FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
libxavs.c
Go to the documentation of this file.
1 /*
2  * AVS encoding using the xavs library
3  * Copyright (C) 2010 Amanda, Y.N. Wu <amanda11192003@gmail.com>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include <stdio.h>
23 #include <stdlib.h>
24 #include <string.h>
25 #include <math.h>
26 #include <stdint.h>
27 #include <float.h>
28 #include <xavs.h>
29 #include "avcodec.h"
30 #include "internal.h"
31 #include "libavutil/internal.h"
32 #include "libavutil/mem.h"
33 #include "libavutil/opt.h"
34 
35 #define END_OF_STREAM 0x001
36 
37 #define XAVS_PART_I8X8 0x002 /* Analyze i8x8 (requires 8x8 transform) */
38 #define XAVS_PART_P8X8 0x010 /* Analyze p16x8, p8x16 and p8x8 */
39 #define XAVS_PART_B8X8 0x100 /* Analyze b16x8, b*/
40 
41 typedef struct XavsContext {
42  AVClass *class;
43  xavs_param_t params;
44  xavs_t *enc;
45  xavs_picture_t pic;
47  int sei_size;
50  float crf;
51  int cqp;
52  int b_bias;
53  float cplxblur;
55  int aud;
57  int mbtree;
59 
60  int64_t *pts_buffer;
62 } XavsContext;
63 
64 static void XAVS_log(void *p, int level, const char *fmt, va_list args)
65 {
66  static const int level_map[] = {
67  [XAVS_LOG_ERROR] = AV_LOG_ERROR,
68  [XAVS_LOG_WARNING] = AV_LOG_WARNING,
69  [XAVS_LOG_INFO] = AV_LOG_INFO,
70  [XAVS_LOG_DEBUG] = AV_LOG_DEBUG
71  };
72 
73  if (level < 0 || level > XAVS_LOG_DEBUG)
74  return;
75 
76  av_vlog(p, level_map[level], fmt, args);
77 }
78 
80  xavs_nal_t *nals, int nnal)
81 {
82  XavsContext *x4 = ctx->priv_data;
83  uint8_t *p;
84  int i, s, ret, size = x4->sei_size + FF_MIN_BUFFER_SIZE;
85 
86  if (!nnal)
87  return 0;
88 
89  for (i = 0; i < nnal; i++)
90  size += nals[i].i_payload;
91 
92  if ((ret = ff_alloc_packet2(ctx, pkt, size)) < 0)
93  return ret;
94  p = pkt->data;
95 
96  /* Write the SEI as part of the first frame. */
97  if (x4->sei_size > 0 && nnal > 0) {
98  memcpy(p, x4->sei, x4->sei_size);
99  p += x4->sei_size;
100  x4->sei_size = 0;
101  }
102 
103  for (i = 0; i < nnal; i++) {
104  s = xavs_nal_encode(p, &size, 1, nals + i);
105  if (s < 0)
106  return -1;
107  p += s;
108  }
109  pkt->size = p - pkt->data;
110 
111  return 1;
112 }
113 
115  const AVFrame *frame, int *got_packet)
116 {
117  XavsContext *x4 = ctx->priv_data;
118  xavs_nal_t *nal;
119  int nnal, i, ret;
120  xavs_picture_t pic_out;
121 
122  x4->pic.img.i_csp = XAVS_CSP_I420;
123  x4->pic.img.i_plane = 3;
124 
125  if (frame) {
126  for (i = 0; i < 3; i++) {
127  x4->pic.img.plane[i] = frame->data[i];
128  x4->pic.img.i_stride[i] = frame->linesize[i];
129  }
130 
131  x4->pic.i_pts = frame->pts;
132  x4->pic.i_type = XAVS_TYPE_AUTO;
133  x4->pts_buffer[ctx->frame_number % (ctx->max_b_frames+1)] = frame->pts;
134  }
135 
136  if (xavs_encoder_encode(x4->enc, &nal, &nnal,
137  frame? &x4->pic: NULL, &pic_out) < 0)
138  return -1;
139 
140  ret = encode_nals(ctx, pkt, nal, nnal);
141 
142  if (ret < 0)
143  return -1;
144 
145  if (!ret) {
146  if (!frame && !(x4->end_of_stream)) {
147  if ((ret = ff_alloc_packet2(ctx, pkt, 4)) < 0)
148  return ret;
149 
150  pkt->data[0] = 0x0;
151  pkt->data[1] = 0x0;
152  pkt->data[2] = 0x01;
153  pkt->data[3] = 0xb1;
154  pkt->dts = 2*x4->pts_buffer[(x4->out_frame_count-1)%(ctx->max_b_frames+1)] -
155  x4->pts_buffer[(x4->out_frame_count-2)%(ctx->max_b_frames+1)];
157  *got_packet = 1;
158  }
159  return 0;
160  }
161 
162  x4->out_pic.pts = pic_out.i_pts;
163  pkt->pts = pic_out.i_pts;
164  if (ctx->has_b_frames) {
165  if (!x4->out_frame_count)
166  pkt->dts = pkt->pts - (x4->pts_buffer[1] - x4->pts_buffer[0]);
167  else
168  pkt->dts = x4->pts_buffer[(x4->out_frame_count-1)%(ctx->max_b_frames+1)];
169  } else
170  pkt->dts = pkt->pts;
171 
172  switch (pic_out.i_type) {
173  case XAVS_TYPE_IDR:
174  case XAVS_TYPE_I:
176  break;
177  case XAVS_TYPE_P:
179  break;
180  case XAVS_TYPE_B:
181  case XAVS_TYPE_BREF:
183  break;
184  }
185 
186  /* There is no IDR frame in AVS JiZhun */
187  /* Sequence header is used as a flag */
188  if (pic_out.i_type == XAVS_TYPE_I) {
189  x4->out_pic.key_frame = 1;
190  pkt->flags |= AV_PKT_FLAG_KEY;
191  }
192 
193  x4->out_pic.quality = (pic_out.i_qpplus1 - 1) * FF_QP2LAMBDA;
194 
195  x4->out_frame_count++;
196  *got_packet = ret;
197  return 0;
198 }
199 
201 {
202  XavsContext *x4 = avctx->priv_data;
203 
204  av_freep(&avctx->extradata);
205  av_free(x4->sei);
206  av_freep(&x4->pts_buffer);
207 
208  if (x4->enc)
209  xavs_encoder_close(x4->enc);
210 
211  return 0;
212 }
213 
214 static av_cold int XAVS_init(AVCodecContext *avctx)
215 {
216  XavsContext *x4 = avctx->priv_data;
217 
218  x4->sei_size = 0;
219  xavs_param_default(&x4->params);
220 
221  x4->params.pf_log = XAVS_log;
222  x4->params.p_log_private = avctx;
223  x4->params.i_keyint_max = avctx->gop_size;
224  if (avctx->bit_rate) {
225  x4->params.rc.i_bitrate = avctx->bit_rate / 1000;
226  x4->params.rc.i_rc_method = XAVS_RC_ABR;
227  }
228  x4->params.rc.i_vbv_buffer_size = avctx->rc_buffer_size / 1000;
229  x4->params.rc.i_vbv_max_bitrate = avctx->rc_max_rate / 1000;
230  x4->params.rc.b_stat_write = avctx->flags & CODEC_FLAG_PASS1;
231  if (avctx->flags & CODEC_FLAG_PASS2) {
232  x4->params.rc.b_stat_read = 1;
233  } else {
234  if (x4->crf >= 0) {
235  x4->params.rc.i_rc_method = XAVS_RC_CRF;
236  x4->params.rc.f_rf_constant = x4->crf;
237  } else if (x4->cqp >= 0) {
238  x4->params.rc.i_rc_method = XAVS_RC_CQP;
239  x4->params.rc.i_qp_constant = x4->cqp;
240  }
241  }
242 
243  if (x4->aud >= 0)
244  x4->params.b_aud = x4->aud;
245  if (x4->mbtree >= 0)
246  x4->params.rc.b_mb_tree = x4->mbtree;
247  if (x4->direct_pred >= 0)
248  x4->params.analyse.i_direct_mv_pred = x4->direct_pred;
249  if (x4->fast_pskip >= 0)
250  x4->params.analyse.b_fast_pskip = x4->fast_pskip;
251  if (x4->mixed_refs >= 0)
252  x4->params.analyse.b_mixed_references = x4->mixed_refs;
253  if (x4->b_bias != INT_MIN)
254  x4->params.i_bframe_bias = x4->b_bias;
255  if (x4->cplxblur >= 0)
256  x4->params.rc.f_complexity_blur = x4->cplxblur;
257 
258  x4->params.i_bframe = avctx->max_b_frames;
259  /* cabac is not included in AVS JiZhun Profile */
260  x4->params.b_cabac = 0;
261 
262  x4->params.i_bframe_adaptive = avctx->b_frame_strategy;
263 
264  avctx->has_b_frames = !!avctx->max_b_frames;
265 
266  /* AVS doesn't allow B picture as reference */
267  /* The max allowed reference frame number of B is 2 */
268  x4->params.i_keyint_min = avctx->keyint_min;
269  if (x4->params.i_keyint_min > x4->params.i_keyint_max)
270  x4->params.i_keyint_min = x4->params.i_keyint_max;
271 
272  x4->params.i_scenecut_threshold = avctx->scenechange_threshold;
273 
274  // x4->params.b_deblocking_filter = avctx->flags & CODEC_FLAG_LOOP_FILTER;
275 
276  x4->params.rc.i_qp_min = avctx->qmin;
277  x4->params.rc.i_qp_max = avctx->qmax;
278  x4->params.rc.i_qp_step = avctx->max_qdiff;
279 
280  x4->params.rc.f_qcompress = avctx->qcompress; /* 0.0 => cbr, 1.0 => constant qp */
281  x4->params.rc.f_qblur = avctx->qblur; /* temporally blur quants */
282 
283  x4->params.i_frame_reference = avctx->refs;
284 
285  x4->params.i_width = avctx->width;
286  x4->params.i_height = avctx->height;
287  x4->params.vui.i_sar_width = avctx->sample_aspect_ratio.num;
288  x4->params.vui.i_sar_height = avctx->sample_aspect_ratio.den;
289  /* This is only used for counting the fps */
290  x4->params.i_fps_num = avctx->time_base.den;
291  x4->params.i_fps_den = avctx->time_base.num;
292  x4->params.analyse.inter = XAVS_ANALYSE_I8x8 |XAVS_ANALYSE_PSUB16x16| XAVS_ANALYSE_BSUB16x16;
293 
294  switch (avctx->me_method) {
295  case ME_EPZS:
296  x4->params.analyse.i_me_method = XAVS_ME_DIA;
297  break;
298  case ME_HEX:
299  x4->params.analyse.i_me_method = XAVS_ME_HEX;
300  break;
301  case ME_UMH:
302  x4->params.analyse.i_me_method = XAVS_ME_UMH;
303  break;
304  case ME_FULL:
305  x4->params.analyse.i_me_method = XAVS_ME_ESA;
306  break;
307  case ME_TESA:
308  x4->params.analyse.i_me_method = XAVS_ME_TESA;
309  break;
310  default:
311  x4->params.analyse.i_me_method = XAVS_ME_HEX;
312  }
313 
314  x4->params.analyse.i_me_range = avctx->me_range;
315  x4->params.analyse.i_subpel_refine = avctx->me_subpel_quality;
316 
317  x4->params.analyse.b_chroma_me = avctx->me_cmp & FF_CMP_CHROMA;
318  /* AVS P2 only enables 8x8 transform */
319  x4->params.analyse.b_transform_8x8 = 1; //avctx->flags2 & CODEC_FLAG2_8X8DCT;
320 
321  x4->params.analyse.i_trellis = avctx->trellis;
322  x4->params.analyse.i_noise_reduction = avctx->noise_reduction;
323 
324  if (avctx->level > 0)
325  x4->params.i_level_idc = avctx->level;
326 
327  x4->params.rc.f_rate_tolerance =
328  (float)avctx->bit_rate_tolerance/avctx->bit_rate;
329 
330  if ((avctx->rc_buffer_size) &&
331  (avctx->rc_initial_buffer_occupancy <= avctx->rc_buffer_size)) {
332  x4->params.rc.f_vbv_buffer_init =
333  (float)avctx->rc_initial_buffer_occupancy / avctx->rc_buffer_size;
334  } else
335  x4->params.rc.f_vbv_buffer_init = 0.9;
336 
337  /* TAG:do we have MB tree RC method */
338  /* what is the RC method we are now using? Default NO */
339  x4->params.rc.f_ip_factor = 1 / fabs(avctx->i_quant_factor);
340  x4->params.rc.f_pb_factor = avctx->b_quant_factor;
341  x4->params.analyse.i_chroma_qp_offset = avctx->chromaoffset;
342 
343  x4->params.analyse.b_psnr = avctx->flags & CODEC_FLAG_PSNR;
344  x4->params.i_log_level = XAVS_LOG_DEBUG;
345  x4->params.i_threads = avctx->thread_count;
346  x4->params.b_interlaced = avctx->flags & CODEC_FLAG_INTERLACED_DCT;
347 
348  if (avctx->flags & CODEC_FLAG_GLOBAL_HEADER)
349  x4->params.b_repeat_headers = 0;
350 
351  x4->enc = xavs_encoder_open(&x4->params);
352  if (!x4->enc)
353  return -1;
354 
355  if (!(x4->pts_buffer = av_mallocz((avctx->max_b_frames+1) * sizeof(*x4->pts_buffer))))
356  return AVERROR(ENOMEM);
357 
358  avctx->coded_frame = &x4->out_pic;
359  /* TAG: Do we have GLOBAL HEADER in AVS */
360  /* We Have PPS and SPS in AVS */
361  if (avctx->flags & CODEC_FLAG_GLOBAL_HEADER) {
362  xavs_nal_t *nal;
363  int nnal, s, i, size;
364  uint8_t *p;
365 
366  s = xavs_encoder_headers(x4->enc, &nal, &nnal);
367 
368  avctx->extradata = p = av_malloc(s);
369  for (i = 0; i < nnal; i++) {
370  /* Don't put the SEI in extradata. */
371  if (nal[i].i_type == NAL_SEI) {
372  x4->sei = av_malloc( 5 + nal[i].i_payload * 4 / 3 );
373  if (xavs_nal_encode(x4->sei, &x4->sei_size, 1, nal + i) < 0)
374  return -1;
375 
376  continue;
377  }
378  size = xavs_nal_encode(p, &s, 1, nal + i);
379  if (size < 0)
380  return -1;
381  p += size;
382  }
383  avctx->extradata_size = p - avctx->extradata;
384  }
385  return 0;
386 }
387 
388 #define OFFSET(x) offsetof(XavsContext, x)
389 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
390 static const AVOption options[] = {
391  { "crf", "Select the quality for constant quality mode", OFFSET(crf), AV_OPT_TYPE_FLOAT, {.dbl = -1 }, -1, FLT_MAX, VE },
392  { "qp", "Constant quantization parameter rate control method",OFFSET(cqp), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, INT_MAX, VE },
393  { "b-bias", "Influences how often B-frames are used", OFFSET(b_bias), AV_OPT_TYPE_INT, {.i64 = INT_MIN}, INT_MIN, INT_MAX, VE },
394  { "cplxblur", "Reduce fluctuations in QP (before curve compression)", OFFSET(cplxblur), AV_OPT_TYPE_FLOAT, {.dbl = -1 }, -1, FLT_MAX, VE},
395  { "direct-pred", "Direct MV prediction mode", OFFSET(direct_pred), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, INT_MAX, VE, "direct-pred" },
396  { "none", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = XAVS_DIRECT_PRED_NONE }, 0, 0, VE, "direct-pred" },
397  { "spatial", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = XAVS_DIRECT_PRED_SPATIAL }, 0, 0, VE, "direct-pred" },
398  { "temporal", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = XAVS_DIRECT_PRED_TEMPORAL }, 0, 0, VE, "direct-pred" },
399  { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = XAVS_DIRECT_PRED_AUTO }, 0, 0, VE, "direct-pred" },
400  { "aud", "Use access unit delimiters.", OFFSET(aud), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, 1, VE},
401  { "mbtree", "Use macroblock tree ratecontrol.", OFFSET(mbtree), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, 1, VE},
402  { "mixed-refs", "One reference per partition, as opposed to one reference per macroblock", OFFSET(mixed_refs), AV_OPT_TYPE_INT, {.i64 = -1}, -1, 1, VE },
403  { "fast-pskip", NULL, OFFSET(fast_pskip), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, 1, VE},
404  { NULL },
405 };
406 
407 static const AVClass class = {
408  .class_name = "libxavs",
409  .item_name = av_default_item_name,
410  .option = options,
412 };
413 
414 static const AVCodecDefault xavs_defaults[] = {
415  { "b", "0" },
416  { NULL },
417 };
418 
420  .name = "libxavs",
421  .type = AVMEDIA_TYPE_VIDEO,
422  .id = AV_CODEC_ID_CAVS,
423  .priv_data_size = sizeof(XavsContext),
424  .init = XAVS_init,
425  .encode2 = XAVS_frame,
426  .close = XAVS_close,
427  .capabilities = CODEC_CAP_DELAY | CODEC_CAP_AUTO_THREADS,
428  .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
429  .long_name = NULL_IF_CONFIG_SMALL("libxavs Chinese AVS (Audio Video Standard)"),
430  .priv_class = &class,
431  .defaults = xavs_defaults,
432 };