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 <= 128; i++) {
178 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
179 scaled_cumul_prob += rac->prob[i];
181 if (scaled_cumul_prob <= 0) {
182 av_log(rac->avctx, AV_LOG_ERROR, "Scaled probabilities invalid\n");
183 return AVERROR_INVALIDDATA;
185 for (; i < 257; i++) {
186 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
187 scaled_cumul_prob += rac->prob[i];
191 cumulative_target = 1 << scale_factor;
193 if (scaled_cumul_prob > cumulative_target) {
194 av_log(rac->avctx, AV_LOG_ERROR,
195 "Scaled probabilities are larger than target!\n");
199 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
201 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
206 /* Comment from reference source:
207 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
208 * // since the compression change is negligible and fixing it
209 * // breaks backwards compatibility
210 * b =- (signed int)b;
220 rac->scale = scale_factor;
222 /* Fill probability array with cumulative probability for each symbol. */
223 for (i = 1; i < 257; i++)
224 rac->prob[i] += rac->prob[i - 1];
229 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
230 uint8_t *diff, int w, int *left,
233 /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
234 * However the &0xFF on the gradient predictor yealds incorrect output
243 for (i = 0; i < w; i++) {
244 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
253 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
254 int width, int stride, int line)
259 /* Left prediction only for first line */
260 L = l->dsp.add_hfyu_left_prediction(buf, buf,
263 /* Left pixel is actually prev_row[width] */
264 L = buf[width - stride - 1];
267 /* Second line, left predict first pixel, the rest of the line is median predicted
268 * NOTE: In the case of RGB this pixel is top predicted */
269 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
271 /* Top left is 2 rows back, last pixel */
272 TL = buf[width - (2 * stride) - 1];
275 add_lag_median_prediction(buf, buf - stride, buf,
280 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
281 int width, int stride, int line,
290 l->dsp.add_hfyu_left_prediction(buf, buf, width, 0);
296 const int HEAD = is_luma ? 4 : 2;
299 L = buf[width - stride - 1];
300 TL = buf[HEAD - stride - 1];
301 for (i = 0; i < HEAD; i++) {
305 for (; i<width; i++) {
306 L = mid_pred(L&0xFF, buf[i-stride], (L + buf[i-stride] - TL)&0xFF) + buf[i];
311 TL = buf[width - (2 * stride) - 1];
312 L = buf[width - stride - 1];
313 l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
318 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
319 uint8_t *dst, int width, int stride,
328 /* Output any zeros remaining from the previous run */
331 int count = FFMIN(l->zeros_rem, width - i);
332 memset(dst + i, 0, count);
334 l->zeros_rem -= count;
338 dst[i] = lag_get_rac(rac);
347 if (l->zeros == esc_count) {
348 int index = lag_get_rac(rac);
353 l->zeros_rem = lag_calc_zero_run(index);
360 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
361 const uint8_t *src, const uint8_t *src_end,
362 int width, int esc_count)
366 uint8_t zero_run = 0;
367 const uint8_t *src_start = src;
368 uint8_t mask1 = -(esc_count < 2);
369 uint8_t mask2 = -(esc_count < 3);
370 uint8_t *end = dst + (width - 2);
374 count = FFMIN(l->zeros_rem, width - i);
375 if (end - dst < count) {
376 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
377 return AVERROR_INVALIDDATA;
380 memset(dst, 0, count);
381 l->zeros_rem -= count;
387 while (!zero_run && dst + i < end) {
389 if (i+2 >= src_end - src)
390 return AVERROR_INVALIDDATA;
392 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
399 l->zeros_rem = lag_calc_zero_run(src[i]);
409 return src - src_start;
414 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
415 int width, int height, int stride,
416 const uint8_t *src, int src_size)
425 const uint8_t *src_end = src + src_size;
427 rac.avctx = l->avctx;
431 return AVERROR_INVALIDDATA;
435 length = width * height;
437 return AVERROR_INVALIDDATA;
438 if (esc_count && AV_RL32(src + 1) < length) {
439 length = AV_RL32(src + 1);
443 init_get_bits(&gb, src + offset, src_size * 8);
445 if (lag_read_prob_header(&rac, &gb) < 0)
448 ff_lag_rac_init(&rac, &gb, length - stride);
450 for (i = 0; i < height; i++)
451 read += lag_decode_line(l, &rac, dst + (i * stride), width,
455 av_log(l->avctx, AV_LOG_WARNING,
456 "Output more bytes than length (%d of %d)\n", read,
458 } else if (esc_count < 8) {
461 /* Zero run coding only, no range coding. */
462 for (i = 0; i < height; i++) {
463 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
464 src_end, width, esc_count);
470 if (src_size < width * height)
471 return AVERROR_INVALIDDATA; // buffer not big enough
472 /* Plane is stored uncompressed */
473 for (i = 0; i < height; i++) {
474 memcpy(dst + (i * stride), src, width);
478 } else if (esc_count == 0xff) {
479 /* Plane is a solid run of given value */
480 for (i = 0; i < height; i++)
481 memset(dst + i * stride, src[1], width);
482 /* Do not apply prediction.
483 Note: memset to 0 above, setting first value to src[1]
484 and applying prediction gives the same result. */
487 av_log(l->avctx, AV_LOG_ERROR,
488 "Invalid zero run escape code! (%#x)\n", esc_count);
492 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
493 for (i = 0; i < height; i++) {
494 lag_pred_line(l, dst, width, stride, i);
498 for (i = 0; i < height; i++) {
499 lag_pred_line_yuy2(l, dst, width, stride, i,
500 width == l->avctx->width);
510 * @param avctx codec context
511 * @param data output AVFrame
512 * @param data_size size of output data or 0 if no picture is returned
513 * @param avpkt input packet
514 * @return number of consumed bytes on success or negative if decode fails
516 static int lag_decode_frame(AVCodecContext *avctx,
517 void *data, int *got_frame, AVPacket *avpkt)
519 const uint8_t *buf = avpkt->data;
520 unsigned int buf_size = avpkt->size;
521 LagarithContext *l = avctx->priv_data;
522 ThreadFrame frame = { .f = data };
523 AVFrame *const p = data;
524 uint8_t frametype = 0;
525 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
527 uint8_t *srcs[4], *dst;
528 int i, j, planes = 3;
535 offset_gu = AV_RL32(buf + 1);
536 offset_bv = AV_RL32(buf + 5);
539 case FRAME_SOLID_RGBA:
540 avctx->pix_fmt = AV_PIX_FMT_RGB32;
541 case FRAME_SOLID_GRAY:
542 if (frametype == FRAME_SOLID_GRAY)
543 if (avctx->bits_per_coded_sample == 24) {
544 avctx->pix_fmt = AV_PIX_FMT_RGB24;
546 avctx->pix_fmt = AV_PIX_FMT_0RGB32;
550 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
554 if (frametype == FRAME_SOLID_RGBA) {
555 for (j = 0; j < avctx->height; j++) {
556 for (i = 0; i < avctx->width; i++)
557 AV_WN32(dst + i * 4, offset_gu);
558 dst += p->linesize[0];
561 for (j = 0; j < avctx->height; j++) {
562 memset(dst, buf[1], avctx->width * planes);
563 dst += p->linesize[0];
567 case FRAME_SOLID_COLOR:
568 if (avctx->bits_per_coded_sample == 24) {
569 avctx->pix_fmt = AV_PIX_FMT_RGB24;
571 avctx->pix_fmt = AV_PIX_FMT_RGB32;
572 offset_gu |= 0xFFU << 24;
575 if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
579 for (j = 0; j < avctx->height; j++) {
580 for (i = 0; i < avctx->width; i++)
581 if (avctx->bits_per_coded_sample == 24) {
582 AV_WB24(dst + i * 3, offset_gu);
584 AV_WN32(dst + i * 4, offset_gu);
586 dst += p->linesize[0];
589 case FRAME_ARITH_RGBA:
590 avctx->pix_fmt = AV_PIX_FMT_RGB32;
593 offs[3] = AV_RL32(buf + 9);
594 case FRAME_ARITH_RGB24:
596 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
597 avctx->pix_fmt = AV_PIX_FMT_RGB24;
599 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
606 if (!l->rgb_planes) {
607 l->rgb_stride = FFALIGN(avctx->width, 16);
608 l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * 4 + 16);
609 if (!l->rgb_planes) {
610 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
611 return AVERROR(ENOMEM);
614 for (i = 0; i < planes; i++)
615 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
616 for (i = 0; i < planes; i++)
617 if (buf_size <= offs[i]) {
618 av_log(avctx, AV_LOG_ERROR,
619 "Invalid frame offsets\n");
620 return AVERROR_INVALIDDATA;
623 for (i = 0; i < planes; i++)
624 lag_decode_arith_plane(l, srcs[i],
625 avctx->width, avctx->height,
626 -l->rgb_stride, buf + offs[i],
629 for (i = 0; i < planes; i++)
630 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
631 for (j = 0; j < avctx->height; j++) {
632 for (i = 0; i < avctx->width; i++) {
639 if (frametype == FRAME_ARITH_RGBA) {
641 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
648 dst += p->linesize[0];
649 for (i = 0; i < planes; i++)
650 srcs[i] += l->rgb_stride;
653 case FRAME_ARITH_YUY2:
654 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
656 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
659 if (offset_ry >= buf_size ||
660 offset_gu >= buf_size ||
661 offset_bv >= buf_size) {
662 av_log(avctx, AV_LOG_ERROR,
663 "Invalid frame offsets\n");
664 return AVERROR_INVALIDDATA;
667 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
668 p->linesize[0], buf + offset_ry,
669 buf_size - offset_ry);
670 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
671 avctx->height, p->linesize[1],
672 buf + offset_gu, buf_size - offset_gu);
673 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
674 avctx->height, p->linesize[2],
675 buf + offset_bv, buf_size - offset_bv);
677 case FRAME_ARITH_YV12:
678 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
680 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
682 if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
683 return AVERROR_INVALIDDATA;
686 if (offset_ry >= buf_size ||
687 offset_gu >= buf_size ||
688 offset_bv >= buf_size) {
689 av_log(avctx, AV_LOG_ERROR,
690 "Invalid frame offsets\n");
691 return AVERROR_INVALIDDATA;
694 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
695 p->linesize[0], buf + offset_ry,
696 buf_size - offset_ry);
697 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
698 avctx->height / 2, p->linesize[2],
699 buf + offset_gu, buf_size - offset_gu);
700 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
701 avctx->height / 2, p->linesize[1],
702 buf + offset_bv, buf_size - offset_bv);
705 av_log(avctx, AV_LOG_ERROR,
706 "Unsupported Lagarith frame type: %#x\n", frametype);
707 return AVERROR_PATCHWELCOME;
715 static av_cold int lag_decode_init(AVCodecContext *avctx)
717 LagarithContext *l = avctx->priv_data;
720 ff_dsputil_init(&l->dsp, avctx);
725 static av_cold int lag_decode_end(AVCodecContext *avctx)
727 LagarithContext *l = avctx->priv_data;
729 av_freep(&l->rgb_planes);
734 AVCodec ff_lagarith_decoder = {
736 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
737 .type = AVMEDIA_TYPE_VIDEO,
738 .id = AV_CODEC_ID_LAGARITH,
739 .priv_data_size = sizeof(LagarithContext),
740 .init = lag_decode_init,
741 .close = lag_decode_end,
742 .decode = lag_decode_frame,
743 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,