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) * 4096 + bb->sym - aa->sym;
85 static int huff_build10(VLC *vlc, uint8_t *len)
91 for (i = 0; i < 1024; i++) {
94 if (len[i] == 0 || len[i] > 32)
95 return AVERROR_INVALIDDATA;
97 AV_QSORT(he, 1024, HuffEntry, huff_cmp_len);
100 for (i = 1023; i >= 0; i--) {
101 he[i].code = code >> (32 - he[i].len);
102 code += 0x80000000u >> (he[i].len - 1);
106 return ff_init_vlc_sparse(vlc, FFMIN(he[1023].len, 12), 1024,
107 &he[0].len, sizeof(he[0]), sizeof(he[0].len),
108 &he[0].code, sizeof(he[0]), sizeof(he[0].code),
109 &he[0].sym, sizeof(he[0]), sizeof(he[0].sym), 0);
112 static int huff_build12(VLC *vlc, uint8_t *len)
118 for (i = 0; i < 4096; i++) {
121 if (len[i] == 0 || len[i] > 32)
122 return AVERROR_INVALIDDATA;
124 AV_QSORT(he, 4096, HuffEntry, huff_cmp_len);
127 for (i = 4095; i >= 0; i--) {
128 he[i].code = code >> (32 - he[i].len);
129 code += 0x80000000u >> (he[i].len - 1);
133 return ff_init_vlc_sparse(vlc, FFMIN(he[4095].len, 12), 4096,
134 &he[0].len, sizeof(he[0]), sizeof(he[0].len),
135 &he[0].code, sizeof(he[0]), sizeof(he[0].code),
136 &he[0].sym, sizeof(he[0]), sizeof(he[0].sym), 0);
139 static int huff_build(VLC *vlc, uint8_t *len)
145 for (i = 0; i < 256; i++) {
148 if (len[i] == 0 || len[i] > 32)
149 return AVERROR_INVALIDDATA;
151 AV_QSORT(he, 256, HuffEntry, huff_cmp_len);
154 for (i = 255; i >= 0; i--) {
155 he[i].code = code >> (32 - he[i].len);
156 code += 0x80000000u >> (he[i].len - 1);
160 return ff_init_vlc_sparse(vlc, FFMIN(he[255].len, 12), 256,
161 &he[0].len, sizeof(he[0]), sizeof(he[0].len),
162 &he[0].code, sizeof(he[0]), sizeof(he[0].code),
163 &he[0].sym, sizeof(he[0]), sizeof(he[0].sym), 0);
166 static void magicyuv_median_pred16(uint16_t *dst, const uint16_t *src1,
167 const uint16_t *diff, intptr_t w,
168 int *left, int *left_top, int max)
176 for (i = 0; i < w; i++) {
177 l = mid_pred(l, src1[i], (l + src1[i] - lt)) + diff[i];
187 static int magy_decode_slice10(AVCodecContext *avctx, void *tdata,
190 MagicYUVContext *s = avctx->priv_data;
191 int interlaced = s->interlaced;
192 const int bps = s->bps;
193 const int max = s->max - 1;
199 for (i = 0; i < s->planes; i++) {
200 int left, lefttop, top;
201 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
202 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
203 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
204 ptrdiff_t fake_stride = (p->linesize[i] / 2) * (1 + interlaced);
205 ptrdiff_t stride = p->linesize[i] / 2;
207 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
208 s->slices[i][j].size);
213 flags = get_bits(&gb, 8);
214 pred = get_bits(&gb, 8);
216 dst = (uint16_t *)p->data[i] + j * sheight * stride;
218 if (get_bits_left(&gb) < bps * width * height)
219 return AVERROR_INVALIDDATA;
220 for (k = 0; k < height; k++) {
221 for (x = 0; x < width; x++)
222 dst[x] = get_bits(&gb, bps);
227 for (k = 0; k < height; k++) {
228 for (x = 0; x < width; x++) {
230 if (get_bits_left(&gb) <= 0)
231 return AVERROR_INVALIDDATA;
233 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
235 return AVERROR_INVALIDDATA;
245 dst = (uint16_t *)p->data[i] + j * sheight * stride;
246 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
249 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
252 for (k = 1 + interlaced; k < height; k++) {
253 s->llviddsp.add_left_pred_int16(dst, dst, max, width, dst[-fake_stride]);
258 dst = (uint16_t *)p->data[i] + j * sheight * stride;
259 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
262 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
265 for (k = 1 + interlaced; k < height; k++) {
266 top = dst[-fake_stride];
269 for (x = 1; x < width; x++) {
270 top = dst[x - fake_stride];
271 lefttop = dst[x - (fake_stride + 1)];
272 left += top - lefttop + dst[x];
279 dst = (uint16_t *)p->data[i] + j * sheight * stride;
280 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
283 s->llviddsp.add_left_pred_int16(dst, dst, max, width, 0);
286 lefttop = left = dst[0];
287 for (k = 1 + interlaced; k < height; k++) {
288 magicyuv_median_pred16(dst, dst - fake_stride, dst, width, &left, &lefttop, max);
289 lefttop = left = dst[0];
294 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
298 if (s->decorrelate) {
299 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
300 int width = avctx->coded_width;
301 uint16_t *r = (uint16_t *)p->data[0] + j * s->slice_height * p->linesize[0] / 2;
302 uint16_t *g = (uint16_t *)p->data[1] + j * s->slice_height * p->linesize[1] / 2;
303 uint16_t *b = (uint16_t *)p->data[2] + j * s->slice_height * p->linesize[2] / 2;
305 for (i = 0; i < height; i++) {
306 for (k = 0; k < width; k++) {
307 b[k] = (b[k] + g[k]) & max;
308 r[k] = (r[k] + g[k]) & max;
310 b += p->linesize[0] / 2;
311 g += p->linesize[1] / 2;
312 r += p->linesize[2] / 2;
319 static int magy_decode_slice(AVCodecContext *avctx, void *tdata,
322 MagicYUVContext *s = avctx->priv_data;
323 int interlaced = s->interlaced;
325 int i, k, x, min_width;
329 for (i = 0; i < s->planes; i++) {
330 int left, lefttop, top;
331 int height = AV_CEIL_RSHIFT(FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height), s->vshift[i]);
332 int width = AV_CEIL_RSHIFT(avctx->coded_width, s->hshift[i]);
333 int sheight = AV_CEIL_RSHIFT(s->slice_height, s->vshift[i]);
334 ptrdiff_t fake_stride = p->linesize[i] * (1 + interlaced);
335 ptrdiff_t stride = p->linesize[i];
337 int ret = init_get_bits8(&gb, s->buf + s->slices[i][j].start,
338 s->slices[i][j].size);
343 flags = get_bits(&gb, 8);
344 pred = get_bits(&gb, 8);
346 dst = p->data[i] + j * sheight * stride;
348 if (get_bits_left(&gb) < 8* width * height)
349 return AVERROR_INVALIDDATA;
350 for (k = 0; k < height; k++) {
351 for (x = 0; x < width; x++)
352 dst[x] = get_bits(&gb, 8);
357 for (k = 0; k < height; k++) {
358 for (x = 0; x < width; x++) {
360 if (get_bits_left(&gb) <= 0)
361 return AVERROR_INVALIDDATA;
363 pix = get_vlc2(&gb, s->vlc[i].table, s->vlc[i].bits, 3);
365 return AVERROR_INVALIDDATA;
375 dst = p->data[i] + j * sheight * stride;
376 s->llviddsp.add_left_pred(dst, dst, width, 0);
379 s->llviddsp.add_left_pred(dst, dst, width, 0);
382 for (k = 1 + interlaced; k < height; k++) {
383 s->llviddsp.add_left_pred(dst, dst, width, dst[-fake_stride]);
388 dst = p->data[i] + j * sheight * stride;
389 s->llviddsp.add_left_pred(dst, dst, width, 0);
392 s->llviddsp.add_left_pred(dst, dst, width, 0);
395 min_width = FFMIN(width, 32);
396 for (k = 1 + interlaced; k < height; k++) {
397 top = dst[-fake_stride];
400 for (x = 1; x < min_width; x++) { /* dsp need aligned 32 */
401 top = dst[x - fake_stride];
402 lefttop = dst[x - (fake_stride + 1)];
403 left += top - lefttop + dst[x];
407 s->llviddsp.add_gradient_pred(dst + 32, fake_stride, width - 32);
412 dst = p->data[i] + j * sheight * stride;
413 s->llviddsp.add_left_pred(dst, dst, width, 0);
416 s->llviddsp.add_left_pred(dst, dst, width, 0);
419 lefttop = left = dst[0];
420 for (k = 1 + interlaced; k < height; k++) {
421 s->llviddsp.add_median_pred(dst, dst - fake_stride,
422 dst, width, &left, &lefttop);
423 lefttop = left = dst[0];
428 avpriv_request_sample(avctx, "Unknown prediction: %d", pred);
432 if (s->decorrelate) {
433 int height = FFMIN(s->slice_height, avctx->coded_height - j * s->slice_height);
434 int width = avctx->coded_width;
435 uint8_t *b = p->data[0] + j * s->slice_height * p->linesize[0];
436 uint8_t *g = p->data[1] + j * s->slice_height * p->linesize[1];
437 uint8_t *r = p->data[2] + j * s->slice_height * p->linesize[2];
439 for (i = 0; i < height; i++) {
440 s->llviddsp.add_bytes(b, g, width);
441 s->llviddsp.add_bytes(r, g, width);
451 static int build_huffman(AVCodecContext *avctx, GetBitContext *gbit, int max)
453 MagicYUVContext *s = avctx->priv_data;
456 while (get_bits_left(gbit) >= 8) {
457 int b = get_bits(gbit, 1);
458 int x = get_bits(gbit, 7);
459 int l = get_bitsz(gbit, b * 8) + 1;
463 av_log(avctx, AV_LOG_ERROR, "Invalid Huffman codes\n");
464 return AVERROR_INVALIDDATA;
472 if (s->huff_build(&s->vlc[i], s->len[i])) {
473 av_log(avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
474 return AVERROR_INVALIDDATA;
477 if (i == s->planes) {
483 if (i != s->planes) {
484 av_log(avctx, AV_LOG_ERROR, "Huffman tables too short\n");
485 return AVERROR_INVALIDDATA;
491 static int magy_decode_frame(AVCodecContext *avctx, void *data,
492 int *got_frame, AVPacket *avpkt)
494 MagicYUVContext *s = avctx->priv_data;
495 ThreadFrame frame = { .f = data };
497 GetByteContext gbyte;
499 uint32_t first_offset, offset, next_offset, header_size, slice_width;
500 int width, height, format, version, table_size;
503 bytestream2_init(&gbyte, avpkt->data, avpkt->size);
504 if (bytestream2_get_le32(&gbyte) != MKTAG('M', 'A', 'G', 'Y'))
505 return AVERROR_INVALIDDATA;
507 header_size = bytestream2_get_le32(&gbyte);
508 if (header_size < 32 || header_size >= avpkt->size) {
509 av_log(avctx, AV_LOG_ERROR,
510 "header or packet too small %"PRIu32"\n", header_size);
511 return AVERROR_INVALIDDATA;
514 version = bytestream2_get_byte(&gbyte);
516 avpriv_request_sample(avctx, "Version %d", version);
517 return AVERROR_PATCHWELCOME;
527 format = bytestream2_get_byte(&gbyte);
530 avctx->pix_fmt = AV_PIX_FMT_GBRP;
534 avctx->pix_fmt = AV_PIX_FMT_GBRAP;
538 avctx->pix_fmt = AV_PIX_FMT_YUV444P;
541 avctx->pix_fmt = AV_PIX_FMT_YUV422P;
546 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
553 avctx->pix_fmt = AV_PIX_FMT_YUVA444P;
556 avctx->pix_fmt = AV_PIX_FMT_GRAY8;
559 avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
565 avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
569 avctx->pix_fmt = AV_PIX_FMT_GBRP10;
574 avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
579 avctx->pix_fmt = AV_PIX_FMT_GBRP12;
584 avctx->pix_fmt = AV_PIX_FMT_GBRAP12;
589 avctx->pix_fmt = AV_PIX_FMT_GRAY10;
593 avpriv_request_sample(avctx, "Format 0x%X", format);
594 return AVERROR_PATCHWELCOME;
596 s->max = 1 << s->bps;
597 s->magy_decode_slice = s->bps == 8 ? magy_decode_slice : magy_decode_slice10;
599 s->huff_build = huff_build;
601 s->huff_build = s->bps == 10 ? huff_build10 : huff_build12;
602 s->planes = av_pix_fmt_count_planes(avctx->pix_fmt);
604 bytestream2_skip(&gbyte, 1);
605 s->color_matrix = bytestream2_get_byte(&gbyte);
606 s->flags = bytestream2_get_byte(&gbyte);
607 s->interlaced = !!(s->flags & 2);
608 bytestream2_skip(&gbyte, 3);
610 width = bytestream2_get_le32(&gbyte);
611 height = bytestream2_get_le32(&gbyte);
612 ret = ff_set_dimensions(avctx, width, height);
616 slice_width = bytestream2_get_le32(&gbyte);
617 if (slice_width != avctx->coded_width) {
618 avpriv_request_sample(avctx, "Slice width %"PRIu32, slice_width);
619 return AVERROR_PATCHWELCOME;
621 s->slice_height = bytestream2_get_le32(&gbyte);
622 if (s->slice_height <= 0 || s->slice_height > INT_MAX - avctx->coded_height) {
623 av_log(avctx, AV_LOG_ERROR,
624 "invalid slice height: %d\n", s->slice_height);
625 return AVERROR_INVALIDDATA;
628 bytestream2_skip(&gbyte, 4);
630 s->nb_slices = (avctx->coded_height + s->slice_height - 1) / s->slice_height;
631 if (s->nb_slices > INT_MAX / sizeof(Slice)) {
632 av_log(avctx, AV_LOG_ERROR,
633 "invalid number of slices: %d\n", s->nb_slices);
634 return AVERROR_INVALIDDATA;
638 if ((s->slice_height >> s->vshift[1]) < 2) {
639 av_log(avctx, AV_LOG_ERROR, "impossible slice height\n");
640 return AVERROR_INVALIDDATA;
642 if ((avctx->coded_height % s->slice_height) && ((avctx->coded_height % s->slice_height) >> s->vshift[1]) < 2) {
643 av_log(avctx, AV_LOG_ERROR, "impossible height\n");
644 return AVERROR_INVALIDDATA;
648 for (i = 0; i < s->planes; i++) {
649 av_fast_malloc(&s->slices[i], &s->slices_size[i], s->nb_slices * sizeof(Slice));
651 return AVERROR(ENOMEM);
653 offset = bytestream2_get_le32(&gbyte);
654 if (offset >= avpkt->size - header_size)
655 return AVERROR_INVALIDDATA;
658 first_offset = offset;
660 for (j = 0; j < s->nb_slices - 1; j++) {
661 s->slices[i][j].start = offset + header_size;
663 next_offset = bytestream2_get_le32(&gbyte);
664 if (next_offset <= offset || next_offset >= avpkt->size - header_size)
665 return AVERROR_INVALIDDATA;
667 s->slices[i][j].size = next_offset - offset;
668 offset = next_offset;
671 s->slices[i][j].start = offset + header_size;
672 s->slices[i][j].size = avpkt->size - s->slices[i][j].start;
675 if (bytestream2_get_byte(&gbyte) != s->planes)
676 return AVERROR_INVALIDDATA;
678 bytestream2_skip(&gbyte, s->nb_slices * s->planes);
680 table_size = header_size + first_offset - bytestream2_tell(&gbyte);
682 return AVERROR_INVALIDDATA;
684 ret = init_get_bits8(&gbit, avpkt->data + bytestream2_tell(&gbyte), table_size);
688 ret = build_huffman(avctx, &gbit, s->max);
692 p->pict_type = AV_PICTURE_TYPE_I;
695 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
698 s->buf = avpkt->data;
700 avctx->execute2(avctx, s->magy_decode_slice, NULL, NULL, s->nb_slices);
702 if (avctx->pix_fmt == AV_PIX_FMT_GBRP ||
703 avctx->pix_fmt == AV_PIX_FMT_GBRAP ||
704 avctx->pix_fmt == AV_PIX_FMT_GBRP10 ||
705 avctx->pix_fmt == AV_PIX_FMT_GBRAP10||
706 avctx->pix_fmt == AV_PIX_FMT_GBRAP12||
707 avctx->pix_fmt == AV_PIX_FMT_GBRP12) {
708 FFSWAP(uint8_t*, p->data[0], p->data[1]);
709 FFSWAP(int, p->linesize[0], p->linesize[1]);
711 switch (s->color_matrix) {
713 p->colorspace = AVCOL_SPC_BT470BG;
716 p->colorspace = AVCOL_SPC_BT709;
719 p->color_range = (s->flags & 4) ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
727 static av_cold int magy_decode_init(AVCodecContext *avctx)
729 MagicYUVContext *s = avctx->priv_data;
730 ff_llviddsp_init(&s->llviddsp);
734 static av_cold int magy_decode_end(AVCodecContext *avctx)
736 MagicYUVContext * const s = avctx->priv_data;
739 for (i = 0; i < FF_ARRAY_ELEMS(s->slices); i++) {
740 av_freep(&s->slices[i]);
741 s->slices_size[i] = 0;
742 ff_free_vlc(&s->vlc[i]);
748 AVCodec ff_magicyuv_decoder = {
750 .long_name = NULL_IF_CONFIG_SMALL("MagicYUV video"),
751 .type = AVMEDIA_TYPE_VIDEO,
752 .id = AV_CODEC_ID_MAGICYUV,
753 .priv_data_size = sizeof(MagicYUVContext),
754 .init = magy_decode_init,
755 .close = magy_decode_end,
756 .decode = magy_decode_frame,
757 .capabilities = AV_CODEC_CAP_DR1 |
758 AV_CODEC_CAP_FRAME_THREADS |
759 AV_CODEC_CAP_SLICE_THREADS,
760 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,