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
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;
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_prediction in dsputil.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->dsp.add_hfyu_left_prediction(buf, buf,
266 /* Left pixel is actually prev_row[width] */
267 L = buf[width - stride - 1];
270 /* Second line, left predict first pixel, the rest of the line is median predicted
271 * NOTE: In the case of RGB this pixel is top predicted */
272 TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
274 /* Top left is 2 rows back, last pixel */
275 TL = buf[width - (2 * stride) - 1];
278 add_lag_median_prediction(buf, buf - stride, buf,
283 static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
284 int width, int stride, int line,
293 l->dsp.add_hfyu_left_prediction(buf, buf, width, 0);
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->dsp.add_hfyu_median_prediction(buf, buf - stride, buf, width,
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);
459 for (i = 0; i < height; i++)
460 read += lag_decode_line(l, &rac, dst + (i * stride), width,
464 av_log(l->avctx, AV_LOG_WARNING,
465 "Output more bytes than length (%d of %"PRIu32")\n", read,
467 } else if (esc_count < 8) {
472 /* Zero run coding only, no range coding. */
473 for (i = 0; i < height; i++) {
474 int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
475 src_end, width, esc_count);
481 if (src_size < width * height)
482 return AVERROR_INVALIDDATA; // buffer not big enough
483 /* Plane is stored uncompressed */
484 for (i = 0; i < height; i++) {
485 memcpy(dst + (i * stride), src, width);
489 } else if (esc_count == 0xff) {
490 /* Plane is a solid run of given value */
491 for (i = 0; i < height; i++)
492 memset(dst + i * stride, src[1], width);
493 /* Do not apply prediction.
494 Note: memset to 0 above, setting first value to src[1]
495 and applying prediction gives the same result. */
498 av_log(l->avctx, AV_LOG_ERROR,
499 "Invalid zero run escape code! (%#x)\n", esc_count);
503 if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
504 for (i = 0; i < height; i++) {
505 lag_pred_line(l, dst, width, stride, i);
509 for (i = 0; i < height; i++) {
510 lag_pred_line_yuy2(l, dst, width, stride, i,
511 width == l->avctx->width);
521 * @param avctx codec context
522 * @param data output AVFrame
523 * @param data_size size of output data or 0 if no picture is returned
524 * @param avpkt input packet
525 * @return number of consumed bytes on success or negative if decode fails
527 static int lag_decode_frame(AVCodecContext *avctx,
528 void *data, int *got_frame, AVPacket *avpkt)
530 const uint8_t *buf = avpkt->data;
531 unsigned int buf_size = avpkt->size;
532 LagarithContext *l = avctx->priv_data;
533 ThreadFrame frame = { .f = data };
534 AVFrame *const p = data;
535 uint8_t frametype = 0;
536 uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
538 uint8_t *srcs[4], *dst;
539 int i, j, planes = 3;
546 offset_gu = AV_RL32(buf + 1);
547 offset_bv = AV_RL32(buf + 5);
550 case FRAME_SOLID_RGBA:
551 avctx->pix_fmt = AV_PIX_FMT_RGB32;
552 case FRAME_SOLID_GRAY:
553 if (frametype == FRAME_SOLID_GRAY)
554 if (avctx->bits_per_coded_sample == 24) {
555 avctx->pix_fmt = AV_PIX_FMT_RGB24;
557 avctx->pix_fmt = AV_PIX_FMT_0RGB32;
561 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
565 if (frametype == FRAME_SOLID_RGBA) {
566 for (j = 0; j < avctx->height; j++) {
567 for (i = 0; i < avctx->width; i++)
568 AV_WN32(dst + i * 4, offset_gu);
569 dst += p->linesize[0];
572 for (j = 0; j < avctx->height; j++) {
573 memset(dst, buf[1], avctx->width * planes);
574 dst += p->linesize[0];
578 case FRAME_SOLID_COLOR:
579 if (avctx->bits_per_coded_sample == 24) {
580 avctx->pix_fmt = AV_PIX_FMT_RGB24;
582 avctx->pix_fmt = AV_PIX_FMT_RGB32;
583 offset_gu |= 0xFFU << 24;
586 if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
590 for (j = 0; j < avctx->height; j++) {
591 for (i = 0; i < avctx->width; i++)
592 if (avctx->bits_per_coded_sample == 24) {
593 AV_WB24(dst + i * 3, offset_gu);
595 AV_WN32(dst + i * 4, offset_gu);
597 dst += p->linesize[0];
600 case FRAME_ARITH_RGBA:
601 avctx->pix_fmt = AV_PIX_FMT_RGB32;
604 offs[3] = AV_RL32(buf + 9);
605 case FRAME_ARITH_RGB24:
607 if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
608 avctx->pix_fmt = AV_PIX_FMT_RGB24;
610 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
617 l->rgb_stride = FFALIGN(avctx->width, 16);
618 av_fast_malloc(&l->rgb_planes, &l->rgb_planes_allocated,
619 l->rgb_stride * avctx->height * planes + 1);
620 if (!l->rgb_planes) {
621 av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
622 return AVERROR(ENOMEM);
624 for (i = 0; i < planes; i++)
625 srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
626 for (i = 0; i < planes; i++)
627 if (buf_size <= offs[i]) {
628 av_log(avctx, AV_LOG_ERROR,
629 "Invalid frame offsets\n");
630 return AVERROR_INVALIDDATA;
633 for (i = 0; i < planes; i++)
634 lag_decode_arith_plane(l, srcs[i],
635 avctx->width, avctx->height,
636 -l->rgb_stride, buf + offs[i],
639 for (i = 0; i < planes; i++)
640 srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
641 for (j = 0; j < avctx->height; j++) {
642 for (i = 0; i < avctx->width; i++) {
649 if (frametype == FRAME_ARITH_RGBA) {
651 AV_WN32(dst + i * 4, MKBETAG(a, r, g, b));
658 dst += p->linesize[0];
659 for (i = 0; i < planes; i++)
660 srcs[i] += l->rgb_stride;
663 case FRAME_ARITH_YUY2:
664 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
666 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
669 if (offset_ry >= buf_size ||
670 offset_gu >= buf_size ||
671 offset_bv >= buf_size) {
672 av_log(avctx, AV_LOG_ERROR,
673 "Invalid frame offsets\n");
674 return AVERROR_INVALIDDATA;
677 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
678 p->linesize[0], buf + offset_ry,
679 buf_size - offset_ry);
680 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
681 avctx->height, p->linesize[1],
682 buf + offset_gu, buf_size - offset_gu);
683 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
684 avctx->height, p->linesize[2],
685 buf + offset_bv, buf_size - offset_bv);
687 case FRAME_ARITH_YV12:
688 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
690 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
692 if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
693 return AVERROR_INVALIDDATA;
696 if (offset_ry >= buf_size ||
697 offset_gu >= buf_size ||
698 offset_bv >= buf_size) {
699 av_log(avctx, AV_LOG_ERROR,
700 "Invalid frame offsets\n");
701 return AVERROR_INVALIDDATA;
704 lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
705 p->linesize[0], buf + offset_ry,
706 buf_size - offset_ry);
707 lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
708 avctx->height / 2, p->linesize[2],
709 buf + offset_gu, buf_size - offset_gu);
710 lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
711 avctx->height / 2, p->linesize[1],
712 buf + offset_bv, buf_size - offset_bv);
715 av_log(avctx, AV_LOG_ERROR,
716 "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
717 return AVERROR_PATCHWELCOME;
725 static av_cold int lag_decode_init(AVCodecContext *avctx)
727 LagarithContext *l = avctx->priv_data;
730 ff_dsputil_init(&l->dsp, avctx);
735 static av_cold int lag_decode_end(AVCodecContext *avctx)
737 LagarithContext *l = avctx->priv_data;
739 av_freep(&l->rgb_planes);
744 AVCodec ff_lagarith_decoder = {
746 .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
747 .type = AVMEDIA_TYPE_VIDEO,
748 .id = AV_CODEC_ID_LAGARITH,
749 .priv_data_size = sizeof(LagarithContext),
750 .init = lag_decode_init,
751 .close = lag_decode_end,
752 .decode = lag_decode_frame,
753 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,