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"
39 #include "proresdata.h"
40 #include "proresdsp.h"
44 const uint8_t *index; ///< pointers to the data of this slice
48 int prev_slice_sf; ///< scalefactor of the previous decoded slice
49 DECLARE_ALIGNED(16, int16_t, blocks)[8 * 4 * 64];
50 DECLARE_ALIGNED(16, int16_t, qmat_luma_scaled)[64];
51 DECLARE_ALIGNED(16, int16_t, qmat_chroma_scaled)[64];
58 int scantable_type; ///< -1 = uninitialized, 0 = progressive, 1/2 = interlaced
60 int frame_type; ///< 0 = progressive, 1 = top-field first, 2 = bottom-field first
61 int pic_format; ///< 2 = 422, 3 = 444
62 uint8_t qmat_luma[64]; ///< dequantization matrix for luma
63 uint8_t qmat_chroma[64]; ///< dequantization matrix for chroma
64 int qmat_changed; ///< 1 - global quantization matrices changed
65 int total_slices; ///< total number of slices in a picture
66 ProresThreadData *slice_data;
70 int num_chroma_blocks; ///< number of chrominance blocks in a macroblock
73 int slice_width_factor;
74 int slice_height_factor;
81 static av_cold int decode_init(AVCodecContext *avctx)
83 ProresContext *ctx = avctx->priv_data;
85 ctx->total_slices = 0;
86 ctx->slice_data = NULL;
88 avctx->bits_per_raw_sample = PRORES_BITS_PER_SAMPLE;
89 ff_proresdsp_init(&ctx->dsp, avctx);
91 ctx->scantable_type = -1; // set scantable type to uninitialized
92 memset(ctx->qmat_luma, 4, 64);
93 memset(ctx->qmat_chroma, 4, 64);
99 static int decode_frame_header(ProresContext *ctx, const uint8_t *buf,
100 const int data_size, AVCodecContext *avctx)
102 int hdr_size, version, width, height, flags;
105 hdr_size = AV_RB16(buf);
106 if (hdr_size > data_size) {
107 av_log(avctx, AV_LOG_ERROR, "frame data too small\n");
108 return AVERROR_INVALIDDATA;
111 version = AV_RB16(buf + 2);
113 av_log(avctx, AV_LOG_ERROR,
114 "unsupported header version: %d\n", version);
115 return AVERROR_INVALIDDATA;
118 width = AV_RB16(buf + 8);
119 height = AV_RB16(buf + 10);
120 if (width != avctx->width || height != avctx->height) {
121 av_log(avctx, AV_LOG_ERROR,
122 "picture dimension changed: old: %d x %d, new: %d x %d\n",
123 avctx->width, avctx->height, width, height);
124 return AVERROR_INVALIDDATA;
127 ctx->frame_type = (buf[12] >> 2) & 3;
128 if (ctx->frame_type > 2) {
129 av_log(avctx, AV_LOG_ERROR,
130 "unsupported frame type: %d\n", ctx->frame_type);
131 return AVERROR_INVALIDDATA;
134 ctx->chroma_factor = (buf[12] >> 6) & 3;
135 ctx->mb_chroma_factor = ctx->chroma_factor + 2;
136 ctx->num_chroma_blocks = (1 << ctx->chroma_factor) >> 1;
137 switch (ctx->chroma_factor) {
139 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
142 avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
145 av_log(avctx, AV_LOG_ERROR,
146 "unsupported picture format: %d\n", ctx->pic_format);
147 return AVERROR_INVALIDDATA;
150 if (ctx->scantable_type != ctx->frame_type) {
151 if (!ctx->frame_type)
152 ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
153 ff_prores_progressive_scan);
155 ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
156 ff_prores_interlaced_scan);
157 ctx->scantable_type = ctx->frame_type;
160 if (ctx->frame_type) { /* if interlaced */
161 ctx->frame->interlaced_frame = 1;
162 ctx->frame->top_field_first = ctx->frame_type & 1;
164 ctx->frame->interlaced_frame = 0;
167 avctx->color_primaries = buf[14];
168 avctx->color_trc = buf[15];
169 avctx->colorspace = buf[16];
171 ctx->alpha_info = buf[17] & 0xf;
173 avpriv_report_missing_feature(avctx, "Alpha channel");
175 ctx->qmat_changed = 0;
179 if (ptr - buf > hdr_size - 64) {
180 av_log(avctx, AV_LOG_ERROR, "header data too small\n");
181 return AVERROR_INVALIDDATA;
183 if (memcmp(ctx->qmat_luma, ptr, 64)) {
184 memcpy(ctx->qmat_luma, ptr, 64);
185 ctx->qmat_changed = 1;
189 memset(ctx->qmat_luma, 4, 64);
190 ctx->qmat_changed = 1;
194 if (ptr - buf > hdr_size - 64) {
195 av_log(avctx, AV_LOG_ERROR, "header data too small\n");
198 if (memcmp(ctx->qmat_chroma, ptr, 64)) {
199 memcpy(ctx->qmat_chroma, ptr, 64);
200 ctx->qmat_changed = 1;
203 memset(ctx->qmat_chroma, 4, 64);
204 ctx->qmat_changed = 1;
211 static int decode_picture_header(ProresContext *ctx, const uint8_t *buf,
212 const int data_size, AVCodecContext *avctx)
214 int i, hdr_size, pic_data_size, num_slices;
215 int slice_width_factor, slice_height_factor;
216 int remainder, num_x_slices;
217 const uint8_t *data_ptr, *index_ptr;
219 hdr_size = data_size > 0 ? buf[0] >> 3 : 0;
220 if (hdr_size < 8 || hdr_size > data_size) {
221 av_log(avctx, AV_LOG_ERROR, "picture header too small\n");
222 return AVERROR_INVALIDDATA;
225 pic_data_size = AV_RB32(buf + 1);
226 if (pic_data_size > data_size) {
227 av_log(avctx, AV_LOG_ERROR, "picture data too small\n");
228 return AVERROR_INVALIDDATA;
231 slice_width_factor = buf[7] >> 4;
232 slice_height_factor = buf[7] & 0xF;
233 if (slice_width_factor > 3 || slice_height_factor) {
234 av_log(avctx, AV_LOG_ERROR,
235 "unsupported slice dimension: %d x %d\n",
236 1 << slice_width_factor, 1 << slice_height_factor);
237 return AVERROR_INVALIDDATA;
240 ctx->slice_width_factor = slice_width_factor;
241 ctx->slice_height_factor = slice_height_factor;
243 ctx->num_x_mbs = (avctx->width + 15) >> 4;
244 ctx->num_y_mbs = (avctx->height +
245 (1 << (4 + ctx->frame->interlaced_frame)) - 1) >>
246 (4 + ctx->frame->interlaced_frame);
248 remainder = ctx->num_x_mbs & ((1 << slice_width_factor) - 1);
249 num_x_slices = (ctx->num_x_mbs >> slice_width_factor) + (remainder & 1) +
250 ((remainder >> 1) & 1) + ((remainder >> 2) & 1);
252 num_slices = num_x_slices * ctx->num_y_mbs;
253 if (num_slices != AV_RB16(buf + 5)) {
254 av_log(avctx, AV_LOG_ERROR, "invalid number of slices\n");
255 return AVERROR_INVALIDDATA;
258 if (ctx->total_slices != num_slices) {
259 av_freep(&ctx->slice_data);
260 ctx->slice_data = av_malloc((num_slices + 1) * sizeof(ctx->slice_data[0]));
261 if (!ctx->slice_data)
262 return AVERROR(ENOMEM);
263 ctx->total_slices = num_slices;
266 if (hdr_size + num_slices * 2 > data_size) {
267 av_log(avctx, AV_LOG_ERROR, "slice table too small\n");
268 return AVERROR_INVALIDDATA;
271 /* parse slice table allowing quick access to the slice data */
272 index_ptr = buf + hdr_size;
273 data_ptr = index_ptr + num_slices * 2;
275 for (i = 0; i < num_slices; i++) {
276 ctx->slice_data[i].index = data_ptr;
277 ctx->slice_data[i].prev_slice_sf = 0;
278 data_ptr += AV_RB16(index_ptr + i * 2);
280 ctx->slice_data[i].index = data_ptr;
281 ctx->slice_data[i].prev_slice_sf = 0;
283 if (data_ptr > buf + data_size) {
284 av_log(avctx, AV_LOG_ERROR, "out of slice data\n");
288 return pic_data_size;
293 * Read an unsigned rice/exp golomb codeword.
295 static inline int decode_vlc_codeword(GetBitContext *gb, unsigned codebook)
297 unsigned int rice_order, exp_order, switch_bits;
298 unsigned int buf, code;
299 int log, prefix_len, len;
302 UPDATE_CACHE(re, gb);
303 buf = GET_CACHE(re, gb);
305 /* number of prefix bits to switch between Rice and expGolomb */
306 switch_bits = (codebook & 3) + 1;
307 rice_order = codebook >> 5; /* rice code order */
308 exp_order = (codebook >> 2) & 7; /* exp golomb code order */
310 log = 31 - av_log2(buf); /* count prefix bits (zeroes) */
312 if (log < switch_bits) { /* ok, we got a rice code */
314 /* shortcut for faster decoding of rice codes without remainder */
316 LAST_SKIP_BITS(re, gb, log + 1);
318 prefix_len = log + 1;
319 code = (log << rice_order) + NEG_USR32(buf << prefix_len, rice_order);
320 LAST_SKIP_BITS(re, gb, prefix_len + rice_order);
322 } else { /* otherwise we got a exp golomb code */
323 len = (log << 1) - switch_bits + exp_order + 1;
324 code = NEG_USR32(buf, len) - (1 << exp_order) + (switch_bits << rice_order);
325 LAST_SKIP_BITS(re, gb, len);
328 CLOSE_READER(re, gb);
333 #define LSB2SIGN(x) (-((x) & 1))
334 #define TOSIGNED(x) (((x) >> 1) ^ LSB2SIGN(x))
337 * Decode DC coefficients for all blocks in a slice.
339 static inline void decode_dc_coeffs(GetBitContext *gb, int16_t *out,
347 code = decode_vlc_codeword(gb, FIRST_DC_CB);
348 out[0] = prev_dc = TOSIGNED(code);
350 out += 64; /* move to the DC coeff of the next block */
353 for (i = 1; i < nblocks; i++, out += 64) {
354 code = decode_vlc_codeword(gb, ff_prores_dc_codebook[FFMIN(FFABS(delta), 3)]);
356 sign = -(((delta >> 15) & 1) ^ (code & 1));
357 delta = (((code + 1) >> 1) ^ sign) - sign;
365 * Decode AC coefficients for all blocks in a slice.
367 static inline void decode_ac_coeffs(GetBitContext *gb, int16_t *out,
368 int blocks_per_slice,
369 int plane_size_factor,
372 int pos, block_mask, run, level, sign, run_cb_index, lev_cb_index;
373 int max_coeffs, bits_left;
375 /* set initial prediction values */
379 max_coeffs = blocks_per_slice << 6;
380 block_mask = blocks_per_slice - 1;
382 for (pos = blocks_per_slice - 1; pos < max_coeffs;) {
383 run_cb_index = ff_prores_run_to_cb_index[FFMIN(run, 15)];
384 lev_cb_index = ff_prores_lev_to_cb_index[FFMIN(level, 9)];
386 bits_left = get_bits_left(gb);
387 if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
390 run = decode_vlc_codeword(gb, ff_prores_ac_codebook[run_cb_index]);
392 bits_left = get_bits_left(gb);
393 if (bits_left <= 0 || (bits_left <= 8 && !show_bits(gb, bits_left)))
396 level = decode_vlc_codeword(gb, ff_prores_ac_codebook[lev_cb_index]) + 1;
399 if (pos >= max_coeffs)
402 sign = get_sbits(gb, 1);
403 out[((pos & block_mask) << 6) + scan[pos >> plane_size_factor]] =
404 (level ^ sign) - sign;
410 * Decode a slice plane (luma or chroma).
412 static void decode_slice_plane(ProresContext *ctx, ProresThreadData *td,
414 int data_size, uint16_t *out_ptr,
415 int linesize, int mbs_per_slice,
416 int blocks_per_mb, int plane_size_factor,
417 const int16_t *qmat, int is_chroma)
421 int mb_num, blocks_per_slice;
423 blocks_per_slice = mbs_per_slice * blocks_per_mb;
425 memset(td->blocks, 0, 8 * 4 * 64 * sizeof(*td->blocks));
427 init_get_bits(&gb, buf, data_size << 3);
429 decode_dc_coeffs(&gb, td->blocks, blocks_per_slice);
431 decode_ac_coeffs(&gb, td->blocks, blocks_per_slice,
432 plane_size_factor, ctx->scantable.permutated);
434 /* inverse quantization, inverse transform and output */
435 block_ptr = td->blocks;
438 for (mb_num = 0; mb_num < mbs_per_slice; mb_num++, out_ptr += blocks_per_mb * 4) {
439 ctx->dsp.idct_put(out_ptr, linesize, block_ptr, qmat);
441 if (blocks_per_mb > 2) {
442 ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat);
445 ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat);
447 if (blocks_per_mb > 2) {
448 ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat);
453 for (mb_num = 0; mb_num < mbs_per_slice; mb_num++, out_ptr += blocks_per_mb * 4) {
454 ctx->dsp.idct_put(out_ptr, linesize, block_ptr, qmat);
456 ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat);
458 if (blocks_per_mb > 2) {
459 ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat);
461 ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat);
469 static int decode_slice(AVCodecContext *avctx, void *tdata)
471 ProresThreadData *td = tdata;
472 ProresContext *ctx = avctx->priv_data;
473 int mb_x_pos = td->x_pos;
474 int mb_y_pos = td->y_pos;
475 int pic_num = ctx->pic_num;
476 int slice_num = td->slice_num;
477 int mbs_per_slice = td->slice_width;
479 uint8_t *y_data, *u_data, *v_data;
480 AVFrame *pic = ctx->frame;
481 int i, sf, slice_width_factor;
482 int slice_data_size, hdr_size, y_data_size, u_data_size, v_data_size;
483 int y_linesize, u_linesize, v_linesize;
485 buf = ctx->slice_data[slice_num].index;
486 slice_data_size = ctx->slice_data[slice_num + 1].index - buf;
488 slice_width_factor = av_log2(mbs_per_slice);
490 y_data = pic->data[0];
491 u_data = pic->data[1];
492 v_data = pic->data[2];
493 y_linesize = pic->linesize[0];
494 u_linesize = pic->linesize[1];
495 v_linesize = pic->linesize[2];
497 if (pic->interlaced_frame) {
498 if (!(pic_num ^ pic->top_field_first)) {
499 y_data += y_linesize;
500 u_data += u_linesize;
501 v_data += v_linesize;
508 if (slice_data_size < 6) {
509 av_log(avctx, AV_LOG_ERROR, "slice data too small\n");
510 return AVERROR_INVALIDDATA;
513 /* parse slice header */
514 hdr_size = buf[0] >> 3;
515 y_data_size = AV_RB16(buf + 2);
516 u_data_size = AV_RB16(buf + 4);
517 v_data_size = hdr_size > 7 ? AV_RB16(buf + 6) :
518 slice_data_size - y_data_size - u_data_size - hdr_size;
520 if (hdr_size + y_data_size + u_data_size + v_data_size > slice_data_size ||
521 v_data_size < 0 || hdr_size < 6) {
522 av_log(avctx, AV_LOG_ERROR, "invalid data size\n");
523 return AVERROR_INVALIDDATA;
526 sf = av_clip(buf[1], 1, 224);
527 sf = sf > 128 ? (sf - 96) << 2 : sf;
529 /* scale quantization matrixes according with slice's scale factor */
530 /* TODO: this can be SIMD-optimized a lot */
531 if (ctx->qmat_changed || sf != td->prev_slice_sf) {
532 td->prev_slice_sf = sf;
533 for (i = 0; i < 64; i++) {
534 td->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf;
535 td->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf;
539 /* decode luma plane */
540 decode_slice_plane(ctx, td, buf + hdr_size, y_data_size,
541 (uint16_t*) (y_data + (mb_y_pos << 4) * y_linesize +
542 (mb_x_pos << 5)), y_linesize,
543 mbs_per_slice, 4, slice_width_factor + 2,
544 td->qmat_luma_scaled, 0);
546 /* decode U chroma plane */
547 decode_slice_plane(ctx, td, buf + hdr_size + y_data_size, u_data_size,
548 (uint16_t*) (u_data + (mb_y_pos << 4) * u_linesize +
549 (mb_x_pos << ctx->mb_chroma_factor)),
550 u_linesize, mbs_per_slice, ctx->num_chroma_blocks,
551 slice_width_factor + ctx->chroma_factor - 1,
552 td->qmat_chroma_scaled, 1);
554 /* decode V chroma plane */
555 decode_slice_plane(ctx, td, buf + hdr_size + y_data_size + u_data_size,
557 (uint16_t*) (v_data + (mb_y_pos << 4) * v_linesize +
558 (mb_x_pos << ctx->mb_chroma_factor)),
559 v_linesize, mbs_per_slice, ctx->num_chroma_blocks,
560 slice_width_factor + ctx->chroma_factor - 1,
561 td->qmat_chroma_scaled, 1);
567 static int decode_picture(ProresContext *ctx, int pic_num,
568 AVCodecContext *avctx)
570 int slice_num, slice_width, x_pos, y_pos;
574 ctx->pic_num = pic_num;
575 for (y_pos = 0; y_pos < ctx->num_y_mbs; y_pos++) {
576 slice_width = 1 << ctx->slice_width_factor;
578 for (x_pos = 0; x_pos < ctx->num_x_mbs && slice_width;
579 x_pos += slice_width) {
580 while (ctx->num_x_mbs - x_pos < slice_width)
583 ctx->slice_data[slice_num].slice_num = slice_num;
584 ctx->slice_data[slice_num].x_pos = x_pos;
585 ctx->slice_data[slice_num].y_pos = y_pos;
586 ctx->slice_data[slice_num].slice_width = slice_width;
592 return avctx->execute(avctx, decode_slice,
593 ctx->slice_data, NULL, slice_num,
594 sizeof(ctx->slice_data[0]));
598 #define MOVE_DATA_PTR(nbytes) buf += (nbytes); buf_size -= (nbytes)
600 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
603 ProresContext *ctx = avctx->priv_data;
604 const uint8_t *buf = avpkt->data;
605 int buf_size = avpkt->size;
606 int frame_hdr_size, pic_num, pic_data_size;
609 ctx->frame->pict_type = AV_PICTURE_TYPE_I;
610 ctx->frame->key_frame = 1;
612 /* check frame atom container */
613 if (buf_size < 28 || buf_size < AV_RB32(buf) ||
614 AV_RB32(buf + 4) != FRAME_ID) {
615 av_log(avctx, AV_LOG_ERROR, "invalid frame\n");
616 return AVERROR_INVALIDDATA;
621 frame_hdr_size = decode_frame_header(ctx, buf, buf_size, avctx);
622 if (frame_hdr_size < 0)
623 return AVERROR_INVALIDDATA;
625 MOVE_DATA_PTR(frame_hdr_size);
627 if (ff_get_buffer(avctx, ctx->frame, 0) < 0)
630 for (pic_num = 0; ctx->frame->interlaced_frame - pic_num + 1; pic_num++) {
631 pic_data_size = decode_picture_header(ctx, buf, buf_size, avctx);
632 if (pic_data_size < 0)
633 return AVERROR_INVALIDDATA;
635 if (decode_picture(ctx, pic_num, avctx))
638 MOVE_DATA_PTR(pic_data_size);
648 static av_cold int decode_close(AVCodecContext *avctx)
650 ProresContext *ctx = avctx->priv_data;
652 av_freep(&ctx->slice_data);
658 AVCodec ff_prores_lgpl_decoder = {
659 .name = "prores_lgpl",
660 .type = AVMEDIA_TYPE_VIDEO,
661 .id = AV_CODEC_ID_PRORES,
662 .priv_data_size = sizeof(ProresContext),
664 .close = decode_close,
665 .decode = decode_frame,
666 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
667 .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)")