lavc: add Ut Video encoder
[libav.git] / libavcodec / utvideodec.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 #include "utvideo.h"
36
37 static int build_huff(const uint8_t *src, VLC *vlc, int *fsym)
38 {
39 int i;
40 HuffEntry he[256];
41 int last;
42 uint32_t codes[256];
43 uint8_t bits[256];
44 uint8_t syms[256];
45 uint32_t code;
46
47 *fsym = -1;
48 for (i = 0; i < 256; i++) {
49 he[i].sym = i;
50 he[i].len = *src++;
51 }
52 qsort(he, 256, sizeof(*he), ff_ut_huff_cmp_len);
53
54 if (!he[0].len) {
55 *fsym = he[0].sym;
56 return 0;
57 }
58 if (he[0].len > 32)
59 return -1;
60
61 last = 255;
62 while (he[last].len == 255 && last)
63 last--;
64
65 code = 1;
66 for (i = last; i >= 0; i--) {
67 codes[i] = code >> (32 - he[i].len);
68 bits[i] = he[i].len;
69 syms[i] = he[i].sym;
70 code += 0x80000000u >> (he[i].len - 1);
71 }
72
73 return ff_init_vlc_sparse(vlc, FFMIN(he[last].len, 9), last + 1,
74 bits, sizeof(*bits), sizeof(*bits),
75 codes, sizeof(*codes), sizeof(*codes),
76 syms, sizeof(*syms), sizeof(*syms), 0);
77 }
78
79 static int decode_plane(UtvideoContext *c, int plane_no,
80 uint8_t *dst, int step, int stride,
81 int width, int height,
82 const uint8_t *src, int use_pred)
83 {
84 int i, j, slice, pix;
85 int sstart, send;
86 VLC vlc;
87 GetBitContext gb;
88 int prev, fsym;
89 const int cmask = ~(!plane_no && c->avctx->pix_fmt == PIX_FMT_YUV420P);
90
91 if (build_huff(src, &vlc, &fsym)) {
92 av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
93 return AVERROR_INVALIDDATA;
94 }
95 if (fsym >= 0) { // build_huff reported a symbol to fill slices with
96 send = 0;
97 for (slice = 0; slice < c->slices; slice++) {
98 uint8_t *dest;
99
100 sstart = send;
101 send = (height * (slice + 1) / c->slices) & cmask;
102 dest = dst + sstart * stride;
103
104 prev = 0x80;
105 for (j = sstart; j < send; j++) {
106 for (i = 0; i < width * step; i += step) {
107 pix = fsym;
108 if (use_pred) {
109 prev += pix;
110 pix = prev;
111 }
112 dest[i] = pix;
113 }
114 dest += stride;
115 }
116 }
117 return 0;
118 }
119
120 src += 256;
121
122 send = 0;
123 for (slice = 0; slice < c->slices; slice++) {
124 uint8_t *dest;
125 int slice_data_start, slice_data_end, slice_size;
126
127 sstart = send;
128 send = (height * (slice + 1) / c->slices) & cmask;
129 dest = dst + sstart * stride;
130
131 // slice offset and size validation was done earlier
132 slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
133 slice_data_end = AV_RL32(src + slice * 4);
134 slice_size = slice_data_end - slice_data_start;
135
136 if (!slice_size) {
137 for (j = sstart; j < send; j++) {
138 for (i = 0; i < width * step; i += step)
139 dest[i] = 0x80;
140 dest += stride;
141 }
142 continue;
143 }
144
145 memcpy(c->slice_bits, src + slice_data_start + c->slices * 4,
146 slice_size);
147 memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
148 c->dsp.bswap_buf((uint32_t *) c->slice_bits, (uint32_t *) c->slice_bits,
149 (slice_data_end - slice_data_start + 3) >> 2);
150 init_get_bits(&gb, c->slice_bits, slice_size * 8);
151
152 prev = 0x80;
153 for (j = sstart; j < send; j++) {
154 for (i = 0; i < width * step; i += step) {
155 if (get_bits_left(&gb) <= 0) {
156 av_log(c->avctx, AV_LOG_ERROR,
157 "Slice decoding ran out of bits\n");
158 goto fail;
159 }
160 pix = get_vlc2(&gb, vlc.table, vlc.bits, 4);
161 if (pix < 0) {
162 av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
163 goto fail;
164 }
165 if (use_pred) {
166 prev += pix;
167 pix = prev;
168 }
169 dest[i] = pix;
170 }
171 dest += stride;
172 }
173 if (get_bits_left(&gb) > 32)
174 av_log(c->avctx, AV_LOG_WARNING,
175 "%d bits left after decoding slice\n", get_bits_left(&gb));
176 }
177
178 ff_free_vlc(&vlc);
179
180 return 0;
181 fail:
182 ff_free_vlc(&vlc);
183 return AVERROR_INVALIDDATA;
184 }
185
186 static void restore_rgb_planes(uint8_t *src, int step, int stride, int width,
187 int height)
188 {
189 int i, j;
190 uint8_t r, g, b;
191
192 for (j = 0; j < height; j++) {
193 for (i = 0; i < width * step; i += step) {
194 r = src[i];
195 g = src[i + 1];
196 b = src[i + 2];
197 src[i] = r + g - 0x80;
198 src[i + 2] = b + g - 0x80;
199 }
200 src += stride;
201 }
202 }
203
204 static void restore_median(uint8_t *src, int step, int stride,
205 int width, int height, int slices, int rmode)
206 {
207 int i, j, slice;
208 int A, B, C;
209 uint8_t *bsrc;
210 int slice_start, slice_height;
211 const int cmask = ~rmode;
212
213 for (slice = 0; slice < slices; slice++) {
214 slice_start = ((slice * height) / slices) & cmask;
215 slice_height = ((((slice + 1) * height) / slices) & cmask) -
216 slice_start;
217
218 bsrc = src + slice_start * stride;
219
220 // first line - left neighbour prediction
221 bsrc[0] += 0x80;
222 A = bsrc[0];
223 for (i = step; i < width * step; i += step) {
224 bsrc[i] += A;
225 A = bsrc[i];
226 }
227 bsrc += stride;
228 if (slice_height == 1)
229 continue;
230 // second line - first element has top prediction, the rest uses median
231 C = bsrc[-stride];
232 bsrc[0] += C;
233 A = bsrc[0];
234 for (i = step; i < width * step; i += step) {
235 B = bsrc[i - stride];
236 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
237 C = B;
238 A = bsrc[i];
239 }
240 bsrc += stride;
241 // the rest of lines use continuous median prediction
242 for (j = 2; j < slice_height; j++) {
243 for (i = 0; i < width * step; i += step) {
244 B = bsrc[i - stride];
245 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
246 C = B;
247 A = bsrc[i];
248 }
249 bsrc += stride;
250 }
251 }
252 }
253
254 /* UtVideo interlaced mode treats every two lines as a single one,
255 * so restoring function should take care of possible padding between
256 * two parts of the same "line".
257 */
258 static void restore_median_il(uint8_t *src, int step, int stride,
259 int width, int height, int slices, int rmode)
260 {
261 int i, j, slice;
262 int A, B, C;
263 uint8_t *bsrc;
264 int slice_start, slice_height;
265 const int cmask = ~(rmode ? 3 : 1);
266 const int stride2 = stride << 1;
267
268 for (slice = 0; slice < slices; slice++) {
269 slice_start = ((slice * height) / slices) & cmask;
270 slice_height = ((((slice + 1) * height) / slices) & cmask) -
271 slice_start;
272 slice_height >>= 1;
273
274 bsrc = src + slice_start * stride;
275
276 // first line - left neighbour prediction
277 bsrc[0] += 0x80;
278 A = bsrc[0];
279 for (i = step; i < width * step; i += step) {
280 bsrc[i] += A;
281 A = bsrc[i];
282 }
283 for (i = 0; i < width * step; i += step) {
284 bsrc[stride + i] += A;
285 A = bsrc[stride + i];
286 }
287 bsrc += stride2;
288 if (slice_height == 1)
289 continue;
290 // second line - first element has top prediction, the rest uses median
291 C = bsrc[-stride2];
292 bsrc[0] += C;
293 A = bsrc[0];
294 for (i = step; i < width * step; i += step) {
295 B = bsrc[i - stride2];
296 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
297 C = B;
298 A = bsrc[i];
299 }
300 for (i = 0; i < width * step; i += step) {
301 B = bsrc[i - stride];
302 bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
303 C = B;
304 A = bsrc[stride + i];
305 }
306 bsrc += stride2;
307 // the rest of lines use continuous median prediction
308 for (j = 2; j < slice_height; j++) {
309 for (i = 0; i < width * step; i += step) {
310 B = bsrc[i - stride2];
311 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
312 C = B;
313 A = bsrc[i];
314 }
315 for (i = 0; i < width * step; i += step) {
316 B = bsrc[i - stride];
317 bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
318 C = B;
319 A = bsrc[i + stride];
320 }
321 bsrc += stride2;
322 }
323 }
324 }
325
326 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
327 AVPacket *avpkt)
328 {
329 const uint8_t *buf = avpkt->data;
330 int buf_size = avpkt->size;
331 UtvideoContext *c = avctx->priv_data;
332 int i, j;
333 const uint8_t *plane_start[5];
334 int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
335 int ret;
336 GetByteContext gb;
337
338 if (c->pic.data[0])
339 ff_thread_release_buffer(avctx, &c->pic);
340
341 c->pic.reference = 1;
342 c->pic.buffer_hints = FF_BUFFER_HINTS_VALID;
343 if ((ret = ff_thread_get_buffer(avctx, &c->pic)) < 0) {
344 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
345 return ret;
346 }
347
348 ff_thread_finish_setup(avctx);
349
350 /* parse plane structure to get frame flags and validate slice offsets */
351 bytestream2_init(&gb, buf, buf_size);
352 for (i = 0; i < c->planes; i++) {
353 plane_start[i] = gb.buffer;
354 if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
355 av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
356 return AVERROR_INVALIDDATA;
357 }
358 bytestream2_skipu(&gb, 256);
359 slice_start = 0;
360 slice_end = 0;
361 for (j = 0; j < c->slices; j++) {
362 slice_end = bytestream2_get_le32u(&gb);
363 slice_size = slice_end - slice_start;
364 if (slice_end <= 0 || slice_size <= 0 ||
365 bytestream2_get_bytes_left(&gb) < slice_end) {
366 av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
367 return AVERROR_INVALIDDATA;
368 }
369 slice_start = slice_end;
370 max_slice_size = FFMAX(max_slice_size, slice_size);
371 }
372 plane_size = slice_end;
373 bytestream2_skipu(&gb, plane_size);
374 }
375 plane_start[c->planes] = gb.buffer;
376 if (bytestream2_get_bytes_left(&gb) < c->frame_info_size) {
377 av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
378 return AVERROR_INVALIDDATA;
379 }
380 c->frame_info = bytestream2_get_le32u(&gb);
381 av_log(avctx, AV_LOG_DEBUG, "frame information flags %X\n", c->frame_info);
382
383 c->frame_pred = (c->frame_info >> 8) & 3;
384
385 if (c->frame_pred == PRED_GRADIENT) {
386 av_log_ask_for_sample(avctx, "Frame uses gradient prediction\n");
387 return AVERROR_PATCHWELCOME;
388 }
389
390 av_fast_malloc(&c->slice_bits, &c->slice_bits_size,
391 max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
392
393 if (!c->slice_bits) {
394 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
395 return AVERROR(ENOMEM);
396 }
397
398 switch (c->avctx->pix_fmt) {
399 case PIX_FMT_RGB24:
400 case PIX_FMT_RGBA:
401 for (i = 0; i < c->planes; i++) {
402 ret = decode_plane(c, i, c->pic.data[0] + ff_ut_rgb_order[i],
403 c->planes, c->pic.linesize[0], avctx->width,
404 avctx->height, plane_start[i],
405 c->frame_pred == PRED_LEFT);
406 if (ret)
407 return ret;
408 if (c->frame_pred == PRED_MEDIAN) {
409 if (!c->interlaced) {
410 restore_median(c->pic.data[0] + ff_ut_rgb_order[i],
411 c->planes, c->pic.linesize[0], avctx->width,
412 avctx->height, c->slices, 0);
413 } else {
414 restore_median_il(c->pic.data[0] + ff_ut_rgb_order[i],
415 c->planes, c->pic.linesize[0],
416 avctx->width, avctx->height, c->slices,
417 0);
418 }
419 }
420 }
421 restore_rgb_planes(c->pic.data[0], c->planes, c->pic.linesize[0],
422 avctx->width, avctx->height);
423 break;
424 case PIX_FMT_YUV420P:
425 for (i = 0; i < 3; i++) {
426 ret = decode_plane(c, i, c->pic.data[i], 1, c->pic.linesize[i],
427 avctx->width >> !!i, avctx->height >> !!i,
428 plane_start[i], c->frame_pred == PRED_LEFT);
429 if (ret)
430 return ret;
431 if (c->frame_pred == PRED_MEDIAN) {
432 if (!c->interlaced) {
433 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
434 avctx->width >> !!i, avctx->height >> !!i,
435 c->slices, !i);
436 } else {
437 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
438 avctx->width >> !!i,
439 avctx->height >> !!i,
440 c->slices, !i);
441 }
442 }
443 }
444 break;
445 case PIX_FMT_YUV422P:
446 for (i = 0; i < 3; i++) {
447 ret = decode_plane(c, i, c->pic.data[i], 1, c->pic.linesize[i],
448 avctx->width >> !!i, avctx->height,
449 plane_start[i], c->frame_pred == PRED_LEFT);
450 if (ret)
451 return ret;
452 if (c->frame_pred == PRED_MEDIAN) {
453 if (!c->interlaced) {
454 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
455 avctx->width >> !!i, avctx->height,
456 c->slices, 0);
457 } else {
458 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
459 avctx->width >> !!i, avctx->height,
460 c->slices, 0);
461 }
462 }
463 }
464 break;
465 }
466
467 c->pic.key_frame = 1;
468 c->pic.pict_type = AV_PICTURE_TYPE_I;
469 c->pic.interlaced_frame = !!c->interlaced;
470
471 *data_size = sizeof(AVFrame);
472 *(AVFrame*)data = c->pic;
473
474 /* always report that the buffer was completely consumed */
475 return buf_size;
476 }
477
478 static av_cold int decode_init(AVCodecContext *avctx)
479 {
480 UtvideoContext * const c = avctx->priv_data;
481
482 c->avctx = avctx;
483
484 ff_dsputil_init(&c->dsp, avctx);
485
486 if (avctx->extradata_size < 16) {
487 av_log(avctx, AV_LOG_ERROR,
488 "Insufficient extradata size %d, should be at least 16\n",
489 avctx->extradata_size);
490 return AVERROR_INVALIDDATA;
491 }
492
493 av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
494 avctx->extradata[3], avctx->extradata[2],
495 avctx->extradata[1], avctx->extradata[0]);
496 av_log(avctx, AV_LOG_DEBUG, "Original format %X\n",
497 AV_RB32(avctx->extradata + 4));
498 c->frame_info_size = AV_RL32(avctx->extradata + 8);
499 c->flags = AV_RL32(avctx->extradata + 12);
500
501 if (c->frame_info_size != 4)
502 av_log_ask_for_sample(avctx, "Frame info is not 4 bytes\n");
503 av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08X\n", c->flags);
504 c->slices = (c->flags >> 24) + 1;
505 c->compression = c->flags & 1;
506 c->interlaced = c->flags & 0x800;
507
508 c->slice_bits_size = 0;
509
510 switch (avctx->codec_tag) {
511 case MKTAG('U', 'L', 'R', 'G'):
512 c->planes = 3;
513 avctx->pix_fmt = PIX_FMT_RGB24;
514 break;
515 case MKTAG('U', 'L', 'R', 'A'):
516 c->planes = 4;
517 avctx->pix_fmt = PIX_FMT_RGBA;
518 break;
519 case MKTAG('U', 'L', 'Y', '0'):
520 c->planes = 3;
521 avctx->pix_fmt = PIX_FMT_YUV420P;
522 break;
523 case MKTAG('U', 'L', 'Y', '2'):
524 c->planes = 3;
525 avctx->pix_fmt = PIX_FMT_YUV422P;
526 break;
527 default:
528 av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
529 avctx->codec_tag);
530 return AVERROR_INVALIDDATA;
531 }
532
533 return 0;
534 }
535
536 static av_cold int decode_end(AVCodecContext *avctx)
537 {
538 UtvideoContext * const c = avctx->priv_data;
539
540 if (c->pic.data[0])
541 ff_thread_release_buffer(avctx, &c->pic);
542
543 av_freep(&c->slice_bits);
544
545 return 0;
546 }
547
548 AVCodec ff_utvideo_decoder = {
549 .name = "utvideo",
550 .type = AVMEDIA_TYPE_VIDEO,
551 .id = AV_CODEC_ID_UTVIDEO,
552 .priv_data_size = sizeof(UtvideoContext),
553 .init = decode_init,
554 .close = decode_end,
555 .decode = decode_frame,
556 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
557 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
558 };