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
33 #include "lagarithrac.h"
34 #include "lossless_videodsp.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 LLVidDSPContext llviddsp;
54 int zeros; /**< number of consecutive zero bytes encountered */
55 int zeros_rem; /**< number of zero bytes remaining to output */
59 * Compute the 52-bit 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 52-bit 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 32-bit integer factor
81 * @param mantissa mantissa of f with exponent 0
82 * @return 32-bit 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 += 1LL << av_log2(h >> 21);
96 static uint8_t lag_calc_zero_run(int8_t x)
98 return (x * 2) ^ (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;
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++)
171 av_log(rac->avctx, AV_LOG_ERROR, "All probabilities are 0!\n");
175 if (nnz == 1 && (show_bits_long(gb, 32) & 0xFFFFFF)) {
176 return AVERROR_INVALIDDATA;
179 /* Scale probabilities so cumulative probability is an even power of 2. */
180 scale_factor = av_log2(cumul_prob);
182 if (cumul_prob & (cumul_prob - 1)) {
183 uint64_t mul = softfloat_reciprocal(cumul_prob);
184 for (i = 1; i <= 128; i++) {
185 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
186 scaled_cumul_prob += rac->prob[i];
188 if (scaled_cumul_prob <= 0) {
189 av_log(rac->avctx, AV_LOG_ERROR, "Scaled probabilities invalid\n");
190 return AVERROR_INVALIDDATA;
192 for (; i < 257; i++) {
193 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
194 scaled_cumul_prob += rac->prob[i];
198 if (scale_factor >= 32U)
199 return AVERROR_INVALIDDATA;
200 cumulative_target = 1U << scale_factor;
202 if (scaled_cumul_prob > cumulative_target) {
203 av_log(rac->avctx, AV_LOG_ERROR,
204 "Scaled probabilities are larger than target!\n");
208 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
210 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
215 /* Comment from reference source:
216 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
217 * // since the compression change is negligible and fixing it
218 * // breaks backwards compatibility
219 * b =- (signed int)b;
229 rac->scale = scale_factor;
231 /* Fill probability array with cumulative probability for each symbol. */
232 for (i = 1; i < 257; i++)
233 rac->prob[i] += rac->prob[i - 1];
238 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
239 uint8_t *diff, int w, int *left,
242 /* This is almost identical to add_hfyu_median_pred in huffyuvdsp.h.
243 * However the &0xFF on the gradient predictor yields incorrect output
252 for (i = 0; i < w; i++) {
253 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
262 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
263 int width, int stride, int line)
268 /* Left prediction only for first line */
269 L = l->llviddsp.add_left_pred(buf, buf, width, 0);
271 /* Left pixel is actually prev_row[width] */
272 L = buf[width - stride - 1];
275 /* Second line, left predict first pixel, the rest of the line is median predicted
276 * NOTE: In the case of RGB this pixel is top predicted */
277 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
279 /* Top left is 2 rows back, last pixel */
280 TL = buf[width - (2 * stride) - 1];
283 add_lag_median_prediction(buf, buf - stride, buf,
288 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
289 int width, int stride, int line,
298 l->llviddsp.add_left_pred(buf, buf, width, 0);
304 const int HEAD = is_luma ? 4 : 2;
307 L = buf[width - stride - 1];
308 TL = buf[HEAD - stride - 1];
309 for (i = 0; i < HEAD; i++) {
313 for (; i < width; i++) {
314 L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
315 TL = buf[i - stride];
319 TL = buf[width - (2 * stride) - 1];
320 L = buf[width - stride - 1];
321 l->llviddsp.add_median_pred(buf, buf - stride, buf, width, &L, &TL);
325 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
326 uint8_t *dst, int width, int stride,
335 /* Output any zeros remaining from the previous run */
338 int count = FFMIN(l->zeros_rem, width - i);
339 memset(dst + i, 0, count);
341 l->zeros_rem -= count;
345 dst[i] = lag_get_rac(rac);
354 if (l->zeros == esc_count) {
355 int index = lag_get_rac(rac);
360 l->zeros_rem = lag_calc_zero_run(index);
367 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
368 const uint8_t *src, const uint8_t *src_end,
369 int width, int esc_count)
373 uint8_t zero_run = 0;
374 const uint8_t *src_start = src;
375 uint8_t mask1 = -(esc_count < 2);
376 uint8_t mask2 = -(esc_count < 3);
377 uint8_t *end = dst + (width - 2);
379 avpriv_request_sample(l->avctx, "zero_run_line");
381 memset(dst, 0, width);
385 count = FFMIN(l->zeros_rem, width - i);
386 if (end - dst < count) {
387 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
388 return AVERROR_INVALIDDATA;
391 memset(dst, 0, count);
392 l->zeros_rem -= count;
398 while (!zero_run && dst + i < end) {
400 if (i+2 >= src_end - src)
401 return AVERROR_INVALIDDATA;
403 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
410 l->zeros_rem = lag_calc_zero_run(src[i]);
420 return src - src_start;
425 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
426 int width, int height, int stride,
427 const uint8_t *src, int src_size)
436 const uint8_t *src_end = src + src_size;
439 rac.avctx = l->avctx;
443 return AVERROR_INVALIDDATA;
447 length = width * height;
449 return AVERROR_INVALIDDATA;
450 if (esc_count && AV_RL32(src + 1) < length) {
451 length = AV_RL32(src + 1);
455 if ((ret = init_get_bits8(&gb, src + offset, src_size - offset)) < 0)
458 if (lag_read_prob_header(&rac, &gb) < 0)
461 ff_lag_rac_init(&rac, &gb, length - stride);
462 for (i = 0; i < height; i++) {
463 if (rac.overread > MAX_OVERREAD)
464 return AVERROR_INVALIDDATA;
465 read += lag_decode_line(l, &rac, dst + (i * stride), width,
470 av_log(l->avctx, AV_LOG_WARNING,
471 "Output more bytes than length (%d of %"PRIu32")\n", read,
473 } else if (esc_count < 8) {
478 /* Zero run coding only, no range coding. */
479 for (i = 0; i < height; i++) {
480 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
481 src_end, width, esc_count);
487 if (src_size < width * height)
488 return AVERROR_INVALIDDATA; // buffer not big enough
489 /* Plane is stored uncompressed */
490 for (i = 0; i < height; i++) {
491 memcpy(dst + (i * stride), src, width);
495 } else if (esc_count == 0xff) {
496 /* Plane is a solid run of given value */
497 for (i = 0; i < height; i++)
498 memset(dst + i * stride, src[1], width);
499 /* Do not apply prediction.
500 Note: memset to 0 above, setting first value to src[1]
501 and applying prediction gives the same result. */
504 av_log(l->avctx, AV_LOG_ERROR,
505 "Invalid zero run escape code! (%#x)\n", esc_count);
509 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
510 for (i = 0; i < height; i++) {
511 lag_pred_line(l, dst, width, stride, i);
515 for (i = 0; i < height; i++) {
516 lag_pred_line_yuy2(l, dst, width, stride, i,
517 width == l->avctx->width);
527 * @param avctx codec context
528 * @param data output AVFrame
529 * @param data_size size of output data or 0 if no picture is returned
530 * @param avpkt input packet
531 * @return number of consumed bytes on success or negative if decode fails
533 static int lag_decode_frame(AVCodecContext *avctx,
534 void *data, int *got_frame, AVPacket *avpkt)
536 const uint8_t *buf = avpkt->data;
537 unsigned int buf_size = avpkt->size;
538 LagarithContext *l = avctx->priv_data;
539 ThreadFrame frame = { .f = data };
540 AVFrame *const p = data;
542 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
545 int i, j, planes = 3;
549 p->pict_type = AV_PICTURE_TYPE_I;
553 offset_gu = AV_RL32(buf + 1);
554 offset_bv = AV_RL32(buf + 5);
557 case FRAME_SOLID_RGBA:
558 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
559 case FRAME_SOLID_GRAY:
560 if (frametype == FRAME_SOLID_GRAY)
561 if (avctx->bits_per_coded_sample == 24) {
562 avctx->pix_fmt = AV_PIX_FMT_GBRP;
564 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
568 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
571 if (frametype == FRAME_SOLID_RGBA) {
572 for (i = 0; i < avctx->height; i++) {
573 memset(p->data[0] + i * p->linesize[0], buf[2], avctx->width);
574 memset(p->data[1] + i * p->linesize[1], buf[1], avctx->width);
575 memset(p->data[2] + i * p->linesize[2], buf[3], avctx->width);
576 memset(p->data[3] + i * p->linesize[3], buf[4], avctx->width);
579 for (i = 0; i < avctx->height; i++) {
580 for (j = 0; j < planes; j++)
581 memset(p->data[j] + i * p->linesize[j], buf[1], avctx->width);
585 case FRAME_SOLID_COLOR:
586 if (avctx->bits_per_coded_sample == 24) {
587 avctx->pix_fmt = AV_PIX_FMT_GBRP;
589 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
592 if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
595 for (i = 0; i < avctx->height; i++) {
596 memset(p->data[0] + i * p->linesize[0], buf[2], avctx->width);
597 memset(p->data[1] + i * p->linesize[1], buf[1], avctx->width);
598 memset(p->data[2] + i * p->linesize[2], buf[3], avctx->width);
599 if (avctx->pix_fmt == AV_PIX_FMT_GBRAP)
600 memset(p->data[3] + i * p->linesize[3], 0xFFu, avctx->width);
603 case FRAME_ARITH_RGBA:
604 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
607 offs[3] = AV_RL32(buf + 9);
608 case FRAME_ARITH_RGB24:
610 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
611 avctx->pix_fmt = AV_PIX_FMT_GBRP;
613 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
620 for (i = 0; i < planes; i++)
621 srcs[i] = p->data[i] + (avctx->height - 1) * p->linesize[i];
622 for (i = 0; i < planes; i++)
623 if (buf_size <= offs[i]) {
624 av_log(avctx, AV_LOG_ERROR,
625 "Invalid frame offsets\n");
626 return AVERROR_INVALIDDATA;
629 for (i = 0; i < planes; i++)
630 lag_decode_arith_plane(l, srcs[i],
631 avctx->width, avctx->height,
632 -p->linesize[i], buf + offs[i],
634 for (i = 0; i < avctx->height; i++) {
635 l->llviddsp.add_bytes(p->data[0] + i * p->linesize[0], p->data[1] + i * p->linesize[1], avctx->width);
636 l->llviddsp.add_bytes(p->data[2] + i * p->linesize[2], p->data[1] + i * p->linesize[1], avctx->width);
638 FFSWAP(uint8_t*, p->data[0], p->data[1]);
639 FFSWAP(int, p->linesize[0], p->linesize[1]);
640 FFSWAP(uint8_t*, p->data[2], p->data[1]);
641 FFSWAP(int, p->linesize[2], p->linesize[1]);
643 case FRAME_ARITH_YUY2:
644 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
646 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
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[1], (avctx->width + 1) / 2,
661 avctx->height, p->linesize[1],
662 buf + offset_gu, buf_size - offset_gu);
663 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
664 avctx->height, p->linesize[2],
665 buf + offset_bv, buf_size - offset_bv);
667 case FRAME_ARITH_YV12:
668 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
670 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
673 if (offset_ry >= buf_size ||
674 offset_gu >= buf_size ||
675 offset_bv >= buf_size) {
676 av_log(avctx, AV_LOG_ERROR,
677 "Invalid frame offsets\n");
678 return AVERROR_INVALIDDATA;
681 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
682 p->linesize[0], buf + offset_ry,
683 buf_size - offset_ry);
684 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
685 (avctx->height + 1) / 2, p->linesize[2],
686 buf + offset_gu, buf_size - offset_gu);
687 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
688 (avctx->height + 1) / 2, p->linesize[1],
689 buf + offset_bv, buf_size - offset_bv);
692 av_log(avctx, AV_LOG_ERROR,
693 "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
694 return AVERROR_PATCHWELCOME;
702 static av_cold int lag_decode_init(AVCodecContext *avctx)
704 LagarithContext *l = avctx->priv_data;
707 ff_llviddsp_init(&l->llviddsp);
713 static av_cold int lag_decode_init_thread_copy(AVCodecContext *avctx)
715 LagarithContext *l = avctx->priv_data;
722 AVCodec ff_lagarith_decoder = {
724 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
725 .type = AVMEDIA_TYPE_VIDEO,
726 .id = AV_CODEC_ID_LAGARITH,
727 .priv_data_size = sizeof(LagarithContext),
728 .init = lag_decode_init,
729 .init_thread_copy = ONLY_IF_THREADS_ENABLED(lag_decode_init_thread_copy),
730 .decode = lag_decode_frame,
731 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,