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);
303 put_bits(pb, exponent, val);
305 exponent = val >> rice_order;
308 put_bits(pb, exponent, 0);
311 put_sbits(pb, rice_order, val);
315 #define GET_SIGN(x) ((x) >> 31)
316 #define MAKE_CODE(x) (((x) << 1) ^ GET_SIGN(x))
318 static void encode_dcs(PutBitContext *pb, DCTELEM *blocks,
319 int blocks_per_slice, int scale)
322 int codebook = 3, code, dc, prev_dc, delta, sign, new_sign;
324 prev_dc = (blocks[0] - 0x4000) / scale;
325 encode_vlc_codeword(pb, FIRST_DC_CB, MAKE_CODE(prev_dc));
330 for (i = 1; i < blocks_per_slice; i++, blocks += 64) {
331 dc = (blocks[0] - 0x4000) / scale;
332 delta = dc - prev_dc;
333 new_sign = GET_SIGN(delta);
334 delta = (delta ^ sign) - sign;
335 code = MAKE_CODE(delta);
336 encode_vlc_codeword(pb, ff_prores_dc_codebook[codebook], code);
337 codebook = (code + (code & 1)) >> 1;
338 codebook = FFMIN(codebook, 3);
344 static void encode_acs(PutBitContext *pb, DCTELEM *blocks,
345 int blocks_per_slice,
346 int plane_size_factor,
347 const uint8_t *scan, const int16_t *qmat)
350 int run, level, run_cb, lev_cb;
351 int max_coeffs, abs_level;
353 max_coeffs = blocks_per_slice << 6;
354 run_cb = ff_prores_run_to_cb_index[4];
355 lev_cb = ff_prores_lev_to_cb_index[2];
358 for (i = 1; i < 64; i++) {
359 for (idx = scan[i]; idx < max_coeffs; idx += 64) {
360 level = blocks[idx] / qmat[scan[i]];
362 abs_level = FFABS(level);
363 encode_vlc_codeword(pb, ff_prores_ac_codebook[run_cb], run);
364 encode_vlc_codeword(pb, ff_prores_ac_codebook[lev_cb],
366 put_sbits(pb, 1, GET_SIGN(level));
368 run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)];
369 lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)];
378 static int encode_slice_plane(ProresContext *ctx, PutBitContext *pb,
379 const uint16_t *src, int linesize,
380 int mbs_per_slice, DCTELEM *blocks,
381 int blocks_per_mb, int plane_size_factor,
384 int blocks_per_slice, saved_pos;
386 saved_pos = put_bits_count(pb);
387 blocks_per_slice = mbs_per_slice * blocks_per_mb;
389 encode_dcs(pb, blocks, blocks_per_slice, qmat[0]);
390 encode_acs(pb, blocks, blocks_per_slice, plane_size_factor,
391 ctx->scantable.permutated, qmat);
394 return (put_bits_count(pb) - saved_pos) >> 3;
397 static int encode_slice(AVCodecContext *avctx, const AVFrame *pic,
399 int sizes[4], int x, int y, int quant,
402 ProresContext *ctx = avctx->priv_data;
406 int slice_width_factor = av_log2(mbs_per_slice);
407 int num_cblocks, pwidth, linesize, line_add;
408 int plane_factor, is_chroma;
411 if (ctx->pictures_per_frame == 1)
414 line_add = ctx->cur_picture_idx ^ !pic->top_field_first;
416 if (ctx->force_quant) {
417 qmat = ctx->quants[0];
418 } else if (quant < MAX_STORED_Q) {
419 qmat = ctx->quants[quant];
421 qmat = ctx->custom_q;
422 for (i = 0; i < 64; i++)
423 qmat[i] = ctx->quant_mat[i] * quant;
426 for (i = 0; i < ctx->num_planes; i++) {
427 is_chroma = (i == 1 || i == 2);
428 plane_factor = slice_width_factor + 2;
430 plane_factor += ctx->chroma_factor - 3;
431 if (!is_chroma || ctx->chroma_factor == CFACTOR_Y444) {
435 pwidth = avctx->width;
440 pwidth = avctx->width >> 1;
443 linesize = pic->linesize[i] * ctx->pictures_per_frame;
444 src = (const uint16_t*)(pic->data[i] + yp * linesize +
445 line_add * pic->linesize[i]) + xp;
447 get_slice_data(ctx, src, linesize, xp, yp,
448 pwidth, avctx->height / ctx->pictures_per_frame,
449 ctx->blocks[0], ctx->emu_buf,
450 mbs_per_slice, num_cblocks, is_chroma);
451 sizes[i] = encode_slice_plane(ctx, pb, src, linesize,
452 mbs_per_slice, ctx->blocks[0],
453 num_cblocks, plane_factor,
455 total_size += sizes[i];
460 static inline int estimate_vlc(unsigned codebook, int val)
462 unsigned int rice_order, exp_order, switch_bits, switch_val;
465 /* number of prefix bits to switch between Rice and expGolomb */
466 switch_bits = (codebook & 3) + 1;
467 rice_order = codebook >> 5; /* rice code order */
468 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
470 switch_val = switch_bits << rice_order;
472 if (val >= switch_val) {
473 val -= switch_val - (1 << exp_order);
474 exponent = av_log2(val);
476 return exponent * 2 - exp_order + switch_bits + 1;
478 return (val >> rice_order) + rice_order + 1;
482 static int estimate_dcs(int *error, DCTELEM *blocks, int blocks_per_slice,
486 int codebook = 3, code, dc, prev_dc, delta, sign, new_sign;
489 prev_dc = (blocks[0] - 0x4000) / scale;
490 bits = estimate_vlc(FIRST_DC_CB, MAKE_CODE(prev_dc));
494 *error += FFABS(blocks[0] - 0x4000) % scale;
496 for (i = 1; i < blocks_per_slice; i++, blocks += 64) {
497 dc = (blocks[0] - 0x4000) / scale;
498 *error += FFABS(blocks[0] - 0x4000) % scale;
499 delta = dc - prev_dc;
500 new_sign = GET_SIGN(delta);
501 delta = (delta ^ sign) - sign;
502 code = MAKE_CODE(delta);
503 bits += estimate_vlc(ff_prores_dc_codebook[codebook], code);
504 codebook = (code + (code & 1)) >> 1;
505 codebook = FFMIN(codebook, 3);
513 static int estimate_acs(int *error, DCTELEM *blocks, int blocks_per_slice,
514 int plane_size_factor,
515 const uint8_t *scan, const int16_t *qmat)
518 int run, level, run_cb, lev_cb;
519 int max_coeffs, abs_level;
522 max_coeffs = blocks_per_slice << 6;
523 run_cb = ff_prores_run_to_cb_index[4];
524 lev_cb = ff_prores_lev_to_cb_index[2];
527 for (i = 1; i < 64; i++) {
528 for (idx = scan[i]; idx < max_coeffs; idx += 64) {
529 level = blocks[idx] / qmat[scan[i]];
530 *error += FFABS(blocks[idx]) % qmat[scan[i]];
532 abs_level = FFABS(level);
533 bits += estimate_vlc(ff_prores_ac_codebook[run_cb], run);
534 bits += estimate_vlc(ff_prores_ac_codebook[lev_cb],
537 run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)];
538 lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)];
549 static int estimate_slice_plane(ProresContext *ctx, int *error, int plane,
550 const uint16_t *src, int linesize,
552 int blocks_per_mb, int plane_size_factor,
553 const int16_t *qmat, ProresThreadData *td)
555 int blocks_per_slice;
558 blocks_per_slice = mbs_per_slice * blocks_per_mb;
560 bits = estimate_dcs(error, td->blocks[plane], blocks_per_slice, qmat[0]);
561 bits += estimate_acs(error, td->blocks[plane], blocks_per_slice,
562 plane_size_factor, ctx->scantable.permutated, qmat);
564 return FFALIGN(bits, 8);
567 static int find_slice_quant(AVCodecContext *avctx, const AVFrame *pic,
568 int trellis_node, int x, int y, int mbs_per_slice,
569 ProresThreadData *td)
571 ProresContext *ctx = avctx->priv_data;
572 int i, q, pq, xp, yp;
574 int slice_width_factor = av_log2(mbs_per_slice);
575 int num_cblocks[MAX_PLANES], pwidth;
576 int plane_factor[MAX_PLANES], is_chroma[MAX_PLANES];
577 const int min_quant = ctx->profile_info->min_quant;
578 const int max_quant = ctx->profile_info->max_quant;
579 int error, bits, bits_limit;
580 int mbs, prev, cur, new_score;
581 int slice_bits[TRELLIS_WIDTH], slice_score[TRELLIS_WIDTH];
584 int linesize[4], line_add;
586 if (ctx->pictures_per_frame == 1)
589 line_add = ctx->cur_picture_idx ^ !pic->top_field_first;
590 mbs = x + mbs_per_slice;
592 for (i = 0; i < ctx->num_planes; i++) {
593 is_chroma[i] = (i == 1 || i == 2);
594 plane_factor[i] = slice_width_factor + 2;
596 plane_factor[i] += ctx->chroma_factor - 3;
597 if (!is_chroma[i] || ctx->chroma_factor == CFACTOR_Y444) {
601 pwidth = avctx->width;
606 pwidth = avctx->width >> 1;
609 linesize[i] = pic->linesize[i] * ctx->pictures_per_frame;
610 src = (const uint16_t*)(pic->data[i] + yp * linesize[i] +
611 line_add * pic->linesize[i]) + xp;
613 get_slice_data(ctx, src, linesize[i], xp, yp,
614 pwidth, avctx->height / ctx->pictures_per_frame,
615 td->blocks[i], td->emu_buf,
616 mbs_per_slice, num_cblocks[i], is_chroma[i]);
619 for (q = min_quant; q < max_quant + 2; q++) {
620 td->nodes[trellis_node + q].prev_node = -1;
621 td->nodes[trellis_node + q].quant = q;
624 // todo: maybe perform coarser quantising to fit into frame size when needed
625 for (q = min_quant; q <= max_quant; q++) {
628 for (i = 0; i < ctx->num_planes; i++) {
629 bits += estimate_slice_plane(ctx, &error, i,
632 num_cblocks[i], plane_factor[i],
635 if (bits > 65000 * 8) {
639 slice_bits[q] = bits;
640 slice_score[q] = error;
642 if (slice_bits[max_quant] <= ctx->bits_per_mb * mbs_per_slice) {
643 slice_bits[max_quant + 1] = slice_bits[max_quant];
644 slice_score[max_quant + 1] = slice_score[max_quant] + 1;
645 overquant = max_quant;
647 for (q = max_quant + 1; q < 128; q++) {
650 if (q < MAX_STORED_Q) {
651 qmat = ctx->quants[q];
654 for (i = 0; i < 64; i++)
655 qmat[i] = ctx->quant_mat[i] * q;
657 for (i = 0; i < ctx->num_planes; i++) {
658 bits += estimate_slice_plane(ctx, &error, i,
661 num_cblocks[i], plane_factor[i],
664 if (bits <= ctx->bits_per_mb * mbs_per_slice)
668 slice_bits[max_quant + 1] = bits;
669 slice_score[max_quant + 1] = error;
672 td->nodes[trellis_node + max_quant + 1].quant = overquant;
674 bits_limit = mbs * ctx->bits_per_mb;
675 for (pq = min_quant; pq < max_quant + 2; pq++) {
676 prev = trellis_node - TRELLIS_WIDTH + pq;
678 for (q = min_quant; q < max_quant + 2; q++) {
679 cur = trellis_node + q;
681 bits = td->nodes[prev].bits + slice_bits[q];
682 error = slice_score[q];
683 if (bits > bits_limit)
686 if (td->nodes[prev].score < SCORE_LIMIT && error < SCORE_LIMIT)
687 new_score = td->nodes[prev].score + error;
689 new_score = SCORE_LIMIT;
690 if (td->nodes[cur].prev_node == -1 ||
691 td->nodes[cur].score >= new_score) {
693 td->nodes[cur].bits = bits;
694 td->nodes[cur].score = new_score;
695 td->nodes[cur].prev_node = prev;
700 error = td->nodes[trellis_node + min_quant].score;
701 pq = trellis_node + min_quant;
702 for (q = min_quant + 1; q < max_quant + 2; q++) {
703 if (td->nodes[trellis_node + q].score <= error) {
704 error = td->nodes[trellis_node + q].score;
705 pq = trellis_node + q;
712 static int find_quant_thread(AVCodecContext *avctx, void *arg,
713 int jobnr, int threadnr)
715 ProresContext *ctx = avctx->priv_data;
716 ProresThreadData *td = ctx->tdata + threadnr;
717 int mbs_per_slice = ctx->mbs_per_slice;
718 int x, y = jobnr, mb, q = 0;
720 for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) {
721 while (ctx->mb_width - x < mbs_per_slice)
723 q = find_slice_quant(avctx, avctx->coded_frame,
724 (mb + 1) * TRELLIS_WIDTH, x, y,
728 for (x = ctx->slices_width - 1; x >= 0; x--) {
729 ctx->slice_q[x + y * ctx->slices_width] = td->nodes[q].quant;
730 q = td->nodes[q].prev_node;
736 static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
737 const AVFrame *pic, int *got_packet)
739 ProresContext *ctx = avctx->priv_data;
740 uint8_t *orig_buf, *buf, *slice_hdr, *slice_sizes, *tmp;
741 uint8_t *picture_size_pos;
743 int x, y, i, mb, q = 0;
744 int sizes[4] = { 0 };
745 int slice_hdr_size = 2 + 2 * (ctx->num_planes - 1);
746 int frame_size, picture_size, slice_size;
750 *avctx->coded_frame = *pic;
751 avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
752 avctx->coded_frame->key_frame = 1;
754 pkt_size = ctx->frame_size_upper_bound + FF_MIN_BUFFER_SIZE;
756 if ((ret = ff_alloc_packet(pkt, pkt_size)) < 0) {
757 av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
761 orig_buf = pkt->data;
764 orig_buf += 4; // frame size
765 bytestream_put_be32 (&orig_buf, FRAME_ID); // frame container ID
770 buf += 2; // frame header size will be stored here
771 bytestream_put_be16 (&buf, 0); // version 1
772 bytestream_put_buffer(&buf, ctx->vendor, 4);
773 bytestream_put_be16 (&buf, avctx->width);
774 bytestream_put_be16 (&buf, avctx->height);
776 frame_flags = ctx->chroma_factor << 6;
777 if (avctx->flags & CODEC_FLAG_INTERLACED_DCT)
778 frame_flags |= pic->top_field_first ? 0x04 : 0x08;
779 bytestream_put_byte (&buf, frame_flags);
781 bytestream_put_byte (&buf, 0); // reserved
782 bytestream_put_byte (&buf, avctx->color_primaries);
783 bytestream_put_byte (&buf, avctx->color_trc);
784 bytestream_put_byte (&buf, avctx->colorspace);
785 bytestream_put_byte (&buf, 0x40); // source format and alpha information
786 bytestream_put_byte (&buf, 0); // reserved
787 if (ctx->quant_sel != QUANT_MAT_DEFAULT) {
788 bytestream_put_byte (&buf, 0x03); // matrix flags - both matrices are present
789 // luma quantisation matrix
790 for (i = 0; i < 64; i++)
791 bytestream_put_byte(&buf, ctx->quant_mat[i]);
792 // chroma quantisation matrix
793 for (i = 0; i < 64; i++)
794 bytestream_put_byte(&buf, ctx->quant_mat[i]);
796 bytestream_put_byte (&buf, 0x00); // matrix flags - default matrices are used
798 bytestream_put_be16 (&tmp, buf - orig_buf); // write back frame header size
800 for (ctx->cur_picture_idx = 0;
801 ctx->cur_picture_idx < ctx->pictures_per_frame;
802 ctx->cur_picture_idx++) {
804 picture_size_pos = buf + 1;
805 bytestream_put_byte (&buf, 0x40); // picture header size (in bits)
806 buf += 4; // picture data size will be stored here
807 bytestream_put_be16 (&buf, ctx->slices_per_picture);
808 bytestream_put_byte (&buf, av_log2(ctx->mbs_per_slice) << 4); // slice width and height in MBs
810 // seek table - will be filled during slice encoding
812 buf += ctx->slices_per_picture * 2;
815 if (!ctx->force_quant) {
816 ret = avctx->execute2(avctx, find_quant_thread, NULL, NULL,
822 for (y = 0; y < ctx->mb_height; y++) {
823 int mbs_per_slice = ctx->mbs_per_slice;
824 for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) {
825 q = ctx->force_quant ? ctx->force_quant
826 : ctx->slice_q[mb + y * ctx->slices_width];
828 while (ctx->mb_width - x < mbs_per_slice)
831 bytestream_put_byte(&buf, slice_hdr_size << 3);
833 buf += slice_hdr_size - 1;
834 init_put_bits(&pb, buf, (pkt_size - (buf - orig_buf)) * 8);
835 encode_slice(avctx, pic, &pb, sizes, x, y, q, mbs_per_slice);
837 bytestream_put_byte(&slice_hdr, q);
838 slice_size = slice_hdr_size + sizes[ctx->num_planes - 1];
839 for (i = 0; i < ctx->num_planes - 1; i++) {
840 bytestream_put_be16(&slice_hdr, sizes[i]);
841 slice_size += sizes[i];
843 bytestream_put_be16(&slice_sizes, slice_size);
844 buf += slice_size - slice_hdr_size;
848 if (ctx->pictures_per_frame == 1)
849 picture_size = buf - picture_size_pos - 6;
851 picture_size = buf - picture_size_pos + 1;
852 bytestream_put_be32(&picture_size_pos, picture_size);
856 frame_size = buf - orig_buf;
857 bytestream_put_be32(&orig_buf, frame_size);
859 pkt->size = frame_size;
860 pkt->flags |= AV_PKT_FLAG_KEY;
866 static av_cold int encode_close(AVCodecContext *avctx)
868 ProresContext *ctx = avctx->priv_data;
871 if (avctx->coded_frame->data[0])
872 avctx->release_buffer(avctx, avctx->coded_frame);
874 av_freep(&avctx->coded_frame);
877 for (i = 0; i < avctx->thread_count; i++)
878 av_free(ctx->tdata[i].nodes);
880 av_freep(&ctx->tdata);
881 av_freep(&ctx->slice_q);
886 static av_cold int encode_init(AVCodecContext *avctx)
888 ProresContext *ctx = avctx->priv_data;
891 int min_quant, max_quant;
892 int interlaced = !!(avctx->flags & CODEC_FLAG_INTERLACED_DCT);
894 avctx->bits_per_raw_sample = 10;
895 avctx->coded_frame = avcodec_alloc_frame();
896 if (!avctx->coded_frame)
897 return AVERROR(ENOMEM);
899 ff_proresdsp_init(&ctx->dsp);
900 ff_init_scantable(ctx->dsp.dct_permutation, &ctx->scantable,
901 interlaced ? ff_prores_interlaced_scan
902 : ff_prores_progressive_scan);
904 mps = ctx->mbs_per_slice;
905 if (mps & (mps - 1)) {
906 av_log(avctx, AV_LOG_ERROR,
907 "there should be an integer power of two MBs per slice\n");
908 return AVERROR(EINVAL);
911 ctx->chroma_factor = avctx->pix_fmt == PIX_FMT_YUV422P10
914 ctx->profile_info = prores_profile_info + ctx->profile;
917 ctx->mb_width = FFALIGN(avctx->width, 16) >> 4;
920 ctx->mb_height = FFALIGN(avctx->height, 32) >> 5;
922 ctx->mb_height = FFALIGN(avctx->height, 16) >> 4;
924 ctx->slices_width = ctx->mb_width / mps;
925 ctx->slices_width += av_popcount(ctx->mb_width - ctx->slices_width * mps);
926 ctx->slices_per_picture = ctx->mb_height * ctx->slices_width;
927 ctx->pictures_per_frame = 1 + interlaced;
929 if (ctx->quant_sel == -1)
930 ctx->quant_mat = prores_quant_matrices[ctx->profile_info->quant];
932 ctx->quant_mat = prores_quant_matrices[ctx->quant_sel];
934 if (strlen(ctx->vendor) != 4) {
935 av_log(avctx, AV_LOG_ERROR, "vendor ID should be 4 bytes\n");
936 return AVERROR_INVALIDDATA;
939 ctx->force_quant = avctx->global_quality / FF_QP2LAMBDA;
940 if (!ctx->force_quant) {
941 if (!ctx->bits_per_mb) {
942 for (i = 0; i < NUM_MB_LIMITS - 1; i++)
943 if (prores_mb_limits[i] >= ctx->mb_width * ctx->mb_height *
944 ctx->pictures_per_frame)
946 ctx->bits_per_mb = ctx->profile_info->br_tab[i];
947 } else if (ctx->bits_per_mb < 128) {
948 av_log(avctx, AV_LOG_ERROR, "too few bits per MB, please set at least 128\n");
949 return AVERROR_INVALIDDATA;
952 min_quant = ctx->profile_info->min_quant;
953 max_quant = ctx->profile_info->max_quant;
954 for (i = min_quant; i < MAX_STORED_Q; i++) {
955 for (j = 0; j < 64; j++)
956 ctx->quants[i][j] = ctx->quant_mat[j] * i;
959 ctx->slice_q = av_malloc(ctx->slices_per_picture * sizeof(*ctx->slice_q));
962 return AVERROR(ENOMEM);
965 ctx->tdata = av_mallocz(avctx->thread_count * sizeof(*ctx->tdata));
968 return AVERROR(ENOMEM);
971 for (j = 0; j < avctx->thread_count; j++) {
972 ctx->tdata[j].nodes = av_malloc((ctx->slices_width + 1)
974 * sizeof(*ctx->tdata->nodes));
975 if (!ctx->tdata[j].nodes) {
977 return AVERROR(ENOMEM);
979 for (i = min_quant; i < max_quant + 2; i++) {
980 ctx->tdata[j].nodes[i].prev_node = -1;
981 ctx->tdata[j].nodes[i].bits = 0;
982 ctx->tdata[j].nodes[i].score = 0;
988 if (ctx->force_quant > 64) {
989 av_log(avctx, AV_LOG_ERROR, "too large quantiser, maximum is 64\n");
990 return AVERROR_INVALIDDATA;
993 for (j = 0; j < 64; j++) {
994 ctx->quants[0][j] = ctx->quant_mat[j] * ctx->force_quant;
995 ls += av_log2((1 << 11) / ctx->quants[0][j]) * 2 + 1;
998 ctx->bits_per_mb = ls * 8;
999 if (ctx->chroma_factor == CFACTOR_Y444)
1000 ctx->bits_per_mb += ls * 4;
1001 if (ctx->num_planes == 4)
1002 ctx->bits_per_mb += ls * 4;
1005 ctx->frame_size_upper_bound = ctx->pictures_per_frame *
1006 ctx->slices_per_picture *
1007 (2 + 2 * ctx->num_planes +
1008 (mps * ctx->bits_per_mb) / 8)
1011 avctx->codec_tag = ctx->profile_info->tag;
1013 av_log(avctx, AV_LOG_DEBUG,
1014 "profile %d, %d slices, interlacing: %s, %d bits per MB\n",
1015 ctx->profile, ctx->slices_per_picture * ctx->pictures_per_frame,
1016 interlaced ? "yes" : "no", ctx->bits_per_mb);
1017 av_log(avctx, AV_LOG_DEBUG, "frame size upper bound: %d\n",
1018 ctx->frame_size_upper_bound);
1023 #define OFFSET(x) offsetof(ProresContext, x)
1024 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1026 static const AVOption options[] = {
1027 { "mbs_per_slice", "macroblocks per slice", OFFSET(mbs_per_slice),
1028 AV_OPT_TYPE_INT, { 8 }, 1, MAX_MBS_PER_SLICE, VE },
1029 { "profile", NULL, OFFSET(profile), AV_OPT_TYPE_INT,
1030 { PRORES_PROFILE_STANDARD },
1031 PRORES_PROFILE_PROXY, PRORES_PROFILE_HQ, VE, "profile" },
1032 { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_PROXY },
1033 0, 0, VE, "profile" },
1034 { "lt", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_LT },
1035 0, 0, VE, "profile" },
1036 { "standard", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_STANDARD },
1037 0, 0, VE, "profile" },
1038 { "hq", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_HQ },
1039 0, 0, VE, "profile" },
1040 { "vendor", "vendor ID", OFFSET(vendor),
1041 AV_OPT_TYPE_STRING, { .str = "Lavc" }, CHAR_MIN, CHAR_MAX, VE },
1042 { "bits_per_mb", "desired bits per macroblock", OFFSET(bits_per_mb),
1043 AV_OPT_TYPE_INT, { 0 }, 0, 8192, VE },
1044 { "quant_mat", "quantiser matrix", OFFSET(quant_sel), AV_OPT_TYPE_INT,
1045 { -1 }, -1, QUANT_MAT_DEFAULT, VE, "quant_mat" },
1046 { "auto", NULL, 0, AV_OPT_TYPE_CONST, { -1 },
1047 0, 0, VE, "quant_mat" },
1048 { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_PROXY },
1049 0, 0, VE, "quant_mat" },
1050 { "lt", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_LT },
1051 0, 0, VE, "quant_mat" },
1052 { "standard", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_STANDARD },
1053 0, 0, VE, "quant_mat" },
1054 { "hq", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_HQ },
1055 0, 0, VE, "quant_mat" },
1056 { "default", NULL, 0, AV_OPT_TYPE_CONST, { QUANT_MAT_DEFAULT },
1057 0, 0, VE, "quant_mat" },
1061 static const AVClass proresenc_class = {
1062 .class_name = "ProRes encoder",
1063 .item_name = av_default_item_name,
1065 .version = LIBAVUTIL_VERSION_INT,
1068 AVCodec ff_prores_encoder = {
1070 .type = AVMEDIA_TYPE_VIDEO,
1071 .id = AV_CODEC_ID_PRORES,
1072 .priv_data_size = sizeof(ProresContext),
1073 .init = encode_init,
1074 .close = encode_close,
1075 .encode2 = encode_frame,
1076 .capabilities = CODEC_CAP_SLICE_THREADS,
1077 .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"),
1078 .pix_fmts = (const enum PixelFormat[]) {
1079 PIX_FMT_YUV422P10, PIX_FMT_YUV444P10, PIX_FMT_NONE
1081 .priv_class = &proresenc_class,