2 * Lagarith lossless decoder
3 * Copyright (c) 2009 Nathan Caldwell <saintdev (at) gmail.com>
5 * This file is part of Libav.
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.
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.
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
24 * Lagarith lossless decoder
25 * @author Nathan Caldwell
33 #include "huffyuvdsp.h"
34 #include "lagarithrac.h"
37 enum LagarithFrameType {
38 FRAME_RAW = 1, /**< uncompressed */
39 FRAME_U_RGB24 = 2, /**< unaligned RGB24 */
40 FRAME_ARITH_YUY2 = 3, /**< arithmetic coded YUY2 */
41 FRAME_ARITH_RGB24 = 4, /**< arithmetic coded RGB24 */
42 FRAME_SOLID_GRAY = 5, /**< solid grayscale color frame */
43 FRAME_SOLID_COLOR = 6, /**< solid non-grayscale color frame */
44 FRAME_OLD_ARITH_RGB = 7, /**< obsolete arithmetic coded RGB (no longer encoded by upstream since version 1.1.0) */
45 FRAME_ARITH_RGBA = 8, /**< arithmetic coded RGBA */
46 FRAME_SOLID_RGBA = 9, /**< solid RGBA color frame */
47 FRAME_ARITH_YV12 = 10, /**< arithmetic coded YV12 */
48 FRAME_REDUCED_RES = 11, /**< reduced resolution YV12 frame */
51 typedef struct LagarithContext {
52 AVCodecContext *avctx;
53 HuffYUVDSPContext hdsp;
54 int zeros; /**< number of consecutive zero bytes encountered */
55 int zeros_rem; /**< number of zero bytes remaining to output */
57 int rgb_planes_allocated;
62 * Compute the 52-bit mantissa of 1/(double)denom.
63 * This crazy format uses floats in an entropy coder and we have to match x86
64 * rounding exactly, thus ordinary floats aren't portable enough.
65 * @param denom denominator
66 * @return 52-bit mantissa
69 static uint64_t softfloat_reciprocal(uint32_t denom)
71 int shift = av_log2(denom - 1) + 1;
72 uint64_t ret = (1ULL << 52) / denom;
73 uint64_t err = (1ULL << 52) - ret * denom;
77 return ret + err / denom;
81 * (uint32_t)(x*f), where f has the given mantissa, and exponent 0
82 * Used in combination with softfloat_reciprocal computes x/(double)denom.
83 * @param x 32-bit integer factor
84 * @param mantissa mantissa of f with exponent 0
85 * @return 32-bit integer value (x*f)
86 * @see softfloat_reciprocal
88 static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
90 uint64_t l = x * (mantissa & 0xffffffff);
91 uint64_t h = x * (mantissa >> 32);
94 l += 1 << av_log2(h >> 21);
99 static uint8_t lag_calc_zero_run(int8_t x)
101 return (x << 1) ^ (x >> 7);
104 static int lag_decode_prob(GetBitContext *gb, uint32_t *value)
106 static const uint8_t series[] = { 1, 2, 3, 5, 8, 13, 21 };
113 for (i = 0; i < 7; i++) {
122 if (bits < 0 || bits > 31) {
125 } else if (bits == 0) {
130 val = get_bits_long(gb, bits);
138 static int lag_read_prob_header(lag_rac *rac, GetBitContext *gb)
140 int i, j, scale_factor;
141 unsigned prob, cumulative_target;
142 unsigned cumul_prob = 0;
143 unsigned scaled_cumul_prob = 0;
146 rac->prob[257] = UINT_MAX;
147 /* Read probabilities from bitstream */
148 for (i = 1; i < 257; i++) {
149 if (lag_decode_prob(gb, &rac->prob[i]) < 0) {
150 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability encountered.\n");
153 if ((uint64_t)cumul_prob + rac->prob[i] > UINT_MAX) {
154 av_log(rac->avctx, AV_LOG_ERROR, "Integer overflow encountered in cumulative probability calculation.\n");
157 cumul_prob += rac->prob[i];
159 if (lag_decode_prob(gb, &prob)) {
160 av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
165 for (j = 0; j < prob; j++)
171 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
175 /* Scale probabilities so cumulative probability is an even power of 2. */
176 scale_factor = av_log2(cumul_prob);
178 if (cumul_prob & (cumul_prob - 1)) {
179 uint64_t mul = softfloat_reciprocal(cumul_prob);
180 for (i = 1; i < 257; i++) {
181 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
182 scaled_cumul_prob += rac->prob[i];
186 cumulative_target = 1 << scale_factor;
188 if (scaled_cumul_prob > cumulative_target) {
189 av_log(rac->avctx, AV_LOG_ERROR,
190 "Scaled probabilities are larger than target!\n");
194 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
196 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
201 /* Comment from reference source:
202 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
203 * // since the compression change is negligible and fixing it
204 * // breaks backwards compatibility
205 * b =- (signed int)b;
215 rac->scale = scale_factor;
217 /* Fill probability array with cumulative probability for each symbol. */
218 for (i = 1; i < 257; i++)
219 rac->prob[i] += rac->prob[i - 1];
224 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
225 uint8_t *diff, int w, int *left,
228 /* This is almost identical to add_hfyu_median_pred in huffyuvdsp.h.
229 * However the &0xFF on the gradient predictor yields incorrect output
238 for (i = 0; i < w; i++) {
239 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
248 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
249 int width, int stride, int line)
254 int i, align_width = (width - 1) & ~31;
255 /* Left prediction only for first line */
256 L = l->hdsp.add_hfyu_left_pred(buf + 1, buf + 1, align_width, buf[0]);
257 for (i = align_width + 1; i < width; i++)
258 buf[i] += buf[i - 1];
260 /* Left pixel is actually prev_row[width] */
261 L = buf[width - stride - 1];
264 /* Second line, left predict first pixel, the rest of the line is median predicted
265 * NOTE: In the case of RGB this pixel is top predicted */
266 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
268 /* Top left is 2 rows back, last pixel */
269 TL = buf[width - (2 * stride) - 1];
272 add_lag_median_prediction(buf, buf - stride, buf,
277 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
278 int width, int stride, int line,
290 align_width = (width - 1) & ~31;
291 l->hdsp.add_hfyu_left_pred(buf + 1, buf + 1, align_width, buf[0]);
293 for (i = align_width + 1; i < width; i++)
294 buf[i] += buf[i - 1];
299 const int HEAD = is_luma ? 4 : 2;
302 L = buf[width - stride - 1];
303 TL = buf[HEAD - stride - 1];
304 for (i = 0; i < HEAD; i++) {
308 for (; i < width; i++) {
309 L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
310 TL = buf[i - stride];
314 TL = buf[width - (2 * stride) - 1];
315 L = buf[width - stride - 1];
316 l->hdsp.add_hfyu_median_pred(buf, buf - stride, buf, width, &L, &TL);
320 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
321 uint8_t *dst, int width, int stride,
330 /* Output any zeros remaining from the previous run */
333 int count = FFMIN(l->zeros_rem, width - i);
334 memset(dst + i, 0, count);
336 l->zeros_rem -= count;
340 dst[i] = lag_get_rac(rac);
349 if (l->zeros == esc_count) {
350 int index = lag_get_rac(rac);
355 l->zeros_rem = lag_calc_zero_run(index);
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)
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);
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;
382 memset(dst, 0, count);
383 l->zeros_rem -= count;
389 while (!zero_run && dst + i < end) {
391 if (src + i >= src_end)
392 return AVERROR_INVALIDDATA;
394 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
401 l->zeros_rem = lag_calc_zero_run(src[i]);
411 return src_start - src;
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)
424 int esc_count = src[0];
427 const uint8_t *src_end = src + src_size;
429 rac.avctx = l->avctx;
433 length = width * height;
434 if (esc_count && AV_RL32(src + 1) < length) {
435 length = AV_RL32(src + 1);
439 init_get_bits(&gb, src + offset, src_size * 8);
441 if (lag_read_prob_header(&rac, &gb) < 0)
444 ff_lag_rac_init(&rac, &gb, length - stride);
446 for (i = 0; i < height; i++)
447 read += lag_decode_line(l, &rac, dst + (i * stride), width,
451 av_log(l->avctx, AV_LOG_WARNING,
452 "Output more bytes than length (%d of %"PRIu32")\n", read,
454 } else if (esc_count < 8) {
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);
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);
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. */
483 av_log(l->avctx, AV_LOG_ERROR,
484 "Invalid zero run escape code! (%#x)\n", esc_count);
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);
494 for (i = 0; i < height; i++) {
495 lag_pred_line_yuy2(l, dst, width, stride, i,
496 width == l->avctx->width);
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
512 static int lag_decode_frame(AVCodecContext *avctx,
513 void *data, int *got_frame, AVPacket *avpkt)
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;
523 uint8_t *srcs[4], *dst;
524 int i, j, planes = 3;
530 offset_gu = AV_RL32(buf + 1);
531 offset_bv = AV_RL32(buf + 5);
534 case FRAME_SOLID_RGBA:
535 avctx->pix_fmt = AV_PIX_FMT_RGB32;
537 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
538 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
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];
549 case FRAME_ARITH_RGBA:
550 avctx->pix_fmt = AV_PIX_FMT_RGB32;
553 offs[3] = AV_RL32(buf + 9);
554 case FRAME_ARITH_RGB24:
556 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
557 avctx->pix_fmt = AV_PIX_FMT_RGB24;
559 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
560 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
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);
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;
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],
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++) {
601 if (frametype == FRAME_ARITH_RGBA) {
603 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
610 dst += p->linesize[0];
611 for (i = 0; i < planes; i++)
612 srcs[i] += l->rgb_stride;
615 case FRAME_ARITH_YUY2:
616 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
618 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
619 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
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;
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);
641 case FRAME_ARITH_YV12:
642 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
644 if (ff_thread_get_buffer(avctx, &frame, 0) < 0) {
645 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
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;
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);
668 av_log(avctx, AV_LOG_ERROR,
669 "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
678 static av_cold int lag_decode_init(AVCodecContext *avctx)
680 LagarithContext *l = avctx->priv_data;
683 ff_huffyuvdsp_init(&l->hdsp);
688 static av_cold int lag_decode_end(AVCodecContext *avctx)
690 LagarithContext *l = avctx->priv_data;
692 av_freep(&l->rgb_planes);
697 AVCodec ff_lagarith_decoder = {
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 = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,