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;
202 const struct prores_profile *profile_info;
206 ProresThreadData *tdata;
209 static void get_slice_data(ProresContext *ctx, const uint16_t *src,
210 int linesize, int x, int y, int w, int h,
211 DCTELEM *blocks, uint16_t *emu_buf,
212 int mbs_per_slice, int blocks_per_mb, int is_chroma)
214 const uint16_t *esrc;
215 const int mb_width = 4 * blocks_per_mb;
219 for (i = 0; i < mbs_per_slice; i++, src += mb_width) {
221 memset(blocks, 0, 64 * (mbs_per_slice - i) * blocks_per_mb
225 if (x + mb_width <= w && y + 16 <= h) {
227 elinesize = linesize;
232 elinesize = 16 * sizeof(*emu_buf);
234 bw = FFMIN(w - x, mb_width);
235 bh = FFMIN(h - y, 16);
237 for (j = 0; j < bh; j++) {
238 memcpy(emu_buf + j * 16,
239 (const uint8_t*)src + j * linesize,
241 pix = emu_buf[j * 16 + bw - 1];
242 for (k = bw; k < mb_width; k++)
243 emu_buf[j * 16 + k] = pix;
246 memcpy(emu_buf + j * 16,
247 emu_buf + (bh - 1) * 16,
248 mb_width * sizeof(*emu_buf));
251 ctx->dsp.fdct(esrc, elinesize, blocks);
253 if (blocks_per_mb > 2) {
254 ctx->dsp.fdct(src + 8, linesize, blocks);
257 ctx->dsp.fdct(src + linesize * 4, linesize, blocks);
259 if (blocks_per_mb > 2) {
260 ctx->dsp.fdct(src + linesize * 4 + 8, linesize, blocks);
264 ctx->dsp.fdct(esrc, elinesize, blocks);
266 ctx->dsp.fdct(src + linesize * 4, linesize, blocks);
268 if (blocks_per_mb > 2) {
269 ctx->dsp.fdct(src + 8, linesize, blocks);
271 ctx->dsp.fdct(src + linesize * 4 + 8, linesize, blocks);
281 * Write an unsigned rice/exp golomb codeword.
283 static inline void encode_vlc_codeword(PutBitContext *pb, unsigned codebook, int val)
285 unsigned int rice_order, exp_order, switch_bits, switch_val;
288 /* number of prefix bits to switch between Rice and expGolomb */
289 switch_bits = (codebook & 3) + 1;
290 rice_order = codebook >> 5; /* rice code order */
291 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
293 switch_val = switch_bits << rice_order;
295 if (val >= switch_val) {
296 val -= switch_val - (1 << exp_order);
297 exponent = av_log2(val);
299 put_bits(pb, exponent - exp_order + switch_bits, 0);
301 put_bits(pb, exponent, val);
303 exponent = val >> rice_order;
306 put_bits(pb, exponent, 0);
309 put_sbits(pb, rice_order, val);
313 #define GET_SIGN(x) ((x) >> 31)
314 #define MAKE_CODE(x) (((x) << 1) ^ GET_SIGN(x))
316 static void encode_dcs(PutBitContext *pb, DCTELEM *blocks,
317 int blocks_per_slice, int scale)
320 int codebook = 3, code, dc, prev_dc, delta, sign, new_sign;
322 prev_dc = (blocks[0] - 0x4000) / scale;
323 encode_vlc_codeword(pb, FIRST_DC_CB, MAKE_CODE(prev_dc));
328 for (i = 1; i < blocks_per_slice; i++, blocks += 64) {
329 dc = (blocks[0] - 0x4000) / scale;
330 delta = dc - prev_dc;
331 new_sign = GET_SIGN(delta);
332 delta = (delta ^ sign) - sign;
333 code = MAKE_CODE(delta);
334 encode_vlc_codeword(pb, ff_prores_dc_codebook[codebook], code);
335 codebook = (code + (code & 1)) >> 1;
336 codebook = FFMIN(codebook, 3);
342 static void encode_acs(PutBitContext *pb, DCTELEM *blocks,
343 int blocks_per_slice,
344 int plane_size_factor,
345 const uint8_t *scan, const int16_t *qmat)
348 int run, level, run_cb, lev_cb;
349 int max_coeffs, abs_level;
351 max_coeffs = blocks_per_slice << 6;
352 run_cb = ff_prores_run_to_cb_index[4];
353 lev_cb = ff_prores_lev_to_cb_index[2];
356 for (i = 1; i < 64; i++) {
357 for (idx = scan[i]; idx < max_coeffs; idx += 64) {
358 level = blocks[idx] / qmat[scan[i]];
360 abs_level = FFABS(level);
361 encode_vlc_codeword(pb, ff_prores_ac_codebook[run_cb], run);
362 encode_vlc_codeword(pb, ff_prores_ac_codebook[lev_cb],
364 put_sbits(pb, 1, GET_SIGN(level));
366 run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)];
367 lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)];
376 static int encode_slice_plane(ProresContext *ctx, PutBitContext *pb,
377 const uint16_t *src, int linesize,
378 int mbs_per_slice, DCTELEM *blocks,
379 int blocks_per_mb, int plane_size_factor,
382 int blocks_per_slice, saved_pos;
384 saved_pos = put_bits_count(pb);
385 blocks_per_slice = mbs_per_slice * blocks_per_mb;
387 encode_dcs(pb, blocks, blocks_per_slice, qmat[0]);
388 encode_acs(pb, blocks, blocks_per_slice, plane_size_factor,
389 ctx->scantable.permutated, qmat);
392 return (put_bits_count(pb) - saved_pos) >> 3;
395 static int encode_slice(AVCodecContext *avctx, const AVFrame *pic,
397 int sizes[4], int x, int y, int quant,
400 ProresContext *ctx = avctx->priv_data;
404 int slice_width_factor = av_log2(mbs_per_slice);
405 int num_cblocks, pwidth;
406 int plane_factor, is_chroma;
409 if (ctx->force_quant) {
410 qmat = ctx->quants[0];
411 } else if (quant < MAX_STORED_Q) {
412 qmat = ctx->quants[quant];
414 qmat = ctx->custom_q;
415 for (i = 0; i < 64; i++)
416 qmat[i] = ctx->quant_mat[i] * quant;
419 for (i = 0; i < ctx->num_planes; i++) {
420 is_chroma = (i == 1 || i == 2);
421 plane_factor = slice_width_factor + 2;
423 plane_factor += ctx->chroma_factor - 3;
424 if (!is_chroma || ctx->chroma_factor == CFACTOR_Y444) {
428 pwidth = avctx->width;
433 pwidth = avctx->width >> 1;
435 src = (const uint16_t*)(pic->data[i] + yp * pic->linesize[i]) + xp;
437 get_slice_data(ctx, src, pic->linesize[i], xp, yp,
438 pwidth, avctx->height, ctx->blocks[0], ctx->emu_buf,
439 mbs_per_slice, num_cblocks, is_chroma);
440 sizes[i] = encode_slice_plane(ctx, pb, src, pic->linesize[i],
441 mbs_per_slice, ctx->blocks[0],
442 num_cblocks, plane_factor,
444 total_size += sizes[i];
449 static inline int estimate_vlc(unsigned codebook, int val)
451 unsigned int rice_order, exp_order, switch_bits, switch_val;
454 /* number of prefix bits to switch between Rice and expGolomb */
455 switch_bits = (codebook & 3) + 1;
456 rice_order = codebook >> 5; /* rice code order */
457 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
459 switch_val = switch_bits << rice_order;
461 if (val >= switch_val) {
462 val -= switch_val - (1 << exp_order);
463 exponent = av_log2(val);
465 return exponent * 2 - exp_order + switch_bits + 1;
467 return (val >> rice_order) + rice_order + 1;
471 static int estimate_dcs(int *error, DCTELEM *blocks, int blocks_per_slice,
475 int codebook = 3, code, dc, prev_dc, delta, sign, new_sign;
478 prev_dc = (blocks[0] - 0x4000) / scale;
479 bits = estimate_vlc(FIRST_DC_CB, MAKE_CODE(prev_dc));
483 *error += FFABS(blocks[0] - 0x4000) % scale;
485 for (i = 1; i < blocks_per_slice; i++, blocks += 64) {
486 dc = (blocks[0] - 0x4000) / scale;
487 *error += FFABS(blocks[0] - 0x4000) % scale;
488 delta = dc - prev_dc;
489 new_sign = GET_SIGN(delta);
490 delta = (delta ^ sign) - sign;
491 code = MAKE_CODE(delta);
492 bits += estimate_vlc(ff_prores_dc_codebook[codebook], code);
493 codebook = (code + (code & 1)) >> 1;
494 codebook = FFMIN(codebook, 3);
502 static int estimate_acs(int *error, DCTELEM *blocks, int blocks_per_slice,
503 int plane_size_factor,
504 const uint8_t *scan, const int16_t *qmat)
507 int run, level, run_cb, lev_cb;
508 int max_coeffs, abs_level;
511 max_coeffs = blocks_per_slice << 6;
512 run_cb = ff_prores_run_to_cb_index[4];
513 lev_cb = ff_prores_lev_to_cb_index[2];
516 for (i = 1; i < 64; i++) {
517 for (idx = scan[i]; idx < max_coeffs; idx += 64) {
518 level = blocks[idx] / qmat[scan[i]];
519 *error += FFABS(blocks[idx]) % qmat[scan[i]];
521 abs_level = FFABS(level);
522 bits += estimate_vlc(ff_prores_ac_codebook[run_cb], run);
523 bits += estimate_vlc(ff_prores_ac_codebook[lev_cb],
526 run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)];
527 lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)];
538 static int estimate_slice_plane(ProresContext *ctx, int *error, int plane,
539 const uint16_t *src, int linesize,
541 int blocks_per_mb, int plane_size_factor,
542 const int16_t *qmat, ProresThreadData *td)
544 int blocks_per_slice;
547 blocks_per_slice = mbs_per_slice * blocks_per_mb;
549 bits = estimate_dcs(error, td->blocks[plane], blocks_per_slice, qmat[0]);
550 bits += estimate_acs(error, td->blocks[plane], blocks_per_slice,
551 plane_size_factor, ctx->scantable.permutated, qmat);
553 return FFALIGN(bits, 8);
556 static int find_slice_quant(AVCodecContext *avctx, const AVFrame *pic,
557 int trellis_node, int x, int y, int mbs_per_slice,
558 ProresThreadData *td)
560 ProresContext *ctx = avctx->priv_data;
561 int i, q, pq, xp, yp;
563 int slice_width_factor = av_log2(mbs_per_slice);
564 int num_cblocks[MAX_PLANES], pwidth;
565 int plane_factor[MAX_PLANES], is_chroma[MAX_PLANES];
566 const int min_quant = ctx->profile_info->min_quant;
567 const int max_quant = ctx->profile_info->max_quant;
568 int error, bits, bits_limit;
569 int mbs, prev, cur, new_score;
570 int slice_bits[TRELLIS_WIDTH], slice_score[TRELLIS_WIDTH];
574 mbs = x + mbs_per_slice;
576 for (i = 0; i < ctx->num_planes; i++) {
577 is_chroma[i] = (i == 1 || i == 2);
578 plane_factor[i] = slice_width_factor + 2;
580 plane_factor[i] += ctx->chroma_factor - 3;
581 if (!is_chroma[i] || ctx->chroma_factor == CFACTOR_Y444) {
585 pwidth = avctx->width;
590 pwidth = avctx->width >> 1;
592 src = (const uint16_t*)(pic->data[i] + yp * pic->linesize[i]) + xp;
594 get_slice_data(ctx, src, pic->linesize[i], xp, yp,
595 pwidth, avctx->height, td->blocks[i], td->emu_buf,
596 mbs_per_slice, num_cblocks[i], is_chroma[i]);
599 for (q = min_quant; q < max_quant + 2; q++) {
600 td->nodes[trellis_node + q].prev_node = -1;
601 td->nodes[trellis_node + q].quant = q;
604 // todo: maybe perform coarser quantising to fit into frame size when needed
605 for (q = min_quant; q <= max_quant; q++) {
608 for (i = 0; i < ctx->num_planes; i++) {
609 bits += estimate_slice_plane(ctx, &error, i,
610 src, pic->linesize[i],
612 num_cblocks[i], plane_factor[i],
615 if (bits > 65000 * 8) {
619 slice_bits[q] = bits;
620 slice_score[q] = error;
622 if (slice_bits[max_quant] <= ctx->bits_per_mb * mbs_per_slice) {
623 slice_bits[max_quant + 1] = slice_bits[max_quant];
624 slice_score[max_quant + 1] = slice_score[max_quant] + 1;
625 overquant = max_quant;
627 for (q = max_quant + 1; q < 128; q++) {
630 if (q < MAX_STORED_Q) {
631 qmat = ctx->quants[q];
634 for (i = 0; i < 64; i++)
635 qmat[i] = ctx->quant_mat[i] * q;
637 for (i = 0; i < ctx->num_planes; i++) {
638 bits += estimate_slice_plane(ctx, &error, i,
639 src, pic->linesize[i],
641 num_cblocks[i], plane_factor[i],
644 if (bits <= ctx->bits_per_mb * mbs_per_slice)
648 slice_bits[max_quant + 1] = bits;
649 slice_score[max_quant + 1] = error;
652 td->nodes[trellis_node + max_quant + 1].quant = overquant;
654 bits_limit = mbs * ctx->bits_per_mb;
655 for (pq = min_quant; pq < max_quant + 2; pq++) {
656 prev = trellis_node - TRELLIS_WIDTH + pq;
658 for (q = min_quant; q < max_quant + 2; q++) {
659 cur = trellis_node + q;
661 bits = td->nodes[prev].bits + slice_bits[q];
662 error = slice_score[q];
663 if (bits > bits_limit)
666 if (td->nodes[prev].score < SCORE_LIMIT && error < SCORE_LIMIT)
667 new_score = td->nodes[prev].score + error;
669 new_score = SCORE_LIMIT;
670 if (td->nodes[cur].prev_node == -1 ||
671 td->nodes[cur].score >= new_score) {
673 td->nodes[cur].bits = bits;
674 td->nodes[cur].score = new_score;
675 td->nodes[cur].prev_node = prev;
680 error = td->nodes[trellis_node + min_quant].score;
681 pq = trellis_node + min_quant;
682 for (q = min_quant + 1; q < max_quant + 2; q++) {
683 if (td->nodes[trellis_node + q].score <= error) {
684 error = td->nodes[trellis_node + q].score;
685 pq = trellis_node + q;
692 static int find_quant_thread(AVCodecContext *avctx, void *arg,
693 int jobnr, int threadnr)
695 ProresContext *ctx = avctx->priv_data;
696 ProresThreadData *td = ctx->tdata + threadnr;
697 int mbs_per_slice = ctx->mbs_per_slice;
698 int x, y = jobnr, mb, q = 0;
700 for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) {
701 while (ctx->mb_width - x < mbs_per_slice)
703 q = find_slice_quant(avctx, avctx->coded_frame,
704 (mb + 1) * TRELLIS_WIDTH, x, y,
708 for (x = ctx->slices_width - 1; x >= 0; x--) {
709 ctx->slice_q[x + y * ctx->slices_width] = td->nodes[q].quant;
710 q = td->nodes[q].prev_node;
716 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
717 const AVFrame *pic, int *got_packet)
719 ProresContext *ctx = avctx->priv_data;
720 uint8_t *orig_buf, *buf, *slice_hdr, *slice_sizes, *tmp;
721 uint8_t *picture_size_pos;
723 int x, y, i, mb, q = 0;
724 int sizes[4] = { 0 };
725 int slice_hdr_size = 2 + 2 * (ctx->num_planes - 1);
726 int frame_size, picture_size, slice_size;
727 int mbs_per_slice = ctx->mbs_per_slice;
730 *avctx->coded_frame = *pic;
731 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
732 avctx->coded_frame->key_frame = 1;
734 pkt_size = ctx->frame_size + FF_MIN_BUFFER_SIZE;
736 if ((ret = ff_alloc_packet(pkt, pkt_size)) < 0) {
737 av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
741 orig_buf = pkt->data;
744 orig_buf += 4; // frame size
745 bytestream_put_be32 (&orig_buf, FRAME_ID); // frame container ID
750 buf += 2; // frame header size will be stored here
751 bytestream_put_be16 (&buf, 0); // version 1
752 bytestream_put_buffer(&buf, ctx->vendor, 4);
753 bytestream_put_be16 (&buf, avctx->width);
754 bytestream_put_be16 (&buf, avctx->height);
755 bytestream_put_byte (&buf, ctx->chroma_factor << 6); // frame flags
756 bytestream_put_byte (&buf, 0); // reserved
757 bytestream_put_byte (&buf, avctx->color_primaries);
758 bytestream_put_byte (&buf, avctx->color_trc);
759 bytestream_put_byte (&buf, avctx->colorspace);
760 bytestream_put_byte (&buf, 0x40); // source format and alpha information
761 bytestream_put_byte (&buf, 0); // reserved
762 if (ctx->quant_sel != QUANT_MAT_DEFAULT) {
763 bytestream_put_byte (&buf, 0x03); // matrix flags - both matrices are present
764 // luma quantisation matrix
765 for (i = 0; i < 64; i++)
766 bytestream_put_byte(&buf, ctx->quant_mat[i]);
767 // chroma quantisation matrix
768 for (i = 0; i < 64; i++)
769 bytestream_put_byte(&buf, ctx->quant_mat[i]);
771 bytestream_put_byte (&buf, 0x00); // matrix flags - default matrices are used
773 bytestream_put_be16 (&tmp, buf - orig_buf); // write back frame header size
776 picture_size_pos = buf + 1;
777 bytestream_put_byte (&buf, 0x40); // picture header size (in bits)
778 buf += 4; // picture data size will be stored here
779 bytestream_put_be16 (&buf, ctx->num_slices); // total number of slices
780 bytestream_put_byte (&buf, av_log2(ctx->mbs_per_slice) << 4); // slice width and height in MBs
782 // seek table - will be filled during slice encoding
784 buf += ctx->num_slices * 2;
787 if (!ctx->force_quant) {
788 ret = avctx->execute2(avctx, find_quant_thread, NULL, NULL,
794 for (y = 0; y < ctx->mb_height; y++) {
795 mbs_per_slice = ctx->mbs_per_slice;
796 for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) {
797 q = ctx->force_quant ? ctx->force_quant
798 : ctx->slice_q[mb + y * ctx->slices_width];
800 while (ctx->mb_width - x < mbs_per_slice)
803 bytestream_put_byte(&buf, slice_hdr_size << 3);
805 buf += slice_hdr_size - 1;
806 init_put_bits(&pb, buf, (pkt_size - (buf - orig_buf)) * 8);
807 encode_slice(avctx, pic, &pb, sizes, x, y, q, mbs_per_slice);
809 bytestream_put_byte(&slice_hdr, q);
810 slice_size = slice_hdr_size + sizes[ctx->num_planes - 1];
811 for (i = 0; i < ctx->num_planes - 1; i++) {
812 bytestream_put_be16(&slice_hdr, sizes[i]);
813 slice_size += sizes[i];
815 bytestream_put_be16(&slice_sizes, slice_size);
816 buf += slice_size - slice_hdr_size;
821 frame_size = buf - orig_buf;
822 picture_size = buf - picture_size_pos - 6;
823 bytestream_put_be32(&orig_buf, frame_size);
824 bytestream_put_be32(&picture_size_pos, picture_size);
826 pkt->size = frame_size;
827 pkt->flags |= AV_PKT_FLAG_KEY;
833 static av_cold int encode_close(AVCodecContext *avctx)
835 ProresContext *ctx = avctx->priv_data;
838 if (avctx->coded_frame->data[0])
839 avctx->release_buffer(avctx, avctx->coded_frame);
841 av_freep(&avctx->coded_frame);
844 for (i = 0; i < avctx->thread_count; i++)
845 av_free(ctx->tdata[i].nodes);
847 av_freep(&ctx->tdata);
848 av_freep(&ctx->slice_q);
853 static av_cold int encode_init(AVCodecContext *avctx)
855 ProresContext *ctx = avctx->priv_data;
858 int min_quant, max_quant;
860 avctx->bits_per_raw_sample = 10;
861 avctx->coded_frame = avcodec_alloc_frame();
862 if (!avctx->coded_frame)
863 return AVERROR(ENOMEM);
865 ff_proresdsp_init(&ctx->dsp);
866 ff_init_scantable(ctx->dsp.dct_permutation, &ctx->scantable,
867 ff_prores_progressive_scan);
869 mps = ctx->mbs_per_slice;
870 if (mps & (mps - 1)) {
871 av_log(avctx, AV_LOG_ERROR,
872 "there should be an integer power of two MBs per slice\n");
873 return AVERROR(EINVAL);
876 ctx->chroma_factor = avctx->pix_fmt == PIX_FMT_YUV422P10
879 ctx->profile_info = prores_profile_info + ctx->profile;
882 ctx->mb_width = FFALIGN(avctx->width, 16) >> 4;
883 ctx->mb_height = FFALIGN(avctx->height, 16) >> 4;
884 ctx->slices_width = ctx->mb_width / mps;
885 ctx->slices_width += av_popcount(ctx->mb_width - ctx->slices_width * mps);
886 ctx->num_slices = ctx->mb_height * ctx->slices_width;
888 if (ctx->quant_sel == -1)
889 ctx->quant_mat = prores_quant_matrices[ctx->profile_info->quant];
891 ctx->quant_mat = prores_quant_matrices[ctx->quant_sel];
893 if (strlen(ctx->vendor) != 4) {
894 av_log(avctx, AV_LOG_ERROR, "vendor ID should be 4 bytes\n");
895 return AVERROR_INVALIDDATA;
898 ctx->force_quant = avctx->global_quality / FF_QP2LAMBDA;
899 if (!ctx->force_quant) {
900 if (!ctx->bits_per_mb) {
901 for (i = 0; i < NUM_MB_LIMITS - 1; i++)
902 if (prores_mb_limits[i] >= ctx->mb_width * ctx->mb_height)
904 ctx->bits_per_mb = ctx->profile_info->br_tab[i];
905 } else if (ctx->bits_per_mb < 128) {
906 av_log(avctx, AV_LOG_ERROR, "too few bits per MB, please set at least 128\n");
907 return AVERROR_INVALIDDATA;
910 min_quant = ctx->profile_info->min_quant;
911 max_quant = ctx->profile_info->max_quant;
912 for (i = min_quant; i < MAX_STORED_Q; i++) {
913 for (j = 0; j < 64; j++)
914 ctx->quants[i][j] = ctx->quant_mat[j] * i;
917 ctx->slice_q = av_malloc(ctx->num_slices * sizeof(*ctx->slice_q));
920 return AVERROR(ENOMEM);
923 ctx->tdata = av_mallocz(avctx->thread_count * sizeof(*ctx->tdata));
926 return AVERROR(ENOMEM);
929 for (j = 0; j < avctx->thread_count; j++) {
930 ctx->tdata[j].nodes = av_malloc((ctx->slices_width + 1)
932 * sizeof(*ctx->tdata->nodes));
933 if (!ctx->tdata[j].nodes) {
935 return AVERROR(ENOMEM);
937 for (i = min_quant; i < max_quant + 2; i++) {
938 ctx->tdata[j].nodes[i].prev_node = -1;
939 ctx->tdata[j].nodes[i].bits = 0;
940 ctx->tdata[j].nodes[i].score = 0;
946 if (ctx->force_quant > 64) {
947 av_log(avctx, AV_LOG_ERROR, "too large quantiser, maximum is 64\n");
948 return AVERROR_INVALIDDATA;
951 for (j = 0; j < 64; j++) {
952 ctx->quants[0][j] = ctx->quant_mat[j] * ctx->force_quant;
953 ls += av_log2((1 << 11) / ctx->quants[0][j]) * 2 + 1;
956 ctx->bits_per_mb = ls * 8;
957 if (ctx->chroma_factor == CFACTOR_Y444)
958 ctx->bits_per_mb += ls * 4;
959 if (ctx->num_planes == 4)
960 ctx->bits_per_mb += ls * 4;
963 ctx->frame_size = ctx->num_slices * (2 + 2 * ctx->num_planes
964 + (2 * mps * ctx->bits_per_mb) / 8)
967 avctx->codec_tag = ctx->profile_info->tag;
969 av_log(avctx, AV_LOG_DEBUG, "profile %d, %d slices, %d bits per MB\n",
970 ctx->profile, ctx->num_slices, ctx->bits_per_mb);
971 av_log(avctx, AV_LOG_DEBUG, "estimated frame size %d\n",
977 #define OFFSET(x) offsetof(ProresContext, x)
978 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
980 static const AVOption options[] = {
981 { "mbs_per_slice", "macroblocks per slice", OFFSET(mbs_per_slice),
982 AV_OPT_TYPE_INT, { 8 }, 1, MAX_MBS_PER_SLICE, VE },
983 { "profile", NULL, OFFSET(profile), AV_OPT_TYPE_INT,
984 { PRORES_PROFILE_STANDARD },
985 PRORES_PROFILE_PROXY, PRORES_PROFILE_HQ, VE, "profile" },
986 { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_PROXY },
987 0, 0, VE, "profile" },
988 { "lt", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_LT },
989 0, 0, VE, "profile" },
990 { "standard", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_STANDARD },
991 0, 0, VE, "profile" },
992 { "hq", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_HQ },
993 0, 0, VE, "profile" },
994 { "vendor", "vendor ID", OFFSET(vendor),
995 AV_OPT_TYPE_STRING, { .str = "Lavc" }, CHAR_MIN, CHAR_MAX, VE },
996 { "bits_per_mb", "desired bits per macroblock", OFFSET(bits_per_mb),
997 AV_OPT_TYPE_INT, { 0 }, 0, 8192, VE },
998 { "quant_mat", "quantiser matrix", OFFSET(quant_sel), AV_OPT_TYPE_INT,
999 { -1 }, -1, QUANT_MAT_DEFAULT, VE, "quant_mat" },
1000 { "auto", NULL, 0, AV_OPT_TYPE_CONST, { -1 },
1001 0, 0, VE, "quant_mat" },
1002 { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_PROXY },
1003 0, 0, VE, "quant_mat" },
1004 { "lt", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_LT },
1005 0, 0, VE, "quant_mat" },
1006 { "standard", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_STANDARD },
1007 0, 0, VE, "quant_mat" },
1008 { "hq", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_HQ },
1009 0, 0, VE, "quant_mat" },
1010 { "default", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_DEFAULT },
1011 0, 0, VE, "quant_mat" },
1015 static const AVClass proresenc_class = {
1016 .class_name = "ProRes encoder",
1017 .item_name = av_default_item_name,
1019 .version = LIBAVUTIL_VERSION_INT,
1022 AVCodec ff_prores_encoder = {
1024 .type = AVMEDIA_TYPE_VIDEO,
1025 .id = CODEC_ID_PRORES,
1026 .priv_data_size = sizeof(ProresContext),
1027 .init = encode_init,
1028 .close = encode_close,
1029 .encode2 = encode_frame,
1030 .capabilities = CODEC_CAP_SLICE_THREADS,
1031 .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"),
1032 .pix_fmts = (const enum PixelFormat[]) {
1033 PIX_FMT_YUV422P10, PIX_FMT_YUV444P10, PIX_FMT_NONE
1035 .priv_class = &proresenc_class,