2 * FFV1 encoder for libavcodec
4 * Copyright (c) 2003-2012 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of Libav.
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.
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.
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
25 * FF Video Codec 1 (a lossless codec) encoder
28 #include "libavutil/avassert.h"
29 #include "libavutil/pixdesc.h"
30 #include "libavutil/crc.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/imgutils.h"
38 #include "rangecoder.h"
43 static void find_best_state(uint8_t best_state[256][256],
44 const uint8_t one_state[256])
49 for (i = 1; i < 256; i++)
50 l2tab[i] = log2(i / 256.0);
52 for (i = 0; i < 256; i++) {
56 for (j = 0; j < 256; j++)
57 best_len[j] = 1 << 30;
59 for (j = FFMAX(i - 10, 1); j < FFMIN(i + 11, 256); j++) {
60 double occ[256] = { 0 };
63 for (k = 0; k < 256; k++) {
64 double newocc[256] = { 0 };
65 for (m = 1; m < 256; m++)
67 len -= occ[m] * (p * l2tab[m] +
68 (1 - p) * l2tab[256 - m]);
70 if (len < best_len[k]) {
74 for (m = 0; m < 256; m++)
76 newocc[one_state[m]] += occ[m] * p;
77 newocc[256 - one_state[256 - m]] += occ[m] * (1 - p);
79 memcpy(occ, newocc, sizeof(occ));
85 static av_always_inline av_flatten void put_symbol_inline(RangeCoder *c,
86 uint8_t *state, int v,
88 uint64_t rc_stat[256][2],
89 uint64_t rc_stat2[32][2])
93 #define put_rac(C, S, B) \
97 rc_stat2[(S) - state][B]++; \
103 const int a = FFABS(v);
104 const int e = av_log2(a);
105 put_rac(c, state + 0, 0);
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);
111 for (i = e - 1; i >= 0; i--)
112 put_rac(c, state + 22 + i, (a >> i) & 1); // 22..31
115 put_rac(c, state + 11 + e, v < 0); // 11..21
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);
121 for (i = e - 1; i >= 0; i--)
122 put_rac(c, state + 22 + FFMIN(i, 9), (a >> i) & 1); // 22..31
125 put_rac(c, state + 11 + 10, v < 0); // 11..21
128 put_rac(c, state + 0, 1);
133 static av_noinline void put_symbol(RangeCoder *c, uint8_t *state,
134 int v, int is_signed)
136 put_symbol_inline(c, state, v, is_signed, NULL, NULL);
139 static inline void put_vlc_symbol(PutBitContext *pb, VlcState *const state,
143 v = fold(v - state->bias, bits);
147 while (i < state->error_sum) { // FIXME: optimize
155 if (k == 0 && 2 * state->drift <= -state->count)
160 code = v ^ ((2 * state->drift + state->count) >> 31);
163 av_dlog(NULL, "v:%d/%d bias:%d error:%d drift:%d count:%d k:%d\n", v, code,
164 state->bias, state->error_sum, state->drift, state->count, k);
165 set_sr_golomb(pb, code, k, 12, bits);
167 update_vlc_state(state, v);
170 static av_always_inline int encode_line(FFV1Context *s, int w,
172 int plane_index, int bits)
174 PlaneContext *const p = &s->plane[plane_index];
175 RangeCoder *const c = &s->c;
177 int run_index = s->run_index;
182 if (c->bytestream_end - c->bytestream < w * 20) {
183 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
184 return AVERROR_INVALIDDATA;
187 if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) < w * 4) {
188 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
189 return AVERROR_INVALIDDATA;
193 for (x = 0; x < w; x++) {
196 context = get_context(p, sample[0] + x, sample[1] + x, sample[2] + x);
197 diff = sample[0][x] - predict(sample[0] + x, sample[1] + x);
204 diff = fold(diff, bits);
207 if (s->flags & CODEC_FLAG_PASS1) {
208 put_symbol_inline(c, p->state[context], diff, 1, s->rc_stat,
209 s->rc_stat2[p->quant_table_index][context]);
211 put_symbol_inline(c, p->state[context], diff, 1, NULL, NULL);
219 while (run_count >= 1 << ff_log2_run[run_index]) {
220 run_count -= 1 << ff_log2_run[run_index];
222 put_bits(&s->pb, 1, 1);
225 put_bits(&s->pb, 1 + ff_log2_run[run_index], run_count);
237 av_dlog(s->avctx, "count:%d index:%d, mode:%d, x:%d pos:%d\n",
238 run_count, run_index, run_mode, x,
239 (int)put_bits_count(&s->pb));
242 put_vlc_symbol(&s->pb, &p->vlc_state[context], diff, bits);
246 while (run_count >= 1 << ff_log2_run[run_index]) {
247 run_count -= 1 << ff_log2_run[run_index];
249 put_bits(&s->pb, 1, 1);
253 put_bits(&s->pb, 1, 1);
255 s->run_index = run_index;
260 static void encode_plane(FFV1Context *s, uint8_t *src, int w, int h,
261 int stride, int plane_index)
264 const int ring_size = s->avctx->context_model ? 3 : 2;
268 memset(s->sample_buffer, 0, ring_size * (w + 6) * sizeof(*s->sample_buffer));
270 for (y = 0; y < h; y++) {
271 for (i = 0; i < ring_size; i++)
272 sample[i] = s->sample_buffer + (w + 6) * ((h + i - y) % ring_size) + 3;
274 sample[0][-1] = sample[1][0];
275 sample[1][w] = sample[1][w - 1];
277 if (s->bits_per_raw_sample <= 8) {
278 for (x = 0; x < w; x++)
279 sample[0][x] = src[x + stride * y];
280 encode_line(s, w, sample, plane_index, 8);
282 if (s->packed_at_lsb) {
283 for (x = 0; x < w; x++)
284 sample[0][x] = ((uint16_t *)(src + stride * y))[x];
286 for (x = 0; x < w; x++)
288 ((uint16_t *)(src + stride * y))[x] >> (16 - s->bits_per_raw_sample);
290 encode_line(s, w, sample, plane_index, s->bits_per_raw_sample);
292 // STOP_TIMER("encode line") }
296 static void encode_rgb_frame(FFV1Context *s, uint8_t *src[3], int w, int h,
300 const int ring_size = s->avctx->context_model ? 3 : 2;
301 int16_t *sample[MAX_PLANES][3];
302 int lbd = s->avctx->bits_per_raw_sample <= 8;
303 int bits = s->avctx->bits_per_raw_sample > 0
304 ? s->avctx->bits_per_raw_sample
306 int offset = 1 << bits;
310 memset(s->sample_buffer, 0, ring_size * MAX_PLANES *
311 (w + 6) * sizeof(*s->sample_buffer));
313 for (y = 0; y < h; y++) {
314 for (i = 0; i < ring_size; i++)
315 for (p = 0; p < MAX_PLANES; p++)
316 sample[p][i] = s->sample_buffer + p * ring_size *
318 ((h + i - y) % ring_size) * (w + 6) + 3;
320 for (x = 0; x < w; x++) {
321 int b, g, r, av_uninit(a);
323 unsigned v = *((uint32_t *)(src[0] + x * 4 + stride[0] * y));
326 r = (v >> 16) & 0xFF;
329 b = *((uint16_t *)(src[0] + x * 2 + stride[0] * y));
330 g = *((uint16_t *)(src[1] + x * 2 + stride[1] * y));
331 r = *((uint16_t *)(src[2] + x * 2 + stride[2] * y));
345 for (p = 0; p < 3 + s->transparency; p++) {
346 sample[p][0][-1] = sample[p][1][0];
347 sample[p][1][w] = sample[p][1][w - 1];
349 encode_line(s, w, sample[p], (p + 1) / 2, 9);
351 encode_line(s, w, sample[p], (p + 1) / 2, bits + 1);
357 static void write_quant_table(RangeCoder *c, int16_t *quant_table)
361 uint8_t state[CONTEXT_SIZE];
362 memset(state, 128, sizeof(state));
364 for (i = 1; i < 128; i++)
365 if (quant_table[i] != quant_table[i - 1]) {
366 put_symbol(c, state, i - last - 1, 0);
369 put_symbol(c, state, i - last - 1, 0);
372 static void write_quant_tables(RangeCoder *c,
373 int16_t quant_table[MAX_CONTEXT_INPUTS][256])
376 for (i = 0; i < 5; i++)
377 write_quant_table(c, quant_table[i]);
380 static void write_header(FFV1Context *f)
382 uint8_t state[CONTEXT_SIZE];
384 RangeCoder *const c = &f->slice_context[0]->c;
386 memset(state, 128, sizeof(state));
388 if (f->version < 2) {
389 put_symbol(c, state, f->version, 0);
390 put_symbol(c, state, f->ac, 0);
392 for (i = 1; i < 256; i++)
394 f->state_transition[i] - c->one_state[i], 1);
396 put_symbol(c, state, f->colorspace, 0); // YUV cs type
398 put_symbol(c, state, f->bits_per_raw_sample, 0);
399 put_rac(c, state, f->chroma_planes);
400 put_symbol(c, state, f->chroma_h_shift, 0);
401 put_symbol(c, state, f->chroma_v_shift, 0);
402 put_rac(c, state, f->transparency);
404 write_quant_tables(c, f->quant_table);
405 } else if (f->version < 3) {
406 put_symbol(c, state, f->slice_count, 0);
407 for (i = 0; i < f->slice_count; i++) {
408 FFV1Context *fs = f->slice_context[i];
410 (fs->slice_x + 1) * f->num_h_slices / f->width, 0);
412 (fs->slice_y + 1) * f->num_v_slices / f->height, 0);
414 (fs->slice_width + 1) * f->num_h_slices / f->width - 1,
417 (fs->slice_height + 1) * f->num_v_slices / f->height - 1,
419 for (j = 0; j < f->plane_count; j++) {
420 put_symbol(c, state, f->plane[j].quant_table_index, 0);
421 av_assert0(f->plane[j].quant_table_index == f->avctx->context_model);
427 static int write_extradata(FFV1Context *f)
429 RangeCoder *const c = &f->c;
430 uint8_t state[CONTEXT_SIZE];
432 uint8_t state2[32][CONTEXT_SIZE];
435 memset(state2, 128, sizeof(state2));
436 memset(state, 128, sizeof(state));
438 f->avctx->extradata_size = 10000 + 4 +
439 (11 * 11 * 5 * 5 * 5 + 11 * 11 * 11) * 32;
440 f->avctx->extradata = av_malloc(f->avctx->extradata_size);
441 ff_init_range_encoder(c, f->avctx->extradata, f->avctx->extradata_size);
442 ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8);
444 put_symbol(c, state, f->version, 0);
445 if (f->version > 2) {
447 f->minor_version = 2;
448 put_symbol(c, state, f->minor_version, 0);
451 put_symbol(c, state, f->ac, 0);
453 for (i = 1; i < 256; i++)
454 put_symbol(c, state, f->state_transition[i] - c->one_state[i], 1);
456 put_symbol(c, state, f->colorspace, 0); // YUV cs type
457 put_symbol(c, state, f->bits_per_raw_sample, 0);
458 put_rac(c, state, f->chroma_planes);
459 put_symbol(c, state, f->chroma_h_shift, 0);
460 put_symbol(c, state, f->chroma_v_shift, 0);
461 put_rac(c, state, f->transparency);
462 put_symbol(c, state, f->num_h_slices - 1, 0);
463 put_symbol(c, state, f->num_v_slices - 1, 0);
465 put_symbol(c, state, f->quant_table_count, 0);
466 for (i = 0; i < f->quant_table_count; i++)
467 write_quant_tables(c, f->quant_tables[i]);
469 for (i = 0; i < f->quant_table_count; i++) {
470 for (j = 0; j < f->context_count[i] * CONTEXT_SIZE; j++)
471 if (f->initial_states[i] && f->initial_states[i][0][j] != 128)
473 if (j < f->context_count[i] * CONTEXT_SIZE) {
474 put_rac(c, state, 1);
475 for (j = 0; j < f->context_count[i]; j++)
476 for (k = 0; k < CONTEXT_SIZE; k++) {
477 int pred = j ? f->initial_states[i][j - 1][k] : 128;
478 put_symbol(c, state2[k],
479 (int8_t)(f->initial_states[i][j][k] - pred), 1);
482 put_rac(c, state, 0);
486 if (f->version > 2) {
487 put_symbol(c, state, f->ec, 0);
490 f->avctx->extradata_size = ff_rac_terminate(c);
492 v = av_crc(av_crc_get_table(AV_CRC_32_IEEE), 0,
493 f->avctx->extradata, f->avctx->extradata_size);
494 AV_WL32(f->avctx->extradata + f->avctx->extradata_size, v);
495 f->avctx->extradata_size += 4;
500 static int sort_stt(FFV1Context *s, uint8_t stt[256])
502 int i, i2, changed, print = 0;
506 for (i = 12; i < 244; i++) {
507 for (i2 = i + 1; i2 < 245 && i2 < i + 4; i2++) {
509 #define COST(old, new) \
510 s->rc_stat[old][0] * -log2((256 - (new)) / 256.0) + \
511 s->rc_stat[old][1] * -log2((new) / 256.0)
513 #define COST2(old, new) \
514 COST(old, new) + COST(256 - (old), 256 - (new))
516 double size0 = COST2(i, i) + COST2(i2, i2);
517 double sizeX = COST2(i, i2) + COST2(i2, i);
518 if (sizeX < size0 && i != 128 && i2 != 128) {
520 FFSWAP(int, stt[i], stt[i2]);
521 FFSWAP(int, s->rc_stat[i][0], s->rc_stat[i2][0]);
522 FFSWAP(int, s->rc_stat[i][1], s->rc_stat[i2][1]);
524 FFSWAP(int, stt[256 - i], stt[256 - i2]);
525 FFSWAP(int, s->rc_stat[256 - i][0], s->rc_stat[256 - i2][0]);
526 FFSWAP(int, s->rc_stat[256 - i][1], s->rc_stat[256 - i2][1]);
528 for (j = 1; j < 256; j++) {
531 else if (stt[j] == i2)
534 if (stt[256 - j] == 256 - i)
535 stt[256 - j] = 256 - i2;
536 else if (stt[256 - j] == 256 - i2)
537 stt[256 - j] = 256 - i;
548 static int init_slices_state(FFV1Context *f)
551 for (i = 0; i < f->slice_count; i++) {
552 FFV1Context *fs = f->slice_context[i];
553 if ((ret = ffv1_init_slice_state(f, fs)) < 0)
554 return AVERROR(ENOMEM);
559 static av_cold int ffv1_encode_init(AVCodecContext *avctx)
561 FFV1Context *s = avctx->priv_data;
562 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
565 ffv1_common_init(avctx);
569 if ((avctx->flags & (CODEC_FLAG_PASS1 | CODEC_FLAG_PASS2)) ||
571 s->version = FFMAX(s->version, 2);
573 if (avctx->level == 3) {
578 s->ec = (s->version >= 3);
581 if (s->version >= 2 &&
582 avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
583 av_log(avctx, AV_LOG_ERROR,
584 "Version %d requested, please set -strict experimental in "
585 "order to enable it\n",
587 return AVERROR(ENOSYS);
590 s->ac = avctx->coder_type > 0 ? 2 : 0;
593 switch (avctx->pix_fmt) {
594 case AV_PIX_FMT_YUV444P9:
595 case AV_PIX_FMT_YUV422P9:
596 case AV_PIX_FMT_YUV420P9:
597 if (!avctx->bits_per_raw_sample)
598 s->bits_per_raw_sample = 9;
599 case AV_PIX_FMT_YUV444P10:
600 case AV_PIX_FMT_YUV420P10:
601 case AV_PIX_FMT_YUV422P10:
602 s->packed_at_lsb = 1;
603 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample)
604 s->bits_per_raw_sample = 10;
605 case AV_PIX_FMT_GRAY16:
606 case AV_PIX_FMT_YUV444P16:
607 case AV_PIX_FMT_YUV422P16:
608 case AV_PIX_FMT_YUV420P16:
609 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample) {
610 s->bits_per_raw_sample = 16;
611 } else if (!s->bits_per_raw_sample) {
612 s->bits_per_raw_sample = avctx->bits_per_raw_sample;
614 if (s->bits_per_raw_sample <= 8) {
615 av_log(avctx, AV_LOG_ERROR, "bits_per_raw_sample invalid\n");
616 return AVERROR_INVALIDDATA;
618 if (!s->ac && avctx->coder_type == -1) {
619 av_log(avctx, AV_LOG_INFO,
620 "bits_per_raw_sample > 8, forcing coder 1\n");
626 "bits_per_raw_sample of more than 8 needs -coder 1 currently\n");
627 return AVERROR_INVALIDDATA;
629 s->version = FFMAX(s->version, 1);
630 case AV_PIX_FMT_GRAY8:
631 case AV_PIX_FMT_YUV444P:
632 case AV_PIX_FMT_YUV440P:
633 case AV_PIX_FMT_YUV422P:
634 case AV_PIX_FMT_YUV420P:
635 case AV_PIX_FMT_YUV411P:
636 case AV_PIX_FMT_YUV410P:
637 s->chroma_planes = desc->nb_components < 3 ? 0 : 1;
640 case AV_PIX_FMT_YUVA444P:
641 case AV_PIX_FMT_YUVA422P:
642 case AV_PIX_FMT_YUVA420P:
643 s->chroma_planes = 1;
647 case AV_PIX_FMT_RGB32:
651 case AV_PIX_FMT_GBRP9:
652 if (!avctx->bits_per_raw_sample)
653 s->bits_per_raw_sample = 9;
654 case AV_PIX_FMT_GBRP10:
655 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample)
656 s->bits_per_raw_sample = 10;
657 case AV_PIX_FMT_GBRP16:
658 if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample)
659 s->bits_per_raw_sample = 16;
660 else if (!s->bits_per_raw_sample)
661 s->bits_per_raw_sample = avctx->bits_per_raw_sample;
663 s->chroma_planes = 1;
664 s->version = FFMAX(s->version, 1);
667 av_log(avctx, AV_LOG_ERROR, "format not supported\n");
668 return AVERROR_INVALIDDATA;
670 if (s->transparency) {
672 avctx, AV_LOG_WARNING,
673 "Storing alpha plane, this will require a recent FFV1 decoder to playback!\n");
675 if (avctx->context_model > 1U) {
676 av_log(avctx, AV_LOG_ERROR,
677 "Invalid context model %d, valid values are 0 and 1\n",
678 avctx->context_model);
679 return AVERROR(EINVAL);
683 for (i = 1; i < 256; i++)
684 s->state_transition[i] = ffv1_ver2_state[i];
686 for (i = 0; i < 256; i++) {
687 s->quant_table_count = 2;
688 if (s->bits_per_raw_sample <= 8) {
689 s->quant_tables[0][0][i] = ffv1_quant11[i];
690 s->quant_tables[0][1][i] = ffv1_quant11[i] * 11;
691 s->quant_tables[0][2][i] = ffv1_quant11[i] * 11 * 11;
692 s->quant_tables[1][0][i] = ffv1_quant11[i];
693 s->quant_tables[1][1][i] = ffv1_quant11[i] * 11;
694 s->quant_tables[1][2][i] = ffv1_quant5[i] * 11 * 11;
695 s->quant_tables[1][3][i] = ffv1_quant5[i] * 5 * 11 * 11;
696 s->quant_tables[1][4][i] = ffv1_quant5[i] * 5 * 5 * 11 * 11;
698 s->quant_tables[0][0][i] = ffv1_quant9_10bit[i];
699 s->quant_tables[0][1][i] = ffv1_quant9_10bit[i] * 11;
700 s->quant_tables[0][2][i] = ffv1_quant9_10bit[i] * 11 * 11;
701 s->quant_tables[1][0][i] = ffv1_quant9_10bit[i];
702 s->quant_tables[1][1][i] = ffv1_quant9_10bit[i] * 11;
703 s->quant_tables[1][2][i] = ffv1_quant5_10bit[i] * 11 * 11;
704 s->quant_tables[1][3][i] = ffv1_quant5_10bit[i] * 5 * 11 * 11;
705 s->quant_tables[1][4][i] = ffv1_quant5_10bit[i] * 5 * 5 * 11 * 11;
708 s->context_count[0] = (11 * 11 * 11 + 1) / 2;
709 s->context_count[1] = (11 * 11 * 5 * 5 * 5 + 1) / 2;
710 memcpy(s->quant_table, s->quant_tables[avctx->context_model],
711 sizeof(s->quant_table));
713 for (i = 0; i < s->plane_count; i++) {
714 PlaneContext *const p = &s->plane[i];
716 memcpy(p->quant_table, s->quant_table, sizeof(p->quant_table));
717 p->quant_table_index = avctx->context_model;
718 p->context_count = s->context_count[p->quant_table_index];
721 if ((ret = ffv1_allocate_initial_states(s)) < 0)
724 avctx->coded_frame = &s->picture;
725 if (!s->transparency)
728 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift,
731 s->picture_number = 0;
733 if (avctx->flags & (CODEC_FLAG_PASS1 | CODEC_FLAG_PASS2)) {
734 for (i = 0; i < s->quant_table_count; i++) {
735 s->rc_stat2[i] = av_mallocz(s->context_count[i] *
736 sizeof(*s->rc_stat2[i]));
738 return AVERROR(ENOMEM);
741 if (avctx->stats_in) {
742 char *p = avctx->stats_in;
743 uint8_t best_state[256][256];
747 av_assert0(s->version >= 2);
750 for (j = 0; j < 256; j++)
751 for (i = 0; i < 2; i++) {
752 s->rc_stat[j][i] = strtol(p, &next, 0);
754 av_log(avctx, AV_LOG_ERROR,
755 "2Pass file invalid at %d %d [%s]\n", j, i, p);
756 return AVERROR_INVALIDDATA;
760 for (i = 0; i < s->quant_table_count; i++)
761 for (j = 0; j < s->context_count[i]; j++) {
762 for (k = 0; k < 32; k++)
763 for (m = 0; m < 2; m++) {
764 s->rc_stat2[i][j][k][m] = strtol(p, &next, 0);
766 av_log(avctx, AV_LOG_ERROR,
767 "2Pass file invalid at %d %d %d %d [%s]\n",
769 return AVERROR_INVALIDDATA;
774 gob_count = strtol(p, &next, 0);
775 if (next == p || gob_count <= 0) {
776 av_log(avctx, AV_LOG_ERROR, "2Pass file invalid\n");
777 return AVERROR_INVALIDDATA;
780 while (*p == '\n' || *p == ' ')
785 sort_stt(s, s->state_transition);
787 find_best_state(best_state, s->state_transition);
789 for (i = 0; i < s->quant_table_count; i++) {
790 for (j = 0; j < s->context_count[i]; j++)
791 for (k = 0; k < 32; k++) {
793 if (s->rc_stat2[i][j][k][0] + s->rc_stat2[i][j][k][1]) {
794 p = 256.0 * s->rc_stat2[i][j][k][1] /
795 (s->rc_stat2[i][j][k][0] + s->rc_stat2[i][j][k][1]);
797 s->initial_states[i][j][k] =
798 best_state[av_clip(round(p), 1, 255)][av_clip((s->rc_stat2[i][j][k][0] +
799 s->rc_stat2[i][j][k][1]) /
805 if (s->version > 1) {
806 for (s->num_v_slices = 2; s->num_v_slices < 9; s->num_v_slices++)
807 for (s->num_h_slices = s->num_v_slices;
808 s->num_h_slices < 2 * s->num_v_slices; s->num_h_slices++)
809 if (avctx->slices == s->num_h_slices * s->num_v_slices &&
810 avctx->slices <= 64 || !avctx->slices)
812 av_log(avctx, AV_LOG_ERROR,
813 "Unsupported number %d of slices requested, please specify a "
814 "supported number with -slices (ex:4,6,9,12,16, ...)\n",
816 return AVERROR(ENOSYS);
821 if ((ret = ffv1_init_slice_contexts(s)) < 0)
823 if ((ret = init_slices_state(s)) < 0)
826 #define STATS_OUT_SIZE 1024 * 1024 * 6
827 if (avctx->flags & CODEC_FLAG_PASS1) {
828 avctx->stats_out = av_mallocz(STATS_OUT_SIZE);
829 for (i = 0; i < s->quant_table_count; i++)
830 for (j = 0; j < s->slice_count; j++) {
831 FFV1Context *sf = s->slice_context[j];
832 av_assert0(!sf->rc_stat2[i]);
833 sf->rc_stat2[i] = av_mallocz(s->context_count[i] *
834 sizeof(*sf->rc_stat2[i]));
835 if (!sf->rc_stat2[i])
836 return AVERROR(ENOMEM);
843 static void encode_slice_header(FFV1Context *f, FFV1Context *fs)
845 RangeCoder *c = &fs->c;
846 uint8_t state[CONTEXT_SIZE];
848 memset(state, 128, sizeof(state));
850 put_symbol(c, state, (fs->slice_x + 1) * f->num_h_slices / f->width, 0);
851 put_symbol(c, state, (fs->slice_y + 1) * f->num_v_slices / f->height, 0);
852 put_symbol(c, state, (fs->slice_width + 1) * f->num_h_slices / f->width - 1,
855 (fs->slice_height + 1) * f->num_v_slices / f->height - 1,
857 for (j = 0; j < f->plane_count; j++) {
858 put_symbol(c, state, f->plane[j].quant_table_index, 0);
859 av_assert0(f->plane[j].quant_table_index == f->avctx->context_model);
861 if (!f->picture.interlaced_frame)
862 put_symbol(c, state, 3, 0);
864 put_symbol(c, state, 1 + !f->picture.top_field_first, 0);
865 put_symbol(c, state, f->picture.sample_aspect_ratio.num, 0);
866 put_symbol(c, state, f->picture.sample_aspect_ratio.den, 0);
869 static int encode_slice(AVCodecContext *c, void *arg)
871 FFV1Context *fs = *(void **)arg;
872 FFV1Context *f = fs->avctx->priv_data;
873 int width = fs->slice_width;
874 int height = fs->slice_height;
877 AVFrame *const p = &f->picture;
878 const int ps = (av_pix_fmt_desc_get(c->pix_fmt)->flags & PIX_FMT_PLANAR)
879 ? (f->bits_per_raw_sample > 8) + 1
883 ffv1_clear_slice_state(f, fs);
884 if (f->version > 2) {
885 encode_slice_header(f, fs);
889 put_rac(&fs->c, (uint8_t[]) { 129 }, 0);
890 fs->ac_byte_count = f->version > 2 || (!x && !y) ? ff_rac_terminate( &fs->c) : 0;
891 init_put_bits(&fs->pb, fs->c.bytestream_start + fs->ac_byte_count,
892 fs->c.bytestream_end - fs->c.bytestream_start - fs->ac_byte_count);
895 if (f->colorspace == 0) {
896 const int chroma_width = -((-width) >> f->chroma_h_shift);
897 const int chroma_height = -((-height) >> f->chroma_v_shift);
898 const int cx = x >> f->chroma_h_shift;
899 const int cy = y >> f->chroma_v_shift;
901 encode_plane(fs, p->data[0] + ps * x + y * p->linesize[0],
902 width, height, p->linesize[0], 0);
904 if (f->chroma_planes) {
905 encode_plane(fs, p->data[1] + ps * cx + cy * p->linesize[1],
906 chroma_width, chroma_height, p->linesize[1], 1);
907 encode_plane(fs, p->data[2] + ps * cx + cy * p->linesize[2],
908 chroma_width, chroma_height, p->linesize[2], 1);
910 if (fs->transparency)
911 encode_plane(fs, p->data[3] + ps * x + y * p->linesize[3], width,
912 height, p->linesize[3], 2);
914 uint8_t *planes[3] = { p->data[0] + ps * x + y * p->linesize[0],
915 p->data[1] + ps * x + y * p->linesize[1],
916 p->data[2] + ps * x + y * p->linesize[2] };
917 encode_rgb_frame(fs, planes, width, height, p->linesize);
924 static int ffv1_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
925 const AVFrame *pict, int *got_packet)
927 FFV1Context *f = avctx->priv_data;
928 RangeCoder *const c = &f->slice_context[0]->c;
929 AVFrame *const p = &f->picture;
931 uint8_t keystate = 128;
935 if ((ret = ff_alloc_packet(pkt, avctx->width * avctx->height *
936 ((8 * 2 + 1 + 1) * 4) / 8 +
937 FF_MIN_BUFFER_SIZE)) < 0) {
938 av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
942 ff_init_range_encoder(c, pkt->data, pkt->size);
943 ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8);
946 p->pict_type = AV_PICTURE_TYPE_I;
948 if (avctx->gop_size == 0 || f->picture_number % avctx->gop_size == 0) {
949 put_rac(c, &keystate, 1);
954 put_rac(c, &keystate, 0);
960 for (i = 1; i < 256; i++) {
961 c->one_state[i] = f->state_transition[i];
962 c->zero_state[256 - i] = 256 - c->one_state[i];
966 for (i = 1; i < f->slice_count; i++) {
967 FFV1Context *fs = f->slice_context[i];
968 uint8_t *start = pkt->data +
969 (pkt->size - used_count) * (int64_t)i / f->slice_count;
970 int len = pkt->size / f->slice_count;
971 ff_init_range_encoder(&fs->c, start, len);
973 avctx->execute(avctx, encode_slice, &f->slice_context[0], NULL,
974 f->slice_count, sizeof(void *));
977 for (i = 0; i < f->slice_count; i++) {
978 FFV1Context *fs = f->slice_context[i];
983 put_rac(&fs->c, &state, 0);
984 bytes = ff_rac_terminate(&fs->c);
986 flush_put_bits(&fs->pb); // FIXME: nicer padding
987 bytes = fs->ac_byte_count + (put_bits_count(&fs->pb) + 7) / 8;
989 if (i > 0 || f->version > 2) {
990 av_assert0(bytes < pkt->size / f->slice_count);
991 memmove(buf_p, fs->c.bytestream_start, bytes);
992 av_assert0(bytes < (1 << 24));
993 AV_WB24(buf_p + bytes, bytes);
999 v = av_crc(av_crc_get_table(AV_CRC_32_IEEE), 0, buf_p, bytes);
1000 AV_WL32(buf_p + bytes, v);
1006 if ((avctx->flags & CODEC_FLAG_PASS1) && (f->picture_number & 31) == 0) {
1008 char *p = avctx->stats_out;
1009 char *end = p + STATS_OUT_SIZE;
1011 memset(f->rc_stat, 0, sizeof(f->rc_stat));
1012 for (i = 0; i < f->quant_table_count; i++)
1013 memset(f->rc_stat2[i], 0, f->context_count[i] * sizeof(*f->rc_stat2[i]));
1015 for (j = 0; j < f->slice_count; j++) {
1016 FFV1Context *fs = f->slice_context[j];
1017 for (i = 0; i < 256; i++) {
1018 f->rc_stat[i][0] += fs->rc_stat[i][0];
1019 f->rc_stat[i][1] += fs->rc_stat[i][1];
1021 for (i = 0; i < f->quant_table_count; i++) {
1022 for (k = 0; k < f->context_count[i]; k++)
1023 for (m = 0; m < 32; m++) {
1024 f->rc_stat2[i][k][m][0] += fs->rc_stat2[i][k][m][0];
1025 f->rc_stat2[i][k][m][1] += fs->rc_stat2[i][k][m][1];
1030 for (j = 0; j < 256; j++) {
1031 snprintf(p, end - p, "%" PRIu64 " %" PRIu64 " ",
1032 f->rc_stat[j][0], f->rc_stat[j][1]);
1035 snprintf(p, end - p, "\n");
1037 for (i = 0; i < f->quant_table_count; i++) {
1038 for (j = 0; j < f->context_count[i]; j++)
1039 for (m = 0; m < 32; m++) {
1040 snprintf(p, end - p, "%" PRIu64 " %" PRIu64 " ",
1041 f->rc_stat2[i][j][m][0], f->rc_stat2[i][j][m][1]);
1045 snprintf(p, end - p, "%d\n", f->gob_count);
1046 } else if (avctx->flags & CODEC_FLAG_PASS1)
1047 avctx->stats_out[0] = '\0';
1049 f->picture_number++;
1050 pkt->size = buf_p - pkt->data;
1051 pkt->flags |= AV_PKT_FLAG_KEY * p->key_frame;
1057 #define OFFSET(x) offsetof(FFV1Context, x)
1058 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1059 static const AVOption options[] = {
1060 { "slicecrc", "Protect slices with CRCs", OFFSET(ec), AV_OPT_TYPE_INT,
1061 { .i64 = -1 }, -1, 1, VE },
1065 static const AVClass class = {
1066 .class_name = "ffv1 encoder",
1067 .item_name = av_default_item_name,
1069 .version = LIBAVUTIL_VERSION_INT,
1072 static const AVCodecDefault ffv1_defaults[] = {
1077 AVCodec ff_ffv1_encoder = {
1079 .type = AVMEDIA_TYPE_VIDEO,
1080 .id = AV_CODEC_ID_FFV1,
1081 .priv_data_size = sizeof(FFV1Context),
1082 .init = ffv1_encode_init,
1083 .encode2 = ffv1_encode_frame,
1084 .close = ffv1_close,
1085 .capabilities = CODEC_CAP_SLICE_THREADS,
1086 .pix_fmts = (const enum AVPixelFormat[]) {
1087 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
1088 AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
1089 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV420P9,
1090 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
1091 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
1093 AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
1094 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
1095 AV_PIX_FMT_GRAY16, AV_PIX_FMT_GRAY8,
1099 .long_name = NULL_IF_CONFIG_SMALL("FFmpeg video codec #1"),
1100 .defaults = ffv1_defaults,
1101 .priv_class = &class,