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 */
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 += 1LL << av_log2(h >> 21);
99 static uint8_t lag_calc_zero_run(int8_t x)
101 return (x * 2) ^ (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 if (scale_factor >= 32U)
195 return AVERROR_INVALIDDATA;
196 cumulative_target = 1U << scale_factor;
198 if (scaled_cumul_prob > cumulative_target) {
199 av_log(rac->avctx, AV_LOG_ERROR,
200 "Scaled probabilities are larger than target!\n");
204 scaled_cumul_prob = cumulative_target - scaled_cumul_prob;
206 for (i = 1; scaled_cumul_prob; i = (i & 0x7f) + 1) {
211 /* Comment from reference source:
212 * if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
213 * // since the compression change is negligible and fixing it
214 * // breaks backwards compatibility
215 * b =- (signed int)b;
225 rac->scale = scale_factor;
227 /* Fill probability array with cumulative probability for each symbol. */
228 for (i = 1; i < 257; i++)
229 rac->prob[i] += rac->prob[i - 1];
234 static void add_lag_median_prediction(uint8_t *dst, uint8_t *src1,
235 uint8_t *diff, int w, int *left,
238 /* This is almost identical to add_hfyu_median_pred in huffyuvdsp.h.
239 * However the &0xFF on the gradient predictor yields incorrect output
248 for (i = 0; i < w; i++) {
249 l = mid_pred(l, src1[i], l + src1[i] - lt) + diff[i];
258 static void lag_pred_line(LagarithContext *l, uint8_t *buf,
259 int width, int stride, int line)
264 /* Left prediction only for first line */
265 L = l->llviddsp.add_left_pred(buf, buf, width, 0);
267 /* Left pixel is actually prev_row[width] */
268 L = buf[width - stride - 1];
271 /* Second line, left predict first pixel, the rest of the line is median predicted
272 * NOTE: In the case of RGB this pixel is top predicted */
273 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
275 /* Top left is 2 rows back, last pixel */
276 TL = buf[width - (2 * stride) - 1];
279 add_lag_median_prediction(buf, buf - stride, buf,
284 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
285 int width, int stride, int line,
294 l->llviddsp.add_left_pred(buf, buf, width, 0);
300 const int HEAD = is_luma ? 4 : 2;
303 L = buf[width - stride - 1];
304 TL = buf[HEAD - stride - 1];
305 for (i = 0; i < HEAD; i++) {
309 for (; i < width; i++) {
310 L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
311 TL = buf[i - stride];
315 TL = buf[width - (2 * stride) - 1];
316 L = buf[width - stride - 1];
317 l->llviddsp.add_median_pred(buf, buf - stride, buf, width, &L, &TL);
321 static int lag_decode_line(LagarithContext *l, lag_rac *rac,
322 uint8_t *dst, int width, int stride,
331 /* Output any zeros remaining from the previous run */
334 int count = FFMIN(l->zeros_rem, width - i);
335 memset(dst + i, 0, count);
337 l->zeros_rem -= count;
341 dst[i] = lag_get_rac(rac);
350 if (l->zeros == esc_count) {
351 int index = lag_get_rac(rac);
356 l->zeros_rem = lag_calc_zero_run(index);
363 static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
364 const uint8_t *src, const uint8_t *src_end,
365 int width, int esc_count)
369 uint8_t zero_run = 0;
370 const uint8_t *src_start = src;
371 uint8_t mask1 = -(esc_count < 2);
372 uint8_t mask2 = -(esc_count < 3);
373 uint8_t *end = dst + (width - 2);
375 avpriv_request_sample(l->avctx, "zero_run_line");
377 memset(dst, 0, width);
381 count = FFMIN(l->zeros_rem, width - i);
382 if (end - dst < count) {
383 av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
384 return AVERROR_INVALIDDATA;
387 memset(dst, 0, count);
388 l->zeros_rem -= count;
394 while (!zero_run && dst + i < end) {
396 if (i+2 >= src_end - src)
397 return AVERROR_INVALIDDATA;
399 !(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
406 l->zeros_rem = lag_calc_zero_run(src[i]);
416 return src - src_start;
421 static int lag_decode_arith_plane(LagarithContext *l, uint8_t *dst,
422 int width, int height, int stride,
423 const uint8_t *src, int src_size)
432 const uint8_t *src_end = src + src_size;
435 rac.avctx = l->avctx;
439 return AVERROR_INVALIDDATA;
443 length = width * height;
445 return AVERROR_INVALIDDATA;
446 if (esc_count && AV_RL32(src + 1) < length) {
447 length = AV_RL32(src + 1);
451 if ((ret = init_get_bits8(&gb, src + offset, src_size - offset)) < 0)
454 if (lag_read_prob_header(&rac, &gb) < 0)
457 ff_lag_rac_init(&rac, &gb, length - stride);
458 for (i = 0; i < height; i++) {
459 if (rac.overread > MAX_OVERREAD)
460 return AVERROR_INVALIDDATA;
461 read += lag_decode_line(l, &rac, dst + (i * stride), width,
466 av_log(l->avctx, AV_LOG_WARNING,
467 "Output more bytes than length (%d of %"PRIu32")\n", read,
469 } else if (esc_count < 8) {
474 /* Zero run coding only, no range coding. */
475 for (i = 0; i < height; i++) {
476 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
477 src_end, width, esc_count);
483 if (src_size < width * height)
484 return AVERROR_INVALIDDATA; // buffer not big enough
485 /* Plane is stored uncompressed */
486 for (i = 0; i < height; i++) {
487 memcpy(dst + (i * stride), src, width);
491 } else if (esc_count == 0xff) {
492 /* Plane is a solid run of given value */
493 for (i = 0; i < height; i++)
494 memset(dst + i * stride, src[1], width);
495 /* Do not apply prediction.
496 Note: memset to 0 above, setting first value to src[1]
497 and applying prediction gives the same result. */
500 av_log(l->avctx, AV_LOG_ERROR,
501 "Invalid zero run escape code! (%#x)\n", esc_count);
505 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
506 for (i = 0; i < height; i++) {
507 lag_pred_line(l, dst, width, stride, i);
511 for (i = 0; i < height; i++) {
512 lag_pred_line_yuy2(l, dst, width, stride, i,
513 width == l->avctx->width);
523 * @param avctx codec context
524 * @param data output AVFrame
525 * @param data_size size of output data or 0 if no picture is returned
526 * @param avpkt input packet
527 * @return number of consumed bytes on success or negative if decode fails
529 static int lag_decode_frame(AVCodecContext *avctx,
530 void *data, int *got_frame, AVPacket *avpkt)
532 const uint8_t *buf = avpkt->data;
533 unsigned int buf_size = avpkt->size;
534 LagarithContext *l = avctx->priv_data;
535 ThreadFrame frame = { .f = data };
536 AVFrame *const p = data;
537 uint8_t frametype = 0;
538 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
540 uint8_t *srcs[4], *dst;
541 int i, j, planes = 3;
548 offset_gu = AV_RL32(buf + 1);
549 offset_bv = AV_RL32(buf + 5);
552 case FRAME_SOLID_RGBA:
553 avctx->pix_fmt = AV_PIX_FMT_RGB32;
554 case FRAME_SOLID_GRAY:
555 if (frametype == FRAME_SOLID_GRAY)
556 if (avctx->bits_per_coded_sample == 24) {
557 avctx->pix_fmt = AV_PIX_FMT_RGB24;
559 avctx->pix_fmt = AV_PIX_FMT_0RGB32;
563 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
567 if (frametype == FRAME_SOLID_RGBA) {
568 for (j = 0; j < avctx->height; j++) {
569 for (i = 0; i < avctx->width; i++)
570 AV_WN32(dst + i * 4, offset_gu);
571 dst += p->linesize[0];
574 for (j = 0; j < avctx->height; j++) {
575 memset(dst, buf[1], avctx->width * planes);
576 dst += p->linesize[0];
580 case FRAME_SOLID_COLOR:
581 if (avctx->bits_per_coded_sample == 24) {
582 avctx->pix_fmt = AV_PIX_FMT_RGB24;
584 avctx->pix_fmt = AV_PIX_FMT_RGB32;
585 offset_gu |= 0xFFU << 24;
588 if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
592 for (j = 0; j < avctx->height; j++) {
593 for (i = 0; i < avctx->width; i++)
594 if (avctx->bits_per_coded_sample == 24) {
595 AV_WB24(dst + i * 3, offset_gu);
597 AV_WN32(dst + i * 4, offset_gu);
599 dst += p->linesize[0];
602 case FRAME_ARITH_RGBA:
603 avctx->pix_fmt = AV_PIX_FMT_RGB32;
606 offs[3] = AV_RL32(buf + 9);
607 case FRAME_ARITH_RGB24:
609 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
610 avctx->pix_fmt = AV_PIX_FMT_RGB24;
612 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
619 l->rgb_stride = FFALIGN(avctx->width, 16);
620 av_fast_malloc(&l->rgb_planes, &l->rgb_planes_allocated,
621 l->rgb_stride * avctx->height * planes + 1);
622 if (!l->rgb_planes) {
623 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
624 return AVERROR(ENOMEM);
626 for (i = 0; i < planes; i++)
627 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
628 for (i = 0; i < planes; i++)
629 if (buf_size <= offs[i]) {
630 av_log(avctx, AV_LOG_ERROR,
631 "Invalid frame offsets\n");
632 return AVERROR_INVALIDDATA;
635 for (i = 0; i < planes; i++)
636 lag_decode_arith_plane(l, srcs[i],
637 avctx->width, avctx->height,
638 -l->rgb_stride, buf + offs[i],
641 for (i = 0; i < planes; i++)
642 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
643 for (j = 0; j < avctx->height; j++) {
644 for (i = 0; i < avctx->width; i++) {
651 if (frametype == FRAME_ARITH_RGBA) {
653 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
660 dst += p->linesize[0];
661 for (i = 0; i < planes; i++)
662 srcs[i] += l->rgb_stride;
665 case FRAME_ARITH_YUY2:
666 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
668 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
671 if (offset_ry >= buf_size ||
672 offset_gu >= buf_size ||
673 offset_bv >= buf_size) {
674 av_log(avctx, AV_LOG_ERROR,
675 "Invalid frame offsets\n");
676 return AVERROR_INVALIDDATA;
679 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
680 p->linesize[0], buf + offset_ry,
681 buf_size - offset_ry);
682 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
683 avctx->height, p->linesize[1],
684 buf + offset_gu, buf_size - offset_gu);
685 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
686 avctx->height, p->linesize[2],
687 buf + offset_bv, buf_size - offset_bv);
689 case FRAME_ARITH_YV12:
690 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
692 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
694 if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
695 return AVERROR_INVALIDDATA;
698 if (offset_ry >= buf_size ||
699 offset_gu >= buf_size ||
700 offset_bv >= buf_size) {
701 av_log(avctx, AV_LOG_ERROR,
702 "Invalid frame offsets\n");
703 return AVERROR_INVALIDDATA;
706 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
707 p->linesize[0], buf + offset_ry,
708 buf_size - offset_ry);
709 lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
710 (avctx->height + 1) / 2, p->linesize[2],
711 buf + offset_gu, buf_size - offset_gu);
712 lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
713 (avctx->height + 1) / 2, p->linesize[1],
714 buf + offset_bv, buf_size - offset_bv);
717 av_log(avctx, AV_LOG_ERROR,
718 "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
719 return AVERROR_PATCHWELCOME;
727 static av_cold int lag_decode_init(AVCodecContext *avctx)
729 LagarithContext *l = avctx->priv_data;
732 ff_llviddsp_init(&l->llviddsp);
738 static av_cold int lag_decode_init_thread_copy(AVCodecContext *avctx)
740 LagarithContext *l = avctx->priv_data;
747 static av_cold int lag_decode_end(AVCodecContext *avctx)
749 LagarithContext *l = avctx->priv_data;
751 av_freep(&l->rgb_planes);
756 AVCodec ff_lagarith_decoder = {
758 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
759 .type = AVMEDIA_TYPE_VIDEO,
760 .id = AV_CODEC_ID_LAGARITH,
761 .priv_data_size = sizeof(LagarithContext),
762 .init = lag_decode_init,
763 .init_thread_copy = ONLY_IF_THREADS_ENABLED(lag_decode_init_thread_copy),
764 .close = lag_decode_end,
765 .decode = lag_decode_frame,
766 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,