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 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
26 * @file mpegvideo_enc.c
27 * The simplest mpeg encoder (well, it was the simplest!).
32 #include "mpegvideo.h"
33 #include "mpegvideo_common.h"
42 static int encode_picture(MpegEncContext *s, int picture_number);
43 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
44 static int sse_mb(MpegEncContext *s);
46 /* enable all paranoid tests for rounding, overflows, etc... */
51 static const uint16_t aanscales[64] = {
52 /* precomputed values scaled up by 14 bits */
53 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
54 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
55 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
56 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
57 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
58 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
59 8867 , 12299, 11585, 10426, 8867, 6967, 4799, 2446,
60 4520 , 6270, 5906, 5315, 4520, 3552, 2446, 1247
63 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
64 static uint8_t default_fcode_tab[MAX_MV*2+1];
66 enum PixelFormat ff_yuv420p_list[2]= {PIX_FMT_YUV420P, -1};
68 static void convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
69 const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
74 for(qscale=qmin; qscale<=qmax; qscale++){
76 if (dsp->fdct == ff_jpeg_fdct_islow
78 || dsp->fdct == ff_faandct
82 const int j= dsp->idct_permutation[i];
83 /* 16 <= qscale * quant_matrix[i] <= 7905 */
84 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
85 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
86 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
88 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
89 (qscale * quant_matrix[j]));
91 } else if (dsp->fdct == fdct_ifast
92 #ifndef FAAN_POSTSCALE
93 || dsp->fdct == ff_faandct
97 const int j= dsp->idct_permutation[i];
98 /* 16 <= qscale * quant_matrix[i] <= 7905 */
99 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
100 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
101 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
103 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
104 (aanscales[i] * qscale * quant_matrix[j]));
108 const int j= dsp->idct_permutation[i];
109 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
110 So 16 <= qscale * quant_matrix[i] <= 7905
111 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
112 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
114 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
115 // qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
116 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
118 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
119 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
123 for(i=intra; i<64; i++){
125 if (dsp->fdct == fdct_ifast
126 #ifndef FAAN_POSTSCALE
127 || dsp->fdct == ff_faandct
130 max= (8191LL*aanscales[i]) >> 14;
132 while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
138 av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
142 static inline void update_qscale(MpegEncContext *s){
143 s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
144 s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
146 s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
149 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
155 put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
161 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
164 dst->pict_type = src->pict_type;
165 dst->quality = src->quality;
166 dst->coded_picture_number = src->coded_picture_number;
167 dst->display_picture_number = src->display_picture_number;
168 // dst->reference = src->reference;
170 dst->interlaced_frame = src->interlaced_frame;
171 dst->top_field_first = src->top_field_first;
173 if(s->avctx->me_threshold){
174 if(!src->motion_val[0])
175 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
177 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
178 if(!src->ref_index[0])
179 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
180 if(src->motion_subsample_log2 != dst->motion_subsample_log2)
181 av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
182 src->motion_subsample_log2, dst->motion_subsample_log2);
184 memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
187 int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
188 int height= ((16*s->mb_height)>>src->motion_subsample_log2);
190 if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
191 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
193 if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
194 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
200 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
201 #define COPY(a) dst->a= src->a
203 COPY(current_picture);
209 COPY(picture_in_gop_number);
210 COPY(gop_picture_number);
211 COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
212 COPY(progressive_frame); //FIXME don't set in encode_header
213 COPY(partitioned_frame); //FIXME don't set in encode_header
218 * sets the given MpegEncContext to defaults for encoding.
219 * the changed fields will not depend upon the prior state of the MpegEncContext.
221 static void MPV_encode_defaults(MpegEncContext *s){
223 MPV_common_defaults(s);
225 for(i=-16; i<16; i++){
226 default_fcode_tab[i + MAX_MV]= 1;
228 s->me.mv_penalty= default_mv_penalty;
229 s->fcode_tab= default_fcode_tab;
232 /* init video encoder */
233 int MPV_encode_init(AVCodecContext *avctx)
235 MpegEncContext *s = avctx->priv_data;
237 int chroma_h_shift, chroma_v_shift;
239 MPV_encode_defaults(s);
241 switch (avctx->codec_id) {
242 case CODEC_ID_MPEG2VIDEO:
243 if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
244 av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
250 if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
251 ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
252 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
257 if(avctx->pix_fmt != PIX_FMT_YUV420P){
258 av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
263 switch (avctx->pix_fmt) {
264 case PIX_FMT_YUVJ422P:
265 case PIX_FMT_YUV422P:
266 s->chroma_format = CHROMA_422;
268 case PIX_FMT_YUVJ420P:
269 case PIX_FMT_YUV420P:
271 s->chroma_format = CHROMA_420;
275 s->bit_rate = avctx->bit_rate;
276 s->width = avctx->width;
277 s->height = avctx->height;
278 if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
279 av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
282 s->gop_size = avctx->gop_size;
284 s->flags= avctx->flags;
285 s->flags2= avctx->flags2;
286 s->max_b_frames= avctx->max_b_frames;
287 s->codec_id= avctx->codec->id;
288 s->luma_elim_threshold = avctx->luma_elim_threshold;
289 s->chroma_elim_threshold= avctx->chroma_elim_threshold;
290 s->strict_std_compliance= avctx->strict_std_compliance;
291 s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
292 s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
293 s->mpeg_quant= avctx->mpeg_quant;
294 s->rtp_mode= !!avctx->rtp_payload_size;
295 s->intra_dc_precision= avctx->intra_dc_precision;
296 s->user_specified_pts = AV_NOPTS_VALUE;
298 if (s->gop_size <= 1) {
305 s->me_method = avctx->me_method;
308 s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
310 s->adaptive_quant= ( s->avctx->lumi_masking
311 || s->avctx->dark_masking
312 || s->avctx->temporal_cplx_masking
313 || s->avctx->spatial_cplx_masking
314 || s->avctx->p_masking
315 || s->avctx->border_masking
316 || (s->flags&CODEC_FLAG_QP_RD))
319 s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
320 s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
321 s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
322 s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
323 s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
325 if(avctx->rc_max_rate && !avctx->rc_buffer_size){
326 av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
330 if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
331 av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
334 if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
335 av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
339 if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
340 av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
344 if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
345 av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
349 if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
350 av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
354 if( s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
355 && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
356 && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
358 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");
361 if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
362 && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
363 av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
367 if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
368 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
372 if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
373 av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
377 if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
378 av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
382 if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
383 av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
387 if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
388 av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
392 if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
393 && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
394 av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
398 if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
399 av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
403 if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){
404 av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
408 if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
409 av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
413 if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
414 av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection arent supported yet, set threshold to 1000000000\n");
418 if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
419 av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
423 if(s->flags & CODEC_FLAG_LOW_DELAY){
424 if (s->codec_id != CODEC_ID_MPEG2VIDEO && s->codec_id != CODEC_ID_MPEG1VIDEO){
425 av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg1/2\n");
428 if (s->max_b_frames != 0){
429 av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
434 if(s->q_scale_type == 1){
435 if(s->codec_id != CODEC_ID_MPEG2VIDEO){
436 av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
439 if(avctx->qmax > 12){
440 av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
445 if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
446 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
447 && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
448 av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
452 if(s->avctx->thread_count > 1)
455 if(!avctx->time_base.den || !avctx->time_base.num){
456 av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
460 i= (INT_MAX/2+128)>>8;
461 if(avctx->me_threshold >= i){
462 av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
465 if(avctx->mb_threshold >= i){
466 av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
470 if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
471 av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
472 avctx->b_frame_strategy = 0;
475 i= ff_gcd(avctx->time_base.den, avctx->time_base.num);
477 av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
478 avctx->time_base.den /= i;
479 avctx->time_base.num /= i;
483 if(s->codec_id==CODEC_ID_MJPEG){
484 s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
485 s->inter_quant_bias= 0;
486 }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
487 s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
488 s->inter_quant_bias= 0;
490 s->intra_quant_bias=0;
491 s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
494 if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
495 s->intra_quant_bias= avctx->intra_quant_bias;
496 if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
497 s->inter_quant_bias= avctx->inter_quant_bias;
499 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
501 if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
502 av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
505 s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
507 switch(avctx->codec->id) {
508 case CODEC_ID_MPEG1VIDEO:
509 s->out_format = FMT_MPEG1;
510 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
511 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
513 case CODEC_ID_MPEG2VIDEO:
514 s->out_format = FMT_MPEG1;
515 s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
516 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
521 s->out_format = FMT_MJPEG;
522 s->intra_only = 1; /* force intra only for jpeg */
523 s->mjpeg_vsample[0] = 2;
524 s->mjpeg_vsample[1] = 2>>chroma_v_shift;
525 s->mjpeg_vsample[2] = 2>>chroma_v_shift;
526 s->mjpeg_hsample[0] = 2;
527 s->mjpeg_hsample[1] = 2>>chroma_h_shift;
528 s->mjpeg_hsample[2] = 2>>chroma_h_shift;
529 if (!(ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER)
530 || ff_mjpeg_encode_init(s) < 0)
536 if (!ENABLE_H261_ENCODER) return -1;
537 if (ff_h261_get_picture_format(s->width, s->height) < 0) {
538 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);
541 s->out_format = FMT_H261;
546 if (h263_get_picture_format(s->width, s->height) == 7) {
547 av_log(avctx, AV_LOG_INFO, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height);
550 s->out_format = FMT_H263;
551 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
556 s->out_format = FMT_H263;
559 s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
560 s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
561 s->modified_quant= s->h263_aic;
562 s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
563 s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
564 s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
565 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
566 s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
569 /* These are just to be sure */
574 s->out_format = FMT_H263;
575 s->h263_flv = 2; /* format = 1; 11-bit codes */
576 s->unrestricted_mv = 1;
577 s->rtp_mode=0; /* don't allow GOB */
582 s->out_format = FMT_H263;
587 s->out_format = FMT_H263;
594 s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
597 s->out_format = FMT_H263;
599 s->unrestricted_mv = 1;
600 s->low_delay= s->max_b_frames ? 0 : 1;
601 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
603 case CODEC_ID_MSMPEG4V1:
604 s->out_format = FMT_H263;
607 s->unrestricted_mv = 1;
608 s->msmpeg4_version= 1;
612 case CODEC_ID_MSMPEG4V2:
613 s->out_format = FMT_H263;
616 s->unrestricted_mv = 1;
617 s->msmpeg4_version= 2;
621 case CODEC_ID_MSMPEG4V3:
622 s->out_format = FMT_H263;
625 s->unrestricted_mv = 1;
626 s->msmpeg4_version= 3;
627 s->flipflop_rounding=1;
632 s->out_format = FMT_H263;
635 s->unrestricted_mv = 1;
636 s->msmpeg4_version= 4;
637 s->flipflop_rounding=1;
642 s->out_format = FMT_H263;
645 s->unrestricted_mv = 1;
646 s->msmpeg4_version= 5;
647 s->flipflop_rounding=1;
655 avctx->has_b_frames= !s->low_delay;
660 if (MPV_common_init(s) < 0)
663 if(s->modified_quant)
664 s->chroma_qscale_table= ff_h263_chroma_qscale_table;
665 s->progressive_frame=
666 s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
667 s->quant_precision=5;
669 ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
670 ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
672 if (ENABLE_H261_ENCODER && s->out_format == FMT_H261)
673 ff_h261_encode_init(s);
674 if (s->out_format == FMT_H263)
676 if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version)
677 ff_msmpeg4_encode_init(s);
678 if ((ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
679 && s->out_format == FMT_MPEG1)
680 ff_mpeg1_encode_init(s);
684 int j= s->dsp.idct_permutation[i];
685 if(s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
686 s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
687 s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
688 }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
690 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
693 s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
694 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
696 if(s->avctx->intra_matrix)
697 s->intra_matrix[j] = s->avctx->intra_matrix[i];
698 if(s->avctx->inter_matrix)
699 s->inter_matrix[j] = s->avctx->inter_matrix[i];
702 /* precompute matrix */
703 /* for mjpeg, we do include qscale in the matrix */
704 if (s->out_format != FMT_MJPEG) {
705 convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
706 s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
707 convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
708 s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
711 if(ff_rate_control_init(s) < 0)
717 int MPV_encode_end(AVCodecContext *avctx)
719 MpegEncContext *s = avctx->priv_data;
721 ff_rate_control_uninit(s);
724 if ((ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
725 ff_mjpeg_encode_close(s);
727 av_freep(&avctx->extradata);
732 static int get_sae(uint8_t *src, int ref, int stride){
738 acc+= FFABS(src[x+y*stride] - ref);
745 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
752 for(y=0; y<h; y+=16){
753 for(x=0; x<w; x+=16){
754 int offset= x + y*stride;
755 int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
756 int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
757 int sae = get_sae(src + offset, mean, stride);
759 acc+= sae + 500 < sad;
766 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
770 const int encoding_delay= s->max_b_frames;
775 pic_arg->display_picture_number= s->input_picture_number++;
777 if(pts != AV_NOPTS_VALUE){
778 if(s->user_specified_pts != AV_NOPTS_VALUE){
780 int64_t last= s->user_specified_pts;
783 av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
787 s->user_specified_pts= pts;
789 if(s->user_specified_pts != AV_NOPTS_VALUE){
790 s->user_specified_pts=
791 pts= s->user_specified_pts + 1;
792 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
794 pts= pic_arg->display_picture_number;
800 if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
801 if(pic_arg->linesize[0] != s->linesize) direct=0;
802 if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
803 if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
805 // av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
808 i= ff_find_unused_picture(s, 1);
810 pic= (AVFrame*)&s->picture[i];
814 pic->data[i]= pic_arg->data[i];
815 pic->linesize[i]= pic_arg->linesize[i];
817 alloc_picture(s, (Picture*)pic, 1);
819 i= ff_find_unused_picture(s, 0);
821 pic= (AVFrame*)&s->picture[i];
824 alloc_picture(s, (Picture*)pic, 0);
826 if( pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
827 && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
828 && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
831 int h_chroma_shift, v_chroma_shift;
832 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
835 int src_stride= pic_arg->linesize[i];
836 int dst_stride= i ? s->uvlinesize : s->linesize;
837 int h_shift= i ? h_chroma_shift : 0;
838 int v_shift= i ? v_chroma_shift : 0;
839 int w= s->width >>h_shift;
840 int h= s->height>>v_shift;
841 uint8_t *src= pic_arg->data[i];
842 uint8_t *dst= pic->data[i];
844 if(!s->avctx->rc_buffer_size)
845 dst +=INPLACE_OFFSET;
847 if(src_stride==dst_stride)
848 memcpy(dst, src, src_stride*h);
859 copy_picture_attributes(s, pic, pic_arg);
860 pic->pts= pts; //we set this here to avoid modifiying pic_arg
863 /* shift buffer entries */
864 for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
865 s->input_picture[i-1]= s->input_picture[i];
867 s->input_picture[encoding_delay]= (Picture*)pic;
872 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
877 for(plane=0; plane<3; plane++){
878 const int stride= p->linesize[plane];
879 const int bw= plane ? 1 : 2;
880 for(y=0; y<s->mb_height*bw; y++){
881 for(x=0; x<s->mb_width*bw; x++){
882 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
883 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);
885 switch(s->avctx->frame_skip_exp){
886 case 0: score= FFMAX(score, v); break;
887 case 1: score+= FFABS(v);break;
888 case 2: score+= v*v;break;
889 case 3: score64+= FFABS(v*v*(int64_t)v);break;
890 case 4: score64+= v*v*(int64_t)(v*v);break;
896 if(score) score64= score;
898 if(score64 < s->avctx->frame_skip_threshold)
900 if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
905 static int estimate_best_b_count(MpegEncContext *s){
906 AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
907 AVCodecContext *c= avcodec_alloc_context();
908 AVFrame input[FF_MAX_B_FRAMES+2];
909 const int scale= s->avctx->brd_scale;
910 int i, j, out_size, p_lambda, b_lambda, lambda2;
911 int outbuf_size= s->width * s->height; //FIXME
912 uint8_t *outbuf= av_malloc(outbuf_size);
913 int64_t best_rd= INT64_MAX;
914 int best_b_count= -1;
916 assert(scale>=0 && scale <=3);
919 p_lambda= s->last_lambda_for[P_TYPE]; //s->next_picture_ptr->quality;
920 b_lambda= s->last_lambda_for[B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
921 if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
922 lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
924 c->width = s->width >> scale;
925 c->height= s->height>> scale;
926 c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
927 c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
928 c->mb_decision= s->avctx->mb_decision;
929 c->me_cmp= s->avctx->me_cmp;
930 c->mb_cmp= s->avctx->mb_cmp;
931 c->me_sub_cmp= s->avctx->me_sub_cmp;
932 c->pix_fmt = PIX_FMT_YUV420P;
933 c->time_base= s->avctx->time_base;
934 c->max_b_frames= s->max_b_frames;
936 if (avcodec_open(c, codec) < 0)
939 for(i=0; i<s->max_b_frames+2; i++){
940 int ysize= c->width*c->height;
941 int csize= (c->width/2)*(c->height/2);
942 Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
944 avcodec_get_frame_defaults(&input[i]);
945 input[i].data[0]= av_malloc(ysize + 2*csize);
946 input[i].data[1]= input[i].data[0] + ysize;
947 input[i].data[2]= input[i].data[1] + csize;
948 input[i].linesize[0]= c->width;
949 input[i].linesize[1]=
950 input[i].linesize[2]= c->width/2;
952 if(pre_input_ptr && (!i || s->input_picture[i-1])) {
953 pre_input= *pre_input_ptr;
955 if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
956 pre_input.data[0]+=INPLACE_OFFSET;
957 pre_input.data[1]+=INPLACE_OFFSET;
958 pre_input.data[2]+=INPLACE_OFFSET;
961 s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
962 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);
963 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);
967 for(j=0; j<s->max_b_frames+1; j++){
970 if(!s->input_picture[j])
973 c->error[0]= c->error[1]= c->error[2]= 0;
975 input[0].pict_type= I_TYPE;
976 input[0].quality= 1 * FF_QP2LAMBDA;
977 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
978 // rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
980 for(i=0; i<s->max_b_frames+1; i++){
981 int is_p= i % (j+1) == j || i==s->max_b_frames;
983 input[i+1].pict_type= is_p ? P_TYPE : B_TYPE;
984 input[i+1].quality= is_p ? p_lambda : b_lambda;
985 out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
986 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
989 /* get the delayed frames */
991 out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
992 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
995 rd += c->error[0] + c->error[1] + c->error[2];
1007 for(i=0; i<s->max_b_frames+2; i++){
1008 av_freep(&input[i].data[0]);
1011 return best_b_count;
1014 static void select_input_picture(MpegEncContext *s){
1017 for(i=1; i<MAX_PICTURE_COUNT; i++)
1018 s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1019 s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1021 /* set next picture type & ordering */
1022 if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1023 if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1024 s->reordered_input_picture[0]= s->input_picture[0];
1025 s->reordered_input_picture[0]->pict_type= I_TYPE;
1026 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1030 if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1031 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1032 //FIXME check that te gop check above is +-1 correct
1033 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1035 if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1037 s->input_picture[0]->data[i]= NULL;
1038 s->input_picture[0]->type= 0;
1040 assert( s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1041 || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1043 s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1047 ff_vbv_update(s, 0);
1053 if(s->flags&CODEC_FLAG_PASS2){
1054 for(i=0; i<s->max_b_frames+1; i++){
1055 int pict_num= s->input_picture[0]->display_picture_number + i;
1057 if(pict_num >= s->rc_context.num_entries)
1059 if(!s->input_picture[i]){
1060 s->rc_context.entry[pict_num-1].new_pict_type = P_TYPE;
1064 s->input_picture[i]->pict_type=
1065 s->rc_context.entry[pict_num].new_pict_type;
1069 if(s->avctx->b_frame_strategy==0){
1070 b_frames= s->max_b_frames;
1071 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1072 }else if(s->avctx->b_frame_strategy==1){
1073 for(i=1; i<s->max_b_frames+1; i++){
1074 if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1075 s->input_picture[i]->b_frame_score=
1076 get_intra_count(s, s->input_picture[i ]->data[0],
1077 s->input_picture[i-1]->data[0], s->linesize) + 1;
1080 for(i=0; i<s->max_b_frames+1; i++){
1081 if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1084 b_frames= FFMAX(0, i-1);
1087 for(i=0; i<b_frames+1; i++){
1088 s->input_picture[i]->b_frame_score=0;
1090 }else if(s->avctx->b_frame_strategy==2){
1091 b_frames= estimate_best_b_count(s);
1093 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1098 //static int b_count=0;
1099 //b_count+= b_frames;
1100 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1102 for(i= b_frames - 1; i>=0; i--){
1103 int type= s->input_picture[i]->pict_type;
1104 if(type && type != B_TYPE)
1107 if(s->input_picture[b_frames]->pict_type == B_TYPE && b_frames == s->max_b_frames){
1108 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1111 if(s->picture_in_gop_number + b_frames >= s->gop_size){
1112 if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1113 b_frames= s->gop_size - s->picture_in_gop_number - 1;
1115 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1117 s->input_picture[b_frames]->pict_type= I_TYPE;
1121 if( (s->flags & CODEC_FLAG_CLOSED_GOP)
1123 && s->input_picture[b_frames]->pict_type== I_TYPE)
1126 s->reordered_input_picture[0]= s->input_picture[b_frames];
1127 if(s->reordered_input_picture[0]->pict_type != I_TYPE)
1128 s->reordered_input_picture[0]->pict_type= P_TYPE;
1129 s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1130 for(i=0; i<b_frames; i++){
1131 s->reordered_input_picture[i+1]= s->input_picture[i];
1132 s->reordered_input_picture[i+1]->pict_type= B_TYPE;
1133 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1138 if(s->reordered_input_picture[0]){
1139 s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0;
1141 copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1143 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1144 // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1146 int i= ff_find_unused_picture(s, 0);
1147 Picture *pic= &s->picture[i];
1149 pic->reference = s->reordered_input_picture[0]->reference;
1150 alloc_picture(s, pic, 0);
1152 /* mark us unused / free shared pic */
1153 if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1154 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1156 s->reordered_input_picture[0]->data[i]= NULL;
1157 s->reordered_input_picture[0]->type= 0;
1159 copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1161 s->current_picture_ptr= pic;
1163 // input is not a shared pix -> reuse buffer for current_pix
1165 assert( s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1166 || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1168 s->current_picture_ptr= s->reordered_input_picture[0];
1170 s->new_picture.data[i]+= INPLACE_OFFSET;
1173 copy_picture(&s->current_picture, s->current_picture_ptr);
1175 s->picture_number= s->new_picture.display_picture_number;
1176 //printf("dpn:%d\n", s->picture_number);
1178 memset(&s->new_picture, 0, sizeof(Picture));
1182 int MPV_encode_picture(AVCodecContext *avctx,
1183 unsigned char *buf, int buf_size, void *data)
1185 MpegEncContext *s = avctx->priv_data;
1186 AVFrame *pic_arg = data;
1187 int i, stuffing_count;
1189 for(i=0; i<avctx->thread_count; i++){
1190 int start_y= s->thread_context[i]->start_mb_y;
1191 int end_y= s->thread_context[i]-> end_mb_y;
1192 int h= s->mb_height;
1193 uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1194 uint8_t *end = buf + (size_t)(((int64_t) buf_size)* end_y/h);
1196 init_put_bits(&s->thread_context[i]->pb, start, end - start);
1199 s->picture_in_gop_number++;
1201 if(load_input_picture(s, pic_arg) < 0)
1204 select_input_picture(s);
1207 if(s->new_picture.data[0]){
1208 s->pict_type= s->new_picture.pict_type;
1210 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1211 MPV_frame_start(s, avctx);
1213 if (encode_picture(s, s->picture_number) < 0)
1216 avctx->real_pict_num = s->picture_number;
1217 avctx->header_bits = s->header_bits;
1218 avctx->mv_bits = s->mv_bits;
1219 avctx->misc_bits = s->misc_bits;
1220 avctx->i_tex_bits = s->i_tex_bits;
1221 avctx->p_tex_bits = s->p_tex_bits;
1222 avctx->i_count = s->i_count;
1223 avctx->p_count = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1224 avctx->skip_count = s->skip_count;
1228 if (ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1229 ff_mjpeg_encode_picture_trailer(s);
1231 if(avctx->rc_buffer_size){
1232 RateControlContext *rcc= &s->rc_context;
1233 int max_size= rcc->buffer_index/3;
1235 if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1236 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1237 if(s->adaptive_quant){
1239 for(i=0; i<s->mb_height*s->mb_stride; i++)
1240 s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1242 s->mb_skipped = 0; //done in MPV_frame_start()
1243 if(s->pict_type==P_TYPE){ //done in encode_picture() so we must undo it
1244 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1245 s->no_rounding ^= 1;
1247 if(s->pict_type!=B_TYPE){
1248 s->time_base= s->last_time_base;
1249 s->last_non_b_time= s->time - s->pp_time;
1251 // av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1252 for(i=0; i<avctx->thread_count; i++){
1253 PutBitContext *pb= &s->thread_context[i]->pb;
1254 init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1259 assert(s->avctx->rc_max_rate);
1262 if(s->flags&CODEC_FLAG_PASS1)
1263 ff_write_pass1_stats(s);
1266 s->current_picture_ptr->error[i]= s->current_picture.error[i];
1267 avctx->error[i] += s->current_picture_ptr->error[i];
1270 if(s->flags&CODEC_FLAG_PASS1)
1271 assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1272 flush_put_bits(&s->pb);
1273 s->frame_bits = put_bits_count(&s->pb);
1275 stuffing_count= ff_vbv_update(s, s->frame_bits);
1277 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1278 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1282 switch(s->codec_id){
1283 case CODEC_ID_MPEG1VIDEO:
1284 case CODEC_ID_MPEG2VIDEO:
1285 while(stuffing_count--){
1286 put_bits(&s->pb, 8, 0);
1289 case CODEC_ID_MPEG4:
1290 put_bits(&s->pb, 16, 0);
1291 put_bits(&s->pb, 16, 0x1C3);
1292 stuffing_count -= 4;
1293 while(stuffing_count--){
1294 put_bits(&s->pb, 8, 0xFF);
1298 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1300 flush_put_bits(&s->pb);
1301 s->frame_bits = put_bits_count(&s->pb);
1304 /* update mpeg1/2 vbv_delay for CBR */
1305 if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1306 && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1309 assert(s->repeat_first_field==0);
1311 vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
1312 assert(vbv_delay < 0xFFFF);
1314 s->vbv_delay_ptr[0] &= 0xF8;
1315 s->vbv_delay_ptr[0] |= vbv_delay>>13;
1316 s->vbv_delay_ptr[1] = vbv_delay>>5;
1317 s->vbv_delay_ptr[2] &= 0x07;
1318 s->vbv_delay_ptr[2] |= vbv_delay<<3;
1320 s->total_bits += s->frame_bits;
1321 avctx->frame_bits = s->frame_bits;
1323 assert((pbBufPtr(&s->pb) == s->pb.buf));
1326 assert((s->frame_bits&7)==0);
1328 return s->frame_bits/8;
1331 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1333 static const char tab[64]=
1345 DCTELEM *block= s->block[n];
1346 const int last_index= s->block_last_index[n];
1351 threshold= -threshold;
1355 /* are all which we could set to zero are allready zero? */
1356 if(last_index<=skip_dc - 1) return;
1358 for(i=0; i<=last_index; i++){
1359 const int j = s->intra_scantable.permutated[i];
1360 const int level = FFABS(block[j]);
1362 if(skip_dc && i==0) continue;
1371 if(score >= threshold) return;
1372 for(i=skip_dc; i<=last_index; i++){
1373 const int j = s->intra_scantable.permutated[i];
1376 if(block[0]) s->block_last_index[n]= 0;
1377 else s->block_last_index[n]= -1;
1380 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1383 const int maxlevel= s->max_qcoeff;
1384 const int minlevel= s->min_qcoeff;
1388 i=1; //skip clipping of intra dc
1392 for(;i<=last_index; i++){
1393 const int j= s->intra_scantable.permutated[i];
1394 int level = block[j];
1396 if (level>maxlevel){
1399 }else if(level<minlevel){
1407 if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1408 av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1411 static void get_vissual_weight(int16_t *weight, uint8_t *ptr, int stride){
1421 for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1422 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1423 int v= ptr[x2 + y2*stride];
1429 weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1434 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1436 int16_t weight[8][64];
1437 DCTELEM orig[8][64];
1438 const int mb_x= s->mb_x;
1439 const int mb_y= s->mb_y;
1442 int dct_offset = s->linesize*8; //default for progressive frames
1443 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1446 for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1448 if(s->adaptive_quant){
1449 const int last_qp= s->qscale;
1450 const int mb_xy= mb_x + mb_y*s->mb_stride;
1452 s->lambda= s->lambda_table[mb_xy];
1455 if(!(s->flags&CODEC_FLAG_QP_RD)){
1456 s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1457 s->dquant= s->qscale - last_qp;
1459 if(s->out_format==FMT_H263){
1460 s->dquant= av_clip(s->dquant, -2, 2);
1462 if(s->codec_id==CODEC_ID_MPEG4){
1464 if(s->pict_type == B_TYPE){
1465 if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1468 if(s->mv_type==MV_TYPE_8X8)
1474 ff_set_qscale(s, last_qp + s->dquant);
1475 }else if(s->flags&CODEC_FLAG_QP_RD)
1476 ff_set_qscale(s, s->qscale + s->dquant);
1478 wrap_y = s->linesize;
1479 wrap_c = s->uvlinesize;
1480 ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1481 ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1482 ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1484 if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1485 uint8_t *ebuf= s->edge_emu_buffer + 32;
1486 ff_emulated_edge_mc(ebuf , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width , s->height);
1488 ff_emulated_edge_mc(ebuf+18*wrap_y , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1489 ptr_cb= ebuf+18*wrap_y;
1490 ff_emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1491 ptr_cr= ebuf+18*wrap_y+8;
1495 if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1496 int progressive_score, interlaced_score;
1498 s->interlaced_dct=0;
1499 progressive_score= s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y, 8)
1500 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1502 if(progressive_score > 0){
1503 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y , NULL, wrap_y*2, 8)
1504 +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y , NULL, wrap_y*2, 8);
1505 if(progressive_score > interlaced_score){
1506 s->interlaced_dct=1;
1510 if (s->chroma_format == CHROMA_422)
1516 s->dsp.get_pixels(s->block[0], ptr_y , wrap_y);
1517 s->dsp.get_pixels(s->block[1], ptr_y + 8, wrap_y);
1518 s->dsp.get_pixels(s->block[2], ptr_y + dct_offset , wrap_y);
1519 s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1521 if(s->flags&CODEC_FLAG_GRAY){
1525 s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1526 s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1527 if(!s->chroma_y_shift){ /* 422 */
1528 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1529 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1533 op_pixels_func (*op_pix)[4];
1534 qpel_mc_func (*op_qpix)[16];
1535 uint8_t *dest_y, *dest_cb, *dest_cr;
1537 dest_y = s->dest[0];
1538 dest_cb = s->dest[1];
1539 dest_cr = s->dest[2];
1541 if ((!s->no_rounding) || s->pict_type==B_TYPE){
1542 op_pix = s->dsp.put_pixels_tab;
1543 op_qpix= s->dsp.put_qpel_pixels_tab;
1545 op_pix = s->dsp.put_no_rnd_pixels_tab;
1546 op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1549 if (s->mv_dir & MV_DIR_FORWARD) {
1550 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1551 op_pix = s->dsp.avg_pixels_tab;
1552 op_qpix= s->dsp.avg_qpel_pixels_tab;
1554 if (s->mv_dir & MV_DIR_BACKWARD) {
1555 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1558 if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1559 int progressive_score, interlaced_score;
1561 s->interlaced_dct=0;
1562 progressive_score= s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y, 8)
1563 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1565 if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1567 if(progressive_score>0){
1568 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y , ptr_y , wrap_y*2, 8)
1569 +s->dsp.ildct_cmp[0](s, dest_y + wrap_y , ptr_y + wrap_y , wrap_y*2, 8);
1571 if(progressive_score > interlaced_score){
1572 s->interlaced_dct=1;
1576 if (s->chroma_format == CHROMA_422)
1582 s->dsp.diff_pixels(s->block[0], ptr_y , dest_y , wrap_y);
1583 s->dsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
1584 s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset , dest_y + dct_offset , wrap_y);
1585 s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1587 if(s->flags&CODEC_FLAG_GRAY){
1591 s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1592 s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1593 if(!s->chroma_y_shift){ /* 422 */
1594 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1595 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1598 /* pre quantization */
1599 if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1601 if(s->dsp.sad[1](NULL, ptr_y , dest_y , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1602 if(s->dsp.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1603 if(s->dsp.sad[1](NULL, ptr_y +dct_offset , dest_y +dct_offset , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1604 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;
1605 if(s->dsp.sad[1](NULL, ptr_cb , dest_cb , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1606 if(s->dsp.sad[1](NULL, ptr_cr , dest_cr , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1607 if(!s->chroma_y_shift){ /* 422 */
1608 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;
1609 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;
1614 if(s->avctx->quantizer_noise_shaping){
1615 if(!skip_dct[0]) get_vissual_weight(weight[0], ptr_y , wrap_y);
1616 if(!skip_dct[1]) get_vissual_weight(weight[1], ptr_y + 8, wrap_y);
1617 if(!skip_dct[2]) get_vissual_weight(weight[2], ptr_y + dct_offset , wrap_y);
1618 if(!skip_dct[3]) get_vissual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1619 if(!skip_dct[4]) get_vissual_weight(weight[4], ptr_cb , wrap_c);
1620 if(!skip_dct[5]) get_vissual_weight(weight[5], ptr_cr , wrap_c);
1621 if(!s->chroma_y_shift){ /* 422 */
1622 if(!skip_dct[6]) get_vissual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1623 if(!skip_dct[7]) get_vissual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1625 memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1628 /* DCT & quantize */
1629 assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1631 for(i=0;i<mb_block_count;i++) {
1634 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1635 // FIXME we could decide to change to quantizer instead of clipping
1636 // JS: I don't think that would be a good idea it could lower quality instead
1637 // of improve it. Just INTRADC clipping deserves changes in quantizer
1638 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1640 s->block_last_index[i]= -1;
1642 if(s->avctx->quantizer_noise_shaping){
1643 for(i=0;i<mb_block_count;i++) {
1645 s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1650 if(s->luma_elim_threshold && !s->mb_intra)
1652 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1653 if(s->chroma_elim_threshold && !s->mb_intra)
1654 for(i=4; i<mb_block_count; i++)
1655 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1657 if(s->flags & CODEC_FLAG_CBP_RD){
1658 for(i=0;i<mb_block_count;i++) {
1659 if(s->block_last_index[i] == -1)
1660 s->coded_score[i]= INT_MAX/256;
1665 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1666 s->block_last_index[4]=
1667 s->block_last_index[5]= 0;
1669 s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1672 //non c quantize code returns incorrect block_last_index FIXME
1673 if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1674 for(i=0; i<mb_block_count; i++){
1676 if(s->block_last_index[i]>0){
1677 for(j=63; j>0; j--){
1678 if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1680 s->block_last_index[i]= j;
1685 /* huffman encode */
1686 switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1687 case CODEC_ID_MPEG1VIDEO:
1688 case CODEC_ID_MPEG2VIDEO:
1689 if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
1690 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1692 case CODEC_ID_MPEG4:
1693 mpeg4_encode_mb(s, s->block, motion_x, motion_y); break;
1694 case CODEC_ID_MSMPEG4V2:
1695 case CODEC_ID_MSMPEG4V3:
1697 if (ENABLE_MSMPEG4_ENCODER)
1698 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1701 if (ENABLE_WMV2_ENCODER)
1702 ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1705 if (ENABLE_H261_ENCODER)
1706 ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1709 case CODEC_ID_H263P:
1713 h263_encode_mb(s, s->block, motion_x, motion_y); break;
1714 case CODEC_ID_MJPEG:
1715 if (ENABLE_MJPEG_ENCODER)
1716 ff_mjpeg_encode_mb(s, s->block);
1723 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1725 if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 6);
1726 else encode_mb_internal(s, motion_x, motion_y, 16, 8);
1729 void ff_copy_bits(PutBitContext *pb, uint8_t *src, int length)
1731 const uint16_t *srcw= (uint16_t*)src;
1732 int words= length>>4;
1733 int bits= length&15;
1736 if(length==0) return;
1739 for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i]));
1740 }else if(put_bits_count(pb)&7){
1741 for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i]));
1743 for(i=0; put_bits_count(pb)&31; i++)
1744 put_bits(pb, 8, src[i]);
1746 memcpy(pbBufPtr(pb), src+i, 2*words-i);
1747 skip_put_bytes(pb, 2*words-i);
1750 put_bits(pb, bits, be2me_16(srcw[words])>>(16-bits));
1753 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1756 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1759 d->mb_skip_run= s->mb_skip_run;
1761 d->last_dc[i]= s->last_dc[i];
1764 d->mv_bits= s->mv_bits;
1765 d->i_tex_bits= s->i_tex_bits;
1766 d->p_tex_bits= s->p_tex_bits;
1767 d->i_count= s->i_count;
1768 d->f_count= s->f_count;
1769 d->b_count= s->b_count;
1770 d->skip_count= s->skip_count;
1771 d->misc_bits= s->misc_bits;
1775 d->qscale= s->qscale;
1776 d->dquant= s->dquant;
1779 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1782 memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1783 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1786 d->mb_skip_run= s->mb_skip_run;
1788 d->last_dc[i]= s->last_dc[i];
1791 d->mv_bits= s->mv_bits;
1792 d->i_tex_bits= s->i_tex_bits;
1793 d->p_tex_bits= s->p_tex_bits;
1794 d->i_count= s->i_count;
1795 d->f_count= s->f_count;
1796 d->b_count= s->b_count;
1797 d->skip_count= s->skip_count;
1798 d->misc_bits= s->misc_bits;
1800 d->mb_intra= s->mb_intra;
1801 d->mb_skipped= s->mb_skipped;
1802 d->mv_type= s->mv_type;
1803 d->mv_dir= s->mv_dir;
1805 if(s->data_partitioning){
1807 d->tex_pb= s->tex_pb;
1811 d->block_last_index[i]= s->block_last_index[i];
1812 d->interlaced_dct= s->interlaced_dct;
1813 d->qscale= s->qscale;
1816 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1817 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1818 int *dmin, int *next_block, int motion_x, int motion_y)
1821 uint8_t *dest_backup[3];
1823 copy_context_before_encode(s, backup, type);
1825 s->block= s->blocks[*next_block];
1826 s->pb= pb[*next_block];
1827 if(s->data_partitioning){
1828 s->pb2 = pb2 [*next_block];
1829 s->tex_pb= tex_pb[*next_block];
1833 memcpy(dest_backup, s->dest, sizeof(s->dest));
1834 s->dest[0] = s->rd_scratchpad;
1835 s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1836 s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1837 assert(s->linesize >= 32); //FIXME
1840 encode_mb(s, motion_x, motion_y);
1842 score= put_bits_count(&s->pb);
1843 if(s->data_partitioning){
1844 score+= put_bits_count(&s->pb2);
1845 score+= put_bits_count(&s->tex_pb);
1848 if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1849 MPV_decode_mb(s, s->block);
1851 score *= s->lambda2;
1852 score += sse_mb(s) << FF_LAMBDA_SHIFT;
1856 memcpy(s->dest, dest_backup, sizeof(s->dest));
1863 copy_context_after_encode(best, s, type);
1867 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1868 uint32_t *sq = ff_squareTbl + 256;
1873 return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1874 else if(w==8 && h==8)
1875 return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1879 acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1888 static int sse_mb(MpegEncContext *s){
1892 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1893 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1896 if(s->avctx->mb_cmp == FF_CMP_NSSE){
1897 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)
1898 +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)
1899 +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);
1901 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)
1902 +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)
1903 +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);
1906 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)
1907 +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)
1908 +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);
1911 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1912 MpegEncContext *s= arg;
1916 s->me.dia_size= s->avctx->pre_dia_size;
1917 s->first_slice_line=1;
1918 for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1919 for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1920 ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1922 s->first_slice_line=0;
1930 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1931 MpegEncContext *s= arg;
1933 ff_check_alignment();
1935 s->me.dia_size= s->avctx->dia_size;
1936 s->first_slice_line=1;
1937 for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1938 s->mb_x=0; //for block init below
1939 ff_init_block_index(s);
1940 for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1941 s->block_index[0]+=2;
1942 s->block_index[1]+=2;
1943 s->block_index[2]+=2;
1944 s->block_index[3]+=2;
1946 /* compute motion vector & mb_type and store in context */
1947 if(s->pict_type==B_TYPE)
1948 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1950 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1952 s->first_slice_line=0;
1957 static int mb_var_thread(AVCodecContext *c, void *arg){
1958 MpegEncContext *s= arg;
1961 ff_check_alignment();
1963 for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1964 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1967 uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1969 int sum = s->dsp.pix_sum(pix, s->linesize);
1971 varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1973 s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1974 s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1975 s->me.mb_var_sum_temp += varc;
1981 static void write_slice_end(MpegEncContext *s){
1982 if(s->codec_id==CODEC_ID_MPEG4){
1983 if(s->partitioned_frame){
1984 ff_mpeg4_merge_partitions(s);
1987 ff_mpeg4_stuffing(&s->pb);
1988 }else if(ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1989 ff_mjpeg_encode_stuffing(&s->pb);
1992 align_put_bits(&s->pb);
1993 flush_put_bits(&s->pb);
1995 if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1996 s->misc_bits+= get_bits_diff(s);
1999 static int encode_thread(AVCodecContext *c, void *arg){
2000 MpegEncContext *s= arg;
2001 int mb_x, mb_y, pdif = 0;
2003 MpegEncContext best_s, backup_s;
2004 uint8_t bit_buf[2][MAX_MB_BYTES];
2005 uint8_t bit_buf2[2][MAX_MB_BYTES];
2006 uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2007 PutBitContext pb[2], pb2[2], tex_pb[2];
2008 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2010 ff_check_alignment();
2013 init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
2014 init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
2015 init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2018 s->last_bits= put_bits_count(&s->pb);
2029 /* init last dc values */
2030 /* note: quant matrix value (8) is implied here */
2031 s->last_dc[i] = 128 << s->intra_dc_precision;
2033 s->current_picture.error[i] = 0;
2036 memset(s->last_mv, 0, sizeof(s->last_mv));
2040 switch(s->codec_id){
2042 case CODEC_ID_H263P:
2044 s->gob_index = ff_h263_get_gob_height(s);
2046 case CODEC_ID_MPEG4:
2047 if(s->partitioned_frame)
2048 ff_mpeg4_init_partitions(s);
2054 s->first_slice_line = 1;
2055 s->ptr_lastgob = s->pb.buf;
2056 for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2057 // printf("row %d at %X\n", s->mb_y, (int)s);
2061 ff_set_qscale(s, s->qscale);
2062 ff_init_block_index(s);
2064 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2065 int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2066 int mb_type= s->mb_type[xy];
2071 if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2072 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2075 if(s->data_partitioning){
2076 if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES
2077 || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2078 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2084 s->mb_y = mb_y; // moved into loop, can get changed by H.261
2085 ff_update_block_index(s);
2087 if(ENABLE_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2088 ff_h261_reorder_mb_index(s);
2089 xy= s->mb_y*s->mb_stride + s->mb_x;
2090 mb_type= s->mb_type[xy];
2093 /* write gob / video packet header */
2095 int current_packet_size, is_gob_start;
2097 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2099 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2101 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2103 switch(s->codec_id){
2105 case CODEC_ID_H263P:
2106 if(!s->h263_slice_structured)
2107 if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2109 case CODEC_ID_MPEG2VIDEO:
2110 if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2111 case CODEC_ID_MPEG1VIDEO:
2112 if(s->mb_skip_run) is_gob_start=0;
2117 if(s->start_mb_y != mb_y || mb_x!=0){
2120 if(s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2121 ff_mpeg4_init_partitions(s);
2125 assert((put_bits_count(&s->pb)&7) == 0);
2126 current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
2128 if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2129 int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2130 int d= 100 / s->avctx->error_rate;
2132 current_packet_size=0;
2133 #ifndef ALT_BITSTREAM_WRITER
2134 s->pb.buf_ptr= s->ptr_lastgob;
2136 assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
2140 if (s->avctx->rtp_callback){
2141 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2142 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2145 switch(s->codec_id){
2146 case CODEC_ID_MPEG4:
2147 ff_mpeg4_encode_video_packet_header(s);
2148 ff_mpeg4_clean_buffers(s);
2150 case CODEC_ID_MPEG1VIDEO:
2151 case CODEC_ID_MPEG2VIDEO:
2152 if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER) {
2153 ff_mpeg1_encode_slice_header(s);
2154 ff_mpeg1_clean_buffers(s);
2158 case CODEC_ID_H263P:
2159 h263_encode_gob_header(s, mb_y);
2163 if(s->flags&CODEC_FLAG_PASS1){
2164 int bits= put_bits_count(&s->pb);
2165 s->misc_bits+= bits - s->last_bits;
2169 s->ptr_lastgob += current_packet_size;
2170 s->first_slice_line=1;
2171 s->resync_mb_x=mb_x;
2172 s->resync_mb_y=mb_y;
2176 if( (s->resync_mb_x == s->mb_x)
2177 && s->resync_mb_y+1 == s->mb_y){
2178 s->first_slice_line=0;
2182 s->dquant=0; //only for QP_RD
2184 if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2186 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2188 copy_context_before_encode(&backup_s, s, -1);
2190 best_s.data_partitioning= s->data_partitioning;
2191 best_s.partitioned_frame= s->partitioned_frame;
2192 if(s->data_partitioning){
2193 backup_s.pb2= s->pb2;
2194 backup_s.tex_pb= s->tex_pb;
2197 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2198 s->mv_dir = MV_DIR_FORWARD;
2199 s->mv_type = MV_TYPE_16X16;
2201 s->mv[0][0][0] = s->p_mv_table[xy][0];
2202 s->mv[0][0][1] = s->p_mv_table[xy][1];
2203 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2204 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2206 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2207 s->mv_dir = MV_DIR_FORWARD;
2208 s->mv_type = MV_TYPE_FIELD;
2211 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2212 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2213 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2215 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2216 &dmin, &next_block, 0, 0);
2218 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2219 s->mv_dir = MV_DIR_FORWARD;
2220 s->mv_type = MV_TYPE_16X16;
2224 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2225 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2227 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2228 s->mv_dir = MV_DIR_FORWARD;
2229 s->mv_type = MV_TYPE_8X8;
2232 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2233 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2235 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2236 &dmin, &next_block, 0, 0);
2238 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2239 s->mv_dir = MV_DIR_FORWARD;
2240 s->mv_type = MV_TYPE_16X16;
2242 s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2243 s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2244 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2245 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2247 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2248 s->mv_dir = MV_DIR_BACKWARD;
2249 s->mv_type = MV_TYPE_16X16;
2251 s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2252 s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2253 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2254 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2256 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2257 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2258 s->mv_type = MV_TYPE_16X16;
2260 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2261 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2262 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2263 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2264 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2265 &dmin, &next_block, 0, 0);
2267 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2268 s->mv_dir = MV_DIR_FORWARD;
2269 s->mv_type = MV_TYPE_FIELD;
2272 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2273 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2274 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2276 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2277 &dmin, &next_block, 0, 0);
2279 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2280 s->mv_dir = MV_DIR_BACKWARD;
2281 s->mv_type = MV_TYPE_FIELD;
2284 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2285 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2286 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2288 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2289 &dmin, &next_block, 0, 0);
2291 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2292 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2293 s->mv_type = MV_TYPE_FIELD;
2295 for(dir=0; dir<2; dir++){
2297 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2298 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2299 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2302 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2303 &dmin, &next_block, 0, 0);
2305 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2307 s->mv_type = MV_TYPE_16X16;
2311 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2312 &dmin, &next_block, 0, 0);
2313 if(s->h263_pred || s->h263_aic){
2315 s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2317 ff_clean_intra_table_entries(s); //old mode?
2321 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2322 if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2323 const int last_qp= backup_s.qscale;
2326 const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2327 static const int dquant_tab[4]={-1,1,-2,2};
2329 assert(backup_s.dquant == 0);
2332 s->mv_dir= best_s.mv_dir;
2333 s->mv_type = MV_TYPE_16X16;
2334 s->mb_intra= best_s.mb_intra;
2335 s->mv[0][0][0] = best_s.mv[0][0][0];
2336 s->mv[0][0][1] = best_s.mv[0][0][1];
2337 s->mv[1][0][0] = best_s.mv[1][0][0];
2338 s->mv[1][0][1] = best_s.mv[1][0][1];
2340 qpi = s->pict_type == B_TYPE ? 2 : 0;
2341 for(; qpi<4; qpi++){
2342 int dquant= dquant_tab[qpi];
2343 qp= last_qp + dquant;
2344 if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2346 backup_s.dquant= dquant;
2347 if(s->mb_intra && s->dc_val[0]){
2349 dc[i]= s->dc_val[0][ s->block_index[i] ];
2350 memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2354 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2355 &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2356 if(best_s.qscale != qp){
2357 if(s->mb_intra && s->dc_val[0]){
2359 s->dc_val[0][ s->block_index[i] ]= dc[i];
2360 memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2367 if(mb_type&CANDIDATE_MB_TYPE_DIRECT){
2368 int mx= s->b_direct_mv_table[xy][0];
2369 int my= s->b_direct_mv_table[xy][1];
2371 backup_s.dquant = 0;
2372 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2374 ff_mpeg4_set_direct_mv(s, mx, my);
2375 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2376 &dmin, &next_block, mx, my);
2378 if(mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2379 backup_s.dquant = 0;
2380 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2382 ff_mpeg4_set_direct_mv(s, 0, 0);
2383 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2384 &dmin, &next_block, 0, 0);
2386 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2389 coded |= s->block_last_index[i];
2392 memcpy(s->mv, best_s.mv, sizeof(s->mv));
2393 if(best_s.mv_dir & MV_DIRECT){
2394 mx=my=0; //FIXME find the one we actually used
2395 ff_mpeg4_set_direct_mv(s, mx, my);
2396 }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2404 s->mv_dir= best_s.mv_dir;
2405 s->mv_type = best_s.mv_type;
2407 /* s->mv[0][0][0] = best_s.mv[0][0][0];
2408 s->mv[0][0][1] = best_s.mv[0][0][1];
2409 s->mv[1][0][0] = best_s.mv[1][0][0];
2410 s->mv[1][0][1] = best_s.mv[1][0][1];*/
2413 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2414 &dmin, &next_block, mx, my);
2419 s->current_picture.qscale_table[xy]= best_s.qscale;
2421 copy_context_after_encode(s, &best_s, -1);
2423 pb_bits_count= put_bits_count(&s->pb);
2424 flush_put_bits(&s->pb);
2425 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2428 if(s->data_partitioning){
2429 pb2_bits_count= put_bits_count(&s->pb2);
2430 flush_put_bits(&s->pb2);
2431 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2432 s->pb2= backup_s.pb2;
2434 tex_pb_bits_count= put_bits_count(&s->tex_pb);
2435 flush_put_bits(&s->tex_pb);
2436 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2437 s->tex_pb= backup_s.tex_pb;
2439 s->last_bits= put_bits_count(&s->pb);
2441 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
2442 ff_h263_update_motion_val(s);
2444 if(next_block==0){ //FIXME 16 vs linesize16
2445 s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad , s->linesize ,16);
2446 s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
2447 s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2450 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2451 MPV_decode_mb(s, s->block);
2453 int motion_x = 0, motion_y = 0;
2454 s->mv_type=MV_TYPE_16X16;
2455 // only one MB-Type possible
2458 case CANDIDATE_MB_TYPE_INTRA:
2461 motion_x= s->mv[0][0][0] = 0;
2462 motion_y= s->mv[0][0][1] = 0;
2464 case CANDIDATE_MB_TYPE_INTER:
2465 s->mv_dir = MV_DIR_FORWARD;
2467 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2468 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2470 case CANDIDATE_MB_TYPE_INTER_I:
2471 s->mv_dir = MV_DIR_FORWARD;
2472 s->mv_type = MV_TYPE_FIELD;
2475 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2476 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2477 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2480 case CANDIDATE_MB_TYPE_INTER4V:
2481 s->mv_dir = MV_DIR_FORWARD;
2482 s->mv_type = MV_TYPE_8X8;
2485 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2486 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2489 case CANDIDATE_MB_TYPE_DIRECT:
2490 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2492 motion_x=s->b_direct_mv_table[xy][0];
2493 motion_y=s->b_direct_mv_table[xy][1];
2494 ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2496 case CANDIDATE_MB_TYPE_DIRECT0:
2497 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2499 ff_mpeg4_set_direct_mv(s, 0, 0);
2501 case CANDIDATE_MB_TYPE_BIDIR:
2502 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2504 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2505 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2506 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2507 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2509 case CANDIDATE_MB_TYPE_BACKWARD:
2510 s->mv_dir = MV_DIR_BACKWARD;
2512 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2513 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2515 case CANDIDATE_MB_TYPE_FORWARD:
2516 s->mv_dir = MV_DIR_FORWARD;
2518 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2519 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2520 // printf(" %d %d ", motion_x, motion_y);
2522 case CANDIDATE_MB_TYPE_FORWARD_I:
2523 s->mv_dir = MV_DIR_FORWARD;
2524 s->mv_type = MV_TYPE_FIELD;
2527 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2528 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2529 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2532 case CANDIDATE_MB_TYPE_BACKWARD_I:
2533 s->mv_dir = MV_DIR_BACKWARD;
2534 s->mv_type = MV_TYPE_FIELD;
2537 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2538 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2539 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2542 case CANDIDATE_MB_TYPE_BIDIR_I:
2543 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2544 s->mv_type = MV_TYPE_FIELD;
2546 for(dir=0; dir<2; dir++){
2548 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2549 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2550 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2555 av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2558 encode_mb(s, motion_x, motion_y);
2560 // RAL: Update last macroblock type
2561 s->last_mv_dir = s->mv_dir;
2563 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
2564 ff_h263_update_motion_val(s);
2566 MPV_decode_mb(s, s->block);
2569 /* clean the MV table in IPS frames for direct mode in B frames */
2570 if(s->mb_intra /* && I,P,S_TYPE */){
2571 s->p_mv_table[xy][0]=0;
2572 s->p_mv_table[xy][1]=0;
2575 if(s->flags&CODEC_FLAG_PSNR){
2579 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2580 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2582 s->current_picture.error[0] += sse(
2583 s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2584 s->dest[0], w, h, s->linesize);
2585 s->current_picture.error[1] += sse(
2586 s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,
2587 s->dest[1], w>>1, h>>1, s->uvlinesize);
2588 s->current_picture.error[2] += sse(
2589 s, s->new_picture .data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,
2590 s->dest[2], w>>1, h>>1, s->uvlinesize);
2593 if(s->out_format == FMT_H263)
2594 ff_h263_loop_filter(s);
2596 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2600 //not beautiful here but we must write it before flushing so it has to be here
2601 if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
2602 msmpeg4_encode_ext_header(s);
2606 /* Send the last GOB if RTP */
2607 if (s->avctx->rtp_callback) {
2608 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2609 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2610 /* Call the RTP callback to send the last GOB */
2612 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2618 #define MERGE(field) dst->field += src->field; src->field=0
2619 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2620 MERGE(me.scene_change_score);
2621 MERGE(me.mc_mb_var_sum_temp);
2622 MERGE(me.mb_var_sum_temp);
2625 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2628 MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2629 MERGE(dct_count[1]);
2639 MERGE(padding_bug_score);
2640 MERGE(current_picture.error[0]);
2641 MERGE(current_picture.error[1]);
2642 MERGE(current_picture.error[2]);
2644 if(dst->avctx->noise_reduction){
2645 for(i=0; i<64; i++){
2646 MERGE(dct_error_sum[0][i]);
2647 MERGE(dct_error_sum[1][i]);
2651 assert(put_bits_count(&src->pb) % 8 ==0);
2652 assert(put_bits_count(&dst->pb) % 8 ==0);
2653 ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2654 flush_put_bits(&dst->pb);
2657 static int estimate_qp(MpegEncContext *s, int dry_run){
2658 if (s->next_lambda){
2659 s->current_picture_ptr->quality=
2660 s->current_picture.quality = s->next_lambda;
2661 if(!dry_run) s->next_lambda= 0;
2662 } else if (!s->fixed_qscale) {
2663 s->current_picture_ptr->quality=
2664 s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2665 if (s->current_picture.quality < 0)
2669 if(s->adaptive_quant){
2670 switch(s->codec_id){
2671 case CODEC_ID_MPEG4:
2672 ff_clean_mpeg4_qscales(s);
2675 case CODEC_ID_H263P:
2677 ff_clean_h263_qscales(s);
2681 s->lambda= s->lambda_table[0];
2684 s->lambda= s->current_picture.quality;
2685 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2690 static int encode_picture(MpegEncContext *s, int picture_number)
2695 s->picture_number = picture_number;
2697 /* Reset the average MB variance */
2698 s->me.mb_var_sum_temp =
2699 s->me.mc_mb_var_sum_temp = 0;
2701 /* we need to initialize some time vars before we can encode b-frames */
2702 // RAL: Condition added for MPEG1VIDEO
2703 if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2704 ff_set_mpeg4_time(s, s->picture_number); //FIXME rename and use has_b_frames or similar
2706 s->me.scene_change_score=0;
2708 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration
2710 if(s->pict_type==I_TYPE){
2711 if(s->msmpeg4_version >= 3) s->no_rounding=1;
2712 else s->no_rounding=0;
2713 }else if(s->pict_type!=B_TYPE){
2714 if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2715 s->no_rounding ^= 1;
2718 if(s->flags & CODEC_FLAG_PASS2){
2719 if (estimate_qp(s,1) < 0)
2721 ff_get_2pass_fcode(s);
2722 }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2723 if(s->pict_type==B_TYPE)
2724 s->lambda= s->last_lambda_for[s->pict_type];
2726 s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2730 s->mb_intra=0; //for the rate distortion & bit compare functions
2731 for(i=1; i<s->avctx->thread_count; i++){
2732 ff_update_duplicate_context(s->thread_context[i], s);
2737 /* Estimate motion for every MB */
2738 if(s->pict_type != I_TYPE){
2739 s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2740 s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2741 if(s->pict_type != B_TYPE && s->avctx->me_threshold==0){
2742 if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){
2743 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2747 s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2748 }else /* if(s->pict_type == I_TYPE) */{
2750 for(i=0; i<s->mb_stride*s->mb_height; i++)
2751 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2753 if(!s->fixed_qscale){
2754 /* finding spatial complexity for I-frame rate control */
2755 s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2758 for(i=1; i<s->avctx->thread_count; i++){
2759 merge_context_after_me(s, s->thread_context[i]);
2761 s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2762 s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
2765 if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
2766 s->pict_type= I_TYPE;
2767 for(i=0; i<s->mb_stride*s->mb_height; i++)
2768 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2769 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2773 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) {
2774 s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2776 if(s->flags & CODEC_FLAG_INTERLACED_ME){
2778 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2779 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2780 s->f_code= FFMAX(s->f_code, FFMAX(a,b));
2783 ff_fix_long_p_mvs(s);
2784 ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2785 if(s->flags & CODEC_FLAG_INTERLACED_ME){
2789 ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2790 s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2795 if(s->pict_type==B_TYPE){
2798 a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2799 b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2800 s->f_code = FFMAX(a, b);
2802 a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2803 b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2804 s->b_code = FFMAX(a, b);
2806 ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2807 ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2808 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2809 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2810 if(s->flags & CODEC_FLAG_INTERLACED_ME){
2812 for(dir=0; dir<2; dir++){
2815 int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2816 : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2817 ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2818 s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2826 if (estimate_qp(s, 0) < 0)
2829 if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2830 s->qscale= 3; //reduce clipping problems
2832 if (s->out_format == FMT_MJPEG) {
2833 /* for mjpeg, we do include qscale in the matrix */
2834 s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2836 int j= s->dsp.idct_permutation[i];
2838 s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2840 convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2841 s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2845 //FIXME var duplication
2846 s->current_picture_ptr->key_frame=
2847 s->current_picture.key_frame= s->pict_type == I_TYPE; //FIXME pic_ptr
2848 s->current_picture_ptr->pict_type=
2849 s->current_picture.pict_type= s->pict_type;
2851 if(s->current_picture.key_frame)
2852 s->picture_in_gop_number=0;
2854 s->last_bits= put_bits_count(&s->pb);
2855 switch(s->out_format) {
2857 if (ENABLE_MJPEG_ENCODER)
2858 ff_mjpeg_encode_picture_header(s);
2861 if (ENABLE_H261_ENCODER)
2862 ff_h261_encode_picture_header(s, picture_number);
2865 if (ENABLE_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2866 ff_wmv2_encode_picture_header(s, picture_number);
2867 else if (ENABLE_MSMPEG4_ENCODER && s->h263_msmpeg4)
2868 msmpeg4_encode_picture_header(s, picture_number);
2869 else if (s->h263_pred)
2870 mpeg4_encode_picture_header(s, picture_number);
2871 else if (ENABLE_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2872 rv10_encode_picture_header(s, picture_number);
2873 else if (ENABLE_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2874 rv20_encode_picture_header(s, picture_number);
2875 else if (s->codec_id == CODEC_ID_FLV1)
2876 ff_flv_encode_picture_header(s, picture_number);
2878 h263_encode_picture_header(s, picture_number);
2881 if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
2882 mpeg1_encode_picture_header(s, picture_number);
2889 bits= put_bits_count(&s->pb);
2890 s->header_bits= bits - s->last_bits;
2892 for(i=1; i<s->avctx->thread_count; i++){
2893 update_duplicate_context_after_me(s->thread_context[i], s);
2895 s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2896 for(i=1; i<s->avctx->thread_count; i++){
2897 merge_context_after_encode(s, s->thread_context[i]);
2903 void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2904 const int intra= s->mb_intra;
2907 s->dct_count[intra]++;
2909 for(i=0; i<64; i++){
2910 int level= block[i];
2914 s->dct_error_sum[intra][i] += level;
2915 level -= s->dct_offset[intra][i];
2916 if(level<0) level=0;
2918 s->dct_error_sum[intra][i] -= level;
2919 level += s->dct_offset[intra][i];
2920 if(level>0) level=0;
2927 int dct_quantize_trellis_c(MpegEncContext *s,
2928 DCTELEM *block, int n,
2929 int qscale, int *overflow){
2931 const uint8_t *scantable= s->intra_scantable.scantable;
2932 const uint8_t *perm_scantable= s->intra_scantable.permutated;
2934 unsigned int threshold1, threshold2;
2946 int coeff_count[64];
2947 int qmul, qadd, start_i, last_non_zero, i, dc;
2948 const int esc_length= s->ac_esc_length;
2950 uint8_t * last_length;
2951 const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2953 s->dsp.fdct (block);
2955 if(s->dct_error_sum)
2956 s->denoise_dct(s, block);
2958 qadd= ((qscale-1)|1)*8;
2969 /* For AIC we skip quant/dequant of INTRADC */
2974 /* note: block[0] is assumed to be positive */
2975 block[0] = (block[0] + (q >> 1)) / q;
2978 qmat = s->q_intra_matrix[qscale];
2979 if(s->mpeg_quant || s->out_format == FMT_MPEG1)
2980 bias= 1<<(QMAT_SHIFT-1);
2981 length = s->intra_ac_vlc_length;
2982 last_length= s->intra_ac_vlc_last_length;
2986 qmat = s->q_inter_matrix[qscale];
2987 length = s->inter_ac_vlc_length;
2988 last_length= s->inter_ac_vlc_last_length;
2992 threshold1= (1<<QMAT_SHIFT) - bias - 1;
2993 threshold2= (threshold1<<1);
2995 for(i=63; i>=start_i; i--) {
2996 const int j = scantable[i];
2997 int level = block[j] * qmat[j];
2999 if(((unsigned)(level+threshold1))>threshold2){
3005 for(i=start_i; i<=last_non_zero; i++) {
3006 const int j = scantable[i];
3007 int level = block[j] * qmat[j];
3009 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
3010 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
3011 if(((unsigned)(level+threshold1))>threshold2){
3013 level= (bias + level)>>QMAT_SHIFT;
3015 coeff[1][i]= level-1;
3016 // coeff[2][k]= level-2;
3018 level= (bias - level)>>QMAT_SHIFT;
3019 coeff[0][i]= -level;
3020 coeff[1][i]= -level+1;
3021 // coeff[2][k]= -level+2;
3023 coeff_count[i]= FFMIN(level, 2);
3024 assert(coeff_count[i]);
3027 coeff[0][i]= (level>>31)|1;
3032 *overflow= s->max_qcoeff < max; //overflow might have happened
3034 if(last_non_zero < start_i){
3035 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3036 return last_non_zero;
3039 score_tab[start_i]= 0;
3040 survivor[0]= start_i;
3043 for(i=start_i; i<=last_non_zero; i++){
3045 const int dct_coeff= FFABS(block[ scantable[i] ]);
3046 const int zero_distoration= dct_coeff*dct_coeff;
3047 int best_score=256*256*256*120;
3048 for(level_index=0; level_index < coeff_count[i]; level_index++){
3050 int level= coeff[level_index][i];
3051 const int alevel= FFABS(level);
3056 if(s->out_format == FMT_H263){
3057 unquant_coeff= alevel*qmul + qadd;
3059 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3061 unquant_coeff = (int)( alevel * qscale * s->intra_matrix[j]) >> 3;
3062 unquant_coeff = (unquant_coeff - 1) | 1;
3064 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3065 unquant_coeff = (unquant_coeff - 1) | 1;
3070 distoration= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distoration;
3072 if((level&(~127)) == 0){
3073 for(j=survivor_count-1; j>=0; j--){
3074 int run= i - survivor[j];
3075 int score= distoration + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3076 score += score_tab[i-run];
3078 if(score < best_score){
3081 level_tab[i+1]= level-64;
3085 if(s->out_format == FMT_H263){
3086 for(j=survivor_count-1; j>=0; j--){
3087 int run= i - survivor[j];
3088 int score= distoration + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3089 score += score_tab[i-run];
3090 if(score < last_score){
3093 last_level= level-64;
3099 distoration += esc_length*lambda;
3100 for(j=survivor_count-1; j>=0; j--){
3101 int run= i - survivor[j];
3102 int score= distoration + score_tab[i-run];
3104 if(score < best_score){
3107 level_tab[i+1]= level-64;
3111 if(s->out_format == FMT_H263){
3112 for(j=survivor_count-1; j>=0; j--){
3113 int run= i - survivor[j];
3114 int score= distoration + score_tab[i-run];
3115 if(score < last_score){
3118 last_level= level-64;
3126 score_tab[i+1]= best_score;
3128 //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3129 if(last_non_zero <= 27){
3130 for(; survivor_count; survivor_count--){
3131 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3135 for(; survivor_count; survivor_count--){
3136 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3141 survivor[ survivor_count++ ]= i+1;
3144 if(s->out_format != FMT_H263){
3145 last_score= 256*256*256*120;
3146 for(i= survivor[0]; i<=last_non_zero + 1; i++){
3147 int score= score_tab[i];
3148 if(i) score += lambda*2; //FIXME exacter?
3150 if(score < last_score){
3153 last_level= level_tab[i];
3154 last_run= run_tab[i];
3159 s->coded_score[n] = last_score;
3161 dc= FFABS(block[0]);
3162 last_non_zero= last_i - 1;
3163 memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3165 if(last_non_zero < start_i)
3166 return last_non_zero;
3168 if(last_non_zero == 0 && start_i == 0){
3170 int best_score= dc * dc;
3172 for(i=0; i<coeff_count[0]; i++){
3173 int level= coeff[i][0];
3174 int alevel= FFABS(level);
3175 int unquant_coeff, score, distortion;
3177 if(s->out_format == FMT_H263){
3178 unquant_coeff= (alevel*qmul + qadd)>>3;
3180 unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3181 unquant_coeff = (unquant_coeff - 1) | 1;
3183 unquant_coeff = (unquant_coeff + 4) >> 3;
3184 unquant_coeff<<= 3 + 3;
3186 distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3188 if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3189 else score= distortion + esc_length*lambda;
3191 if(score < best_score){
3193 best_level= level - 64;
3196 block[0]= best_level;
3197 s->coded_score[n] = best_score - dc*dc;
3198 if(best_level == 0) return -1;
3199 else return last_non_zero;
3205 block[ perm_scantable[last_non_zero] ]= last_level;
3208 for(; i>start_i; i -= run_tab[i] + 1){
3209 block[ perm_scantable[i-1] ]= level_tab[i];
3212 return last_non_zero;
3215 //#define REFINE_STATS 1
3216 static int16_t basis[64][64];
3218 static void build_basis(uint8_t *perm){
3225 double s= 0.25*(1<<BASIS_SHIFT);
3227 int perm_index= perm[index];
3228 if(i==0) s*= sqrt(0.5);
3229 if(j==0) s*= sqrt(0.5);
3230 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)));
3237 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3238 DCTELEM *block, int16_t *weight, DCTELEM *orig,
3241 DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3243 const uint8_t *scantable= s->intra_scantable.scantable;
3244 const uint8_t *perm_scantable= s->intra_scantable.permutated;
3245 // unsigned int threshold1, threshold2;
3250 int qmul, qadd, start_i, last_non_zero, i, dc;
3252 uint8_t * last_length;
3254 int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3257 static int after_last=0;
3258 static int to_zero=0;
3259 static int from_zero=0;
3262 static int messed_sign=0;
3265 if(basis[0][0] == 0)
3266 build_basis(s->dsp.idct_permutation);
3277 /* For AIC we skip quant/dequant of INTRADC */
3281 q <<= RECON_SHIFT-3;
3282 /* note: block[0] is assumed to be positive */
3284 // block[0] = (block[0] + (q >> 1)) / q;
3286 qmat = s->q_intra_matrix[qscale];
3287 // if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3288 // bias= 1<<(QMAT_SHIFT-1);
3289 length = s->intra_ac_vlc_length;
3290 last_length= s->intra_ac_vlc_last_length;
3294 qmat = s->q_inter_matrix[qscale];
3295 length = s->inter_ac_vlc_length;
3296 last_length= s->inter_ac_vlc_last_length;
3298 last_non_zero = s->block_last_index[n];
3303 dc += (1<<(RECON_SHIFT-1));
3304 for(i=0; i<64; i++){
3305 rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME use orig dirrectly instead of copying to rem[]
3308 STOP_TIMER("memset rem[]")}
3311 for(i=0; i<64; i++){
3316 w= FFABS(weight[i]) + qns*one;
3317 w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3320 // w=weight[i] = (63*qns + (w/2)) / w;
3326 lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3332 for(i=start_i; i<=last_non_zero; i++){
3333 int j= perm_scantable[i];
3334 const int level= block[j];
3338 if(level<0) coeff= qmul*level - qadd;
3339 else coeff= qmul*level + qadd;
3340 run_tab[rle_index++]=run;
3343 s->dsp.add_8x8basis(rem, basis[j], coeff);
3349 if(last_non_zero>0){
3350 STOP_TIMER("init rem[]")
3357 int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3360 int run2, best_unquant_change=0, analyze_gradient;
3364 analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3366 if(analyze_gradient){
3370 for(i=0; i<64; i++){
3373 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3376 STOP_TIMER("rem*w*w")}
3386 const int level= block[0];
3387 int change, old_coeff;
3389 assert(s->mb_intra);
3393 for(change=-1; change<=1; change+=2){
3394 int new_level= level + change;
3395 int score, new_coeff;
3397 new_coeff= q*new_level;
3398 if(new_coeff >= 2048 || new_coeff < 0)
3401 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3402 if(score<best_score){
3405 best_change= change;
3406 best_unquant_change= new_coeff - old_coeff;
3413 run2= run_tab[rle_index++];
3417 for(i=start_i; i<64; i++){
3418 int j= perm_scantable[i];
3419 const int level= block[j];
3420 int change, old_coeff;
3422 if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3426 if(level<0) old_coeff= qmul*level - qadd;
3427 else old_coeff= qmul*level + qadd;
3428 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3432 assert(run2>=0 || i >= last_non_zero );
3435 for(change=-1; change<=1; change+=2){
3436 int new_level= level + change;
3437 int score, new_coeff, unquant_change;
3440 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3444 if(new_level<0) new_coeff= qmul*new_level - qadd;
3445 else new_coeff= qmul*new_level + qadd;
3446 if(new_coeff >= 2048 || new_coeff <= -2048)
3448 //FIXME check for overflow
3451 if(level < 63 && level > -63){
3452 if(i < last_non_zero)
3453 score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
3454 - length[UNI_AC_ENC_INDEX(run, level+64)];
3456 score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3457 - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3460 assert(FFABS(new_level)==1);
3462 if(analyze_gradient){
3463 int g= d1[ scantable[i] ];
3464 if(g && (g^new_level) >= 0)
3468 if(i < last_non_zero){
3469 int next_i= i + run2 + 1;
3470 int next_level= block[ perm_scantable[next_i] ] + 64;
3472 if(next_level&(~127))
3475 if(next_i < last_non_zero)
3476 score += length[UNI_AC_ENC_INDEX(run, 65)]
3477 + length[UNI_AC_ENC_INDEX(run2, next_level)]
3478 - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3480 score += length[UNI_AC_ENC_INDEX(run, 65)]
3481 + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3482 - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3484 score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3486 score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3487 - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3493 assert(FFABS(level)==1);
3495 if(i < last_non_zero){
3496 int next_i= i + run2 + 1;
3497 int next_level= block[ perm_scantable[next_i] ] + 64;
3499 if(next_level&(~127))
3502 if(next_i < last_non_zero)
3503 score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3504 - length[UNI_AC_ENC_INDEX(run2, next_level)]
3505 - length[UNI_AC_ENC_INDEX(run, 65)];
3507 score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3508 - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3509 - length[UNI_AC_ENC_INDEX(run, 65)];
3511 score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3513 score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3514 - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3521 unquant_change= new_coeff - old_coeff;
3522 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3524 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3525 if(score<best_score){
3528 best_change= change;
3529 best_unquant_change= unquant_change;
3533 prev_level= level + 64;
3534 if(prev_level&(~127))
3543 STOP_TIMER("iterative step")}
3547 int j= perm_scantable[ best_coeff ];
3549 block[j] += best_change;
3551 if(best_coeff > last_non_zero){
3552 last_non_zero= best_coeff;
3560 if(block[j] - best_change){
3561 if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3573 for(; last_non_zero>=start_i; last_non_zero--){
3574 if(block[perm_scantable[last_non_zero]])
3580 if(256*256*256*64 % count == 0){
3581 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);
3586 for(i=start_i; i<=last_non_zero; i++){
3587 int j= perm_scantable[i];
3588 const int level= block[j];
3591 run_tab[rle_index++]=run;
3598 s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3604 if(last_non_zero>0){
3605 STOP_TIMER("iterative search")
3610 return last_non_zero;
3613 int dct_quantize_c(MpegEncContext *s,
3614 DCTELEM *block, int n,
3615 int qscale, int *overflow)
3617 int i, j, level, last_non_zero, q, start_i;
3619 const uint8_t *scantable= s->intra_scantable.scantable;
3622 unsigned int threshold1, threshold2;
3624 s->dsp.fdct (block);
3626 if(s->dct_error_sum)
3627 s->denoise_dct(s, block);
3637 /* For AIC we skip quant/dequant of INTRADC */
3640 /* note: block[0] is assumed to be positive */
3641 block[0] = (block[0] + (q >> 1)) / q;
3644 qmat = s->q_intra_matrix[qscale];
3645 bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3649 qmat = s->q_inter_matrix[qscale];
3650 bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3652 threshold1= (1<<QMAT_SHIFT) - bias - 1;
3653 threshold2= (threshold1<<1);
3654 for(i=63;i>=start_i;i--) {
3656 level = block[j] * qmat[j];
3658 if(((unsigned)(level+threshold1))>threshold2){
3665 for(i=start_i; i<=last_non_zero; i++) {
3667 level = block[j] * qmat[j];
3669 // if( bias+level >= (1<<QMAT_SHIFT)
3670 // || bias-level >= (1<<QMAT_SHIFT)){
3671 if(((unsigned)(level+threshold1))>threshold2){
3673 level= (bias + level)>>QMAT_SHIFT;
3676 level= (bias - level)>>QMAT_SHIFT;
3684 *overflow= s->max_qcoeff < max; //overflow might have happened
3686 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3687 if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3688 ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3690 return last_non_zero;
3693 AVCodec h263_encoder = {
3697 sizeof(MpegEncContext),
3701 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3704 AVCodec h263p_encoder = {
3708 sizeof(MpegEncContext),
3712 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3715 AVCodec flv_encoder = {
3719 sizeof(MpegEncContext),
3723 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3726 AVCodec rv10_encoder = {
3730 sizeof(MpegEncContext),
3734 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3737 AVCodec rv20_encoder = {
3741 sizeof(MpegEncContext),
3745 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3748 AVCodec mpeg4_encoder = {
3752 sizeof(MpegEncContext),
3756 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3757 .capabilities= CODEC_CAP_DELAY,
3760 AVCodec msmpeg4v1_encoder = {
3764 sizeof(MpegEncContext),
3768 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3771 AVCodec msmpeg4v2_encoder = {
3775 sizeof(MpegEncContext),
3779 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3782 AVCodec msmpeg4v3_encoder = {
3786 sizeof(MpegEncContext),
3790 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3793 AVCodec wmv1_encoder = {
3797 sizeof(MpegEncContext),
3801 .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},