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 LONG_BITSTREAM_READER // some ProRes vlc codes require up to 28 bits to be read at once
35 #include "libavutil/intmath.h"
37 #include "proresdata.h"
38 #include "proresdsp.h"
42 const uint8_t *index; ///< pointers to the data of this slice
46 int prev_slice_sf; ///< scalefactor of the previous decoded slice
47 DECLARE_ALIGNED(16, DCTELEM, blocks)[8 * 4 * 64];
48 DECLARE_ALIGNED(16, int16_t, qmat_luma_scaled)[64];
49 DECLARE_ALIGNED(16, int16_t, qmat_chroma_scaled)[64];
56 int scantable_type; ///< -1 = uninitialized, 0 = progressive, 1/2 = interlaced
58 int frame_type; ///< 0 = progressive, 1 = top-field first, 2 = bottom-field first
59 int pic_format; ///< 2 = 422, 3 = 444
60 uint8_t qmat_luma[64]; ///< dequantization matrix for luma
61 uint8_t qmat_chroma[64]; ///< dequantization matrix for chroma
62 int qmat_changed; ///< 1 - global quantization matrices changed
63 int total_slices; ///< total number of slices in a picture
64 ProresThreadData *slice_data;
68 int num_chroma_blocks; ///< number of chrominance blocks in a macroblock
71 int slice_width_factor;
72 int slice_height_factor;
79 static av_cold int decode_init(AVCodecContext *avctx)
81 ProresContext *ctx = avctx->priv_data;
83 ctx->total_slices = 0;
84 ctx->slice_data = NULL;
86 avctx->bits_per_raw_sample = PRORES_BITS_PER_SAMPLE;
87 ff_proresdsp_init(&ctx->dsp);
89 avctx->coded_frame = &ctx->picture;
90 avcodec_get_frame_defaults(&ctx->picture);
91 ctx->picture.type = AV_PICTURE_TYPE_I;
92 ctx->picture.key_frame = 1;
94 ctx->scantable_type = -1; // set scantable type to uninitialized
95 memset(ctx->qmat_luma, 4, 64);
96 memset(ctx->qmat_chroma, 4, 64);
102 static int decode_frame_header(ProresContext *ctx, const uint8_t *buf,
103 const int data_size, AVCodecContext *avctx)
105 int hdr_size, version, width, height, flags;
108 hdr_size = AV_RB16(buf);
109 if (hdr_size > data_size) {
110 av_log(avctx, AV_LOG_ERROR, "frame data too small\n");
111 return AVERROR_INVALIDDATA;
114 version = AV_RB16(buf + 2);
116 av_log(avctx, AV_LOG_ERROR,
117 "unsupported header version: %d\n", version);
118 return AVERROR_INVALIDDATA;
121 width = AV_RB16(buf + 8);
122 height = AV_RB16(buf + 10);
123 if (width != avctx->width || height != avctx->height) {
124 av_log(avctx, AV_LOG_ERROR,
125 "picture dimension changed: old: %d x %d, new: %d x %d\n",
126 avctx->width, avctx->height, width, height);
127 return AVERROR_INVALIDDATA;
130 ctx->frame_type = (buf[12] >> 2) & 3;
131 if (ctx->frame_type > 2) {
132 av_log(avctx, AV_LOG_ERROR,
133 "unsupported frame type: %d\n", ctx->frame_type);
134 return AVERROR_INVALIDDATA;
137 ctx->chroma_factor = (buf[12] >> 6) & 3;
138 ctx->mb_chroma_factor = ctx->chroma_factor + 2;
139 ctx->num_chroma_blocks = (1 << ctx->chroma_factor) >> 1;
140 switch (ctx->chroma_factor) {
142 avctx->pix_fmt = PIX_FMT_YUV422P10;
145 avctx->pix_fmt = PIX_FMT_YUV444P10;
148 av_log(avctx, AV_LOG_ERROR,
149 "unsupported picture format: %d\n", ctx->pic_format);
150 return AVERROR_INVALIDDATA;
153 if (ctx->scantable_type != ctx->frame_type) {
154 if (!ctx->frame_type)
155 ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
156 ff_prores_progressive_scan);
158 ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
159 ff_prores_interlaced_scan);
160 ctx->scantable_type = ctx->frame_type;
163 if (ctx->frame_type) { /* if interlaced */
164 ctx->picture.interlaced_frame = 1;
165 ctx->picture.top_field_first = ctx->frame_type & 1;
168 ctx->alpha_info = buf[17] & 0xf;
170 av_log_missing_feature(avctx, "alpha channel", 0);
172 ctx->qmat_changed = 0;
176 if (ptr - buf > hdr_size - 64) {
177 av_log(avctx, AV_LOG_ERROR, "header data too small\n");
178 return AVERROR_INVALIDDATA;
180 if (memcmp(ctx->qmat_luma, ptr, 64)) {
181 memcpy(ctx->qmat_luma, ptr, 64);
182 ctx->qmat_changed = 1;
186 memset(ctx->qmat_luma, 4, 64);
187 ctx->qmat_changed = 1;
191 if (ptr - buf > hdr_size - 64) {
192 av_log(avctx, AV_LOG_ERROR, "header data too small\n");
195 if (memcmp(ctx->qmat_chroma, ptr, 64)) {
196 memcpy(ctx->qmat_chroma, ptr, 64);
197 ctx->qmat_changed = 1;
200 memset(ctx->qmat_chroma, 4, 64);
201 ctx->qmat_changed = 1;
208 static int decode_picture_header(ProresContext *ctx, const uint8_t *buf,
209 const int data_size, AVCodecContext *avctx)
211 int i, hdr_size, pic_data_size, num_slices;
212 int slice_width_factor, slice_height_factor;
213 int remainder, num_x_slices;
214 const uint8_t *data_ptr, *index_ptr;
216 hdr_size = data_size > 0 ? buf[0] >> 3 : 0;
217 if (hdr_size < 8 || hdr_size > data_size) {
218 av_log(avctx, AV_LOG_ERROR, "picture header too small\n");
219 return AVERROR_INVALIDDATA;
222 pic_data_size = AV_RB32(buf + 1);
223 if (pic_data_size > data_size) {
224 av_log(avctx, AV_LOG_ERROR, "picture data too small\n");
225 return AVERROR_INVALIDDATA;
228 slice_width_factor = buf[7] >> 4;
229 slice_height_factor = buf[7] & 0xF;
230 if (slice_width_factor > 3 || slice_height_factor) {
231 av_log(avctx, AV_LOG_ERROR,
232 "unsupported slice dimension: %d x %d\n",
233 1 << slice_width_factor, 1 << slice_height_factor);
234 return AVERROR_INVALIDDATA;
237 ctx->slice_width_factor = slice_width_factor;
238 ctx->slice_height_factor = slice_height_factor;
240 ctx->num_x_mbs = (avctx->width + 15) >> 4;
241 ctx->num_y_mbs = (avctx->height +
242 (1 << (4 + ctx->picture.interlaced_frame)) - 1) >>
243 (4 + ctx->picture.interlaced_frame);
245 remainder = ctx->num_x_mbs & ((1 << slice_width_factor) - 1);
246 num_x_slices = (ctx->num_x_mbs >> slice_width_factor) + (remainder & 1) +
247 ((remainder >> 1) & 1) + ((remainder >> 2) & 1);
249 num_slices = num_x_slices * ctx->num_y_mbs;
250 if (num_slices != AV_RB16(buf + 5)) {
251 av_log(avctx, AV_LOG_ERROR, "invalid number of slices\n");
252 return AVERROR_INVALIDDATA;
255 if (ctx->total_slices != num_slices) {
256 av_freep(&ctx->slice_data);
257 ctx->slice_data = av_malloc((num_slices + 1) * sizeof(ctx->slice_data[0]));
258 if (!ctx->slice_data)
259 return AVERROR(ENOMEM);
260 ctx->total_slices = num_slices;
263 if (hdr_size + num_slices * 2 > data_size) {
264 av_log(avctx, AV_LOG_ERROR, "slice table too small\n");
265 return AVERROR_INVALIDDATA;
268 /* parse slice table allowing quick access to the slice data */
269 index_ptr = buf + hdr_size;
270 data_ptr = index_ptr + num_slices * 2;
272 for (i = 0; i < num_slices; i++) {
273 ctx->slice_data[i].index = data_ptr;
274 ctx->slice_data[i].prev_slice_sf = 0;
275 data_ptr += AV_RB16(index_ptr + i * 2);
277 ctx->slice_data[i].index = data_ptr;
278 ctx->slice_data[i].prev_slice_sf = 0;
280 if (data_ptr > buf + data_size) {
281 av_log(avctx, AV_LOG_ERROR, "out of slice data\n");
285 return pic_data_size;
290 * Read an unsigned rice/exp golomb codeword.
292 static inline int decode_vlc_codeword(GetBitContext *gb, unsigned codebook)
294 unsigned int rice_order, exp_order, switch_bits;
295 unsigned int buf, code;
296 int log, prefix_len, len;
299 UPDATE_CACHE(re, gb);
300 buf = GET_CACHE(re, gb);
302 /* number of prefix bits to switch between Rice and expGolomb */
303 switch_bits = (codebook & 3) + 1;
304 rice_order = codebook >> 5; /* rice code order */
305 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
307 log = 31 - av_log2(buf); /* count prefix bits (zeroes) */
309 if (log < switch_bits) { /* ok, we got a rice code */
311 /* shortcut for faster decoding of rice codes without remainder */
313 LAST_SKIP_BITS(re, gb, log + 1);
315 prefix_len = log + 1;
316 code = (log << rice_order) + NEG_USR32(buf << prefix_len, rice_order);
317 LAST_SKIP_BITS(re, gb, prefix_len + rice_order);
319 } else { /* otherwise we got a exp golomb code */
320 len = (log << 1) - switch_bits + exp_order + 1;
321 code = NEG_USR32(buf, len) - (1 << exp_order) + (switch_bits << rice_order);
322 LAST_SKIP_BITS(re, gb, len);
325 CLOSE_READER(re, gb);
330 #define LSB2SIGN(x) (-((x) & 1))
331 #define TOSIGNED(x) (((x) >> 1) ^ LSB2SIGN(x))
334 * Decode DC coefficients for all blocks in a slice.
336 static inline void decode_dc_coeffs(GetBitContext *gb, DCTELEM *out,
344 code = decode_vlc_codeword(gb, FIRST_DC_CB);
345 out[0] = prev_dc = TOSIGNED(code);
347 out += 64; /* move to the DC coeff of the next block */
350 for (i = 1; i < nblocks; i++, out += 64) {
351 code = decode_vlc_codeword(gb, ff_prores_dc_codebook[FFMIN(FFABS(delta), 3)]);
353 sign = -(((delta >> 15) & 1) ^ (code & 1));
354 delta = (((code + 1) >> 1) ^ sign) - sign;
362 * Decode AC coefficients for all blocks in a slice.
364 static inline void decode_ac_coeffs(GetBitContext *gb, DCTELEM *out,
365 int blocks_per_slice,
366 int plane_size_factor,
369 int pos, block_mask, run, level, sign, run_cb_index, lev_cb_index;
370 int max_coeffs, bits_left;
372 /* set initial prediction values */
376 max_coeffs = blocks_per_slice << 6;
377 block_mask = blocks_per_slice - 1;
379 for (pos = blocks_per_slice - 1; pos < max_coeffs;) {
380 run_cb_index = ff_prores_run_to_cb_index[FFMIN(run, 15)];
381 lev_cb_index = ff_prores_lev_to_cb_index[FFMIN(level, 9)];
383 bits_left = get_bits_left(gb);
384 if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
387 run = decode_vlc_codeword(gb, ff_prores_ac_codebook[run_cb_index]);
389 bits_left = get_bits_left(gb);
390 if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
393 level = decode_vlc_codeword(gb, ff_prores_ac_codebook[lev_cb_index]) + 1;
396 if (pos >= max_coeffs)
399 sign = get_sbits(gb, 1);
400 out[((pos & block_mask) << 6) + scan[pos >> plane_size_factor]] =
401 (level ^ sign) - sign;
407 * Decode a slice plane (luma or chroma).
409 static void decode_slice_plane(ProresContext *ctx, ProresThreadData *td,
411 int data_size, uint16_t *out_ptr,
412 int linesize, int mbs_per_slice,
413 int blocks_per_mb, int plane_size_factor,
418 int mb_num, blocks_per_slice;
420 blocks_per_slice = mbs_per_slice * blocks_per_mb;
422 memset(td->blocks, 0, 8 * 4 * 64 * sizeof(*td->blocks));
424 init_get_bits(&gb, buf, data_size << 3);
426 decode_dc_coeffs(&gb, td->blocks, blocks_per_slice);
428 decode_ac_coeffs(&gb, td->blocks, blocks_per_slice,
429 plane_size_factor, ctx->scantable.permutated);
431 /* inverse quantization, inverse transform and output */
432 block_ptr = td->blocks;
434 for (mb_num = 0; mb_num < mbs_per_slice; mb_num++, out_ptr += blocks_per_mb * 4) {
435 ctx->dsp.idct_put(out_ptr, linesize, block_ptr, qmat);
437 if (blocks_per_mb > 2) {
438 ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat);
441 ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat);
443 if (blocks_per_mb > 2) {
444 ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat);
451 static int decode_slice(AVCodecContext *avctx, void *tdata)
453 ProresThreadData *td = tdata;
454 ProresContext *ctx = avctx->priv_data;
455 int mb_x_pos = td->x_pos;
456 int mb_y_pos = td->y_pos;
457 int pic_num = ctx->pic_num;
458 int slice_num = td->slice_num;
459 int mbs_per_slice = td->slice_width;
461 uint8_t *y_data, *u_data, *v_data;
462 AVFrame *pic = avctx->coded_frame;
463 int i, sf, slice_width_factor;
464 int slice_data_size, hdr_size, y_data_size, u_data_size, v_data_size;
465 int y_linesize, u_linesize, v_linesize;
467 buf = ctx->slice_data[slice_num].index;
468 slice_data_size = ctx->slice_data[slice_num + 1].index - buf;
470 slice_width_factor = av_log2(mbs_per_slice);
472 y_data = pic->data[0];
473 u_data = pic->data[1];
474 v_data = pic->data[2];
475 y_linesize = pic->linesize[0];
476 u_linesize = pic->linesize[1];
477 v_linesize = pic->linesize[2];
479 if (pic->interlaced_frame) {
480 if (!(pic_num ^ pic->top_field_first)) {
481 y_data += y_linesize;
482 u_data += u_linesize;
483 v_data += v_linesize;
490 if (slice_data_size < 6) {
491 av_log(avctx, AV_LOG_ERROR, "slice data too small\n");
492 return AVERROR_INVALIDDATA;
495 /* parse slice header */
496 hdr_size = buf[0] >> 3;
497 y_data_size = AV_RB16(buf + 2);
498 u_data_size = AV_RB16(buf + 4);
499 v_data_size = hdr_size > 7 ? AV_RB16(buf + 6) :
500 slice_data_size - y_data_size - u_data_size - hdr_size;
502 if (hdr_size + y_data_size + u_data_size + v_data_size > slice_data_size ||
503 v_data_size < 0 || hdr_size < 6) {
504 av_log(avctx, AV_LOG_ERROR, "invalid data size\n");
505 return AVERROR_INVALIDDATA;
508 sf = av_clip(buf[1], 1, 224);
509 sf = sf > 128 ? (sf - 96) << 2 : sf;
511 /* scale quantization matrixes according with slice's scale factor */
512 /* TODO: this can be SIMD-optimized a lot */
513 if (ctx->qmat_changed || sf != td->prev_slice_sf) {
514 td->prev_slice_sf = sf;
515 for (i = 0; i < 64; i++) {
516 td->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf;
517 td->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf;
521 /* decode luma plane */
522 decode_slice_plane(ctx, td, buf + hdr_size, y_data_size,
523 (uint16_t*) (y_data + (mb_y_pos << 4) * y_linesize +
524 (mb_x_pos << 5)), y_linesize,
525 mbs_per_slice, 4, slice_width_factor + 2,
526 td->qmat_luma_scaled);
528 /* decode U chroma plane */
529 decode_slice_plane(ctx, td, buf + hdr_size + y_data_size, u_data_size,
530 (uint16_t*) (u_data + (mb_y_pos << 4) * u_linesize +
531 (mb_x_pos << ctx->mb_chroma_factor)),
532 u_linesize, mbs_per_slice, ctx->num_chroma_blocks,
533 slice_width_factor + ctx->chroma_factor - 1,
534 td->qmat_chroma_scaled);
536 /* decode V chroma plane */
537 decode_slice_plane(ctx, td, buf + hdr_size + y_data_size + u_data_size,
539 (uint16_t*) (v_data + (mb_y_pos << 4) * v_linesize +
540 (mb_x_pos << ctx->mb_chroma_factor)),
541 v_linesize, mbs_per_slice, ctx->num_chroma_blocks,
542 slice_width_factor + ctx->chroma_factor - 1,
543 td->qmat_chroma_scaled);
549 static int decode_picture(ProresContext *ctx, int pic_num,
550 AVCodecContext *avctx)
552 int slice_num, slice_width, x_pos, y_pos;
556 ctx->pic_num = pic_num;
557 for (y_pos = 0; y_pos < ctx->num_y_mbs; y_pos++) {
558 slice_width = 1 << ctx->slice_width_factor;
560 for (x_pos = 0; x_pos < ctx->num_x_mbs && slice_width;
561 x_pos += slice_width) {
562 while (ctx->num_x_mbs - x_pos < slice_width)
565 ctx->slice_data[slice_num].slice_num = slice_num;
566 ctx->slice_data[slice_num].x_pos = x_pos;
567 ctx->slice_data[slice_num].y_pos = y_pos;
568 ctx->slice_data[slice_num].slice_width = slice_width;
574 return avctx->execute(avctx, decode_slice,
575 ctx->slice_data, NULL, slice_num,
576 sizeof(ctx->slice_data[0]));
580 #define MOVE_DATA_PTR(nbytes) buf += (nbytes); buf_size -= (nbytes)
582 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
585 ProresContext *ctx = avctx->priv_data;
586 AVFrame *picture = avctx->coded_frame;
587 const uint8_t *buf = avpkt->data;
588 int buf_size = avpkt->size;
589 int frame_hdr_size, pic_num, pic_data_size;
591 /* check frame atom container */
592 if (buf_size < 28 || buf_size < AV_RB32(buf) ||
593 AV_RB32(buf + 4) != FRAME_ID) {
594 av_log(avctx, AV_LOG_ERROR, "invalid frame\n");
595 return AVERROR_INVALIDDATA;
600 frame_hdr_size = decode_frame_header(ctx, buf, buf_size, avctx);
601 if (frame_hdr_size < 0)
602 return AVERROR_INVALIDDATA;
604 MOVE_DATA_PTR(frame_hdr_size);
606 if (picture->data[0])
607 avctx->release_buffer(avctx, picture);
609 picture->reference = 0;
610 if (avctx->get_buffer(avctx, picture) < 0)
613 for (pic_num = 0; ctx->picture.interlaced_frame - pic_num + 1; pic_num++) {
614 pic_data_size = decode_picture_header(ctx, buf, buf_size, avctx);
615 if (pic_data_size < 0)
616 return AVERROR_INVALIDDATA;
618 if (decode_picture(ctx, pic_num, avctx))
621 MOVE_DATA_PTR(pic_data_size);
624 *data_size = sizeof(AVPicture);
625 *(AVFrame*) data = *avctx->coded_frame;
631 static av_cold int decode_close(AVCodecContext *avctx)
633 ProresContext *ctx = avctx->priv_data;
635 if (ctx->picture.data[0])
636 avctx->release_buffer(avctx, &ctx->picture);
638 av_freep(&ctx->slice_data);
644 AVCodec ff_prores_decoder = {
646 .type = AVMEDIA_TYPE_VIDEO,
647 .id = CODEC_ID_PRORES,
648 .priv_data_size = sizeof(ProresContext),
650 .close = decode_close,
651 .decode = decode_frame,
652 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
653 .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)")