lagarith: reallocate rgb_planes when needed
[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 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_planes_allocated;
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 negligible and fixing it
202 * // breaks backwards compatibility
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 int i, align_width = (width - 1) & ~31;
253 /* Left prediction only for first line */
254 L = l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1,
255 align_width, buf[0]);
256 for (i = align_width + 1; i < width; i++)
257 buf[i] += buf[i - 1];
258 } else {
259 /* Left pixel is actually prev_row[width] */
260 L = buf[width - stride - 1];
261
262 if (line == 1) {
263 /* Second line, left predict first pixel, the rest of the line is median predicted
264 * NOTE: In the case of RGB this pixel is top predicted */
265 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
266 } else {
267 /* Top left is 2 rows back, last pixel */
268 TL = buf[width - (2 * stride) - 1];
269 }
270
271 add_lag_median_prediction(buf, buf - stride, buf,
272 width, &L, &TL);
273 }
274 }
275
276 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
277 int width, int stride, int line,
278 int is_luma)
279 {
280 int L, TL;
281
282 if (!line) {
283 int i, align_width;
284 if (is_luma) {
285 buf++;
286 width--;
287 }
288
289 align_width = (width - 1) & ~31;
290 l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1, align_width, buf[0]);
291
292 for (i = align_width + 1; i < width; i++)
293 buf[i] += buf[i - 1];
294
295 return;
296 }
297 if (line == 1) {
298 const int HEAD = is_luma ? 4 : 2;
299 int i;
300
301 L = buf[width - stride - 1];
302 TL = buf[HEAD - stride - 1];
303 for (i = 0; i < HEAD; i++) {
304 L += buf[i];
305 buf[i] = L;
306 }
307 for (; i < width; i++) {
308 L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
309 TL = buf[i - stride];
310 buf[i] = L;
311 }
312 } else {
313 TL = buf[width - (2 * stride) - 1];
314 L = buf[width - stride - 1];
315 l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
316 &L, &TL);
317 }
318 }
319
320 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
321 uint8_t *dst, int width, int stride,
322 int esc_count)
323 {
324 int i = 0;
325 int ret = 0;
326
327 if (!esc_count)
328 esc_count = -1;
329
330 /* Output any zeros remaining from the previous run */
331 handle_zeros:
332 if (l->zeros_rem) {
333 int count = FFMIN(l->zeros_rem, width - i);
334 memset(dst + i, 0, count);
335 i += count;
336 l->zeros_rem -= count;
337 }
338
339 while (i < width) {
340 dst[i] = lag_get_rac(rac);
341 ret++;
342
343 if (dst[i])
344 l->zeros = 0;
345 else
346 l->zeros++;
347
348 i++;
349 if (l->zeros == esc_count) {
350 int index = lag_get_rac(rac);
351 ret++;
352
353 l->zeros = 0;
354
355 l->zeros_rem = lag_calc_zero_run(index);
356 goto handle_zeros;
357 }
358 }
359 return ret;
360 }
361
362 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
363 const uint8_t *src, const uint8_t *src_end,
364 int width, int esc_count)
365 {
366 int i = 0;
367 int count;
368 uint8_t zero_run = 0;
369 const uint8_t *src_start = src;
370 uint8_t mask1 = -(esc_count < 2);
371 uint8_t mask2 = -(esc_count < 3);
372 uint8_t *end = dst + (width - 2);
373
374 output_zeros:
375 if (l->zeros_rem) {
376 count = FFMIN(l->zeros_rem, width - i);
377 if (end - dst < count) {
378 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
379 return AVERROR_INVALIDDATA;
380 }
381
382 memset(dst, 0, count);
383 l->zeros_rem -= count;
384 dst += count;
385 }
386
387 while (dst < end) {
388 i = 0;
389 while (!zero_run && dst + i < end) {
390 i++;
391 if (src + i >= src_end)
392 return AVERROR_INVALIDDATA;
393 zero_run =
394 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
395 }
396 if (zero_run) {
397 zero_run = 0;
398 i += esc_count;
399 memcpy(dst, src, i);
400 dst += i;
401 l->zeros_rem = lag_calc_zero_run(src[i]);
402
403 src += i + 1;
404 goto output_zeros;
405 } else {
406 memcpy(dst, src, i);
407 src += i;
408 dst += i;
409 }
410 }
411 return src_start - src;
412 }
413
414
415
416 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
417 int width, int height, int stride,
418 const uint8_t *src, int src_size)
419 {
420 int i = 0;
421 int read = 0;
422 uint32_t length;
423 uint32_t offset = 1;
424 int esc_count = src[0];
425 GetBitContext gb;
426 lag_rac rac;
427 const uint8_t *src_end = src + src_size;
428
429 rac.avctx = l->avctx;
430 l->zeros = 0;
431
432 if (esc_count < 4) {
433 length = width * height;
434 if (esc_count && AV_RL32(src + 1) < length) {
435 length = AV_RL32(src + 1);
436 offset += 4;
437 }
438
439 init_get_bits(&gb, src + offset, src_size * 8);
440
441 if (lag_read_prob_header(&rac, &gb) < 0)
442 return -1;
443
444 ff_lag_rac_init(&rac, &gb, length - stride);
445
446 for (i = 0; i < height; i++)
447 read += lag_decode_line(l, &rac, dst + (i * stride), width,
448 stride, esc_count);
449
450 if (read > length)
451 av_log(l->avctx, AV_LOG_WARNING,
452 "Output more bytes than length (%d of %d)\n", read,
453 length);
454 } else if (esc_count < 8) {
455 esc_count -= 4;
456 if (esc_count > 0) {
457 /* Zero run coding only, no range coding. */
458 for (i = 0; i < height; i++) {
459 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
460 src_end, width, esc_count);
461 if (res < 0)
462 return res;
463 src += res;
464 }
465 } else {
466 if (src_size < width * height)
467 return AVERROR_INVALIDDATA; // buffer not big enough
468 /* Plane is stored uncompressed */
469 for (i = 0; i < height; i++) {
470 memcpy(dst + (i * stride), src, width);
471 src += width;
472 }
473 }
474 } else if (esc_count == 0xff) {
475 /* Plane is a solid run of given value */
476 for (i = 0; i < height; i++)
477 memset(dst + i * stride, src[1], width);
478 /* Do not apply prediction.
479 Note: memset to 0 above, setting first value to src[1]
480 and applying prediction gives the same result. */
481 return 0;
482 } else {
483 av_log(l->avctx, AV_LOG_ERROR,
484 "Invalid zero run escape code! (%#x)\n", esc_count);
485 return -1;
486 }
487
488 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
489 for (i = 0; i < height; i++) {
490 lag_pred_line(l, dst, width, stride, i);
491 dst += stride;
492 }
493 } else {
494 for (i = 0; i < height; i++) {
495 lag_pred_line_yuy2(l, dst, width, stride, i,
496 width == l->avctx->width);
497 dst += stride;
498 }
499 }
500
501 return 0;
502 }
503
504 /**
505 * Decode a frame.
506 * @param avctx codec context
507 * @param data output AVFrame
508 * @param data_size size of output data or 0 if no picture is returned
509 * @param avpkt input packet
510 * @return number of consumed bytes on success or negative if decode fails
511 */
512 static int lag_decode_frame(AVCodecContext *avctx,
513 void *data, int *got_frame, AVPacket *avpkt)
514 {
515 const uint8_t *buf = avpkt->data;
516 int buf_size = avpkt->size;
517 LagarithContext *l = avctx->priv_data;
518 ThreadFrame frame = { .f = data };
519 AVFrame *const p = data;
520 uint8_t frametype = 0;
521 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
522 uint32_t offs[4];
523 uint8_t *srcs[4], *dst;
524 int i, j, planes = 3;
525
526 p->key_frame = 1;
527
528 frametype = buf[0];
529
530 offset_gu = AV_RL32(buf + 1);
531 offset_bv = AV_RL32(buf + 5);
532
533 switch (frametype) {
534 case FRAME_SOLID_RGBA:
535 avctx->pix_fmt = AV_PIX_FMT_RGB32;
536
537 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
538 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
539 return -1;
540 }
541
542 dst = p->data[0];
543 for (j = 0; j < avctx->height; j++) {
544 for (i = 0; i < avctx->width; i++)
545 AV_WN32(dst + i * 4, offset_gu);
546 dst += p->linesize[0];
547 }
548 break;
549 case FRAME_ARITH_RGBA:
550 avctx->pix_fmt = AV_PIX_FMT_RGB32;
551 planes = 4;
552 offset_ry += 4;
553 offs[3] = AV_RL32(buf + 9);
554 case FRAME_ARITH_RGB24:
555 case FRAME_U_RGB24:
556 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
557 avctx->pix_fmt = AV_PIX_FMT_RGB24;
558
559 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
560 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
561 return -1;
562 }
563
564 offs[0] = offset_bv;
565 offs[1] = offset_gu;
566 offs[2] = offset_ry;
567
568 l->rgb_stride = FFALIGN(avctx->width, 16);
569 av_fast_malloc(&l->rgb_planes, &l->rgb_planes_allocated,
570 l->rgb_stride * avctx->height * planes + 1);
571 if (!l->rgb_planes) {
572 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
573 return AVERROR(ENOMEM);
574 }
575 for (i = 0; i < planes; i++)
576 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
577 if (offset_ry >= buf_size ||
578 offset_gu >= buf_size ||
579 offset_bv >= buf_size ||
580 (planes == 4 && offs[3] >= buf_size)) {
581 av_log(avctx, AV_LOG_ERROR,
582 "Invalid frame offsets\n");
583 return AVERROR_INVALIDDATA;
584 }
585 for (i = 0; i < planes; i++)
586 lag_decode_arith_plane(l, srcs[i],
587 avctx->width, avctx->height,
588 -l->rgb_stride, buf + offs[i],
589 buf_size - offs[i]);
590 dst = p->data[0];
591 for (i = 0; i < planes; i++)
592 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
593 for (j = 0; j < avctx->height; j++) {
594 for (i = 0; i < avctx->width; i++) {
595 uint8_t r, g, b, a;
596 r = srcs[0][i];
597 g = srcs[1][i];
598 b = srcs[2][i];
599 r += g;
600 b += g;
601 if (frametype == FRAME_ARITH_RGBA) {
602 a = srcs[3][i];
603 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
604 } else {
605 dst[i * 3 + 0] = r;
606 dst[i * 3 + 1] = g;
607 dst[i * 3 + 2] = b;
608 }
609 }
610 dst += p->linesize[0];
611 for (i = 0; i < planes; i++)
612 srcs[i] += l->rgb_stride;
613 }
614 break;
615 case FRAME_ARITH_YUY2:
616 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
617
618 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
619 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
620 return -1;
621 }
622
623 if (offset_ry >= buf_size ||
624 offset_gu >= buf_size ||
625 offset_bv >= buf_size) {
626 av_log(avctx, AV_LOG_ERROR,
627 "Invalid frame offsets\n");
628 return AVERROR_INVALIDDATA;
629 }
630
631 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
632 p->linesize[0], buf + offset_ry,
633 buf_size - offset_ry);
634 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
635 avctx->height, p->linesize[1],
636 buf + offset_gu, buf_size - offset_gu);
637 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
638 avctx->height, p->linesize[2],
639 buf + offset_bv, buf_size - offset_bv);
640 break;
641 case FRAME_ARITH_YV12:
642 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
643
644 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
645 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
646 return -1;
647 }
648
649 if (offset_ry >= buf_size ||
650 offset_gu >= buf_size ||
651 offset_bv >= buf_size) {
652 av_log(avctx, AV_LOG_ERROR,
653 "Invalid frame offsets\n");
654 return AVERROR_INVALIDDATA;
655 }
656
657 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
658 p->linesize[0], buf + offset_ry,
659 buf_size - offset_ry);
660 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
661 avctx->height / 2, p->linesize[2],
662 buf + offset_gu, buf_size - offset_gu);
663 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
664 avctx->height / 2, p->linesize[1],
665 buf + offset_bv, buf_size - offset_bv);
666 break;
667 default:
668 av_log(avctx, AV_LOG_ERROR,
669 "Unsupported Lagarith frame type: %#x\n", frametype);
670 return -1;
671 }
672
673 *got_frame = 1;
674
675 return buf_size;
676 }
677
678 static av_cold int lag_decode_init(AVCodecContext *avctx)
679 {
680 LagarithContext *l = avctx->priv_data;
681 l->avctx = avctx;
682
683 ff_dsputil_init(&l->dsp, avctx);
684
685 return 0;
686 }
687
688 static av_cold int lag_decode_end(AVCodecContext *avctx)
689 {
690 LagarithContext *l = avctx->priv_data;
691
692 av_freep(&l->rgb_planes);
693
694 return 0;
695 }
696
697 AVCodec ff_lagarith_decoder = {
698 .name = "lagarith",
699 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
700 .type = AVMEDIA_TYPE_VIDEO,
701 .id = AV_CODEC_ID_LAGARITH,
702 .priv_data_size = sizeof(LagarithContext),
703 .init = lag_decode_init,
704 .close = lag_decode_end,
705 .decode = lag_decode_frame,
706 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
707 };