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 ctx->rows = av_mallocz_array(avctx->thread_count, sizeof(RowContext));
92 return AVERROR(ENOMEM);
97 static int dnxhd_init_vlc(DNXHDContext *ctx, uint32_t cid)
99 if (cid != ctx->cid) {
102 if ((index = ff_dnxhd_get_cid_table(cid)) < 0) {
103 av_log(ctx->avctx, AV_LOG_ERROR, "unsupported cid %d\n", cid);
104 return AVERROR(ENOSYS);
106 if (ff_dnxhd_cid_table[index].bit_depth != ctx->bit_depth) {
107 av_log(ctx->avctx, AV_LOG_ERROR, "bit depth mismatches %d %d\n", ff_dnxhd_cid_table[index].bit_depth, ctx->bit_depth);
108 return AVERROR_INVALIDDATA;
110 ctx->cid_table = &ff_dnxhd_cid_table[index];
111 av_log(ctx->avctx, AV_LOG_VERBOSE, "Profile cid %d.\n", cid);
113 ff_free_vlc(&ctx->ac_vlc);
114 ff_free_vlc(&ctx->dc_vlc);
115 ff_free_vlc(&ctx->run_vlc);
117 init_vlc(&ctx->ac_vlc, DNXHD_VLC_BITS, 257,
118 ctx->cid_table->ac_bits, 1, 1,
119 ctx->cid_table->ac_codes, 2, 2, 0);
120 init_vlc(&ctx->dc_vlc, DNXHD_DC_VLC_BITS, ctx->bit_depth + 4,
121 ctx->cid_table->dc_bits, 1, 1,
122 ctx->cid_table->dc_codes, 1, 1, 0);
123 init_vlc(&ctx->run_vlc, DNXHD_VLC_BITS, 62,
124 ctx->cid_table->run_bits, 1, 1,
125 ctx->cid_table->run_codes, 2, 2, 0);
127 ff_init_scantable(ctx->idsp.idct_permutation, &ctx->scantable,
134 static av_cold int dnxhd_decode_init_thread_copy(AVCodecContext *avctx)
136 DNXHDContext *ctx = avctx->priv_data;
138 // make sure VLC tables will be loaded when cid is parsed
141 ctx->rows = av_mallocz_array(avctx->thread_count, sizeof(RowContext));
143 return AVERROR(ENOMEM);
148 static int dnxhd_decode_header(DNXHDContext *ctx, AVFrame *frame,
149 const uint8_t *buf, int buf_size,
152 static const uint8_t header_prefix[] = { 0x00, 0x00, 0x02, 0x80, 0x01 };
153 static const uint8_t header_prefix444[] = { 0x00, 0x00, 0x02, 0x80, 0x02 };
155 int old_bit_depth = ctx->bit_depth;
157 if (buf_size < 0x280) {
158 av_log(ctx->avctx, AV_LOG_ERROR,
159 "buffer too small (%d < 640).\n", buf_size);
160 return AVERROR_INVALIDDATA;
163 if (memcmp(buf, header_prefix, 5) && memcmp(buf, header_prefix444, 5)) {
164 av_log(ctx->avctx, AV_LOG_ERROR,
165 "unknown header 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X\n",
166 buf[0], buf[1], buf[2], buf[3], buf[4]);
167 return AVERROR_INVALIDDATA;
169 if (buf[5] & 2) { /* interlaced */
170 ctx->cur_field = buf[5] & 1;
171 frame->interlaced_frame = 1;
172 frame->top_field_first = first_field ^ ctx->cur_field;
173 av_log(ctx->avctx, AV_LOG_DEBUG,
174 "interlaced %d, cur field %d\n", buf[5] & 3, ctx->cur_field);
178 ctx->mbaff = buf[0x6] & 32;
180 ctx->height = AV_RB16(buf + 0x18);
181 ctx->width = AV_RB16(buf + 0x1a);
183 ff_dlog(ctx->avctx, "width %d, height %d\n", ctx->width, ctx->height);
185 if (buf[0x21] == 0x58) { /* 10 bit */
186 ctx->bit_depth = ctx->avctx->bits_per_raw_sample = 10;
188 if (buf[0x4] == 0x2) {
189 ctx->decode_dct_block = dnxhd_decode_dct_block_10_444;
190 ctx->pix_fmt = AV_PIX_FMT_YUV444P10;
193 ctx->decode_dct_block = dnxhd_decode_dct_block_10;
194 ctx->pix_fmt = AV_PIX_FMT_YUV422P10;
197 } else if (buf[0x21] == 0x38) { /* 8 bit */
198 ctx->bit_depth = ctx->avctx->bits_per_raw_sample = 8;
200 ctx->pix_fmt = AV_PIX_FMT_YUV422P;
202 ctx->decode_dct_block = dnxhd_decode_dct_block_8;
204 av_log(ctx->avctx, AV_LOG_ERROR,
205 "invalid bit depth value (%d).\n", buf[0x21]);
206 return AVERROR_INVALIDDATA;
208 if (ctx->bit_depth != old_bit_depth) {
209 ff_blockdsp_init(&ctx->bdsp, ctx->avctx);
210 ff_idctdsp_init(&ctx->idsp, ctx->avctx);
213 cid = AV_RB32(buf + 0x28);
214 ff_dlog(ctx->avctx, "compression id %d\n", cid);
216 if ((ret = dnxhd_init_vlc(ctx, cid)) < 0)
218 if (ctx->mbaff && ctx->cid_table->cid != 1260)
219 av_log(ctx->avctx, AV_LOG_WARNING,
220 "Adaptive MB interlace flag in an unsupported profile.\n");
222 ctx->act = buf[0x2C] & 7;
223 if (ctx->act && ctx->cid_table->cid != 1256)
224 av_log(ctx->avctx, AV_LOG_WARNING,
225 "Adaptive color transform in an unsupported profile.\n");
227 // make sure profile size constraints are respected
228 // DNx100 allows 1920->1440 and 1280->960 subsampling
229 if (ctx->width != ctx->cid_table->width) {
230 av_reduce(&ctx->avctx->sample_aspect_ratio.num,
231 &ctx->avctx->sample_aspect_ratio.den,
232 ctx->width, ctx->cid_table->width, 255);
233 ctx->width = ctx->cid_table->width;
236 if (buf_size < ctx->cid_table->coding_unit_size) {
237 av_log(ctx->avctx, AV_LOG_ERROR, "incorrect frame size (%d < %d).\n",
238 buf_size, ctx->cid_table->coding_unit_size);
239 return AVERROR_INVALIDDATA;
242 ctx->mb_width = ctx->width >> 4;
243 ctx->mb_height = buf[0x16d];
246 "mb width %d, mb height %d\n", ctx->mb_width, ctx->mb_height);
248 if ((ctx->height + 15) >> 4 == ctx->mb_height && frame->interlaced_frame)
251 if (ctx->mb_height > 68 ||
252 (ctx->mb_height << frame->interlaced_frame) > (ctx->height + 15) >> 4) {
253 av_log(ctx->avctx, AV_LOG_ERROR,
254 "mb height too big: %d\n", ctx->mb_height);
255 return AVERROR_INVALIDDATA;
258 for (i = 0; i < ctx->mb_height; i++) {
259 ctx->mb_scan_index[i] = AV_RB32(buf + 0x170 + (i << 2));
260 ff_dlog(ctx->avctx, "mb scan index %d\n", ctx->mb_scan_index[i]);
261 if (buf_size < ctx->mb_scan_index[i] + 0x280LL) {
262 av_log(ctx->avctx, AV_LOG_ERROR,
263 "invalid mb scan index (%d < %d).\n",
264 buf_size, ctx->mb_scan_index[i] + 0x280);
265 return AVERROR_INVALIDDATA;
272 static av_always_inline int dnxhd_decode_dct_block(const DNXHDContext *ctx,
279 int i, j, index1, index2, len, flags;
280 int level, component, sign;
282 const uint8_t *weight_matrix;
283 const uint8_t *ac_level = ctx->cid_table->ac_level;
284 const uint8_t *ac_flags = ctx->cid_table->ac_flags;
285 int16_t *block = row->blocks[n];
286 const int eob_index = ctx->cid_table->eob_index;
288 OPEN_READER(bs, &row->gb);
290 ctx->bdsp.clear_block(block);
294 component = 1 + (n & 1);
295 scale = row->chroma_scale;
296 weight_matrix = ctx->cid_table->chroma_weight;
299 scale = row->luma_scale;
300 weight_matrix = ctx->cid_table->luma_weight;
303 component = (n >> 1) % 3;
305 scale = row->chroma_scale;
306 weight_matrix = ctx->cid_table->chroma_weight;
308 scale = row->luma_scale;
309 weight_matrix = ctx->cid_table->luma_weight;
313 UPDATE_CACHE(bs, &row->gb);
314 GET_VLC(len, bs, &row->gb, ctx->dc_vlc.table, DNXHD_DC_VLC_BITS, 1);
316 level = GET_CACHE(bs, &row->gb);
317 LAST_SKIP_BITS(bs, &row->gb, len);
319 level = (NEG_USR32(sign ^ level, len) ^ sign) - sign;
320 row->last_dc[component] += level;
322 block[0] = row->last_dc[component];
326 UPDATE_CACHE(bs, &row->gb);
327 GET_VLC(index1, bs, &row->gb, ctx->ac_vlc.table,
330 while (index1 != eob_index) {
331 level = ac_level[index1];
332 flags = ac_flags[index1];
334 sign = SHOW_SBITS(bs, &row->gb, 1);
335 SKIP_BITS(bs, &row->gb, 1);
338 level += SHOW_UBITS(bs, &row->gb, index_bits) << 7;
339 SKIP_BITS(bs, &row->gb, index_bits);
343 UPDATE_CACHE(bs, &row->gb);
344 GET_VLC(index2, bs, &row->gb, ctx->run_vlc.table,
346 i += ctx->cid_table->run[index2];
350 av_log(ctx->avctx, AV_LOG_ERROR, "ac tex damaged %d, %d\n", n, i);
355 j = ctx->scantable.permutated[i];
357 if (level_bias < 32 || weight_matrix[i] != level_bias)
359 level >>= level_shift;
361 block[j] = (level ^ sign) - sign;
363 UPDATE_CACHE(bs, &row->gb);
364 GET_VLC(index1, bs, &row->gb, ctx->ac_vlc.table,
368 CLOSE_READER(bs, &row->gb);
372 static int dnxhd_decode_dct_block_8(const DNXHDContext *ctx,
373 RowContext *row, int n)
375 return dnxhd_decode_dct_block(ctx, row, n, 4, 32, 6);
378 static int dnxhd_decode_dct_block_10(const DNXHDContext *ctx,
379 RowContext *row, int n)
381 return dnxhd_decode_dct_block(ctx, row, n, 6, 8, 4);
384 static int dnxhd_decode_dct_block_10_444(const DNXHDContext *ctx,
385 RowContext *row, int n)
387 return dnxhd_decode_dct_block(ctx, row, n, 6, 32, 6);
390 static int dnxhd_decode_macroblock(const DNXHDContext *ctx, RowContext *row,
391 AVFrame *frame, int x, int y)
393 int shift1 = ctx->bit_depth == 10;
394 int dct_linesize_luma = frame->linesize[0];
395 int dct_linesize_chroma = frame->linesize[1];
396 uint8_t *dest_y, *dest_u, *dest_v;
397 int dct_y_offset, dct_x_offset;
399 int interlaced_mb = 0;
402 interlaced_mb = get_bits1(&row->gb);
403 qscale = get_bits(&row->gb, 10);
405 qscale = get_bits(&row->gb, 11);
406 act = get_bits1(&row->gb);
408 static int warned = 0;
411 av_log(ctx->avctx, AV_LOG_ERROR,
412 "Unsupported adaptive color transform, patch welcome.\n");
416 if (qscale != row->last_qscale) {
417 for (i = 0; i < 64; i++) {
418 row->luma_scale[i] = qscale * ctx->cid_table->luma_weight[i];
419 row->chroma_scale[i] = qscale * ctx->cid_table->chroma_weight[i];
421 row->last_qscale = qscale;
424 for (i = 0; i < 8 + 4 * ctx->is_444; i++) {
425 if (ctx->decode_dct_block(ctx, row, i) < 0)
426 return AVERROR_INVALIDDATA;
429 if (frame->interlaced_frame) {
430 dct_linesize_luma <<= 1;
431 dct_linesize_chroma <<= 1;
434 dest_y = frame->data[0] + ((y * dct_linesize_luma) << 4) + (x << (4 + shift1));
435 dest_u = frame->data[1] + ((y * dct_linesize_chroma) << 4) + (x << (3 + shift1 + ctx->is_444));
436 dest_v = frame->data[2] + ((y * dct_linesize_chroma) << 4) + (x << (3 + shift1 + ctx->is_444));
438 if (frame->interlaced_frame && ctx->cur_field) {
439 dest_y += frame->linesize[0];
440 dest_u += frame->linesize[1];
441 dest_v += frame->linesize[2];
444 dct_linesize_luma <<= 1;
445 dct_linesize_chroma <<= 1;
448 dct_y_offset = interlaced_mb ? frame->linesize[0] : (dct_linesize_luma << 3);
449 dct_x_offset = 8 << shift1;
451 ctx->idsp.idct_put(dest_y, dct_linesize_luma, row->blocks[0]);
452 ctx->idsp.idct_put(dest_y + dct_x_offset, dct_linesize_luma, row->blocks[1]);
453 ctx->idsp.idct_put(dest_y + dct_y_offset, dct_linesize_luma, row->blocks[4]);
454 ctx->idsp.idct_put(dest_y + dct_y_offset + dct_x_offset, dct_linesize_luma, row->blocks[5]);
456 if (!(ctx->avctx->flags & AV_CODEC_FLAG_GRAY)) {
457 dct_y_offset = interlaced_mb ? frame->linesize[1] : (dct_linesize_chroma << 3);
458 ctx->idsp.idct_put(dest_u, dct_linesize_chroma, row->blocks[2]);
459 ctx->idsp.idct_put(dest_v, dct_linesize_chroma, row->blocks[3]);
460 ctx->idsp.idct_put(dest_u + dct_y_offset, dct_linesize_chroma, row->blocks[6]);
461 ctx->idsp.idct_put(dest_v + dct_y_offset, dct_linesize_chroma, row->blocks[7]);
464 ctx->idsp.idct_put(dest_y, dct_linesize_luma, row->blocks[0]);
465 ctx->idsp.idct_put(dest_y + dct_x_offset, dct_linesize_luma, row->blocks[1]);
466 ctx->idsp.idct_put(dest_y + dct_y_offset, dct_linesize_luma, row->blocks[6]);
467 ctx->idsp.idct_put(dest_y + dct_y_offset + dct_x_offset, dct_linesize_luma, row->blocks[7]);
469 if (!(ctx->avctx->flags & AV_CODEC_FLAG_GRAY)) {
470 dct_y_offset = interlaced_mb ? frame->linesize[1] : (dct_linesize_chroma << 3);
471 ctx->idsp.idct_put(dest_u, dct_linesize_chroma, row->blocks[2]);
472 ctx->idsp.idct_put(dest_u + dct_x_offset, dct_linesize_chroma, row->blocks[3]);
473 ctx->idsp.idct_put(dest_u + dct_y_offset, dct_linesize_chroma, row->blocks[8]);
474 ctx->idsp.idct_put(dest_u + dct_y_offset + dct_x_offset, dct_linesize_chroma, row->blocks[9]);
475 ctx->idsp.idct_put(dest_v, dct_linesize_chroma, row->blocks[4]);
476 ctx->idsp.idct_put(dest_v + dct_x_offset, dct_linesize_chroma, row->blocks[5]);
477 ctx->idsp.idct_put(dest_v + dct_y_offset, dct_linesize_chroma, row->blocks[10]);
478 ctx->idsp.idct_put(dest_v + dct_y_offset + dct_x_offset, dct_linesize_chroma, row->blocks[11]);
485 static int dnxhd_decode_row(AVCodecContext *avctx, void *data,
486 int rownb, int threadnb)
488 const DNXHDContext *ctx = avctx->priv_data;
489 uint32_t offset = ctx->mb_scan_index[rownb];
490 RowContext *row = ctx->rows + threadnb;
495 row->last_dc[2] = 1 << (ctx->bit_depth + 2); // for levels +2^(bitdepth-1)
496 init_get_bits(&row->gb, ctx->buf + offset, (ctx->buf_size - offset) << 3);
497 for (x = 0; x < ctx->mb_width; x++) {
499 int ret = dnxhd_decode_macroblock(ctx, row, data, x, rownb);
504 //STOP_TIMER("decode macroblock");
510 static int dnxhd_decode_frame(AVCodecContext *avctx, void *data,
511 int *got_frame, AVPacket *avpkt)
513 const uint8_t *buf = avpkt->data;
514 int buf_size = avpkt->size;
515 DNXHDContext *ctx = avctx->priv_data;
516 ThreadFrame frame = { .f = data };
517 AVFrame *picture = data;
521 ff_dlog(avctx, "frame size %d\n", buf_size);
524 if ((ret = dnxhd_decode_header(ctx, picture, buf, buf_size, first_field)) < 0)
527 if ((avctx->width || avctx->height) &&
528 (ctx->width != avctx->width || ctx->height != avctx->height)) {
529 av_log(avctx, AV_LOG_WARNING, "frame size changed: %dx%d -> %dx%d\n",
530 avctx->width, avctx->height, ctx->width, ctx->height);
533 if (avctx->pix_fmt != AV_PIX_FMT_NONE && avctx->pix_fmt != ctx->pix_fmt) {
534 av_log(avctx, AV_LOG_WARNING, "pix_fmt changed: %s -> %s\n",
535 av_get_pix_fmt_name(avctx->pix_fmt), av_get_pix_fmt_name(ctx->pix_fmt));
539 avctx->pix_fmt = ctx->pix_fmt;
540 ret = ff_set_dimensions(avctx, ctx->width, ctx->height);
545 if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
547 picture->pict_type = AV_PICTURE_TYPE_I;
548 picture->key_frame = 1;
551 ctx->buf_size = buf_size - 0x280;
552 ctx->buf = buf + 0x280;
553 avctx->execute2(avctx, dnxhd_decode_row, picture, NULL, ctx->mb_height);
555 if (first_field && picture->interlaced_frame) {
556 buf += ctx->cid_table->coding_unit_size;
557 buf_size -= ctx->cid_table->coding_unit_size;
559 goto decode_coding_unit;
563 for (i = 0; i < avctx->thread_count; i++) {
564 ret += ctx->rows[i].errors;
565 ctx->rows[i].errors = 0;
569 av_log(ctx->avctx, AV_LOG_ERROR, "%d lines with errors\n", ret);
570 return AVERROR_INVALIDDATA;
577 static av_cold int dnxhd_decode_close(AVCodecContext *avctx)
579 DNXHDContext *ctx = avctx->priv_data;
581 ff_free_vlc(&ctx->ac_vlc);
582 ff_free_vlc(&ctx->dc_vlc);
583 ff_free_vlc(&ctx->run_vlc);
585 av_freep(&ctx->rows);
590 AVCodec ff_dnxhd_decoder = {
592 .long_name = NULL_IF_CONFIG_SMALL("VC3/DNxHD"),
593 .type = AVMEDIA_TYPE_VIDEO,
594 .id = AV_CODEC_ID_DNXHD,
595 .priv_data_size = sizeof(DNXHDContext),
596 .init = dnxhd_decode_init,
597 .close = dnxhd_decode_close,
598 .decode = dnxhd_decode_frame,
599 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
600 AV_CODEC_CAP_SLICE_THREADS,
601 .init_thread_copy = ONLY_IF_THREADS_ENABLED(dnxhd_decode_init_thread_copy),