Libav
|
00001 /* 00002 * The simplest mpeg encoder (well, it was the simplest!) 00003 * Copyright (c) 2000,2001 Fabrice Bellard 00004 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> 00005 * 00006 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at> 00007 * 00008 * This file is part of FFmpeg. 00009 * 00010 * FFmpeg is free software; you can redistribute it and/or 00011 * modify it under the terms of the GNU Lesser General Public 00012 * License as published by the Free Software Foundation; either 00013 * version 2.1 of the License, or (at your option) any later version. 00014 * 00015 * FFmpeg is distributed in the hope that it will be useful, 00016 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00017 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00018 * Lesser General Public License for more details. 00019 * 00020 * You should have received a copy of the GNU Lesser General Public 00021 * License along with FFmpeg; if not, write to the Free Software 00022 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 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 "aandcttab.h" 00040 #include "flv.h" 00041 #include "mpeg4video.h" 00042 #include "internal.h" 00043 #include <limits.h> 00044 00045 //#undef NDEBUG 00046 //#include <assert.h> 00047 00048 static int encode_picture(MpegEncContext *s, int picture_number); 00049 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale); 00050 static int sse_mb(MpegEncContext *s); 00051 00052 /* enable all paranoid tests for rounding, overflows, etc... */ 00053 //#define PARANOID 00054 00055 //#define DEBUG 00056 00057 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1]; 00058 static uint8_t default_fcode_tab[MAX_MV*2+1]; 00059 00060 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64], 00061 const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra) 00062 { 00063 int qscale; 00064 int shift=0; 00065 00066 for(qscale=qmin; qscale<=qmax; qscale++){ 00067 int i; 00068 if (dsp->fdct == ff_jpeg_fdct_islow 00069 #ifdef FAAN_POSTSCALE 00070 || dsp->fdct == ff_faandct 00071 #endif 00072 ) { 00073 for(i=0;i<64;i++) { 00074 const int j= dsp->idct_permutation[i]; 00075 /* 16 <= qscale * quant_matrix[i] <= 7905 */ 00076 /* 19952 <= ff_aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ 00077 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */ 00078 /* 3444240 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */ 00079 00080 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / 00081 (qscale * quant_matrix[j])); 00082 } 00083 } else if (dsp->fdct == fdct_ifast 00084 #ifndef FAAN_POSTSCALE 00085 || dsp->fdct == ff_faandct 00086 #endif 00087 ) { 00088 for(i=0;i<64;i++) { 00089 const int j= dsp->idct_permutation[i]; 00090 /* 16 <= qscale * quant_matrix[i] <= 7905 */ 00091 /* 19952 <= ff_aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ 00092 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ 00093 /* 3444240 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */ 00094 00095 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) / 00096 (ff_aanscales[i] * qscale * quant_matrix[j])); 00097 } 00098 } else { 00099 for(i=0;i<64;i++) { 00100 const int j= dsp->idct_permutation[i]; 00101 /* We can safely suppose that 16 <= quant_matrix[i] <= 255 00102 So 16 <= qscale * quant_matrix[i] <= 7905 00103 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 00104 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 00105 */ 00106 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j])); 00107 // qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); 00108 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]); 00109 00110 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1; 00111 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]); 00112 } 00113 } 00114 00115 for(i=intra; i<64; i++){ 00116 int64_t max= 8191; 00117 if (dsp->fdct == fdct_ifast 00118 #ifndef FAAN_POSTSCALE 00119 || dsp->fdct == ff_faandct 00120 #endif 00121 ) { 00122 max = (8191LL*ff_aanscales[i]) >> 14; 00123 } 00124 while(((max * qmat[qscale][i]) >> shift) > INT_MAX){ 00125 shift++; 00126 } 00127 } 00128 } 00129 if(shift){ 00130 av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift); 00131 } 00132 } 00133 00134 static inline void update_qscale(MpegEncContext *s){ 00135 s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); 00136 s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax); 00137 00138 s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; 00139 } 00140 00141 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){ 00142 int i; 00143 00144 if(matrix){ 00145 put_bits(pb, 1, 1); 00146 for(i=0;i<64;i++) { 00147 put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]); 00148 } 00149 }else 00150 put_bits(pb, 1, 0); 00151 } 00152 00156 void ff_init_qscale_tab(MpegEncContext *s){ 00157 int8_t * const qscale_table= s->current_picture.qscale_table; 00158 int i; 00159 00160 for(i=0; i<s->mb_num; i++){ 00161 unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ]; 00162 int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); 00163 qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax); 00164 } 00165 } 00166 00167 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){ 00168 int i; 00169 00170 dst->pict_type = src->pict_type; 00171 dst->quality = src->quality; 00172 dst->coded_picture_number = src->coded_picture_number; 00173 dst->display_picture_number = src->display_picture_number; 00174 // dst->reference = src->reference; 00175 dst->pts = src->pts; 00176 dst->interlaced_frame = src->interlaced_frame; 00177 dst->top_field_first = src->top_field_first; 00178 00179 if(s->avctx->me_threshold){ 00180 if(!src->motion_val[0]) 00181 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n"); 00182 if(!src->mb_type) 00183 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n"); 00184 if(!src->ref_index[0]) 00185 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n"); 00186 if(src->motion_subsample_log2 != dst->motion_subsample_log2) 00187 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n", 00188 src->motion_subsample_log2, dst->motion_subsample_log2); 00189 00190 memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0])); 00191 00192 for(i=0; i<2; i++){ 00193 int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1; 00194 int height= ((16*s->mb_height)>>src->motion_subsample_log2); 00195 00196 if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){ 00197 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t)); 00198 } 00199 if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){ 00200 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t)); 00201 } 00202 } 00203 } 00204 } 00205 00206 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){ 00207 #define COPY(a) dst->a= src->a 00208 COPY(pict_type); 00209 COPY(current_picture); 00210 COPY(f_code); 00211 COPY(b_code); 00212 COPY(qscale); 00213 COPY(lambda); 00214 COPY(lambda2); 00215 COPY(picture_in_gop_number); 00216 COPY(gop_picture_number); 00217 COPY(frame_pred_frame_dct); //FIXME don't set in encode_header 00218 COPY(progressive_frame); //FIXME don't set in encode_header 00219 COPY(partitioned_frame); //FIXME don't set in encode_header 00220 #undef COPY 00221 } 00222 00227 static void MPV_encode_defaults(MpegEncContext *s){ 00228 int i; 00229 MPV_common_defaults(s); 00230 00231 for(i=-16; i<16; i++){ 00232 default_fcode_tab[i + MAX_MV]= 1; 00233 } 00234 s->me.mv_penalty= default_mv_penalty; 00235 s->fcode_tab= default_fcode_tab; 00236 } 00237 00238 /* init video encoder */ 00239 av_cold int MPV_encode_init(AVCodecContext *avctx) 00240 { 00241 MpegEncContext *s = avctx->priv_data; 00242 int i; 00243 int chroma_h_shift, chroma_v_shift; 00244 00245 MPV_encode_defaults(s); 00246 00247 switch (avctx->codec_id) { 00248 case CODEC_ID_MPEG2VIDEO: 00249 if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){ 00250 av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n"); 00251 return -1; 00252 } 00253 break; 00254 case CODEC_ID_LJPEG: 00255 case CODEC_ID_MJPEG: 00256 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_RGB32 && 00257 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){ 00258 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n"); 00259 return -1; 00260 } 00261 break; 00262 default: 00263 if(avctx->pix_fmt != PIX_FMT_YUV420P){ 00264 av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n"); 00265 return -1; 00266 } 00267 } 00268 00269 switch (avctx->pix_fmt) { 00270 case PIX_FMT_YUVJ422P: 00271 case PIX_FMT_YUV422P: 00272 s->chroma_format = CHROMA_422; 00273 break; 00274 case PIX_FMT_YUVJ420P: 00275 case PIX_FMT_YUV420P: 00276 default: 00277 s->chroma_format = CHROMA_420; 00278 break; 00279 } 00280 00281 s->bit_rate = avctx->bit_rate; 00282 s->width = avctx->width; 00283 s->height = avctx->height; 00284 if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){ 00285 av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n"); 00286 avctx->gop_size=600; 00287 } 00288 s->gop_size = avctx->gop_size; 00289 s->avctx = avctx; 00290 s->flags= avctx->flags; 00291 s->flags2= avctx->flags2; 00292 s->max_b_frames= avctx->max_b_frames; 00293 s->codec_id= avctx->codec->id; 00294 s->luma_elim_threshold = avctx->luma_elim_threshold; 00295 s->chroma_elim_threshold= avctx->chroma_elim_threshold; 00296 s->strict_std_compliance= avctx->strict_std_compliance; 00297 s->data_partitioning= avctx->flags & CODEC_FLAG_PART; 00298 s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0; 00299 s->mpeg_quant= avctx->mpeg_quant; 00300 s->rtp_mode= !!avctx->rtp_payload_size; 00301 s->intra_dc_precision= avctx->intra_dc_precision; 00302 s->user_specified_pts = AV_NOPTS_VALUE; 00303 00304 if (s->gop_size <= 1) { 00305 s->intra_only = 1; 00306 s->gop_size = 12; 00307 } else { 00308 s->intra_only = 0; 00309 } 00310 00311 s->me_method = avctx->me_method; 00312 00313 /* Fixed QSCALE */ 00314 s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE); 00315 00316 s->adaptive_quant= ( s->avctx->lumi_masking 00317 || s->avctx->dark_masking 00318 || s->avctx->temporal_cplx_masking 00319 || s->avctx->spatial_cplx_masking 00320 || s->avctx->p_masking 00321 || s->avctx->border_masking 00322 || (s->flags&CODEC_FLAG_QP_RD)) 00323 && !s->fixed_qscale; 00324 00325 s->obmc= !!(s->flags & CODEC_FLAG_OBMC); 00326 s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER); 00327 s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN); 00328 s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC); 00329 s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT); 00330 00331 if(avctx->rc_max_rate && !avctx->rc_buffer_size){ 00332 av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n"); 00333 return -1; 00334 } 00335 00336 if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){ 00337 av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n"); 00338 } 00339 00340 if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){ 00341 av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n"); 00342 return -1; 00343 } 00344 00345 if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){ 00346 av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n"); 00347 return -1; 00348 } 00349 00350 if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){ 00351 av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n"); 00352 } 00353 00354 if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){ 00355 av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n"); 00356 return -1; 00357 } 00358 00359 if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){ 00360 av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n"); 00361 return -1; 00362 } 00363 00364 if( s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate 00365 && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO) 00366 && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){ 00367 00368 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"); 00369 } 00370 00371 if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4 00372 && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){ 00373 av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n"); 00374 return -1; 00375 } 00376 00377 if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){ 00378 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n"); 00379 return -1; 00380 } 00381 00382 if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){ 00383 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n"); 00384 return -1; 00385 } 00386 00387 if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){ 00388 av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n"); 00389 return -1; 00390 } 00391 00392 if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){ 00393 av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n"); 00394 return -1; 00395 } 00396 00397 if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){ 00398 av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n"); 00399 return -1; 00400 } 00401 00402 if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 || 00403 s->codec_id == CODEC_ID_H263P) && 00404 (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) { 00405 av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n", 00406 avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den); 00407 return -1; 00408 } 00409 00410 if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN)) 00411 && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){ 00412 av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n"); 00413 return -1; 00414 } 00415 00416 if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too 00417 av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n"); 00418 return -1; 00419 } 00420 00421 if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){ 00422 av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n"); 00423 return -1; 00424 } 00425 00426 if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){ 00427 av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n"); 00428 return -1; 00429 } 00430 00431 if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){ 00432 av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n"); 00433 return -1; 00434 } 00435 00436 if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){ 00437 av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n"); 00438 return -1; 00439 } 00440 00441 if(s->flags & CODEC_FLAG_LOW_DELAY){ 00442 if (s->codec_id != CODEC_ID_MPEG2VIDEO){ 00443 av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n"); 00444 return -1; 00445 } 00446 if (s->max_b_frames != 0){ 00447 av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n"); 00448 return -1; 00449 } 00450 } 00451 00452 if(s->q_scale_type == 1){ 00453 if(s->codec_id != CODEC_ID_MPEG2VIDEO){ 00454 av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n"); 00455 return -1; 00456 } 00457 if(avctx->qmax > 12){ 00458 av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n"); 00459 return -1; 00460 } 00461 } 00462 00463 if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4 00464 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO 00465 && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){ 00466 av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n"); 00467 return -1; 00468 } 00469 00470 if(s->avctx->thread_count < 1){ 00471 av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n"); 00472 return -1; 00473 } 00474 00475 if(s->avctx->thread_count > 1) 00476 s->rtp_mode= 1; 00477 00478 if(!avctx->time_base.den || !avctx->time_base.num){ 00479 av_log(avctx, AV_LOG_ERROR, "framerate not set\n"); 00480 return -1; 00481 } 00482 00483 i= (INT_MAX/2+128)>>8; 00484 if(avctx->me_threshold >= i){ 00485 av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1); 00486 return -1; 00487 } 00488 if(avctx->mb_threshold >= i){ 00489 av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1); 00490 return -1; 00491 } 00492 00493 if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){ 00494 av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n"); 00495 avctx->b_frame_strategy = 0; 00496 } 00497 00498 i= av_gcd(avctx->time_base.den, avctx->time_base.num); 00499 if(i > 1){ 00500 av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n"); 00501 avctx->time_base.den /= i; 00502 avctx->time_base.num /= i; 00503 // return -1; 00504 } 00505 00506 if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){ 00507 s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x 00508 s->inter_quant_bias= 0; 00509 }else{ 00510 s->intra_quant_bias=0; 00511 s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x 00512 } 00513 00514 if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS) 00515 s->intra_quant_bias= avctx->intra_quant_bias; 00516 if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS) 00517 s->inter_quant_bias= avctx->inter_quant_bias; 00518 00519 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift); 00520 00521 if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){ 00522 av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n"); 00523 return -1; 00524 } 00525 s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1; 00526 00527 switch(avctx->codec->id) { 00528 case CODEC_ID_MPEG1VIDEO: 00529 s->out_format = FMT_MPEG1; 00530 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY); 00531 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); 00532 break; 00533 case CODEC_ID_MPEG2VIDEO: 00534 s->out_format = FMT_MPEG1; 00535 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY); 00536 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); 00537 s->rtp_mode= 1; 00538 break; 00539 case CODEC_ID_LJPEG: 00540 case CODEC_ID_MJPEG: 00541 s->out_format = FMT_MJPEG; 00542 s->intra_only = 1; /* force intra only for jpeg */ 00543 if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){ 00544 s->mjpeg_vsample[0] = s->mjpeg_hsample[0] = 00545 s->mjpeg_vsample[1] = s->mjpeg_hsample[1] = 00546 s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1; 00547 }else{ 00548 s->mjpeg_vsample[0] = 2; 00549 s->mjpeg_vsample[1] = 2>>chroma_v_shift; 00550 s->mjpeg_vsample[2] = 2>>chroma_v_shift; 00551 s->mjpeg_hsample[0] = 2; 00552 s->mjpeg_hsample[1] = 2>>chroma_h_shift; 00553 s->mjpeg_hsample[2] = 2>>chroma_h_shift; 00554 } 00555 if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) 00556 || ff_mjpeg_encode_init(s) < 0) 00557 return -1; 00558 avctx->delay=0; 00559 s->low_delay=1; 00560 break; 00561 case CODEC_ID_H261: 00562 if (!CONFIG_H261_ENCODER) return -1; 00563 if (ff_h261_get_picture_format(s->width, s->height) < 0) { 00564 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); 00565 return -1; 00566 } 00567 s->out_format = FMT_H261; 00568 avctx->delay=0; 00569 s->low_delay=1; 00570 break; 00571 case CODEC_ID_H263: 00572 if (!CONFIG_H263_ENCODER) return -1; 00573 if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 7) { 00574 av_log(avctx, AV_LOG_INFO, "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); 00575 return -1; 00576 } 00577 s->out_format = FMT_H263; 00578 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0; 00579 avctx->delay=0; 00580 s->low_delay=1; 00581 break; 00582 case CODEC_ID_H263P: 00583 s->out_format = FMT_H263; 00584 s->h263_plus = 1; 00585 /* Fx */ 00586 s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0; 00587 s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0; 00588 s->modified_quant= s->h263_aic; 00589 s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0; 00590 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0; 00591 s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0; 00592 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus; 00593 s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0; 00594 00595 /* /Fx */ 00596 /* These are just to be sure */ 00597 avctx->delay=0; 00598 s->low_delay=1; 00599 break; 00600 case CODEC_ID_FLV1: 00601 s->out_format = FMT_H263; 00602 s->h263_flv = 2; /* format = 1; 11-bit codes */ 00603 s->unrestricted_mv = 1; 00604 s->rtp_mode=0; /* don't allow GOB */ 00605 avctx->delay=0; 00606 s->low_delay=1; 00607 break; 00608 case CODEC_ID_RV10: 00609 s->out_format = FMT_H263; 00610 avctx->delay=0; 00611 s->low_delay=1; 00612 break; 00613 case CODEC_ID_RV20: 00614 s->out_format = FMT_H263; 00615 avctx->delay=0; 00616 s->low_delay=1; 00617 s->modified_quant=1; 00618 s->h263_aic=1; 00619 s->h263_plus=1; 00620 s->loop_filter=1; 00621 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus; 00622 break; 00623 case CODEC_ID_MPEG4: 00624 s->out_format = FMT_H263; 00625 s->h263_pred = 1; 00626 s->unrestricted_mv = 1; 00627 s->low_delay= s->max_b_frames ? 0 : 1; 00628 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1); 00629 break; 00630 case CODEC_ID_MSMPEG4V1: 00631 s->out_format = FMT_H263; 00632 s->h263_msmpeg4 = 1; 00633 s->h263_pred = 1; 00634 s->unrestricted_mv = 1; 00635 s->msmpeg4_version= 1; 00636 avctx->delay=0; 00637 s->low_delay=1; 00638 break; 00639 case CODEC_ID_MSMPEG4V2: 00640 s->out_format = FMT_H263; 00641 s->h263_msmpeg4 = 1; 00642 s->h263_pred = 1; 00643 s->unrestricted_mv = 1; 00644 s->msmpeg4_version= 2; 00645 avctx->delay=0; 00646 s->low_delay=1; 00647 break; 00648 case CODEC_ID_MSMPEG4V3: 00649 s->out_format = FMT_H263; 00650 s->h263_msmpeg4 = 1; 00651 s->h263_pred = 1; 00652 s->unrestricted_mv = 1; 00653 s->msmpeg4_version= 3; 00654 s->flipflop_rounding=1; 00655 avctx->delay=0; 00656 s->low_delay=1; 00657 break; 00658 case CODEC_ID_WMV1: 00659 s->out_format = FMT_H263; 00660 s->h263_msmpeg4 = 1; 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_msmpeg4 = 1; 00671 s->h263_pred = 1; 00672 s->unrestricted_mv = 1; 00673 s->msmpeg4_version= 5; 00674 s->flipflop_rounding=1; 00675 avctx->delay=0; 00676 s->low_delay=1; 00677 break; 00678 default: 00679 return -1; 00680 } 00681 00682 avctx->has_b_frames= !s->low_delay; 00683 00684 s->encoding = 1; 00685 00686 s->progressive_frame= 00687 s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN)); 00688 00689 /* init */ 00690 if (MPV_common_init(s) < 0) 00691 return -1; 00692 00693 if(!s->dct_quantize) 00694 s->dct_quantize = dct_quantize_c; 00695 if(!s->denoise_dct) 00696 s->denoise_dct = denoise_dct_c; 00697 s->fast_dct_quantize = s->dct_quantize; 00698 if(avctx->trellis) 00699 s->dct_quantize = dct_quantize_trellis_c; 00700 00701 if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant) 00702 s->chroma_qscale_table= ff_h263_chroma_qscale_table; 00703 00704 s->quant_precision=5; 00705 00706 ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp); 00707 ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp); 00708 00709 if (CONFIG_H261_ENCODER && s->out_format == FMT_H261) 00710 ff_h261_encode_init(s); 00711 if (CONFIG_H263_ENCODER && s->out_format == FMT_H263) 00712 h263_encode_init(s); 00713 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version) 00714 ff_msmpeg4_encode_init(s); 00715 if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) 00716 && s->out_format == FMT_MPEG1) 00717 ff_mpeg1_encode_init(s); 00718 00719 /* init q matrix */ 00720 for(i=0;i<64;i++) { 00721 int j= s->dsp.idct_permutation[i]; 00722 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){ 00723 s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i]; 00724 s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i]; 00725 }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){ 00726 s->intra_matrix[j] = 00727 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i]; 00728 }else 00729 { /* mpeg1/2 */ 00730 s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i]; 00731 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i]; 00732 } 00733 if(s->avctx->intra_matrix) 00734 s->intra_matrix[j] = s->avctx->intra_matrix[i]; 00735 if(s->avctx->inter_matrix) 00736 s->inter_matrix[j] = s->avctx->inter_matrix[i]; 00737 } 00738 00739 /* precompute matrix */ 00740 /* for mjpeg, we do include qscale in the matrix */ 00741 if (s->out_format != FMT_MJPEG) { 00742 ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16, 00743 s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1); 00744 ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16, 00745 s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0); 00746 } 00747 00748 if(ff_rate_control_init(s) < 0) 00749 return -1; 00750 00751 return 0; 00752 } 00753 00754 av_cold int MPV_encode_end(AVCodecContext *avctx) 00755 { 00756 MpegEncContext *s = avctx->priv_data; 00757 00758 ff_rate_control_uninit(s); 00759 00760 MPV_common_end(s); 00761 if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG) 00762 ff_mjpeg_encode_close(s); 00763 00764 av_freep(&avctx->extradata); 00765 00766 return 0; 00767 } 00768 00769 static int get_sae(uint8_t *src, int ref, int stride){ 00770 int x,y; 00771 int acc=0; 00772 00773 for(y=0; y<16; y++){ 00774 for(x=0; x<16; x++){ 00775 acc+= FFABS(src[x+y*stride] - ref); 00776 } 00777 } 00778 00779 return acc; 00780 } 00781 00782 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){ 00783 int x, y, w, h; 00784 int acc=0; 00785 00786 w= s->width &~15; 00787 h= s->height&~15; 00788 00789 for(y=0; y<h; y+=16){ 00790 for(x=0; x<w; x+=16){ 00791 int offset= x + y*stride; 00792 int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16); 00793 int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8; 00794 int sae = get_sae(src + offset, mean, stride); 00795 00796 acc+= sae + 500 < sad; 00797 } 00798 } 00799 return acc; 00800 } 00801 00802 00803 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){ 00804 AVFrame *pic=NULL; 00805 int64_t pts; 00806 int i; 00807 const int encoding_delay= s->max_b_frames; 00808 int direct=1; 00809 00810 if(pic_arg){ 00811 pts= pic_arg->pts; 00812 pic_arg->display_picture_number= s->input_picture_number++; 00813 00814 if(pts != AV_NOPTS_VALUE){ 00815 if(s->user_specified_pts != AV_NOPTS_VALUE){ 00816 int64_t time= pts; 00817 int64_t last= s->user_specified_pts; 00818 00819 if(time <= last){ 00820 av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts); 00821 return -1; 00822 } 00823 } 00824 s->user_specified_pts= pts; 00825 }else{ 00826 if(s->user_specified_pts != AV_NOPTS_VALUE){ 00827 s->user_specified_pts= 00828 pts= s->user_specified_pts + 1; 00829 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts); 00830 }else{ 00831 pts= pic_arg->display_picture_number; 00832 } 00833 } 00834 } 00835 00836 if(pic_arg){ 00837 if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0; 00838 if(pic_arg->linesize[0] != s->linesize) direct=0; 00839 if(pic_arg->linesize[1] != s->uvlinesize) direct=0; 00840 if(pic_arg->linesize[2] != s->uvlinesize) direct=0; 00841 00842 // av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize); 00843 00844 if(direct){ 00845 i= ff_find_unused_picture(s, 1); 00846 00847 pic= (AVFrame*)&s->picture[i]; 00848 pic->reference= 3; 00849 00850 for(i=0; i<4; i++){ 00851 pic->data[i]= pic_arg->data[i]; 00852 pic->linesize[i]= pic_arg->linesize[i]; 00853 } 00854 if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){ 00855 return -1; 00856 } 00857 }else{ 00858 i= ff_find_unused_picture(s, 0); 00859 00860 pic= (AVFrame*)&s->picture[i]; 00861 pic->reference= 3; 00862 00863 if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){ 00864 return -1; 00865 } 00866 00867 if( pic->data[0] + INPLACE_OFFSET == pic_arg->data[0] 00868 && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1] 00869 && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){ 00870 // empty 00871 }else{ 00872 int h_chroma_shift, v_chroma_shift; 00873 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift); 00874 00875 for(i=0; i<3; i++){ 00876 int src_stride= pic_arg->linesize[i]; 00877 int dst_stride= i ? s->uvlinesize : s->linesize; 00878 int h_shift= i ? h_chroma_shift : 0; 00879 int v_shift= i ? v_chroma_shift : 0; 00880 int w= s->width >>h_shift; 00881 int h= s->height>>v_shift; 00882 uint8_t *src= pic_arg->data[i]; 00883 uint8_t *dst= pic->data[i]; 00884 00885 if(!s->avctx->rc_buffer_size) 00886 dst +=INPLACE_OFFSET; 00887 00888 if(src_stride==dst_stride) 00889 memcpy(dst, src, src_stride*h); 00890 else{ 00891 while(h--){ 00892 memcpy(dst, src, w); 00893 dst += dst_stride; 00894 src += src_stride; 00895 } 00896 } 00897 } 00898 } 00899 } 00900 copy_picture_attributes(s, pic, pic_arg); 00901 pic->pts= pts; //we set this here to avoid modifiying pic_arg 00902 } 00903 00904 /* shift buffer entries */ 00905 for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++) 00906 s->input_picture[i-1]= s->input_picture[i]; 00907 00908 s->input_picture[encoding_delay]= (Picture*)pic; 00909 00910 return 0; 00911 } 00912 00913 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){ 00914 int x, y, plane; 00915 int score=0; 00916 int64_t score64=0; 00917 00918 for(plane=0; plane<3; plane++){ 00919 const int stride= p->linesize[plane]; 00920 const int bw= plane ? 1 : 2; 00921 for(y=0; y<s->mb_height*bw; y++){ 00922 for(x=0; x<s->mb_width*bw; x++){ 00923 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16; 00924 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); 00925 00926 switch(s->avctx->frame_skip_exp){ 00927 case 0: score= FFMAX(score, v); break; 00928 case 1: score+= FFABS(v);break; 00929 case 2: score+= v*v;break; 00930 case 3: score64+= FFABS(v*v*(int64_t)v);break; 00931 case 4: score64+= v*v*(int64_t)(v*v);break; 00932 } 00933 } 00934 } 00935 } 00936 00937 if(score) score64= score; 00938 00939 if(score64 < s->avctx->frame_skip_threshold) 00940 return 1; 00941 if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8)) 00942 return 1; 00943 return 0; 00944 } 00945 00946 static int estimate_best_b_count(MpegEncContext *s){ 00947 AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id); 00948 AVCodecContext *c= avcodec_alloc_context(); 00949 AVFrame input[FF_MAX_B_FRAMES+2]; 00950 const int scale= s->avctx->brd_scale; 00951 int i, j, out_size, p_lambda, b_lambda, lambda2; 00952 int outbuf_size= s->width * s->height; //FIXME 00953 uint8_t *outbuf= av_malloc(outbuf_size); 00954 int64_t best_rd= INT64_MAX; 00955 int best_b_count= -1; 00956 00957 assert(scale>=0 && scale <=3); 00958 00959 // emms_c(); 00960 p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality; 00961 b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset; 00962 if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else 00963 lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT; 00964 00965 c->width = s->width >> scale; 00966 c->height= s->height>> scale; 00967 c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/; 00968 c->flags|= s->avctx->flags & CODEC_FLAG_QPEL; 00969 c->mb_decision= s->avctx->mb_decision; 00970 c->me_cmp= s->avctx->me_cmp; 00971 c->mb_cmp= s->avctx->mb_cmp; 00972 c->me_sub_cmp= s->avctx->me_sub_cmp; 00973 c->pix_fmt = PIX_FMT_YUV420P; 00974 c->time_base= s->avctx->time_base; 00975 c->max_b_frames= s->max_b_frames; 00976 00977 if (avcodec_open(c, codec) < 0) 00978 return -1; 00979 00980 for(i=0; i<s->max_b_frames+2; i++){ 00981 int ysize= c->width*c->height; 00982 int csize= (c->width/2)*(c->height/2); 00983 Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr; 00984 00985 avcodec_get_frame_defaults(&input[i]); 00986 input[i].data[0]= av_malloc(ysize + 2*csize); 00987 input[i].data[1]= input[i].data[0] + ysize; 00988 input[i].data[2]= input[i].data[1] + csize; 00989 input[i].linesize[0]= c->width; 00990 input[i].linesize[1]= 00991 input[i].linesize[2]= c->width/2; 00992 00993 if(pre_input_ptr && (!i || s->input_picture[i-1])) { 00994 pre_input= *pre_input_ptr; 00995 00996 if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) { 00997 pre_input.data[0]+=INPLACE_OFFSET; 00998 pre_input.data[1]+=INPLACE_OFFSET; 00999 pre_input.data[2]+=INPLACE_OFFSET; 01000 } 01001 01002 s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height); 01003 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); 01004 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); 01005 } 01006 } 01007 01008 for(j=0; j<s->max_b_frames+1; j++){ 01009 int64_t rd=0; 01010 01011 if(!s->input_picture[j]) 01012 break; 01013 01014 c->error[0]= c->error[1]= c->error[2]= 0; 01015 01016 input[0].pict_type= FF_I_TYPE; 01017 input[0].quality= 1 * FF_QP2LAMBDA; 01018 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]); 01019 // rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT; 01020 01021 for(i=0; i<s->max_b_frames+1; i++){ 01022 int is_p= i % (j+1) == j || i==s->max_b_frames; 01023 01024 input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE; 01025 input[i+1].quality= is_p ? p_lambda : b_lambda; 01026 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]); 01027 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3); 01028 } 01029 01030 /* get the delayed frames */ 01031 while(out_size){ 01032 out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL); 01033 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3); 01034 } 01035 01036 rd += c->error[0] + c->error[1] + c->error[2]; 01037 01038 if(rd < best_rd){ 01039 best_rd= rd; 01040 best_b_count= j; 01041 } 01042 } 01043 01044 av_freep(&outbuf); 01045 avcodec_close(c); 01046 av_freep(&c); 01047 01048 for(i=0; i<s->max_b_frames+2; i++){ 01049 av_freep(&input[i].data[0]); 01050 } 01051 01052 return best_b_count; 01053 } 01054 01055 static int select_input_picture(MpegEncContext *s){ 01056 int i; 01057 01058 for(i=1; i<MAX_PICTURE_COUNT; i++) 01059 s->reordered_input_picture[i-1]= s->reordered_input_picture[i]; 01060 s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL; 01061 01062 /* set next picture type & ordering */ 01063 if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){ 01064 if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){ 01065 s->reordered_input_picture[0]= s->input_picture[0]; 01066 s->reordered_input_picture[0]->pict_type= FF_I_TYPE; 01067 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++; 01068 }else{ 01069 int b_frames; 01070 01071 if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){ 01072 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){ 01073 //FIXME check that te gop check above is +-1 correct 01074 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts); 01075 01076 if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){ 01077 for(i=0; i<4; i++) 01078 s->input_picture[0]->data[i]= NULL; 01079 s->input_picture[0]->type= 0; 01080 }else{ 01081 assert( s->input_picture[0]->type==FF_BUFFER_TYPE_USER 01082 || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL); 01083 01084 s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]); 01085 } 01086 01087 emms_c(); 01088 ff_vbv_update(s, 0); 01089 01090 goto no_output_pic; 01091 } 01092 } 01093 01094 if(s->flags&CODEC_FLAG_PASS2){ 01095 for(i=0; i<s->max_b_frames+1; i++){ 01096 int pict_num= s->input_picture[0]->display_picture_number + i; 01097 01098 if(pict_num >= s->rc_context.num_entries) 01099 break; 01100 if(!s->input_picture[i]){ 01101 s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE; 01102 break; 01103 } 01104 01105 s->input_picture[i]->pict_type= 01106 s->rc_context.entry[pict_num].new_pict_type; 01107 } 01108 } 01109 01110 if(s->avctx->b_frame_strategy==0){ 01111 b_frames= s->max_b_frames; 01112 while(b_frames && !s->input_picture[b_frames]) b_frames--; 01113 }else if(s->avctx->b_frame_strategy==1){ 01114 for(i=1; i<s->max_b_frames+1; i++){ 01115 if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){ 01116 s->input_picture[i]->b_frame_score= 01117 get_intra_count(s, s->input_picture[i ]->data[0], 01118 s->input_picture[i-1]->data[0], s->linesize) + 1; 01119 } 01120 } 01121 for(i=0; i<s->max_b_frames+1; i++){ 01122 if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break; 01123 } 01124 01125 b_frames= FFMAX(0, i-1); 01126 01127 /* reset scores */ 01128 for(i=0; i<b_frames+1; i++){ 01129 s->input_picture[i]->b_frame_score=0; 01130 } 01131 }else if(s->avctx->b_frame_strategy==2){ 01132 b_frames= estimate_best_b_count(s); 01133 }else{ 01134 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n"); 01135 b_frames=0; 01136 } 01137 01138 emms_c(); 01139 //static int b_count=0; 01140 //b_count+= b_frames; 01141 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count); 01142 01143 for(i= b_frames - 1; i>=0; i--){ 01144 int type= s->input_picture[i]->pict_type; 01145 if(type && type != FF_B_TYPE) 01146 b_frames= i; 01147 } 01148 if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){ 01149 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n"); 01150 } 01151 01152 if(s->picture_in_gop_number + b_frames >= s->gop_size){ 01153 if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){ 01154 b_frames= s->gop_size - s->picture_in_gop_number - 1; 01155 }else{ 01156 if(s->flags & CODEC_FLAG_CLOSED_GOP) 01157 b_frames=0; 01158 s->input_picture[b_frames]->pict_type= FF_I_TYPE; 01159 } 01160 } 01161 01162 if( (s->flags & CODEC_FLAG_CLOSED_GOP) 01163 && b_frames 01164 && s->input_picture[b_frames]->pict_type== FF_I_TYPE) 01165 b_frames--; 01166 01167 s->reordered_input_picture[0]= s->input_picture[b_frames]; 01168 if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE) 01169 s->reordered_input_picture[0]->pict_type= FF_P_TYPE; 01170 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++; 01171 for(i=0; i<b_frames; i++){ 01172 s->reordered_input_picture[i+1]= s->input_picture[i]; 01173 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE; 01174 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++; 01175 } 01176 } 01177 } 01178 no_output_pic: 01179 if(s->reordered_input_picture[0]){ 01180 s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0; 01181 01182 ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]); 01183 01184 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){ 01185 // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable 01186 01187 int i= ff_find_unused_picture(s, 0); 01188 Picture *pic= &s->picture[i]; 01189 01190 pic->reference = s->reordered_input_picture[0]->reference; 01191 if(ff_alloc_picture(s, pic, 0) < 0){ 01192 return -1; 01193 } 01194 01195 /* mark us unused / free shared pic */ 01196 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL) 01197 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]); 01198 for(i=0; i<4; i++) 01199 s->reordered_input_picture[0]->data[i]= NULL; 01200 s->reordered_input_picture[0]->type= 0; 01201 01202 copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]); 01203 01204 s->current_picture_ptr= pic; 01205 }else{ 01206 // input is not a shared pix -> reuse buffer for current_pix 01207 01208 assert( s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER 01209 || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL); 01210 01211 s->current_picture_ptr= s->reordered_input_picture[0]; 01212 for(i=0; i<4; i++){ 01213 s->new_picture.data[i]+= INPLACE_OFFSET; 01214 } 01215 } 01216 ff_copy_picture(&s->current_picture, s->current_picture_ptr); 01217 01218 s->picture_number= s->new_picture.display_picture_number; 01219 //printf("dpn:%d\n", s->picture_number); 01220 }else{ 01221 memset(&s->new_picture, 0, sizeof(Picture)); 01222 } 01223 return 0; 01224 } 01225 01226 int MPV_encode_picture(AVCodecContext *avctx, 01227 unsigned char *buf, int buf_size, void *data) 01228 { 01229 MpegEncContext *s = avctx->priv_data; 01230 AVFrame *pic_arg = data; 01231 int i, stuffing_count; 01232 01233 for(i=0; i<avctx->thread_count; i++){ 01234 int start_y= s->thread_context[i]->start_mb_y; 01235 int end_y= s->thread_context[i]-> end_mb_y; 01236 int h= s->mb_height; 01237 uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h); 01238 uint8_t *end = buf + (size_t)(((int64_t) buf_size)* end_y/h); 01239 01240 init_put_bits(&s->thread_context[i]->pb, start, end - start); 01241 } 01242 01243 s->picture_in_gop_number++; 01244 01245 if(load_input_picture(s, pic_arg) < 0) 01246 return -1; 01247 01248 if(select_input_picture(s) < 0){ 01249 return -1; 01250 } 01251 01252 /* output? */ 01253 if(s->new_picture.data[0]){ 01254 s->pict_type= s->new_picture.pict_type; 01255 //emms_c(); 01256 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale); 01257 MPV_frame_start(s, avctx); 01258 vbv_retry: 01259 if (encode_picture(s, s->picture_number) < 0) 01260 return -1; 01261 01262 avctx->header_bits = s->header_bits; 01263 avctx->mv_bits = s->mv_bits; 01264 avctx->misc_bits = s->misc_bits; 01265 avctx->i_tex_bits = s->i_tex_bits; 01266 avctx->p_tex_bits = s->p_tex_bits; 01267 avctx->i_count = s->i_count; 01268 avctx->p_count = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx 01269 avctx->skip_count = s->skip_count; 01270 01271 MPV_frame_end(s); 01272 01273 if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG) 01274 ff_mjpeg_encode_picture_trailer(s); 01275 01276 if(avctx->rc_buffer_size){ 01277 RateControlContext *rcc= &s->rc_context; 01278 int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use; 01279 01280 if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){ 01281 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale); 01282 if(s->adaptive_quant){ 01283 int i; 01284 for(i=0; i<s->mb_height*s->mb_stride; i++) 01285 s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale); 01286 } 01287 s->mb_skipped = 0; //done in MPV_frame_start() 01288 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it 01289 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4) 01290 s->no_rounding ^= 1; 01291 } 01292 if(s->pict_type!=FF_B_TYPE){ 01293 s->time_base= s->last_time_base; 01294 s->last_non_b_time= s->time - s->pp_time; 01295 } 01296 // av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda); 01297 for(i=0; i<avctx->thread_count; i++){ 01298 PutBitContext *pb= &s->thread_context[i]->pb; 01299 init_put_bits(pb, pb->buf, pb->buf_end - pb->buf); 01300 } 01301 goto vbv_retry; 01302 } 01303 01304 assert(s->avctx->rc_max_rate); 01305 } 01306 01307 if(s->flags&CODEC_FLAG_PASS1) 01308 ff_write_pass1_stats(s); 01309 01310 for(i=0; i<4; i++){ 01311 s->current_picture_ptr->error[i]= s->current_picture.error[i]; 01312 avctx->error[i] += s->current_picture_ptr->error[i]; 01313 } 01314 01315 if(s->flags&CODEC_FLAG_PASS1) 01316 assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb)); 01317 flush_put_bits(&s->pb); 01318 s->frame_bits = put_bits_count(&s->pb); 01319 01320 stuffing_count= ff_vbv_update(s, s->frame_bits); 01321 if(stuffing_count){ 01322 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){ 01323 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n"); 01324 return -1; 01325 } 01326 01327 switch(s->codec_id){ 01328 case CODEC_ID_MPEG1VIDEO: 01329 case CODEC_ID_MPEG2VIDEO: 01330 while(stuffing_count--){ 01331 put_bits(&s->pb, 8, 0); 01332 } 01333 break; 01334 case CODEC_ID_MPEG4: 01335 put_bits(&s->pb, 16, 0); 01336 put_bits(&s->pb, 16, 0x1C3); 01337 stuffing_count -= 4; 01338 while(stuffing_count--){ 01339 put_bits(&s->pb, 8, 0xFF); 01340 } 01341 break; 01342 default: 01343 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n"); 01344 } 01345 flush_put_bits(&s->pb); 01346 s->frame_bits = put_bits_count(&s->pb); 01347 } 01348 01349 /* update mpeg1/2 vbv_delay for CBR */ 01350 if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1 01351 && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){ 01352 int vbv_delay, min_delay; 01353 double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base); 01354 int minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1); 01355 double bits = s->rc_context.buffer_index + minbits - inbits; 01356 01357 if(bits<0) 01358 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n"); 01359 01360 assert(s->repeat_first_field==0); 01361 01362 vbv_delay= bits * 90000 / s->avctx->rc_max_rate; 01363 min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate; 01364 01365 vbv_delay= FFMAX(vbv_delay, min_delay); 01366 01367 assert(vbv_delay < 0xFFFF); 01368 01369 s->vbv_delay_ptr[0] &= 0xF8; 01370 s->vbv_delay_ptr[0] |= vbv_delay>>13; 01371 s->vbv_delay_ptr[1] = vbv_delay>>5; 01372 s->vbv_delay_ptr[2] &= 0x07; 01373 s->vbv_delay_ptr[2] |= vbv_delay<<3; 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 /* Are all we could set to zero already zero? */ 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; //skip clipping of intra dc 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 //FIXME optimize 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; //default for progressive frames 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 == FF_B_TYPE){ 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 ff_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 ff_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 ff_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){ /* 422 */ 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==FF_B_TYPE){ 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){ /* 422 */ 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 /* pre quantization */ 01654 if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){ 01655 //FIXME optimize 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){ /* 422 */ 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){ /* 422 */ 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 /* DCT & quantize */ 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 // FIXME we could decide to change to quantizer instead of clipping 01691 // JS: I don't think that would be a good idea it could lower quality instead 01692 // of improve it. Just INTRADC clipping deserves changes in quantizer 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 //non c quantize code returns incorrect block_last_index FIXME 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 /* huffman encode */ 01741 switch(s->codec_id){ //FIXME funct ptr could be slightly faster 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)); //FIXME is memcpy faster then a loop? 01792 01793 /* mpeg1 */ 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 /* statistics */ 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)); //FIXME is memcpy faster then a loop? 01821 01822 /* mpeg1 */ 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 /* statistics */ 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); //FIXME 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; //for block init below 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 /* compute motion vector & mb_type and store in context */ 01986 if(s->pict_type==FF_B_TYPE) 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 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y); 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 /* init last dc values */ 02070 /* note: quant matrix value (8) is implied here */ 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 // printf("row %d at %X\n", s->mb_y, (int)s); 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; // removed const, H261 needs to adjust this 02107 int mb_type= s->mb_type[xy]; 02108 // int d; 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; // moved into loop, can get changed by H.261 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 /* write gob / video packet header */ 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; //only for QP_RD 02227 02228 if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or 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); //old mode? 02362 } 02363 } 02364 02365 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){ 02366 if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD 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 //FIXME intra 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 == FF_B_TYPE ? 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 /* wrong but unused */, 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; //FIXME find the one we actually used 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 /* s->mv[0][0][0] = best_s.mv[0][0][0]; 02452 s->mv[0][0][1] = best_s.mv[0][0][1]; 02453 s->mv[1][0][0] = best_s.mv[1][0][0]; 02454 s->mv[1][0][1] = best_s.mv[1][0][1];*/ 02455 backup_s.dquant= 0; 02456 s->skipdct=1; 02457 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, 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!=FF_B_TYPE) 02487 ff_h263_update_motion_val(s); 02488 02489 if(next_block==0){ //FIXME 16 vs linesize16 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 // only one MB-Type possible 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 // printf(" %d %d ", motion_x, motion_y); 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 // RAL: Update last macroblock type 02610 s->last_mv_dir = s->mv_dir; 02611 02612 if (CONFIG_H263_ENCODER && 02613 s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE) 02614 ff_h263_update_motion_val(s); 02615 02616 MPV_decode_mb(s, s->block); 02617 } 02618 02619 /* clean the MV table in IPS frames for direct mode in B frames */ 02620 if(s->mb_intra /* && I,P,S_TYPE */){ 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 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb)); 02647 } 02648 } 02649 02650 //not beautiful here but we must write it before flushing so it has to be here 02651 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE) 02652 msmpeg4_encode_ext_header(s); 02653 02654 write_slice_end(s); 02655 02656 /* Send the last GOB if RTP */ 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 /* Call the RTP callback to send the last GOB */ 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]); //note, the other dct vars are not part of the context 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 //FIXME broken 02737 }else 02738 s->lambda= s->current_picture.quality; 02739 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality); 02740 update_qscale(s); 02741 return 0; 02742 } 02743 02744 /* must be called before writing the header */ 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==FF_B_TYPE){ 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 02764 s->picture_number = picture_number; 02765 02766 /* Reset the average MB variance */ 02767 s->me.mb_var_sum_temp = 02768 s->me.mc_mb_var_sum_temp = 0; 02769 02770 /* we need to initialize some time vars before we can encode b-frames */ 02771 // RAL: Condition added for MPEG1VIDEO 02772 if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4)) 02773 set_frame_distances(s); 02774 if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4) 02775 ff_set_mpeg4_time(s); 02776 02777 s->me.scene_change_score=0; 02778 02779 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion 02780 02781 if(s->pict_type==FF_I_TYPE){ 02782 if(s->msmpeg4_version >= 3) s->no_rounding=1; 02783 else s->no_rounding=0; 02784 }else if(s->pict_type!=FF_B_TYPE){ 02785 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4) 02786 s->no_rounding ^= 1; 02787 } 02788 02789 if(s->flags & CODEC_FLAG_PASS2){ 02790 if (estimate_qp(s,1) < 0) 02791 return -1; 02792 ff_get_2pass_fcode(s); 02793 }else if(!(s->flags & CODEC_FLAG_QSCALE)){ 02794 if(s->pict_type==FF_B_TYPE) 02795 s->lambda= s->last_lambda_for[s->pict_type]; 02796 else 02797 s->lambda= s->last_lambda_for[s->last_non_b_pict_type]; 02798 update_qscale(s); 02799 } 02800 02801 s->mb_intra=0; //for the rate distortion & bit compare functions 02802 for(i=1; i<s->avctx->thread_count; i++){ 02803 ff_update_duplicate_context(s->thread_context[i], s); 02804 } 02805 02806 if(ff_init_me(s)<0) 02807 return -1; 02808 02809 /* Estimate motion for every MB */ 02810 if(s->pict_type != FF_I_TYPE){ 02811 s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8; 02812 s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8; 02813 if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){ 02814 if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){ 02815 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*)); 02816 } 02817 } 02818 02819 s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*)); 02820 }else /* if(s->pict_type == FF_I_TYPE) */{ 02821 /* I-Frame */ 02822 for(i=0; i<s->mb_stride*s->mb_height; i++) 02823 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA; 02824 02825 if(!s->fixed_qscale){ 02826 /* finding spatial complexity for I-frame rate control */ 02827 s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*)); 02828 } 02829 } 02830 for(i=1; i<s->avctx->thread_count; i++){ 02831 merge_context_after_me(s, s->thread_context[i]); 02832 } 02833 s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp; 02834 s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp; 02835 emms_c(); 02836 02837 if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){ 02838 s->pict_type= FF_I_TYPE; 02839 for(i=0; i<s->mb_stride*s->mb_height; i++) 02840 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA; 02841 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum); 02842 } 02843 02844 if(!s->umvplus){ 02845 if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) { 02846 s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER); 02847 02848 if(s->flags & CODEC_FLAG_INTERLACED_ME){ 02849 int a,b; 02850 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select 02851 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I); 02852 s->f_code= FFMAX3(s->f_code, a, b); 02853 } 02854 02855 ff_fix_long_p_mvs(s); 02856 ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0); 02857 if(s->flags & CODEC_FLAG_INTERLACED_ME){ 02858 int j; 02859 for(i=0; i<2; i++){ 02860 for(j=0; j<2; j++) 02861 ff_fix_long_mvs(s, s->p_field_select_table[i], j, 02862 s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0); 02863 } 02864 } 02865 } 02866 02867 if(s->pict_type==FF_B_TYPE){ 02868 int a, b; 02869 02870 a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD); 02871 b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR); 02872 s->f_code = FFMAX(a, b); 02873 02874 a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD); 02875 b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR); 02876 s->b_code = FFMAX(a, b); 02877 02878 ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1); 02879 ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1); 02880 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1); 02881 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1); 02882 if(s->flags & CODEC_FLAG_INTERLACED_ME){ 02883 int dir, j; 02884 for(dir=0; dir<2; dir++){ 02885 for(i=0; i<2; i++){ 02886 for(j=0; j<2; j++){ 02887 int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I) 02888 : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I); 02889 ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j, 02890 s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1); 02891 } 02892 } 02893 } 02894 } 02895 } 02896 } 02897 02898 if (estimate_qp(s, 0) < 0) 02899 return -1; 02900 02901 if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE)) 02902 s->qscale= 3; //reduce clipping problems 02903 02904 if (s->out_format == FMT_MJPEG) { 02905 /* for mjpeg, we do include qscale in the matrix */ 02906 for(i=1;i<64;i++){ 02907 int j= s->dsp.idct_permutation[i]; 02908 02909 s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3); 02910 } 02911 s->y_dc_scale_table= 02912 s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision]; 02913 s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8]; 02914 ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16, 02915 s->intra_matrix, s->intra_quant_bias, 8, 8, 1); 02916 s->qscale= 8; 02917 } 02918 02919 //FIXME var duplication 02920 s->current_picture_ptr->key_frame= 02921 s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr 02922 s->current_picture_ptr->pict_type= 02923 s->current_picture.pict_type= s->pict_type; 02924 02925 if(s->current_picture.key_frame) 02926 s->picture_in_gop_number=0; 02927 02928 s->last_bits= put_bits_count(&s->pb); 02929 switch(s->out_format) { 02930 case FMT_MJPEG: 02931 if (CONFIG_MJPEG_ENCODER) 02932 ff_mjpeg_encode_picture_header(s); 02933 break; 02934 case FMT_H261: 02935 if (CONFIG_H261_ENCODER) 02936 ff_h261_encode_picture_header(s, picture_number); 02937 break; 02938 case FMT_H263: 02939 if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2) 02940 ff_wmv2_encode_picture_header(s, picture_number); 02941 else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4) 02942 msmpeg4_encode_picture_header(s, picture_number); 02943 else if (CONFIG_MPEG4_ENCODER && s->h263_pred) 02944 mpeg4_encode_picture_header(s, picture_number); 02945 else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10) 02946 rv10_encode_picture_header(s, picture_number); 02947 else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20) 02948 rv20_encode_picture_header(s, picture_number); 02949 else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1) 02950 ff_flv_encode_picture_header(s, picture_number); 02951 else if (CONFIG_H263_ENCODER) 02952 h263_encode_picture_header(s, picture_number); 02953 break; 02954 case FMT_MPEG1: 02955 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) 02956 mpeg1_encode_picture_header(s, picture_number); 02957 break; 02958 case FMT_H264: 02959 break; 02960 default: 02961 assert(0); 02962 } 02963 bits= put_bits_count(&s->pb); 02964 s->header_bits= bits - s->last_bits; 02965 02966 for(i=1; i<s->avctx->thread_count; i++){ 02967 update_duplicate_context_after_me(s->thread_context[i], s); 02968 } 02969 s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*)); 02970 for(i=1; i<s->avctx->thread_count; i++){ 02971 merge_context_after_encode(s, s->thread_context[i]); 02972 } 02973 emms_c(); 02974 return 0; 02975 } 02976 02977 void denoise_dct_c(MpegEncContext *s, DCTELEM *block){ 02978 const int intra= s->mb_intra; 02979 int i; 02980 02981 s->dct_count[intra]++; 02982 02983 for(i=0; i<64; i++){ 02984 int level= block[i]; 02985 02986 if(level){ 02987 if(level>0){ 02988 s->dct_error_sum[intra][i] += level; 02989 level -= s->dct_offset[intra][i]; 02990 if(level<0) level=0; 02991 }else{ 02992 s->dct_error_sum[intra][i] -= level; 02993 level += s->dct_offset[intra][i]; 02994 if(level>0) level=0; 02995 } 02996 block[i]= level; 02997 } 02998 } 02999 } 03000 03001 int dct_quantize_trellis_c(MpegEncContext *s, 03002 DCTELEM *block, int n, 03003 int qscale, int *overflow){ 03004 const int *qmat; 03005 const uint8_t *scantable= s->intra_scantable.scantable; 03006 const uint8_t *perm_scantable= s->intra_scantable.permutated; 03007 int max=0; 03008 unsigned int threshold1, threshold2; 03009 int bias=0; 03010 int run_tab[65]; 03011 int level_tab[65]; 03012 int score_tab[65]; 03013 int survivor[65]; 03014 int survivor_count; 03015 int last_run=0; 03016 int last_level=0; 03017 int last_score= 0; 03018 int last_i; 03019 int coeff[2][64]; 03020 int coeff_count[64]; 03021 int qmul, qadd, start_i, last_non_zero, i, dc; 03022 const int esc_length= s->ac_esc_length; 03023 uint8_t * length; 03024 uint8_t * last_length; 03025 const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6); 03026 03027 s->dsp.fdct (block); 03028 03029 if(s->dct_error_sum) 03030 s->denoise_dct(s, block); 03031 qmul= qscale*16; 03032 qadd= ((qscale-1)|1)*8; 03033 03034 if (s->mb_intra) { 03035 int q; 03036 if (!s->h263_aic) { 03037 if (n < 4) 03038 q = s->y_dc_scale; 03039 else 03040 q = s->c_dc_scale; 03041 q = q << 3; 03042 } else{ 03043 /* For AIC we skip quant/dequant of INTRADC */ 03044 q = 1 << 3; 03045 qadd=0; 03046 } 03047 03048 /* note: block[0] is assumed to be positive */ 03049 block[0] = (block[0] + (q >> 1)) / q; 03050 start_i = 1; 03051 last_non_zero = 0; 03052 qmat = s->q_intra_matrix[qscale]; 03053 if(s->mpeg_quant || s->out_format == FMT_MPEG1) 03054 bias= 1<<(QMAT_SHIFT-1); 03055 length = s->intra_ac_vlc_length; 03056 last_length= s->intra_ac_vlc_last_length; 03057 } else { 03058 start_i = 0; 03059 last_non_zero = -1; 03060 qmat = s->q_inter_matrix[qscale]; 03061 length = s->inter_ac_vlc_length; 03062 last_length= s->inter_ac_vlc_last_length; 03063 } 03064 last_i= start_i; 03065 03066 threshold1= (1<<QMAT_SHIFT) - bias - 1; 03067 threshold2= (threshold1<<1); 03068 03069 for(i=63; i>=start_i; i--) { 03070 const int j = scantable[i]; 03071 int level = block[j] * qmat[j]; 03072 03073 if(((unsigned)(level+threshold1))>threshold2){ 03074 last_non_zero = i; 03075 break; 03076 } 03077 } 03078 03079 for(i=start_i; i<=last_non_zero; i++) { 03080 const int j = scantable[i]; 03081 int level = block[j] * qmat[j]; 03082 03083 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) 03084 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ 03085 if(((unsigned)(level+threshold1))>threshold2){ 03086 if(level>0){ 03087 level= (bias + level)>>QMAT_SHIFT; 03088 coeff[0][i]= level; 03089 coeff[1][i]= level-1; 03090 // coeff[2][k]= level-2; 03091 }else{ 03092 level= (bias - level)>>QMAT_SHIFT; 03093 coeff[0][i]= -level; 03094 coeff[1][i]= -level+1; 03095 // coeff[2][k]= -level+2; 03096 } 03097 coeff_count[i]= FFMIN(level, 2); 03098 assert(coeff_count[i]); 03099 max |=level; 03100 }else{ 03101 coeff[0][i]= (level>>31)|1; 03102 coeff_count[i]= 1; 03103 } 03104 } 03105 03106 *overflow= s->max_qcoeff < max; //overflow might have happened 03107 03108 if(last_non_zero < start_i){ 03109 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM)); 03110 return last_non_zero; 03111 } 03112 03113 score_tab[start_i]= 0; 03114 survivor[0]= start_i; 03115 survivor_count= 1; 03116 03117 for(i=start_i; i<=last_non_zero; i++){ 03118 int level_index, j, zero_distortion; 03119 int dct_coeff= FFABS(block[ scantable[i] ]); 03120 int best_score=256*256*256*120; 03121 03122 if ( s->dsp.fdct == fdct_ifast 03123 #ifndef FAAN_POSTSCALE 03124 || s->dsp.fdct == ff_faandct 03125 #endif 03126 ) 03127 dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12; 03128 zero_distortion= dct_coeff*dct_coeff; 03129 03130 for(level_index=0; level_index < coeff_count[i]; level_index++){ 03131 int distortion; 03132 int level= coeff[level_index][i]; 03133 const int alevel= FFABS(level); 03134 int unquant_coeff; 03135 03136 assert(level); 03137 03138 if(s->out_format == FMT_H263){ 03139 unquant_coeff= alevel*qmul + qadd; 03140 }else{ //MPEG1 03141 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize 03142 if(s->mb_intra){ 03143 unquant_coeff = (int)( alevel * qscale * s->intra_matrix[j]) >> 3; 03144 unquant_coeff = (unquant_coeff - 1) | 1; 03145 }else{ 03146 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4; 03147 unquant_coeff = (unquant_coeff - 1) | 1; 03148 } 03149 unquant_coeff<<= 3; 03150 } 03151 03152 distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion; 03153 level+=64; 03154 if((level&(~127)) == 0){ 03155 for(j=survivor_count-1; j>=0; j--){ 03156 int run= i - survivor[j]; 03157 int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda; 03158 score += score_tab[i-run]; 03159 03160 if(score < best_score){ 03161 best_score= score; 03162 run_tab[i+1]= run; 03163 level_tab[i+1]= level-64; 03164 } 03165 } 03166 03167 if(s->out_format == FMT_H263){ 03168 for(j=survivor_count-1; j>=0; j--){ 03169 int run= i - survivor[j]; 03170 int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda; 03171 score += score_tab[i-run]; 03172 if(score < last_score){ 03173 last_score= score; 03174 last_run= run; 03175 last_level= level-64; 03176 last_i= i+1; 03177 } 03178 } 03179 } 03180 }else{ 03181 distortion += esc_length*lambda; 03182 for(j=survivor_count-1; j>=0; j--){ 03183 int run= i - survivor[j]; 03184 int score= distortion + score_tab[i-run]; 03185 03186 if(score < best_score){ 03187 best_score= score; 03188 run_tab[i+1]= run; 03189 level_tab[i+1]= level-64; 03190 } 03191 } 03192 03193 if(s->out_format == FMT_H263){ 03194 for(j=survivor_count-1; j>=0; j--){ 03195 int run= i - survivor[j]; 03196 int score= distortion + score_tab[i-run]; 03197 if(score < last_score){ 03198 last_score= score; 03199 last_run= run; 03200 last_level= level-64; 03201 last_i= i+1; 03202 } 03203 } 03204 } 03205 } 03206 } 03207 03208 score_tab[i+1]= best_score; 03209 03210 //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level 03211 if(last_non_zero <= 27){ 03212 for(; survivor_count; survivor_count--){ 03213 if(score_tab[ survivor[survivor_count-1] ] <= best_score) 03214 break; 03215 } 03216 }else{ 03217 for(; survivor_count; survivor_count--){ 03218 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda) 03219 break; 03220 } 03221 } 03222 03223 survivor[ survivor_count++ ]= i+1; 03224 } 03225 03226 if(s->out_format != FMT_H263){ 03227 last_score= 256*256*256*120; 03228 for(i= survivor[0]; i<=last_non_zero + 1; i++){ 03229 int score= score_tab[i]; 03230 if(i) score += lambda*2; //FIXME exacter? 03231 03232 if(score < last_score){ 03233 last_score= score; 03234 last_i= i; 03235 last_level= level_tab[i]; 03236 last_run= run_tab[i]; 03237 } 03238 } 03239 } 03240 03241 s->coded_score[n] = last_score; 03242 03243 dc= FFABS(block[0]); 03244 last_non_zero= last_i - 1; 03245 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM)); 03246 03247 if(last_non_zero < start_i) 03248 return last_non_zero; 03249 03250 if(last_non_zero == 0 && start_i == 0){ 03251 int best_level= 0; 03252 int best_score= dc * dc; 03253 03254 for(i=0; i<coeff_count[0]; i++){ 03255 int level= coeff[i][0]; 03256 int alevel= FFABS(level); 03257 int unquant_coeff, score, distortion; 03258 03259 if(s->out_format == FMT_H263){ 03260 unquant_coeff= (alevel*qmul + qadd)>>3; 03261 }else{ //MPEG1 03262 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4; 03263 unquant_coeff = (unquant_coeff - 1) | 1; 03264 } 03265 unquant_coeff = (unquant_coeff + 4) >> 3; 03266 unquant_coeff<<= 3 + 3; 03267 03268 distortion= (unquant_coeff - dc) * (unquant_coeff - dc); 03269 level+=64; 03270 if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda; 03271 else score= distortion + esc_length*lambda; 03272 03273 if(score < best_score){ 03274 best_score= score; 03275 best_level= level - 64; 03276 } 03277 } 03278 block[0]= best_level; 03279 s->coded_score[n] = best_score - dc*dc; 03280 if(best_level == 0) return -1; 03281 else return last_non_zero; 03282 } 03283 03284 i= last_i; 03285 assert(last_level); 03286 03287 block[ perm_scantable[last_non_zero] ]= last_level; 03288 i -= last_run + 1; 03289 03290 for(; i>start_i; i -= run_tab[i] + 1){ 03291 block[ perm_scantable[i-1] ]= level_tab[i]; 03292 } 03293 03294 return last_non_zero; 03295 } 03296 03297 //#define REFINE_STATS 1 03298 static int16_t basis[64][64]; 03299 03300 static void build_basis(uint8_t *perm){ 03301 int i, j, x, y; 03302 emms_c(); 03303 for(i=0; i<8; i++){ 03304 for(j=0; j<8; j++){ 03305 for(y=0; y<8; y++){ 03306 for(x=0; x<8; x++){ 03307 double s= 0.25*(1<<BASIS_SHIFT); 03308 int index= 8*i + j; 03309 int perm_index= perm[index]; 03310 if(i==0) s*= sqrt(0.5); 03311 if(j==0) s*= sqrt(0.5); 03312 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))); 03313 } 03314 } 03315 } 03316 } 03317 } 03318 03319 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise? 03320 DCTELEM *block, int16_t *weight, DCTELEM *orig, 03321 int n, int qscale){ 03322 int16_t rem[64]; 03323 LOCAL_ALIGNED_16(DCTELEM, d1, [64]); 03324 const uint8_t *scantable= s->intra_scantable.scantable; 03325 const uint8_t *perm_scantable= s->intra_scantable.permutated; 03326 // unsigned int threshold1, threshold2; 03327 // int bias=0; 03328 int run_tab[65]; 03329 int prev_run=0; 03330 int prev_level=0; 03331 int qmul, qadd, start_i, last_non_zero, i, dc; 03332 uint8_t * length; 03333 uint8_t * last_length; 03334 int lambda; 03335 int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true 03336 #ifdef REFINE_STATS 03337 static int count=0; 03338 static int after_last=0; 03339 static int to_zero=0; 03340 static int from_zero=0; 03341 static int raise=0; 03342 static int lower=0; 03343 static int messed_sign=0; 03344 #endif 03345 03346 if(basis[0][0] == 0) 03347 build_basis(s->dsp.idct_permutation); 03348 03349 qmul= qscale*2; 03350 qadd= (qscale-1)|1; 03351 if (s->mb_intra) { 03352 if (!s->h263_aic) { 03353 if (n < 4) 03354 q = s->y_dc_scale; 03355 else 03356 q = s->c_dc_scale; 03357 } else{ 03358 /* For AIC we skip quant/dequant of INTRADC */ 03359 q = 1; 03360 qadd=0; 03361 } 03362 q <<= RECON_SHIFT-3; 03363 /* note: block[0] is assumed to be positive */ 03364 dc= block[0]*q; 03365 // block[0] = (block[0] + (q >> 1)) / q; 03366 start_i = 1; 03367 // if(s->mpeg_quant || s->out_format == FMT_MPEG1) 03368 // bias= 1<<(QMAT_SHIFT-1); 03369 length = s->intra_ac_vlc_length; 03370 last_length= s->intra_ac_vlc_last_length; 03371 } else { 03372 dc= 0; 03373 start_i = 0; 03374 length = s->inter_ac_vlc_length; 03375 last_length= s->inter_ac_vlc_last_length; 03376 } 03377 last_non_zero = s->block_last_index[n]; 03378 03379 #ifdef REFINE_STATS 03380 {START_TIMER 03381 #endif 03382 dc += (1<<(RECON_SHIFT-1)); 03383 for(i=0; i<64; i++){ 03384 rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME use orig dirrectly instead of copying to rem[] 03385 } 03386 #ifdef REFINE_STATS 03387 STOP_TIMER("memset rem[]")} 03388 #endif 03389 sum=0; 03390 for(i=0; i<64; i++){ 03391 int one= 36; 03392 int qns=4; 03393 int w; 03394 03395 w= FFABS(weight[i]) + qns*one; 03396 w= 15 + (48*qns*one + w/2)/w; // 16 .. 63 03397 03398 weight[i] = w; 03399 // w=weight[i] = (63*qns + (w/2)) / w; 03400 03401 assert(w>0); 03402 assert(w<(1<<6)); 03403 sum += w*w; 03404 } 03405 lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6); 03406 #ifdef REFINE_STATS 03407 {START_TIMER 03408 #endif 03409 run=0; 03410 rle_index=0; 03411 for(i=start_i; i<=last_non_zero; i++){ 03412 int j= perm_scantable[i]; 03413 const int level= block[j]; 03414 int coeff; 03415 03416 if(level){ 03417 if(level<0) coeff= qmul*level - qadd; 03418 else coeff= qmul*level + qadd; 03419 run_tab[rle_index++]=run; 03420 run=0; 03421 03422 s->dsp.add_8x8basis(rem, basis[j], coeff); 03423 }else{ 03424 run++; 03425 } 03426 } 03427 #ifdef REFINE_STATS 03428 if(last_non_zero>0){ 03429 STOP_TIMER("init rem[]") 03430 } 03431 } 03432 03433 {START_TIMER 03434 #endif 03435 for(;;){ 03436 int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0); 03437 int best_coeff=0; 03438 int best_change=0; 03439 int run2, best_unquant_change=0, analyze_gradient; 03440 #ifdef REFINE_STATS 03441 {START_TIMER 03442 #endif 03443 analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3; 03444 03445 if(analyze_gradient){ 03446 #ifdef REFINE_STATS 03447 {START_TIMER 03448 #endif 03449 for(i=0; i<64; i++){ 03450 int w= weight[i]; 03451 03452 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12); 03453 } 03454 #ifdef REFINE_STATS 03455 STOP_TIMER("rem*w*w")} 03456 {START_TIMER 03457 #endif 03458 s->dsp.fdct(d1); 03459 #ifdef REFINE_STATS 03460 STOP_TIMER("dct")} 03461 #endif 03462 } 03463 03464 if(start_i){ 03465 const int level= block[0]; 03466 int change, old_coeff; 03467 03468 assert(s->mb_intra); 03469 03470 old_coeff= q*level; 03471 03472 for(change=-1; change<=1; change+=2){ 03473 int new_level= level + change; 03474 int score, new_coeff; 03475 03476 new_coeff= q*new_level; 03477 if(new_coeff >= 2048 || new_coeff < 0) 03478 continue; 03479 03480 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff); 03481 if(score<best_score){ 03482 best_score= score; 03483 best_coeff= 0; 03484 best_change= change; 03485 best_unquant_change= new_coeff - old_coeff; 03486 } 03487 } 03488 } 03489 03490 run=0; 03491 rle_index=0; 03492 run2= run_tab[rle_index++]; 03493 prev_level=0; 03494 prev_run=0; 03495 03496 for(i=start_i; i<64; i++){ 03497 int j= perm_scantable[i]; 03498 const int level= block[j]; 03499 int change, old_coeff; 03500 03501 if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1) 03502 break; 03503 03504 if(level){ 03505 if(level<0) old_coeff= qmul*level - qadd; 03506 else old_coeff= qmul*level + qadd; 03507 run2= run_tab[rle_index++]; //FIXME ! maybe after last 03508 }else{ 03509 old_coeff=0; 03510 run2--; 03511 assert(run2>=0 || i >= last_non_zero ); 03512 } 03513 03514 for(change=-1; change<=1; change+=2){ 03515 int new_level= level + change; 03516 int score, new_coeff, unquant_change; 03517 03518 score=0; 03519 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level)) 03520 continue; 03521 03522 if(new_level){ 03523 if(new_level<0) new_coeff= qmul*new_level - qadd; 03524 else new_coeff= qmul*new_level + qadd; 03525 if(new_coeff >= 2048 || new_coeff <= -2048) 03526 continue; 03527 //FIXME check for overflow 03528 03529 if(level){ 03530 if(level < 63 && level > -63){ 03531 if(i < last_non_zero) 03532 score += length[UNI_AC_ENC_INDEX(run, new_level+64)] 03533 - length[UNI_AC_ENC_INDEX(run, level+64)]; 03534 else 03535 score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)] 03536 - last_length[UNI_AC_ENC_INDEX(run, level+64)]; 03537 } 03538 }else{ 03539 assert(FFABS(new_level)==1); 03540 03541 if(analyze_gradient){ 03542 int g= d1[ scantable[i] ]; 03543 if(g && (g^new_level) >= 0) 03544 continue; 03545 } 03546 03547 if(i < last_non_zero){ 03548 int next_i= i + run2 + 1; 03549 int next_level= block[ perm_scantable[next_i] ] + 64; 03550 03551 if(next_level&(~127)) 03552 next_level= 0; 03553 03554 if(next_i < last_non_zero) 03555 score += length[UNI_AC_ENC_INDEX(run, 65)] 03556 + length[UNI_AC_ENC_INDEX(run2, next_level)] 03557 - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]; 03558 else 03559 score += length[UNI_AC_ENC_INDEX(run, 65)] 03560 + last_length[UNI_AC_ENC_INDEX(run2, next_level)] 03561 - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]; 03562 }else{ 03563 score += last_length[UNI_AC_ENC_INDEX(run, 65)]; 03564 if(prev_level){ 03565 score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)] 03566 - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]; 03567 } 03568 } 03569 } 03570 }else{ 03571 new_coeff=0; 03572 assert(FFABS(level)==1); 03573 03574 if(i < last_non_zero){ 03575 int next_i= i + run2 + 1; 03576 int next_level= block[ perm_scantable[next_i] ] + 64; 03577 03578 if(next_level&(~127)) 03579 next_level= 0; 03580 03581 if(next_i < last_non_zero) 03582 score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)] 03583 - length[UNI_AC_ENC_INDEX(run2, next_level)] 03584 - length[UNI_AC_ENC_INDEX(run, 65)]; 03585 else 03586 score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)] 03587 - last_length[UNI_AC_ENC_INDEX(run2, next_level)] 03588 - length[UNI_AC_ENC_INDEX(run, 65)]; 03589 }else{ 03590 score += -last_length[UNI_AC_ENC_INDEX(run, 65)]; 03591 if(prev_level){ 03592 score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)] 03593 - length[UNI_AC_ENC_INDEX(prev_run, prev_level)]; 03594 } 03595 } 03596 } 03597 03598 score *= lambda; 03599 03600 unquant_change= new_coeff - old_coeff; 03601 assert((score < 100*lambda && score > -100*lambda) || lambda==0); 03602 03603 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change); 03604 if(score<best_score){ 03605 best_score= score; 03606 best_coeff= i; 03607 best_change= change; 03608 best_unquant_change= unquant_change; 03609 } 03610 } 03611 if(level){ 03612 prev_level= level + 64; 03613 if(prev_level&(~127)) 03614 prev_level= 0; 03615 prev_run= run; 03616 run=0; 03617 }else{ 03618 run++; 03619 } 03620 } 03621 #ifdef REFINE_STATS 03622 STOP_TIMER("iterative step")} 03623 #endif 03624 03625 if(best_change){ 03626 int j= perm_scantable[ best_coeff ]; 03627 03628 block[j] += best_change; 03629 03630 if(best_coeff > last_non_zero){ 03631 last_non_zero= best_coeff; 03632 assert(block[j]); 03633 #ifdef REFINE_STATS 03634 after_last++; 03635 #endif 03636 }else{ 03637 #ifdef REFINE_STATS 03638 if(block[j]){ 03639 if(block[j] - best_change){ 03640 if(FFABS(block[j]) > FFABS(block[j] - best_change)){ 03641 raise++; 03642 }else{ 03643 lower++; 03644 } 03645 }else{ 03646 from_zero++; 03647 } 03648 }else{ 03649 to_zero++; 03650 } 03651 #endif 03652 for(; last_non_zero>=start_i; last_non_zero--){ 03653 if(block[perm_scantable[last_non_zero]]) 03654 break; 03655 } 03656 } 03657 #ifdef REFINE_STATS 03658 count++; 03659 if(256*256*256*64 % count == 0){ 03660 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); 03661 } 03662 #endif 03663 run=0; 03664 rle_index=0; 03665 for(i=start_i; i<=last_non_zero; i++){ 03666 int j= perm_scantable[i]; 03667 const int level= block[j]; 03668 03669 if(level){ 03670 run_tab[rle_index++]=run; 03671 run=0; 03672 }else{ 03673 run++; 03674 } 03675 } 03676 03677 s->dsp.add_8x8basis(rem, basis[j], best_unquant_change); 03678 }else{ 03679 break; 03680 } 03681 } 03682 #ifdef REFINE_STATS 03683 if(last_non_zero>0){ 03684 STOP_TIMER("iterative search") 03685 } 03686 } 03687 #endif 03688 03689 return last_non_zero; 03690 } 03691 03692 int dct_quantize_c(MpegEncContext *s, 03693 DCTELEM *block, int n, 03694 int qscale, int *overflow) 03695 { 03696 int i, j, level, last_non_zero, q, start_i; 03697 const int *qmat; 03698 const uint8_t *scantable= s->intra_scantable.scantable; 03699 int bias; 03700 int max=0; 03701 unsigned int threshold1, threshold2; 03702 03703 s->dsp.fdct (block); 03704 03705 if(s->dct_error_sum) 03706 s->denoise_dct(s, block); 03707 03708 if (s->mb_intra) { 03709 if (!s->h263_aic) { 03710 if (n < 4) 03711 q = s->y_dc_scale; 03712 else 03713 q = s->c_dc_scale; 03714 q = q << 3; 03715 } else 03716 /* For AIC we skip quant/dequant of INTRADC */ 03717 q = 1 << 3; 03718 03719 /* note: block[0] is assumed to be positive */ 03720 block[0] = (block[0] + (q >> 1)) / q; 03721 start_i = 1; 03722 last_non_zero = 0; 03723 qmat = s->q_intra_matrix[qscale]; 03724 bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT); 03725 } else { 03726 start_i = 0; 03727 last_non_zero = -1; 03728 qmat = s->q_inter_matrix[qscale]; 03729 bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT); 03730 } 03731 threshold1= (1<<QMAT_SHIFT) - bias - 1; 03732 threshold2= (threshold1<<1); 03733 for(i=63;i>=start_i;i--) { 03734 j = scantable[i]; 03735 level = block[j] * qmat[j]; 03736 03737 if(((unsigned)(level+threshold1))>threshold2){ 03738 last_non_zero = i; 03739 break; 03740 }else{ 03741 block[j]=0; 03742 } 03743 } 03744 for(i=start_i; i<=last_non_zero; i++) { 03745 j = scantable[i]; 03746 level = block[j] * qmat[j]; 03747 03748 // if( bias+level >= (1<<QMAT_SHIFT) 03749 // || bias-level >= (1<<QMAT_SHIFT)){ 03750 if(((unsigned)(level+threshold1))>threshold2){ 03751 if(level>0){ 03752 level= (bias + level)>>QMAT_SHIFT; 03753 block[j]= level; 03754 }else{ 03755 level= (bias - level)>>QMAT_SHIFT; 03756 block[j]= -level; 03757 } 03758 max |=level; 03759 }else{ 03760 block[j]=0; 03761 } 03762 } 03763 *overflow= s->max_qcoeff < max; //overflow might have happened 03764 03765 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */ 03766 if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM) 03767 ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero); 03768 03769 return last_non_zero; 03770 } 03771 03772 AVCodec h263_encoder = { 03773 "h263", 03774 AVMEDIA_TYPE_VIDEO, 03775 CODEC_ID_H263, 03776 sizeof(MpegEncContext), 03777 MPV_encode_init, 03778 MPV_encode_picture, 03779 MPV_encode_end, 03780 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE}, 03781 .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"), 03782 }; 03783 03784 AVCodec h263p_encoder = { 03785 "h263p", 03786 AVMEDIA_TYPE_VIDEO, 03787 CODEC_ID_H263P, 03788 sizeof(MpegEncContext), 03789 MPV_encode_init, 03790 MPV_encode_picture, 03791 MPV_encode_end, 03792 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE}, 03793 .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"), 03794 }; 03795 03796 AVCodec msmpeg4v1_encoder = { 03797 "msmpeg4v1", 03798 AVMEDIA_TYPE_VIDEO, 03799 CODEC_ID_MSMPEG4V1, 03800 sizeof(MpegEncContext), 03801 MPV_encode_init, 03802 MPV_encode_picture, 03803 MPV_encode_end, 03804 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE}, 03805 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"), 03806 }; 03807 03808 AVCodec msmpeg4v2_encoder = { 03809 "msmpeg4v2", 03810 AVMEDIA_TYPE_VIDEO, 03811 CODEC_ID_MSMPEG4V2, 03812 sizeof(MpegEncContext), 03813 MPV_encode_init, 03814 MPV_encode_picture, 03815 MPV_encode_end, 03816 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE}, 03817 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"), 03818 }; 03819 03820 AVCodec msmpeg4v3_encoder = { 03821 "msmpeg4", 03822 AVMEDIA_TYPE_VIDEO, 03823 CODEC_ID_MSMPEG4V3, 03824 sizeof(MpegEncContext), 03825 MPV_encode_init, 03826 MPV_encode_picture, 03827 MPV_encode_end, 03828 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE}, 03829 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"), 03830 }; 03831 03832 AVCodec wmv1_encoder = { 03833 "wmv1", 03834 AVMEDIA_TYPE_VIDEO, 03835 CODEC_ID_WMV1, 03836 sizeof(MpegEncContext), 03837 MPV_encode_init, 03838 MPV_encode_picture, 03839 MPV_encode_end, 03840 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE}, 03841 .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"), 03842 };