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 #include "libavutil/pixdesc.h"
26 #include "libavutil/qsort.h"
29 #include "bytestream.h"
31 #include "huffyuvdsp.h"
33 #include "lossless_videodsp.h"
36 typedef struct Slice {
41 typedef enum Prediction {
47 typedef struct HuffEntry {
53 typedef struct MagicYUVContext {
58 int planes; // number of encoded planes in bitstream
59 int decorrelate; // postprocessing work
60 int color_matrix; // video color matrix
62 int interlaced; // video is interlaced
63 uint8_t *buf; // pointer to AVPacket->data
66 Slice *slices[4]; // slice bitstream positions for each plane
67 unsigned int slices_size[4]; // slice sizes for each plane
68 uint8_t len[4][1024]; // table of code lengths for each plane
69 VLC vlc[4]; // VLC for each plane
70 int (*huff_build)(VLC *vlc, uint8_t *len);
71 int (*magy_decode_slice)(AVCodecContext *avctx, void *tdata,
73 LLVidDSPContext llviddsp;
76 static int huff_cmp_len(const void *a, const void *b)
78 const HuffEntry *aa = a, *bb = b;
79 return (aa->len - bb->len) * 256 + aa->sym - bb->sym;
82 static int huff_cmp_len10(const void *a, const void *b)
84 const HuffEntry *aa = a, *bb = b;
85 return (aa->len - bb->len) * 1024 + aa->sym - bb->sym;
88 static int huff_build10(VLC *vlc, uint8_t *len)
97 for (i = 0; i < 1024; i++) {
101 return AVERROR_INVALIDDATA;
103 AV_QSORT(he, 1024, HuffEntry, huff_cmp_len10);
106 for (i = 1023; i >= 0; i--) {
107 codes[i] = code >> (32 - he[i].len);
110 code += 0x80000000u >> (he[i].len - 1);
114 return ff_init_vlc_sparse(vlc, FFMIN(he[1023].len, 12), 1024,
115 bits, sizeof(*bits), sizeof(*bits),
116 codes, sizeof(*codes), sizeof(*codes),
117 syms, sizeof(*syms), sizeof(*syms), 0);
120 static int huff_build(VLC *vlc, uint8_t *len)
129 for (i = 0; i < 256; i++) {
133 return AVERROR_INVALIDDATA;
135 AV_QSORT(he, 256, HuffEntry, huff_cmp_len);
138 for (i = 255; i >= 0; i--) {
139 codes[i] = code >> (32 - he[i].len);
142 code += 0x80000000u >> (he[i].len - 1);
146 return ff_init_vlc_sparse(vlc, FFMIN(he[255].len, 12), 256,
147 bits, sizeof(*bits), sizeof(*bits),
148 codes, sizeof(*codes), sizeof(*codes),
149 syms, sizeof(*syms), sizeof(*syms), 0);
152 static void magicyuv_median_pred10(uint16_t *dst, const uint16_t *src1,
153 const uint16_t *diff, intptr_t w,
154 int *left, int *left_top)
162 for (i = 0; i < w; i++) {
163 l = mid_pred(l, src1[i], (l + src1[i] - lt)) + diff[i];
173 static int magy_decode_slice10(AVCodecContext *avctx, void *tdata,
176 MagicYUVContext *s = avctx->priv_data;
177 int interlaced = s->interlaced;
183 for (i = 0; i < s->planes; i++) {
184 int left, lefttop, top;
185 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
186 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
187 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
188 ptrdiff_t fake_stride = (p->linesize[i] / 2) * (1 + interlaced);
189 ptrdiff_t stride = p->linesize[i] / 2;
191 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
192 s->slices[i][j].size);
197 flags = get_bits(&gb, 8);
198 pred = get_bits(&gb, 8);
200 dst = (uint16_t *)p->data[i] + j * sheight * stride;
202 for (k = 0; k < height; k++) {
203 for (x = 0; x < width; x++)
204 dst[x] = get_bits(&gb, 10);
209 for (k = 0; k < height; k++) {
210 for (x = 0; x < width; x++) {
212 if (get_bits_left(&gb) <= 0)
213 return AVERROR_INVALIDDATA;
215 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
217 return AVERROR_INVALIDDATA;
227 dst = (uint16_t *)p->data[i] + j * sheight * stride;
228 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
231 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
234 for (k = 1 + interlaced; k < height; k++) {
235 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, dst[-fake_stride]);
240 dst = (uint16_t *)p->data[i] + j * sheight * stride;
241 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
245 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
249 for (k = 1 + interlaced; k < height; k++) {
250 top = dst[-fake_stride];
252 dst[0] = left & 0x3FF;
253 for (x = 1; x < width; x++) {
254 top = dst[x - fake_stride];
255 lefttop = dst[x - (fake_stride + 1)];
256 left += top - lefttop + dst[x];
257 dst[x] = left & 0x3FF;
263 dst = (uint16_t *)p->data[i] + j * sheight * stride;
264 lefttop = left = dst[0];
265 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
268 lefttop = left = dst[0];
269 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
272 for (k = 1 + interlaced; k < height; k++) {
273 magicyuv_median_pred10(dst, dst - fake_stride, dst, width, &left, &lefttop);
274 lefttop = left = dst[0];
279 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
283 if (s->decorrelate) {
284 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
285 int width = avctx->coded_width;
286 uint16_t *r = (uint16_t *)p->data[0] + j * s->slice_height * p->linesize[0] / 2;
287 uint16_t *g = (uint16_t *)p->data[1] + j * s->slice_height * p->linesize[1] / 2;
288 uint16_t *b = (uint16_t *)p->data[2] + j * s->slice_height * p->linesize[2] / 2;
290 for (i = 0; i < height; i++) {
291 for (k = 0; k < width; k++) {
292 b[k] = (b[k] + g[k]) & 0x3FF;
293 r[k] = (r[k] + g[k]) & 0x3FF;
295 b += p->linesize[0] / 2;
296 g += p->linesize[1] / 2;
297 r += p->linesize[2] / 2;
304 static int magy_decode_slice(AVCodecContext *avctx, void *tdata,
307 MagicYUVContext *s = avctx->priv_data;
308 int interlaced = s->interlaced;
314 for (i = 0; i < s->planes; i++) {
315 int left, lefttop, top;
316 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
317 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
318 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
319 ptrdiff_t fake_stride = p->linesize[i] * (1 + interlaced);
320 ptrdiff_t stride = p->linesize[i];
322 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
323 s->slices[i][j].size);
328 flags = get_bits(&gb, 8);
329 pred = get_bits(&gb, 8);
331 dst = p->data[i] + j * sheight * stride;
333 for (k = 0; k < height; k++) {
334 for (x = 0; x < width; x++)
335 dst[x] = get_bits(&gb, 8);
340 for (k = 0; k < height; k++) {
341 for (x = 0; x < width; x++) {
343 if (get_bits_left(&gb) <= 0)
344 return AVERROR_INVALIDDATA;
346 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
348 return AVERROR_INVALIDDATA;
358 dst = p->data[i] + j * sheight * stride;
359 s->llviddsp.add_left_pred(dst, dst, width, 0);
362 s->llviddsp.add_left_pred(dst, dst, width, 0);
365 for (k = 1 + interlaced; k < height; k++) {
366 s->llviddsp.add_left_pred(dst, dst, width, dst[-fake_stride]);
371 dst = p->data[i] + j * sheight * stride;
372 s->llviddsp.add_left_pred(dst, dst, width, 0);
376 s->llviddsp.add_left_pred(dst, dst, width, 0);
380 for (k = 1 + interlaced; k < height; k++) {
381 top = dst[-fake_stride];
384 for (x = 1; x < width; x++) {
385 top = dst[x - fake_stride];
386 lefttop = dst[x - (fake_stride + 1)];
387 left += top - lefttop + dst[x];
394 dst = p->data[i] + j * sheight * stride;
395 lefttop = left = dst[0];
396 s->llviddsp.add_left_pred(dst, dst, width, 0);
399 lefttop = left = dst[0];
400 s->llviddsp.add_left_pred(dst, dst, width, 0);
403 for (k = 1 + interlaced; k < height; k++) {
404 s->llviddsp.add_median_pred(dst, dst - fake_stride,
405 dst, width, &left, &lefttop);
406 lefttop = left = dst[0];
411 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
415 if (s->decorrelate) {
416 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
417 int width = avctx->coded_width;
418 uint8_t *b = p->data[0] + j * s->slice_height * p->linesize[0];
419 uint8_t *g = p->data[1] + j * s->slice_height * p->linesize[1];
420 uint8_t *r = p->data[2] + j * s->slice_height * p->linesize[2];
422 for (i = 0; i < height; i++) {
423 s->llviddsp.add_bytes(b, g, width);
424 s->llviddsp.add_bytes(r, g, width);
434 static int build_huffman(AVCodecContext *avctx, GetBitContext *gbit, int max)
436 MagicYUVContext *s = avctx->priv_data;
439 memset(s->len, 0, sizeof(s->len));
440 while (get_bits_left(gbit) >= 8) {
441 int b = get_bits(gbit, 4);
442 int x = get_bits(gbit, 4);
443 int l = get_bitsz(gbit, b) + 1;
445 for (k = 0; k < l; k++)
447 s->len[i][j + k] = x;
452 if (s->huff_build(&s->vlc[i], s->len[i])) {
453 av_log(avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
454 return AVERROR_INVALIDDATA;
457 if (i == s->planes) {
460 } else if (j > max) {
461 return AVERROR_INVALIDDATA;
465 if (i != s->planes) {
466 av_log(avctx, AV_LOG_ERROR, "Huffman tables too short\n");
467 return AVERROR_INVALIDDATA;
473 static int magy_decode_frame(AVCodecContext *avctx, void *data,
474 int *got_frame, AVPacket *avpkt)
476 MagicYUVContext *s = avctx->priv_data;
477 ThreadFrame frame = { .f = data };
479 GetByteContext gbyte;
481 uint32_t first_offset, offset, next_offset, header_size, slice_width;
482 int width, height, format, version, table_size;
485 bytestream2_init(&gbyte, avpkt->data, avpkt->size);
486 if (bytestream2_get_le32(&gbyte) != MKTAG('M', 'A', 'G', 'Y'))
487 return AVERROR_INVALIDDATA;
489 header_size = bytestream2_get_le32(&gbyte);
490 if (header_size < 32 || header_size >= avpkt->size) {
491 av_log(avctx, AV_LOG_ERROR,
492 "header or packet too small %"PRIu32"\n", header_size);
493 return AVERROR_INVALIDDATA;
496 version = bytestream2_get_byte(&gbyte);
498 avpriv_request_sample(avctx, "Version %d", version);
499 return AVERROR_PATCHWELCOME;
508 s->huff_build = huff_build;
509 s->magy_decode_slice = magy_decode_slice;
511 format = bytestream2_get_byte(&gbyte);
514 avctx->pix_fmt = AV_PIX_FMT_GBRP;
518 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
522 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
525 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
530 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
537 avctx->pix_fmt = AV_PIX_FMT_YUVA444P;
540 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
543 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
547 s->huff_build = huff_build10;
548 s->magy_decode_slice = magy_decode_slice10;
551 avctx->pix_fmt = AV_PIX_FMT_GBRP10;
554 s->huff_build = huff_build10;
555 s->magy_decode_slice = magy_decode_slice10;
558 avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
561 s->huff_build = huff_build10;
562 s->magy_decode_slice = magy_decode_slice10;
565 avctx->pix_fmt = AV_PIX_FMT_GRAY10;
567 s->huff_build = huff_build10;
568 s->magy_decode_slice = magy_decode_slice10;
571 avpriv_request_sample(avctx, "Format 0x%X", format);
572 return AVERROR_PATCHWELCOME;
574 s->planes = av_pix_fmt_count_planes(avctx->pix_fmt);
576 bytestream2_skip(&gbyte, 1);
577 s->color_matrix = bytestream2_get_byte(&gbyte);
578 s->flags = bytestream2_get_byte(&gbyte);
579 s->interlaced = !!(s->flags & 2);
580 bytestream2_skip(&gbyte, 3);
582 width = bytestream2_get_le32(&gbyte);
583 height = bytestream2_get_le32(&gbyte);
584 ret = ff_set_dimensions(avctx, width, height);
588 slice_width = bytestream2_get_le32(&gbyte);
589 if (slice_width != avctx->coded_width) {
590 avpriv_request_sample(avctx, "Slice width %"PRIu32, slice_width);
591 return AVERROR_PATCHWELCOME;
593 s->slice_height = bytestream2_get_le32(&gbyte);
594 if (s->slice_height <= 0 || s->slice_height > INT_MAX - avctx->coded_height) {
595 av_log(avctx, AV_LOG_ERROR,
596 "invalid slice height: %d\n", s->slice_height);
597 return AVERROR_INVALIDDATA;
600 bytestream2_skip(&gbyte, 4);
602 s->nb_slices = (avctx->coded_height + s->slice_height - 1) / s->slice_height;
603 if (s->nb_slices > INT_MAX / sizeof(Slice)) {
604 av_log(avctx, AV_LOG_ERROR,
605 "invalid number of slices: %d\n", s->nb_slices);
606 return AVERROR_INVALIDDATA;
609 for (i = 0; i < s->planes; i++) {
610 av_fast_malloc(&s->slices[i], &s->slices_size[i], s->nb_slices * sizeof(Slice));
612 return AVERROR(ENOMEM);
614 offset = bytestream2_get_le32(&gbyte);
615 if (offset >= avpkt->size - header_size)
616 return AVERROR_INVALIDDATA;
619 first_offset = offset;
621 for (j = 0; j < s->nb_slices - 1; j++) {
622 s->slices[i][j].start = offset + header_size;
624 next_offset = bytestream2_get_le32(&gbyte);
625 if (next_offset <= offset || next_offset >= avpkt->size - header_size)
626 return AVERROR_INVALIDDATA;
628 s->slices[i][j].size = next_offset - offset;
629 offset = next_offset;
632 s->slices[i][j].start = offset + header_size;
633 s->slices[i][j].size = avpkt->size - s->slices[i][j].start;
636 if (bytestream2_get_byte(&gbyte) != s->planes)
637 return AVERROR_INVALIDDATA;
639 bytestream2_skip(&gbyte, s->nb_slices * s->planes);
641 table_size = header_size + first_offset - bytestream2_tell(&gbyte);
643 return AVERROR_INVALIDDATA;
645 ret = init_get_bits8(&gbit, avpkt->data + bytestream2_tell(&gbyte), table_size);
649 ret = build_huffman(avctx, &gbit, s->max);
653 p->pict_type = AV_PICTURE_TYPE_I;
656 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
659 s->buf = avpkt->data;
661 avctx->execute2(avctx, s->magy_decode_slice, NULL, NULL, s->nb_slices);
663 if (avctx->pix_fmt == AV_PIX_FMT_GBRP ||
664 avctx->pix_fmt == AV_PIX_FMT_GBRAP ||
665 avctx->pix_fmt == AV_PIX_FMT_GBRP10 ||
666 avctx->pix_fmt == AV_PIX_FMT_GBRAP10) {
667 FFSWAP(uint8_t*, p->data[0], p->data[1]);
668 FFSWAP(int, p->linesize[0], p->linesize[1]);
670 switch (s->color_matrix) {
672 p->colorspace = AVCOL_SPC_BT470BG;
675 p->colorspace = AVCOL_SPC_BT709;
678 p->color_range = (s->flags & 4) ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
687 static int magy_init_thread_copy(AVCodecContext *avctx)
689 MagicYUVContext *s = avctx->priv_data;
692 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
694 s->slices_size[i] = 0;
701 static av_cold int magy_decode_init(AVCodecContext *avctx)
703 MagicYUVContext *s = avctx->priv_data;
704 ff_llviddsp_init(&s->llviddsp);
708 static av_cold int magy_decode_end(AVCodecContext *avctx)
710 MagicYUVContext * const s = avctx->priv_data;
713 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
714 av_freep(&s->slices[i]);
715 s->slices_size[i] = 0;
716 ff_free_vlc(&s->vlc[i]);
722 AVCodec ff_magicyuv_decoder = {
724 .long_name = NULL_IF_CONFIG_SMALL("MagicYUV video"),
725 .type = AVMEDIA_TYPE_VIDEO,
726 .id = AV_CODEC_ID_MAGICYUV,
727 .priv_data_size = sizeof(MagicYUVContext),
728 .init = magy_decode_init,
729 .init_thread_copy = ONLY_IF_THREADS_ENABLED(magy_init_thread_copy),
730 .close = magy_decode_end,
731 .decode = magy_decode_frame,
732 .capabilities = AV_CODEC_CAP_DR1 |
733 AV_CODEC_CAP_FRAME_THREADS |
734 AV_CODEC_CAP_SLICE_THREADS,
735 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,