2 * The simplest mpeg encoder (well, it was the simplest!)
3 * Copyright (c) 2000,2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * non linear quantizers with large QPs and VBV with restrictive qmin fixes sponsored by NOA GmbH
31 * The simplest mpeg encoder (well, it was the simplest!).
36 #include "libavutil/internal.h"
37 #include "libavutil/intmath.h"
38 #include "libavutil/mathematics.h"
39 #include "libavutil/mem_internal.h"
40 #include "libavutil/pixdesc.h"
41 #include "libavutil/opt.h"
42 #include "libavutil/thread.h"
47 #include "mpegvideo.h"
48 #include "mpegvideodata.h"
52 #include "mjpegenc_common.h"
54 #include "mpegutils.h"
56 #include "speedhqenc.h"
58 #include "pixblockdsp.h"
62 #include "aandcttab.h"
64 #include "mpeg4video.h"
66 #include "bytestream.h"
69 #include "packet_internal.h"
73 #define QUANT_BIAS_SHIFT 8
75 #define QMAT_SHIFT_MMX 16
78 static int encode_picture(MpegEncContext *s, int picture_number);
79 static int dct_quantize_refine(MpegEncContext *s, int16_t *block, int16_t *weight, int16_t *orig, int n, int qscale);
80 static int sse_mb(MpegEncContext *s);
81 static void denoise_dct_c(MpegEncContext *s, int16_t *block);
82 static int dct_quantize_trellis_c(MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow);
84 static uint8_t default_mv_penalty[MAX_FCODE + 1][MAX_DMV * 2 + 1];
85 static uint8_t default_fcode_tab[MAX_MV * 2 + 1];
87 const AVOption ff_mpv_generic_options[] = {
92 void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
93 uint16_t (*qmat16)[2][64],
94 const uint16_t *quant_matrix,
95 int bias, int qmin, int qmax, int intra)
97 FDCTDSPContext *fdsp = &s->fdsp;
101 for (qscale = qmin; qscale <= qmax; qscale++) {
105 if (s->q_scale_type) qscale2 = ff_mpeg2_non_linear_qscale[qscale];
106 else qscale2 = qscale << 1;
108 if (fdsp->fdct == ff_jpeg_fdct_islow_8 ||
110 fdsp->fdct == ff_faandct ||
111 #endif /* CONFIG_FAANDCT */
112 fdsp->fdct == ff_jpeg_fdct_islow_10) {
113 for (i = 0; i < 64; i++) {
114 const int j = s->idsp.idct_permutation[i];
115 int64_t den = (int64_t) qscale2 * quant_matrix[j];
116 /* 16 <= qscale * quant_matrix[i] <= 7905
117 * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
118 * 19952 <= x <= 249205026
119 * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
120 * 3444240 >= (1 << 36) / (x) >= 275 */
122 qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
124 } else if (fdsp->fdct == ff_fdct_ifast) {
125 for (i = 0; i < 64; i++) {
126 const int j = s->idsp.idct_permutation[i];
127 int64_t den = ff_aanscales[i] * (int64_t) qscale2 * quant_matrix[j];
128 /* 16 <= qscale * quant_matrix[i] <= 7905
129 * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
130 * 19952 <= x <= 249205026
131 * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
132 * 3444240 >= (1 << 36) / (x) >= 275 */
134 qmat[qscale][i] = (int)((UINT64_C(2) << (QMAT_SHIFT + 14)) / den);
137 for (i = 0; i < 64; i++) {
138 const int j = s->idsp.idct_permutation[i];
139 int64_t den = (int64_t) qscale2 * quant_matrix[j];
140 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
141 * Assume x = qscale * quant_matrix[i]
143 * so (1 << 19) / 16 >= (1 << 19) / (x) >= (1 << 19) / 7905
144 * so 32768 >= (1 << 19) / (x) >= 67 */
145 qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
146 //qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) /
147 // (qscale * quant_matrix[i]);
148 qmat16[qscale][0][i] = (2 << QMAT_SHIFT_MMX) / den;
150 if (qmat16[qscale][0][i] == 0 ||
151 qmat16[qscale][0][i] == 128 * 256)
152 qmat16[qscale][0][i] = 128 * 256 - 1;
153 qmat16[qscale][1][i] =
154 ROUNDED_DIV(bias * (1<<(16 - QUANT_BIAS_SHIFT)),
155 qmat16[qscale][0][i]);
159 for (i = intra; i < 64; i++) {
161 if (fdsp->fdct == ff_fdct_ifast) {
162 max = (8191LL * ff_aanscales[i]) >> 14;
164 while (((max * qmat[qscale][i]) >> shift) > INT_MAX) {
170 av_log(s->avctx, AV_LOG_INFO,
171 "Warning, QMAT_SHIFT is larger than %d, overflows possible\n",
176 static inline void update_qscale(MpegEncContext *s)
178 if (s->q_scale_type == 1 && 0) {
180 int bestdiff=INT_MAX;
183 for (i = 0 ; i<FF_ARRAY_ELEMS(ff_mpeg2_non_linear_qscale); i++) {
184 int diff = FFABS((ff_mpeg2_non_linear_qscale[i]<<(FF_LAMBDA_SHIFT + 6)) - (int)s->lambda * 139);
185 if (ff_mpeg2_non_linear_qscale[i] < s->avctx->qmin ||
186 (ff_mpeg2_non_linear_qscale[i] > s->avctx->qmax && !s->vbv_ignore_qmax))
188 if (diff < bestdiff) {
195 s->qscale = (s->lambda * 139 + FF_LAMBDA_SCALE * 64) >>
196 (FF_LAMBDA_SHIFT + 7);
197 s->qscale = av_clip(s->qscale, s->avctx->qmin, s->vbv_ignore_qmax ? 31 : s->avctx->qmax);
200 s->lambda2 = (s->lambda * s->lambda + FF_LAMBDA_SCALE / 2) >>
204 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix)
210 for (i = 0; i < 64; i++) {
211 put_bits(pb, 8, matrix[ff_zigzag_direct[i]]);
218 * init s->current_picture.qscale_table from s->lambda_table
220 void ff_init_qscale_tab(MpegEncContext *s)
222 int8_t * const qscale_table = s->current_picture.qscale_table;
225 for (i = 0; i < s->mb_num; i++) {
226 unsigned int lam = s->lambda_table[s->mb_index2xy[i]];
227 int qp = (lam * 139 + FF_LAMBDA_SCALE * 64) >> (FF_LAMBDA_SHIFT + 7);
228 qscale_table[s->mb_index2xy[i]] = av_clip(qp, s->avctx->qmin,
233 static void update_duplicate_context_after_me(MpegEncContext *dst,
236 #define COPY(a) dst->a= src->a
238 COPY(current_picture);
244 COPY(picture_in_gop_number);
245 COPY(gop_picture_number);
246 COPY(frame_pred_frame_dct); // FIXME don't set in encode_header
247 COPY(progressive_frame); // FIXME don't set in encode_header
248 COPY(partitioned_frame); // FIXME don't set in encode_header
252 static void mpv_encode_init_static(void)
254 for (int i = -16; i < 16; i++)
255 default_fcode_tab[i + MAX_MV] = 1;
259 * Set the given MpegEncContext to defaults for encoding.
260 * the changed fields will not depend upon the prior state of the MpegEncContext.
262 static void mpv_encode_defaults(MpegEncContext *s)
264 static AVOnce init_static_once = AV_ONCE_INIT;
266 ff_mpv_common_defaults(s);
268 ff_thread_once(&init_static_once, mpv_encode_init_static);
270 s->me.mv_penalty = default_mv_penalty;
271 s->fcode_tab = default_fcode_tab;
273 s->input_picture_number = 0;
274 s->picture_in_gop_number = 0;
277 av_cold int ff_dct_encode_init(MpegEncContext *s)
280 ff_dct_encode_init_x86(s);
282 if (CONFIG_H263_ENCODER)
283 ff_h263dsp_init(&s->h263dsp);
284 if (!s->dct_quantize)
285 s->dct_quantize = ff_dct_quantize_c;
287 s->denoise_dct = denoise_dct_c;
288 s->fast_dct_quantize = s->dct_quantize;
289 if (s->avctx->trellis)
290 s->dct_quantize = dct_quantize_trellis_c;
295 /* init video encoder */
296 av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
298 MpegEncContext *s = avctx->priv_data;
299 AVCPBProperties *cpb_props;
300 int i, ret, format_supported;
302 mpv_encode_defaults(s);
304 switch (avctx->codec_id) {
305 case AV_CODEC_ID_MPEG2VIDEO:
306 if (avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
307 avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
308 av_log(avctx, AV_LOG_ERROR,
309 "only YUV420 and YUV422 are supported\n");
310 return AVERROR(EINVAL);
313 case AV_CODEC_ID_MJPEG:
314 case AV_CODEC_ID_AMV:
315 format_supported = 0;
316 /* JPEG color space */
317 if (avctx->pix_fmt == AV_PIX_FMT_YUVJ420P ||
318 avctx->pix_fmt == AV_PIX_FMT_YUVJ422P ||
319 avctx->pix_fmt == AV_PIX_FMT_YUVJ444P ||
320 (avctx->color_range == AVCOL_RANGE_JPEG &&
321 (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
322 avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
323 avctx->pix_fmt == AV_PIX_FMT_YUV444P)))
324 format_supported = 1;
325 /* MPEG color space */
326 else if (avctx->strict_std_compliance <= FF_COMPLIANCE_UNOFFICIAL &&
327 (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
328 avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
329 avctx->pix_fmt == AV_PIX_FMT_YUV444P))
330 format_supported = 1;
332 if (!format_supported) {
333 av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
334 return AVERROR(EINVAL);
337 case AV_CODEC_ID_SPEEDHQ:
338 if (avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
339 avctx->pix_fmt != AV_PIX_FMT_YUV422P &&
340 avctx->pix_fmt != AV_PIX_FMT_YUV444P) {
341 av_log(avctx, AV_LOG_ERROR,
342 "only YUV420/YUV422/YUV444 are supported (no alpha support yet)\n");
343 return AVERROR(EINVAL);
347 if (avctx->pix_fmt != AV_PIX_FMT_YUV420P) {
348 av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
349 return AVERROR(EINVAL);
353 switch (avctx->pix_fmt) {
354 case AV_PIX_FMT_YUVJ444P:
355 case AV_PIX_FMT_YUV444P:
356 s->chroma_format = CHROMA_444;
358 case AV_PIX_FMT_YUVJ422P:
359 case AV_PIX_FMT_YUV422P:
360 s->chroma_format = CHROMA_422;
362 case AV_PIX_FMT_YUVJ420P:
363 case AV_PIX_FMT_YUV420P:
365 s->chroma_format = CHROMA_420;
369 avctx->bits_per_raw_sample = av_clip(avctx->bits_per_raw_sample, 0, 8);
371 #if FF_API_PRIVATE_OPT
372 FF_DISABLE_DEPRECATION_WARNINGS
373 if (avctx->rtp_payload_size)
374 s->rtp_payload_size = avctx->rtp_payload_size;
375 if (avctx->me_penalty_compensation)
376 s->me_penalty_compensation = avctx->me_penalty_compensation;
378 s->me_pre = avctx->pre_me;
379 FF_ENABLE_DEPRECATION_WARNINGS
382 s->bit_rate = avctx->bit_rate;
383 s->width = avctx->width;
384 s->height = avctx->height;
385 if (avctx->gop_size > 600 &&
386 avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
387 av_log(avctx, AV_LOG_WARNING,
388 "keyframe interval too large!, reducing it from %d to %d\n",
389 avctx->gop_size, 600);
390 avctx->gop_size = 600;
392 s->gop_size = avctx->gop_size;
394 if (avctx->max_b_frames > MAX_B_FRAMES) {
395 av_log(avctx, AV_LOG_ERROR, "Too many B-frames requested, maximum "
396 "is %d.\n", MAX_B_FRAMES);
397 avctx->max_b_frames = MAX_B_FRAMES;
399 s->max_b_frames = avctx->max_b_frames;
400 s->codec_id = avctx->codec->id;
401 s->strict_std_compliance = avctx->strict_std_compliance;
402 s->quarter_sample = (avctx->flags & AV_CODEC_FLAG_QPEL) != 0;
403 s->rtp_mode = !!s->rtp_payload_size;
404 s->intra_dc_precision = avctx->intra_dc_precision;
406 // workaround some differences between how applications specify dc precision
407 if (s->intra_dc_precision < 0) {
408 s->intra_dc_precision += 8;
409 } else if (s->intra_dc_precision >= 8)
410 s->intra_dc_precision -= 8;
412 if (s->intra_dc_precision < 0) {
413 av_log(avctx, AV_LOG_ERROR,
414 "intra dc precision must be positive, note some applications use"
415 " 0 and some 8 as base meaning 8bit, the value must not be smaller than that\n");
416 return AVERROR(EINVAL);
419 if (avctx->codec_id == AV_CODEC_ID_AMV || (avctx->active_thread_type & FF_THREAD_SLICE))
422 if (s->intra_dc_precision > (avctx->codec_id == AV_CODEC_ID_MPEG2VIDEO ? 3 : 0)) {
423 av_log(avctx, AV_LOG_ERROR, "intra dc precision too large\n");
424 return AVERROR(EINVAL);
426 s->user_specified_pts = AV_NOPTS_VALUE;
428 if (s->gop_size <= 1) {
436 s->fixed_qscale = !!(avctx->flags & AV_CODEC_FLAG_QSCALE);
438 s->adaptive_quant = (avctx->lumi_masking ||
439 avctx->dark_masking ||
440 avctx->temporal_cplx_masking ||
441 avctx->spatial_cplx_masking ||
444 (s->mpv_flags & FF_MPV_FLAG_QP_RD)) &&
447 s->loop_filter = !!(avctx->flags & AV_CODEC_FLAG_LOOP_FILTER);
449 if (avctx->rc_max_rate && !avctx->rc_buffer_size) {
450 switch(avctx->codec_id) {
451 case AV_CODEC_ID_MPEG1VIDEO:
452 case AV_CODEC_ID_MPEG2VIDEO:
453 avctx->rc_buffer_size = FFMAX(avctx->rc_max_rate, 15000000) * 112LL / 15000000 * 16384;
455 case AV_CODEC_ID_MPEG4:
456 case AV_CODEC_ID_MSMPEG4V1:
457 case AV_CODEC_ID_MSMPEG4V2:
458 case AV_CODEC_ID_MSMPEG4V3:
459 if (avctx->rc_max_rate >= 15000000) {
460 avctx->rc_buffer_size = 320 + (avctx->rc_max_rate - 15000000LL) * (760-320) / (38400000 - 15000000);
461 } else if(avctx->rc_max_rate >= 2000000) {
462 avctx->rc_buffer_size = 80 + (avctx->rc_max_rate - 2000000LL) * (320- 80) / (15000000 - 2000000);
463 } else if(avctx->rc_max_rate >= 384000) {
464 avctx->rc_buffer_size = 40 + (avctx->rc_max_rate - 384000LL) * ( 80- 40) / ( 2000000 - 384000);
466 avctx->rc_buffer_size = 40;
467 avctx->rc_buffer_size *= 16384;
470 if (avctx->rc_buffer_size) {
471 av_log(avctx, AV_LOG_INFO, "Automatically choosing VBV buffer size of %d kbyte\n", avctx->rc_buffer_size/8192);
475 if ((!avctx->rc_max_rate) != (!avctx->rc_buffer_size)) {
476 av_log(avctx, AV_LOG_ERROR, "Either both buffer size and max rate or neither must be specified\n");
477 return AVERROR(EINVAL);
480 if (avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate) {
481 av_log(avctx, AV_LOG_INFO,
482 "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
485 if (avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate) {
486 av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
487 return AVERROR(EINVAL);
490 if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
491 av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
492 return AVERROR(EINVAL);
495 if (avctx->rc_max_rate &&
496 avctx->rc_max_rate == avctx->bit_rate &&
497 avctx->rc_max_rate != avctx->rc_min_rate) {
498 av_log(avctx, AV_LOG_INFO,
499 "impossible bitrate constraints, this will fail\n");
502 if (avctx->rc_buffer_size &&
503 avctx->bit_rate * (int64_t)avctx->time_base.num >
504 avctx->rc_buffer_size * (int64_t)avctx->time_base.den) {
505 av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
506 return AVERROR(EINVAL);
509 if (!s->fixed_qscale &&
510 avctx->bit_rate * av_q2d(avctx->time_base) >
511 avctx->bit_rate_tolerance) {
512 av_log(avctx, AV_LOG_WARNING,
513 "bitrate tolerance %d too small for bitrate %"PRId64", overriding\n", avctx->bit_rate_tolerance, avctx->bit_rate);
514 avctx->bit_rate_tolerance = 5 * avctx->bit_rate * av_q2d(avctx->time_base);
517 if (avctx->rc_max_rate &&
518 avctx->rc_min_rate == avctx->rc_max_rate &&
519 (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
520 s->codec_id == AV_CODEC_ID_MPEG2VIDEO) &&
521 90000LL * (avctx->rc_buffer_size - 1) >
522 avctx->rc_max_rate * 0xFFFFLL) {
523 av_log(avctx, AV_LOG_INFO,
524 "Warning vbv_delay will be set to 0xFFFF (=VBR) as the "
525 "specified vbv buffer is too large for the given bitrate!\n");
528 if ((avctx->flags & AV_CODEC_FLAG_4MV) && s->codec_id != AV_CODEC_ID_MPEG4 &&
529 s->codec_id != AV_CODEC_ID_H263 && s->codec_id != AV_CODEC_ID_H263P &&
530 s->codec_id != AV_CODEC_ID_FLV1) {
531 av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
532 return AVERROR(EINVAL);
535 if (s->obmc && avctx->mb_decision != FF_MB_DECISION_SIMPLE) {
536 av_log(avctx, AV_LOG_ERROR,
537 "OBMC is only supported with simple mb decision\n");
538 return AVERROR(EINVAL);
541 if (s->quarter_sample && s->codec_id != AV_CODEC_ID_MPEG4) {
542 av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
543 return AVERROR(EINVAL);
546 if (s->max_b_frames &&
547 s->codec_id != AV_CODEC_ID_MPEG4 &&
548 s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
549 s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
550 av_log(avctx, AV_LOG_ERROR, "B-frames not supported by codec\n");
551 return AVERROR(EINVAL);
553 if (s->max_b_frames < 0) {
554 av_log(avctx, AV_LOG_ERROR,
555 "max b frames must be 0 or positive for mpegvideo based encoders\n");
556 return AVERROR(EINVAL);
559 if ((s->codec_id == AV_CODEC_ID_MPEG4 ||
560 s->codec_id == AV_CODEC_ID_H263 ||
561 s->codec_id == AV_CODEC_ID_H263P) &&
562 (avctx->sample_aspect_ratio.num > 255 ||
563 avctx->sample_aspect_ratio.den > 255)) {
564 av_log(avctx, AV_LOG_WARNING,
565 "Invalid pixel aspect ratio %i/%i, limit is 255/255 reducing\n",
566 avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
567 av_reduce(&avctx->sample_aspect_ratio.num, &avctx->sample_aspect_ratio.den,
568 avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den, 255);
571 if ((s->codec_id == AV_CODEC_ID_H263 ||
572 s->codec_id == AV_CODEC_ID_H263P) &&
573 (avctx->width > 2048 ||
574 avctx->height > 1152 )) {
575 av_log(avctx, AV_LOG_ERROR, "H.263 does not support resolutions above 2048x1152\n");
576 return AVERROR(EINVAL);
578 if ((s->codec_id == AV_CODEC_ID_H263 ||
579 s->codec_id == AV_CODEC_ID_H263P) &&
580 ((avctx->width &3) ||
581 (avctx->height&3) )) {
582 av_log(avctx, AV_LOG_ERROR, "w/h must be a multiple of 4\n");
583 return AVERROR(EINVAL);
586 if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO &&
587 (avctx->width > 4095 ||
588 avctx->height > 4095 )) {
589 av_log(avctx, AV_LOG_ERROR, "MPEG-1 does not support resolutions above 4095x4095\n");
590 return AVERROR(EINVAL);
593 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO &&
594 (avctx->width > 16383 ||
595 avctx->height > 16383 )) {
596 av_log(avctx, AV_LOG_ERROR, "MPEG-2 does not support resolutions above 16383x16383\n");
597 return AVERROR(EINVAL);
600 if (s->codec_id == AV_CODEC_ID_RV10 &&
602 avctx->height&15 )) {
603 av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 16\n");
604 return AVERROR(EINVAL);
607 if (s->codec_id == AV_CODEC_ID_RV20 &&
610 av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 4\n");
611 return AVERROR(EINVAL);
614 if ((s->codec_id == AV_CODEC_ID_WMV1 ||
615 s->codec_id == AV_CODEC_ID_WMV2) &&
617 av_log(avctx, AV_LOG_ERROR, "width must be multiple of 2\n");
618 return AVERROR(EINVAL);
621 if ((avctx->flags & (AV_CODEC_FLAG_INTERLACED_DCT | AV_CODEC_FLAG_INTERLACED_ME)) &&
622 s->codec_id != AV_CODEC_ID_MPEG4 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
623 av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
624 return AVERROR(EINVAL);
627 #if FF_API_PRIVATE_OPT
628 FF_DISABLE_DEPRECATION_WARNINGS
629 if (avctx->mpeg_quant)
630 s->mpeg_quant = avctx->mpeg_quant;
631 FF_ENABLE_DEPRECATION_WARNINGS
634 // FIXME mpeg2 uses that too
635 if (s->mpeg_quant && ( s->codec_id != AV_CODEC_ID_MPEG4
636 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO)) {
637 av_log(avctx, AV_LOG_ERROR,
638 "mpeg2 style quantization not supported by codec\n");
639 return AVERROR(EINVAL);
642 if ((s->mpv_flags & FF_MPV_FLAG_CBP_RD) && !avctx->trellis) {
643 av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
644 return AVERROR(EINVAL);
647 if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
648 avctx->mb_decision != FF_MB_DECISION_RD) {
649 av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
650 return AVERROR(EINVAL);
653 if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
654 (s->codec_id == AV_CODEC_ID_AMV ||
655 s->codec_id == AV_CODEC_ID_MJPEG)) {
656 // Used to produce garbage with MJPEG.
657 av_log(avctx, AV_LOG_ERROR,
658 "QP RD is no longer compatible with MJPEG or AMV\n");
659 return AVERROR(EINVAL);
662 #if FF_API_PRIVATE_OPT
663 FF_DISABLE_DEPRECATION_WARNINGS
664 if (avctx->scenechange_threshold)
665 s->scenechange_threshold = avctx->scenechange_threshold;
666 FF_ENABLE_DEPRECATION_WARNINGS
669 if (s->scenechange_threshold < 1000000000 &&
670 (avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)) {
671 av_log(avctx, AV_LOG_ERROR,
672 "closed gop with scene change detection are not supported yet, "
673 "set threshold to 1000000000\n");
674 return AVERROR_PATCHWELCOME;
677 if (avctx->flags & AV_CODEC_FLAG_LOW_DELAY) {
678 if (s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
679 s->strict_std_compliance >= FF_COMPLIANCE_NORMAL) {
680 av_log(avctx, AV_LOG_ERROR,
681 "low delay forcing is only available for mpeg2, "
682 "set strict_std_compliance to 'unofficial' or lower in order to allow it\n");
683 return AVERROR(EINVAL);
685 if (s->max_b_frames != 0) {
686 av_log(avctx, AV_LOG_ERROR,
687 "B-frames cannot be used with low delay\n");
688 return AVERROR(EINVAL);
692 if (s->q_scale_type == 1) {
693 if (avctx->qmax > 28) {
694 av_log(avctx, AV_LOG_ERROR,
695 "non linear quant only supports qmax <= 28 currently\n");
696 return AVERROR_PATCHWELCOME;
700 if (avctx->slices > 1 &&
701 (avctx->codec_id == AV_CODEC_ID_FLV1 || avctx->codec_id == AV_CODEC_ID_H261)) {
702 av_log(avctx, AV_LOG_ERROR, "Multiple slices are not supported by this codec\n");
703 return AVERROR(EINVAL);
706 if (avctx->thread_count > 1 &&
707 s->codec_id != AV_CODEC_ID_MPEG4 &&
708 s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
709 s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
710 s->codec_id != AV_CODEC_ID_MJPEG &&
711 (s->codec_id != AV_CODEC_ID_H263P)) {
712 av_log(avctx, AV_LOG_ERROR,
713 "multi threaded encoding not supported by codec\n");
714 return AVERROR_PATCHWELCOME;
717 if (avctx->thread_count < 1) {
718 av_log(avctx, AV_LOG_ERROR,
719 "automatic thread number detection not supported by codec, "
721 return AVERROR_PATCHWELCOME;
724 if (!avctx->time_base.den || !avctx->time_base.num) {
725 av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
726 return AVERROR(EINVAL);
729 #if FF_API_PRIVATE_OPT
730 FF_DISABLE_DEPRECATION_WARNINGS
731 if (avctx->b_frame_strategy)
732 s->b_frame_strategy = avctx->b_frame_strategy;
733 if (avctx->b_sensitivity != 40)
734 s->b_sensitivity = avctx->b_sensitivity;
735 FF_ENABLE_DEPRECATION_WARNINGS
738 if (s->b_frame_strategy && (avctx->flags & AV_CODEC_FLAG_PASS2)) {
739 av_log(avctx, AV_LOG_INFO,
740 "notice: b_frame_strategy only affects the first pass\n");
741 s->b_frame_strategy = 0;
744 i = av_gcd(avctx->time_base.den, avctx->time_base.num);
746 av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
747 avctx->time_base.den /= i;
748 avctx->time_base.num /= i;
752 if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG1VIDEO || s->codec_id == AV_CODEC_ID_MPEG2VIDEO || s->codec_id == AV_CODEC_ID_MJPEG || s->codec_id == AV_CODEC_ID_AMV || s->codec_id == AV_CODEC_ID_SPEEDHQ) {
753 // (a + x * 3 / 8) / x
754 s->intra_quant_bias = 3 << (QUANT_BIAS_SHIFT - 3);
755 s->inter_quant_bias = 0;
757 s->intra_quant_bias = 0;
759 s->inter_quant_bias = -(1 << (QUANT_BIAS_SHIFT - 2));
762 if (avctx->qmin > avctx->qmax || avctx->qmin <= 0) {
763 av_log(avctx, AV_LOG_ERROR, "qmin and or qmax are invalid, they must be 0 < min <= max\n");
764 return AVERROR(EINVAL);
767 av_log(avctx, AV_LOG_DEBUG, "intra_quant_bias = %d inter_quant_bias = %d\n",s->intra_quant_bias,s->inter_quant_bias);
769 if (avctx->codec_id == AV_CODEC_ID_MPEG4 &&
770 avctx->time_base.den > (1 << 16) - 1) {
771 av_log(avctx, AV_LOG_ERROR,
772 "timebase %d/%d not supported by MPEG 4 standard, "
773 "the maximum admitted value for the timebase denominator "
774 "is %d\n", avctx->time_base.num, avctx->time_base.den,
776 return AVERROR(EINVAL);
778 s->time_increment_bits = av_log2(avctx->time_base.den - 1) + 1;
780 switch (avctx->codec->id) {
781 case AV_CODEC_ID_MPEG1VIDEO:
782 s->out_format = FMT_MPEG1;
783 s->low_delay = !!(avctx->flags & AV_CODEC_FLAG_LOW_DELAY);
784 avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
786 case AV_CODEC_ID_MPEG2VIDEO:
787 s->out_format = FMT_MPEG1;
788 s->low_delay = !!(avctx->flags & AV_CODEC_FLAG_LOW_DELAY);
789 avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
792 case AV_CODEC_ID_MJPEG:
793 case AV_CODEC_ID_AMV:
794 s->out_format = FMT_MJPEG;
795 s->intra_only = 1; /* force intra only for jpeg */
796 if (!CONFIG_MJPEG_ENCODER)
797 return AVERROR_ENCODER_NOT_FOUND;
798 if ((ret = ff_mjpeg_encode_init(s)) < 0)
803 case AV_CODEC_ID_SPEEDHQ:
804 s->out_format = FMT_SPEEDHQ;
805 s->intra_only = 1; /* force intra only for SHQ */
806 if (!CONFIG_SPEEDHQ_ENCODER)
807 return AVERROR_ENCODER_NOT_FOUND;
808 if ((ret = ff_speedhq_encode_init(s)) < 0)
813 case AV_CODEC_ID_H261:
814 if (!CONFIG_H261_ENCODER)
815 return AVERROR_ENCODER_NOT_FOUND;
816 if (ff_h261_get_picture_format(s->width, s->height) < 0) {
817 av_log(avctx, AV_LOG_ERROR,
818 "The specified picture size of %dx%d is not valid for the "
819 "H.261 codec.\nValid sizes are 176x144, 352x288\n",
820 s->width, s->height);
821 return AVERROR(EINVAL);
823 s->out_format = FMT_H261;
826 s->rtp_mode = 0; /* Sliced encoding not supported */
828 case AV_CODEC_ID_H263:
829 if (!CONFIG_H263_ENCODER)
830 return AVERROR_ENCODER_NOT_FOUND;
831 if (ff_match_2uint16(ff_h263_format, FF_ARRAY_ELEMS(ff_h263_format),
832 s->width, s->height) == 8) {
833 av_log(avctx, AV_LOG_ERROR,
834 "The specified picture size of %dx%d is not valid for "
835 "the H.263 codec.\nValid sizes are 128x96, 176x144, "
836 "352x288, 704x576, and 1408x1152. "
837 "Try H.263+.\n", s->width, s->height);
838 return AVERROR(EINVAL);
840 s->out_format = FMT_H263;
844 case AV_CODEC_ID_H263P:
845 s->out_format = FMT_H263;
848 s->h263_aic = (avctx->flags & AV_CODEC_FLAG_AC_PRED) ? 1 : 0;
849 s->modified_quant = s->h263_aic;
850 s->loop_filter = (avctx->flags & AV_CODEC_FLAG_LOOP_FILTER) ? 1 : 0;
851 s->unrestricted_mv = s->obmc || s->loop_filter || s->umvplus;
854 /* These are just to be sure */
858 case AV_CODEC_ID_FLV1:
859 s->out_format = FMT_H263;
860 s->h263_flv = 2; /* format = 1; 11-bit codes */
861 s->unrestricted_mv = 1;
862 s->rtp_mode = 0; /* don't allow GOB */
866 case AV_CODEC_ID_RV10:
867 s->out_format = FMT_H263;
871 case AV_CODEC_ID_RV20:
872 s->out_format = FMT_H263;
875 s->modified_quant = 1;
879 s->unrestricted_mv = 0;
881 case AV_CODEC_ID_MPEG4:
882 s->out_format = FMT_H263;
884 s->unrestricted_mv = 1;
885 s->low_delay = s->max_b_frames ? 0 : 1;
886 avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
888 case AV_CODEC_ID_MSMPEG4V2:
889 s->out_format = FMT_H263;
891 s->unrestricted_mv = 1;
892 s->msmpeg4_version = 2;
896 case AV_CODEC_ID_MSMPEG4V3:
897 s->out_format = FMT_H263;
899 s->unrestricted_mv = 1;
900 s->msmpeg4_version = 3;
901 s->flipflop_rounding = 1;
905 case AV_CODEC_ID_WMV1:
906 s->out_format = FMT_H263;
908 s->unrestricted_mv = 1;
909 s->msmpeg4_version = 4;
910 s->flipflop_rounding = 1;
914 case AV_CODEC_ID_WMV2:
915 s->out_format = FMT_H263;
917 s->unrestricted_mv = 1;
918 s->msmpeg4_version = 5;
919 s->flipflop_rounding = 1;
924 return AVERROR(EINVAL);
927 #if FF_API_PRIVATE_OPT
928 FF_DISABLE_DEPRECATION_WARNINGS
929 if (avctx->noise_reduction)
930 s->noise_reduction = avctx->noise_reduction;
931 FF_ENABLE_DEPRECATION_WARNINGS
934 avctx->has_b_frames = !s->low_delay;
938 s->progressive_frame =
939 s->progressive_sequence = !(avctx->flags & (AV_CODEC_FLAG_INTERLACED_DCT |
940 AV_CODEC_FLAG_INTERLACED_ME) ||
945 if ((ret = ff_mpv_common_init(s)) < 0)
948 ff_fdctdsp_init(&s->fdsp, avctx);
949 ff_me_cmp_init(&s->mecc, avctx);
950 ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
951 ff_pixblockdsp_init(&s->pdsp, avctx);
952 ff_qpeldsp_init(&s->qdsp);
954 if (s->msmpeg4_version) {
955 int ac_stats_size = 2 * 2 * (MAX_LEVEL + 1) * (MAX_RUN + 1) * 2 * sizeof(int);
956 if (!(s->ac_stats = av_mallocz(ac_stats_size)))
957 return AVERROR(ENOMEM);
960 if (!(avctx->stats_out = av_mallocz(256)) ||
961 !FF_ALLOCZ_TYPED_ARRAY(s->q_intra_matrix, 32) ||
962 !FF_ALLOCZ_TYPED_ARRAY(s->q_chroma_intra_matrix, 32) ||
963 !FF_ALLOCZ_TYPED_ARRAY(s->q_inter_matrix, 32) ||
964 !FF_ALLOCZ_TYPED_ARRAY(s->q_intra_matrix16, 32) ||
965 !FF_ALLOCZ_TYPED_ARRAY(s->q_chroma_intra_matrix16, 32) ||
966 !FF_ALLOCZ_TYPED_ARRAY(s->q_inter_matrix16, 32) ||
967 !FF_ALLOCZ_TYPED_ARRAY(s->input_picture, MAX_PICTURE_COUNT) ||
968 !FF_ALLOCZ_TYPED_ARRAY(s->reordered_input_picture, MAX_PICTURE_COUNT))
969 return AVERROR(ENOMEM);
971 if (s->noise_reduction) {
972 if (!FF_ALLOCZ_TYPED_ARRAY(s->dct_offset, 2))
973 return AVERROR(ENOMEM);
976 ff_dct_encode_init(s);
978 if ((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
979 s->chroma_qscale_table = ff_h263_chroma_qscale_table;
981 if (s->slice_context_count > 1) {
984 if (avctx->codec_id == AV_CODEC_ID_H263P)
985 s->h263_slice_structured = 1;
988 s->quant_precision = 5;
990 #if FF_API_PRIVATE_OPT
991 FF_DISABLE_DEPRECATION_WARNINGS
992 if (avctx->frame_skip_threshold)
993 s->frame_skip_threshold = avctx->frame_skip_threshold;
994 if (avctx->frame_skip_factor)
995 s->frame_skip_factor = avctx->frame_skip_factor;
996 if (avctx->frame_skip_exp)
997 s->frame_skip_exp = avctx->frame_skip_exp;
998 if (avctx->frame_skip_cmp != FF_CMP_DCTMAX)
999 s->frame_skip_cmp = avctx->frame_skip_cmp;
1000 FF_ENABLE_DEPRECATION_WARNINGS
1003 ff_set_cmp(&s->mecc, s->mecc.ildct_cmp, avctx->ildct_cmp);
1004 ff_set_cmp(&s->mecc, s->mecc.frame_skip_cmp, s->frame_skip_cmp);
1006 if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
1007 ff_h261_encode_init(s);
1008 if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
1009 ff_h263_encode_init(s);
1010 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
1011 ff_msmpeg4_encode_init(s);
1012 if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1013 && s->out_format == FMT_MPEG1)
1014 ff_mpeg1_encode_init(s);
1017 for (i = 0; i < 64; i++) {
1018 int j = s->idsp.idct_permutation[i];
1019 if (CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4 &&
1021 s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
1022 s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
1023 } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1024 s->intra_matrix[j] =
1025 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1026 } else if (CONFIG_SPEEDHQ_ENCODER && s->codec_id == AV_CODEC_ID_SPEEDHQ) {
1027 s->intra_matrix[j] =
1028 s->inter_matrix[j] = ff_mpeg1_default_intra_matrix[i];
1031 s->chroma_intra_matrix[j] =
1032 s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
1033 s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1035 if (avctx->intra_matrix)
1036 s->intra_matrix[j] = avctx->intra_matrix[i];
1037 if (avctx->inter_matrix)
1038 s->inter_matrix[j] = avctx->inter_matrix[i];
1041 /* precompute matrix */
1042 /* for mjpeg, we do include qscale in the matrix */
1043 if (s->out_format != FMT_MJPEG) {
1044 ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
1045 s->intra_matrix, s->intra_quant_bias, avctx->qmin,
1047 ff_convert_matrix(s, s->q_inter_matrix, s->q_inter_matrix16,
1048 s->inter_matrix, s->inter_quant_bias, avctx->qmin,
1052 if ((ret = ff_rate_control_init(s)) < 0)
1055 #if FF_API_PRIVATE_OPT
1056 FF_DISABLE_DEPRECATION_WARNINGS
1057 if (avctx->brd_scale)
1058 s->brd_scale = avctx->brd_scale;
1060 if (avctx->prediction_method)
1061 s->pred = avctx->prediction_method + 1;
1062 FF_ENABLE_DEPRECATION_WARNINGS
1065 if (s->b_frame_strategy == 2) {
1066 for (i = 0; i < s->max_b_frames + 2; i++) {
1067 s->tmp_frames[i] = av_frame_alloc();
1068 if (!s->tmp_frames[i])
1069 return AVERROR(ENOMEM);
1071 s->tmp_frames[i]->format = AV_PIX_FMT_YUV420P;
1072 s->tmp_frames[i]->width = s->width >> s->brd_scale;
1073 s->tmp_frames[i]->height = s->height >> s->brd_scale;
1075 ret = av_frame_get_buffer(s->tmp_frames[i], 0);
1081 cpb_props = ff_add_cpb_side_data(avctx);
1083 return AVERROR(ENOMEM);
1084 cpb_props->max_bitrate = avctx->rc_max_rate;
1085 cpb_props->min_bitrate = avctx->rc_min_rate;
1086 cpb_props->avg_bitrate = avctx->bit_rate;
1087 cpb_props->buffer_size = avctx->rc_buffer_size;
1092 av_cold int ff_mpv_encode_end(AVCodecContext *avctx)
1094 MpegEncContext *s = avctx->priv_data;
1097 ff_rate_control_uninit(s);
1099 ff_mpv_common_end(s);
1100 if (CONFIG_MJPEG_ENCODER &&
1101 s->out_format == FMT_MJPEG)
1102 ff_mjpeg_encode_close(s);
1104 av_freep(&avctx->extradata);
1106 for (i = 0; i < FF_ARRAY_ELEMS(s->tmp_frames); i++)
1107 av_frame_free(&s->tmp_frames[i]);
1109 ff_free_picture_tables(&s->new_picture);
1110 ff_mpeg_unref_picture(avctx, &s->new_picture);
1112 av_freep(&avctx->stats_out);
1113 av_freep(&s->ac_stats);
1115 if(s->q_chroma_intra_matrix != s->q_intra_matrix ) av_freep(&s->q_chroma_intra_matrix);
1116 if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
1117 s->q_chroma_intra_matrix= NULL;
1118 s->q_chroma_intra_matrix16= NULL;
1119 av_freep(&s->q_intra_matrix);
1120 av_freep(&s->q_inter_matrix);
1121 av_freep(&s->q_intra_matrix16);
1122 av_freep(&s->q_inter_matrix16);
1123 av_freep(&s->input_picture);
1124 av_freep(&s->reordered_input_picture);
1125 av_freep(&s->dct_offset);
1130 static int get_sae(uint8_t *src, int ref, int stride)
1135 for (y = 0; y < 16; y++) {
1136 for (x = 0; x < 16; x++) {
1137 acc += FFABS(src[x + y * stride] - ref);
1144 static int get_intra_count(MpegEncContext *s, uint8_t *src,
1145 uint8_t *ref, int stride)
1151 h = s->height & ~15;
1153 for (y = 0; y < h; y += 16) {
1154 for (x = 0; x < w; x += 16) {
1155 int offset = x + y * stride;
1156 int sad = s->mecc.sad[0](NULL, src + offset, ref + offset,
1158 int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8;
1159 int sae = get_sae(src + offset, mean, stride);
1161 acc += sae + 500 < sad;
1167 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
1169 return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 1,
1170 s->chroma_x_shift, s->chroma_y_shift, s->out_format,
1171 s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
1172 &s->linesize, &s->uvlinesize);
1175 static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
1177 Picture *pic = NULL;
1179 int i, display_picture_number = 0, ret;
1180 int encoding_delay = s->max_b_frames ? s->max_b_frames
1181 : (s->low_delay ? 0 : 1);
1182 int flush_offset = 1;
1187 display_picture_number = s->input_picture_number++;
1189 if (pts != AV_NOPTS_VALUE) {
1190 if (s->user_specified_pts != AV_NOPTS_VALUE) {
1191 int64_t last = s->user_specified_pts;
1194 av_log(s->avctx, AV_LOG_ERROR,
1195 "Invalid pts (%"PRId64") <= last (%"PRId64")\n",
1197 return AVERROR(EINVAL);
1200 if (!s->low_delay && display_picture_number == 1)
1201 s->dts_delta = pts - last;
1203 s->user_specified_pts = pts;
1205 if (s->user_specified_pts != AV_NOPTS_VALUE) {
1206 s->user_specified_pts =
1207 pts = s->user_specified_pts + 1;
1208 av_log(s->avctx, AV_LOG_INFO,
1209 "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n",
1212 pts = display_picture_number;
1216 if (!pic_arg->buf[0] ||
1217 pic_arg->linesize[0] != s->linesize ||
1218 pic_arg->linesize[1] != s->uvlinesize ||
1219 pic_arg->linesize[2] != s->uvlinesize)
1221 if ((s->width & 15) || (s->height & 15))
1223 if (((intptr_t)(pic_arg->data[0])) & (STRIDE_ALIGN-1))
1225 if (s->linesize & (STRIDE_ALIGN-1))
1228 ff_dlog(s->avctx, "%d %d %"PTRDIFF_SPECIFIER" %"PTRDIFF_SPECIFIER"\n", pic_arg->linesize[0],
1229 pic_arg->linesize[1], s->linesize, s->uvlinesize);
1231 i = ff_find_unused_picture(s->avctx, s->picture, direct);
1235 pic = &s->picture[i];
1239 if ((ret = av_frame_ref(pic->f, pic_arg)) < 0)
1242 ret = alloc_picture(s, pic, direct);
1247 if (pic->f->data[0] + INPLACE_OFFSET == pic_arg->data[0] &&
1248 pic->f->data[1] + INPLACE_OFFSET == pic_arg->data[1] &&
1249 pic->f->data[2] + INPLACE_OFFSET == pic_arg->data[2]) {
1252 int h_chroma_shift, v_chroma_shift;
1253 av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1257 for (i = 0; i < 3; i++) {
1258 int src_stride = pic_arg->linesize[i];
1259 int dst_stride = i ? s->uvlinesize : s->linesize;
1260 int h_shift = i ? h_chroma_shift : 0;
1261 int v_shift = i ? v_chroma_shift : 0;
1262 int w = s->width >> h_shift;
1263 int h = s->height >> v_shift;
1264 uint8_t *src = pic_arg->data[i];
1265 uint8_t *dst = pic->f->data[i];
1268 if ( s->codec_id == AV_CODEC_ID_MPEG2VIDEO
1269 && !s->progressive_sequence
1270 && FFALIGN(s->height, 32) - s->height > 16)
1273 if (!s->avctx->rc_buffer_size)
1274 dst += INPLACE_OFFSET;
1276 if (src_stride == dst_stride)
1277 memcpy(dst, src, src_stride * h);
1280 uint8_t *dst2 = dst;
1282 memcpy(dst2, src, w);
1287 if ((s->width & 15) || (s->height & (vpad-1))) {
1288 s->mpvencdsp.draw_edges(dst, dst_stride,
1298 ret = av_frame_copy_props(pic->f, pic_arg);
1302 pic->f->display_picture_number = display_picture_number;
1303 pic->f->pts = pts; // we set this here to avoid modifying pic_arg
1305 /* Flushing: When we have not received enough input frames,
1306 * ensure s->input_picture[0] contains the first picture */
1307 for (flush_offset = 0; flush_offset < encoding_delay + 1; flush_offset++)
1308 if (s->input_picture[flush_offset])
1311 if (flush_offset <= 1)
1314 encoding_delay = encoding_delay - flush_offset + 1;
1317 /* shift buffer entries */
1318 for (i = flush_offset; i < MAX_PICTURE_COUNT /*s->encoding_delay + 1*/; i++)
1319 s->input_picture[i - flush_offset] = s->input_picture[i];
1321 s->input_picture[encoding_delay] = (Picture*) pic;
1326 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref)
1330 int64_t score64 = 0;
1332 for (plane = 0; plane < 3; plane++) {
1333 const int stride = p->f->linesize[plane];
1334 const int bw = plane ? 1 : 2;
1335 for (y = 0; y < s->mb_height * bw; y++) {
1336 for (x = 0; x < s->mb_width * bw; x++) {
1337 int off = p->shared ? 0 : 16;
1338 uint8_t *dptr = p->f->data[plane] + 8 * (x + y * stride) + off;
1339 uint8_t *rptr = ref->f->data[plane] + 8 * (x + y * stride);
1340 int v = s->mecc.frame_skip_cmp[1](s, dptr, rptr, stride, 8);
1342 switch (FFABS(s->frame_skip_exp)) {
1343 case 0: score = FFMAX(score, v); break;
1344 case 1: score += FFABS(v); break;
1345 case 2: score64 += v * (int64_t)v; break;
1346 case 3: score64 += FFABS(v * (int64_t)v * v); break;
1347 case 4: score64 += (v * (int64_t)v) * (v * (int64_t)v); break;
1356 if (s->frame_skip_exp < 0)
1357 score64 = pow(score64 / (double)(s->mb_width * s->mb_height),
1358 -1.0/s->frame_skip_exp);
1360 if (score64 < s->frame_skip_threshold)
1362 if (score64 < ((s->frame_skip_factor * (int64_t) s->lambda) >> 8))
1367 static int encode_frame(AVCodecContext *c, AVFrame *frame, AVPacket *pkt)
1372 ret = avcodec_send_frame(c, frame);
1377 ret = avcodec_receive_packet(c, pkt);
1380 av_packet_unref(pkt);
1381 } else if (ret < 0 && ret != AVERROR(EAGAIN) && ret != AVERROR_EOF)
1388 static int estimate_best_b_count(MpegEncContext *s)
1390 const AVCodec *codec = avcodec_find_encoder(s->avctx->codec_id);
1392 const int scale = s->brd_scale;
1393 int width = s->width >> scale;
1394 int height = s->height >> scale;
1395 int i, j, out_size, p_lambda, b_lambda, lambda2;
1396 int64_t best_rd = INT64_MAX;
1397 int best_b_count = -1;
1400 av_assert0(scale >= 0 && scale <= 3);
1402 pkt = av_packet_alloc();
1404 return AVERROR(ENOMEM);
1407 //s->next_picture_ptr->quality;
1408 p_lambda = s->last_lambda_for[AV_PICTURE_TYPE_P];
1409 //p_lambda * FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
1410 b_lambda = s->last_lambda_for[AV_PICTURE_TYPE_B];
1411 if (!b_lambda) // FIXME we should do this somewhere else
1412 b_lambda = p_lambda;
1413 lambda2 = (b_lambda * b_lambda + (1 << FF_LAMBDA_SHIFT) / 2) >>
1416 for (i = 0; i < s->max_b_frames + 2; i++) {
1417 Picture pre_input, *pre_input_ptr = i ? s->input_picture[i - 1] :
1418 s->next_picture_ptr;
1421 if (pre_input_ptr && (!i || s->input_picture[i - 1])) {
1422 pre_input = *pre_input_ptr;
1423 memcpy(data, pre_input_ptr->f->data, sizeof(data));
1425 if (!pre_input.shared && i) {
1426 data[0] += INPLACE_OFFSET;
1427 data[1] += INPLACE_OFFSET;
1428 data[2] += INPLACE_OFFSET;
1431 s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0],
1432 s->tmp_frames[i]->linesize[0],
1434 pre_input.f->linesize[0],
1436 s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1],
1437 s->tmp_frames[i]->linesize[1],
1439 pre_input.f->linesize[1],
1440 width >> 1, height >> 1);
1441 s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2],
1442 s->tmp_frames[i]->linesize[2],
1444 pre_input.f->linesize[2],
1445 width >> 1, height >> 1);
1449 for (j = 0; j < s->max_b_frames + 1; j++) {
1453 if (!s->input_picture[j])
1456 c = avcodec_alloc_context3(NULL);
1458 ret = AVERROR(ENOMEM);
1464 c->flags = AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_PSNR;
1465 c->flags |= s->avctx->flags & AV_CODEC_FLAG_QPEL;
1466 c->mb_decision = s->avctx->mb_decision;
1467 c->me_cmp = s->avctx->me_cmp;
1468 c->mb_cmp = s->avctx->mb_cmp;
1469 c->me_sub_cmp = s->avctx->me_sub_cmp;
1470 c->pix_fmt = AV_PIX_FMT_YUV420P;
1471 c->time_base = s->avctx->time_base;
1472 c->max_b_frames = s->max_b_frames;
1474 ret = avcodec_open2(c, codec, NULL);
1479 s->tmp_frames[0]->pict_type = AV_PICTURE_TYPE_I;
1480 s->tmp_frames[0]->quality = 1 * FF_QP2LAMBDA;
1482 out_size = encode_frame(c, s->tmp_frames[0], pkt);
1488 //rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1490 for (i = 0; i < s->max_b_frames + 1; i++) {
1491 int is_p = i % (j + 1) == j || i == s->max_b_frames;
1493 s->tmp_frames[i + 1]->pict_type = is_p ?
1494 AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1495 s->tmp_frames[i + 1]->quality = is_p ? p_lambda : b_lambda;
1497 out_size = encode_frame(c, s->tmp_frames[i + 1], pkt);
1503 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1506 /* get the delayed frames */
1507 out_size = encode_frame(c, NULL, pkt);
1512 rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1514 rd += c->error[0] + c->error[1] + c->error[2];
1522 avcodec_free_context(&c);
1523 av_packet_unref(pkt);
1530 av_packet_free(&pkt);
1532 return best_b_count;
1535 static int select_input_picture(MpegEncContext *s)
1539 for (i = 1; i < MAX_PICTURE_COUNT; i++)
1540 s->reordered_input_picture[i - 1] = s->reordered_input_picture[i];
1541 s->reordered_input_picture[MAX_PICTURE_COUNT - 1] = NULL;
1543 /* set next picture type & ordering */
1544 if (!s->reordered_input_picture[0] && s->input_picture[0]) {
1545 if (s->frame_skip_threshold || s->frame_skip_factor) {
1546 if (s->picture_in_gop_number < s->gop_size &&
1547 s->next_picture_ptr &&
1548 skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
1549 // FIXME check that the gop check above is +-1 correct
1550 av_frame_unref(s->input_picture[0]->f);
1552 ff_vbv_update(s, 0);
1558 if (/*s->picture_in_gop_number >= s->gop_size ||*/
1559 !s->next_picture_ptr || s->intra_only) {
1560 s->reordered_input_picture[0] = s->input_picture[0];
1561 s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_I;
1562 s->reordered_input_picture[0]->f->coded_picture_number =
1563 s->coded_picture_number++;
1567 if (s->avctx->flags & AV_CODEC_FLAG_PASS2) {
1568 for (i = 0; i < s->max_b_frames + 1; i++) {
1569 int pict_num = s->input_picture[0]->f->display_picture_number + i;
1571 if (pict_num >= s->rc_context.num_entries)
1573 if (!s->input_picture[i]) {
1574 s->rc_context.entry[pict_num - 1].new_pict_type = AV_PICTURE_TYPE_P;
1578 s->input_picture[i]->f->pict_type =
1579 s->rc_context.entry[pict_num].new_pict_type;
1583 if (s->b_frame_strategy == 0) {
1584 b_frames = s->max_b_frames;
1585 while (b_frames && !s->input_picture[b_frames])
1587 } else if (s->b_frame_strategy == 1) {
1588 for (i = 1; i < s->max_b_frames + 1; i++) {
1589 if (s->input_picture[i] &&
1590 s->input_picture[i]->b_frame_score == 0) {
1591 s->input_picture[i]->b_frame_score =
1593 s->input_picture[i ]->f->data[0],
1594 s->input_picture[i - 1]->f->data[0],
1598 for (i = 0; i < s->max_b_frames + 1; i++) {
1599 if (!s->input_picture[i] ||
1600 s->input_picture[i]->b_frame_score - 1 >
1601 s->mb_num / s->b_sensitivity)
1605 b_frames = FFMAX(0, i - 1);
1608 for (i = 0; i < b_frames + 1; i++) {
1609 s->input_picture[i]->b_frame_score = 0;
1611 } else if (s->b_frame_strategy == 2) {
1612 b_frames = estimate_best_b_count(s);
1619 for (i = b_frames - 1; i >= 0; i--) {
1620 int type = s->input_picture[i]->f->pict_type;
1621 if (type && type != AV_PICTURE_TYPE_B)
1624 if (s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_B &&
1625 b_frames == s->max_b_frames) {
1626 av_log(s->avctx, AV_LOG_ERROR,
1627 "warning, too many B-frames in a row\n");
1630 if (s->picture_in_gop_number + b_frames >= s->gop_size) {
1631 if ((s->mpv_flags & FF_MPV_FLAG_STRICT_GOP) &&
1632 s->gop_size > s->picture_in_gop_number) {
1633 b_frames = s->gop_size - s->picture_in_gop_number - 1;
1635 if (s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)
1637 s->input_picture[b_frames]->f->pict_type = AV_PICTURE_TYPE_I;
1641 if ((s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) && b_frames &&
1642 s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_I)
1645 s->reordered_input_picture[0] = s->input_picture[b_frames];
1646 if (s->reordered_input_picture[0]->f->pict_type != AV_PICTURE_TYPE_I)
1647 s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_P;
1648 s->reordered_input_picture[0]->f->coded_picture_number =
1649 s->coded_picture_number++;
1650 for (i = 0; i < b_frames; i++) {
1651 s->reordered_input_picture[i + 1] = s->input_picture[i];
1652 s->reordered_input_picture[i + 1]->f->pict_type =
1654 s->reordered_input_picture[i + 1]->f->coded_picture_number =
1655 s->coded_picture_number++;
1660 ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1662 if (s->reordered_input_picture[0]) {
1663 s->reordered_input_picture[0]->reference =
1664 s->reordered_input_picture[0]->f->pict_type !=
1665 AV_PICTURE_TYPE_B ? 3 : 0;
1667 if ((ret = ff_mpeg_ref_picture(s->avctx, &s->new_picture, s->reordered_input_picture[0])))
1670 if (s->reordered_input_picture[0]->shared || s->avctx->rc_buffer_size) {
1671 // input is a shared pix, so we can't modify it -> allocate a new
1672 // one & ensure that the shared one is reuseable
1675 int i = ff_find_unused_picture(s->avctx, s->picture, 0);
1678 pic = &s->picture[i];
1680 pic->reference = s->reordered_input_picture[0]->reference;
1681 if (alloc_picture(s, pic, 0) < 0) {
1685 ret = av_frame_copy_props(pic->f, s->reordered_input_picture[0]->f);
1689 /* mark us unused / free shared pic */
1690 av_frame_unref(s->reordered_input_picture[0]->f);
1691 s->reordered_input_picture[0]->shared = 0;
1693 s->current_picture_ptr = pic;
1695 // input is not a shared pix -> reuse buffer for current_pix
1696 s->current_picture_ptr = s->reordered_input_picture[0];
1697 for (i = 0; i < 4; i++) {
1698 if (s->new_picture.f->data[i])
1699 s->new_picture.f->data[i] += INPLACE_OFFSET;
1702 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1703 if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1704 s->current_picture_ptr)) < 0)
1707 s->picture_number = s->new_picture.f->display_picture_number;
1712 static void frame_end(MpegEncContext *s)
1714 if (s->unrestricted_mv &&
1715 s->current_picture.reference &&
1717 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
1718 int hshift = desc->log2_chroma_w;
1719 int vshift = desc->log2_chroma_h;
1720 s->mpvencdsp.draw_edges(s->current_picture.f->data[0],
1721 s->current_picture.f->linesize[0],
1722 s->h_edge_pos, s->v_edge_pos,
1723 EDGE_WIDTH, EDGE_WIDTH,
1724 EDGE_TOP | EDGE_BOTTOM);
1725 s->mpvencdsp.draw_edges(s->current_picture.f->data[1],
1726 s->current_picture.f->linesize[1],
1727 s->h_edge_pos >> hshift,
1728 s->v_edge_pos >> vshift,
1729 EDGE_WIDTH >> hshift,
1730 EDGE_WIDTH >> vshift,
1731 EDGE_TOP | EDGE_BOTTOM);
1732 s->mpvencdsp.draw_edges(s->current_picture.f->data[2],
1733 s->current_picture.f->linesize[2],
1734 s->h_edge_pos >> hshift,
1735 s->v_edge_pos >> vshift,
1736 EDGE_WIDTH >> hshift,
1737 EDGE_WIDTH >> vshift,
1738 EDGE_TOP | EDGE_BOTTOM);
1743 s->last_pict_type = s->pict_type;
1744 s->last_lambda_for [s->pict_type] = s->current_picture_ptr->f->quality;
1745 if (s->pict_type!= AV_PICTURE_TYPE_B)
1746 s->last_non_b_pict_type = s->pict_type;
1748 #if FF_API_CODED_FRAME
1749 FF_DISABLE_DEPRECATION_WARNINGS
1750 av_frame_unref(s->avctx->coded_frame);
1751 av_frame_copy_props(s->avctx->coded_frame, s->current_picture.f);
1752 FF_ENABLE_DEPRECATION_WARNINGS
1754 #if FF_API_ERROR_FRAME
1755 FF_DISABLE_DEPRECATION_WARNINGS
1756 memcpy(s->current_picture.f->error, s->current_picture.encoding_error,
1757 sizeof(s->current_picture.encoding_error));
1758 FF_ENABLE_DEPRECATION_WARNINGS
1762 static void update_noise_reduction(MpegEncContext *s)
1766 for (intra = 0; intra < 2; intra++) {
1767 if (s->dct_count[intra] > (1 << 16)) {
1768 for (i = 0; i < 64; i++) {
1769 s->dct_error_sum[intra][i] >>= 1;
1771 s->dct_count[intra] >>= 1;
1774 for (i = 0; i < 64; i++) {
1775 s->dct_offset[intra][i] = (s->noise_reduction *
1776 s->dct_count[intra] +
1777 s->dct_error_sum[intra][i] / 2) /
1778 (s->dct_error_sum[intra][i] + 1);
1783 static int frame_start(MpegEncContext *s)
1787 /* mark & release old frames */
1788 if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1789 s->last_picture_ptr != s->next_picture_ptr &&
1790 s->last_picture_ptr->f->buf[0]) {
1791 ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1794 s->current_picture_ptr->f->pict_type = s->pict_type;
1795 s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1797 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1798 if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1799 s->current_picture_ptr)) < 0)
1802 if (s->pict_type != AV_PICTURE_TYPE_B) {
1803 s->last_picture_ptr = s->next_picture_ptr;
1805 s->next_picture_ptr = s->current_picture_ptr;
1808 if (s->last_picture_ptr) {
1809 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1810 if (s->last_picture_ptr->f->buf[0] &&
1811 (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1812 s->last_picture_ptr)) < 0)
1815 if (s->next_picture_ptr) {
1816 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1817 if (s->next_picture_ptr->f->buf[0] &&
1818 (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1819 s->next_picture_ptr)) < 0)
1823 if (s->picture_structure!= PICT_FRAME) {
1825 for (i = 0; i < 4; i++) {
1826 if (s->picture_structure == PICT_BOTTOM_FIELD) {
1827 s->current_picture.f->data[i] +=
1828 s->current_picture.f->linesize[i];
1830 s->current_picture.f->linesize[i] *= 2;
1831 s->last_picture.f->linesize[i] *= 2;
1832 s->next_picture.f->linesize[i] *= 2;
1836 if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1837 s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1838 s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1839 } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1840 s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1841 s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1843 s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1844 s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1847 if (s->dct_error_sum) {
1848 av_assert2(s->noise_reduction && s->encoding);
1849 update_noise_reduction(s);
1855 int ff_mpv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
1856 const AVFrame *pic_arg, int *got_packet)
1858 MpegEncContext *s = avctx->priv_data;
1859 int i, stuffing_count, ret;
1860 int context_count = s->slice_context_count;
1862 s->vbv_ignore_qmax = 0;
1864 s->picture_in_gop_number++;
1866 if (load_input_picture(s, pic_arg) < 0)
1869 if (select_input_picture(s) < 0) {
1874 if (s->new_picture.f->data[0]) {
1875 int growing_buffer = context_count == 1 && !pkt->data && !s->data_partitioning;
1876 int pkt_size = growing_buffer ? FFMAX(s->mb_width*s->mb_height*64+10000, avctx->internal->byte_buffer_size) - AV_INPUT_BUFFER_PADDING_SIZE
1878 s->mb_width*s->mb_height*(MAX_MB_BYTES+100)+10000;
1879 if ((ret = ff_alloc_packet2(avctx, pkt, pkt_size, 0)) < 0)
1882 s->mb_info_ptr = av_packet_new_side_data(pkt,
1883 AV_PKT_DATA_H263_MB_INFO,
1884 s->mb_width*s->mb_height*12);
1885 s->prev_mb_info = s->last_mb_info = s->mb_info_size = 0;
1888 for (i = 0; i < context_count; i++) {
1889 int start_y = s->thread_context[i]->start_mb_y;
1890 int end_y = s->thread_context[i]-> end_mb_y;
1891 int h = s->mb_height;
1892 uint8_t *start = pkt->data + (size_t)(((int64_t) pkt->size) * start_y / h);
1893 uint8_t *end = pkt->data + (size_t)(((int64_t) pkt->size) * end_y / h);
1895 init_put_bits(&s->thread_context[i]->pb, start, end - start);
1898 s->pict_type = s->new_picture.f->pict_type;
1900 ret = frame_start(s);
1904 ret = encode_picture(s, s->picture_number);
1905 if (growing_buffer) {
1906 av_assert0(s->pb.buf == avctx->internal->byte_buffer);
1907 pkt->data = s->pb.buf;
1908 pkt->size = avctx->internal->byte_buffer_size;
1913 #if FF_API_STAT_BITS
1914 FF_DISABLE_DEPRECATION_WARNINGS
1915 avctx->header_bits = s->header_bits;
1916 avctx->mv_bits = s->mv_bits;
1917 avctx->misc_bits = s->misc_bits;
1918 avctx->i_tex_bits = s->i_tex_bits;
1919 avctx->p_tex_bits = s->p_tex_bits;
1920 avctx->i_count = s->i_count;
1921 // FIXME f/b_count in avctx
1922 avctx->p_count = s->mb_num - s->i_count - s->skip_count;
1923 avctx->skip_count = s->skip_count;
1924 FF_ENABLE_DEPRECATION_WARNINGS
1929 if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1930 ff_mjpeg_encode_picture_trailer(&s->pb, s->header_bits);
1932 if (avctx->rc_buffer_size) {
1933 RateControlContext *rcc = &s->rc_context;
1934 int max_size = FFMAX(rcc->buffer_index * avctx->rc_max_available_vbv_use, rcc->buffer_index - 500);
1935 int hq = (avctx->mb_decision == FF_MB_DECISION_RD || avctx->trellis);
1936 int min_step = hq ? 1 : (1<<(FF_LAMBDA_SHIFT + 7))/139;
1938 if (put_bits_count(&s->pb) > max_size &&
1939 s->lambda < s->lmax) {
1940 s->next_lambda = FFMAX(s->lambda + min_step, s->lambda *
1941 (s->qscale + 1) / s->qscale);
1942 if (s->adaptive_quant) {
1944 for (i = 0; i < s->mb_height * s->mb_stride; i++)
1945 s->lambda_table[i] =
1946 FFMAX(s->lambda_table[i] + min_step,
1947 s->lambda_table[i] * (s->qscale + 1) /
1950 s->mb_skipped = 0; // done in frame_start()
1951 // done in encode_picture() so we must undo it
1952 if (s->pict_type == AV_PICTURE_TYPE_P) {
1953 if (s->flipflop_rounding ||
1954 s->codec_id == AV_CODEC_ID_H263P ||
1955 s->codec_id == AV_CODEC_ID_MPEG4)
1956 s->no_rounding ^= 1;
1958 if (s->pict_type != AV_PICTURE_TYPE_B) {
1959 s->time_base = s->last_time_base;
1960 s->last_non_b_time = s->time - s->pp_time;
1962 for (i = 0; i < context_count; i++) {
1963 PutBitContext *pb = &s->thread_context[i]->pb;
1964 init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1966 s->vbv_ignore_qmax = 1;
1967 av_log(avctx, AV_LOG_VERBOSE, "reencoding frame due to VBV\n");
1971 av_assert0(avctx->rc_max_rate);
1974 if (avctx->flags & AV_CODEC_FLAG_PASS1)
1975 ff_write_pass1_stats(s);
1977 for (i = 0; i < 4; i++) {
1978 s->current_picture_ptr->encoding_error[i] = s->current_picture.encoding_error[i];
1979 avctx->error[i] += s->current_picture_ptr->encoding_error[i];
1981 ff_side_data_set_encoder_stats(pkt, s->current_picture.f->quality,
1982 s->current_picture_ptr->encoding_error,
1983 (avctx->flags&AV_CODEC_FLAG_PSNR) ? 4 : 0,
1986 if (avctx->flags & AV_CODEC_FLAG_PASS1)
1987 assert(put_bits_count(&s->pb) == s->header_bits + s->mv_bits +
1988 s->misc_bits + s->i_tex_bits +
1990 flush_put_bits(&s->pb);
1991 s->frame_bits = put_bits_count(&s->pb);
1993 stuffing_count = ff_vbv_update(s, s->frame_bits);
1994 s->stuffing_bits = 8*stuffing_count;
1995 if (stuffing_count) {
1996 if (put_bytes_left(&s->pb, 0) < stuffing_count + 50) {
1997 av_log(avctx, AV_LOG_ERROR, "stuffing too large\n");
2001 switch (s->codec_id) {
2002 case AV_CODEC_ID_MPEG1VIDEO:
2003 case AV_CODEC_ID_MPEG2VIDEO:
2004 while (stuffing_count--) {
2005 put_bits(&s->pb, 8, 0);
2008 case AV_CODEC_ID_MPEG4:
2009 put_bits(&s->pb, 16, 0);
2010 put_bits(&s->pb, 16, 0x1C3);
2011 stuffing_count -= 4;
2012 while (stuffing_count--) {
2013 put_bits(&s->pb, 8, 0xFF);
2017 av_log(avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
2019 flush_put_bits(&s->pb);
2020 s->frame_bits = put_bits_count(&s->pb);
2023 /* update MPEG-1/2 vbv_delay for CBR */
2024 if (avctx->rc_max_rate &&
2025 avctx->rc_min_rate == avctx->rc_max_rate &&
2026 s->out_format == FMT_MPEG1 &&
2027 90000LL * (avctx->rc_buffer_size - 1) <=
2028 avctx->rc_max_rate * 0xFFFFLL) {
2029 AVCPBProperties *props;
2032 int vbv_delay, min_delay;
2033 double inbits = avctx->rc_max_rate *
2034 av_q2d(avctx->time_base);
2035 int minbits = s->frame_bits - 8 *
2036 (s->vbv_delay_ptr - s->pb.buf - 1);
2037 double bits = s->rc_context.buffer_index + minbits - inbits;
2040 av_log(avctx, AV_LOG_ERROR,
2041 "Internal error, negative bits\n");
2043 av_assert1(s->repeat_first_field == 0);
2045 vbv_delay = bits * 90000 / avctx->rc_max_rate;
2046 min_delay = (minbits * 90000LL + avctx->rc_max_rate - 1) /
2049 vbv_delay = FFMAX(vbv_delay, min_delay);
2051 av_assert0(vbv_delay < 0xFFFF);
2053 s->vbv_delay_ptr[0] &= 0xF8;
2054 s->vbv_delay_ptr[0] |= vbv_delay >> 13;
2055 s->vbv_delay_ptr[1] = vbv_delay >> 5;
2056 s->vbv_delay_ptr[2] &= 0x07;
2057 s->vbv_delay_ptr[2] |= vbv_delay << 3;
2059 props = av_cpb_properties_alloc(&props_size);
2061 return AVERROR(ENOMEM);
2062 props->vbv_delay = vbv_delay * 300;
2064 ret = av_packet_add_side_data(pkt, AV_PKT_DATA_CPB_PROPERTIES,
2065 (uint8_t*)props, props_size);
2071 #if FF_API_VBV_DELAY
2072 FF_DISABLE_DEPRECATION_WARNINGS
2073 avctx->vbv_delay = vbv_delay * 300;
2074 FF_ENABLE_DEPRECATION_WARNINGS
2077 s->total_bits += s->frame_bits;
2078 #if FF_API_STAT_BITS
2079 FF_DISABLE_DEPRECATION_WARNINGS
2080 avctx->frame_bits = s->frame_bits;
2081 FF_ENABLE_DEPRECATION_WARNINGS
2085 pkt->pts = s->current_picture.f->pts;
2086 if (!s->low_delay && s->pict_type != AV_PICTURE_TYPE_B) {
2087 if (!s->current_picture.f->coded_picture_number)
2088 pkt->dts = pkt->pts - s->dts_delta;
2090 pkt->dts = s->reordered_pts;
2091 s->reordered_pts = pkt->pts;
2093 pkt->dts = pkt->pts;
2094 if (s->current_picture.f->key_frame)
2095 pkt->flags |= AV_PKT_FLAG_KEY;
2097 av_packet_shrink_side_data(pkt, AV_PKT_DATA_H263_MB_INFO, s->mb_info_size);
2102 /* release non-reference frames */
2103 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
2104 if (!s->picture[i].reference)
2105 ff_mpeg_unref_picture(avctx, &s->picture[i]);
2108 av_assert1((s->frame_bits & 7) == 0);
2110 pkt->size = s->frame_bits / 8;
2111 *got_packet = !!pkt->size;
2115 static inline void dct_single_coeff_elimination(MpegEncContext *s,
2116 int n, int threshold)
2118 static const char tab[64] = {
2119 3, 2, 2, 1, 1, 1, 1, 1,
2120 1, 1, 1, 1, 1, 1, 1, 1,
2121 1, 1, 1, 1, 1, 1, 1, 1,
2122 0, 0, 0, 0, 0, 0, 0, 0,
2123 0, 0, 0, 0, 0, 0, 0, 0,
2124 0, 0, 0, 0, 0, 0, 0, 0,
2125 0, 0, 0, 0, 0, 0, 0, 0,
2126 0, 0, 0, 0, 0, 0, 0, 0
2131 int16_t *block = s->block[n];
2132 const int last_index = s->block_last_index[n];
2135 if (threshold < 0) {
2137 threshold = -threshold;
2141 /* Are all we could set to zero already zero? */
2142 if (last_index <= skip_dc - 1)
2145 for (i = 0; i <= last_index; i++) {
2146 const int j = s->intra_scantable.permutated[i];
2147 const int level = FFABS(block[j]);
2149 if (skip_dc && i == 0)
2153 } else if (level > 1) {
2159 if (score >= threshold)
2161 for (i = skip_dc; i <= last_index; i++) {
2162 const int j = s->intra_scantable.permutated[i];
2166 s->block_last_index[n] = 0;
2168 s->block_last_index[n] = -1;
2171 static inline void clip_coeffs(MpegEncContext *s, int16_t *block,
2175 const int maxlevel = s->max_qcoeff;
2176 const int minlevel = s->min_qcoeff;
2180 i = 1; // skip clipping of intra dc
2184 for (; i <= last_index; i++) {
2185 const int j = s->intra_scantable.permutated[i];
2186 int level = block[j];
2188 if (level > maxlevel) {
2191 } else if (level < minlevel) {
2199 if (overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
2200 av_log(s->avctx, AV_LOG_INFO,
2201 "warning, clipping %d dct coefficients to %d..%d\n",
2202 overflow, minlevel, maxlevel);
2205 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride)
2209 for (y = 0; y < 8; y++) {
2210 for (x = 0; x < 8; x++) {
2216 for (y2 = FFMAX(y - 1, 0); y2 < FFMIN(8, y + 2); y2++) {
2217 for (x2= FFMAX(x - 1, 0); x2 < FFMIN(8, x + 2); x2++) {
2218 int v = ptr[x2 + y2 * stride];
2224 weight[x + 8 * y]= (36 * ff_sqrt(count * sqr - sum * sum)) / count;
2229 static av_always_inline void encode_mb_internal(MpegEncContext *s,
2230 int motion_x, int motion_y,
2231 int mb_block_height,
2235 int16_t weight[12][64];
2236 int16_t orig[12][64];
2237 const int mb_x = s->mb_x;
2238 const int mb_y = s->mb_y;
2241 int dct_offset = s->linesize * 8; // default for progressive frames
2242 int uv_dct_offset = s->uvlinesize * 8;
2243 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2244 ptrdiff_t wrap_y, wrap_c;
2246 for (i = 0; i < mb_block_count; i++)
2247 skip_dct[i] = s->skipdct;
2249 if (s->adaptive_quant) {
2250 const int last_qp = s->qscale;
2251 const int mb_xy = mb_x + mb_y * s->mb_stride;
2253 s->lambda = s->lambda_table[mb_xy];
2256 if (!(s->mpv_flags & FF_MPV_FLAG_QP_RD)) {
2257 s->qscale = s->current_picture_ptr->qscale_table[mb_xy];
2258 s->dquant = s->qscale - last_qp;
2260 if (s->out_format == FMT_H263) {
2261 s->dquant = av_clip(s->dquant, -2, 2);
2263 if (s->codec_id == AV_CODEC_ID_MPEG4) {
2265 if (s->pict_type == AV_PICTURE_TYPE_B) {
2266 if (s->dquant & 1 || s->mv_dir & MV_DIRECT)
2269 if (s->mv_type == MV_TYPE_8X8)
2275 ff_set_qscale(s, last_qp + s->dquant);
2276 } else if (s->mpv_flags & FF_MPV_FLAG_QP_RD)
2277 ff_set_qscale(s, s->qscale + s->dquant);
2279 wrap_y = s->linesize;
2280 wrap_c = s->uvlinesize;
2281 ptr_y = s->new_picture.f->data[0] +
2282 (mb_y * 16 * wrap_y) + mb_x * 16;
2283 ptr_cb = s->new_picture.f->data[1] +
2284 (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
2285 ptr_cr = s->new_picture.f->data[2] +
2286 (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
2288 if((mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) && s->codec_id != AV_CODEC_ID_AMV){
2289 uint8_t *ebuf = s->sc.edge_emu_buffer + 38 * wrap_y;
2290 int cw = (s->width + s->chroma_x_shift) >> s->chroma_x_shift;
2291 int ch = (s->height + s->chroma_y_shift) >> s->chroma_y_shift;
2292 s->vdsp.emulated_edge_mc(ebuf, ptr_y,
2294 16, 16, mb_x * 16, mb_y * 16,
2295 s->width, s->height);
2297 s->vdsp.emulated_edge_mc(ebuf + 16 * wrap_y, ptr_cb,
2299 mb_block_width, mb_block_height,
2300 mb_x * mb_block_width, mb_y * mb_block_height,
2302 ptr_cb = ebuf + 16 * wrap_y;
2303 s->vdsp.emulated_edge_mc(ebuf + 16 * wrap_y + 16, ptr_cr,
2305 mb_block_width, mb_block_height,
2306 mb_x * mb_block_width, mb_y * mb_block_height,
2308 ptr_cr = ebuf + 16 * wrap_y + 16;
2312 if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
2313 int progressive_score, interlaced_score;
2315 s->interlaced_dct = 0;
2316 progressive_score = s->mecc.ildct_cmp[4](s, ptr_y, NULL, wrap_y, 8) +
2317 s->mecc.ildct_cmp[4](s, ptr_y + wrap_y * 8,
2318 NULL, wrap_y, 8) - 400;
2320 if (progressive_score > 0) {
2321 interlaced_score = s->mecc.ildct_cmp[4](s, ptr_y,
2322 NULL, wrap_y * 2, 8) +
2323 s->mecc.ildct_cmp[4](s, ptr_y + wrap_y,
2324 NULL, wrap_y * 2, 8);
2325 if (progressive_score > interlaced_score) {
2326 s->interlaced_dct = 1;
2328 dct_offset = wrap_y;
2329 uv_dct_offset = wrap_c;
2331 if (s->chroma_format == CHROMA_422 ||
2332 s->chroma_format == CHROMA_444)
2338 s->pdsp.get_pixels(s->block[0], ptr_y, wrap_y);
2339 s->pdsp.get_pixels(s->block[1], ptr_y + 8, wrap_y);
2340 s->pdsp.get_pixels(s->block[2], ptr_y + dct_offset, wrap_y);
2341 s->pdsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
2343 if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
2347 s->pdsp.get_pixels(s->block[4], ptr_cb, wrap_c);
2348 s->pdsp.get_pixels(s->block[5], ptr_cr, wrap_c);
2349 if (!s->chroma_y_shift && s->chroma_x_shift) { /* 422 */
2350 s->pdsp.get_pixels(s->block[6], ptr_cb + uv_dct_offset, wrap_c);
2351 s->pdsp.get_pixels(s->block[7], ptr_cr + uv_dct_offset, wrap_c);
2352 } else if (!s->chroma_y_shift && !s->chroma_x_shift) { /* 444 */
2353 s->pdsp.get_pixels(s->block[ 6], ptr_cb + 8, wrap_c);
2354 s->pdsp.get_pixels(s->block[ 7], ptr_cr + 8, wrap_c);
2355 s->pdsp.get_pixels(s->block[ 8], ptr_cb + uv_dct_offset, wrap_c);
2356 s->pdsp.get_pixels(s->block[ 9], ptr_cr + uv_dct_offset, wrap_c);
2357 s->pdsp.get_pixels(s->block[10], ptr_cb + uv_dct_offset + 8, wrap_c);
2358 s->pdsp.get_pixels(s->block[11], ptr_cr + uv_dct_offset + 8, wrap_c);
2362 op_pixels_func (*op_pix)[4];
2363 qpel_mc_func (*op_qpix)[16];
2364 uint8_t *dest_y, *dest_cb, *dest_cr;
2366 dest_y = s->dest[0];
2367 dest_cb = s->dest[1];
2368 dest_cr = s->dest[2];
2370 if ((!s->no_rounding) || s->pict_type == AV_PICTURE_TYPE_B) {
2371 op_pix = s->hdsp.put_pixels_tab;
2372 op_qpix = s->qdsp.put_qpel_pixels_tab;
2374 op_pix = s->hdsp.put_no_rnd_pixels_tab;
2375 op_qpix = s->qdsp.put_no_rnd_qpel_pixels_tab;
2378 if (s->mv_dir & MV_DIR_FORWARD) {
2379 ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0,
2380 s->last_picture.f->data,
2382 op_pix = s->hdsp.avg_pixels_tab;
2383 op_qpix = s->qdsp.avg_qpel_pixels_tab;
2385 if (s->mv_dir & MV_DIR_BACKWARD) {
2386 ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1,
2387 s->next_picture.f->data,
2391 if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
2392 int progressive_score, interlaced_score;
2394 s->interlaced_dct = 0;
2395 progressive_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y, wrap_y, 8) +
2396 s->mecc.ildct_cmp[0](s, dest_y + wrap_y * 8,
2400 if (s->avctx->ildct_cmp == FF_CMP_VSSE)
2401 progressive_score -= 400;
2403 if (progressive_score > 0) {
2404 interlaced_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y,
2406 s->mecc.ildct_cmp[0](s, dest_y + wrap_y,
2410 if (progressive_score > interlaced_score) {
2411 s->interlaced_dct = 1;
2413 dct_offset = wrap_y;
2414 uv_dct_offset = wrap_c;
2416 if (s->chroma_format == CHROMA_422)
2422 s->pdsp.diff_pixels(s->block[0], ptr_y, dest_y, wrap_y);
2423 s->pdsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
2424 s->pdsp.diff_pixels(s->block[2], ptr_y + dct_offset,
2425 dest_y + dct_offset, wrap_y);
2426 s->pdsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8,
2427 dest_y + dct_offset + 8, wrap_y);
2429 if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
2433 s->pdsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
2434 s->pdsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
2435 if (!s->chroma_y_shift) { /* 422 */
2436 s->pdsp.diff_pixels(s->block[6], ptr_cb + uv_dct_offset,
2437 dest_cb + uv_dct_offset, wrap_c);
2438 s->pdsp.diff_pixels(s->block[7], ptr_cr + uv_dct_offset,
2439 dest_cr + uv_dct_offset, wrap_c);
2442 /* pre quantization */
2443 if (s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] <
2444 2 * s->qscale * s->qscale) {
2446 if (s->mecc.sad[1](NULL, ptr_y, dest_y, wrap_y, 8) < 20 * s->qscale)
2448 if (s->mecc.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20 * s->qscale)
2450 if (s->mecc.sad[1](NULL, ptr_y + dct_offset, dest_y + dct_offset,
2451 wrap_y, 8) < 20 * s->qscale)
2453 if (s->mecc.sad[1](NULL, ptr_y + dct_offset + 8, dest_y + dct_offset + 8,
2454 wrap_y, 8) < 20 * s->qscale)
2456 if (s->mecc.sad[1](NULL, ptr_cb, dest_cb, wrap_c, 8) < 20 * s->qscale)
2458 if (s->mecc.sad[1](NULL, ptr_cr, dest_cr, wrap_c, 8) < 20 * s->qscale)
2460 if (!s->chroma_y_shift) { /* 422 */
2461 if (s->mecc.sad[1](NULL, ptr_cb + uv_dct_offset,
2462 dest_cb + uv_dct_offset,
2463 wrap_c, 8) < 20 * s->qscale)
2465 if (s->mecc.sad[1](NULL, ptr_cr + uv_dct_offset,
2466 dest_cr + uv_dct_offset,
2467 wrap_c, 8) < 20 * s->qscale)
2473 if (s->quantizer_noise_shaping) {
2475 get_visual_weight(weight[0], ptr_y , wrap_y);
2477 get_visual_weight(weight[1], ptr_y + 8, wrap_y);
2479 get_visual_weight(weight[2], ptr_y + dct_offset , wrap_y);
2481 get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
2483 get_visual_weight(weight[4], ptr_cb , wrap_c);
2485 get_visual_weight(weight[5], ptr_cr , wrap_c);
2486 if (!s->chroma_y_shift) { /* 422 */
2488 get_visual_weight(weight[6], ptr_cb + uv_dct_offset,
2491 get_visual_weight(weight[7], ptr_cr + uv_dct_offset,
2494 memcpy(orig[0], s->block[0], sizeof(int16_t) * 64 * mb_block_count);
2497 /* DCT & quantize */
2498 av_assert2(s->out_format != FMT_MJPEG || s->qscale == 8);
2500 for (i = 0; i < mb_block_count; i++) {
2503 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
2504 // FIXME we could decide to change to quantizer instead of
2506 // JS: I don't think that would be a good idea it could lower
2507 // quality instead of improve it. Just INTRADC clipping
2508 // deserves changes in quantizer
2510 clip_coeffs(s, s->block[i], s->block_last_index[i]);
2512 s->block_last_index[i] = -1;
2514 if (s->quantizer_noise_shaping) {
2515 for (i = 0; i < mb_block_count; i++) {
2517 s->block_last_index[i] =
2518 dct_quantize_refine(s, s->block[i], weight[i],
2519 orig[i], i, s->qscale);
2524 if (s->luma_elim_threshold && !s->mb_intra)
2525 for (i = 0; i < 4; i++)
2526 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
2527 if (s->chroma_elim_threshold && !s->mb_intra)
2528 for (i = 4; i < mb_block_count; i++)
2529 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
2531 if (s->mpv_flags & FF_MPV_FLAG_CBP_RD) {
2532 for (i = 0; i < mb_block_count; i++) {
2533 if (s->block_last_index[i] == -1)
2534 s->coded_score[i] = INT_MAX / 256;
2539 if ((s->avctx->flags & AV_CODEC_FLAG_GRAY) && s->mb_intra) {
2540 s->block_last_index[4] =
2541 s->block_last_index[5] = 0;
2543 s->block[5][0] = (1024 + s->c_dc_scale / 2) / s->c_dc_scale;
2544 if (!s->chroma_y_shift) { /* 422 / 444 */
2545 for (i=6; i<12; i++) {
2546 s->block_last_index[i] = 0;
2547 s->block[i][0] = s->block[4][0];
2552 // non c quantize code returns incorrect block_last_index FIXME
2553 if (s->alternate_scan && s->dct_quantize != ff_dct_quantize_c) {
2554 for (i = 0; i < mb_block_count; i++) {
2556 if (s->block_last_index[i] > 0) {
2557 for (j = 63; j > 0; j--) {
2558 if (s->block[i][s->intra_scantable.permutated[j]])
2561 s->block_last_index[i] = j;
2566 /* huffman encode */
2567 switch(s->codec_id){ //FIXME funct ptr could be slightly faster
2568 case AV_CODEC_ID_MPEG1VIDEO:
2569 case AV_CODEC_ID_MPEG2VIDEO:
2570 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2571 ff_mpeg1_encode_mb(s, s->block, motion_x, motion_y);
2573 case AV_CODEC_ID_MPEG4:
2574 if (CONFIG_MPEG4_ENCODER)
2575 ff_mpeg4_encode_mb(s, s->block, motion_x, motion_y);
2577 case AV_CODEC_ID_MSMPEG4V2:
2578 case AV_CODEC_ID_MSMPEG4V3:
2579 case AV_CODEC_ID_WMV1:
2580 if (CONFIG_MSMPEG4_ENCODER)
2581 ff_msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
2583 case AV_CODEC_ID_WMV2:
2584 if (CONFIG_WMV2_ENCODER)
2585 ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
2587 case AV_CODEC_ID_H261:
2588 if (CONFIG_H261_ENCODER)
2589 ff_h261_encode_mb(s, s->block, motion_x, motion_y);
2591 case AV_CODEC_ID_H263:
2592 case AV_CODEC_ID_H263P:
2593 case AV_CODEC_ID_FLV1:
2594 case AV_CODEC_ID_RV10:
2595 case AV_CODEC_ID_RV20:
2596 if (CONFIG_H263_ENCODER)
2597 ff_h263_encode_mb(s, s->block, motion_x, motion_y);
2599 case AV_CODEC_ID_MJPEG:
2600 case AV_CODEC_ID_AMV:
2601 if (CONFIG_MJPEG_ENCODER)
2602 ff_mjpeg_encode_mb(s, s->block);
2604 case AV_CODEC_ID_SPEEDHQ:
2605 if (CONFIG_SPEEDHQ_ENCODER)
2606 ff_speedhq_encode_mb(s, s->block);
2613 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
2615 if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 8, 6);
2616 else if (s->chroma_format == CHROMA_422) encode_mb_internal(s, motion_x, motion_y, 16, 8, 8);
2617 else encode_mb_internal(s, motion_x, motion_y, 16, 16, 12);
2620 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
2623 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2626 d->mb_skip_run= s->mb_skip_run;
2628 d->last_dc[i] = s->last_dc[i];
2631 d->mv_bits= s->mv_bits;
2632 d->i_tex_bits= s->i_tex_bits;
2633 d->p_tex_bits= s->p_tex_bits;
2634 d->i_count= s->i_count;
2635 d->f_count= s->f_count;
2636 d->b_count= s->b_count;
2637 d->skip_count= s->skip_count;
2638 d->misc_bits= s->misc_bits;
2642 d->qscale= s->qscale;
2643 d->dquant= s->dquant;
2645 d->esc3_level_length= s->esc3_level_length;
2648 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
2651 memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
2652 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2655 d->mb_skip_run= s->mb_skip_run;
2657 d->last_dc[i] = s->last_dc[i];
2660 d->mv_bits= s->mv_bits;
2661 d->i_tex_bits= s->i_tex_bits;
2662 d->p_tex_bits= s->p_tex_bits;
2663 d->i_count= s->i_count;
2664 d->f_count= s->f_count;
2665 d->b_count= s->b_count;
2666 d->skip_count= s->skip_count;
2667 d->misc_bits= s->misc_bits;
2669 d->mb_intra= s->mb_intra;
2670 d->mb_skipped= s->mb_skipped;
2671 d->mv_type= s->mv_type;
2672 d->mv_dir= s->mv_dir;
2674 if(s->data_partitioning){
2676 d->tex_pb= s->tex_pb;
2680 d->block_last_index[i]= s->block_last_index[i];
2681 d->interlaced_dct= s->interlaced_dct;
2682 d->qscale= s->qscale;
2684 d->esc3_level_length= s->esc3_level_length;
2687 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
2688 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
2689 int *dmin, int *next_block, int motion_x, int motion_y)
2692 uint8_t *dest_backup[3];
2694 copy_context_before_encode(s, backup, type);
2696 s->block= s->blocks[*next_block];
2697 s->pb= pb[*next_block];
2698 if(s->data_partitioning){
2699 s->pb2 = pb2 [*next_block];
2700 s->tex_pb= tex_pb[*next_block];
2704 memcpy(dest_backup, s->dest, sizeof(s->dest));
2705 s->dest[0] = s->sc.rd_scratchpad;
2706 s->dest[1] = s->sc.rd_scratchpad + 16*s->linesize;
2707 s->dest[2] = s->sc.rd_scratchpad + 16*s->linesize + 8;
2708 av_assert0(s->linesize >= 32); //FIXME
2711 encode_mb(s, motion_x, motion_y);
2713 score= put_bits_count(&s->pb);
2714 if(s->data_partitioning){
2715 score+= put_bits_count(&s->pb2);
2716 score+= put_bits_count(&s->tex_pb);
2719 if(s->avctx->mb_decision == FF_MB_DECISION_RD){
2720 ff_mpv_reconstruct_mb(s, s->block);
2722 score *= s->lambda2;
2723 score += sse_mb(s) << FF_LAMBDA_SHIFT;
2727 memcpy(s->dest, dest_backup, sizeof(s->dest));
2734 copy_context_after_encode(best, s, type);
2738 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
2739 const uint32_t *sq = ff_square_tab + 256;
2744 return s->mecc.sse[0](NULL, src1, src2, stride, 16);
2745 else if(w==8 && h==8)
2746 return s->mecc.sse[1](NULL, src1, src2, stride, 8);
2750 acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
2759 static int sse_mb(MpegEncContext *s){
2763 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2764 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2767 if(s->avctx->mb_cmp == FF_CMP_NSSE){
2768 return s->mecc.nsse[0](s, s->new_picture.f->data[0] + s->mb_x * 16 + s->mb_y * s->linesize * 16, s->dest[0], s->linesize, 16) +
2769 s->mecc.nsse[1](s, s->new_picture.f->data[1] + s->mb_x * 8 + s->mb_y * s->uvlinesize * 8, s->dest[1], s->uvlinesize, 8) +
2770 s->mecc.nsse[1](s, s->new_picture.f->data[2] + s->mb_x * 8 + s->mb_y * s->uvlinesize * 8, s->dest[2], s->uvlinesize, 8);
2772 return s->mecc.sse[0](NULL, s->new_picture.f->data[0] + s->mb_x * 16 + s->mb_y * s->linesize * 16, s->dest[0], s->linesize, 16) +
2773 s->mecc.sse[1](NULL, s->new_picture.f->data[1] + s->mb_x * 8 + s->mb_y * s->uvlinesize * 8, s->dest[1], s->uvlinesize, 8) +
2774 s->mecc.sse[1](NULL, s->new_picture.f->data[2] + s->mb_x * 8 + s->mb_y * s->uvlinesize * 8, s->dest[2], s->uvlinesize, 8);
2777 return sse(s, s->new_picture.f->data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
2778 +sse(s, s->new_picture.f->data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
2779 +sse(s, s->new_picture.f->data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
2782 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
2783 MpegEncContext *s= *(void**)arg;
2787 s->me.dia_size= s->avctx->pre_dia_size;
2788 s->first_slice_line=1;
2789 for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
2790 for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
2791 ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2793 s->first_slice_line=0;
2801 static int estimate_motion_thread(AVCodecContext *c, void *arg){
2802 MpegEncContext *s= *(void**)arg;
2804 s->me.dia_size= s->avctx->dia_size;
2805 s->first_slice_line=1;
2806 for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
2807 s->mb_x=0; //for block init below
2808 ff_init_block_index(s);
2809 for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
2810 s->block_index[0]+=2;
2811 s->block_index[1]+=2;
2812 s->block_index[2]+=2;
2813 s->block_index[3]+=2;
2815 /* compute motion vector & mb_type and store in context */
2816 if(s->pict_type==AV_PICTURE_TYPE_B)
2817 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
2819 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2821 s->first_slice_line=0;
2826 static int mb_var_thread(AVCodecContext *c, void *arg){
2827 MpegEncContext *s= *(void**)arg;
2830 for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2831 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2834 uint8_t *pix = s->new_picture.f->data[0] + (yy * s->linesize) + xx;
2836 int sum = s->mpvencdsp.pix_sum(pix, s->linesize);
2838 varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) -
2839 (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8;
2841 s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2842 s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2843 s->me.mb_var_sum_temp += varc;
2849 static void write_slice_end(MpegEncContext *s){
2850 if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4){
2851 if(s->partitioned_frame){
2852 ff_mpeg4_merge_partitions(s);
2855 ff_mpeg4_stuffing(&s->pb);
2856 }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2857 ff_mjpeg_encode_stuffing(s);
2858 } else if (CONFIG_SPEEDHQ_ENCODER && s->out_format == FMT_SPEEDHQ) {
2859 ff_speedhq_end_slice(s);
2862 flush_put_bits(&s->pb);
2864 if ((s->avctx->flags & AV_CODEC_FLAG_PASS1) && !s->partitioned_frame)
2865 s->misc_bits+= get_bits_diff(s);
2868 static void write_mb_info(MpegEncContext *s)
2870 uint8_t *ptr = s->mb_info_ptr + s->mb_info_size - 12;
2871 int offset = put_bits_count(&s->pb);
2872 int mba = s->mb_x + s->mb_width * (s->mb_y % s->gob_index);
2873 int gobn = s->mb_y / s->gob_index;
2875 if (CONFIG_H263_ENCODER)
2876 ff_h263_pred_motion(s, 0, 0, &pred_x, &pred_y);
2877 bytestream_put_le32(&ptr, offset);
2878 bytestream_put_byte(&ptr, s->qscale);
2879 bytestream_put_byte(&ptr, gobn);
2880 bytestream_put_le16(&ptr, mba);
2881 bytestream_put_byte(&ptr, pred_x); /* hmv1 */
2882 bytestream_put_byte(&ptr, pred_y); /* vmv1 */
2883 /* 4MV not implemented */
2884 bytestream_put_byte(&ptr, 0); /* hmv2 */
2885 bytestream_put_byte(&ptr, 0); /* vmv2 */
2888 static void update_mb_info(MpegEncContext *s, int startcode)
2892 if (put_bytes_count(&s->pb, 0) - s->prev_mb_info >= s->mb_info) {
2893 s->mb_info_size += 12;
2894 s->prev_mb_info = s->last_mb_info;
2897 s->prev_mb_info = put_bytes_count(&s->pb, 0);
2898 /* This might have incremented mb_info_size above, and we return without
2899 * actually writing any info into that slot yet. But in that case,
2900 * this will be called again at the start of the after writing the
2901 * start code, actually writing the mb info. */
2905 s->last_mb_info = put_bytes_count(&s->pb, 0);
2906 if (!s->mb_info_size)
2907 s->mb_info_size += 12;
2911 int ff_mpv_reallocate_putbitbuffer(MpegEncContext *s, size_t threshold, size_t size_increase)
2913 if (put_bytes_left(&s->pb, 0) < threshold
2914 && s->slice_context_count == 1
2915 && s->pb.buf == s->avctx->internal->byte_buffer) {
2916 int lastgob_pos = s->ptr_lastgob - s->pb.buf;
2917 int vbv_pos = s->vbv_delay_ptr - s->pb.buf;
2919 uint8_t *new_buffer = NULL;
2920 int new_buffer_size = 0;
2922 if ((s->avctx->internal->byte_buffer_size + size_increase) >= INT_MAX/8) {
2923 av_log(s->avctx, AV_LOG_ERROR, "Cannot reallocate putbit buffer\n");
2924 return AVERROR(ENOMEM);
2929 av_fast_padded_malloc(&new_buffer, &new_buffer_size,
2930 s->avctx->internal->byte_buffer_size + size_increase);
2932 return AVERROR(ENOMEM);
2934 memcpy(new_buffer, s->avctx->internal->byte_buffer, s->avctx->internal->byte_buffer_size);
2935 av_free(s->avctx->internal->byte_buffer);
2936 s->avctx->internal->byte_buffer = new_buffer;
2937 s->avctx->internal->byte_buffer_size = new_buffer_size;
2938 rebase_put_bits(&s->pb, new_buffer, new_buffer_size);
2939 s->ptr_lastgob = s->pb.buf + lastgob_pos;
2940 s->vbv_delay_ptr = s->pb.buf + vbv_pos;
2942 if (put_bytes_left(&s->pb, 0) < threshold)
2943 return AVERROR(EINVAL);
2947 static int encode_thread(AVCodecContext *c, void *arg){
2948 MpegEncContext *s= *(void**)arg;
2949 int mb_x, mb_y, mb_y_order;
2950 int chr_h= 16>>s->chroma_y_shift;
2952 MpegEncContext best_s = { 0 }, backup_s;
2953 uint8_t bit_buf[2][MAX_MB_BYTES];
2954 uint8_t bit_buf2[2][MAX_MB_BYTES];
2955 uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2956 PutBitContext pb[2], pb2[2], tex_pb[2];
2959 init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
2960 init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
2961 init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2964 s->last_bits= put_bits_count(&s->pb);
2975 /* init last dc values */
2976 /* note: quant matrix value (8) is implied here */
2977 s->last_dc[i] = 128 << s->intra_dc_precision;
2979 s->current_picture.encoding_error[i] = 0;
2981 if(s->codec_id==AV_CODEC_ID_AMV){
2982 s->last_dc[0] = 128*8/13;
2983 s->last_dc[1] = 128*8/14;
2984 s->last_dc[2] = 128*8/14;
2987 memset(s->last_mv, 0, sizeof(s->last_mv));
2991 switch(s->codec_id){
2992 case AV_CODEC_ID_H263:
2993 case AV_CODEC_ID_H263P:
2994 case AV_CODEC_ID_FLV1:
2995 if (CONFIG_H263_ENCODER)
2996 s->gob_index = H263_GOB_HEIGHT(s->height);
2998 case AV_CODEC_ID_MPEG4:
2999 if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
3000 ff_mpeg4_init_partitions(s);
3006 s->first_slice_line = 1;
3007 s->ptr_lastgob = s->pb.buf;
3008 for (mb_y_order = s->start_mb_y; mb_y_order < s->end_mb_y; mb_y_order++) {
3009 if (CONFIG_SPEEDHQ_ENCODER && s->codec_id == AV_CODEC_ID_SPEEDHQ) {
3011 mb_y = ff_speedhq_mb_y_order_to_mb(mb_y_order, s->mb_height, &first_in_slice);
3012 if (first_in_slice && mb_y_order != s->start_mb_y)
3013 ff_speedhq_end_slice(s);
3014 s->last_dc[0] = s->last_dc[1] = s->last_dc[2] = 1024 << s->intra_dc_precision;
3021 ff_set_qscale(s, s->qscale);
3022 ff_init_block_index(s);
3024 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
3025 int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
3026 int mb_type= s->mb_type[xy];
3030 int size_increase = s->avctx->internal->byte_buffer_size/4
3031 + s->mb_width*MAX_MB_BYTES;
3033 ff_mpv_reallocate_putbitbuffer(s, MAX_MB_BYTES, size_increase);
3034 if (put_bytes_left(&s->pb, 0) < MAX_MB_BYTES){
3035 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
3038 if(s->data_partitioning){
3039 if (put_bytes_left(&s->pb2, 0) < MAX_MB_BYTES ||
3040 put_bytes_left(&s->tex_pb, 0) < MAX_MB_BYTES) {
3041 av_log(s->avctx, AV_LOG_ERROR, "encoded partitioned frame too large\n");
3047 s->mb_y = mb_y; // moved into loop, can get changed by H.261
3048 ff_update_block_index(s);
3050 if(CONFIG_H261_ENCODER && s->codec_id == AV_CODEC_ID_H261){
3051 ff_h261_reorder_mb_index(s);
3052 xy= s->mb_y*s->mb_stride + s->mb_x;
3053 mb_type= s->mb_type[xy];
3056 /* write gob / video packet header */
3058 int current_packet_size, is_gob_start;
3060 current_packet_size = put_bytes_count(&s->pb, 1)
3061 - (s->ptr_lastgob - s->pb.buf);
3063 is_gob_start = s->rtp_payload_size &&
3064 current_packet_size >= s->rtp_payload_size &&
3067 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
3069 switch(s->codec_id){
3070 case AV_CODEC_ID_H263:
3071 case AV_CODEC_ID_H263P:
3072 if(!s->h263_slice_structured)
3073 if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
3075 case AV_CODEC_ID_MPEG2VIDEO:
3076 if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
3077 case AV_CODEC_ID_MPEG1VIDEO:
3078 if(s->mb_skip_run) is_gob_start=0;
3080 case AV_CODEC_ID_MJPEG:
3081 if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
3086 if(s->start_mb_y != mb_y || mb_x!=0){
3089 if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4 && s->partitioned_frame){
3090 ff_mpeg4_init_partitions(s);
3094 av_assert2((put_bits_count(&s->pb)&7) == 0);
3095 current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
3097 if (s->error_rate && s->resync_mb_x + s->resync_mb_y > 0) {
3098 int r = put_bytes_count(&s->pb, 0) + s->picture_number + 16 + s->mb_x + s->mb_y;
3099 int d = 100 / s->error_rate;
3101 current_packet_size=0;
3102 s->pb.buf_ptr= s->ptr_lastgob;
3103 av_assert1(put_bits_ptr(&s->pb) == s->ptr_lastgob);
3107 #if FF_API_RTP_CALLBACK
3108 FF_DISABLE_DEPRECATION_WARNINGS
3109 if (s->avctx->rtp_callback){
3110 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
3111 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
3113 FF_ENABLE_DEPRECATION_WARNINGS
3115 update_mb_info(s, 1);
3117 switch(s->codec_id){
3118 case AV_CODEC_ID_MPEG4:
3119 if (CONFIG_MPEG4_ENCODER) {
3120 ff_mpeg4_encode_video_packet_header(s);
3121 ff_mpeg4_clean_buffers(s);
3124 case AV_CODEC_ID_MPEG1VIDEO:
3125 case AV_CODEC_ID_MPEG2VIDEO:
3126 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
3127 ff_mpeg1_encode_slice_header(s);
3128 ff_mpeg1_clean_buffers(s);
3131 case AV_CODEC_ID_H263:
3132 case AV_CODEC_ID_H263P:
3133 if (CONFIG_H263_ENCODER)
3134 ff_h263_encode_gob_header(s, mb_y);
3138 if (s->avctx->flags & AV_CODEC_FLAG_PASS1) {
3139 int bits= put_bits_count(&s->pb);
3140 s->misc_bits+= bits - s->last_bits;
3144 s->ptr_lastgob += current_packet_size;
3145 s->first_slice_line=1;
3146 s->resync_mb_x=mb_x;
3147 s->resync_mb_y=mb_y;
3151 if( (s->resync_mb_x == s->mb_x)
3152 && s->resync_mb_y+1 == s->mb_y){
3153 s->first_slice_line=0;
3157 s->dquant=0; //only for QP_RD
3159 update_mb_info(s, 0);
3161 if (mb_type & (mb_type-1) || (s->mpv_flags & FF_MPV_FLAG_QP_RD)) { // more than 1 MB type possible or FF_MPV_FLAG_QP_RD
3163 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
3165 copy_context_before_encode(&backup_s, s, -1);
3167 best_s.data_partitioning= s->data_partitioning;
3168 best_s.partitioned_frame= s->partitioned_frame;
3169 if(s->data_partitioning){
3170 backup_s.pb2= s->pb2;
3171 backup_s.tex_pb= s->tex_pb;
3174 if(mb_type&CANDIDATE_MB_TYPE_INTER){
3175 s->mv_dir = MV_DIR_FORWARD;
3176 s->mv_type = MV_TYPE_16X16;
3178 s->mv[0][0][0] = s->p_mv_table[xy][0];
3179 s->mv[0][0][1] = s->p_mv_table[xy][1];
3180 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
3181 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3183 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
3184 s->mv_dir = MV_DIR_FORWARD;
3185 s->mv_type = MV_TYPE_FIELD;
3188 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
3189 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
3190 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
3192 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
3193 &dmin, &next_block, 0, 0);
3195 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
3196 s->mv_dir = MV_DIR_FORWARD;
3197 s->mv_type = MV_TYPE_16X16;
3201 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
3202 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3204 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
3205 s->mv_dir = MV_DIR_FORWARD;
3206 s->mv_type = MV_TYPE_8X8;
3209 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
3210 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
3212 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
3213 &dmin, &next_block, 0, 0);
3215 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
3216 s->mv_dir = MV_DIR_FORWARD;
3217 s->mv_type = MV_TYPE_16X16;
3219 s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
3220 s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
3221 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
3222 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3224 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
3225 s->mv_dir = MV_DIR_BACKWARD;
3226 s->mv_type = MV_TYPE_16X16;
3228 s->mv[1][0][0] = s->b_back_mv_table[xy][0];
3229 s->mv[1][0][1] = s->b_back_mv_table[xy][1];
3230 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
3231 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
3233 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
3234 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3235 s->mv_type = MV_TYPE_16X16;
3237 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
3238 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
3239 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
3240 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
3241 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
3242 &dmin, &next_block, 0, 0);
3244 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
3245 s->mv_dir = MV_DIR_FORWARD;
3246 s->mv_type = MV_TYPE_FIELD;
3249 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
3250 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
3251 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
3253 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
3254 &dmin, &next_block, 0, 0);
3256 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
3257 s->mv_dir = MV_DIR_BACKWARD;
3258 s->mv_type = MV_TYPE_FIELD;
3261 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
3262 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
3263 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
3265 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
3266 &dmin, &next_block, 0, 0);
3268 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
3269 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3270 s->mv_type = MV_TYPE_FIELD;
3272 for(dir=0; dir<2; dir++){
3274 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
3275 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
3276 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
3279 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
3280 &dmin, &next_block, 0, 0);
3282 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
3284 s->mv_type = MV_TYPE_16X16;
3288 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
3289 &dmin, &next_block, 0, 0);
3290 if(s->h263_pred || s->h263_aic){
3292 s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
3294 ff_clean_intra_table_entries(s); //old mode?
3298 if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) && dmin < INT_MAX) {
3299 if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
3300 const int last_qp= backup_s.qscale;
3303 const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
3304 static const int dquant_tab[4]={-1,1,-2,2};
3305 int storecoefs = s->mb_intra && s->dc_val[0];
3307 av_assert2(backup_s.dquant == 0);
3310 s->mv_dir= best_s.mv_dir;
3311 s->mv_type = MV_TYPE_16X16;
3312 s->mb_intra= best_s.mb_intra;
3313 s->mv[0][0][0] = best_s.mv[0][0][0];
3314 s->mv[0][0][1] = best_s.mv[0][0][1];
3315 s->mv[1][0][0] = best_s.mv[1][0][0];
3316 s->mv[1][0][1] = best_s.mv[1][0][1];
3318 qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
3319 for(; qpi<4; qpi++){
3320 int dquant= dquant_tab[qpi];
3321 qp= last_qp + dquant;
3322 if(qp < s->avctx->qmin || qp > s->avctx->qmax)
3324 backup_s.dquant= dquant;
3327 dc[i]= s->dc_val[0][ s->block_index[i] ];
3328 memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(int16_t)*16);
3332 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
3333 &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
3334 if(best_s.qscale != qp){
3337 s->dc_val[0][ s->block_index[i] ]= dc[i];
3338 memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(int16_t)*16);
3345 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
3346 int mx= s->b_direct_mv_table[xy][0];
3347 int my= s->b_direct_mv_table[xy][1];
3349 backup_s.dquant = 0;
3350 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
3352 ff_mpeg4_set_direct_mv(s, mx, my);
3353 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
3354 &dmin, &next_block, mx, my);
3356 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
3357 backup_s.dquant = 0;
3358 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
3360 ff_mpeg4_set_direct_mv(s, 0, 0);
3361 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
3362 &dmin, &next_block, 0, 0);
3364 if (!best_s.mb_intra && s->mpv_flags & FF_MPV_FLAG_SKIP_RD) {
3367 coded |= s->block_last_index[i];
3370 memcpy(s->mv, best_s.mv, sizeof(s->mv));
3371 if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
3372 mx=my=0; //FIXME find the one we actually used
3373 ff_mpeg4_set_direct_mv(s, mx, my);
3374 }else if(best_s.mv_dir&MV_DIR_BACKWARD){
3382 s->mv_dir= best_s.mv_dir;
3383 s->mv_type = best_s.mv_type;
3385 /* s->mv[0][0][0] = best_s.mv[0][0][0];
3386 s->mv[0][0][1] = best_s.mv[0][0][1];
3387 s->mv[1][0][0] = best_s.mv[1][0][0];
3388 s->mv[1][0][1] = best_s.mv[1][0][1];*/
3391 encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
3392 &dmin, &next_block, mx, my);
3397 s->current_picture.qscale_table[xy] = best_s.qscale;
3399 copy_context_after_encode(s, &best_s, -1);
3401 pb_bits_count= put_bits_count(&s->pb);
3402 flush_put_bits(&s->pb);
3403 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
3406 if(s->data_partitioning){
3407 pb2_bits_count= put_bits_count(&s->pb2);
3408 flush_put_bits(&s->pb2);
3409 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
3410 s->pb2= backup_s.pb2;
3412 tex_pb_bits_count= put_bits_count(&s->tex_pb);
3413 flush_put_bits(&s->tex_pb);
3414 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
3415 s->tex_pb= backup_s.tex_pb;
3417 s->last_bits= put_bits_count(&s->pb);
3419 if (CONFIG_H263_ENCODER &&
3420 s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
3421 ff_h263_update_motion_val(s);
3423 if(next_block==0){ //FIXME 16 vs linesize16
3424 s->hdsp.put_pixels_tab[0][0](s->dest[0], s->sc.rd_scratchpad , s->linesize ,16);
3425 s->hdsp.put_pixels_tab[1][0](s->dest[1], s->sc.rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
3426 s->hdsp.put_pixels_tab[1][0](s->dest[2], s->sc.rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
3429 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
3430 ff_mpv_reconstruct_mb(s, s->block);
3432 int motion_x = 0, motion_y = 0;
3433 s->mv_type=MV_TYPE_16X16;
3434 // only one MB-Type possible
3437 case CANDIDATE_MB_TYPE_INTRA:
3440 motion_x= s->mv[0][0][0] = 0;
3441 motion_y= s->mv[0][0][1] = 0;
3443 case CANDIDATE_MB_TYPE_INTER:
3444 s->mv_dir = MV_DIR_FORWARD;
3446 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
3447 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
3449 case CANDIDATE_MB_TYPE_INTER_I:
3450 s->mv_dir = MV_DIR_FORWARD;
3451 s->mv_type = MV_TYPE_FIELD;
3454 j= s->field_select[0][i] = s->p_field_select_table[i][xy];
3455 s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
3456 s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
3459 case CANDIDATE_MB_TYPE_INTER4V:
3460 s->mv_dir = MV_DIR_FORWARD;
3461 s->mv_type = MV_TYPE_8X8;
3464 s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
3465 s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
3468 case CANDIDATE_MB_TYPE_DIRECT:
3469 if (CONFIG_MPEG4_ENCODER) {
3470 s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
3472 motion_x=s->b_direct_mv_table[xy][0];
3473 motion_y=s->b_direct_mv_table[xy][1];
3474 ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
3477 case CANDIDATE_MB_TYPE_DIRECT0:
3478 if (CONFIG_MPEG4_ENCODER) {
3479 s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
3481 ff_mpeg4_set_direct_mv(s, 0, 0);
3484 case CANDIDATE_MB_TYPE_BIDIR:
3485 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3487 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
3488 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
3489 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
3490 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
3492 case CANDIDATE_MB_TYPE_BACKWARD:
3493 s->mv_dir = MV_DIR_BACKWARD;
3495 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
3496 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
3498 case CANDIDATE_MB_TYPE_FORWARD:
3499 s->mv_dir = MV_DIR_FORWARD;
3501 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
3502 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
3504 case CANDIDATE_MB_TYPE_FORWARD_I:
3505 s->mv_dir = MV_DIR_FORWARD;
3506 s->mv_type = MV_TYPE_FIELD;
3509 j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
3510 s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
3511 s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
3514 case CANDIDATE_MB_TYPE_BACKWARD_I:
3515 s->mv_dir = MV_DIR_BACKWARD;
3516 s->mv_type = MV_TYPE_FIELD;
3519 j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
3520 s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
3521 s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
3524 case CANDIDATE_MB_TYPE_BIDIR_I:
3525 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3526 s->mv_type = MV_TYPE_FIELD;
3528 for(dir=0; dir<2; dir++){
3530 j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
3531 s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
3532 s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
3537 av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
3540 encode_mb(s, motion_x, motion_y);
3542 // RAL: Update last macroblock type
3543 s->last_mv_dir = s->mv_dir;
3545 if (CONFIG_H263_ENCODER &&
3546 s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
3547 ff_h263_update_motion_val(s);
3549 ff_mpv_reconstruct_mb(s, s->block);
3552 /* clean the MV table in IPS frames for direct mode in B-frames */
3553 if(s->mb_intra /* && I,P,S_TYPE */){
3554 s->p_mv_table[xy][0]=0;
3555 s->p_mv_table[xy][1]=0;
3558 if (s->avctx->flags & AV_CODEC_FLAG_PSNR) {
3562 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
3563 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
3565 s->current_picture.encoding_error[0] += sse(
3566 s, s->new_picture.f->data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
3567 s->dest[0], w, h, s->linesize);
3568 s->current_picture.encoding_error[1] += sse(
3569 s, s->new_picture.f->data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
3570 s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
3571 s->current_picture.encoding_error[2] += sse(
3572 s, s->new_picture.f->data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
3573 s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
3576 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
3577 ff_h263_loop_filter(s);
3579 ff_dlog(s->avctx, "MB %d %d bits\n",
3580 s->mb_x + s->mb_y * s->mb_stride, put_bits_count(&s->pb));
3584 //not beautiful here but we must write it before flushing so it has to be here
3585 if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
3586 ff_msmpeg4_encode_ext_header(s);
3590 #if FF_API_RTP_CALLBACK
3591 FF_DISABLE_DEPRECATION_WARNINGS
3592 /* Send the last GOB if RTP */
3593 if (s->avctx->rtp_callback) {
3594 int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
3595 int pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
3596 /* Call the RTP callback to send the last GOB */
3598 s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
3600 FF_ENABLE_DEPRECATION_WARNINGS
3606 #define MERGE(field) dst->field += src->field; src->field=0
3607 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
3608 MERGE(me.scene_change_score);
3609 MERGE(me.mc_mb_var_sum_temp);
3610 MERGE(me.mb_var_sum_temp);
3613 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
3616 MERGE(dct_count[0]); //note, the other dct vars are not part of the context
3617 MERGE(dct_count[1]);
3626 MERGE(er.error_count);
3627 MERGE(padding_bug_score);
3628 MERGE(current_picture.encoding_error[0]);
3629 MERGE(current_picture.encoding_error[1]);
3630 MERGE(current_picture.encoding_error[2]);
3632 if (dst->noise_reduction){
3633 for(i=0; i<64; i++){
3634 MERGE(dct_error_sum[0][i]);
3635 MERGE(dct_error_sum[1][i]);
3639 av_assert1(put_bits_count(&src->pb) % 8 ==0);
3640 av_assert1(put_bits_count(&dst->pb) % 8 ==0);
3641 ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
3642 flush_put_bits(&dst->pb);
3645 static int estimate_qp(MpegEncContext *s, int dry_run){
3646 if (s->next_lambda){
3647 s->current_picture_ptr->f->quality =
3648 s->current_picture.f->quality = s->next_lambda;
3649 if(!dry_run) s->next_lambda= 0;
3650 } else if (!s->fixed_qscale) {
3651 int quality = ff_rate_estimate_qscale(s, dry_run);
3652 s->current_picture_ptr->f->quality =
3653 s->current_picture.f->quality = quality;
3654 if (s->current_picture.f->quality < 0)
3658 if(s->adaptive_quant){
3659 switch(s->codec_id){
3660 case AV_CODEC_ID_MPEG4:
3661 if (CONFIG_MPEG4_ENCODER)
3662 ff_clean_mpeg4_qscales(s);
3664 case AV_CODEC_ID_H263:
3665 case AV_CODEC_ID_H263P:
3666 case AV_CODEC_ID_FLV1:
3667 if (CONFIG_H263_ENCODER)
3668 ff_clean_h263_qscales(s);
3671 ff_init_qscale_tab(s);
3674 s->lambda= s->lambda_table[0];
3677 s->lambda = s->current_picture.f->quality;
3682 /* must be called before writing the header */
3683 static void set_frame_distances(MpegEncContext * s){
3684 av_assert1(s->current_picture_ptr->f->pts != AV_NOPTS_VALUE);
3685 s->time = s->current_picture_ptr->f->pts * s->avctx->time_base.num;
3687 if(s->pict_type==AV_PICTURE_TYPE_B){
3688 s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
3689 av_assert1(s->pb_time > 0 && s->pb_time < s->pp_time);
3691 s->pp_time= s->time - s->last_non_b_time;
3692 s->last_non_b_time= s->time;
3693 av_assert1(s->picture_number==0 || s->pp_time > 0);
3697 static int encode_picture(MpegEncContext *s, int picture_number)
3701 int context_count = s->slice_context_count;
3703 s->picture_number = picture_number;
3705 /* Reset the average MB variance */
3706 s->me.mb_var_sum_temp =
3707 s->me.mc_mb_var_sum_temp = 0;
3709 /* we need to initialize some time vars before we can encode B-frames */
3710 // RAL: Condition added for MPEG1VIDEO
3711 if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO || s->codec_id == AV_CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
3712 set_frame_distances(s);
3713 if(CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4)
3714 ff_set_mpeg4_time(s);
3716 s->me.scene_change_score=0;
3718 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
3720 if(s->pict_type==AV_PICTURE_TYPE_I){
3721 if(s->msmpeg4_version >= 3) s->no_rounding=1;
3722 else s->no_rounding=0;
3723 }else if(s->pict_type!=AV_PICTURE_TYPE_B){
3724 if(s->flipflop_rounding || s->codec_id == AV_CODEC_ID_H263P || s->codec_id == AV_CODEC_ID_MPEG4)
3725 s->no_rounding ^= 1;
3728 if (s->avctx->flags & AV_CODEC_FLAG_PASS2) {
3729 if (estimate_qp(s,1) < 0)
3731 ff_get_2pass_fcode(s);
3732 } else if (!(s->avctx->flags & AV_CODEC_FLAG_QSCALE)) {
3733 if(s->pict_type==AV_PICTURE_TYPE_B)
3734 s->lambda= s->last_lambda_for[s->pict_type];
3736 s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
3740 if(s->codec_id != AV_CODEC_ID_AMV && s->codec_id != AV_CODEC_ID_MJPEG){
3741 if(s->q_chroma_intra_matrix != s->q_intra_matrix ) av_freep(&s->q_chroma_intra_matrix);
3742 if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
3743 s->q_chroma_intra_matrix = s->q_intra_matrix;
3744 s->q_chroma_intra_matrix16 = s->q_intra_matrix16;
3747 s->mb_intra=0; //for the rate distortion & bit compare functions
3748 for(i=1; i<context_count; i++){
3749 ret = ff_update_duplicate_context(s->thread_context[i], s);
3757 /* Estimate motion for every MB */
3758 if(s->pict_type != AV_PICTURE_TYPE_I){
3759 s->lambda = (s->lambda * s->me_penalty_compensation + 128) >> 8;
3760 s->lambda2 = (s->lambda2 * (int64_t) s->me_penalty_compensation + 128) >> 8;
3761 if (s->pict_type != AV_PICTURE_TYPE_B) {
3762 if ((s->me_pre && s->last_non_b_pict_type == AV_PICTURE_TYPE_I) ||
3764 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3768 s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3769 }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
3771 for(i=0; i<s->mb_stride*s->mb_height; i++)
3772 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3774 if(!s->fixed_qscale){
3775 /* finding spatial complexity for I-frame rate control */
3776 s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3779 for(i=1; i<context_count; i++){
3780 merge_context_after_me(s, s->thread_context[i]);
3782 s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
3783 s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
3786 if (s->me.scene_change_score > s->scenechange_threshold &&
3787 s->pict_type == AV_PICTURE_TYPE_P) {
3788 s->pict_type= AV_PICTURE_TYPE_I;
3789 for(i=0; i<s->mb_stride*s->mb_height; i++)
3790 s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3791 if(s->msmpeg4_version >= 3)
3793 ff_dlog(s, "Scene change detected, encoding as I Frame %"PRId64" %"PRId64"\n",
3794 s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
3798 if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
3799 s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
3801 if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3803 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
3804 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
3805 s->f_code= FFMAX3(s->f_code, a, b);
3808 ff_fix_long_p_mvs(s, s->intra_penalty ? CANDIDATE_MB_TYPE_INTER : CANDIDATE_MB_TYPE_INTRA);
3809 ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, !!s->intra_penalty);
3810 if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3814 ff_fix_long_mvs(s, s->p_field_select_table[i], j,
3815 s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, !!s->intra_penalty);
3820 if(s->pict_type==AV_PICTURE_TYPE_B){
3823 a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
3824 b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
3825 s->f_code = FFMAX(a, b);
3827 a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
3828 b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
3829 s->b_code = FFMAX(a, b);
3831 ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
3832 ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
3833 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
3834 ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
3835 if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3837 for(dir=0; dir<2; dir++){
3840 int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
3841 : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
3842 ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
3843 s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
3851 if (estimate_qp(s, 0) < 0)
3854 if (s->qscale < 3 && s->max_qcoeff <= 128 &&
3855 s->pict_type == AV_PICTURE_TYPE_I &&
3856 !(s->avctx->flags & AV_CODEC_FLAG_QSCALE))
3857 s->qscale= 3; //reduce clipping problems
3859 if (s->out_format == FMT_MJPEG) {
3860 const uint16_t * luma_matrix = ff_mpeg1_default_intra_matrix;
3861 const uint16_t *chroma_matrix = ff_mpeg1_default_intra_matrix;
3863 if (s->avctx->intra_matrix) {
3865 luma_matrix = s->avctx->intra_matrix;
3867 if (s->avctx->chroma_intra_matrix)
3868 chroma_matrix = s->avctx->chroma_intra_matrix;
3870 /* for mjpeg, we do include qscale in the matrix */
3872 int j = s->idsp.idct_permutation[i];
3874 s->chroma_intra_matrix[j] = av_clip_uint8((chroma_matrix[i] * s->qscale) >> 3);
3875 s-> intra_matrix[j] = av_clip_uint8(( luma_matrix[i] * s->qscale) >> 3);
3877 s->y_dc_scale_table=
3878 s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
3879 s->chroma_intra_matrix[0] =
3880 s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
3881 ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
3882 s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
3883 ff_convert_matrix(s, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
3884 s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
3887 if(s->codec_id == AV_CODEC_ID_AMV){
3888 static const uint8_t y[32]={13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13};
3889 static const uint8_t c[32]={14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14};
3891 int j= s->idsp.idct_permutation[ff_zigzag_direct[i]];
3893 s->intra_matrix[j] = sp5x_qscale_five_quant_table[0][i];
3894 s->chroma_intra_matrix[j] = sp5x_qscale_five_quant_table[1][i];
3896 s->y_dc_scale_table= y;
3897 s->c_dc_scale_table= c;
3898 s->intra_matrix[0] = 13;
3899 s->chroma_intra_matrix[0] = 14;
3900 ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
3901 s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
3902 ff_convert_matrix(s, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
3903 s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
3907 if (s->out_format == FMT_SPEEDHQ) {
3908 s->y_dc_scale_table=
3909 s->c_dc_scale_table= ff_mpeg2_dc_scale_table[3];
3912 //FIXME var duplication
3913 s->current_picture_ptr->f->key_frame =
3914 s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
3915 s->current_picture_ptr->f->pict_type =
3916 s->current_picture.f->pict_type = s->pict_type;
3918 if (s->current_picture.f->key_frame)
3919 s->picture_in_gop_number=0;
3921 s->mb_x = s->mb_y = 0;
3922 s->last_bits= put_bits_count(&s->pb);
3923 switch(s->out_format) {
3925 if (CONFIG_MJPEG_ENCODER && s->huffman != HUFFMAN_TABLE_OPTIMAL)
3926 ff_mjpeg_encode_picture_header(s->avctx, &s->pb, &s->intra_scantable,
3927 s->pred, s->intra_matrix, s->chroma_intra_matrix);
3930 if (CONFIG_SPEEDHQ_ENCODER)
3931 ff_speedhq_encode_picture_header(s);
3934 if (CONFIG_H261_ENCODER)
3935 ff_h261_encode_picture_header(s, picture_number);
3938 if (CONFIG_WMV2_ENCODER && s->codec_id == AV_CODEC_ID_WMV2)
3939 ff_wmv2_encode_picture_header(s, picture_number);
3940 else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
3941 ff_msmpeg4_encode_picture_header(s, picture_number);
3942 else if (CONFIG_MPEG4_ENCODER && s->h263_pred) {
3943 ret = ff_mpeg4_encode_picture_header(s, picture_number);
3946 } else if (CONFIG_RV10_ENCODER && s->codec_id == AV_CODEC_ID_RV10) {
3947 ret = ff_rv10_encode_picture_header(s, picture_number);
3951 else if (CONFIG_RV20_ENCODER && s->codec_id == AV_CODEC_ID_RV20)
3952 ff_rv20_encode_picture_header(s, picture_number);
3953 else if (CONFIG_FLV_ENCODER && s->codec_id == AV_CODEC_ID_FLV1)
3954 ff_flv_encode_picture_header(s, picture_number);
3955 else if (CONFIG_H263_ENCODER)
3956 ff_h263_encode_picture_header(s, picture_number);
3959 if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
3960 ff_mpeg1_encode_picture_header(s, picture_number);
3965 bits= put_bits_count(&s->pb);
3966 s->header_bits= bits - s->last_bits;
3968 for(i=1; i<context_count; i++){
3969 update_duplicate_context_after_me(s->thread_context[i], s);
3971 s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3972 for(i=1; i<context_count; i++){
3973 if (s->pb.buf_end == s->thread_context[i]->pb.buf)
3974 set_put_bits_buffer_size(&s->pb, FFMIN(s->thread_context[i]->pb.buf_end - s->pb.buf, INT_MAX/8-BUF_BITS));
3975 merge_context_after_encode(s, s->thread_context[i]);
3981 static void denoise_dct_c(MpegEncContext *s, int16_t *block){
3982 const int intra= s->mb_intra;
3985 s->dct_count[intra]++;
3987 for(i=0; i<64; i++){
3988 int level= block[i];
3992 s->dct_error_sum[intra][i] += level;
3993 level -= s->dct_offset[intra][i];
3994 if(level<0) level=0;
3996 s->dct_error_sum[intra][i] -= level;
3997 level += s->dct_offset[intra][i];
3998 if(level>0) level=0;
4005 static int dct_quantize_trellis_c(MpegEncContext *s,
4006 int16_t *block, int n,
4007 int qscale, int *overflow){
4009 const uint16_t *matrix;
4010 const uint8_t *scantable;
4011 const uint8_t *perm_scantable;
4013 unsigned int threshold1, threshold2;
4025 int coeff_count[64];
4026 int qmul, qadd, start_i, last_non_zero, i, dc;
4027 const int esc_length= s->ac_esc_length;
4029 uint8_t * last_length;
4030 const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
4033 s->fdsp.fdct(block);
4035 if(s->dct_error_sum)
4036 s->denoise_dct(s, block);
4038 qadd= ((qscale-1)|1)*8;
4040 if (s->q_scale_type) mpeg2_qscale = ff_mpeg2_non_linear_qscale[qscale];
4041 else mpeg2_qscale = qscale << 1;
4045 scantable= s->intra_scantable.scantable;
4046 perm_scantable= s->intra_scantable.permutated;
4054 /* For AIC we skip quant/dequant of INTRADC */
4059 /* note: block[0] is assumed to be positive */
4060 block[0] = (block[0] + (q >> 1)) / q;
4063 qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
4064 matrix = n < 4 ? s->intra_matrix : s->chroma_intra_matrix;
4065 if(s->mpeg_quant || s->out_format == FMT_MPEG1 || s->out_format == FMT_MJPEG)
4066 bias= 1<<(QMAT_SHIFT-1);
4068 if (n > 3 && s->intra_chroma_ac_vlc_length) {
4069 length = s->intra_chroma_ac_vlc_length;
4070 last_length= s->intra_chroma_ac_vlc_last_length;
4072 length = s->intra_ac_vlc_length;
4073 last_length= s->intra_ac_vlc_last_length;
4076 scantable= s->inter_scantable.scantable;
4077 perm_scantable= s->inter_scantable.permutated;
4080 qmat = s->q_inter_matrix[qscale];
4081 matrix = s->inter_matrix;
4082 length = s->inter_ac_vlc_length;
4083 last_length= s->inter_ac_vlc_last_length;
4087 threshold1= (1<<QMAT_SHIFT) - bias - 1;
4088 threshold2= (threshold1<<1);
4090 for(i=63; i>=start_i; i--) {
4091 const int j = scantable[i];
4092 int level = block[j] * qmat[j];
4094 if(((unsigned)(level+threshold1))>threshold2){
4100 for(i=start_i; i<=last_non_zero; i++) {
4101 const int j = scantable[i];
4102 int level = block[j] * qmat[j];
4104 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
4105 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
4106 if(((unsigned)(level+threshold1))>threshold2){
4108 level= (bias + level)>>QMAT_SHIFT;
4110 coeff[1][i]= level-1;
4111 // coeff[2][k]= level-2;
4113 level= (bias - level)>>QMAT_SHIFT;
4114 coeff[0][i]= -level;
4115 coeff[1][i]= -level+1;
4116 // coeff[2][k]= -level+2;
4118 coeff_count[i]= FFMIN(level, 2);
4119 av_assert2(coeff_count[i]);
4122 coeff[0][i]= (level>>31)|1;
4127 *overflow= s->max_qcoeff < max; //overflow might have happened
4129 if(last_non_zero < start_i){
4130 memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
4131 return last_non_zero;
4134 score_tab[start_i]= 0;
4135 survivor[0]= start_i;
4138 for(i=start_i; i<=last_non_zero; i++){
4139 int level_index, j, zero_distortion;
4140 int dct_coeff= FFABS(block[ scantable[i] ]);
4141 int best_score=256*256*256*120;
4143 if (s->fdsp.fdct == ff_fdct_ifast)
4144 dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
4145 zero_distortion= dct_coeff*dct_coeff;
4147 for(level_index=0; level_index < coeff_count[i]; level_index++){
4149 int level= coeff[level_index][i];
4150 const int alevel= FFABS(level);
4155 if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4156 unquant_coeff= alevel*qmul + qadd;
4157 } else if(s->out_format == FMT_MJPEG) {
4158 j = s->idsp.idct_permutation[scantable[i]];
4159 unquant_coeff = alevel * matrix[j] * 8;
4161 j = s->idsp.idct_permutation[scantable[i]]; // FIXME: optimize
4163 unquant_coeff = (int)( alevel * mpeg2_qscale * matrix[j]) >> 4;
4164 unquant_coeff = (unquant_coeff - 1) | 1;
4166 unquant_coeff = ((( alevel << 1) + 1) * mpeg2_qscale * ((int) matrix[j])) >> 5;
4167 unquant_coeff = (unquant_coeff - 1) | 1;
4172 distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
4174 if((level&(~127)) == 0){
4175 for(j=survivor_count-1; j>=0; j--){
4176 int run= i - survivor[j];
4177 int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
4178 score += score_tab[i-run];
4180 if(score < best_score){
4183 level_tab[i+1]= level-64;
4187 if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4188 for(j=survivor_count-1; j>=0; j--){
4189 int run= i - survivor[j];
4190 int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
4191 score += score_tab[i-run];
4192 if(score < last_score){
4195 last_level= level-64;
4201 distortion += esc_length*lambda;
4202 for(j=survivor_count-1; j>=0; j--){
4203 int run= i - survivor[j];
4204 int score= distortion + score_tab[i-run];
4206 if(score < best_score){
4209 level_tab[i+1]= level-64;
4213 if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4214 for(j=survivor_count-1; j>=0; j--){
4215 int run= i - survivor[j];
4216 int score= distortion + score_tab[i-run];
4217 if(score < last_score){
4220 last_level= level-64;
4228 score_tab[i+1]= best_score;
4230 // Note: there is a vlc code in MPEG-4 which is 1 bit shorter then another one with a shorter run and the same level
4231 if(last_non_zero <= 27){
4232 for(; survivor_count; survivor_count--){
4233 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
4237 for(; survivor_count; survivor_count--){
4238 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
4243 survivor[ survivor_count++ ]= i+1;
4246 if(s->out_format != FMT_H263 && s->out_format != FMT_H261){
4247 last_score= 256*256*256*120;
4248 for(i= survivor[0]; i<=last_non_zero + 1; i++){
4249 int score= score_tab[i];
4251 score += lambda * 2; // FIXME more exact?
4253 if(score < last_score){
4256 last_level= level_tab[i];
4257 last_run= run_tab[i];
4262 s->coded_score[n] = last_score;
4264 dc= FFABS(block[0]);
4265 last_non_zero= last_i - 1;
4266 memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
4268 if(last_non_zero < start_i)
4269 return last_non_zero;
4271 if(last_non_zero == 0 && start_i == 0){
4273 int best_score= dc * dc;
4275 for(i=0; i<coeff_count[0]; i++){
4276 int level= coeff[i][0];
4277 int alevel= FFABS(level);
4278 int unquant_coeff, score, distortion;
4280 if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4281 unquant_coeff= (alevel*qmul + qadd)>>3;
4283 unquant_coeff = ((( alevel << 1) + 1) * mpeg2_qscale * ((int) matrix[0])) >> 5;
4284 unquant_coeff = (unquant_coeff - 1) | 1;
4286 unquant_coeff = (unquant_coeff + 4) >> 3;
4287 unquant_coeff<<= 3 + 3;
4289 distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
4291 if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
4292 else score= distortion + esc_length*lambda;
4294 if(score < best_score){
4296 best_level= level - 64;
4299 block[0]= best_level;
4300 s->coded_score[n] = best_score - dc*dc;
4301 if(best_level == 0) return -1;
4302 else return last_non_zero;
4306 av_assert2(last_level);
4308 block[ perm_scantable[last_non_zero] ]= last_level;
4311 for(; i>start_i; i -= run_tab[i] + 1){
4312 block[ perm_scantable[i-1] ]= level_tab[i];
4315 return last_non_zero;
4318 static int16_t basis[64][64];
4320 static void build_basis(uint8_t *perm){
4327 double s= 0.25*(1<<BASIS_SHIFT);
4329 int perm_index= perm[index];
4330 if(i==0) s*= sqrt(0.5);
4331 if(j==0) s*= sqrt(0.5);
4332 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)));
4339 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
4340 int16_t *block, int16_t *weight, int16_t *orig,
4343 LOCAL_ALIGNED_16(int16_t, d1, [64]);
4344 const uint8_t *scantable;
4345 const uint8_t *perm_scantable;
4346 // unsigned int threshold1, threshold2;
4351 int qmul, qadd, start_i, last_non_zero, i, dc;
4353 uint8_t * last_length;
4355 int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
4357 if(basis[0][0] == 0)
4358 build_basis(s->idsp.idct_permutation);
4363 scantable= s->intra_scantable.scantable;
4364 perm_scantable= s->intra_scantable.permutated;
4371 /* For AIC we skip quant/dequant of INTRADC */
4375 q <<= RECON_SHIFT-3;
4376 /* note: block[0] is assumed to be positive */
4378 // block[0] = (block[0] + (q >> 1)) / q;
4380 // if(s->mpeg_quant || s->out_format == FMT_MPEG1)
4381 // bias= 1<<(QMAT_SHIFT-1);
4382 if (n > 3 && s->intra_chroma_ac_vlc_length) {
4383 length = s->intra_chroma_ac_vlc_length;
4384 last_length= s->intra_chroma_ac_vlc_last_length;
4386 length = s->intra_ac_vlc_length;
4387 last_length= s->intra_ac_vlc_last_length;
4390 scantable= s->inter_scantable.scantable;
4391 perm_scantable= s->inter_scantable.permutated;
4394 length = s->inter_ac_vlc_length;
4395 last_length= s->inter_ac_vlc_last_length;
4397 last_non_zero = s->block_last_index[n];
4399 dc += (1<<(RECON_SHIFT-1));
4400 for(i=0; i<64; i++){
4401 rem[i] = dc - (orig[i] << RECON_SHIFT); // FIXME use orig directly instead of copying to rem[]
4405 for(i=0; i<64; i++){
4410 w= FFABS(weight[i]) + qns*one;
4411 w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
4414 // w=weight[i] = (63*qns + (w/2)) / w;
4417 av_assert2(w<(1<<6));
4420 lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
4424 for(i=start_i; i<=last_non_zero; i++){
4425 int j= perm_scantable[i];
4426 const int level= block[j];
4430 if(level<0) coeff= qmul*level - qadd;
4431 else coeff= qmul*level + qadd;
4432 run_tab[rle_index++]=run;
4435 s->mpvencdsp.add_8x8basis(rem, basis[j], coeff);
4442 int best_score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0], 0);
4445 int run2, best_unquant_change=0, analyze_gradient;
4446 analyze_gradient = last_non_zero > 2 || s->quantizer_noise_shaping >= 3;
4448 if(analyze_gradient){
4449 for(i=0; i<64; i++){
4452 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
4458 const int level= block[0];
4459 int change, old_coeff;
4461 av_assert2(s->mb_intra);
4465 for(change=-1; change<=1; change+=2){
4466 int new_level= level + change;
4467 int score, new_coeff;
4469 new_coeff= q*new_level;
4470 if(new_coeff >= 2048 || new_coeff < 0)
4473 score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0],
4474 new_coeff - old_coeff);
4475 if(score<best_score){
4478 best_change= change;
4479 best_unquant_change= new_coeff - old_coeff;
4486 run2= run_tab[rle_index++];
4490 for(i=start_i; i<64; i++){
4491 int j= perm_scantable[i];
4492 const int level= block[j];
4493 int change, old_coeff;
4495 if(s->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
4499 if(level<0) old_coeff= qmul*level - qadd;
4500 else old_coeff= qmul*level + qadd;
4501 run2= run_tab[rle_index++]; //FIXME ! maybe after last
4505 av_assert2(run2>=0 || i >= last_non_zero );
4508 for(change=-1; change<=1; change+=2){
4509 int new_level= level + change;
4510 int score, new_coeff, unquant_change;
4513 if(s->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
4517 if(new_level<0) new_coeff= qmul*new_level - qadd;
4518 else new_coeff= qmul*new_level + qadd;
4519 if(new_coeff >= 2048 || new_coeff <= -2048)
4521 //FIXME check for overflow
4524 if(level < 63 && level > -63){
4525 if(i < last_non_zero)
4526 score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
4527 - length[UNI_AC_ENC_INDEX(run, level+64)];
4529 score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
4530 - last_length[UNI_AC_ENC_INDEX(run, level+64)];
4533 av_assert2(FFABS(new_level)==1);
4535 if(analyze_gradient){
4536 int g= d1[ scantable[i] ];
4537 if(g && (g^new_level) >= 0)
4541 if(i < last_non_zero){
4542 int next_i= i + run2 + 1;
4543 int next_level= block[ perm_scantable[next_i] ] + 64;
4545 if(next_level&(~127))
4548 if(next_i < last_non_zero)
4549 score += length[UNI_AC_ENC_INDEX(run, 65)]
4550 + length[UNI_AC_ENC_INDEX(run2, next_level)]
4551 - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
4553 score += length[UNI_AC_ENC_INDEX(run, 65)]
4554 + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
4555 - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
4557 score += last_length[UNI_AC_ENC_INDEX(run, 65)];
4559 score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
4560 - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
4566 av_assert2(FFABS(level)==1);
4568 if(i < last_non_zero){
4569 int next_i= i + run2 + 1;
4570 int next_level= block[ perm_scantable[next_i] ] + 64;
4572 if(next_level&(~127))
4575 if(next_i < last_non_zero)
4576 score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
4577 - length[UNI_AC_ENC_INDEX(run2, next_level)]
4578 - length[UNI_AC_ENC_INDEX(run, 65)];
4580 score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
4581 - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
4582 - length[UNI_AC_ENC_INDEX(run, 65)];
4584 score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
4586 score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
4587 - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
4594 unquant_change= new_coeff - old_coeff;
4595 av_assert2((score < 100*lambda && score > -100*lambda) || lambda==0);
4597 score += s->mpvencdsp.try_8x8basis(rem, weight, basis[j],
4599 if(score<best_score){
4602 best_change= change;
4603 best_unquant_change= unquant_change;
4607 prev_level= level + 64;
4608 if(prev_level&(~127))
4618 int j= perm_scantable[ best_coeff ];
4620 block[j] += best_change;
4622 if(best_coeff > last_non_zero){
4623 last_non_zero= best_coeff;
4624 av_assert2(block[j]);
4626 for(; last_non_zero>=start_i; last_non_zero--){
4627 if(block[perm_scantable[last_non_zero]])
4634 for(i=start_i; i<=last_non_zero; i++){
4635 int j= perm_scantable[i];
4636 const int level= block[j];
4639 run_tab[rle_index++]=run;
4646 s->mpvencdsp.add_8x8basis(rem, basis[j], best_unquant_change);
4652 return last_non_zero;
4656 * Permute an 8x8 block according to permutation.
4657 * @param block the block which will be permuted according to
4658 * the given permutation vector
4659 * @param permutation the permutation vector
4660 * @param last the last non zero coefficient in scantable order, used to
4661 * speed the permutation up
4662 * @param scantable the used scantable, this is only used to speed the
4663 * permutation up, the block is not (inverse) permutated
4664 * to scantable order!
4666 void ff_block_permute(int16_t *block, uint8_t *permutation,
4667 const uint8_t *scantable, int last)
4674 //FIXME it is ok but not clean and might fail for some permutations
4675 // if (permutation[1] == 1)
4678 for (i = 0; i <= last; i++) {
4679 const int j = scantable[i];
4684 for (i = 0; i <= last; i++) {
4685 const int j = scantable[i];
4686 const int perm_j = permutation[j];
4687 block[perm_j] = temp[j];
4691 int ff_dct_quantize_c(MpegEncContext *s,
4692 int16_t *block, int n,
4693 int qscale, int *overflow)
4695 int i, j, level, last_non_zero, q, start_i;
4697 const uint8_t *scantable;
4700 unsigned int threshold1, threshold2;
4702 s->fdsp.fdct(block);
4704 if(s->dct_error_sum)
4705 s->denoise_dct(s, block);
4708 scantable= s->intra_scantable.scantable;
4716 /* For AIC we skip quant/dequant of INTRADC */
4719 /* note: block[0] is assumed to be positive */
4720 block[0] = (block[0] + (q >> 1)) / q;
4723 qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
4724 bias= s->intra_quant_bias*(1<<(QMAT_SHIFT - QUANT_BIAS_SHIFT));
4726 scantable= s->inter_scantable.scantable;
4729 qmat = s->q_inter_matrix[qscale];
4730 bias= s->inter_quant_bias*(1<<(QMAT_SHIFT - QUANT_BIAS_SHIFT));
4732 threshold1= (1<<QMAT_SHIFT) - bias - 1;
4733 threshold2= (threshold1<<1);
4734 for(i=63;i>=start_i;i--) {
4736 level = block[j] * qmat[j];
4738 if(((unsigned)(level+threshold1))>threshold2){
4745 for(i=start_i; i<=last_non_zero; i++) {
4747 level = block[j] * qmat[j];
4749 // if( bias+level >= (1<<QMAT_SHIFT)
4750 // || bias-level >= (1<<QMAT_SHIFT)){
4751 if(((unsigned)(level+threshold1))>threshold2){
4753 level= (bias + level)>>QMAT_SHIFT;
4756 level= (bias - level)>>QMAT_SHIFT;
4764 *overflow= s->max_qcoeff < max; //overflow might have happened
4766 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
4767 if (s->idsp.perm_type != FF_IDCT_PERM_NONE)
4768 ff_block_permute(block, s->idsp.idct_permutation,
4769 scantable, last_non_zero);
4771 return last_non_zero;
4774 #define OFFSET(x) offsetof(MpegEncContext, x)
4775 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
4776 static const AVOption h263_options[] = {
4777 { "obmc", "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
4778 { "mb_info", "emit macroblock info for RFC 2190 packetization, the parameter value is the maximum payload size", OFFSET(mb_info), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
4783 static const AVClass h263_class = {
4784 .class_name = "H.263 encoder",
4785 .item_name = av_default_item_name,
4786 .option = h263_options,
4787 .version = LIBAVUTIL_VERSION_INT,
4790 AVCodec ff_h263_encoder = {
4792 .long_name = NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
4793 .type = AVMEDIA_TYPE_VIDEO,
4794 .id = AV_CODEC_ID_H263,
4795 .priv_data_size = sizeof(MpegEncContext),
4796 .init = ff_mpv_encode_init,
4797 .encode2 = ff_mpv_encode_picture,
4798 .close = ff_mpv_encode_end,
4799 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
4800 .pix_fmts= (const enum AVPixelFormat[]){AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE},
4801 .priv_class = &h263_class,
4804 static const AVOption h263p_options[] = {
4805 { "umv", "Use unlimited motion vectors.", OFFSET(umvplus), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
4806 { "aiv", "Use alternative inter VLC.", OFFSET(alt_inter_vlc), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
4807 { "obmc", "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
4808 { "structured_slices", "Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE},
4812 static const AVClass h263p_class = {
4813 .class_name = "H.263p encoder",
4814 .item_name = av_default_item_name,
4815 .option = h263p_options,
4816 .version = LIBAVUTIL_VERSION_INT,
4819 AVCodec ff_h263p_encoder = {
4821 .long_name = NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
4822 .type = AVMEDIA_TYPE_VIDEO,
4823 .id = AV_CODEC_ID_H263P,
4824 .priv_data_size = sizeof(MpegEncContext),
4825 .init = ff_mpv_encode_init,
4826 .encode2 = ff_mpv_encode_picture,
4827 .close = ff_mpv_encode_end,
4828 .capabilities = AV_CODEC_CAP_SLICE_THREADS,
4829 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
4830 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4831 .priv_class = &h263p_class,
4834 static const AVClass msmpeg4v2_class = {
4835 .class_name = "msmpeg4v2 encoder",
4836 .item_name = av_default_item_name,
4837 .option = ff_mpv_generic_options,
4838 .version = LIBAVUTIL_VERSION_INT,
4841 AVCodec ff_msmpeg4v2_encoder = {
4842 .name = "msmpeg4v2",
4843 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
4844 .type = AVMEDIA_TYPE_VIDEO,
4845 .id = AV_CODEC_ID_MSMPEG4V2,
4846 .priv_data_size = sizeof(MpegEncContext),
4847 .init = ff_mpv_encode_init,
4848 .encode2 = ff_mpv_encode_picture,
4849 .close = ff_mpv_encode_end,
4850 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
4851 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4852 .priv_class = &msmpeg4v2_class,
4855 static const AVClass msmpeg4v3_class = {
4856 .class_name = "msmpeg4v3 encoder",
4857 .item_name = av_default_item_name,
4858 .option = ff_mpv_generic_options,
4859 .version = LIBAVUTIL_VERSION_INT,
4862 AVCodec ff_msmpeg4v3_encoder = {
4864 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
4865 .type = AVMEDIA_TYPE_VIDEO,
4866 .id = AV_CODEC_ID_MSMPEG4V3,
4867 .priv_data_size = sizeof(MpegEncContext),
4868 .init = ff_mpv_encode_init,
4869 .encode2 = ff_mpv_encode_picture,
4870 .close = ff_mpv_encode_end,
4871 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
4872 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4873 .priv_class = &msmpeg4v3_class,
4876 static const AVClass wmv1_class = {
4877 .class_name = "wmv1 encoder",
4878 .item_name = av_default_item_name,
4879 .option = ff_mpv_generic_options,
4880 .version = LIBAVUTIL_VERSION_INT,
4883 AVCodec ff_wmv1_encoder = {
4885 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
4886 .type = AVMEDIA_TYPE_VIDEO,
4887 .id = AV_CODEC_ID_WMV1,
4888 .priv_data_size = sizeof(MpegEncContext),
4889 .init = ff_mpv_encode_init,
4890 .encode2 = ff_mpv_encode_picture,
4891 .close = ff_mpv_encode_end,
4892 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
4893 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
4894 .priv_class = &wmv1_class,