2 * The simplest mpeg encoder (well, it was the simplest!)
3 * Copyright (c) 2000,2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 * The simplest mpeg encoder (well, it was the simplest!).
30 #include "libavutil/intmath.h"
33 #include "mpegvideo.h"
34 #include "mpegvideo_common.h"
40 #include "aandcttab.h"
42 #include "mpeg4video.h"
49 static int encode_picture(MpegEncContext *s, int picture_number);
50 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
51 static int sse_mb(MpegEncContext *s);
52 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
53 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
55 /* enable all paranoid tests for rounding, overflows, etc... */
60 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
61 static uint8_t default_fcode_tab[MAX_MV*2+1];
63 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
64 const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
69 for(qscale=qmin; qscale<=qmax; qscale++){
71 if (dsp->fdct == ff_jpeg_fdct_islow
73 || dsp->fdct == ff_faandct
77 const int j= dsp->idct_permutation[i];
78 /* 16 <= qscale * quant_matrix[i] <= 7905 */
79 /* 19952 <= ff_aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
80 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
81 /* 3444240 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
83 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
84 (qscale * quant_matrix[j]));
86 } else if (dsp->fdct == fdct_ifast
87 #ifndef FAAN_POSTSCALE
88 || dsp->fdct == ff_faandct
92 const int j= dsp->idct_permutation[i];
93 /* 16 <= qscale * quant_matrix[i] <= 7905 */
94 /* 19952 <= ff_aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
95 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
96 /* 3444240 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
98 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
99 (ff_aanscales[i] * qscale * quant_matrix[j]));
103 const int j= dsp->idct_permutation[i];
104 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
105 So 16 <= qscale * quant_matrix[i] <= 7905
106 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
107 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
109 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
110 // qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
111 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
113 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
114 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
118 for(i=intra; i<64; i++){
120 if (dsp->fdct == fdct_ifast
121 #ifndef FAAN_POSTSCALE
122 || dsp->fdct == ff_faandct
125 max = (8191LL*ff_aanscales[i]) >> 14;
127 while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
133 av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
137 static inline void update_qscale(MpegEncContext *s){
138 s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
139 s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
141 s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
144 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
150 put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
157 * init s->current_picture.qscale_table from s->lambda_table
159 void ff_init_qscale_tab(MpegEncContext *s){
160 int8_t * const qscale_table= s->current_picture.qscale_table;
163 for(i=0; i<s->mb_num; i++){
164 unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
165 int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
166 qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
170 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
173 dst->pict_type = src->pict_type;
174 dst->quality = src->quality;
175 dst->coded_picture_number = src->coded_picture_number;
176 dst->display_picture_number = src->display_picture_number;
177 // dst->reference = src->reference;
179 dst->interlaced_frame = src->interlaced_frame;
180 dst->top_field_first = src->top_field_first;
182 if(s->avctx->me_threshold){
183 if(!src->motion_val[0])
184 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
186 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
187 if(!src->ref_index[0])
188 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
189 if(src->motion_subsample_log2 != dst->motion_subsample_log2)
190 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
191 src->motion_subsample_log2, dst->motion_subsample_log2);
193 memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
196 int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
197 int height= ((16*s->mb_height)>>src->motion_subsample_log2);
199 if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
200 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
202 if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
203 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t));
209 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
210 #define COPY(a) dst->a= src->a
212 COPY(current_picture);
218 COPY(picture_in_gop_number);
219 COPY(gop_picture_number);
220 COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
221 COPY(progressive_frame); //FIXME don't set in encode_header
222 COPY(partitioned_frame); //FIXME don't set in encode_header
227 * sets the given MpegEncContext to defaults for encoding.
228 * the changed fields will not depend upon the prior state of the MpegEncContext.
230 static void MPV_encode_defaults(MpegEncContext *s){
232 MPV_common_defaults(s);
234 for(i=-16; i<16; i++){
235 default_fcode_tab[i + MAX_MV]= 1;
237 s->me.mv_penalty= default_mv_penalty;
238 s->fcode_tab= default_fcode_tab;
241 /* init video encoder */
242 av_cold int MPV_encode_init(AVCodecContext *avctx)
244 MpegEncContext *s = avctx->priv_data;
246 int chroma_h_shift, chroma_v_shift;
248 MPV_encode_defaults(s);
250 switch (avctx->codec_id) {
251 case CODEC_ID_MPEG2VIDEO:
252 if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
253 av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
258 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_YUVJ444P && avctx->pix_fmt != PIX_FMT_BGRA &&
259 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P && avctx->pix_fmt != PIX_FMT_YUV444P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
260 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
265 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
266 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
267 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
272 if(avctx->pix_fmt != PIX_FMT_YUV420P){
273 av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
278 switch (avctx->pix_fmt) {
279 case PIX_FMT_YUVJ422P:
280 case PIX_FMT_YUV422P:
281 s->chroma_format = CHROMA_422;
283 case PIX_FMT_YUVJ420P:
284 case PIX_FMT_YUV420P:
286 s->chroma_format = CHROMA_420;
290 s->bit_rate = avctx->bit_rate;
291 s->width = avctx->width;
292 s->height = avctx->height;
293 if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
294 av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
297 s->gop_size = avctx->gop_size;
299 s->flags= avctx->flags;
300 s->flags2= avctx->flags2;
301 s->max_b_frames= avctx->max_b_frames;
302 s->codec_id= avctx->codec->id;
303 s->luma_elim_threshold = avctx->luma_elim_threshold;
304 s->chroma_elim_threshold= avctx->chroma_elim_threshold;
305 s->strict_std_compliance= avctx->strict_std_compliance;
306 s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
307 s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
308 s->mpeg_quant= avctx->mpeg_quant;
309 s->rtp_mode= !!avctx->rtp_payload_size;
310 s->intra_dc_precision= avctx->intra_dc_precision;
311 s->user_specified_pts = AV_NOPTS_VALUE;
313 if (s->gop_size <= 1) {
320 s->me_method = avctx->me_method;
323 s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
325 s->adaptive_quant= ( s->avctx->lumi_masking
326 || s->avctx->dark_masking
327 || s->avctx->temporal_cplx_masking
328 || s->avctx->spatial_cplx_masking
329 || s->avctx->p_masking
330 || s->avctx->border_masking
331 || (s->flags&CODEC_FLAG_QP_RD))
334 s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
335 s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
336 s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
337 s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
338 s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
340 if(avctx->rc_max_rate && !avctx->rc_buffer_size){
341 av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
345 if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
346 av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
349 if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
350 av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
354 if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
355 av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
359 if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
360 av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
363 if(avctx->rc_buffer_size && avctx->bit_rate*(int64_t)avctx->time_base.num > avctx->rc_buffer_size * (int64_t)avctx->time_base.den){
364 av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
368 if(!s->fixed_qscale && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
369 av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
373 if( s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
374 && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
375 && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
377 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");
380 if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
381 && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
382 av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
386 if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
387 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
391 if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
392 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
396 if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
397 av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
401 if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
402 av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
406 if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
407 av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
411 if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
412 s->codec_id == CODEC_ID_H263P) &&
413 (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
414 av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
415 avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
419 if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
420 && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
421 av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
425 if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
426 av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
430 if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
431 av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
435 if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
436 av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
440 if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
441 av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
445 if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
446 av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
450 if(s->flags & CODEC_FLAG_LOW_DELAY){
451 if (s->codec_id != CODEC_ID_MPEG2VIDEO){
452 av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
455 if (s->max_b_frames != 0){
456 av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
461 if(s->q_scale_type == 1){
462 if(s->codec_id != CODEC_ID_MPEG2VIDEO){
463 av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
466 if(avctx->qmax > 12){
467 av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
472 if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
473 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
474 && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
475 av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
479 if(s->avctx->thread_count < 1){
480 av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
484 if(s->avctx->thread_count > 1)
487 if(!avctx->time_base.den || !avctx->time_base.num){
488 av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
492 i= (INT_MAX/2+128)>>8;
493 if(avctx->me_threshold >= i){
494 av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
497 if(avctx->mb_threshold >= i){
498 av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
502 if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
503 av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
504 avctx->b_frame_strategy = 0;
507 i= av_gcd(avctx->time_base.den, avctx->time_base.num);
509 av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
510 avctx->time_base.den /= i;
511 avctx->time_base.num /= i;
515 if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
516 s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
517 s->inter_quant_bias= 0;
519 s->intra_quant_bias=0;
520 s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
523 if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
524 s->intra_quant_bias= avctx->intra_quant_bias;
525 if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
526 s->inter_quant_bias= avctx->inter_quant_bias;
528 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
530 if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
531 av_log(avctx, AV_LOG_ERROR, "timebase %d/%d not supported by MPEG 4 standard, "
532 "the maximum admitted value for the timebase denominator is %d\n",
533 s->avctx->time_base.num, s->avctx->time_base.den, (1<<16)-1);
536 s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
538 switch(avctx->codec->id) {
539 case CODEC_ID_MPEG1VIDEO:
540 s->out_format = FMT_MPEG1;
541 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
542 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
544 case CODEC_ID_MPEG2VIDEO:
545 s->out_format = FMT_MPEG1;
546 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
547 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
552 s->out_format = FMT_MJPEG;
553 s->intra_only = 1; /* force intra only for jpeg */
554 if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
555 s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
556 s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
557 s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
559 s->mjpeg_vsample[0] = 2;
560 s->mjpeg_vsample[1] = 2>>chroma_v_shift;
561 s->mjpeg_vsample[2] = 2>>chroma_v_shift;
562 s->mjpeg_hsample[0] = 2;
563 s->mjpeg_hsample[1] = 2>>chroma_h_shift;
564 s->mjpeg_hsample[2] = 2>>chroma_h_shift;
566 if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
567 || ff_mjpeg_encode_init(s) < 0)
573 if (!CONFIG_H261_ENCODER) return -1;
574 if (ff_h261_get_picture_format(s->width, s->height) < 0) {
575 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);
578 s->out_format = FMT_H261;
583 if (!CONFIG_H263_ENCODER) return -1;
584 if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 8) {
585 av_log(avctx, AV_LOG_ERROR, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height);
588 s->out_format = FMT_H263;
589 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
594 s->out_format = FMT_H263;
597 s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
598 s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
599 s->modified_quant= s->h263_aic;
600 s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
601 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
602 s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
603 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
604 s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
607 /* These are just to be sure */
612 s->out_format = FMT_H263;
613 s->h263_flv = 2; /* format = 1; 11-bit codes */
614 s->unrestricted_mv = 1;
615 s->rtp_mode=0; /* don't allow GOB */
620 s->out_format = FMT_H263;
625 s->out_format = FMT_H263;
632 s->unrestricted_mv= 0;
635 s->out_format = FMT_H263;
637 s->unrestricted_mv = 1;
638 s->low_delay= s->max_b_frames ? 0 : 1;
639 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
641 case CODEC_ID_MSMPEG4V2:
642 s->out_format = FMT_H263;
644 s->unrestricted_mv = 1;
645 s->msmpeg4_version= 2;
649 case CODEC_ID_MSMPEG4V3:
650 s->out_format = FMT_H263;
652 s->unrestricted_mv = 1;
653 s->msmpeg4_version= 3;
654 s->flipflop_rounding=1;
659 s->out_format = FMT_H263;
661 s->unrestricted_mv = 1;
662 s->msmpeg4_version= 4;
663 s->flipflop_rounding=1;
668 s->out_format = FMT_H263;
670 s->unrestricted_mv = 1;
671 s->msmpeg4_version= 5;
672 s->flipflop_rounding=1;
680 avctx->has_b_frames= !s->low_delay;
684 s->progressive_frame=
685 s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
688 if (MPV_common_init(s) < 0)
692 s->dct_quantize = dct_quantize_c;
694 s->denoise_dct = denoise_dct_c;
695 s->fast_dct_quantize = s->dct_quantize;
697 s->dct_quantize = dct_quantize_trellis_c;
699 if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
700 s->chroma_qscale_table= ff_h263_chroma_qscale_table;
702 s->quant_precision=5;
704 ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
705 ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
707 if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
708 ff_h261_encode_init(s);
709 if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
711 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
712 ff_msmpeg4_encode_init(s);
713 if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
714 && s->out_format == FMT_MPEG1)
715 ff_mpeg1_encode_init(s);
719 int j= s->dsp.idct_permutation[i];
720 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
721 s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
722 s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
723 }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
725 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
728 s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
729 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
731 if(s->avctx->intra_matrix)
732 s->intra_matrix[j] = s->avctx->intra_matrix[i];
733 if(s->avctx->inter_matrix)
734 s->inter_matrix[j] = s->avctx->inter_matrix[i];
737 /* precompute matrix */
738 /* for mjpeg, we do include qscale in the matrix */
739 if (s->out_format != FMT_MJPEG) {
740 ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
741 s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
742 ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
743 s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
746 if(ff_rate_control_init(s) < 0)
752 av_cold int MPV_encode_end(AVCodecContext *avctx)
754 MpegEncContext *s = avctx->priv_data;
756 ff_rate_control_uninit(s);
759 if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
760 ff_mjpeg_encode_close(s);
762 av_freep(&avctx->extradata);
767 static int get_sae(uint8_t *src, int ref, int stride){
773 acc+= FFABS(src[x+y*stride] - ref);
780 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
787 for(y=0; y<h; y+=16){
788 for(x=0; x<w; x+=16){
789 int offset= x + y*stride;
790 int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
791 int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
792 int sae = get_sae(src + offset, mean, stride);
794 acc+= sae + 500 < sad;
801 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
805 const int encoding_delay= s->max_b_frames;
810 pic_arg->display_picture_number= s->input_picture_number++;
812 if(pts != AV_NOPTS_VALUE){
813 if(s->user_specified_pts != AV_NOPTS_VALUE){
815 int64_t last= s->user_specified_pts;
818 av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
822 s->user_specified_pts= pts;
824 if(s->user_specified_pts != AV_NOPTS_VALUE){
825 s->user_specified_pts=
826 pts= s->user_specified_pts + 1;
827 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
829 pts= pic_arg->display_picture_number;
835 if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
836 if(pic_arg->linesize[0] != s->linesize) direct=0;
837 if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
838 if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
840 // av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
843 i= ff_find_unused_picture(s, 1);
845 pic= (AVFrame*)&s->picture[i];
849 pic->data[i]= pic_arg->data[i];
850 pic->linesize[i]= pic_arg->linesize[i];
852 if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){
856 i= ff_find_unused_picture(s, 0);
858 pic= (AVFrame*)&s->picture[i];
861 if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){
865 if( pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
866 && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
867 && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
870 int h_chroma_shift, v_chroma_shift;
871 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
874 int src_stride= pic_arg->linesize[i];
875 int dst_stride= i ? s->uvlinesize : s->linesize;
876 int h_shift= i ? h_chroma_shift : 0;
877 int v_shift= i ? v_chroma_shift : 0;
878 int w= s->width >>h_shift;
879 int h= s->height>>v_shift;
880 uint8_t *src= pic_arg->data[i];
881 uint8_t *dst= pic->data[i];
883 if(!s->avctx->rc_buffer_size)
884 dst +=INPLACE_OFFSET;
886 if(src_stride==dst_stride)
887 memcpy(dst, src, src_stride*h);
898 copy_picture_attributes(s, pic, pic_arg);
899 pic->pts= pts; //we set this here to avoid modifiying pic_arg
902 /* shift buffer entries */
903 for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
904 s->input_picture[i-1]= s->input_picture[i];
906 s->input_picture[encoding_delay]= (Picture*)pic;
911 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
916 for(plane=0; plane<3; plane++){
917 const int stride= p->linesize[plane];
918 const int bw= plane ? 1 : 2;
919 for(y=0; y<s->mb_height*bw; y++){
920 for(x=0; x<s->mb_width*bw; x++){
921 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
922 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);
924 switch(s->avctx->frame_skip_exp){
925 case 0: score= FFMAX(score, v); break;
926 case 1: score+= FFABS(v);break;
927 case 2: score+= v*v;break;
928 case 3: score64+= FFABS(v*v*(int64_t)v);break;
929 case 4: score64+= v*v*(int64_t)(v*v);break;
935 if(score) score64= score;
937 if(score64 < s->avctx->frame_skip_threshold)
939 if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
944 static int estimate_best_b_count(MpegEncContext *s){
945 AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
946 AVCodecContext *c= avcodec_alloc_context();
947 AVFrame input[FF_MAX_B_FRAMES+2];
948 const int scale= s->avctx->brd_scale;
949 int i, j, out_size, p_lambda, b_lambda, lambda2;
950 int outbuf_size= s->width * s->height; //FIXME
951 uint8_t *outbuf= av_malloc(outbuf_size);
952 int64_t best_rd= INT64_MAX;
953 int best_b_count= -1;
955 assert(scale>=0 && scale <=3);
958 p_lambda= s->last_lambda_for[AV_PICTURE_TYPE_P]; //s->next_picture_ptr->quality;
959 b_lambda= s->last_lambda_for[AV_PICTURE_TYPE_B]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
960 if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
961 lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
963 c->width = s->width >> scale;
964 c->height= s->height>> scale;
965 c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
966 c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
967 c->mb_decision= s->avctx->mb_decision;
968 c->me_cmp= s->avctx->me_cmp;
969 c->mb_cmp= s->avctx->mb_cmp;
970 c->me_sub_cmp= s->avctx->me_sub_cmp;
971 c->pix_fmt = PIX_FMT_YUV420P;
972 c->time_base= s->avctx->time_base;
973 c->max_b_frames= s->max_b_frames;
975 if (avcodec_open(c, codec) < 0)
978 for(i=0; i<s->max_b_frames+2; i++){
979 int ysize= c->width*c->height;
980 int csize= (c->width/2)*(c->height/2);
981 Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
983 avcodec_get_frame_defaults(&input[i]);
984 input[i].data[0]= av_malloc(ysize + 2*csize);
985 input[i].data[1]= input[i].data[0] + ysize;
986 input[i].data[2]= input[i].data[1] + csize;
987 input[i].linesize[0]= c->width;
988 input[i].linesize[1]=
989 input[i].linesize[2]= c->width/2;
991 if(pre_input_ptr && (!i || s->input_picture[i-1])) {
992 pre_input= *pre_input_ptr;
994 if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
995 pre_input.data[0]+=INPLACE_OFFSET;
996 pre_input.data[1]+=INPLACE_OFFSET;
997 pre_input.data[2]+=INPLACE_OFFSET;
1000 s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
1001 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);
1002 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);
1006 for(j=0; j<s->max_b_frames+1; j++){
1009 if(!s->input_picture[j])
1012 c->error[0]= c->error[1]= c->error[2]= 0;
1014 input[0].pict_type= AV_PICTURE_TYPE_I;
1015 input[0].quality= 1 * FF_QP2LAMBDA;
1016 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1017 // rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1019 for(i=0; i<s->max_b_frames+1; i++){
1020 int is_p= i % (j+1) == j || i==s->max_b_frames;
1022 input[i+1].pict_type= is_p ? AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1023 input[i+1].quality= is_p ? p_lambda : b_lambda;
1024 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1025 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1028 /* get the delayed frames */
1030 out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1031 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1034 rd += c->error[0] + c->error[1] + c->error[2];
1046 for(i=0; i<s->max_b_frames+2; i++){
1047 av_freep(&input[i].data[0]);
1050 return best_b_count;
1053 static int select_input_picture(MpegEncContext *s){
1056 for(i=1; i<MAX_PICTURE_COUNT; i++)
1057 s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1058 s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1060 /* set next picture type & ordering */
1061 if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1062 if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1063 s->reordered_input_picture[0]= s->input_picture[0];
1064 s->reordered_input_picture[0]->pict_type= AV_PICTURE_TYPE_I;
1065 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1069 if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1070 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1071 //FIXME check that te gop check above is +-1 correct
1072 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1074 if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1076 s->input_picture[0]->data[i]= NULL;
1077 s->input_picture[0]->type= 0;
1079 assert( s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1080 || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1082 s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1086 ff_vbv_update(s, 0);
1092 if(s->flags&CODEC_FLAG_PASS2){
1093 for(i=0; i<s->max_b_frames+1; i++){
1094 int pict_num= s->input_picture[0]->display_picture_number + i;
1096 if(pict_num >= s->rc_context.num_entries)
1098 if(!s->input_picture[i]){
1099 s->rc_context.entry[pict_num-1].new_pict_type = AV_PICTURE_TYPE_P;
1103 s->input_picture[i]->pict_type=
1104 s->rc_context.entry[pict_num].new_pict_type;
1108 if(s->avctx->b_frame_strategy==0){
1109 b_frames= s->max_b_frames;
1110 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1111 }else if(s->avctx->b_frame_strategy==1){
1112 for(i=1; i<s->max_b_frames+1; i++){
1113 if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1114 s->input_picture[i]->b_frame_score=
1115 get_intra_count(s, s->input_picture[i ]->data[0],
1116 s->input_picture[i-1]->data[0], s->linesize) + 1;
1119 for(i=0; i<s->max_b_frames+1; i++){
1120 if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1123 b_frames= FFMAX(0, i-1);
1126 for(i=0; i<b_frames+1; i++){
1127 s->input_picture[i]->b_frame_score=0;
1129 }else if(s->avctx->b_frame_strategy==2){
1130 b_frames= estimate_best_b_count(s);
1132 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1137 //static int b_count=0;
1138 //b_count+= b_frames;
1139 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1141 for(i= b_frames - 1; i>=0; i--){
1142 int type= s->input_picture[i]->pict_type;
1143 if(type && type != AV_PICTURE_TYPE_B)
1146 if(s->input_picture[b_frames]->pict_type == AV_PICTURE_TYPE_B && b_frames == s->max_b_frames){
1147 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1150 if(s->picture_in_gop_number + b_frames >= s->gop_size){
1151 if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1152 b_frames= s->gop_size - s->picture_in_gop_number - 1;
1154 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1156 s->input_picture[b_frames]->pict_type= AV_PICTURE_TYPE_I;
1160 if( (s->flags & CODEC_FLAG_CLOSED_GOP)
1162 && s->input_picture[b_frames]->pict_type== AV_PICTURE_TYPE_I)
1165 s->reordered_input_picture[0]= s->input_picture[b_frames];
1166 if(s->reordered_input_picture[0]->pict_type != AV_PICTURE_TYPE_I)
1167 s->reordered_input_picture[0]->pict_type= AV_PICTURE_TYPE_P;
1168 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1169 for(i=0; i<b_frames; i++){
1170 s->reordered_input_picture[i+1]= s->input_picture[i];
1171 s->reordered_input_picture[i+1]->pict_type= AV_PICTURE_TYPE_B;
1172 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1177 if(s->reordered_input_picture[0]){
1178 s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=AV_PICTURE_TYPE_B ? 3 : 0;
1180 ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1182 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1183 // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1185 int i= ff_find_unused_picture(s, 0);
1186 Picture *pic= &s->picture[i];
1188 pic->reference = s->reordered_input_picture[0]->reference;
1189 if(ff_alloc_picture(s, pic, 0) < 0){
1193 /* mark us unused / free shared pic */
1194 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1195 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1197 s->reordered_input_picture[0]->data[i]= NULL;
1198 s->reordered_input_picture[0]->type= 0;
1200 copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1202 s->current_picture_ptr= pic;
1204 // input is not a shared pix -> reuse buffer for current_pix
1206 assert( s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1207 || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1209 s->current_picture_ptr= s->reordered_input_picture[0];
1211 s->new_picture.data[i]+= INPLACE_OFFSET;
1214 ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1216 s->picture_number= s->new_picture.display_picture_number;
1217 //printf("dpn:%d\n", s->picture_number);
1219 memset(&s->new_picture, 0, sizeof(Picture));
1224 int MPV_encode_picture(AVCodecContext *avctx,
1225 unsigned char *buf, int buf_size, void *data)
1227 MpegEncContext *s = avctx->priv_data;
1228 AVFrame *pic_arg = data;
1229 int i, stuffing_count, context_count = avctx->thread_count;
1231 for(i=0; i<context_count; i++){
1232 int start_y= s->thread_context[i]->start_mb_y;
1233 int end_y= s->thread_context[i]-> end_mb_y;
1234 int h= s->mb_height;
1235 uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1236 uint8_t *end = buf + (size_t)(((int64_t) buf_size)* end_y/h);
1238 init_put_bits(&s->thread_context[i]->pb, start, end - start);
1241 s->picture_in_gop_number++;
1243 if(load_input_picture(s, pic_arg) < 0)
1246 if(select_input_picture(s) < 0){
1251 if(s->new_picture.data[0]){
1252 s->pict_type= s->new_picture.pict_type;
1254 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1255 MPV_frame_start(s, avctx);
1257 if (encode_picture(s, s->picture_number) < 0)
1260 avctx->header_bits = s->header_bits;
1261 avctx->mv_bits = s->mv_bits;
1262 avctx->misc_bits = s->misc_bits;
1263 avctx->i_tex_bits = s->i_tex_bits;
1264 avctx->p_tex_bits = s->p_tex_bits;
1265 avctx->i_count = s->i_count;
1266 avctx->p_count = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1267 avctx->skip_count = s->skip_count;
1271 if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1272 ff_mjpeg_encode_picture_trailer(s);
1274 if(avctx->rc_buffer_size){
1275 RateControlContext *rcc= &s->rc_context;
1276 int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1278 if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1279 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1280 if(s->adaptive_quant){
1282 for(i=0; i<s->mb_height*s->mb_stride; i++)
1283 s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1285 s->mb_skipped = 0; //done in MPV_frame_start()
1286 if(s->pict_type==AV_PICTURE_TYPE_P){ //done in encode_picture() so we must undo it
1287 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1288 s->no_rounding ^= 1;
1290 if(s->pict_type!=AV_PICTURE_TYPE_B){
1291 s->time_base= s->last_time_base;
1292 s->last_non_b_time= s->time - s->pp_time;
1294 // av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1295 for(i=0; i<context_count; i++){
1296 PutBitContext *pb= &s->thread_context[i]->pb;
1297 init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1302 assert(s->avctx->rc_max_rate);
1305 if(s->flags&CODEC_FLAG_PASS1)
1306 ff_write_pass1_stats(s);
1309 s->current_picture_ptr->error[i]= s->current_picture.error[i];
1310 avctx->error[i] += s->current_picture_ptr->error[i];
1313 if(s->flags&CODEC_FLAG_PASS1)
1314 assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1315 flush_put_bits(&s->pb);
1316 s->frame_bits = put_bits_count(&s->pb);
1318 stuffing_count= ff_vbv_update(s, s->frame_bits);
1320 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1321 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1325 switch(s->codec_id){
1326 case CODEC_ID_MPEG1VIDEO:
1327 case CODEC_ID_MPEG2VIDEO:
1328 while(stuffing_count--){
1329 put_bits(&s->pb, 8, 0);
1332 case CODEC_ID_MPEG4:
1333 put_bits(&s->pb, 16, 0);
1334 put_bits(&s->pb, 16, 0x1C3);
1335 stuffing_count -= 4;
1336 while(stuffing_count--){
1337 put_bits(&s->pb, 8, 0xFF);
1341 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1343 flush_put_bits(&s->pb);
1344 s->frame_bits = put_bits_count(&s->pb);
1347 /* update mpeg1/2 vbv_delay for CBR */
1348 if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1349 && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1350 int vbv_delay, min_delay;
1351 double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1352 int minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1353 double bits = s->rc_context.buffer_index + minbits - inbits;
1356 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1358 assert(s->repeat_first_field==0);
1360 vbv_delay= bits * 90000 / s->avctx->rc_max_rate;
1361 min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1363 vbv_delay= FFMAX(vbv_delay, min_delay);
1365 assert(vbv_delay < 0xFFFF);
1367 s->vbv_delay_ptr[0] &= 0xF8;
1368 s->vbv_delay_ptr[0] |= vbv_delay>>13;
1369 s->vbv_delay_ptr[1] = vbv_delay>>5;
1370 s->vbv_delay_ptr[2] &= 0x07;
1371 s->vbv_delay_ptr[2] |= vbv_delay<<3;
1372 avctx->vbv_delay = vbv_delay*300;
1374 s->total_bits += s->frame_bits;
1375 avctx->frame_bits = s->frame_bits;
1377 assert((put_bits_ptr(&s->pb) == s->pb.buf));
1380 assert((s->frame_bits&7)==0);
1382 return s->frame_bits/8;
1385 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1387 static const char tab[64]=
1399 DCTELEM *block= s->block[n];
1400 const int last_index= s->block_last_index[n];
1405 threshold= -threshold;
1409 /* Are all we could set to zero already zero? */
1410 if(last_index<=skip_dc - 1) return;
1412 for(i=0; i<=last_index; i++){
1413 const int j = s->intra_scantable.permutated[i];
1414 const int level = FFABS(block[j]);
1416 if(skip_dc && i==0) continue;
1425 if(score >= threshold) return;
1426 for(i=skip_dc; i<=last_index; i++){
1427 const int j = s->intra_scantable.permutated[i];
1430 if(block[0]) s->block_last_index[n]= 0;
1431 else s->block_last_index[n]= -1;
1434 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1437 const int maxlevel= s->max_qcoeff;
1438 const int minlevel= s->min_qcoeff;
1442 i=1; //skip clipping of intra dc
1446 for(;i<=last_index; i++){
1447 const int j= s->intra_scantable.permutated[i];
1448 int level = block[j];
1450 if (level>maxlevel){
1453 }else if(level<minlevel){
1461 if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1462 av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1465 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1475 for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1476 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1477 int v= ptr[x2 + y2*stride];
1483 weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1488 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1490 int16_t weight[8][64];
1491 DCTELEM orig[8][64];
1492 const int mb_x= s->mb_x;
1493 const int mb_y= s->mb_y;
1496 int dct_offset = s->linesize*8; //default for progressive frames
1497 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1500 for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1502 if(s->adaptive_quant){
1503 const int last_qp= s->qscale;
1504 const int mb_xy= mb_x + mb_y*s->mb_stride;
1506 s->lambda= s->lambda_table[mb_xy];
1509 if(!(s->flags&CODEC_FLAG_QP_RD)){
1510 s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1511 s->dquant= s->qscale - last_qp;
1513 if(s->out_format==FMT_H263){
1514 s->dquant= av_clip(s->dquant, -2, 2);
1516 if(s->codec_id==CODEC_ID_MPEG4){
1518 if(s->pict_type == AV_PICTURE_TYPE_B){
1519 if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1522 if(s->mv_type==MV_TYPE_8X8)
1528 ff_set_qscale(s, last_qp + s->dquant);
1529 }else if(s->flags&CODEC_FLAG_QP_RD)
1530 ff_set_qscale(s, s->qscale + s->dquant);
1532 wrap_y = s->linesize;
1533 wrap_c = s->uvlinesize;
1534 ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1535 ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1536 ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1538 if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1539 uint8_t *ebuf= s->edge_emu_buffer + 32;
1540 s->dsp.emulated_edge_mc(ebuf , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width , s->height);
1542 s->dsp.emulated_edge_mc(ebuf+18*wrap_y , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1543 ptr_cb= ebuf+18*wrap_y;
1544 s->dsp.emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1545 ptr_cr= ebuf+18*wrap_y+8;
1549 if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1550 int progressive_score, interlaced_score;
1552 s->interlaced_dct=0;
1553 progressive_score= s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y, 8)
1554 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1556 if(progressive_score > 0){
1557 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y*2, 8)
1558 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y , NULL, wrap_y*2, 8);
1559 if(progressive_score > interlaced_score){
1560 s->interlaced_dct=1;
1564 if (s->chroma_format == CHROMA_422)
1570 s->dsp.get_pixels(s->block[0], ptr_y , wrap_y);
1571 s->dsp.get_pixels(s->block[1], ptr_y + 8, wrap_y);
1572 s->dsp.get_pixels(s->block[2], ptr_y + dct_offset , wrap_y);
1573 s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1575 if(s->flags&CODEC_FLAG_GRAY){
1579 s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1580 s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1581 if(!s->chroma_y_shift){ /* 422 */
1582 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1583 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1587 op_pixels_func (*op_pix)[4];
1588 qpel_mc_func (*op_qpix)[16];
1589 uint8_t *dest_y, *dest_cb, *dest_cr;
1591 dest_y = s->dest[0];
1592 dest_cb = s->dest[1];
1593 dest_cr = s->dest[2];
1595 if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
1596 op_pix = s->dsp.put_pixels_tab;
1597 op_qpix= s->dsp.put_qpel_pixels_tab;
1599 op_pix = s->dsp.put_no_rnd_pixels_tab;
1600 op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1603 if (s->mv_dir & MV_DIR_FORWARD) {
1604 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1605 op_pix = s->dsp.avg_pixels_tab;
1606 op_qpix= s->dsp.avg_qpel_pixels_tab;
1608 if (s->mv_dir & MV_DIR_BACKWARD) {
1609 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1612 if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1613 int progressive_score, interlaced_score;
1615 s->interlaced_dct=0;
1616 progressive_score= s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y, 8)
1617 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1619 if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1621 if(progressive_score>0){
1622 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y*2, 8)
1623 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y , ptr_y + wrap_y , wrap_y*2, 8);
1625 if(progressive_score > interlaced_score){
1626 s->interlaced_dct=1;
1630 if (s->chroma_format == CHROMA_422)
1636 s->dsp.diff_pixels(s->block[0], ptr_y , dest_y , wrap_y);
1637 s->dsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
1638 s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset , dest_y + dct_offset , wrap_y);
1639 s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1641 if(s->flags&CODEC_FLAG_GRAY){
1645 s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1646 s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1647 if(!s->chroma_y_shift){ /* 422 */
1648 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1649 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1652 /* pre quantization */
1653 if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1655 if(s->dsp.sad[1](NULL, ptr_y , dest_y , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1656 if(s->dsp.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1657 if(s->dsp.sad[1](NULL, ptr_y +dct_offset , dest_y +dct_offset , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1658 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;
1659 if(s->dsp.sad[1](NULL, ptr_cb , dest_cb , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1660 if(s->dsp.sad[1](NULL, ptr_cr , dest_cr , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1661 if(!s->chroma_y_shift){ /* 422 */
1662 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;
1663 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;
1668 if(s->avctx->quantizer_noise_shaping){
1669 if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y , wrap_y);
1670 if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y + 8, wrap_y);
1671 if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset , wrap_y);
1672 if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1673 if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb , wrap_c);
1674 if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr , wrap_c);
1675 if(!s->chroma_y_shift){ /* 422 */
1676 if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1677 if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1679 memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1682 /* DCT & quantize */
1683 assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1685 for(i=0;i<mb_block_count;i++) {
1688 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1689 // FIXME we could decide to change to quantizer instead of clipping
1690 // JS: I don't think that would be a good idea it could lower quality instead
1691 // of improve it. Just INTRADC clipping deserves changes in quantizer
1692 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1694 s->block_last_index[i]= -1;
1696 if(s->avctx->quantizer_noise_shaping){
1697 for(i=0;i<mb_block_count;i++) {
1699 s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1704 if(s->luma_elim_threshold && !s->mb_intra)
1706 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1707 if(s->chroma_elim_threshold && !s->mb_intra)
1708 for(i=4; i<mb_block_count; i++)
1709 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1711 if(s->flags & CODEC_FLAG_CBP_RD){
1712 for(i=0;i<mb_block_count;i++) {
1713 if(s->block_last_index[i] == -1)
1714 s->coded_score[i]= INT_MAX/256;
1719 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1720 s->block_last_index[4]=
1721 s->block_last_index[5]= 0;
1723 s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1726 //non c quantize code returns incorrect block_last_index FIXME
1727 if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1728 for(i=0; i<mb_block_count; i++){
1730 if(s->block_last_index[i]>0){
1731 for(j=63; j>0; j--){
1732 if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1734 s->block_last_index[i]= j;
1739 /* huffman encode */
1740 switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1741 case CODEC_ID_MPEG1VIDEO:
1742 case CODEC_ID_MPEG2VIDEO:
1743 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1744 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1746 case CODEC_ID_MPEG4:
1747 if (CONFIG_MPEG4_ENCODER)
1748 mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1750 case CODEC_ID_MSMPEG4V2:
1751 case CODEC_ID_MSMPEG4V3:
1753 if (CONFIG_MSMPEG4_ENCODER)
1754 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1757 if (CONFIG_WMV2_ENCODER)
1758 ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1761 if (CONFIG_H261_ENCODER)
1762 ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1765 case CODEC_ID_H263P:
1769 if (CONFIG_H263_ENCODER)
1770 h263_encode_mb(s, s->block, motion_x, motion_y);
1772 case CODEC_ID_MJPEG:
1773 if (CONFIG_MJPEG_ENCODER)
1774 ff_mjpeg_encode_mb(s, s->block);
1781 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1783 if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 6);
1784 else encode_mb_internal(s, motion_x, motion_y, 16, 8);
1787 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1790 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1793 d->mb_skip_run= s->mb_skip_run;
1795 d->last_dc[i]= s->last_dc[i];
1798 d->mv_bits= s->mv_bits;
1799 d->i_tex_bits= s->i_tex_bits;
1800 d->p_tex_bits= s->p_tex_bits;
1801 d->i_count= s->i_count;
1802 d->f_count= s->f_count;
1803 d->b_count= s->b_count;
1804 d->skip_count= s->skip_count;
1805 d->misc_bits= s->misc_bits;
1809 d->qscale= s->qscale;
1810 d->dquant= s->dquant;
1812 d->esc3_level_length= s->esc3_level_length;
1815 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1818 memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1819 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1822 d->mb_skip_run= s->mb_skip_run;
1824 d->last_dc[i]= s->last_dc[i];
1827 d->mv_bits= s->mv_bits;
1828 d->i_tex_bits= s->i_tex_bits;
1829 d->p_tex_bits= s->p_tex_bits;
1830 d->i_count= s->i_count;
1831 d->f_count= s->f_count;
1832 d->b_count= s->b_count;
1833 d->skip_count= s->skip_count;
1834 d->misc_bits= s->misc_bits;
1836 d->mb_intra= s->mb_intra;
1837 d->mb_skipped= s->mb_skipped;
1838 d->mv_type= s->mv_type;
1839 d->mv_dir= s->mv_dir;
1841 if(s->data_partitioning){
1843 d->tex_pb= s->tex_pb;
1847 d->block_last_index[i]= s->block_last_index[i];
1848 d->interlaced_dct= s->interlaced_dct;
1849 d->qscale= s->qscale;
1851 d->esc3_level_length= s->esc3_level_length;
1854 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1855 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1856 int *dmin, int *next_block, int motion_x, int motion_y)
1859 uint8_t *dest_backup[3];
1861 copy_context_before_encode(s, backup, type);
1863 s->block= s->blocks[*next_block];
1864 s->pb= pb[*next_block];
1865 if(s->data_partitioning){
1866 s->pb2 = pb2 [*next_block];
1867 s->tex_pb= tex_pb[*next_block];
1871 memcpy(dest_backup, s->dest, sizeof(s->dest));
1872 s->dest[0] = s->rd_scratchpad;
1873 s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1874 s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1875 assert(s->linesize >= 32); //FIXME
1878 encode_mb(s, motion_x, motion_y);
1880 score= put_bits_count(&s->pb);
1881 if(s->data_partitioning){
1882 score+= put_bits_count(&s->pb2);
1883 score+= put_bits_count(&s->tex_pb);
1886 if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1887 MPV_decode_mb(s, s->block);
1889 score *= s->lambda2;
1890 score += sse_mb(s) << FF_LAMBDA_SHIFT;
1894 memcpy(s->dest, dest_backup, sizeof(s->dest));
1901 copy_context_after_encode(best, s, type);
1905 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1906 uint32_t *sq = ff_squareTbl + 256;
1911 return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1912 else if(w==8 && h==8)
1913 return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1917 acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1926 static int sse_mb(MpegEncContext *s){
1930 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1931 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1934 if(s->avctx->mb_cmp == FF_CMP_NSSE){
1935 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)
1936 +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)
1937 +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);
1939 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)
1940 +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)
1941 +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);
1944 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)
1945 +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)
1946 +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);
1949 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1950 MpegEncContext *s= *(void**)arg;
1954 s->me.dia_size= s->avctx->pre_dia_size;
1955 s->first_slice_line=1;
1956 for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1957 for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1958 ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1960 s->first_slice_line=0;
1968 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1969 MpegEncContext *s= *(void**)arg;
1971 ff_check_alignment();
1973 s->me.dia_size= s->avctx->dia_size;
1974 s->first_slice_line=1;
1975 for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1976 s->mb_x=0; //for block init below
1977 ff_init_block_index(s);
1978 for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1979 s->block_index[0]+=2;
1980 s->block_index[1]+=2;
1981 s->block_index[2]+=2;
1982 s->block_index[3]+=2;
1984 /* compute motion vector & mb_type and store in context */
1985 if(s->pict_type==AV_PICTURE_TYPE_B)
1986 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1988 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1990 s->first_slice_line=0;
1995 static int mb_var_thread(AVCodecContext *c, void *arg){
1996 MpegEncContext *s= *(void**)arg;
1999 ff_check_alignment();
2001 for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2002 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2005 uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
2007 int sum = s->dsp.pix_sum(pix, s->linesize);
2009 varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2011 s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2012 s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2013 s->me.mb_var_sum_temp += varc;
2019 static void write_slice_end(MpegEncContext *s){
2020 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2021 if(s->partitioned_frame){
2022 ff_mpeg4_merge_partitions(s);
2025 ff_mpeg4_stuffing(&s->pb);
2026 }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2027 ff_mjpeg_encode_stuffing(&s->pb);
2030 align_put_bits(&s->pb);
2031 flush_put_bits(&s->pb);
2033 if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2034 s->misc_bits+= get_bits_diff(s);
2037 static int encode_thread(AVCodecContext *c, void *arg){
2038 MpegEncContext *s= *(void**)arg;
2039 int mb_x, mb_y, pdif = 0;
2040 int chr_h= 16>>s->chroma_y_shift;
2042 MpegEncContext best_s, backup_s;
2043 uint8_t bit_buf[2][MAX_MB_BYTES];
2044 uint8_t bit_buf2[2][MAX_MB_BYTES];
2045 uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2046 PutBitContext pb[2], pb2[2], tex_pb[2];
2047 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2049 ff_check_alignment();
2052 init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
2053 init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
2054 init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2057 s->last_bits= put_bits_count(&s->pb);
2068 /* init last dc values */
2069 /* note: quant matrix value (8) is implied here */
2070 s->last_dc[i] = 128 << s->intra_dc_precision;
2072 s->current_picture.error[i] = 0;
2075 memset(s->last_mv, 0, sizeof(s->last_mv));
2079 switch(s->codec_id){
2081 case CODEC_ID_H263P:
2083 if (CONFIG_H263_ENCODER)
2084 s->gob_index = ff_h263_get_gob_height(s);
2086 case CODEC_ID_MPEG4:
2087 if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2088 ff_mpeg4_init_partitions(s);
2094 s->first_slice_line = 1;
2095 s->ptr_lastgob = s->pb.buf;
2096 for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2097 // printf("row %d at %X\n", s->mb_y, (int)s);
2101 ff_set_qscale(s, s->qscale);
2102 ff_init_block_index(s);
2104 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2105 int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2106 int mb_type= s->mb_type[xy];
2111 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2112 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2115 if(s->data_partitioning){
2116 if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES
2117 || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2118 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2124 s->mb_y = mb_y; // moved into loop, can get changed by H.261
2125 ff_update_block_index(s);
2127 if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2128 ff_h261_reorder_mb_index(s);
2129 xy= s->mb_y*s->mb_stride + s->mb_x;
2130 mb_type= s->mb_type[xy];
2133 /* write gob / video packet header */
2135 int current_packet_size, is_gob_start;
2137 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2139 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2141 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2143 switch(s->codec_id){
2145 case CODEC_ID_H263P:
2146 if(!s->h263_slice_structured)
2147 if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2149 case CODEC_ID_MPEG2VIDEO:
2150 if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2151 case CODEC_ID_MPEG1VIDEO:
2152 if(s->mb_skip_run) is_gob_start=0;
2157 if(s->start_mb_y != mb_y || mb_x!=0){
2160 if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2161 ff_mpeg4_init_partitions(s);
2165 assert((put_bits_count(&s->pb)&7) == 0);
2166 current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2168 if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2169 int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2170 int d= 100 / s->avctx->error_rate;
2172 current_packet_size=0;
2173 #ifndef ALT_BITSTREAM_WRITER
2174 s->pb.buf_ptr= s->ptr_lastgob;
2176 assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2180 if (s->avctx->rtp_callback){
2181 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2182 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2185 switch(s->codec_id){
2186 case CODEC_ID_MPEG4:
2187 if (CONFIG_MPEG4_ENCODER) {
2188 ff_mpeg4_encode_video_packet_header(s);
2189 ff_mpeg4_clean_buffers(s);
2192 case CODEC_ID_MPEG1VIDEO:
2193 case CODEC_ID_MPEG2VIDEO:
2194 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2195 ff_mpeg1_encode_slice_header(s);
2196 ff_mpeg1_clean_buffers(s);
2200 case CODEC_ID_H263P:
2201 if (CONFIG_H263_ENCODER)
2202 h263_encode_gob_header(s, mb_y);
2206 if(s->flags&CODEC_FLAG_PASS1){
2207 int bits= put_bits_count(&s->pb);
2208 s->misc_bits+= bits - s->last_bits;
2212 s->ptr_lastgob += current_packet_size;
2213 s->first_slice_line=1;
2214 s->resync_mb_x=mb_x;
2215 s->resync_mb_y=mb_y;
2219 if( (s->resync_mb_x == s->mb_x)
2220 && s->resync_mb_y+1 == s->mb_y){
2221 s->first_slice_line=0;
2225 s->dquant=0; //only for QP_RD
2227 if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2229 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2231 copy_context_before_encode(&backup_s, s, -1);
2233 best_s.data_partitioning= s->data_partitioning;
2234 best_s.partitioned_frame= s->partitioned_frame;
2235 if(s->data_partitioning){
2236 backup_s.pb2= s->pb2;
2237 backup_s.tex_pb= s->tex_pb;
2240 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2241 s->mv_dir = MV_DIR_FORWARD;
2242 s->mv_type = MV_TYPE_16X16;
2244 s->mv[0][0][0] = s->p_mv_table[xy][0];
2245 s->mv[0][0][1] = s->p_mv_table[xy][1];
2246 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2247 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2249 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2250 s->mv_dir = MV_DIR_FORWARD;
2251 s->mv_type = MV_TYPE_FIELD;
2254 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2255 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2256 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2258 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2259 &dmin, &next_block, 0, 0);
2261 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2262 s->mv_dir = MV_DIR_FORWARD;
2263 s->mv_type = MV_TYPE_16X16;
2267 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2268 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2270 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2271 s->mv_dir = MV_DIR_FORWARD;
2272 s->mv_type = MV_TYPE_8X8;
2275 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2276 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2278 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2279 &dmin, &next_block, 0, 0);
2281 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2282 s->mv_dir = MV_DIR_FORWARD;
2283 s->mv_type = MV_TYPE_16X16;
2285 s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2286 s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2287 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2288 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2290 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2291 s->mv_dir = MV_DIR_BACKWARD;
2292 s->mv_type = MV_TYPE_16X16;
2294 s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2295 s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2296 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2297 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2299 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2300 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2301 s->mv_type = MV_TYPE_16X16;
2303 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2304 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2305 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2306 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2307 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2308 &dmin, &next_block, 0, 0);
2310 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2311 s->mv_dir = MV_DIR_FORWARD;
2312 s->mv_type = MV_TYPE_FIELD;
2315 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2316 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2317 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2319 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2320 &dmin, &next_block, 0, 0);
2322 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2323 s->mv_dir = MV_DIR_BACKWARD;
2324 s->mv_type = MV_TYPE_FIELD;
2327 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2328 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2329 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2331 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2332 &dmin, &next_block, 0, 0);
2334 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2335 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2336 s->mv_type = MV_TYPE_FIELD;
2338 for(dir=0; dir<2; dir++){
2340 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2341 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2342 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2345 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2346 &dmin, &next_block, 0, 0);
2348 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2350 s->mv_type = MV_TYPE_16X16;
2354 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2355 &dmin, &next_block, 0, 0);
2356 if(s->h263_pred || s->h263_aic){
2358 s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2360 ff_clean_intra_table_entries(s); //old mode?
2364 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2365 if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2366 const int last_qp= backup_s.qscale;
2369 const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2370 static const int dquant_tab[4]={-1,1,-2,2};
2372 assert(backup_s.dquant == 0);
2375 s->mv_dir= best_s.mv_dir;
2376 s->mv_type = MV_TYPE_16X16;
2377 s->mb_intra= best_s.mb_intra;
2378 s->mv[0][0][0] = best_s.mv[0][0][0];
2379 s->mv[0][0][1] = best_s.mv[0][0][1];
2380 s->mv[1][0][0] = best_s.mv[1][0][0];
2381 s->mv[1][0][1] = best_s.mv[1][0][1];
2383 qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
2384 for(; qpi<4; qpi++){
2385 int dquant= dquant_tab[qpi];
2386 qp= last_qp + dquant;
2387 if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2389 backup_s.dquant= dquant;
2390 if(s->mb_intra && s->dc_val[0]){
2392 dc[i]= s->dc_val[0][ s->block_index[i] ];
2393 memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2397 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2398 &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2399 if(best_s.qscale != qp){
2400 if(s->mb_intra && s->dc_val[0]){
2402 s->dc_val[0][ s->block_index[i] ]= dc[i];
2403 memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2410 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2411 int mx= s->b_direct_mv_table[xy][0];
2412 int my= s->b_direct_mv_table[xy][1];
2414 backup_s.dquant = 0;
2415 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2417 ff_mpeg4_set_direct_mv(s, mx, my);
2418 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2419 &dmin, &next_block, mx, my);
2421 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2422 backup_s.dquant = 0;
2423 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2425 ff_mpeg4_set_direct_mv(s, 0, 0);
2426 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2427 &dmin, &next_block, 0, 0);
2429 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2432 coded |= s->block_last_index[i];
2435 memcpy(s->mv, best_s.mv, sizeof(s->mv));
2436 if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2437 mx=my=0; //FIXME find the one we actually used
2438 ff_mpeg4_set_direct_mv(s, mx, my);
2439 }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2447 s->mv_dir= best_s.mv_dir;
2448 s->mv_type = best_s.mv_type;
2450 /* s->mv[0][0][0] = best_s.mv[0][0][0];
2451 s->mv[0][0][1] = best_s.mv[0][0][1];
2452 s->mv[1][0][0] = best_s.mv[1][0][0];
2453 s->mv[1][0][1] = best_s.mv[1][0][1];*/
2456 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2457 &dmin, &next_block, mx, my);
2462 s->current_picture.qscale_table[xy]= best_s.qscale;
2464 copy_context_after_encode(s, &best_s, -1);
2466 pb_bits_count= put_bits_count(&s->pb);
2467 flush_put_bits(&s->pb);
2468 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2471 if(s->data_partitioning){
2472 pb2_bits_count= put_bits_count(&s->pb2);
2473 flush_put_bits(&s->pb2);
2474 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2475 s->pb2= backup_s.pb2;
2477 tex_pb_bits_count= put_bits_count(&s->tex_pb);
2478 flush_put_bits(&s->tex_pb);
2479 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2480 s->tex_pb= backup_s.tex_pb;
2482 s->last_bits= put_bits_count(&s->pb);
2484 if (CONFIG_H263_ENCODER &&
2485 s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2486 ff_h263_update_motion_val(s);
2488 if(next_block==0){ //FIXME 16 vs linesize16
2489 s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad , s->linesize ,16);
2490 s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
2491 s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2494 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2495 MPV_decode_mb(s, s->block);
2497 int motion_x = 0, motion_y = 0;
2498 s->mv_type=MV_TYPE_16X16;
2499 // only one MB-Type possible
2502 case CANDIDATE_MB_TYPE_INTRA:
2505 motion_x= s->mv[0][0][0] = 0;
2506 motion_y= s->mv[0][0][1] = 0;
2508 case CANDIDATE_MB_TYPE_INTER:
2509 s->mv_dir = MV_DIR_FORWARD;
2511 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2512 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2514 case CANDIDATE_MB_TYPE_INTER_I:
2515 s->mv_dir = MV_DIR_FORWARD;
2516 s->mv_type = MV_TYPE_FIELD;
2519 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2520 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2521 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2524 case CANDIDATE_MB_TYPE_INTER4V:
2525 s->mv_dir = MV_DIR_FORWARD;
2526 s->mv_type = MV_TYPE_8X8;
2529 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2530 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2533 case CANDIDATE_MB_TYPE_DIRECT:
2534 if (CONFIG_MPEG4_ENCODER) {
2535 s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2537 motion_x=s->b_direct_mv_table[xy][0];
2538 motion_y=s->b_direct_mv_table[xy][1];
2539 ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2542 case CANDIDATE_MB_TYPE_DIRECT0:
2543 if (CONFIG_MPEG4_ENCODER) {
2544 s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2546 ff_mpeg4_set_direct_mv(s, 0, 0);
2549 case CANDIDATE_MB_TYPE_BIDIR:
2550 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2552 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2553 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2554 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2555 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2557 case CANDIDATE_MB_TYPE_BACKWARD:
2558 s->mv_dir = MV_DIR_BACKWARD;
2560 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2561 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2563 case CANDIDATE_MB_TYPE_FORWARD:
2564 s->mv_dir = MV_DIR_FORWARD;
2566 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2567 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2568 // printf(" %d %d ", motion_x, motion_y);
2570 case CANDIDATE_MB_TYPE_FORWARD_I:
2571 s->mv_dir = MV_DIR_FORWARD;
2572 s->mv_type = MV_TYPE_FIELD;
2575 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2576 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2577 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2580 case CANDIDATE_MB_TYPE_BACKWARD_I:
2581 s->mv_dir = MV_DIR_BACKWARD;
2582 s->mv_type = MV_TYPE_FIELD;
2585 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2586 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2587 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2590 case CANDIDATE_MB_TYPE_BIDIR_I:
2591 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2592 s->mv_type = MV_TYPE_FIELD;
2594 for(dir=0; dir<2; dir++){
2596 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2597 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2598 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2603 av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2606 encode_mb(s, motion_x, motion_y);
2608 // RAL: Update last macroblock type
2609 s->last_mv_dir = s->mv_dir;
2611 if (CONFIG_H263_ENCODER &&
2612 s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2613 ff_h263_update_motion_val(s);
2615 MPV_decode_mb(s, s->block);
2618 /* clean the MV table in IPS frames for direct mode in B frames */
2619 if(s->mb_intra /* && I,P,S_TYPE */){
2620 s->p_mv_table[xy][0]=0;
2621 s->p_mv_table[xy][1]=0;
2624 if(s->flags&CODEC_FLAG_PSNR){
2628 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2629 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2631 s->current_picture.error[0] += sse(
2632 s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2633 s->dest[0], w, h, s->linesize);
2634 s->current_picture.error[1] += sse(
2635 s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
2636 s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2637 s->current_picture.error[2] += sse(
2638 s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
2639 s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2642 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2643 ff_h263_loop_filter(s);
2645 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2649 //not beautiful here but we must write it before flushing so it has to be here
2650 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
2651 msmpeg4_encode_ext_header(s);
2655 /* Send the last GOB if RTP */
2656 if (s->avctx->rtp_callback) {
2657 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2658 pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2659 /* Call the RTP callback to send the last GOB */
2661 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2667 #define MERGE(field) dst->field += src->field; src->field=0
2668 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2669 MERGE(me.scene_change_score);
2670 MERGE(me.mc_mb_var_sum_temp);
2671 MERGE(me.mb_var_sum_temp);
2674 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2677 MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2678 MERGE(dct_count[1]);
2688 MERGE(padding_bug_score);
2689 MERGE(current_picture.error[0]);
2690 MERGE(current_picture.error[1]);
2691 MERGE(current_picture.error[2]);
2693 if(dst->avctx->noise_reduction){
2694 for(i=0; i<64; i++){
2695 MERGE(dct_error_sum[0][i]);
2696 MERGE(dct_error_sum[1][i]);
2700 assert(put_bits_count(&src->pb) % 8 ==0);
2701 assert(put_bits_count(&dst->pb) % 8 ==0);
2702 ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2703 flush_put_bits(&dst->pb);
2706 static int estimate_qp(MpegEncContext *s, int dry_run){
2707 if (s->next_lambda){
2708 s->current_picture_ptr->quality=
2709 s->current_picture.quality = s->next_lambda;
2710 if(!dry_run) s->next_lambda= 0;
2711 } else if (!s->fixed_qscale) {
2712 s->current_picture_ptr->quality=
2713 s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2714 if (s->current_picture.quality < 0)
2718 if(s->adaptive_quant){
2719 switch(s->codec_id){
2720 case CODEC_ID_MPEG4:
2721 if (CONFIG_MPEG4_ENCODER)
2722 ff_clean_mpeg4_qscales(s);
2725 case CODEC_ID_H263P:
2727 if (CONFIG_H263_ENCODER)
2728 ff_clean_h263_qscales(s);
2731 ff_init_qscale_tab(s);
2734 s->lambda= s->lambda_table[0];
2737 s->lambda= s->current_picture.quality;
2738 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2743 /* must be called before writing the header */
2744 static void set_frame_distances(MpegEncContext * s){
2745 assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2746 s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2748 if(s->pict_type==AV_PICTURE_TYPE_B){
2749 s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2750 assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2752 s->pp_time= s->time - s->last_non_b_time;
2753 s->last_non_b_time= s->time;
2754 assert(s->picture_number==0 || s->pp_time > 0);
2758 static int encode_picture(MpegEncContext *s, int picture_number)
2762 int context_count = s->avctx->thread_count;
2764 s->picture_number = picture_number;
2766 /* Reset the average MB variance */
2767 s->me.mb_var_sum_temp =
2768 s->me.mc_mb_var_sum_temp = 0;
2770 /* we need to initialize some time vars before we can encode b-frames */
2771 // RAL: Condition added for MPEG1VIDEO
2772 if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
2773 set_frame_distances(s);
2774 if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2775 ff_set_mpeg4_time(s);
2777 s->me.scene_change_score=0;
2779 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2781 if(s->pict_type==AV_PICTURE_TYPE_I){
2782 if(s->msmpeg4_version >= 3) s->no_rounding=1;
2783 else s->no_rounding=0;
2784 }else if(s->pict_type!=AV_PICTURE_TYPE_B){
2785 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2786 s->no_rounding ^= 1;
2789 if(s->flags & CODEC_FLAG_PASS2){
2790 if (estimate_qp(s,1) < 0)
2792 ff_get_2pass_fcode(s);
2793 }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2794 if(s->pict_type==AV_PICTURE_TYPE_B)
2795 s->lambda= s->last_lambda_for[s->pict_type];
2797 s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2801 s->mb_intra=0; //for the rate distortion & bit compare functions
2802 for(i=1; i<context_count; i++){
2803 ff_update_duplicate_context(s->thread_context[i], s);
2809 /* Estimate motion for every MB */
2810 if(s->pict_type != AV_PICTURE_TYPE_I){
2811 s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2812 s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2813 if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
2814 if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
2815 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2819 s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2820 }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
2822 for(i=0; i<s->mb_stride*s->mb_height; i++)
2823 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2825 if(!s->fixed_qscale){
2826 /* finding spatial complexity for I-frame rate control */
2827 s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2830 for(i=1; i<context_count; i++){
2831 merge_context_after_me(s, s->thread_context[i]);
2833 s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2834 s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
2837 if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == AV_PICTURE_TYPE_P){
2838 s->pict_type= AV_PICTURE_TYPE_I;
2839 for(i=0; i<s->mb_stride*s->mb_height; i++)
2840 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2841 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2845 if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
2846 s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2848 if(s->flags & CODEC_FLAG_INTERLACED_ME){
2850 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2851 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2852 s->f_code= FFMAX3(s->f_code, a, b);
2855 ff_fix_long_p_mvs(s);
2856 ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2857 if(s->flags & CODEC_FLAG_INTERLACED_ME){
2861 ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2862 s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2867 if(s->pict_type==AV_PICTURE_TYPE_B){
2870 a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2871 b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2872 s->f_code = FFMAX(a, b);
2874 a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2875 b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2876 s->b_code = FFMAX(a, b);
2878 ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2879 ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2880 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2881 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2882 if(s->flags & CODEC_FLAG_INTERLACED_ME){
2884 for(dir=0; dir<2; dir++){
2887 int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2888 : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2889 ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2890 s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2898 if (estimate_qp(s, 0) < 0)
2901 if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==AV_PICTURE_TYPE_I && !(s->flags & CODEC_FLAG_QSCALE))
2902 s->qscale= 3; //reduce clipping problems
2904 if (s->out_format == FMT_MJPEG) {
2905 /* for mjpeg, we do include qscale in the matrix */
2907 int j= s->dsp.idct_permutation[i];
2909 s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2911 s->y_dc_scale_table=
2912 s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
2913 s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
2914 ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2915 s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2919 //FIXME var duplication
2920 s->current_picture_ptr->key_frame=
2921 s->current_picture.key_frame= s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
2922 s->current_picture_ptr->pict_type=
2923 s->current_picture.pict_type= s->pict_type;
2925 if(s->current_picture.key_frame)
2926 s->picture_in_gop_number=0;
2928 s->last_bits= put_bits_count(&s->pb);
2929 switch(s->out_format) {
2931 if (CONFIG_MJPEG_ENCODER)
2932 ff_mjpeg_encode_picture_header(s);
2935 if (CONFIG_H261_ENCODER)
2936 ff_h261_encode_picture_header(s, picture_number);
2939 if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2940 ff_wmv2_encode_picture_header(s, picture_number);
2941 else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
2942 msmpeg4_encode_picture_header(s, picture_number);
2943 else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2944 mpeg4_encode_picture_header(s, picture_number);
2945 else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2946 rv10_encode_picture_header(s, picture_number);
2947 else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2948 rv20_encode_picture_header(s, picture_number);
2949 else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2950 ff_flv_encode_picture_header(s, picture_number);
2951 else if (CONFIG_H263_ENCODER)
2952 h263_encode_picture_header(s, picture_number);
2955 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2956 mpeg1_encode_picture_header(s, picture_number);
2963 bits= put_bits_count(&s->pb);
2964 s->header_bits= bits - s->last_bits;
2966 for(i=1; i<context_count; i++){
2967 update_duplicate_context_after_me(s->thread_context[i], s);
2969 s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2970 for(i=1; i<context_count; i++){
2971 merge_context_after_encode(s, s->thread_context[i]);
2977 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2978 const int intra= s->mb_intra;
2981 s->dct_count[intra]++;
2983 for(i=0; i<64; i++){
2984 int level= block[i];
2988 s->dct_error_sum[intra][i] += level;
2989 level -= s->dct_offset[intra][i];
2990 if(level<0) level=0;
2992 s->dct_error_sum[intra][i] -= level;
2993 level += s->dct_offset[intra][i];
2994 if(level>0) level=0;
3001 static int dct_quantize_trellis_c(MpegEncContext *s,
3002 DCTELEM *block, int n,
3003 int qscale, int *overflow){
3005 const uint8_t *scantable= s->intra_scantable.scantable;
3006 const uint8_t *perm_scantable= s->intra_scantable.permutated;
3008 unsigned int threshold1, threshold2;
3020 int coeff_count[64];
3021 int qmul, qadd, start_i, last_non_zero, i, dc;
3022 const int esc_length= s->ac_esc_length;
3024 uint8_t * last_length;
3025 const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3027 s->dsp.fdct (block);
3029 if(s->dct_error_sum)
3030 s->denoise_dct(s, block);
3032 qadd= ((qscale-1)|1)*8;
3043 /* For AIC we skip quant/dequant of INTRADC */
3048 /* note: block[0] is assumed to be positive */
3049 block[0] = (block[0] + (q >> 1)) / q;
3052 qmat = s->q_intra_matrix[qscale];
3053 if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3054 bias= 1<<(QMAT_SHIFT-1);
3055 length = s->intra_ac_vlc_length;
3056 last_length= s->intra_ac_vlc_last_length;
3060 qmat = s->q_inter_matrix[qscale];
3061 length = s->inter_ac_vlc_length;
3062 last_length= s->inter_ac_vlc_last_length;
3066 threshold1= (1<<QMAT_SHIFT) - bias - 1;
3067 threshold2= (threshold1<<1);
3069 for(i=63; i>=start_i; i--) {
3070 const int j = scantable[i];
3071 int level = block[j] * qmat[j];
3073 if(((unsigned)(level+threshold1))>threshold2){
3079 for(i=start_i; i<=last_non_zero; i++) {
3080 const int j = scantable[i];
3081 int level = block[j] * qmat[j];
3083 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
3084 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
3085 if(((unsigned)(level+threshold1))>threshold2){
3087 level= (bias + level)>>QMAT_SHIFT;
3089 coeff[1][i]= level-1;
3090 // coeff[2][k]= level-2;
3092 level= (bias - level)>>QMAT_SHIFT;
3093 coeff[0][i]= -level;
3094 coeff[1][i]= -level+1;
3095 // coeff[2][k]= -level+2;
3097 coeff_count[i]= FFMIN(level, 2);
3098 assert(coeff_count[i]);
3101 coeff[0][i]= (level>>31)|1;
3106 *overflow= s->max_qcoeff < max; //overflow might have happened
3108 if(last_non_zero < start_i){
3109 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3110 return last_non_zero;
3113 score_tab[start_i]= 0;
3114 survivor[0]= start_i;
3117 for(i=start_i; i<=last_non_zero; i++){
3118 int level_index, j, zero_distortion;
3119 int dct_coeff= FFABS(block[ scantable[i] ]);
3120 int best_score=256*256*256*120;
3122 if ( s->dsp.fdct == fdct_ifast
3123 #ifndef FAAN_POSTSCALE
3124 || s->dsp.fdct == ff_faandct
3127 dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3128 zero_distortion= dct_coeff*dct_coeff;
3130 for(level_index=0; level_index < coeff_count[i]; level_index++){
3132 int level= coeff[level_index][i];
3133 const int alevel= FFABS(level);
3138 if(s->out_format == FMT_H263){
3139 unquant_coeff= alevel*qmul + qadd;
3141 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3143 unquant_coeff = (int)( alevel * qscale * s->intra_matrix[j]) >> 3;
3144 unquant_coeff = (unquant_coeff - 1) | 1;
3146 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3147 unquant_coeff = (unquant_coeff - 1) | 1;
3152 distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3154 if((level&(~127)) == 0){
3155 for(j=survivor_count-1; j>=0; j--){
3156 int run= i - survivor[j];
3157 int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3158 score += score_tab[i-run];
3160 if(score < best_score){
3163 level_tab[i+1]= level-64;
3167 if(s->out_format == FMT_H263){
3168 for(j=survivor_count-1; j>=0; j--){
3169 int run= i - survivor[j];
3170 int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3171 score += score_tab[i-run];
3172 if(score < last_score){
3175 last_level= level-64;
3181 distortion += esc_length*lambda;
3182 for(j=survivor_count-1; j>=0; j--){
3183 int run= i - survivor[j];
3184 int score= distortion + score_tab[i-run];
3186 if(score < best_score){
3189 level_tab[i+1]= level-64;
3193 if(s->out_format == FMT_H263){
3194 for(j=survivor_count-1; j>=0; j--){
3195 int run= i - survivor[j];
3196 int score= distortion + score_tab[i-run];
3197 if(score < last_score){
3200 last_level= level-64;
3208 score_tab[i+1]= best_score;
3210 //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3211 if(last_non_zero <= 27){
3212 for(; survivor_count; survivor_count--){
3213 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3217 for(; survivor_count; survivor_count--){
3218 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3223 survivor[ survivor_count++ ]= i+1;
3226 if(s->out_format != FMT_H263){
3227 last_score= 256*256*256*120;
3228 for(i= survivor[0]; i<=last_non_zero + 1; i++){
3229 int score= score_tab[i];
3230 if(i) score += lambda*2; //FIXME exacter?
3232 if(score < last_score){
3235 last_level= level_tab[i];
3236 last_run= run_tab[i];
3241 s->coded_score[n] = last_score;
3243 dc= FFABS(block[0]);
3244 last_non_zero= last_i - 1;
3245 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3247 if(last_non_zero < start_i)
3248 return last_non_zero;
3250 if(last_non_zero == 0 && start_i == 0){
3252 int best_score= dc * dc;
3254 for(i=0; i<coeff_count[0]; i++){
3255 int level= coeff[i][0];
3256 int alevel= FFABS(level);
3257 int unquant_coeff, score, distortion;
3259 if(s->out_format == FMT_H263){
3260 unquant_coeff= (alevel*qmul + qadd)>>3;
3262 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3263 unquant_coeff = (unquant_coeff - 1) | 1;
3265 unquant_coeff = (unquant_coeff + 4) >> 3;
3266 unquant_coeff<<= 3 + 3;
3268 distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3270 if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3271 else score= distortion + esc_length*lambda;
3273 if(score < best_score){
3275 best_level= level - 64;
3278 block[0]= best_level;
3279 s->coded_score[n] = best_score - dc*dc;
3280 if(best_level == 0) return -1;
3281 else return last_non_zero;
3287 block[ perm_scantable[last_non_zero] ]= last_level;
3290 for(; i>start_i; i -= run_tab[i] + 1){
3291 block[ perm_scantable[i-1] ]= level_tab[i];
3294 return last_non_zero;
3297 //#define REFINE_STATS 1
3298 static int16_t basis[64][64];
3300 static void build_basis(uint8_t *perm){
3307 double s= 0.25*(1<<BASIS_SHIFT);
3309 int perm_index= perm[index];
3310 if(i==0) s*= sqrt(0.5);
3311 if(j==0) s*= sqrt(0.5);
3312 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)));
3319 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3320 DCTELEM *block, int16_t *weight, DCTELEM *orig,
3323 LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
3324 const uint8_t *scantable= s->intra_scantable.scantable;
3325 const uint8_t *perm_scantable= s->intra_scantable.permutated;
3326 // unsigned int threshold1, threshold2;
3331 int qmul, qadd, start_i, last_non_zero, i, dc;
3333 uint8_t * last_length;
3335 int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3338 static int after_last=0;
3339 static int to_zero=0;
3340 static int from_zero=0;
3343 static int messed_sign=0;
3346 if(basis[0][0] == 0)
3347 build_basis(s->dsp.idct_permutation);
3358 /* For AIC we skip quant/dequant of INTRADC */
3362 q <<= RECON_SHIFT-3;
3363 /* note: block[0] is assumed to be positive */
3365 // block[0] = (block[0] + (q >> 1)) / q;
3367 // if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3368 // bias= 1<<(QMAT_SHIFT-1);
3369 length = s->intra_ac_vlc_length;
3370 last_length= s->intra_ac_vlc_last_length;
3374 length = s->inter_ac_vlc_length;
3375 last_length= s->inter_ac_vlc_last_length;
3377 last_non_zero = s->block_last_index[n];
3382 dc += (1<<(RECON_SHIFT-1));
3383 for(i=0; i<64; i++){
3384 rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME use orig dirrectly instead of copying to rem[]
3387 STOP_TIMER("memset rem[]")}
3390 for(i=0; i<64; i++){
3395 w= FFABS(weight[i]) + qns*one;
3396 w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3399 // w=weight[i] = (63*qns + (w/2)) / w;
3405 lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3411 for(i=start_i; i<=last_non_zero; i++){
3412 int j= perm_scantable[i];
3413 const int level= block[j];
3417 if(level<0) coeff= qmul*level - qadd;
3418 else coeff= qmul*level + qadd;
3419 run_tab[rle_index++]=run;
3422 s->dsp.add_8x8basis(rem, basis[j], coeff);
3428 if(last_non_zero>0){
3429 STOP_TIMER("init rem[]")
3436 int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3439 int run2, best_unquant_change=0, analyze_gradient;
3443 analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3445 if(analyze_gradient){
3449 for(i=0; i<64; i++){
3452 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3455 STOP_TIMER("rem*w*w")}
3465 const int level= block[0];
3466 int change, old_coeff;
3468 assert(s->mb_intra);
3472 for(change=-1; change<=1; change+=2){
3473 int new_level= level + change;
3474 int score, new_coeff;
3476 new_coeff= q*new_level;
3477 if(new_coeff >= 2048 || new_coeff < 0)
3480 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3481 if(score<best_score){
3484 best_change= change;
3485 best_unquant_change= new_coeff - old_coeff;
3492 run2= run_tab[rle_index++];
3496 for(i=start_i; i<64; i++){
3497 int j= perm_scantable[i];
3498 const int level= block[j];
3499 int change, old_coeff;
3501 if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3505 if(level<0) old_coeff= qmul*level - qadd;
3506 else old_coeff= qmul*level + qadd;
3507 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3511 assert(run2>=0 || i >= last_non_zero );
3514 for(change=-1; change<=1; change+=2){
3515 int new_level= level + change;
3516 int score, new_coeff, unquant_change;
3519 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3523 if(new_level<0) new_coeff= qmul*new_level - qadd;
3524 else new_coeff= qmul*new_level + qadd;
3525 if(new_coeff >= 2048 || new_coeff <= -2048)
3527 //FIXME check for overflow
3530 if(level < 63 && level > -63){
3531 if(i < last_non_zero)
3532 score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
3533 - length[UNI_AC_ENC_INDEX(run, level+64)];
3535 score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3536 - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3539 assert(FFABS(new_level)==1);
3541 if(analyze_gradient){
3542 int g= d1[ scantable[i] ];
3543 if(g && (g^new_level) >= 0)
3547 if(i < last_non_zero){
3548 int next_i= i + run2 + 1;
3549 int next_level= block[ perm_scantable[next_i] ] + 64;
3551 if(next_level&(~127))
3554 if(next_i < last_non_zero)
3555 score += length[UNI_AC_ENC_INDEX(run, 65)]
3556 + length[UNI_AC_ENC_INDEX(run2, next_level)]
3557 - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3559 score += length[UNI_AC_ENC_INDEX(run, 65)]
3560 + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3561 - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3563 score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3565 score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3566 - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3572 assert(FFABS(level)==1);
3574 if(i < last_non_zero){
3575 int next_i= i + run2 + 1;
3576 int next_level= block[ perm_scantable[next_i] ] + 64;
3578 if(next_level&(~127))
3581 if(next_i < last_non_zero)
3582 score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3583 - length[UNI_AC_ENC_INDEX(run2, next_level)]
3584 - length[UNI_AC_ENC_INDEX(run, 65)];
3586 score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3587 - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3588 - length[UNI_AC_ENC_INDEX(run, 65)];
3590 score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3592 score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3593 - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3600 unquant_change= new_coeff - old_coeff;
3601 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3603 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3604 if(score<best_score){
3607 best_change= change;
3608 best_unquant_change= unquant_change;
3612 prev_level= level + 64;
3613 if(prev_level&(~127))
3622 STOP_TIMER("iterative step")}
3626 int j= perm_scantable[ best_coeff ];
3628 block[j] += best_change;
3630 if(best_coeff > last_non_zero){
3631 last_non_zero= best_coeff;
3639 if(block[j] - best_change){
3640 if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3652 for(; last_non_zero>=start_i; last_non_zero--){
3653 if(block[perm_scantable[last_non_zero]])
3659 if(256*256*256*64 % count == 0){
3660 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);
3665 for(i=start_i; i<=last_non_zero; i++){
3666 int j= perm_scantable[i];
3667 const int level= block[j];
3670 run_tab[rle_index++]=run;
3677 s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3683 if(last_non_zero>0){
3684 STOP_TIMER("iterative search")
3689 return last_non_zero;
3692 int dct_quantize_c(MpegEncContext *s,
3693 DCTELEM *block, int n,
3694 int qscale, int *overflow)
3696 int i, j, level, last_non_zero, q, start_i;
3698 const uint8_t *scantable= s->intra_scantable.scantable;
3701 unsigned int threshold1, threshold2;
3703 s->dsp.fdct (block);
3705 if(s->dct_error_sum)
3706 s->denoise_dct(s, block);
3716 /* For AIC we skip quant/dequant of INTRADC */
3719 /* note: block[0] is assumed to be positive */
3720 block[0] = (block[0] + (q >> 1)) / q;
3723 qmat = s->q_intra_matrix[qscale];
3724 bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3728 qmat = s->q_inter_matrix[qscale];
3729 bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3731 threshold1= (1<<QMAT_SHIFT) - bias - 1;
3732 threshold2= (threshold1<<1);
3733 for(i=63;i>=start_i;i--) {
3735 level = block[j] * qmat[j];
3737 if(((unsigned)(level+threshold1))>threshold2){
3744 for(i=start_i; i<=last_non_zero; i++) {
3746 level = block[j] * qmat[j];
3748 // if( bias+level >= (1<<QMAT_SHIFT)
3749 // || bias-level >= (1<<QMAT_SHIFT)){
3750 if(((unsigned)(level+threshold1))>threshold2){
3752 level= (bias + level)>>QMAT_SHIFT;
3755 level= (bias - level)>>QMAT_SHIFT;
3763 *overflow= s->max_qcoeff < max; //overflow might have happened
3765 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3766 if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3767 ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3769 return last_non_zero;
3772 AVCodec ff_h263_encoder = {
3776 sizeof(MpegEncContext),
3780 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3781 .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
3784 AVCodec ff_h263p_encoder = {
3788 sizeof(MpegEncContext),
3792 .capabilities = CODEC_CAP_SLICE_THREADS,
3793 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3794 .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
3797 AVCodec ff_msmpeg4v2_encoder = {
3801 sizeof(MpegEncContext),
3805 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3806 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3809 AVCodec ff_msmpeg4v3_encoder = {
3813 sizeof(MpegEncContext),
3817 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3818 .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3821 AVCodec ff_wmv1_encoder = {
3825 sizeof(MpegEncContext),
3829 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3830 .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),