FFmpeg
mpegvideo_enc.c
Go to the documentation of this file.
1 /*
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>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of FFmpeg.
9  *
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.
14  *
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.
19  *
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
23  */
24 
25 /*
26  * non linear quantizers with large QPs and VBV with restrictive qmin fixes sponsored by NOA GmbH
27  */
28 
29 /**
30  * @file
31  * The simplest mpeg encoder (well, it was the simplest!).
32  */
33 
34 #include <stdint.h>
35 
36 #include "libavutil/internal.h"
37 #include "libavutil/intmath.h"
38 #include "libavutil/mathematics.h"
39 #include "libavutil/pixdesc.h"
40 #include "libavutil/opt.h"
41 #include "avcodec.h"
42 #include "dct.h"
43 #include "idctdsp.h"
44 #include "mpeg12.h"
45 #include "mpegvideo.h"
46 #include "mpegvideodata.h"
47 #include "h261.h"
48 #include "h263.h"
49 #include "h263data.h"
50 #include "mjpegenc_common.h"
51 #include "mathops.h"
52 #include "mpegutils.h"
53 #include "mjpegenc.h"
54 #include "msmpeg4.h"
55 #include "pixblockdsp.h"
56 #include "qpeldsp.h"
57 #include "faandct.h"
58 #include "thread.h"
59 #include "aandcttab.h"
60 #include "flv.h"
61 #include "mpeg4video.h"
62 #include "internal.h"
63 #include "bytestream.h"
64 #include "wmv2.h"
65 #include "rv10.h"
66 #include "libxvid.h"
67 #include <limits.h>
68 #include "sp5x.h"
69 
70 #define QUANT_BIAS_SHIFT 8
71 
72 #define QMAT_SHIFT_MMX 16
73 #define QMAT_SHIFT 21
74 
75 static int encode_picture(MpegEncContext *s, int picture_number);
76 static int dct_quantize_refine(MpegEncContext *s, int16_t *block, int16_t *weight, int16_t *orig, int n, int qscale);
77 static int sse_mb(MpegEncContext *s);
78 static void denoise_dct_c(MpegEncContext *s, int16_t *block);
79 static int dct_quantize_trellis_c(MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow);
80 
81 static uint8_t default_mv_penalty[MAX_FCODE + 1][MAX_DMV * 2 + 1];
82 static uint8_t default_fcode_tab[MAX_MV * 2 + 1];
83 
84 const AVOption ff_mpv_generic_options[] = {
85  FF_MPV_COMMON_OPTS
86  { NULL },
87 };
88 
89 void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
90  uint16_t (*qmat16)[2][64],
91  const uint16_t *quant_matrix,
92  int bias, int qmin, int qmax, int intra)
93 {
94  FDCTDSPContext *fdsp = &s->fdsp;
95  int qscale;
96  int shift = 0;
97 
98  for (qscale = qmin; qscale <= qmax; qscale++) {
99  int i;
100  int qscale2;
101 
102  if (s->q_scale_type) qscale2 = ff_mpeg2_non_linear_qscale[qscale];
103  else qscale2 = qscale << 1;
104 
105  if (fdsp->fdct == ff_jpeg_fdct_islow_8 ||
106 #if CONFIG_FAANDCT
107  fdsp->fdct == ff_faandct ||
108 #endif /* CONFIG_FAANDCT */
109  fdsp->fdct == ff_jpeg_fdct_islow_10) {
110  for (i = 0; i < 64; i++) {
111  const int j = s->idsp.idct_permutation[i];
112  int64_t den = (int64_t) qscale2 * quant_matrix[j];
113  /* 16 <= qscale * quant_matrix[i] <= 7905
114  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
115  * 19952 <= x <= 249205026
116  * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
117  * 3444240 >= (1 << 36) / (x) >= 275 */
118 
119  qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
120  }
121  } else if (fdsp->fdct == ff_fdct_ifast) {
122  for (i = 0; i < 64; i++) {
123  const int j = s->idsp.idct_permutation[i];
124  int64_t den = ff_aanscales[i] * (int64_t) qscale2 * quant_matrix[j];
125  /* 16 <= qscale * quant_matrix[i] <= 7905
126  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
127  * 19952 <= x <= 249205026
128  * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
129  * 3444240 >= (1 << 36) / (x) >= 275 */
130 
131  qmat[qscale][i] = (int)((UINT64_C(2) << (QMAT_SHIFT + 14)) / den);
132  }
133  } else {
134  for (i = 0; i < 64; i++) {
135  const int j = s->idsp.idct_permutation[i];
136  int64_t den = (int64_t) qscale2 * quant_matrix[j];
137  /* We can safely suppose that 16 <= quant_matrix[i] <= 255
138  * Assume x = qscale * quant_matrix[i]
139  * So 16 <= x <= 7905
140  * so (1 << 19) / 16 >= (1 << 19) / (x) >= (1 << 19) / 7905
141  * so 32768 >= (1 << 19) / (x) >= 67 */
142  qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
143  //qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) /
144  // (qscale * quant_matrix[i]);
145  qmat16[qscale][0][i] = (2 << QMAT_SHIFT_MMX) / den;
146 
147  if (qmat16[qscale][0][i] == 0 ||
148  qmat16[qscale][0][i] == 128 * 256)
149  qmat16[qscale][0][i] = 128 * 256 - 1;
150  qmat16[qscale][1][i] =
151  ROUNDED_DIV(bias * (1<<(16 - QUANT_BIAS_SHIFT)),
152  qmat16[qscale][0][i]);
153  }
154  }
155 
156  for (i = intra; i < 64; i++) {
157  int64_t max = 8191;
158  if (fdsp->fdct == ff_fdct_ifast) {
159  max = (8191LL * ff_aanscales[i]) >> 14;
160  }
161  while (((max * qmat[qscale][i]) >> shift) > INT_MAX) {
162  shift++;
163  }
164  }
165  }
166  if (shift) {
167  av_log(s->avctx, AV_LOG_INFO,
168  "Warning, QMAT_SHIFT is larger than %d, overflows possible\n",
169  QMAT_SHIFT - shift);
170  }
171 }
172 
173 static inline void update_qscale(MpegEncContext *s)
174 {
175  if (s->q_scale_type == 1 && 0) {
176  int i;
177  int bestdiff=INT_MAX;
178  int best = 1;
179 
180  for (i = 0 ; i<FF_ARRAY_ELEMS(ff_mpeg2_non_linear_qscale); i++) {
181  int diff = FFABS((ff_mpeg2_non_linear_qscale[i]<<(FF_LAMBDA_SHIFT + 6)) - (int)s->lambda * 139);
182  if (ff_mpeg2_non_linear_qscale[i] < s->avctx->qmin ||
183  (ff_mpeg2_non_linear_qscale[i] > s->avctx->qmax && !s->vbv_ignore_qmax))
184  continue;
185  if (diff < bestdiff) {
186  bestdiff = diff;
187  best = i;
188  }
189  }
190  s->qscale = best;
191  } else {
192  s->qscale = (s->lambda * 139 + FF_LAMBDA_SCALE * 64) >>
193  (FF_LAMBDA_SHIFT + 7);
194  s->qscale = av_clip(s->qscale, s->avctx->qmin, s->vbv_ignore_qmax ? 31 : s->avctx->qmax);
195  }
196 
197  s->lambda2 = (s->lambda * s->lambda + FF_LAMBDA_SCALE / 2) >>
198  FF_LAMBDA_SHIFT;
199 }
200 
201 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix)
202 {
203  int i;
204 
205  if (matrix) {
206  put_bits(pb, 1, 1);
207  for (i = 0; i < 64; i++) {
208  put_bits(pb, 8, matrix[ff_zigzag_direct[i]]);
209  }
210  } else
211  put_bits(pb, 1, 0);
212 }
213 
214 /**
215  * init s->current_picture.qscale_table from s->lambda_table
216  */
217 void ff_init_qscale_tab(MpegEncContext *s)
218 {
219  int8_t * const qscale_table = s->current_picture.qscale_table;
220  int i;
221 
222  for (i = 0; i < s->mb_num; i++) {
223  unsigned int lam = s->lambda_table[s->mb_index2xy[i]];
224  int qp = (lam * 139 + FF_LAMBDA_SCALE * 64) >> (FF_LAMBDA_SHIFT + 7);
225  qscale_table[s->mb_index2xy[i]] = av_clip(qp, s->avctx->qmin,
226  s->avctx->qmax);
227  }
228 }
229 
230 static void update_duplicate_context_after_me(MpegEncContext *dst,
231  MpegEncContext *src)
232 {
233 #define COPY(a) dst->a= src->a
234  COPY(pict_type);
235  COPY(current_picture);
236  COPY(f_code);
237  COPY(b_code);
238  COPY(qscale);
239  COPY(lambda);
240  COPY(lambda2);
241  COPY(picture_in_gop_number);
242  COPY(gop_picture_number);
243  COPY(frame_pred_frame_dct); // FIXME don't set in encode_header
244  COPY(progressive_frame); // FIXME don't set in encode_header
245  COPY(partitioned_frame); // FIXME don't set in encode_header
246 #undef COPY
247 }
248 
249 /**
250  * Set the given MpegEncContext to defaults for encoding.
251  * the changed fields will not depend upon the prior state of the MpegEncContext.
252  */
253 static void mpv_encode_defaults(MpegEncContext *s)
254 {
255  int i;
256  ff_mpv_common_defaults(s);
257 
258  for (i = -16; i < 16; i++) {
259  default_fcode_tab[i + MAX_MV] = 1;
260  }
261  s->me.mv_penalty = default_mv_penalty;
262  s->fcode_tab = default_fcode_tab;
263 
264  s->input_picture_number = 0;
265  s->picture_in_gop_number = 0;
266 }
267 
268 av_cold int ff_dct_encode_init(MpegEncContext *s)
269 {
270  if (ARCH_X86)
271  ff_dct_encode_init_x86(s);
272 
273  if (CONFIG_H263_ENCODER)
274  ff_h263dsp_init(&s->h263dsp);
275  if (!s->dct_quantize)
276  s->dct_quantize = ff_dct_quantize_c;
277  if (!s->denoise_dct)
278  s->denoise_dct = denoise_dct_c;
279  s->fast_dct_quantize = s->dct_quantize;
280  if (s->avctx->trellis)
281  s->dct_quantize = dct_quantize_trellis_c;
282 
283  return 0;
284 }
285 
286 /* init video encoder */
287 av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
288 {
289  MpegEncContext *s = avctx->priv_data;
290  AVCPBProperties *cpb_props;
291  int i, ret, format_supported;
292 
293  mpv_encode_defaults(s);
294 
295  switch (avctx->codec_id) {
296  case AV_CODEC_ID_MPEG2VIDEO:
297  if (avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
298  avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
299  av_log(avctx, AV_LOG_ERROR,
300  "only YUV420 and YUV422 are supported\n");
301  return -1;
302  }
303  break;
304  case AV_CODEC_ID_MJPEG:
305  case AV_CODEC_ID_AMV:
306  format_supported = 0;
307  /* JPEG color space */
308  if (avctx->pix_fmt == AV_PIX_FMT_YUVJ420P ||
309  avctx->pix_fmt == AV_PIX_FMT_YUVJ422P ||
310  avctx->pix_fmt == AV_PIX_FMT_YUVJ444P ||
311  (avctx->color_range == AVCOL_RANGE_JPEG &&
312  (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
313  avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
314  avctx->pix_fmt == AV_PIX_FMT_YUV444P)))
315  format_supported = 1;
316  /* MPEG color space */
317  else if (avctx->strict_std_compliance <= FF_COMPLIANCE_UNOFFICIAL &&
318  (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
319  avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
320  avctx->pix_fmt == AV_PIX_FMT_YUV444P))
321  format_supported = 1;
322 
323  if (!format_supported) {
324  av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
325  return -1;
326  }
327  break;
328  default:
329  if (avctx->pix_fmt != AV_PIX_FMT_YUV420P) {
330  av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
331  return -1;
332  }
333  }
334 
335  switch (avctx->pix_fmt) {
336  case AV_PIX_FMT_YUVJ444P:
337  case AV_PIX_FMT_YUV444P:
338  s->chroma_format = CHROMA_444;
339  break;
340  case AV_PIX_FMT_YUVJ422P:
341  case AV_PIX_FMT_YUV422P:
342  s->chroma_format = CHROMA_422;
343  break;
344  case AV_PIX_FMT_YUVJ420P:
345  case AV_PIX_FMT_YUV420P:
346  default:
347  s->chroma_format = CHROMA_420;
348  break;
349  }
350 
351  avctx->bits_per_raw_sample = av_clip(avctx->bits_per_raw_sample, 0, 8);
352 
353 #if FF_API_PRIVATE_OPT
354 FF_DISABLE_DEPRECATION_WARNINGS
355  if (avctx->rtp_payload_size)
356  s->rtp_payload_size = avctx->rtp_payload_size;
357  if (avctx->me_penalty_compensation)
358  s->me_penalty_compensation = avctx->me_penalty_compensation;
359  if (avctx->pre_me)
360  s->me_pre = avctx->pre_me;
361 FF_ENABLE_DEPRECATION_WARNINGS
362 #endif
363 
364  s->bit_rate = avctx->bit_rate;
365  s->width = avctx->width;
366  s->height = avctx->height;
367  if (avctx->gop_size > 600 &&
368  avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
369  av_log(avctx, AV_LOG_WARNING,
370  "keyframe interval too large!, reducing it from %d to %d\n",
371  avctx->gop_size, 600);
372  avctx->gop_size = 600;
373  }
374  s->gop_size = avctx->gop_size;
375  s->avctx = avctx;
376  if (avctx->max_b_frames > MAX_B_FRAMES) {
377  av_log(avctx, AV_LOG_ERROR, "Too many B-frames requested, maximum "
378  "is %d.\n", MAX_B_FRAMES);
379  avctx->max_b_frames = MAX_B_FRAMES;
380  }
381  s->max_b_frames = avctx->max_b_frames;
382  s->codec_id = avctx->codec->id;
383  s->strict_std_compliance = avctx->strict_std_compliance;
384  s->quarter_sample = (avctx->flags & AV_CODEC_FLAG_QPEL) != 0;
385  s->rtp_mode = !!s->rtp_payload_size;
386  s->intra_dc_precision = avctx->intra_dc_precision;
387 
388  // workaround some differences between how applications specify dc precision
389  if (s->intra_dc_precision < 0) {
390  s->intra_dc_precision += 8;
391  } else if (s->intra_dc_precision >= 8)
392  s->intra_dc_precision -= 8;
393 
394  if (s->intra_dc_precision < 0) {
395  av_log(avctx, AV_LOG_ERROR,
396  "intra dc precision must be positive, note some applications use"
397  " 0 and some 8 as base meaning 8bit, the value must not be smaller than that\n");
398  return AVERROR(EINVAL);
399  }
400 
401  if (avctx->codec_id == AV_CODEC_ID_AMV || (avctx->active_thread_type & FF_THREAD_SLICE))
402  s->huffman = 0;
403 
404  if (s->intra_dc_precision > (avctx->codec_id == AV_CODEC_ID_MPEG2VIDEO ? 3 : 0)) {
405  av_log(avctx, AV_LOG_ERROR, "intra dc precision too large\n");
406  return AVERROR(EINVAL);
407  }
408  s->user_specified_pts = AV_NOPTS_VALUE;
409 
410  if (s->gop_size <= 1) {
411  s->intra_only = 1;
412  s->gop_size = 12;
413  } else {
414  s->intra_only = 0;
415  }
416 
417  /* Fixed QSCALE */
418  s->fixed_qscale = !!(avctx->flags & AV_CODEC_FLAG_QSCALE);
419 
420  s->adaptive_quant = (s->avctx->lumi_masking ||
421  s->avctx->dark_masking ||
422  s->avctx->temporal_cplx_masking ||
423  s->avctx->spatial_cplx_masking ||
424  s->avctx->p_masking ||
425  s->border_masking ||
426  (s->mpv_flags & FF_MPV_FLAG_QP_RD)) &&
427  !s->fixed_qscale;
428 
429  s->loop_filter = !!(s->avctx->flags & AV_CODEC_FLAG_LOOP_FILTER);
430 
431  if (avctx->rc_max_rate && !avctx->rc_buffer_size) {
432  switch(avctx->codec_id) {
433  case AV_CODEC_ID_MPEG1VIDEO:
434  case AV_CODEC_ID_MPEG2VIDEO:
435  avctx->rc_buffer_size = FFMAX(avctx->rc_max_rate, 15000000) * 112LL / 15000000 * 16384;
436  break;
437  case AV_CODEC_ID_MPEG4:
438  case AV_CODEC_ID_MSMPEG4V1:
439  case AV_CODEC_ID_MSMPEG4V2:
440  case AV_CODEC_ID_MSMPEG4V3:
441  if (avctx->rc_max_rate >= 15000000) {
442  avctx->rc_buffer_size = 320 + (avctx->rc_max_rate - 15000000LL) * (760-320) / (38400000 - 15000000);
443  } else if(avctx->rc_max_rate >= 2000000) {
444  avctx->rc_buffer_size = 80 + (avctx->rc_max_rate - 2000000LL) * (320- 80) / (15000000 - 2000000);
445  } else if(avctx->rc_max_rate >= 384000) {
446  avctx->rc_buffer_size = 40 + (avctx->rc_max_rate - 384000LL) * ( 80- 40) / ( 2000000 - 384000);
447  } else
448  avctx->rc_buffer_size = 40;
449  avctx->rc_buffer_size *= 16384;
450  break;
451  }
452  if (avctx->rc_buffer_size) {
453  av_log(avctx, AV_LOG_INFO, "Automatically choosing VBV buffer size of %d kbyte\n", avctx->rc_buffer_size/8192);
454  }
455  }
456 
457  if ((!avctx->rc_max_rate) != (!avctx->rc_buffer_size)) {
458  av_log(avctx, AV_LOG_ERROR, "Either both buffer size and max rate or neither must be specified\n");
459  return -1;
460  }
461 
462  if (avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate) {
463  av_log(avctx, AV_LOG_INFO,
464  "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
465  }
466 
467  if (avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate) {
468  av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
469  return -1;
470  }
471 
472  if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
473  av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
474  return -1;
475  }
476 
477  if (avctx->rc_max_rate &&
478  avctx->rc_max_rate == avctx->bit_rate &&
479  avctx->rc_max_rate != avctx->rc_min_rate) {
480  av_log(avctx, AV_LOG_INFO,
481  "impossible bitrate constraints, this will fail\n");
482  }
483 
484  if (avctx->rc_buffer_size &&
485  avctx->bit_rate * (int64_t)avctx->time_base.num >
486  avctx->rc_buffer_size * (int64_t)avctx->time_base.den) {
487  av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
488  return -1;
489  }
490 
491  if (!s->fixed_qscale &&
492  avctx->bit_rate * av_q2d(avctx->time_base) >
493  avctx->bit_rate_tolerance) {
494  av_log(avctx, AV_LOG_WARNING,
495  "bitrate tolerance %d too small for bitrate %"PRId64", overriding\n", avctx->bit_rate_tolerance, avctx->bit_rate);
496  avctx->bit_rate_tolerance = 5 * avctx->bit_rate * av_q2d(avctx->time_base);
497  }
498 
499  if (s->avctx->rc_max_rate &&
500  s->avctx->rc_min_rate == s->avctx->rc_max_rate &&
501  (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
502  s->codec_id == AV_CODEC_ID_MPEG2VIDEO) &&
503  90000LL * (avctx->rc_buffer_size - 1) >
504  s->avctx->rc_max_rate * 0xFFFFLL) {
505  av_log(avctx, AV_LOG_INFO,
506  "Warning vbv_delay will be set to 0xFFFF (=VBR) as the "
507  "specified vbv buffer is too large for the given bitrate!\n");
508  }
509 
510  if ((s->avctx->flags & AV_CODEC_FLAG_4MV) && s->codec_id != AV_CODEC_ID_MPEG4 &&
511  s->codec_id != AV_CODEC_ID_H263 && s->codec_id != AV_CODEC_ID_H263P &&
512  s->codec_id != AV_CODEC_ID_FLV1) {
513  av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
514  return -1;
515  }
516 
517  if (s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE) {
518  av_log(avctx, AV_LOG_ERROR,
519  "OBMC is only supported with simple mb decision\n");
520  return -1;
521  }
522 
523  if (s->quarter_sample && s->codec_id != AV_CODEC_ID_MPEG4) {
524  av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
525  return -1;
526  }
527 
528  if (s->max_b_frames &&
529  s->codec_id != AV_CODEC_ID_MPEG4 &&
530  s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
531  s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
532  av_log(avctx, AV_LOG_ERROR, "B-frames not supported by codec\n");
533  return -1;
534  }
535  if (s->max_b_frames < 0) {
536  av_log(avctx, AV_LOG_ERROR,
537  "max b frames must be 0 or positive for mpegvideo based encoders\n");
538  return -1;
539  }
540 
541  if ((s->codec_id == AV_CODEC_ID_MPEG4 ||
542  s->codec_id == AV_CODEC_ID_H263 ||
543  s->codec_id == AV_CODEC_ID_H263P) &&
544  (avctx->sample_aspect_ratio.num > 255 ||
545  avctx->sample_aspect_ratio.den > 255)) {
546  av_log(avctx, AV_LOG_WARNING,
547  "Invalid pixel aspect ratio %i/%i, limit is 255/255 reducing\n",
548  avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
549  av_reduce(&avctx->sample_aspect_ratio.num, &avctx->sample_aspect_ratio.den,
550  avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den, 255);
551  }
552 
553  if ((s->codec_id == AV_CODEC_ID_H263 ||
554  s->codec_id == AV_CODEC_ID_H263P) &&
555  (avctx->width > 2048 ||
556  avctx->height > 1152 )) {
557  av_log(avctx, AV_LOG_ERROR, "H.263 does not support resolutions above 2048x1152\n");
558  return -1;
559  }
560  if ((s->codec_id == AV_CODEC_ID_H263 ||
561  s->codec_id == AV_CODEC_ID_H263P) &&
562  ((avctx->width &3) ||
563  (avctx->height&3) )) {
564  av_log(avctx, AV_LOG_ERROR, "w/h must be a multiple of 4\n");
565  return -1;
566  }
567 
568  if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO &&
569  (avctx->width > 4095 ||
570  avctx->height > 4095 )) {
571  av_log(avctx, AV_LOG_ERROR, "MPEG-1 does not support resolutions above 4095x4095\n");
572  return -1;
573  }
574 
575  if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO &&
576  (avctx->width > 16383 ||
577  avctx->height > 16383 )) {
578  av_log(avctx, AV_LOG_ERROR, "MPEG-2 does not support resolutions above 16383x16383\n");
579  return -1;
580  }
581 
582  if (s->codec_id == AV_CODEC_ID_RV10 &&
583  (avctx->width &15 ||
584  avctx->height&15 )) {
585  av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 16\n");
586  return AVERROR(EINVAL);
587  }
588 
589  if (s->codec_id == AV_CODEC_ID_RV20 &&
590  (avctx->width &3 ||
591  avctx->height&3 )) {
592  av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 4\n");
593  return AVERROR(EINVAL);
594  }
595 
596  if ((s->codec_id == AV_CODEC_ID_WMV1 ||
597  s->codec_id == AV_CODEC_ID_WMV2) &&
598  avctx->width & 1) {
599  av_log(avctx, AV_LOG_ERROR, "width must be multiple of 2\n");
600  return -1;
601  }
602 
603  if ((s->avctx->flags & (AV_CODEC_FLAG_INTERLACED_DCT | AV_CODEC_FLAG_INTERLACED_ME)) &&
604  s->codec_id != AV_CODEC_ID_MPEG4 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
605  av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
606  return -1;
607  }
608 
609 #if FF_API_PRIVATE_OPT
610  FF_DISABLE_DEPRECATION_WARNINGS
611  if (avctx->mpeg_quant)
612  s->mpeg_quant = avctx->mpeg_quant;
613  FF_ENABLE_DEPRECATION_WARNINGS
614 #endif
615 
616  // FIXME mpeg2 uses that too
617  if (s->mpeg_quant && ( s->codec_id != AV_CODEC_ID_MPEG4
618  && s->codec_id != AV_CODEC_ID_MPEG2VIDEO)) {
619  av_log(avctx, AV_LOG_ERROR,
620  "mpeg2 style quantization not supported by codec\n");
621  return -1;
622  }
623 
624  if ((s->mpv_flags & FF_MPV_FLAG_CBP_RD) && !avctx->trellis) {
625  av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
626  return -1;
627  }
628 
629  if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
630  s->avctx->mb_decision != FF_MB_DECISION_RD) {
631  av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
632  return -1;
633  }
634 
635  if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
636  (s->codec_id == AV_CODEC_ID_AMV ||
637  s->codec_id == AV_CODEC_ID_MJPEG)) {
638  // Used to produce garbage with MJPEG.
639  av_log(avctx, AV_LOG_ERROR,
640  "QP RD is no longer compatible with MJPEG or AMV\n");
641  return -1;
642  }
643 
644 #if FF_API_PRIVATE_OPT
645 FF_DISABLE_DEPRECATION_WARNINGS
646  if (avctx->scenechange_threshold)
647  s->scenechange_threshold = avctx->scenechange_threshold;
648 FF_ENABLE_DEPRECATION_WARNINGS
649 #endif
650 
651  if (s->scenechange_threshold < 1000000000 &&
652  (s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)) {
653  av_log(avctx, AV_LOG_ERROR,
654  "closed gop with scene change detection are not supported yet, "
655  "set threshold to 1000000000\n");
656  return -1;
657  }
658 
659  if (s->avctx->flags & AV_CODEC_FLAG_LOW_DELAY) {
660  if (s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
661  s->strict_std_compliance >= FF_COMPLIANCE_NORMAL) {
662  av_log(avctx, AV_LOG_ERROR,
663  "low delay forcing is only available for mpeg2, "
664  "set strict_std_compliance to 'unofficial' or lower in order to allow it\n");
665  return -1;
666  }
667  if (s->max_b_frames != 0) {
668  av_log(avctx, AV_LOG_ERROR,
669  "B-frames cannot be used with low delay\n");
670  return -1;
671  }
672  }
673 
674  if (s->q_scale_type == 1) {
675  if (avctx->qmax > 28) {
676  av_log(avctx, AV_LOG_ERROR,
677  "non linear quant only supports qmax <= 28 currently\n");
678  return -1;
679  }
680  }
681 
682  if (avctx->slices > 1 &&
683  (avctx->codec_id == AV_CODEC_ID_FLV1 || avctx->codec_id == AV_CODEC_ID_H261)) {
684  av_log(avctx, AV_LOG_ERROR, "Multiple slices are not supported by this codec\n");
685  return AVERROR(EINVAL);
686  }
687 
688  if (s->avctx->thread_count > 1 &&
689  s->codec_id != AV_CODEC_ID_MPEG4 &&
690  s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
691  s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
692  s->codec_id != AV_CODEC_ID_MJPEG &&
693  (s->codec_id != AV_CODEC_ID_H263P)) {
694  av_log(avctx, AV_LOG_ERROR,
695  "multi threaded encoding not supported by codec\n");
696  return -1;
697  }
698 
699  if (s->avctx->thread_count < 1) {
700  av_log(avctx, AV_LOG_ERROR,
701  "automatic thread number detection not supported by codec, "
702  "patch welcome\n");
703  return -1;
704  }
705 
706  if (!avctx->time_base.den || !avctx->time_base.num) {
707  av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
708  return -1;
709  }
710 
711 #if FF_API_PRIVATE_OPT
712 FF_DISABLE_DEPRECATION_WARNINGS
713  if (avctx->b_frame_strategy)
714  s->b_frame_strategy = avctx->b_frame_strategy;
715  if (avctx->b_sensitivity != 40)
716  s->b_sensitivity = avctx->b_sensitivity;
717 FF_ENABLE_DEPRECATION_WARNINGS
718 #endif
719 
720  if (s->b_frame_strategy && (avctx->flags & AV_CODEC_FLAG_PASS2)) {
721  av_log(avctx, AV_LOG_INFO,
722  "notice: b_frame_strategy only affects the first pass\n");
723  s->b_frame_strategy = 0;
724  }
725 
726  i = av_gcd(avctx->time_base.den, avctx->time_base.num);
727  if (i > 1) {
728  av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
729  avctx->time_base.den /= i;
730  avctx->time_base.num /= i;
731  //return -1;
732  }
733 
734  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) {
735  // (a + x * 3 / 8) / x
736  s->intra_quant_bias = 3 << (QUANT_BIAS_SHIFT - 3);
737  s->inter_quant_bias = 0;
738  } else {
739  s->intra_quant_bias = 0;
740  // (a - x / 4) / x
741  s->inter_quant_bias = -(1 << (QUANT_BIAS_SHIFT - 2));
742  }
743 
744  if (avctx->qmin > avctx->qmax || avctx->qmin <= 0) {
745  av_log(avctx, AV_LOG_ERROR, "qmin and or qmax are invalid, they must be 0 < min <= max\n");
746  return AVERROR(EINVAL);
747  }
748 
749  av_log(avctx, AV_LOG_DEBUG, "intra_quant_bias = %d inter_quant_bias = %d\n",s->intra_quant_bias,s->inter_quant_bias);
750 
751  if (avctx->codec_id == AV_CODEC_ID_MPEG4 &&
752  s->avctx->time_base.den > (1 << 16) - 1) {
753  av_log(avctx, AV_LOG_ERROR,
754  "timebase %d/%d not supported by MPEG 4 standard, "
755  "the maximum admitted value for the timebase denominator "
756  "is %d\n", s->avctx->time_base.num, s->avctx->time_base.den,
757  (1 << 16) - 1);
758  return -1;
759  }
760  s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
761 
762  switch (avctx->codec->id) {
763  case AV_CODEC_ID_MPEG1VIDEO:
764  s->out_format = FMT_MPEG1;
765  s->low_delay = !!(s->avctx->flags & AV_CODEC_FLAG_LOW_DELAY);
766  avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
767  break;
768  case AV_CODEC_ID_MPEG2VIDEO:
769  s->out_format = FMT_MPEG1;
770  s->low_delay = !!(s->avctx->flags & AV_CODEC_FLAG_LOW_DELAY);
771  avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
772  s->rtp_mode = 1;
773  break;
774  case AV_CODEC_ID_MJPEG:
775  case AV_CODEC_ID_AMV:
776  s->out_format = FMT_MJPEG;
777  s->intra_only = 1; /* force intra only for jpeg */
778  if (!CONFIG_MJPEG_ENCODER ||
779  ff_mjpeg_encode_init(s) < 0)
780  return -1;
781  avctx->delay = 0;
782  s->low_delay = 1;
783  break;
784  case AV_CODEC_ID_H261:
785  if (!CONFIG_H261_ENCODER)
786  return -1;
787  if (ff_h261_get_picture_format(s->width, s->height) < 0) {
788  av_log(avctx, AV_LOG_ERROR,
789  "The specified picture size of %dx%d is not valid for the "
790  "H.261 codec.\nValid sizes are 176x144, 352x288\n",
791  s->width, s->height);
792  return -1;
793  }
794  s->out_format = FMT_H261;
795  avctx->delay = 0;
796  s->low_delay = 1;
797  s->rtp_mode = 0; /* Sliced encoding not supported */
798  break;
799  case AV_CODEC_ID_H263:
800  if (!CONFIG_H263_ENCODER)
801  return -1;
802  if (ff_match_2uint16(ff_h263_format, FF_ARRAY_ELEMS(ff_h263_format),
803  s->width, s->height) == 8) {
804  av_log(avctx, AV_LOG_ERROR,
805  "The specified picture size of %dx%d is not valid for "
806  "the H.263 codec.\nValid sizes are 128x96, 176x144, "
807  "352x288, 704x576, and 1408x1152. "
808  "Try H.263+.\n", s->width, s->height);
809  return -1;
810  }
811  s->out_format = FMT_H263;
812  avctx->delay = 0;
813  s->low_delay = 1;
814  break;
815  case AV_CODEC_ID_H263P:
816  s->out_format = FMT_H263;
817  s->h263_plus = 1;
818  /* Fx */
819  s->h263_aic = (avctx->flags & AV_CODEC_FLAG_AC_PRED) ? 1 : 0;
820  s->modified_quant = s->h263_aic;
821  s->loop_filter = (avctx->flags & AV_CODEC_FLAG_LOOP_FILTER) ? 1 : 0;
822  s->unrestricted_mv = s->obmc || s->loop_filter || s->umvplus;
823 
824  /* /Fx */
825  /* These are just to be sure */
826  avctx->delay = 0;
827  s->low_delay = 1;
828  break;
829  case AV_CODEC_ID_FLV1:
830  s->out_format = FMT_H263;
831  s->h263_flv = 2; /* format = 1; 11-bit codes */
832  s->unrestricted_mv = 1;
833  s->rtp_mode = 0; /* don't allow GOB */
834  avctx->delay = 0;
835  s->low_delay = 1;
836  break;
837  case AV_CODEC_ID_RV10:
838  s->out_format = FMT_H263;
839  avctx->delay = 0;
840  s->low_delay = 1;
841  break;
842  case AV_CODEC_ID_RV20:
843  s->out_format = FMT_H263;
844  avctx->delay = 0;
845  s->low_delay = 1;
846  s->modified_quant = 1;
847  s->h263_aic = 1;
848  s->h263_plus = 1;
849  s->loop_filter = 1;
850  s->unrestricted_mv = 0;
851  break;
852  case AV_CODEC_ID_MPEG4:
853  s->out_format = FMT_H263;
854  s->h263_pred = 1;
855  s->unrestricted_mv = 1;
856  s->low_delay = s->max_b_frames ? 0 : 1;
857  avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
858  break;
859  case AV_CODEC_ID_MSMPEG4V2:
860  s->out_format = FMT_H263;
861  s->h263_pred = 1;
862  s->unrestricted_mv = 1;
863  s->msmpeg4_version = 2;
864  avctx->delay = 0;
865  s->low_delay = 1;
866  break;
867  case AV_CODEC_ID_MSMPEG4V3:
868  s->out_format = FMT_H263;
869  s->h263_pred = 1;
870  s->unrestricted_mv = 1;
871  s->msmpeg4_version = 3;
872  s->flipflop_rounding = 1;
873  avctx->delay = 0;
874  s->low_delay = 1;
875  break;
876  case AV_CODEC_ID_WMV1:
877  s->out_format = FMT_H263;
878  s->h263_pred = 1;
879  s->unrestricted_mv = 1;
880  s->msmpeg4_version = 4;
881  s->flipflop_rounding = 1;
882  avctx->delay = 0;
883  s->low_delay = 1;
884  break;
885  case AV_CODEC_ID_WMV2:
886  s->out_format = FMT_H263;
887  s->h263_pred = 1;
888  s->unrestricted_mv = 1;
889  s->msmpeg4_version = 5;
890  s->flipflop_rounding = 1;
891  avctx->delay = 0;
892  s->low_delay = 1;
893  break;
894  default:
895  return -1;
896  }
897 
898 #if FF_API_PRIVATE_OPT
899  FF_DISABLE_DEPRECATION_WARNINGS
900  if (avctx->noise_reduction)
901  s->noise_reduction = avctx->noise_reduction;
902  FF_ENABLE_DEPRECATION_WARNINGS
903 #endif
904 
905  avctx->has_b_frames = !s->low_delay;
906 
907  s->encoding = 1;
908 
909  s->progressive_frame =
910  s->progressive_sequence = !(avctx->flags & (AV_CODEC_FLAG_INTERLACED_DCT |
911  AV_CODEC_FLAG_INTERLACED_ME) ||
912  s->alternate_scan);
913 
914  /* init */
915  ff_mpv_idct_init(s);
916  if (ff_mpv_common_init(s) < 0)
917  return -1;
918 
919  ff_fdctdsp_init(&s->fdsp, avctx);
920  ff_me_cmp_init(&s->mecc, avctx);
921  ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
922  ff_pixblockdsp_init(&s->pdsp, avctx);
923  ff_qpeldsp_init(&s->qdsp);
924 
925  if (s->msmpeg4_version) {
926  FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
927  2 * 2 * (MAX_LEVEL + 1) *
928  (MAX_RUN + 1) * 2 * sizeof(int), fail);
929  }
930  FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
931 
932  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix, 64 * 32 * sizeof(int), fail);
933  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix, 64 * 32 * sizeof(int), fail);
934  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix, 64 * 32 * sizeof(int), fail);
935  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
936  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
937  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
938  FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture,
939  MAX_PICTURE_COUNT * sizeof(Picture *), fail);
940  FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture,
941  MAX_PICTURE_COUNT * sizeof(Picture *), fail);
942 
943 
944  if (s->noise_reduction) {
945  FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset,
946  2 * 64 * sizeof(uint16_t), fail);
947  }
948 
949  ff_dct_encode_init(s);
950 
951  if ((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
952  s->chroma_qscale_table = ff_h263_chroma_qscale_table;
953 
954  if (s->slice_context_count > 1) {
955  s->rtp_mode = 1;
956 
957  if (avctx->codec_id == AV_CODEC_ID_H263P)
958  s->h263_slice_structured = 1;
959  }
960 
961  s->quant_precision = 5;
962 
963 #if FF_API_PRIVATE_OPT
964 FF_DISABLE_DEPRECATION_WARNINGS
965  if (avctx->frame_skip_threshold)
966  s->frame_skip_threshold = avctx->frame_skip_threshold;
967  if (avctx->frame_skip_factor)
968  s->frame_skip_factor = avctx->frame_skip_factor;
969  if (avctx->frame_skip_exp)
970  s->frame_skip_exp = avctx->frame_skip_exp;
971  if (avctx->frame_skip_cmp != FF_CMP_DCTMAX)
972  s->frame_skip_cmp = avctx->frame_skip_cmp;
973 FF_ENABLE_DEPRECATION_WARNINGS
974 #endif
975 
976  ff_set_cmp(&s->mecc, s->mecc.ildct_cmp, s->avctx->ildct_cmp);
977  ff_set_cmp(&s->mecc, s->mecc.frame_skip_cmp, s->frame_skip_cmp);
978 
979  if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
980  ff_h261_encode_init(s);
981  if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
982  ff_h263_encode_init(s);
983  if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
984  if ((ret = ff_msmpeg4_encode_init(s)) < 0)
985  return ret;
986  if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
987  && s->out_format == FMT_MPEG1)
988  ff_mpeg1_encode_init(s);
989 
990  /* init q matrix */
991  for (i = 0; i < 64; i++) {
992  int j = s->idsp.idct_permutation[i];
993  if (CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4 &&
994  s->mpeg_quant) {
995  s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
996  s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
997  } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
998  s->intra_matrix[j] =
999  s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1000  } else {
1001  /* MPEG-1/2 */
1002  s->chroma_intra_matrix[j] =
1003  s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
1004  s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1005  }
1006  if (s->avctx->intra_matrix)
1007  s->intra_matrix[j] = s->avctx->intra_matrix[i];
1008  if (s->avctx->inter_matrix)
1009  s->inter_matrix[j] = s->avctx->inter_matrix[i];
1010  }
1011 
1012  /* precompute matrix */
1013  /* for mjpeg, we do include qscale in the matrix */
1014  if (s->out_format != FMT_MJPEG) {
1015  ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
1016  s->intra_matrix, s->intra_quant_bias, avctx->qmin,
1017  31, 1);
1018  ff_convert_matrix(s, s->q_inter_matrix, s->q_inter_matrix16,
1019  s->inter_matrix, s->inter_quant_bias, avctx->qmin,
1020  31, 0);
1021  }
1022 
1023  if (ff_rate_control_init(s) < 0)
1024  return -1;
1025 
1026 #if FF_API_PRIVATE_OPT
1027  FF_DISABLE_DEPRECATION_WARNINGS
1028  if (avctx->brd_scale)
1029  s->brd_scale = avctx->brd_scale;
1030 
1031  if (avctx->prediction_method)
1032  s->pred = avctx->prediction_method + 1;
1033  FF_ENABLE_DEPRECATION_WARNINGS
1034 #endif
1035 
1036  if (s->b_frame_strategy == 2) {
1037  for (i = 0; i < s->max_b_frames + 2; i++) {
1038  s->tmp_frames[i] = av_frame_alloc();
1039  if (!s->tmp_frames[i])
1040  return AVERROR(ENOMEM);
1041 
1042  s->tmp_frames[i]->format = AV_PIX_FMT_YUV420P;
1043  s->tmp_frames[i]->width = s->width >> s->brd_scale;
1044  s->tmp_frames[i]->height = s->height >> s->brd_scale;
1045 
1046  ret = av_frame_get_buffer(s->tmp_frames[i], 32);
1047  if (ret < 0)
1048  return ret;
1049  }
1050  }
1051 
1052  cpb_props = ff_add_cpb_side_data(avctx);
1053  if (!cpb_props)
1054  return AVERROR(ENOMEM);
1055  cpb_props->max_bitrate = avctx->rc_max_rate;
1056  cpb_props->min_bitrate = avctx->rc_min_rate;
1057  cpb_props->avg_bitrate = avctx->bit_rate;
1058  cpb_props->buffer_size = avctx->rc_buffer_size;
1059 
1060  return 0;
1061 fail:
1062  ff_mpv_encode_end(avctx);
1063  return AVERROR_UNKNOWN;
1064 }
1065 
1066 av_cold int ff_mpv_encode_end(AVCodecContext *avctx)
1067 {
1068  MpegEncContext *s = avctx->priv_data;
1069  int i;
1070 
1071  ff_rate_control_uninit(s);
1072 
1073  ff_mpv_common_end(s);
1074  if (CONFIG_MJPEG_ENCODER &&
1075  s->out_format == FMT_MJPEG)
1076  ff_mjpeg_encode_close(s);
1077 
1078  av_freep(&avctx->extradata);
1079 
1080  for (i = 0; i < FF_ARRAY_ELEMS(s->tmp_frames); i++)
1081  av_frame_free(&s->tmp_frames[i]);
1082 
1083  ff_free_picture_tables(&s->new_picture);
1084  ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1085 
1086  av_freep(&s->avctx->stats_out);
1087  av_freep(&s->ac_stats);
1088 
1089  if(s->q_chroma_intra_matrix != s->q_intra_matrix ) av_freep(&s->q_chroma_intra_matrix);
1090  if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
1091  s->q_chroma_intra_matrix= NULL;
1092  s->q_chroma_intra_matrix16= NULL;
1093  av_freep(&s->q_intra_matrix);
1094  av_freep(&s->q_inter_matrix);
1095  av_freep(&s->q_intra_matrix16);
1096  av_freep(&s->q_inter_matrix16);
1097  av_freep(&s->input_picture);
1098  av_freep(&s->reordered_input_picture);
1099  av_freep(&s->dct_offset);
1100 
1101  return 0;
1102 }
1103 
1104 static int get_sae(uint8_t *src, int ref, int stride)
1105 {
1106  int x,y;
1107  int acc = 0;
1108 
1109  for (y = 0; y < 16; y++) {
1110  for (x = 0; x < 16; x++) {
1111  acc += FFABS(src[x + y * stride] - ref);
1112  }
1113  }
1114 
1115  return acc;
1116 }
1117 
1118 static int get_intra_count(MpegEncContext *s, uint8_t *src,
1119  uint8_t *ref, int stride)
1120 {
1121  int x, y, w, h;
1122  int acc = 0;
1123 
1124  w = s->width & ~15;
1125  h = s->height & ~15;
1126 
1127  for (y = 0; y < h; y += 16) {
1128  for (x = 0; x < w; x += 16) {
1129  int offset = x + y * stride;
1130  int sad = s->mecc.sad[0](NULL, src + offset, ref + offset,
1131  stride, 16);
1132  int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8;
1133  int sae = get_sae(src + offset, mean, stride);
1134 
1135  acc += sae + 500 < sad;
1136  }
1137  }
1138  return acc;
1139 }
1140 
1141 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
1142 {
1143  return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 1,
1144  s->chroma_x_shift, s->chroma_y_shift, s->out_format,
1145  s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
1146  &s->linesize, &s->uvlinesize);
1147 }
1148 
1149 static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
1150 {
1151  Picture *pic = NULL;
1152  int64_t pts;
1153  int i, display_picture_number = 0, ret;
1154  int encoding_delay = s->max_b_frames ? s->max_b_frames
1155  : (s->low_delay ? 0 : 1);
1156  int flush_offset = 1;
1157  int direct = 1;
1158 
1159  if (pic_arg) {
1160  pts = pic_arg->pts;
1161  display_picture_number = s->input_picture_number++;
1162 
1163  if (pts != AV_NOPTS_VALUE) {
1164  if (s->user_specified_pts != AV_NOPTS_VALUE) {
1165  int64_t last = s->user_specified_pts;
1166 
1167  if (pts <= last) {
1168  av_log(s->avctx, AV_LOG_ERROR,
1169  "Invalid pts (%"PRId64") <= last (%"PRId64")\n",
1170  pts, last);
1171  return AVERROR(EINVAL);
1172  }
1173 
1174  if (!s->low_delay && display_picture_number == 1)
1175  s->dts_delta = pts - last;
1176  }
1177  s->user_specified_pts = pts;
1178  } else {
1179  if (s->user_specified_pts != AV_NOPTS_VALUE) {
1180  s->user_specified_pts =
1181  pts = s->user_specified_pts + 1;
1182  av_log(s->avctx, AV_LOG_INFO,
1183  "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n",
1184  pts);
1185  } else {
1186  pts = display_picture_number;
1187  }
1188  }
1189 
1190  if (!pic_arg->buf[0] ||
1191  pic_arg->linesize[0] != s->linesize ||
1192  pic_arg->linesize[1] != s->uvlinesize ||
1193  pic_arg->linesize[2] != s->uvlinesize)
1194  direct = 0;
1195  if ((s->width & 15) || (s->height & 15))
1196  direct = 0;
1197  if (((intptr_t)(pic_arg->data[0])) & (STRIDE_ALIGN-1))
1198  direct = 0;
1199  if (s->linesize & (STRIDE_ALIGN-1))
1200  direct = 0;
1201 
1202  ff_dlog(s->avctx, "%d %d %"PTRDIFF_SPECIFIER" %"PTRDIFF_SPECIFIER"\n", pic_arg->linesize[0],
1203  pic_arg->linesize[1], s->linesize, s->uvlinesize);
1204 
1205  i = ff_find_unused_picture(s->avctx, s->picture, direct);
1206  if (i < 0)
1207  return i;
1208 
1209  pic = &s->picture[i];
1210  pic->reference = 3;
1211 
1212  if (direct) {
1213  if ((ret = av_frame_ref(pic->f, pic_arg)) < 0)
1214  return ret;
1215  }
1216  ret = alloc_picture(s, pic, direct);
1217  if (ret < 0)
1218  return ret;
1219 
1220  if (!direct) {
1221  if (pic->f->data[0] + INPLACE_OFFSET == pic_arg->data[0] &&
1222  pic->f->data[1] + INPLACE_OFFSET == pic_arg->data[1] &&
1223  pic->f->data[2] + INPLACE_OFFSET == pic_arg->data[2]) {
1224  // empty
1225  } else {
1226  int h_chroma_shift, v_chroma_shift;
1227  av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1228  &h_chroma_shift,
1229  &v_chroma_shift);
1230 
1231  for (i = 0; i < 3; i++) {
1232  int src_stride = pic_arg->linesize[i];
1233  int dst_stride = i ? s->uvlinesize : s->linesize;
1234  int h_shift = i ? h_chroma_shift : 0;
1235  int v_shift = i ? v_chroma_shift : 0;
1236  int w = s->width >> h_shift;
1237  int h = s->height >> v_shift;
1238  uint8_t *src = pic_arg->data[i];
1239  uint8_t *dst = pic->f->data[i];
1240  int vpad = 16;
1241 
1242  if ( s->codec_id == AV_CODEC_ID_MPEG2VIDEO
1243  && !s->progressive_sequence
1244  && FFALIGN(s->height, 32) - s->height > 16)
1245  vpad = 32;
1246 
1247  if (!s->avctx->rc_buffer_size)
1248  dst += INPLACE_OFFSET;
1249 
1250  if (src_stride == dst_stride)
1251  memcpy(dst, src, src_stride * h);
1252  else {
1253  int h2 = h;
1254  uint8_t *dst2 = dst;
1255  while (h2--) {
1256  memcpy(dst2, src, w);
1257  dst2 += dst_stride;
1258  src += src_stride;
1259  }
1260  }
1261  if ((s->width & 15) || (s->height & (vpad-1))) {
1262  s->mpvencdsp.draw_edges(dst, dst_stride,
1263  w, h,
1264  16 >> h_shift,
1265  vpad >> v_shift,
1266  EDGE_BOTTOM);
1267  }
1268  }
1269  emms_c();
1270  }
1271  }
1272  ret = av_frame_copy_props(pic->f, pic_arg);
1273  if (ret < 0)
1274  return ret;
1275 
1276  pic->f->display_picture_number = display_picture_number;
1277  pic->f->pts = pts; // we set this here to avoid modifying pic_arg
1278  } else {
1279  /* Flushing: When we have not received enough input frames,
1280  * ensure s->input_picture[0] contains the first picture */
1281  for (flush_offset = 0; flush_offset < encoding_delay + 1; flush_offset++)
1282  if (s->input_picture[flush_offset])
1283  break;
1284 
1285  if (flush_offset <= 1)
1286  flush_offset = 1;
1287  else
1288  encoding_delay = encoding_delay - flush_offset + 1;
1289  }
1290 
1291  /* shift buffer entries */
1292  for (i = flush_offset; i < MAX_PICTURE_COUNT /*s->encoding_delay + 1*/; i++)
1293  s->input_picture[i - flush_offset] = s->input_picture[i];
1294 
1295  s->input_picture[encoding_delay] = (Picture*) pic;
1296 
1297  return 0;
1298 }
1299 
1300 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref)
1301 {
1302  int x, y, plane;
1303  int score = 0;
1304  int64_t score64 = 0;
1305 
1306  for (plane = 0; plane < 3; plane++) {
1307  const int stride = p->f->linesize[plane];
1308  const int bw = plane ? 1 : 2;
1309  for (y = 0; y < s->mb_height * bw; y++) {
1310  for (x = 0; x < s->mb_width * bw; x++) {
1311  int off = p->shared ? 0 : 16;
1312  uint8_t *dptr = p->f->data[plane] + 8 * (x + y * stride) + off;
1313  uint8_t *rptr = ref->f->data[plane] + 8 * (x + y * stride);
1314  int v = s->mecc.frame_skip_cmp[1](s, dptr, rptr, stride, 8);
1315 
1316  switch (FFABS(s->frame_skip_exp)) {
1317  case 0: score = FFMAX(score, v); break;
1318  case 1: score += FFABS(v); break;
1319  case 2: score64 += v * (int64_t)v; break;
1320  case 3: score64 += FFABS(v * (int64_t)v * v); break;
1321  case 4: score64 += (v * (int64_t)v) * (v * (int64_t)v); break;
1322  }
1323  }
1324  }
1325  }
1326  emms_c();
1327 
1328  if (score)
1329  score64 = score;
1330  if (s->frame_skip_exp < 0)
1331  score64 = pow(score64 / (double)(s->mb_width * s->mb_height),
1332  -1.0/s->frame_skip_exp);
1333 
1334  if (score64 < s->frame_skip_threshold)
1335  return 1;
1336  if (score64 < ((s->frame_skip_factor * (int64_t) s->lambda) >> 8))
1337  return 1;
1338  return 0;
1339 }
1340 
1341 static int encode_frame(AVCodecContext *c, AVFrame *frame)
1342 {
1343  AVPacket pkt = { 0 };
1344  int ret;
1345  int size = 0;
1346 
1347  av_init_packet(&pkt);
1348 
1349  ret = avcodec_send_frame(c, frame);
1350  if (ret < 0)
1351  return ret;
1352 
1353  do {
1354  ret = avcodec_receive_packet(c, &pkt);
1355  if (ret >= 0) {
1356  size += pkt.size;
1357  av_packet_unref(&pkt);
1358  } else if (ret < 0 && ret != AVERROR(EAGAIN) && ret != AVERROR_EOF)
1359  return ret;
1360  } while (ret >= 0);
1361 
1362  return size;
1363 }
1364 
1365 static int estimate_best_b_count(MpegEncContext *s)
1366 {
1367  const AVCodec *codec = avcodec_find_encoder(s->avctx->codec_id);
1368  const int scale = s->brd_scale;
1369  int width = s->width >> scale;
1370  int height = s->height >> scale;
1371  int i, j, out_size, p_lambda, b_lambda, lambda2;
1372  int64_t best_rd = INT64_MAX;
1373  int best_b_count = -1;
1374  int ret = 0;
1375 
1376  av_assert0(scale >= 0 && scale <= 3);
1377 
1378  //emms_c();
1379  //s->next_picture_ptr->quality;
1380  p_lambda = s->last_lambda_for[AV_PICTURE_TYPE_P];
1381  //p_lambda * FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
1382  b_lambda = s->last_lambda_for[AV_PICTURE_TYPE_B];
1383  if (!b_lambda) // FIXME we should do this somewhere else
1384  b_lambda = p_lambda;
1385  lambda2 = (b_lambda * b_lambda + (1 << FF_LAMBDA_SHIFT) / 2) >>
1386  FF_LAMBDA_SHIFT;
1387 
1388  for (i = 0; i < s->max_b_frames + 2; i++) {
1389  Picture pre_input, *pre_input_ptr = i ? s->input_picture[i - 1] :
1390  s->next_picture_ptr;
1391  uint8_t *data[4];
1392 
1393  if (pre_input_ptr && (!i || s->input_picture[i - 1])) {
1394  pre_input = *pre_input_ptr;
1395  memcpy(data, pre_input_ptr->f->data, sizeof(data));
1396 
1397  if (!pre_input.shared && i) {
1398  data[0] += INPLACE_OFFSET;
1399  data[1] += INPLACE_OFFSET;
1400  data[2] += INPLACE_OFFSET;
1401  }
1402 
1403  s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0],
1404  s->tmp_frames[i]->linesize[0],
1405  data[0],
1406  pre_input.f->linesize[0],
1407  width, height);
1408  s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1],
1409  s->tmp_frames[i]->linesize[1],
1410  data[1],
1411  pre_input.f->linesize[1],
1412  width >> 1, height >> 1);
1413  s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2],
1414  s->tmp_frames[i]->linesize[2],
1415  data[2],
1416  pre_input.f->linesize[2],
1417  width >> 1, height >> 1);
1418  }
1419  }
1420 
1421  for (j = 0; j < s->max_b_frames + 1; j++) {
1422  AVCodecContext *c;
1423  int64_t rd = 0;
1424 
1425  if (!s->input_picture[j])
1426  break;
1427 
1428  c = avcodec_alloc_context3(NULL);
1429  if (!c)
1430  return AVERROR(ENOMEM);
1431 
1432  c->width = width;
1433  c->height = height;
1434  c->flags = AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_PSNR;
1435  c->flags |= s->avctx->flags & AV_CODEC_FLAG_QPEL;
1436  c->mb_decision = s->avctx->mb_decision;
1437  c->me_cmp = s->avctx->me_cmp;
1438  c->mb_cmp = s->avctx->mb_cmp;
1439  c->me_sub_cmp = s->avctx->me_sub_cmp;
1440  c->pix_fmt = AV_PIX_FMT_YUV420P;
1441  c->time_base = s->avctx->time_base;
1442  c->max_b_frames = s->max_b_frames;
1443 
1444  ret = avcodec_open2(c, codec, NULL);
1445  if (ret < 0)
1446  goto fail;
1447 
1448  s->tmp_frames[0]->pict_type = AV_PICTURE_TYPE_I;
1449  s->tmp_frames[0]->quality = 1 * FF_QP2LAMBDA;
1450 
1451  out_size = encode_frame(c, s->tmp_frames[0]);
1452  if (out_size < 0) {
1453  ret = out_size;
1454  goto fail;
1455  }
1456 
1457  //rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1458 
1459  for (i = 0; i < s->max_b_frames + 1; i++) {
1460  int is_p = i % (j + 1) == j || i == s->max_b_frames;
1461 
1462  s->tmp_frames[i + 1]->pict_type = is_p ?
1463  AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1464  s->tmp_frames[i + 1]->quality = is_p ? p_lambda : b_lambda;
1465 
1466  out_size = encode_frame(c, s->tmp_frames[i + 1]);
1467  if (out_size < 0) {
1468  ret = out_size;
1469  goto fail;
1470  }
1471 
1472  rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1473  }
1474 
1475  /* get the delayed frames */
1476  out_size = encode_frame(c, NULL);
1477  if (out_size < 0) {
1478  ret = out_size;
1479  goto fail;
1480  }
1481  rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1482 
1483  rd += c->error[0] + c->error[1] + c->error[2];
1484 
1485  if (rd < best_rd) {
1486  best_rd = rd;
1487  best_b_count = j;
1488  }
1489 
1490 fail:
1491  avcodec_free_context(&c);
1492  if (ret < 0)
1493  return ret;
1494  }
1495 
1496  return best_b_count;
1497 }
1498 
1499 static int select_input_picture(MpegEncContext *s)
1500 {
1501  int i, ret;
1502 
1503  for (i = 1; i < MAX_PICTURE_COUNT; i++)
1504  s->reordered_input_picture[i - 1] = s->reordered_input_picture[i];
1505  s->reordered_input_picture[MAX_PICTURE_COUNT - 1] = NULL;
1506 
1507  /* set next picture type & ordering */
1508  if (!s->reordered_input_picture[0] && s->input_picture[0]) {
1509  if (s->frame_skip_threshold || s->frame_skip_factor) {
1510  if (s->picture_in_gop_number < s->gop_size &&
1511  s->next_picture_ptr &&
1512  skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
1513  // FIXME check that the gop check above is +-1 correct
1514  av_frame_unref(s->input_picture[0]->f);
1515 
1516  ff_vbv_update(s, 0);
1517 
1518  goto no_output_pic;
1519  }
1520  }
1521 
1522  if (/*s->picture_in_gop_number >= s->gop_size ||*/
1523  !s->next_picture_ptr || s->intra_only) {
1524  s->reordered_input_picture[0] = s->input_picture[0];
1525  s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_I;
1526  s->reordered_input_picture[0]->f->coded_picture_number =
1527  s->coded_picture_number++;
1528  } else {
1529  int b_frames = 0;
1530 
1531  if (s->avctx->flags & AV_CODEC_FLAG_PASS2) {
1532  for (i = 0; i < s->max_b_frames + 1; i++) {
1533  int pict_num = s->input_picture[0]->f->display_picture_number + i;
1534 
1535  if (pict_num >= s->rc_context.num_entries)
1536  break;
1537  if (!s->input_picture[i]) {
1538  s->rc_context.entry[pict_num - 1].new_pict_type = AV_PICTURE_TYPE_P;
1539  break;
1540  }
1541 
1542  s->input_picture[i]->f->pict_type =
1543  s->rc_context.entry[pict_num].new_pict_type;
1544  }
1545  }
1546 
1547  if (s->b_frame_strategy == 0) {
1548  b_frames = s->max_b_frames;
1549  while (b_frames && !s->input_picture[b_frames])
1550  b_frames--;
1551  } else if (s->b_frame_strategy == 1) {
1552  for (i = 1; i < s->max_b_frames + 1; i++) {
1553  if (s->input_picture[i] &&
1554  s->input_picture[i]->b_frame_score == 0) {
1555  s->input_picture[i]->b_frame_score =
1556  get_intra_count(s,
1557  s->input_picture[i ]->f->data[0],
1558  s->input_picture[i - 1]->f->data[0],
1559  s->linesize) + 1;
1560  }
1561  }
1562  for (i = 0; i < s->max_b_frames + 1; i++) {
1563  if (!s->input_picture[i] ||
1564  s->input_picture[i]->b_frame_score - 1 >
1565  s->mb_num / s->b_sensitivity)
1566  break;
1567  }
1568 
1569  b_frames = FFMAX(0, i - 1);
1570 
1571  /* reset scores */
1572  for (i = 0; i < b_frames + 1; i++) {
1573  s->input_picture[i]->b_frame_score = 0;
1574  }
1575  } else if (s->b_frame_strategy == 2) {
1576  b_frames = estimate_best_b_count(s);
1577  if (b_frames < 0)
1578  return b_frames;
1579  }
1580 
1581  emms_c();
1582 
1583  for (i = b_frames - 1; i >= 0; i--) {
1584  int type = s->input_picture[i]->f->pict_type;
1585  if (type && type != AV_PICTURE_TYPE_B)
1586  b_frames = i;
1587  }
1588  if (s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_B &&
1589  b_frames == s->max_b_frames) {
1590  av_log(s->avctx, AV_LOG_ERROR,
1591  "warning, too many B-frames in a row\n");
1592  }
1593 
1594  if (s->picture_in_gop_number + b_frames >= s->gop_size) {
1595  if ((s->mpv_flags & FF_MPV_FLAG_STRICT_GOP) &&
1596  s->gop_size > s->picture_in_gop_number) {
1597  b_frames = s->gop_size - s->picture_in_gop_number - 1;
1598  } else {
1599  if (s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)
1600  b_frames = 0;
1601  s->input_picture[b_frames]->f->pict_type = AV_PICTURE_TYPE_I;
1602  }
1603  }
1604 
1605  if ((s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) && b_frames &&
1606  s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_I)
1607  b_frames--;
1608 
1609  s->reordered_input_picture[0] = s->input_picture[b_frames];
1610  if (s->reordered_input_picture[0]->f->pict_type != AV_PICTURE_TYPE_I)
1611  s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_P;
1612  s->reordered_input_picture[0]->f->coded_picture_number =
1613  s->coded_picture_number++;
1614  for (i = 0; i < b_frames; i++) {
1615  s->reordered_input_picture[i + 1] = s->input_picture[i];
1616  s->reordered_input_picture[i + 1]->f->pict_type =
1617  AV_PICTURE_TYPE_B;
1618  s->reordered_input_picture[i + 1]->f->coded_picture_number =
1619  s->coded_picture_number++;
1620  }
1621  }
1622  }
1623 no_output_pic:
1624  ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1625 
1626  if (s->reordered_input_picture[0]) {
1627  s->reordered_input_picture[0]->reference =
1628  s->reordered_input_picture[0]->f->pict_type !=
1629  AV_PICTURE_TYPE_B ? 3 : 0;
1630 
1631  if ((ret = ff_mpeg_ref_picture(s->avctx, &s->new_picture, s->reordered_input_picture[0])))
1632  return ret;
1633 
1634  if (s->reordered_input_picture[0]->shared || s->avctx->rc_buffer_size) {
1635  // input is a shared pix, so we can't modify it -> allocate a new
1636  // one & ensure that the shared one is reuseable
1637 
1638  Picture *pic;
1639  int i = ff_find_unused_picture(s->avctx, s->picture, 0);
1640  if (i < 0)
1641  return i;
1642  pic = &s->picture[i];
1643 
1644  pic->reference = s->reordered_input_picture[0]->reference;
1645  if (alloc_picture(s, pic, 0) < 0) {
1646  return -1;
1647  }
1648 
1649  ret = av_frame_copy_props(pic->f, s->reordered_input_picture[0]->f);
1650  if (ret < 0)
1651  return ret;
1652 
1653  /* mark us unused / free shared pic */
1654  av_frame_unref(s->reordered_input_picture[0]->f);
1655  s->reordered_input_picture[0]->shared = 0;
1656 
1657  s->current_picture_ptr = pic;
1658  } else {
1659  // input is not a shared pix -> reuse buffer for current_pix
1660  s->current_picture_ptr = s->reordered_input_picture[0];
1661  for (i = 0; i < 4; i++) {
1662  s->new_picture.f->data[i] += INPLACE_OFFSET;
1663  }
1664  }
1665  ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1666  if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1667  s->current_picture_ptr)) < 0)
1668  return ret;
1669 
1670  s->picture_number = s->new_picture.f->display_picture_number;
1671  }
1672  return 0;
1673 }
1674 
1675 static void frame_end(MpegEncContext *s)
1676 {
1677  if (s->unrestricted_mv &&
1678  s->current_picture.reference &&
1679  !s->intra_only) {
1680  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
1681  int hshift = desc->log2_chroma_w;
1682  int vshift = desc->log2_chroma_h;
1683  s->mpvencdsp.draw_edges(s->current_picture.f->data[0],
1684  s->current_picture.f->linesize[0],
1685  s->h_edge_pos, s->v_edge_pos,
1686  EDGE_WIDTH, EDGE_WIDTH,
1687  EDGE_TOP | EDGE_BOTTOM);
1688  s->mpvencdsp.draw_edges(s->current_picture.f->data[1],
1689  s->current_picture.f->linesize[1],
1690  s->h_edge_pos >> hshift,
1691  s->v_edge_pos >> vshift,
1692  EDGE_WIDTH >> hshift,
1693  EDGE_WIDTH >> vshift,
1694  EDGE_TOP | EDGE_BOTTOM);
1695  s->mpvencdsp.draw_edges(s->current_picture.f->data[2],
1696  s->current_picture.f->linesize[2],
1697  s->h_edge_pos >> hshift,
1698  s->v_edge_pos >> vshift,
1699  EDGE_WIDTH >> hshift,
1700  EDGE_WIDTH >> vshift,
1701  EDGE_TOP | EDGE_BOTTOM);
1702  }
1703 
1704  emms_c();
1705 
1706  s->last_pict_type = s->pict_type;
1707  s->last_lambda_for [s->pict_type] = s->current_picture_ptr->f->quality;
1708  if (s->pict_type!= AV_PICTURE_TYPE_B)
1709  s->last_non_b_pict_type = s->pict_type;
1710 
1711 #if FF_API_CODED_FRAME
1712 FF_DISABLE_DEPRECATION_WARNINGS
1713  av_frame_unref(s->avctx->coded_frame);
1714  av_frame_copy_props(s->avctx->coded_frame, s->current_picture.f);
1715 FF_ENABLE_DEPRECATION_WARNINGS
1716 #endif
1717 #if FF_API_ERROR_FRAME
1718 FF_DISABLE_DEPRECATION_WARNINGS
1719  memcpy(s->current_picture.f->error, s->current_picture.encoding_error,
1720  sizeof(s->current_picture.encoding_error));
1721 FF_ENABLE_DEPRECATION_WARNINGS
1722 #endif
1723 }
1724 
1725 static void update_noise_reduction(MpegEncContext *s)
1726 {
1727  int intra, i;
1728 
1729  for (intra = 0; intra < 2; intra++) {
1730  if (s->dct_count[intra] > (1 << 16)) {
1731  for (i = 0; i < 64; i++) {
1732  s->dct_error_sum[intra][i] >>= 1;
1733  }
1734  s->dct_count[intra] >>= 1;
1735  }
1736 
1737  for (i = 0; i < 64; i++) {
1738  s->dct_offset[intra][i] = (s->noise_reduction *
1739  s->dct_count[intra] +
1740  s->dct_error_sum[intra][i] / 2) /
1741  (s->dct_error_sum[intra][i] + 1);
1742  }
1743  }
1744 }
1745 
1746 static int frame_start(MpegEncContext *s)
1747 {
1748  int ret;
1749 
1750  /* mark & release old frames */
1751  if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1752  s->last_picture_ptr != s->next_picture_ptr &&
1753  s->last_picture_ptr->f->buf[0]) {
1754  ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1755  }
1756 
1757  s->current_picture_ptr->f->pict_type = s->pict_type;
1758  s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1759 
1760  ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1761  if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1762  s->current_picture_ptr)) < 0)
1763  return ret;
1764 
1765  if (s->pict_type != AV_PICTURE_TYPE_B) {
1766  s->last_picture_ptr = s->next_picture_ptr;
1767  if (!s->droppable)
1768  s->next_picture_ptr = s->current_picture_ptr;
1769  }
1770 
1771  if (s->last_picture_ptr) {
1772  ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1773  if (s->last_picture_ptr->f->buf[0] &&
1774  (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1775  s->last_picture_ptr)) < 0)
1776  return ret;
1777  }
1778  if (s->next_picture_ptr) {
1779  ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1780  if (s->next_picture_ptr->f->buf[0] &&
1781  (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1782  s->next_picture_ptr)) < 0)
1783  return ret;
1784  }
1785 
1786  if (s->picture_structure!= PICT_FRAME) {
1787  int i;
1788  for (i = 0; i < 4; i++) {
1789  if (s->picture_structure == PICT_BOTTOM_FIELD) {
1790  s->current_picture.f->data[i] +=
1791  s->current_picture.f->linesize[i];
1792  }
1793  s->current_picture.f->linesize[i] *= 2;
1794  s->last_picture.f->linesize[i] *= 2;
1795  s->next_picture.f->linesize[i] *= 2;
1796  }
1797  }
1798 
1799  if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1800  s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1801  s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1802  } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1803  s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1804  s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1805  } else {
1806  s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1807  s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1808  }
1809 
1810  if (s->dct_error_sum) {
1811  av_assert2(s->noise_reduction && s->encoding);
1812  update_noise_reduction(s);
1813  }
1814 
1815  return 0;
1816 }
1817 
1818 int ff_mpv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
1819  const AVFrame *pic_arg, int *got_packet)
1820 {
1821  MpegEncContext *s = avctx->priv_data;
1822  int i, stuffing_count, ret;
1823  int context_count = s->slice_context_count;
1824 
1825  s->vbv_ignore_qmax = 0;
1826 
1827  s->picture_in_gop_number++;
1828 
1829  if (load_input_picture(s, pic_arg) < 0)
1830  return -1;
1831 
1832  if (select_input_picture(s) < 0) {
1833  return -1;
1834  }
1835 
1836  /* output? */
1837  if (s->new_picture.f->data[0]) {
1838  int growing_buffer = context_count == 1 && !pkt->data && !s->data_partitioning;
1839  int pkt_size = growing_buffer ? FFMAX(s->mb_width*s->mb_height*64+10000, avctx->internal->byte_buffer_size) - AV_INPUT_BUFFER_PADDING_SIZE
1840  :
1841  s->mb_width*s->mb_height*(MAX_MB_BYTES+100)+10000;
1842  if ((ret = ff_alloc_packet2(avctx, pkt, pkt_size, 0)) < 0)
1843  return ret;
1844  if (s->mb_info) {
1845  s->mb_info_ptr = av_packet_new_side_data(pkt,
1846  AV_PKT_DATA_H263_MB_INFO,
1847  s->mb_width*s->mb_height*12);
1848  s->prev_mb_info = s->last_mb_info = s->mb_info_size = 0;
1849  }
1850 
1851  for (i = 0; i < context_count; i++) {
1852  int start_y = s->thread_context[i]->start_mb_y;
1853  int end_y = s->thread_context[i]-> end_mb_y;
1854  int h = s->mb_height;
1855  uint8_t *start = pkt->data + (size_t)(((int64_t) pkt->size) * start_y / h);
1856  uint8_t *end = pkt->data + (size_t)(((int64_t) pkt->size) * end_y / h);
1857 
1858  init_put_bits(&s->thread_context[i]->pb, start, end - start);
1859  }
1860 
1861  s->pict_type = s->new_picture.f->pict_type;
1862  //emms_c();
1863  ret = frame_start(s);
1864  if (ret < 0)
1865  return ret;
1866 vbv_retry:
1867  ret = encode_picture(s, s->picture_number);
1868  if (growing_buffer) {
1869  av_assert0(s->pb.buf == avctx->internal->byte_buffer);
1870  pkt->data = s->pb.buf;
1871  pkt->size = avctx->internal->byte_buffer_size;
1872  }
1873  if (ret < 0)
1874  return -1;
1875 
1876 #if FF_API_STAT_BITS
1877 FF_DISABLE_DEPRECATION_WARNINGS
1878  avctx->header_bits = s->header_bits;
1879  avctx->mv_bits = s->mv_bits;
1880  avctx->misc_bits = s->misc_bits;
1881  avctx->i_tex_bits = s->i_tex_bits;
1882  avctx->p_tex_bits = s->p_tex_bits;
1883  avctx->i_count = s->i_count;
1884  // FIXME f/b_count in avctx
1885  avctx->p_count = s->mb_num - s->i_count - s->skip_count;
1886  avctx->skip_count = s->skip_count;
1887 FF_ENABLE_DEPRECATION_WARNINGS
1888 #endif
1889 
1890  frame_end(s);
1891 
1892  if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1893  ff_mjpeg_encode_picture_trailer(&s->pb, s->header_bits);
1894 
1895  if (avctx->rc_buffer_size) {
1896  RateControlContext *rcc = &s->rc_context;
1897  int max_size = FFMAX(rcc->buffer_index * avctx->rc_max_available_vbv_use, rcc->buffer_index - 500);
1898  int hq = (s->avctx->mb_decision == FF_MB_DECISION_RD || s->avctx->trellis);
1899  int min_step = hq ? 1 : (1<<(FF_LAMBDA_SHIFT + 7))/139;
1900 
1901  if (put_bits_count(&s->pb) > max_size &&
1902  s->lambda < s->lmax) {
1903  s->next_lambda = FFMAX(s->lambda + min_step, s->lambda *
1904  (s->qscale + 1) / s->qscale);
1905  if (s->adaptive_quant) {
1906  int i;
1907  for (i = 0; i < s->mb_height * s->mb_stride; i++)
1908  s->lambda_table[i] =
1909  FFMAX(s->lambda_table[i] + min_step,
1910  s->lambda_table[i] * (s->qscale + 1) /
1911  s->qscale);
1912  }
1913  s->mb_skipped = 0; // done in frame_start()
1914  // done in encode_picture() so we must undo it
1915  if (s->pict_type == AV_PICTURE_TYPE_P) {
1916  if (s->flipflop_rounding ||
1917  s->codec_id == AV_CODEC_ID_H263P ||
1918  s->codec_id == AV_CODEC_ID_MPEG4)
1919  s->no_rounding ^= 1;
1920  }
1921  if (s->pict_type != AV_PICTURE_TYPE_B) {
1922  s->time_base = s->last_time_base;
1923  s->last_non_b_time = s->time - s->pp_time;
1924  }
1925  for (i = 0; i < context_count; i++) {
1926  PutBitContext *pb = &s->thread_context[i]->pb;
1927  init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1928  }
1929  s->vbv_ignore_qmax = 1;
1930  av_log(s->avctx, AV_LOG_VERBOSE, "reencoding frame due to VBV\n");
1931  goto vbv_retry;
1932  }
1933 
1934  av_assert0(s->avctx->rc_max_rate);
1935  }
1936 
1937  if (s->avctx->flags & AV_CODEC_FLAG_PASS1)
1938  ff_write_pass1_stats(s);
1939 
1940  for (i = 0; i < 4; i++) {
1941  s->current_picture_ptr->encoding_error[i] = s->current_picture.encoding_error[i];
1942  avctx->error[i] += s->current_picture_ptr->encoding_error[i];
1943  }
1944  ff_side_data_set_encoder_stats(pkt, s->current_picture.f->quality,
1945  s->current_picture_ptr->encoding_error,
1946  (s->avctx->flags&AV_CODEC_FLAG_PSNR) ? 4 : 0,
1947  s->pict_type);
1948 
1949  if (s->avctx->flags & AV_CODEC_FLAG_PASS1)
1950  assert(put_bits_count(&s->pb) == s->header_bits + s->mv_bits +
1951  s->misc_bits + s->i_tex_bits +
1952  s->p_tex_bits);
1953  flush_put_bits(&s->pb);
1954  s->frame_bits = put_bits_count(&s->pb);
1955 
1956  stuffing_count = ff_vbv_update(s, s->frame_bits);
1957  s->stuffing_bits = 8*stuffing_count;
1958  if (stuffing_count) {
1959  if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
1960  stuffing_count + 50) {
1961  av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1962  return -1;
1963  }
1964 
1965  switch (s->codec_id) {
1966  case AV_CODEC_ID_MPEG1VIDEO:
1967  case AV_CODEC_ID_MPEG2VIDEO:
1968  while (stuffing_count--) {
1969  put_bits(&s->pb, 8, 0);
1970  }
1971  break;
1972  case AV_CODEC_ID_MPEG4:
1973  put_bits(&s->pb, 16, 0);
1974  put_bits(&s->pb, 16, 0x1C3);
1975  stuffing_count -= 4;
1976  while (stuffing_count--) {
1977  put_bits(&s->pb, 8, 0xFF);
1978  }
1979  break;
1980  default:
1981  av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1982  }
1983  flush_put_bits(&s->pb);
1984  s->frame_bits = put_bits_count(&s->pb);
1985  }
1986 
1987  /* update MPEG-1/2 vbv_delay for CBR */
1988  if (s->avctx->rc_max_rate &&
1989  s->avctx->rc_min_rate == s->avctx->rc_max_rate &&
1990  s->out_format == FMT_MPEG1 &&
1991  90000LL * (avctx->rc_buffer_size - 1) <=
1992  s->avctx->rc_max_rate * 0xFFFFLL) {
1993  AVCPBProperties *props;
1994  size_t props_size;
1995 
1996  int vbv_delay, min_delay;
1997  double inbits = s->avctx->rc_max_rate *
1998  av_q2d(s->avctx->time_base);
1999  int minbits = s->frame_bits - 8 *
2000  (s->vbv_delay_ptr - s->pb.buf - 1);
2001  double bits = s->rc_context.buffer_index + minbits - inbits;
2002 
2003  if (bits < 0)
2004  av_log(s->avctx, AV_LOG_ERROR,
2005  "Internal error, negative bits\n");
2006 
2007  av_assert1(s->repeat_first_field == 0);
2008 
2009  vbv_delay = bits * 90000 / s->avctx->rc_max_rate;
2010  min_delay = (minbits * 90000LL + s->avctx->rc_max_rate - 1) /
2011  s->avctx->rc_max_rate;
2012 
2013  vbv_delay = FFMAX(vbv_delay, min_delay);
2014 
2015  av_assert0(vbv_delay < 0xFFFF);
2016 
2017  s->vbv_delay_ptr[0] &= 0xF8;
2018  s->vbv_delay_ptr[0] |= vbv_delay >> 13;
2019  s->vbv_delay_ptr[1] = vbv_delay >> 5;
2020  s->vbv_delay_ptr[2] &= 0x07;
2021  s->vbv_delay_ptr[2] |= vbv_delay << 3;
2022 
2023  props = av_cpb_properties_alloc(&props_size);
2024  if (!props)
2025  return AVERROR(ENOMEM);
2026  props->vbv_delay = vbv_delay * 300;
2027 
2028  ret = av_packet_add_side_data(pkt, AV_PKT_DATA_CPB_PROPERTIES,
2029  (uint8_t*)props, props_size);
2030  if (ret < 0) {
2031  av_freep(&props);
2032  return ret;
2033  }
2034 
2035 #if FF_API_VBV_DELAY
2036 FF_DISABLE_DEPRECATION_WARNINGS
2037  avctx->vbv_delay = vbv_delay * 300;
2038 FF_ENABLE_DEPRECATION_WARNINGS
2039 #endif
2040  }
2041  s->total_bits += s->frame_bits;
2042 #if FF_API_STAT_BITS
2043 FF_DISABLE_DEPRECATION_WARNINGS
2044  avctx->frame_bits = s->frame_bits;
2045 FF_ENABLE_DEPRECATION_WARNINGS
2046 #endif
2047 
2048 
2049  pkt->pts = s->current_picture.f->pts;
2050  if (!s->low_delay && s->pict_type != AV_PICTURE_TYPE_B) {
2051  if (!s->current_picture.f->coded_picture_number)
2052  pkt->dts = pkt->pts - s->dts_delta;
2053  else
2054  pkt->dts = s->reordered_pts;
2055  s->reordered_pts = pkt->pts;
2056  } else
2057  pkt->dts = pkt->pts;
2058  if (s->current_picture.f->key_frame)
2059  pkt->flags |= AV_PKT_FLAG_KEY;
2060  if (s->mb_info)
2061  av_packet_shrink_side_data(pkt, AV_PKT_DATA_H263_MB_INFO, s->mb_info_size);
2062  } else {
2063  s->frame_bits = 0;
2064  }
2065 
2066  /* release non-reference frames */
2067  for (i = 0; i < MAX_PICTURE_COUNT; i++) {
2068  if (!s->picture[i].reference)
2069  ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2070  }
2071 
2072  av_assert1((s->frame_bits & 7) == 0);
2073 
2074  pkt->size = s->frame_bits / 8;
2075  *got_packet = !!pkt->size;
2076  return 0;
2077 }
2078 
2079 static inline void dct_single_coeff_elimination(MpegEncContext *s,
2080  int n, int threshold)
2081 {
2082  static const char tab[64] = {
2083  3, 2, 2, 1, 1, 1, 1, 1,
2084  1, 1, 1, 1, 1, 1, 1, 1,
2085  1, 1, 1, 1, 1, 1, 1, 1,
2086  0, 0, 0, 0, 0, 0, 0, 0,
2087  0, 0, 0, 0, 0, 0, 0, 0,
2088  0, 0, 0, 0, 0, 0, 0, 0,
2089  0, 0, 0, 0, 0, 0, 0, 0,
2090  0, 0, 0, 0, 0, 0, 0, 0
2091  };
2092  int score = 0;
2093  int run = 0;
2094  int i;
2095  int16_t *block = s->block[n];
2096  const int last_index = s->block_last_index[n];
2097  int skip_dc;
2098 
2099  if (threshold < 0) {
2100  skip_dc = 0;
2101  threshold = -threshold;
2102  } else
2103  skip_dc = 1;
2104 
2105  /* Are all we could set to zero already zero? */
2106  if (last_index <= skip_dc - 1)
2107  return;
2108 
2109  for (i = 0; i <= last_index; i++) {
2110  const int j = s->intra_scantable.permutated[i];
2111  const int level = FFABS(block[j]);
2112  if (level == 1) {
2113  if (skip_dc && i == 0)
2114  continue;
2115  score += tab[run];
2116  run = 0;
2117  } else if (level > 1) {
2118  return;
2119  } else {
2120  run++;
2121  }
2122  }
2123  if (score >= threshold)
2124  return;
2125  for (i = skip_dc; i <= last_index; i++) {
2126  const int j = s->intra_scantable.permutated[i];
2127  block[j] = 0;
2128  }
2129  if (block[0])
2130  s->block_last_index[n] = 0;
2131  else
2132  s->block_last_index[n] = -1;
2133 }
2134 
2135 static inline void clip_coeffs(MpegEncContext *s, int16_t *block,
2136  int last_index)
2137 {
2138  int i;
2139  const int maxlevel = s->max_qcoeff;
2140  const int minlevel = s->min_qcoeff;
2141  int overflow = 0;
2142 
2143  if (s->mb_intra) {
2144  i = 1; // skip clipping of intra dc
2145  } else
2146  i = 0;
2147 
2148  for (; i <= last_index; i++) {
2149  const int j = s->intra_scantable.permutated[i];
2150  int level = block[j];
2151 
2152  if (level > maxlevel) {
2153  level = maxlevel;
2154  overflow++;
2155  } else if (level < minlevel) {
2156  level = minlevel;
2157  overflow++;
2158  }
2159 
2160  block[j] = level;
2161  }
2162 
2163  if (overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
2164  av_log(s->avctx, AV_LOG_INFO,
2165  "warning, clipping %d dct coefficients to %d..%d\n",
2166  overflow, minlevel, maxlevel);
2167 }
2168 
2169 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride)
2170 {
2171  int x, y;
2172  // FIXME optimize
2173  for (y = 0; y < 8; y++) {
2174  for (x = 0; x < 8; x++) {
2175  int x2, y2;
2176  int sum = 0;
2177  int sqr = 0;
2178  int count = 0;
2179 
2180  for (y2 = FFMAX(y - 1, 0); y2 < FFMIN(8, y + 2); y2++) {
2181  for (x2= FFMAX(x - 1, 0); x2 < FFMIN(8, x + 2); x2++) {
2182  int v = ptr[x2 + y2 * stride];
2183  sum += v;
2184  sqr += v * v;
2185  count++;
2186  }
2187  }
2188  weight[x + 8 * y]= (36 * ff_sqrt(count * sqr - sum * sum)) / count;
2189  }
2190  }
2191 }
2192 
2193 static av_always_inline void encode_mb_internal(MpegEncContext *s,
2194  int motion_x, int motion_y,
2195  int mb_block_height,
2196  int mb_block_width,
2197  int mb_block_count)
2198 {
2199  int16_t weight[12][64];
2200  int16_t orig[12][64];
2201  const int mb_x = s->mb_x;
2202  const int mb_y = s->mb_y;
2203  int i;
2204  int skip_dct[12];
2205  int dct_offset = s->linesize * 8; // default for progressive frames
2206  int uv_dct_offset = s->uvlinesize * 8;
2207  uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2208  ptrdiff_t wrap_y, wrap_c;
2209 
2210  for (i = 0; i < mb_block_count; i++)
2211  skip_dct[i] = s->skipdct;
2212 
2213  if (s->adaptive_quant) {
2214  const int last_qp = s->qscale;
2215  const int mb_xy = mb_x + mb_y * s->mb_stride;
2216 
2217  s->lambda = s->lambda_table[mb_xy];
2218  update_qscale(s);
2219 
2220  if (!(s->mpv_flags & FF_MPV_FLAG_QP_RD)) {
2221  s->qscale = s->current_picture_ptr->qscale_table[mb_xy];
2222  s->dquant = s->qscale - last_qp;
2223 
2224  if (s->out_format == FMT_H263) {
2225  s->dquant = av_clip(s->dquant, -2, 2);
2226 
2227  if (s->codec_id == AV_CODEC_ID_MPEG4) {
2228  if (!s->mb_intra) {
2229  if (s->pict_type == AV_PICTURE_TYPE_B) {
2230  if (s->dquant & 1 || s->mv_dir & MV_DIRECT)
2231  s->dquant = 0;
2232  }
2233  if (s->mv_type == MV_TYPE_8X8)
2234  s->dquant = 0;
2235  }
2236  }
2237  }
2238  }
2239  ff_set_qscale(s, last_qp + s->dquant);
2240  } else if (s->mpv_flags & FF_MPV_FLAG_QP_RD)
2241  ff_set_qscale(s, s->qscale + s->dquant);
2242 
2243  wrap_y = s->linesize;
2244  wrap_c = s->uvlinesize;
2245  ptr_y = s->new_picture.f->data[0] +
2246  (mb_y * 16 * wrap_y) + mb_x * 16;
2247  ptr_cb = s->new_picture.f->data[1] +
2248  (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
2249  ptr_cr = s->new_picture.f->data[2] +
2250  (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
2251 
2252  if((mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) && s->codec_id != AV_CODEC_ID_AMV){
2253  uint8_t *ebuf = s->sc.edge_emu_buffer + 38 * wrap_y;
2254  int cw = (s->width + s->chroma_x_shift) >> s->chroma_x_shift;
2255  int ch = (s->height + s->chroma_y_shift) >> s->chroma_y_shift;
2256  s->vdsp.emulated_edge_mc(ebuf, ptr_y,
2257  wrap_y, wrap_y,
2258  16, 16, mb_x * 16, mb_y * 16,
2259  s->width, s->height);
2260  ptr_y = ebuf;
2261  s->vdsp.emulated_edge_mc(ebuf + 16 * wrap_y, ptr_cb,
2262  wrap_c, wrap_c,
2263  mb_block_width, mb_block_height,
2264  mb_x * mb_block_width, mb_y * mb_block_height,
2265  cw, ch);
2266  ptr_cb = ebuf + 16 * wrap_y;
2267  s->vdsp.emulated_edge_mc(ebuf + 16 * wrap_y + 16, ptr_cr,
2268  wrap_c, wrap_c,
2269  mb_block_width, mb_block_height,
2270  mb_x * mb_block_width, mb_y * mb_block_height,
2271  cw, ch);
2272  ptr_cr = ebuf + 16 * wrap_y + 16;
2273  }
2274 
2275  if (s->mb_intra) {
2276  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
2277  int progressive_score, interlaced_score;
2278 
2279  s->interlaced_dct = 0;
2280  progressive_score = s->mecc.ildct_cmp[4](s, ptr_y, NULL, wrap_y, 8) +
2281  s->mecc.ildct_cmp[4](s, ptr_y + wrap_y * 8,
2282  NULL, wrap_y, 8) - 400;
2283 
2284  if (progressive_score > 0) {
2285  interlaced_score = s->mecc.ildct_cmp[4](s, ptr_y,
2286  NULL, wrap_y * 2, 8) +
2287  s->mecc.ildct_cmp[4](s, ptr_y + wrap_y,
2288  NULL, wrap_y * 2, 8);
2289  if (progressive_score > interlaced_score) {
2290  s->interlaced_dct = 1;
2291 
2292  dct_offset = wrap_y;
2293  uv_dct_offset = wrap_c;
2294  wrap_y <<= 1;
2295  if (s->chroma_format == CHROMA_422 ||
2296  s->chroma_format == CHROMA_444)
2297  wrap_c <<= 1;
2298  }
2299  }
2300  }
2301 
2302  s->pdsp.get_pixels(s->block[0], ptr_y, wrap_y);
2303  s->pdsp.get_pixels(s->block[1], ptr_y + 8, wrap_y);
2304  s->pdsp.get_pixels(s->block[2], ptr_y + dct_offset, wrap_y);
2305  s->pdsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
2306 
2307  if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
2308  skip_dct[4] = 1;
2309  skip_dct[5] = 1;
2310  } else {
2311  s->pdsp.get_pixels(s->block[4], ptr_cb, wrap_c);
2312  s->pdsp.get_pixels(s->block[5], ptr_cr, wrap_c);
2313  if (!s->chroma_y_shift && s->chroma_x_shift) { /* 422 */
2314  s->pdsp.get_pixels(s->block[6], ptr_cb + uv_dct_offset, wrap_c);
2315  s->pdsp.get_pixels(s->block[7], ptr_cr + uv_dct_offset, wrap_c);
2316  } else if (!s->chroma_y_shift && !s->chroma_x_shift) { /* 444 */
2317  s->pdsp.get_pixels(s->block[ 6], ptr_cb + 8, wrap_c);
2318  s->pdsp.get_pixels(s->block[ 7], ptr_cr + 8, wrap_c);
2319  s->pdsp.get_pixels(s->block[ 8], ptr_cb + uv_dct_offset, wrap_c);
2320  s->pdsp.get_pixels(s->block[ 9], ptr_cr + uv_dct_offset, wrap_c);
2321  s->pdsp.get_pixels(s->block[10], ptr_cb + uv_dct_offset + 8, wrap_c);
2322  s->pdsp.get_pixels(s->block[11], ptr_cr + uv_dct_offset + 8, wrap_c);
2323  }
2324  }
2325  } else {
2326  op_pixels_func (*op_pix)[4];
2327  qpel_mc_func (*op_qpix)[16];
2328  uint8_t *dest_y, *dest_cb, *dest_cr;
2329 
2330  dest_y = s->dest[0];
2331  dest_cb = s->dest[1];
2332  dest_cr = s->dest[2];
2333 
2334  if ((!s->no_rounding) || s->pict_type == AV_PICTURE_TYPE_B) {
2335  op_pix = s->hdsp.put_pixels_tab;
2336  op_qpix = s->qdsp.put_qpel_pixels_tab;
2337  } else {
2338  op_pix = s->hdsp.put_no_rnd_pixels_tab;
2339  op_qpix = s->qdsp.put_no_rnd_qpel_pixels_tab;
2340  }
2341 
2342  if (s->mv_dir & MV_DIR_FORWARD) {
2343  ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0,
2344  s->last_picture.f->data,
2345  op_pix, op_qpix);
2346  op_pix = s->hdsp.avg_pixels_tab;
2347  op_qpix = s->qdsp.avg_qpel_pixels_tab;
2348  }
2349  if (s->mv_dir & MV_DIR_BACKWARD) {
2350  ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1,
2351  s->next_picture.f->data,
2352  op_pix, op_qpix);
2353  }
2354 
2355  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
2356  int progressive_score, interlaced_score;
2357 
2358  s->interlaced_dct = 0;
2359  progressive_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y, wrap_y, 8) +
2360  s->mecc.ildct_cmp[0](s, dest_y + wrap_y * 8,
2361  ptr_y + wrap_y * 8,
2362  wrap_y, 8) - 400;
2363 
2364  if (s->avctx->ildct_cmp == FF_CMP_VSSE)
2365  progressive_score -= 400;
2366 
2367  if (progressive_score > 0) {
2368  interlaced_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y,
2369  wrap_y * 2, 8) +
2370  s->mecc.ildct_cmp[0](s, dest_y + wrap_y,
2371  ptr_y + wrap_y,
2372  wrap_y * 2, 8);
2373 
2374  if (progressive_score > interlaced_score) {
2375  s->interlaced_dct = 1;
2376 
2377  dct_offset = wrap_y;
2378  uv_dct_offset = wrap_c;
2379  wrap_y <<= 1;
2380  if (s->chroma_format == CHROMA_422)
2381  wrap_c <<= 1;
2382  }
2383  }
2384  }
2385 
2386  s->pdsp.diff_pixels(s->block[0], ptr_y, dest_y, wrap_y);
2387  s->pdsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
2388  s->pdsp.diff_pixels(s->block[2], ptr_y + dct_offset,
2389  dest_y + dct_offset, wrap_y);
2390  s->pdsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8,
2391  dest_y + dct_offset + 8, wrap_y);
2392 
2393  if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
2394  skip_dct[4] = 1;
2395  skip_dct[5] = 1;
2396  } else {
2397  s->pdsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
2398  s->pdsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
2399  if (!s->chroma_y_shift) { /* 422 */
2400  s->pdsp.diff_pixels(s->block[6], ptr_cb + uv_dct_offset,
2401  dest_cb + uv_dct_offset, wrap_c);
2402  s->pdsp.diff_pixels(s->block[7], ptr_cr + uv_dct_offset,
2403  dest_cr + uv_dct_offset, wrap_c);
2404  }
2405  }
2406  /* pre quantization */
2407  if (s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] <
2408  2 * s->qscale * s->qscale) {
2409  // FIXME optimize
2410  if (s->mecc.sad[1](NULL, ptr_y, dest_y, wrap_y, 8) < 20 * s->qscale)
2411  skip_dct[0] = 1;
2412  if (s->mecc.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20 * s->qscale)
2413  skip_dct[1] = 1;
2414  if (s->mecc.sad[1](NULL, ptr_y + dct_offset, dest_y + dct_offset,
2415  wrap_y, 8) < 20 * s->qscale)
2416  skip_dct[2] = 1;
2417  if (s->mecc.sad[1](NULL, ptr_y + dct_offset + 8, dest_y + dct_offset + 8,
2418  wrap_y, 8) < 20 * s->qscale)
2419  skip_dct[3] = 1;
2420  if (s->mecc.sad[1](NULL, ptr_cb, dest_cb, wrap_c, 8) < 20 * s->qscale)
2421  skip_dct[4] = 1;
2422  if (s->mecc.sad[1](NULL, ptr_cr, dest_cr, wrap_c, 8) < 20 * s->qscale)
2423  skip_dct[5] = 1;
2424  if (!s->chroma_y_shift) { /* 422 */
2425  if (s->mecc.sad[1](NULL, ptr_cb + uv_dct_offset,
2426  dest_cb + uv_dct_offset,
2427  wrap_c, 8) < 20 * s->qscale)
2428  skip_dct[6] = 1;
2429  if (s->mecc.sad[1](NULL, ptr_cr + uv_dct_offset,
2430  dest_cr + uv_dct_offset,
2431  wrap_c, 8) < 20 * s->qscale)
2432  skip_dct[7] = 1;
2433  }
2434  }
2435  }
2436 
2437  if (s->quantizer_noise_shaping) {
2438  if (!skip_dct[0])
2439  get_visual_weight(weight[0], ptr_y , wrap_y);
2440  if (!skip_dct[1])
2441  get_visual_weight(weight[1], ptr_y + 8, wrap_y);
2442  if (!skip_dct[2])
2443  get_visual_weight(weight[2], ptr_y + dct_offset , wrap_y);
2444  if (!skip_dct[3])
2445  get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
2446  if (!skip_dct[4])
2447  get_visual_weight(weight[4], ptr_cb , wrap_c);
2448  if (!skip_dct[5])
2449  get_visual_weight(weight[5], ptr_cr , wrap_c);
2450  if (!s->chroma_y_shift) { /* 422 */
2451  if (!skip_dct[6])
2452  get_visual_weight(weight[6], ptr_cb + uv_dct_offset,
2453  wrap_c);
2454  if (!skip_dct[7])
2455  get_visual_weight(weight[7], ptr_cr + uv_dct_offset,
2456  wrap_c);
2457  }
2458  memcpy(orig[0], s->block[0], sizeof(int16_t) * 64 * mb_block_count);
2459  }
2460 
2461  /* DCT & quantize */
2462  av_assert2(s->out_format != FMT_MJPEG || s->qscale == 8);
2463  {
2464  for (i = 0; i < mb_block_count; i++) {
2465  if (!skip_dct[i]) {
2466  int overflow;
2467  s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
2468  // FIXME we could decide to change to quantizer instead of
2469  // clipping
2470  // JS: I don't think that would be a good idea it could lower
2471  // quality instead of improve it. Just INTRADC clipping
2472  // deserves changes in quantizer
2473  if (overflow)
2474  clip_coeffs(s, s->block[i], s->block_last_index[i]);
2475  } else
2476  s->block_last_index[i] = -1;
2477  }
2478  if (s->quantizer_noise_shaping) {
2479  for (i = 0; i < mb_block_count; i++) {
2480  if (!skip_dct[i]) {
2481  s->block_last_index[i] =
2482  dct_quantize_refine(s, s->block[i], weight[i],
2483  orig[i], i, s->qscale);
2484  }
2485  }
2486  }
2487 
2488  if (s->luma_elim_threshold && !s->mb_intra)
2489  for (i = 0; i < 4; i++)
2490  dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
2491  if (s->chroma_elim_threshold && !s->mb_intra)
2492  for (i = 4; i < mb_block_count; i++)
2493  dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
2494 
2495  if (s->mpv_flags & FF_MPV_FLAG_CBP_RD) {
2496  for (i = 0; i < mb_block_count; i++) {
2497  if (s->block_last_index[i] == -1)
2498  s->coded_score[i] = INT_MAX / 256;
2499  }
2500  }
2501  }
2502 
2503  if ((s->avctx->flags & AV_CODEC_FLAG_GRAY) && s->mb_intra) {
2504  s->block_last_index[4] =
2505  s->block_last_index[5] = 0;
2506  s->block[4][0] =
2507  s->block[5][0] = (1024 + s->c_dc_scale / 2) / s->c_dc_scale;
2508  if (!s->chroma_y_shift) { /* 422 / 444 */
2509  for (i=6; i<12; i++) {
2510  s->block_last_index[i] = 0;
2511  s->block[i][0] = s->block[4][0];
2512  }
2513  }
2514  }
2515 
2516  // non c quantize code returns incorrect block_last_index FIXME
2517  if (s->alternate_scan && s->dct_quantize != ff_dct_quantize_c) {
2518  for (i = 0; i < mb_block_count; i++) {
2519  int j;
2520  if (s->block_last_index[i] > 0) {
2521  for (j = 63; j > 0; j--) {
2522  if (s->block[i][s->intra_scantable.permutated[j]])
2523  break;
2524  }
2525  s->block_last_index[i] = j;
2526  }
2527  }
2528  }
2529 
2530  /* huffman encode */
2531  switch(s->codec_id){ //FIXME funct ptr could be slightly faster
2532  case AV_CODEC_ID_MPEG1VIDEO:
2533  case AV_CODEC_ID_MPEG2VIDEO:
2534  if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2535  ff_mpeg1_encode_mb(s, s->block, motion_x, motion_y);
2536  break;
2537  case AV_CODEC_ID_MPEG4:
2538  if (CONFIG_MPEG4_ENCODER)
2539  ff_mpeg4_encode_mb(s, s->block, motion_x, motion_y);
2540  break;
2541  case AV_CODEC_ID_MSMPEG4V2:
2542  case AV_CODEC_ID_MSMPEG4V3:
2543  case AV_CODEC_ID_WMV1:
2544  if (CONFIG_MSMPEG4_ENCODER)
2545  ff_msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
2546  break;
2547  case AV_CODEC_ID_WMV2:
2548  if (CONFIG_WMV2_ENCODER)
2549  ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
2550  break;
2551  case AV_CODEC_ID_H261:
2552  if (CONFIG_H261_ENCODER)
2553  ff_h261_encode_mb(s, s->block, motion_x, motion_y);
2554  break;
2555  case AV_CODEC_ID_H263:
2556  case AV_CODEC_ID_H263P:
2557  case AV_CODEC_ID_FLV1:
2558  case AV_CODEC_ID_RV10:
2559  case AV_CODEC_ID_RV20:
2560  if (CONFIG_H263_ENCODER)
2561  ff_h263_encode_mb(s, s->block, motion_x, motion_y);
2562  break;
2563  case AV_CODEC_ID_MJPEG:
2564  case AV_CODEC_ID_AMV:
2565  if (CONFIG_MJPEG_ENCODER)
2566  ff_mjpeg_encode_mb(s, s->block);
2567  break;
2568  default:
2569  av_assert1(0);
2570  }
2571 }
2572 
2573 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
2574 {
2575  if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 8, 6);
2576  else if (s->chroma_format == CHROMA_422) encode_mb_internal(s, motion_x, motion_y, 16, 8, 8);
2577  else encode_mb_internal(s, motion_x, motion_y, 16, 16, 12);
2578 }
2579 
2580 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
2581  int i;
2582 
2583  memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2584 
2585  /* MPEG-1 */
2586  d->mb_skip_run= s->mb_skip_run;
2587  for(i=0; i<3; i++)
2588  d->last_dc[i] = s->last_dc[i];
2589 
2590  /* statistics */
2591  d->mv_bits= s->mv_bits;
2592  d->i_tex_bits= s->i_tex_bits;
2593  d->p_tex_bits= s->p_tex_bits;
2594  d->i_count= s->i_count;
2595  d->f_count= s->f_count;
2596  d->b_count= s->b_count;
2597  d->skip_count= s->skip_count;
2598  d->misc_bits= s->misc_bits;
2599  d->last_bits= 0;
2600 
2601  d->mb_skipped= 0;
2602  d->qscale= s->qscale;
2603  d->dquant= s->dquant;
2604 
2605  d->esc3_level_length= s->esc3_level_length;
2606 }
2607 
2608 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
2609  int i;
2610 
2611  memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
2612  memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2613 
2614  /* MPEG-1 */
2615  d->mb_skip_run= s->mb_skip_run;
2616  for(i=0; i<3; i++)
2617  d->last_dc[i] = s->last_dc[i];
2618 
2619  /* statistics */
2620  d->mv_bits= s->mv_bits;
2621  d->i_tex_bits= s->i_tex_bits;
2622  d->p_tex_bits= s->p_tex_bits;
2623  d->i_count= s->i_count;
2624  d->f_count= s->f_count;
2625  d->b_count= s->b_count;
2626  d->skip_count= s->skip_count;
2627  d->misc_bits= s->misc_bits;
2628 
2629  d->mb_intra= s->mb_intra;
2630  d->mb_skipped= s->mb_skipped;
2631  d->mv_type= s->mv_type;
2632  d->mv_dir= s->mv_dir;
2633  d->pb= s->pb;
2634  if(s->data_partitioning){
2635  d->pb2= s->pb2;
2636  d->tex_pb= s->tex_pb;
2637  }
2638  d->block= s->block;
2639  for(i=0; i<8; i++)
2640  d->block_last_index[i]= s->block_last_index[i];
2641  d->interlaced_dct= s->interlaced_dct;
2642  d->qscale= s->qscale;
2643 
2644  d->esc3_level_length= s->esc3_level_length;
2645 }
2646 
2647 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
2648  PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
2649  int *dmin, int *next_block, int motion_x, int motion_y)
2650 {
2651  int score;
2652  uint8_t *dest_backup[3];
2653 
2654  copy_context_before_encode(s, backup, type);
2655 
2656  s->block= s->blocks[*next_block];
2657  s->pb= pb[*next_block];
2658  if(s->data_partitioning){
2659  s->pb2 = pb2 [*next_block];
2660  s->tex_pb= tex_pb[*next_block];
2661  }
2662 
2663  if(*next_block){
2664  memcpy(dest_backup, s->dest, sizeof(s->dest));
2665  s->dest[0] = s->sc.rd_scratchpad;
2666  s->dest[1] = s->sc.rd_scratchpad + 16*s->linesize;
2667  s->dest[2] = s->sc.rd_scratchpad + 16*s->linesize + 8;
2668  av_assert0(s->linesize >= 32); //FIXME
2669  }
2670 
2671  encode_mb(s, motion_x, motion_y);
2672 
2673  score= put_bits_count(&s->pb);
2674  if(s->data_partitioning){
2675  score+= put_bits_count(&s->pb2);
2676  score+= put_bits_count(&s->tex_pb);
2677  }
2678 
2679  if(s->avctx->mb_decision == FF_MB_DECISION_RD){
2680  ff_mpv_reconstruct_mb(s, s->block);
2681 
2682  score *= s->lambda2;
2683  score += sse_mb(s) << FF_LAMBDA_SHIFT;
2684  }
2685 
2686  if(*next_block){
2687  memcpy(s->dest, dest_backup, sizeof(s->dest));
2688  }
2689 
2690  if(score<*dmin){
2691  *dmin= score;
2692  *next_block^=1;
2693 
2694  copy_context_after_encode(best, s, type);
2695  }
2696 }
2697 
2698 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
2699  const uint32_t *sq = ff_square_tab + 256;
2700  int acc=0;
2701  int x,y;
2702 
2703  if(w==16 && h==16)
2704  return s->mecc.sse[0](NULL, src1, src2, stride, 16);
2705  else if(w==8 && h==8)
2706  return s->mecc.sse[1](NULL, src1, src2, stride, 8);
2707 
2708  for(y=0; y<h; y++){
2709  for(x=0; x<w; x++){
2710  acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
2711  }
2712  }
2713 
2714  av_assert2(acc>=0);
2715 
2716  return acc;
2717 }
2718 
2719 static int sse_mb(MpegEncContext *s){
2720  int w= 16;
2721  int h= 16;
2722 
2723  if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2724  if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2725 
2726  if(w==16 && h==16)
2727  if(s->avctx->mb_cmp == FF_CMP_NSSE){
2728  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) +
2729  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) +
2730  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);
2731  }else{
2732  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) +
2733  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) +
2734  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);
2735  }
2736  else
2737  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)
2738  +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)
2739  +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);
2740 }
2741 
2742 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
2743  MpegEncContext *s= *(void**)arg;
2744 
2745 
2746  s->me.pre_pass=1;
2747  s->me.dia_size= s->avctx->pre_dia_size;
2748  s->first_slice_line=1;
2749  for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
2750  for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
2751  ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2752  }
2753  s->first_slice_line=0;
2754  }
2755 
2756  s->me.pre_pass=0;
2757 
2758  return 0;
2759 }
2760 
2761 static int estimate_motion_thread(AVCodecContext *c, void *arg){
2762  MpegEncContext *s= *(void**)arg;
2763 
2764  ff_check_alignment();
2765 
2766  s->me.dia_size= s->avctx->dia_size;
2767  s->first_slice_line=1;
2768  for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
2769  s->mb_x=0; //for block init below
2770  ff_init_block_index(s);
2771  for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
2772  s->block_index[0]+=2;
2773  s->block_index[1]+=2;
2774  s->block_index[2]+=2;
2775  s->block_index[3]+=2;
2776 
2777  /* compute motion vector & mb_type and store in context */
2778  if(s->pict_type==AV_PICTURE_TYPE_B)
2779  ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
2780  else
2781  ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2782  }
2783  s->first_slice_line=0;
2784  }
2785  return 0;
2786 }
2787 
2788 static int mb_var_thread(AVCodecContext *c, void *arg){
2789  MpegEncContext *s= *(void**)arg;
2790  int mb_x, mb_y;
2791 
2792  ff_check_alignment();
2793 
2794  for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2795  for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2796  int xx = mb_x * 16;
2797  int yy = mb_y * 16;
2798  uint8_t *pix = s->new_picture.f->data[0] + (yy * s->linesize) + xx;
2799  int varc;
2800  int sum = s->mpvencdsp.pix_sum(pix, s->linesize);
2801 
2802  varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) -
2803  (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8;
2804 
2805  s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2806  s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2807  s->me.mb_var_sum_temp += varc;
2808  }
2809  }
2810  return 0;
2811 }
2812 
2813 static void write_slice_end(MpegEncContext *s){
2814  if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4){
2815  if(s->partitioned_frame){
2816  ff_mpeg4_merge_partitions(s);
2817  }
2818 
2819  ff_mpeg4_stuffing(&s->pb);
2820  }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2821  ff_mjpeg_encode_stuffing(s);
2822  }
2823 
2824  avpriv_align_put_bits(&s->pb);
2825  flush_put_bits(&s->pb);
2826 
2827  if ((s->avctx->flags & AV_CODEC_FLAG_PASS1) && !s->partitioned_frame)
2828  s->misc_bits+= get_bits_diff(s);
2829 }
2830 
2831 static void write_mb_info(MpegEncContext *s)
2832 {
2833  uint8_t *ptr = s->mb_info_ptr + s->mb_info_size - 12;
2834  int offset = put_bits_count(&s->pb);
2835  int mba = s->mb_x + s->mb_width * (s->mb_y % s->gob_index);
2836  int gobn = s->mb_y / s->gob_index;
2837  int pred_x, pred_y;
2838  if (CONFIG_H263_ENCODER)
2839  ff_h263_pred_motion(s, 0, 0, &pred_x, &pred_y);
2840  bytestream_put_le32(&ptr, offset);
2841  bytestream_put_byte(&ptr, s->qscale);
2842  bytestream_put_byte(&ptr, gobn);
2843  bytestream_put_le16(&ptr, mba);
2844  bytestream_put_byte(&ptr, pred_x); /* hmv1 */
2845  bytestream_put_byte(&ptr, pred_y); /* vmv1 */
2846  /* 4MV not implemented */
2847  bytestream_put_byte(&ptr, 0); /* hmv2 */
2848  bytestream_put_byte(&ptr, 0); /* vmv2 */
2849 }
2850 
2851 static void update_mb_info(MpegEncContext *s, int startcode)
2852 {
2853  if (!s->mb_info)
2854  return;
2855  if (put_bits_count(&s->pb) - s->prev_mb_info*8 >= s->mb_info*8) {
2856  s->mb_info_size += 12;
2857  s->prev_mb_info = s->last_mb_info;
2858  }
2859  if (startcode) {
2860  s->prev_mb_info = put_bits_count(&s->pb)/8;
2861  /* This might have incremented mb_info_size above, and we return without
2862  * actually writing any info into that slot yet. But in that case,
2863  * this will be called again at the start of the after writing the
2864  * start code, actually writing the mb info. */
2865  return;
2866  }
2867 
2868  s->last_mb_info = put_bits_count(&s->pb)/8;
2869  if (!s->mb_info_size)
2870  s->mb_info_size += 12;
2871  write_mb_info(s);
2872 }
2873 
2874 int ff_mpv_reallocate_putbitbuffer(MpegEncContext *s, size_t threshold, size_t size_increase)
2875 {
2876  if ( s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < threshold
2877  && s->slice_context_count == 1
2878  && s->pb.buf == s->avctx->internal->byte_buffer) {
2879  int lastgob_pos = s->ptr_lastgob - s->pb.buf;
2880  int vbv_pos = s->vbv_delay_ptr - s->pb.buf;
2881 
2882  uint8_t *new_buffer = NULL;
2883  int new_buffer_size = 0;
2884 
2885  if ((s->avctx->internal->byte_buffer_size + size_increase) >= INT_MAX/8) {
2886  av_log(s->avctx, AV_LOG_ERROR, "Cannot reallocate putbit buffer\n");
2887  return AVERROR(ENOMEM);
2888  }
2889 
2890  emms_c();
2891 
2892  av_fast_padded_malloc(&new_buffer, &new_buffer_size,
2893  s->avctx->internal->byte_buffer_size + size_increase);
2894  if (!new_buffer)
2895  return AVERROR(ENOMEM);
2896 
2897  memcpy(new_buffer, s->avctx->internal->byte_buffer, s->avctx->internal->byte_buffer_size);
2898  av_free(s->avctx->internal->byte_buffer);
2899  s->avctx->internal->byte_buffer = new_buffer;
2900  s->avctx->internal->byte_buffer_size = new_buffer_size;
2901  rebase_put_bits(&s->pb, new_buffer, new_buffer_size);
2902  s->ptr_lastgob = s->pb.buf + lastgob_pos;
2903  s->vbv_delay_ptr = s->pb.buf + vbv_pos;
2904  }
2905  if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < threshold)
2906  return AVERROR(EINVAL);
2907  return 0;
2908 }
2909 
2910 static int encode_thread(AVCodecContext *c, void *arg){
2911  MpegEncContext *s= *(void**)arg;
2912  int mb_x, mb_y;
2913  int chr_h= 16>>s->chroma_y_shift;
2914  int i, j;
2915  MpegEncContext best_s = { 0 }, backup_s;
2916  uint8_t bit_buf[2][MAX_MB_BYTES];
2917  uint8_t bit_buf2[2][MAX_MB_BYTES];
2918  uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2919  PutBitContext pb[2], pb2[2], tex_pb[2];
2920 
2921  ff_check_alignment();
2922 
2923  for(i=0; i<2; i++){
2924  init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
2925  init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
2926  init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2927  }
2928 
2929  s->last_bits= put_bits_count(&s->pb);
2930  s->mv_bits=0;
2931  s->misc_bits=0;
2932  s->i_tex_bits=0;
2933  s->p_tex_bits=0;
2934  s->i_count=0;
2935  s->f_count=0;
2936  s->b_count=0;
2937  s->skip_count=0;
2938 
2939  for(i=0; i<3; i++){
2940  /* init last dc values */
2941  /* note: quant matrix value (8) is implied here */
2942  s->last_dc[i] = 128 << s->intra_dc_precision;
2943 
2944  s->current_picture.encoding_error[i] = 0;
2945  }
2946  if(s->codec_id==AV_CODEC_ID_AMV){
2947  s->last_dc[0] = 128*8/13;
2948  s->last_dc[1] = 128*8/14;
2949  s->last_dc[2] = 128*8/14;
2950  }
2951  s->mb_skip_run = 0;
2952  memset(s->last_mv, 0, sizeof(s->last_mv));
2953 
2954  s->last_mv_dir = 0;
2955 
2956  switch(s->codec_id){
2957  case AV_CODEC_ID_H263:
2958  case AV_CODEC_ID_H263P:
2959  case AV_CODEC_ID_FLV1:
2960  if (CONFIG_H263_ENCODER)
2961  s->gob_index = H263_GOB_HEIGHT(s->height);
2962  break;
2963  case AV_CODEC_ID_MPEG4:
2964  if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2965  ff_mpeg4_init_partitions(s);
2966  break;
2967  }
2968 
2969  s->resync_mb_x=0;
2970  s->resync_mb_y=0;
2971  s->first_slice_line = 1;
2972  s->ptr_lastgob = s->pb.buf;
2973  for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2974  s->mb_x=0;
2975  s->mb_y= mb_y;
2976 
2977  ff_set_qscale(s, s->qscale);
2978  ff_init_block_index(s);
2979 
2980  for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2981  int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2982  int mb_type= s->mb_type[xy];
2983 // int d;
2984  int dmin= INT_MAX;
2985  int dir;
2986  int size_increase = s->avctx->internal->byte_buffer_size/4
2987  + s->mb_width*MAX_MB_BYTES;
2988 
2989  ff_mpv_reallocate_putbitbuffer(s, MAX_MB_BYTES, size_increase);
2990  if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2991  av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2992  return -1;
2993  }
2994  if(s->data_partitioning){
2995  if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES
2996  || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2997  av_log(s->avctx, AV_LOG_ERROR, "encoded partitioned frame too large\n");
2998  return -1;
2999  }
3000  }
3001 
3002  s->mb_x = mb_x;
3003  s->mb_y = mb_y; // moved into loop, can get changed by H.261
3004  ff_update_block_index(s);
3005 
3006  if(CONFIG_H261_ENCODER && s->codec_id == AV_CODEC_ID_H261){
3007  ff_h261_reorder_mb_index(s);
3008  xy= s->mb_y*s->mb_stride + s->mb_x;
3009  mb_type= s->mb_type[xy];
3010  }
3011 
3012  /* write gob / video packet header */
3013  if(s->rtp_mode){
3014  int current_packet_size, is_gob_start;
3015 
3016  current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
3017 
3018  is_gob_start = s->rtp_payload_size &&
3019  current_packet_size >= s->rtp_payload_size &&
3020  mb_y + mb_x > 0;
3021 
3022  if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
3023 
3024  switch(s->codec_id){
3025  case AV_CODEC_ID_H263:
3026  case AV_CODEC_ID_H263P:
3027  if(!s->h263_slice_structured)
3028  if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
3029  break;
3030  case AV_CODEC_ID_MPEG2VIDEO:
3031  if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
3032  case AV_CODEC_ID_MPEG1VIDEO:
3033  if(s->mb_skip_run) is_gob_start=0;
3034  break;
3035  case AV_CODEC_ID_MJPEG:
3036  if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
3037  break;
3038  }
3039 
3040  if(is_gob_start){
3041  if(s->start_mb_y != mb_y || mb_x!=0){
3042  write_slice_end(s);
3043 
3044  if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4 && s->partitioned_frame){
3045  ff_mpeg4_init_partitions(s);
3046  }
3047  }
3048 
3049  av_assert2((put_bits_count(&s->pb)&7) == 0);
3050  current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
3051 
3052  if (s->error_rate && s->resync_mb_x + s->resync_mb_y > 0) {
3053  int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
3054  int d = 100 / s->error_rate;
3055  if(r % d == 0){
3056  current_packet_size=0;
3057  s->pb.buf_ptr= s->ptr_lastgob;
3058  av_assert1(put_bits_ptr(&s->pb) == s->ptr_lastgob);
3059  }
3060  }
3061 
3062 #if FF_API_RTP_CALLBACK
3063 FF_DISABLE_DEPRECATION_WARNINGS
3064  if (s->avctx->rtp_callback){
3065  int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
3066  s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
3067  }
3068 FF_ENABLE_DEPRECATION_WARNINGS
3069 #endif
3070  update_mb_info(s, 1);
3071 
3072  switch(s->codec_id){
3073  case AV_CODEC_ID_MPEG4:
3074  if (CONFIG_MPEG4_ENCODER) {
3075  ff_mpeg4_encode_video_packet_header(s);
3076  ff_mpeg4_clean_buffers(s);
3077  }
3078  break;
3079  case AV_CODEC_ID_MPEG1VIDEO:
3080  case AV_CODEC_ID_MPEG2VIDEO:
3081  if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
3082  ff_mpeg1_encode_slice_header(s);
3083  ff_mpeg1_clean_buffers(s);
3084  }
3085  break;
3086  case AV_CODEC_ID_H263:
3087  case AV_CODEC_ID_H263P:
3088  if (CONFIG_H263_ENCODER)
3089  ff_h263_encode_gob_header(s, mb_y);
3090  break;
3091  }
3092 
3093  if (s->avctx->flags & AV_CODEC_FLAG_PASS1) {
3094  int bits= put_bits_count(&s->pb);
3095  s->misc_bits+= bits - s->last_bits;
3096  s->last_bits= bits;
3097  }
3098 
3099  s->ptr_lastgob += current_packet_size;
3100  s->first_slice_line=1;
3101  s->resync_mb_x=mb_x;
3102  s->resync_mb_y=mb_y;
3103  }
3104  }
3105 
3106  if( (s->resync_mb_x == s->mb_x)
3107  && s->resync_mb_y+1 == s->mb_y){
3108  s->first_slice_line=0;
3109  }
3110 
3111  s->mb_skipped=0;
3112  s->dquant=0; //only for QP_RD
3113 
3114  update_mb_info(s, 0);
3115 
3116  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
3117  int next_block=0;
3118  int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
3119 
3120  copy_context_before_encode(&backup_s, s, -1);
3121  backup_s.pb= s->pb;
3122  best_s.data_partitioning= s->data_partitioning;
3123  best_s.partitioned_frame= s->partitioned_frame;
3124  if(s->data_partitioning){
3125  backup_s.pb2= s->pb2;
3126  backup_s.tex_pb= s->tex_pb;
3127  }
3128 
3129  if(mb_type&CANDIDATE_MB_TYPE_INTER){
3130  s->mv_dir = MV_DIR_FORWARD;
3131  s->mv_type = MV_TYPE_16X16;
3132  s->mb_intra= 0;
3133  s->mv[0][0][0] = s->p_mv_table[xy][0];
3134  s->mv[0][0][1] = s->p_mv_table[xy][1];
3135  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
3136  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3137  }
3138  if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
3139  s->mv_dir = MV_DIR_FORWARD;
3140  s->mv_type = MV_TYPE_FIELD;
3141  s->mb_intra= 0;
3142  for(i=0; i<2; i++){
3143  j= s->field_select[0][i] = s->p_field_select_table[i][xy];
3144  s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
3145  s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
3146  }
3147  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
3148  &dmin, &next_block, 0, 0);
3149  }
3150  if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
3151  s->mv_dir = MV_DIR_FORWARD;
3152  s->mv_type = MV_TYPE_16X16;
3153  s->mb_intra= 0;
3154  s->mv[0][0][0] = 0;
3155  s->mv[0][0][1] = 0;
3156  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
3157  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3158  }
3159  if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
3160  s->mv_dir = MV_DIR_FORWARD;
3161  s->mv_type = MV_TYPE_8X8;
3162  s->mb_intra= 0;
3163  for(i=0; i<4; i++){
3164  s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
3165  s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
3166  }
3167  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
3168  &dmin, &next_block, 0, 0);
3169  }
3170  if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
3171  s->mv_dir = MV_DIR_FORWARD;
3172  s->mv_type = MV_TYPE_16X16;
3173  s->mb_intra= 0;
3174  s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
3175  s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
3176  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
3177  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3178  }
3179  if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
3180  s->mv_dir = MV_DIR_BACKWARD;
3181  s->mv_type = MV_TYPE_16X16;
3182  s->mb_intra= 0;
3183  s->mv[1][0][0] = s->b_back_mv_table[xy][0];
3184  s->mv[1][0][1] = s->b_back_mv_table[xy][1];
3185  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
3186  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
3187  }
3188  if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
3189  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3190  s->mv_type = MV_TYPE_16X16;
3191  s->mb_intra= 0;
3192  s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
3193  s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
3194  s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
3195  s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
3196  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
3197  &dmin, &next_block, 0, 0);
3198  }
3199  if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
3200  s->mv_dir = MV_DIR_FORWARD;
3201  s->mv_type = MV_TYPE_FIELD;
3202  s->mb_intra= 0;
3203  for(i=0; i<2; i++){
3204  j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
3205  s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
3206  s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
3207  }
3208  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
3209  &dmin, &next_block, 0, 0);
3210  }
3211  if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
3212  s->mv_dir = MV_DIR_BACKWARD;
3213  s->mv_type = MV_TYPE_FIELD;
3214  s->mb_intra= 0;
3215  for(i=0; i<2; i++){
3216  j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
3217  s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
3218  s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
3219  }
3220  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
3221  &dmin, &next_block, 0, 0);
3222  }
3223  if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
3224  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3225  s->mv_type = MV_TYPE_FIELD;
3226  s->mb_intra= 0;
3227  for(dir=0; dir<2; dir++){
3228  for(i=0; i<2; i++){
3229  j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
3230  s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
3231  s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
3232  }
3233  }
3234  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
3235  &dmin, &next_block, 0, 0);
3236  }
3237  if(mb_type&CANDIDATE_MB_TYPE_INTRA){
3238  s->mv_dir = 0;
3239  s->mv_type = MV_TYPE_16X16;
3240  s->mb_intra= 1;
3241  s->mv[0][0][0] = 0;
3242  s->mv[0][0][1] = 0;
3243  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
3244  &dmin, &next_block, 0, 0);
3245  if(s->h263_pred || s->h263_aic){
3246  if(best_s.mb_intra)
3247  s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
3248  else
3249  ff_clean_intra_table_entries(s); //old mode?
3250  }
3251  }
3252 
3253  if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) && dmin < INT_MAX) {
3254  if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
3255  const int last_qp= backup_s.qscale;
3256  int qpi, qp, dc[6];
3257  int16_t ac[6][16];
3258  const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
3259  static const int dquant_tab[4]={-1,1,-2,2};
3260  int storecoefs = s->mb_intra && s->dc_val[0];
3261 
3262  av_assert2(backup_s.dquant == 0);
3263 
3264  //FIXME intra
3265  s->mv_dir= best_s.mv_dir;
3266  s->mv_type = MV_TYPE_16X16;
3267  s->mb_intra= best_s.mb_intra;
3268  s->mv[0][0][0] = best_s.mv[0][0][0];
3269  s->mv[0][0][1] = best_s.mv[0][0][1];
3270  s->mv[1][0][0] = best_s.mv[1][0][0];
3271  s->mv[1][0][1] = best_s.mv[1][0][1];
3272 
3273  qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
3274  for(; qpi<4; qpi++){
3275  int dquant= dquant_tab[qpi];
3276  qp= last_qp + dquant;
3277  if(qp < s->avctx->qmin || qp > s->avctx->qmax)
3278  continue;
3279  backup_s.dquant= dquant;
3280  if(storecoefs){
3281  for(i=0; i<6; i++){
3282  dc[i]= s->dc_val[0][ s->block_index[i] ];
3283  memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(int16_t)*16);
3284  }
3285  }
3286 
3287  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
3288  &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
3289  if(best_s.qscale != qp){
3290  if(storecoefs){
3291  for(i=0; i<6; i++){
3292  s->dc_val[0][ s->block_index[i] ]= dc[i];
3293  memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(int16_t)*16);
3294  }
3295  }
3296  }
3297  }
3298  }
3299  }
3300  if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
3301  int mx= s->b_direct_mv_table[xy][0];
3302  int my= s->b_direct_mv_table[xy][1];
3303 
3304  backup_s.dquant = 0;
3305  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
3306  s->mb_intra= 0;
3307  ff_mpeg4_set_direct_mv(s, mx, my);
3308  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
3309  &dmin, &next_block, mx, my);
3310  }
3311  if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
3312  backup_s.dquant = 0;
3313  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
3314  s->mb_intra= 0;
3315  ff_mpeg4_set_direct_mv(s, 0, 0);
3316  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
3317  &dmin, &next_block, 0, 0);
3318  }
3319  if (!best_s.mb_intra && s->mpv_flags & FF_MPV_FLAG_SKIP_RD) {
3320  int coded=0;
3321  for(i=0; i<6; i++)
3322  coded |= s->block_last_index[i];
3323  if(coded){
3324  int mx,my;
3325  memcpy(s->mv, best_s.mv, sizeof(s->mv));
3326  if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
3327  mx=my=0; //FIXME find the one we actually used
3328  ff_mpeg4_set_direct_mv(s, mx, my);
3329  }else if(best_s.mv_dir&MV_DIR_BACKWARD){
3330  mx= s->mv[1][0][0];
3331  my= s->mv[1][0][1];
3332  }else{
3333  mx= s->mv[0][0][0];
3334  my= s->mv[0][0][1];
3335  }
3336 
3337  s->mv_dir= best_s.mv_dir;
3338  s->mv_type = best_s.mv_type;
3339  s->mb_intra= 0;
3340 /* s->mv[0][0][0] = best_s.mv[0][0][0];
3341  s->mv[0][0][1] = best_s.mv[0][0][1];
3342  s->mv[1][0][0] = best_s.mv[1][0][0];
3343  s->mv[1][0][1] = best_s.mv[1][0][1];*/
3344  backup_s.dquant= 0;
3345  s->skipdct=1;
3346  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
3347  &dmin, &next_block, mx, my);
3348  s->skipdct=0;
3349  }
3350  }
3351 
3352  s->current_picture.qscale_table[xy] = best_s.qscale;
3353 
3354  copy_context_after_encode(s, &best_s, -1);
3355 
3356  pb_bits_count= put_bits_count(&s->pb);
3357  flush_put_bits(&s->pb);
3358  avpriv_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
3359  s->pb= backup_s.pb;
3360 
3361  if(s->data_partitioning){
3362  pb2_bits_count= put_bits_count(&s->pb2);
3363  flush_put_bits(&s->pb2);
3364  avpriv_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
3365  s->pb2= backup_s.pb2;
3366 
3367  tex_pb_bits_count= put_bits_count(&s->tex_pb);
3368  flush_put_bits(&s->tex_pb);
3369  avpriv_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
3370  s->tex_pb= backup_s.tex_pb;
3371  }
3372  s->last_bits= put_bits_count(&s->pb);
3373 
3374  if (CONFIG_H263_ENCODER &&
3375  s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
3376  ff_h263_update_motion_val(s);
3377 
3378  if(next_block==0){ //FIXME 16 vs linesize16
3379  s->hdsp.put_pixels_tab[0][0](s->dest[0], s->sc.rd_scratchpad , s->linesize ,16);
3380  s->hdsp.put_pixels_tab[1][0](s->dest[1], s->sc.rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
3381  s->hdsp.put_pixels_tab[1][0](s->dest[2], s->sc.rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
3382  }
3383 
3384  if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
3385  ff_mpv_reconstruct_mb(s, s->block);
3386  } else {
3387  int motion_x = 0, motion_y = 0;
3388  s->mv_type=MV_TYPE_16X16;
3389  // only one MB-Type possible
3390 
3391  switch(mb_type){
3392  case CANDIDATE_MB_TYPE_INTRA:
3393  s->mv_dir = 0;
3394  s->mb_intra= 1;
3395  motion_x= s->mv[0][0][0] = 0;
3396  motion_y= s->mv[0][0][1] = 0;
3397  break;
3398  case CANDIDATE_MB_TYPE_INTER:
3399  s->mv_dir = MV_DIR_FORWARD;
3400  s->mb_intra= 0;
3401  motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
3402  motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
3403  break;
3404  case CANDIDATE_MB_TYPE_INTER_I:
3405  s->mv_dir = MV_DIR_FORWARD;
3406  s->mv_type = MV_TYPE_FIELD;
3407  s->mb_intra= 0;
3408  for(i=0; i<2; i++){
3409  j= s->field_select[0][i] = s->p_field_select_table[i][xy];
3410  s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
3411  s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
3412  }
3413  break;
3414  case CANDIDATE_MB_TYPE_INTER4V:
3415  s->mv_dir = MV_DIR_FORWARD;
3416  s->mv_type = MV_TYPE_8X8;
3417  s->mb_intra= 0;
3418  for(i=0; i<4; i++){
3419  s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
3420  s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
3421  }
3422  break;
3423  case CANDIDATE_MB_TYPE_DIRECT:
3424  if (CONFIG_MPEG4_ENCODER) {
3425  s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
3426  s->mb_intra= 0;
3427  motion_x=s->b_direct_mv_table[xy][0];
3428  motion_y=s->b_direct_mv_table[xy][1];
3429  ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
3430  }
3431  break;
3432  case CANDIDATE_MB_TYPE_DIRECT0:
3433  if (CONFIG_MPEG4_ENCODER) {
3434  s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
3435  s->mb_intra= 0;
3436  ff_mpeg4_set_direct_mv(s, 0, 0);
3437  }
3438  break;
3439  case CANDIDATE_MB_TYPE_BIDIR:
3440  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3441  s->mb_intra= 0;
3442  s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
3443  s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
3444  s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
3445  s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
3446  break;
3447  case CANDIDATE_MB_TYPE_BACKWARD:
3448  s->mv_dir = MV_DIR_BACKWARD;
3449  s->mb_intra= 0;
3450  motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
3451  motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
3452  break;
3453  case CANDIDATE_MB_TYPE_FORWARD:
3454  s->mv_dir = MV_DIR_FORWARD;
3455  s->mb_intra= 0;
3456  motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
3457  motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
3458  break;
3459  case CANDIDATE_MB_TYPE_FORWARD_I:
3460  s->mv_dir = MV_DIR_FORWARD;
3461  s->mv_type = MV_TYPE_FIELD;
3462  s->mb_intra= 0;
3463  for(i=0; i<2; i++){
3464  j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
3465  s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
3466  s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
3467  }
3468  break;
3469  case CANDIDATE_MB_TYPE_BACKWARD_I:
3470  s->mv_dir = MV_DIR_BACKWARD;
3471  s->mv_type = MV_TYPE_FIELD;
3472  s->mb_intra= 0;
3473  for(i=0; i<2; i++){
3474  j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
3475  s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
3476  s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
3477  }
3478  break;
3479  case CANDIDATE_MB_TYPE_BIDIR_I:
3480  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3481  s->mv_type = MV_TYPE_FIELD;
3482  s->mb_intra= 0;
3483  for(dir=0; dir<2; dir++){
3484  for(i=0; i<2; i++){
3485  j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
3486  s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
3487  s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
3488  }
3489  }
3490  break;
3491  default:
3492  av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
3493  }
3494 
3495  encode_mb(s, motion_x, motion_y);
3496 
3497  // RAL: Update last macroblock type
3498  s->last_mv_dir = s->mv_dir;
3499 
3500  if (CONFIG_H263_ENCODER &&
3501  s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
3502  ff_h263_update_motion_val(s);
3503 
3504  ff_mpv_reconstruct_mb(s, s->block);
3505  }
3506 
3507  /* clean the MV table in IPS frames for direct mode in B-frames */
3508  if(s->mb_intra /* && I,P,S_TYPE */){
3509  s->p_mv_table[xy][0]=0;
3510  s->p_mv_table[xy][1]=0;
3511  }
3512 
3513  if (s->avctx->flags & AV_CODEC_FLAG_PSNR) {
3514  int w= 16;
3515  int h= 16;
3516 
3517  if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
3518  if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
3519 
3520  s->current_picture.encoding_error[0] += sse(
3521  s, s->new_picture.f->data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
3522  s->dest[0], w, h, s->linesize);
3523  s->current_picture.encoding_error[1] += sse(
3524  s, s->new_picture.f->data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
3525  s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
3526  s->current_picture.encoding_error[2] += sse(
3527  s, s->new_picture.f->data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
3528  s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
3529  }
3530  if(s->loop_filter){
3531  if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
3532  ff_h263_loop_filter(s);
3533  }
3534  ff_dlog(s->avctx, "MB %d %d bits\n",
3535  s->mb_x + s->mb_y * s->mb_stride, put_bits_count(&s->pb));
3536  }
3537  }
3538 
3539  //not beautiful here but we must write it before flushing so it has to be here
3540  if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
3541  ff_msmpeg4_encode_ext_header(s);
3542 
3543  write_slice_end(s);
3544 
3545 #if FF_API_RTP_CALLBACK
3546 FF_DISABLE_DEPRECATION_WARNINGS
3547  /* Send the last GOB if RTP */
3548  if (s->avctx->rtp_callback) {
3549  int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
3550  int pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
3551  /* Call the RTP callback to send the last GOB */
3552  emms_c();
3553  s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
3554  }
3555 FF_ENABLE_DEPRECATION_WARNINGS
3556 #endif
3557 
3558  return 0;
3559 }
3560 
3561 #define MERGE(field) dst->field += src->field; src->field=0
3562 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
3563  MERGE(me.scene_change_score);
3564  MERGE(me.mc_mb_var_sum_temp);
3565  MERGE(me.mb_var_sum_temp);
3566 }
3567 
3568 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
3569  int i;
3570 
3571  MERGE(dct_count[0]); //note, the other dct vars are not part of the context
3572  MERGE(dct_count[1]);
3573  MERGE(mv_bits);
3574  MERGE(i_tex_bits);
3575  MERGE(p_tex_bits);
3576  MERGE(i_count);
3577  MERGE(f_count);
3578  MERGE(b_count);
3579  MERGE(skip_count);
3580  MERGE(misc_bits);
3581  MERGE(er.error_count);
3582  MERGE(padding_bug_score);
3583  MERGE(current_picture.encoding_error[0]);
3584  MERGE(current_picture.encoding_error[1]);
3585  MERGE(current_picture.encoding_error[2]);
3586 
3587  if (dst->noise_reduction){
3588  for(i=0; i<64; i++){
3589  MERGE(dct_error_sum[0][i]);
3590  MERGE(dct_error_sum[1][i]);
3591  }
3592  }
3593 
3594  av_assert1(put_bits_count(&src->pb) % 8 ==0);
3595  av_assert1(put_bits_count(&dst->pb) % 8 ==0);
3596  avpriv_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
3597  flush_put_bits(&dst->pb);
3598 }
3599 
3600 static int estimate_qp(MpegEncContext *s, int dry_run){
3601  if (s->next_lambda){
3602  s->current_picture_ptr->f->quality =
3603  s->current_picture.f->quality = s->next_lambda;
3604  if(!dry_run) s->next_lambda= 0;
3605  } else if (!s->fixed_qscale) {
3606  int quality = ff_rate_estimate_qscale(s, dry_run);
3607  s->current_picture_ptr->f->quality =
3608  s->current_picture.f->quality = quality;
3609  if (s->current_picture.f->quality < 0)
3610  return -1;
3611  }
3612 
3613  if(s->adaptive_quant){
3614  switch(s->codec_id){
3615  case AV_CODEC_ID_MPEG4:
3616  if (CONFIG_MPEG4_ENCODER)
3617  ff_clean_mpeg4_qscales(s);
3618  break;
3619  case AV_CODEC_ID_H263:
3620  case AV_CODEC_ID_H263P:
3621  case AV_CODEC_ID_FLV1:
3622  if (CONFIG_H263_ENCODER)
3623  ff_clean_h263_qscales(s);
3624  break;
3625  default:
3626  ff_init_qscale_tab(s);
3627  }
3628 
3629  s->lambda= s->lambda_table[0];
3630  //FIXME broken
3631  }else
3632  s->lambda = s->current_picture.f->quality;
3633  update_qscale(s);
3634  return 0;
3635 }
3636 
3637 /* must be called before writing the header */
3638 static void set_frame_distances(MpegEncContext * s){
3639  av_assert1(s->current_picture_ptr->f->pts != AV_NOPTS_VALUE);
3640  s->time = s->current_picture_ptr->f->pts * s->avctx->time_base.num;
3641 
3642  if(s->pict_type==AV_PICTURE_TYPE_B){
3643  s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
3644  av_assert1(s->pb_time > 0 && s->pb_time < s->pp_time);
3645  }else{
3646  s->pp_time= s->time - s->last_non_b_time;
3647  s->last_non_b_time= s->time;
3648  av_assert1(s->picture_number==0 || s->pp_time > 0);
3649  }
3650 }
3651 
3652 static int encode_picture(MpegEncContext *s, int picture_number)
3653 {
3654  int i, ret;
3655  int bits;
3656  int context_count = s->slice_context_count;
3657 
3658  s->picture_number = picture_number;
3659 
3660  /* Reset the average MB variance */
3661  s->me.mb_var_sum_temp =
3662  s->me.mc_mb_var_sum_temp = 0;
3663 
3664  /* we need to initialize some time vars before we can encode B-frames */
3665  // RAL: Condition added for MPEG1VIDEO
3666  if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO || s->codec_id == AV_CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
3667  set_frame_distances(s);
3668  if(CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4)
3669  ff_set_mpeg4_time(s);
3670 
3671  s->me.scene_change_score=0;
3672 
3673 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
3674 
3675  if(s->pict_type==AV_PICTURE_TYPE_I){
3676  if(s->msmpeg4_version >= 3) s->no_rounding=1;
3677  else s->no_rounding=0;
3678  }else if(s->pict_type!=AV_PICTURE_TYPE_B){
3679  if(s->flipflop_rounding || s->codec_id == AV_CODEC_ID_H263P || s->codec_id == AV_CODEC_ID_MPEG4)
3680  s->no_rounding ^= 1;
3681  }
3682 
3683  if (s->avctx->flags & AV_CODEC_FLAG_PASS2) {
3684  if (estimate_qp(s,1) < 0)
3685  return -1;
3686  ff_get_2pass_fcode(s);
3687  } else if (!(s->avctx->flags & AV_CODEC_FLAG_QSCALE)) {
3688  if(s->pict_type==AV_PICTURE_TYPE_B)
3689  s->lambda= s->last_lambda_for[s->pict_type];
3690  else
3691  s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
3692  update_qscale(s);
3693  }
3694 
3695  if(s->codec_id != AV_CODEC_ID_AMV && s->codec_id != AV_CODEC_ID_MJPEG){
3696  if(s->q_chroma_intra_matrix != s->q_intra_matrix ) av_freep(&s->q_chroma_intra_matrix);
3697  if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
3698  s->q_chroma_intra_matrix = s->q_intra_matrix;
3699  s->q_chroma_intra_matrix16 = s->q_intra_matrix16;
3700  }
3701 
3702  s->mb_intra=0; //for the rate distortion & bit compare functions
3703  for(i=1; i<context_count; i++){
3704  ret = ff_update_duplicate_context(s->thread_context[i], s);
3705  if (ret < 0)
3706  return ret;
3707  }
3708 
3709  if(ff_init_me(s)<0)
3710  return -1;
3711 
3712  /* Estimate motion for every MB */
3713  if(s->pict_type != AV_PICTURE_TYPE_I){
3714  s->lambda = (s->lambda * s->me_penalty_compensation + 128) >> 8;
3715  s->lambda2 = (s->lambda2 * (int64_t) s->me_penalty_compensation + 128) >> 8;
3716  if (s->pict_type != AV_PICTURE_TYPE_B) {
3717  if ((s->me_pre && s->last_non_b_pict_type == AV_PICTURE_TYPE_I) ||
3718  s->me_pre == 2) {
3719  s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3720  }
3721  }
3722 
3723  s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3724  }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
3725  /* I-Frame */
3726  for(i=0; i<s->mb_stride*s->mb_height; i++)
3727  s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3728 
3729  if(!s->fixed_qscale){
3730  /* finding spatial complexity for I-frame rate control */
3731  s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3732  }
3733  }
3734  for(i=1; i<context_count; i++){
3735  merge_context_after_me(s, s->thread_context[i]);
3736  }
3737  s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
3738  s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
3739  emms_c();
3740 
3741  if (s->me.scene_change_score > s->scenechange_threshold &&
3742  s->pict_type == AV_PICTURE_TYPE_P) {
3743  s->pict_type= AV_PICTURE_TYPE_I;
3744  for(i=0; i<s->mb_stride*s->mb_height; i++)
3745  s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3746  if(s->msmpeg4_version >= 3)
3747  s->no_rounding=1;
3748  ff_dlog(s, "Scene change detected, encoding as I Frame %"PRId64" %"PRId64"\n",
3749  s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
3750  }
3751 
3752  if(!s->umvplus){
3753  if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
3754  s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
3755 
3756  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3757  int a,b;
3758  a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
3759  b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
3760  s->f_code= FFMAX3(s->f_code, a, b);
3761  }
3762 
3763  ff_fix_long_p_mvs(s, s->intra_penalty ? CANDIDATE_MB_TYPE_INTER : CANDIDATE_MB_TYPE_INTRA);
3764  ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, !!s->intra_penalty);
3765  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3766  int j;
3767  for(i=0; i<2; i++){
3768  for(j=0; j<2; j++)
3769  ff_fix_long_mvs(s, s->p_field_select_table[i], j,
3770  s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, !!s->intra_penalty);
3771  }
3772  }
3773  }
3774 
3775  if(s->pict_type==AV_PICTURE_TYPE_B){
3776  int a, b;
3777 
3778  a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
3779  b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
3780  s->f_code = FFMAX(a, b);
3781 
3782  a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
3783  b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
3784  s->b_code = FFMAX(a, b);
3785 
3786  ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
3787  ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
3788  ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
3789  ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
3790  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3791  int dir, j;
3792  for(dir=0; dir<2; dir++){
3793  for(i=0; i<2; i++){
3794  for(j=0; j<2; j++){
3795  int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
3796  : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
3797  ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
3798  s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
3799  }
3800  }
3801  }
3802  }
3803  }
3804  }
3805 
3806  if (estimate_qp(s, 0) < 0)
3807  return -1;
3808 
3809  if (s->qscale < 3 && s->max_qcoeff <= 128 &&
3810  s->pict_type == AV_PICTURE_TYPE_I &&
3811  !(s->avctx->flags & AV_CODEC_FLAG_QSCALE))
3812  s->qscale= 3; //reduce clipping problems
3813 
3814  if (s->out_format == FMT_MJPEG) {
3815  const uint16_t * luma_matrix = ff_mpeg1_default_intra_matrix;
3816  const uint16_t *chroma_matrix = ff_mpeg1_default_intra_matrix;
3817 
3818  if (s->avctx->intra_matrix) {
3819  chroma_matrix =
3820  luma_matrix = s->avctx->intra_matrix;
3821  }
3822  if (s->avctx->chroma_intra_matrix)
3823  chroma_matrix = s->avctx->chroma_intra_matrix;
3824 
3825  /* for mjpeg, we do include qscale in the matrix */
3826  for(i=1;i<64;i++){
3827  int j = s->idsp.idct_permutation[i];
3828 
3829  s->chroma_intra_matrix[j] = av_clip_uint8((chroma_matrix[i] * s->qscale) >> 3);
3830  s-> intra_matrix[j] = av_clip_uint8(( luma_matrix[i] * s->qscale) >> 3);
3831  }
3832  s->y_dc_scale_table=
3833  s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
3834  s->chroma_intra_matrix[0] =
3835  s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
3836  ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
3837  s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
3838  ff_convert_matrix(s, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
3839  s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
3840  s->qscale= 8;
3841  }
3842  if(s->codec_id == AV_CODEC_ID_AMV){
3843  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};
3844  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};
3845  for(i=1;i<64;i++){
3846  int j= s->idsp.idct_permutation[ff_zigzag_direct[i]];
3847 
3848  s->intra_matrix[j] = sp5x_quant_table[5*2+0][i];
3849  s->chroma_intra_matrix[j] = sp5x_quant_table[5*2+1][i];
3850  }
3851  s->y_dc_scale_table= y;
3852  s->c_dc_scale_table= c;
3853  s->intra_matrix[0] = 13;
3854  s->chroma_intra_matrix[0] = 14;
3855  ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
3856  s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
3857  ff_convert_matrix(s, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
3858  s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
3859  s->qscale= 8;
3860  }
3861 
3862  //FIXME var duplication
3863  s->current_picture_ptr->f->key_frame =
3864  s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
3865  s->current_picture_ptr->f->pict_type =
3866  s->current_picture.f->pict_type = s->pict_type;
3867 
3868  if (s->current_picture.f->key_frame)
3869  s->picture_in_gop_number=0;
3870 
3871  s->mb_x = s->mb_y = 0;
3872  s->last_bits= put_bits_count(&s->pb);
3873  switch(s->out_format) {
3874  case FMT_MJPEG:
3875  if (CONFIG_MJPEG_ENCODER && s->huffman != HUFFMAN_TABLE_OPTIMAL)
3876  ff_mjpeg_encode_picture_header(s->avctx, &s->pb, &s->intra_scantable,
3877  s->pred, s->intra_matrix, s->chroma_intra_matrix);
3878  break;
3879  case FMT_H261:
3880  if (CONFIG_H261_ENCODER)
3881  ff_h261_encode_picture_header(s, picture_number);
3882  break;
3883  case FMT_H263:
3884  if (CONFIG_WMV2_ENCODER && s->codec_id == AV_CODEC_ID_WMV2)
3885  ff_wmv2_encode_picture_header(s, picture_number);
3886  else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
3887  ff_msmpeg4_encode_picture_header(s, picture_number);
3888  else if (CONFIG_MPEG4_ENCODER && s->h263_pred) {
3889  ret = ff_mpeg4_encode_picture_header(s, picture_number);
3890  if (ret < 0)
3891  return ret;
3892  } else if (CONFIG_RV10_ENCODER && s->codec_id == AV_CODEC_ID_RV10) {
3893  ret = ff_rv10_encode_picture_header(s, picture_number);
3894  if (ret < 0)
3895  return ret;
3896  }
3897  else if (CONFIG_RV20_ENCODER && s->codec_id == AV_CODEC_ID_RV20)
3898  ff_rv20_encode_picture_header(s, picture_number);
3899  else if (CONFIG_FLV_ENCODER && s->codec_id == AV_CODEC_ID_FLV1)
3900  ff_flv_encode_picture_header(s, picture_number);
3901  else if (CONFIG_H263_ENCODER)
3902  ff_h263_encode_picture_header(s, picture_number);
3903  break;
3904  case FMT_MPEG1:
3905  if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
3906  ff_mpeg1_encode_picture_header(s, picture_number);
3907  break;
3908  default:
3909  av_assert0(0);
3910  }
3911  bits= put_bits_count(&s->pb);
3912  s->header_bits= bits - s->last_bits;
3913 
3914  for(i=1; i<context_count; i++){
3915  update_duplicate_context_after_me(s->thread_context[i], s);
3916  }
3917  s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3918  for(i=1; i<context_count; i++){
3919  if (s->pb.buf_end == s->thread_context[i]->pb.buf)
3920  set_put_bits_buffer_size(&s->pb, FFMIN(s->thread_context[i]->pb.buf_end - s->pb.buf, INT_MAX/8-32));
3921  merge_context_after_encode(s, s->thread_context[i]);
3922  }
3923  emms_c();
3924  return 0;
3925 }
3926 
3927 static void denoise_dct_c(MpegEncContext *s, int16_t *block){
3928  const int intra= s->mb_intra;
3929  int i;
3930 
3931  s->dct_count[intra]++;
3932 
3933  for(i=0; i<64; i++){
3934  int level= block[i];
3935 
3936  if(level){
3937  if(level>0){
3938  s->dct_error_sum[intra][i] += level;
3939  level -= s->dct_offset[intra][i];
3940  if(level<0) level=0;
3941  }else{
3942  s->dct_error_sum[intra][i] -= level;
3943  level += s->dct_offset[intra][i];
3944  if(level>0) level=0;
3945  }
3946  block[i]= level;
3947  }
3948  }
3949 }
3950 
3951 static int dct_quantize_trellis_c(MpegEncContext *s,
3952  int16_t *block, int n,
3953  int qscale, int *overflow){
3954  const int *qmat;
3955  const uint16_t *matrix;
3956  const uint8_t *scantable;
3957  const uint8_t *perm_scantable;
3958  int max=0;
3959  unsigned int threshold1, threshold2;
3960  int bias=0;
3961  int run_tab[65];
3962  int level_tab[65];
3963  int score_tab[65];
3964  int survivor[65];
3965  int survivor_count;
3966  int last_run=0;
3967  int last_level=0;
3968  int last_score= 0;
3969  int last_i;
3970  int coeff[2][64];
3971  int coeff_count[64];
3972  int qmul, qadd, start_i, last_non_zero, i, dc;
3973  const int esc_length= s->ac_esc_length;
3974  uint8_t * length;
3975  uint8_t * last_length;
3976  const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3977  int mpeg2_qscale;
3978 
3979  s->fdsp.fdct(block);
3980 
3981  if(s->dct_error_sum)
3982  s->denoise_dct(s, block);
3983  qmul= qscale*16;
3984  qadd= ((qscale-1)|1)*8;
3985 
3986  if (s->q_scale_type) mpeg2_qscale = ff_mpeg2_non_linear_qscale[qscale];
3987  else mpeg2_qscale = qscale << 1;
3988 
3989  if (s->mb_intra) {
3990  int q;
3991  scantable= s->intra_scantable.scantable;
3992  perm_scantable= s->intra_scantable.permutated;
3993  if (!s->h263_aic) {
3994  if (n < 4)
3995  q = s->y_dc_scale;
3996  else
3997  q = s->c_dc_scale;
3998  q = q << 3;
3999  } else{
4000  /* For AIC we skip quant/dequant of INTRADC */
4001  q = 1 << 3;
4002  qadd=0;
4003  }
4004 
4005  /* note: block[0] is assumed to be positive */
4006  block[0] = (block[0] + (q >> 1)) / q;
4007  start_i = 1;
4008  last_non_zero = 0;
4009  qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
4010  matrix = n < 4 ? s->intra_matrix : s->chroma_intra_matrix;
4011  if(s->mpeg_quant || s->out_format == FMT_MPEG1 || s->out_format == FMT_MJPEG)
4012  bias= 1<<(QMAT_SHIFT-1);
4013 
4014  if (n > 3 && s->intra_chroma_ac_vlc_length) {
4015  length = s->intra_chroma_ac_vlc_length;
4016  last_length= s->intra_chroma_ac_vlc_last_length;
4017  } else {
4018  length = s->intra_ac_vlc_length;
4019  last_length= s->intra_ac_vlc_last_length;
4020  }
4021  } else {
4022  scantable= s->inter_scantable.scantable;
4023  perm_scantable= s->inter_scantable.permutated;
4024  start_i = 0;
4025  last_non_zero = -1;
4026  qmat = s->q_inter_matrix[qscale];
4027  matrix = s->inter_matrix;
4028  length = s->inter_ac_vlc_length;
4029  last_length= s->inter_ac_vlc_last_length;
4030  }
4031  last_i= start_i;
4032 
4033  threshold1= (1<<QMAT_SHIFT) - bias - 1;
4034  threshold2= (threshold1<<1);
4035 
4036  for(i=63; i>=start_i; i--) {
4037  const int j = scantable[i];
4038  int level = block[j] * qmat[j];
4039 
4040  if(((unsigned)(level+threshold1))>threshold2){
4041  last_non_zero = i;
4042  break;
4043  }
4044  }
4045 
4046  for(i=start_i; i<=last_non_zero; i++) {
4047  const int j = scantable[i];
4048  int level = block[j] * qmat[j];
4049 
4050 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
4051 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
4052  if(((unsigned)(level+threshold1))>threshold2){
4053  if(level>0){
4054  level= (bias + level)>>QMAT_SHIFT;
4055  coeff[0][i]= level;
4056  coeff[1][i]= level-1;
4057 // coeff[2][k]= level-2;
4058  }else{
4059  level= (bias - level)>>QMAT_SHIFT;
4060  coeff[0][i]= -level;
4061  coeff[1][i]= -level+1;
4062 // coeff[2][k]= -level+2;
4063  }
4064  coeff_count[i]= FFMIN(level, 2);
4065  av_assert2(coeff_count[i]);
4066  max |=level;
4067  }else{
4068  coeff[0][i]= (level>>31)|1;
4069  coeff_count[i]= 1;
4070  }
4071  }
4072 
4073  *overflow= s->max_qcoeff < max; //overflow might have happened
4074 
4075  if(last_non_zero < start_i){
4076  memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
4077  return last_non_zero;
4078  }
4079 
4080  score_tab[start_i]= 0;
4081  survivor[0]= start_i;
4082  survivor_count= 1;
4083 
4084  for(i=start_i; i<=last_non_zero; i++){
4085  int level_index, j, zero_distortion;
4086  int dct_coeff= FFABS(block[ scantable[i] ]);
4087  int best_score=256*256*256*120;
4088 
4089  if (s->fdsp.fdct == ff_fdct_ifast)
4090  dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
4091  zero_distortion= dct_coeff*dct_coeff;
4092 
4093  for(level_index=0; level_index < coeff_count[i]; level_index++){
4094  int distortion;
4095  int level= coeff[level_index][i];
4096  const int alevel= FFABS(level);
4097  int unquant_coeff;
4098 
4099  av_assert2(level);
4100 
4101  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4102  unquant_coeff= alevel*qmul + qadd;
4103  } else if(s->out_format == FMT_MJPEG) {
4104  j = s->idsp.idct_permutation[scantable[i]];
4105  unquant_coeff = alevel * matrix[j] * 8;
4106  }else{ // MPEG-1
4107  j = s->idsp.idct_permutation[scantable[i]]; // FIXME: optimize
4108  if(s->mb_intra){
4109  unquant_coeff = (int)( alevel * mpeg2_qscale * matrix[j]) >> 4;
4110  unquant_coeff = (unquant_coeff - 1) | 1;
4111  }else{
4112  unquant_coeff = ((( alevel << 1) + 1) * mpeg2_qscale * ((int) matrix[j])) >> 5;
4113  unquant_coeff = (unquant_coeff - 1) | 1;
4114  }
4115  unquant_coeff<<= 3;
4116  }
4117 
4118  distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
4119  level+=64;
4120  if((level&(~127)) == 0){
4121  for(j=survivor_count-1; j>=0; j--){
4122  int run= i - survivor[j];
4123  int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
4124  score += score_tab[i-run];
4125 
4126  if(score < best_score){
4127  best_score= score;
4128  run_tab[i+1]= run;
4129  level_tab[i+1]= level-64;
4130  }
4131  }
4132 
4133  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4134  for(j=survivor_count-1; j>=0; j--){
4135  int run= i - survivor[j];
4136  int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
4137  score += score_tab[i-run];
4138  if(score < last_score){
4139  last_score= score;
4140  last_run= run;
4141  last_level= level-64;
4142  last_i= i+1;
4143  }
4144  }
4145  }
4146  }else{
4147  distortion += esc_length*lambda;
4148  for(j=survivor_count-1; j>=0; j--){
4149  int run= i - survivor[j];
4150  int score= distortion + score_tab[i-run];
4151 
4152  if(score < best_score){
4153  best_score= score;
4154  run_tab[i+1]= run;
4155  level_tab[i+1]= level-64;
4156  }
4157  }
4158 
4159  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4160  for(j=survivor_count-1; j>=0; j--){
4161  int run= i - survivor[j];
4162  int score= distortion + score_tab[i-run];
4163  if(score < last_score){
4164  last_score= score;
4165  last_run= run;
4166  last_level= level-64;
4167  last_i= i+1;
4168  }
4169  }
4170  }
4171  }
4172  }
4173 
4174  score_tab[i+1]= best_score;
4175 
4176  // 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
4177  if(last_non_zero <= 27){
4178  for(; survivor_count; survivor_count--){
4179  if(score_tab[ survivor[survivor_count-1] ] <= best_score)
4180  break;
4181  }
4182  }else{
4183  for(; survivor_count; survivor_count--){
4184  if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
4185  break;
4186  }
4187  }
4188 
4189  survivor[ survivor_count++ ]= i+1;
4190  }
4191 
4192  if(s->out_format != FMT_H263 && s->out_format != FMT_H261){
4193  last_score= 256*256*256*120;
4194  for(i= survivor[0]; i<=last_non_zero + 1; i++){
4195  int score= score_tab[i];
4196  if (i)
4197  score += lambda * 2; // FIXME more exact?
4198 
4199  if(score < last_score){
4200  last_score= score;
4201  last_i= i;
4202  last_level= level_tab[i];
4203  last_run= run_tab[i];
4204  }
4205  }
4206  }
4207 
4208  s->coded_score[n] = last_score;
4209 
4210  dc= FFABS(block[0]);
4211  last_non_zero= last_i - 1;
4212  memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
4213 
4214  if(last_non_zero < start_i)
4215  return last_non_zero;
4216 
4217  if(last_non_zero == 0 && start_i == 0){
4218  int best_level= 0;
4219  int best_score= dc * dc;
4220 
4221  for(i=0; i<coeff_count[0]; i++){
4222  int level= coeff[i][0];
4223  int alevel= FFABS(level);
4224  int unquant_coeff, score, distortion;
4225 
4226  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4227  unquant_coeff= (alevel*qmul + qadd)>>3;
4228  } else{ // MPEG-1
4229  unquant_coeff = ((( alevel << 1) + 1) * mpeg2_qscale * ((int) matrix[0])) >> 5;
4230  unquant_coeff = (unquant_coeff - 1) | 1;
4231  }
4232  unquant_coeff = (unquant_coeff + 4) >> 3;
4233  unquant_coeff<<= 3 + 3;
4234 
4235  distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
4236  level+=64;
4237  if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
4238  else score= distortion + esc_length*lambda;
4239 
4240  if(score < best_score){
4241  best_score= score;
4242  best_level= level - 64;
4243  }
4244  }
4245  block[0]= best_level;
4246  s->coded_score[n] = best_score - dc*dc;
4247  if(best_level == 0) return -1;
4248  else return last_non_zero;
4249  }
4250 
4251  i= last_i;
4252  av_assert2(last_level);
4253 
4254  block[ perm_scantable[last_non_zero] ]= last_level;
4255  i -= last_run + 1;
4256 
4257  for(; i>start_i; i -= run_tab[i] + 1){
4258  block[ perm_scantable[i-1] ]= level_tab[i];
4259  }
4260 
4261  return last_non_zero;
4262 }
4263 
4264 static int16_t basis[64][64];
4265 
4266 static void build_basis(uint8_t *perm){
4267  int i, j, x, y;
4268  emms_c();
4269  for(i=0; i<8; i++){
4270  for(j=0; j<8; j++){
4271  for(y=0; y<8; y++){
4272  for(x=0; x<8; x++){
4273  double s= 0.25*(1<<BASIS_SHIFT);
4274  int index= 8*i + j;
4275  int perm_index= perm[index];
4276  if(i==0) s*= sqrt(0.5);
4277  if(j==0) s*= sqrt(0.5);
4278  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)));
4279  }
4280  }
4281  }
4282  }
4283 }
4284 
4285 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
4286  int16_t *block, int16_t *weight, int16_t *orig,
4287  int n, int qscale){
4288  int16_t rem[64];
4289  LOCAL_ALIGNED_16(int16_t, d1, [64]);
4290  const uint8_t *scantable;
4291  const uint8_t *perm_scantable;
4292 // unsigned int threshold1, threshold2;
4293 // int bias=0;
4294  int run_tab[65];
4295  int prev_run=0;
4296  int prev_level=0;
4297  int qmul, qadd, start_i, last_non_zero, i, dc;
4298  uint8_t * length;
4299  uint8_t * last_length;
4300  int lambda;
4301  int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
4302 
4303  if(basis[0][0] == 0)
4304  build_basis(s->idsp.idct_permutation);
4305 
4306  qmul= qscale*2;
4307  qadd= (qscale-1)|1;
4308  if (s->mb_intra) {
4309  scantable= s->intra_scantable.scantable;
4310  perm_scantable= s->intra_scantable.permutated;
4311  if (!s->h263_aic) {
4312  if (n < 4)
4313  q = s->y_dc_scale;
4314  else
4315  q = s->c_dc_scale;
4316  } else{
4317  /* For AIC we skip quant/dequant of INTRADC */
4318  q = 1;
4319  qadd=0;
4320  }
4321  q <<= RECON_SHIFT-3;
4322  /* note: block[0] is assumed to be positive */
4323  dc= block[0]*q;
4324 // block[0] = (block[0] + (q >> 1)) / q;
4325  start_i = 1;
4326 // if(s->mpeg_quant || s->out_format == FMT_MPEG1)
4327 // bias= 1<<(QMAT_SHIFT-1);
4328  if (n > 3 && s->intra_chroma_ac_vlc_length) {
4329  length = s->intra_chroma_ac_vlc_length;
4330  last_length= s->intra_chroma_ac_vlc_last_length;
4331  } else {
4332  length = s->intra_ac_vlc_length;
4333  last_length= s->intra_ac_vlc_last_length;
4334  }
4335  } else {
4336  scantable= s->inter_scantable.scantable;
4337  perm_scantable= s->inter_scantable.permutated;
4338  dc= 0;
4339  start_i = 0;
4340  length = s->inter_ac_vlc_length;
4341  last_length= s->inter_ac_vlc_last_length;
4342  }
4343  last_non_zero = s->block_last_index[n];
4344 
4345  dc += (1<<(RECON_SHIFT-1));
4346  for(i=0; i<64; i++){
4347  rem[i] = dc - (orig[i] << RECON_SHIFT); // FIXME use orig directly instead of copying to rem[]
4348  }
4349 
4350  sum=0;
4351  for(i=0; i<64; i++){
4352  int one= 36;
4353  int qns=4;
4354  int w;
4355 
4356  w= FFABS(weight[i]) + qns*one;
4357  w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
4358 
4359  weight[i] = w;
4360 // w=weight[i] = (63*qns + (w/2)) / w;
4361 
4362  av_assert2(w>0);
4363  av_assert2(w<(1<<6));
4364  sum += w*w;
4365  }
4366  lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
4367 
4368  run=0;
4369  rle_index=0;
4370  for(i=start_i; i<=last_non_zero; i++){
4371  int j= perm_scantable[i];
4372  const int level= block[j];
4373  int coeff;
4374 
4375  if(level){
4376  if(level<0) coeff= qmul*level - qadd;
4377  else coeff= qmul*level + qadd;
4378  run_tab[rle_index++]=run;
4379  run=0;
4380 
4381  s->mpvencdsp.add_8x8basis(rem, basis[j], coeff);
4382  }else{
4383  run++;
4384  }
4385  }
4386 
4387  for(;;){
4388  int best_score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0], 0);
4389  int best_coeff=0;
4390  int best_change=0;
4391  int run2, best_unquant_change=0, analyze_gradient;
4392  analyze_gradient = last_non_zero > 2 || s->quantizer_noise_shaping >= 3;
4393 
4394  if(analyze_gradient){
4395  for(i=0; i<64; i++){
4396  int w= weight[i];
4397 
4398  d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
4399  }
4400  s->fdsp.fdct(d1);
4401  }
4402 
4403  if(start_i){
4404  const int level= block[0];
4405  int change, old_coeff;
4406 
4407  av_assert2(s->mb_intra);
4408 
4409  old_coeff= q*level;
4410 
4411  for(change=-1; change<=1; change+=2){
4412  int new_level= level + change;
4413  int score, new_coeff;
4414 
4415  new_coeff= q*new_level;
4416  if(new_coeff >= 2048 || new_coeff < 0)
4417  continue;
4418 
4419  score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0],
4420  new_coeff - old_coeff);
4421  if(score<best_score){
4422  best_score= score;
4423  best_coeff= 0;
4424  best_change= change;
4425  best_unquant_change= new_coeff - old_coeff;
4426  }
4427  }
4428  }
4429 
4430  run=0;
4431  rle_index=0;
4432  run2= run_tab[rle_index++];
4433  prev_level=0;
4434  prev_run=0;
4435 
4436  for(i=start_i; i<64; i++){
4437  int j= perm_scantable[i];
4438  const int level= block[j];
4439  int change, old_coeff;
4440 
4441  if(s->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
4442  break;
4443 
4444  if(level){
4445  if(level<0) old_coeff= qmul*level - qadd;
4446  else old_coeff= qmul*level + qadd;
4447  run2= run_tab[rle_index++]; //FIXME ! maybe after last
4448  }else{
4449  old_coeff=0;
4450  run2--;
4451  av_assert2(run2>=0 || i >= last_non_zero );
4452  }
4453 
4454  for(change=-1; change<=1; change+=2){
4455  int new_level= level + change;
4456  int score, new_coeff, unquant_change;
4457 
4458  score=0;
4459  if(s->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
4460  continue;
4461 
4462  if(new_level){
4463  if(new_level<0) new_coeff= qmul*new_level - qadd;
4464  else new_coeff= qmul*new_level + qadd;
4465  if(new_coeff >= 2048 || new_coeff <= -2048)
4466  continue;
4467  //FIXME check for overflow
4468 
4469  if(level){
4470  if(level < 63 && level > -63){
4471  if(i < last_non_zero)
4472  score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
4473  - length[UNI_AC_ENC_INDEX(run, level+64)];
4474  else
4475  score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
4476  - last_length[UNI_AC_ENC_INDEX(run, level+64)];
4477  }
4478  }else{
4479  av_assert2(FFABS(new_level)==1);
4480 
4481  if(analyze_gradient){
4482  int g= d1[ scantable[i] ];
4483  if(g && (g^new_level) >= 0)
4484  continue;
4485  }
4486 
4487  if(i < last_non_zero){
4488  int next_i= i + run2 + 1;
4489  int next_level= block[ perm_scantable[next_i] ] + 64;
4490 
4491  if(next_level&(~127))
4492  next_level= 0;
4493