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 AV_QSORT(he, 1024, HuffEntry, huff_cmp_len10);
104 for (i = 1023; i >= 0; i--) {
105 codes[i] = code >> (32 - he[i].len);
108 code += 0x80000000u >> (he[i].len - 1);
112 return ff_init_vlc_sparse(vlc, FFMIN(he[1023].len, 12), 1024,
113 bits, sizeof(*bits), sizeof(*bits),
114 codes, sizeof(*codes), sizeof(*codes),
115 syms, sizeof(*syms), sizeof(*syms), 0);
118 static int huff_build(VLC *vlc, uint8_t *len)
127 for (i = 0; i < 256; i++) {
131 AV_QSORT(he, 256, HuffEntry, huff_cmp_len);
134 for (i = 255; i >= 0; i--) {
135 codes[i] = code >> (32 - he[i].len);
138 code += 0x80000000u >> (he[i].len - 1);
142 return ff_init_vlc_sparse(vlc, FFMIN(he[255].len, 12), 256,
143 bits, sizeof(*bits), sizeof(*bits),
144 codes, sizeof(*codes), sizeof(*codes),
145 syms, sizeof(*syms), sizeof(*syms), 0);
148 static void magicyuv_median_pred10(uint16_t *dst, const uint16_t *src1,
149 const uint16_t *diff, intptr_t w,
150 int *left, int *left_top)
158 for (i = 0; i < w; i++) {
159 l = mid_pred(l, src1[i], (l + src1[i] - lt)) + diff[i];
169 static int magy_decode_slice10(AVCodecContext *avctx, void *tdata,
172 MagicYUVContext *s = avctx->priv_data;
173 int interlaced = s->interlaced;
179 for (i = 0; i < s->planes; i++) {
180 int left, lefttop, top;
181 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
182 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
183 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
184 ptrdiff_t fake_stride = (p->linesize[i] / 2) * (1 + interlaced);
185 ptrdiff_t stride = p->linesize[i] / 2;
187 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
188 s->slices[i][j].size);
193 flags = get_bits(&gb, 8);
194 pred = get_bits(&gb, 8);
196 dst = (uint16_t *)p->data[i] + j * sheight * stride;
198 for (k = 0; k < height; k++) {
199 for (x = 0; x < width; x++)
200 dst[x] = get_bits(&gb, 10);
205 for (k = 0; k < height; k++) {
206 for (x = 0; x < width; x++) {
208 if (get_bits_left(&gb) <= 0)
209 return AVERROR_INVALIDDATA;
211 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
213 return AVERROR_INVALIDDATA;
223 dst = (uint16_t *)p->data[i] + j * sheight * stride;
224 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
227 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
230 for (k = 1 + interlaced; k < height; k++) {
231 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, dst[-fake_stride]);
236 dst = (uint16_t *)p->data[i] + j * sheight * stride;
237 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
241 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
245 for (k = 1 + interlaced; k < height; k++) {
246 top = dst[-fake_stride];
248 dst[0] = left & 0x3FF;
249 for (x = 1; x < width; x++) {
250 top = dst[x - fake_stride];
251 lefttop = dst[x - (fake_stride + 1)];
252 left += top - lefttop + dst[x];
253 dst[x] = left & 0x3FF;
259 dst = (uint16_t *)p->data[i] + j * sheight * stride;
260 lefttop = left = dst[0];
261 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
264 lefttop = left = dst[0];
265 s->llviddsp.add_left_pred_int16(dst, dst, 1023, width, 0);
268 for (k = 1 + interlaced; k < height; k++) {
269 magicyuv_median_pred10(dst, dst - fake_stride, dst, width, &left, &lefttop);
270 lefttop = left = dst[0];
275 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
279 if (s->decorrelate) {
280 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
281 int width = avctx->coded_width;
282 uint16_t *r = (uint16_t *)p->data[0] + j * s->slice_height * p->linesize[0] / 2;
283 uint16_t *g = (uint16_t *)p->data[1] + j * s->slice_height * p->linesize[1] / 2;
284 uint16_t *b = (uint16_t *)p->data[2] + j * s->slice_height * p->linesize[2] / 2;
286 for (i = 0; i < height; i++) {
287 for (k = 0; k < width; k++) {
288 b[k] = (b[k] + g[k]) & 0x3FF;
289 r[k] = (r[k] + g[k]) & 0x3FF;
291 b += p->linesize[0] / 2;
292 g += p->linesize[1] / 2;
293 r += p->linesize[2] / 2;
300 static int magy_decode_slice(AVCodecContext *avctx, void *tdata,
303 MagicYUVContext *s = avctx->priv_data;
304 int interlaced = s->interlaced;
310 for (i = 0; i < s->planes; i++) {
311 int left, lefttop, top;
312 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
313 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
314 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
315 ptrdiff_t fake_stride = p->linesize[i] * (1 + interlaced);
316 ptrdiff_t stride = p->linesize[i];
318 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
319 s->slices[i][j].size);
324 flags = get_bits(&gb, 8);
325 pred = get_bits(&gb, 8);
327 dst = p->data[i] + j * sheight * stride;
329 for (k = 0; k < height; k++) {
330 for (x = 0; x < width; x++)
331 dst[x] = get_bits(&gb, 8);
336 for (k = 0; k < height; k++) {
337 for (x = 0; x < width; x++) {
339 if (get_bits_left(&gb) <= 0)
340 return AVERROR_INVALIDDATA;
342 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
344 return AVERROR_INVALIDDATA;
354 dst = p->data[i] + j * sheight * stride;
355 s->llviddsp.add_left_pred(dst, dst, width, 0);
358 s->llviddsp.add_left_pred(dst, dst, width, 0);
361 for (k = 1 + interlaced; k < height; k++) {
362 s->llviddsp.add_left_pred(dst, dst, width, dst[-fake_stride]);
367 dst = p->data[i] + j * sheight * stride;
368 s->llviddsp.add_left_pred(dst, dst, width, 0);
372 s->llviddsp.add_left_pred(dst, dst, width, 0);
376 for (k = 1 + interlaced; k < height; k++) {
377 top = dst[-fake_stride];
380 for (x = 1; x < width; x++) {
381 top = dst[x - fake_stride];
382 lefttop = dst[x - (fake_stride + 1)];
383 left += top - lefttop + dst[x];
390 dst = p->data[i] + j * sheight * stride;
391 lefttop = left = dst[0];
392 s->llviddsp.add_left_pred(dst, dst, width, 0);
395 lefttop = left = dst[0];
396 s->llviddsp.add_left_pred(dst, dst, width, 0);
399 for (k = 1 + interlaced; k < height; k++) {
400 s->llviddsp.add_median_pred(dst, dst - fake_stride,
401 dst, width, &left, &lefttop);
402 lefttop = left = dst[0];
407 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
411 if (s->decorrelate) {
412 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
413 int width = avctx->coded_width;
414 uint8_t *b = p->data[0] + j * s->slice_height * p->linesize[0];
415 uint8_t *g = p->data[1] + j * s->slice_height * p->linesize[1];
416 uint8_t *r = p->data[2] + j * s->slice_height * p->linesize[2];
418 for (i = 0; i < height; i++) {
419 s->llviddsp.add_bytes(b, g, width);
420 s->llviddsp.add_bytes(r, g, width);
430 static int build_huffman(AVCodecContext *avctx, GetBitContext *gbit, int max)
432 MagicYUVContext *s = avctx->priv_data;
435 memset(s->len, 0, sizeof(s->len));
436 while (get_bits_left(gbit) >= 8) {
437 int b = get_bits(gbit, 4);
438 int x = get_bits(gbit, 4);
439 int l = get_bitsz(gbit, b) + 1;
441 for (k = 0; k < l; k++)
443 s->len[i][j + k] = x;
448 if (s->huff_build(&s->vlc[i], s->len[i])) {
449 av_log(avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
450 return AVERROR_INVALIDDATA;
453 if (i == s->planes) {
456 } else if (j > max) {
457 return AVERROR_INVALIDDATA;
461 if (i != s->planes) {
462 av_log(avctx, AV_LOG_ERROR, "Huffman tables too short\n");
463 return AVERROR_INVALIDDATA;
469 static int magy_decode_frame(AVCodecContext *avctx, void *data,
470 int *got_frame, AVPacket *avpkt)
472 MagicYUVContext *s = avctx->priv_data;
473 ThreadFrame frame = { .f = data };
475 GetByteContext gbyte;
477 uint32_t first_offset, offset, next_offset, header_size, slice_width;
478 int width, height, format, version, table_size;
481 bytestream2_init(&gbyte, avpkt->data, avpkt->size);
482 if (bytestream2_get_le32(&gbyte) != MKTAG('M', 'A', 'G', 'Y'))
483 return AVERROR_INVALIDDATA;
485 header_size = bytestream2_get_le32(&gbyte);
486 if (header_size < 32 || header_size >= avpkt->size) {
487 av_log(avctx, AV_LOG_ERROR,
488 "header or packet too small %"PRIu32"\n", header_size);
489 return AVERROR_INVALIDDATA;
492 version = bytestream2_get_byte(&gbyte);
494 avpriv_request_sample(avctx, "Version %d", version);
495 return AVERROR_PATCHWELCOME;
504 s->huff_build = huff_build;
505 s->magy_decode_slice = magy_decode_slice;
507 format = bytestream2_get_byte(&gbyte);
510 avctx->pix_fmt = AV_PIX_FMT_GBRP;
514 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
518 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
521 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
526 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
533 avctx->pix_fmt = AV_PIX_FMT_YUVA444P;
536 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
539 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
543 s->huff_build = huff_build10;
544 s->magy_decode_slice = magy_decode_slice10;
547 avctx->pix_fmt = AV_PIX_FMT_GBRP10;
550 s->huff_build = huff_build10;
551 s->magy_decode_slice = magy_decode_slice10;
554 avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
557 s->huff_build = huff_build10;
558 s->magy_decode_slice = magy_decode_slice10;
561 avctx->pix_fmt = AV_PIX_FMT_GRAY10;
563 s->huff_build = huff_build10;
564 s->magy_decode_slice = magy_decode_slice10;
567 avpriv_request_sample(avctx, "Format 0x%X", format);
568 return AVERROR_PATCHWELCOME;
570 s->planes = av_pix_fmt_count_planes(avctx->pix_fmt);
572 bytestream2_skip(&gbyte, 1);
573 s->color_matrix = bytestream2_get_byte(&gbyte);
574 s->flags = bytestream2_get_byte(&gbyte);
575 s->interlaced = !!(s->flags & 2);
576 bytestream2_skip(&gbyte, 3);
578 width = bytestream2_get_le32(&gbyte);
579 height = bytestream2_get_le32(&gbyte);
580 ret = ff_set_dimensions(avctx, width, height);
584 slice_width = bytestream2_get_le32(&gbyte);
585 if (slice_width != avctx->coded_width) {
586 avpriv_request_sample(avctx, "Slice width %"PRIu32, slice_width);
587 return AVERROR_PATCHWELCOME;
589 s->slice_height = bytestream2_get_le32(&gbyte);
590 if (s->slice_height <= 0 || s->slice_height > INT_MAX - avctx->coded_height) {
591 av_log(avctx, AV_LOG_ERROR,
592 "invalid slice height: %d\n", s->slice_height);
593 return AVERROR_INVALIDDATA;
596 bytestream2_skip(&gbyte, 4);
598 s->nb_slices = (avctx->coded_height + s->slice_height - 1) / s->slice_height;
599 if (s->nb_slices > INT_MAX / sizeof(Slice)) {
600 av_log(avctx, AV_LOG_ERROR,
601 "invalid number of slices: %d\n", s->nb_slices);
602 return AVERROR_INVALIDDATA;
605 for (i = 0; i < s->planes; i++) {
606 av_fast_malloc(&s->slices[i], &s->slices_size[i], s->nb_slices * sizeof(Slice));
608 return AVERROR(ENOMEM);
610 offset = bytestream2_get_le32(&gbyte);
611 if (offset >= avpkt->size - header_size)
612 return AVERROR_INVALIDDATA;
615 first_offset = offset;
617 for (j = 0; j < s->nb_slices - 1; j++) {
618 s->slices[i][j].start = offset + header_size;
620 next_offset = bytestream2_get_le32(&gbyte);
621 if (next_offset <= offset || next_offset >= avpkt->size - header_size)
622 return AVERROR_INVALIDDATA;
624 s->slices[i][j].size = next_offset - offset;
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_byte(&gbyte) != s->planes)
633 return AVERROR_INVALIDDATA;
635 bytestream2_skip(&gbyte, s->nb_slices * s->planes);
637 table_size = header_size + first_offset - bytestream2_tell(&gbyte);
639 return AVERROR_INVALIDDATA;
641 ret = init_get_bits8(&gbit, avpkt->data + bytestream2_tell(&gbyte), table_size);
645 ret = build_huffman(avctx, &gbit, s->max);
649 p->pict_type = AV_PICTURE_TYPE_I;
652 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
655 s->buf = avpkt->data;
657 avctx->execute2(avctx, s->magy_decode_slice, NULL, NULL, s->nb_slices);
659 if (avctx->pix_fmt == AV_PIX_FMT_GBRP ||
660 avctx->pix_fmt == AV_PIX_FMT_GBRAP ||
661 avctx->pix_fmt == AV_PIX_FMT_GBRP10 ||
662 avctx->pix_fmt == AV_PIX_FMT_GBRAP10) {
663 FFSWAP(uint8_t*, p->data[0], p->data[1]);
664 FFSWAP(int, p->linesize[0], p->linesize[1]);
666 switch (s->color_matrix) {
668 p->colorspace = AVCOL_SPC_BT470BG;
671 p->colorspace = AVCOL_SPC_BT709;
674 p->color_range = (s->flags & 4) ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
683 static int magy_init_thread_copy(AVCodecContext *avctx)
685 MagicYUVContext *s = avctx->priv_data;
688 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
690 s->slices_size[i] = 0;
697 static av_cold int magy_decode_init(AVCodecContext *avctx)
699 MagicYUVContext *s = avctx->priv_data;
700 ff_llviddsp_init(&s->llviddsp);
704 static av_cold int magy_decode_end(AVCodecContext *avctx)
706 MagicYUVContext * const s = avctx->priv_data;
709 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
710 av_freep(&s->slices[i]);
711 s->slices_size[i] = 0;
712 ff_free_vlc(&s->vlc[i]);
718 AVCodec ff_magicyuv_decoder = {
720 .long_name = NULL_IF_CONFIG_SMALL("MagicYUV video"),
721 .type = AVMEDIA_TYPE_VIDEO,
722 .id = AV_CODEC_ID_MAGICYUV,
723 .priv_data_size = sizeof(MagicYUVContext),
724 .init = magy_decode_init,
725 .init_thread_copy = ONLY_IF_THREADS_ENABLED(magy_init_thread_copy),
726 .close = magy_decode_end,
727 .decode = magy_decode_frame,
728 .capabilities = AV_CODEC_CAP_DR1 |
729 AV_CODEC_CAP_FRAME_THREADS |
730 AV_CODEC_CAP_SLICE_THREADS,
731 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,