aarch64: Add assembly support for -fsanitize=hwaddress tagged globals.
[libav.git] / libavcodec / ffv1enc.c
1 /*
2 * FFV1 encoder for libavcodec
3 *
4 * Copyright (c) 2003-2012 Michael Niedermayer <michaelni@gmx.at>
5 *
6 * This file is part of Libav.
7 *
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 /**
24 * @file
25 * FF Video Codec 1 (a lossless codec) encoder
26 */
27
28 #include "libavutil/attributes.h"
29 #include "libavutil/avassert.h"
30 #include "libavutil/pixdesc.h"
31 #include "libavutil/crc.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/imgutils.h"
34
35 #include "avcodec.h"
36 #include "golomb.h"
37 #include "internal.h"
38 #include "put_bits.h"
39 #include "rangecoder.h"
40 #include "mathops.h"
41 #include "ffv1.h"
42
43 static void find_best_state(uint8_t best_state[256][256],
44 const uint8_t one_state[256])
45 {
46 int i, j, k, m;
47 double l2tab[256];
48
49 for (i = 1; i < 256; i++)
50 l2tab[i] = log2(i / 256.0);
51
52 for (i = 0; i < 256; i++) {
53 double best_len[256];
54 double p = i / 256.0;
55
56 for (j = 0; j < 256; j++)
57 best_len[j] = 1 << 30;
58
59 for (j = FFMAX(i - 10, 1); j < FFMIN(i + 11, 256); j++) {
60 double occ[256] = { 0 };
61 double len = 0;
62 occ[j] = 1.0;
63 for (k = 0; k < 256; k++) {
64 double newocc[256] = { 0 };
65 for (m = 1; m < 256; m++)
66 if (occ[m]) {
67 len -= occ[m] * (p * l2tab[m] +
68 (1 - p) * l2tab[256 - m]);
69 }
70 if (len < best_len[k]) {
71 best_len[k] = len;
72 best_state[i][k] = j;
73 }
74 for (m = 1; m < 256; m++)
75 if (occ[m]) {
76 newocc[one_state[m]] += occ[m] * p;
77 newocc[256 - one_state[256 - m]] += occ[m] * (1 - p);
78 }
79 memcpy(occ, newocc, sizeof(occ));
80 }
81 }
82 }
83 }
84
85 static av_always_inline av_flatten void put_symbol_inline(RangeCoder *c,
86 uint8_t *state, int v,
87 int is_signed,
88 uint64_t rc_stat[256][2],
89 uint64_t rc_stat2[32][2])
90 {
91 int i;
92
93 #define put_rac(C, S, B) \
94 do { \
95 if (rc_stat) { \
96 rc_stat[*(S)][B]++; \
97 rc_stat2[(S) - state][B]++; \
98 } \
99 put_rac(C, S, B); \
100 } while (0)
101
102 if (v) {
103 const int a = FFABS(v);
104 const int e = av_log2(a);
105 put_rac(c, state + 0, 0);
106 if (e <= 9) {
107 for (i = 0; i < e; i++)
108 put_rac(c, state + 1 + i, 1); // 1..10
109 put_rac(c, state + 1 + i, 0);
110
111 for (i = e - 1; i >= 0; i--)
112 put_rac(c, state + 22 + i, (a >> i) & 1); // 22..31
113
114 if (is_signed)
115 put_rac(c, state + 11 + e, v < 0); // 11..21
116 } else {
117 for (i = 0; i < e; i++)
118 put_rac(c, state + 1 + FFMIN(i, 9), 1); // 1..10
119 put_rac(c, state + 1 + 9, 0);
120
121 for (i = e - 1; i >= 0; i--)
122 put_rac(c, state + 22 + FFMIN(i, 9), (a >> i) & 1); // 22..31
123
124 if (is_signed)
125 put_rac(c, state + 11 + 10, v < 0); // 11..21
126 }
127 } else {
128 put_rac(c, state + 0, 1);
129 }
130 #undef put_rac
131 }
132
133 static av_noinline void put_symbol(RangeCoder *c, uint8_t *state,
134 int v, int is_signed)
135 {
136 put_symbol_inline(c, state, v, is_signed, NULL, NULL);
137 }
138
139 static inline void put_vlc_symbol(PutBitContext *pb, VlcState *const state,
140 int v, int bits)
141 {
142 int i, k, code;
143 v = fold(v - state->bias, bits);
144
145 i = state->count;
146 k = 0;
147 while (i < state->error_sum) { // FIXME: optimize
148 k++;
149 i += i;
150 }
151
152 assert(k <= 13);
153
154 code = v ^ ((2 * state->drift + state->count) >> 31);
155
156 ff_dlog(NULL, "v:%d/%d bias:%d error:%d drift:%d count:%d k:%d\n", v, code,
157 state->bias, state->error_sum, state->drift, state->count, k);
158 set_sr_golomb(pb, code, k, 12, bits);
159
160 update_vlc_state(state, v);
161 }
162
163 static av_always_inline int encode_line(FFV1Context *s, int w,
164 int16_t *sample[3],
165 int plane_index, int bits)
166 {
167 PlaneContext *const p = &s->plane[plane_index];
168 RangeCoder *const c = &s->c;
169 int x;
170 int run_index = s->run_index;
171 int run_count = 0;
172 int run_mode = 0;
173
174 if (s->ac != AC_GOLOMB_RICE) {
175 if (c->bytestream_end - c->bytestream < w * 20) {
176 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
177 return AVERROR_INVALIDDATA;
178 }
179 } else {
180 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < w * 4) {
181 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
182 return AVERROR_INVALIDDATA;
183 }
184 }
185
186 for (x = 0; x < w; x++) {
187 int diff, context;
188
189 context = get_context(p, sample[0] + x, sample[1] + x, sample[2] + x);
190 diff = sample[0][x] - predict(sample[0] + x, sample[1] + x);
191
192 if (context < 0) {
193 context = -context;
194 diff = -diff;
195 }
196
197 diff = fold(diff, bits);
198
199 if (s->ac != AC_GOLOMB_RICE) {
200 if (s->flags & AV_CODEC_FLAG_PASS1) {
201 put_symbol_inline(c, p->state[context], diff, 1, s->rc_stat,
202 s->rc_stat2[p->quant_table_index][context]);
203 } else {
204 put_symbol_inline(c, p->state[context], diff, 1, NULL, NULL);
205 }
206 } else {
207 if (context == 0)
208 run_mode = 1;
209
210 if (run_mode) {
211 if (diff) {
212 while (run_count >= 1 << ff_log2_run[run_index]) {
213 run_count -= 1 << ff_log2_run[run_index];
214 run_index++;
215 put_bits(&s->pb, 1, 1);
216 }
217
218 put_bits(&s->pb, 1 + ff_log2_run[run_index], run_count);
219 if (run_index)
220 run_index--;
221 run_count = 0;
222 run_mode = 0;
223 if (diff > 0)
224 diff--;
225 } else {
226 run_count++;
227 }
228 }
229
230 ff_dlog(s->avctx, "count:%d index:%d, mode:%d, x:%d pos:%d\n",
231 run_count, run_index, run_mode, x,
232 (int)put_bits_count(&s->pb));
233
234 if (run_mode == 0)
235 put_vlc_symbol(&s->pb, &p->vlc_state[context], diff, bits);
236 }
237 }
238 if (run_mode) {
239 while (run_count >= 1 << ff_log2_run[run_index]) {
240 run_count -= 1 << ff_log2_run[run_index];
241 run_index++;
242 put_bits(&s->pb, 1, 1);
243 }
244
245 if (run_count)
246 put_bits(&s->pb, 1, 1);
247 }
248 s->run_index = run_index;
249
250 return 0;
251 }
252
253 static void encode_plane(FFV1Context *s, uint8_t *src, int w, int h,
254 int stride, int plane_index)
255 {
256 int x, y, i;
257 const int ring_size = s->context_model ? 3 : 2;
258 int16_t *sample[3];
259 s->run_index = 0;
260
261 memset(s->sample_buffer, 0, ring_size * (w + 6) * sizeof(*s->sample_buffer));
262
263 for (y = 0; y < h; y++) {
264 for (i = 0; i < ring_size; i++)
265 sample[i] = s->sample_buffer + (w + 6) * ((h + i - y) % ring_size) + 3;
266
267 sample[0][-1] = sample[1][0];
268 sample[1][w] = sample[1][w - 1];
269 // { START_TIMER
270 if (s->bits_per_raw_sample <= 8) {
271 for (x = 0; x < w; x++)
272 sample[0][x] = src[x + stride * y];
273 encode_line(s, w, sample, plane_index, 8);
274 } else {
275 if (s->packed_at_lsb) {
276 for (x = 0; x < w; x++)
277 sample[0][x] = ((uint16_t *)(src + stride * y))[x];
278 } else {
279 for (x = 0; x < w; x++)
280 sample[0][x] =
281 ((uint16_t *)(src + stride * y))[x] >> (16 - s->bits_per_raw_sample);
282 }
283 encode_line(s, w, sample, plane_index, s->bits_per_raw_sample);
284 }
285 // STOP_TIMER("encode line") }
286 }
287 }
288
289 static void encode_rgb_frame(FFV1Context *s, const uint8_t *src[3],
290 int w, int h, const int stride[3])
291 {
292 int x, y, p, i;
293 const int ring_size = s->context_model ? 3 : 2;
294 int16_t *sample[MAX_PLANES][3];
295 int lbd = s->avctx->bits_per_raw_sample <= 8;
296 int bits = s->avctx->bits_per_raw_sample > 0
297 ? s->avctx->bits_per_raw_sample
298 : 8;
299 int offset = 1 << bits;
300
301 s->run_index = 0;
302
303 memset(s->sample_buffer, 0, ring_size * MAX_PLANES *
304 (w + 6) * sizeof(*s->sample_buffer));
305
306 for (y = 0; y < h; y++) {
307 for (i = 0; i < ring_size; i++)
308 for (p = 0; p < MAX_PLANES; p++)
309 sample[p][i] = s->sample_buffer + p * ring_size *
310 (w + 6) +
311 ((h + i - y) % ring_size) * (w + 6) + 3;
312
313 for (x = 0; x < w; x++) {
314 int b, g, r, av_uninit(a);
315 if (lbd) {
316 unsigned v = *((const uint32_t *)(src[0] + x * 4 + stride[0] * y));
317 b = v & 0xFF;
318 g = (v >> 8) & 0xFF;
319 r = (v >> 16) & 0xFF;
320 a = v >> 24;
321 } else {
322 b = *((const uint16_t *)(src[0] + x * 2 + stride[0] * y));
323 g = *((const uint16_t *)(src[1] + x * 2 + stride[1] * y));
324 r = *((const uint16_t *)(src[2] + x * 2 + stride[2] * y));
325 }
326
327 b -= g;
328 r -= g;
329 g += (b + r) >> 2;
330 b += offset;
331 r += offset;
332
333 sample[0][0][x] = g;
334 sample[1][0][x] = b;
335 sample[2][0][x] = r;
336 sample[3][0][x] = a;
337 }
338 for (p = 0; p < 3 + s->transparency; p++) {
339 sample[p][0][-1] = sample[p][1][0];
340 sample[p][1][w] = sample[p][1][w - 1];
341 if (lbd)
342 encode_line(s, w, sample[p], (p + 1) / 2, 9);
343 else
344 encode_line(s, w, sample[p], (p + 1) / 2, bits + 1);
345 }
346 }
347 }
348
349
350 static void write_quant_table(RangeCoder *c, int16_t *quant_table)
351 {
352 int last = 0;
353 int i;
354 uint8_t state[CONTEXT_SIZE];
355 memset(state, 128, sizeof(state));
356
357 for (i = 1; i < 128; i++)
358 if (quant_table[i] != quant_table[i - 1]) {
359 put_symbol(c, state, i - last - 1, 0);
360 last = i;
361 }
362 put_symbol(c, state, i - last - 1, 0);
363 }
364
365 static void write_quant_tables(RangeCoder *c,
366 int16_t quant_table[MAX_CONTEXT_INPUTS][256])
367 {
368 int i;
369 for (i = 0; i < 5; i++)
370 write_quant_table(c, quant_table[i]);
371 }
372
373 static void write_header(FFV1Context *f)
374 {
375 uint8_t state[CONTEXT_SIZE];
376 int i;
377 RangeCoder *const c = &f->slice_context[0]->c;
378
379 memset(state, 128, sizeof(state));
380
381 if (f->version < 2) {
382 put_symbol(c, state, f->version, 0);
383 put_symbol(c, state, f->ac, 0);
384 if (f->ac == AC_RANGE_CUSTOM_TAB) {
385 for (i = 1; i < 256; i++)
386 put_symbol(c, state,
387 f->state_transition[i] - c->one_state[i], 1);
388 }
389 put_symbol(c, state, f->colorspace, 0); // YUV cs type
390 if (f->version > 0)
391 put_symbol(c, state, f->bits_per_raw_sample, 0);
392 put_rac(c, state, f->chroma_planes);
393 put_symbol(c, state, f->chroma_h_shift, 0);
394 put_symbol(c, state, f->chroma_v_shift, 0);
395 put_rac(c, state, f->transparency);
396
397 write_quant_tables(c, f->quant_table);
398 }
399 }
400
401 static int write_extradata(FFV1Context *f)
402 {
403 RangeCoder *const c = &f->c;
404 uint8_t state[CONTEXT_SIZE];
405 int i, j, k;
406 uint8_t state2[32][CONTEXT_SIZE];
407 unsigned v;
408
409 memset(state2, 128, sizeof(state2));
410 memset(state, 128, sizeof(state));
411
412 f->avctx->extradata_size = 10000 + 4 +
413 (11 * 11 * 5 * 5 * 5 + 11 * 11 * 11) * 32;
414 f->avctx->extradata = av_malloc(f->avctx->extradata_size);
415 ff_init_range_encoder(c, f->avctx->extradata, f->avctx->extradata_size);
416 ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8);
417
418 put_symbol(c, state, f->version, 0);
419 if (f->version > 1) {
420 if (f->version == 3)
421 f->minor_version = 2;
422 put_symbol(c, state, f->minor_version, 0);
423 }
424
425 put_symbol(c, state, f->ac, 0);
426 if (f->ac == AC_RANGE_CUSTOM_TAB)
427 for (i = 1; i < 256; i++)
428 put_symbol(c, state, f->state_transition[i] - c->one_state[i], 1);
429
430 put_symbol(c, state, f->colorspace, 0); // YUV cs type
431 put_symbol(c, state, f->bits_per_raw_sample, 0);
432 put_rac(c, state, f->chroma_planes);
433 put_symbol(c, state, f->chroma_h_shift, 0);
434 put_symbol(c, state, f->chroma_v_shift, 0);
435 put_rac(c, state, f->transparency);
436 put_symbol(c, state, f->num_h_slices - 1, 0);
437 put_symbol(c, state, f->num_v_slices - 1, 0);
438
439 put_symbol(c, state, f->quant_table_count, 0);
440 for (i = 0; i < f->quant_table_count; i++)
441 write_quant_tables(c, f->quant_tables[i]);
442
443 for (i = 0; i < f->quant_table_count; i++) {
444 for (j = 0; j < f->context_count[i] * CONTEXT_SIZE; j++)
445 if (f->initial_states[i] && f->initial_states[i][0][j] != 128)
446 break;
447 if (j < f->context_count[i] * CONTEXT_SIZE) {
448 put_rac(c, state, 1);
449 for (j = 0; j < f->context_count[i]; j++)
450 for (k = 0; k < CONTEXT_SIZE; k++) {
451 int pred = j ? f->initial_states[i][j - 1][k] : 128;
452 put_symbol(c, state2[k],
453 (int8_t)(f->initial_states[i][j][k] - pred), 1);
454 }
455 } else {
456 put_rac(c, state, 0);
457 }
458 }
459
460 if (f->version > 2) {
461 put_symbol(c, state, f->ec, 0);
462 }
463
464 f->avctx->extradata_size = ff_rac_terminate(c);
465
466 v = av_crc(av_crc_get_table(AV_CRC_32_IEEE), 0,
467 f->avctx->extradata, f->avctx->extradata_size);
468 AV_WL32(f->avctx->extradata + f->avctx->extradata_size, v);
469 f->avctx->extradata_size += 4;
470
471 return 0;
472 }
473
474 static int sort_stt(FFV1Context *s, uint8_t stt[256])
475 {
476 int i, i2, changed, print = 0;
477
478 do {
479 changed = 0;
480 for (i = 12; i < 244; i++) {
481 for (i2 = i + 1; i2 < 245 && i2 < i + 4; i2++) {
482
483 #define COST(old, new) \
484 s->rc_stat[old][0] * -log2((256 - (new)) / 256.0) + \
485 s->rc_stat[old][1] * -log2((new) / 256.0)
486
487 #define COST2(old, new) \
488 COST(old, new) + COST(256 - (old), 256 - (new))
489
490 double size0 = COST2(i, i) + COST2(i2, i2);
491 double sizeX = COST2(i, i2) + COST2(i2, i);
492 if (sizeX < size0 && i != 128 && i2 != 128) {
493 int j;
494 FFSWAP(int, stt[i], stt[i2]);
495 FFSWAP(int, s->rc_stat[i][0], s->rc_stat[i2][0]);
496 FFSWAP(int, s->rc_stat[i][1], s->rc_stat[i2][1]);
497 if (i != 256 - i2) {
498 FFSWAP(int, stt[256 - i], stt[256 - i2]);
499 FFSWAP(int, s->rc_stat[256 - i][0], s->rc_stat[256 - i2][0]);
500 FFSWAP(int, s->rc_stat[256 - i][1], s->rc_stat[256 - i2][1]);
501 }
502 for (j = 1; j < 256; j++) {
503 if (stt[j] == i)
504 stt[j] = i2;
505 else if (stt[j] == i2)
506 stt[j] = i;
507 if (i != 256 - i2) {
508 if (stt[256 - j] == 256 - i)
509 stt[256 - j] = 256 - i2;
510 else if (stt[256 - j] == 256 - i2)
511 stt[256 - j] = 256 - i;
512 }
513 }
514 print = changed = 1;
515 }
516 }
517 }
518 } while (changed);
519 return print;
520 }
521
522 static av_cold int init_slices_state(FFV1Context *f)
523 {
524 int i, ret;
525 for (i = 0; i < f->slice_count; i++) {
526 FFV1Context *fs = f->slice_context[i];
527 if ((ret = ffv1_init_slice_state(f, fs)) < 0)
528 return AVERROR(ENOMEM);
529 }
530 return 0;
531 }
532
533 static av_cold int ffv1_encode_init(AVCodecContext *avctx)
534 {
535 FFV1Context *s = avctx->priv_data;
536 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
537 int i, j, k, m, ret;
538
539 ffv1_common_init(avctx);
540
541 s->version = 0;
542
543 switch (avctx->level) {
544 case 3:
545 break;
546 case 2:
547 av_log(avctx, AV_LOG_ERROR,
548 "Version 2 had been deemed non-standard and deprecated "
549 "the support for it had been removed\n");
550 return AVERROR(ENOSYS);
551 case 1:
552 case 0:
553 if (avctx->flags & (AV_CODEC_FLAG_PASS1 | AV_CODEC_FLAG_PASS2)) {
554 av_log(avctx, AV_LOG_ERROR,
555 "Multiple pass encoding requires version 3.\n");
556 return AVERROR(ENOSYS);
557 }
558 if (avctx->slices > 1) {
559 av_log(avctx, AV_LOG_ERROR,
560 "Multiple slices support requires version 3.\n");
561 return AVERROR(ENOSYS);
562 }
563 break;
564 case FF_LEVEL_UNKNOWN:
565 if ((avctx->flags & (AV_CODEC_FLAG_PASS1 | AV_CODEC_FLAG_PASS2)) ||
566 avctx->slices > 1)
567 s->version = 3;
568 else
569 s->version = 0;
570 break;
571 default:
572 av_log(avctx, AV_LOG_ERROR, "Version %d not supported\n",
573 avctx->level);
574 return AVERROR(ENOSYS);
575 }
576
577 if (s->ec < 0) {
578 s->ec = (s->version >= 3);
579 }
580
581 #if FF_API_CODER_TYPE
582 FF_DISABLE_DEPRECATION_WARNINGS
583 if (avctx->coder_type != -1)
584 s->ac = avctx->coder_type > 0 ? AC_RANGE_CUSTOM_TAB : AC_GOLOMB_RICE;
585 FF_ENABLE_DEPRECATION_WARNINGS
586 #endif
587
588 s->plane_count = 3;
589 switch (avctx->pix_fmt) {
590 case AV_PIX_FMT_YUV444P9:
591 case AV_PIX_FMT_YUV422P9:
592 case AV_PIX_FMT_YUV420P9:
593 if (!avctx->bits_per_raw_sample)
594 s->bits_per_raw_sample = 9;
595 case AV_PIX_FMT_YUV444P10:
596 case AV_PIX_FMT_YUV420P10:
597 case AV_PIX_FMT_YUV422P10:
598 s->packed_at_lsb = 1;
599 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample)
600 s->bits_per_raw_sample = 10;
601 case AV_PIX_FMT_GRAY16:
602 case AV_PIX_FMT_YUV444P16:
603 case AV_PIX_FMT_YUV422P16:
604 case AV_PIX_FMT_YUV420P16:
605 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample) {
606 s->bits_per_raw_sample = 16;
607 } else if (!s->bits_per_raw_sample) {
608 s->bits_per_raw_sample = avctx->bits_per_raw_sample;
609 }
610 if (s->bits_per_raw_sample <= 8) {
611 av_log(avctx, AV_LOG_ERROR, "bits_per_raw_sample invalid\n");
612 return AVERROR_INVALIDDATA;
613 }
614 if (s->ac == AC_GOLOMB_RICE) {
615 av_log(avctx, AV_LOG_INFO,
616 "bits_per_raw_sample > 8, forcing range coder\n");
617 s->ac = AC_RANGE_CUSTOM_TAB;
618 }
619 s->version = FFMAX(s->version, 1);
620 case AV_PIX_FMT_GRAY8:
621 case AV_PIX_FMT_YUV444P:
622 case AV_PIX_FMT_YUV440P:
623 case AV_PIX_FMT_YUV422P:
624 case AV_PIX_FMT_YUV420P:
625 case AV_PIX_FMT_YUV411P:
626 case AV_PIX_FMT_YUV410P:
627 s->chroma_planes = desc->nb_components < 3 ? 0 : 1;
628 s->colorspace = 0;
629 break;
630 case AV_PIX_FMT_YUVA444P:
631 case AV_PIX_FMT_YUVA422P:
632 case AV_PIX_FMT_YUVA420P:
633 s->chroma_planes = 1;
634 s->colorspace = 0;
635 s->transparency = 1;
636 break;
637 case AV_PIX_FMT_RGB32:
638 s->colorspace = 1;
639 s->transparency = 1;
640 break;
641 case AV_PIX_FMT_GBRP9:
642 if (!avctx->bits_per_raw_sample)
643 s->bits_per_raw_sample = 9;
644 case AV_PIX_FMT_GBRP10:
645 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample)
646 s->bits_per_raw_sample = 10;
647 case AV_PIX_FMT_GBRP16:
648 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample)
649 s->bits_per_raw_sample = 16;
650 else if (!s->bits_per_raw_sample)
651 s->bits_per_raw_sample = avctx->bits_per_raw_sample;
652 s->colorspace = 1;
653 s->chroma_planes = 1;
654 s->version = FFMAX(s->version, 1);
655 break;
656 default:
657 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
658 return AVERROR_INVALIDDATA;
659 }
660 if (s->transparency) {
661 av_log(
662 avctx, AV_LOG_WARNING,
663 "Storing alpha plane, this will require a recent FFV1 decoder to playback!\n");
664 }
665 #if FF_API_PRIVATE_OPT
666 FF_DISABLE_DEPRECATION_WARNINGS
667 if (avctx->context_model)
668 s->context_model = avctx->context_model;
669 if (avctx->context_model > 1U) {
670 av_log(avctx, AV_LOG_ERROR,
671 "Invalid context model %d, valid values are 0 and 1\n",
672 avctx->context_model);
673 return AVERROR(EINVAL);
674 }
675 FF_ENABLE_DEPRECATION_WARNINGS
676 #endif
677
678 if (s->ac == AC_RANGE_CUSTOM_TAB)
679 for (i = 1; i < 256; i++)
680 s->state_transition[i] = ffv1_ver2_state[i];
681
682 for (i = 0; i < 256; i++) {
683 s->quant_table_count = 2;
684 if (s->bits_per_raw_sample <= 8) {
685 s->quant_tables[0][0][i] = ffv1_quant11[i];
686 s->quant_tables[0][1][i] = ffv1_quant11[i] * 11;
687 s->quant_tables[0][2][i] = ffv1_quant11[i] * 11 * 11;
688 s->quant_tables[1][0][i] = ffv1_quant11[i];
689 s->quant_tables[1][1][i] = ffv1_quant11[i] * 11;
690 s->quant_tables[1][2][i] = ffv1_quant5[i] * 11 * 11;
691 s->quant_tables[1][3][i] = ffv1_quant5[i] * 5 * 11 * 11;
692 s->quant_tables[1][4][i] = ffv1_quant5[i] * 5 * 5 * 11 * 11;
693 } else {
694 s->quant_tables[0][0][i] = ffv1_quant9_10bit[i];
695 s->quant_tables[0][1][i] = ffv1_quant9_10bit[i] * 11;
696 s->quant_tables[0][2][i] = ffv1_quant9_10bit[i] * 11 * 11;
697 s->quant_tables[1][0][i] = ffv1_quant9_10bit[i];
698 s->quant_tables[1][1][i] = ffv1_quant9_10bit[i] * 11;
699 s->quant_tables[1][2][i] = ffv1_quant5_10bit[i] * 11 * 11;
700 s->quant_tables[1][3][i] = ffv1_quant5_10bit[i] * 5 * 11 * 11;
701 s->quant_tables[1][4][i] = ffv1_quant5_10bit[i] * 5 * 5 * 11 * 11;
702 }
703 }
704 s->context_count[0] = (11 * 11 * 11 + 1) / 2;
705 s->context_count[1] = (11 * 11 * 5 * 5 * 5 + 1) / 2;
706 memcpy(s->quant_table, s->quant_tables[s->context_model],
707 sizeof(s->quant_table));
708
709 for (i = 0; i < s->plane_count; i++) {
710 PlaneContext *const p = &s->plane[i];
711
712 memcpy(p->quant_table, s->quant_table, sizeof(p->quant_table));
713 p->quant_table_index = s->context_model;
714 p->context_count = s->context_count[p->quant_table_index];
715 }
716
717 if ((ret = ffv1_allocate_initial_states(s)) < 0)
718 return ret;
719
720 #if FF_API_CODED_FRAME
721 FF_DISABLE_DEPRECATION_WARNINGS
722 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
723 FF_ENABLE_DEPRECATION_WARNINGS
724 #endif
725
726 if (!s->transparency)
727 s->plane_count = 2;
728
729 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift,
730 &s->chroma_v_shift);
731
732 s->picture_number = 0;
733
734 if (avctx->flags & (AV_CODEC_FLAG_PASS1 | AV_CODEC_FLAG_PASS2)) {
735 for (i = 0; i < s->quant_table_count; i++) {
736 s->rc_stat2[i] = av_mallocz(s->context_count[i] *
737 sizeof(*s->rc_stat2[i]));
738 if (!s->rc_stat2[i])
739 return AVERROR(ENOMEM);
740 }
741 }
742 if (avctx->stats_in) {
743 char *p = avctx->stats_in;
744 uint8_t best_state[256][256];
745 int gob_count = 0;
746 char *next;
747
748 av_assert0(s->version > 2);
749
750 for (;; ) {
751 for (j = 0; j < 256; j++)
752 for (i = 0; i < 2; i++) {
753 s->rc_stat[j][i] = strtol(p, &next, 0);
754 if (next == p) {
755 av_log(avctx, AV_LOG_ERROR,
756 "2Pass file invalid at %d %d [%s]\n", j, i, p);
757 return AVERROR_INVALIDDATA;
758 }
759 p = next;
760 }
761 for (i = 0; i < s->quant_table_count; i++)
762 for (j = 0; j < s->context_count[i]; j++) {
763 for (k = 0; k < 32; k++)
764 for (m = 0; m < 2; m++) {
765 s->rc_stat2[i][j][k][m] = strtol(p, &next, 0);
766 if (next == p) {
767 av_log(avctx, AV_LOG_ERROR,
768 "2Pass file invalid at %d %d %d %d [%s]\n",
769 i, j, k, m, p);
770 return AVERROR_INVALIDDATA;
771 }
772 p = next;
773 }
774 }
775 gob_count = strtol(p, &next, 0);
776 if (next == p || gob_count <= 0) {
777 av_log(avctx, AV_LOG_ERROR, "2Pass file invalid\n");
778 return AVERROR_INVALIDDATA;
779 }
780 p = next;
781 while (*p == '\n' || *p == ' ')
782 p++;
783 if (p[0] == 0)
784 break;
785 }
786 sort_stt(s, s->state_transition);
787
788 find_best_state(best_state, s->state_transition);
789
790 for (i = 0; i < s->quant_table_count; i++) {
791 for (j = 0; j < s->context_count[i]; j++)
792 for (k = 0; k < 32; k++) {
793 double p = 128;
794 if (s->rc_stat2[i][j][k][0] + s->rc_stat2[i][j][k][1]) {
795 p = 256.0 * s->rc_stat2[i][j][k][1] /
796 (s->rc_stat2[i][j][k][0] + s->rc_stat2[i][j][k][1]);
797 }
798 s->initial_states[i][j][k] =
799 best_state[av_clip(round(p), 1, 255)][av_clip((s->rc_stat2[i][j][k][0] +
800 s->rc_stat2[i][j][k][1]) /
801 gob_count, 0, 255)];
802 }
803 }
804 }
805
806 if (s->version > 1) {
807 for (s->num_v_slices = 2; s->num_v_slices < 9; s->num_v_slices++)
808 for (s->num_h_slices = s->num_v_slices;
809 s->num_h_slices < 2 * s->num_v_slices; s->num_h_slices++)
810 if (avctx->slices == s->num_h_slices * s->num_v_slices &&
811 avctx->slices <= 64 || !avctx->slices)
812 goto slices_ok;
813 av_log(avctx, AV_LOG_ERROR,
814 "Unsupported number %d of slices requested, please specify a "
815 "supported number with -slices (ex:4,6,9,12,16, ...)\n",
816 avctx->slices);
817 return AVERROR(ENOSYS);
818 slices_ok:
819 write_extradata(s);
820 }
821
822 if ((ret = ffv1_init_slice_contexts(s)) < 0)
823 return ret;
824 if ((ret = init_slices_state(s)) < 0)
825 return ret;
826
827 #define STATS_OUT_SIZE 1024 * 1024 * 6
828 if (avctx->flags & AV_CODEC_FLAG_PASS1) {
829 avctx->stats_out = av_mallocz(STATS_OUT_SIZE);
830 for (i = 0; i < s->quant_table_count; i++)
831 for (j = 0; j < s->slice_count; j++) {
832 FFV1Context *sf = s->slice_context[j];
833 av_assert0(!sf->rc_stat2[i]);
834 sf->rc_stat2[i] = av_mallocz(s->context_count[i] *
835 sizeof(*sf->rc_stat2[i]));
836 if (!sf->rc_stat2[i])
837 return AVERROR(ENOMEM);
838 }
839 }
840
841 return 0;
842 }
843
844 static void encode_slice_header(FFV1Context *f, FFV1Context *fs)
845 {
846 RangeCoder *c = &fs->c;
847 uint8_t state[CONTEXT_SIZE];
848 int j;
849 memset(state, 128, sizeof(state));
850
851 put_symbol(c, state, (fs->slice_x + 1) * f->num_h_slices / f->width, 0);
852 put_symbol(c, state, (fs->slice_y + 1) * f->num_v_slices / f->height, 0);
853 put_symbol(c, state, (fs->slice_width + 1) * f->num_h_slices / f->width - 1,
854 0);
855 put_symbol(c, state,
856 (fs->slice_height + 1) * f->num_v_slices / f->height - 1,
857 0);
858 for (j = 0; j < f->plane_count; j++) {
859 put_symbol(c, state, f->plane[j].quant_table_index, 0);
860 av_assert0(f->plane[j].quant_table_index == f->context_model);
861 }
862 if (!f->frame->interlaced_frame)
863 put_symbol(c, state, 3, 0);
864 else
865 put_symbol(c, state, 1 + !f->frame->top_field_first, 0);
866 put_symbol(c, state, f->frame->sample_aspect_ratio.num, 0);
867 put_symbol(c, state, f->frame->sample_aspect_ratio.den, 0);
868 }
869
870 static int encode_slice(AVCodecContext *c, void *arg)
871 {
872 FFV1Context *fs = *(void **)arg;
873 FFV1Context *f = fs->avctx->priv_data;
874 int width = fs->slice_width;
875 int height = fs->slice_height;
876 int x = fs->slice_x;
877 int y = fs->slice_y;
878 const AVFrame *const p = f->frame;
879 const int ps = (av_pix_fmt_desc_get(c->pix_fmt)->flags & AV_PIX_FMT_FLAG_PLANAR)
880 ? (f->bits_per_raw_sample > 8) + 1
881 : 4;
882
883 if (f->key_frame)
884 ffv1_clear_slice_state(f, fs);
885 if (f->version > 2) {
886 encode_slice_header(f, fs);
887 }
888 if (fs->ac == AC_GOLOMB_RICE) {
889 if (f->version > 2)
890 put_rac(&fs->c, (uint8_t[]) { 129 }, 0);
891 fs->ac_byte_count = f->version > 2 || (!x && !y) ? ff_rac_terminate( &fs->c) : 0;
892 init_put_bits(&fs->pb, fs->c.bytestream_start + fs->ac_byte_count,
893 fs->c.bytestream_end - fs->c.bytestream_start - fs->ac_byte_count);
894 }
895
896 if (f->colorspace == 0) {
897 const int chroma_width = AV_CEIL_RSHIFT(width, f->chroma_h_shift);
898 const int chroma_height = AV_CEIL_RSHIFT(height, f->chroma_v_shift);
899 const int cx = x >> f->chroma_h_shift;
900 const int cy = y >> f->chroma_v_shift;
901
902 encode_plane(fs, p->data[0] + ps * x + y * p->linesize[0],
903 width, height, p->linesize[0], 0);
904
905 if (f->chroma_planes) {
906 encode_plane(fs, p->data[1] + ps * cx + cy * p->linesize[1],
907 chroma_width, chroma_height, p->linesize[1], 1);
908 encode_plane(fs, p->data[2] + ps * cx + cy * p->linesize[2],
909 chroma_width, chroma_height, p->linesize[2], 1);
910 }
911 if (fs->transparency)
912 encode_plane(fs, p->data[3] + ps * x + y * p->linesize[3], width,
913 height, p->linesize[3], 2);
914 } else {
915 const uint8_t *planes[3] = { p->data[0] + ps * x + y * p->linesize[0],
916 p->data[1] + ps * x + y * p->linesize[1],
917 p->data[2] + ps * x + y * p->linesize[2] };
918 encode_rgb_frame(fs, planes, width, height, p->linesize);
919 }
920 emms_c();
921
922 return 0;
923 }
924
925 static int ffv1_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
926 const AVFrame *pict, int *got_packet)
927 {
928 FFV1Context *f = avctx->priv_data;
929 RangeCoder *const c = &f->slice_context[0]->c;
930 int used_count = 0;
931 uint8_t keystate = 128;
932 uint8_t *buf_p;
933 int i, ret;
934
935 f->frame = pict;
936
937 if ((ret = ff_alloc_packet(pkt, avctx->width * avctx->height *
938 ((8 * 2 + 1 + 1) * 4) / 8 +
939 AV_INPUT_BUFFER_MIN_SIZE)) < 0) {
940 av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
941 return ret;
942 }
943
944 ff_init_range_encoder(c, pkt->data, pkt->size);
945 ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8);
946
947 if (avctx->gop_size == 0 || f->picture_number % avctx->gop_size == 0) {
948 put_rac(c, &keystate, 1);
949 f->key_frame = 1;
950 f->gob_count++;
951 write_header(f);
952 } else {
953 put_rac(c, &keystate, 0);
954 f->key_frame = 0;
955 }
956
957 if (f->ac == AC_RANGE_CUSTOM_TAB) {
958 int i;
959 for (i = 1; i < 256; i++) {
960 c->one_state[i] = f->state_transition[i];
961 c->zero_state[256 - i] = 256 - c->one_state[i];
962 }
963 }
964
965 for (i = 1; i < f->slice_count; i++) {
966 FFV1Context *fs = f->slice_context[i];
967 uint8_t *start = pkt->data +
968 (pkt->size - used_count) * (int64_t)i / f->slice_count;
969 int len = pkt->size / f->slice_count;
970 ff_init_range_encoder(&fs->c, start, len);
971 }
972 avctx->execute(avctx, encode_slice, &f->slice_context[0], NULL,
973 f->slice_count, sizeof(void *));
974
975 buf_p = pkt->data;
976 for (i = 0; i < f->slice_count; i++) {
977 FFV1Context *fs = f->slice_context[i];
978 int bytes;
979
980 if (fs->ac != AC_GOLOMB_RICE) {
981 uint8_t state = 129;
982 put_rac(&fs->c, &state, 0);
983 bytes = ff_rac_terminate(&fs->c);
984 } else {
985 flush_put_bits(&fs->pb); // FIXME: nicer padding
986 bytes = fs->ac_byte_count + (put_bits_count(&fs->pb) + 7) / 8;
987 }
988 if (i > 0 || f->version > 2) {
989 av_assert0(bytes < pkt->size / f->slice_count);
990 memmove(buf_p, fs->c.bytestream_start, bytes);
991 av_assert0(bytes < (1 << 24));
992 AV_WB24(buf_p + bytes, bytes);
993 bytes += 3;
994 }
995 if (f->ec) {
996 unsigned v;
997 buf_p[bytes++] = 0;
998 v = av_crc(av_crc_get_table(AV_CRC_32_IEEE), 0, buf_p, bytes);
999 AV_WL32(buf_p + bytes, v);
1000 bytes += 4;
1001 }
1002 buf_p += bytes;
1003 }
1004
1005 if ((avctx->flags & AV_CODEC_FLAG_PASS1) && (f->picture_number & 31) == 0) {
1006 int j, k, m;
1007 char *p = avctx->stats_out;
1008 char *end = p + STATS_OUT_SIZE;
1009
1010 memset(f->rc_stat, 0, sizeof(f->rc_stat));
1011 for (i = 0; i < f->quant_table_count; i++)
1012 memset(f->rc_stat2[i], 0, f->context_count[i] * sizeof(*f->rc_stat2[i]));
1013
1014 for (j = 0; j < f->slice_count; j++) {
1015 FFV1Context *fs = f->slice_context[j];
1016 for (i = 0; i < 256; i++) {
1017 f->rc_stat[i][0] += fs->rc_stat[i][0];
1018 f->rc_stat[i][1] += fs->rc_stat[i][1];
1019 }
1020 for (i = 0; i < f->quant_table_count; i++) {
1021 for (k = 0; k < f->context_count[i]; k++)
1022 for (m = 0; m < 32; m++) {
1023 f->rc_stat2[i][k][m][0] += fs->rc_stat2[i][k][m][0];
1024 f->rc_stat2[i][k][m][1] += fs->rc_stat2[i][k][m][1];
1025 }
1026 }
1027 }
1028
1029 for (j = 0; j < 256; j++) {
1030 snprintf(p, end - p, "%" PRIu64 " %" PRIu64 " ",
1031 f->rc_stat[j][0], f->rc_stat[j][1]);
1032 p += strlen(p);
1033 }
1034 snprintf(p, end - p, "\n");
1035
1036 for (i = 0; i < f->quant_table_count; i++) {
1037 for (j = 0; j < f->context_count[i]; j++)
1038 for (m = 0; m < 32; m++) {
1039 snprintf(p, end - p, "%" PRIu64 " %" PRIu64 " ",
1040 f->rc_stat2[i][j][m][0], f->rc_stat2[i][j][m][1]);
1041 p += strlen(p);
1042 }
1043 }
1044 snprintf(p, end - p, "%d\n", f->gob_count);
1045 } else if (avctx->flags & AV_CODEC_FLAG_PASS1)
1046 avctx->stats_out[0] = '\0';
1047
1048 #if FF_API_CODED_FRAME
1049 FF_DISABLE_DEPRECATION_WARNINGS
1050 avctx->coded_frame->key_frame = f->key_frame;
1051 FF_ENABLE_DEPRECATION_WARNINGS
1052 #endif
1053
1054 f->picture_number++;
1055 pkt->size = buf_p - pkt->data;
1056 pkt->flags |= AV_PKT_FLAG_KEY * f->key_frame;
1057 *got_packet = 1;
1058
1059 return 0;
1060 }
1061
1062 static av_cold int ffv1_encode_close(AVCodecContext *avctx)
1063 {
1064 ffv1_close(avctx);
1065 return 0;
1066 }
1067
1068 #define OFFSET(x) offsetof(FFV1Context, x)
1069 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1070 static const AVOption options[] = {
1071 { "slicecrc", "Protect slices with CRCs", OFFSET(ec), AV_OPT_TYPE_INT,
1072 { .i64 = -1 }, -1, 1, VE },
1073 { "coder", "Coder type", OFFSET(ac), AV_OPT_TYPE_INT,
1074 { .i64 = AC_GOLOMB_RICE }, 0, 2, VE, "coder" },
1075 { "rice", "Golomb rice", 0, AV_OPT_TYPE_CONST,
1076 { .i64 = AC_GOLOMB_RICE }, INT_MIN, INT_MAX, VE, "coder" },
1077 { "range_def", "Range with default table", 0, AV_OPT_TYPE_CONST,
1078 { .i64 = AC_RANGE_DEFAULT_TAB }, INT_MIN, INT_MAX, VE, "coder" },
1079 { "range_tab", "Range with custom table", 0, AV_OPT_TYPE_CONST,
1080 { .i64 = AC_RANGE_CUSTOM_TAB }, INT_MIN, INT_MAX, VE, "coder" },
1081 { "context", "Context model", OFFSET(context_model), AV_OPT_TYPE_INT,
1082 { .i64 = 0 }, 0, 1, VE },
1083
1084 { NULL }
1085 };
1086
1087 static const AVClass class = {
1088 .class_name = "ffv1 encoder",
1089 .item_name = av_default_item_name,
1090 .option = options,
1091 .version = LIBAVUTIL_VERSION_INT,
1092 };
1093
1094 #if FF_API_CODER_TYPE
1095 static const AVCodecDefault ffv1_defaults[] = {
1096 { "coder", "-1" },
1097 { NULL },
1098 };
1099 #endif
1100
1101 AVCodec ff_ffv1_encoder = {
1102 .name = "ffv1",
1103 .long_name = NULL_IF_CONFIG_SMALL("FFmpeg video codec #1"),
1104 .type = AVMEDIA_TYPE_VIDEO,
1105 .id = AV_CODEC_ID_FFV1,
1106 .priv_data_size = sizeof(FFV1Context),
1107 .init = ffv1_encode_init,
1108 .encode2 = ffv1_encode_frame,
1109 .close = ffv1_encode_close,
1110 .capabilities = AV_CODEC_CAP_SLICE_THREADS,
1111 .pix_fmts = (const enum AVPixelFormat[]) {
1112 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
1113 AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
1114 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV420P9,
1115 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
1116 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
1117 AV_PIX_FMT_RGB32,
1118 AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
1119 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
1120 AV_PIX_FMT_GRAY16, AV_PIX_FMT_GRAY8,
1121 AV_PIX_FMT_NONE
1122
1123 },
1124 #if FF_API_CODER_TYPE
1125 .defaults = ffv1_defaults,
1126 #endif
1127 .priv_class = &class,
1128 };