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 if (scale_factor > 23)
230 return AVERROR_INVALIDDATA;
232 rac->scale = scale_factor;
234 /* Fill probability array with cumulative probability for each symbol. */
235 for (i = 1; i < 257; i++)
236 rac->prob[i] += rac->prob[i - 1];
241 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
242 uint8_t *diff, int w, int *left,
245 /* This is almost identical to add_hfyu_median_pred in huffyuvdsp.h.
246 * However the &0xFF on the gradient predictor yields incorrect output
255 for (i = 0; i < w; i++) {
256 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
265 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
266 int width, int stride, int line)
271 /* Left prediction only for first line */
272 L = l->llviddsp.add_left_pred(buf, buf, width, 0);
274 /* Left pixel is actually prev_row[width] */
275 L = buf[width - stride - 1];
278 /* Second line, left predict first pixel, the rest of the line is median predicted
279 * NOTE: In the case of RGB this pixel is top predicted */
280 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
282 /* Top left is 2 rows back, last pixel */
283 TL = buf[width - (2 * stride) - 1];
286 add_lag_median_prediction(buf, buf - stride, buf,
291 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
292 int width, int stride, int line,
301 l->llviddsp.add_left_pred(buf, buf, width, 0);
307 const int HEAD = is_luma ? 4 : 2;
310 L = buf[width - stride - 1];
311 TL = buf[HEAD - stride - 1];
312 for (i = 0; i < HEAD; i++) {
316 for (; i < width; i++) {
317 L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
318 TL = buf[i - stride];
322 TL = buf[width - (2 * stride) - 1];
323 L = buf[width - stride - 1];
324 l->llviddsp.add_median_pred(buf, buf - stride, buf, width, &L, &TL);
328 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
329 uint8_t *dst, int width, int stride,
338 /* Output any zeros remaining from the previous run */
341 int count = FFMIN(l->zeros_rem, width - i);
342 memset(dst + i, 0, count);
344 l->zeros_rem -= count;
348 dst[i] = lag_get_rac(rac);
357 if (l->zeros == esc_count) {
358 int index = lag_get_rac(rac);
363 l->zeros_rem = lag_calc_zero_run(index);
370 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
371 const uint8_t *src, const uint8_t *src_end,
372 int width, int esc_count)
376 uint8_t zero_run = 0;
377 const uint8_t *src_start = src;
378 uint8_t mask1 = -(esc_count < 2);
379 uint8_t mask2 = -(esc_count < 3);
380 uint8_t *end = dst + (width - 2);
382 avpriv_request_sample(l->avctx, "zero_run_line");
384 memset(dst, 0, width);
388 count = FFMIN(l->zeros_rem, width - i);
389 if (end - dst < count) {
390 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
391 return AVERROR_INVALIDDATA;
394 memset(dst, 0, count);
395 l->zeros_rem -= count;
401 while (!zero_run && dst + i < end) {
403 if (i+2 >= src_end - src)
404 return AVERROR_INVALIDDATA;
406 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
413 l->zeros_rem = lag_calc_zero_run(src[i]);
423 return src - src_start;
428 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
429 int width, int height, int stride,
430 const uint8_t *src, int src_size)
439 const uint8_t *src_end = src + src_size;
442 rac.avctx = l->avctx;
446 return AVERROR_INVALIDDATA;
450 length = width * height;
452 return AVERROR_INVALIDDATA;
453 if (esc_count && AV_RL32(src + 1) < length) {
454 length = AV_RL32(src + 1);
458 if ((ret = init_get_bits8(&gb, src + offset, src_size - offset)) < 0)
461 if (lag_read_prob_header(&rac, &gb) < 0)
464 ff_lag_rac_init(&rac, &gb, length - stride);
465 for (i = 0; i < height; i++) {
466 if (rac.overread > MAX_OVERREAD)
467 return AVERROR_INVALIDDATA;
468 read += lag_decode_line(l, &rac, dst + (i * stride), width,
473 av_log(l->avctx, AV_LOG_WARNING,
474 "Output more bytes than length (%d of %"PRIu32")\n", read,
476 } else if (esc_count < 8) {
481 /* Zero run coding only, no range coding. */
482 for (i = 0; i < height; i++) {
483 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
484 src_end, width, esc_count);
490 if (src_size < width * height)
491 return AVERROR_INVALIDDATA; // buffer not big enough
492 /* Plane is stored uncompressed */
493 for (i = 0; i < height; i++) {
494 memcpy(dst + (i * stride), src, width);
498 } else if (esc_count == 0xff) {
499 /* Plane is a solid run of given value */
500 for (i = 0; i < height; i++)
501 memset(dst + i * stride, src[1], width);
502 /* Do not apply prediction.
503 Note: memset to 0 above, setting first value to src[1]
504 and applying prediction gives the same result. */
507 av_log(l->avctx, AV_LOG_ERROR,
508 "Invalid zero run escape code! (%#x)\n", esc_count);
512 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
513 for (i = 0; i < height; i++) {
514 lag_pred_line(l, dst, width, stride, i);
518 for (i = 0; i < height; i++) {
519 lag_pred_line_yuy2(l, dst, width, stride, i,
520 width == l->avctx->width);
530 * @param avctx codec context
531 * @param data output AVFrame
532 * @param data_size size of output data or 0 if no picture is returned
533 * @param avpkt input packet
534 * @return number of consumed bytes on success or negative if decode fails
536 static int lag_decode_frame(AVCodecContext *avctx,
537 void *data, int *got_frame, AVPacket *avpkt)
539 const uint8_t *buf = avpkt->data;
540 unsigned int buf_size = avpkt->size;
541 LagarithContext *l = avctx->priv_data;
542 ThreadFrame frame = { .f = data };
543 AVFrame *const p = data;
545 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
548 int i, j, planes = 3;
552 p->pict_type = AV_PICTURE_TYPE_I;
556 offset_gu = AV_RL32(buf + 1);
557 offset_bv = AV_RL32(buf + 5);
560 case FRAME_SOLID_RGBA:
561 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
562 case FRAME_SOLID_GRAY:
563 if (frametype == FRAME_SOLID_GRAY)
564 if (avctx->bits_per_coded_sample == 24) {
565 avctx->pix_fmt = AV_PIX_FMT_GBRP;
567 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
571 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
574 if (frametype == FRAME_SOLID_RGBA) {
575 for (i = 0; i < avctx->height; i++) {
576 memset(p->data[0] + i * p->linesize[0], buf[2], avctx->width);
577 memset(p->data[1] + i * p->linesize[1], buf[1], avctx->width);
578 memset(p->data[2] + i * p->linesize[2], buf[3], avctx->width);
579 memset(p->data[3] + i * p->linesize[3], buf[4], avctx->width);
582 for (i = 0; i < avctx->height; i++) {
583 for (j = 0; j < planes; j++)
584 memset(p->data[j] + i * p->linesize[j], buf[1], avctx->width);
588 case FRAME_SOLID_COLOR:
589 if (avctx->bits_per_coded_sample == 24) {
590 avctx->pix_fmt = AV_PIX_FMT_GBRP;
592 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
595 if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
598 for (i = 0; i < avctx->height; i++) {
599 memset(p->data[0] + i * p->linesize[0], buf[2], avctx->width);
600 memset(p->data[1] + i * p->linesize[1], buf[1], avctx->width);
601 memset(p->data[2] + i * p->linesize[2], buf[3], avctx->width);
602 if (avctx->pix_fmt == AV_PIX_FMT_GBRAP)
603 memset(p->data[3] + i * p->linesize[3], 0xFFu, avctx->width);
606 case FRAME_ARITH_RGBA:
607 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
610 offs[3] = AV_RL32(buf + 9);
611 case FRAME_ARITH_RGB24:
613 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
614 avctx->pix_fmt = AV_PIX_FMT_GBRP;
616 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
623 for (i = 0; i < planes; i++)
624 srcs[i] = p->data[i] + (avctx->height - 1) * p->linesize[i];
625 for (i = 0; i < planes; i++)
626 if (buf_size <= offs[i]) {
627 av_log(avctx, AV_LOG_ERROR,
628 "Invalid frame offsets\n");
629 return AVERROR_INVALIDDATA;
632 for (i = 0; i < planes; i++)
633 lag_decode_arith_plane(l, srcs[i],
634 avctx->width, avctx->height,
635 -p->linesize[i], buf + offs[i],
637 for (i = 0; i < avctx->height; i++) {
638 l->llviddsp.add_bytes(p->data[0] + i * p->linesize[0], p->data[1] + i * p->linesize[1], avctx->width);
639 l->llviddsp.add_bytes(p->data[2] + i * p->linesize[2], p->data[1] + i * p->linesize[1], avctx->width);
641 FFSWAP(uint8_t*, p->data[0], p->data[1]);
642 FFSWAP(int, p->linesize[0], p->linesize[1]);
643 FFSWAP(uint8_t*, p->data[2], p->data[1]);
644 FFSWAP(int, p->linesize[2], p->linesize[1]);
646 case FRAME_ARITH_YUY2:
647 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
649 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
652 if (offset_ry >= buf_size ||
653 offset_gu >= buf_size ||
654 offset_bv >= buf_size) {
655 av_log(avctx, AV_LOG_ERROR,
656 "Invalid frame offsets\n");
657 return AVERROR_INVALIDDATA;
660 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
661 p->linesize[0], buf + offset_ry,
662 buf_size - offset_ry);
663 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
664 avctx->height, p->linesize[1],
665 buf + offset_gu, buf_size - offset_gu);
666 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
667 avctx->height, p->linesize[2],
668 buf + offset_bv, buf_size - offset_bv);
670 case FRAME_ARITH_YV12:
671 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
673 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
676 if (offset_ry >= buf_size ||
677 offset_gu >= buf_size ||
678 offset_bv >= buf_size) {
679 av_log(avctx, AV_LOG_ERROR,
680 "Invalid frame offsets\n");
681 return AVERROR_INVALIDDATA;
684 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
685 p->linesize[0], buf + offset_ry,
686 buf_size - offset_ry);
687 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
688 (avctx->height + 1) / 2, p->linesize[2],
689 buf + offset_gu, buf_size - offset_gu);
690 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
691 (avctx->height + 1) / 2, p->linesize[1],
692 buf + offset_bv, buf_size - offset_bv);
695 av_log(avctx, AV_LOG_ERROR,
696 "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
697 return AVERROR_PATCHWELCOME;
705 static av_cold int lag_decode_init(AVCodecContext *avctx)
707 LagarithContext *l = avctx->priv_data;
710 ff_llviddsp_init(&l->llviddsp);
715 AVCodec ff_lagarith_decoder = {
717 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
718 .type = AVMEDIA_TYPE_VIDEO,
719 .id = AV_CODEC_ID_LAGARITH,
720 .priv_data_size = sizeof(LagarithContext),
721 .init = lag_decode_init,
722 .decode = lag_decode_frame,
723 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,