378a1f7c1635cd59137c6e7df6097232cfc9951d
[libav.git] / libavcodec / utvideo.c
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"
34 #include "thread.h"
35
36 enum {
37 PRED_NONE = 0,
38 PRED_LEFT,
39 PRED_GRADIENT,
40 PRED_MEDIAN,
41 };
42
43 typedef 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
59 typedef struct HuffEntry {
60 uint8_t sym;
61 uint8_t len;
62 } HuffEntry;
63
64 static 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
70 static int build_huff(const uint8_t *src, VLC *vlc, int *fsym)
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
80 *fsym = -1;
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
87 if (!he[0].len) {
88 *fsym = he[0].sym;
89 return 0;
90 }
91 if (he[0].len > 32)
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 ff_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
112 static 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 use_pred)
116 {
117 int i, j, slice, pix;
118 int sstart, send;
119 VLC vlc;
120 GetBitContext gb;
121 int prev, fsym;
122 const int cmask = ~(!plane_no && c->avctx->pix_fmt == PIX_FMT_YUV420P);
123
124 if (build_huff(src, &vlc, &fsym)) {
125 av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
126 return AVERROR_INVALIDDATA;
127 }
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 }
152
153 src += 256;
154
155 send = 0;
156 for (slice = 0; slice < c->slices; slice++) {
157 uint8_t *dest;
158 int slice_data_start, slice_data_end, slice_size;
159
160 sstart = send;
161 send = (height * (slice + 1) / c->slices) & cmask;
162 dest = dst + sstart * stride;
163
164 // slice offset and size validation was done earlier
165 slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
166 slice_data_end = AV_RL32(src + slice * 4);
167 slice_size = slice_data_end - slice_data_start;
168
169 if (!slice_size) {
170 for (j = sstart; j < send; j++) {
171 for (i = 0; i < width * step; i += step)
172 dest[i] = 0x80;
173 dest += stride;
174 }
175 continue;
176 }
177
178 memcpy(c->slice_bits, src + slice_data_start + c->slices * 4, slice_size);
179 memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
180 c->dsp.bswap_buf((uint32_t*)c->slice_bits, (uint32_t*)c->slice_bits,
181 (slice_data_end - slice_data_start + 3) >> 2);
182 init_get_bits(&gb, c->slice_bits, slice_size * 8);
183
184 prev = 0x80;
185 for (j = sstart; j < send; j++) {
186 for (i = 0; i < width * step; i += step) {
187 if (get_bits_left(&gb) <= 0) {
188 av_log(c->avctx, AV_LOG_ERROR, "Slice decoding ran out of bits\n");
189 goto fail;
190 }
191 pix = get_vlc2(&gb, vlc.table, vlc.bits, 4);
192 if (pix < 0) {
193 av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
194 goto fail;
195 }
196 if (use_pred) {
197 prev += pix;
198 pix = prev;
199 }
200 dest[i] = pix;
201 }
202 dest += stride;
203 }
204 if (get_bits_left(&gb) > 32)
205 av_log(c->avctx, AV_LOG_WARNING, "%d bits left after decoding slice\n",
206 get_bits_left(&gb));
207 }
208
209 ff_free_vlc(&vlc);
210
211 return 0;
212 fail:
213 ff_free_vlc(&vlc);
214 return AVERROR_INVALIDDATA;
215 }
216
217 static const int rgb_order[4] = { 1, 2, 0, 3 };
218
219 static void restore_rgb_planes(uint8_t *src, int step, int stride, int width, int height)
220 {
221 int i, j;
222 uint8_t r, g, b;
223
224 for (j = 0; j < height; j++) {
225 for (i = 0; i < width * step; i += step) {
226 r = src[i];
227 g = src[i + 1];
228 b = src[i + 2];
229 src[i] = r + g - 0x80;
230 src[i + 2] = b + g - 0x80;
231 }
232 src += stride;
233 }
234 }
235
236 static void restore_median(uint8_t *src, int step, int stride,
237 int width, int height, int slices, int rmode)
238 {
239 int i, j, slice;
240 int A, B, C;
241 uint8_t *bsrc;
242 int slice_start, slice_height;
243 const int cmask = ~rmode;
244
245 for (slice = 0; slice < slices; slice++) {
246 slice_start = ((slice * height) / slices) & cmask;
247 slice_height = ((((slice + 1) * height) / slices) & cmask) - slice_start;
248
249 bsrc = src + slice_start * stride;
250
251 // first line - left neighbour prediction
252 bsrc[0] += 0x80;
253 A = bsrc[0];
254 for (i = step; i < width * step; i += step) {
255 bsrc[i] += A;
256 A = bsrc[i];
257 }
258 bsrc += stride;
259 if (slice_height == 1)
260 continue;
261 // second line - first element has top predition, the rest uses median
262 C = bsrc[-stride];
263 bsrc[0] += C;
264 A = bsrc[0];
265 for (i = step; i < width * step; i += step) {
266 B = bsrc[i - stride];
267 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
268 C = B;
269 A = bsrc[i];
270 }
271 bsrc += stride;
272 // the rest of lines use continuous median prediction
273 for (j = 2; j < slice_height; j++) {
274 for (i = 0; i < width * step; i += step) {
275 B = bsrc[i - stride];
276 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
277 C = B;
278 A = bsrc[i];
279 }
280 bsrc += stride;
281 }
282 }
283 }
284
285 /* UtVideo interlaced mode treats every two lines as a single one,
286 * so restoring function should take care of possible padding between
287 * two parts of the same "line".
288 */
289 static void restore_median_il(uint8_t *src, int step, int stride,
290 int width, int height, int slices, int rmode)
291 {
292 int i, j, slice;
293 int A, B, C;
294 uint8_t *bsrc;
295 int slice_start, slice_height;
296 const int cmask = ~(rmode ? 3 : 1);
297 const int stride2 = stride << 1;
298
299 for (slice = 0; slice < slices; slice++) {
300 slice_start = ((slice * height) / slices) & cmask;
301 slice_height = ((((slice + 1) * height) / slices) & cmask) - slice_start;
302 slice_height >>= 1;
303
304 bsrc = src + slice_start * stride;
305
306 // first line - left neighbour prediction
307 bsrc[0] += 0x80;
308 A = bsrc[0];
309 for (i = step; i < width * step; i += step) {
310 bsrc[i] += A;
311 A = bsrc[i];
312 }
313 for (i = 0; i < width * step; i += step) {
314 bsrc[stride + i] += A;
315 A = bsrc[stride + i];
316 }
317 bsrc += stride2;
318 if (slice_height == 1)
319 continue;
320 // second line - first element has top predition, the rest uses median
321 C = bsrc[-stride2];
322 bsrc[0] += C;
323 A = bsrc[0];
324 for (i = step; i < width * step; i += step) {
325 B = bsrc[i - stride2];
326 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
327 C = B;
328 A = bsrc[i];
329 }
330 for (i = 0; i < width * step; i += step) {
331 B = bsrc[i - stride];
332 bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
333 C = B;
334 A = bsrc[stride + i];
335 }
336 bsrc += stride2;
337 // the rest of lines use continuous median prediction
338 for (j = 2; j < slice_height; j++) {
339 for (i = 0; i < width * step; i += step) {
340 B = bsrc[i - stride2];
341 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
342 C = B;
343 A = bsrc[i];
344 }
345 for (i = 0; i < width * step; i += step) {
346 B = bsrc[i - stride];
347 bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
348 C = B;
349 A = bsrc[i + stride];
350 }
351 bsrc += stride2;
352 }
353 }
354 }
355
356 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
357 {
358 const uint8_t *buf = avpkt->data;
359 int buf_size = avpkt->size;
360 UtvideoContext *c = avctx->priv_data;
361 int i, j;
362 const uint8_t *plane_start[5];
363 int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
364 int ret;
365 GetByteContext gb;
366
367 if (c->pic.data[0])
368 ff_thread_release_buffer(avctx, &c->pic);
369
370 c->pic.reference = 1;
371 c->pic.buffer_hints = FF_BUFFER_HINTS_VALID;
372 if ((ret = ff_thread_get_buffer(avctx, &c->pic)) < 0) {
373 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
374 return ret;
375 }
376
377 ff_thread_finish_setup(avctx);
378
379 /* parse plane structure to retrieve frame flags and validate slice offsets */
380 bytestream2_init(&gb, buf, buf_size);
381 for (i = 0; i < c->planes; i++) {
382 plane_start[i] = gb.buffer;
383 if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
384 av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
385 return AVERROR_INVALIDDATA;
386 }
387 bytestream2_skipu(&gb, 256);
388 slice_start = 0;
389 slice_end = 0;
390 for (j = 0; j < c->slices; j++) {
391 slice_end = bytestream2_get_le32u(&gb);
392 slice_size = slice_end - slice_start;
393 if (slice_end <= 0 || slice_size <= 0 ||
394 bytestream2_get_bytes_left(&gb) < slice_end) {
395 av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
396 return AVERROR_INVALIDDATA;
397 }
398 slice_start = slice_end;
399 max_slice_size = FFMAX(max_slice_size, slice_size);
400 }
401 plane_size = slice_end;
402 bytestream2_skipu(&gb, plane_size);
403 }
404 plane_start[c->planes] = gb.buffer;
405 if (bytestream2_get_bytes_left(&gb) < c->frame_info_size) {
406 av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
407 return AVERROR_INVALIDDATA;
408 }
409 c->frame_info = bytestream2_get_le32u(&gb);
410 av_log(avctx, AV_LOG_DEBUG, "frame information flags %X\n", c->frame_info);
411
412 c->frame_pred = (c->frame_info >> 8) & 3;
413
414 if (c->frame_pred == PRED_GRADIENT) {
415 av_log_ask_for_sample(avctx, "Frame uses gradient prediction\n");
416 return AVERROR_PATCHWELCOME;
417 }
418
419 av_fast_malloc(&c->slice_bits, &c->slice_bits_size,
420 max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
421
422 if (!c->slice_bits) {
423 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
424 return AVERROR(ENOMEM);
425 }
426
427 switch (c->avctx->pix_fmt) {
428 case PIX_FMT_RGB24:
429 case PIX_FMT_RGBA:
430 for (i = 0; i < c->planes; i++) {
431 ret = decode_plane(c, i, c->pic.data[0] + rgb_order[i], c->planes,
432 c->pic.linesize[0], avctx->width, avctx->height,
433 plane_start[i], c->frame_pred == PRED_LEFT);
434 if (ret)
435 return ret;
436 if (c->frame_pred == PRED_MEDIAN)
437 restore_median(c->pic.data[0] + rgb_order[i], c->planes,
438 c->pic.linesize[0], avctx->width, avctx->height,
439 c->slices, 0);
440 }
441 restore_rgb_planes(c->pic.data[0], c->planes, c->pic.linesize[0],
442 avctx->width, avctx->height);
443 break;
444 case PIX_FMT_YUV420P:
445 for (i = 0; i < 3; i++) {
446 ret = decode_plane(c, i, c->pic.data[i], 1,
447 c->pic.linesize[i], avctx->width >> !!i, avctx->height >> !!i,
448 plane_start[i], c->frame_pred == PRED_LEFT);
449 if (ret)
450 return ret;
451 if (c->frame_pred == PRED_MEDIAN) {
452 if (!c->interlaced) {
453 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
454 avctx->width >> !!i, avctx->height >> !!i,
455 c->slices, !i);
456 } else {
457 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
458 avctx->width >> !!i,
459 avctx->height >> !!i,
460 c->slices, !i);
461 }
462 }
463 }
464 break;
465 case PIX_FMT_YUV422P:
466 for (i = 0; i < 3; i++) {
467 ret = decode_plane(c, i, c->pic.data[i], 1,
468 c->pic.linesize[i], avctx->width >> !!i, avctx->height,
469 plane_start[i], c->frame_pred == PRED_LEFT);
470 if (ret)
471 return ret;
472 if (c->frame_pred == PRED_MEDIAN) {
473 if (!c->interlaced) {
474 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
475 avctx->width >> !!i, avctx->height,
476 c->slices, 0);
477 } else {
478 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
479 avctx->width >> !!i, avctx->height,
480 c->slices, 0);
481 }
482 }
483 }
484 break;
485 }
486
487 c->pic.key_frame = 1;
488 c->pic.pict_type = AV_PICTURE_TYPE_I;
489 *data_size = sizeof(AVFrame);
490 *(AVFrame*)data = c->pic;
491
492 /* always report that the buffer was completely consumed */
493 return buf_size;
494 }
495
496 static av_cold int decode_init(AVCodecContext *avctx)
497 {
498 UtvideoContext * const c = avctx->priv_data;
499
500 c->avctx = avctx;
501
502 ff_dsputil_init(&c->dsp, avctx);
503
504 if (avctx->extradata_size < 16) {
505 av_log(avctx, AV_LOG_ERROR, "Insufficient extradata size %d, should be at least 16\n",
506 avctx->extradata_size);
507 return AVERROR_INVALIDDATA;
508 }
509
510 av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
511 avctx->extradata[3], avctx->extradata[2],
512 avctx->extradata[1], avctx->extradata[0]);
513 av_log(avctx, AV_LOG_DEBUG, "Original format %X\n", AV_RB32(avctx->extradata + 4));
514 c->frame_info_size = AV_RL32(avctx->extradata + 8);
515 c->flags = AV_RL32(avctx->extradata + 12);
516
517 if (c->frame_info_size != 4)
518 av_log_ask_for_sample(avctx, "Frame info is not 4 bytes\n");
519 av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08X\n", c->flags);
520 c->slices = (c->flags >> 24) + 1;
521 c->compression = c->flags & 1;
522 c->interlaced = c->flags & 0x800;
523
524 c->slice_bits_size = 0;
525
526 switch (avctx->codec_tag) {
527 case MKTAG('U', 'L', 'R', 'G'):
528 c->planes = 3;
529 avctx->pix_fmt = PIX_FMT_RGB24;
530 break;
531 case MKTAG('U', 'L', 'R', 'A'):
532 c->planes = 4;
533 avctx->pix_fmt = PIX_FMT_RGBA;
534 break;
535 case MKTAG('U', 'L', 'Y', '0'):
536 c->planes = 3;
537 avctx->pix_fmt = PIX_FMT_YUV420P;
538 break;
539 case MKTAG('U', 'L', 'Y', '2'):
540 c->planes = 3;
541 avctx->pix_fmt = PIX_FMT_YUV422P;
542 break;
543 default:
544 av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
545 avctx->codec_tag);
546 return AVERROR_INVALIDDATA;
547 }
548
549 return 0;
550 }
551
552 static av_cold int decode_end(AVCodecContext *avctx)
553 {
554 UtvideoContext * const c = avctx->priv_data;
555
556 if (c->pic.data[0])
557 ff_thread_release_buffer(avctx, &c->pic);
558
559 av_freep(&c->slice_bits);
560
561 return 0;
562 }
563
564 AVCodec ff_utvideo_decoder = {
565 .name = "utvideo",
566 .type = AVMEDIA_TYPE_VIDEO,
567 .id = CODEC_ID_UTVIDEO,
568 .priv_data_size = sizeof(UtvideoContext),
569 .init = decode_init,
570 .close = decode_end,
571 .decode = decode_frame,
572 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
573 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
574 };