3 * Copyright (c) 2007 SmartJog S.A., Baptiste Coudurier <baptiste dot coudurier at smartjog dot com>
4 * Copyright (c) 2011 MirriAd Ltd
5 * Copyright (c) 2015 Christophe Gisquet
7 * 10 bit support added by MirriAd Ltd, Joseph Artsimovich <joseph@mirriad.com>
8 * Slice multithreading and MB interlaced support added by Christophe Gisquet
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libavutil/imgutils.h"
28 #include "libavutil/timer.h"
31 #define UNCHECKED_BITSTREAM_READER 1
33 #include "dnxhddata.h"
38 typedef struct RowContext {
39 DECLARE_ALIGNED(16, int16_t, blocks)[12][64];
48 typedef struct DNXHDContext {
49 AVCodecContext *avctx;
54 int64_t cid; ///< compression id
55 unsigned int width, height;
56 enum AVPixelFormat pix_fmt;
57 unsigned int mb_width, mb_height;
58 uint32_t mb_scan_index[68]; /* max for 1080p */
59 int cur_field; ///< current interlaced field
60 VLC ac_vlc, dc_vlc, run_vlc;
63 const CIDEntry *cid_table;
64 int bit_depth; // 8, 10 or 0 if not initialized at all.
68 int (*decode_dct_block)(const struct DNXHDContext *ctx,
69 RowContext *row, int n);
72 #define DNXHD_VLC_BITS 9
73 #define DNXHD_DC_VLC_BITS 7
75 static int dnxhd_decode_dct_block_8(const DNXHDContext *ctx,
76 RowContext *row, int n);
77 static int dnxhd_decode_dct_block_10(const DNXHDContext *ctx,
78 RowContext *row, int n);
79 static int dnxhd_decode_dct_block_10_444(const DNXHDContext *ctx,
80 RowContext *row, int n);
82 static av_cold int dnxhd_decode_init(AVCodecContext *avctx)
84 DNXHDContext *ctx = avctx->priv_data;
88 avctx->colorspace = AVCOL_SPC_BT709;
90 avctx->coded_width = FFALIGN(avctx->width, 16);
91 avctx->coded_height = FFALIGN(avctx->height, 16);
93 ctx->rows = av_mallocz_array(avctx->thread_count, sizeof(RowContext));
95 return AVERROR(ENOMEM);
100 static int dnxhd_init_vlc(DNXHDContext *ctx, uint32_t cid)
102 if (cid != ctx->cid) {
105 if ((index = ff_dnxhd_get_cid_table(cid)) < 0) {
106 av_log(ctx->avctx, AV_LOG_ERROR, "unsupported cid %d\n", cid);
107 return AVERROR(ENOSYS);
109 if (ff_dnxhd_cid_table[index].bit_depth != ctx->bit_depth) {
110 av_log(ctx->avctx, AV_LOG_ERROR, "bit depth mismatches %d %d\n", ff_dnxhd_cid_table[index].bit_depth, ctx->bit_depth);
111 return AVERROR_INVALIDDATA;
113 ctx->cid_table = &ff_dnxhd_cid_table[index];
114 av_log(ctx->avctx, AV_LOG_VERBOSE, "Profile cid %d.\n", cid);
116 ff_free_vlc(&ctx->ac_vlc);
117 ff_free_vlc(&ctx->dc_vlc);
118 ff_free_vlc(&ctx->run_vlc);
120 init_vlc(&ctx->ac_vlc, DNXHD_VLC_BITS, 257,
121 ctx->cid_table->ac_bits, 1, 1,
122 ctx->cid_table->ac_codes, 2, 2, 0);
123 init_vlc(&ctx->dc_vlc, DNXHD_DC_VLC_BITS, ctx->bit_depth + 4,
124 ctx->cid_table->dc_bits, 1, 1,
125 ctx->cid_table->dc_codes, 1, 1, 0);
126 init_vlc(&ctx->run_vlc, DNXHD_VLC_BITS, 62,
127 ctx->cid_table->run_bits, 1, 1,
128 ctx->cid_table->run_codes, 2, 2, 0);
135 static av_cold int dnxhd_decode_init_thread_copy(AVCodecContext *avctx)
137 DNXHDContext *ctx = avctx->priv_data;
139 // make sure VLC tables will be loaded when cid is parsed
142 ctx->rows = av_mallocz_array(avctx->thread_count, sizeof(RowContext));
144 return AVERROR(ENOMEM);
149 static int dnxhd_decode_header(DNXHDContext *ctx, AVFrame *frame,
150 const uint8_t *buf, int buf_size,
153 static const uint8_t header_prefix[] = { 0x00, 0x00, 0x02, 0x80, 0x01 };
154 static const uint8_t header_prefix444[] = { 0x00, 0x00, 0x02, 0x80, 0x02 };
156 int old_bit_depth = ctx->bit_depth;
158 if (buf_size < 0x280) {
159 av_log(ctx->avctx, AV_LOG_ERROR,
160 "buffer too small (%d < 640).\n", buf_size);
161 return AVERROR_INVALIDDATA;
164 if (memcmp(buf, header_prefix, 5) && memcmp(buf, header_prefix444, 5)) {
165 av_log(ctx->avctx, AV_LOG_ERROR,
166 "unknown header 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X\n",
167 buf[0], buf[1], buf[2], buf[3], buf[4]);
168 return AVERROR_INVALIDDATA;
170 if (buf[5] & 2) { /* interlaced */
171 ctx->cur_field = buf[5] & 1;
172 frame->interlaced_frame = 1;
173 frame->top_field_first = first_field ^ ctx->cur_field;
174 av_log(ctx->avctx, AV_LOG_DEBUG,
175 "interlaced %d, cur field %d\n", buf[5] & 3, ctx->cur_field);
179 ctx->mbaff = buf[0x6] & 32;
181 ctx->height = AV_RB16(buf + 0x18);
182 ctx->width = AV_RB16(buf + 0x1a);
184 ff_dlog(ctx->avctx, "width %d, height %d\n", ctx->width, ctx->height);
186 if (buf[0x21] == 0x58) { /* 10 bit */
187 ctx->bit_depth = ctx->avctx->bits_per_raw_sample = 10;
189 if (buf[0x4] == 0x2) {
190 ctx->decode_dct_block = dnxhd_decode_dct_block_10_444;
191 ctx->pix_fmt = AV_PIX_FMT_YUV444P10;
194 ctx->decode_dct_block = dnxhd_decode_dct_block_10;
195 ctx->pix_fmt = AV_PIX_FMT_YUV422P10;
198 } else if (buf[0x21] == 0x38) { /* 8 bit */
199 ctx->bit_depth = ctx->avctx->bits_per_raw_sample = 8;
201 ctx->pix_fmt = AV_PIX_FMT_YUV422P;
203 ctx->decode_dct_block = dnxhd_decode_dct_block_8;
205 av_log(ctx->avctx, AV_LOG_ERROR,
206 "invalid bit depth value (%d).\n", buf[0x21]);
207 return AVERROR_INVALIDDATA;
209 if (ctx->bit_depth != old_bit_depth) {
210 ff_blockdsp_init(&ctx->bdsp, ctx->avctx);
211 ff_idctdsp_init(&ctx->idsp, ctx->avctx);
212 ff_init_scantable(ctx->idsp.idct_permutation, &ctx->scantable,
216 cid = AV_RB32(buf + 0x28);
217 ff_dlog(ctx->avctx, "compression id %d\n", cid);
219 if ((ret = dnxhd_init_vlc(ctx, cid)) < 0)
221 if (ctx->mbaff && ctx->cid_table->cid != 1260)
222 av_log(ctx->avctx, AV_LOG_WARNING,
223 "Adaptive MB interlace flag in an unsupported profile.\n");
225 ctx->act = buf[0x2C] & 7;
226 if (ctx->act && ctx->cid_table->cid != 1256)
227 av_log(ctx->avctx, AV_LOG_WARNING,
228 "Adaptive color transform in an unsupported profile.\n");
230 // make sure profile size constraints are respected
231 // DNx100 allows 1920->1440 and 1280->960 subsampling
232 if (ctx->width != ctx->cid_table->width) {
233 av_reduce(&ctx->avctx->sample_aspect_ratio.num,
234 &ctx->avctx->sample_aspect_ratio.den,
235 ctx->width, ctx->cid_table->width, 255);
236 ctx->width = ctx->cid_table->width;
239 if (buf_size < ctx->cid_table->coding_unit_size) {
240 av_log(ctx->avctx, AV_LOG_ERROR, "incorrect frame size (%d < %d).\n",
241 buf_size, ctx->cid_table->coding_unit_size);
242 return AVERROR_INVALIDDATA;
245 ctx->mb_width = ctx->width >> 4;
246 ctx->mb_height = buf[0x16d];
249 "mb width %d, mb height %d\n", ctx->mb_width, ctx->mb_height);
251 if ((ctx->height + 15) >> 4 == ctx->mb_height && frame->interlaced_frame)
254 if (ctx->mb_height > 68 ||
255 (ctx->mb_height << frame->interlaced_frame) > (ctx->height + 15) >> 4) {
256 av_log(ctx->avctx, AV_LOG_ERROR,
257 "mb height too big: %d\n", ctx->mb_height);
258 return AVERROR_INVALIDDATA;
261 for (i = 0; i < ctx->mb_height; i++) {
262 ctx->mb_scan_index[i] = AV_RB32(buf + 0x170 + (i << 2));
263 ff_dlog(ctx->avctx, "mb scan index %d\n", ctx->mb_scan_index[i]);
264 if (buf_size < ctx->mb_scan_index[i] + 0x280LL) {
265 av_log(ctx->avctx, AV_LOG_ERROR,
266 "invalid mb scan index (%d < %d).\n",
267 buf_size, ctx->mb_scan_index[i] + 0x280);
268 return AVERROR_INVALIDDATA;
275 static av_always_inline int dnxhd_decode_dct_block(const DNXHDContext *ctx,
282 int i, j, index1, index2, len, flags;
283 int level, component, sign;
285 const uint8_t *weight_matrix;
286 const uint8_t *ac_level = ctx->cid_table->ac_level;
287 const uint8_t *ac_flags = ctx->cid_table->ac_flags;
288 int16_t *block = row->blocks[n];
289 const int eob_index = ctx->cid_table->eob_index;
291 OPEN_READER(bs, &row->gb);
293 ctx->bdsp.clear_block(block);
297 component = 1 + (n & 1);
298 scale = row->chroma_scale;
299 weight_matrix = ctx->cid_table->chroma_weight;
302 scale = row->luma_scale;
303 weight_matrix = ctx->cid_table->luma_weight;
306 component = (n >> 1) % 3;
308 scale = row->chroma_scale;
309 weight_matrix = ctx->cid_table->chroma_weight;
311 scale = row->luma_scale;
312 weight_matrix = ctx->cid_table->luma_weight;
316 UPDATE_CACHE(bs, &row->gb);
317 GET_VLC(len, bs, &row->gb, ctx->dc_vlc.table, DNXHD_DC_VLC_BITS, 1);
319 level = GET_CACHE(bs, &row->gb);
320 LAST_SKIP_BITS(bs, &row->gb, len);
322 level = (NEG_USR32(sign ^ level, len) ^ sign) - sign;
323 row->last_dc[component] += level;
325 block[0] = row->last_dc[component];
329 UPDATE_CACHE(bs, &row->gb);
330 GET_VLC(index1, bs, &row->gb, ctx->ac_vlc.table,
333 while (index1 != eob_index) {
334 level = ac_level[index1];
335 flags = ac_flags[index1];
337 sign = SHOW_SBITS(bs, &row->gb, 1);
338 SKIP_BITS(bs, &row->gb, 1);
341 level += SHOW_UBITS(bs, &row->gb, index_bits) << 7;
342 SKIP_BITS(bs, &row->gb, index_bits);
346 UPDATE_CACHE(bs, &row->gb);
347 GET_VLC(index2, bs, &row->gb, ctx->run_vlc.table,
349 i += ctx->cid_table->run[index2];
353 av_log(ctx->avctx, AV_LOG_ERROR, "ac tex damaged %d, %d\n", n, i);
358 j = ctx->scantable.permutated[i];
360 level += scale[i] >> 1;
361 if (level_bias < 32 || weight_matrix[i] != level_bias)
362 level += level_bias; // 1<<(level_shift-1)
363 level >>= level_shift;
365 block[j] = (level ^ sign) - sign;
367 UPDATE_CACHE(bs, &row->gb);
368 GET_VLC(index1, bs, &row->gb, ctx->ac_vlc.table,
372 CLOSE_READER(bs, &row->gb);
376 static int dnxhd_decode_dct_block_8(const DNXHDContext *ctx,
377 RowContext *row, int n)
379 return dnxhd_decode_dct_block(ctx, row, n, 4, 32, 6);
382 static int dnxhd_decode_dct_block_10(const DNXHDContext *ctx,
383 RowContext *row, int n)
385 return dnxhd_decode_dct_block(ctx, row, n, 6, 8, 4);
388 static int dnxhd_decode_dct_block_10_444(const DNXHDContext *ctx,
389 RowContext *row, int n)
391 return dnxhd_decode_dct_block(ctx, row, n, 6, 32, 6);
394 static int dnxhd_decode_macroblock(const DNXHDContext *ctx, RowContext *row,
395 AVFrame *frame, int x, int y)
397 int shift1 = ctx->bit_depth == 10;
398 int dct_linesize_luma = frame->linesize[0];
399 int dct_linesize_chroma = frame->linesize[1];
400 uint8_t *dest_y, *dest_u, *dest_v;
401 int dct_y_offset, dct_x_offset;
403 int interlaced_mb = 0;
406 interlaced_mb = get_bits1(&row->gb);
407 qscale = get_bits(&row->gb, 10);
409 qscale = get_bits(&row->gb, 11);
411 act = get_bits1(&row->gb);
413 static int warned = 0;
416 av_log(ctx->avctx, AV_LOG_ERROR,
417 "Unsupported adaptive color transform, patch welcome.\n");
421 if (qscale != row->last_qscale) {
422 for (i = 0; i < 64; i++) {
423 row->luma_scale[i] = qscale * ctx->cid_table->luma_weight[i];
424 row->chroma_scale[i] = qscale * ctx->cid_table->chroma_weight[i];
426 row->last_qscale = qscale;
429 for (i = 0; i < 8 + 4 * ctx->is_444; i++) {
430 if (ctx->decode_dct_block(ctx, row, i) < 0)
431 return AVERROR_INVALIDDATA;
434 if (frame->interlaced_frame) {
435 dct_linesize_luma <<= 1;
436 dct_linesize_chroma <<= 1;
439 dest_y = frame->data[0] + ((y * dct_linesize_luma) << 4) + (x << (4 + shift1));
440 dest_u = frame->data[1] + ((y * dct_linesize_chroma) << 4) + (x << (3 + shift1 + ctx->is_444));
441 dest_v = frame->data[2] + ((y * dct_linesize_chroma) << 4) + (x << (3 + shift1 + ctx->is_444));
443 if (frame->interlaced_frame && ctx->cur_field) {
444 dest_y += frame->linesize[0];
445 dest_u += frame->linesize[1];
446 dest_v += frame->linesize[2];
449 dct_linesize_luma <<= 1;
450 dct_linesize_chroma <<= 1;
453 dct_y_offset = interlaced_mb ? frame->linesize[0] : (dct_linesize_luma << 3);
454 dct_x_offset = 8 << shift1;
456 ctx->idsp.idct_put(dest_y, dct_linesize_luma, row->blocks[0]);
457 ctx->idsp.idct_put(dest_y + dct_x_offset, dct_linesize_luma, row->blocks[1]);
458 ctx->idsp.idct_put(dest_y + dct_y_offset, dct_linesize_luma, row->blocks[4]);
459 ctx->idsp.idct_put(dest_y + dct_y_offset + dct_x_offset, dct_linesize_luma, row->blocks[5]);
461 if (!(ctx->avctx->flags & AV_CODEC_FLAG_GRAY)) {
462 dct_y_offset = interlaced_mb ? frame->linesize[1] : (dct_linesize_chroma << 3);
463 ctx->idsp.idct_put(dest_u, dct_linesize_chroma, row->blocks[2]);
464 ctx->idsp.idct_put(dest_v, dct_linesize_chroma, row->blocks[3]);
465 ctx->idsp.idct_put(dest_u + dct_y_offset, dct_linesize_chroma, row->blocks[6]);
466 ctx->idsp.idct_put(dest_v + dct_y_offset, dct_linesize_chroma, row->blocks[7]);
469 ctx->idsp.idct_put(dest_y, dct_linesize_luma, row->blocks[0]);
470 ctx->idsp.idct_put(dest_y + dct_x_offset, dct_linesize_luma, row->blocks[1]);
471 ctx->idsp.idct_put(dest_y + dct_y_offset, dct_linesize_luma, row->blocks[6]);
472 ctx->idsp.idct_put(dest_y + dct_y_offset + dct_x_offset, dct_linesize_luma, row->blocks[7]);
474 if (!(ctx->avctx->flags & AV_CODEC_FLAG_GRAY)) {
475 dct_y_offset = interlaced_mb ? frame->linesize[1] : (dct_linesize_chroma << 3);
476 ctx->idsp.idct_put(dest_u, dct_linesize_chroma, row->blocks[2]);
477 ctx->idsp.idct_put(dest_u + dct_x_offset, dct_linesize_chroma, row->blocks[3]);
478 ctx->idsp.idct_put(dest_u + dct_y_offset, dct_linesize_chroma, row->blocks[8]);
479 ctx->idsp.idct_put(dest_u + dct_y_offset + dct_x_offset, dct_linesize_chroma, row->blocks[9]);
480 ctx->idsp.idct_put(dest_v, dct_linesize_chroma, row->blocks[4]);
481 ctx->idsp.idct_put(dest_v + dct_x_offset, dct_linesize_chroma, row->blocks[5]);
482 ctx->idsp.idct_put(dest_v + dct_y_offset, dct_linesize_chroma, row->blocks[10]);
483 ctx->idsp.idct_put(dest_v + dct_y_offset + dct_x_offset, dct_linesize_chroma, row->blocks[11]);
490 static int dnxhd_decode_row(AVCodecContext *avctx, void *data,
491 int rownb, int threadnb)
493 const DNXHDContext *ctx = avctx->priv_data;
494 uint32_t offset = ctx->mb_scan_index[rownb];
495 RowContext *row = ctx->rows + threadnb;
500 row->last_dc[2] = 1 << (ctx->bit_depth + 2); // for levels +2^(bitdepth-1)
501 init_get_bits(&row->gb, ctx->buf + offset, (ctx->buf_size - offset) << 3);
502 for (x = 0; x < ctx->mb_width; x++) {
504 int ret = dnxhd_decode_macroblock(ctx, row, data, x, rownb);
509 //STOP_TIMER("decode macroblock");
515 static int dnxhd_decode_frame(AVCodecContext *avctx, void *data,
516 int *got_frame, AVPacket *avpkt)
518 const uint8_t *buf = avpkt->data;
519 int buf_size = avpkt->size;
520 DNXHDContext *ctx = avctx->priv_data;
521 ThreadFrame frame = { .f = data };
522 AVFrame *picture = data;
526 ff_dlog(avctx, "frame size %d\n", buf_size);
529 if ((ret = dnxhd_decode_header(ctx, picture, buf, buf_size, first_field)) < 0)
532 if ((avctx->width || avctx->height) &&
533 (ctx->width != avctx->width || ctx->height != avctx->height)) {
534 av_log(avctx, AV_LOG_WARNING, "frame size changed: %dx%d -> %dx%d\n",
535 avctx->width, avctx->height, ctx->width, ctx->height);
538 if (avctx->pix_fmt != AV_PIX_FMT_NONE && avctx->pix_fmt != ctx->pix_fmt) {
539 av_log(avctx, AV_LOG_WARNING, "pix_fmt changed: %s -> %s\n",
540 av_get_pix_fmt_name(avctx->pix_fmt), av_get_pix_fmt_name(ctx->pix_fmt));
544 avctx->pix_fmt = ctx->pix_fmt;
545 ret = ff_set_dimensions(avctx, ctx->width, ctx->height);
550 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
552 picture->pict_type = AV_PICTURE_TYPE_I;
553 picture->key_frame = 1;
556 ctx->buf_size = buf_size - 0x280;
557 ctx->buf = buf + 0x280;
558 avctx->execute2(avctx, dnxhd_decode_row, picture, NULL, ctx->mb_height);
560 if (first_field && picture->interlaced_frame) {
561 buf += ctx->cid_table->coding_unit_size;
562 buf_size -= ctx->cid_table->coding_unit_size;
564 goto decode_coding_unit;
568 for (i = 0; i < avctx->thread_count; i++) {
569 ret += ctx->rows[i].errors;
570 ctx->rows[i].errors = 0;
574 av_log(ctx->avctx, AV_LOG_ERROR, "%d lines with errors\n", ret);
575 return AVERROR_INVALIDDATA;
582 static av_cold int dnxhd_decode_close(AVCodecContext *avctx)
584 DNXHDContext *ctx = avctx->priv_data;
586 ff_free_vlc(&ctx->ac_vlc);
587 ff_free_vlc(&ctx->dc_vlc);
588 ff_free_vlc(&ctx->run_vlc);
590 av_freep(&ctx->rows);
595 AVCodec ff_dnxhd_decoder = {
597 .long_name = NULL_IF_CONFIG_SMALL("VC3/DNxHD"),
598 .type = AVMEDIA_TYPE_VIDEO,
599 .id = AV_CODEC_ID_DNXHD,
600 .priv_data_size = sizeof(DNXHDContext),
601 .init = dnxhd_decode_init,
602 .close = dnxhd_decode_close,
603 .decode = dnxhd_decode_frame,
604 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
605 AV_CODEC_CAP_SLICE_THREADS,
606 .init_thread_copy = ONLY_IF_THREADS_ENABLED(dnxhd_decode_init_thread_copy),