2 * Lagarith lossless decoder
3 * Copyright (c) 2009 Nathan Caldwell <saintdev (at) gmail.com>
5 * This file is part of FFmpeg.
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.
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.
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
24 * Lagarith lossless decoder
25 * @author Nathan Caldwell
32 #include "lagarithrac.h"
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 */
49 typedef struct LagarithContext {
50 AVCodecContext *avctx;
52 int zeros; /**< number of consecutive zero bytes encountered */
53 int zeros_rem; /**< number of zero bytes remaining to output */
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
66 static uint64_t softfloat_reciprocal(uint32_t denom)
68 int shift = av_log2(denom - 1) + 1;
69 uint64_t ret = (1ULL << 52) / denom;
70 uint64_t err = (1ULL << 52) - ret * denom;
74 return ret + err / denom;
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
85 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
87 uint64_t l = x * (mantissa & 0xffffffff);
88 uint64_t h = x * (mantissa >> 32);
91 l += 1 << av_log2(h >> 21);
96 static uint8_t lag_calc_zero_run(int8_t x)
98 return (x << 1) ^ (x >> 7);
101 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
103 static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
110 for (i = 0; i < 7; i++) {
119 if (bits < 0 || bits > 31) {
122 } else if (bits == 0) {
127 val = get_bits_long(gb, bits);
135 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
137 int i, j, scale_factor;
138 unsigned prob, cumulative_target;
139 unsigned cumul_prob = 0;
140 unsigned scaled_cumul_prob = 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");
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");
154 cumul_prob += rac->prob[i];
156 if (lag_decode_prob(gb, &prob)) {
157 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
162 for (j = 0; j < prob; j++)
168 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
172 /* Scale probabilities so cumulative probability is an even power of 2. */
173 scale_factor = av_log2(cumul_prob);
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];
183 cumulative_target = 1 << scale_factor;
185 if (scaled_cumul_prob > cumulative_target) {
186 av_log(rac->avctx, AV_LOG_ERROR,
187 "Scaled probabilities are larger than target!\n");
191 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
193 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
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 negligible and fixing it
201 * // breaks backwards compatibility
202 * b =- (signed int)b;
212 rac->scale = scale_factor;
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];
221 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
222 uint8_t *diff, int w, int *left,
225 /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
226 * However the &0xFF on the gradient predictor yealds incorrect output
235 for (i = 0; i < w; i++) {
236 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
245 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
246 int width, int stride, int line)
251 /* Left prediction only for first line */
252 L = l->dsp.add_hfyu_left_prediction(buf, buf,
255 /* Left pixel is actually prev_row[width] */
256 L = buf[width - stride - 1];
259 /* Second line, left predict first pixel, the rest of the line is median predicted
260 * NOTE: In the case of RGB this pixel is top predicted */
261 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
263 /* Top left is 2 rows back, last pixel */
264 TL = buf[width - (2 * stride) - 1];
267 add_lag_median_prediction(buf, buf - stride, buf,
272 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
273 int width, int stride, int line,
282 l->dsp.add_hfyu_left_prediction(buf, buf, width, 0);
288 const int HEAD = is_luma ? 4 : 2;
291 L = buf[width - stride - 1];
292 TL = buf[HEAD - stride - 1];
293 for (i = 0; i < HEAD; i++) {
297 for (; i<width; i++) {
298 L = mid_pred(L&0xFF, buf[i-stride], (L + buf[i-stride] - TL)&0xFF) + buf[i];
303 TL = buf[width - (2 * stride) - 1];
304 L = buf[width - stride - 1];
305 l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
310 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
311 uint8_t *dst, int width, int stride,
320 /* Output any zeros remaining from the previous run */
323 int count = FFMIN(l->zeros_rem, width - i);
324 memset(dst + i, 0, count);
326 l->zeros_rem -= count;
330 dst[i] = lag_get_rac(rac);
339 if (l->zeros == esc_count) {
340 int index = lag_get_rac(rac);
345 l->zeros_rem = lag_calc_zero_run(index);
352 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
353 const uint8_t *src, const uint8_t *src_end,
354 int width, int esc_count)
358 uint8_t zero_run = 0;
359 const uint8_t *src_start = src;
360 uint8_t mask1 = -(esc_count < 2);
361 uint8_t mask2 = -(esc_count < 3);
362 uint8_t *end = dst + (width - 2);
366 count = FFMIN(l->zeros_rem, width - i);
367 if (end - dst < count) {
368 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
369 return AVERROR_INVALIDDATA;
372 memset(dst, 0, count);
373 l->zeros_rem -= count;
379 while (!zero_run && dst + i < end) {
381 if (i+2 >= src_end - src)
382 return AVERROR_INVALIDDATA;
384 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
391 l->zeros_rem = lag_calc_zero_run(src[i]);
401 return src - src_start;
406 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
407 int width, int height, int stride,
408 const uint8_t *src, int src_size)
417 const uint8_t *src_end = src + src_size;
419 rac.avctx = l->avctx;
423 return AVERROR_INVALIDDATA;
427 length = width * height;
429 return AVERROR_INVALIDDATA;
430 if (esc_count && AV_RL32(src + 1) < length) {
431 length = AV_RL32(src + 1);
435 init_get_bits(&gb, src + offset, src_size * 8);
437 if (lag_read_prob_header(&rac, &gb) < 0)
440 ff_lag_rac_init(&rac, &gb, length - stride);
442 for (i = 0; i < height; i++)
443 read += lag_decode_line(l, &rac, dst + (i * stride), width,
447 av_log(l->avctx, AV_LOG_WARNING,
448 "Output more bytes than length (%d of %d)\n", read,
450 } else if (esc_count < 8) {
453 /* Zero run coding only, no range coding. */
454 for (i = 0; i < height; i++) {
455 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
456 src_end, width, esc_count);
462 if (src_size < width * height)
463 return AVERROR_INVALIDDATA; // buffer not big enough
464 /* Plane is stored uncompressed */
465 for (i = 0; i < height; i++) {
466 memcpy(dst + (i * stride), src, width);
470 } else if (esc_count == 0xff) {
471 /* Plane is a solid run of given value */
472 for (i = 0; i < height; i++)
473 memset(dst + i * stride, src[1], width);
474 /* Do not apply prediction.
475 Note: memset to 0 above, setting first value to src[1]
476 and applying prediction gives the same result. */
479 av_log(l->avctx, AV_LOG_ERROR,
480 "Invalid zero run escape code! (%#x)\n", esc_count);
484 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
485 for (i = 0; i < height; i++) {
486 lag_pred_line(l, dst, width, stride, i);
490 for (i = 0; i < height; i++) {
491 lag_pred_line_yuy2(l, dst, width, stride, i,
492 width == l->avctx->width);
502 * @param avctx codec context
503 * @param data output AVFrame
504 * @param data_size size of output data or 0 if no picture is returned
505 * @param avpkt input packet
506 * @return number of consumed bytes on success or negative if decode fails
508 static int lag_decode_frame(AVCodecContext *avctx,
509 void *data, int *got_frame, AVPacket *avpkt)
511 const uint8_t *buf = avpkt->data;
512 unsigned int buf_size = avpkt->size;
513 LagarithContext *l = avctx->priv_data;
514 ThreadFrame frame = { .f = data };
515 AVFrame *const p = data;
516 uint8_t frametype = 0;
517 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
519 uint8_t *srcs[4], *dst;
520 int i, j, planes = 3;
527 offset_gu = AV_RL32(buf + 1);
528 offset_bv = AV_RL32(buf + 5);
531 case FRAME_SOLID_RGBA:
532 avctx->pix_fmt = AV_PIX_FMT_RGB32;
533 case FRAME_SOLID_GRAY:
534 if (frametype == FRAME_SOLID_GRAY)
535 if (avctx->bits_per_coded_sample == 24) {
536 avctx->pix_fmt = AV_PIX_FMT_RGB24;
538 avctx->pix_fmt = AV_PIX_FMT_0RGB32;
542 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
546 if (frametype == FRAME_SOLID_RGBA) {
547 for (j = 0; j < avctx->height; j++) {
548 for (i = 0; i < avctx->width; i++)
549 AV_WN32(dst + i * 4, offset_gu);
550 dst += p->linesize[0];
553 for (j = 0; j < avctx->height; j++) {
554 memset(dst, buf[1], avctx->width * planes);
555 dst += p->linesize[0];
559 case FRAME_ARITH_RGBA:
560 avctx->pix_fmt = AV_PIX_FMT_RGB32;
563 offs[3] = AV_RL32(buf + 9);
564 case FRAME_ARITH_RGB24:
566 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
567 avctx->pix_fmt = AV_PIX_FMT_RGB24;
569 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
576 if (!l->rgb_planes) {
577 l->rgb_stride = FFALIGN(avctx->width, 16);
578 l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * 4 + 16);
579 if (!l->rgb_planes) {
580 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
581 return AVERROR(ENOMEM);
584 for (i = 0; i < planes; i++)
585 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
586 for (i = 0; i < planes; i++)
587 if (buf_size <= offs[i]) {
588 av_log(avctx, AV_LOG_ERROR,
589 "Invalid frame offsets\n");
590 return AVERROR_INVALIDDATA;
593 for (i = 0; i < planes; i++)
594 lag_decode_arith_plane(l, srcs[i],
595 avctx->width, avctx->height,
596 -l->rgb_stride, buf + offs[i],
599 for (i = 0; i < planes; i++)
600 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
601 for (j = 0; j < avctx->height; j++) {
602 for (i = 0; i < avctx->width; i++) {
609 if (frametype == FRAME_ARITH_RGBA) {
611 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
618 dst += p->linesize[0];
619 for (i = 0; i < planes; i++)
620 srcs[i] += l->rgb_stride;
623 case FRAME_ARITH_YUY2:
624 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
626 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
629 if (offset_ry >= buf_size ||
630 offset_gu >= buf_size ||
631 offset_bv >= buf_size) {
632 av_log(avctx, AV_LOG_ERROR,
633 "Invalid frame offsets\n");
634 return AVERROR_INVALIDDATA;
637 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
638 p->linesize[0], buf + offset_ry,
639 buf_size - offset_ry);
640 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
641 avctx->height, p->linesize[1],
642 buf + offset_gu, buf_size - offset_gu);
643 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
644 avctx->height, p->linesize[2],
645 buf + offset_bv, buf_size - offset_bv);
647 case FRAME_ARITH_YV12:
648 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
650 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
652 if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
653 return AVERROR_INVALIDDATA;
656 if (offset_ry >= buf_size ||
657 offset_gu >= buf_size ||
658 offset_bv >= buf_size) {
659 av_log(avctx, AV_LOG_ERROR,
660 "Invalid frame offsets\n");
661 return AVERROR_INVALIDDATA;
664 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
665 p->linesize[0], buf + offset_ry,
666 buf_size - offset_ry);
667 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
668 avctx->height / 2, p->linesize[2],
669 buf + offset_gu, buf_size - offset_gu);
670 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
671 avctx->height / 2, p->linesize[1],
672 buf + offset_bv, buf_size - offset_bv);
675 av_log(avctx, AV_LOG_ERROR,
676 "Unsupported Lagarith frame type: %#x\n", frametype);
677 return AVERROR_PATCHWELCOME;
685 static av_cold int lag_decode_init(AVCodecContext *avctx)
687 LagarithContext *l = avctx->priv_data;
690 ff_dsputil_init(&l->dsp, avctx);
695 static av_cold int lag_decode_end(AVCodecContext *avctx)
697 LagarithContext *l = avctx->priv_data;
699 av_freep(&l->rgb_planes);
704 AVCodec ff_lagarith_decoder = {
706 .type = AVMEDIA_TYPE_VIDEO,
707 .id = AV_CODEC_ID_LAGARITH,
708 .priv_data_size = sizeof(LagarithContext),
709 .init = lag_decode_init,
710 .close = lag_decode_end,
711 .decode = lag_decode_frame,
712 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
713 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),