s->new_picture.data[0]){
01253 s->pict_type= s->new_picture.pict_type;
01254
01255
01256 MPV_frame_start(s, avctx);
01257 vbv_retry:
01258 if (encode_picture(s, s->picture_number) < 0)
01259 return -1;
01260
01261 avctx->header_bits = s->header_bits;
01262 avctx->mv_bits = s->mv_bits;
01263 avctx->misc_bits = s->misc_bits;
01264 avctx->i_tex_bits = s->i_tex_bits;
01265 avctx->p_tex_bits = s->p_tex_bits;
01266 avctx->i_count = s->i_count;
01267 avctx->p_count = s->mb_num - s->i_count - s->skip_count;
01268 avctx->skip_count = s->skip_count;
01269
01270 MPV_frame_end(s);
01271
01272 if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
01273 ff_mjpeg_encode_picture_trailer(s);
01274
01275 if(avctx->rc_buffer_size){
01276 RateControlContext *rcc= &s->rc_context;
01277 int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
01278
01279 if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
01280 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
01281 if(s->adaptive_quant){
01282 int i;
01283 for(i=0; i<s->mb_height*s->mb_stride; i++)
01284 s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
01285 }
01286 s->mb_skipped = 0;
01287 if(s->pict_type==AV_PICTURE_TYPE_P){
01288 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
01289 s->no_rounding ^= 1;
01290 }
01291 if(s->pict_type!=AV_PICTURE_TYPE_B){
01292 s->time_base= s->last_time_base;
01293 s->last_non_b_time= s->time - s->pp_time;
01294 }
01295
01296 for(i=0; i<context_count; i++){
01297 PutBitContext *pb= &s->thread_context[i]->pb;
01298 init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
01299 }
01300 goto vbv_retry;
01301 }
01302
01303 assert(s->avctx->rc_max_rate);
01304 }
01305
01306 if(s->flags&CODEC_FLAG_PASS1)
01307 ff_write_pass1_stats(s);
01308
01309 for(i=0; i<4; i++){
01310 s->current_picture_ptr->error[i]= s->current_picture.error[i];
01311 avctx->error[i] += s->current_picture_ptr->error[i];
01312 }
01313
01314 if(s->flags&CODEC_FLAG_PASS1)
01315 assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
01316 flush_put_bits(&s->pb);
01317 s->frame_bits = put_bits_count(&s->pb);
01318
01319 stuffing_count= ff_vbv_update(s, s->frame_bits);
01320 if(stuffing_count){
01321 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
01322 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
01323 return -1;
01324 }
01325
01326 switch(s->codec_id){
01327 case CODEC_ID_MPEG1VIDEO:
01328 case CODEC_ID_MPEG2VIDEO:
01329 while(stuffing_count--){
01330 put_bits(&s->pb, 8, 0);
01331 }
01332 break;
01333 case CODEC_ID_MPEG4:
01334 put_bits(&s->pb, 16, 0);
01335 put_bits(&s->pb, 16, 0x1C3);
01336 stuffing_count -= 4;
01337 while(stuffing_count--){
01338 put_bits(&s->pb, 8, 0xFF);
01339 }
01340 break;
01341 default:
01342 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
01343 }
01344 flush_put_bits(&s->pb);
01345 s->frame_bits = put_bits_count(&s->pb);
01346 }
01347
01348
01349 if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
01350 && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
01351 int vbv_delay, min_delay;
01352 double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
01353 int minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
01354 double bits = s->rc_context.buffer_index + minbits - inbits;
01355
01356 if(bits<0)
01357 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
01358
01359 assert(s->repeat_first_field==0);
01360
01361 vbv_delay= bits * 90000 / s->avctx->rc_max_rate;
01362 min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
01363
01364 vbv_delay= FFMAX(vbv_delay, min_delay);
01365
01366 assert(vbv_delay < 0xFFFF);
01367
01368 s->vbv_delay_ptr[0] &= 0xF8;
01369 s->vbv_delay_ptr[0] |= vbv_delay>>13;
01370 s->vbv_delay_ptr[1] = vbv_delay>>5;
01371 s->vbv_delay_ptr[2] &= 0x07;
01372 s->vbv_delay_ptr[2] |= vbv_delay<<3;
01373 avctx->vbv_delay = vbv_delay*300;
01374 }
01375 s->total_bits += s->frame_bits;
01376 avctx->frame_bits = s->frame_bits;
01377 }else{
01378 assert((put_bits_ptr(&s->pb) == s->pb.buf));
01379 s->frame_bits=0;
01380 }
01381 assert((s->frame_bits&7)==0);
01382
01383 return s->frame_bits/8;
01384 }
01385
01386 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
01387 {
01388 static const char tab[64]=
01389 {3,2,2,1,1,1,1,1,
01390 1,1,1,1,1,1,1,1,
01391 1,1,1,1,1,1,1,1,
01392 0,0,0,0,0,0,0,0,
01393 0,0,0,0,0,0,0,0,
01394 0,0,0,0,0,0,0,0,
01395 0,0,0,0,0,0,0,0,
01396 0,0,0,0,0,0,0,0};
01397 int score=0;
01398 int run=0;
01399 int i;
01400 DCTELEM *block= s->block[n];
01401 const int last_index= s->block_last_index[n];
01402 int skip_dc;
01403
01404 if(threshold<0){
01405 skip_dc=0;
01406 threshold= -threshold;
01407 }else
01408 skip_dc=1;
01409
01410
01411 if(last_index<=skip_dc - 1) return;
01412
01413 for(i=0; i<=last_index; i++){
01414 const int j = s->intra_scantable.permutated[i];
01415 const int level = FFABS(block[j]);
01416 if(level==1){
01417 if(skip_dc && i==0) continue;
01418 score+= tab[run];
01419 run=0;
01420 }else if(level>1){
01421 return;
01422 }else{
01423 run++;
01424 }
01425 }
01426 if(score >= threshold) return;
01427 for(i=skip_dc; i<=last_index; i++){
01428 const int j = s->intra_scantable.permutated[i];
01429 block[j]=0;
01430 }
01431 if(block[0]) s->block_last_index[n]= 0;
01432 else s->block_last_index[n]= -1;
01433 }
01434
01435 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
01436 {
01437 int i;
01438 const int maxlevel= s->max_qcoeff;
01439 const int minlevel= s->min_qcoeff;
01440 int overflow=0;
01441
01442 if(s->mb_intra){
01443 i=1;
01444 }else
01445 i=0;
01446
01447 for(;i<=last_index; i++){
01448 const int j= s->intra_scantable.permutated[i];
01449 int level = block[j];
01450
01451 if (level>maxlevel){
01452 level=maxlevel;
01453 overflow++;
01454 }else if(level<minlevel){
01455 level=minlevel;
01456 overflow++;
01457 }
01458
01459 block[j]= level;
01460 }
01461
01462 if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
01463 av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
01464 }
01465
01466 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
01467 int x, y;
01468
01469 for(y=0; y<8; y++){
01470 for(x=0; x<8; x++){
01471 int x2, y2;
01472 int sum=0;
01473 int sqr=0;
01474 int count=0;
01475
01476 for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
01477 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
01478 int v= ptr[x2 + y2*stride];
01479 sum += v;
01480 sqr += v*v;
01481 count++;
01482 }
01483 }
01484 weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
01485 }
01486 }
01487 }
01488
01489 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
01490 {
01491 int16_t weight[8][64];
01492 DCTELEM orig[8][64];
01493 const int mb_x= s->mb_x;
01494 const int mb_y= s->mb_y;
01495 int i;
01496 int skip_dct[8];
01497 int dct_offset = s->linesize*8;
01498 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
01499 int wrap_y, wrap_c;
01500
01501 for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
01502
01503 if(s->adaptive_quant){
01504 const int last_qp= s->qscale;
01505 const int mb_xy= mb_x + mb_y*s->mb_stride;
01506
01507 s->lambda= s->lambda_table[mb_xy];
01508 update_qscale(s);
01509
01510 if(!(s->flags&CODEC_FLAG_QP_RD)){
01511 s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
01512 s->dquant= s->qscale - last_qp;
01513
01514 if(s->out_format==FMT_H263){
01515 s->dquant= av_clip(s->dquant, -2, 2);
01516
01517 if(s->codec_id==CODEC_ID_MPEG4){
01518 if(!s->mb_intra){
01519 if(s->pict_type == AV_PICTURE_TYPE_B){
01520 if(s->dquant&1 || s->mv_dir&MV_DIRECT)
01521 s->dquant= 0;
01522 }
01523 if(s->mv_type==MV_TYPE_8X8)
01524 s->dquant=0;
01525 }
01526 }
01527 }
01528 }
01529 ff_set_qscale(s, last_qp + s->dquant);
01530 }else if(s->flags&CODEC_FLAG_QP_RD)
01531 ff_set_qscale(s, s->qscale + s->dquant);
01532
01533 wrap_y = s->linesize;
01534 wrap_c = s->uvlinesize;
01535 ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
01536 ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
01537 ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
01538
01539 if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
01540 uint8_t *ebuf= s->edge_emu_buffer + 32;
01541 s->dsp.emulated_edge_mc(ebuf , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width , s->height);
01542 ptr_y= ebuf;
01543 s->dsp.emulated_edge_mc(ebuf+18*wrap_y , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
01544 ptr_cb= ebuf+18*wrap_y;
01545 s->dsp.emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
01546 ptr_cr= ebuf+18*wrap_y+8;
01547 }
01548
01549 if (s->mb_intra) {
01550 if(s->flags&CODEC_FLAG_INTERLACED_DCT){
01551 int progressive_score, interlaced_score;
01552
01553 s->interlaced_dct=0;
01554 progressive_score= s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y, 8)
01555 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
01556
01557 if(progressive_score > 0){
01558 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y*2, 8)
01559 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y , NULL, wrap_y*2, 8);
01560 if(progressive_score > interlaced_score){
01561 s->interlaced_dct=1;
01562
01563 dct_offset= wrap_y;
01564 wrap_y<<=1;
01565 if (s->chroma_format == CHROMA_422)
01566 wrap_c<<=1;
01567 }
01568 }
01569 }
01570
01571 s->dsp.get_pixels(s->block[0], ptr_y , wrap_y);
01572 s->dsp.get_pixels(s->block[1], ptr_y + 8, wrap_y);
01573 s->dsp.get_pixels(s->block[2], ptr_y + dct_offset , wrap_y);
01574 s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
01575
01576 if(s->flags&CODEC_FLAG_GRAY){
01577 skip_dct[4]= 1;
01578 skip_dct[5]= 1;
01579 }else{
01580 s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
01581 s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
01582 if(!s->chroma_y_shift){
01583 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
01584 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
01585 }
01586 }
01587 }else{
01588 op_pixels_func (*op_pix)[4];
01589 qpel_mc_func (*op_qpix)[16];
01590 uint8_t *dest_y, *dest_cb, *dest_cr;
01591
01592 dest_y = s->dest[0];
01593 dest_cb = s->dest[1];
01594 dest_cr = s->dest[2];
01595
01596 if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
01597 op_pix = s->dsp.put_pixels_tab;
01598 op_qpix= s->dsp.put_qpel_pixels_tab;
01599 }else{
01600 op_pix = s->dsp.put_no_rnd_pixels_tab;
01601 op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
01602 }
01603
01604 if (s->mv_dir & MV_DIR_FORWARD) {
01605 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
01606 op_pix = s->dsp.avg_pixels_tab;
01607 op_qpix= s->dsp.avg_qpel_pixels_tab;
01608 }
01609 if (s->mv_dir & MV_DIR_BACKWARD) {
01610 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
01611 }
01612
01613 if(s->flags&CODEC_FLAG_INTERLACED_DCT){
01614 int progressive_score, interlaced_score;
01615
01616 s->interlaced_dct=0;
01617 progressive_score= s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y, 8)
01618 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
01619
01620 if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
01621
01622 if(progressive_score>0){
01623 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y*2, 8)
01624 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y , ptr_y + wrap_y , wrap_y*2, 8);
01625
01626 if(progressive_score > interlaced_score){
01627 s->interlaced_dct=1;
01628
01629 dct_offset= wrap_y;
01630 wrap_y<<=1;
01631 if (s->chroma_format == CHROMA_422)
01632 wrap_c<<=1;
01633 }
01634 }
01635 }
01636
01637 s->dsp.diff_pixels(s->block[0], ptr_y , dest_y , wrap_y);
01638 s->dsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
01639 s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset , dest_y + dct_offset , wrap_y);
01640 s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
01641
01642 if(s->flags&CODEC_FLAG_GRAY){
01643 skip_dct[4]= 1;
01644 skip_dct[5]= 1;
01645 }else{
01646 s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
01647 s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
01648 if(!s->chroma_y_shift){
01649 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
01650 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
01651 }
01652 }
01653
01654 if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
01655
01656 if(s->dsp.sad[1](NULL, ptr_y , dest_y , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
01657 if(s->dsp.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
01658 if(s->dsp.sad[1](NULL, ptr_y +dct_offset , dest_y +dct_offset , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
01659 if(s->dsp.sad[1](NULL, ptr_y +dct_offset+ 8, dest_y +dct_offset+ 8, wrap_y, 8) < 20*s->qscale) skip_dct[3]= 1;
01660 if(s->dsp.sad[1](NULL, ptr_cb , dest_cb , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
01661 if(s->dsp.sad[1](NULL, ptr_cr , dest_cr , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
01662 if(!s->chroma_y_shift){
01663 if(s->dsp.sad[1](NULL, ptr_cb +(dct_offset>>1), dest_cb +(dct_offset>>1), wrap_c, 8) < 20*s->qscale) skip_dct[6]= 1;
01664 if(s->dsp.sad[1](NULL, ptr_cr +(dct_offset>>1), dest_cr +(dct_offset>>1), wrap_c, 8) < 20*s->qscale) skip_dct[7]= 1;
01665 }
01666 }
01667 }
01668
01669 if(s->avctx->quantizer_noise_shaping){
01670 if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y , wrap_y);
01671 if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y + 8, wrap_y);
01672 if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset , wrap_y);
01673 if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
01674 if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb , wrap_c);
01675 if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr , wrap_c);
01676 if(!s->chroma_y_shift){
01677 if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
01678 if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
01679 }
01680 memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
01681 }
01682
01683
01684 assert(s->out_format!=FMT_MJPEG || s->qscale==8);
01685 {
01686 for(i=0;i<mb_block_count;i++) {
01687 if(!skip_dct[i]){
01688 int overflow;
01689 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
01690
01691
01692
01693 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
01694 }else
01695 s->block_last_index[i]= -1;
01696 }
01697 if(s->avctx->quantizer_noise_shaping){
01698 for(i=0;i<mb_block_count;i++) {
01699 if(!skip_dct[i]){
01700 s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
01701 }
01702 }
01703 }
01704
01705 if(s->luma_elim_threshold && !s->mb_intra)
01706 for(i=0; i<4; i++)
01707 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
01708 if(s->chroma_elim_threshold && !s->mb_intra)
01709 for(i=4; i<mb_block_count; i++)
01710 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
01711
01712 if(s->flags & CODEC_FLAG_CBP_RD){
01713 for(i=0;i<mb_block_count;i++) {
01714 if(s->block_last_index[i] == -1)
01715 s->coded_score[i]= INT_MAX/256;
01716 }
01717 }
01718 }
01719
01720 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
01721 s->block_last_index[4]=
01722 s->block_last_index[5]= 0;
01723 s->block[4][0]=
01724 s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
01725 }
01726
01727
01728 if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
01729 for(i=0; i<mb_block_count; i++){
01730 int j;
01731 if(s->block_last_index[i]>0){
01732 for(j=63; j>0; j--){
01733 if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
01734 }
01735 s->block_last_index[i]= j;
01736 }
01737 }
01738 }
01739
01740
01741 switch(s->codec_id){
01742 case CODEC_ID_MPEG1VIDEO:
01743 case CODEC_ID_MPEG2VIDEO:
01744 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
01745 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
01746 break;
01747 case CODEC_ID_MPEG4:
01748 if (CONFIG_MPEG4_ENCODER)
01749 mpeg4_encode_mb(s, s->block, motion_x, motion_y);
01750 break;
01751 case CODEC_ID_MSMPEG4V2:
01752 case CODEC_ID_MSMPEG4V3:
01753 case CODEC_ID_WMV1:
01754 if (CONFIG_MSMPEG4_ENCODER)
01755 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
01756 break;
01757 case CODEC_ID_WMV2:
01758 if (CONFIG_WMV2_ENCODER)
01759 ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
01760 break;
01761 case CODEC_ID_H261:
01762 if (CONFIG_H261_ENCODER)
01763 ff_h261_encode_mb(s, s->block, motion_x, motion_y);
01764 break;
01765 case CODEC_ID_H263:
01766 case CODEC_ID_H263P:
01767 case CODEC_ID_FLV1:
01768 case CODEC_ID_RV10:
01769 case CODEC_ID_RV20:
01770 if (CONFIG_H263_ENCODER)
01771 h263_encode_mb(s, s->block, motion_x, motion_y);
01772 break;
01773 case CODEC_ID_MJPEG:
01774 if (CONFIG_MJPEG_ENCODER)
01775 ff_mjpeg_encode_mb(s, s->block);
01776 break;
01777 default:
01778 assert(0);
01779 }
01780 }
01781
01782 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
01783 {
01784 if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 6);
01785 else encode_mb_internal(s, motion_x, motion_y, 16, 8);
01786 }
01787
01788 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
01789 int i;
01790
01791 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int));
01792
01793
01794 d->mb_skip_run= s->mb_skip_run;
01795 for(i=0; i<3; i++)
01796 d->last_dc[i]= s->last_dc[i];
01797
01798
01799 d->mv_bits= s->mv_bits;
01800 d->i_tex_bits= s->i_tex_bits;
01801 d->p_tex_bits= s->p_tex_bits;
01802 d->i_count= s->i_count;
01803 d->f_count= s->f_count;
01804 d->b_count= s->b_count;
01805 d->skip_count= s->skip_count;
01806 d->misc_bits= s->misc_bits;
01807 d->last_bits= 0;
01808
01809 d->mb_skipped= 0;
01810 d->qscale= s->qscale;
01811 d->dquant= s->dquant;
01812
01813 d->esc3_level_length= s->esc3_level_length;
01814 }
01815
01816 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
01817 int i;
01818
01819 memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
01820 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int));
01821
01822
01823 d->mb_skip_run= s->mb_skip_run;
01824 for(i=0; i<3; i++)
01825 d->last_dc[i]= s->last_dc[i];
01826
01827
01828 d->mv_bits= s->mv_bits;
01829 d->i_tex_bits= s->i_tex_bits;
01830 d->p_tex_bits= s->p_tex_bits;
01831 d->i_count= s->i_count;
01832 d->f_count= s->f_count;
01833 d->b_count= s->b_count;
01834 d->skip_count= s->skip_count;
01835 d->misc_bits= s->misc_bits;
01836
01837 d->mb_intra= s->mb_intra;
01838 d->mb_skipped= s->mb_skipped;
01839 d->mv_type= s->mv_type;
01840 d->mv_dir= s->mv_dir;
01841 d->pb= s->pb;
01842 if(s->data_partitioning){
01843 d->pb2= s->pb2;
01844 d->tex_pb= s->tex_pb;
01845 }
01846 d->block= s->block;
01847 for(i=0; i<8; i++)
01848 d->block_last_index[i]= s->block_last_index[i];
01849 d->interlaced_dct= s->interlaced_dct;
01850 d->qscale= s->qscale;
01851
01852 d->esc3_level_length= s->esc3_level_length;
01853 }
01854
01855 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
01856 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
01857 int *dmin, int *next_block, int motion_x, int motion_y)
01858 {
01859 int score;
01860 uint8_t *dest_backup[3];
01861
01862 copy_context_before_encode(s, backup, type);
01863
01864 s->block= s->blocks[*next_block];
01865 s->pb= pb[*next_block];
01866 if(s->data_partitioning){
01867 s->pb2 = pb2 [*next_block];
01868 s->tex_pb= tex_pb[*next_block];
01869 }
01870
01871 if(*next_block){
01872 memcpy(dest_backup, s->dest, sizeof(s->dest));
01873 s->dest[0] = s->rd_scratchpad;
01874 s->dest[1] = s->rd_scratchpad + 16*s->linesize;
01875 s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
01876 assert(s->linesize >= 32);
01877 }
01878
01879 encode_mb(s, motion_x, motion_y);
01880
01881 score= put_bits_count(&s->pb);
01882 if(s->data_partitioning){
01883 score+= put_bits_count(&s->pb2);
01884 score+= put_bits_count(&s->tex_pb);
01885 }
01886
01887 if(s->avctx->mb_decision == FF_MB_DECISION_RD){
01888 MPV_decode_mb(s, s->block);
01889
01890 score *= s->lambda2;
01891 score += sse_mb(s) << FF_LAMBDA_SHIFT;
01892 }
01893
01894 if(*next_block){
01895 memcpy(s->dest, dest_backup, sizeof(s->dest));
01896 }
01897
01898 if(score<*dmin){
01899 *dmin= score;
01900 *next_block^=1;
01901
01902 copy_context_after_encode(best, s, type);
01903 }
01904 }
01905
01906 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
01907 uint32_t *sq = ff_squareTbl + 256;
01908 int acc=0;
01909 int x,y;
01910
01911 if(w==16 && h==16)
01912 return s->dsp.sse[0](NULL, src1, src2, stride, 16);
01913 else if(w==8 && h==8)
01914 return s->dsp.sse[1](NULL, src1, src2, stride, 8);
01915
01916 for(y=0; y<h; y++){
01917 for(x=0; x<w; x++){
01918 acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
01919 }
01920 }
01921
01922 assert(acc>=0);
01923
01924 return acc;
01925 }
01926
01927 static int sse_mb(MpegEncContext *s){
01928 int w= 16;
01929 int h= 16;
01930
01931 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
01932 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
01933
01934 if(w==16 && h==16)
01935 if(s->avctx->mb_cmp == FF_CMP_NSSE){
01936 return s->dsp.nsse[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
01937 +s->dsp.nsse[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
01938 +s->dsp.nsse[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
01939 }else{
01940 return s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
01941 +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
01942 +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
01943 }
01944 else
01945 return sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
01946 +sse(s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
01947 +sse(s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
01948 }
01949
01950 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
01951 MpegEncContext *s= *(void**)arg;
01952
01953
01954 s->me.pre_pass=1;
01955 s->me.dia_size= s->avctx->pre_dia_size;
01956 s->first_slice_line=1;
01957 for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
01958 for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
01959 ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
01960 }
01961 s->first_slice_line=0;
01962 }
01963
01964 s->me.pre_pass=0;
01965
01966 return 0;
01967 }
01968
01969 static int estimate_motion_thread(AVCodecContext *c, void *arg){
01970 MpegEncContext *s= *(void**)arg;
01971
01972 ff_check_alignment();
01973
01974 s->me.dia_size= s->avctx->dia_size;
01975 s->first_slice_line=1;
01976 for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
01977 s->mb_x=0;
01978 ff_init_block_index(s);
01979 for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
01980 s->block_index[0]+=2;
01981 s->block_index[1]+=2;
01982 s->block_index[2]+=2;
01983 s->block_index[3]+=2;
01984
01985
01986 if(s->pict_type==AV_PICTURE_TYPE_B)
01987 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
01988 else
01989 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
01990 }
01991 s->first_slice_line=0;
01992 }
01993 return 0;
01994 }
01995
01996 static int mb_var_thread(AVCodecContext *c, void *arg){
01997 MpegEncContext *s= *(void**)arg;
01998 int mb_x, mb_y;
01999
02000 ff_check_alignment();
02001
02002 for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
02003 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
02004 int xx = mb_x * 16;
02005 int yy = mb_y * 16;
02006 uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
02007 int varc;
02008 int sum = s->dsp.pix_sum(pix, s->linesize);
02009
02010 varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500 + 128)>>8;
02011
02012 s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
02013 s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
02014 s->me.mb_var_sum_temp += varc;
02015 }
02016 }
02017 return 0;
02018 }
02019
02020 static void write_slice_end(MpegEncContext *s){
02021 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
02022 if(s->partitioned_frame){
02023 ff_mpeg4_merge_partitions(s);
02024 }
02025
02026 ff_mpeg4_stuffing(&s->pb);
02027 }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
02028 ff_mjpeg_encode_stuffing(&s->pb);
02029 }
02030
02031 align_put_bits(&s->pb);
02032 flush_put_bits(&s->pb);
02033
02034 if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
02035 s->misc_bits+= get_bits_diff(s);
02036 }
02037
02038 static int encode_thread(AVCodecContext *c, void *arg){
02039 MpegEncContext *s= *(void**)arg;
02040 int mb_x, mb_y, pdif = 0;
02041 int chr_h= 16>>s->chroma_y_shift;
02042 int i, j;
02043 MpegEncContext best_s, backup_s;
02044 uint8_t bit_buf[2][MAX_MB_BYTES];
02045 uint8_t bit_buf2[2][MAX_MB_BYTES];
02046 uint8_t bit_buf_tex[2][MAX_MB_BYTES];
02047 PutBitContext pb[2], pb2[2], tex_pb[2];
02048
02049
02050 ff_check_alignment();
02051
02052 for(i=0; i<2; i++){
02053 init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
02054 init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
02055 init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
02056 }
02057
02058 s->last_bits= put_bits_count(&s->pb);
02059 s->mv_bits=0;
02060 s->misc_bits=0;
02061 s->i_tex_bits=0;
02062 s->p_tex_bits=0;
02063 s->i_count=0;
02064 s->f_count=0;
02065 s->b_count=0;
02066 s->skip_count=0;
02067
02068 for(i=0; i<3; i++){
02069
02070
02071 s->last_dc[i] = 128 << s->intra_dc_precision;
02072
02073 s->current_picture.error[i] = 0;
02074 }
02075 s->mb_skip_run = 0;
02076 memset(s->last_mv, 0, sizeof(s->last_mv));
02077
02078 s->last_mv_dir = 0;
02079
02080 switch(s->codec_id){
02081 case CODEC_ID_H263:
02082 case CODEC_ID_H263P:
02083 case CODEC_ID_FLV1:
02084 if (CONFIG_H263_ENCODER)
02085 s->gob_index = ff_h263_get_gob_height(s);
02086 break;
02087 case CODEC_ID_MPEG4:
02088 if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
02089 ff_mpeg4_init_partitions(s);
02090 break;
02091 }
02092
02093 s->resync_mb_x=0;
02094 s->resync_mb_y=0;
02095 s->first_slice_line = 1;
02096 s->ptr_lastgob = s->pb.buf;
02097 for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
02098
02099 s->mb_x=0;
02100 s->mb_y= mb_y;
02101
02102 ff_set_qscale(s, s->qscale);
02103 ff_init_block_index(s);
02104
02105 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
02106 int xy= mb_y*s->mb_stride + mb_x;
02107 int mb_type= s->mb_type[xy];
02108
02109 int dmin= INT_MAX;
02110 int dir;
02111
02112 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
02113 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
02114 return -1;
02115 }
02116 if(s->data_partitioning){
02117 if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES
02118 || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
02119 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
02120 return -1;
02121 }
02122 }
02123
02124 s->mb_x = mb_x;
02125 s->mb_y = mb_y;
02126 ff_update_block_index(s);
02127
02128 if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
02129 ff_h261_reorder_mb_index(s);
02130 xy= s->mb_y*s->mb_stride + s->mb_x;
02131 mb_type= s->mb_type[xy];
02132 }
02133
02134
02135 if(s->rtp_mode){
02136 int current_packet_size, is_gob_start;
02137
02138 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
02139
02140 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
02141
02142 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
02143
02144 switch(s->codec_id){
02145 case CODEC_ID_H263:
02146 case CODEC_ID_H263P:
02147 if(!s->h263_slice_structured)
02148 if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
02149 break;
02150 case CODEC_ID_MPEG2VIDEO:
02151 if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
02152 case CODEC_ID_MPEG1VIDEO:
02153 if(s->mb_skip_run) is_gob_start=0;
02154 break;
02155 }
02156
02157 if(is_gob_start){
02158 if(s->start_mb_y != mb_y || mb_x!=0){
02159 write_slice_end(s);
02160
02161 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
02162 ff_mpeg4_init_partitions(s);
02163 }
02164 }
02165
02166 assert((put_bits_count(&s->pb)&7) == 0);
02167 current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
02168
02169 if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
02170 int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
02171 int d= 100 / s->avctx->error_rate;
02172 if(r % d == 0){
02173 current_packet_size=0;
02174 #ifndef ALT_BITSTREAM_WRITER
02175 s->pb.buf_ptr= s->ptr_lastgob;
02176 #endif
02177 assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
02178 }
02179 }
02180
02181 if (s->avctx->rtp_callback){
02182 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
02183 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
02184 }
02185
02186 switch(s->codec_id){
02187 case CODEC_ID_MPEG4:
02188 if (CONFIG_MPEG4_ENCODER) {
02189 ff_mpeg4_encode_video_packet_header(s);
02190 ff_mpeg4_clean_buffers(s);
02191 }
02192 break;
02193 case CODEC_ID_MPEG1VIDEO:
02194 case CODEC_ID_MPEG2VIDEO:
02195 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
02196 ff_mpeg1_encode_slice_header(s);
02197 ff_mpeg1_clean_buffers(s);
02198 }
02199 break;
02200 case CODEC_ID_H263:
02201 case CODEC_ID_H263P:
02202 if (CONFIG_H263_ENCODER)
02203 h263_encode_gob_header(s, mb_y);
02204 break;
02205 }
02206
02207 if(s->flags&CODEC_FLAG_PASS1){
02208 int bits= put_bits_count(&s->pb);
02209 s->misc_bits+= bits - s->last_bits;
02210 s->last_bits= bits;
02211 }
02212
02213 s->ptr_lastgob += current_packet_size;
02214 s->first_slice_line=1;
02215 s->resync_mb_x=mb_x;
02216 s->resync_mb_y=mb_y;
02217 }
02218 }
02219
02220 if( (s->resync_mb_x == s->mb_x)
02221 && s->resync_mb_y+1 == s->mb_y){
02222 s->first_slice_line=0;
02223 }
02224
02225 s->mb_skipped=0;
02226 s->dquant=0;
02227
02228 if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){
02229 int next_block=0;
02230 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
02231
02232 copy_context_before_encode(&backup_s, s, -1);
02233 backup_s.pb= s->pb;
02234 best_s.data_partitioning= s->data_partitioning;
02235 best_s.partitioned_frame= s->partitioned_frame;
02236 if(s->data_partitioning){
02237 backup_s.pb2= s->pb2;
02238 backup_s.tex_pb= s->tex_pb;
02239 }
02240
02241 if(mb_type&CANDIDATE_MB_TYPE_INTER){
02242 s->mv_dir = MV_DIR_FORWARD;
02243 s->mv_type = MV_TYPE_16X16;
02244 s->mb_intra= 0;
02245 s->mv[0][0][0] = s->p_mv_table[xy][0];
02246 s->mv[0][0][1] = s->p_mv_table[xy][1];
02247 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
02248 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
02249 }
02250 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
02251 s->mv_dir = MV_DIR_FORWARD;
02252 s->mv_type = MV_TYPE_FIELD;
02253 s->mb_intra= 0;
02254 for(i=0; i<2; i++){
02255 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
02256 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
02257 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
02258 }
02259 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
02260 &dmin, &next_block, 0, 0);
02261 }
02262 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
02263 s->mv_dir = MV_DIR_FORWARD;
02264 s->mv_type = MV_TYPE_16X16;
02265 s->mb_intra= 0;
02266 s->mv[0][0][0] = 0;
02267 s->mv[0][0][1] = 0;
02268 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
02269 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
02270 }
02271 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
02272 s->mv_dir = MV_DIR_FORWARD;
02273 s->mv_type = MV_TYPE_8X8;
02274 s->mb_intra= 0;
02275 for(i=0; i<4; i++){
02276 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
02277 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
02278 }
02279 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
02280 &dmin, &next_block, 0, 0);
02281 }
02282 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
02283 s->mv_dir = MV_DIR_FORWARD;
02284 s->mv_type = MV_TYPE_16X16;
02285 s->mb_intra= 0;
02286 s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
02287 s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
02288 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
02289 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
02290 }
02291 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
02292 s->mv_dir = MV_DIR_BACKWARD;
02293 s->mv_type = MV_TYPE_16X16;
02294 s->mb_intra= 0;
02295 s->mv[1][0][0] = s->b_back_mv_table[xy][0];
02296 s->mv[1][0][1] = s->b_back_mv_table[xy][1];
02297 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
02298 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
02299 }
02300 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
02301 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
02302 s->mv_type = MV_TYPE_16X16;
02303 s->mb_intra= 0;
02304 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
02305 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
02306 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
02307 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
02308 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
02309 &dmin, &next_block, 0, 0);
02310 }
02311 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
02312 s->mv_dir = MV_DIR_FORWARD;
02313 s->mv_type = MV_TYPE_FIELD;
02314 s->mb_intra= 0;
02315 for(i=0; i<2; i++){
02316 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
02317 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
02318 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
02319 }
02320 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
02321 &dmin, &next_block, 0, 0);
02322 }
02323 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
02324 s->mv_dir = MV_DIR_BACKWARD;
02325 s->mv_type = MV_TYPE_FIELD;
02326 s->mb_intra= 0;
02327 for(i=0; i<2; i++){
02328 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
02329 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
02330 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
02331 }
02332 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
02333 &dmin, &next_block, 0, 0);
02334 }
02335 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
02336 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
02337 s->mv_type = MV_TYPE_FIELD;
02338 s->mb_intra= 0;
02339 for(dir=0; dir<2; dir++){
02340 for(i=0; i<2; i++){
02341 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
02342 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
02343 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
02344 }
02345 }
02346 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
02347 &dmin, &next_block, 0, 0);
02348 }
02349 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
02350 s->mv_dir = 0;
02351 s->mv_type = MV_TYPE_16X16;
02352 s->mb_intra= 1;
02353 s->mv[0][0][0] = 0;
02354 s->mv[0][0][1] = 0;
02355 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
02356 &dmin, &next_block, 0, 0);
02357 if(s->h263_pred || s->h263_aic){
02358 if(best_s.mb_intra)
02359 s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
02360 else
02361 ff_clean_intra_table_entries(s);
02362 }
02363 }
02364
02365 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
02366 if(best_s.mv_type==MV_TYPE_16X16){
02367 const int last_qp= backup_s.qscale;
02368 int qpi, qp, dc[6];
02369 DCTELEM ac[6][16];
02370 const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
02371 static const int dquant_tab[4]={-1,1,-2,2};
02372
02373 assert(backup_s.dquant == 0);
02374
02375
02376 s->mv_dir= best_s.mv_dir;
02377 s->mv_type = MV_TYPE_16X16;
02378 s->mb_intra= best_s.mb_intra;
02379 s->mv[0][0][0] = best_s.mv[0][0][0];
02380 s->mv[0][0][1] = best_s.mv[0][0][1];
02381 s->mv[1][0][0] = best_s.mv[1][0][0];
02382 s->mv[1][0][1] = best_s.mv[1][0][1];
02383
02384 qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
02385 for(; qpi<4; qpi++){
02386 int dquant= dquant_tab[qpi];
02387 qp= last_qp + dquant;
02388 if(qp < s->avctx->qmin || qp > s->avctx->qmax)
02389 continue;
02390 backup_s.dquant= dquant;
02391 if(s->mb_intra && s->dc_val[0]){
02392 for(i=0; i<6; i++){
02393 dc[i]= s->dc_val[0][ s->block_index[i] ];
02394 memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
02395 }
02396 }
02397
02398 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER , pb, pb2, tex_pb,
02399 &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
02400 if(best_s.qscale != qp){
02401 if(s->mb_intra && s->dc_val[0]){
02402 for(i=0; i<6; i++){
02403 s->dc_val[0][ s->block_index[i] ]= dc[i];
02404 memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
02405 }
02406 }
02407 }
02408 }
02409 }
02410 }
02411 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
02412 int mx= s->b_direct_mv_table[xy][0];
02413 int my= s->b_direct_mv_table[xy][1];
02414
02415 backup_s.dquant = 0;
02416 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
02417 s->mb_intra= 0;
02418 ff_mpeg4_set_direct_mv(s, mx, my);
02419 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
02420 &dmin, &next_block, mx, my);
02421 }
02422 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
02423 backup_s.dquant = 0;
02424 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
02425 s->mb_intra= 0;
02426 ff_mpeg4_set_direct_mv(s, 0, 0);
02427 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
02428 &dmin, &next_block, 0, 0);
02429 }
02430 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
02431 int coded=0;
02432 for(i=0; i<6; i++)
02433 coded |= s->block_last_index[i];
02434 if(coded){
02435 int mx,my;
02436 memcpy(s->mv, best_s.mv, sizeof(s->mv));
02437 if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
02438 mx=my=0;
02439 ff_mpeg4_set_direct_mv(s, mx, my);
02440 }else if(best_s.mv_dir&MV_DIR_BACKWARD){
02441 mx= s->mv[1][0][0];
02442 my= s->mv[1][0][1];
02443 }else{
02444 mx= s->mv[0][0][0];
02445 my= s->mv[0][0][1];
02446 }
02447
02448 s->mv_dir= best_s.mv_dir;
02449 s->mv_type = best_s.mv_type;
02450 s->mb_intra= 0;
02451
02452
02453
02454
02455 backup_s.dquant= 0;
02456 s->skipdct=1;
02457 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER , pb, pb2, tex_pb,
02458 &dmin, &next_block, mx, my);
02459 s->skipdct=0;
02460 }
02461 }
02462
02463 s->current_picture.qscale_table[xy]= best_s.qscale;
02464
02465 copy_context_after_encode(s, &best_s, -1);
02466
02467 pb_bits_count= put_bits_count(&s->pb);
02468 flush_put_bits(&s->pb);
02469 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
02470 s->pb= backup_s.pb;
02471
02472 if(s->data_partitioning){
02473 pb2_bits_count= put_bits_count(&s->pb2);
02474 flush_put_bits(&s->pb2);
02475 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
02476 s->pb2= backup_s.pb2;
02477
02478 tex_pb_bits_count= put_bits_count(&s->tex_pb);
02479 flush_put_bits(&s->tex_pb);
02480 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
02481 s->tex_pb= backup_s.tex_pb;
02482 }
02483 s->last_bits= put_bits_count(&s->pb);
02484
02485 if (CONFIG_H263_ENCODER &&
02486 s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
02487 ff_h263_update_motion_val(s);
02488
02489 if(next_block==0){
02490 s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad , s->linesize ,16);
02491 s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
02492 s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
02493 }
02494
02495 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
02496 MPV_decode_mb(s, s->block);
02497 } else {
02498 int motion_x = 0, motion_y = 0;
02499 s->mv_type=MV_TYPE_16X16;
02500
02501
02502 switch(mb_type){
02503 case CANDIDATE_MB_TYPE_INTRA:
02504 s->mv_dir = 0;
02505 s->mb_intra= 1;
02506 motion_x= s->mv[0][0][0] = 0;
02507 motion_y= s->mv[0][0][1] = 0;
02508 break;
02509 case CANDIDATE_MB_TYPE_INTER:
02510 s->mv_dir = MV_DIR_FORWARD;
02511 s->mb_intra= 0;
02512 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
02513 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
02514 break;
02515 case CANDIDATE_MB_TYPE_INTER_I:
02516 s->mv_dir = MV_DIR_FORWARD;
02517 s->mv_type = MV_TYPE_FIELD;
02518 s->mb_intra= 0;
02519 for(i=0; i<2; i++){
02520 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
02521 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
02522 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
02523 }
02524 break;
02525 case CANDIDATE_MB_TYPE_INTER4V:
02526 s->mv_dir = MV_DIR_FORWARD;
02527 s->mv_type = MV_TYPE_8X8;
02528 s->mb_intra= 0;
02529 for(i=0; i<4; i++){
02530 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
02531 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
02532 }
02533 break;
02534 case CANDIDATE_MB_TYPE_DIRECT:
02535 if (CONFIG_MPEG4_ENCODER) {
02536 s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
02537 s->mb_intra= 0;
02538 motion_x=s->b_direct_mv_table[xy][0];
02539 motion_y=s->b_direct_mv_table[xy][1];
02540 ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
02541 }
02542 break;
02543 case CANDIDATE_MB_TYPE_DIRECT0:
02544 if (CONFIG_MPEG4_ENCODER) {
02545 s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
02546 s->mb_intra= 0;
02547 ff_mpeg4_set_direct_mv(s, 0, 0);
02548 }
02549 break;
02550 case CANDIDATE_MB_TYPE_BIDIR:
02551 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
02552 s->mb_intra= 0;
02553 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
02554 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
02555 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
02556 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
02557 break;
02558 case CANDIDATE_MB_TYPE_BACKWARD:
02559 s->mv_dir = MV_DIR_BACKWARD;
02560 s->mb_intra= 0;
02561 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
02562 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
02563 break;
02564 case CANDIDATE_MB_TYPE_FORWARD:
02565 s->mv_dir = MV_DIR_FORWARD;
02566 s->mb_intra= 0;
02567 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
02568 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
02569
02570 break;
02571 case CANDIDATE_MB_TYPE_FORWARD_I:
02572 s->mv_dir = MV_DIR_FORWARD;
02573 s->mv_type = MV_TYPE_FIELD;
02574 s->mb_intra= 0;
02575 for(i=0; i<2; i++){
02576 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
02577 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
02578 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
02579 }
02580 break;
02581 case CANDIDATE_MB_TYPE_BACKWARD_I:
02582 s->mv_dir = MV_DIR_BACKWARD;
02583 s->mv_type = MV_TYPE_FIELD;
02584 s->mb_intra= 0;
02585 for(i=0; i<2; i++){
02586 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
02587 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
02588 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
02589 }
02590 break;
02591 case CANDIDATE_MB_TYPE_BIDIR_I:
02592 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
02593 s->mv_type = MV_TYPE_FIELD;
02594 s->mb_intra= 0;
02595 for(dir=0; dir<2; dir++){
02596 for(i=0; i<2; i++){
02597 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
02598 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
02599 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
02600 }
02601 }
02602 break;
02603 default:
02604 av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
02605 }
02606
02607 encode_mb(s, motion_x, motion_y);
02608
02609
02610 s->last_mv_dir = s->mv_dir;
02611
02612 if (CONFIG_H263_ENCODER &&
02613 s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
02614 ff_h263_update_motion_val(s);
02615
02616 MPV_decode_mb(s, s->block);
02617 }
02618
02619
02620 if(s->mb_intra ){
02621 s->p_mv_table[xy][0]=0;
02622 s->p_mv_table[xy][1]=0;
02623 }
02624
02625 if(s->flags&CODEC_FLAG_PSNR){
02626 int w= 16;
02627 int h= 16;
02628
02629 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
02630 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
02631
02632 s->current_picture.error[0] += sse(
02633 s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
02634 s->dest[0], w, h, s->linesize);
02635 s->current_picture.error[1] += sse(
02636 s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
02637 s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
02638 s->current_picture.error[2] += sse(
02639 s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
02640 s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
02641 }
02642 if(s->loop_filter){
02643 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
02644 ff_h263_loop_filter(s);
02645 }
02646
02647 }
02648 }
02649
02650
02651 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
02652 msmpeg4_encode_ext_header(s);
02653
02654 write_slice_end(s);
02655
02656
02657 if (s->avctx->rtp_callback) {
02658 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
02659 pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
02660
02661 emms_c();
02662 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
02663 }
02664
02665 return 0;
02666 }
02667
02668 #define MERGE(field) dst->field += src->field; src->field=0
02669 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
02670 MERGE(me.scene_change_score);
02671 MERGE(me.mc_mb_var_sum_temp);
02672 MERGE(me.mb_var_sum_temp);
02673 }
02674
02675 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
02676 int i;
02677
02678 MERGE(dct_count[0]);
02679 MERGE(dct_count[1]);
02680 MERGE(mv_bits);
02681 MERGE(i_tex_bits);
02682 MERGE(p_tex_bits);
02683 MERGE(i_count);
02684 MERGE(f_count);
02685 MERGE(b_count);
02686 MERGE(skip_count);
02687 MERGE(misc_bits);
02688 MERGE(error_count);
02689 MERGE(padding_bug_score);
02690 MERGE(current_picture.error[0]);
02691 MERGE(current_picture.error[1]);
02692 MERGE(current_picture.error[2]);
02693
02694 if(dst->avctx->noise_reduction){
02695 for(i=0; i<64; i++){
02696 MERGE(dct_error_sum[0][i]);
02697 MERGE(dct_error_sum[1][i]);
02698 }
02699 }
02700
02701 assert(put_bits_count(&src->pb) % 8 ==0);
02702 assert(put_bits_count(&dst->pb) % 8 ==0);
02703 ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
02704 flush_put_bits(&dst->pb);
02705 }
02706
02707 static int estimate_qp(MpegEncContext *s, int dry_run){
02708 if (s->next_lambda){
02709 s->current_picture_ptr->quality=
02710 s->current_picture.quality = s->next_lambda;
02711 if(!dry_run) s->next_lambda= 0;
02712 } else if (!s->fixed_qscale) {
02713 s->current_picture_ptr->quality=
02714 s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
02715 if (s->current_picture.quality < 0)
02716 return -1;
02717 }
02718
02719 if(s->adaptive_quant){
02720 switch(s->codec_id){
02721 case CODEC_ID_MPEG4:
02722 if (CONFIG_MPEG4_ENCODER)
02723 ff_clean_mpeg4_qscales(s);
02724 break;
02725 case CODEC_ID_H263:
02726 case CODEC_ID_H263P:
02727 case CODEC_ID_FLV1:
02728 if (CONFIG_H263_ENCODER)
02729 ff_clean_h263_qscales(s);
02730 break;
02731 default:
02732 ff_init_qscale_tab(s);
02733 }
02734
02735 s->lambda= s->lambda_table[0];
02736
02737 }else
02738 s->lambda= s->current_picture.quality;
02739
02740 update_qscale(s);
02741 return 0;
02742 }
02743
02744
02745 static void set_frame_distances(MpegEncContext * s){
02746 assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
02747 s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
02748
02749 if(s->pict_type==AV_PICTURE_TYPE_B){
02750 s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
02751 assert(s->pb_time > 0 && s->pb_time < s->pp_time);
02752 }else{
02753 s->pp_time= s->time - s->last_non_b_time;
02754 s->last_non_b_time= s->time;
02755 assert(s->picture_number==0 || s->pp_time > 0);
02756 }
02757 }
02758
02759 static int encode_picture(MpegEncContext *s, int picture_number)
02760 {
02761 int i;
02762 int bits;
02763 int context_count = s->avctx->thread_count;
02764
02765 s->picture_number = picture_number;
02766
02767
02768 s->me.mb_var_sum_temp =
02769 s->me.mc_mb_var_sum_temp = 0;
02770
02771
02772
02773 if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
02774 set_frame_distances(s);
02775 if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
02776 ff_set_mpeg4_time(s);
02777
02778 s->me.scene_change_score=0;
02779
02780
02781
02782 if(s->pict_type==AV_PICTURE_TYPE_I){
02783 if(s->msmpeg4_version >= 3) s->no_rounding=1;
02784 else s->no_rounding=0;
02785 }else if(s->pict_type!=AV_PICTURE_TYPE_B){
02786 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
02787 s->no_rounding ^= 1;
02788 }
02789
02790 if(s->flags & CODEC_FLAG_PASS2){
02791 if (estimate_qp(s,1) < 0)
02792 return -1;
02793 ff_get_2pass_fcode(s);
02794 }else if(!(s->flags & CODEC_FLAG_QSCALE)){
02795 if(s->pict_type==AV_PICTURE_TYPE_B)
02796 s->lambda= s->last_lambda_for[s->pict_type];
02797 else
02798 s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
02799 update_qscale(s);
02800 }
02801
02802 s->mb_intra=0;
02803 for(i=1; i<context_count; i++){
02804 ff_update_duplicate_context(s->thread_context[i], s);
02805 }
02806
02807 if(ff_init_me(s)<0)
02808 return -1;
02809
02810
02811 if(s->pict_type != AV_PICTURE_TYPE_I){
02812 s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
02813 s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
02814 if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
02815 if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
02816 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
02817 }
02818 }
02819
02820 s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
02821 }else {
02822
02823 for(i=0; i<s->mb_stride*s->mb_height; i++)
02824 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
02825
02826 if(!s->fixed_qscale){
02827
02828 s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
02829 }
02830 }
02831 for(i=1; i<context_count; i++){
02832 merge_context_after_me(s, s->thread_context[i]);
02833 }
02834 s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
02835 s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
02836 emms_c();
02837
02838 if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == AV_PICTURE_TYPE_P){
02839 s->pict_type= AV_PICTURE_TYPE_I;
02840 for(i=0; i<s->mb_stride*s->mb_height; i++)
02841 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
02842
02843 }
02844
02845 if(!s->umvplus){
02846 if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
02847 s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
02848
02849 if(s->flags & CODEC_FLAG_INTERLACED_ME){
02850 int a,b;
02851 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I);
02852 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
02853 s->f_code= FFMAX3(s->f_code, a, b);
02854 }
02855
02856 ff_fix_long_p_mvs(s);
02857 ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
02858 if(s->flags & CODEC_FLAG_INTERLACED_ME){
02859 int j;
02860 for(i=0; i<2; i++){
02861 for(j=0; j<2; j++)
02862 ff_fix_long_mvs(s, s->p_field_select_table[i], j,
02863 s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
02864 }
02865 }
02866 }
02867
02868 if(s->pict_type==AV_PICTURE_TYPE_B){
02869 int a, b;
02870
02871 a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
02872 b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
02873 s->f_code = FFMAX(a, b);
02874
02875 a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
02876 b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
02877 s->b_code = FFMAX(a, b);
02878
02879 ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
02880 ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
02881 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
02882 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
02883 if(s->flags & CODEC_FLAG_INTERLACED_ME){
02884 int dir, j;
02885 for(dir=0; dir<2; dir++){
02886 for(i=0; i<2; i++){
02887 for(j=0; j<2; j++){
02888 int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
02889 : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
02890 ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
02891 s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
02892 }
02893 }
02894 }
02895 }
02896 }
02897 }
02898
02899 if (estimate_qp(s, 0) < 0)
02900 return -1;
02901
02902 if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==AV_PICTURE_TYPE_I && !(s->flags & CODEC_FLAG_QSCALE))
02903 s->qscale= 3;
02904
02905 if (s->out_format == FMT_MJPEG) {
02906
02907 for(i=1;i<64;i++){
02908 int j= s->dsp.idct_permutation[i];
02909
02910 s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
02911 }
02912 s->y_dc_scale_table=
02913 s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
02914 s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
02915 ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
02916 s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
02917 s->qscale= 8;
02918 }
02919
02920
02921 s->current_picture_ptr->key_frame=
02922 s->current_picture.key_frame= s->pict_type == AV_PICTURE_TYPE_I;
02923 s->current_picture_ptr->pict_type=
02924 s->current_picture.pict_type= s->pict_type;
02925
02926 if(s->current_picture.key_frame)
02927 s->picture_in_gop_number=0;
02928
02929 s->last_bits= put_bits_count(&s->pb);
02930 switch(s->out_format) {
02931 case FMT_MJPEG:
02932 if (CONFIG_MJPEG_ENCODER)
02933 ff_mjpeg_encode_picture_header(s);
02934 break;
02935 case FMT_H261:
02936 if (CONFIG_H261_ENCODER)
02937 ff_h261_encode_picture_header(s, picture_number);
02938 break;
02939 case FMT_H263:
02940 if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
02941 ff_wmv2_encode_picture_header(s, picture_number);
02942 else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
02943 msmpeg4_encode_picture_header(s, picture_number);
02944 else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
02945 mpeg4_encode_picture_header(s, picture_number);
02946 else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
02947 rv10_encode_picture_header(s, picture_number);
02948 else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
02949 rv20_encode_picture_header(s, picture_number);
02950 else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
02951 ff_flv_encode_picture_header(s, picture_number);
02952 else if (CONFIG_H263_ENCODER)
02953 h263_encode_picture_header(s, picture_number);
02954 break;
02955 case FMT_MPEG1:
02956 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
02957 mpeg1_encode_picture_header(s, picture_number);
02958 break;
02959 case FMT_H264:
02960 break;
02961 default:
02962 assert(0);
02963 }
02964 bits= put_bits_count(&s->pb);
02965 s->header_bits= bits - s->last_bits;
02966
02967 for(i=1; i<context_count; i++){
02968 update_duplicate_context_after_me(s->thread_context[i], s);
02969 }
02970 s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
02971 for(i=1; i<context_count; i++){
02972 merge_context_after_encode(s, s->thread_context[i]);
02973 }
02974 emms_c();
02975 return 0;
02976 }
02977
02978 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
02979 const int intra= s->mb_intra;
02980 int i;
02981
02982 s->dct_count[intra]++;
02983
02984 for(i=0; i<64; i++){
02985 int level= block[i];
02986
02987 if(level){
02988 if(level>0){
02989 s->dct_error_sum[intra][i] += level;
02990 level -= s->dct_offset[intra][i];
02991 if(level<0) level=0;
02992 }else{
02993 s->dct_error_sum[intra][i] -= level;
02994 level += s->dct_offset[intra][i];
02995 if(level>0) level=0;
02996 }
02997 block[i]= level;
02998 }
02999 }
03000 }
03001
03002 static int dct_quantize_trellis_c(MpegEncContext *s,
03003 DCTELEM *block, int n,
03004 int qscale, int *overflow){
03005 const int *qmat;
03006 const uint8_t *scantable= s->intra_scantable.scantable;
03007 const uint8_t *perm_scantable= s->intra_scantable.permutated;
03008 int max=0;
03009 unsigned int threshold1, threshold2;
03010 int bias=0;
03011 int run_tab[65];
03012 int level_tab[65];
03013 int score_tab[65];
03014 int survivor[65];
03015 int survivor_count;
03016 int last_run=0;
03017 int last_level=0;
03018 int last_score= 0;
03019 int last_i;
03020 int coeff[2][64];
03021 int coeff_count[64];
03022 int qmul, qadd, start_i, last_non_zero, i, dc;
03023 const int esc_length= s->ac_esc_length;
03024 uint8_t * length;
03025 uint8_t * last_length;
03026 const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
03027
03028 s->dsp.fdct (block);
03029
03030 if(s->dct_error_sum)
03031 s->denoise_dct(s, block);
03032 qmul= qscale*16;
03033 qadd= ((qscale-1)|1)*8;
03034
03035 if (s->mb_intra) {
03036 int q;
03037 if (!s->h263_aic) {
03038 if (n < 4)
03039 q = s->y_dc_scale;
03040 else
03041 q = s->c_dc_scale;
03042 q = q << 3;
03043 } else{
03044
03045 q = 1 << 3;
03046 qadd=0;
03047 }
03048
03049
03050 block[0] = (block[0] + (q >> 1)) / q;
03051 start_i = 1;
03052 last_non_zero = 0;
03053 qmat = s->q_intra_matrix[qscale];
03054 if(s->mpeg_quant || s->out_format == FMT_MPEG1)
03055 bias= 1<<(QMAT_SHIFT-1);
03056 length = s->intra_ac_vlc_length;
03057 last_length= s->intra_ac_vlc_last_length;
03058 } else {
03059 start_i = 0;
03060 last_non_zero = -1;
03061 qmat = s->q_inter_matrix[qscale];
03062 length = s->inter_ac_vlc_length;
03063 last_length= s->inter_ac_vlc_last_length;
03064 }
03065 last_i= start_i;
03066
03067 threshold1= (1<<QMAT_SHIFT) - bias - 1;
03068 threshold2= (threshold1<<1);
03069
03070 for(i=63; i>=start_i; i--) {
03071 const int j = scantable[i];
03072 int level = block[j] * qmat[j];
03073
03074 if(((unsigned)(level+threshold1))>threshold2){
03075 last_non_zero = i;
03076 break;
03077 }
03078 }
03079
03080 for(i=start_i; i<=last_non_zero; i++) {
03081 const int j = scantable[i];
03082 int level = block[j] * qmat[j];
03083
03084
03085
03086 if(((unsigned)(level+threshold1))>threshold2){
03087 if(level>0){
03088 level= (bias + level)>>QMAT_SHIFT;
03089 coeff[0][i]= level;
03090 coeff[1][i]= level-1;
03091
03092 }else{
03093 level= (bias - level)>>QMAT_SHIFT;
03094 coeff[0][i]= -level;
03095 coeff[1][i]= -level+1;
03096
03097 }
03098 coeff_count[i]= FFMIN(level, 2);
03099 assert(coeff_count[i]);
03100 max |=level;
03101 }else{
03102 coeff[0][i]= (level>>31)|1;
03103 coeff_count[i]= 1;
03104 }
03105 }
03106
03107 *overflow= s->max_qcoeff < max;
03108
03109 if(last_non_zero < start_i){
03110 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
03111 return last_non_zero;
03112 }
03113
03114 score_tab[start_i]= 0;
03115 survivor[0]= start_i;
03116 survivor_count= 1;
03117
03118 for(i=start_i; i<=last_non_zero; i++){
03119 int level_index, j, zero_distortion;
03120 int dct_coeff= FFABS(block[ scantable[i] ]);
03121 int best_score=256*256*256*120;
03122
03123 if ( s->dsp.fdct == fdct_ifast
03124 #ifndef FAAN_POSTSCALE
03125 || s->dsp.fdct == ff_faandct
03126 #endif
03127 )
03128 dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
03129 zero_distortion= dct_coeff*dct_coeff;
03130
03131 for(level_index=0; level_index < coeff_count[i]; level_index++){
03132 int distortion;
03133 int level= coeff[level_index][i];
03134 const int alevel= FFABS(level);
03135 int unquant_coeff;
03136
03137 assert(level);
03138
03139 if(s->out_format == FMT_H263){
03140 unquant_coeff= alevel*qmul + qadd;
03141 }else{
03142 j= s->dsp.idct_permutation[ scantable[i] ];
03143 if(s->mb_intra){
03144 unquant_coeff = (int)( alevel * qscale * s->intra_matrix[j]) >> 3;
03145 unquant_coeff = (unquant_coeff - 1) | 1;
03146 }else{
03147 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
03148 unquant_coeff = (unquant_coeff - 1) | 1;
03149 }
03150 unquant_coeff<<= 3;
03151 }
03152
03153 distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
03154 level+=64;
03155 if((level&(~127)) == 0){
03156 for(j=survivor_count-1; j>=0; j--){
03157 int run= i - survivor[j];
03158 int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
03159 score += score_tab[i-run];
03160
03161 if(score < best_score){
03162 best_score= score;
03163 run_tab[i+1]= run;
03164 level_tab[i+1]= level-64;
03165 }
03166 }
03167
03168 if(s->out_format == FMT_H263){
03169 for(j=survivor_count-1; j>=0; j--){
03170 int run= i - survivor[j];
03171 int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
03172 score += score_tab[i-run];
03173 if(score < last_score){
03174 last_score= score;
03175 last_run= run;
03176 last_level= level-64;
03177 last_i= i+1;
03178 }
03179 }
03180 }
03181 }else{
03182 distortion += esc_length*lambda;
03183 for(j=survivor_count-1; j>=0; j--){
03184 int run= i - survivor[j];
03185 int score= distortion + score_tab[i-run];
03186
03187 if(score < best_score){
03188 best_score= score;
03189 run_tab[i+1]= run;
03190 level_tab[i+1]= level-64;
03191 }
03192 }
03193
03194 if(s->out_format == FMT_H263){
03195 for(j=survivor_count-1; j>=0; j--){
03196 int run= i - survivor[j];
03197 int score= distortion + score_tab[i-run];
03198 if(score < last_score){
03199 last_score= score;
03200 last_run= run;
03201 last_level= level-64;
03202 last_i= i+1;
03203 }
03204 }
03205 }
03206 }
03207 }
03208
03209 score_tab[i+1]= best_score;
03210
03211
03212 if(last_non_zero <= 27){
03213 for(; survivor_count; survivor_count--){
03214 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
03215 break;
03216 }
03217 }else{
03218 for(; survivor_count; survivor_count--){
03219 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
03220 break;
03221 }
03222 }
03223
03224 survivor[ survivor_count++ ]= i+1;
03225 }
03226
03227 if(s->out_format != FMT_H263){
03228 last_score= 256*256*256*120;
03229 for(i= survivor[0]; i<=last_non_zero + 1; i++){
03230 int score= score_tab[i];
03231 if(i) score += lambda*2;
03232
03233 if(score < last_score){
03234 last_score= score;
03235 last_i= i;
03236 last_level= level_tab[i];
03237 last_run= run_tab[i];
03238 }
03239 }
03240 }
03241
03242 s->coded_score[n] = last_score;
03243
03244 dc= FFABS(block[0]);
03245 last_non_zero= last_i - 1;
03246 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
03247
03248 if(last_non_zero < start_i)
03249 return last_non_zero;
03250
03251 if(last_non_zero == 0 && start_i == 0){
03252 int best_level= 0;
03253 int best_score= dc * dc;
03254
03255 for(i=0; i<coeff_count[0]; i++){
03256 int level= coeff[i][0];
03257 int alevel= FFABS(level);
03258 int unquant_coeff, score, distortion;
03259
03260 if(s->out_format == FMT_H263){
03261 unquant_coeff= (alevel*qmul + qadd)>>3;
03262 }else{
03263 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
03264 unquant_coeff = (unquant_coeff - 1) | 1;
03265 }
03266 unquant_coeff = (unquant_coeff + 4) >> 3;
03267 unquant_coeff<<= 3 + 3;
03268
03269 distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
03270 level+=64;
03271 if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
03272 else score= distortion + esc_length*lambda;
03273
03274 if(score < best_score){
03275 best_score= score;
03276 best_level= level - 64;
03277 }
03278 }
03279 block[0]= best_level;
03280 s->coded_score[n] = best_score - dc*dc;
03281 if(best_level == 0) return -1;
03282 else return last_non_zero;
03283 }
03284
03285 i= last_i;
03286 assert(last_level);
03287
03288 block[ perm_scantable[last_non_zero] ]= last_level;
03289 i -= last_run + 1;
03290
03291 for(; i>start_i; i -= run_tab[i] + 1){
03292 block[ perm_scantable[i-1] ]= level_tab[i];
03293 }
03294
03295 return last_non_zero;
03296 }
03297
03298
03299 static int16_t basis[64][64];
03300
03301 static void build_basis(uint8_t *perm){
03302 int i, j, x, y;
03303 emms_c();
03304 for(i=0; i<8; i++){
03305 for(j=0; j<8; j++){
03306 for(y=0; y<8; y++){
03307 for(x=0; x<8; x++){
03308 double s= 0.25*(1<<BASIS_SHIFT);
03309 int index= 8*i + j;
03310 int perm_index= perm[index];
03311 if(i==0) s*= sqrt(0.5);
03312 if(j==0) s*= sqrt(0.5);
03313 basis[perm_index][8*x + y]= lrintf(s * cos((M_PI/8.0)*i*(x+0.5)) * cos((M_PI/8.0)*j*(y+0.5)));
03314 }
03315 }
03316 }
03317 }
03318 }
03319
03320 static int dct_quantize_refine(MpegEncContext *s,
03321 DCTELEM *block, int16_t *weight, DCTELEM *orig,
03322 int n, int qscale){
03323 int16_t rem[64];
03324 LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
03325 const uint8_t *scantable= s->intra_scantable.scantable;
03326 const uint8_t *perm_scantable= s->intra_scantable.permutated;
03327
03328
03329 int run_tab[65];
03330 int prev_run=0;
03331 int prev_level=0;
03332 int qmul, qadd, start_i, last_non_zero, i, dc;
03333 uint8_t * length;
03334 uint8_t * last_length;
03335 int lambda;
03336 int rle_index, run, q = 1, sum;
03337 #ifdef REFINE_STATS
03338 static int count=0;
03339 static int after_last=0;
03340 static int to_zero=0;
03341 static int from_zero=0;
03342 static int raise=0;
03343 static int lower=0;
03344 static int messed_sign=0;
03345 #endif
03346
03347 if(basis[0][0] == 0)
03348 build_basis(s->dsp.idct_permutation);
03349
03350 qmul= qscale*2;
03351 qadd= (qscale-1)|1;
03352 if (s->mb_intra) {
03353 if (!s->h263_aic) {
03354 if (n < 4)
03355 q = s->y_dc_scale;
03356 else
03357 q = s->c_dc_scale;
03358 } else{
03359
03360 q = 1;
03361 qadd=0;
03362 }
03363 q <<= RECON_SHIFT-3;
03364
03365 dc= block[0]*q;
03366
03367 start_i = 1;
03368
03369
03370 length = s->intra_ac_vlc_length;
03371 last_length= s->intra_ac_vlc_last_length;
03372 } else {
03373 dc= 0;
03374 start_i = 0;
03375 length = s->inter_ac_vlc_length;
03376 last_length= s->inter_ac_vlc_last_length;
03377 }
03378 last_non_zero = s->block_last_index[n];
03379
03380 #ifdef REFINE_STATS
03381 {START_TIMER
03382 #endif
03383 dc += (1<<(RECON_SHIFT-1));
03384 for(i=0; i<64; i++){
03385 rem[i]= dc - (orig[i]<<RECON_SHIFT);
03386 }
03387 #ifdef REFINE_STATS
03388 STOP_TIMER("memset rem[]")}
03389 #endif
03390 sum=0;
03391 for(i=0; i<64; i++){
03392 int one= 36;
03393 int qns=4;
03394 int w;
03395
03396 w= FFABS(weight[i]) + qns*one;
03397 w= 15 + (48*qns*one + w/2)/w;
03398
03399 weight[i] = w;
03400
03401
03402 assert(w>0);
03403 assert(w<(1<<6));
03404 sum += w*w;
03405 }
03406 lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
03407 #ifdef REFINE_STATS
03408 {START_TIMER
03409 #endif
03410 run=0;
03411 rle_index=0;
03412 for(i=start_i; i<=last_non_zero; i++){
03413 int j= perm_scantable[i];
03414 const int level= block[j];
03415 int coeff;
03416
03417 if(level){
03418 if(level<0) coeff= qmul*level - qadd;
03419 else coeff= qmul*level + qadd;
03420 run_tab[rle_index++]=run;
03421 run=0;
03422
03423 s->dsp.add_8x8basis(rem, basis[j], coeff);
03424 }else{
03425 run++;
03426 }
03427 }
03428 #ifdef REFINE_STATS
03429 if(last_non_zero>0){
03430 STOP_TIMER("init rem[]")
03431 }
03432 }
03433
03434 {START_TIMER
03435 #endif
03436 for(;;){
03437 int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
03438 int best_coeff=0;
03439 int best_change=0;
03440 int run2, best_unquant_change=0, analyze_gradient;
03441 #ifdef REFINE_STATS
03442 {START_TIMER
03443 #endif
03444 analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
03445
03446 if(analyze_gradient){
03447 #ifdef REFINE_STATS
03448 {START_TIMER
03449 #endif
03450 for(i=0; i<64; i++){
03451 int w= weight[i];
03452
03453 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
03454 }
03455 #ifdef REFINE_STATS
03456 STOP_TIMER("rem*w*w")}
03457 {START_TIMER
03458 #endif
03459 s->dsp.fdct(d1);
03460 #ifdef REFINE_STATS
03461 STOP_TIMER("dct")}
03462 #endif
03463 }
03464
03465 if(start_i){
03466 const int level= block[0];
03467 int change, old_coeff;
03468
03469 assert(s->mb_intra);
03470
03471 old_coeff= q*level;
03472
03473 for(change=-1; change<=1; change+=2){
03474 int new_level= level + change;
03475 int score, new_coeff;
03476
03477 new_coeff= q*new_level;
03478 if(new_coeff >= 2048 || new_coeff < 0)
03479 continue;
03480
03481 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
03482 if(score<best_score){
03483 best_score= score;
03484 best_coeff= 0;
03485 best_change= change;
03486 best_unquant_change= new_coeff - old_coeff;
03487 }
03488 }
03489 }
03490
03491 run=0;
03492 rle_index=0;
03493 run2= run_tab[rle_index++];
03494 prev_level=0;
03495 prev_run=0;
03496
03497 for(i=start_i; i<64; i++){
03498 int j= perm_scantable[i];
03499 const int level= block[j];
03500 int change, old_coeff;
03501
03502 if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
03503 break;
03504
03505 if(level){
03506 if(level<0) old_coeff= qmul*level - qadd;
03507 else old_coeff= qmul*level + qadd;
03508 run2= run_tab[rle_index++];
03509 }else{
03510 old_coeff=0;
03511 run2--;
03512 assert(run2>=0 || i >= last_non_zero );
03513 }
03514
03515 for(change=-1; change<=1; change+=2){
03516 int new_level= level + change;
03517 int score, new_coeff, unquant_change;
03518
03519 score=0;
03520 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
03521 continue;
03522
03523 if(new_level){
03524 if(new_level<0) new_coeff= qmul*new_level - qadd;
03525 else new_coeff= qmul*new_level + qadd;
03526 if(new_coeff >= 2048 || new_coeff <= -2048)
03527 continue;
03528
03529
03530 if(level){
03531 if(level < 63 && level > -63){
03532 if(i < last_non_zero)
03533 score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
03534 - length[UNI_AC_ENC_INDEX(run, level+64)];
03535 else
03536 score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
03537 - last_length[UNI_AC_ENC_INDEX(run, level+64)];
03538 }
03539 }else{
03540 assert(FFABS(new_level)==1);
03541
03542 if(analyze_gradient){
03543 int g= d1[ scantable[i] ];
03544 if(g && (g^new_level) >= 0)
03545 continue;
03546 }
03547
03548 if(i < last_non_zero){
03549 int next_i= i + run2 + 1;
03550 int next_level= block[ perm_scantable[next_i] ] + 64;
03551
03552 if(next_level&(~127))
03553 next_level= 0;
03554
03555 if(next_i < last_non_zero)
03556 score += length[UNI_AC_ENC_INDEX(run, 65)]
03557 + length[UNI_AC_ENC_INDEX(run2, next_level)]
03558 - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
03559 else
03560 score += length[UNI_AC_ENC_INDEX(run, 65)]
03561 + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
03562 - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
03563 }else{
03564 score += last_length[UNI_AC_ENC_INDEX(run, 65)];
03565 if(prev_level){
03566 score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
03567 - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
03568 }
03569 }
03570 }
03571 }else{
03572 new_coeff=0;
03573 assert(FFABS(level)==1);
03574
03575 if(i < last_non_zero){
03576 int next_i= i + run2 + 1;
03577 int next_level= block[ perm_scantable[next_i] ] + 64;
03578
03579 if(next_level&(~127))
03580 next_level= 0;
03581
03582 if(next_i < last_non_zero)
03583 score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
03584 - length[UNI_AC_ENC_INDEX(run2, next_level)]
03585 - length[UNI_AC_ENC_INDEX(run, 65)];
03586 else
03587 score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
03588 - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
03589 - length[UNI_AC_ENC_INDEX(run, 65)];
03590 }else{
03591 score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
03592 if(prev_level){
03593 score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
03594 - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
03595 }
03596 }
03597 }
03598
03599 score *= lambda;
03600
03601 unquant_change= new_coeff - old_coeff;
03602 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
03603
03604 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
03605 if(score<best_score){
03606 best_score= score;
03607 best_coeff= i;
03608 best_change= change;
03609 best_unquant_change= unquant_change;
03610 }
03611 }
03612 if(level){
03613 prev_level= level + 64;
03614 if(prev_level&(~127))
03615 prev_level= 0;
03616 prev_run= run;
03617 run=0;
03618 }else{
03619 run++;
03620 }
03621 }
03622 #ifdef REFINE_STATS
03623 STOP_TIMER("iterative step")}
03624 #endif
03625
03626 if(best_change){
03627 int j= perm_scantable[ best_coeff ];
03628
03629 block[j] += best_change;
03630
03631 if(best_coeff > last_non_zero){
03632 last_non_zero= best_coeff;
03633 assert(block[j]);
03634 #ifdef REFINE_STATS
03635 after_last++;
03636 #endif
03637 }else{
03638 #ifdef REFINE_STATS
03639 if(block[j]){
03640 if(block[j] - best_change){
03641 if(FFABS(block[j]) > FFABS(block[j] - best_change)){
03642 raise++;
03643 }else{
03644 lower++;
03645 }
03646 }else{
03647 from_zero++;
03648 }
03649 }else{
03650 to_zero++;
03651 }
03652 #endif
03653 for(; last_non_zero>=start_i; last_non_zero--){
03654 if(block[perm_scantable[last_non_zero]])
03655 break;
03656 }
03657 }
03658 #ifdef REFINE_STATS
03659 count++;
03660 if(256*256*256*64 % count == 0){
03661 printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
03662 }
03663 #endif
03664 run=0;
03665 rle_index=0;
03666 for(i=start_i; i<=last_non_zero; i++){
03667 int j= perm_scantable[i];
03668 const int level= block[j];
03669
03670 if(level){
03671 run_tab[rle_index++]=run;
03672 run=0;
03673 }else{
03674 run++;
03675 }
03676 }
03677
03678 s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
03679 }else{
03680 break;
03681 }
03682 }
03683 #ifdef REFINE_STATS
03684 if(last_non_zero>0){
03685 STOP_TIMER("iterative search")
03686 }
03687 }
03688 #endif
03689
03690 return last_non_zero;
03691 }
03692
03693 int dct_quantize_c(MpegEncContext *s,
03694 DCTELEM *block, int n,
03695 int qscale, int *overflow)
03696 {
03697 int i, j, level, last_non_zero, q, start_i;
03698 const int *qmat;
03699 const uint8_t *scantable= s->intra_scantable.scantable;
03700 int bias;
03701 int max=0;
03702 unsigned int threshold1, threshold2;
03703
03704 s->dsp.fdct (block);
03705
03706 if(s->dct_error_sum)
03707 s->denoise_dct(s, block);
03708
03709 if (s->mb_intra) {
03710 if (!s->h263_aic) {
03711 if (n < 4)
03712 q = s->y_dc_scale;
03713 else
03714 q = s->c_dc_scale;
03715 q = q << 3;
03716 } else
03717
03718 q = 1 << 3;
03719
03720
03721 block[0] = (block[0] + (q >> 1)) / q;
03722 start_i = 1;
03723 last_non_zero = 0;
03724 qmat = s->q_intra_matrix[qscale];
03725 bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
03726 } else {
03727 start_i = 0;
03728 last_non_zero = -1;
03729 qmat = s->q_inter_matrix[qscale];
03730 bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
03731 }
03732 threshold1= (1<<QMAT_SHIFT) - bias - 1;
03733 threshold2= (threshold1<<1);
03734 for(i=63;i>=start_i;i--) {
03735 j = scantable[i];
03736 level = block[j] * qmat[j];
03737
03738 if(((unsigned)(level+threshold1))>threshold2){
03739 last_non_zero = i;
03740 break;
03741 }else{
03742 block[j]=0;
03743 }
03744 }
03745 for(i=start_i; i<=last_non_zero; i++) {
03746 j = scantable[i];
03747 level = block[j] * qmat[j];
03748
03749
03750
03751 if(((unsigned)(level+threshold1))>threshold2){
03752 if(level>0){
03753 level= (bias + level)>>QMAT_SHIFT;
03754 block[j]= level;
03755 }else{
03756 level= (bias - level)>>QMAT_SHIFT;
03757 block[j]= -level;
03758 }
03759 max |=level;
03760 }else{
03761 block[j]=0;
03762 }
03763 }
03764 *overflow= s->max_qcoeff < max;
03765
03766
03767 if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
03768 ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
03769
03770 return last_non_zero;
03771 }
03772
03773 AVCodec ff_h263_encoder = {
03774 "h263",
03775 AVMEDIA_TYPE_VIDEO,
03776 CODEC_ID_H263,
03777 sizeof(MpegEncContext),
03778 MPV_encode_init,
03779 MPV_encode_picture,
03780 MPV_encode_end,
03781 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
03782 .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
03783 };
03784
03785 AVCodec ff_h263p_encoder = {
03786 "h263p",
03787 AVMEDIA_TYPE_VIDEO,
03788 CODEC_ID_H263P,
03789 sizeof(MpegEncContext),
03790 MPV_encode_init,
03791 MPV_encode_picture,
03792 MPV_encode_end,
03793 .capabilities = CODEC_CAP_SLICE_THREADS,
03794 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
03795 .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
03796 };
03797
03798 AVCodec ff_msmpeg4v2_encoder = {
03799 "msmpeg4v2",
03800 AVMEDIA_TYPE_VIDEO,
03801 CODEC_ID_MSMPEG4V2,
03802 sizeof(MpegEncContext),
03803 MPV_encode_init,
03804 MPV_encode_picture,
03805 MPV_encode_end,
03806 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
03807 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
03808 };
03809
03810 AVCodec ff_msmpeg4v3_encoder = {
03811 "msmpeg4",
03812 AVMEDIA_TYPE_VIDEO,
03813 CODEC_ID_MSMPEG4V3,
03814 sizeof(MpegEncContext),
03815 MPV_encode_init,
03816 MPV_encode_picture,
03817 MPV_encode_end,
03818 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
03819 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
03820 };
03821
03822 AVCodec ff_wmv1_encoder = {
03823 "wmv1",
03824 AVMEDIA_TYPE_VIDEO,
03825 CODEC_ID_WMV1,
03826 sizeof(MpegEncContext),
03827 MPV_encode_init,
03828 MPV_encode_picture,
03829 MPV_encode_end,
03830 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
03831 .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
03832 };