2 * Apple ProRes compatible decoder
4 * Copyright (c) 2010-2011 Maxim Poliakovski
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 * This is a decoder for Apple ProRes 422 SD/HQ/LT/Proxy and ProRes 4444.
26 * It is used for storing and editing high definition video data in Apple's Final Cut Pro.
28 * @see http://wiki.multimedia.cx/index.php?title=Apple_ProRes
31 #define A32_BITSTREAM_READER // some ProRes vlc codes require up to 28 bits to be read at once
35 #include "libavutil/intmath.h"
37 #include "proresdsp.h"
41 const uint8_t *index; ///< pointers to the data of this slice
45 DECLARE_ALIGNED(16, DCTELEM, blocks[8 * 4 * 64]);
52 int scantable_type; ///< -1 = uninitialized, 0 = progressive, 1/2 = interlaced
54 int frame_type; ///< 0 = progressive, 1 = top-field first, 2 = bottom-field first
55 int pic_format; ///< 2 = 422, 3 = 444
56 uint8_t qmat_luma[64]; ///< dequantization matrix for luma
57 uint8_t qmat_chroma[64]; ///< dequantization matrix for chroma
58 int qmat_changed; ///< 1 - global quantization matrices changed
59 int prev_slice_sf; ///< scalefactor of the previous decoded slice
60 DECLARE_ALIGNED(16, int16_t, qmat_luma_scaled[64]);
61 DECLARE_ALIGNED(16, int16_t, qmat_chroma_scaled[64]);
62 int total_slices; ///< total number of slices in a picture
63 ProresThreadData *slice_data;
67 int num_chroma_blocks; ///< number of chrominance blocks in a macroblock
70 int slice_width_factor;
71 int slice_height_factor;
78 static const uint8_t progressive_scan[64] = {
79 0, 1, 8, 9, 2, 3, 10, 11,
80 16, 17, 24, 25, 18, 19, 26, 27,
81 4, 5, 12, 20, 13, 6, 7, 14,
82 21, 28, 29, 22, 15, 23, 30, 31,
83 32, 33, 40, 48, 41, 34, 35, 42,
84 49, 56, 57, 50, 43, 36, 37, 44,
85 51, 58, 59, 52, 45, 38, 39, 46,
86 53, 60, 61, 54, 47, 55, 62, 63
89 static const uint8_t interlaced_scan[64] = {
90 0, 8, 1, 9, 16, 24, 17, 25,
91 2, 10, 3, 11, 18, 26, 19, 27,
92 32, 40, 33, 34, 41, 48, 56, 49,
93 42, 35, 43, 50, 57, 58, 51, 59,
94 4, 12, 5, 6, 13, 20, 28, 21,
95 14, 7, 15, 22, 29, 36, 44, 37,
96 30, 23, 31, 38, 45, 52, 60, 53,
97 46, 39, 47, 54, 61, 62, 55, 63
101 static av_cold int decode_init(AVCodecContext *avctx)
103 ProresContext *ctx = avctx->priv_data;
105 ctx->total_slices = 0;
106 ctx->slice_data = NULL;
108 avctx->pix_fmt = PIX_FMT_YUV422P10; // set default pixel format
110 avctx->bits_per_raw_sample = PRORES_BITS_PER_SAMPLE;
111 ff_proresdsp_init(&ctx->dsp, avctx);
113 avctx->coded_frame = &ctx->picture;
114 avcodec_get_frame_defaults(&ctx->picture);
115 ctx->picture.type = AV_PICTURE_TYPE_I;
116 ctx->picture.key_frame = 1;
118 ctx->scantable_type = -1; // set scantable type to uninitialized
119 memset(ctx->qmat_luma, 4, 64);
120 memset(ctx->qmat_chroma, 4, 64);
121 ctx->prev_slice_sf = 0;
127 static int decode_frame_header(ProresContext *ctx, const uint8_t *buf,
128 const int data_size, AVCodecContext *avctx)
130 int hdr_size, version, width, height, flags;
133 hdr_size = AV_RB16(buf);
134 if (hdr_size > data_size) {
135 av_log(avctx, AV_LOG_ERROR, "frame data too small\n");
136 return AVERROR_INVALIDDATA;
139 version = AV_RB16(buf + 2);
141 av_log(avctx, AV_LOG_ERROR,
142 "unsupported header version: %d\n", version);
143 return AVERROR_INVALIDDATA;
146 width = AV_RB16(buf + 8);
147 height = AV_RB16(buf + 10);
148 if (width != avctx->width || height != avctx->height) {
149 av_log(avctx, AV_LOG_ERROR,
150 "picture dimension changed: old: %d x %d, new: %d x %d\n",
151 avctx->width, avctx->height, width, height);
152 return AVERROR_INVALIDDATA;
155 ctx->frame_type = (buf[12] >> 2) & 3;
156 if (ctx->frame_type > 2) {
157 av_log(avctx, AV_LOG_ERROR,
158 "unsupported frame type: %d\n", ctx->frame_type);
159 return AVERROR_INVALIDDATA;
162 ctx->chroma_factor = (buf[12] >> 6) & 3;
163 ctx->mb_chroma_factor = ctx->chroma_factor + 2;
164 ctx->num_chroma_blocks = (1 << ctx->chroma_factor) >> 1;
165 switch (ctx->chroma_factor) {
167 avctx->pix_fmt = PIX_FMT_YUV422P10;
170 avctx->pix_fmt = PIX_FMT_YUV444P10;
173 av_log(avctx, AV_LOG_ERROR,
174 "unsupported picture format: %d\n", ctx->pic_format);
175 return AVERROR_INVALIDDATA;
178 if (ctx->scantable_type != ctx->frame_type) {
179 if (!ctx->frame_type)
180 ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
183 ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
185 ctx->scantable_type = ctx->frame_type;
188 if (ctx->frame_type) { /* if interlaced */
189 ctx->picture.interlaced_frame = 1;
190 ctx->picture.top_field_first = ctx->frame_type & 1;
193 ctx->alpha_info = buf[17] & 0xf;
195 av_log_missing_feature(avctx, "alpha channel", 0);
197 ctx->qmat_changed = 0;
201 if (ptr - buf > hdr_size - 64) {
202 av_log(avctx, AV_LOG_ERROR, "header data too small\n");
203 return AVERROR_INVALIDDATA;
205 if (memcmp(ctx->qmat_luma, ptr, 64)) {
206 memcpy(ctx->qmat_luma, ptr, 64);
207 ctx->qmat_changed = 1;
211 memset(ctx->qmat_luma, 4, 64);
212 ctx->qmat_changed = 1;
216 if (ptr - buf > hdr_size - 64) {
217 av_log(avctx, AV_LOG_ERROR, "header data too small\n");
220 if (memcmp(ctx->qmat_chroma, ptr, 64)) {
221 memcpy(ctx->qmat_chroma, ptr, 64);
222 ctx->qmat_changed = 1;
225 memset(ctx->qmat_chroma, 4, 64);
226 ctx->qmat_changed = 1;
233 static int decode_picture_header(ProresContext *ctx, const uint8_t *buf,
234 const int data_size, AVCodecContext *avctx)
236 int i, hdr_size, pic_data_size, num_slices;
237 int slice_width_factor, slice_height_factor;
238 int remainder, num_x_slices;
239 const uint8_t *data_ptr, *index_ptr;
241 hdr_size = data_size > 0 ? buf[0] >> 3 : 0;
242 if (hdr_size < 8 || hdr_size > data_size) {
243 av_log(avctx, AV_LOG_ERROR, "picture header too small\n");
244 return AVERROR_INVALIDDATA;
247 pic_data_size = AV_RB32(buf + 1);
248 if (pic_data_size > data_size) {
249 av_log(avctx, AV_LOG_ERROR, "picture data too small\n");
250 return AVERROR_INVALIDDATA;
253 slice_width_factor = buf[7] >> 4;
254 slice_height_factor = buf[7] & 0xF;
255 if (slice_width_factor > 3 || slice_height_factor) {
256 av_log(avctx, AV_LOG_ERROR,
257 "unsupported slice dimension: %d x %d\n",
258 1 << slice_width_factor, 1 << slice_height_factor);
259 return AVERROR_INVALIDDATA;
262 ctx->slice_width_factor = slice_width_factor;
263 ctx->slice_height_factor = slice_height_factor;
265 ctx->num_x_mbs = (avctx->width + 15) >> 4;
266 ctx->num_y_mbs = (avctx->height +
267 (1 << (4 + ctx->picture.interlaced_frame)) - 1) >>
268 (4 + ctx->picture.interlaced_frame);
270 remainder = ctx->num_x_mbs & ((1 << slice_width_factor) - 1);
271 num_x_slices = (ctx->num_x_mbs >> slice_width_factor) + (remainder & 1) +
272 ((remainder >> 1) & 1) + ((remainder >> 2) & 1);
274 num_slices = num_x_slices * ctx->num_y_mbs;
275 if (num_slices != AV_RB16(buf + 5)) {
276 av_log(avctx, AV_LOG_ERROR, "invalid number of slices\n");
277 return AVERROR_INVALIDDATA;
280 if (ctx->total_slices != num_slices) {
281 av_freep(&ctx->slice_data);
282 ctx->slice_data = av_malloc((num_slices + 1) * sizeof(ctx->slice_data[0]));
283 if (!ctx->slice_data)
284 return AVERROR(ENOMEM);
285 ctx->total_slices = num_slices;
288 if (hdr_size + num_slices * 2 > data_size) {
289 av_log(avctx, AV_LOG_ERROR, "slice table too small\n");
290 return AVERROR_INVALIDDATA;
293 /* parse slice table allowing quick access to the slice data */
294 index_ptr = buf + hdr_size;
295 data_ptr = index_ptr + num_slices * 2;
297 for (i = 0; i < num_slices; i++) {
298 ctx->slice_data[i].index = data_ptr;
299 data_ptr += AV_RB16(index_ptr + i * 2);
301 ctx->slice_data[i].index = data_ptr;
303 if (data_ptr > buf + data_size) {
304 av_log(avctx, AV_LOG_ERROR, "out of slice data\n");
308 return pic_data_size;
313 * Read an unsigned rice/exp golomb codeword.
315 static inline int decode_vlc_codeword(GetBitContext *gb, uint8_t codebook)
317 unsigned int rice_order, exp_order, switch_bits;
318 unsigned int buf, code;
319 int log, prefix_len, len;
322 UPDATE_CACHE(re, gb);
323 buf = GET_CACHE(re, gb);
325 /* number of prefix bits to switch between Rice and expGolomb */
326 switch_bits = (codebook & 3) + 1;
327 rice_order = codebook >> 5; /* rice code order */
328 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
330 log = 31 - av_log2(buf); /* count prefix bits (zeroes) */
332 if (log < switch_bits) { /* ok, we got a rice code */
334 /* shortcut for faster decoding of rice codes without remainder */
336 LAST_SKIP_BITS(re, gb, log + 1);
338 prefix_len = log + 1;
339 code = (log << rice_order) + NEG_USR32(buf << prefix_len, rice_order);
340 LAST_SKIP_BITS(re, gb, prefix_len + rice_order);
342 } else { /* otherwise we got a exp golomb code */
343 len = (log << 1) - switch_bits + exp_order + 1;
344 code = NEG_USR32(buf, len) - (1 << exp_order) + (switch_bits << rice_order);
345 LAST_SKIP_BITS(re, gb, len);
348 CLOSE_READER(re, gb);
353 #define LSB2SIGN(x) (-((x) & 1))
354 #define TOSIGNED(x) (((x) >> 1) ^ LSB2SIGN(x))
356 #define FIRST_DC_CB 0xB8 // rice_order = 5, exp_golomb_order = 6, switch_bits = 0
358 static uint8_t dc_codebook[4] = {
359 0x04, // rice_order = 0, exp_golomb_order = 1, switch_bits = 0
360 0x28, // rice_order = 1, exp_golomb_order = 2, switch_bits = 0
361 0x4D, // rice_order = 2, exp_golomb_order = 3, switch_bits = 1
362 0x70 // rice_order = 3, exp_golomb_order = 4, switch_bits = 0
367 * Decode DC coefficients for all blocks in a slice.
369 static inline void decode_dc_coeffs(GetBitContext *gb, DCTELEM *out,
377 code = decode_vlc_codeword(gb, FIRST_DC_CB);
378 out[0] = prev_dc = TOSIGNED(code);
380 out += 64; /* move to the DC coeff of the next block */
383 for (i = 1; i < nblocks; i++, out += 64) {
384 code = decode_vlc_codeword(gb, dc_codebook[FFMIN(FFABS(delta), 3)]);
386 sign = -(((delta >> 15) & 1) ^ (code & 1));
387 delta = (((code + 1) >> 1) ^ sign) - sign;
394 static uint8_t ac_codebook[7] = {
395 0x04, // rice_order = 0, exp_golomb_order = 1, switch_bits = 0
396 0x28, // rice_order = 1, exp_golomb_order = 2, switch_bits = 0
397 0x4C, // rice_order = 2, exp_golomb_order = 3, switch_bits = 0
398 0x05, // rice_order = 0, exp_golomb_order = 1, switch_bits = 1
399 0x29, // rice_order = 1, exp_golomb_order = 2, switch_bits = 1
400 0x06, // rice_order = 0, exp_golomb_order = 1, switch_bits = 2
401 0x0A, // rice_order = 0, exp_golomb_order = 2, switch_bits = 2
405 * Lookup tables for adaptive switching between codebooks
406 * according with previous run/level value.
408 static uint8_t run_to_cb_index[16] =
409 { 5, 5, 3, 3, 0, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 2 };
411 static uint8_t lev_to_cb_index[10] = { 0, 6, 3, 5, 0, 1, 1, 1, 1, 2 };
415 * Decode AC coefficients for all blocks in a slice.
417 static inline void decode_ac_coeffs(GetBitContext *gb, DCTELEM *out,
418 int blocks_per_slice,
419 int plane_size_factor,
422 int pos, block_mask, run, level, sign, run_cb_index, lev_cb_index;
423 int max_coeffs, bits_left;
425 /* set initial prediction values */
429 max_coeffs = blocks_per_slice << 6;
430 block_mask = blocks_per_slice - 1;
432 for (pos = blocks_per_slice - 1; pos < max_coeffs;) {
433 run_cb_index = run_to_cb_index[FFMIN(run, 15)];
434 lev_cb_index = lev_to_cb_index[FFMIN(level, 9)];
436 bits_left = get_bits_left(gb);
437 if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
440 run = decode_vlc_codeword(gb, ac_codebook[run_cb_index]);
442 bits_left = get_bits_left(gb);
443 if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
446 level = decode_vlc_codeword(gb, ac_codebook[lev_cb_index]) + 1;
449 if (pos >= max_coeffs)
452 sign = get_sbits(gb, 1);
453 out[((pos & block_mask) << 6) + scan[pos >> plane_size_factor]] =
454 (level ^ sign) - sign;
460 * Decode a slice plane (luma or chroma).
462 static void decode_slice_plane(ProresContext *ctx, ProresThreadData *td,
464 int data_size, uint16_t *out_ptr,
465 int linesize, int mbs_per_slice,
466 int blocks_per_mb, int plane_size_factor,
471 int mb_num, blocks_per_slice;
473 blocks_per_slice = mbs_per_slice * blocks_per_mb;
475 memset(td->blocks, 0, 8 * 4 * 64 * sizeof(*td->blocks));
477 init_get_bits(&gb, buf, data_size << 3);
479 decode_dc_coeffs(&gb, td->blocks, blocks_per_slice);
481 decode_ac_coeffs(&gb, td->blocks, blocks_per_slice,
482 plane_size_factor, ctx->scantable.permutated);
484 /* inverse quantization, inverse transform and output */
485 block_ptr = td->blocks;
487 for (mb_num = 0; mb_num < mbs_per_slice; mb_num++, out_ptr += blocks_per_mb * 4) {
488 ctx->dsp.idct_put(out_ptr, linesize, block_ptr, qmat);
490 if (blocks_per_mb > 2) {
491 ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat);
494 ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat);
496 if (blocks_per_mb > 2) {
497 ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat);
504 static int decode_slice(AVCodecContext *avctx, ProresThreadData *td)
506 ProresContext *ctx = avctx->priv_data;
507 int mb_x_pos = td->x_pos;
508 int mb_y_pos = td->y_pos;
509 int pic_num = ctx->pic_num;
510 int slice_num = td->slice_num;
511 int mbs_per_slice = td->slice_width;
513 uint8_t *y_data, *u_data, *v_data;
514 AVFrame *pic = avctx->coded_frame;
515 int i, sf, slice_width_factor;
516 int slice_data_size, hdr_size, y_data_size, u_data_size, v_data_size;
517 int y_linesize, u_linesize, v_linesize;
519 buf = ctx->slice_data[slice_num].index;
520 slice_data_size = ctx->slice_data[slice_num + 1].index - buf;
522 slice_width_factor = av_log2(mbs_per_slice);
524 y_data = pic->data[0];
525 u_data = pic->data[1];
526 v_data = pic->data[2];
527 y_linesize = pic->linesize[0];
528 u_linesize = pic->linesize[1];
529 v_linesize = pic->linesize[2];
531 if (pic->interlaced_frame) {
532 if (!(pic_num ^ pic->top_field_first)) {
533 y_data += y_linesize;
534 u_data += u_linesize;
535 v_data += v_linesize;
542 if (slice_data_size < 6) {
543 av_log(avctx, AV_LOG_ERROR, "slice data too small\n");
544 return AVERROR_INVALIDDATA;
547 /* parse slice header */
548 hdr_size = buf[0] >> 3;
549 y_data_size = AV_RB16(buf + 2);
550 u_data_size = AV_RB16(buf + 4);
551 v_data_size = slice_data_size - y_data_size - u_data_size - hdr_size;
553 if (v_data_size < 0 || hdr_size < 6) {
554 av_log(avctx, AV_LOG_ERROR, "invalid data size\n");
555 return AVERROR_INVALIDDATA;
558 sf = av_clip(buf[1], 1, 224);
559 sf = sf > 128 ? (sf - 96) << 2 : sf;
561 /* scale quantization matrixes according with slice's scale factor */
562 /* TODO: this can be SIMD-optimized alot */
563 if (ctx->qmat_changed || sf != ctx->prev_slice_sf) {
564 ctx->prev_slice_sf = sf;
565 for (i = 0; i < 64; i++) {
566 ctx->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf;
567 ctx->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf;
571 /* decode luma plane */
572 decode_slice_plane(ctx, td, buf + hdr_size, y_data_size,
573 (uint16_t*) (y_data + (mb_y_pos << 4) * y_linesize +
574 (mb_x_pos << 5)), y_linesize,
575 mbs_per_slice, 4, slice_width_factor + 2,
576 ctx->qmat_luma_scaled);
578 /* decode U chroma plane */
579 decode_slice_plane(ctx, td, buf + hdr_size + y_data_size, u_data_size,
580 (uint16_t*) (u_data + (mb_y_pos << 4) * u_linesize +
581 (mb_x_pos << ctx->mb_chroma_factor)),
582 u_linesize, mbs_per_slice, ctx->num_chroma_blocks,
583 slice_width_factor + ctx->chroma_factor - 1,
584 ctx->qmat_chroma_scaled);
586 /* decode V chroma plane */
587 decode_slice_plane(ctx, td, buf + hdr_size + y_data_size + u_data_size,
589 (uint16_t*) (v_data + (mb_y_pos << 4) * v_linesize +
590 (mb_x_pos << ctx->mb_chroma_factor)),
591 v_linesize, mbs_per_slice, ctx->num_chroma_blocks,
592 slice_width_factor + ctx->chroma_factor - 1,
593 ctx->qmat_chroma_scaled);
599 static int decode_picture(ProresContext *ctx, int pic_num,
600 AVCodecContext *avctx)
602 int slice_num, slice_width, x_pos, y_pos;
606 ctx->pic_num = pic_num;
607 for (y_pos = 0; y_pos < ctx->num_y_mbs; y_pos++) {
608 slice_width = 1 << ctx->slice_width_factor;
610 for (x_pos = 0; x_pos < ctx->num_x_mbs && slice_width;
611 x_pos += slice_width) {
612 while (ctx->num_x_mbs - x_pos < slice_width)
615 ctx->slice_data[slice_num].slice_num = slice_num;
616 ctx->slice_data[slice_num].x_pos = x_pos;
617 ctx->slice_data[slice_num].y_pos = y_pos;
618 ctx->slice_data[slice_num].slice_width = slice_width;
624 return avctx->execute(avctx, (void *) decode_slice,
625 ctx->slice_data, NULL, slice_num,
626 sizeof(ctx->slice_data[0]));
630 #define FRAME_ID MKBETAG('i', 'c', 'p', 'f')
631 #define MOVE_DATA_PTR(nbytes) buf += (nbytes); buf_size -= (nbytes)
633 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
636 ProresContext *ctx = avctx->priv_data;
637 AVFrame *picture = avctx->coded_frame;
638 const uint8_t *buf = avpkt->data;
639 int buf_size = avpkt->size;
640 int frame_hdr_size, pic_num, pic_data_size;
642 /* check frame atom container */
643 if (buf_size < 28 || buf_size < AV_RB32(buf) ||
644 AV_RB32(buf + 4) != FRAME_ID) {
645 av_log(avctx, AV_LOG_ERROR, "invalid frame\n");
646 return AVERROR_INVALIDDATA;
651 frame_hdr_size = decode_frame_header(ctx, buf, buf_size, avctx);
652 if (frame_hdr_size < 0)
653 return AVERROR_INVALIDDATA;
655 MOVE_DATA_PTR(frame_hdr_size);
657 if (picture->data[0])
658 avctx->release_buffer(avctx, picture);
660 picture->reference = 0;
661 if (avctx->get_buffer(avctx, picture) < 0)
664 for (pic_num = 0; ctx->picture.interlaced_frame - pic_num + 1; pic_num++) {
665 pic_data_size = decode_picture_header(ctx, buf, buf_size, avctx);
666 if (pic_data_size < 0)
667 return AVERROR_INVALIDDATA;
669 if (decode_picture(ctx, pic_num, avctx))
672 MOVE_DATA_PTR(pic_data_size);
675 *data_size = sizeof(AVPicture);
676 *(AVFrame*) data = *avctx->coded_frame;
682 static av_cold int decode_close(AVCodecContext *avctx)
684 ProresContext *ctx = avctx->priv_data;
686 if (ctx->picture.data[0])
687 avctx->release_buffer(avctx, &ctx->picture);
689 av_freep(&ctx->slice_data);
695 AVCodec ff_prores_lgpl_decoder = {
696 .name = "prores_lgpl",
697 .type = AVMEDIA_TYPE_VIDEO,
698 .id = CODEC_ID_PRORES,
699 .priv_data_size = sizeof(ProresContext),
701 .close = decode_close,
702 .decode = decode_frame,
703 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
704 .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)")