3 * Copyright (c) 2016 Paul B Mahol
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
25 #define CACHED_BITSTREAM_READER !ARCH_X86_32
27 #include "libavutil/pixdesc.h"
30 #include "bytestream.h"
32 #include "huffyuvdsp.h"
34 #include "lossless_videodsp.h"
37 typedef struct Slice {
42 typedef enum Prediction {
48 typedef struct HuffEntry {
53 typedef struct MagicYUVContext {
59 int planes; // number of encoded planes in bitstream
60 int decorrelate; // postprocessing work
61 int color_matrix; // video color matrix
63 int interlaced; // video is interlaced
64 const uint8_t *buf; // pointer to AVPacket->data
67 Slice *slices[4]; // slice bitstream positions for each plane
68 unsigned int slices_size[4]; // slice sizes for each plane
69 VLC vlc[4]; // VLC for each plane
70 int (*magy_decode_slice)(AVCodecContext *avctx, void *tdata,
72 LLVidDSPContext llviddsp;
75 static int huff_build(HuffEntry he[], uint16_t codes_count[33],
76 VLC *vlc, int nb_elems)
78 unsigned nb_codes = 0, max = 0;
80 for (int i = 32; i > 0; i--) {
81 uint16_t curr = codes_count[i]; // # of leafs of length i
82 codes_count[i] = nb_codes / 2; // # of non-leaf nodes on level i
83 nb_codes = codes_count[i] + curr; // # of nodes on level i
88 for (unsigned i = 0; i < nb_elems; i++) {
89 he[i].code = codes_count[he[i].len];
90 codes_count[he[i].len]++;
93 return init_vlc(vlc, FFMIN(max, 12), nb_elems,
94 &he[0].len, sizeof(he[0]), sizeof(he[0].len),
95 &he[0].code, sizeof(he[0]), sizeof(he[0].code), 0);
98 static void magicyuv_median_pred16(uint16_t *dst, const uint16_t *src1,
99 const uint16_t *diff, intptr_t w,
100 int *left, int *left_top, int max)
108 for (i = 0; i < w; i++) {
109 l = mid_pred(l, src1[i], (l + src1[i] - lt)) + diff[i];
119 static int magy_decode_slice10(AVCodecContext *avctx, void *tdata,
122 MagicYUVContext *s = avctx->priv_data;
123 int interlaced = s->interlaced;
124 const int bps = s->bps;
125 const int max = s->max - 1;
131 for (i = 0; i < s->planes; i++) {
132 int left, lefttop, top;
133 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
134 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
135 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
136 ptrdiff_t fake_stride = (p->linesize[i] / 2) * (1 + interlaced);
137 ptrdiff_t stride = p->linesize[i] / 2;
139 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
140 s->slices[i][j].size);
145 flags = get_bits(&gb, 8);
146 pred = get_bits(&gb, 8);
148 dst = (uint16_t *)p->data[i] + j * sheight * stride;
150 if (get_bits_left(&gb) < bps * width * height)
151 return AVERROR_INVALIDDATA;
152 for (k = 0; k < height; k++) {
153 for (x = 0; x < width; x++)
154 dst[x] = get_bits(&gb, bps);
159 for (k = 0; k < height; k++) {
160 for (x = 0; x < width; x++) {
162 if (get_bits_left(&gb) <= 0)
163 return AVERROR_INVALIDDATA;
165 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
167 return AVERROR_INVALIDDATA;
177 dst = (uint16_t *)p->data[i] + j * sheight * stride;
178 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
181 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
184 for (k = 1 + interlaced; k < height; k++) {
185 s->llviddsp.add_left_pred_int16(dst, dst, max, width, dst[-fake_stride]);
190 dst = (uint16_t *)p->data[i] + j * sheight * stride;
191 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
194 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
197 for (k = 1 + interlaced; k < height; k++) {
198 top = dst[-fake_stride];
201 for (x = 1; x < width; x++) {
202 top = dst[x - fake_stride];
203 lefttop = dst[x - (fake_stride + 1)];
204 left += top - lefttop + dst[x];
211 dst = (uint16_t *)p->data[i] + j * sheight * stride;
212 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
215 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
218 lefttop = left = dst[0];
219 for (k = 1 + interlaced; k < height; k++) {
220 magicyuv_median_pred16(dst, dst - fake_stride, dst, width, &left, &lefttop, max);
221 lefttop = left = dst[0];
226 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
230 if (s->decorrelate) {
231 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
232 int width = avctx->coded_width;
233 uint16_t *r = (uint16_t *)p->data[0] + j * s->slice_height * p->linesize[0] / 2;
234 uint16_t *g = (uint16_t *)p->data[1] + j * s->slice_height * p->linesize[1] / 2;
235 uint16_t *b = (uint16_t *)p->data[2] + j * s->slice_height * p->linesize[2] / 2;
237 for (i = 0; i < height; i++) {
238 for (k = 0; k < width; k++) {
239 b[k] = (b[k] + g[k]) & max;
240 r[k] = (r[k] + g[k]) & max;
242 b += p->linesize[0] / 2;
243 g += p->linesize[1] / 2;
244 r += p->linesize[2] / 2;
251 static int magy_decode_slice(AVCodecContext *avctx, void *tdata,
254 MagicYUVContext *s = avctx->priv_data;
255 int interlaced = s->interlaced;
257 int i, k, x, min_width;
261 for (i = 0; i < s->planes; i++) {
262 int left, lefttop, top;
263 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
264 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
265 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
266 ptrdiff_t fake_stride = p->linesize[i] * (1 + interlaced);
267 ptrdiff_t stride = p->linesize[i];
268 const uint8_t *slice = s->buf + s->slices[i][j].start;
271 flags = bytestream_get_byte(&slice);
272 pred = bytestream_get_byte(&slice);
274 dst = p->data[i] + j * sheight * stride;
276 if (s->slices[i][j].size - 2 < width * height)
277 return AVERROR_INVALIDDATA;
278 for (k = 0; k < height; k++) {
279 bytestream_get_buffer(&slice, dst, width);
283 int ret = init_get_bits8(&gb, slice, s->slices[i][j].size - 2);
288 for (k = 0; k < height; k++) {
289 for (x = 0; x < width; x++) {
291 if (get_bits_left(&gb) <= 0)
292 return AVERROR_INVALIDDATA;
294 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
296 return AVERROR_INVALIDDATA;
306 dst = p->data[i] + j * sheight * stride;
307 s->llviddsp.add_left_pred(dst, dst, width, 0);
310 s->llviddsp.add_left_pred(dst, dst, width, 0);
313 for (k = 1 + interlaced; k < height; k++) {
314 s->llviddsp.add_left_pred(dst, dst, width, dst[-fake_stride]);
319 dst = p->data[i] + j * sheight * stride;
320 s->llviddsp.add_left_pred(dst, dst, width, 0);
323 s->llviddsp.add_left_pred(dst, dst, width, 0);
326 min_width = FFMIN(width, 32);
327 for (k = 1 + interlaced; k < height; k++) {
328 top = dst[-fake_stride];
331 for (x = 1; x < min_width; x++) { /* dsp need aligned 32 */
332 top = dst[x - fake_stride];
333 lefttop = dst[x - (fake_stride + 1)];
334 left += top - lefttop + dst[x];
338 s->llviddsp.add_gradient_pred(dst + 32, fake_stride, width - 32);
343 dst = p->data[i] + j * sheight * stride;
344 s->llviddsp.add_left_pred(dst, dst, width, 0);
347 s->llviddsp.add_left_pred(dst, dst, width, 0);
350 lefttop = left = dst[0];
351 for (k = 1 + interlaced; k < height; k++) {
352 s->llviddsp.add_median_pred(dst, dst - fake_stride,
353 dst, width, &left, &lefttop);
354 lefttop = left = dst[0];
359 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
363 if (s->decorrelate) {
364 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
365 int width = avctx->coded_width;
366 uint8_t *b = p->data[0] + j * s->slice_height * p->linesize[0];
367 uint8_t *g = p->data[1] + j * s->slice_height * p->linesize[1];
368 uint8_t *r = p->data[2] + j * s->slice_height * p->linesize[2];
370 for (i = 0; i < height; i++) {
371 s->llviddsp.add_bytes(b, g, width);
372 s->llviddsp.add_bytes(r, g, width);
382 static int build_huffman(AVCodecContext *avctx, const uint8_t *table,
383 int table_size, int max)
385 MagicYUVContext *s = avctx->priv_data;
388 uint16_t length_count[33] = { 0 };
391 bytestream2_init(&gb, table, table_size);
393 while (bytestream2_get_bytes_left(&gb) > 0) {
394 int b = bytestream2_peek_byteu(&gb) & 0x80;
395 int x = bytestream2_get_byteu(&gb) & ~0x80;
399 if (bytestream2_get_bytes_left(&gb) <= 0)
401 l += bytestream2_get_byteu(&gb);
404 if (k > max || x == 0 || x > 32) {
405 av_log(avctx, AV_LOG_ERROR, "Invalid Huffman codes\n");
406 return AVERROR_INVALIDDATA;
409 length_count[x] += l;
415 if (huff_build(he, length_count, &s->vlc[i], max)) {
416 av_log(avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
417 return AVERROR_INVALIDDATA;
420 if (i == s->planes) {
423 memset(length_count, 0, sizeof(length_count));
427 if (i != s->planes) {
428 av_log(avctx, AV_LOG_ERROR, "Huffman tables too short\n");
429 return AVERROR_INVALIDDATA;
435 static int magy_decode_frame(AVCodecContext *avctx, void *data,
436 int *got_frame, AVPacket *avpkt)
438 MagicYUVContext *s = avctx->priv_data;
439 ThreadFrame frame = { .f = data };
442 uint32_t first_offset, offset, next_offset, header_size, slice_width;
443 int width, height, format, version, table_size;
446 if (avpkt->size < 36)
447 return AVERROR_INVALIDDATA;
449 bytestream2_init(&gb, avpkt->data, avpkt->size);
450 if (bytestream2_get_le32u(&gb) != MKTAG('M', 'A', 'G', 'Y'))
451 return AVERROR_INVALIDDATA;
453 header_size = bytestream2_get_le32u(&gb);
454 if (header_size < 32 || header_size >= avpkt->size) {
455 av_log(avctx, AV_LOG_ERROR,
456 "header or packet too small %"PRIu32"\n", header_size);
457 return AVERROR_INVALIDDATA;
460 version = bytestream2_get_byteu(&gb);
462 avpriv_request_sample(avctx, "Version %d", version);
463 return AVERROR_PATCHWELCOME;
473 format = bytestream2_get_byteu(&gb);
476 avctx->pix_fmt = AV_PIX_FMT_GBRP;
480 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
484 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
487 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
492 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
499 avctx->pix_fmt = AV_PIX_FMT_YUVA444P;
502 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
505 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
511 avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
515 avctx->pix_fmt = AV_PIX_FMT_GBRP10;
520 avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
525 avctx->pix_fmt = AV_PIX_FMT_GBRP12;
530 avctx->pix_fmt = AV_PIX_FMT_GBRAP12;
535 avctx->pix_fmt = AV_PIX_FMT_GRAY10;
539 avctx->pix_fmt = AV_PIX_FMT_YUV420P10;
547 avpriv_request_sample(avctx, "Format 0x%X", format);
548 return AVERROR_PATCHWELCOME;
550 s->max = 1 << s->bps;
551 s->magy_decode_slice = s->bps == 8 ? magy_decode_slice : magy_decode_slice10;
552 s->planes = av_pix_fmt_count_planes(avctx->pix_fmt);
554 bytestream2_skipu(&gb, 1);
555 s->color_matrix = bytestream2_get_byteu(&gb);
556 s->flags = bytestream2_get_byteu(&gb);
557 s->interlaced = !!(s->flags & 2);
558 bytestream2_skipu(&gb, 3);
560 width = bytestream2_get_le32u(&gb);
561 height = bytestream2_get_le32u(&gb);
562 ret = ff_set_dimensions(avctx, width, height);
566 slice_width = bytestream2_get_le32u(&gb);
567 if (slice_width != avctx->coded_width) {
568 avpriv_request_sample(avctx, "Slice width %"PRIu32, slice_width);
569 return AVERROR_PATCHWELCOME;
571 s->slice_height = bytestream2_get_le32u(&gb);
572 if (s->slice_height <= 0 || s->slice_height > INT_MAX - avctx->coded_height) {
573 av_log(avctx, AV_LOG_ERROR,
574 "invalid slice height: %d\n", s->slice_height);
575 return AVERROR_INVALIDDATA;
578 bytestream2_skipu(&gb, 4);
580 s->nb_slices = (avctx->coded_height + s->slice_height - 1) / s->slice_height;
581 if (s->nb_slices > INT_MAX / FFMAX(sizeof(Slice), 4 * 5)) {
582 av_log(avctx, AV_LOG_ERROR,
583 "invalid number of slices: %d\n", s->nb_slices);
584 return AVERROR_INVALIDDATA;
588 if ((s->slice_height >> s->vshift[1]) < 2) {
589 av_log(avctx, AV_LOG_ERROR, "impossible slice height\n");
590 return AVERROR_INVALIDDATA;
592 if ((avctx->coded_height % s->slice_height) && ((avctx->coded_height % s->slice_height) >> s->vshift[1]) < 2) {
593 av_log(avctx, AV_LOG_ERROR, "impossible height\n");
594 return AVERROR_INVALIDDATA;
598 if (bytestream2_get_bytes_left(&gb) <= s->nb_slices * s->planes * 5)
599 return AVERROR_INVALIDDATA;
600 for (i = 0; i < s->planes; i++) {
601 av_fast_malloc(&s->slices[i], &s->slices_size[i], s->nb_slices * sizeof(Slice));
603 return AVERROR(ENOMEM);
605 offset = bytestream2_get_le32u(&gb);
606 if (offset >= avpkt->size - header_size)
607 return AVERROR_INVALIDDATA;
610 first_offset = offset;
612 for (j = 0; j < s->nb_slices - 1; j++) {
613 s->slices[i][j].start = offset + header_size;
615 next_offset = bytestream2_get_le32u(&gb);
616 if (next_offset <= offset || next_offset >= avpkt->size - header_size)
617 return AVERROR_INVALIDDATA;
619 s->slices[i][j].size = next_offset - offset;
620 if (s->slices[i][j].size < 2)
621 return AVERROR_INVALIDDATA;
622 offset = next_offset;
625 s->slices[i][j].start = offset + header_size;
626 s->slices[i][j].size = avpkt->size - s->slices[i][j].start;
628 if (s->slices[i][j].size < 2)
629 return AVERROR_INVALIDDATA;
632 if (bytestream2_get_byteu(&gb) != s->planes)
633 return AVERROR_INVALIDDATA;
635 bytestream2_skipu(&gb, s->nb_slices * s->planes);
637 table_size = header_size + first_offset - bytestream2_tell(&gb);
639 return AVERROR_INVALIDDATA;
641 ret = build_huffman(avctx, avpkt->data + bytestream2_tell(&gb),
646 p->pict_type = AV_PICTURE_TYPE_I;
649 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
652 s->buf = avpkt->data;
654 avctx->execute2(avctx, s->magy_decode_slice, NULL, NULL, s->nb_slices);
656 if (avctx->pix_fmt == AV_PIX_FMT_GBRP ||
657 avctx->pix_fmt == AV_PIX_FMT_GBRAP ||
658 avctx->pix_fmt == AV_PIX_FMT_GBRP10 ||
659 avctx->pix_fmt == AV_PIX_FMT_GBRAP10||
660 avctx->pix_fmt == AV_PIX_FMT_GBRAP12||
661 avctx->pix_fmt == AV_PIX_FMT_GBRP12) {
662 FFSWAP(uint8_t*, p->data[0], p->data[1]);
663 FFSWAP(int, p->linesize[0], p->linesize[1]);
665 switch (s->color_matrix) {
667 p->colorspace = AVCOL_SPC_BT470BG;
670 p->colorspace = AVCOL_SPC_BT709;
673 p->color_range = (s->flags & 4) ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
681 static av_cold int magy_decode_init(AVCodecContext *avctx)
683 MagicYUVContext *s = avctx->priv_data;
684 ff_llviddsp_init(&s->llviddsp);
688 static av_cold int magy_decode_end(AVCodecContext *avctx)
690 MagicYUVContext * const s = avctx->priv_data;
693 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
694 av_freep(&s->slices[i]);
695 s->slices_size[i] = 0;
696 ff_free_vlc(&s->vlc[i]);
702 AVCodec ff_magicyuv_decoder = {
704 .long_name = NULL_IF_CONFIG_SMALL("MagicYUV video"),
705 .type = AVMEDIA_TYPE_VIDEO,
706 .id = AV_CODEC_ID_MAGICYUV,
707 .priv_data_size = sizeof(MagicYUVContext),
708 .init = magy_decode_init,
709 .close = magy_decode_end,
710 .decode = magy_decode_frame,
711 .capabilities = AV_CODEC_CAP_DR1 |
712 AV_CODEC_CAP_FRAME_THREADS |
713 AV_CODEC_CAP_SLICE_THREADS,
714 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,