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"
28 #include "libavutil/qsort.h"
31 #include "bytestream.h"
33 #include "huffyuvdsp.h"
35 #include "lossless_videodsp.h"
38 typedef struct Slice {
43 typedef enum Prediction {
49 typedef struct HuffEntry {
55 typedef struct MagicYUVContext {
61 int planes; // number of encoded planes in bitstream
62 int decorrelate; // postprocessing work
63 int color_matrix; // video color matrix
65 int interlaced; // video is interlaced
66 const uint8_t *buf; // pointer to AVPacket->data
69 Slice *slices[4]; // slice bitstream positions for each plane
70 unsigned int slices_size[4]; // slice sizes for each plane
71 VLC vlc[4]; // VLC for each plane
72 int (*magy_decode_slice)(AVCodecContext *avctx, void *tdata,
74 LLVidDSPContext llviddsp;
77 static int huff_cmp_len(const void *a, const void *b)
79 const HuffEntry *aa = a, *bb = b;
80 return (bb->len - aa->len) * 4096 + aa->sym - bb->sym;
83 static int huff_build(HuffEntry he[], VLC *vlc, int nb_elems)
87 AV_QSORT(he, nb_elems, HuffEntry, huff_cmp_len);
90 for (unsigned i = 0; i < nb_elems; i++) {
91 he[i].code = code >> (32 - he[i].len);
92 code += 0x80000000u >> (he[i].len - 1);
96 return ff_init_vlc_sparse(vlc, FFMIN(he[0].len, 12), nb_elems,
97 &he[0].len, sizeof(he[0]), sizeof(he[0].len),
98 &he[0].code, sizeof(he[0]), sizeof(he[0].code),
99 &he[0].sym, sizeof(he[0]), sizeof(he[0].sym), 0);
102 static void magicyuv_median_pred16(uint16_t *dst, const uint16_t *src1,
103 const uint16_t *diff, intptr_t w,
104 int *left, int *left_top, int max)
112 for (i = 0; i < w; i++) {
113 l = mid_pred(l, src1[i], (l + src1[i] - lt)) + diff[i];
123 static int magy_decode_slice10(AVCodecContext *avctx, void *tdata,
126 MagicYUVContext *s = avctx->priv_data;
127 int interlaced = s->interlaced;
128 const int bps = s->bps;
129 const int max = s->max - 1;
135 for (i = 0; i < s->planes; i++) {
136 int left, lefttop, top;
137 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
138 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
139 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
140 ptrdiff_t fake_stride = (p->linesize[i] / 2) * (1 + interlaced);
141 ptrdiff_t stride = p->linesize[i] / 2;
143 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
144 s->slices[i][j].size);
149 flags = get_bits(&gb, 8);
150 pred = get_bits(&gb, 8);
152 dst = (uint16_t *)p->data[i] + j * sheight * stride;
154 if (get_bits_left(&gb) < bps * width * height)
155 return AVERROR_INVALIDDATA;
156 for (k = 0; k < height; k++) {
157 for (x = 0; x < width; x++)
158 dst[x] = get_bits(&gb, bps);
163 for (k = 0; k < height; k++) {
164 for (x = 0; x < width; x++) {
166 if (get_bits_left(&gb) <= 0)
167 return AVERROR_INVALIDDATA;
169 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
171 return AVERROR_INVALIDDATA;
181 dst = (uint16_t *)p->data[i] + j * sheight * stride;
182 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
185 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
188 for (k = 1 + interlaced; k < height; k++) {
189 s->llviddsp.add_left_pred_int16(dst, dst, max, width, dst[-fake_stride]);
194 dst = (uint16_t *)p->data[i] + j * sheight * stride;
195 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
198 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
201 for (k = 1 + interlaced; k < height; k++) {
202 top = dst[-fake_stride];
205 for (x = 1; x < width; x++) {
206 top = dst[x - fake_stride];
207 lefttop = dst[x - (fake_stride + 1)];
208 left += top - lefttop + dst[x];
215 dst = (uint16_t *)p->data[i] + j * sheight * stride;
216 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
219 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
222 lefttop = left = dst[0];
223 for (k = 1 + interlaced; k < height; k++) {
224 magicyuv_median_pred16(dst, dst - fake_stride, dst, width, &left, &lefttop, max);
225 lefttop = left = dst[0];
230 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
234 if (s->decorrelate) {
235 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
236 int width = avctx->coded_width;
237 uint16_t *r = (uint16_t *)p->data[0] + j * s->slice_height * p->linesize[0] / 2;
238 uint16_t *g = (uint16_t *)p->data[1] + j * s->slice_height * p->linesize[1] / 2;
239 uint16_t *b = (uint16_t *)p->data[2] + j * s->slice_height * p->linesize[2] / 2;
241 for (i = 0; i < height; i++) {
242 for (k = 0; k < width; k++) {
243 b[k] = (b[k] + g[k]) & max;
244 r[k] = (r[k] + g[k]) & max;
246 b += p->linesize[0] / 2;
247 g += p->linesize[1] / 2;
248 r += p->linesize[2] / 2;
255 static int magy_decode_slice(AVCodecContext *avctx, void *tdata,
258 MagicYUVContext *s = avctx->priv_data;
259 int interlaced = s->interlaced;
261 int i, k, x, min_width;
265 for (i = 0; i < s->planes; i++) {
266 int left, lefttop, top;
267 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
268 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
269 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
270 ptrdiff_t fake_stride = p->linesize[i] * (1 + interlaced);
271 ptrdiff_t stride = p->linesize[i];
272 const uint8_t *slice = s->buf + s->slices[i][j].start;
275 flags = bytestream_get_byte(&slice);
276 pred = bytestream_get_byte(&slice);
278 dst = p->data[i] + j * sheight * stride;
280 if (s->slices[i][j].size - 2 < width * height)
281 return AVERROR_INVALIDDATA;
282 for (k = 0; k < height; k++) {
283 bytestream_get_buffer(&slice, dst, width);
287 int ret = init_get_bits8(&gb, slice, s->slices[i][j].size - 2);
292 for (k = 0; k < height; k++) {
293 for (x = 0; x < width; x++) {
295 if (get_bits_left(&gb) <= 0)
296 return AVERROR_INVALIDDATA;
298 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
300 return AVERROR_INVALIDDATA;
310 dst = p->data[i] + j * sheight * stride;
311 s->llviddsp.add_left_pred(dst, dst, width, 0);
314 s->llviddsp.add_left_pred(dst, dst, width, 0);
317 for (k = 1 + interlaced; k < height; k++) {
318 s->llviddsp.add_left_pred(dst, dst, width, dst[-fake_stride]);
323 dst = p->data[i] + j * sheight * stride;
324 s->llviddsp.add_left_pred(dst, dst, width, 0);
327 s->llviddsp.add_left_pred(dst, dst, width, 0);
330 min_width = FFMIN(width, 32);
331 for (k = 1 + interlaced; k < height; k++) {
332 top = dst[-fake_stride];
335 for (x = 1; x < min_width; x++) { /* dsp need aligned 32 */
336 top = dst[x - fake_stride];
337 lefttop = dst[x - (fake_stride + 1)];
338 left += top - lefttop + dst[x];
342 s->llviddsp.add_gradient_pred(dst + 32, fake_stride, width - 32);
347 dst = p->data[i] + j * sheight * stride;
348 s->llviddsp.add_left_pred(dst, dst, width, 0);
351 s->llviddsp.add_left_pred(dst, dst, width, 0);
354 lefttop = left = dst[0];
355 for (k = 1 + interlaced; k < height; k++) {
356 s->llviddsp.add_median_pred(dst, dst - fake_stride,
357 dst, width, &left, &lefttop);
358 lefttop = left = dst[0];
363 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
367 if (s->decorrelate) {
368 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
369 int width = avctx->coded_width;
370 uint8_t *b = p->data[0] + j * s->slice_height * p->linesize[0];
371 uint8_t *g = p->data[1] + j * s->slice_height * p->linesize[1];
372 uint8_t *r = p->data[2] + j * s->slice_height * p->linesize[2];
374 for (i = 0; i < height; i++) {
375 s->llviddsp.add_bytes(b, g, width);
376 s->llviddsp.add_bytes(r, g, width);
386 static int build_huffman(AVCodecContext *avctx, const uint8_t *table,
387 int table_size, int max)
389 MagicYUVContext *s = avctx->priv_data;
394 bytestream2_init(&gb, table, table_size);
396 while (bytestream2_get_bytes_left(&gb) > 0) {
397 int b = bytestream2_peek_byteu(&gb) & 0x80;
398 int x = bytestream2_get_byteu(&gb) & ~0x80;
402 if (bytestream2_get_bytes_left(&gb) <= 0)
404 l += bytestream2_get_byteu(&gb);
407 if (k > max || x == 0 || x > 32) {
408 av_log(avctx, AV_LOG_ERROR, "Invalid Huffman codes\n");
409 return AVERROR_INVALIDDATA;
419 if (huff_build(he, &s->vlc[i], max)) {
420 av_log(avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
421 return AVERROR_INVALIDDATA;
424 if (i == s->planes) {
430 if (i != s->planes) {
431 av_log(avctx, AV_LOG_ERROR, "Huffman tables too short\n");
432 return AVERROR_INVALIDDATA;
438 static int magy_decode_frame(AVCodecContext *avctx, void *data,
439 int *got_frame, AVPacket *avpkt)
441 MagicYUVContext *s = avctx->priv_data;
442 ThreadFrame frame = { .f = data };
445 uint32_t first_offset, offset, next_offset, header_size, slice_width;
446 int width, height, format, version, table_size;
449 if (avpkt->size < 36)
450 return AVERROR_INVALIDDATA;
452 bytestream2_init(&gb, avpkt->data, avpkt->size);
453 if (bytestream2_get_le32u(&gb) != MKTAG('M', 'A', 'G', 'Y'))
454 return AVERROR_INVALIDDATA;
456 header_size = bytestream2_get_le32u(&gb);
457 if (header_size < 32 || header_size >= avpkt->size) {
458 av_log(avctx, AV_LOG_ERROR,
459 "header or packet too small %"PRIu32"\n", header_size);
460 return AVERROR_INVALIDDATA;
463 version = bytestream2_get_byteu(&gb);
465 avpriv_request_sample(avctx, "Version %d", version);
466 return AVERROR_PATCHWELCOME;
476 format = bytestream2_get_byteu(&gb);
479 avctx->pix_fmt = AV_PIX_FMT_GBRP;
483 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
487 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
490 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
495 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
502 avctx->pix_fmt = AV_PIX_FMT_YUVA444P;
505 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
508 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
514 avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
518 avctx->pix_fmt = AV_PIX_FMT_GBRP10;
523 avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
528 avctx->pix_fmt = AV_PIX_FMT_GBRP12;
533 avctx->pix_fmt = AV_PIX_FMT_GBRAP12;
538 avctx->pix_fmt = AV_PIX_FMT_GRAY10;
542 avctx->pix_fmt = AV_PIX_FMT_YUV420P10;
550 avpriv_request_sample(avctx, "Format 0x%X", format);
551 return AVERROR_PATCHWELCOME;
553 s->max = 1 << s->bps;
554 s->magy_decode_slice = s->bps == 8 ? magy_decode_slice : magy_decode_slice10;
555 s->planes = av_pix_fmt_count_planes(avctx->pix_fmt);
557 bytestream2_skipu(&gb, 1);
558 s->color_matrix = bytestream2_get_byteu(&gb);
559 s->flags = bytestream2_get_byteu(&gb);
560 s->interlaced = !!(s->flags & 2);
561 bytestream2_skipu(&gb, 3);
563 width = bytestream2_get_le32u(&gb);
564 height = bytestream2_get_le32u(&gb);
565 ret = ff_set_dimensions(avctx, width, height);
569 slice_width = bytestream2_get_le32u(&gb);
570 if (slice_width != avctx->coded_width) {
571 avpriv_request_sample(avctx, "Slice width %"PRIu32, slice_width);
572 return AVERROR_PATCHWELCOME;
574 s->slice_height = bytestream2_get_le32u(&gb);
575 if (s->slice_height <= 0 || s->slice_height > INT_MAX - avctx->coded_height) {
576 av_log(avctx, AV_LOG_ERROR,
577 "invalid slice height: %d\n", s->slice_height);
578 return AVERROR_INVALIDDATA;
581 bytestream2_skipu(&gb, 4);
583 s->nb_slices = (avctx->coded_height + s->slice_height - 1) / s->slice_height;
584 if (s->nb_slices > INT_MAX / FFMAX(sizeof(Slice), 4 * 5)) {
585 av_log(avctx, AV_LOG_ERROR,
586 "invalid number of slices: %d\n", s->nb_slices);
587 return AVERROR_INVALIDDATA;
591 if ((s->slice_height >> s->vshift[1]) < 2) {
592 av_log(avctx, AV_LOG_ERROR, "impossible slice height\n");
593 return AVERROR_INVALIDDATA;
595 if ((avctx->coded_height % s->slice_height) && ((avctx->coded_height % s->slice_height) >> s->vshift[1]) < 2) {
596 av_log(avctx, AV_LOG_ERROR, "impossible height\n");
597 return AVERROR_INVALIDDATA;
601 if (bytestream2_get_bytes_left(&gb) <= s->nb_slices * s->planes * 5)
602 return AVERROR_INVALIDDATA;
603 for (i = 0; i < s->planes; i++) {
604 av_fast_malloc(&s->slices[i], &s->slices_size[i], s->nb_slices * sizeof(Slice));
606 return AVERROR(ENOMEM);
608 offset = bytestream2_get_le32u(&gb);
609 if (offset >= avpkt->size - header_size)
610 return AVERROR_INVALIDDATA;
613 first_offset = offset;
615 for (j = 0; j < s->nb_slices - 1; j++) {
616 s->slices[i][j].start = offset + header_size;
618 next_offset = bytestream2_get_le32u(&gb);
619 if (next_offset <= offset || next_offset >= avpkt->size - header_size)
620 return AVERROR_INVALIDDATA;
622 s->slices[i][j].size = next_offset - offset;
623 if (s->slices[i][j].size < 2)
624 return AVERROR_INVALIDDATA;
625 offset = next_offset;
628 s->slices[i][j].start = offset + header_size;
629 s->slices[i][j].size = avpkt->size - s->slices[i][j].start;
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,