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 "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 52bit 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 52bit 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 32bit integer factor
84 * @param mantissa mantissa of f with exponent 0
85 * @return 32bit 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 <= 128; i++) {
181 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
182 scaled_cumul_prob += rac->prob[i];
184 if (scaled_cumul_prob <= 0) {
185 av_log(rac->avctx, AV_LOG_ERROR, "Scaled probabilities invalid\n");
186 return AVERROR_INVALIDDATA;
188 for (; i < 257; i++) {
189 rac->prob[i] = softfloat_mul(rac->prob[i], mul);
190 scaled_cumul_prob += rac->prob[i];
194 cumulative_target = 1 << scale_factor;
196 if (scaled_cumul_prob > cumulative_target) {
197 av_log(rac->avctx, AV_LOG_ERROR,
198 "Scaled probabilities are larger than target!\n");
202 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
204 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
209 /* Comment from reference source:
210 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
211 * // since the compression change is negligible and fixing it
212 * // breaks backwards compatibility
213 * b =- (signed int)b;
223 rac->scale = scale_factor;
225 /* Fill probability array with cumulative probability for each symbol. */
226 for (i = 1; i < 257; i++)
227 rac->prob[i] += rac->prob[i - 1];
232 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
233 uint8_t *diff, int w, int *left,
236 /* This is almost identical to add_hfyu_median_pred in huffyuvdsp.h.
237 * However the &0xFF on the gradient predictor yealds incorrect output
246 for (i = 0; i < w; i++) {
247 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
256 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
257 int width, int stride, int line)
262 /* Left prediction only for first line */
263 L = l->hdsp.add_hfyu_left_pred(buf, buf, width, 0);
265 /* Left pixel is actually prev_row[width] */
266 L = buf[width - stride - 1];
269 /* Second line, left predict first pixel, the rest of the line is median predicted
270 * NOTE: In the case of RGB this pixel is top predicted */
271 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
273 /* Top left is 2 rows back, last pixel */
274 TL = buf[width - (2 * stride) - 1];
277 add_lag_median_prediction(buf, buf - stride, buf,
282 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
283 int width, int stride, int line,
292 l->hdsp.add_hfyu_left_pred(buf, buf, width, 0);
298 const int HEAD = is_luma ? 4 : 2;
301 L = buf[width - stride - 1];
302 TL = buf[HEAD - stride - 1];
303 for (i = 0; i < HEAD; i++) {
307 for (; i < width; i++) {
308 L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
309 TL = buf[i - stride];
313 TL = buf[width - (2 * stride) - 1];
314 L = buf[width - stride - 1];
315 l->hdsp.add_hfyu_median_pred(buf, buf - stride, buf, width, &L, &TL);
319 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
320 uint8_t *dst, int width, int stride,
329 /* Output any zeros remaining from the previous run */
332 int count = FFMIN(l->zeros_rem, width - i);
333 memset(dst + i, 0, count);
335 l->zeros_rem -= count;
339 dst[i] = lag_get_rac(rac);
348 if (l->zeros == esc_count) {
349 int index = lag_get_rac(rac);
354 l->zeros_rem = lag_calc_zero_run(index);
361 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
362 const uint8_t *src, const uint8_t *src_end,
363 int width, int esc_count)
367 uint8_t zero_run = 0;
368 const uint8_t *src_start = src;
369 uint8_t mask1 = -(esc_count < 2);
370 uint8_t mask2 = -(esc_count < 3);
371 uint8_t *end = dst + (width - 2);
373 avpriv_request_sample(l->avctx, "zero_run_line");
375 memset(dst, 0, width);
379 count = FFMIN(l->zeros_rem, width - i);
380 if (end - dst < count) {
381 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
382 return AVERROR_INVALIDDATA;
385 memset(dst, 0, count);
386 l->zeros_rem -= count;
392 while (!zero_run && dst + i < end) {
394 if (i+2 >= src_end - src)
395 return AVERROR_INVALIDDATA;
397 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
404 l->zeros_rem = lag_calc_zero_run(src[i]);
414 return src - src_start;
419 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
420 int width, int height, int stride,
421 const uint8_t *src, int src_size)
430 const uint8_t *src_end = src + src_size;
433 rac.avctx = l->avctx;
437 return AVERROR_INVALIDDATA;
441 length = width * height;
443 return AVERROR_INVALIDDATA;
444 if (esc_count && AV_RL32(src + 1) < length) {
445 length = AV_RL32(src + 1);
449 if ((ret = init_get_bits8(&gb, src + offset, src_size - offset)) < 0)
452 if (lag_read_prob_header(&rac, &gb) < 0)
455 ff_lag_rac_init(&rac, &gb, length - stride);
457 for (i = 0; i < height; i++)
458 read += lag_decode_line(l, &rac, dst + (i * stride), width,
462 av_log(l->avctx, AV_LOG_WARNING,
463 "Output more bytes than length (%d of %"PRIu32")\n", read,
465 } else if (esc_count < 8) {
470 /* Zero run coding only, no range coding. */
471 for (i = 0; i < height; i++) {
472 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
473 src_end, width, esc_count);
479 if (src_size < width * height)
480 return AVERROR_INVALIDDATA; // buffer not big enough
481 /* Plane is stored uncompressed */
482 for (i = 0; i < height; i++) {
483 memcpy(dst + (i * stride), src, width);
487 } else if (esc_count == 0xff) {
488 /* Plane is a solid run of given value */
489 for (i = 0; i < height; i++)
490 memset(dst + i * stride, src[1], width);
491 /* Do not apply prediction.
492 Note: memset to 0 above, setting first value to src[1]
493 and applying prediction gives the same result. */
496 av_log(l->avctx, AV_LOG_ERROR,
497 "Invalid zero run escape code! (%#x)\n", esc_count);
501 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
502 for (i = 0; i < height; i++) {
503 lag_pred_line(l, dst, width, stride, i);
507 for (i = 0; i < height; i++) {
508 lag_pred_line_yuy2(l, dst, width, stride, i,
509 width == l->avctx->width);
519 * @param avctx codec context
520 * @param data output AVFrame
521 * @param data_size size of output data or 0 if no picture is returned
522 * @param avpkt input packet
523 * @return number of consumed bytes on success or negative if decode fails
525 static int lag_decode_frame(AVCodecContext *avctx,
526 void *data, int *got_frame, AVPacket *avpkt)
528 const uint8_t *buf = avpkt->data;
529 unsigned int buf_size = avpkt->size;
530 LagarithContext *l = avctx->priv_data;
531 ThreadFrame frame = { .f = data };
532 AVFrame *const p = data;
533 uint8_t frametype = 0;
534 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
536 uint8_t *srcs[4], *dst;
537 int i, j, planes = 3;
544 offset_gu = AV_RL32(buf + 1);
545 offset_bv = AV_RL32(buf + 5);
548 case FRAME_SOLID_RGBA:
549 avctx->pix_fmt = AV_PIX_FMT_RGB32;
550 case FRAME_SOLID_GRAY:
551 if (frametype == FRAME_SOLID_GRAY)
552 if (avctx->bits_per_coded_sample == 24) {
553 avctx->pix_fmt = AV_PIX_FMT_RGB24;
555 avctx->pix_fmt = AV_PIX_FMT_0RGB32;
559 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
563 if (frametype == FRAME_SOLID_RGBA) {
564 for (j = 0; j < avctx->height; j++) {
565 for (i = 0; i < avctx->width; i++)
566 AV_WN32(dst + i * 4, offset_gu);
567 dst += p->linesize[0];
570 for (j = 0; j < avctx->height; j++) {
571 memset(dst, buf[1], avctx->width * planes);
572 dst += p->linesize[0];
576 case FRAME_SOLID_COLOR:
577 if (avctx->bits_per_coded_sample == 24) {
578 avctx->pix_fmt = AV_PIX_FMT_RGB24;
580 avctx->pix_fmt = AV_PIX_FMT_RGB32;
581 offset_gu |= 0xFFU << 24;
584 if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
588 for (j = 0; j < avctx->height; j++) {
589 for (i = 0; i < avctx->width; i++)
590 if (avctx->bits_per_coded_sample == 24) {
591 AV_WB24(dst + i * 3, offset_gu);
593 AV_WN32(dst + i * 4, offset_gu);
595 dst += p->linesize[0];
598 case FRAME_ARITH_RGBA:
599 avctx->pix_fmt = AV_PIX_FMT_RGB32;
602 offs[3] = AV_RL32(buf + 9);
603 case FRAME_ARITH_RGB24:
605 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
606 avctx->pix_fmt = AV_PIX_FMT_RGB24;
608 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
615 l->rgb_stride = FFALIGN(avctx->width, 16);
616 av_fast_malloc(&l->rgb_planes, &l->rgb_planes_allocated,
617 l->rgb_stride * avctx->height * planes + 1);
618 if (!l->rgb_planes) {
619 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
620 return AVERROR(ENOMEM);
622 for (i = 0; i < planes; i++)
623 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
624 for (i = 0; i < planes; i++)
625 if (buf_size <= offs[i]) {
626 av_log(avctx, AV_LOG_ERROR,
627 "Invalid frame offsets\n");
628 return AVERROR_INVALIDDATA;
631 for (i = 0; i < planes; i++)
632 lag_decode_arith_plane(l, srcs[i],
633 avctx->width, avctx->height,
634 -l->rgb_stride, buf + offs[i],
637 for (i = 0; i < planes; i++)
638 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
639 for (j = 0; j < avctx->height; j++) {
640 for (i = 0; i < avctx->width; i++) {
647 if (frametype == FRAME_ARITH_RGBA) {
649 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
656 dst += p->linesize[0];
657 for (i = 0; i < planes; i++)
658 srcs[i] += l->rgb_stride;
661 case FRAME_ARITH_YUY2:
662 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
664 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
667 if (offset_ry >= buf_size ||
668 offset_gu >= buf_size ||
669 offset_bv >= buf_size) {
670 av_log(avctx, AV_LOG_ERROR,
671 "Invalid frame offsets\n");
672 return AVERROR_INVALIDDATA;
675 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
676 p->linesize[0], buf + offset_ry,
677 buf_size - offset_ry);
678 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
679 avctx->height, p->linesize[1],
680 buf + offset_gu, buf_size - offset_gu);
681 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
682 avctx->height, p->linesize[2],
683 buf + offset_bv, buf_size - offset_bv);
685 case FRAME_ARITH_YV12:
686 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
688 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
690 if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
691 return AVERROR_INVALIDDATA;
694 if (offset_ry >= buf_size ||
695 offset_gu >= buf_size ||
696 offset_bv >= buf_size) {
697 av_log(avctx, AV_LOG_ERROR,
698 "Invalid frame offsets\n");
699 return AVERROR_INVALIDDATA;
702 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
703 p->linesize[0], buf + offset_ry,
704 buf_size - offset_ry);
705 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
706 (avctx->height + 1) / 2, p->linesize[2],
707 buf + offset_gu, buf_size - offset_gu);
708 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
709 (avctx->height + 1) / 2, p->linesize[1],
710 buf + offset_bv, buf_size - offset_bv);
713 av_log(avctx, AV_LOG_ERROR,
714 "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
715 return AVERROR_PATCHWELCOME;
723 static av_cold int lag_decode_init(AVCodecContext *avctx)
725 LagarithContext *l = avctx->priv_data;
728 ff_huffyuvdsp_init(&l->hdsp);
733 static av_cold int lag_decode_end(AVCodecContext *avctx)
735 LagarithContext *l = avctx->priv_data;
737 av_freep(&l->rgb_planes);
742 AVCodec ff_lagarith_decoder = {
744 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
745 .type = AVMEDIA_TYPE_VIDEO,
746 .id = AV_CODEC_ID_LAGARITH,
747 .priv_data_size = sizeof(LagarithContext),
748 .init = lag_decode_init,
749 .close = lag_decode_end,
750 .decode = lag_decode_frame,
751 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,