1 /*
2 * Interface to xvidcore for mpeg4 encoding
3 * Copyright (c) 2004 Adam Thayer <krevnik@comcast.net>
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 /**
23 * @file
24 * Interface to xvidcore for MPEG-4 compliant encoding.
25 * @author Adam Thayer (krevnik@comcast.net)
26 */
27
28 #include <xvid.h>
29
34
39
40 #if HAVE_UNISTD_H
41 #include <unistd.h>
42 #endif
43
44 #if HAVE_IO_H
45 #include <io.h>
46 #endif
47
48 /**
49 * Buffer management macros.
50 */
51 #define BUFFER_SIZE 1024
52 #define BUFFER_REMAINING(x) (BUFFER_SIZE - strlen(x))
53 #define BUFFER_CAT(x) (&((x)[strlen(x)]))
54
55 /**
56 * Structure for the private Xvid context.
57 * This stores all the private context for the codec.
58 */
67 int qscale;
/**< Do we use constant scale? */
75 int lumi_aq;
/**< Lumi masking as an aq method */
77 int ssim;
/**< SSIM information display mode */
78 int ssim_acc;
/**< SSIM accuracy. 0: accurate. 4: fast. */
80 };
81
82 /**
83 * Structure for the private first-pass plugin.
84 */
88 };
89
91
92 /*
93 * Xvid 2-Pass Kludge Section
94 *
95 * Xvid's default 2-pass doesn't allow us to create data as we need to, so
96 * this section spends time replacing the first pass plugin so we can write
97 * statistic information as libavcodec requests in. We have another kludge
98 * that allows us to pass data to the second pass in Xvid without a custom
99 * rate-control plugin.
100 */
101
102 /**
103 * Initialize the two-pass plugin and context.
104 *
105 * @param param Input construction parameter structure
106 * @param handle Private context handle
107 * @return Returns XVID_ERR_xxxx on failure, or 0 on success.
108 */
110 {
113
114 /* Do a quick bounds check */
116 return XVID_ERR_FAIL;
117
118 /* We use snprintf() */
119 /* This is because we can safely prevent a buffer overflow */
120 log[0] = 0;
122 "# ffmpeg 2-pass log file, using xvid codec\n");
124 "# Do not modify. libxvidcore version: %d.%d.%d\n\n",
125 XVID_VERSION_MAJOR(XVID_VERSION),
126 XVID_VERSION_MINOR(XVID_VERSION),
127 XVID_VERSION_PATCH(XVID_VERSION));
128
130 return 0;
131 }
132
133 /**
134 * Destroy the two-pass plugin context.
135 *
136 * @param ref Context pointer for the plugin
137 * @param param Destrooy context
138 * @return Returns 0, success guaranteed
139 */
141 xvid_plg_destroy_t *param)
142 {
143 /* Currently cannot think of anything to do on destruction */
144 /* Still, the framework should be here for reference/use */
147 return 0;
148 }
149
150 /**
151 * Enable fast encode mode during the first pass.
152 *
153 * @param ref Context pointer for the plugin
154 * @param param Frame data
155 * @return Returns 0, success guaranteed
156 */
158 xvid_plg_data_t *param)
159 {
160 int motion_remove;
161 int motion_replacements;
162 int vop_remove;
163
164 /* Nothing to do here, result is changed too much */
165 if (param->zone && param->zone->mode == XVID_ZONE_QUANT)
166 return 0;
167
168 /* We can implement a 'turbo' first pass mode here */
169 param->quant = 2;
170
171 /* Init values */
172 motion_remove = ~XVID_ME_CHROMA_PVOP &
173 ~XVID_ME_CHROMA_BVOP &
174 ~XVID_ME_EXTSEARCH16 &
175 ~XVID_ME_ADVANCEDDIAMOND16;
176 motion_replacements = XVID_ME_FAST_MODEINTERPOLATE |
177 XVID_ME_SKIP_DELTASEARCH |
178 XVID_ME_FASTREFINE16 |
179 XVID_ME_BFRAME_EARLYSTOP;
180 vop_remove = ~XVID_VOP_MODEDECISION_RD &
181 ~XVID_VOP_FAST_MODEDECISION_RD &
182 ~XVID_VOP_TRELLISQUANT &
183 ~XVID_VOP_INTER4V &
184 ~XVID_VOP_HQACPRED;
185
186 param->vol_flags &= ~XVID_VOL_GMC;
187 param->vop_flags &= vop_remove;
188 param->motion_flags &= motion_remove;
189 param->motion_flags |= motion_replacements;
190
191 return 0;
192 }
193
194 /**
195 * Capture statistic data and write it during first pass.
196 *
197 * @param ref Context pointer for the plugin
198 * @param param Statistic data
199 * @return Returns XVID_ERR_xxxx on failure, or 0 on success
200 */
202 xvid_plg_data_t *param)
203 {
205 const char *frame_types = " ipbs";
206 char frame_type;
207
208 /* Quick bounds check */
209 if (!log)
210 return XVID_ERR_FAIL;
211
212 /* Convert the type given to us into a character */
213 if (param->type < 5 && param->type > 0)
214 frame_type = frame_types[param->type];
215 else
216 return XVID_ERR_FAIL;
217
219 "%c %d %d %d %d %d %d\n",
220 frame_type, param->stats.quant, param->stats.kblks,
221 param->stats.mblks, param->stats.ublks,
222 param->stats.length, param->stats.hlength);
223
224 return 0;
225 }
226
227 /**
228 * Dispatch function for our custom plugin.
229 * This handles the dispatch for the Xvid plugin. It passes data
230 * on to other functions for actual processing.
231 *
232 * @param ref Context pointer for the plugin
233 * @param cmd The task given for us to complete
234 * @param p1 First parameter (varies)
235 * @param p2 Second parameter (varies)
236 * @return Returns XVID_ERR_xxxx on failure, or 0 on success
237 */
239 {
240 switch (cmd) {
241 case XVID_PLG_INFO:
242 case XVID_PLG_FRAME:
243 return 0;
244 case XVID_PLG_BEFORE:
246 case XVID_PLG_CREATE:
248 case XVID_PLG_AFTER:
250 case XVID_PLG_DESTROY:
252 default:
253 return XVID_ERR_FAIL;
254 }
255 }
256
257 /**
258 * Routine to create a global VO/VOL header for MP4 container.
259 * What we do here is extract the header from the Xvid bitstream
260 * as it is encoded. We also strip the repeated headers from the
261 * bitstream when a global header is requested for MPEG-4 ISO
262 * compliance.
263 *
264 * @param avctx AVCodecContext pointer to context
265 * @param frame Pointer to encoded frame data
266 * @param header_len Length of header to search
267 * @param frame_len Length of encoded frame data
268 * @return Returns new length of frame data
269 */
271 unsigned int header_len,
272 unsigned int frame_len)
273 {
274 int vo_len = 0, i;
275
276 for (i = 0; i < header_len - 3; i++) {
277 if (pkt->
data[i] == 0x00 &&
278 pkt->
data[i + 1] == 0x00 &&
279 pkt->
data[i + 2] == 0x01 &&
280 pkt->
data[i + 3] == 0xB6) {
281 vo_len = i;
282 break;
283 }
284 }
285
286 if (vo_len > 0) {
287 /* We need to store the header, so extract it */
294 }
295 /* Less dangerous now, memmove properly copies the two
296 * chunks of overlapping data */
297 memmove(pkt->
data, &pkt->
data[vo_len], frame_len - vo_len);
298 pkt->
size = frame_len - vo_len;
299 }
300 return 0;
301 }
302
303 /**
304 * Routine to correct a possibly erroneous framerate being fed to us.
305 * Xvid currently chokes on framerates where the ticks per frame is
306 * extremely large. This function works to correct problems in this area
307 * by estimating a new framerate and taking the simpler fraction of
308 * the two presented.
309 *
310 * @param avctx Context that contains the framerate to correct.
311 */
313 {
314 int frate, fbase;
315 int est_frate, est_fbase;
316 int gcd;
317 float est_fps, fps;
318
321
322 gcd =
av_gcd(frate, fbase);
323 if (gcd > 1) {
324 frate /= gcd;
325 fbase /= gcd;
326 }
327
328 if (frate <= 65000 && fbase <= 65000) {
331 return;
332 }
333
334 fps = (float) frate / (float) fbase;
335 est_fps =
roundf(fps * 1000.0) / 1000.0;
336
337 est_frate = (int) est_fps;
338 if (est_fps > (int) est_fps) {
339 est_frate = (est_frate + 1) * 1000;
340 est_fbase = (int)
roundf((
float) est_frate / est_fps);
341 } else
342 est_fbase = 1;
343
344 gcd =
av_gcd(est_frate, est_fbase);
345 if (gcd > 1) {
346 est_frate /= gcd;
347 est_fbase /= gcd;
348 }
349
350 if (fbase > est_fbase) {
354 "Xvid: framerate re-estimated: %.2f, %.3f%% correction\n",
355 est_fps, (((est_fps - fps) / fps) * 100.0));
356 } else {
359 }
360 }
361
363 {
364 int xerr, i,
ret = -1;
365 int xvid_flags = avctx->
flags;
367 uint16_t *intra, *inter;
368 int fd;
369
370 xvid_plugin_single_t single = { 0 };
372 xvid_plugin_2pass2_t rc2pass2 = { 0 };
373 xvid_plugin_lumimasking_t masking_l = { 0 }; /* For lumi masking */
374 xvid_plugin_lumimasking_t masking_v = { 0 }; /* For variance AQ */
375 xvid_plugin_ssim_t ssim = { 0 };
376 xvid_gbl_init_t xvid_gbl_init = { 0 };
377 xvid_enc_create_t xvid_enc_create = { 0 };
378 xvid_enc_plugin_t plugins[4];
379
381
382 /* Bring in VOP flags from ffmpeg command-line */
383 x->
vop_flags = XVID_VOP_HALFPEL;
/* Bare minimum quality */
385 x->
vop_flags |= XVID_VOP_INTER4V;
/* Level 3 */
387 x->
vop_flags |= XVID_VOP_TRELLISQUANT;
/* Level 5 */
389 x->
vop_flags |= XVID_VOP_HQACPRED;
/* Level 6 */
392
393 /* Decide which ME quality setting to use */
397 x->
me_flags |= XVID_ME_EXTSEARCH16 |
398 XVID_ME_EXTSEARCH8;
399
401 x->
me_flags |= XVID_ME_ADVANCEDDIAMOND8 |
402 XVID_ME_HALFPELREFINE8 |
403 XVID_ME_CHROMA_PVOP |
404 XVID_ME_CHROMA_BVOP;
405
406 case ME_LOG:
/* Quality 2 */
409 x->
me_flags |= XVID_ME_ADVANCEDDIAMOND16 |
410 XVID_ME_HALFPELREFINE16;
411
413 default:
414 break;
415 }
416
417 /* Decide how we should decide blocks */
419 case 2:
420 x->
vop_flags |= XVID_VOP_MODEDECISION_RD;
421 x->
me_flags |= XVID_ME_HALFPELREFINE8_RD |
422 XVID_ME_QUARTERPELREFINE8_RD |
423 XVID_ME_EXTSEARCH_RD |
424 XVID_ME_CHECKPREDICTION_RD;
425 case 1:
426 if (!(x->
vop_flags & XVID_VOP_MODEDECISION_RD))
427 x->
vop_flags |= XVID_VOP_FAST_MODEDECISION_RD;
428 x->
me_flags |= XVID_ME_HALFPELREFINE16_RD |
429 XVID_ME_QUARTERPELREFINE16_RD;
430 default:
431 break;
432 }
433
434 /* Bring in VOL flags from ffmpeg command-line */
435 #if FF_API_GMC
438 #endif
439
444 }
447 x->
me_flags |= XVID_ME_QUARTERPELREFINE16;
449 x->
me_flags |= XVID_ME_QUARTERPELREFINE8;
450 }
451
452 xvid_gbl_init.version = XVID_VERSION;
453 xvid_gbl_init.debug = 0;
454 xvid_gbl_init.cpu_flags = 0;
455
456 /* Initialize */
457 xvid_global(
NULL, XVID_GBL_INIT, &xvid_gbl_init,
NULL);
458
459 /* Create the encoder reference */
460 xvid_enc_create.version = XVID_VERSION;
461
462 /* Store the desired frame size */
463 xvid_enc_create.width =
465 xvid_enc_create.height =
467
468 /* Xvid can determine the proper profile to use */
469 /* xvid_enc_create.profile = XVID_PROFILE_S_L3; */
470
471 /* We don't use zones */
472 xvid_enc_create.zones =
NULL;
473 xvid_enc_create.num_zones = 0;
474
476 #if (XVID_VERSION <= 0x010303) && (XVID_VERSION >= 0x010300)
477 /* workaround for a bug in libxvidcore */
478 if (avctx->
height <= 16) {
480 xvid_enc_create.num_threads = 0;
481 } else {
483 "Too small height for threads > 1.");
485 goto fail;
486 }
487 }
488 #endif
489
490 xvid_enc_create.plugins = plugins;
491 xvid_enc_create.num_plugins = 0;
492
493 /* Initialize Buffers */
497
499 rc2pass1.
version = XVID_VERSION;
505 "Xvid: Cannot allocate 2-pass log buffers\n");
507 goto fail;
508 }
511
513 plugins[xvid_enc_create.num_plugins].param = &rc2pass1;
514 xvid_enc_create.num_plugins++;
516 rc2pass2.
version = XVID_VERSION;
518
520 if (fd < 0) {
522 ret = fd;
523 goto fail;
524 }
526
529 "Xvid: No 2-pass information loaded for second pass\n");
531 goto fail;
532 }
533
535 if (ret == -1)
537 else if (strlen(avctx->
stats_in) > ret) {
540 }
541 if (ret < 0)
542 goto fail;
543
545 plugins[xvid_enc_create.num_plugins].func = xvid_plugin_2pass2;
546 plugins[xvid_enc_create.num_plugins].param = &rc2pass2;
547 xvid_enc_create.num_plugins++;
549 /* Single Pass Bitrate Control! */
550 single.version = XVID_VERSION;
552
553 plugins[xvid_enc_create.num_plugins].func = xvid_plugin_single;
554 plugins[xvid_enc_create.num_plugins].param = &single;
555 xvid_enc_create.num_plugins++;
556 }
557
560
561 /* Luminance Masking */
563 masking_l.method = 0;
564 plugins[xvid_enc_create.num_plugins].func = xvid_plugin_lumimasking;
565
566 /* The old behavior is that when avctx->lumi_masking is specified,
567 * plugins[...].param = NULL. Trying to keep the old behavior here. */
568 plugins[xvid_enc_create.num_plugins].param =
570 xvid_enc_create.num_plugins++;
571 }
572
573 /* Variance AQ */
575 masking_v.method = 1;
576 plugins[xvid_enc_create.num_plugins].func = xvid_plugin_lumimasking;
577 plugins[xvid_enc_create.num_plugins].param = &masking_v;
578 xvid_enc_create.num_plugins++;
579 }
580
583 "Both lumi_aq and variance_aq are enabled. The resulting quality"
584 "will be the worse one of the two effects made by the AQ.\n");
585
586 /* SSIM */
588 plugins[xvid_enc_create.num_plugins].func = xvid_plugin_ssim;
589 ssim.b_printstat = x->
ssim == 2;
591 ssim.cpu_flags = xvid_gbl_init.cpu_flags;
592 ssim.b_visualize = 0;
593 plugins[xvid_enc_create.num_plugins].param = &ssim;
594 xvid_enc_create.num_plugins++;
595 }
596
597 /* Frame Rate and Key Frames */
602 xvid_enc_create.max_key_interval = avctx->
gop_size;
603 else
604 xvid_enc_create.max_key_interval = 240; /* Xvid's best default */
605
606 /* Quants */
607 if (xvid_flags & CODEC_FLAG_QSCALE)
609 else
611
612 xvid_enc_create.min_quant[0] = avctx->
qmin;
613 xvid_enc_create.min_quant[1] = avctx->
qmin;
614 xvid_enc_create.min_quant[2] = avctx->
qmin;
615 xvid_enc_create.max_quant[0] = avctx->
qmax;
616 xvid_enc_create.max_quant[1] = avctx->
qmax;
617 xvid_enc_create.max_quant[2] = avctx->
qmax;
618
619 /* Quant Matrices */
626
632 goto fail;
633 }
634 } else
641 goto fail;
642 }
643 } else
645
646 for (i = 0; i < 64; i++) {
647 if (intra)
649 if (inter)
651 }
652 }
653
654 /* Misc Settings */
655 xvid_enc_create.frame_drop_ratio = 0;
656 xvid_enc_create.global = 0;
658 xvid_enc_create.global |= XVID_GLOBAL_CLOSED_GOP;
659
660 /* Determines which codec mode we are operating in */
664 /* In this case, we are claiming to be MPEG4 */
667 } else {
668 /* We are claiming to be Xvid */
672 }
673
674 /* Bframes */
679 xvid_enc_create.global |= XVID_GLOBAL_PACKED;
680
682
683 /* Create encoder context */
684 xerr = xvid_encore(
NULL, XVID_ENC_CREATE, &xvid_enc_create,
NULL);
685 if (xerr) {
687 goto fail;
688 }
689
694 goto fail;
695 }
696
697 return 0;
698 fail:
701 }
702
704 const AVFrame *picture,
int *got_packet)
705 {
706 int xerr, i,
ret, user_packet = !!pkt->
data;
709 int mb_width = (avctx->
width + 15) / 16;
710 int mb_height = (avctx->
height + 15) / 16;
711 char *tmp;
712
713 xvid_enc_frame_t xvid_enc_frame = { 0 };
714 xvid_enc_stats_t xvid_enc_stats = { 0 };
715
718
719 /* Start setting up the frame */
720 xvid_enc_frame.version = XVID_VERSION;
721 xvid_enc_stats.version = XVID_VERSION;
722
723 /* Let Xvid know where to put the frame. */
724 xvid_enc_frame.bitstream = pkt->
data;
725 xvid_enc_frame.length = pkt->
size;
726
727 /* Initialize input image fields */
730 "Xvid: Color spaces other than 420P not supported\n");
732 }
733
734 xvid_enc_frame.input.csp = XVID_CSP_PLANAR; /* YUV420P */
735
736 for (i = 0; i < 4; i++) {
737 xvid_enc_frame.input.plane[i] = picture->
data[i];
738 xvid_enc_frame.input.stride[i] = picture->
linesize[i];
739 }
740
741 /* Encoder Flags */
744 xvid_enc_frame.motion = x->
me_flags;
745 xvid_enc_frame.type =
749 XVID_TYPE_AUTO;
750
751 /* Pixel aspect ratio setting */
755 "Invalid pixel aspect ratio %i/%i, limit is 255/255 reducing\n",
759 }
760 xvid_enc_frame.par = XVID_PAR_EXT;
763
764 /* Quant Setting */
767 else
768 xvid_enc_frame.quant = 0;
769
770 /* Matrices */
773
774 /* Encode */
776 &xvid_enc_frame, &xvid_enc_stats);
777
778 /* Two-pass log buffer swapping */
787 }
788 }
789
790 if (xerr > 0) {
791 *got_packet = 1;
792
794 if (xvid_enc_stats.type == XVID_TYPE_PVOP)
796 else if (xvid_enc_stats.type == XVID_TYPE_BVOP)
798 else if (xvid_enc_stats.type == XVID_TYPE_SVOP)
800 else
802 if (xvid_enc_frame.out_flags & XVID_KEYFRAME) {
807 xvid_enc_stats.hlength, xerr);
808 } else
810
812
813 return 0;
814 } else {
815 if (!user_packet)
817 if (!xerr)
818 return 0;
820 "Xvid: Encoding Error Occurred: %i\n", xerr);
822 }
823 }
824
826 {
828
832
838 }
843 }
848
849 return 0;
850 }
851
852 #define OFFSET(x) offsetof(struct xvid_context, x)
853 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
864 };
865
871 };
872
884 };