blob: bc5bdb5a4d5ee129bb0222bf2285a5013b6aa691
1 | /* |
2 | * Apple ProRes compatible decoder |
3 | * |
4 | * Copyright (c) 2010-2011 Maxim Poliakovski |
5 | * |
6 | * This file is part of FFmpeg. |
7 | * |
8 | * FFmpeg 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. |
12 | * |
13 | * FFmpeg 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. |
17 | * |
18 | * You should have received a copy of the GNU Lesser General Public |
19 | * License along with FFmpeg; if not, write to the Free Software |
20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
21 | */ |
22 | |
23 | /** |
24 | * @file |
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. |
27 | * |
28 | * @see http://wiki.multimedia.cx/index.php?title=Apple_ProRes |
29 | */ |
30 | |
31 | #define LONG_BITSTREAM_READER // some ProRes vlc codes require up to 28 bits to be read at once |
32 | |
33 | #include <stdint.h> |
34 | |
35 | #include "libavutil/intmath.h" |
36 | #include "avcodec.h" |
37 | #include "idctdsp.h" |
38 | #include "internal.h" |
39 | #include "proresdata.h" |
40 | #include "proresdsp.h" |
41 | #include "get_bits.h" |
42 | |
43 | typedef struct ProresThreadData { |
44 | const uint8_t *index; ///< pointers to the data of this slice |
45 | int slice_num; |
46 | int x_pos, y_pos; |
47 | int slice_width; |
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]; |
52 | } ProresThreadData; |
53 | |
54 | typedef struct ProresContext { |
55 | ProresDSPContext dsp; |
56 | AVFrame *frame; |
57 | ScanTable scantable; |
58 | int scantable_type; ///< -1 = uninitialized, 0 = progressive, 1/2 = interlaced |
59 | |
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; |
67 | int pic_num; |
68 | int chroma_factor; |
69 | int mb_chroma_factor; |
70 | int num_chroma_blocks; ///< number of chrominance blocks in a macroblock |
71 | int num_x_slices; |
72 | int num_y_slices; |
73 | int slice_width_factor; |
74 | int slice_height_factor; |
75 | int num_x_mbs; |
76 | int num_y_mbs; |
77 | int alpha_info; |
78 | } ProresContext; |
79 | |
80 | |
81 | static av_cold int decode_init(AVCodecContext *avctx) |
82 | { |
83 | ProresContext *ctx = avctx->priv_data; |
84 | |
85 | ctx->total_slices = 0; |
86 | ctx->slice_data = NULL; |
87 | |
88 | avctx->bits_per_raw_sample = PRORES_BITS_PER_SAMPLE; |
89 | ff_proresdsp_init(&ctx->dsp, avctx); |
90 | |
91 | ctx->scantable_type = -1; // set scantable type to uninitialized |
92 | memset(ctx->qmat_luma, 4, 64); |
93 | memset(ctx->qmat_chroma, 4, 64); |
94 | |
95 | return 0; |
96 | } |
97 | |
98 | |
99 | static int decode_frame_header(ProresContext *ctx, const uint8_t *buf, |
100 | const int data_size, AVCodecContext *avctx) |
101 | { |
102 | int hdr_size, version, width, height, flags; |
103 | const uint8_t *ptr; |
104 | |
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; |
109 | } |
110 | |
111 | version = AV_RB16(buf + 2); |
112 | if (version >= 2) { |
113 | av_log(avctx, AV_LOG_ERROR, |
114 | "unsupported header version: %d\n", version); |
115 | return AVERROR_INVALIDDATA; |
116 | } |
117 | |
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; |
125 | } |
126 | |
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; |
132 | } |
133 | |
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 | ctx->alpha_info = buf[17] & 0xf; |
138 | |
139 | if (ctx->alpha_info > 2) { |
140 | av_log(avctx, AV_LOG_ERROR, "Invalid alpha mode %d\n", ctx->alpha_info); |
141 | return AVERROR_INVALIDDATA; |
142 | } |
143 | if (avctx->skip_alpha) ctx->alpha_info = 0; |
144 | |
145 | switch (ctx->chroma_factor) { |
146 | case 2: |
147 | avctx->pix_fmt = ctx->alpha_info ? AV_PIX_FMT_YUVA422P10 |
148 | : AV_PIX_FMT_YUV422P10; |
149 | break; |
150 | case 3: |
151 | avctx->pix_fmt = ctx->alpha_info ? AV_PIX_FMT_YUVA444P10 |
152 | : AV_PIX_FMT_YUV444P10; |
153 | break; |
154 | default: |
155 | av_log(avctx, AV_LOG_ERROR, |
156 | "unsupported picture format: %d\n", ctx->pic_format); |
157 | return AVERROR_INVALIDDATA; |
158 | } |
159 | |
160 | if (ctx->scantable_type != ctx->frame_type) { |
161 | if (!ctx->frame_type) |
162 | ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable, |
163 | ff_prores_progressive_scan); |
164 | else |
165 | ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable, |
166 | ff_prores_interlaced_scan); |
167 | ctx->scantable_type = ctx->frame_type; |
168 | } |
169 | |
170 | if (ctx->frame_type) { /* if interlaced */ |
171 | ctx->frame->interlaced_frame = 1; |
172 | ctx->frame->top_field_first = ctx->frame_type & 1; |
173 | } else { |
174 | ctx->frame->interlaced_frame = 0; |
175 | } |
176 | |
177 | avctx->color_primaries = buf[14]; |
178 | avctx->color_trc = buf[15]; |
179 | avctx->colorspace = buf[16]; |
180 | |
181 | ctx->qmat_changed = 0; |
182 | ptr = buf + 20; |
183 | flags = buf[19]; |
184 | if (flags & 2) { |
185 | if (ptr - buf > hdr_size - 64) { |
186 | av_log(avctx, AV_LOG_ERROR, "header data too small\n"); |
187 | return AVERROR_INVALIDDATA; |
188 | } |
189 | if (memcmp(ctx->qmat_luma, ptr, 64)) { |
190 | memcpy(ctx->qmat_luma, ptr, 64); |
191 | ctx->qmat_changed = 1; |
192 | } |
193 | ptr += 64; |
194 | } else { |
195 | memset(ctx->qmat_luma, 4, 64); |
196 | ctx->qmat_changed = 1; |
197 | } |
198 | |
199 | if (flags & 1) { |
200 | if (ptr - buf > hdr_size - 64) { |
201 | av_log(avctx, AV_LOG_ERROR, "header data too small\n"); |
202 | return -1; |
203 | } |
204 | if (memcmp(ctx->qmat_chroma, ptr, 64)) { |
205 | memcpy(ctx->qmat_chroma, ptr, 64); |
206 | ctx->qmat_changed = 1; |
207 | } |
208 | } else { |
209 | memset(ctx->qmat_chroma, 4, 64); |
210 | ctx->qmat_changed = 1; |
211 | } |
212 | |
213 | return hdr_size; |
214 | } |
215 | |
216 | |
217 | static int decode_picture_header(ProresContext *ctx, const uint8_t *buf, |
218 | const int data_size, AVCodecContext *avctx) |
219 | { |
220 | int i, hdr_size, pic_data_size, num_slices; |
221 | int slice_width_factor, slice_height_factor; |
222 | int remainder, num_x_slices; |
223 | const uint8_t *data_ptr, *index_ptr; |
224 | |
225 | hdr_size = data_size > 0 ? buf[0] >> 3 : 0; |
226 | if (hdr_size < 8 || hdr_size > data_size) { |
227 | av_log(avctx, AV_LOG_ERROR, "picture header too small\n"); |
228 | return AVERROR_INVALIDDATA; |
229 | } |
230 | |
231 | pic_data_size = AV_RB32(buf + 1); |
232 | if (pic_data_size > data_size) { |
233 | av_log(avctx, AV_LOG_ERROR, "picture data too small\n"); |
234 | return AVERROR_INVALIDDATA; |
235 | } |
236 | |
237 | slice_width_factor = buf[7] >> 4; |
238 | slice_height_factor = buf[7] & 0xF; |
239 | if (slice_width_factor > 3 || slice_height_factor) { |
240 | av_log(avctx, AV_LOG_ERROR, |
241 | "unsupported slice dimension: %d x %d\n", |
242 | 1 << slice_width_factor, 1 << slice_height_factor); |
243 | return AVERROR_INVALIDDATA; |
244 | } |
245 | |
246 | ctx->slice_width_factor = slice_width_factor; |
247 | ctx->slice_height_factor = slice_height_factor; |
248 | |
249 | ctx->num_x_mbs = (avctx->width + 15) >> 4; |
250 | ctx->num_y_mbs = (avctx->height + |
251 | (1 << (4 + ctx->frame->interlaced_frame)) - 1) >> |
252 | (4 + ctx->frame->interlaced_frame); |
253 | |
254 | remainder = av_mod_uintp2(ctx->num_x_mbs, slice_width_factor); |
255 | num_x_slices = (ctx->num_x_mbs >> slice_width_factor) + (remainder & 1) + |
256 | ((remainder >> 1) & 1) + ((remainder >> 2) & 1); |
257 | |
258 | num_slices = num_x_slices * ctx->num_y_mbs; |
259 | if (num_slices != AV_RB16(buf + 5)) { |
260 | av_log(avctx, AV_LOG_ERROR, "invalid number of slices\n"); |
261 | return AVERROR_INVALIDDATA; |
262 | } |
263 | |
264 | if (ctx->total_slices != num_slices) { |
265 | av_freep(&ctx->slice_data); |
266 | ctx->slice_data = av_malloc_array(num_slices + 1, sizeof(ctx->slice_data[0])); |
267 | if (!ctx->slice_data) |
268 | return AVERROR(ENOMEM); |
269 | ctx->total_slices = num_slices; |
270 | } |
271 | |
272 | if (hdr_size + num_slices * 2 > data_size) { |
273 | av_log(avctx, AV_LOG_ERROR, "slice table too small\n"); |
274 | return AVERROR_INVALIDDATA; |
275 | } |
276 | |
277 | /* parse slice table allowing quick access to the slice data */ |
278 | index_ptr = buf + hdr_size; |
279 | data_ptr = index_ptr + num_slices * 2; |
280 | |
281 | for (i = 0; i < num_slices; i++) { |
282 | ctx->slice_data[i].index = data_ptr; |
283 | ctx->slice_data[i].prev_slice_sf = 0; |
284 | data_ptr += AV_RB16(index_ptr + i * 2); |
285 | } |
286 | ctx->slice_data[i].index = data_ptr; |
287 | ctx->slice_data[i].prev_slice_sf = 0; |
288 | |
289 | if (data_ptr > buf + data_size) { |
290 | av_log(avctx, AV_LOG_ERROR, "out of slice data\n"); |
291 | return -1; |
292 | } |
293 | |
294 | return pic_data_size; |
295 | } |
296 | |
297 | |
298 | /** |
299 | * Read an unsigned rice/exp golomb codeword. |
300 | */ |
301 | static inline int decode_vlc_codeword(GetBitContext *gb, unsigned codebook) |
302 | { |
303 | unsigned int rice_order, exp_order, switch_bits; |
304 | unsigned int buf, code; |
305 | int log, prefix_len, len; |
306 | |
307 | OPEN_READER(re, gb); |
308 | UPDATE_CACHE(re, gb); |
309 | buf = GET_CACHE(re, gb); |
310 | |
311 | /* number of prefix bits to switch between Rice and expGolomb */ |
312 | switch_bits = (codebook & 3) + 1; |
313 | rice_order = codebook >> 5; /* rice code order */ |
314 | exp_order = (codebook >> 2) & 7; /* exp golomb code order */ |
315 | |
316 | log = 31 - av_log2(buf); /* count prefix bits (zeroes) */ |
317 | |
318 | if (log < switch_bits) { /* ok, we got a rice code */ |
319 | if (!rice_order) { |
320 | /* shortcut for faster decoding of rice codes without remainder */ |
321 | code = log; |
322 | LAST_SKIP_BITS(re, gb, log + 1); |
323 | } else { |
324 | prefix_len = log + 1; |
325 | code = (log << rice_order) + NEG_USR32(buf << prefix_len, rice_order); |
326 | LAST_SKIP_BITS(re, gb, prefix_len + rice_order); |
327 | } |
328 | } else { /* otherwise we got a exp golomb code */ |
329 | len = (log << 1) - switch_bits + exp_order + 1; |
330 | code = NEG_USR32(buf, len) - (1 << exp_order) + (switch_bits << rice_order); |
331 | LAST_SKIP_BITS(re, gb, len); |
332 | } |
333 | |
334 | CLOSE_READER(re, gb); |
335 | |
336 | return code; |
337 | } |
338 | |
339 | #define LSB2SIGN(x) (-((x) & 1)) |
340 | #define TOSIGNED(x) (((x) >> 1) ^ LSB2SIGN(x)) |
341 | |
342 | /** |
343 | * Decode DC coefficients for all blocks in a slice. |
344 | */ |
345 | static inline void decode_dc_coeffs(GetBitContext *gb, int16_t *out, |
346 | int nblocks) |
347 | { |
348 | int16_t prev_dc; |
349 | int i, sign; |
350 | int16_t delta; |
351 | unsigned int code; |
352 | |
353 | code = decode_vlc_codeword(gb, FIRST_DC_CB); |
354 | out[0] = prev_dc = TOSIGNED(code); |
355 | |
356 | out += 64; /* move to the DC coeff of the next block */ |
357 | delta = 3; |
358 | |
359 | for (i = 1; i < nblocks; i++, out += 64) { |
360 | code = decode_vlc_codeword(gb, ff_prores_dc_codebook[FFMIN(FFABS(delta), 3)]); |
361 | |
362 | sign = -(((delta >> 15) & 1) ^ (code & 1)); |
363 | delta = (((code + 1) >> 1) ^ sign) - sign; |
364 | prev_dc += delta; |
365 | out[0] = prev_dc; |
366 | } |
367 | } |
368 | |
369 | #define MAX_PADDING 16 |
370 | |
371 | /** |
372 | * Decode AC coefficients for all blocks in a slice. |
373 | */ |
374 | static inline int decode_ac_coeffs(GetBitContext *gb, int16_t *out, |
375 | int blocks_per_slice, |
376 | int plane_size_factor, |
377 | const uint8_t *scan) |
378 | { |
379 | int pos, block_mask, run, level, sign, run_cb_index, lev_cb_index; |
380 | int max_coeffs, bits_left; |
381 | |
382 | /* set initial prediction values */ |
383 | run = 4; |
384 | level = 2; |
385 | |
386 | max_coeffs = blocks_per_slice << 6; |
387 | block_mask = blocks_per_slice - 1; |
388 | |
389 | for (pos = blocks_per_slice - 1; pos < max_coeffs;) { |
390 | run_cb_index = ff_prores_run_to_cb_index[FFMIN(run, 15)]; |
391 | lev_cb_index = ff_prores_lev_to_cb_index[FFMIN(level, 9)]; |
392 | |
393 | bits_left = get_bits_left(gb); |
394 | if (bits_left <= 0 || (bits_left <= MAX_PADDING && !show_bits(gb, bits_left))) |
395 | return 0; |
396 | |
397 | run = decode_vlc_codeword(gb, ff_prores_ac_codebook[run_cb_index]); |
398 | if (run < 0) |
399 | return AVERROR_INVALIDDATA; |
400 | |
401 | bits_left = get_bits_left(gb); |
402 | if (bits_left <= 0 || (bits_left <= MAX_PADDING && !show_bits(gb, bits_left))) |
403 | return AVERROR_INVALIDDATA; |
404 | |
405 | level = decode_vlc_codeword(gb, ff_prores_ac_codebook[lev_cb_index]) + 1; |
406 | if (level < 0) |
407 | return AVERROR_INVALIDDATA; |
408 | |
409 | pos += run + 1; |
410 | if (pos >= max_coeffs) |
411 | break; |
412 | |
413 | sign = get_sbits(gb, 1); |
414 | out[((pos & block_mask) << 6) + scan[pos >> plane_size_factor]] = |
415 | (level ^ sign) - sign; |
416 | } |
417 | |
418 | return 0; |
419 | } |
420 | |
421 | |
422 | /** |
423 | * Decode a slice plane (luma or chroma). |
424 | */ |
425 | static int decode_slice_plane(ProresContext *ctx, ProresThreadData *td, |
426 | const uint8_t *buf, |
427 | int data_size, uint16_t *out_ptr, |
428 | int linesize, int mbs_per_slice, |
429 | int blocks_per_mb, int plane_size_factor, |
430 | const int16_t *qmat, int is_chroma) |
431 | { |
432 | GetBitContext gb; |
433 | int16_t *block_ptr; |
434 | int mb_num, blocks_per_slice, ret; |
435 | |
436 | blocks_per_slice = mbs_per_slice * blocks_per_mb; |
437 | |
438 | memset(td->blocks, 0, 8 * 4 * 64 * sizeof(*td->blocks)); |
439 | |
440 | init_get_bits(&gb, buf, data_size << 3); |
441 | |
442 | decode_dc_coeffs(&gb, td->blocks, blocks_per_slice); |
443 | |
444 | ret = decode_ac_coeffs(&gb, td->blocks, blocks_per_slice, |
445 | plane_size_factor, ctx->scantable.permutated); |
446 | if (ret < 0) |
447 | return ret; |
448 | |
449 | /* inverse quantization, inverse transform and output */ |
450 | block_ptr = td->blocks; |
451 | |
452 | if (!is_chroma) { |
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); |
455 | block_ptr += 64; |
456 | if (blocks_per_mb > 2) { |
457 | ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat); |
458 | block_ptr += 64; |
459 | } |
460 | ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat); |
461 | block_ptr += 64; |
462 | if (blocks_per_mb > 2) { |
463 | ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat); |
464 | block_ptr += 64; |
465 | } |
466 | } |
467 | } else { |
468 | for (mb_num = 0; mb_num < mbs_per_slice; mb_num++, out_ptr += blocks_per_mb * 4) { |
469 | ctx->dsp.idct_put(out_ptr, linesize, block_ptr, qmat); |
470 | block_ptr += 64; |
471 | ctx->dsp.idct_put(out_ptr + linesize * 4, linesize, block_ptr, qmat); |
472 | block_ptr += 64; |
473 | if (blocks_per_mb > 2) { |
474 | ctx->dsp.idct_put(out_ptr + 8, linesize, block_ptr, qmat); |
475 | block_ptr += 64; |
476 | ctx->dsp.idct_put(out_ptr + linesize * 4 + 8, linesize, block_ptr, qmat); |
477 | block_ptr += 64; |
478 | } |
479 | } |
480 | } |
481 | return 0; |
482 | } |
483 | |
484 | |
485 | static void unpack_alpha(GetBitContext *gb, uint16_t *dst, int num_coeffs, |
486 | const int num_bits) |
487 | { |
488 | const int mask = (1 << num_bits) - 1; |
489 | int i, idx, val, alpha_val; |
490 | |
491 | idx = 0; |
492 | alpha_val = mask; |
493 | do { |
494 | do { |
495 | if (get_bits1(gb)) |
496 | val = get_bits(gb, num_bits); |
497 | else { |
498 | int sign; |
499 | val = get_bits(gb, num_bits == 16 ? 7 : 4); |
500 | sign = val & 1; |
501 | val = (val + 2) >> 1; |
502 | if (sign) |
503 | val = -val; |
504 | } |
505 | alpha_val = (alpha_val + val) & mask; |
506 | if (num_bits == 16) |
507 | dst[idx++] = alpha_val >> 6; |
508 | else |
509 | dst[idx++] = (alpha_val << 2) | (alpha_val >> 6); |
510 | if (idx >= num_coeffs) { |
511 | break; |
512 | } |
513 | } while (get_bits1(gb)); |
514 | val = get_bits(gb, 4); |
515 | if (!val) |
516 | val = get_bits(gb, 11); |
517 | if (idx + val > num_coeffs) |
518 | val = num_coeffs - idx; |
519 | if (num_bits == 16) |
520 | for (i = 0; i < val; i++) |
521 | dst[idx++] = alpha_val >> 6; |
522 | else |
523 | for (i = 0; i < val; i++) |
524 | dst[idx++] = (alpha_val << 2) | (alpha_val >> 6); |
525 | } while (idx < num_coeffs); |
526 | } |
527 | |
528 | /** |
529 | * Decode alpha slice plane. |
530 | */ |
531 | static void decode_alpha_plane(ProresContext *ctx, ProresThreadData *td, |
532 | const uint8_t *buf, int data_size, |
533 | uint16_t *out_ptr, int linesize, |
534 | int mbs_per_slice) |
535 | { |
536 | GetBitContext gb; |
537 | int i; |
538 | uint16_t *block_ptr; |
539 | |
540 | memset(td->blocks, 0, 8 * 4 * 64 * sizeof(*td->blocks)); |
541 | |
542 | init_get_bits(&gb, buf, data_size << 3); |
543 | |
544 | if (ctx->alpha_info == 2) |
545 | unpack_alpha(&gb, td->blocks, mbs_per_slice * 4 * 64, 16); |
546 | else |
547 | unpack_alpha(&gb, td->blocks, mbs_per_slice * 4 * 64, 8); |
548 | |
549 | block_ptr = td->blocks; |
550 | |
551 | for (i = 0; i < 16; i++) { |
552 | memcpy(out_ptr, block_ptr, 16 * mbs_per_slice * sizeof(*out_ptr)); |
553 | out_ptr += linesize >> 1; |
554 | block_ptr += 16 * mbs_per_slice; |
555 | } |
556 | } |
557 | |
558 | static int decode_slice(AVCodecContext *avctx, void *tdata) |
559 | { |
560 | ProresThreadData *td = tdata; |
561 | ProresContext *ctx = avctx->priv_data; |
562 | int mb_x_pos = td->x_pos; |
563 | int mb_y_pos = td->y_pos; |
564 | int pic_num = ctx->pic_num; |
565 | int slice_num = td->slice_num; |
566 | int mbs_per_slice = td->slice_width; |
567 | const uint8_t *buf; |
568 | uint8_t *y_data, *u_data, *v_data, *a_data; |
569 | AVFrame *pic = ctx->frame; |
570 | int i, sf, slice_width_factor; |
571 | int slice_data_size, hdr_size; |
572 | int y_data_size, u_data_size, v_data_size, a_data_size; |
573 | int y_linesize, u_linesize, v_linesize, a_linesize; |
574 | int coff[4]; |
575 | int ret; |
576 | |
577 | buf = ctx->slice_data[slice_num].index; |
578 | slice_data_size = ctx->slice_data[slice_num + 1].index - buf; |
579 | |
580 | slice_width_factor = av_log2(mbs_per_slice); |
581 | |
582 | y_data = pic->data[0]; |
583 | u_data = pic->data[1]; |
584 | v_data = pic->data[2]; |
585 | a_data = pic->data[3]; |
586 | y_linesize = pic->linesize[0]; |
587 | u_linesize = pic->linesize[1]; |
588 | v_linesize = pic->linesize[2]; |
589 | a_linesize = pic->linesize[3]; |
590 | |
591 | if (pic->interlaced_frame) { |
592 | if (!(pic_num ^ pic->top_field_first)) { |
593 | y_data += y_linesize; |
594 | u_data += u_linesize; |
595 | v_data += v_linesize; |
596 | if (a_data) |
597 | a_data += a_linesize; |
598 | } |
599 | y_linesize <<= 1; |
600 | u_linesize <<= 1; |
601 | v_linesize <<= 1; |
602 | a_linesize <<= 1; |
603 | } |
604 | y_data += (mb_y_pos << 4) * y_linesize + (mb_x_pos << 5); |
605 | u_data += (mb_y_pos << 4) * u_linesize + (mb_x_pos << ctx->mb_chroma_factor); |
606 | v_data += (mb_y_pos << 4) * v_linesize + (mb_x_pos << ctx->mb_chroma_factor); |
607 | if (a_data) |
608 | a_data += (mb_y_pos << 4) * a_linesize + (mb_x_pos << 5); |
609 | |
610 | if (slice_data_size < 6) { |
611 | av_log(avctx, AV_LOG_ERROR, "slice data too small\n"); |
612 | return AVERROR_INVALIDDATA; |
613 | } |
614 | |
615 | /* parse slice header */ |
616 | hdr_size = buf[0] >> 3; |
617 | coff[0] = hdr_size; |
618 | y_data_size = AV_RB16(buf + 2); |
619 | coff[1] = coff[0] + y_data_size; |
620 | u_data_size = AV_RB16(buf + 4); |
621 | coff[2] = coff[1] + u_data_size; |
622 | v_data_size = hdr_size > 7 ? AV_RB16(buf + 6) : slice_data_size - coff[2]; |
623 | coff[3] = coff[2] + v_data_size; |
624 | a_data_size = ctx->alpha_info ? slice_data_size - coff[3] : 0; |
625 | |
626 | /* if V or alpha component size is negative that means that previous |
627 | component sizes are too large */ |
628 | if (v_data_size < 0 || a_data_size < 0 || hdr_size < 6 || coff[3] > slice_data_size) { |
629 | av_log(avctx, AV_LOG_ERROR, "invalid data size\n"); |
630 | return AVERROR_INVALIDDATA; |
631 | } |
632 | |
633 | sf = av_clip(buf[1], 1, 224); |
634 | sf = sf > 128 ? (sf - 96) << 2 : sf; |
635 | |
636 | /* scale quantization matrixes according with slice's scale factor */ |
637 | /* TODO: this can be SIMD-optimized a lot */ |
638 | if (ctx->qmat_changed || sf != td->prev_slice_sf) { |
639 | td->prev_slice_sf = sf; |
640 | for (i = 0; i < 64; i++) { |
641 | td->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf; |
642 | td->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf; |
643 | } |
644 | } |
645 | |
646 | /* decode luma plane */ |
647 | ret = decode_slice_plane(ctx, td, buf + coff[0], y_data_size, |
648 | (uint16_t*) y_data, y_linesize, |
649 | mbs_per_slice, 4, slice_width_factor + 2, |
650 | td->qmat_luma_scaled, 0); |
651 | |
652 | if (ret < 0) |
653 | return ret; |
654 | |
655 | /* decode U chroma plane */ |
656 | ret = decode_slice_plane(ctx, td, buf + coff[1], u_data_size, |
657 | (uint16_t*) u_data, u_linesize, |
658 | mbs_per_slice, ctx->num_chroma_blocks, |
659 | slice_width_factor + ctx->chroma_factor - 1, |
660 | td->qmat_chroma_scaled, 1); |
661 | if (ret < 0) |
662 | return ret; |
663 | |
664 | /* decode V chroma plane */ |
665 | ret = decode_slice_plane(ctx, td, buf + coff[2], v_data_size, |
666 | (uint16_t*) v_data, v_linesize, |
667 | mbs_per_slice, ctx->num_chroma_blocks, |
668 | slice_width_factor + ctx->chroma_factor - 1, |
669 | td->qmat_chroma_scaled, 1); |
670 | if (ret < 0) |
671 | return ret; |
672 | |
673 | /* decode alpha plane if available */ |
674 | if (a_data && a_data_size) |
675 | decode_alpha_plane(ctx, td, buf + coff[3], a_data_size, |
676 | (uint16_t*) a_data, a_linesize, |
677 | mbs_per_slice); |
678 | |
679 | return 0; |
680 | } |
681 | |
682 | |
683 | static int decode_picture(ProresContext *ctx, int pic_num, |
684 | AVCodecContext *avctx) |
685 | { |
686 | int slice_num, slice_width, x_pos, y_pos; |
687 | |
688 | slice_num = 0; |
689 | |
690 | ctx->pic_num = pic_num; |
691 | for (y_pos = 0; y_pos < ctx->num_y_mbs; y_pos++) { |
692 | slice_width = 1 << ctx->slice_width_factor; |
693 | |
694 | for (x_pos = 0; x_pos < ctx->num_x_mbs && slice_width; |
695 | x_pos += slice_width) { |
696 | while (ctx->num_x_mbs - x_pos < slice_width) |
697 | slice_width >>= 1; |
698 | |
699 | ctx->slice_data[slice_num].slice_num = slice_num; |
700 | ctx->slice_data[slice_num].x_pos = x_pos; |
701 | ctx->slice_data[slice_num].y_pos = y_pos; |
702 | ctx->slice_data[slice_num].slice_width = slice_width; |
703 | |
704 | slice_num++; |
705 | } |
706 | } |
707 | |
708 | return avctx->execute(avctx, decode_slice, |
709 | ctx->slice_data, NULL, slice_num, |
710 | sizeof(ctx->slice_data[0])); |
711 | } |
712 | |
713 | |
714 | #define MOVE_DATA_PTR(nbytes) buf += (nbytes); buf_size -= (nbytes) |
715 | |
716 | static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame, |
717 | AVPacket *avpkt) |
718 | { |
719 | ProresContext *ctx = avctx->priv_data; |
720 | const uint8_t *buf = avpkt->data; |
721 | int buf_size = avpkt->size; |
722 | int frame_hdr_size, pic_num, pic_data_size; |
723 | |
724 | ctx->frame = data; |
725 | ctx->frame->pict_type = AV_PICTURE_TYPE_I; |
726 | ctx->frame->key_frame = 1; |
727 | |
728 | /* check frame atom container */ |
729 | if (buf_size < 28 || buf_size < AV_RB32(buf) || |
730 | AV_RB32(buf + 4) != FRAME_ID) { |
731 | av_log(avctx, AV_LOG_ERROR, "invalid frame\n"); |
732 | return AVERROR_INVALIDDATA; |
733 | } |
734 | |
735 | MOVE_DATA_PTR(8); |
736 | |
737 | frame_hdr_size = decode_frame_header(ctx, buf, buf_size, avctx); |
738 | if (frame_hdr_size < 0) |
739 | return AVERROR_INVALIDDATA; |
740 | |
741 | MOVE_DATA_PTR(frame_hdr_size); |
742 | |
743 | if (ff_get_buffer(avctx, ctx->frame, 0) < 0) |
744 | return -1; |
745 | |
746 | for (pic_num = 0; ctx->frame->interlaced_frame - pic_num + 1; pic_num++) { |
747 | pic_data_size = decode_picture_header(ctx, buf, buf_size, avctx); |
748 | if (pic_data_size < 0) |
749 | return AVERROR_INVALIDDATA; |
750 | |
751 | if (decode_picture(ctx, pic_num, avctx)) |
752 | return -1; |
753 | |
754 | MOVE_DATA_PTR(pic_data_size); |
755 | } |
756 | |
757 | ctx->frame = NULL; |
758 | *got_frame = 1; |
759 | |
760 | return avpkt->size; |
761 | } |
762 | |
763 | |
764 | static av_cold int decode_close(AVCodecContext *avctx) |
765 | { |
766 | ProresContext *ctx = avctx->priv_data; |
767 | |
768 | av_freep(&ctx->slice_data); |
769 | |
770 | return 0; |
771 | } |
772 | |
773 | |
774 | AVCodec ff_prores_lgpl_decoder = { |
775 | .name = "prores_lgpl", |
776 | .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"), |
777 | .type = AVMEDIA_TYPE_VIDEO, |
778 | .id = AV_CODEC_ID_PRORES, |
779 | .priv_data_size = sizeof(ProresContext), |
780 | .init = decode_init, |
781 | .close = decode_close, |
782 | .decode = decode_frame, |
783 | .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS, |
784 | }; |
785 |