f8db42e0ea3b0b610ffe3d3fb9351355c4d8d2d4
[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 #include "thread.h"
34
35 enum LagarithFrameType {
36 FRAME_RAW = 1, /**< uncompressed */
37 FRAME_U_RGB24 = 2, /**< unaligned RGB24 */
38 FRAME_ARITH_YUY2 = 3, /**< arithmetic coded YUY2 */
39 FRAME_ARITH_RGB24 = 4, /**< arithmetic coded RGB24 */
40 FRAME_SOLID_GRAY = 5, /**< solid grayscale color frame */
41 FRAME_SOLID_COLOR = 6, /**< solid non-grayscale color frame */
42 FRAME_OLD_ARITH_RGB = 7, /**< obsolete arithmetic coded RGB (no longer encoded by upstream since version 1.1.0) */
43 FRAME_ARITH_RGBA = 8, /**< arithmetic coded RGBA */
44 FRAME_SOLID_RGBA = 9, /**< solid RGBA color frame */
45 FRAME_ARITH_YV12 = 10, /**< arithmetic coded YV12 */
46 FRAME_REDUCED_RES = 11, /**< reduced resolution YV12 frame */
47 };
48
49 typedef struct LagarithContext {
50 AVCodecContext *avctx;
51 AVFrame picture;
52 DSPContext dsp;
53 int zeros; /**< number of consecutive zero bytes encountered */
54 int zeros_rem; /**< number of zero bytes remaining to output */
55 uint8_t *rgb_planes;
56 int rgb_stride;
57 } LagarithContext;
58
59 /**
60 * Compute the 52bit mantissa of 1/(double)denom.
61 * This crazy format uses floats in an entropy coder and we have to match x86
62 * rounding exactly, thus ordinary floats aren't portable enough.
63 * @param denom denominator
64 * @return 52bit mantissa
65 * @see softfloat_mul
66 */
67 static uint64_t softfloat_reciprocal(uint32_t denom)
68 {
69 int shift = av_log2(denom - 1) + 1;
70 uint64_t ret = (1ULL << 52) / denom;
71 uint64_t err = (1ULL << 52) - ret * denom;
72 ret <<= shift;
73 err <<= shift;
74 err += denom / 2;
75 return ret + err / denom;
76 }
77
78 /**
79 * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
80 * Used in combination with softfloat_reciprocal computes x/(double)denom.
81 * @param x 32bit integer factor
82 * @param mantissa mantissa of f with exponent 0
83 * @return 32bit integer value (x*f)
84 * @see softfloat_reciprocal
85 */
86 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
87 {
88 uint64_t l = x * (mantissa & 0xffffffff);
89 uint64_t h = x * (mantissa >> 32);
90 h += l >> 32;
91 l &= 0xffffffff;
92 l += 1 << av_log2(h >> 21);
93 h += l >> 32;
94 return h >> 20;
95 }
96
97 static uint8_t lag_calc_zero_run(int8_t x)
98 {
99 return (x << 1) ^ (x >> 7);
100 }
101
102 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
103 {
104 static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
105 int i;
106 int bit = 0;
107 int bits = 0;
108 int prevbit = 0;
109 unsigned val;
110
111 for (i = 0; i < 7; i++) {
112 if (prevbit && bit)
113 break;
114 prevbit = bit;
115 bit = get_bits1(gb);
116 if (bit && !prevbit)
117 bits += series[i];
118 }
119 bits--;
120 if (bits < 0 || bits > 31) {
121 *value = 0;
122 return -1;
123 } else if (bits == 0) {
124 *value = 0;
125 return 0;
126 }
127
128 val = get_bits_long(gb, bits);
129 val |= 1 << bits;
130
131 *value = val - 1;
132
133 return 0;
134 }
135
136 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
137 {
138 int i, j, scale_factor;
139 unsigned prob, cumulative_target;
140 unsigned cumul_prob = 0;
141 unsigned scaled_cumul_prob = 0;
142
143 rac->prob[0] = 0;
144 rac->prob[257] = UINT_MAX;
145 /* Read probabilities from bitstream */
146 for (i = 1; i < 257; i++) {
147 if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
148 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
149 return -1;
150 }
151 if ((uint64_t)cumul_prob + rac->prob[i] > UINT_MAX) {
152 av_log(rac->avctx, AV_LOG_ERROR, "Integer overflow encountered in cumulative probability calculation.\n");
153 return -1;
154 }
155 cumul_prob += rac->prob[i];
156 if (!rac->prob[i]) {
157 if (lag_decode_prob(gb, &prob)) {
158 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
159 return -1;
160 }
161 if (prob > 257 - i)
162 prob = 257 - i;
163 for (j = 0; j < prob; j++)
164 rac->prob[++i] = 0;
165 }
166 }
167
168 if (!cumul_prob) {
169 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
170 return -1;
171 }
172
173 /* Scale probabilities so cumulative probability is an even power of 2. */
174 scale_factor = av_log2(cumul_prob);
175
176 if (cumul_prob & (cumul_prob - 1)) {
177 uint64_t mul = softfloat_reciprocal(cumul_prob);
178 for (i = 1; i < 257; i++) {
179 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
180 scaled_cumul_prob += rac->prob[i];
181 }
182
183 scale_factor++;
184 cumulative_target = 1 << scale_factor;
185
186 if (scaled_cumul_prob > cumulative_target) {
187 av_log(rac->avctx, AV_LOG_ERROR,
188 "Scaled probabilities are larger than target!\n");
189 return -1;
190 }
191
192 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
193
194 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
195 if (rac->prob[i]) {
196 rac->prob[i]++;
197 scaled_cumul_prob--;
198 }
199 /* Comment from reference source:
200 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
201 * // since the compression change is negligable and fixing it
202 * // breaks backwards compatibilty
203 * b =- (signed int)b;
204 * b &= 0xFF;
205 * } else {
206 * b++;
207 * b &= 0x7f;
208 * }
209 */
210 }
211 }
212
213 rac->scale = scale_factor;
214
215 /* Fill probability array with cumulative probability for each symbol. */
216 for (i = 1; i < 257; i++)
217 rac->prob[i] += rac->prob[i - 1];
218
219 return 0;
220 }
221
222 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
223 uint8_t *diff, int w, int *left,
224 int *left_top)
225 {
226 /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
227 * However the &0xFF on the gradient predictor yealds incorrect output
228 * for lagarith.
229 */
230 int i;
231 uint8_t l, lt;
232
233 l = *left;
234 lt = *left_top;
235
236 for (i = 0; i < w; i++) {
237 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
238 lt = src1[i];
239 dst[i] = l;
240 }
241
242 *left = l;
243 *left_top = lt;
244 }
245
246 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
247 int width, int stride, int line)
248 {
249 int L, TL;
250
251 if (!line) {
252 /* Left prediction only for first line */
253 L = l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1,
254 width - 1, buf[0]);
255 } else {
256 /* Left pixel is actually prev_row[width] */
257 L = buf[width - stride - 1];
258
259 if (line == 1) {
260 /* Second line, left predict first pixel, the rest of the line is median predicted
261 * NOTE: In the case of RGB this pixel is top predicted */
262 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
263 } else {
264 /* Top left is 2 rows back, last pixel */
265 TL = buf[width - (2 * stride) - 1];
266 }
267
268 add_lag_median_prediction(buf, buf - stride, buf,
269 width, &L, &TL);
270 }
271 }
272
273 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
274 int width, int stride, int line,
275 int is_luma)
276 {
277 int L, TL;
278
279 if (!line) {
280 if (is_luma) {
281 buf++;
282 width--;
283 }
284 l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1, width - 1, buf[0]);
285 return;
286 }
287 if (line == 1) {
288 const int HEAD = is_luma ? 4 : 2;
289 int i;
290
291 L = buf[width - stride - 1];
292 TL = buf[HEAD - stride - 1];
293 for (i = 0; i < HEAD; i++) {
294 L += buf[i];
295 buf[i] = L;
296 }
297 buf += HEAD;
298 width -= HEAD;
299 } else {
300 TL = buf[width - (2 * stride) - 1];
301 L = buf[width - stride - 1];
302 }
303 l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
304 &L, &TL);
305 }
306
307 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
308 uint8_t *dst, int width, int stride,
309 int esc_count)
310 {
311 int i = 0;
312 int ret = 0;
313
314 if (!esc_count)
315 esc_count = -1;
316
317 /* Output any zeros remaining from the previous run */
318 handle_zeros:
319 if (l->zeros_rem) {
320 int count = FFMIN(l->zeros_rem, width - i);
321 memset(dst + i, 0, count);
322 i += count;
323 l->zeros_rem -= count;
324 }
325
326 while (i < width) {
327 dst[i] = lag_get_rac(rac);
328 ret++;
329
330 if (dst[i])
331 l->zeros = 0;
332 else
333 l->zeros++;
334
335 i++;
336 if (l->zeros == esc_count) {
337 int index = lag_get_rac(rac);
338 ret++;
339
340 l->zeros = 0;
341
342 l->zeros_rem = lag_calc_zero_run(index);
343 goto handle_zeros;
344 }
345 }
346 return ret;
347 }
348
349 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
350 const uint8_t *src, const uint8_t *src_end,
351 int width, int esc_count)
352 {
353 int i = 0;
354 int count;
355 uint8_t zero_run = 0;
356 const uint8_t *src_start = src;
357 uint8_t mask1 = -(esc_count < 2);
358 uint8_t mask2 = -(esc_count < 3);
359 uint8_t *end = dst + (width - 2);
360
361 output_zeros:
362 if (l->zeros_rem) {
363 count = FFMIN(l->zeros_rem, width - i);
364 if (end - dst < count) {
365 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
366 return AVERROR_INVALIDDATA;
367 }
368
369 memset(dst, 0, count);
370 l->zeros_rem -= count;
371 dst += count;
372 }
373
374 while (dst < end) {
375 i = 0;
376 while (!zero_run && dst + i < end) {
377 i++;
378 if (src + i >= src_end)
379 return AVERROR_INVALIDDATA;
380 zero_run =
381 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
382 }
383 if (zero_run) {
384 zero_run = 0;
385 i += esc_count;
386 memcpy(dst, src, i);
387 dst += i;
388 l->zeros_rem = lag_calc_zero_run(src[i]);
389
390 src += i + 1;
391 goto output_zeros;
392 } else {
393 memcpy(dst, src, i);
394 src += i;
395 dst += i;
396 }
397 }
398 return src_start - src;
399 }
400
401
402
403 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
404 int width, int height, int stride,
405 const uint8_t *src, int src_size)
406 {
407 int i = 0;
408 int read = 0;
409 uint32_t length;
410 uint32_t offset = 1;
411 int esc_count = src[0];
412 GetBitContext gb;
413 lag_rac rac;
414 const uint8_t *src_end = src + src_size;
415
416 rac.avctx = l->avctx;
417 l->zeros = 0;
418
419 if (esc_count < 4) {
420 length = width * height;
421 if (esc_count && AV_RL32(src + 1) < length) {
422 length = AV_RL32(src + 1);
423 offset += 4;
424 }
425
426 init_get_bits(&gb, src + offset, src_size * 8);
427
428 if (lag_read_prob_header(&rac, &gb) < 0)
429 return -1;
430
431 ff_lag_rac_init(&rac, &gb, length - stride);
432
433 for (i = 0; i < height; i++)
434 read += lag_decode_line(l, &rac, dst + (i * stride), width,
435 stride, esc_count);
436
437 if (read > length)
438 av_log(l->avctx, AV_LOG_WARNING,
439 "Output more bytes than length (%d of %d)\n", read,
440 length);
441 } else if (esc_count < 8) {
442 esc_count -= 4;
443 if (esc_count > 0) {
444 /* Zero run coding only, no range coding. */
445 for (i = 0; i < height; i++) {
446 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
447 src_end, width, esc_count);
448 if (res < 0)
449 return res;
450 src += res;
451 }
452 } else {
453 if (src_size < width * height)
454 return AVERROR_INVALIDDATA; // buffer not big enough
455 /* Plane is stored uncompressed */
456 for (i = 0; i < height; i++) {
457 memcpy(dst + (i * stride), src, width);
458 src += width;
459 }
460 }
461 } else if (esc_count == 0xff) {
462 /* Plane is a solid run of given value */
463 for (i = 0; i < height; i++)
464 memset(dst + i * stride, src[1], width);
465 /* Do not apply prediction.
466 Note: memset to 0 above, setting first value to src[1]
467 and applying prediction gives the same result. */
468 return 0;
469 } else {
470 av_log(l->avctx, AV_LOG_ERROR,
471 "Invalid zero run escape code! (%#x)\n", esc_count);
472 return -1;
473 }
474
475 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
476 for (i = 0; i < height; i++) {
477 lag_pred_line(l, dst, width, stride, i);
478 dst += stride;
479 }
480 } else {
481 for (i = 0; i < height; i++) {
482 lag_pred_line_yuy2(l, dst, width, stride, i,
483 width == l->avctx->width);
484 dst += stride;
485 }
486 }
487
488 return 0;
489 }
490
491 /**
492 * Decode a frame.
493 * @param avctx codec context
494 * @param data output AVFrame
495 * @param data_size size of output data or 0 if no picture is returned
496 * @param avpkt input packet
497 * @return number of consumed bytes on success or negative if decode fails
498 */
499 static int lag_decode_frame(AVCodecContext *avctx,
500 void *data, int *data_size, AVPacket *avpkt)
501 {
502 const uint8_t *buf = avpkt->data;
503 int buf_size = avpkt->size;
504 LagarithContext *l = avctx->priv_data;
505 AVFrame *const p = &l->picture;
506 uint8_t frametype = 0;
507 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
508 uint32_t offs[4];
509 uint8_t *srcs[4], *dst;
510 int i, j, planes = 3;
511
512 AVFrame *picture = data;
513
514 if (p->data[0])
515 ff_thread_release_buffer(avctx, p);
516
517 p->reference = 0;
518 p->key_frame = 1;
519
520 frametype = buf[0];
521
522 offset_gu = AV_RL32(buf + 1);
523 offset_bv = AV_RL32(buf + 5);
524
525 switch (frametype) {
526 case FRAME_SOLID_RGBA:
527 avctx->pix_fmt = AV_PIX_FMT_RGB32;
528
529 if (ff_thread_get_buffer(avctx, p) < 0) {
530 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
531 return -1;
532 }
533
534 dst = p->data[0];
535 for (j = 0; j < avctx->height; j++) {
536 for (i = 0; i < avctx->width; i++)
537 AV_WN32(dst + i * 4, offset_gu);
538 dst += p->linesize[0];
539 }
540 break;
541 case FRAME_ARITH_RGBA:
542 avctx->pix_fmt = AV_PIX_FMT_RGB32;
543 planes = 4;
544 offset_ry += 4;
545 offs[3] = AV_RL32(buf + 9);
546 case FRAME_ARITH_RGB24:
547 case FRAME_U_RGB24:
548 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
549 avctx->pix_fmt = AV_PIX_FMT_RGB24;
550
551 if (ff_thread_get_buffer(avctx, p) < 0) {
552 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
553 return -1;
554 }
555
556 offs[0] = offset_bv;
557 offs[1] = offset_gu;
558 offs[2] = offset_ry;
559
560 if (!l->rgb_planes) {
561 l->rgb_stride = FFALIGN(avctx->width, 16);
562 l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * planes + 1);
563 if (!l->rgb_planes) {
564 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
565 return AVERROR(ENOMEM);
566 }
567 }
568 for (i = 0; i < planes; i++)
569 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
570 if (offset_ry >= buf_size ||
571 offset_gu >= buf_size ||
572 offset_bv >= buf_size ||
573 (planes == 4 && offs[3] >= buf_size)) {
574 av_log(avctx, AV_LOG_ERROR,
575 "Invalid frame offsets\n");
576 return AVERROR_INVALIDDATA;
577 }
578 for (i = 0; i < planes; i++)
579 lag_decode_arith_plane(l, srcs[i],
580 avctx->width, avctx->height,
581 -l->rgb_stride, buf + offs[i],
582 buf_size - offs[i]);
583 dst = p->data[0];
584 for (i = 0; i < planes; i++)
585 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
586 for (j = 0; j < avctx->height; j++) {
587 for (i = 0; i < avctx->width; i++) {
588 uint8_t r, g, b, a;
589 r = srcs[0][i];
590 g = srcs[1][i];
591 b = srcs[2][i];
592 r += g;
593 b += g;
594 if (frametype == FRAME_ARITH_RGBA) {
595 a = srcs[3][i];
596 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
597 } else {
598 dst[i * 3 + 0] = r;
599 dst[i * 3 + 1] = g;
600 dst[i * 3 + 2] = b;
601 }
602 }
603 dst += p->linesize[0];
604 for (i = 0; i < planes; i++)
605 srcs[i] += l->rgb_stride;
606 }
607 break;
608 case FRAME_ARITH_YUY2:
609 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
610
611 if (ff_thread_get_buffer(avctx, p) < 0) {
612 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
613 return -1;
614 }
615
616 if (offset_ry >= buf_size ||
617 offset_gu >= buf_size ||
618 offset_bv >= buf_size) {
619 av_log(avctx, AV_LOG_ERROR,
620 "Invalid frame offsets\n");
621 return AVERROR_INVALIDDATA;
622 }
623
624 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
625 p->linesize[0], buf + offset_ry,
626 buf_size - offset_ry);
627 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
628 avctx->height, p->linesize[1],
629 buf + offset_gu, buf_size - offset_gu);
630 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
631 avctx->height, p->linesize[2],
632 buf + offset_bv, buf_size - offset_bv);
633 break;
634 case FRAME_ARITH_YV12:
635 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
636
637 if (ff_thread_get_buffer(avctx, p) < 0) {
638 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
639 return -1;
640 }
641
642 if (offset_ry >= buf_size ||
643 offset_gu >= buf_size ||
644 offset_bv >= buf_size) {
645 av_log(avctx, AV_LOG_ERROR,
646 "Invalid frame offsets\n");
647 return AVERROR_INVALIDDATA;
648 }
649
650 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
651 p->linesize[0], buf + offset_ry,
652 buf_size - offset_ry);
653 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
654 avctx->height / 2, p->linesize[2],
655 buf + offset_gu, buf_size - offset_gu);
656 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
657 avctx->height / 2, p->linesize[1],
658 buf + offset_bv, buf_size - offset_bv);
659 break;
660 default:
661 av_log(avctx, AV_LOG_ERROR,
662 "Unsupported Lagarith frame type: %#x\n", frametype);
663 return -1;
664 }
665
666 *picture = *p;
667 *data_size = sizeof(AVFrame);
668
669 return buf_size;
670 }
671
672 static av_cold int lag_decode_init(AVCodecContext *avctx)
673 {
674 LagarithContext *l = avctx->priv_data;
675 l->avctx = avctx;
676
677 ff_dsputil_init(&l->dsp, avctx);
678
679 return 0;
680 }
681
682 static av_cold int lag_decode_end(AVCodecContext *avctx)
683 {
684 LagarithContext *l = avctx->priv_data;
685
686 if (l->picture.data[0])
687 ff_thread_release_buffer(avctx, &l->picture);
688 av_freep(&l->rgb_planes);
689
690 return 0;
691 }
692
693 AVCodec ff_lagarith_decoder = {
694 .name = "lagarith",
695 .type = AVMEDIA_TYPE_VIDEO,
696 .id = AV_CODEC_ID_LAGARITH,
697 .priv_data_size = sizeof(LagarithContext),
698 .init = lag_decode_init,
699 .close = lag_decode_end,
700 .decode = lag_decode_frame,
701 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
702 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
703 };