1bf61bd23d72c9e58fb9e8a8b9363d28d6f4fe34
[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 FFmpeg.
6 *
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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 libavcodec/lagarith.c
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 } LagarithContext;
55
56 /**
57 * Compute the 52bit mantissa of 1/(double)denom.
58 * This crazy format uses floats in an entropy coder and we have to match x86
59 * rounding exactly, thus ordinary floats aren't portable enough.
60 * @param denom denominator
61 * @return 52bit mantissa
62 * @see softfloat_mul
63 */
64 static uint64_t softfloat_reciprocal(uint32_t denom)
65 {
66 int shift = av_log2(denom - 1) + 1;
67 uint64_t ret = (1ULL << 52) / denom;
68 uint64_t err = (1ULL << 52) - ret * denom;
69 ret <<= shift;
70 err <<= shift;
71 err += denom / 2;
72 return ret + err / denom;
73 }
74
75 /**
76 * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
77 * Used in combination with softfloat_reciprocal computes x/(double)denom.
78 * @param x 32bit integer factor
79 * @param mantissa mantissa of f with exponent 0
80 * @return 32bit integer value (x*f)
81 * @see softfloat_reciprocal
82 */
83 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
84 {
85 uint64_t l = x * (mantissa & 0xffffffff);
86 uint64_t h = x * (mantissa >> 32);
87 h += l >> 32;
88 l &= 0xffffffff;
89 l += 1 << av_log2(h >> 21);
90 h += l >> 32;
91 return h >> 20;
92 }
93
94 static uint8_t lag_calc_zero_run(int8_t x)
95 {
96 return (x << 1) ^ (x >> 7);
97 }
98
99 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
100 {
101 static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
102 int i;
103 int bit = 0;
104 int bits = 0;
105 int prevbit = 0;
106 unsigned val;
107
108 for (i = 0; i < 7; i++) {
109 if (prevbit && bit)
110 break;
111 prevbit = bit;
112 bit = get_bits1(gb);
113 if (bit && !prevbit)
114 bits += series[i];
115 }
116 bits--;
117 if (bits < 0 || bits > 31) {
118 *value = 0;
119 return -1;
120 } else if (bits == 0) {
121 *value = 0;
122 return 0;
123 }
124
125 val = get_bits_long(gb, bits);
126 val |= 1 << bits;
127
128 *value = val - 1;
129
130 return 0;
131 }
132
133 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
134 {
135 int i, j, scale_factor;
136 unsigned prob, cumulative_target;
137 unsigned cumul_prob = 0;
138 unsigned scaled_cumul_prob = 0;
139
140 rac->prob[0] = 0;
141 rac->prob[257] = UINT_MAX;
142 /* Read probabilities from bitstream */
143 for (i = 1; i < 257; i++) {
144 if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
145 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
146 return -1;
147 }
148 if ((uint64_t)cumul_prob + rac->prob[i] > UINT_MAX) {
149 av_log(rac->avctx, AV_LOG_ERROR, "Integer overflow encountered in cumulative probability calculation.\n");
150 return -1;
151 }
152 cumul_prob += rac->prob[i];
153 if (!rac->prob[i]) {
154 if (lag_decode_prob(gb, &prob)) {
155 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
156 return -1;
157 }
158 if (prob > 257 - i)
159 prob = 257 - i;
160 for (j = 0; j < prob; j++)
161 rac->prob[++i] = 0;
162 }
163 }
164
165 if (!cumul_prob) {
166 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
167 return -1;
168 }
169
170 /* Scale probabilities so cumulative probability is an even power of 2. */
171 scale_factor = av_log2(cumul_prob);
172
173 if (cumul_prob & (cumul_prob - 1)) {
174 uint64_t mul = softfloat_reciprocal(cumul_prob);
175 for (i = 1; i < 257; i++) {
176 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
177 scaled_cumul_prob += rac->prob[i];
178 }
179
180 scale_factor++;
181 cumulative_target = 1 << scale_factor;
182
183 if (scaled_cumul_prob > cumulative_target) {
184 av_log(rac->avctx, AV_LOG_ERROR,
185 "Scaled probabilities are larger than target!\n");
186 return -1;
187 }
188
189 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
190
191 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
192 if (rac->prob[i]) {
193 rac->prob[i]++;
194 scaled_cumul_prob--;
195 }
196 /* Comment from reference source:
197 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
198 * // since the compression change is negligable and fixing it
199 * // breaks backwards compatibilty
200 * b =- (signed int)b;
201 * b &= 0xFF;
202 * } else {
203 * b++;
204 * b &= 0x7f;
205 * }
206 */
207 }
208 }
209
210 rac->scale = scale_factor;
211
212 /* Fill probability array with cumulative probability for each symbol. */
213 for (i = 1; i < 257; i++)
214 rac->prob[i] += rac->prob[i - 1];
215
216 return 0;
217 }
218
219 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
220 uint8_t *diff, int w, int *left,
221 int *left_top)
222 {
223 /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
224 * However the &0xFF on the gradient predictor yealds incorrect output
225 * for lagarith.
226 */
227 int i;
228 uint8_t l, lt;
229
230 l = *left;
231 lt = *left_top;
232
233 for (i = 0; i < w; i++) {
234 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
235 lt = src1[i];
236 dst[i] = l;
237 }
238
239 *left = l;
240 *left_top = lt;
241 }
242
243 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
244 int width, int stride, int line)
245 {
246 int L, TL;
247
248 if (!line) {
249 /* Left prediction only for first line */
250 L = l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1,
251 width - 1, buf[0]);
252 return;
253 } else if (line == 1) {
254 /* Second line, left predict first pixel, the rest of the line is median predicted */
255 /* FIXME: In the case of RGB this pixel is top predicted */
256 TL = buf[-stride];
257 } else {
258 /* Top left is 2 rows back, last pixel */
259 TL = buf[width - (2 * stride) - 1];
260 }
261 /* Left pixel is actually prev_row[width] */
262 L = buf[width - stride - 1];
263
264 add_lag_median_prediction(buf, buf - stride, buf,
265 width, &L, &TL);
266 }
267
268 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
269 uint8_t *dst, int width, int stride,
270 int esc_count)
271 {
272 int i = 0;
273 int ret = 0;
274
275 if (!esc_count)
276 esc_count = -1;
277
278 /* Output any zeros remaining from the previous run */
279 handle_zeros:
280 if (l->zeros_rem) {
281 int count = FFMIN(l->zeros_rem, width - i);
282 memset(dst + i, 0, count);
283 i += count;
284 l->zeros_rem -= count;
285 }
286
287 while (i < width) {
288 dst[i] = lag_get_rac(rac);
289 ret++;
290
291 if (dst[i])
292 l->zeros = 0;
293 else
294 l->zeros++;
295
296 i++;
297 if (l->zeros == esc_count) {
298 int index = lag_get_rac(rac);
299 ret++;
300
301 l->zeros = 0;
302
303 l->zeros_rem = lag_calc_zero_run(index);
304 goto handle_zeros;
305 }
306 }
307 return ret;
308 }
309
310 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
311 const uint8_t *src, int width,
312 int esc_count)
313 {
314 int i = 0;
315 int count;
316 uint8_t zero_run = 0;
317 const uint8_t *start = src;
318 uint8_t mask1 = -(esc_count < 2);
319 uint8_t mask2 = -(esc_count < 3);
320 uint8_t *end = dst + (width - 2);
321
322 output_zeros:
323 if (l->zeros_rem) {
324 count = FFMIN(l->zeros_rem, width - i);
325 memset(dst, 0, count);
326 l->zeros_rem -= count;
327 dst += count;
328 }
329
330 while (dst < end) {
331 i = 0;
332 while (!zero_run && dst + i < end) {
333 i++;
334 zero_run =
335 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
336 }
337 if (zero_run) {
338 zero_run = 0;
339 i += esc_count;
340 memcpy(dst, src, i);
341 dst += i;
342 l->zeros_rem = lag_calc_zero_run(src[i]);
343
344 src += i + 1;
345 goto output_zeros;
346 } else {
347 memcpy(dst, src, i);
348 src += i;
349 }
350 }
351 return start - src;
352 }
353
354
355
356 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
357 int width, int height, int stride,
358 const uint8_t *src, int src_size)
359 {
360 int i = 0;
361 int read = 0;
362 uint32_t length;
363 uint32_t offset = 1;
364 int esc_count = src[0];
365 GetBitContext gb;
366 lag_rac rac;
367
368 rac.avctx = l->avctx;
369 l->zeros = 0;
370
371 if (esc_count < 4) {
372 length = width * height;
373 if (esc_count && AV_RL32(src + 1) < length) {
374 length = AV_RL32(src + 1);
375 offset += 4;
376 }
377
378 init_get_bits(&gb, src + offset, src_size * 8);
379
380 if (lag_read_prob_header(&rac, &gb) < 0)
381 return -1;
382
383 lag_rac_init(&rac, &gb, length - stride);
384
385 for (i = 0; i < height; i++)
386 read += lag_decode_line(l, &rac, dst + (i * stride), width,
387 stride, esc_count);
388
389 if (read > length)
390 av_log(l->avctx, AV_LOG_WARNING,
391 "Output more bytes than length (%d of %d)\n", read,
392 length);
393 } else if (esc_count < 8) {
394 esc_count -= 4;
395 if (esc_count > 0) {
396 /* Zero run coding only, no range coding. */
397 for (i = 0; i < height; i++)
398 src += lag_decode_zero_run_line(l, dst + (i * stride), src,
399 width, esc_count);
400 } else {
401 /* Plane is stored uncompressed */
402 for (i = 0; i < height; i++) {
403 memcpy(dst + (i * stride), src, width);
404 src += width;
405 }
406 }
407 } else if (esc_count == 0xff) {
408 /* Plane is a solid run of given value */
409 for (i = 0; i < height; i++)
410 memset(dst + i * stride, src[1], width);
411 /* Do not apply prediction.
412 Note: memset to 0 above, setting first value to src[1]
413 and applying prediction gives the same result. */
414 return 0;
415 } else {
416 av_log(l->avctx, AV_LOG_ERROR,
417 "Invalid zero run escape code! (%#x)\n", esc_count);
418 return -1;
419 }
420
421 for (i = 0; i < height; i++) {
422 lag_pred_line(l, dst, width, stride, i);
423 dst += stride;
424 }
425
426 return 0;
427 }
428
429 /**
430 * Decode a frame.
431 * @param avctx codec context
432 * @param data output AVFrame
433 * @param data_size size of output data or 0 if no picture is returned
434 * @param avpkt input packet
435 * @return number of consumed bytes on success or negative if decode fails
436 */
437 static int lag_decode_frame(AVCodecContext *avctx,
438 void *data, int *data_size, AVPacket *avpkt)
439 {
440 const uint8_t *buf = avpkt->data;
441 int buf_size = avpkt->size;
442 LagarithContext *l = avctx->priv_data;
443 AVFrame *const p = &l->picture;
444 uint8_t frametype = 0;
445 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
446
447 AVFrame *picture = data;
448
449 if (p->data[0])
450 avctx->release_buffer(avctx, p);
451
452 p->reference = 0;
453 p->key_frame = 1;
454
455 frametype = buf[0];
456
457 offset_gu = AV_RL32(buf + 1);
458 offset_bv = AV_RL32(buf + 5);
459
460 switch (frametype) {
461 case FRAME_ARITH_YV12:
462 avctx->pix_fmt = PIX_FMT_YUV420P;
463
464 if (avctx->get_buffer(avctx, p) < 0) {
465 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
466 return -1;
467 }
468
469 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
470 p->linesize[0], buf + offset_ry,
471 buf_size);
472 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
473 avctx->height / 2, p->linesize[2],
474 buf + offset_gu, buf_size);
475 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
476 avctx->height / 2, p->linesize[1],
477 buf + offset_bv, buf_size);
478 break;
479 default:
480 av_log(avctx, AV_LOG_ERROR,
481 "Unsupported Lagarith frame type: %#x\n", frametype);
482 return -1;
483 }
484
485 *picture = *p;
486 *data_size = sizeof(AVFrame);
487
488 return buf_size;
489 }
490
491 static av_cold int lag_decode_init(AVCodecContext *avctx)
492 {
493 LagarithContext *l = avctx->priv_data;
494 l->avctx = avctx;
495
496 dsputil_init(&l->dsp, avctx);
497
498 return 0;
499 }
500
501 static av_cold int lag_decode_end(AVCodecContext *avctx)
502 {
503 LagarithContext *l = avctx->priv_data;
504
505 if (l->picture.data[0])
506 avctx->release_buffer(avctx, &l->picture);
507
508 return 0;
509 }
510
511 AVCodec lagarith_decoder = {
512 "lagarith",
513 CODEC_TYPE_VIDEO,
514 CODEC_ID_LAGARITH,
515 sizeof(LagarithContext),
516 lag_decode_init,
517 NULL,
518 lag_decode_end,
519 lag_decode_frame,
520 CODEC_CAP_DR1,
521 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
522 };