3a47f4c632b5db76ed186a1e62ed4532fef6ce03
[libav.git] / libavcodec / lagarith.c
1 /*
2 * Lagarith lossless decoder
3 * Copyright (c) 2009 Nathan Caldwell <saintdev (at) gmail.com>
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 * Lagarith lossless decoder
25 * @author Nathan Caldwell
26 */
27
28 #include "avcodec.h"
29 #include "get_bits.h"
30 #include "mathops.h"
31 #include "dsputil.h"
32 #include "lagarithrac.h"
33
34 enum LagarithFrameType {
35 FRAME_RAW = 1, /**< uncompressed */
36 FRAME_U_RGB24 = 2, /**< unaligned RGB24 */
37 FRAME_ARITH_YUY2 = 3, /**< arithmetic coded YUY2 */
38 FRAME_ARITH_RGB24 = 4, /**< arithmetic coded RGB24 */
39 FRAME_SOLID_GRAY = 5, /**< solid grayscale color frame */
40 FRAME_SOLID_COLOR = 6, /**< solid non-grayscale color frame */
41 FRAME_OLD_ARITH_RGB = 7, /**< obsolete arithmetic coded RGB (no longer encoded by upstream since version 1.1.0) */
42 FRAME_ARITH_RGBA = 8, /**< arithmetic coded RGBA */
43 FRAME_SOLID_RGBA = 9, /**< solid RGBA color frame */
44 FRAME_ARITH_YV12 = 10, /**< arithmetic coded YV12 */
45 FRAME_REDUCED_RES = 11, /**< reduced resolution YV12 frame */
46 };
47
48 typedef struct LagarithContext {
49 AVCodecContext *avctx;
50 AVFrame picture;
51 DSPContext dsp;
52 int zeros; /**< number of consecutive zero bytes encountered */
53 int zeros_rem; /**< number of zero bytes remaining to output */
54 uint8_t *rgb_planes;
55 int rgb_stride;
56 } LagarithContext;
57
58 /**
59 * Compute the 52bit mantissa of 1/(double)denom.
60 * This crazy format uses floats in an entropy coder and we have to match x86
61 * rounding exactly, thus ordinary floats aren't portable enough.
62 * @param denom denominator
63 * @return 52bit mantissa
64 * @see softfloat_mul
65 */
66 static uint64_t softfloat_reciprocal(uint32_t denom)
67 {
68 int shift = av_log2(denom - 1) + 1;
69 uint64_t ret = (1ULL << 52) / denom;
70 uint64_t err = (1ULL << 52) - ret * denom;
71 ret <<= shift;
72 err <<= shift;
73 err += denom / 2;
74 return ret + err / denom;
75 }
76
77 /**
78 * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
79 * Used in combination with softfloat_reciprocal computes x/(double)denom.
80 * @param x 32bit integer factor
81 * @param mantissa mantissa of f with exponent 0
82 * @return 32bit integer value (x*f)
83 * @see softfloat_reciprocal
84 */
85 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
86 {
87 uint64_t l = x * (mantissa & 0xffffffff);
88 uint64_t h = x * (mantissa >> 32);
89 h += l >> 32;
90 l &= 0xffffffff;
91 l += 1 << av_log2(h >> 21);
92 h += l >> 32;
93 return h >> 20;
94 }
95
96 static uint8_t lag_calc_zero_run(int8_t x)
97 {
98 return (x << 1) ^ (x >> 7);
99 }
100
101 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
102 {
103 static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
104 int i;
105 int bit = 0;
106 int bits = 0;
107 int prevbit = 0;
108 unsigned val;
109
110 for (i = 0; i < 7; i++) {
111 if (prevbit && bit)
112 break;
113 prevbit = bit;
114 bit = get_bits1(gb);
115 if (bit && !prevbit)
116 bits += series[i];
117 }
118 bits--;
119 if (bits < 0 || bits > 31) {
120 *value = 0;
121 return -1;
122 } else if (bits == 0) {
123 *value = 0;
124 return 0;
125 }
126
127 val = get_bits_long(gb, bits);
128 val |= 1 << bits;
129
130 *value = val - 1;
131
132 return 0;
133 }
134
135 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
136 {
137 int i, j, scale_factor;
138 unsigned prob, cumulative_target;
139 unsigned cumul_prob = 0;
140 unsigned scaled_cumul_prob = 0;
141
142 rac->prob[0] = 0;
143 rac->prob[257] = UINT_MAX;
144 /* Read probabilities from bitstream */
145 for (i = 1; i < 257; i++) {
146 if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
147 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
148 return -1;
149 }
150 if ((uint64_t)cumul_prob + rac->prob[i] > UINT_MAX) {
151 av_log(rac->avctx, AV_LOG_ERROR, "Integer overflow encountered in cumulative probability calculation.\n");
152 return -1;
153 }
154 cumul_prob += rac->prob[i];
155 if (!rac->prob[i]) {
156 if (lag_decode_prob(gb, &prob)) {
157 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
158 return -1;
159 }
160 if (prob > 257 - i)
161 prob = 257 - i;
162 for (j = 0; j < prob; j++)
163 rac->prob[++i] = 0;
164 }
165 }
166
167 if (!cumul_prob) {
168 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
169 return -1;
170 }
171
172 /* Scale probabilities so cumulative probability is an even power of 2. */
173 scale_factor = av_log2(cumul_prob);
174
175 if (cumul_prob & (cumul_prob - 1)) {
176 uint64_t mul = softfloat_reciprocal(cumul_prob);
177 for (i = 1; i < 257; i++) {
178 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
179 scaled_cumul_prob += rac->prob[i];
180 }
181
182 scale_factor++;
183 cumulative_target = 1 << scale_factor;
184
185 if (scaled_cumul_prob > cumulative_target) {
186 av_log(rac->avctx, AV_LOG_ERROR,
187 "Scaled probabilities are larger than target!\n");
188 return -1;
189 }
190
191 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
192
193 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
194 if (rac->prob[i]) {
195 rac->prob[i]++;
196 scaled_cumul_prob--;
197 }
198 /* Comment from reference source:
199 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
200 * // since the compression change is negligable and fixing it
201 * // breaks backwards compatibilty
202 * b =- (signed int)b;
203 * b &= 0xFF;
204 * } else {
205 * b++;
206 * b &= 0x7f;
207 * }
208 */
209 }
210 }
211
212 rac->scale = scale_factor;
213
214 /* Fill probability array with cumulative probability for each symbol. */
215 for (i = 1; i < 257; i++)
216 rac->prob[i] += rac->prob[i - 1];
217
218 return 0;
219 }
220
221 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
222 uint8_t *diff, int w, int *left,
223 int *left_top)
224 {
225 /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
226 * However the &0xFF on the gradient predictor yealds incorrect output
227 * for lagarith.
228 */
229 int i;
230 uint8_t l, lt;
231
232 l = *left;
233 lt = *left_top;
234
235 for (i = 0; i < w; i++) {
236 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
237 lt = src1[i];
238 dst[i] = l;
239 }
240
241 *left = l;
242 *left_top = lt;
243 }
244
245 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
246 int width, int stride, int line)
247 {
248 int L, TL;
249
250 /* Left pixel is actually prev_row[width] */
251 L = buf[width - stride - 1];
252 if (!line) {
253 /* Left prediction only for first line */
254 L = l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1,
255 width - 1, buf[0]);
256 return;
257 } else if (line == 1) {
258 /* Second line, left predict first pixel, the rest of the line is median predicted
259 * NOTE: In the case of RGB this pixel is top predicted */
260 TL = l->avctx->pix_fmt == PIX_FMT_YUV420P ? buf[-stride] : L;
261 } else {
262 /* Top left is 2 rows back, last pixel */
263 TL = buf[width - (2 * stride) - 1];
264 }
265
266 add_lag_median_prediction(buf, buf - stride, buf,
267 width, &L, &TL);
268 }
269
270 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
271 uint8_t *dst, int width, int stride,
272 int esc_count)
273 {
274 int i = 0;
275 int ret = 0;
276
277 if (!esc_count)
278 esc_count = -1;
279
280 /* Output any zeros remaining from the previous run */
281 handle_zeros:
282 if (l->zeros_rem) {
283 int count = FFMIN(l->zeros_rem, width - i);
284 memset(dst + i, 0, count);
285 i += count;
286 l->zeros_rem -= count;
287 }
288
289 while (i < width) {
290 dst[i] = lag_get_rac(rac);
291 ret++;
292
293 if (dst[i])
294 l->zeros = 0;
295 else
296 l->zeros++;
297
298 i++;
299 if (l->zeros == esc_count) {
300 int index = lag_get_rac(rac);
301 ret++;
302
303 l->zeros = 0;
304
305 l->zeros_rem = lag_calc_zero_run(index);
306 goto handle_zeros;
307 }
308 }
309 return ret;
310 }
311
312 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
313 const uint8_t *src, int width,
314 int esc_count)
315 {
316 int i = 0;
317 int count;
318 uint8_t zero_run = 0;
319 const uint8_t *start = src;
320 uint8_t mask1 = -(esc_count < 2);
321 uint8_t mask2 = -(esc_count < 3);
322 uint8_t *end = dst + (width - 2);
323
324 output_zeros:
325 if (l->zeros_rem) {
326 count = FFMIN(l->zeros_rem, width - i);
327 memset(dst, 0, count);
328 l->zeros_rem -= count;
329 dst += count;
330 }
331
332 while (dst < end) {
333 i = 0;
334 while (!zero_run && dst + i < end) {
335 i++;
336 zero_run =
337 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
338 }
339 if (zero_run) {
340 zero_run = 0;
341 i += esc_count;
342 memcpy(dst, src, i);
343 dst += i;
344 l->zeros_rem = lag_calc_zero_run(src[i]);
345
346 src += i + 1;
347 goto output_zeros;
348 } else {
349 memcpy(dst, src, i);
350 src += i;
351 }
352 }
353 return start - src;
354 }
355
356
357
358 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
359 int width, int height, int stride,
360 const uint8_t *src, int src_size)
361 {
362 int i = 0;
363 int read = 0;
364 uint32_t length;
365 uint32_t offset = 1;
366 int esc_count = src[0];
367 GetBitContext gb;
368 lag_rac rac;
369
370 rac.avctx = l->avctx;
371 l->zeros = 0;
372
373 if (esc_count < 4) {
374 length = width * height;
375 if (esc_count && AV_RL32(src + 1) < length) {
376 length = AV_RL32(src + 1);
377 offset += 4;
378 }
379
380 init_get_bits(&gb, src + offset, src_size * 8);
381
382 if (lag_read_prob_header(&rac, &gb) < 0)
383 return -1;
384
385 ff_lag_rac_init(&rac, &gb, length - stride);
386
387 for (i = 0; i < height; i++)
388 read += lag_decode_line(l, &rac, dst + (i * stride), width,
389 stride, esc_count);
390
391 if (read > length)
392 av_log(l->avctx, AV_LOG_WARNING,
393 "Output more bytes than length (%d of %d)\n", read,
394 length);
395 } else if (esc_count < 8) {
396 esc_count -= 4;
397 if (esc_count > 0) {
398 /* Zero run coding only, no range coding. */
399 for (i = 0; i < height; i++)
400 src += lag_decode_zero_run_line(l, dst + (i * stride), src,
401 width, esc_count);
402 } else {
403 /* Plane is stored uncompressed */
404 for (i = 0; i < height; i++) {
405 memcpy(dst + (i * stride), src, width);
406 src += width;
407 }
408 }
409 } else if (esc_count == 0xff) {
410 /* Plane is a solid run of given value */
411 for (i = 0; i < height; i++)
412 memset(dst + i * stride, src[1], width);
413 /* Do not apply prediction.
414 Note: memset to 0 above, setting first value to src[1]
415 and applying prediction gives the same result. */
416 return 0;
417 } else {
418 av_log(l->avctx, AV_LOG_ERROR,
419 "Invalid zero run escape code! (%#x)\n", esc_count);
420 return -1;
421 }
422
423 for (i = 0; i < height; i++) {
424 lag_pred_line(l, dst, width, stride, i);
425 dst += stride;
426 }
427
428 return 0;
429 }
430
431 /**
432 * Decode a frame.
433 * @param avctx codec context
434 * @param data output AVFrame
435 * @param data_size size of output data or 0 if no picture is returned
436 * @param avpkt input packet
437 * @return number of consumed bytes on success or negative if decode fails
438 */
439 static int lag_decode_frame(AVCodecContext *avctx,
440 void *data, int *data_size, AVPacket *avpkt)
441 {
442 const uint8_t *buf = avpkt->data;
443 int buf_size = avpkt->size;
444 LagarithContext *l = avctx->priv_data;
445 AVFrame *const p = &l->picture;
446 uint8_t frametype = 0;
447 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
448 int offs[4];
449 uint8_t *srcs[4], *dst;
450 int i, j, planes = 3;
451
452 AVFrame *picture = data;
453
454 if (p->data[0])
455 avctx->release_buffer(avctx, p);
456
457 p->reference = 0;
458 p->key_frame = 1;
459
460 frametype = buf[0];
461
462 offset_gu = AV_RL32(buf + 1);
463 offset_bv = AV_RL32(buf + 5);
464
465 switch (frametype) {
466 case FRAME_SOLID_RGBA:
467 avctx->pix_fmt = PIX_FMT_RGB32;
468
469 if (avctx->get_buffer(avctx, p) < 0) {
470 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
471 return -1;
472 }
473
474 dst = p->data[0];
475 for (j = 0; j < avctx->height; j++) {
476 for (i = 0; i < avctx->width; i++)
477 AV_WN32(dst + i * 4, offset_gu);
478 dst += p->linesize[0];
479 }
480 break;
481 case FRAME_ARITH_RGBA:
482 avctx->pix_fmt = PIX_FMT_RGB32;
483 planes = 4;
484 offset_ry += 4;
485 offs[3] = AV_RL32(buf + 9);
486 case FRAME_ARITH_RGB24:
487 if (frametype == FRAME_ARITH_RGB24)
488 avctx->pix_fmt = PIX_FMT_RGB24;
489
490 if (avctx->get_buffer(avctx, p) < 0) {
491 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
492 return -1;
493 }
494
495 offs[0] = offset_bv;
496 offs[1] = offset_gu;
497 offs[2] = offset_ry;
498
499 if (!l->rgb_planes) {
500 l->rgb_stride = FFALIGN(avctx->width, 16);
501 l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * planes);
502 if (!l->rgb_planes) {
503 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
504 return AVERROR(ENOMEM);
505 }
506 }
507 for (i = 0; i < planes; i++)
508 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
509 for (i = 0; i < planes; i++)
510 lag_decode_arith_plane(l, srcs[i],
511 avctx->width, avctx->height,
512 -l->rgb_stride, buf + offs[i],
513 buf_size);
514 dst = p->data[0];
515 for (i = 0; i < planes; i++)
516 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
517 for (j = 0; j < avctx->height; j++) {
518 for (i = 0; i < avctx->width; i++) {
519 uint8_t r, g, b, a;
520 r = srcs[0][i];
521 g = srcs[1][i];
522 b = srcs[2][i];
523 r += g;
524 b += g;
525 if (frametype == FRAME_ARITH_RGBA) {
526 a = srcs[3][i];
527 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
528 } else {
529 dst[i * 3 + 0] = r;
530 dst[i * 3 + 1] = g;
531 dst[i * 3 + 2] = b;
532 }
533 }
534 dst += p->linesize[0];
535 for (i = 0; i < planes; i++)
536 srcs[i] += l->rgb_stride;
537 }
538 break;
539 case FRAME_ARITH_YV12:
540 avctx->pix_fmt = PIX_FMT_YUV420P;
541
542 if (avctx->get_buffer(avctx, p) < 0) {
543 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
544 return -1;
545 }
546
547 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
548 p->linesize[0], buf + offset_ry,
549 buf_size);
550 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
551 avctx->height / 2, p->linesize[2],
552 buf + offset_gu, buf_size);
553 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
554 avctx->height / 2, p->linesize[1],
555 buf + offset_bv, buf_size);
556 break;
557 default:
558 av_log(avctx, AV_LOG_ERROR,
559 "Unsupported Lagarith frame type: %#x\n", frametype);
560 return -1;
561 }
562
563 *picture = *p;
564 *data_size = sizeof(AVFrame);
565
566 return buf_size;
567 }
568
569 static av_cold int lag_decode_init(AVCodecContext *avctx)
570 {
571 LagarithContext *l = avctx->priv_data;
572 l->avctx = avctx;
573
574 ff_dsputil_init(&l->dsp, avctx);
575
576 return 0;
577 }
578
579 static av_cold int lag_decode_end(AVCodecContext *avctx)
580 {
581 LagarithContext *l = avctx->priv_data;
582
583 if (l->picture.data[0])
584 avctx->release_buffer(avctx, &l->picture);
585 av_freep(&l->rgb_planes);
586
587 return 0;
588 }
589
590 AVCodec ff_lagarith_decoder = {
591 .name = "lagarith",
592 .type = AVMEDIA_TYPE_VIDEO,
593 .id = CODEC_ID_LAGARITH,
594 .priv_data_size = sizeof(LagarithContext),
595 .init = lag_decode_init,
596 .close = lag_decode_end,
597 .decode = lag_decode_frame,
598 .capabilities = CODEC_CAP_DR1,
599 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
600 };