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 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 uint8_t len[4][4096]; // table of code lengths for each plane
72 VLC vlc[4]; // VLC for each plane
73 int (*huff_build)(VLC *vlc, uint8_t *len);
74 int (*magy_decode_slice)(AVCodecContext *avctx, void *tdata,
76 LLVidDSPContext llviddsp;
79 static int huff_cmp_len(const void *a, const void *b)
81 const HuffEntry *aa = a, *bb = b;
82 return (aa->len - bb->len) * 256 + aa->sym - bb->sym;
85 static int huff_cmp_len10(const void *a, const void *b)
87 const HuffEntry *aa = a, *bb = b;
88 return (aa->len - bb->len) * 1024 + aa->sym - bb->sym;
91 static int huff_cmp_len12(const void *a, const void *b)
93 const HuffEntry *aa = a, *bb = b;
94 return (aa->len - bb->len) * 4096 + aa->sym - bb->sym;
97 static int huff_build10(VLC *vlc, uint8_t *len)
100 uint32_t codes[1024];
106 for (i = 0; i < 1024; i++) {
107 he[i].sym = 1023 - i;
109 if (len[i] == 0 || len[i] > 32)
110 return AVERROR_INVALIDDATA;
112 AV_QSORT(he, 1024, HuffEntry, huff_cmp_len10);
115 for (i = 1023; i >= 0; i--) {
116 codes[i] = code >> (32 - he[i].len);
118 syms[i] = 1023 - he[i].sym;
119 code += 0x80000000u >> (he[i].len - 1);
123 return ff_init_vlc_sparse(vlc, FFMIN(he[1023].len, 12), 1024,
124 bits, sizeof(*bits), sizeof(*bits),
125 codes, sizeof(*codes), sizeof(*codes),
126 syms, sizeof(*syms), sizeof(*syms), 0);
129 static int huff_build12(VLC *vlc, uint8_t *len)
132 uint32_t codes[4096];
138 for (i = 0; i < 4096; i++) {
139 he[i].sym = 4095 - i;
141 if (len[i] == 0 || len[i] > 32)
142 return AVERROR_INVALIDDATA;
144 AV_QSORT(he, 4096, HuffEntry, huff_cmp_len12);
147 for (i = 4095; i >= 0; i--) {
148 codes[i] = code >> (32 - he[i].len);
150 syms[i] = 4095 - he[i].sym;
151 code += 0x80000000u >> (he[i].len - 1);
155 return ff_init_vlc_sparse(vlc, FFMIN(he[4095].len, 12), 4096,
156 bits, sizeof(*bits), sizeof(*bits),
157 codes, sizeof(*codes), sizeof(*codes),
158 syms, sizeof(*syms), sizeof(*syms), 0);
161 static int huff_build(VLC *vlc, uint8_t *len)
170 for (i = 0; i < 256; i++) {
173 if (len[i] == 0 || len[i] > 32)
174 return AVERROR_INVALIDDATA;
176 AV_QSORT(he, 256, HuffEntry, huff_cmp_len);
179 for (i = 255; i >= 0; i--) {
180 codes[i] = code >> (32 - he[i].len);
182 syms[i] = 255 - he[i].sym;
183 code += 0x80000000u >> (he[i].len - 1);
187 return ff_init_vlc_sparse(vlc, FFMIN(he[255].len, 12), 256,
188 bits, sizeof(*bits), sizeof(*bits),
189 codes, sizeof(*codes), sizeof(*codes),
190 syms, sizeof(*syms), sizeof(*syms), 0);
193 static void magicyuv_median_pred16(uint16_t *dst, const uint16_t *src1,
194 const uint16_t *diff, intptr_t w,
195 int *left, int *left_top, int max)
203 for (i = 0; i < w; i++) {
204 l = mid_pred(l, src1[i], (l + src1[i] - lt)) + diff[i];
214 static int magy_decode_slice10(AVCodecContext *avctx, void *tdata,
217 MagicYUVContext *s = avctx->priv_data;
218 int interlaced = s->interlaced;
219 const int bps = s->bps;
220 const int max = s->max - 1;
226 for (i = 0; i < s->planes; i++) {
227 int left, lefttop, top;
228 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
229 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
230 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
231 ptrdiff_t fake_stride = (p->linesize[i] / 2) * (1 + interlaced);
232 ptrdiff_t stride = p->linesize[i] / 2;
234 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
235 s->slices[i][j].size);
240 flags = get_bits(&gb, 8);
241 pred = get_bits(&gb, 8);
243 dst = (uint16_t *)p->data[i] + j * sheight * stride;
245 if (get_bits_left(&gb) < bps * width * height)
246 return AVERROR_INVALIDDATA;
247 for (k = 0; k < height; k++) {
248 for (x = 0; x < width; x++)
249 dst[x] = get_bits(&gb, bps);
254 for (k = 0; k < height; k++) {
255 for (x = 0; x < width; x++) {
257 if (get_bits_left(&gb) <= 0)
258 return AVERROR_INVALIDDATA;
260 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
262 return AVERROR_INVALIDDATA;
272 dst = (uint16_t *)p->data[i] + j * sheight * stride;
273 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
276 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
279 for (k = 1 + interlaced; k < height; k++) {
280 s->llviddsp.add_left_pred_int16(dst, dst, max, width, dst[-fake_stride]);
285 dst = (uint16_t *)p->data[i] + j * sheight * stride;
286 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
289 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
292 for (k = 1 + interlaced; k < height; k++) {
293 top = dst[-fake_stride];
296 for (x = 1; x < width; x++) {
297 top = dst[x - fake_stride];
298 lefttop = dst[x - (fake_stride + 1)];
299 left += top - lefttop + dst[x];
306 dst = (uint16_t *)p->data[i] + j * sheight * stride;
307 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
310 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
313 lefttop = left = dst[0];
314 for (k = 1 + interlaced; k < height; k++) {
315 magicyuv_median_pred16(dst, dst - fake_stride, dst, width, &left, &lefttop, max);
316 lefttop = left = dst[0];
321 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
325 if (s->decorrelate) {
326 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
327 int width = avctx->coded_width;
328 uint16_t *r = (uint16_t *)p->data[0] + j * s->slice_height * p->linesize[0] / 2;
329 uint16_t *g = (uint16_t *)p->data[1] + j * s->slice_height * p->linesize[1] / 2;
330 uint16_t *b = (uint16_t *)p->data[2] + j * s->slice_height * p->linesize[2] / 2;
332 for (i = 0; i < height; i++) {
333 for (k = 0; k < width; k++) {
334 b[k] = (b[k] + g[k]) & max;
335 r[k] = (r[k] + g[k]) & max;
337 b += p->linesize[0] / 2;
338 g += p->linesize[1] / 2;
339 r += p->linesize[2] / 2;
346 static int magy_decode_slice(AVCodecContext *avctx, void *tdata,
349 MagicYUVContext *s = avctx->priv_data;
350 int interlaced = s->interlaced;
352 int i, k, x, min_width;
356 for (i = 0; i < s->planes; i++) {
357 int left, lefttop, top;
358 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
359 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
360 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
361 ptrdiff_t fake_stride = p->linesize[i] * (1 + interlaced);
362 ptrdiff_t stride = p->linesize[i];
364 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
365 s->slices[i][j].size);
370 flags = get_bits(&gb, 8);
371 pred = get_bits(&gb, 8);
373 dst = p->data[i] + j * sheight * stride;
375 if (get_bits_left(&gb) < 8* width * height)
376 return AVERROR_INVALIDDATA;
377 for (k = 0; k < height; k++) {
378 for (x = 0; x < width; x++)
379 dst[x] = get_bits(&gb, 8);
384 for (k = 0; k < height; k++) {
385 for (x = 0; x < width; x++) {
387 if (get_bits_left(&gb) <= 0)
388 return AVERROR_INVALIDDATA;
390 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
392 return AVERROR_INVALIDDATA;
402 dst = p->data[i] + j * sheight * stride;
403 s->llviddsp.add_left_pred(dst, dst, width, 0);
406 s->llviddsp.add_left_pred(dst, dst, width, 0);
409 for (k = 1 + interlaced; k < height; k++) {
410 s->llviddsp.add_left_pred(dst, dst, width, dst[-fake_stride]);
415 dst = p->data[i] + j * sheight * stride;
416 s->llviddsp.add_left_pred(dst, dst, width, 0);
419 s->llviddsp.add_left_pred(dst, dst, width, 0);
422 min_width = FFMIN(width, 32);
423 for (k = 1 + interlaced; k < height; k++) {
424 top = dst[-fake_stride];
427 for (x = 1; x < min_width; x++) { /* dsp need aligned 32 */
428 top = dst[x - fake_stride];
429 lefttop = dst[x - (fake_stride + 1)];
430 left += top - lefttop + dst[x];
434 s->llviddsp.add_gradient_pred(dst + 32, fake_stride, width - 32);
439 dst = p->data[i] + j * sheight * stride;
440 s->llviddsp.add_left_pred(dst, dst, width, 0);
443 s->llviddsp.add_left_pred(dst, dst, width, 0);
446 lefttop = left = dst[0];
447 for (k = 1 + interlaced; k < height; k++) {
448 s->llviddsp.add_median_pred(dst, dst - fake_stride,
449 dst, width, &left, &lefttop);
450 lefttop = left = dst[0];
455 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
459 if (s->decorrelate) {
460 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
461 int width = avctx->coded_width;
462 uint8_t *b = p->data[0] + j * s->slice_height * p->linesize[0];
463 uint8_t *g = p->data[1] + j * s->slice_height * p->linesize[1];
464 uint8_t *r = p->data[2] + j * s->slice_height * p->linesize[2];
466 for (i = 0; i < height; i++) {
467 s->llviddsp.add_bytes(b, g, width);
468 s->llviddsp.add_bytes(r, g, width);
478 static int build_huffman(AVCodecContext *avctx, GetBitContext *gbit, int max)
480 MagicYUVContext *s = avctx->priv_data;
483 memset(s->len, 0, sizeof(s->len));
484 while (get_bits_left(gbit) >= 8) {
485 int b = get_bits(gbit, 1);
486 int x = get_bits(gbit, 7);
487 int l = get_bitsz(gbit, b * 8) + 1;
489 for (k = 0; k < l; k++)
491 s->len[i][j + k] = x;
496 if (s->huff_build(&s->vlc[i], s->len[i])) {
497 av_log(avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
498 return AVERROR_INVALIDDATA;
501 if (i == s->planes) {
504 } else if (j > max) {
505 av_log(avctx, AV_LOG_ERROR, "Invalid Huffman codes\n");
506 return AVERROR_INVALIDDATA;
510 if (i != s->planes) {
511 av_log(avctx, AV_LOG_ERROR, "Huffman tables too short\n");
512 return AVERROR_INVALIDDATA;
518 static int magy_decode_frame(AVCodecContext *avctx, void *data,
519 int *got_frame, AVPacket *avpkt)
521 MagicYUVContext *s = avctx->priv_data;
522 ThreadFrame frame = { .f = data };
524 GetByteContext gbyte;
526 uint32_t first_offset, offset, next_offset, header_size, slice_width;
527 int width, height, format, version, table_size;
530 bytestream2_init(&gbyte, avpkt->data, avpkt->size);
531 if (bytestream2_get_le32(&gbyte) != MKTAG('M', 'A', 'G', 'Y'))
532 return AVERROR_INVALIDDATA;
534 header_size = bytestream2_get_le32(&gbyte);
535 if (header_size < 32 || header_size >= avpkt->size) {
536 av_log(avctx, AV_LOG_ERROR,
537 "header or packet too small %"PRIu32"\n", header_size);
538 return AVERROR_INVALIDDATA;
541 version = bytestream2_get_byte(&gbyte);
543 avpriv_request_sample(avctx, "Version %d", version);
544 return AVERROR_PATCHWELCOME;
554 format = bytestream2_get_byte(&gbyte);
557 avctx->pix_fmt = AV_PIX_FMT_GBRP;
561 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
565 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
568 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
573 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
580 avctx->pix_fmt = AV_PIX_FMT_YUVA444P;
583 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
586 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
592 avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
596 avctx->pix_fmt = AV_PIX_FMT_GBRP10;
601 avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
606 avctx->pix_fmt = AV_PIX_FMT_GBRP12;
611 avctx->pix_fmt = AV_PIX_FMT_GBRAP12;
616 avctx->pix_fmt = AV_PIX_FMT_GRAY10;
620 avpriv_request_sample(avctx, "Format 0x%X", format);
621 return AVERROR_PATCHWELCOME;
623 s->max = 1 << s->bps;
624 s->magy_decode_slice = s->bps == 8 ? magy_decode_slice : magy_decode_slice10;
626 s->huff_build = huff_build;
628 s->huff_build = s->bps == 10 ? huff_build10 : huff_build12;
629 s->planes = av_pix_fmt_count_planes(avctx->pix_fmt);
631 bytestream2_skip(&gbyte, 1);
632 s->color_matrix = bytestream2_get_byte(&gbyte);
633 s->flags = bytestream2_get_byte(&gbyte);
634 s->interlaced = !!(s->flags & 2);
635 bytestream2_skip(&gbyte, 3);
637 width = bytestream2_get_le32(&gbyte);
638 height = bytestream2_get_le32(&gbyte);
639 ret = ff_set_dimensions(avctx, width, height);
643 slice_width = bytestream2_get_le32(&gbyte);
644 if (slice_width != avctx->coded_width) {
645 avpriv_request_sample(avctx, "Slice width %"PRIu32, slice_width);
646 return AVERROR_PATCHWELCOME;
648 s->slice_height = bytestream2_get_le32(&gbyte);
649 if (s->slice_height <= 0 || s->slice_height > INT_MAX - avctx->coded_height) {
650 av_log(avctx, AV_LOG_ERROR,
651 "invalid slice height: %d\n", s->slice_height);
652 return AVERROR_INVALIDDATA;
655 bytestream2_skip(&gbyte, 4);
657 s->nb_slices = (avctx->coded_height + s->slice_height - 1) / s->slice_height;
658 if (s->nb_slices > INT_MAX / sizeof(Slice)) {
659 av_log(avctx, AV_LOG_ERROR,
660 "invalid number of slices: %d\n", s->nb_slices);
661 return AVERROR_INVALIDDATA;
665 if ((s->slice_height >> s->vshift[1]) < 2) {
666 av_log(avctx, AV_LOG_ERROR, "impossible slice height\n");
667 return AVERROR_INVALIDDATA;
669 if ((avctx->coded_height % s->slice_height) && ((avctx->coded_height % s->slice_height) >> s->vshift[1]) < 2) {
670 av_log(avctx, AV_LOG_ERROR, "impossible height\n");
671 return AVERROR_INVALIDDATA;
675 for (i = 0; i < s->planes; i++) {
676 av_fast_malloc(&s->slices[i], &s->slices_size[i], s->nb_slices * sizeof(Slice));
678 return AVERROR(ENOMEM);
680 offset = bytestream2_get_le32(&gbyte);
681 if (offset >= avpkt->size - header_size)
682 return AVERROR_INVALIDDATA;
685 first_offset = offset;
687 for (j = 0; j < s->nb_slices - 1; j++) {
688 s->slices[i][j].start = offset + header_size;
690 next_offset = bytestream2_get_le32(&gbyte);
691 if (next_offset <= offset || next_offset >= avpkt->size - header_size)
692 return AVERROR_INVALIDDATA;
694 s->slices[i][j].size = next_offset - offset;
695 offset = next_offset;
698 s->slices[i][j].start = offset + header_size;
699 s->slices[i][j].size = avpkt->size - s->slices[i][j].start;
702 if (bytestream2_get_byte(&gbyte) != s->planes)
703 return AVERROR_INVALIDDATA;
705 bytestream2_skip(&gbyte, s->nb_slices * s->planes);
707 table_size = header_size + first_offset - bytestream2_tell(&gbyte);
709 return AVERROR_INVALIDDATA;
711 ret = init_get_bits8(&gbit, avpkt->data + bytestream2_tell(&gbyte), table_size);
715 ret = build_huffman(avctx, &gbit, s->max);
719 p->pict_type = AV_PICTURE_TYPE_I;
722 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
725 s->buf = avpkt->data;
727 avctx->execute2(avctx, s->magy_decode_slice, NULL, NULL, s->nb_slices);
729 if (avctx->pix_fmt == AV_PIX_FMT_GBRP ||
730 avctx->pix_fmt == AV_PIX_FMT_GBRAP ||
731 avctx->pix_fmt == AV_PIX_FMT_GBRP10 ||
732 avctx->pix_fmt == AV_PIX_FMT_GBRAP10||
733 avctx->pix_fmt == AV_PIX_FMT_GBRAP12||
734 avctx->pix_fmt == AV_PIX_FMT_GBRP12) {
735 FFSWAP(uint8_t*, p->data[0], p->data[1]);
736 FFSWAP(int, p->linesize[0], p->linesize[1]);
738 switch (s->color_matrix) {
740 p->colorspace = AVCOL_SPC_BT470BG;
743 p->colorspace = AVCOL_SPC_BT709;
746 p->color_range = (s->flags & 4) ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
754 static av_cold int magy_decode_init(AVCodecContext *avctx)
756 MagicYUVContext *s = avctx->priv_data;
757 ff_llviddsp_init(&s->llviddsp);
761 static av_cold int magy_decode_end(AVCodecContext *avctx)
763 MagicYUVContext * const s = avctx->priv_data;
766 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
767 av_freep(&s->slices[i]);
768 s->slices_size[i] = 0;
769 ff_free_vlc(&s->vlc[i]);
775 AVCodec ff_magicyuv_decoder = {
777 .long_name = NULL_IF_CONFIG_SMALL("MagicYUV video"),
778 .type = AVMEDIA_TYPE_VIDEO,
779 .id = AV_CODEC_ID_MAGICYUV,
780 .priv_data_size = sizeof(MagicYUVContext),
781 .init = magy_decode_init,
782 .close = magy_decode_end,
783 .decode = magy_decode_frame,
784 .capabilities = AV_CODEC_CAP_DR1 |
785 AV_CODEC_CAP_FRAME_THREADS |
786 AV_CODEC_CAP_SLICE_THREADS,
787 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,