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 "libavutil/timer.h"
42 #include "avcodec.h"
43 #include "dct.h"
44 #include "idctdsp.h"
45 #include "mpeg12.h"
46 #include "mpegvideo.h"
47 #include "mpegvideodata.h"
48 #include "h261.h"
49 #include "h263.h"
50 #include "h263data.h"
51 #include "mjpegenc_common.h"
52 #include "mathops.h"
53 #include "mpegutils.h"
54 #include "mjpegenc.h"
55 #include "msmpeg4.h"
56 #include "pixblockdsp.h"
57 #include "qpeldsp.h"
58 #include "faandct.h"
59 #include "thread.h"
60 #include "aandcttab.h"
61 #include "flv.h"
62 #include "mpeg4video.h"
63 #include "internal.h"
64 #include "bytestream.h"
65 #include "wmv2.h"
66 #include "rv10.h"
67 #include "libxvid.h"
68 #include <limits.h>
69 #include "sp5x.h"
70 
71 #define QUANT_BIAS_SHIFT 8
72 
73 #define QMAT_SHIFT_MMX 16
74 #define QMAT_SHIFT 21
75 
76 static int encode_picture(MpegEncContext *s, int picture_number);
77 static int dct_quantize_refine(MpegEncContext *s, int16_t *block, int16_t *weight, int16_t *orig, int n, int qscale);
78 static int sse_mb(MpegEncContext *s);
79 static void denoise_dct_c(MpegEncContext *s, int16_t *block);
80 static int dct_quantize_trellis_c(MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow);
81 
82 static uint8_t default_mv_penalty[MAX_FCODE + 1][MAX_DMV * 2 + 1];
83 static uint8_t default_fcode_tab[MAX_MV * 2 + 1];
84 
85 const AVOption ff_mpv_generic_options[] = {
86  FF_MPV_COMMON_OPTS
87  { NULL },
88 };
89 
90 void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
91  uint16_t (*qmat16)[2][64],
92  const uint16_t *quant_matrix,
93  int bias, int qmin, int qmax, int intra)
94 {
95  FDCTDSPContext *fdsp = &s->fdsp;
96  int qscale;
97  int shift = 0;
98 
99  for (qscale = qmin; qscale <= qmax; qscale++) {
100  int i;
101  int qscale2;
102 
103  if (s->q_scale_type) qscale2 = ff_mpeg2_non_linear_qscale[qscale];
104  else qscale2 = qscale << 1;
105 
106  if (fdsp->fdct == ff_jpeg_fdct_islow_8 ||
107 #if CONFIG_FAANDCT
108  fdsp->fdct == ff_faandct ||
109 #endif /* CONFIG_FAANDCT */
110  fdsp->fdct == ff_jpeg_fdct_islow_10) {
111  for (i = 0; i < 64; i++) {
112  const int j = s->idsp.idct_permutation[i];
113  int64_t den = (int64_t) qscale2 * quant_matrix[j];
114  /* 16 <= qscale * quant_matrix[i] <= 7905
115  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
116  * 19952 <= x <= 249205026
117  * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
118  * 3444240 >= (1 << 36) / (x) >= 275 */
119 
120  qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
121  }
122  } else if (fdsp->fdct == ff_fdct_ifast) {
123  for (i = 0; i < 64; i++) {
124  const int j = s->idsp.idct_permutation[i];
125  int64_t den = ff_aanscales[i] * (int64_t) qscale2 * quant_matrix[j];
126  /* 16 <= qscale * quant_matrix[i] <= 7905
127  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
128  * 19952 <= x <= 249205026
129  * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
130  * 3444240 >= (1 << 36) / (x) >= 275 */
131 
132  qmat[qscale][i] = (int)((UINT64_C(2) << (QMAT_SHIFT + 14)) / den);
133  }
134  } else {
135  for (i = 0; i < 64; i++) {
136  const int j = s->idsp.idct_permutation[i];
137  int64_t den = (int64_t) qscale2 * quant_matrix[j];
138  /* We can safely suppose that 16 <= quant_matrix[i] <= 255
139  * Assume x = qscale * quant_matrix[i]
140  * So 16 <= x <= 7905
141  * so (1 << 19) / 16 >= (1 << 19) / (x) >= (1 << 19) / 7905
142  * so 32768 >= (1 << 19) / (x) >= 67 */
143  qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
144  //qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) /
145  // (qscale * quant_matrix[i]);
146  qmat16[qscale][0][i] = (2 << QMAT_SHIFT_MMX) / den;
147 
148  if (qmat16[qscale][0][i] == 0 ||
149  qmat16[qscale][0][i] == 128 * 256)
150  qmat16[qscale][0][i] = 128 * 256 - 1;
151  qmat16[qscale][1][i] =
152  ROUNDED_DIV(bias * (1<<(16 - QUANT_BIAS_SHIFT)),
153  qmat16[qscale][0][i]);
154  }
155  }
156 
157  for (i = intra; i < 64; i++) {
158  int64_t max = 8191;
159  if (fdsp->fdct == ff_fdct_ifast) {
160  max = (8191LL * ff_aanscales[i]) >> 14;
161  }
162  while (((max * qmat[qscale][i]) >> shift) > INT_MAX) {
163  shift++;
164  }
165  }
166  }
167  if (shift) {
168  av_log(NULL, AV_LOG_INFO,
169  "Warning, QMAT_SHIFT is larger than %d, overflows possible\n",
170  QMAT_SHIFT - shift);
171  }
172 }
173 
174 static inline void update_qscale(MpegEncContext *s)
175 {
176  if (s->q_scale_type == 1 && 0) {
177  int i;
178  int bestdiff=INT_MAX;
179  int best = 1;
180 
181  for (i = 0 ; i<FF_ARRAY_ELEMS(ff_mpeg2_non_linear_qscale); i++) {
182  int diff = FFABS((ff_mpeg2_non_linear_qscale[i]<<(FF_LAMBDA_SHIFT + 6)) - (int)s->lambda * 139);
183  if (ff_mpeg2_non_linear_qscale[i] < s->avctx->qmin ||
184  (ff_mpeg2_non_linear_qscale[i] > s->avctx->qmax && !s->vbv_ignore_qmax))
185  continue;
186  if (diff < bestdiff) {
187  bestdiff = diff;
188  best = i;
189  }
190  }
191  s->qscale = best;
192  } else {
193  s->qscale = (s->lambda * 139 + FF_LAMBDA_SCALE * 64) >>
194  (FF_LAMBDA_SHIFT + 7);
195  s->qscale = av_clip(s->qscale, s->avctx->qmin, s->vbv_ignore_qmax ? 31 : s->avctx->qmax);
196  }
197 
198  s->lambda2 = (s->lambda * s->lambda + FF_LAMBDA_SCALE / 2) >>
199  FF_LAMBDA_SHIFT;
200 }
201 
202 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix)
203 {
204  int i;
205 
206  if (matrix) {
207  put_bits(pb, 1, 1);
208  for (i = 0; i < 64; i++) {
209  put_bits(pb, 8, matrix[ff_zigzag_direct[i]]);
210  }
211  } else
212  put_bits(pb, 1, 0);
213 }
214 
215 /**
216  * init s->current_picture.qscale_table from s->lambda_table
217  */
218 void ff_init_qscale_tab(MpegEncContext *s)
219 {
220  int8_t * const qscale_table = s->current_picture.qscale_table;
221  int i;
222 
223  for (i = 0; i < s->mb_num; i++) {
224  unsigned int lam = s->lambda_table[s->mb_index2xy[i]];
225  int qp = (lam * 139 + FF_LAMBDA_SCALE * 64) >> (FF_LAMBDA_SHIFT + 7);
226  qscale_table[s->mb_index2xy[i]] = av_clip(qp, s->avctx->qmin,
227  s->avctx->qmax);
228  }
229 }
230 
231 static void update_duplicate_context_after_me(MpegEncContext *dst,
232  MpegEncContext *src)
233 {
234 #define COPY(a) dst->a= src->a
235  COPY(pict_type);
236  COPY(current_picture);
237  COPY(f_code);
238  COPY(b_code);
239  COPY(qscale);
240  COPY(lambda);
241  COPY(lambda2);
242  COPY(picture_in_gop_number);
243  COPY(gop_picture_number);
244  COPY(frame_pred_frame_dct); // FIXME don't set in encode_header
245  COPY(progressive_frame); // FIXME don't set in encode_header
246  COPY(partitioned_frame); // FIXME don't set in encode_header
247 #undef COPY
248 }
249 
250 /**
251  * Set the given MpegEncContext to defaults for encoding.
252  * the changed fields will not depend upon the prior state of the MpegEncContext.
253  */
254 static void mpv_encode_defaults(MpegEncContext *s)
255 {
256  int i;
257  ff_mpv_common_defaults(s);
258 
259  for (i = -16; i < 16; i++) {
260  default_fcode_tab[i + MAX_MV] = 1;
261  }
262  s->me.mv_penalty = default_mv_penalty;
263  s->fcode_tab = default_fcode_tab;
264 
265  s->input_picture_number = 0;
266  s->picture_in_gop_number = 0;
267 }
268 
269 av_cold int ff_dct_encode_init(MpegEncContext *s)
270 {
271  if (ARCH_X86)
272  ff_dct_encode_init_x86(s);
273 
274  if (CONFIG_H263_ENCODER)
275  ff_h263dsp_init(&s->h263dsp);
276  if (!s->dct_quantize)
277  s->dct_quantize = ff_dct_quantize_c;
278  if (!s->denoise_dct)
279  s->denoise_dct = denoise_dct_c;
280  s->fast_dct_quantize = s->dct_quantize;
281  if (s->avctx->trellis)
282  s->dct_quantize = dct_quantize_trellis_c;
283 
284  return 0;
285 }
286 
287 /* init video encoder */
288 av_cold int ff_mpv_encode_init(AVCodecContext *avctx)
289 {
290  MpegEncContext *s = avctx->priv_data;
291  AVCPBProperties *cpb_props;
292  int i, ret, format_supported;
293 
294  mpv_encode_defaults(s);
295 
296  switch (avctx->codec_id) {
297  case AV_CODEC_ID_MPEG2VIDEO:
298  if (avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
299  avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
300  av_log(avctx, AV_LOG_ERROR,
301  "only YUV420 and YUV422 are supported\n");
302  return -1;
303  }
304  break;
305  case AV_CODEC_ID_MJPEG:
306  case AV_CODEC_ID_AMV:
307  format_supported = 0;
308  /* JPEG color space */
309  if (avctx->pix_fmt == AV_PIX_FMT_YUVJ420P ||
310  avctx->pix_fmt == AV_PIX_FMT_YUVJ422P ||
311  avctx->pix_fmt == AV_PIX_FMT_YUVJ444P ||
312  (avctx->color_range == AVCOL_RANGE_JPEG &&
313  (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
314  avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
315  avctx->pix_fmt == AV_PIX_FMT_YUV444P)))
316  format_supported = 1;
317  /* MPEG color space */
318  else if (avctx->strict_std_compliance <= FF_COMPLIANCE_UNOFFICIAL &&
319  (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
320  avctx->pix_fmt == AV_PIX_FMT_YUV422P ||
321  avctx->pix_fmt == AV_PIX_FMT_YUV444P))
322  format_supported = 1;
323 
324  if (!format_supported) {
325  av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
326  return -1;
327  }
328  break;
329  default:
330  if (avctx->pix_fmt != AV_PIX_FMT_YUV420P) {
331  av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
332  return -1;
333  }
334  }
335 
336  switch (avctx->pix_fmt) {
337  case AV_PIX_FMT_YUVJ444P:
338  case AV_PIX_FMT_YUV444P:
339  s->chroma_format = CHROMA_444;
340  break;
341  case AV_PIX_FMT_YUVJ422P:
342  case AV_PIX_FMT_YUV422P:
343  s->chroma_format = CHROMA_422;
344  break;
345  case AV_PIX_FMT_YUVJ420P:
346  case AV_PIX_FMT_YUV420P:
347  default:
348  s->chroma_format = CHROMA_420;
349  break;
350  }
351 
352  avctx->bits_per_raw_sample = av_clip(avctx->bits_per_raw_sample, 0, 8);
353 
354 #if FF_API_PRIVATE_OPT
355 FF_DISABLE_DEPRECATION_WARNINGS
356  if (avctx->rtp_payload_size)
357  s->rtp_payload_size = avctx->rtp_payload_size;
358  if (avctx->me_penalty_compensation)
359  s->me_penalty_compensation = avctx->me_penalty_compensation;
360  if (avctx->pre_me)
361  s->me_pre = avctx->pre_me;
362 FF_ENABLE_DEPRECATION_WARNINGS
363 #endif
364 
365  s->bit_rate = avctx->bit_rate;
366  s->width = avctx->width;
367  s->height = avctx->height;
368  if (avctx->gop_size > 600 &&
369  avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
370  av_log(avctx, AV_LOG_WARNING,
371  "keyframe interval too large!, reducing it from %d to %d\n",
372  avctx->gop_size, 600);
373  avctx->gop_size = 600;
374  }
375  s->gop_size = avctx->gop_size;
376  s->avctx = avctx;
377  if (avctx->max_b_frames > MAX_B_FRAMES) {
378  av_log(avctx, AV_LOG_ERROR, "Too many B-frames requested, maximum "
379  "is %d.\n", MAX_B_FRAMES);
380  avctx->max_b_frames = MAX_B_FRAMES;
381  }
382  s->max_b_frames = avctx->max_b_frames;
383  s->codec_id = avctx->codec->id;
384  s->strict_std_compliance = avctx->strict_std_compliance;
385  s->quarter_sample = (avctx->flags & AV_CODEC_FLAG_QPEL) != 0;
386  s->rtp_mode = !!s->rtp_payload_size;
387  s->intra_dc_precision = avctx->intra_dc_precision;
388 
389  // workaround some differences between how applications specify dc precision
390  if (s->intra_dc_precision < 0) {
391  s->intra_dc_precision += 8;
392  } else if (s->intra_dc_precision >= 8)
393  s->intra_dc_precision -= 8;
394 
395  if (s->intra_dc_precision < 0) {
396  av_log(avctx, AV_LOG_ERROR,
397  "intra dc precision must be positive, note some applications use"
398  " 0 and some 8 as base meaning 8bit, the value must not be smaller than that\n");
399  return AVERROR(EINVAL);
400  }
401 
402  if (avctx->codec_id == AV_CODEC_ID_AMV || (avctx->active_thread_type & FF_THREAD_SLICE))
403  s->huffman = 0;
404 
405  if (s->intra_dc_precision > (avctx->codec_id == AV_CODEC_ID_MPEG2VIDEO ? 3 : 0)) {
406  av_log(avctx, AV_LOG_ERROR, "intra dc precision too large\n");
407  return AVERROR(EINVAL);
408  }
409  s->user_specified_pts = AV_NOPTS_VALUE;
410 
411  if (s->gop_size <= 1) {
412  s->intra_only = 1;
413  s->gop_size = 12;
414  } else {
415  s->intra_only = 0;
416  }
417 
418  /* Fixed QSCALE */
419  s->fixed_qscale = !!(avctx->flags & AV_CODEC_FLAG_QSCALE);
420 
421  s->adaptive_quant = (s->avctx->lumi_masking ||
422  s->avctx->dark_masking ||
423  s->avctx->temporal_cplx_masking ||
424  s->avctx->spatial_cplx_masking ||
425  s->avctx->p_masking ||
426  s->border_masking ||
427  (s->mpv_flags & FF_MPV_FLAG_QP_RD)) &&
428  !s->fixed_qscale;
429 
430  s->loop_filter = !!(s->avctx->flags & AV_CODEC_FLAG_LOOP_FILTER);
431 
432  if (avctx->rc_max_rate && !avctx->rc_buffer_size) {
433  switch(avctx->codec_id) {
434  case AV_CODEC_ID_MPEG1VIDEO:
435  case AV_CODEC_ID_MPEG2VIDEO:
436  avctx->rc_buffer_size = FFMAX(avctx->rc_max_rate, 15000000) * 112LL / 15000000 * 16384;
437  break;
438  case AV_CODEC_ID_MPEG4:
439  case AV_CODEC_ID_MSMPEG4V1:
440  case AV_CODEC_ID_MSMPEG4V2:
441  case AV_CODEC_ID_MSMPEG4V3:
442  if (avctx->rc_max_rate >= 15000000) {
443  avctx->rc_buffer_size = 320 + (avctx->rc_max_rate - 15000000LL) * (760-320) / (38400000 - 15000000);
444  } else if(avctx->rc_max_rate >= 2000000) {
445  avctx->rc_buffer_size = 80 + (avctx->rc_max_rate - 2000000LL) * (320- 80) / (15000000 - 2000000);
446  } else if(avctx->rc_max_rate >= 384000) {
447  avctx->rc_buffer_size = 40 + (avctx->rc_max_rate - 384000LL) * ( 80- 40) / ( 2000000 - 384000);
448  } else
449  avctx->rc_buffer_size = 40;
450  avctx->rc_buffer_size *= 16384;
451  break;
452  }
453  if (avctx->rc_buffer_size) {
454  av_log(avctx, AV_LOG_INFO, "Automatically choosing VBV buffer size of %d kbyte\n", avctx->rc_buffer_size/8192);
455  }
456  }
457 
458  if ((!avctx->rc_max_rate) != (!avctx->rc_buffer_size)) {
459  av_log(avctx, AV_LOG_ERROR, "Either both buffer size and max rate or neither must be specified\n");
460  return -1;
461  }
462 
463  if (avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate) {
464  av_log(avctx, AV_LOG_INFO,
465  "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
466  }
467 
468  if (avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate) {
469  av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
470  return -1;
471  }
472 
473  if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
474  av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
475  return -1;
476  }
477 
478  if (avctx->rc_max_rate &&
479  avctx->rc_max_rate == avctx->bit_rate &&
480  avctx->rc_max_rate != avctx->rc_min_rate) {
481  av_log(avctx, AV_LOG_INFO,
482  "impossible bitrate constraints, this will fail\n");
483  }
484 
485  if (avctx->rc_buffer_size &&
486  avctx->bit_rate * (int64_t)avctx->time_base.num >
487  avctx->rc_buffer_size * (int64_t)avctx->time_base.den) {
488  av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
489  return -1;
490  }
491 
492  if (!s->fixed_qscale &&
493  avctx->bit_rate * av_q2d(avctx->time_base) >
494  avctx->bit_rate_tolerance) {
495  av_log(avctx, AV_LOG_WARNING,
496  "bitrate tolerance %d too small for bitrate %"PRId64", overriding\n", avctx->bit_rate_tolerance, avctx->bit_rate);
497  avctx->bit_rate_tolerance = 5 * avctx->bit_rate * av_q2d(avctx->time_base);
498  }
499 
500  if (s->avctx->rc_max_rate &&
501  s->avctx->rc_min_rate == s->avctx->rc_max_rate &&
502  (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
503  s->codec_id == AV_CODEC_ID_MPEG2VIDEO) &&
504  90000LL * (avctx->rc_buffer_size - 1) >
505  s->avctx->rc_max_rate * 0xFFFFLL) {
506  av_log(avctx, AV_LOG_INFO,
507  "Warning vbv_delay will be set to 0xFFFF (=VBR) as the "
508  "specified vbv buffer is too large for the given bitrate!\n");
509  }
510 
511  if ((s->avctx->flags & AV_CODEC_FLAG_4MV) && s->codec_id != AV_CODEC_ID_MPEG4 &&
512  s->codec_id != AV_CODEC_ID_H263 && s->codec_id != AV_CODEC_ID_H263P &&
513  s->codec_id != AV_CODEC_ID_FLV1) {
514  av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
515  return -1;
516  }
517 
518  if (s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE) {
519  av_log(avctx, AV_LOG_ERROR,
520  "OBMC is only supported with simple mb decision\n");
521  return -1;
522  }
523 
524  if (s->quarter_sample && s->codec_id != AV_CODEC_ID_MPEG4) {
525  av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
526  return -1;
527  }
528 
529  if (s->max_b_frames &&
530  s->codec_id != AV_CODEC_ID_MPEG4 &&
531  s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
532  s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
533  av_log(avctx, AV_LOG_ERROR, "B-frames not supported by codec\n");
534  return -1;
535  }
536  if (s->max_b_frames < 0) {
537  av_log(avctx, AV_LOG_ERROR,
538  "max b frames must be 0 or positive for mpegvideo based encoders\n");
539  return -1;
540  }
541 
542  if ((s->codec_id == AV_CODEC_ID_MPEG4 ||
543  s->codec_id == AV_CODEC_ID_H263 ||
544  s->codec_id == AV_CODEC_ID_H263P) &&
545  (avctx->sample_aspect_ratio.num > 255 ||
546  avctx->sample_aspect_ratio.den > 255)) {
547  av_log(avctx, AV_LOG_WARNING,
548  "Invalid pixel aspect ratio %i/%i, limit is 255/255 reducing\n",
549  avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
550  av_reduce(&avctx->sample_aspect_ratio.num, &avctx->sample_aspect_ratio.den,
551  avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den, 255);
552  }
553 
554  if ((s->codec_id == AV_CODEC_ID_H263 ||
555  s->codec_id == AV_CODEC_ID_H263P) &&
556  (avctx->width > 2048 ||
557  avctx->height > 1152 )) {
558  av_log(avctx, AV_LOG_ERROR, "H.263 does not support resolutions above 2048x1152\n");
559  return -1;
560  }
561  if ((s->codec_id == AV_CODEC_ID_H263 ||
562  s->codec_id == AV_CODEC_ID_H263P) &&
563  ((avctx->width &3) ||
564  (avctx->height&3) )) {
565  av_log(avctx, AV_LOG_ERROR, "w/h must be a multiple of 4\n");
566  return -1;
567  }
568 
569  if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO &&
570  (avctx->width > 4095 ||
571  avctx->height > 4095 )) {
572  av_log(avctx, AV_LOG_ERROR, "MPEG-1 does not support resolutions above 4095x4095\n");
573  return -1;
574  }
575 
576  if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO &&
577  (avctx->width > 16383 ||
578  avctx->height > 16383 )) {
579  av_log(avctx, AV_LOG_ERROR, "MPEG-2 does not support resolutions above 16383x16383\n");
580  return -1;
581  }
582 
583  if (s->codec_id == AV_CODEC_ID_RV10 &&
584  (avctx->width &15 ||
585  avctx->height&15 )) {
586  av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 16\n");
587  return AVERROR(EINVAL);
588  }
589 
590  if (s->codec_id == AV_CODEC_ID_RV20 &&
591  (avctx->width &3 ||
592  avctx->height&3 )) {
593  av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 4\n");
594  return AVERROR(EINVAL);
595  }
596 
597  if ((s->codec_id == AV_CODEC_ID_WMV1 ||
598  s->codec_id == AV_CODEC_ID_WMV2) &&
599  avctx->width & 1) {
600  av_log(avctx, AV_LOG_ERROR, "width must be multiple of 2\n");
601  return -1;
602  }
603 
604  if ((s->avctx->flags & (AV_CODEC_FLAG_INTERLACED_DCT | AV_CODEC_FLAG_INTERLACED_ME)) &&
605  s->codec_id != AV_CODEC_ID_MPEG4 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
606  av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
607  return -1;
608  }
609 
610 #if FF_API_PRIVATE_OPT
611  FF_DISABLE_DEPRECATION_WARNINGS
612  if (avctx->mpeg_quant)
613  s->mpeg_quant = avctx->mpeg_quant;
614  FF_ENABLE_DEPRECATION_WARNINGS
615 #endif
616 
617  // FIXME mpeg2 uses that too
618  if (s->mpeg_quant && ( s->codec_id != AV_CODEC_ID_MPEG4
619  && s->codec_id != AV_CODEC_ID_MPEG2VIDEO)) {
620  av_log(avctx, AV_LOG_ERROR,
621  "mpeg2 style quantization not supported by codec\n");
622  return -1;
623  }
624 
625  if ((s->mpv_flags & FF_MPV_FLAG_CBP_RD) && !avctx->trellis) {
626  av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
627  return -1;
628  }
629 
630  if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
631  s->avctx->mb_decision != FF_MB_DECISION_RD) {
632  av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
633  return -1;
634  }
635 
636  if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) &&
637  (s->codec_id == AV_CODEC_ID_AMV ||
638  s->codec_id == AV_CODEC_ID_MJPEG)) {
639  // Used to produce garbage with MJPEG.
640  av_log(avctx, AV_LOG_ERROR,
641  "QP RD is no longer compatible with MJPEG or AMV\n");
642  return -1;
643  }
644 
645 #if FF_API_PRIVATE_OPT
646 FF_DISABLE_DEPRECATION_WARNINGS
647  if (avctx->scenechange_threshold)
648  s->scenechange_threshold = avctx->scenechange_threshold;
649 FF_ENABLE_DEPRECATION_WARNINGS
650 #endif
651 
652  if (s->scenechange_threshold < 1000000000 &&
653  (s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)) {
654  av_log(avctx, AV_LOG_ERROR,
655  "closed gop with scene change detection are not supported yet, "
656  "set threshold to 1000000000\n");
657  return -1;
658  }
659 
660  if (s->avctx->flags & AV_CODEC_FLAG_LOW_DELAY) {
661  if (s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
662  s->strict_std_compliance >= FF_COMPLIANCE_NORMAL) {
663  av_log(avctx, AV_LOG_ERROR,
664  "low delay forcing is only available for mpeg2, "
665  "set strict_std_compliance to 'unofficial' or lower in order to allow it\n");
666  return -1;
667  }
668  if (s->max_b_frames != 0) {
669  av_log(avctx, AV_LOG_ERROR,
670  "B-frames cannot be used with low delay\n");
671  return -1;
672  }
673  }
674 
675  if (s->q_scale_type == 1) {
676  if (avctx->qmax > 28) {
677  av_log(avctx, AV_LOG_ERROR,
678  "non linear quant only supports qmax <= 28 currently\n");
679  return -1;
680  }
681  }
682 
683  if (avctx->slices > 1 &&
684  (avctx->codec_id == AV_CODEC_ID_FLV1 || avctx->codec_id == AV_CODEC_ID_H261)) {
685  av_log(avctx, AV_LOG_ERROR, "Multiple slices are not supported by this codec\n");
686  return AVERROR(EINVAL);
687  }
688 
689  if (s->avctx->thread_count > 1 &&
690  s->codec_id != AV_CODEC_ID_MPEG4 &&
691  s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
692  s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
693  s->codec_id != AV_CODEC_ID_MJPEG &&
694  (s->codec_id != AV_CODEC_ID_H263P)) {
695  av_log(avctx, AV_LOG_ERROR,
696  "multi threaded encoding not supported by codec\n");
697  return -1;
698  }
699 
700  if (s->avctx->thread_count < 1) {
701  av_log(avctx, AV_LOG_ERROR,
702  "automatic thread number detection not supported by codec, "
703  "patch welcome\n");
704  return -1;
705  }
706 
707  if (!avctx->time_base.den || !avctx->time_base.num) {
708  av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
709  return -1;
710  }
711 
712 #if FF_API_PRIVATE_OPT
713 FF_DISABLE_DEPRECATION_WARNINGS
714  if (avctx->b_frame_strategy)
715  s->b_frame_strategy = avctx->b_frame_strategy;
716  if (avctx->b_sensitivity != 40)
717  s->b_sensitivity = avctx->b_sensitivity;
718 FF_ENABLE_DEPRECATION_WARNINGS
719 #endif
720 
721  if (s->b_frame_strategy && (avctx->flags & AV_CODEC_FLAG_PASS2)) {
722  av_log(avctx, AV_LOG_INFO,
723  "notice: b_frame_strategy only affects the first pass\n");
724  s->b_frame_strategy = 0;
725  }
726 
727  i = av_gcd(avctx->time_base.den, avctx->time_base.num);
728  if (i > 1) {
729  av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
730  avctx->time_base.den /= i;
731  avctx->time_base.num /= i;
732  //return -1;
733  }
734 
735  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) {
736  // (a + x * 3 / 8) / x
737  s->intra_quant_bias = 3 << (QUANT_BIAS_SHIFT - 3);
738  s->inter_quant_bias = 0;
739  } else {
740  s->intra_quant_bias = 0;
741  // (a - x / 4) / x
742  s->inter_quant_bias = -(1 << (QUANT_BIAS_SHIFT - 2));
743  }
744 
745  if (avctx->qmin > avctx->qmax || avctx->qmin <= 0) {
746  av_log(avctx, AV_LOG_ERROR, "qmin and or qmax are invalid, they must be 0 < min <= max\n");
747  return AVERROR(EINVAL);
748  }
749 
750  av_log(avctx, AV_LOG_DEBUG, "intra_quant_bias = %d inter_quant_bias = %d\n",s->intra_quant_bias,s->inter_quant_bias);
751 
752  if (avctx->codec_id == AV_CODEC_ID_MPEG4 &&
753  s->avctx->time_base.den > (1 << 16) - 1) {
754  av_log(avctx, AV_LOG_ERROR,
755  "timebase %d/%d not supported by MPEG 4 standard, "
756  "the maximum admitted value for the timebase denominator "
757  "is %d\n", s->avctx->time_base.num, s->avctx->time_base.den,
758  (1 << 16) - 1);
759  return -1;
760  }
761  s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
762 
763  switch (avctx->codec->id) {
764  case AV_CODEC_ID_MPEG1VIDEO:
765  s->out_format = FMT_MPEG1;
766  s->low_delay = !!(s->avctx->flags & AV_CODEC_FLAG_LOW_DELAY);
767  avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
768  break;
769  case AV_CODEC_ID_MPEG2VIDEO:
770  s->out_format = FMT_MPEG1;
771  s->low_delay = !!(s->avctx->flags & AV_CODEC_FLAG_LOW_DELAY);
772  avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
773  s->rtp_mode = 1;
774  break;
775  case AV_CODEC_ID_MJPEG:
776  case AV_CODEC_ID_AMV:
777  s->out_format = FMT_MJPEG;
778  s->intra_only = 1; /* force intra only for jpeg */
779  if (!CONFIG_MJPEG_ENCODER ||
780  ff_mjpeg_encode_init(s) < 0)
781  return -1;
782  avctx->delay = 0;
783  s->low_delay = 1;
784  break;
785  case AV_CODEC_ID_H261:
786  if (!CONFIG_H261_ENCODER)
787  return -1;
788  if (ff_h261_get_picture_format(s->width, s->height) < 0) {
789  av_log(avctx, AV_LOG_ERROR,
790  "The specified picture size of %dx%d is not valid for the "
791  "H.261 codec.\nValid sizes are 176x144, 352x288\n",
792  s->width, s->height);
793  return -1;
794  }
795  s->out_format = FMT_H261;
796  avctx->delay = 0;
797  s->low_delay = 1;
798  s->rtp_mode = 0; /* Sliced encoding not supported */
799  break;
800  case AV_CODEC_ID_H263:
801  if (!CONFIG_H263_ENCODER)
802  return -1;
803  if (ff_match_2uint16(ff_h263_format, FF_ARRAY_ELEMS(ff_h263_format),
804  s->width, s->height) == 8) {
805  av_log(avctx, AV_LOG_ERROR,
806  "The specified picture size of %dx%d is not valid for "
807  "the H.263 codec.\nValid sizes are 128x96, 176x144, "
808  "352x288, 704x576, and 1408x1152. "
809  "Try H.263+.\n", s->width, s->height);
810  return -1;
811  }
812  s->out_format = FMT_H263;
813  avctx->delay = 0;
814  s->low_delay = 1;
815  break;
816  case AV_CODEC_ID_H263P:
817  s->out_format = FMT_H263;
818  s->h263_plus = 1;
819  /* Fx */
820  s->h263_aic = (avctx->flags & AV_CODEC_FLAG_AC_PRED) ? 1 : 0;
821  s->modified_quant = s->h263_aic;
822  s->loop_filter = (avctx->flags & AV_CODEC_FLAG_LOOP_FILTER) ? 1 : 0;
823  s->unrestricted_mv = s->obmc || s->loop_filter || s->umvplus;
824 
825  /* /Fx */
826  /* These are just to be sure */
827  avctx->delay = 0;
828  s->low_delay = 1;
829  break;
830  case AV_CODEC_ID_FLV1:
831  s->out_format = FMT_H263;
832  s->h263_flv = 2; /* format = 1; 11-bit codes */
833  s->unrestricted_mv = 1;
834  s->rtp_mode = 0; /* don't allow GOB */
835  avctx->delay = 0;
836  s->low_delay = 1;
837  break;
838  case AV_CODEC_ID_RV10:
839  s->out_format = FMT_H263;
840  avctx->delay = 0;
841  s->low_delay = 1;
842  break;
843  case AV_CODEC_ID_RV20:
844  s->out_format = FMT_H263;
845  avctx->delay = 0;
846  s->low_delay = 1;
847  s->modified_quant = 1;
848  s->h263_aic = 1;
849  s->h263_plus = 1;
850  s->loop_filter = 1;
851  s->unrestricted_mv = 0;
852  break;
853  case AV_CODEC_ID_MPEG4:
854  s->out_format = FMT_H263;
855  s->h263_pred = 1;
856  s->unrestricted_mv = 1;
857  s->low_delay = s->max_b_frames ? 0 : 1;
858  avctx->delay = s->low_delay ? 0 : (s->max_b_frames + 1);
859  break;
860  case AV_CODEC_ID_MSMPEG4V2:
861  s->out_format = FMT_H263;
862  s->h263_pred = 1;
863  s->unrestricted_mv = 1;
864  s->msmpeg4_version = 2;
865  avctx->delay = 0;
866  s->low_delay = 1;
867  break;
868  case AV_CODEC_ID_MSMPEG4V3:
869  s->out_format = FMT_H263;
870  s->h263_pred = 1;
871  s->unrestricted_mv = 1;
872  s->msmpeg4_version = 3;
873  s->flipflop_rounding = 1;
874  avctx->delay = 0;
875  s->low_delay = 1;
876  break;
877  case AV_CODEC_ID_WMV1:
878  s->out_format = FMT_H263;
879  s->h263_pred = 1;
880  s->unrestricted_mv = 1;
881  s->msmpeg4_version = 4;
882  s->flipflop_rounding = 1;
883  avctx->delay = 0;
884  s->low_delay = 1;
885  break;
886  case AV_CODEC_ID_WMV2:
887  s->out_format = FMT_H263;
888  s->h263_pred = 1;
889  s->unrestricted_mv = 1;
890  s->msmpeg4_version = 5;
891  s->flipflop_rounding = 1;
892  avctx->delay = 0;
893  s->low_delay = 1;
894  break;
895  default:
896  return -1;
897  }
898 
899 #if FF_API_PRIVATE_OPT
900  FF_DISABLE_DEPRECATION_WARNINGS
901  if (avctx->noise_reduction)
902  s->noise_reduction = avctx->noise_reduction;
903  FF_ENABLE_DEPRECATION_WARNINGS
904 #endif
905 
906  avctx->has_b_frames = !s->low_delay;
907 
908  s->encoding = 1;
909 
910  s->progressive_frame =
911  s->progressive_sequence = !(avctx->flags & (AV_CODEC_FLAG_INTERLACED_DCT |
912  AV_CODEC_FLAG_INTERLACED_ME) ||
913  s->alternate_scan);
914 
915  /* init */
916  ff_mpv_idct_init(s);
917  if (ff_mpv_common_init(s) < 0)
918  return -1;
919 
920  ff_fdctdsp_init(&s->fdsp, avctx);
921  ff_me_cmp_init(&s->mecc, avctx);
922  ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
923  ff_pixblockdsp_init(&s->pdsp, avctx);
924  ff_qpeldsp_init(&s->qdsp);
925 
926  if (s->msmpeg4_version) {
927  FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
928  2 * 2 * (MAX_LEVEL + 1) *
929  (MAX_RUN + 1) * 2 * sizeof(int), fail);
930  }
931  FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
932 
933  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix, 64 * 32 * sizeof(int), fail);
934  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix, 64 * 32 * sizeof(int), fail);
935  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix, 64 * 32 * sizeof(int), fail);
936  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
937  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
938  FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
939  FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture,
940  MAX_PICTURE_COUNT * sizeof(Picture *), fail);
941  FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture,
942  MAX_PICTURE_COUNT * sizeof(Picture *), fail);
943 
944 
945  if (s->noise_reduction) {
946  FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset,
947  2 * 64 * sizeof(uint16_t), fail);
948  }
949 
950  ff_dct_encode_init(s);
951 
952  if ((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
953  s->chroma_qscale_table = ff_h263_chroma_qscale_table;
954 
955  if (s->slice_context_count > 1) {
956  s->rtp_mode = 1;
957 
958  if (avctx->codec_id == AV_CODEC_ID_H263P)
959  s->h263_slice_structured = 1;
960  }
961 
962  s->quant_precision = 5;
963 
964 #if FF_API_PRIVATE_OPT
965 FF_DISABLE_DEPRECATION_WARNINGS
966  if (avctx->frame_skip_threshold)
967  s->frame_skip_threshold = avctx->frame_skip_threshold;
968  if (avctx->frame_skip_factor)
969  s->frame_skip_factor = avctx->frame_skip_factor;
970  if (avctx->frame_skip_exp)
971  s->frame_skip_exp = avctx->frame_skip_exp;
972  if (avctx->frame_skip_cmp != FF_CMP_DCTMAX)
973  s->frame_skip_cmp = avctx->frame_skip_cmp;
974 FF_ENABLE_DEPRECATION_WARNINGS
975 #endif
976 
977  ff_set_cmp(&s->mecc, s->mecc.ildct_cmp, s->avctx->ildct_cmp);
978  ff_set_cmp(&s->mecc, s->mecc.frame_skip_cmp, s->frame_skip_cmp);
979 
980  if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
981  ff_h261_encode_init(s);
982  if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
983  ff_h263_encode_init(s);
984  if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
985  if ((ret = ff_msmpeg4_encode_init(s)) < 0)
986  return ret;
987  if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
988  && s->out_format == FMT_MPEG1)
989  ff_mpeg1_encode_init(s);
990 
991  /* init q matrix */
992  for (i = 0; i < 64; i++) {
993  int j = s->idsp.idct_permutation[i];
994  if (CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4 &&
995  s->mpeg_quant) {
996  s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
997  s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
998  } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
999  s->intra_matrix[j] =
1000  s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1001  } else {
1002  /* MPEG-1/2 */
1003  s->chroma_intra_matrix[j] =
1004  s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
1005  s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1006  }
1007  if (s->avctx->intra_matrix)
1008  s->intra_matrix[j] = s->avctx->intra_matrix[i];
1009  if (s->avctx->inter_matrix)
1010  s->inter_matrix[j] = s->avctx->inter_matrix[i];
1011  }
1012 
1013  /* precompute matrix */
1014  /* for mjpeg, we do include qscale in the matrix */
1015  if (s->out_format != FMT_MJPEG) {
1016  ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
1017  s->intra_matrix, s->intra_quant_bias, avctx->qmin,
1018  31, 1);
1019  ff_convert_matrix(s, s->q_inter_matrix, s->q_inter_matrix16,
1020  s->inter_matrix, s->inter_quant_bias, avctx->qmin,
1021  31, 0);
1022  }
1023 
1024  if (ff_rate_control_init(s) < 0)
1025  return -1;
1026 
1027 #if FF_API_PRIVATE_OPT
1028  FF_DISABLE_DEPRECATION_WARNINGS
1029  if (avctx->brd_scale)
1030  s->brd_scale = avctx->brd_scale;
1031 
1032  if (avctx->prediction_method)
1033  s->pred = avctx->prediction_method + 1;
1034  FF_ENABLE_DEPRECATION_WARNINGS
1035 #endif
1036 
1037  if (s->b_frame_strategy == 2) {
1038  for (i = 0; i < s->max_b_frames + 2; i++) {
1039  s->tmp_frames[i] = av_frame_alloc();
1040  if (!s->tmp_frames[i])
1041  return AVERROR(ENOMEM);
1042 
1043  s->tmp_frames[i]->format = AV_PIX_FMT_YUV420P;
1044  s->tmp_frames[i]->width = s->width >> s->brd_scale;
1045  s->tmp_frames[i]->height = s->height >> s->brd_scale;
1046 
1047  ret = av_frame_get_buffer(s->tmp_frames[i], 32);
1048  if (ret < 0)
1049  return ret;
1050  }
1051  }
1052 
1053  cpb_props = ff_add_cpb_side_data(avctx);
1054  if (!cpb_props)
1055  return AVERROR(ENOMEM);
1056  cpb_props->max_bitrate = avctx->rc_max_rate;
1057  cpb_props->min_bitrate = avctx->rc_min_rate;
1058  cpb_props->avg_bitrate = avctx->bit_rate;
1059  cpb_props->buffer_size = avctx->rc_buffer_size;
1060 
1061  return 0;
1062 fail:
1063  ff_mpv_encode_end(avctx);
1064  return AVERROR_UNKNOWN;
1065 }
1066 
1067 av_cold int ff_mpv_encode_end(AVCodecContext *avctx)
1068 {
1069  MpegEncContext *s = avctx->priv_data;
1070  int i;
1071 
1072  ff_rate_control_uninit(s);
1073 
1074  ff_mpv_common_end(s);
1075  if (CONFIG_MJPEG_ENCODER &&
1076  s->out_format == FMT_MJPEG)
1077  ff_mjpeg_encode_close(s);
1078 
1079  av_freep(&avctx->extradata);
1080 
1081  for (i = 0; i < FF_ARRAY_ELEMS(s->tmp_frames); i++)
1082  av_frame_free(&s->tmp_frames[i]);
1083 
1084  ff_free_picture_tables(&s->new_picture);
1085  ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1086 
1087  av_freep(&s->avctx->stats_out);
1088  av_freep(&s->ac_stats);
1089 
1090  if(s->q_chroma_intra_matrix != s->q_intra_matrix ) av_freep(&s->q_chroma_intra_matrix);
1091  if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
1092  s->q_chroma_intra_matrix= NULL;
1093  s->q_chroma_intra_matrix16= NULL;
1094  av_freep(&s->q_intra_matrix);
1095  av_freep(&s->q_inter_matrix);
1096  av_freep(&s->q_intra_matrix16);
1097  av_freep(&s->q_inter_matrix16);
1098  av_freep(&s->input_picture);
1099  av_freep(&s->reordered_input_picture);
1100  av_freep(&s->dct_offset);
1101 
1102  return 0;
1103 }
1104 
1105 static int get_sae(uint8_t *src, int ref, int stride)
1106 {
1107  int x,y;
1108  int acc = 0;
1109 
1110  for (y = 0; y < 16; y++) {
1111  for (x = 0; x < 16; x++) {
1112  acc += FFABS(src[x + y * stride] - ref);
1113  }
1114  }
1115 
1116  return acc;
1117 }
1118 
1119 static int get_intra_count(MpegEncContext *s, uint8_t *src,
1120  uint8_t *ref, int stride)
1121 {
1122  int x, y, w, h;
1123  int acc = 0;
1124 
1125  w = s->width & ~15;
1126  h = s->height & ~15;
1127 
1128  for (y = 0; y < h; y += 16) {
1129  for (x = 0; x < w; x += 16) {
1130  int offset = x + y * stride;
1131  int sad = s->mecc.sad[0](NULL, src + offset, ref + offset,
1132  stride, 16);
1133  int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8;
1134  int sae = get_sae(src + offset, mean, stride);
1135 
1136  acc += sae + 500 < sad;
1137  }
1138  }
1139  return acc;
1140 }
1141 
1142 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
1143 {
1144  return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 1,
1145  s->chroma_x_shift, s->chroma_y_shift, s->out_format,
1146  s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
1147  &s->linesize, &s->uvlinesize);
1148 }
1149 
1150 static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
1151 {
1152  Picture *pic = NULL;
1153  int64_t pts;
1154  int i, display_picture_number = 0, ret;
1155  int encoding_delay = s->max_b_frames ? s->max_b_frames
1156  : (s->low_delay ? 0 : 1);
1157  int flush_offset = 1;
1158  int direct = 1;
1159 
1160  if (pic_arg) {
1161  pts = pic_arg->pts;
1162  display_picture_number = s->input_picture_number++;
1163 
1164  if (pts != AV_NOPTS_VALUE) {
1165  if (s->user_specified_pts != AV_NOPTS_VALUE) {
1166  int64_t last = s->user_specified_pts;
1167 
1168  if (pts <= last) {
1169  av_log(s->avctx, AV_LOG_ERROR,
1170  "Invalid pts (%"PRId64") <= last (%"PRId64")\n",
1171  pts, last);
1172  return AVERROR(EINVAL);
1173  }
1174 
1175  if (!s->low_delay && display_picture_number == 1)
1176  s->dts_delta = pts - last;
1177  }
1178  s->user_specified_pts = pts;
1179  } else {
1180  if (s->user_specified_pts != AV_NOPTS_VALUE) {
1181  s->user_specified_pts =
1182  pts = s->user_specified_pts + 1;
1183  av_log(s->avctx, AV_LOG_INFO,
1184  "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n",
1185  pts);
1186  } else {
1187  pts = display_picture_number;
1188  }
1189  }
1190 
1191  if (!pic_arg->buf[0] ||
1192  pic_arg->linesize[0] != s->linesize ||
1193  pic_arg->linesize[1] != s->uvlinesize ||
1194  pic_arg->linesize[2] != s->uvlinesize)
1195  direct = 0;
1196  if ((s->width & 15) || (s->height & 15))
1197  direct = 0;
1198  if (((intptr_t)(pic_arg->data[0])) & (STRIDE_ALIGN-1))
1199  direct = 0;
1200  if (s->linesize & (STRIDE_ALIGN-1))
1201  direct = 0;
1202 
1203  ff_dlog(s->avctx, "%d %d %"PTRDIFF_SPECIFIER" %"PTRDIFF_SPECIFIER"\n", pic_arg->linesize[0],
1204  pic_arg->linesize[1], s->linesize, s->uvlinesize);
1205 
1206  i = ff_find_unused_picture(s->avctx, s->picture, direct);
1207  if (i < 0)
1208  return i;
1209 
1210  pic = &s->picture[i];
1211  pic->reference = 3;
1212 
1213  if (direct) {
1214  if ((ret = av_frame_ref(pic->f, pic_arg)) < 0)
1215  return ret;
1216  }
1217  ret = alloc_picture(s, pic, direct);
1218  if (ret < 0)
1219  return ret;
1220 
1221  if (!direct) {
1222  if (pic->f->data[0] + INPLACE_OFFSET == pic_arg->data[0] &&
1223  pic->f->data[1] + INPLACE_OFFSET == pic_arg->data[1] &&
1224  pic->f->data[2] + INPLACE_OFFSET == pic_arg->data[2]) {
1225  // empty
1226  } else {
1227  int h_chroma_shift, v_chroma_shift;
1228  av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1229  &h_chroma_shift,
1230  &v_chroma_shift);
1231 
1232  for (i = 0; i < 3; i++) {
1233  int src_stride = pic_arg->linesize[i];
1234  int dst_stride = i ? s->uvlinesize : s->linesize;
1235  int h_shift = i ? h_chroma_shift : 0;
1236  int v_shift = i ? v_chroma_shift : 0;
1237  int w = s->width >> h_shift;
1238  int h = s->height >> v_shift;
1239  uint8_t *src = pic_arg->data[i];
1240  uint8_t *dst = pic->f->data[i];
1241  int vpad = 16;
1242 
1243  if ( s->codec_id == AV_CODEC_ID_MPEG2VIDEO
1244  && !s->progressive_sequence
1245  && FFALIGN(s->height, 32) - s->height > 16)
1246  vpad = 32;
1247 
1248  if (!s->avctx->rc_buffer_size)
1249  dst += INPLACE_OFFSET;
1250 
1251  if (src_stride == dst_stride)
1252  memcpy(dst, src, src_stride * h);
1253  else {
1254  int h2 = h;
1255  uint8_t *dst2 = dst;
1256  while (h2--) {
1257  memcpy(dst2, src, w);
1258  dst2 += dst_stride;
1259  src += src_stride;
1260  }
1261  }
1262  if ((s->width & 15) || (s->height & (vpad-1))) {
1263  s->mpvencdsp.draw_edges(dst, dst_stride,
1264  w, h,
1265  16 >> h_shift,
1266  vpad >> v_shift,
1267  EDGE_BOTTOM);
1268  }
1269  }
1270  emms_c();
1271  }
1272  }
1273  ret = av_frame_copy_props(pic->f, pic_arg);
1274  if (ret < 0)
1275  return ret;
1276 
1277  pic->f->display_picture_number = display_picture_number;
1278  pic->f->pts = pts; // we set this here to avoid modifying pic_arg
1279  } else {
1280  /* Flushing: When we have not received enough input frames,
1281  * ensure s->input_picture[0] contains the first picture */
1282  for (flush_offset = 0; flush_offset < encoding_delay + 1; flush_offset++)
1283  if (s->input_picture[flush_offset])
1284  break;
1285 
1286  if (flush_offset <= 1)
1287  flush_offset = 1;
1288  else
1289  encoding_delay = encoding_delay - flush_offset + 1;
1290  }
1291 
1292  /* shift buffer entries */
1293  for (i = flush_offset; i < MAX_PICTURE_COUNT /*s->encoding_delay + 1*/; i++)
1294  s->input_picture[i - flush_offset] = s->input_picture[i];
1295 
1296  s->input_picture[encoding_delay] = (Picture*) pic;
1297 
1298  return 0;
1299 }
1300 
1301 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref)
1302 {
1303  int x, y, plane;
1304  int score = 0;
1305  int64_t score64 = 0;
1306 
1307  for (plane = 0; plane < 3; plane++) {
1308  const int stride = p->f->linesize[plane];
1309  const int bw = plane ? 1 : 2;
1310  for (y = 0; y < s->mb_height * bw; y++) {
1311  for (x = 0; x < s->mb_width * bw; x++) {
1312  int off = p->shared ? 0 : 16;
1313  uint8_t *dptr = p->f->data[plane] + 8 * (x + y * stride) + off;
1314  uint8_t *rptr = ref->f->data[plane] + 8 * (x + y * stride);
1315  int v = s->mecc.frame_skip_cmp[1](s, dptr, rptr, stride, 8);
1316 
1317  switch (FFABS(s->frame_skip_exp)) {
1318  case 0: score = FFMAX(score, v); break;
1319  case 1: score += FFABS(v); break;
1320  case 2: score64 += v * (int64_t)v; break;
1321  case 3: score64 += FFABS(v * (int64_t)v * v); break;
1322  case 4: score64 += (v * (int64_t)v) * (v * (int64_t)v); break;
1323  }
1324  }
1325  }
1326  }
1327  emms_c();
1328 
1329  if (score)
1330  score64 = score;
1331  if (s->frame_skip_exp < 0)
1332  score64 = pow(score64 / (double)(s->mb_width * s->mb_height),
1333  -1.0/s->frame_skip_exp);
1334 
1335  if (score64 < s->frame_skip_threshold)
1336  return 1;
1337  if (score64 < ((s->frame_skip_factor * (int64_t) s->lambda) >> 8))
1338  return 1;
1339  return 0;
1340 }
1341 
1342 static int encode_frame(AVCodecContext *c, AVFrame *frame)
1343 {
1344  AVPacket pkt = { 0 };
1345  int ret;
1346  int size = 0;
1347 
1348  av_init_packet(&pkt);
1349 
1350  ret = avcodec_send_frame(c, frame);
1351  if (ret < 0)
1352  return ret;
1353 
1354  do {
1355  ret = avcodec_receive_packet(c, &pkt);
1356  if (ret >= 0) {
1357  size += pkt.size;
1358  av_packet_unref(&pkt);
1359  } else if (ret < 0 && ret != AVERROR(EAGAIN) && ret != AVERROR_EOF)
1360  return ret;
1361  } while (ret >= 0);
1362 
1363  return size;
1364 }
1365 
1366 static int estimate_best_b_count(MpegEncContext *s)
1367 {
1368  const AVCodec *codec = avcodec_find_encoder(s->avctx->codec_id);
1369  const int scale = s->brd_scale;
1370  int width = s->width >> scale;
1371  int height = s->height >> scale;
1372  int i, j, out_size, p_lambda, b_lambda, lambda2;
1373  int64_t best_rd = INT64_MAX;
1374  int best_b_count = -1;
1375  int ret = 0;
1376 
1377  av_assert0(scale >= 0 && scale <= 3);
1378 
1379  //emms_c();
1380  //s->next_picture_ptr->quality;
1381  p_lambda = s->last_lambda_for[AV_PICTURE_TYPE_P];
1382  //p_lambda * FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
1383  b_lambda = s->last_lambda_for[AV_PICTURE_TYPE_B];
1384  if (!b_lambda) // FIXME we should do this somewhere else
1385  b_lambda = p_lambda;
1386  lambda2 = (b_lambda * b_lambda + (1 << FF_LAMBDA_SHIFT) / 2) >>
1387  FF_LAMBDA_SHIFT;
1388 
1389  for (i = 0; i < s->max_b_frames + 2; i++) {
1390  Picture pre_input, *pre_input_ptr = i ? s->input_picture[i - 1] :
1391  s->next_picture_ptr;
1392  uint8_t *data[4];
1393 
1394  if (pre_input_ptr && (!i || s->input_picture[i - 1])) {
1395  pre_input = *pre_input_ptr;
1396  memcpy(data, pre_input_ptr->f->data, sizeof(data));
1397 
1398  if (!pre_input.shared && i) {
1399  data[0] += INPLACE_OFFSET;
1400  data[1] += INPLACE_OFFSET;
1401  data[2] += INPLACE_OFFSET;
1402  }
1403 
1404  s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0],
1405  s->tmp_frames[i]->linesize[0],
1406  data[0],
1407  pre_input.f->linesize[0],
1408  width, height);
1409  s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1],
1410  s->tmp_frames[i]->linesize[1],
1411  data[1],
1412  pre_input.f->linesize[1],
1413  width >> 1, height >> 1);
1414  s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2],
1415  s->tmp_frames[i]->linesize[2],
1416  data[2],
1417  pre_input.f->linesize[2],
1418  width >> 1, height >> 1);
1419  }
1420  }
1421 
1422  for (j = 0; j < s->max_b_frames + 1; j++) {
1423  AVCodecContext *c;
1424  int64_t rd = 0;
1425 
1426  if (!s->input_picture[j])
1427  break;
1428 
1429  c = avcodec_alloc_context3(NULL);
1430  if (!c)
1431  return AVERROR(ENOMEM);
1432 
1433  c->width = width;
1434  c->height = height;
1435  c->flags = AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_PSNR;
1436  c->flags |= s->avctx->flags & AV_CODEC_FLAG_QPEL;
1437  c->mb_decision = s->avctx->mb_decision;
1438  c->me_cmp = s->avctx->me_cmp;
1439  c->mb_cmp = s->avctx->mb_cmp;
1440  c->me_sub_cmp = s->avctx->me_sub_cmp;
1441  c->pix_fmt = AV_PIX_FMT_YUV420P;
1442  c->time_base = s->avctx->time_base;
1443  c->max_b_frames = s->max_b_frames;
1444 
1445  ret = avcodec_open2(c, codec, NULL);
1446  if (ret < 0)
1447  goto fail;
1448 
1449  s->tmp_frames[0]->pict_type = AV_PICTURE_TYPE_I;
1450  s->tmp_frames[0]->quality = 1 * FF_QP2LAMBDA;
1451 
1452  out_size = encode_frame(c, s->tmp_frames[0]);
1453  if (out_size < 0) {
1454  ret = out_size;
1455  goto fail;
1456  }
1457 
1458  //rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1459 
1460  for (i = 0; i < s->max_b_frames + 1; i++) {
1461  int is_p = i % (j + 1) == j || i == s->max_b_frames;
1462 
1463  s->tmp_frames[i + 1]->pict_type = is_p ?
1464  AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1465  s->tmp_frames[i + 1]->quality = is_p ? p_lambda : b_lambda;
1466 
1467  out_size = encode_frame(c, s->tmp_frames[i + 1]);
1468  if (out_size < 0) {
1469  ret = out_size;
1470  goto fail;
1471  }
1472 
1473  rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1474  }
1475 
1476  /* get the delayed frames */
1477  out_size = encode_frame(c, NULL);
1478  if (out_size < 0) {
1479  ret = out_size;
1480  goto fail;
1481  }
1482  rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1483 
1484  rd += c->error[0] + c->error[1] + c->error[2];
1485 
1486  if (rd < best_rd) {
1487  best_rd = rd;
1488  best_b_count = j;
1489  }
1490 
1491 fail:
1492  avcodec_free_context(&c);
1493  if (ret < 0)
1494  return ret;
1495  }
1496 
1497  return best_b_count;
1498 }
1499 
1500 static int select_input_picture(MpegEncContext *s)
1501 {
1502  int i, ret;
1503 
1504  for (i = 1; i < MAX_PICTURE_COUNT; i++)
1505  s->reordered_input_picture[i - 1] = s->reordered_input_picture[i];
1506  s->reordered_input_picture[MAX_PICTURE_COUNT - 1] = NULL;
1507 
1508  /* set next picture type & ordering */
1509  if (!s->reordered_input_picture[0] && s->input_picture[0]) {
1510  if (s->frame_skip_threshold || s->frame_skip_factor) {
1511  if (s->picture_in_gop_number < s->gop_size &&
1512  s->next_picture_ptr &&
1513  skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
1514  // FIXME check that the gop check above is +-1 correct
1515  av_frame_unref(s->input_picture[0]->f);
1516 
1517  ff_vbv_update(s, 0);
1518 
1519  goto no_output_pic;
1520  }
1521  }
1522 
1523  if (/*s->picture_in_gop_number >= s->gop_size ||*/
1524  !s->next_picture_ptr || s->intra_only) {
1525  s->reordered_input_picture[0] = s->input_picture[0];
1526  s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_I;
1527  s->reordered_input_picture[0]->f->coded_picture_number =
1528  s->coded_picture_number++;
1529  } else {
1530  int b_frames = 0;
1531 
1532  if (s->avctx->flags & AV_CODEC_FLAG_PASS2) {
1533  for (i = 0; i < s->max_b_frames + 1; i++) {
1534  int pict_num = s->input_picture[0]->f->display_picture_number + i;
1535 
1536  if (pict_num >= s->rc_context.num_entries)
1537  break;
1538  if (!s->input_picture[i]) {
1539  s->rc_context.entry[pict_num - 1].new_pict_type = AV_PICTURE_TYPE_P;
1540  break;
1541  }
1542 
1543  s->input_picture[i]->f->pict_type =
1544  s->rc_context.entry[pict_num].new_pict_type;
1545  }
1546  }
1547 
1548  if (s->b_frame_strategy == 0) {
1549  b_frames = s->max_b_frames;
1550  while (b_frames && !s->input_picture[b_frames])
1551  b_frames--;
1552  } else if (s->b_frame_strategy == 1) {
1553  for (i = 1; i < s->max_b_frames + 1; i++) {
1554  if (s->input_picture[i] &&
1555  s->input_picture[i]->b_frame_score == 0) {
1556  s->input_picture[i]->b_frame_score =
1557  get_intra_count(s,
1558  s->input_picture[i ]->f->data[0],
1559  s->input_picture[i - 1]->f->data[0],
1560  s->linesize) + 1;
1561  }
1562  }
1563  for (i = 0; i < s->max_b_frames + 1; i++) {
1564  if (!s->input_picture[i] ||
1565  s->input_picture[i]->b_frame_score - 1 >
1566  s->mb_num / s->b_sensitivity)
1567  break;
1568  }
1569 
1570  b_frames = FFMAX(0, i - 1);
1571 
1572  /* reset scores */
1573  for (i = 0; i < b_frames + 1; i++) {
1574  s->input_picture[i]->b_frame_score = 0;
1575  }
1576  } else if (s->b_frame_strategy == 2) {
1577  b_frames = estimate_best_b_count(s);
1578  if (b_frames < 0)
1579  return b_frames;
1580  }
1581 
1582  emms_c();
1583 
1584  for (i = b_frames - 1; i >= 0; i--) {
1585  int type = s->input_picture[i]->f->pict_type;
1586  if (type && type != AV_PICTURE_TYPE_B)
1587  b_frames = i;
1588  }
1589  if (s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_B &&
1590  b_frames == s->max_b_frames) {
1591  av_log(s->avctx, AV_LOG_ERROR,
1592  "warning, too many B-frames in a row\n");
1593  }
1594 
1595  if (s->picture_in_gop_number + b_frames >= s->gop_size) {
1596  if ((s->mpv_flags & FF_MPV_FLAG_STRICT_GOP) &&
1597  s->gop_size > s->picture_in_gop_number) {
1598  b_frames = s->gop_size - s->picture_in_gop_number - 1;
1599  } else {
1600  if (s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP)
1601  b_frames = 0;
1602  s->input_picture[b_frames]->f->pict_type = AV_PICTURE_TYPE_I;
1603  }
1604  }
1605 
1606  if ((s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP) && b_frames &&
1607  s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_I)
1608  b_frames--;
1609 
1610  s->reordered_input_picture[0] = s->input_picture[b_frames];
1611  if (s->reordered_input_picture[0]->f->pict_type != AV_PICTURE_TYPE_I)
1612  s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_P;
1613  s->reordered_input_picture[0]->f->coded_picture_number =
1614  s->coded_picture_number++;
1615  for (i = 0; i < b_frames; i++) {
1616  s->reordered_input_picture[i + 1] = s->input_picture[i];
1617  s->reordered_input_picture[i + 1]->f->pict_type =
1618  AV_PICTURE_TYPE_B;
1619  s->reordered_input_picture[i + 1]->f->coded_picture_number =
1620  s->coded_picture_number++;
1621  }
1622  }
1623  }
1624 no_output_pic:
1625  ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1626 
1627  if (s->reordered_input_picture[0]) {
1628  s->reordered_input_picture[0]->reference =
1629  s->reordered_input_picture[0]->f->pict_type !=
1630  AV_PICTURE_TYPE_B ? 3 : 0;
1631 
1632  if ((ret = ff_mpeg_ref_picture(s->avctx, &s->new_picture, s->reordered_input_picture[0])))
1633  return ret;
1634 
1635  if (s->reordered_input_picture[0]->shared || s->avctx->rc_buffer_size) {
1636  // input is a shared pix, so we can't modify it -> allocate a new
1637  // one & ensure that the shared one is reuseable
1638 
1639  Picture *pic;
1640  int i = ff_find_unused_picture(s->avctx, s->picture, 0);
1641  if (i < 0)
1642  return i;
1643  pic = &s->picture[i];
1644 
1645  pic->reference = s->reordered_input_picture[0]->reference;
1646  if (alloc_picture(s, pic, 0) < 0) {
1647  return -1;
1648  }
1649 
1650  ret = av_frame_copy_props(pic->f, s->reordered_input_picture[0]->f);
1651  if (ret < 0)
1652  return ret;
1653 
1654  /* mark us unused / free shared pic */
1655  av_frame_unref(s->reordered_input_picture[0]->f);
1656  s->reordered_input_picture[0]->shared = 0;
1657 
1658  s->current_picture_ptr = pic;
1659  } else {
1660  // input is not a shared pix -> reuse buffer for current_pix
1661  s->current_picture_ptr = s->reordered_input_picture[0];
1662  for (i = 0; i < 4; i++) {
1663  s->new_picture.f->data[i] += INPLACE_OFFSET;
1664  }
1665  }
1666  ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1667  if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1668  s->current_picture_ptr)) < 0)
1669  return ret;
1670 
1671  s->picture_number = s->new_picture.f->display_picture_number;
1672  }
1673  return 0;
1674 }
1675 
1676 static void frame_end(MpegEncContext *s)
1677 {
1678  if (s->unrestricted_mv &&
1679  s->current_picture.reference &&
1680  !s->intra_only) {
1681  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
1682  int hshift = desc->log2_chroma_w;
1683  int vshift = desc->log2_chroma_h;
1684  s->mpvencdsp.draw_edges(s->current_picture.f->data[0],
1685  s->current_picture.f->linesize[0],
1686  s->h_edge_pos, s->v_edge_pos,
1687  EDGE_WIDTH, EDGE_WIDTH,
1688  EDGE_TOP | EDGE_BOTTOM);
1689  s->mpvencdsp.draw_edges(s->current_picture.f->data[1],
1690  s->current_picture.f->linesize[1],
1691  s->h_edge_pos >> hshift,
1692  s->v_edge_pos >> vshift,
1693  EDGE_WIDTH >> hshift,
1694  EDGE_WIDTH >> vshift,
1695  EDGE_TOP | EDGE_BOTTOM);
1696  s->mpvencdsp.draw_edges(s->current_picture.f->data[2],
1697  s->current_picture.f->linesize[2],
1698  s->h_edge_pos >> hshift,
1699  s->v_edge_pos >> vshift,
1700  EDGE_WIDTH >> hshift,
1701  EDGE_WIDTH >> vshift,
1702  EDGE_TOP | EDGE_BOTTOM);
1703  }
1704 
1705  emms_c();
1706 
1707  s->last_pict_type = s->pict_type;
1708  s->last_lambda_for [s->pict_type] = s->current_picture_ptr->f->quality;
1709  if (s->pict_type!= AV_PICTURE_TYPE_B)
1710  s->last_non_b_pict_type = s->pict_type;
1711 
1712 #if FF_API_CODED_FRAME
1713 FF_DISABLE_DEPRECATION_WARNINGS
1714  av_frame_unref(s->avctx->coded_frame);
1715  av_frame_copy_props(s->avctx->coded_frame, s->current_picture.f);
1716 FF_ENABLE_DEPRECATION_WARNINGS
1717 #endif
1718 #if FF_API_ERROR_FRAME
1719 FF_DISABLE_DEPRECATION_WARNINGS
1720  memcpy(s->current_picture.f->error, s->current_picture.encoding_error,
1721  sizeof(s->current_picture.encoding_error));
1722 FF_ENABLE_DEPRECATION_WARNINGS
1723 #endif
1724 }
1725 
1726 static void update_noise_reduction(MpegEncContext *s)
1727 {
1728  int intra, i;
1729 
1730  for (intra = 0; intra < 2; intra++) {
1731  if (s->dct_count[intra] > (1 << 16)) {
1732  for (i = 0; i < 64; i++) {
1733  s->dct_error_sum[intra][i] >>= 1;
1734  }
1735  s->dct_count[intra] >>= 1;
1736  }
1737 
1738  for (i = 0; i < 64; i++) {
1739  s->dct_offset[intra][i] = (s->noise_reduction *
1740  s->dct_count[intra] +
1741  s->dct_error_sum[intra][i] / 2) /
1742  (s->dct_error_sum[intra][i] + 1);
1743  }
1744  }
1745 }
1746 
1747 static int frame_start(MpegEncContext *s)
1748 {
1749  int ret;
1750 
1751  /* mark & release old frames */
1752  if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1753  s->last_picture_ptr != s->next_picture_ptr &&
1754  s->last_picture_ptr->f->buf[0]) {
1755  ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1756  }
1757 
1758  s->current_picture_ptr->f->pict_type = s->pict_type;
1759  s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1760 
1761  ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1762  if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1763  s->current_picture_ptr)) < 0)
1764  return ret;
1765 
1766  if (s->pict_type != AV_PICTURE_TYPE_B) {
1767  s->last_picture_ptr = s->next_picture_ptr;
1768  if (!s->droppable)
1769  s->next_picture_ptr = s->current_picture_ptr;
1770  }
1771 
1772  if (s->last_picture_ptr) {
1773  ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1774  if (s->last_picture_ptr->f->buf[0] &&
1775  (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1776  s->last_picture_ptr)) < 0)
1777  return ret;
1778  }
1779  if (s->next_picture_ptr) {
1780  ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1781  if (s->next_picture_ptr->f->buf[0] &&
1782  (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1783  s->next_picture_ptr)) < 0)
1784  return ret;
1785  }
1786 
1787  if (s->picture_structure!= PICT_FRAME) {
1788  int i;
1789  for (i = 0; i < 4; i++) {
1790  if (s->picture_structure == PICT_BOTTOM_FIELD) {
1791  s->current_picture.f->data[i] +=
1792  s->current_picture.f->linesize[i];
1793  }
1794  s->current_picture.f->linesize[i] *= 2;
1795  s->last_picture.f->linesize[i] *= 2;
1796  s->next_picture.f->linesize[i] *= 2;
1797  }
1798  }
1799 
1800  if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1801  s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1802  s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1803  } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1804  s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1805  s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1806  } else {
1807  s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1808  s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1809  }
1810 
1811  if (s->dct_error_sum) {
1812  av_assert2(s->noise_reduction && s->encoding);
1813  update_noise_reduction(s);
1814  }
1815 
1816  return 0;
1817 }
1818 
1819 int ff_mpv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
1820  const AVFrame *pic_arg, int *got_packet)
1821 {
1822  MpegEncContext *s = avctx->priv_data;
1823  int i, stuffing_count, ret;
1824  int context_count = s->slice_context_count;
1825 
1826  s->vbv_ignore_qmax = 0;
1827 
1828  s->picture_in_gop_number++;
1829 
1830  if (load_input_picture(s, pic_arg) < 0)
1831  return -1;
1832 
1833  if (select_input_picture(s) < 0) {
1834  return -1;
1835  }
1836 
1837  /* output? */
1838  if (s->new_picture.f->data[0]) {
1839  int growing_buffer = context_count == 1 && !pkt->data && !s->data_partitioning;
1840  int pkt_size = growing_buffer ? FFMAX(s->mb_width*s->mb_height*64+10000, avctx->internal->byte_buffer_size) - AV_INPUT_BUFFER_PADDING_SIZE
1841  :
1842  s->mb_width*s->mb_height*(MAX_MB_BYTES+100)+10000;
1843  if ((ret = ff_alloc_packet2(avctx, pkt, pkt_size, 0)) < 0)
1844  return ret;
1845  if (s->mb_info) {
1846  s->mb_info_ptr = av_packet_new_side_data(pkt,
1847  AV_PKT_DATA_H263_MB_INFO,
1848  s->mb_width*s->mb_height*12);
1849  s->prev_mb_info = s->last_mb_info = s->mb_info_size = 0;
1850  }
1851 
1852  for (i = 0; i < context_count; i++) {
1853  int start_y = s->thread_context[i]->start_mb_y;
1854  int end_y = s->thread_context[i]-> end_mb_y;
1855  int h = s->mb_height;
1856  uint8_t *start = pkt->data + (size_t)(((int64_t) pkt->size) * start_y / h);
1857  uint8_t *end = pkt->data + (size_t)(((int64_t) pkt->size) * end_y / h);
1858 
1859  init_put_bits(&s->thread_context[i]->pb, start, end - start);
1860  }
1861 
1862  s->pict_type = s->new_picture.f->pict_type;
1863  //emms_c();
1864  ret = frame_start(s);
1865  if (ret < 0)
1866  return ret;
1867 vbv_retry:
1868  ret = encode_picture(s, s->picture_number);
1869  if (growing_buffer) {
1870  av_assert0(s->pb.buf == avctx->internal->byte_buffer);
1871  pkt->data = s->pb.buf;
1872  pkt->size = avctx->internal->byte_buffer_size;
1873  }
1874  if (ret < 0)
1875  return -1;
1876 
1877 #if FF_API_STAT_BITS
1878 FF_DISABLE_DEPRECATION_WARNINGS
1879  avctx->header_bits = s->header_bits;
1880  avctx->mv_bits = s->mv_bits;
1881  avctx->misc_bits = s->misc_bits;
1882  avctx->i_tex_bits = s->i_tex_bits;
1883  avctx->p_tex_bits = s->p_tex_bits;
1884  avctx->i_count = s->i_count;
1885  // FIXME f/b_count in avctx
1886  avctx->p_count = s->mb_num - s->i_count - s->skip_count;
1887  avctx->skip_count = s->skip_count;
1888 FF_ENABLE_DEPRECATION_WARNINGS
1889 #endif
1890 
1891  frame_end(s);
1892 
1893  if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1894  ff_mjpeg_encode_picture_trailer(&s->pb, s->header_bits);
1895 
1896  if (avctx->rc_buffer_size) {
1897  RateControlContext *rcc = &s->rc_context;
1898  int max_size = FFMAX(rcc->buffer_index * avctx->rc_max_available_vbv_use, rcc->buffer_index - 500);
1899  int hq = (s->avctx->mb_decision == FF_MB_DECISION_RD || s->avctx->trellis);
1900  int min_step = hq ? 1 : (1<<(FF_LAMBDA_SHIFT + 7))/139;
1901 
1902  if (put_bits_count(&s->pb) > max_size &&
1903  s->lambda < s->lmax) {
1904  s->next_lambda = FFMAX(s->lambda + min_step, s->lambda *
1905  (s->qscale + 1) / s->qscale);
1906  if (s->adaptive_quant) {
1907  int i;
1908  for (i = 0; i < s->mb_height * s->mb_stride; i++)
1909  s->lambda_table[i] =
1910  FFMAX(s->lambda_table[i] + min_step,
1911  s->lambda_table[i] * (s->qscale + 1) /
1912  s->qscale);
1913  }
1914  s->mb_skipped = 0; // done in frame_start()
1915  // done in encode_picture() so we must undo it
1916  if (s->pict_type == AV_PICTURE_TYPE_P) {
1917  if (s->flipflop_rounding ||
1918  s->codec_id == AV_CODEC_ID_H263P ||
1919  s->codec_id == AV_CODEC_ID_MPEG4)
1920  s->no_rounding ^= 1;
1921  }
1922  if (s->pict_type != AV_PICTURE_TYPE_B) {
1923  s->time_base = s->last_time_base;
1924  s->last_non_b_time = s->time - s->pp_time;
1925  }
1926  for (i = 0; i < context_count; i++) {
1927  PutBitContext *pb = &s->thread_context[i]->pb;
1928  init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1929  }
1930  s->vbv_ignore_qmax = 1;
1931  av_log(s->avctx, AV_LOG_VERBOSE, "reencoding frame due to VBV\n");
1932  goto vbv_retry;
1933  }
1934 
1935  av_assert0(s->avctx->rc_max_rate);
1936  }
1937 
1938  if (s->avctx->flags & AV_CODEC_FLAG_PASS1)
1939  ff_write_pass1_stats(s);
1940 
1941  for (i = 0; i < 4; i++) {
1942  s->current_picture_ptr->encoding_error[i] = s->current_picture.encoding_error[i];
1943  avctx->error[i] += s->current_picture_ptr->encoding_error[i];
1944  }
1945  ff_side_data_set_encoder_stats(pkt, s->current_picture.f->quality,
1946  s->current_picture_ptr->encoding_error,
1947  (s->avctx->flags&AV_CODEC_FLAG_PSNR) ? 4 : 0,
1948  s->pict_type);
1949 
1950  if (s->avctx->flags & AV_CODEC_FLAG_PASS1)
1951  assert(put_bits_count(&s->pb) == s->header_bits + s->mv_bits +
1952  s->misc_bits + s->i_tex_bits +
1953  s->p_tex_bits);
1954  flush_put_bits(&s->pb);
1955  s->frame_bits = put_bits_count(&s->pb);
1956 
1957  stuffing_count = ff_vbv_update(s, s->frame_bits);
1958  s->stuffing_bits = 8*stuffing_count;
1959  if (stuffing_count) {
1960  if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
1961  stuffing_count + 50) {
1962  av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1963  return -1;
1964  }
1965 
1966  switch (s->codec_id) {
1967  case AV_CODEC_ID_MPEG1VIDEO:
1968  case AV_CODEC_ID_MPEG2VIDEO:
1969  while (stuffing_count--) {
1970  put_bits(&s->pb, 8, 0);
1971  }
1972  break;
1973  case AV_CODEC_ID_MPEG4:
1974  put_bits(&s->pb, 16, 0);
1975  put_bits(&s->pb, 16, 0x1C3);
1976  stuffing_count -= 4;
1977  while (stuffing_count--) {
1978  put_bits(&s->pb, 8, 0xFF);
1979  }
1980  break;
1981  default:
1982  av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1983  }
1984  flush_put_bits(&s->pb);
1985  s->frame_bits = put_bits_count(&s->pb);
1986  }
1987 
1988  /* update MPEG-1/2 vbv_delay for CBR */
1989  if (s->avctx->rc_max_rate &&
1990  s->avctx->rc_min_rate == s->avctx->rc_max_rate &&
1991  s->out_format == FMT_MPEG1 &&
1992  90000LL * (avctx->rc_buffer_size - 1) <=
1993  s->avctx->rc_max_rate * 0xFFFFLL) {
1994  AVCPBProperties *props;
1995  size_t props_size;
1996 
1997  int vbv_delay, min_delay;
1998  double inbits = s->avctx->rc_max_rate *
1999  av_q2d(s->avctx->time_base);
2000  int minbits = s->frame_bits - 8 *
2001  (s->vbv_delay_ptr - s->pb.buf - 1);
2002  double bits = s->rc_context.buffer_index + minbits - inbits;
2003 
2004  if (bits < 0)
2005  av_log(s->avctx, AV_LOG_ERROR,
2006  "Internal error, negative bits\n");
2007 
2008  av_assert1(s->repeat_first_field == 0);
2009 
2010  vbv_delay = bits * 90000 / s->avctx->rc_max_rate;
2011  min_delay = (minbits * 90000LL + s->avctx->rc_max_rate - 1) /
2012  s->avctx->rc_max_rate;
2013 
2014  vbv_delay = FFMAX(vbv_delay, min_delay);
2015 
2016  av_assert0(vbv_delay < 0xFFFF);
2017 
2018  s->vbv_delay_ptr[0] &= 0xF8;
2019  s->vbv_delay_ptr[0] |= vbv_delay >> 13;
2020  s->vbv_delay_ptr[1] = vbv_delay >> 5;
2021  s->vbv_delay_ptr[2] &= 0x07;
2022  s->vbv_delay_ptr[2] |= vbv_delay << 3;
2023 
2024  props = av_cpb_properties_alloc(&props_size);
2025  if (!props)
2026  return AVERROR(ENOMEM);
2027  props->vbv_delay = vbv_delay * 300;
2028 
2029  ret = av_packet_add_side_data(pkt, AV_PKT_DATA_CPB_PROPERTIES,
2030  (uint8_t*)props, props_size);
2031  if (ret < 0) {
2032  av_freep(&props);
2033  return ret;
2034  }
2035 
2036 #if FF_API_VBV_DELAY
2037 FF_DISABLE_DEPRECATION_WARNINGS
2038  avctx->vbv_delay = vbv_delay * 300;
2039 FF_ENABLE_DEPRECATION_WARNINGS
2040 #endif
2041  }
2042  s->total_bits += s->frame_bits;
2043 #if FF_API_STAT_BITS
2044 FF_DISABLE_DEPRECATION_WARNINGS
2045  avctx->frame_bits = s->frame_bits;
2046 FF_ENABLE_DEPRECATION_WARNINGS
2047 #endif
2048 
2049 
2050  pkt->pts = s->current_picture.f->pts;
2051  if (!s->low_delay && s->pict_type != AV_PICTURE_TYPE_B) {
2052  if (!s->current_picture.f->coded_picture_number)
2053  pkt->dts = pkt->pts - s->dts_delta;
2054  else
2055  pkt->dts = s->reordered_pts;
2056  s->reordered_pts = pkt->pts;
2057  } else
2058  pkt->dts = pkt->pts;
2059  if (s->current_picture.f->key_frame)
2060  pkt->flags |= AV_PKT_FLAG_KEY;
2061  if (s->mb_info)
2062  av_packet_shrink_side_data(pkt, AV_PKT_DATA_H263_MB_INFO, s->mb_info_size);
2063  } else {
2064  s->frame_bits = 0;
2065  }
2066 
2067  /* release non-reference frames */
2068  for (i = 0; i < MAX_PICTURE_COUNT; i++) {
2069  if (!s->picture[i].reference)
2070  ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2071  }
2072 
2073  av_assert1((s->frame_bits & 7) == 0);
2074 
2075  pkt->size = s->frame_bits / 8;
2076  *got_packet = !!pkt->size;
2077  return 0;
2078 }
2079 
2080 static inline void dct_single_coeff_elimination(MpegEncContext *s,
2081  int n, int threshold)
2082 {
2083  static const char tab[64] = {
2084  3, 2, 2, 1, 1, 1, 1, 1,
2085  1, 1, 1, 1, 1, 1, 1, 1,
2086  1, 1, 1, 1, 1, 1, 1, 1,
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  0, 0, 0, 0, 0, 0, 0, 0
2092  };
2093  int score = 0;
2094  int run = 0;
2095  int i;
2096  int16_t *block = s->block[n];
2097  const int last_index = s->block_last_index[n];
2098  int skip_dc;
2099 
2100  if (threshold < 0) {
2101  skip_dc = 0;
2102  threshold = -threshold;
2103  } else
2104  skip_dc = 1;
2105 
2106  /* Are all we could set to zero already zero? */
2107  if (last_index <= skip_dc - 1)
2108  return;
2109 
2110  for (i = 0; i <= last_index; i++) {
2111  const int j = s->intra_scantable.permutated[i];
2112  const int level = FFABS(block[j]);
2113  if (level == 1) {
2114  if (skip_dc && i == 0)
2115  continue;
2116  score += tab[run];
2117  run = 0;
2118  } else if (level > 1) {
2119  return;
2120  } else {
2121  run++;
2122  }
2123  }
2124  if (score >= threshold)
2125  return;
2126  for (i = skip_dc; i <= last_index; i++) {
2127  const int j = s->intra_scantable.permutated[i];
2128  block[j] = 0;
2129  }
2130  if (block[0])
2131  s->block_last_index[n] = 0;
2132  else
2133  s->block_last_index[n] = -1;
2134 }
2135 
2136 static inline void clip_coeffs(MpegEncContext *s, int16_t *block,
2137  int last_index)
2138 {
2139  int i;
2140  const int maxlevel = s->max_qcoeff;
2141  const int minlevel = s->min_qcoeff;
2142  int overflow = 0;
2143 
2144  if (s->mb_intra) {
2145  i = 1; // skip clipping of intra dc
2146  } else
2147  i = 0;
2148 
2149  for (; i <= last_index; i++) {
2150  const int j = s->intra_scantable.permutated[i];
2151  int level = block[j];
2152 
2153  if (level > maxlevel) {
2154  level = maxlevel;
2155  overflow++;
2156  } else if (level < minlevel) {
2157  level = minlevel;
2158  overflow++;
2159  }
2160 
2161  block[j] = level;
2162  }
2163 
2164  if (overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
2165  av_log(s->avctx, AV_LOG_INFO,
2166  "warning, clipping %d dct coefficients to %d..%d\n",
2167  overflow, minlevel, maxlevel);
2168 }
2169 
2170 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride)
2171 {
2172  int x, y;
2173  // FIXME optimize
2174  for (y = 0; y < 8; y++) {
2175  for (x = 0; x < 8; x++) {
2176  int x2, y2;
2177  int sum = 0;
2178  int sqr = 0;
2179  int count = 0;
2180 
2181  for (y2 = FFMAX(y - 1, 0); y2 < FFMIN(8, y + 2); y2++) {
2182  for (x2= FFMAX(x - 1, 0); x2 < FFMIN(8, x + 2); x2++) {
2183  int v = ptr[x2 + y2 * stride];
2184  sum += v;
2185  sqr += v * v;
2186  count++;
2187  }
2188  }
2189  weight[x + 8 * y]= (36 * ff_sqrt(count * sqr - sum * sum)) / count;
2190  }
2191  }
2192 }
2193 
2194 static av_always_inline void encode_mb_internal(MpegEncContext *s,
2195  int motion_x, int motion_y,
2196  int mb_block_height,
2197  int mb_block_width,
2198  int mb_block_count)
2199 {
2200  int16_t weight[12][64];
2201  int16_t orig[12][64];
2202  const int mb_x = s->mb_x;
2203  const int mb_y = s->mb_y;
2204  int i;
2205  int skip_dct[12];
2206  int dct_offset = s->linesize * 8; // default for progressive frames
2207  int uv_dct_offset = s->uvlinesize * 8;
2208  uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2209  ptrdiff_t wrap_y, wrap_c;
2210 
2211  for (i = 0; i < mb_block_count; i++)
2212  skip_dct[i] = s->skipdct;
2213 
2214  if (s->adaptive_quant) {
2215  const int last_qp = s->qscale;
2216  const int mb_xy = mb_x + mb_y * s->mb_stride;
2217 
2218  s->lambda = s->lambda_table[mb_xy];
2219  update_qscale(s);
2220 
2221  if (!(s->mpv_flags & FF_MPV_FLAG_QP_RD)) {
2222  s->qscale = s->current_picture_ptr->qscale_table[mb_xy];
2223  s->dquant = s->qscale - last_qp;
2224 
2225  if (s->out_format == FMT_H263) {
2226  s->dquant = av_clip(s->dquant, -2, 2);
2227 
2228  if (s->codec_id == AV_CODEC_ID_MPEG4) {
2229  if (!s->mb_intra) {
2230  if (s->pict_type == AV_PICTURE_TYPE_B) {
2231  if (s->dquant & 1 || s->mv_dir & MV_DIRECT)
2232  s->dquant = 0;
2233  }
2234  if (s->mv_type == MV_TYPE_8X8)
2235  s->dquant = 0;
2236  }
2237  }
2238  }
2239  }
2240  ff_set_qscale(s, last_qp + s->dquant);
2241  } else if (s->mpv_flags & FF_MPV_FLAG_QP_RD)
2242  ff_set_qscale(s, s->qscale + s->dquant);
2243 
2244  wrap_y = s->linesize;
2245  wrap_c = s->uvlinesize;
2246  ptr_y = s->new_picture.f->data[0] +
2247  (mb_y * 16 * wrap_y) + mb_x * 16;
2248  ptr_cb = s->new_picture.f->data[1] +
2249  (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
2250  ptr_cr = s->new_picture.f->data[2] +
2251  (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
2252 
2253  if((mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) && s->codec_id != AV_CODEC_ID_AMV){
2254  uint8_t *ebuf = s->sc.edge_emu_buffer + 38 * wrap_y;
2255  int cw = (s->width + s->chroma_x_shift) >> s->chroma_x_shift;
2256  int ch = (s->height + s->chroma_y_shift) >> s->chroma_y_shift;
2257  s->vdsp.emulated_edge_mc(ebuf, ptr_y,
2258  wrap_y, wrap_y,
2259  16, 16, mb_x * 16, mb_y * 16,
2260  s->width, s->height);
2261  ptr_y = ebuf;
2262  s->vdsp.emulated_edge_mc(ebuf + 16 * wrap_y, ptr_cb,
2263  wrap_c, wrap_c,
2264  mb_block_width, mb_block_height,
2265  mb_x * mb_block_width, mb_y * mb_block_height,
2266  cw, ch);
2267  ptr_cb = ebuf + 16 * wrap_y;
2268  s->vdsp.emulated_edge_mc(ebuf + 16 * wrap_y + 16, ptr_cr,
2269  wrap_c, wrap_c,
2270  mb_block_width, mb_block_height,
2271  mb_x * mb_block_width, mb_y * mb_block_height,
2272  cw, ch);
2273  ptr_cr = ebuf + 16 * wrap_y + 16;
2274  }
2275 
2276  if (s->mb_intra) {
2277  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
2278  int progressive_score, interlaced_score;
2279 
2280  s->interlaced_dct = 0;
2281  progressive_score = s->mecc.ildct_cmp[4](s, ptr_y, NULL, wrap_y, 8) +
2282  s->mecc.ildct_cmp[4](s, ptr_y + wrap_y * 8,
2283  NULL, wrap_y, 8) - 400;
2284 
2285  if (progressive_score > 0) {
2286  interlaced_score = s->mecc.ildct_cmp[4](s, ptr_y,
2287  NULL, wrap_y * 2, 8) +
2288  s->mecc.ildct_cmp[4](s, ptr_y + wrap_y,
2289  NULL, wrap_y * 2, 8);
2290  if (progressive_score > interlaced_score) {
2291  s->interlaced_dct = 1;
2292 
2293  dct_offset = wrap_y;
2294  uv_dct_offset = wrap_c;
2295  wrap_y <<= 1;
2296  if (s->chroma_format == CHROMA_422 ||
2297  s->chroma_format == CHROMA_444)
2298  wrap_c <<= 1;
2299  }
2300  }
2301  }
2302 
2303  s->pdsp.get_pixels(s->block[0], ptr_y, wrap_y);
2304  s->pdsp.get_pixels(s->block[1], ptr_y + 8, wrap_y);
2305  s->pdsp.get_pixels(s->block[2], ptr_y + dct_offset, wrap_y);
2306  s->pdsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
2307 
2308  if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
2309  skip_dct[4] = 1;
2310  skip_dct[5] = 1;
2311  } else {
2312  s->pdsp.get_pixels(s->block[4], ptr_cb, wrap_c);
2313  s->pdsp.get_pixels(s->block[5], ptr_cr, wrap_c);
2314  if (!s->chroma_y_shift && s->chroma_x_shift) { /* 422 */
2315  s->pdsp.get_pixels(s->block[6], ptr_cb + uv_dct_offset, wrap_c);
2316  s->pdsp.get_pixels(s->block[7], ptr_cr + uv_dct_offset, wrap_c);
2317  } else if (!s->chroma_y_shift && !s->chroma_x_shift) { /* 444 */
2318  s->pdsp.get_pixels(s->block[ 6], ptr_cb + 8, wrap_c);
2319  s->pdsp.get_pixels(s->block[ 7], ptr_cr + 8, wrap_c);
2320  s->pdsp.get_pixels(s->block[ 8], ptr_cb + uv_dct_offset, wrap_c);
2321  s->pdsp.get_pixels(s->block[ 9], ptr_cr + uv_dct_offset, wrap_c);
2322  s->pdsp.get_pixels(s->block[10], ptr_cb + uv_dct_offset + 8, wrap_c);
2323  s->pdsp.get_pixels(s->block[11], ptr_cr + uv_dct_offset + 8, wrap_c);
2324  }
2325  }
2326  } else {
2327  op_pixels_func (*op_pix)[4];
2328  qpel_mc_func (*op_qpix)[16];
2329  uint8_t *dest_y, *dest_cb, *dest_cr;
2330 
2331  dest_y = s->dest[0];
2332  dest_cb = s->dest[1];
2333  dest_cr = s->dest[2];
2334 
2335  if ((!s->no_rounding) || s->pict_type == AV_PICTURE_TYPE_B) {
2336  op_pix = s->hdsp.put_pixels_tab;
2337  op_qpix = s->qdsp.put_qpel_pixels_tab;
2338  } else {
2339  op_pix = s->hdsp.put_no_rnd_pixels_tab;
2340  op_qpix = s->qdsp.put_no_rnd_qpel_pixels_tab;
2341  }
2342 
2343  if (s->mv_dir & MV_DIR_FORWARD) {
2344  ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0,
2345  s->last_picture.f->data,
2346  op_pix, op_qpix);
2347  op_pix = s->hdsp.avg_pixels_tab;
2348  op_qpix = s->qdsp.avg_qpel_pixels_tab;
2349  }
2350  if (s->mv_dir & MV_DIR_BACKWARD) {
2351  ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1,
2352  s->next_picture.f->data,
2353  op_pix, op_qpix);
2354  }
2355 
2356  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
2357  int progressive_score, interlaced_score;
2358 
2359  s->interlaced_dct = 0;
2360  progressive_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y, wrap_y, 8) +
2361  s->mecc.ildct_cmp[0](s, dest_y + wrap_y * 8,
2362  ptr_y + wrap_y * 8,
2363  wrap_y, 8) - 400;
2364 
2365  if (s->avctx->ildct_cmp == FF_CMP_VSSE)
2366  progressive_score -= 400;
2367 
2368  if (progressive_score > 0) {
2369  interlaced_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y,
2370  wrap_y * 2, 8) +
2371  s->mecc.ildct_cmp[0](s, dest_y + wrap_y,
2372  ptr_y + wrap_y,
2373  wrap_y * 2, 8);
2374 
2375  if (progressive_score > interlaced_score) {
2376  s->interlaced_dct = 1;
2377 
2378  dct_offset = wrap_y;
2379  uv_dct_offset = wrap_c;
2380  wrap_y <<= 1;
2381  if (s->chroma_format == CHROMA_422)
2382  wrap_c <<= 1;
2383  }
2384  }
2385  }
2386 
2387  s->pdsp.diff_pixels(s->block[0], ptr_y, dest_y, wrap_y);
2388  s->pdsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
2389  s->pdsp.diff_pixels(s->block[2], ptr_y + dct_offset,
2390  dest_y + dct_offset, wrap_y);
2391  s->pdsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8,
2392  dest_y + dct_offset + 8, wrap_y);
2393 
2394  if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
2395  skip_dct[4] = 1;
2396  skip_dct[5] = 1;
2397  } else {
2398  s->pdsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
2399  s->pdsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
2400  if (!s->chroma_y_shift) { /* 422 */
2401  s->pdsp.diff_pixels(s->block[6], ptr_cb + uv_dct_offset,
2402  dest_cb + uv_dct_offset, wrap_c);
2403  s->pdsp.diff_pixels(s->block[7], ptr_cr + uv_dct_offset,
2404  dest_cr + uv_dct_offset, wrap_c);
2405  }
2406  }
2407  /* pre quantization */
2408  if (s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] <
2409  2 * s->qscale * s->qscale) {
2410  // FIXME optimize
2411  if (s->mecc.sad[1](NULL, ptr_y, dest_y, wrap_y, 8) < 20 * s->qscale)
2412  skip_dct[0] = 1;
2413  if (s->mecc.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20 * s->qscale)
2414  skip_dct[1] = 1;
2415  if (s->mecc.sad[1](NULL, ptr_y + dct_offset, dest_y + dct_offset,
2416  wrap_y, 8) < 20 * s->qscale)
2417  skip_dct[2] = 1;
2418  if (s->mecc.sad[1](NULL, ptr_y + dct_offset + 8, dest_y + dct_offset + 8,
2419  wrap_y, 8) < 20 * s->qscale)
2420  skip_dct[3] = 1;
2421  if (s->mecc.sad[1](NULL, ptr_cb, dest_cb, wrap_c, 8) < 20 * s->qscale)
2422  skip_dct[4] = 1;
2423  if (s->mecc.sad[1](NULL, ptr_cr, dest_cr, wrap_c, 8) < 20 * s->qscale)
2424  skip_dct[5] = 1;
2425  if (!s->chroma_y_shift) { /* 422 */
2426  if (s->mecc.sad[1](NULL, ptr_cb + uv_dct_offset,
2427  dest_cb + uv_dct_offset,
2428  wrap_c, 8) < 20 * s->qscale)
2429  skip_dct[6] = 1;
2430  if (s->mecc.sad[1](NULL, ptr_cr + uv_dct_offset,
2431  dest_cr + uv_dct_offset,
2432  wrap_c, 8) < 20 * s->qscale)
2433  skip_dct[7] = 1;
2434  }
2435  }
2436  }
2437 
2438  if (s->quantizer_noise_shaping) {
2439  if (!skip_dct[0])
2440  get_visual_weight(weight[0], ptr_y , wrap_y);
2441  if (!skip_dct[1])
2442  get_visual_weight(weight[1], ptr_y + 8, wrap_y);
2443  if (!skip_dct[2])
2444  get_visual_weight(weight[2], ptr_y + dct_offset , wrap_y);
2445  if (!skip_dct[3])
2446  get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
2447  if (!skip_dct[4])
2448  get_visual_weight(weight[4], ptr_cb , wrap_c);
2449  if (!skip_dct[5])
2450  get_visual_weight(weight[5], ptr_cr , wrap_c);
2451  if (!s->chroma_y_shift) { /* 422 */
2452  if (!skip_dct[6])
2453  get_visual_weight(weight[6], ptr_cb + uv_dct_offset,
2454  wrap_c);
2455  if (!skip_dct[7])
2456  get_visual_weight(weight[7], ptr_cr + uv_dct_offset,
2457  wrap_c);
2458  }
2459  memcpy(orig[0], s->block[0], sizeof(int16_t) * 64 * mb_block_count);
2460  }
2461 
2462  /* DCT & quantize */
2463  av_assert2(s->out_format != FMT_MJPEG || s->qscale == 8);
2464  {
2465  for (i = 0; i < mb_block_count; i++) {
2466  if (!skip_dct[i]) {
2467  int overflow;
2468  s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
2469  // FIXME we could decide to change to quantizer instead of
2470  // clipping
2471  // JS: I don't think that would be a good idea it could lower
2472  // quality instead of improve it. Just INTRADC clipping
2473  // deserves changes in quantizer
2474  if (overflow)
2475  clip_coeffs(s, s->block[i], s->block_last_index[i]);
2476  } else
2477  s->block_last_index[i] = -1;
2478  }
2479  if (s->quantizer_noise_shaping) {
2480  for (i = 0; i < mb_block_count; i++) {
2481  if (!skip_dct[i]) {
2482  s->block_last_index[i] =
2483  dct_quantize_refine(s, s->block[i], weight[i],
2484  orig[i], i, s->qscale);
2485  }
2486  }
2487  }
2488 
2489  if (s->luma_elim_threshold && !s->mb_intra)
2490  for (i = 0; i < 4; i++)
2491  dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
2492  if (s->chroma_elim_threshold && !s->mb_intra)
2493  for (i = 4; i < mb_block_count; i++)
2494  dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
2495 
2496  if (s->mpv_flags & FF_MPV_FLAG_CBP_RD) {
2497  for (i = 0; i < mb_block_count; i++) {
2498  if (s->block_last_index[i] == -1)
2499  s->coded_score[i] = INT_MAX / 256;
2500  }
2501  }
2502  }
2503 
2504  if ((s->avctx->flags & AV_CODEC_FLAG_GRAY) && s->mb_intra) {
2505  s->block_last_index[4] =
2506  s->block_last_index[5] = 0;
2507  s->block[4][0] =
2508  s->block[5][0] = (1024 + s->c_dc_scale / 2) / s->c_dc_scale;
2509  if (!s->chroma_y_shift) { /* 422 / 444 */
2510  for (i=6; i<12; i++) {
2511  s->block_last_index[i] = 0;
2512  s->block[i][0] = s->block[4][0];
2513  }
2514  }
2515  }
2516 
2517  // non c quantize code returns incorrect block_last_index FIXME
2518  if (s->alternate_scan && s->dct_quantize != ff_dct_quantize_c) {
2519  for (i = 0; i < mb_block_count; i++) {
2520  int j;
2521  if (s->block_last_index[i] > 0) {
2522  for (j = 63; j > 0; j--) {
2523  if (s->block[i][s->intra_scantable.permutated[j]])
2524  break;
2525  }
2526  s->block_last_index[i] = j;
2527  }
2528  }
2529  }
2530 
2531  /* huffman encode */
2532  switch(s->codec_id){ //FIXME funct ptr could be slightly faster
2533  case AV_CODEC_ID_MPEG1VIDEO:
2534  case AV_CODEC_ID_MPEG2VIDEO:
2535  if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2536  ff_mpeg1_encode_mb(s, s->block, motion_x, motion_y);
2537  break;
2538  case AV_CODEC_ID_MPEG4:
2539  if (CONFIG_MPEG4_ENCODER)
2540  ff_mpeg4_encode_mb(s, s->block, motion_x, motion_y);
2541  break;
2542  case AV_CODEC_ID_MSMPEG4V2:
2543  case AV_CODEC_ID_MSMPEG4V3:
2544  case AV_CODEC_ID_WMV1:
2545  if (CONFIG_MSMPEG4_ENCODER)
2546  ff_msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
2547  break;
2548  case AV_CODEC_ID_WMV2:
2549  if (CONFIG_WMV2_ENCODER)
2550  ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
2551  break;
2552  case AV_CODEC_ID_H261:
2553  if (CONFIG_H261_ENCODER)
2554  ff_h261_encode_mb(s, s->block, motion_x, motion_y);
2555  break;
2556  case AV_CODEC_ID_H263:
2557  case AV_CODEC_ID_H263P:
2558  case AV_CODEC_ID_FLV1:
2559  case AV_CODEC_ID_RV10:
2560  case AV_CODEC_ID_RV20:
2561  if (CONFIG_H263_ENCODER)
2562  ff_h263_encode_mb(s, s->block, motion_x, motion_y);
2563  break;
2564  case AV_CODEC_ID_MJPEG:
2565  case AV_CODEC_ID_AMV:
2566  if (CONFIG_MJPEG_ENCODER)
2567  ff_mjpeg_encode_mb(s, s->block);
2568  break;
2569  default:
2570  av_assert1(0);
2571  }
2572 }
2573 
2574 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
2575 {
2576  if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y, 8, 8, 6);
2577  else if (s->chroma_format == CHROMA_422) encode_mb_internal(s, motion_x, motion_y, 16, 8, 8);
2578  else encode_mb_internal(s, motion_x, motion_y, 16, 16, 12);
2579 }
2580 
2581 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
2582  int i;
2583 
2584  memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2585 
2586  /* MPEG-1 */
2587  d->mb_skip_run= s->mb_skip_run;
2588  for(i=0; i<3; i++)
2589  d->last_dc[i] = s->last_dc[i];
2590 
2591  /* statistics */
2592  d->mv_bits= s->mv_bits;
2593  d->i_tex_bits= s->i_tex_bits;
2594  d->p_tex_bits= s->p_tex_bits;
2595  d->i_count= s->i_count;
2596  d->f_count= s->f_count;
2597  d->b_count= s->b_count;
2598  d->skip_count= s->skip_count;
2599  d->misc_bits= s->misc_bits;
2600  d->last_bits= 0;
2601 
2602  d->mb_skipped= 0;
2603  d->qscale= s->qscale;
2604  d->dquant= s->dquant;
2605 
2606  d->esc3_level_length= s->esc3_level_length;
2607 }
2608 
2609 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
2610  int i;
2611 
2612  memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
2613  memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
2614 
2615  /* MPEG-1 */
2616  d->mb_skip_run= s->mb_skip_run;
2617  for(i=0; i<3; i++)
2618  d->last_dc[i] = s->last_dc[i];
2619 
2620  /* statistics */
2621  d->mv_bits= s->mv_bits;
2622  d->i_tex_bits= s->i_tex_bits;
2623  d->p_tex_bits= s->p_tex_bits;
2624  d->i_count= s->i_count;
2625  d->f_count= s->f_count;
2626  d->b_count= s->b_count;
2627  d->skip_count= s->skip_count;
2628  d->misc_bits= s->misc_bits;
2629 
2630  d->mb_intra= s->mb_intra;
2631  d->mb_skipped= s->mb_skipped;
2632  d->mv_type= s->mv_type;
2633  d->mv_dir= s->mv_dir;
2634  d->pb= s->pb;
2635  if(s->data_partitioning){
2636  d->pb2= s->pb2;
2637  d->tex_pb= s->tex_pb;
2638  }
2639  d->block= s->block;
2640  for(i=0; i<8; i++)
2641  d->block_last_index[i]= s->block_last_index[i];
2642  d->interlaced_dct= s->interlaced_dct;
2643  d->qscale= s->qscale;
2644 
2645  d->esc3_level_length= s->esc3_level_length;
2646 }
2647 
2648 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
2649  PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
2650  int *dmin, int *next_block, int motion_x, int motion_y)
2651 {
2652  int score;
2653  uint8_t *dest_backup[3];
2654 
2655  copy_context_before_encode(s, backup, type);
2656 
2657  s->block= s->blocks[*next_block];
2658  s->pb= pb[*next_block];
2659  if(s->data_partitioning){
2660  s->pb2 = pb2 [*next_block];
2661  s->tex_pb= tex_pb[*next_block];
2662  }
2663 
2664  if(*next_block){
2665  memcpy(dest_backup, s->dest, sizeof(s->dest));
2666  s->dest[0] = s->sc.rd_scratchpad;
2667  s->dest[1] = s->sc.rd_scratchpad + 16*s->linesize;
2668  s->dest[2] = s->sc.rd_scratchpad + 16*s->linesize + 8;
2669  av_assert0(s->linesize >= 32); //FIXME
2670  }
2671 
2672  encode_mb(s, motion_x, motion_y);
2673 
2674  score= put_bits_count(&s->pb);
2675  if(s->data_partitioning){
2676  score+= put_bits_count(&s->pb2);
2677  score+= put_bits_count(&s->tex_pb);
2678  }
2679 
2680  if(s->avctx->mb_decision == FF_MB_DECISION_RD){
2681  ff_mpv_reconstruct_mb(s, s->block);
2682 
2683  score *= s->lambda2;
2684  score += sse_mb(s) << FF_LAMBDA_SHIFT;
2685  }
2686 
2687  if(*next_block){
2688  memcpy(s->dest, dest_backup, sizeof(s->dest));
2689  }
2690 
2691  if(score<*dmin){
2692  *dmin= score;
2693  *next_block^=1;
2694 
2695  copy_context_after_encode(best, s, type);
2696  }
2697 }
2698 
2699 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
2700  const uint32_t *sq = ff_square_tab + 256;
2701  int acc=0;
2702  int x,y;
2703 
2704  if(w==16 && h==16)
2705  return s->mecc.sse[0](NULL, src1, src2, stride, 16);
2706  else if(w==8 && h==8)
2707  return s->mecc.sse[1](NULL, src1, src2, stride, 8);
2708 
2709  for(y=0; y<h; y++){
2710  for(x=0; x<w; x++){
2711  acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
2712  }
2713  }
2714 
2715  av_assert2(acc>=0);
2716 
2717  return acc;
2718 }
2719 
2720 static int sse_mb(MpegEncContext *s){
2721  int w= 16;
2722  int h= 16;
2723 
2724  if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2725  if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2726 
2727  if(w==16 && h==16)
2728  if(s->avctx->mb_cmp == FF_CMP_NSSE){
2729  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) +
2730  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) +
2731  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);
2732  }else{
2733  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) +
2734  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) +
2735  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);
2736  }
2737  else
2738  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)
2739  +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)
2740  +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);
2741 }
2742 
2743 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
2744  MpegEncContext *s= *(void**)arg;
2745 
2746 
2747  s->me.pre_pass=1;
2748  s->me.dia_size= s->avctx->pre_dia_size;
2749  s->first_slice_line=1;
2750  for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
2751  for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
2752  ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2753  }
2754  s->first_slice_line=0;
2755  }
2756 
2757  s->me.pre_pass=0;
2758 
2759  return 0;
2760 }
2761 
2762 static int estimate_motion_thread(AVCodecContext *c, void *arg){
2763  MpegEncContext *s= *(void**)arg;
2764 
2765  ff_check_alignment();
2766 
2767  s->me.dia_size= s->avctx->dia_size;
2768  s->first_slice_line=1;
2769  for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
2770  s->mb_x=0; //for block init below
2771  ff_init_block_index(s);
2772  for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
2773  s->block_index[0]+=2;
2774  s->block_index[1]+=2;
2775  s->block_index[2]+=2;
2776  s->block_index[3]+=2;
2777 
2778  /* compute motion vector & mb_type and store in context */
2779  if(s->pict_type==AV_PICTURE_TYPE_B)
2780  ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
2781  else
2782  ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2783  }
2784  s->first_slice_line=0;
2785  }
2786  return 0;
2787 }
2788 
2789 static int mb_var_thread(AVCodecContext *c, void *arg){
2790  MpegEncContext *s= *(void**)arg;
2791  int mb_x, mb_y;
2792 
2793  ff_check_alignment();
2794 
2795  for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2796  for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2797  int xx = mb_x * 16;
2798  int yy = mb_y * 16;
2799  uint8_t *pix = s->new_picture.f->data[0] + (yy * s->linesize) + xx;
2800  int varc;
2801  int sum = s->mpvencdsp.pix_sum(pix, s->linesize);
2802 
2803  varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) -
2804  (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8;
2805 
2806  s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2807  s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2808  s->me.mb_var_sum_temp += varc;
2809  }
2810  }
2811  return 0;
2812 }
2813 
2814 static void write_slice_end(MpegEncContext *s){
2815  if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4){
2816  if(s->partitioned_frame){
2817  ff_mpeg4_merge_partitions(s);
2818  }
2819 
2820  ff_mpeg4_stuffing(&s->pb);
2821  }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2822  ff_mjpeg_encode_stuffing(s);
2823  }
2824 
2825  avpriv_align_put_bits(&s->pb);
2826  flush_put_bits(&s->pb);
2827 
2828  if ((s->avctx->flags & AV_CODEC_FLAG_PASS1) && !s->partitioned_frame)
2829  s->misc_bits+= get_bits_diff(s);
2830 }
2831 
2832 static void write_mb_info(MpegEncContext *s)
2833 {
2834  uint8_t *ptr = s->mb_info_ptr + s->mb_info_size - 12;
2835  int offset = put_bits_count(&s->pb);
2836  int mba = s->mb_x + s->mb_width * (s->mb_y % s->gob_index);
2837  int gobn = s->mb_y / s->gob_index;
2838  int pred_x, pred_y;
2839  if (CONFIG_H263_ENCODER)
2840  ff_h263_pred_motion(s, 0, 0, &pred_x, &pred_y);
2841  bytestream_put_le32(&ptr, offset);
2842  bytestream_put_byte(&ptr, s->qscale);
2843  bytestream_put_byte(&ptr, gobn);
2844  bytestream_put_le16(&ptr, mba);
2845  bytestream_put_byte(&ptr, pred_x); /* hmv1 */
2846  bytestream_put_byte(&ptr, pred_y); /* vmv1 */
2847  /* 4MV not implemented */
2848  bytestream_put_byte(&ptr, 0); /* hmv2 */
2849  bytestream_put_byte(&ptr, 0); /* vmv2 */
2850 }
2851 
2852 static void update_mb_info(MpegEncContext *s, int startcode)
2853 {
2854  if (!s->mb_info)
2855  return;
2856  if (put_bits_count(&s->pb) - s->prev_mb_info*8 >= s->mb_info*8) {
2857  s->mb_info_size += 12;
2858  s->prev_mb_info = s->last_mb_info;
2859  }
2860  if (startcode) {
2861  s->prev_mb_info = put_bits_count(&s->pb)/8;
2862  /* This might have incremented mb_info_size above, and we return without
2863  * actually writing any info into that slot yet. But in that case,
2864  * this will be called again at the start of the after writing the
2865  * start code, actually writing the mb info. */
2866  return;
2867  }
2868 
2869  s->last_mb_info = put_bits_count(&s->pb)/8;
2870  if (!s->mb_info_size)
2871  s->mb_info_size += 12;
2872  write_mb_info(s);
2873 }
2874 
2875 int ff_mpv_reallocate_putbitbuffer(MpegEncContext *s, size_t threshold, size_t size_increase)
2876 {
2877  if ( s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < threshold
2878  && s->slice_context_count == 1
2879  && s->pb.buf == s->avctx->internal->byte_buffer) {
2880  int lastgob_pos = s->ptr_lastgob - s->pb.buf;
2881  int vbv_pos = s->vbv_delay_ptr - s->pb.buf;
2882 
2883  uint8_t *new_buffer = NULL;
2884  int new_buffer_size = 0;
2885 
2886  if ((s->avctx->internal->byte_buffer_size + size_increase) >= INT_MAX/8) {
2887  av_log(s->avctx, AV_LOG_ERROR, "Cannot reallocate putbit buffer\n");
2888  return AVERROR(ENOMEM);
2889  }
2890 
2891  emms_c();
2892 
2893  av_fast_padded_malloc(&new_buffer, &new_buffer_size,
2894  s->avctx->internal->byte_buffer_size + size_increase);
2895  if (!new_buffer)
2896  return AVERROR(ENOMEM);
2897 
2898  memcpy(new_buffer, s->avctx->internal->byte_buffer, s->avctx->internal->byte_buffer_size);
2899  av_free(s->avctx->internal->byte_buffer);
2900  s->avctx->internal->byte_buffer = new_buffer;
2901  s->avctx->internal->byte_buffer_size = new_buffer_size;
2902  rebase_put_bits(&s->pb, new_buffer, new_buffer_size);
2903  s->ptr_lastgob = s->pb.buf + lastgob_pos;
2904  s->vbv_delay_ptr = s->pb.buf + vbv_pos;
2905  }
2906  if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < threshold)
2907  return AVERROR(EINVAL);
2908  return 0;
2909 }
2910 
2911 static int encode_thread(AVCodecContext *c, void *arg){
2912  MpegEncContext *s= *(void**)arg;
2913  int mb_x, mb_y;
2914  int chr_h= 16>>s->chroma_y_shift;
2915  int i, j;
2916  MpegEncContext best_s = { 0 }, backup_s;
2917  uint8_t bit_buf[2][MAX_MB_BYTES];
2918  uint8_t bit_buf2[2][MAX_MB_BYTES];
2919  uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2920  PutBitContext pb[2], pb2[2], tex_pb[2];
2921 
2922  ff_check_alignment();
2923 
2924  for(i=0; i<2; i++){
2925  init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
2926  init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
2927  init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2928  }
2929 
2930  s->last_bits= put_bits_count(&s->pb);
2931  s->mv_bits=0;
2932  s->misc_bits=0;
2933  s->i_tex_bits=0;
2934  s->p_tex_bits=0;
2935  s->i_count=0;
2936  s->f_count=0;
2937  s->b_count=0;
2938  s->skip_count=0;
2939 
2940  for(i=0; i<3; i++){
2941  /* init last dc values */
2942  /* note: quant matrix value (8) is implied here */
2943  s->last_dc[i] = 128 << s->intra_dc_precision;
2944 
2945  s->current_picture.encoding_error[i] = 0;
2946  }
2947  if(s->codec_id==AV_CODEC_ID_AMV){
2948  s->last_dc[0] = 128*8/13;
2949  s->last_dc[1] = 128*8/14;
2950  s->last_dc[2] = 128*8/14;
2951  }
2952  s->mb_skip_run = 0;
2953  memset(s->last_mv, 0, sizeof(s->last_mv));
2954 
2955  s->last_mv_dir = 0;
2956 
2957  switch(s->codec_id){
2958  case AV_CODEC_ID_H263:
2959  case AV_CODEC_ID_H263P:
2960  case AV_CODEC_ID_FLV1:
2961  if (CONFIG_H263_ENCODER)
2962  s->gob_index = H263_GOB_HEIGHT(s->height);
2963  break;
2964  case AV_CODEC_ID_MPEG4:
2965  if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2966  ff_mpeg4_init_partitions(s);
2967  break;
2968  }
2969 
2970  s->resync_mb_x=0;
2971  s->resync_mb_y=0;
2972  s->first_slice_line = 1;
2973  s->ptr_lastgob = s->pb.buf;
2974  for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2975  s->mb_x=0;
2976  s->mb_y= mb_y;
2977 
2978  ff_set_qscale(s, s->qscale);
2979  ff_init_block_index(s);
2980 
2981  for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2982  int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2983  int mb_type= s->mb_type[xy];
2984 // int d;
2985  int dmin= INT_MAX;
2986  int dir;
2987  int size_increase = s->avctx->internal->byte_buffer_size/4
2988  + s->mb_width*MAX_MB_BYTES;
2989 
2990  ff_mpv_reallocate_putbitbuffer(s, MAX_MB_BYTES, size_increase);
2991  if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2992  av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2993  return -1;
2994  }
2995  if(s->data_partitioning){
2996  if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES
2997  || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2998  av_log(s->avctx, AV_LOG_ERROR, "encoded partitioned frame too large\n");
2999  return -1;
3000  }
3001  }
3002 
3003  s->mb_x = mb_x;
3004  s->mb_y = mb_y; // moved into loop, can get changed by H.261
3005  ff_update_block_index(s);
3006 
3007  if(CONFIG_H261_ENCODER && s->codec_id == AV_CODEC_ID_H261){
3008  ff_h261_reorder_mb_index(s);
3009  xy= s->mb_y*s->mb_stride + s->mb_x;
3010  mb_type= s->mb_type[xy];
3011  }
3012 
3013  /* write gob / video packet header */
3014  if(s->rtp_mode){
3015  int current_packet_size, is_gob_start;
3016 
3017  current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
3018 
3019  is_gob_start = s->rtp_payload_size &&
3020  current_packet_size >= s->rtp_payload_size &&
3021  mb_y + mb_x > 0;
3022 
3023  if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
3024 
3025  switch(s->codec_id){
3026  case AV_CODEC_ID_H263:
3027  case AV_CODEC_ID_H263P:
3028  if(!s->h263_slice_structured)
3029  if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
3030  break;
3031  case AV_CODEC_ID_MPEG2VIDEO:
3032  if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
3033  case AV_CODEC_ID_MPEG1VIDEO:
3034  if(s->mb_skip_run) is_gob_start=0;
3035  break;
3036  case AV_CODEC_ID_MJPEG:
3037  if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
3038  break;
3039  }
3040 
3041  if(is_gob_start){
3042  if(s->start_mb_y != mb_y || mb_x!=0){
3043  write_slice_end(s);
3044 
3045  if(CONFIG_MPEG4_ENCODER && s->codec_id==AV_CODEC_ID_MPEG4 && s->partitioned_frame){
3046  ff_mpeg4_init_partitions(s);
3047  }
3048  }
3049 
3050  av_assert2((put_bits_count(&s->pb)&7) == 0);
3051  current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
3052 
3053  if (s->error_rate && s->resync_mb_x + s->resync_mb_y > 0) {
3054  int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
3055  int d = 100 / s->error_rate;
3056  if(r % d == 0){
3057  current_packet_size=0;
3058  s->pb.buf_ptr= s->ptr_lastgob;
3059  av_assert1(put_bits_ptr(&s->pb) == s->ptr_lastgob);
3060  }
3061  }
3062 
3063 #if FF_API_RTP_CALLBACK
3064 FF_DISABLE_DEPRECATION_WARNINGS
3065  if (s->avctx->rtp_callback){
3066  int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
3067  s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
3068  }
3069 FF_ENABLE_DEPRECATION_WARNINGS
3070 #endif
3071  update_mb_info(s, 1);
3072 
3073  switch(s->codec_id){
3074  case AV_CODEC_ID_MPEG4:
3075  if (CONFIG_MPEG4_ENCODER) {
3076  ff_mpeg4_encode_video_packet_header(s);
3077  ff_mpeg4_clean_buffers(s);
3078  }
3079  break;
3080  case AV_CODEC_ID_MPEG1VIDEO:
3081  case AV_CODEC_ID_MPEG2VIDEO:
3082  if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
3083  ff_mpeg1_encode_slice_header(s);
3084  ff_mpeg1_clean_buffers(s);
3085  }
3086  break;
3087  case AV_CODEC_ID_H263:
3088  case AV_CODEC_ID_H263P:
3089  if (CONFIG_H263_ENCODER)
3090  ff_h263_encode_gob_header(s, mb_y);
3091  break;
3092  }
3093 
3094  if (s->avctx->flags & AV_CODEC_FLAG_PASS1) {
3095  int bits= put_bits_count(&s->pb);
3096  s->misc_bits+= bits - s->last_bits;
3097  s->last_bits= bits;
3098  }
3099 
3100  s->ptr_lastgob += current_packet_size;
3101  s->first_slice_line=1;
3102  s->resync_mb_x=mb_x;
3103  s->resync_mb_y=mb_y;
3104  }
3105  }
3106 
3107  if( (s->resync_mb_x == s->mb_x)
3108  && s->resync_mb_y+1 == s->mb_y){
3109  s->first_slice_line=0;
3110  }
3111 
3112  s->mb_skipped=0;
3113  s->dquant=0; //only for QP_RD
3114 
3115  update_mb_info(s, 0);
3116 
3117  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
3118  int next_block=0;
3119  int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
3120 
3121  copy_context_before_encode(&backup_s, s, -1);
3122  backup_s.pb= s->pb;
3123  best_s.data_partitioning= s->data_partitioning;
3124  best_s.partitioned_frame= s->partitioned_frame;
3125  if(s->data_partitioning){
3126  backup_s.pb2= s->pb2;
3127  backup_s.tex_pb= s->tex_pb;
3128  }
3129 
3130  if(mb_type&CANDIDATE_MB_TYPE_INTER){
3131  s->mv_dir = MV_DIR_FORWARD;
3132  s->mv_type = MV_TYPE_16X16;
3133  s->mb_intra= 0;
3134  s->mv[0][0][0] = s->p_mv_table[xy][0];
3135  s->mv[0][0][1] = s->p_mv_table[xy][1];
3136  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
3137  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3138  }
3139  if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
3140  s->mv_dir = MV_DIR_FORWARD;
3141  s->mv_type = MV_TYPE_FIELD;
3142  s->mb_intra= 0;
3143  for(i=0; i<2; i++){
3144  j= s->field_select[0][i] = s->p_field_select_table[i][xy];
3145  s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
3146  s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
3147  }
3148  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
3149  &dmin, &next_block, 0, 0);
3150  }
3151  if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
3152  s->mv_dir = MV_DIR_FORWARD;
3153  s->mv_type = MV_TYPE_16X16;
3154  s->mb_intra= 0;
3155  s->mv[0][0][0] = 0;
3156  s->mv[0][0][1] = 0;
3157  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
3158  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3159  }
3160  if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
3161  s->mv_dir = MV_DIR_FORWARD;
3162  s->mv_type = MV_TYPE_8X8;
3163  s->mb_intra= 0;
3164  for(i=0; i<4; i++){
3165  s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
3166  s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
3167  }
3168  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
3169  &dmin, &next_block, 0, 0);
3170  }
3171  if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
3172  s->mv_dir = MV_DIR_FORWARD;
3173  s->mv_type = MV_TYPE_16X16;
3174  s->mb_intra= 0;
3175  s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
3176  s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
3177  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
3178  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
3179  }
3180  if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
3181  s->mv_dir = MV_DIR_BACKWARD;
3182  s->mv_type = MV_TYPE_16X16;
3183  s->mb_intra= 0;
3184  s->mv[1][0][0] = s->b_back_mv_table[xy][0];
3185  s->mv[1][0][1] = s->b_back_mv_table[xy][1];
3186  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
3187  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
3188  }
3189  if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
3190  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3191  s->mv_type = MV_TYPE_16X16;
3192  s->mb_intra= 0;
3193  s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
3194  s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
3195  s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
3196  s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
3197  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
3198  &dmin, &next_block, 0, 0);
3199  }
3200  if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
3201  s->mv_dir = MV_DIR_FORWARD;
3202  s->mv_type = MV_TYPE_FIELD;
3203  s->mb_intra= 0;
3204  for(i=0; i<2; i++){
3205  j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
3206  s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
3207  s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
3208  }
3209  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
3210  &dmin, &next_block, 0, 0);
3211  }
3212  if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
3213  s->mv_dir = MV_DIR_BACKWARD;
3214  s->mv_type = MV_TYPE_FIELD;
3215  s->mb_intra= 0;
3216  for(i=0; i<2; i++){
3217  j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
3218  s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
3219  s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
3220  }
3221  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
3222  &dmin, &next_block, 0, 0);
3223  }
3224  if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
3225  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3226  s->mv_type = MV_TYPE_FIELD;
3227  s->mb_intra= 0;
3228  for(dir=0; dir<2; dir++){
3229  for(i=0; i<2; i++){
3230  j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
3231  s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
3232  s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
3233  }
3234  }
3235  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
3236  &dmin, &next_block, 0, 0);
3237  }
3238  if(mb_type&CANDIDATE_MB_TYPE_INTRA){
3239  s->mv_dir = 0;
3240  s->mv_type = MV_TYPE_16X16;
3241  s->mb_intra= 1;
3242  s->mv[0][0][0] = 0;
3243  s->mv[0][0][1] = 0;
3244  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
3245  &dmin, &next_block, 0, 0);
3246  if(s->h263_pred || s->h263_aic){
3247  if(best_s.mb_intra)
3248  s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
3249  else
3250  ff_clean_intra_table_entries(s); //old mode?
3251  }
3252  }
3253 
3254  if ((s->mpv_flags & FF_MPV_FLAG_QP_RD) && dmin < INT_MAX) {
3255  if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
3256  const int last_qp= backup_s.qscale;
3257  int qpi, qp, dc[6];
3258  int16_t ac[6][16];
3259  const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
3260  static const int dquant_tab[4]={-1,1,-2,2};
3261  int storecoefs = s->mb_intra && s->dc_val[0];
3262 
3263  av_assert2(backup_s.dquant == 0);
3264 
3265  //FIXME intra
3266  s->mv_dir= best_s.mv_dir;
3267  s->mv_type = MV_TYPE_16X16;
3268  s->mb_intra= best_s.mb_intra;
3269  s->mv[0][0][0] = best_s.mv[0][0][0];
3270  s->mv[0][0][1] = best_s.mv[0][0][1];
3271  s->mv[1][0][0] = best_s.mv[1][0][0];
3272  s->mv[1][0][1] = best_s.mv[1][0][1];
3273 
3274  qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
3275  for(; qpi<4; qpi++){
3276  int dquant= dquant_tab[qpi];
3277  qp= last_qp + dquant;
3278  if(qp < s->avctx->qmin || qp > s->avctx->qmax)
3279  continue;
3280  backup_s.dquant= dquant;
3281  if(storecoefs){
3282  for(i=0; i<6; i++){
3283  dc[i]= s->dc_val[0][ s->block_index[i] ];
3284  memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(int16_t)*16);
3285  }
3286  }
3287 
3288  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
3289  &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
3290  if(best_s.qscale != qp){
3291  if(storecoefs){
3292  for(i=0; i<6; i++){
3293  s->dc_val[0][ s->block_index[i] ]= dc[i];
3294  memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(int16_t)*16);
3295  }
3296  }
3297  }
3298  }
3299  }
3300  }
3301  if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
3302  int mx= s->b_direct_mv_table[xy][0];
3303  int my= s->b_direct_mv_table[xy][1];
3304 
3305  backup_s.dquant = 0;
3306  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
3307  s->mb_intra= 0;
3308  ff_mpeg4_set_direct_mv(s, mx, my);
3309  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
3310  &dmin, &next_block, mx, my);
3311  }
3312  if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
3313  backup_s.dquant = 0;
3314  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
3315  s->mb_intra= 0;
3316  ff_mpeg4_set_direct_mv(s, 0, 0);
3317  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
3318  &dmin, &next_block, 0, 0);
3319  }
3320  if (!best_s.mb_intra && s->mpv_flags & FF_MPV_FLAG_SKIP_RD) {
3321  int coded=0;
3322  for(i=0; i<6; i++)
3323  coded |= s->block_last_index[i];
3324  if(coded){
3325  int mx,my;
3326  memcpy(s->mv, best_s.mv, sizeof(s->mv));
3327  if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
3328  mx=my=0; //FIXME find the one we actually used
3329  ff_mpeg4_set_direct_mv(s, mx, my);
3330  }else if(best_s.mv_dir&MV_DIR_BACKWARD){
3331  mx= s->mv[1][0][0];
3332  my= s->mv[1][0][1];
3333  }else{
3334  mx= s->mv[0][0][0];
3335  my= s->mv[0][0][1];
3336  }
3337 
3338  s->mv_dir= best_s.mv_dir;
3339  s->mv_type = best_s.mv_type;
3340  s->mb_intra= 0;
3341 /* s->mv[0][0][0] = best_s.mv[0][0][0];
3342  s->mv[0][0][1] = best_s.mv[0][0][1];
3343  s->mv[1][0][0] = best_s.mv[1][0][0];
3344  s->mv[1][0][1] = best_s.mv[1][0][1];*/
3345  backup_s.dquant= 0;
3346  s->skipdct=1;
3347  encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
3348  &dmin, &next_block, mx, my);
3349  s->skipdct=0;
3350  }
3351  }
3352 
3353  s->current_picture.qscale_table[xy] = best_s.qscale;
3354 
3355  copy_context_after_encode(s, &best_s, -1);
3356 
3357  pb_bits_count= put_bits_count(&s->pb);
3358  flush_put_bits(&s->pb);
3359  avpriv_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
3360  s->pb= backup_s.pb;
3361 
3362  if(s->data_partitioning){
3363  pb2_bits_count= put_bits_count(&s->pb2);
3364  flush_put_bits(&s->pb2);
3365  avpriv_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
3366  s->pb2= backup_s.pb2;
3367 
3368  tex_pb_bits_count= put_bits_count(&s->tex_pb);
3369  flush_put_bits(&s->tex_pb);
3370  avpriv_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
3371  s->tex_pb= backup_s.tex_pb;
3372  }
3373  s->last_bits= put_bits_count(&s->pb);
3374 
3375  if (CONFIG_H263_ENCODER &&
3376  s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
3377  ff_h263_update_motion_val(s);
3378 
3379  if(next_block==0){ //FIXME 16 vs linesize16
3380  s->hdsp.put_pixels_tab[0][0](s->dest[0], s->sc.rd_scratchpad , s->linesize ,16);
3381  s->hdsp.put_pixels_tab[1][0](s->dest[1], s->sc.rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
3382  s->hdsp.put_pixels_tab[1][0](s->dest[2], s->sc.rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
3383  }
3384 
3385  if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
3386  ff_mpv_reconstruct_mb(s, s->block);
3387  } else {
3388  int motion_x = 0, motion_y = 0;
3389  s->mv_type=MV_TYPE_16X16;
3390  // only one MB-Type possible
3391 
3392  switch(mb_type){
3393  case CANDIDATE_MB_TYPE_INTRA:
3394  s->mv_dir = 0;
3395  s->mb_intra= 1;
3396  motion_x= s->mv[0][0][0] = 0;
3397  motion_y= s->mv[0][0][1] = 0;
3398  break;
3399  case CANDIDATE_MB_TYPE_INTER:
3400  s->mv_dir = MV_DIR_FORWARD;
3401  s->mb_intra= 0;
3402  motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
3403  motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
3404  break;
3405  case CANDIDATE_MB_TYPE_INTER_I:
3406  s->mv_dir = MV_DIR_FORWARD;
3407  s->mv_type = MV_TYPE_FIELD;
3408  s->mb_intra= 0;
3409  for(i=0; i<2; i++){
3410  j= s->field_select[0][i] = s->p_field_select_table[i][xy];
3411  s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
3412  s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
3413  }
3414  break;
3415  case CANDIDATE_MB_TYPE_INTER4V:
3416  s->mv_dir = MV_DIR_FORWARD;
3417  s->mv_type = MV_TYPE_8X8;
3418  s->mb_intra= 0;
3419  for(i=0; i<4; i++){
3420  s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
3421  s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
3422  }
3423  break;
3424  case CANDIDATE_MB_TYPE_DIRECT:
3425  if (CONFIG_MPEG4_ENCODER) {
3426  s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
3427  s->mb_intra= 0;
3428  motion_x=s->b_direct_mv_table[xy][0];
3429  motion_y=s->b_direct_mv_table[xy][1];
3430  ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
3431  }
3432  break;
3433  case CANDIDATE_MB_TYPE_DIRECT0:
3434  if (CONFIG_MPEG4_ENCODER) {
3435  s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
3436  s->mb_intra= 0;
3437  ff_mpeg4_set_direct_mv(s, 0, 0);
3438  }
3439  break;
3440  case CANDIDATE_MB_TYPE_BIDIR:
3441  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3442  s->mb_intra= 0;
3443  s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
3444  s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
3445  s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
3446  s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
3447  break;
3448  case CANDIDATE_MB_TYPE_BACKWARD:
3449  s->mv_dir = MV_DIR_BACKWARD;
3450  s->mb_intra= 0;
3451  motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
3452  motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
3453  break;
3454  case CANDIDATE_MB_TYPE_FORWARD:
3455  s->mv_dir = MV_DIR_FORWARD;
3456  s->mb_intra= 0;
3457  motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
3458  motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
3459  break;
3460  case CANDIDATE_MB_TYPE_FORWARD_I:
3461  s->mv_dir = MV_DIR_FORWARD;
3462  s->mv_type = MV_TYPE_FIELD;
3463  s->mb_intra= 0;
3464  for(i=0; i<2; i++){
3465  j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
3466  s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
3467  s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
3468  }
3469  break;
3470  case CANDIDATE_MB_TYPE_BACKWARD_I:
3471  s->mv_dir = MV_DIR_BACKWARD;
3472  s->mv_type = MV_TYPE_FIELD;
3473  s->mb_intra= 0;
3474  for(i=0; i<2; i++){
3475  j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
3476  s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
3477  s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
3478  }
3479  break;
3480  case CANDIDATE_MB_TYPE_BIDIR_I:
3481  s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
3482  s->mv_type = MV_TYPE_FIELD;
3483  s->mb_intra= 0;
3484  for(dir=0; dir<2; dir++){
3485  for(i=0; i<2; i++){
3486  j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
3487  s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
3488  s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
3489  }
3490  }
3491  break;
3492  default:
3493  av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
3494  }
3495 
3496  encode_mb(s, motion_x, motion_y);
3497 
3498  // RAL: Update last macroblock type
3499  s->last_mv_dir = s->mv_dir;
3500 
3501  if (CONFIG_H263_ENCODER &&
3502  s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
3503  ff_h263_update_motion_val(s);
3504 
3505  ff_mpv_reconstruct_mb(s, s->block);
3506  }
3507 
3508  /* clean the MV table in IPS frames for direct mode in B-frames */
3509  if(s->mb_intra /* && I,P,S_TYPE */){
3510  s->p_mv_table[xy][0]=0;
3511  s->p_mv_table[xy][1]=0;
3512  }
3513 
3514  if (s->avctx->flags & AV_CODEC_FLAG_PSNR) {
3515  int w= 16;
3516  int h= 16;
3517 
3518  if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
3519  if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
3520 
3521  s->current_picture.encoding_error[0] += sse(
3522  s, s->new_picture.f->data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
3523  s->dest[0], w, h, s->linesize);
3524  s->current_picture.encoding_error[1] += sse(
3525  s, s->new_picture.f->data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
3526  s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
3527  s->current_picture.encoding_error[2] += sse(
3528  s, s->new_picture.f->data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*chr_h,
3529  s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
3530  }
3531  if(s->loop_filter){
3532  if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
3533  ff_h263_loop_filter(s);
3534  }
3535  ff_dlog(s->avctx, "MB %d %d bits\n",
3536  s->mb_x + s->mb_y * s->mb_stride, put_bits_count(&s->pb));
3537  }
3538  }
3539 
3540  //not beautiful here but we must write it before flushing so it has to be here
3541  if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
3542  ff_msmpeg4_encode_ext_header(s);
3543 
3544  write_slice_end(s);
3545 
3546 #if FF_API_RTP_CALLBACK
3547 FF_DISABLE_DEPRECATION_WARNINGS
3548  /* Send the last GOB if RTP */
3549  if (s->avctx->rtp_callback) {
3550  int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
3551  int pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
3552  /* Call the RTP callback to send the last GOB */
3553  emms_c();
3554  s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
3555  }
3556 FF_ENABLE_DEPRECATION_WARNINGS
3557 #endif
3558 
3559  return 0;
3560 }
3561 
3562 #define MERGE(field) dst->field += src->field; src->field=0
3563 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
3564  MERGE(me.scene_change_score);
3565  MERGE(me.mc_mb_var_sum_temp);
3566  MERGE(me.mb_var_sum_temp);
3567 }
3568 
3569 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
3570  int i;
3571 
3572  MERGE(dct_count[0]); //note, the other dct vars are not part of the context
3573  MERGE(dct_count[1]);
3574  MERGE(mv_bits);
3575  MERGE(i_tex_bits);
3576  MERGE(p_tex_bits);
3577  MERGE(i_count);
3578  MERGE(f_count);
3579  MERGE(b_count);
3580  MERGE(skip_count);
3581  MERGE(misc_bits);
3582  MERGE(er.error_count);
3583  MERGE(padding_bug_score);
3584  MERGE(current_picture.encoding_error[0]);
3585  MERGE(current_picture.encoding_error[1]);
3586  MERGE(current_picture.encoding_error[2]);
3587 
3588  if (dst->noise_reduction){
3589  for(i=0; i<64; i++){
3590  MERGE(dct_error_sum[0][i]);
3591  MERGE(dct_error_sum[1][i]);
3592  }
3593  }
3594 
3595  av_assert1(put_bits_count(&src->pb) % 8 ==0);
3596  av_assert1(put_bits_count(&dst->pb) % 8 ==0);
3597  avpriv_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
3598  flush_put_bits(&dst->pb);
3599 }
3600 
3601 static int estimate_qp(MpegEncContext *s, int dry_run){
3602  if (s->next_lambda){
3603  s->current_picture_ptr->f->quality =
3604  s->current_picture.f->quality = s->next_lambda;
3605  if(!dry_run) s->next_lambda= 0;
3606  } else if (!s->fixed_qscale) {
3607  int quality = ff_rate_estimate_qscale(s, dry_run);
3608  s->current_picture_ptr->f->quality =
3609  s->current_picture.f->quality = quality;
3610  if (s->current_picture.f->quality < 0)
3611  return -1;
3612  }
3613 
3614  if(s->adaptive_quant){
3615  switch(s->codec_id){
3616  case AV_CODEC_ID_MPEG4:
3617  if (CONFIG_MPEG4_ENCODER)
3618  ff_clean_mpeg4_qscales(s);
3619  break;
3620  case AV_CODEC_ID_H263:
3621  case AV_CODEC_ID_H263P:
3622  case AV_CODEC_ID_FLV1:
3623  if (CONFIG_H263_ENCODER)
3624  ff_clean_h263_qscales(s);
3625  break;
3626  default:
3627  ff_init_qscale_tab(s);
3628  }
3629 
3630  s->lambda= s->lambda_table[0];
3631  //FIXME broken
3632  }else
3633  s->lambda = s->current_picture.f->quality;
3634  update_qscale(s);
3635  return 0;
3636 }
3637 
3638 /* must be called before writing the header */
3639 static void set_frame_distances(MpegEncContext * s){
3640  av_assert1(s->current_picture_ptr->f->pts != AV_NOPTS_VALUE);
3641  s->time = s->current_picture_ptr->f->pts * s->avctx->time_base.num;
3642 
3643  if(s->pict_type==AV_PICTURE_TYPE_B){
3644  s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
3645  av_assert1(s->pb_time > 0 && s->pb_time < s->pp_time);
3646  }else{
3647  s->pp_time= s->time - s->last_non_b_time;
3648  s->last_non_b_time= s->time;
3649  av_assert1(s->picture_number==0 || s->pp_time > 0);
3650  }
3651 }
3652 
3653 static int encode_picture(MpegEncContext *s, int picture_number)
3654 {
3655  int i, ret;
3656  int bits;
3657  int context_count = s->slice_context_count;
3658 
3659  s->picture_number = picture_number;
3660 
3661  /* Reset the average MB variance */
3662  s->me.mb_var_sum_temp =
3663  s->me.mc_mb_var_sum_temp = 0;
3664 
3665  /* we need to initialize some time vars before we can encode B-frames */
3666  // RAL: Condition added for MPEG1VIDEO
3667  if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO || s->codec_id == AV_CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
3668  set_frame_distances(s);
3669  if(CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4)
3670  ff_set_mpeg4_time(s);
3671 
3672  s->me.scene_change_score=0;
3673 
3674 // s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
3675 
3676  if(s->pict_type==AV_PICTURE_TYPE_I){
3677  if(s->msmpeg4_version >= 3) s->no_rounding=1;
3678  else s->no_rounding=0;
3679  }else if(s->pict_type!=AV_PICTURE_TYPE_B){
3680  if(s->flipflop_rounding || s->codec_id == AV_CODEC_ID_H263P || s->codec_id == AV_CODEC_ID_MPEG4)
3681  s->no_rounding ^= 1;
3682  }
3683 
3684  if (s->avctx->flags & AV_CODEC_FLAG_PASS2) {
3685  if (estimate_qp(s,1) < 0)
3686  return -1;
3687  ff_get_2pass_fcode(s);
3688  } else if (!(s->avctx->flags & AV_CODEC_FLAG_QSCALE)) {
3689  if(s->pict_type==AV_PICTURE_TYPE_B)
3690  s->lambda= s->last_lambda_for[s->pict_type];
3691  else
3692  s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
3693  update_qscale(s);
3694  }
3695 
3696  if(s->codec_id != AV_CODEC_ID_AMV && s->codec_id != AV_CODEC_ID_MJPEG){
3697  if(s->q_chroma_intra_matrix != s->q_intra_matrix ) av_freep(&s->q_chroma_intra_matrix);
3698  if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
3699  s->q_chroma_intra_matrix = s->q_intra_matrix;
3700  s->q_chroma_intra_matrix16 = s->q_intra_matrix16;
3701  }
3702 
3703  s->mb_intra=0; //for the rate distortion & bit compare functions
3704  for(i=1; i<context_count; i++){
3705  ret = ff_update_duplicate_context(s->thread_context[i], s);
3706  if (ret < 0)
3707  return ret;
3708  }
3709 
3710  if(ff_init_me(s)<0)
3711  return -1;
3712 
3713  /* Estimate motion for every MB */
3714  if(s->pict_type != AV_PICTURE_TYPE_I){
3715  s->lambda = (s->lambda * s->me_penalty_compensation + 128) >> 8;
3716  s->lambda2 = (s->lambda2 * (int64_t) s->me_penalty_compensation + 128) >> 8;
3717  if (s->pict_type != AV_PICTURE_TYPE_B) {
3718  if ((s->me_pre && s->last_non_b_pict_type == AV_PICTURE_TYPE_I) ||
3719  s->me_pre == 2) {
3720  s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3721  }
3722  }
3723 
3724  s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3725  }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
3726  /* I-Frame */
3727  for(i=0; i<s->mb_stride*s->mb_height; i++)
3728  s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3729 
3730  if(!s->fixed_qscale){
3731  /* finding spatial complexity for I-frame rate control */
3732  s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3733  }
3734  }
3735  for(i=1; i<context_count; i++){
3736  merge_context_after_me(s, s->thread_context[i]);
3737  }
3738  s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
3739  s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
3740  emms_c();
3741 
3742  if (s->me.scene_change_score > s->scenechange_threshold &&
3743  s->pict_type == AV_PICTURE_TYPE_P) {
3744  s->pict_type= AV_PICTURE_TYPE_I;
3745  for(i=0; i<s->mb_stride*s->mb_height; i++)
3746  s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
3747  if(s->msmpeg4_version >= 3)
3748  s->no_rounding=1;
3749  ff_dlog(s, "Scene change detected, encoding as I Frame %"PRId64" %"PRId64"\n",
3750  s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
3751  }
3752 
3753  if(!s->umvplus){
3754  if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
3755  s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
3756 
3757  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3758  int a,b;
3759  a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
3760  b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
3761  s->f_code= FFMAX3(s->f_code, a, b);
3762  }
3763 
3764  ff_fix_long_p_mvs(s);
3765  ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
3766  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3767  int j;
3768  for(i=0; i<2; i++){
3769  for(j=0; j<2; j++)
3770  ff_fix_long_mvs(s, s->p_field_select_table[i], j,
3771  s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
3772  }
3773  }
3774  }
3775 
3776  if(s->pict_type==AV_PICTURE_TYPE_B){
3777  int a, b;
3778 
3779  a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
3780  b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
3781  s->f_code = FFMAX(a, b);
3782 
3783  a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
3784  b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
3785  s->b_code = FFMAX(a, b);
3786 
3787  ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
3788  ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
3789  ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
3790  ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
3791  if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
3792  int dir, j;
3793  for(dir=0; dir<2; dir++){
3794  for(i=0; i<2; i++){
3795  for(j=0; j<2; j++){
3796  int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
3797  : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
3798  ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
3799  s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
3800  }
3801  }
3802  }
3803  }
3804  }
3805  }
3806 
3807  if (estimate_qp(s, 0) < 0)
3808  return -1;
3809 
3810  if (s->qscale < 3 && s->max_qcoeff <= 128 &&
3811  s->pict_type == AV_PICTURE_TYPE_I &&
3812  !(s->avctx->flags & AV_CODEC_FLAG_QSCALE))
3813  s->qscale= 3; //reduce clipping problems
3814 
3815  if (s->out_format == FMT_MJPEG) {
3816  const uint16_t * luma_matrix = ff_mpeg1_default_intra_matrix;
3817  const uint16_t *chroma_matrix = ff_mpeg1_default_intra_matrix;
3818 
3819  if (s->avctx->intra_matrix) {
3820  chroma_matrix =
3821  luma_matrix = s->avctx->intra_matrix;
3822  }
3823  if (s->avctx->chroma_intra_matrix)
3824  chroma_matrix = s->avctx->chroma_intra_matrix;
3825 
3826  /* for mjpeg, we do include qscale in the matrix */
3827  for(i=1;i<64;i++){
3828  int j = s->idsp.idct_permutation[i];
3829 
3830  s->chroma_intra_matrix[j] = av_clip_uint8((chroma_matrix[i] * s->qscale) >> 3);
3831  s-> intra_matrix[j] = av_clip_uint8(( luma_matrix[i] * s->qscale) >> 3);
3832  }
3833  s->y_dc_scale_table=
3834  s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
3835  s->chroma_intra_matrix[0] =
3836  s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
3837  ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
3838  s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
3839  ff_convert_matrix(s, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
3840  s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
3841  s->qscale= 8;
3842  }
3843  if(s->codec_id == AV_CODEC_ID_AMV){
3844  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};
3845  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};
3846  for(i=1;i<64;i++){
3847  int j= s->idsp.idct_permutation[ff_zigzag_direct[i]];
3848 
3849  s->intra_matrix[j] = sp5x_quant_table[5*2+0][i];
3850  s->chroma_intra_matrix[j] = sp5x_quant_table[5*2+1][i];
3851  }
3852  s->y_dc_scale_table= y;
3853  s->c_dc_scale_table= c;
3854  s->intra_matrix[0] = 13;
3855  s->chroma_intra_matrix[0] = 14;
3856  ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
3857  s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
3858  ff_convert_matrix(s, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
3859  s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
3860  s->qscale= 8;
3861  }
3862 
3863  //FIXME var duplication
3864  s->current_picture_ptr->f->key_frame =
3865  s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
3866  s->current_picture_ptr->f->pict_type =
3867  s->current_picture.f->pict_type = s->pict_type;
3868 
3869  if (s->current_picture.f->key_frame)
3870  s->picture_in_gop_number=0;
3871 
3872  s->mb_x = s->mb_y = 0;
3873  s->last_bits= put_bits_count(&s->pb);
3874  switch(s->out_format) {
3875  case FMT_MJPEG:
3876  if (CONFIG_MJPEG_ENCODER && s->huffman != HUFFMAN_TABLE_OPTIMAL)
3877  ff_mjpeg_encode_picture_header(s->avctx, &s->pb, &s->intra_scantable,
3878  s->pred, s->intra_matrix, s->chroma_intra_matrix);
3879  break;
3880  case FMT_H261:
3881  if (CONFIG_H261_ENCODER)
3882  ff_h261_encode_picture_header(s, picture_number);
3883  break;
3884  case FMT_H263:
3885  if (CONFIG_WMV2_ENCODER && s->codec_id == AV_CODEC_ID_WMV2)
3886  ff_wmv2_encode_picture_header(s, picture_number);
3887  else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
3888  ff_msmpeg4_encode_picture_header(s, picture_number);
3889  else if (CONFIG_MPEG4_ENCODER && s->h263_pred) {
3890  ret = ff_mpeg4_encode_picture_header(s, picture_number);
3891  if (ret < 0)
3892  return ret;
3893  } else if (CONFIG_RV10_ENCODER && s->codec_id == AV_CODEC_ID_RV10) {
3894  ret = ff_rv10_encode_picture_header(s, picture_number);
3895  if (ret < 0)
3896  return ret;
3897  }
3898  else if (CONFIG_RV20_ENCODER && s->codec_id == AV_CODEC_ID_RV20)
3899  ff_rv20_encode_picture_header(s, picture_number);
3900  else if (CONFIG_FLV_ENCODER && s->codec_id == AV_CODEC_ID_FLV1)
3901  ff_flv_encode_picture_header(s, picture_number);
3902  else if (CONFIG_H263_ENCODER)
3903  ff_h263_encode_picture_header(s, picture_number);
3904  break;
3905  case FMT_MPEG1:
3906  if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
3907  ff_mpeg1_encode_picture_header(s, picture_number);
3908  break;
3909  default:
3910  av_assert0(0);
3911  }
3912  bits= put_bits_count(&s->pb);
3913  s->header_bits= bits - s->last_bits;
3914 
3915  for(i=1; i<context_count; i++){
3916  update_duplicate_context_after_me(s->thread_context[i], s);
3917  }
3918  s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
3919  for(i=1; i<context_count; i++){
3920  if (s->pb.buf_end == s->thread_context[i]->pb.buf)
3921  set_put_bits_buffer_size(&s->pb, FFMIN(s->thread_context[i]->pb.buf_end - s->pb.buf, INT_MAX/8-32));
3922  merge_context_after_encode(s, s->thread_context[i]);
3923  }
3924  emms_c();
3925  return 0;
3926 }
3927 
3928 static void denoise_dct_c(MpegEncContext *s, int16_t *block){
3929  const int intra= s->mb_intra;
3930  int i;
3931 
3932  s->dct_count[intra]++;
3933 
3934  for(i=0; i<64; i++){
3935  int level= block[i];
3936 
3937  if(level){
3938  if(level>0){
3939  s->dct_error_sum[intra][i] += level;
3940  level -= s->dct_offset[intra][i];
3941  if(level<0) level=0;
3942  }else{
3943  s->dct_error_sum[intra][i] -= level;
3944  level += s->dct_offset[intra][i];
3945  if(level>0) level=0;
3946  }
3947  block[i]= level;
3948  }
3949  }
3950 }
3951 
3952 static int dct_quantize_trellis_c(MpegEncContext *s,
3953  int16_t *block, int n,
3954  int qscale, int *overflow){
3955  const int *qmat;
3956  const uint16_t *matrix;
3957  const uint8_t *scantable;
3958  const uint8_t *perm_scantable;
3959  int max=0;
3960  unsigned int threshold1, threshold2;
3961  int bias=0;
3962  int run_tab[65];
3963  int level_tab[65];
3964  int score_tab[65];
3965  int survivor[65];
3966  int survivor_count;
3967  int last_run=0;
3968  int last_level=0;
3969  int last_score= 0;
3970  int last_i;
3971  int coeff[2][64];
3972  int coeff_count[64];
3973  int qmul, qadd, start_i, last_non_zero, i, dc;
3974  const int esc_length= s->ac_esc_length;
3975  uint8_t * length;
3976  uint8_t * last_length;
3977  const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3978  int mpeg2_qscale;
3979 
3980  s->fdsp.fdct(block);
3981 
3982  if(s->dct_error_sum)
3983  s->denoise_dct(s, block);
3984  qmul= qscale*16;
3985  qadd= ((qscale-1)|1)*8;
3986 
3987  if (s->q_scale_type) mpeg2_qscale = ff_mpeg2_non_linear_qscale[qscale];
3988  else mpeg2_qscale = qscale << 1;
3989 
3990  if (s->mb_intra) {
3991  int q;
3992  scantable= s->intra_scantable.scantable;
3993  perm_scantable= s->intra_scantable.permutated;
3994  if (!s->h263_aic) {
3995  if (n < 4)
3996  q = s->y_dc_scale;
3997  else
3998  q = s->c_dc_scale;
3999  q = q << 3;
4000  } else{
4001  /* For AIC we skip quant/dequant of INTRADC */
4002  q = 1 << 3;
4003  qadd=0;
4004  }
4005 
4006  /* note: block[0] is assumed to be positive */
4007  block[0] = (block[0] + (q >> 1)) / q;
4008  start_i = 1;
4009  last_non_zero = 0;
4010  qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
4011  matrix = n < 4 ? s->intra_matrix : s->chroma_intra_matrix;
4012  if(s->mpeg_quant || s->out_format == FMT_MPEG1 || s->out_format == FMT_MJPEG)
4013  bias= 1<<(QMAT_SHIFT-1);
4014 
4015  if (n > 3 && s->intra_chroma_ac_vlc_length) {
4016  length = s->intra_chroma_ac_vlc_length;
4017  last_length= s->intra_chroma_ac_vlc_last_length;
4018  } else {
4019  length = s->intra_ac_vlc_length;
4020  last_length= s->intra_ac_vlc_last_length;
4021  }
4022  } else {
4023  scantable= s->inter_scantable.scantable;
4024  perm_scantable= s->inter_scantable.permutated;
4025  start_i = 0;
4026  last_non_zero = -1;
4027  qmat = s->q_inter_matrix[qscale];
4028  matrix = s->inter_matrix;
4029  length = s->inter_ac_vlc_length;
4030  last_length= s->inter_ac_vlc_last_length;
4031  }
4032  last_i= start_i;
4033 
4034  threshold1= (1<<QMAT_SHIFT) - bias - 1;
4035  threshold2= (threshold1<<1);
4036 
4037  for(i=63; i>=start_i; i--) {
4038  const int j = scantable[i];
4039  int level = block[j] * qmat[j];
4040 
4041  if(((unsigned)(level+threshold1))>threshold2){
4042  last_non_zero = i;
4043  break;
4044  }
4045  }
4046 
4047  for(i=start_i; i<=last_non_zero; i++) {
4048  const int j = scantable[i];
4049  int level = block[j] * qmat[j];
4050 
4051 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
4052 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
4053  if(((unsigned)(level+threshold1))>threshold2){
4054  if(level>0){
4055  level= (bias + level)>>QMAT_SHIFT;
4056  coeff[0][i]= level;
4057  coeff[1][i]= level-1;
4058 // coeff[2][k]= level-2;
4059  }else{
4060  level= (bias - level)>>QMAT_SHIFT;
4061  coeff[0][i]= -level;
4062  coeff[1][i]= -level+1;
4063 // coeff[2][k]= -level+2;
4064  }
4065  coeff_count[i]= FFMIN(level, 2);
4066  av_assert2(coeff_count[i]);
4067  max |=level;
4068  }else{
4069  coeff[0][i]= (level>>31)|1;
4070  coeff_count[i]= 1;
4071  }
4072  }
4073 
4074  *overflow= s->max_qcoeff < max; //overflow might have happened
4075 
4076  if(last_non_zero < start_i){
4077  memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
4078  return last_non_zero;
4079  }
4080 
4081  score_tab[start_i]= 0;
4082  survivor[0]= start_i;
4083  survivor_count= 1;
4084 
4085  for(i=start_i; i<=last_non_zero; i++){
4086  int level_index, j, zero_distortion;
4087  int dct_coeff= FFABS(block[ scantable[i] ]);
4088  int best_score=256*256*256*120;
4089 
4090  if (s->fdsp.fdct == ff_fdct_ifast)
4091  dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
4092  zero_distortion= dct_coeff*dct_coeff;
4093 
4094  for(level_index=0; level_index < coeff_count[i]; level_index++){
4095  int distortion;
4096  int level= coeff[level_index][i];
4097  const int alevel= FFABS(level);
4098  int unquant_coeff;
4099 
4100  av_assert2(level);
4101 
4102  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4103  unquant_coeff= alevel*qmul + qadd;
4104  } else if(s->out_format == FMT_MJPEG) {
4105  j = s->idsp.idct_permutation[scantable[i]];
4106  unquant_coeff = alevel * matrix[j] * 8;
4107  }else{ // MPEG-1
4108  j = s->idsp.idct_permutation[scantable[i]]; // FIXME: optimize
4109  if(s->mb_intra){
4110  unquant_coeff = (int)( alevel * mpeg2_qscale * matrix[j]) >> 4;
4111  unquant_coeff = (unquant_coeff - 1) | 1;
4112  }else{
4113  unquant_coeff = ((( alevel << 1) + 1) * mpeg2_qscale * ((int) matrix[j])) >> 5;
4114  unquant_coeff = (unquant_coeff - 1) | 1;
4115  }
4116  unquant_coeff<<= 3;
4117  }
4118 
4119  distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
4120  level+=64;
4121  if((level&(~127)) == 0){
4122  for(j=survivor_count-1; j>=0; j--){
4123  int run= i - survivor[j];
4124  int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
4125  score += score_tab[i-run];
4126 
4127  if(score < best_score){
4128  best_score= score;
4129  run_tab[i+1]= run;
4130  level_tab[i+1]= level-64;
4131  }
4132  }
4133 
4134  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4135  for(j=survivor_count-1; j>=0; j--){
4136  int run= i - survivor[j];
4137  int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
4138  score += score_tab[i-run];
4139  if(score < last_score){
4140  last_score= score;
4141  last_run= run;
4142  last_level= level-64;
4143  last_i= i+1;
4144  }
4145  }
4146  }
4147  }else{
4148  distortion += esc_length*lambda;
4149  for(j=survivor_count-1; j>=0; j--){
4150  int run= i - survivor[j];
4151  int score= distortion + score_tab[i-run];
4152 
4153  if(score < best_score){
4154  best_score= score;
4155  run_tab[i+1]= run;
4156  level_tab[i+1]= level-64;
4157  }
4158  }
4159 
4160  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4161  for(j=survivor_count-1; j>=0; j--){
4162  int run= i - survivor[j];
4163  int score= distortion + score_tab[i-run];
4164  if(score < last_score){
4165  last_score= score;
4166  last_run= run;
4167  last_level= level-64;
4168  last_i= i+1;
4169  }
4170  }
4171  }
4172  }
4173  }
4174 
4175  score_tab[i+1]= best_score;
4176 
4177  // 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
4178  if(last_non_zero <= 27){
4179  for(; survivor_count; survivor_count--){
4180  if(score_tab[ survivor[survivor_count-1] ] <= best_score)
4181  break;
4182  }
4183  }else{
4184  for(; survivor_count; survivor_count--){
4185  if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
4186  break;
4187  }
4188  }
4189 
4190  survivor[ survivor_count++ ]= i+1;
4191  }
4192 
4193  if(s->out_format != FMT_H263 && s->out_format != FMT_H261){
4194  last_score= 256*256*256*120;
4195  for(i= survivor[0]; i<=last_non_zero + 1; i++){
4196  int score= score_tab[i];
4197  if (i)
4198  score += lambda * 2; // FIXME more exact?
4199 
4200  if(score < last_score){
4201  last_score= score;
4202  last_i= i;
4203  last_level= level_tab[i];
4204  last_run= run_tab[i];
4205  }
4206  }
4207  }
4208 
4209  s->coded_score[n] = last_score;
4210 
4211  dc= FFABS(block[0]);
4212  last_non_zero= last_i - 1;
4213  memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
4214 
4215  if(last_non_zero < start_i)
4216  return last_non_zero;
4217 
4218  if(last_non_zero == 0 && start_i == 0){
4219  int best_level= 0;
4220  int best_score= dc * dc;
4221 
4222  for(i=0; i<coeff_count[0]; i++){
4223  int level= coeff[i][0];
4224  int alevel= FFABS(level);
4225  int unquant_coeff, score, distortion;
4226 
4227  if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
4228  unquant_coeff= (alevel*qmul + qadd)>>3;
4229  } else{ // MPEG-1
4230  unquant_coeff = ((( alevel << 1) + 1) * mpeg2_qscale * ((int) matrix[0])) >> 5;
4231  unquant_coeff = (unquant_coeff - 1) | 1;
4232  }
4233  unquant_coeff = (unquant_coeff + 4) >> 3;
4234  unquant_coeff<<= 3 + 3;
4235 
4236  distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
4237  level+=64;
4238  if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
4239  else score= distortion + esc_length*lambda;
4240 
4241  if(score < best_score){
4242  best_score= score;
4243  best_level= level - 64;
4244  }
4245  }
4246  block[0]= best_level;
4247  s->coded_score[n] = best_score - dc*dc;
4248  if(best_level == 0) return -1;
4249  else return last_non_zero;
4250  }
4251 
4252  i= last_i;
4253  av_assert2(last_level);
4254 
4255  block[ perm_scantable[last_non_zero] ]= last_level;
4256  i -= last_run + 1;
4257 
4258  for(; i>start_i; i -= run_tab[i] + 1){
4259  block[ perm_scantable[i-1] ]= level_tab[i];
4260  }
4261 
4262  return last_non_zero;
4263 }
4264 
4265 //#define REFINE_STATS 1
4266 static int16_t basis[64][64];
4267 
4268 static void build_basis(uint8_t *perm){
4269  int i, j, x, y;
4270  emms_c();
4271  for(i=0; i<8; i++){
4272  for(j=0; j<8; j++){
4273  for(y=0; y<8; y++){
4274  for(x=0; x<8; x++){
4275  double s= 0.25*(1<<BASIS_SHIFT);
4276  int index= 8*i + j;
4277  int perm_index= perm[index];
4278  if(i==0) s*= sqrt(0.5);
4279  if(j==0) s*= sqrt(0.5);
4280  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)));
4281  }
4282  }
4283  }
4284  }
4285 }
4286 
4287 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
4288  int16_t *block, int16_t *weight, int16_t *orig,
4289  int n, int qscale){
4290  int16_t rem[64];
4291  LOCAL_ALIGNED_16(int16_t, d1, [64]);
4292  const uint8_t *scantable;
4293  const uint8_t *perm_scantable;
4294 // unsigned int threshold1, threshold2;
4295 // int bias=0;
4296  int run_tab[65];
4297  int prev_run=0;
4298  int prev_level=0;
4299  int qmul, qadd, start_i, last_non_zero, i, dc;
4300  uint8_t * length;
4301  uint8_t * last_length;
4302  int lambda;
4303  int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
4304 #ifdef REFINE_STATS
4305 static int count=0;
4306 static int after_last=0;
4307 static int to_zero=0;
4308 static int from_zero=0;
4309 static int raise=0;
4310 static int lower=0;
4311 static int messed_sign=0;
4312 #endif
4313 
4314  if(basis[0][0] == 0)
4315  build_basis(s->idsp.idct_permutation);
4316 
4317  qmul= qscale*2;
4318  qadd= (qscale-1)|1;
4319  if (s->mb_intra) {
4320  scantable= s->intra_scantable.scantable;
4321  perm_scantable= s->intra_scantable.permutated;
4322  if (!s->h263_aic) {
4323  if (n < 4)
4324  q = s->y_dc_scale;
4325  else
4326  q = s->c_dc_scale;
4327  } else{
4328  /* For AIC we skip quant/dequant of INTRADC */
4329  q = 1;
4330  qadd=0;
4331  }
4332  q <<= RECON_SHIFT-3;
4333  /* note: block[0] is assumed to be positive */
4334  dc= block[0]*q;
4335 // block[0] = (block[0] + (q >> 1)) / q;
4336  start_i = 1;
4337 // if(s->mpeg_quant || s->out_format == FMT_MPEG1)
4338 // bias= 1<<(QMAT_SHIFT-1);
4339  if (n > 3 && s->intra_chroma_ac_vlc_length) {
4340  length = s->intra_chroma_ac_vlc_length;
4341  last_length= s->intra_chroma_ac_vlc_last_length;
4342  } else {
4343  length = s->intra_ac_vlc_length;
4344  last_length= s->intra_ac_vlc_last_length;
4345  }
4346  } else {
4347  scantable= s->inter_scantable.scantable;
4348  perm_scantable= s->inter_scantable.permutated;
4349  dc= 0;
4350  start_i = 0;
4351  length = s->inter_ac_vlc_length;
4352  last_length= s->inter_ac_vlc_last_length;
4353  }
4354  last_non_zero = s->block_last_index[n];
4355 
4356 #ifdef REFINE_STATS
4357 {START_TIMER
4358 #endif
4359  dc += (1<<(RECON_SHIFT-1));
4360  for(i=0; i<64; i++){
4361  rem[i] = dc - (orig[i] << RECON_SHIFT); // FIXME use orig directly instead of copying to rem[]
4362  }
4363 #ifdef REFINE_STATS
4364 STOP_TIMER("memset rem[]")}
4365 #endif
4366  sum=0;
4367  for(i=0; i<64; i++){
4368  int one= 36;
4369  int qns=4;
4370  int w;
4371 
4372  w= FFABS(weight[i]) + qns*one;
4373  w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
4374 
4375  weight[i] = w;
4376 // w=weight[i] = (63*qns + (w/2)) / w;
4377 
4378  av_assert2(w>0);
4379  av_assert2(w<(1<<6));
4380  sum += w*w;
4381  }
4382  lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
4383 #ifdef REFINE_STATS
4384 {START_TIMER
4385 #endif
4386  run=0;
4387  rle_index=0;
4388  for(i=start_i; i<=last_non_zero; i++){
4389  int j= perm_scantable[i];
4390  const int level= block[j];
4391  int coeff;
4392 
4393  if(level){
4394  if(level<0) coeff= qmul*level - qadd;
4395  else coeff= qmul*level + qadd;
4396  run_tab[rle_index++]=run;
4397  run=0;
4398 
4399  s->mpvencdsp.add_8x8basis(rem, basis[j], coeff);
4400  }else{
4401  run++;
4402  }
4403  }
4404 #ifdef REFINE_STATS
4405 if(last_non_zero>0){
4406 STOP_TIMER("init rem[]")
4407 }
4408 }
4409 
4410 {START_TIMER
4411 #endif
4412  for(;;){
4413  int best_score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0], 0);
4414  int best_coeff=0;
4415  int best_change=0;
4416  int run2, best_unquant_change=0, analyze_gradient;
4417 #ifdef REFINE_STATS
4418 {START_TIMER
4419 #endif
4420  analyze_gradient = last_non_zero > 2 || s->quantizer_noise_shaping >= 3;
4421 
4422  if(analyze_gradient){
4423 #ifdef REFINE_STATS
4424 {START_TIMER
4425 #endif
4426  for(i=0; i<64; i++){
4427  int w= weight[i];
4428 
4429  d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
4430  }
4431 #ifdef REFINE_STATS
4432 STOP_TIMER("rem*w*w")}
4433 {START_TIMER
4434 #endif
4435  s->fdsp.fdct(d1);
4436 #ifdef REFINE_STATS
4437 STOP_TIMER("dct")}
4438 #endif
4439  }
4440 
4441  if(start_i){
4442  const int level= block[0];
4443  int change, old_coeff;
4444 
4445  av_assert2(s->mb_intra);
4446 
4447  old_coeff= q*level;
4448 
4449  for(change=-1; change<=1; change+=2){
4450  int new_level= level + change;
4451  int score, new_coeff;
4452 
4453  new_coeff= q*new_level;
4454  if(new_coeff >= 2048 || new_coeff < 0)
4455  continue;
4456 
4457  score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0],
4458  new_coeff - old_coeff);
4459  if(score<best_score){
4460  best_score= score;
4461  best_coeff= 0;
4462  best_change= change;
4463  best_unquant_change= new_coeff - old_coeff;
4464  }
4465  }
4466  }
4467 
4468  run=0;
4469  rle_index=0;
4470  run2= run_tab[rle_index++];
4471  prev_level=0;
4472  prev_run=0;
4473 
4474  for(i=start_i; i<64; i++){
4475  int j= perm_scantable[i];
4476  const int level= block[j];
4477  int change, old_coeff;
4478 
4479  if(s->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
4480  break;
4481 
4482  if(level){
4483  if(level<0) old_coeff= qmul*level - qadd;
4484  else old_coeff= qmul*level + qadd;
4485  run2= run_tab[rle_index++]; //FIXME ! maybe after last
4486  }else{
4487  old_coeff=0;
4488  run2--;
4489  av_assert2(run2>=0 || i >= last_non_zero );
4490  }
4491 
4492  for(change=-1; change<=1; change+=2){
4493  int new_level= level + change;
4494  int score, new_coeff, unquant_change;
4495 
4496  score=0;
4497  if(s->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
4498  continue;
4499 
4500  if(new_level){
4501  if(new_level<0) new_coeff= qmul*new_level - qadd;
4502  else new_coeff= qmul*new_level + qadd;
4503  if(new_coeff >= 2048 || new_coeff <= -2048)
4504  continue;
4505  //FIXME check for overflow
4506 
4507  if(level){
4508  if(level < 63 && level > -63){
4509  if(i < last_non_zero)
4510  score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
4511  - length[UNI_AC_ENC_INDEX(run, level+64)];
4512  else
4513  score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
4514  - last_length[UNI_AC_ENC_INDEX(run, level+64)];
4515  }
4516  }else{
4517  av_assert2(FFABS(new_level)==1);
4518 
4519  if(analyze_gradient){
4520  int g= d1[ scantable[i] ];
4521  if(g && (g^new_level) >= 0)
4522  continue;
4523  }
4524 
4525  if(i < last_non_zero){
4526  int next_i= i + run2 + 1;
4527  int next_level= block[ perm_scantable[next_i] ] + 64;
4528 
4529  if(next_level&(~127))
4530  next_level= 0;
4531 
4532  if(next_i < last_non_zero)
4533  score += length[UNI_AC_ENC_INDEX(run, 65)]
4534  + length[UNI_AC_ENC_INDEX(run2, next_level)]
4535  - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
4536  else
4537  score += length[UNI_AC_ENC_INDEX(run, 65)]
4538  + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
4539  - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
4540  }else{
4541  score += last_length[UNI_AC_ENC_INDEX(run, 65)];
4542  if(prev_level){
4543  score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
4544  - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
4545  }
4546  }
4547  }
4548  }else{
4549  new_coeff=0;
4550  av_assert2(FFABS(level)==1);
4551 
4552  if(i < last_non_zero){
4553  int next_i= i + run2 + 1;
4554  int next_level= block[ perm_scantable[next_i] ] + 64;
4555 
4556  if(next_level&(~127))
4557  next_level= 0;
4558 
4559  if(next_i < last_non_zero)
4560  score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
4561  - length[UNI_AC_ENC_INDEX(run2, next_level)]
4562  - length[UNI_AC_ENC_INDEX(run, 65)];
4563  else
4564  score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
4565  - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
4566  - length[UNI_AC_ENC_INDEX(run, 65)];
4567  }else{
4568  score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
4569  if(prev_level){
4570  score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
4571  - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
4572  }
4573  }
4574  }
4575 
4576  score *= lambda;
4577 
4578  unquant_change= new_coeff - old_coeff;
4579  av_assert2((score < 100*lambda && score > -100*lambda) || lambda==0);
4580 
4581  score += s->mpvencdsp.try_8x8basis(rem, weight, basis[j],
4582  unquant_change);
4583  if(score<best_score){
4584  best_score= score;
4585  best_coeff= i;
4586  best_change= change;
4587  best_unquant_change= unquant_change;
4588