4 * Copyright (c) 2012 Konstantin Shishkov
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
23 #include "libavutil/opt.h"
26 #include "bytestream.h"
28 #include "proresdsp.h"
29 #include "proresdata.h"
31 #define CFACTOR_Y422 2
32 #define CFACTOR_Y444 3
34 #define MAX_MBS_PER_SLICE 8
36 #define MAX_PLANES 3 // should be increased to 4 when there's PIX_FMT_YUV444AP10
39 PRORES_PROFILE_PROXY = 0,
41 PRORES_PROFILE_STANDARD,
53 static const uint8_t prores_quant_matrices[][64] = {
55 4, 7, 9, 11, 13, 14, 15, 63,
56 7, 7, 11, 12, 14, 15, 63, 63,
57 9, 11, 13, 14, 15, 63, 63, 63,
58 11, 11, 13, 14, 63, 63, 63, 63,
59 11, 13, 14, 63, 63, 63, 63, 63,
60 13, 14, 63, 63, 63, 63, 63, 63,
61 13, 63, 63, 63, 63, 63, 63, 63,
62 63, 63, 63, 63, 63, 63, 63, 63,
65 4, 5, 6, 7, 9, 11, 13, 15,
66 5, 5, 7, 8, 11, 13, 15, 17,
67 6, 7, 9, 11, 13, 15, 15, 17,
68 7, 7, 9, 11, 13, 15, 17, 19,
69 7, 9, 11, 13, 14, 16, 19, 23,
70 9, 11, 13, 14, 16, 19, 23, 29,
71 9, 11, 13, 15, 17, 21, 28, 35,
72 11, 13, 16, 17, 21, 28, 35, 41,
75 4, 4, 5, 5, 6, 7, 7, 9,
76 4, 4, 5, 6, 7, 7, 9, 9,
77 5, 5, 6, 7, 7, 9, 9, 10,
78 5, 5, 6, 7, 7, 9, 9, 10,
79 5, 6, 7, 7, 8, 9, 10, 12,
80 6, 7, 7, 8, 9, 10, 12, 15,
81 6, 7, 7, 9, 10, 11, 14, 17,
82 7, 7, 9, 10, 11, 14, 17, 21,
85 4, 4, 4, 4, 4, 4, 4, 4,
86 4, 4, 4, 4, 4, 4, 4, 4,
87 4, 4, 4, 4, 4, 4, 4, 4,
88 4, 4, 4, 4, 4, 4, 4, 5,
89 4, 4, 4, 4, 4, 4, 5, 5,
90 4, 4, 4, 4, 4, 5, 5, 6,
91 4, 4, 4, 4, 5, 5, 6, 7,
92 4, 4, 4, 4, 5, 6, 7, 7,
95 4, 4, 4, 4, 4, 4, 4, 4,
96 4, 4, 4, 4, 4, 4, 4, 4,
97 4, 4, 4, 4, 4, 4, 4, 4,
98 4, 4, 4, 4, 4, 4, 4, 4,
99 4, 4, 4, 4, 4, 4, 4, 4,
100 4, 4, 4, 4, 4, 4, 4, 4,
101 4, 4, 4, 4, 4, 4, 4, 4,
102 4, 4, 4, 4, 4, 4, 4, 4,
106 #define NUM_MB_LIMITS 4
107 static const int prores_mb_limits[NUM_MB_LIMITS] = {
108 1620, // up to 720x576
109 2700, // up to 960x720
110 6075, // up to 1440x1080
111 9216, // up to 2048x1152
114 static const struct prores_profile {
115 const char *full_name;
119 int br_tab[NUM_MB_LIMITS];
121 } prores_profile_info[4] = {
123 .full_name = "proxy",
124 .tag = MKTAG('a', 'p', 'c', 'o'),
127 .br_tab = { 300, 242, 220, 194 },
128 .quant = QUANT_MAT_PROXY,
132 .tag = MKTAG('a', 'p', 'c', 's'),
135 .br_tab = { 720, 560, 490, 440 },
136 .quant = QUANT_MAT_LT,
139 .full_name = "standard",
140 .tag = MKTAG('a', 'p', 'c', 'n'),
143 .br_tab = { 1050, 808, 710, 632 },
144 .quant = QUANT_MAT_STANDARD,
147 .full_name = "high quality",
148 .tag = MKTAG('a', 'p', 'c', 'h'),
151 .br_tab = { 1566, 1216, 1070, 950 },
152 .quant = QUANT_MAT_HQ,
154 // for 4444 profile bitrate numbers are { 2350, 1828, 1600, 1425 }
157 #define TRELLIS_WIDTH 16
158 #define SCORE_LIMIT INT_MAX / 2
167 #define MAX_STORED_Q 16
169 typedef struct ProresThreadData {
170 DECLARE_ALIGNED(16, DCTELEM, blocks)[MAX_PLANES][64 * 4 * MAX_MBS_PER_SLICE];
171 DECLARE_ALIGNED(16, uint16_t, emu_buf)[16 * 16];
172 int16_t custom_q[64];
173 struct TrellisNode *nodes;
176 typedef struct ProresContext {
178 DECLARE_ALIGNED(16, DCTELEM, blocks)[MAX_PLANES][64 * 4 * MAX_MBS_PER_SLICE];
179 DECLARE_ALIGNED(16, uint16_t, emu_buf)[16*16];
180 int16_t quants[MAX_STORED_Q][64];
181 int16_t custom_q[64];
182 const uint8_t *quant_mat;
184 ProresDSPContext dsp;
187 int mb_width, mb_height;
189 int num_chroma_blocks, chroma_factor;
191 int slices_per_picture;
192 int pictures_per_frame; // 1 for progressive, 2 for interlaced
201 int frame_size_upper_bound;
204 const struct prores_profile *profile_info;
208 ProresThreadData *tdata;
211 static void get_slice_data(ProresContext *ctx, const uint16_t *src,
212 int linesize, int x, int y, int w, int h,
213 DCTELEM *blocks, uint16_t *emu_buf,
214 int mbs_per_slice, int blocks_per_mb, int is_chroma)
216 const uint16_t *esrc;
217 const int mb_width = 4 * blocks_per_mb;
221 for (i = 0; i < mbs_per_slice; i++, src += mb_width) {
223 memset(blocks, 0, 64 * (mbs_per_slice - i) * blocks_per_mb
227 if (x + mb_width <= w && y + 16 <= h) {
229 elinesize = linesize;
234 elinesize = 16 * sizeof(*emu_buf);
236 bw = FFMIN(w - x, mb_width);
237 bh = FFMIN(h - y, 16);
239 for (j = 0; j < bh; j++) {
240 memcpy(emu_buf + j * 16,
241 (const uint8_t*)src + j * linesize,
243 pix = emu_buf[j * 16 + bw - 1];
244 for (k = bw; k < mb_width; k++)
245 emu_buf[j * 16 + k] = pix;
248 memcpy(emu_buf + j * 16,
249 emu_buf + (bh - 1) * 16,
250 mb_width * sizeof(*emu_buf));
253 ctx->dsp.fdct(esrc, elinesize, blocks);
255 if (blocks_per_mb > 2) {
256 ctx->dsp.fdct(esrc + 8, elinesize, blocks);
259 ctx->dsp.fdct(esrc + elinesize * 4, elinesize, blocks);
261 if (blocks_per_mb > 2) {
262 ctx->dsp.fdct(esrc + elinesize * 4 + 8, elinesize, blocks);
266 ctx->dsp.fdct(esrc, elinesize, blocks);
268 ctx->dsp.fdct(esrc + elinesize * 4, elinesize, blocks);
270 if (blocks_per_mb > 2) {
271 ctx->dsp.fdct(esrc + 8, elinesize, blocks);
273 ctx->dsp.fdct(esrc + elinesize * 4 + 8, elinesize, blocks);
283 * Write an unsigned rice/exp golomb codeword.
285 static inline void encode_vlc_codeword(PutBitContext *pb, unsigned codebook, int val)
287 unsigned int rice_order, exp_order, switch_bits, switch_val;
290 /* number of prefix bits to switch between Rice and expGolomb */
291 switch_bits = (codebook & 3) + 1;
292 rice_order = codebook >> 5; /* rice code order */
293 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
295 switch_val = switch_bits << rice_order;
297 if (val >= switch_val) {
298 val -= switch_val - (1 << exp_order);
299 exponent = av_log2(val);
301 put_bits(pb, exponent - exp_order + switch_bits, 0);
302 put_bits(pb, exponent + 1, val);
304 exponent = val >> rice_order;
307 put_bits(pb, exponent, 0);
310 put_sbits(pb, rice_order, val);
314 #define GET_SIGN(x) ((x) >> 31)
315 #define MAKE_CODE(x) (((x) << 1) ^ GET_SIGN(x))
317 static void encode_dcs(PutBitContext *pb, DCTELEM *blocks,
318 int blocks_per_slice, int scale)
321 int codebook = 3, code, dc, prev_dc, delta, sign, new_sign;
323 prev_dc = (blocks[0] - 0x4000) / scale;
324 encode_vlc_codeword(pb, FIRST_DC_CB, MAKE_CODE(prev_dc));
329 for (i = 1; i < blocks_per_slice; i++, blocks += 64) {
330 dc = (blocks[0] - 0x4000) / scale;
331 delta = dc - prev_dc;
332 new_sign = GET_SIGN(delta);
333 delta = (delta ^ sign) - sign;
334 code = MAKE_CODE(delta);
335 encode_vlc_codeword(pb, ff_prores_dc_codebook[codebook], code);
336 codebook = (code + (code & 1)) >> 1;
337 codebook = FFMIN(codebook, 3);
343 static void encode_acs(PutBitContext *pb, DCTELEM *blocks,
344 int blocks_per_slice,
345 int plane_size_factor,
346 const uint8_t *scan, const int16_t *qmat)
349 int run, level, run_cb, lev_cb;
350 int max_coeffs, abs_level;
352 max_coeffs = blocks_per_slice << 6;
353 run_cb = ff_prores_run_to_cb_index[4];
354 lev_cb = ff_prores_lev_to_cb_index[2];
357 for (i = 1; i < 64; i++) {
358 for (idx = scan[i]; idx < max_coeffs; idx += 64) {
359 level = blocks[idx] / qmat[scan[i]];
361 abs_level = FFABS(level);
362 encode_vlc_codeword(pb, ff_prores_ac_codebook[run_cb], run);
363 encode_vlc_codeword(pb, ff_prores_ac_codebook[lev_cb],
365 put_sbits(pb, 1, GET_SIGN(level));
367 run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)];
368 lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)];
377 static int encode_slice_plane(ProresContext *ctx, PutBitContext *pb,
378 const uint16_t *src, int linesize,
379 int mbs_per_slice, DCTELEM *blocks,
380 int blocks_per_mb, int plane_size_factor,
383 int blocks_per_slice, saved_pos;
385 saved_pos = put_bits_count(pb);
386 blocks_per_slice = mbs_per_slice * blocks_per_mb;
388 encode_dcs(pb, blocks, blocks_per_slice, qmat[0]);
389 encode_acs(pb, blocks, blocks_per_slice, plane_size_factor,
390 ctx->scantable.permutated, qmat);
393 return (put_bits_count(pb) - saved_pos) >> 3;
396 static int encode_slice(AVCodecContext *avctx, const AVFrame *pic,
398 int sizes[4], int x, int y, int quant,
401 ProresContext *ctx = avctx->priv_data;
405 int slice_width_factor = av_log2(mbs_per_slice);
406 int num_cblocks, pwidth, linesize, line_add;
407 int plane_factor, is_chroma;
410 if (ctx->pictures_per_frame == 1)
413 line_add = ctx->cur_picture_idx ^ !pic->top_field_first;
415 if (ctx->force_quant) {
416 qmat = ctx->quants[0];
417 } else if (quant < MAX_STORED_Q) {
418 qmat = ctx->quants[quant];
420 qmat = ctx->custom_q;
421 for (i = 0; i < 64; i++)
422 qmat[i] = ctx->quant_mat[i] * quant;
425 for (i = 0; i < ctx->num_planes; i++) {
426 is_chroma = (i == 1 || i == 2);
427 plane_factor = slice_width_factor + 2;
429 plane_factor += ctx->chroma_factor - 3;
430 if (!is_chroma || ctx->chroma_factor == CFACTOR_Y444) {
434 pwidth = avctx->width;
439 pwidth = avctx->width >> 1;
442 linesize = pic->linesize[i] * ctx->pictures_per_frame;
443 src = (const uint16_t*)(pic->data[i] + yp * linesize +
444 line_add * pic->linesize[i]) + xp;
446 get_slice_data(ctx, src, linesize, xp, yp,
447 pwidth, avctx->height / ctx->pictures_per_frame,
448 ctx->blocks[0], ctx->emu_buf,
449 mbs_per_slice, num_cblocks, is_chroma);
450 sizes[i] = encode_slice_plane(ctx, pb, src, linesize,
451 mbs_per_slice, ctx->blocks[0],
452 num_cblocks, plane_factor,
454 total_size += sizes[i];
459 static inline int estimate_vlc(unsigned codebook, int val)
461 unsigned int rice_order, exp_order, switch_bits, switch_val;
464 /* number of prefix bits to switch between Rice and expGolomb */
465 switch_bits = (codebook & 3) + 1;
466 rice_order = codebook >> 5; /* rice code order */
467 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
469 switch_val = switch_bits << rice_order;
471 if (val >= switch_val) {
472 val -= switch_val - (1 << exp_order);
473 exponent = av_log2(val);
475 return exponent * 2 - exp_order + switch_bits + 1;
477 return (val >> rice_order) + rice_order + 1;
481 static int estimate_dcs(int *error, DCTELEM *blocks, int blocks_per_slice,
485 int codebook = 3, code, dc, prev_dc, delta, sign, new_sign;
488 prev_dc = (blocks[0] - 0x4000) / scale;
489 bits = estimate_vlc(FIRST_DC_CB, MAKE_CODE(prev_dc));
493 *error += FFABS(blocks[0] - 0x4000) % scale;
495 for (i = 1; i < blocks_per_slice; i++, blocks += 64) {
496 dc = (blocks[0] - 0x4000) / scale;
497 *error += FFABS(blocks[0] - 0x4000) % scale;
498 delta = dc - prev_dc;
499 new_sign = GET_SIGN(delta);
500 delta = (delta ^ sign) - sign;
501 code = MAKE_CODE(delta);
502 bits += estimate_vlc(ff_prores_dc_codebook[codebook], code);
503 codebook = (code + (code & 1)) >> 1;
504 codebook = FFMIN(codebook, 3);
512 static int estimate_acs(int *error, DCTELEM *blocks, int blocks_per_slice,
513 int plane_size_factor,
514 const uint8_t *scan, const int16_t *qmat)
517 int run, level, run_cb, lev_cb;
518 int max_coeffs, abs_level;
521 max_coeffs = blocks_per_slice << 6;
522 run_cb = ff_prores_run_to_cb_index[4];
523 lev_cb = ff_prores_lev_to_cb_index[2];
526 for (i = 1; i < 64; i++) {
527 for (idx = scan[i]; idx < max_coeffs; idx += 64) {
528 level = blocks[idx] / qmat[scan[i]];
529 *error += FFABS(blocks[idx]) % qmat[scan[i]];
531 abs_level = FFABS(level);
532 bits += estimate_vlc(ff_prores_ac_codebook[run_cb], run);
533 bits += estimate_vlc(ff_prores_ac_codebook[lev_cb],
536 run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)];
537 lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)];
548 static int estimate_slice_plane(ProresContext *ctx, int *error, int plane,
549 const uint16_t *src, int linesize,
551 int blocks_per_mb, int plane_size_factor,
552 const int16_t *qmat, ProresThreadData *td)
554 int blocks_per_slice;
557 blocks_per_slice = mbs_per_slice * blocks_per_mb;
559 bits = estimate_dcs(error, td->blocks[plane], blocks_per_slice, qmat[0]);
560 bits += estimate_acs(error, td->blocks[plane], blocks_per_slice,
561 plane_size_factor, ctx->scantable.permutated, qmat);
563 return FFALIGN(bits, 8);
566 static int find_slice_quant(AVCodecContext *avctx, const AVFrame *pic,
567 int trellis_node, int x, int y, int mbs_per_slice,
568 ProresThreadData *td)
570 ProresContext *ctx = avctx->priv_data;
571 int i, q, pq, xp, yp;
573 int slice_width_factor = av_log2(mbs_per_slice);
574 int num_cblocks[MAX_PLANES], pwidth;
575 int plane_factor[MAX_PLANES], is_chroma[MAX_PLANES];
576 const int min_quant = ctx->profile_info->min_quant;
577 const int max_quant = ctx->profile_info->max_quant;
578 int error, bits, bits_limit;
579 int mbs, prev, cur, new_score;
580 int slice_bits[TRELLIS_WIDTH], slice_score[TRELLIS_WIDTH];
583 int linesize[4], line_add;
585 if (ctx->pictures_per_frame == 1)
588 line_add = ctx->cur_picture_idx ^ !pic->top_field_first;
589 mbs = x + mbs_per_slice;
591 for (i = 0; i < ctx->num_planes; i++) {
592 is_chroma[i] = (i == 1 || i == 2);
593 plane_factor[i] = slice_width_factor + 2;
595 plane_factor[i] += ctx->chroma_factor - 3;
596 if (!is_chroma[i] || ctx->chroma_factor == CFACTOR_Y444) {
600 pwidth = avctx->width;
605 pwidth = avctx->width >> 1;
608 linesize[i] = pic->linesize[i] * ctx->pictures_per_frame;
609 src = (const uint16_t*)(pic->data[i] + yp * linesize[i] +
610 line_add * pic->linesize[i]) + xp;
612 get_slice_data(ctx, src, linesize[i], xp, yp,
613 pwidth, avctx->height / ctx->pictures_per_frame,
614 td->blocks[i], td->emu_buf,
615 mbs_per_slice, num_cblocks[i], is_chroma[i]);
618 for (q = min_quant; q < max_quant + 2; q++) {
619 td->nodes[trellis_node + q].prev_node = -1;
620 td->nodes[trellis_node + q].quant = q;
623 // todo: maybe perform coarser quantising to fit into frame size when needed
624 for (q = min_quant; q <= max_quant; q++) {
627 for (i = 0; i < ctx->num_planes; i++) {
628 bits += estimate_slice_plane(ctx, &error, i,
631 num_cblocks[i], plane_factor[i],
634 if (bits > 65000 * 8) {
638 slice_bits[q] = bits;
639 slice_score[q] = error;
641 if (slice_bits[max_quant] <= ctx->bits_per_mb * mbs_per_slice) {
642 slice_bits[max_quant + 1] = slice_bits[max_quant];
643 slice_score[max_quant + 1] = slice_score[max_quant] + 1;
644 overquant = max_quant;
646 for (q = max_quant + 1; q < 128; q++) {
649 if (q < MAX_STORED_Q) {
650 qmat = ctx->quants[q];
653 for (i = 0; i < 64; i++)
654 qmat[i] = ctx->quant_mat[i] * q;
656 for (i = 0; i < ctx->num_planes; i++) {
657 bits += estimate_slice_plane(ctx, &error, i,
660 num_cblocks[i], plane_factor[i],
663 if (bits <= ctx->bits_per_mb * mbs_per_slice)
667 slice_bits[max_quant + 1] = bits;
668 slice_score[max_quant + 1] = error;
671 td->nodes[trellis_node + max_quant + 1].quant = overquant;
673 bits_limit = mbs * ctx->bits_per_mb;
674 for (pq = min_quant; pq < max_quant + 2; pq++) {
675 prev = trellis_node - TRELLIS_WIDTH + pq;
677 for (q = min_quant; q < max_quant + 2; q++) {
678 cur = trellis_node + q;
680 bits = td->nodes[prev].bits + slice_bits[q];
681 error = slice_score[q];
682 if (bits > bits_limit)
685 if (td->nodes[prev].score < SCORE_LIMIT && error < SCORE_LIMIT)
686 new_score = td->nodes[prev].score + error;
688 new_score = SCORE_LIMIT;
689 if (td->nodes[cur].prev_node == -1 ||
690 td->nodes[cur].score >= new_score) {
692 td->nodes[cur].bits = bits;
693 td->nodes[cur].score = new_score;
694 td->nodes[cur].prev_node = prev;
699 error = td->nodes[trellis_node + min_quant].score;
700 pq = trellis_node + min_quant;
701 for (q = min_quant + 1; q < max_quant + 2; q++) {
702 if (td->nodes[trellis_node + q].score <= error) {
703 error = td->nodes[trellis_node + q].score;
704 pq = trellis_node + q;
711 static int find_quant_thread(AVCodecContext *avctx, void *arg,
712 int jobnr, int threadnr)
714 ProresContext *ctx = avctx->priv_data;
715 ProresThreadData *td = ctx->tdata + threadnr;
716 int mbs_per_slice = ctx->mbs_per_slice;
717 int x, y = jobnr, mb, q = 0;
719 for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) {
720 while (ctx->mb_width - x < mbs_per_slice)
722 q = find_slice_quant(avctx, avctx->coded_frame,
723 (mb + 1) * TRELLIS_WIDTH, x, y,
727 for (x = ctx->slices_width - 1; x >= 0; x--) {
728 ctx->slice_q[x + y * ctx->slices_width] = td->nodes[q].quant;
729 q = td->nodes[q].prev_node;
735 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
736 const AVFrame *pic, int *got_packet)
738 ProresContext *ctx = avctx->priv_data;
739 uint8_t *orig_buf, *buf, *slice_hdr, *slice_sizes, *tmp;
740 uint8_t *picture_size_pos;
742 int x, y, i, mb, q = 0;
743 int sizes[4] = { 0 };
744 int slice_hdr_size = 2 + 2 * (ctx->num_planes - 1);
745 int frame_size, picture_size, slice_size;
749 *avctx->coded_frame = *pic;
750 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
751 avctx->coded_frame->key_frame = 1;
753 pkt_size = ctx->frame_size_upper_bound + FF_MIN_BUFFER_SIZE;
755 if ((ret = ff_alloc_packet(pkt, pkt_size)) < 0) {
756 av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
760 orig_buf = pkt->data;
763 orig_buf += 4; // frame size
764 bytestream_put_be32 (&orig_buf, FRAME_ID); // frame container ID
769 buf += 2; // frame header size will be stored here
770 bytestream_put_be16 (&buf, 0); // version 1
771 bytestream_put_buffer(&buf, ctx->vendor, 4);
772 bytestream_put_be16 (&buf, avctx->width);
773 bytestream_put_be16 (&buf, avctx->height);
775 frame_flags = ctx->chroma_factor << 6;
776 if (avctx->flags & CODEC_FLAG_INTERLACED_DCT)
777 frame_flags |= pic->top_field_first ? 0x04 : 0x08;
778 bytestream_put_byte (&buf, frame_flags);
780 bytestream_put_byte (&buf, 0); // reserved
781 bytestream_put_byte (&buf, avctx->color_primaries);
782 bytestream_put_byte (&buf, avctx->color_trc);
783 bytestream_put_byte (&buf, avctx->colorspace);
784 bytestream_put_byte (&buf, 0x40); // source format and alpha information
785 bytestream_put_byte (&buf, 0); // reserved
786 if (ctx->quant_sel != QUANT_MAT_DEFAULT) {
787 bytestream_put_byte (&buf, 0x03); // matrix flags - both matrices are present
788 // luma quantisation matrix
789 for (i = 0; i < 64; i++)
790 bytestream_put_byte(&buf, ctx->quant_mat[i]);
791 // chroma quantisation matrix
792 for (i = 0; i < 64; i++)
793 bytestream_put_byte(&buf, ctx->quant_mat[i]);
795 bytestream_put_byte (&buf, 0x00); // matrix flags - default matrices are used
797 bytestream_put_be16 (&tmp, buf - orig_buf); // write back frame header size
799 for (ctx->cur_picture_idx = 0;
800 ctx->cur_picture_idx < ctx->pictures_per_frame;
801 ctx->cur_picture_idx++) {
803 picture_size_pos = buf + 1;
804 bytestream_put_byte (&buf, 0x40); // picture header size (in bits)
805 buf += 4; // picture data size will be stored here
806 bytestream_put_be16 (&buf, ctx->slices_per_picture);
807 bytestream_put_byte (&buf, av_log2(ctx->mbs_per_slice) << 4); // slice width and height in MBs
809 // seek table - will be filled during slice encoding
811 buf += ctx->slices_per_picture * 2;
814 if (!ctx->force_quant) {
815 ret = avctx->execute2(avctx, find_quant_thread, NULL, NULL,
821 for (y = 0; y < ctx->mb_height; y++) {
822 int mbs_per_slice = ctx->mbs_per_slice;
823 for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) {
824 q = ctx->force_quant ? ctx->force_quant
825 : ctx->slice_q[mb + y * ctx->slices_width];
827 while (ctx->mb_width - x < mbs_per_slice)
830 bytestream_put_byte(&buf, slice_hdr_size << 3);
832 buf += slice_hdr_size - 1;
833 init_put_bits(&pb, buf, (pkt_size - (buf - orig_buf)) * 8);
834 encode_slice(avctx, pic, &pb, sizes, x, y, q, mbs_per_slice);
836 bytestream_put_byte(&slice_hdr, q);
837 slice_size = slice_hdr_size + sizes[ctx->num_planes - 1];
838 for (i = 0; i < ctx->num_planes - 1; i++) {
839 bytestream_put_be16(&slice_hdr, sizes[i]);
840 slice_size += sizes[i];
842 bytestream_put_be16(&slice_sizes, slice_size);
843 buf += slice_size - slice_hdr_size;
847 if (ctx->pictures_per_frame == 1)
848 picture_size = buf - picture_size_pos - 6;
850 picture_size = buf - picture_size_pos + 1;
851 bytestream_put_be32(&picture_size_pos, picture_size);
855 frame_size = buf - orig_buf;
856 bytestream_put_be32(&orig_buf, frame_size);
858 pkt->size = frame_size;
859 pkt->flags |= AV_PKT_FLAG_KEY;
865 static av_cold int encode_close(AVCodecContext *avctx)
867 ProresContext *ctx = avctx->priv_data;
870 if (avctx->coded_frame->data[0])
871 avctx->release_buffer(avctx, avctx->coded_frame);
873 av_freep(&avctx->coded_frame);
876 for (i = 0; i < avctx->thread_count; i++)
877 av_free(ctx->tdata[i].nodes);
879 av_freep(&ctx->tdata);
880 av_freep(&ctx->slice_q);
885 static av_cold int encode_init(AVCodecContext *avctx)
887 ProresContext *ctx = avctx->priv_data;
890 int min_quant, max_quant;
891 int interlaced = !!(avctx->flags & CODEC_FLAG_INTERLACED_DCT);
893 avctx->bits_per_raw_sample = 10;
894 avctx->coded_frame = avcodec_alloc_frame();
895 if (!avctx->coded_frame)
896 return AVERROR(ENOMEM);
898 ff_proresdsp_init(&ctx->dsp);
899 ff_init_scantable(ctx->dsp.dct_permutation, &ctx->scantable,
900 interlaced ? ff_prores_interlaced_scan
901 : ff_prores_progressive_scan);
903 mps = ctx->mbs_per_slice;
904 if (mps & (mps - 1)) {
905 av_log(avctx, AV_LOG_ERROR,
906 "there should be an integer power of two MBs per slice\n");
907 return AVERROR(EINVAL);
910 ctx->chroma_factor = avctx->pix_fmt == PIX_FMT_YUV422P10
913 ctx->profile_info = prores_profile_info + ctx->profile;
916 ctx->mb_width = FFALIGN(avctx->width, 16) >> 4;
919 ctx->mb_height = FFALIGN(avctx->height, 32) >> 5;
921 ctx->mb_height = FFALIGN(avctx->height, 16) >> 4;
923 ctx->slices_width = ctx->mb_width / mps;
924 ctx->slices_width += av_popcount(ctx->mb_width - ctx->slices_width * mps);
925 ctx->slices_per_picture = ctx->mb_height * ctx->slices_width;
926 ctx->pictures_per_frame = 1 + interlaced;
928 if (ctx->quant_sel == -1)
929 ctx->quant_mat = prores_quant_matrices[ctx->profile_info->quant];
931 ctx->quant_mat = prores_quant_matrices[ctx->quant_sel];
933 if (strlen(ctx->vendor) != 4) {
934 av_log(avctx, AV_LOG_ERROR, "vendor ID should be 4 bytes\n");
935 return AVERROR_INVALIDDATA;
938 ctx->force_quant = avctx->global_quality / FF_QP2LAMBDA;
939 if (!ctx->force_quant) {
940 if (!ctx->bits_per_mb) {
941 for (i = 0; i < NUM_MB_LIMITS - 1; i++)
942 if (prores_mb_limits[i] >= ctx->mb_width * ctx->mb_height *
943 ctx->pictures_per_frame)
945 ctx->bits_per_mb = ctx->profile_info->br_tab[i];
946 } else if (ctx->bits_per_mb < 128) {
947 av_log(avctx, AV_LOG_ERROR, "too few bits per MB, please set at least 128\n");
948 return AVERROR_INVALIDDATA;
951 min_quant = ctx->profile_info->min_quant;
952 max_quant = ctx->profile_info->max_quant;
953 for (i = min_quant; i < MAX_STORED_Q; i++) {
954 for (j = 0; j < 64; j++)
955 ctx->quants[i][j] = ctx->quant_mat[j] * i;
958 ctx->slice_q = av_malloc(ctx->slices_per_picture * sizeof(*ctx->slice_q));
961 return AVERROR(ENOMEM);
964 ctx->tdata = av_mallocz(avctx->thread_count * sizeof(*ctx->tdata));
967 return AVERROR(ENOMEM);
970 for (j = 0; j < avctx->thread_count; j++) {
971 ctx->tdata[j].nodes = av_malloc((ctx->slices_width + 1)
973 * sizeof(*ctx->tdata->nodes));
974 if (!ctx->tdata[j].nodes) {
976 return AVERROR(ENOMEM);
978 for (i = min_quant; i < max_quant + 2; i++) {
979 ctx->tdata[j].nodes[i].prev_node = -1;
980 ctx->tdata[j].nodes[i].bits = 0;
981 ctx->tdata[j].nodes[i].score = 0;
987 if (ctx->force_quant > 64) {
988 av_log(avctx, AV_LOG_ERROR, "too large quantiser, maximum is 64\n");
989 return AVERROR_INVALIDDATA;
992 for (j = 0; j < 64; j++) {
993 ctx->quants[0][j] = ctx->quant_mat[j] * ctx->force_quant;
994 ls += av_log2((1 << 11) / ctx->quants[0][j]) * 2 + 1;
997 ctx->bits_per_mb = ls * 8;
998 if (ctx->chroma_factor == CFACTOR_Y444)
999 ctx->bits_per_mb += ls * 4;
1000 if (ctx->num_planes == 4)
1001 ctx->bits_per_mb += ls * 4;
1004 ctx->frame_size_upper_bound = ctx->pictures_per_frame *
1005 ctx->slices_per_picture *
1006 (2 + 2 * ctx->num_planes +
1007 (mps * ctx->bits_per_mb) / 8)
1010 avctx->codec_tag = ctx->profile_info->tag;
1012 av_log(avctx, AV_LOG_DEBUG,
1013 "profile %d, %d slices, interlacing: %s, %d bits per MB\n",
1014 ctx->profile, ctx->slices_per_picture * ctx->pictures_per_frame,
1015 interlaced ? "yes" : "no", ctx->bits_per_mb);
1016 av_log(avctx, AV_LOG_DEBUG, "frame size upper bound: %d\n",
1017 ctx->frame_size_upper_bound);
1022 #define OFFSET(x) offsetof(ProresContext, x)
1023 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1025 static const AVOption options[] = {
1026 { "mbs_per_slice", "macroblocks per slice", OFFSET(mbs_per_slice),
1027 AV_OPT_TYPE_INT, { 8 }, 1, MAX_MBS_PER_SLICE, VE },
1028 { "profile", NULL, OFFSET(profile), AV_OPT_TYPE_INT,
1029 { PRORES_PROFILE_STANDARD },
1030 PRORES_PROFILE_PROXY, PRORES_PROFILE_HQ, VE, "profile" },
1031 { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_PROXY },
1032 0, 0, VE, "profile" },
1033 { "lt", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_LT },
1034 0, 0, VE, "profile" },
1035 { "standard", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_STANDARD },
1036 0, 0, VE, "profile" },
1037 { "hq", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_HQ },
1038 0, 0, VE, "profile" },
1039 { "vendor", "vendor ID", OFFSET(vendor),
1040 AV_OPT_TYPE_STRING, { .str = "Lavc" }, CHAR_MIN, CHAR_MAX, VE },
1041 { "bits_per_mb", "desired bits per macroblock", OFFSET(bits_per_mb),
1042 AV_OPT_TYPE_INT, { 0 }, 0, 8192, VE },
1043 { "quant_mat", "quantiser matrix", OFFSET(quant_sel), AV_OPT_TYPE_INT,
1044 { -1 }, -1, QUANT_MAT_DEFAULT, VE, "quant_mat" },
1045 { "auto", NULL, 0, AV_OPT_TYPE_CONST, { -1 },
1046 0, 0, VE, "quant_mat" },
1047 { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_PROXY },
1048 0, 0, VE, "quant_mat" },
1049 { "lt", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_LT },
1050 0, 0, VE, "quant_mat" },
1051 { "standard", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_STANDARD },
1052 0, 0, VE, "quant_mat" },
1053 { "hq", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_HQ },
1054 0, 0, VE, "quant_mat" },
1055 { "default", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_DEFAULT },
1056 0, 0, VE, "quant_mat" },
1060 static const AVClass proresenc_class = {
1061 .class_name = "ProRes encoder",
1062 .item_name = av_default_item_name,
1064 .version = LIBAVUTIL_VERSION_INT,
1067 AVCodec ff_prores_encoder = {
1069 .type = AVMEDIA_TYPE_VIDEO,
1070 .id = AV_CODEC_ID_PRORES,
1071 .priv_data_size = sizeof(ProresContext),
1072 .init = encode_init,
1073 .close = encode_close,
1074 .encode2 = encode_frame,
1075 .capabilities = CODEC_CAP_SLICE_THREADS,
1076 .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"),
1077 .pix_fmts = (const enum PixelFormat[]) {
1078 PIX_FMT_YUV422P10, PIX_FMT_YUV444P10, PIX_FMT_NONE
1080 .priv_class = &proresenc_class,