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;
53 int zeros; /**< number of consecutive zero bytes encountered */
54 int zeros_rem; /**< number of zero bytes remaining to output */
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
67 static uint64_t softfloat_reciprocal(uint32_t denom)
69 int shift = av_log2(denom - 1) + 1;
70 uint64_t ret = (1ULL << 52) / denom;
71 uint64_t err = (1ULL << 52) - ret * denom;
75 return ret + err / denom;
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
86 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
88 uint64_t l = x * (mantissa & 0xffffffff);
89 uint64_t h = x * (mantissa >> 32);
92 l += 1 << av_log2(h >> 21);
97 static uint8_t lag_calc_zero_run(int8_t x)
99 return (x << 1) ^ (x >> 7);
102 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
104 static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
111 for (i = 0; i < 7; i++) {
120 if (bits < 0 || bits > 31) {
123 } else if (bits == 0) {
128 val = get_bits_long(gb, bits);
136 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
138 int i, j, scale_factor;
139 unsigned prob, cumulative_target;
140 unsigned cumul_prob = 0;
141 unsigned scaled_cumul_prob = 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");
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");
155 cumul_prob += rac->prob[i];
157 if (lag_decode_prob(gb, &prob)) {
158 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
163 for (j = 0; j < prob; j++)
169 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
173 /* Scale probabilities so cumulative probability is an even power of 2. */
174 scale_factor = av_log2(cumul_prob);
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];
184 cumulative_target = 1 << scale_factor;
186 if (scaled_cumul_prob > cumulative_target) {
187 av_log(rac->avctx, AV_LOG_ERROR,
188 "Scaled probabilities are larger than target!\n");
192 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
194 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
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 compatibility
203 * b =- (signed int)b;
213 rac->scale = scale_factor;
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];
222 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
223 uint8_t *diff, int w, int *left,
226 /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
227 * However the &0xFF on the gradient predictor yealds incorrect output
236 for (i = 0; i < w; i++) {
237 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
246 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
247 int width, int stride, int line)
252 /* Left prediction only for first line */
253 L = l->dsp.add_hfyu_left_prediction(buf, buf,
256 /* Left pixel is actually prev_row[width] */
257 L = buf[width - stride - 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 == PIX_FMT_YUV420P ? buf[-stride] : L;
264 /* Top left is 2 rows back, last pixel */
265 TL = buf[width - (2 * stride) - 1];
268 add_lag_median_prediction(buf, buf - stride, buf,
273 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
274 int width, int stride, int line,
283 l->dsp.add_hfyu_left_prediction(buf, buf, width, 0);
289 const int HEAD = is_luma ? 4 : 2;
292 L = buf[width - stride - 1];
293 TL = buf[HEAD - stride - 1];
294 for (i = 0; i < HEAD; i++) {
298 for (; i<width; i++) {
299 L = mid_pred(L&0xFF, buf[i-stride], (L + buf[i-stride] - TL)&0xFF) + buf[i];
304 TL = buf[width - (2 * stride) - 1];
305 L = buf[width - stride - 1];
306 l->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
311 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
312 uint8_t *dst, int width, int stride,
321 /* Output any zeros remaining from the previous run */
324 int count = FFMIN(l->zeros_rem, width - i);
325 memset(dst + i, 0, count);
327 l->zeros_rem -= count;
331 dst[i] = lag_get_rac(rac);
340 if (l->zeros == esc_count) {
341 int index = lag_get_rac(rac);
346 l->zeros_rem = lag_calc_zero_run(index);
353 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
354 const uint8_t *src, const uint8_t *src_end,
355 int width, int esc_count)
359 uint8_t zero_run = 0;
360 const uint8_t *src_start = src;
361 uint8_t mask1 = -(esc_count < 2);
362 uint8_t mask2 = -(esc_count < 3);
363 uint8_t *end = dst + (width - 2);
367 count = FFMIN(l->zeros_rem, width - i);
368 if(end - dst < count) {
369 av_log(l->avctx, AV_LOG_ERROR, "too many zeros remaining\n");
370 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 != 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 *data_size, AVPacket *avpkt)
511 const uint8_t *buf = avpkt->data;
512 unsigned int buf_size = avpkt->size;
513 LagarithContext *l = avctx->priv_data;
514 AVFrame *const p = &l->picture;
515 uint8_t frametype = 0;
516 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
518 uint8_t *srcs[4], *dst;
519 int i, j, planes = 3;
521 AVFrame *picture = data;
524 ff_thread_release_buffer(avctx, p);
531 offset_gu = AV_RL32(buf + 1);
532 offset_bv = AV_RL32(buf + 5);
535 case FRAME_SOLID_RGBA:
536 avctx->pix_fmt = PIX_FMT_RGB32;
538 if (ff_thread_get_buffer(avctx, p) < 0) {
539 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
544 for (j = 0; j < avctx->height; j++) {
545 for (i = 0; i < avctx->width; i++)
546 AV_WN32(dst + i * 4, offset_gu);
547 dst += p->linesize[0];
550 case FRAME_ARITH_RGBA:
551 avctx->pix_fmt = PIX_FMT_RGB32;
554 offs[3] = AV_RL32(buf + 9);
555 case FRAME_ARITH_RGB24:
557 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
558 avctx->pix_fmt = PIX_FMT_RGB24;
560 if (ff_thread_get_buffer(avctx, p) < 0) {
561 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
569 if (!l->rgb_planes) {
570 l->rgb_stride = FFALIGN(avctx->width, 16);
571 l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * planes + 16);
572 if (!l->rgb_planes) {
573 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
574 return AVERROR(ENOMEM);
577 for (i = 0; i < planes; i++)
578 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
579 for (i = 0; i < planes; i++)
580 if (buf_size <= offs[i]) {
581 av_log(avctx, AV_LOG_ERROR,
582 "Invalid frame offsets\n");
583 return AVERROR_INVALIDDATA;
586 for (i = 0; i < planes; i++)
587 lag_decode_arith_plane(l, srcs[i],
588 avctx->width, avctx->height,
589 -l->rgb_stride, buf + offs[i],
592 for (i = 0; i < planes; i++)
593 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
594 for (j = 0; j < avctx->height; j++) {
595 for (i = 0; i < avctx->width; i++) {
602 if (frametype == FRAME_ARITH_RGBA) {
604 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
611 dst += p->linesize[0];
612 for (i = 0; i < planes; i++)
613 srcs[i] += l->rgb_stride;
616 case FRAME_ARITH_YUY2:
617 avctx->pix_fmt = PIX_FMT_YUV422P;
619 if (ff_thread_get_buffer(avctx, p) < 0) {
620 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
624 if (offset_ry >= buf_size ||
625 offset_gu >= buf_size ||
626 offset_bv >= buf_size) {
627 av_log(avctx, AV_LOG_ERROR,
628 "Invalid frame offsets\n");
629 return AVERROR_INVALIDDATA;
632 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
633 p->linesize[0], buf + offset_ry,
634 buf_size - offset_ry);
635 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
636 avctx->height, p->linesize[1],
637 buf + offset_gu, buf_size - offset_gu);
638 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
639 avctx->height, p->linesize[2],
640 buf + offset_bv, buf_size - offset_bv);
642 case FRAME_ARITH_YV12:
643 avctx->pix_fmt = PIX_FMT_YUV420P;
645 if (ff_thread_get_buffer(avctx, p) < 0) {
646 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
649 if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
650 return AVERROR_INVALIDDATA;
653 if (offset_ry >= buf_size ||
654 offset_gu >= buf_size ||
655 offset_bv >= buf_size) {
656 av_log(avctx, AV_LOG_ERROR,
657 "Invalid frame offsets\n");
658 return AVERROR_INVALIDDATA;
661 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
662 p->linesize[0], buf + offset_ry,
663 buf_size - offset_ry);
664 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
665 avctx->height / 2, p->linesize[2],
666 buf + offset_gu, buf_size - offset_gu);
667 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
668 avctx->height / 2, p->linesize[1],
669 buf + offset_bv, buf_size - offset_bv);
672 av_log(avctx, AV_LOG_ERROR,
673 "Unsupported Lagarith frame type: %#x\n", frametype);
678 *data_size = sizeof(AVFrame);
683 static av_cold int lag_decode_init(AVCodecContext *avctx)
685 LagarithContext *l = avctx->priv_data;
688 ff_dsputil_init(&l->dsp, avctx);
693 static av_cold int lag_decode_end(AVCodecContext *avctx)
695 LagarithContext *l = avctx->priv_data;
697 if (l->picture.data[0])
698 ff_thread_release_buffer(avctx, &l->picture);
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"),