utvideo: frame multithreading.
[libav.git] / libavcodec / utvideo.c
CommitLineData
0d8506b8
KS
1/*
2 * Ut Video decoder
3 * Copyright (c) 2011 Konstantin Shishkov
4 *
5 * This file is part of Libav.
6 *
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file
24 * Ut Video decoder
25 */
26
27#include <stdlib.h>
28
29#include "libavutil/intreadwrite.h"
30#include "avcodec.h"
31#include "bytestream.h"
32#include "get_bits.h"
33#include "dsputil.h"
12e984ae 34#include "thread.h"
0d8506b8
KS
35
36enum {
37 PRED_NONE = 0,
38 PRED_LEFT,
39 PRED_GRADIENT,
40 PRED_MEDIAN,
41};
42
43typedef struct UtvideoContext {
44 AVCodecContext *avctx;
45 AVFrame pic;
46 DSPContext dsp;
47
48 uint32_t frame_info_size, flags, frame_info;
49 int planes;
50 int slices;
51 int compression;
52 int interlaced;
53 int frame_pred;
54
55 uint8_t *slice_bits;
56 int slice_bits_size;
57} UtvideoContext;
58
59typedef struct HuffEntry {
60 uint8_t sym;
61 uint8_t len;
62} HuffEntry;
63
64static int huff_cmp(const void *a, const void *b)
65{
66 const HuffEntry *aa = a, *bb = b;
67 return (aa->len - bb->len)*256 + aa->sym - bb->sym;
68}
69
46e1af3b 70static int build_huff(const uint8_t *src, VLC *vlc, int *fsym)
0d8506b8
KS
71{
72 int i;
73 HuffEntry he[256];
74 int last;
75 uint32_t codes[256];
76 uint8_t bits[256];
77 uint8_t syms[256];
78 uint32_t code;
79
46e1af3b 80 *fsym = -1;
0d8506b8
KS
81 for (i = 0; i < 256; i++) {
82 he[i].sym = i;
83 he[i].len = *src++;
84 }
85 qsort(he, 256, sizeof(*he), huff_cmp);
86
46e1af3b
KS
87 if (!he[0].len) {
88 *fsym = he[0].sym;
89 return 0;
90 }
91 if (he[0].len > 32)
0d8506b8
KS
92 return -1;
93
94 last = 255;
95 while (he[last].len == 255 && last)
96 last--;
97
98 code = 1;
99 for (i = last; i >= 0; i--) {
100 codes[i] = code >> (32 - he[i].len);
101 bits[i] = he[i].len;
102 syms[i] = he[i].sym;
103 code += 0x80000000u >> (he[i].len - 1);
104 }
105
106 return init_vlc_sparse(vlc, FFMIN(he[last].len, 9), last + 1,
107 bits, sizeof(*bits), sizeof(*bits),
108 codes, sizeof(*codes), sizeof(*codes),
109 syms, sizeof(*syms), sizeof(*syms), 0);
110}
111
112static int decode_plane(UtvideoContext *c, int plane_no,
113 uint8_t *dst, int step, int stride,
114 int width, int height,
115 const uint8_t *src, int src_size, int use_pred)
116{
117 int i, j, slice, pix;
118 int sstart, send;
119 VLC vlc;
120 GetBitContext gb;
46e1af3b 121 int prev, fsym;
9a173575 122 const int cmask = ~(!plane_no && c->avctx->pix_fmt == PIX_FMT_YUV420P);
0d8506b8 123
46e1af3b 124 if (build_huff(src, &vlc, &fsym)) {
0d8506b8
KS
125 av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
126 return AVERROR_INVALIDDATA;
127 }
46e1af3b
KS
128 if (fsym >= 0) { // build_huff reported a symbol to fill slices with
129 send = 0;
130 for (slice = 0; slice < c->slices; slice++) {
131 uint8_t *dest;
132
133 sstart = send;
134 send = (height * (slice + 1) / c->slices) & cmask;
135 dest = dst + sstart * stride;
136
137 prev = 0x80;
138 for (j = sstart; j < send; j++) {
139 for (i = 0; i < width * step; i += step) {
140 pix = fsym;
141 if (use_pred) {
142 prev += pix;
143 pix = prev;
144 }
145 dest[i] = pix;
146 }
147 dest += stride;
148 }
149 }
150 return 0;
151 }
0d8506b8
KS
152
153 src += 256;
154 src_size -= 256;
155
156 send = 0;
157 for (slice = 0; slice < c->slices; slice++) {
158 uint8_t *dest;
159 int slice_data_start, slice_data_end, slice_size;
160
161 sstart = send;
9a173575 162 send = (height * (slice + 1) / c->slices) & cmask;
0d8506b8
KS
163 dest = dst + sstart * stride;
164
165 // slice offset and size validation was done earlier
166 slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
167 slice_data_end = AV_RL32(src + slice * 4);
168 slice_size = slice_data_end - slice_data_start;
169
170 if (!slice_size) {
171 for (j = sstart; j < send; j++) {
172 for (i = 0; i < width * step; i += step)
173 dest[i] = 0x80;
174 dest += stride;
175 }
176 continue;
177 }
178
179 memcpy(c->slice_bits, src + slice_data_start + c->slices * 4, slice_size);
180 memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
181 c->dsp.bswap_buf((uint32_t*)c->slice_bits, (uint32_t*)c->slice_bits,
182 (slice_data_end - slice_data_start + 3) >> 2);
183 init_get_bits(&gb, c->slice_bits, slice_size * 8);
184
185 prev = 0x80;
186 for (j = sstart; j < send; j++) {
187 for (i = 0; i < width * step; i += step) {
188 if (get_bits_left(&gb) <= 0) {
189 av_log(c->avctx, AV_LOG_ERROR, "Slice decoding ran out of bits\n");
190 goto fail;
191 }
192 pix = get_vlc2(&gb, vlc.table, vlc.bits, 4);
193 if (pix < 0) {
194 av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
195 goto fail;
196 }
197 if (use_pred) {
198 prev += pix;
199 pix = prev;
200 }
201 dest[i] = pix;
202 }
203 dest += stride;
204 }
205 if (get_bits_left(&gb) > 32)
206 av_log(c->avctx, AV_LOG_WARNING, "%d bits left after decoding slice\n",
207 get_bits_left(&gb));
208 }
209
210 free_vlc(&vlc);
211
212 return 0;
213fail:
214 free_vlc(&vlc);
215 return AVERROR_INVALIDDATA;
216}
217
218static const int rgb_order[4] = { 1, 2, 0, 3 };
219
220static void restore_rgb_planes(uint8_t *src, int step, int stride, int width, int height)
221{
222 int i, j;
223 uint8_t r, g, b;
224
225 for (j = 0; j < height; j++) {
226 for (i = 0; i < width * step; i += step) {
227 r = src[i];
228 g = src[i + 1];
229 b = src[i + 2];
230 src[i] = r + g - 0x80;
231 src[i + 2] = b + g - 0x80;
232 }
233 src += stride;
234 }
235}
236
237static void restore_median(uint8_t *src, int step, int stride,
9a173575 238 int width, int height, int slices, int rmode)
0d8506b8
KS
239{
240 int i, j, slice;
241 int A, B, C;
242 uint8_t *bsrc;
243 int slice_start, slice_height;
9a173575 244 const int cmask = ~rmode;
0d8506b8
KS
245
246 for (slice = 0; slice < slices; slice++) {
9a173575
KS
247 slice_start = ((slice * height) / slices) & cmask;
248 slice_height = ((((slice + 1) * height) / slices) & cmask) - slice_start;
0d8506b8
KS
249
250 bsrc = src + slice_start * stride;
251
252 // first line - left neighbour prediction
253 bsrc[0] += 0x80;
254 A = bsrc[0];
255 for (i = step; i < width * step; i += step) {
256 bsrc[i] += A;
257 A = bsrc[i];
258 }
259 bsrc += stride;
260 if (slice_height == 1)
261 continue;
262 // second line - first element has top predition, the rest uses median
263 C = bsrc[-stride];
264 bsrc[0] += C;
265 A = bsrc[0];
266 for (i = step; i < width * step; i += step) {
267 B = bsrc[i - stride];
268 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
269 C = B;
270 A = bsrc[i];
271 }
272 bsrc += stride;
273 // the rest of lines use continuous median prediction
274 for (j = 2; j < slice_height; j++) {
275 for (i = 0; i < width * step; i += step) {
276 B = bsrc[i - stride];
277 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
278 C = B;
279 A = bsrc[i];
280 }
281 bsrc += stride;
282 }
283 }
284}
285
490dcda6
KS
286/* UtVideo interlaced mode treats every two lines as a single one,
287 * so restoring function should take care of possible padding between
288 * two parts of the same "line".
289 */
290static void restore_median_il(uint8_t *src, int step, int stride,
291 int width, int height, int slices, int rmode)
292{
293 int i, j, slice;
294 int A, B, C;
295 uint8_t *bsrc;
296 int slice_start, slice_height;
297 const int cmask = ~(rmode ? 3 : 1);
298 const int stride2 = stride << 1;
299
300 for (slice = 0; slice < slices; slice++) {
301 slice_start = ((slice * height) / slices) & cmask;
302 slice_height = ((((slice + 1) * height) / slices) & cmask) - slice_start;
303 slice_height >>= 1;
304
305 bsrc = src + slice_start * stride;
306
307 // first line - left neighbour prediction
308 bsrc[0] += 0x80;
309 A = bsrc[0];
310 for (i = step; i < width * step; i += step) {
311 bsrc[i] += A;
312 A = bsrc[i];
313 }
314 for (i = 0; i < width * step; i += step) {
315 bsrc[stride + i] += A;
316 A = bsrc[stride + i];
317 }
318 bsrc += stride2;
319 if (slice_height == 1)
320 continue;
321 // second line - first element has top predition, the rest uses median
322 C = bsrc[-stride2];
323 bsrc[0] += C;
324 A = bsrc[0];
325 for (i = step; i < width * step; i += step) {
326 B = bsrc[i - stride2];
327 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
328 C = B;
329 A = bsrc[i];
330 }
331 for (i = 0; i < width * step; i += step) {
332 B = bsrc[i - stride];
333 bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
334 C = B;
335 A = bsrc[stride + i];
336 }
337 bsrc += stride2;
338 // the rest of lines use continuous median prediction
339 for (j = 2; j < slice_height; j++) {
340 for (i = 0; i < width * step; i += step) {
341 B = bsrc[i - stride2];
342 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
343 C = B;
344 A = bsrc[i];
345 }
346 for (i = 0; i < width * step; i += step) {
347 B = bsrc[i - stride];
348 bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
349 C = B;
350 A = bsrc[i + stride];
351 }
352 bsrc += stride2;
353 }
354 }
355}
356
0d8506b8
KS
357static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
358{
359 const uint8_t *buf = avpkt->data;
360 int buf_size = avpkt->size;
361 const uint8_t *buf_end = buf + buf_size;
362 UtvideoContext *c = avctx->priv_data;
363 const uint8_t *ptr;
364 int i, j;
365 const uint8_t *plane_start[5];
366 int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
367 int ret;
368
369 if (c->pic.data[0])
12e984ae 370 ff_thread_release_buffer(avctx, &c->pic);
0d8506b8
KS
371
372 c->pic.reference = 1;
373 c->pic.buffer_hints = FF_BUFFER_HINTS_VALID;
12e984ae 374 if ((ret = ff_thread_get_buffer(avctx, &c->pic)) < 0) {
0d8506b8
KS
375 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
376 return ret;
377 }
378
12e984ae
AD
379 ff_thread_finish_setup(avctx);
380
0d8506b8
KS
381 /* parse plane structure to retrieve frame flags and validate slice offsets */
382 ptr = buf;
383 for (i = 0; i < c->planes; i++) {
384 plane_start[i] = ptr;
385 if (buf_end - ptr < 256 + 4 * c->slices) {
386 av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
387 return AVERROR_INVALIDDATA;
388 }
389 ptr += 256;
390 slice_start = 0;
391 slice_end = 0;
392 for (j = 0; j < c->slices; j++) {
393 slice_end = bytestream_get_le32(&ptr);
394 slice_size = slice_end - slice_start;
395 if (slice_size < 0) {
396 av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
397 return AVERROR_INVALIDDATA;
398 }
399 slice_start = slice_end;
400 max_slice_size = FFMAX(max_slice_size, slice_size);
401 }
402 plane_size = slice_end;
403 if (buf_end - ptr < plane_size) {
404 av_log(avctx, AV_LOG_ERROR, "Plane size is bigger than available data\n");
405 return AVERROR_INVALIDDATA;
406 }
407 ptr += plane_size;
408 }
409 plane_start[c->planes] = ptr;
410 if (buf_end - ptr < c->frame_info_size) {
411 av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
412 return AVERROR_INVALIDDATA;
413 }
414 c->frame_info = AV_RL32(ptr);
415 av_log(avctx, AV_LOG_DEBUG, "frame information flags %X\n", c->frame_info);
416
417 c->frame_pred = (c->frame_info >> 8) & 3;
418
419 if (c->frame_pred == PRED_GRADIENT) {
420 av_log_ask_for_sample(avctx, "Frame uses gradient prediction\n");
421 return AVERROR_PATCHWELCOME;
422 }
423
424 av_fast_malloc(&c->slice_bits, &c->slice_bits_size,
425 max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
426
427 if (!c->slice_bits) {
428 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
429 return AVERROR(ENOMEM);
430 }
431
432 switch (c->avctx->pix_fmt) {
433 case PIX_FMT_RGB24:
434 case PIX_FMT_RGBA:
435 for (i = 0; i < c->planes; i++) {
436 ret = decode_plane(c, i, c->pic.data[0] + rgb_order[i], c->planes,
437 c->pic.linesize[0], avctx->width, avctx->height,
438 plane_start[i], plane_start[i + 1] - plane_start[i],
439 c->frame_pred == PRED_LEFT);
440 if (ret)
441 return ret;
442 if (c->frame_pred == PRED_MEDIAN)
443 restore_median(c->pic.data[0] + rgb_order[i], c->planes,
444 c->pic.linesize[0], avctx->width, avctx->height,
9a173575 445 c->slices, 0);
0d8506b8
KS
446 }
447 restore_rgb_planes(c->pic.data[0], c->planes, c->pic.linesize[0],
448 avctx->width, avctx->height);
449 break;
450 case PIX_FMT_YUV420P:
451 for (i = 0; i < 3; i++) {
452 ret = decode_plane(c, i, c->pic.data[i], 1,
453 c->pic.linesize[i], avctx->width >> !!i, avctx->height >> !!i,
454 plane_start[i], plane_start[i + 1] - plane_start[i],
455 c->frame_pred == PRED_LEFT);
456 if (ret)
457 return ret;
490dcda6
KS
458 if (c->frame_pred == PRED_MEDIAN) {
459 if (!c->interlaced) {
460 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
461 avctx->width >> !!i, avctx->height >> !!i,
462 c->slices, !i);
463 } else {
464 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
465 avctx->width >> !!i,
466 avctx->height >> !!i,
467 c->slices, !i);
468 }
469 }
0d8506b8
KS
470 }
471 break;
472 case PIX_FMT_YUV422P:
473 for (i = 0; i < 3; i++) {
474 ret = decode_plane(c, i, c->pic.data[i], 1,
475 c->pic.linesize[i], avctx->width >> !!i, avctx->height,
476 plane_start[i], plane_start[i + 1] - plane_start[i],
477 c->frame_pred == PRED_LEFT);
478 if (ret)
479 return ret;
490dcda6
KS
480 if (c->frame_pred == PRED_MEDIAN) {
481 if (!c->interlaced) {
482 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
483 avctx->width >> !!i, avctx->height,
484 c->slices, 0);
485 } else {
486 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
487 avctx->width >> !!i, avctx->height,
488 c->slices, 0);
489 }
490 }
0d8506b8
KS
491 }
492 break;
493 }
494
495 *data_size = sizeof(AVFrame);
496 *(AVFrame*)data = c->pic;
497
498 /* always report that the buffer was completely consumed */
499 return buf_size;
500}
501
502static av_cold int decode_init(AVCodecContext *avctx)
503{
504 UtvideoContext * const c = avctx->priv_data;
505
506 c->avctx = avctx;
507
508 dsputil_init(&c->dsp, avctx);
509
510 if (avctx->extradata_size < 16) {
511 av_log(avctx, AV_LOG_ERROR, "Insufficient extradata size %d, should be at least 16\n",
512 avctx->extradata_size);
513 return AVERROR_INVALIDDATA;
514 }
515
516 av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
517 avctx->extradata[3], avctx->extradata[2],
518 avctx->extradata[1], avctx->extradata[0]);
519 av_log(avctx, AV_LOG_DEBUG, "Original format %X\n", AV_RB32(avctx->extradata + 4));
520 c->frame_info_size = AV_RL32(avctx->extradata + 8);
521 c->flags = AV_RL32(avctx->extradata + 12);
522
523 if (c->frame_info_size != 4)
524 av_log_ask_for_sample(avctx, "Frame info is not 4 bytes\n");
525 av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08X\n", c->flags);
526 c->slices = (c->flags >> 24) + 1;
527 c->compression = c->flags & 1;
528 c->interlaced = c->flags & 0x800;
529
530 c->slice_bits_size = 0;
531
532 switch (avctx->codec_tag) {
533 case MKTAG('U', 'L', 'R', 'G'):
534 c->planes = 3;
535 avctx->pix_fmt = PIX_FMT_RGB24;
536 break;
537 case MKTAG('U', 'L', 'R', 'A'):
538 c->planes = 4;
539 avctx->pix_fmt = PIX_FMT_RGBA;
540 break;
541 case MKTAG('U', 'L', 'Y', '0'):
542 c->planes = 3;
543 avctx->pix_fmt = PIX_FMT_YUV420P;
544 break;
545 case MKTAG('U', 'L', 'Y', '2'):
546 c->planes = 3;
547 avctx->pix_fmt = PIX_FMT_YUV422P;
548 break;
549 default:
550 av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
551 avctx->codec_tag);
552 return AVERROR_INVALIDDATA;
553 }
554
555 return 0;
556}
557
558static av_cold int decode_end(AVCodecContext *avctx)
559{
560 UtvideoContext * const c = avctx->priv_data;
561
562 if (c->pic.data[0])
12e984ae 563 ff_thread_release_buffer(avctx, &c->pic);
0d8506b8
KS
564
565 av_freep(&c->slice_bits);
566
567 return 0;
568}
569
570AVCodec ff_utvideo_decoder = {
571 .name = "utvideo",
572 .type = AVMEDIA_TYPE_VIDEO,
573 .id = CODEC_ID_UTVIDEO,
574 .priv_data_size = sizeof(UtvideoContext),
575 .init = decode_init,
576 .close = decode_end,
577 .decode = decode_frame,
12e984ae 578 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
0d8506b8
KS
579 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
580};
581