vf_*_vaapi: Support increasing hardware frame pool size
[libav.git] / libavcodec / dnxhddec.c
1 /*
2 * VC3/DNxHD decoder.
3 * Copyright (c) 2007 SmartJog S.A., Baptiste Coudurier <baptiste dot coudurier at smartjog dot com>
4 * Copyright (c) 2011 MirriAd Ltd
5 *
6 * 10 bit support added by MirriAd Ltd, Joseph Artsimovich <joseph@mirriad.com>
7 *
8 * This file is part of Libav.
9 *
10 * Libav is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
14 *
15 * Libav is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with Libav; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25 #include "libavutil/imgutils.h"
26 #include "libavutil/timer.h"
27 #include "avcodec.h"
28 #include "blockdsp.h"
29 #include "get_bits.h"
30 #include "dnxhddata.h"
31 #include "idctdsp.h"
32 #include "internal.h"
33 #include "thread.h"
34
35 typedef struct DNXHDContext {
36 AVCodecContext *avctx;
37 GetBitContext gb;
38 BlockDSPContext bdsp;
39 int cid; ///< compression id
40 unsigned int width, height;
41 unsigned int mb_width, mb_height;
42 uint32_t mb_scan_index[68]; /* max for 1080p */
43 int cur_field; ///< current interlaced field
44 VLC ac_vlc, dc_vlc, run_vlc;
45 int last_dc[3];
46 IDCTDSPContext idsp;
47 DECLARE_ALIGNED(16, int16_t, blocks)[12][64];
48 ScanTable scantable;
49 const CIDEntry *cid_table;
50 int bit_depth; // 8, 10 or 0 if not initialized at all.
51 int is_444;
52 int mbaff;
53 void (*decode_dct_block)(struct DNXHDContext *ctx, int16_t *block,
54 int n, int qscale);
55 } DNXHDContext;
56
57 #define DNXHD_VLC_BITS 9
58 #define DNXHD_DC_VLC_BITS 7
59
60 static void dnxhd_decode_dct_block_8(DNXHDContext *ctx, int16_t *block,
61 int n, int qscale);
62 static void dnxhd_decode_dct_block_10(DNXHDContext *ctx, int16_t *block,
63 int n, int qscale);
64 static void dnxhd_decode_dct_block_10_444(DNXHDContext *ctx, int16_t *block,
65 int n, int qscale);
66
67 static av_cold int dnxhd_decode_init(AVCodecContext *avctx)
68 {
69 DNXHDContext *ctx = avctx->priv_data;
70
71 ctx->avctx = avctx;
72 return 0;
73 }
74
75 static int dnxhd_init_vlc(DNXHDContext *ctx, int cid)
76 {
77 if (cid != ctx->cid) {
78 int index;
79
80 if ((index = ff_dnxhd_get_cid_table(cid)) < 0) {
81 av_log(ctx->avctx, AV_LOG_ERROR, "unsupported cid %d\n", cid);
82 return AVERROR(ENOSYS);
83 }
84 ctx->cid_table = &ff_dnxhd_cid_table[index];
85 av_log(ctx->avctx, AV_LOG_VERBOSE, "Profile cid %d.\n", cid);
86
87 ff_free_vlc(&ctx->ac_vlc);
88 ff_free_vlc(&ctx->dc_vlc);
89 ff_free_vlc(&ctx->run_vlc);
90
91 init_vlc(&ctx->ac_vlc, DNXHD_VLC_BITS, 257,
92 ctx->cid_table->ac_bits, 1, 1,
93 ctx->cid_table->ac_codes, 2, 2, 0);
94 init_vlc(&ctx->dc_vlc, DNXHD_DC_VLC_BITS, ctx->bit_depth + 4,
95 ctx->cid_table->dc_bits, 1, 1,
96 ctx->cid_table->dc_codes, 1, 1, 0);
97 init_vlc(&ctx->run_vlc, DNXHD_VLC_BITS, 62,
98 ctx->cid_table->run_bits, 1, 1,
99 ctx->cid_table->run_codes, 2, 2, 0);
100
101 ff_init_scantable(ctx->idsp.idct_permutation, &ctx->scantable,
102 ff_zigzag_direct);
103 ctx->cid = cid;
104 }
105 return 0;
106 }
107
108 static int dnxhd_decode_header(DNXHDContext *ctx, AVFrame *frame,
109 const uint8_t *buf, int buf_size,
110 int first_field)
111 {
112 static const uint8_t header_prefix[] = { 0x00, 0x00, 0x02, 0x80, 0x01 };
113 static const uint8_t header_prefix444[] = { 0x00, 0x00, 0x02, 0x80, 0x02 };
114 int i, cid, ret;
115 int old_bit_depth = ctx->bit_depth;
116
117 if (buf_size < 0x280) {
118 av_log(ctx->avctx, AV_LOG_ERROR, "buffer too small (%d < 640).\n",
119 buf_size);
120 return AVERROR_INVALIDDATA;
121 }
122
123 if (memcmp(buf, header_prefix, 5) && memcmp(buf, header_prefix444, 5)) {
124 av_log(ctx->avctx, AV_LOG_ERROR,
125 "unknown header 0x%02"PRIX8" 0x%02"PRIX8" 0x%02"PRIX8" 0x%02"PRIX8" 0x%02"PRIX8"\n",
126 buf[0], buf[1], buf[2], buf[3], buf[4]);
127 return AVERROR_INVALIDDATA;
128 }
129 if (buf[5] & 2) { /* interlaced */
130 ctx->cur_field = buf[5] & 1;
131 frame->interlaced_frame = 1;
132 frame->top_field_first = first_field ^ ctx->cur_field;
133 av_log(ctx->avctx, AV_LOG_DEBUG,
134 "interlaced %"PRId8", cur field %d\n", buf[5] & 3, ctx->cur_field);
135 }
136 ctx->mbaff = buf[0x6] & 32;
137
138 ctx->height = AV_RB16(buf + 0x18);
139 ctx->width = AV_RB16(buf + 0x1a);
140
141 ff_dlog(ctx->avctx, "width %u, height %u\n", ctx->width, ctx->height);
142
143 if (buf[0x21] == 0x58) { /* 10 bit */
144 ctx->bit_depth = ctx->avctx->bits_per_raw_sample = 10;
145
146 if (buf[0x4] == 0x2) {
147 ctx->decode_dct_block = dnxhd_decode_dct_block_10_444;
148 ctx->avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
149 ctx->is_444 = 1;
150 } else {
151 ctx->decode_dct_block = dnxhd_decode_dct_block_10;
152 ctx->avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
153 }
154 } else if (buf[0x21] == 0x38) { /* 8 bit */
155 ctx->bit_depth = ctx->avctx->bits_per_raw_sample = 8;
156
157 ctx->avctx->pix_fmt = AV_PIX_FMT_YUV422P;
158 ctx->decode_dct_block = dnxhd_decode_dct_block_8;
159 } else {
160 av_log(ctx->avctx, AV_LOG_ERROR, "invalid bit depth value (%"PRId8").\n",
161 buf[0x21]);
162 return AVERROR_INVALIDDATA;
163 }
164 if (ctx->bit_depth != old_bit_depth) {
165 ff_blockdsp_init(&ctx->bdsp);
166 ff_idctdsp_init(&ctx->idsp, ctx->avctx);
167 }
168
169 cid = AV_RB32(buf + 0x28);
170 ff_dlog(ctx->avctx, "compression id %d\n", cid);
171
172 if ((ret = dnxhd_init_vlc(ctx, cid)) < 0)
173 return ret;
174 if (ctx->mbaff && ctx->cid_table->cid != 1260)
175 av_log(ctx->avctx, AV_LOG_WARNING,
176 "Adaptive MB interlace flag in an unsupported profile.\n");
177
178 // make sure profile size constraints are respected
179 // DNx100 allows 1920->1440 and 1280->960 subsampling
180 if (ctx->width != ctx->cid_table->width) {
181 av_reduce(&ctx->avctx->sample_aspect_ratio.num,
182 &ctx->avctx->sample_aspect_ratio.den,
183 ctx->width, ctx->cid_table->width, 255);
184 ctx->width = ctx->cid_table->width;
185 }
186
187 if (buf_size < ctx->cid_table->coding_unit_size) {
188 av_log(ctx->avctx, AV_LOG_ERROR, "incorrect frame size (%d < %u).\n",
189 buf_size, ctx->cid_table->coding_unit_size);
190 return AVERROR_INVALIDDATA;
191 }
192
193 ctx->mb_width = ctx->width >> 4;
194 ctx->mb_height = buf[0x16d];
195
196 ff_dlog(ctx->avctx,
197 "mb width %u, mb height %u\n", ctx->mb_width, ctx->mb_height);
198
199 if ((ctx->height + 15) >> 4 == ctx->mb_height && frame->interlaced_frame)
200 ctx->height <<= 1;
201
202 if (ctx->mb_height > 68 ||
203 (ctx->mb_height << frame->interlaced_frame) > (ctx->height + 15) >> 4) {
204 av_log(ctx->avctx, AV_LOG_ERROR,
205 "mb height too big: %u\n", ctx->mb_height);
206 return AVERROR_INVALIDDATA;
207 }
208
209 for (i = 0; i < ctx->mb_height; i++) {
210 ctx->mb_scan_index[i] = AV_RB32(buf + 0x170 + (i << 2));
211 ff_dlog(ctx->avctx, "mb scan index %"PRIu32"\n", ctx->mb_scan_index[i]);
212 if (buf_size < ctx->mb_scan_index[i] + 0x280) {
213 av_log(ctx->avctx, AV_LOG_ERROR,
214 "invalid mb scan index (%d < %"PRIu32").\n",
215 buf_size, ctx->mb_scan_index[i] + 0x280);
216 return AVERROR_INVALIDDATA;
217 }
218 }
219
220 return 0;
221 }
222
223 static av_always_inline void dnxhd_decode_dct_block(DNXHDContext *ctx,
224 int16_t *block, int n,
225 int qscale,
226 int index_bits,
227 int level_bias,
228 int level_shift)
229 {
230 int i, j, index1, index2, len;
231 int level, component, sign;
232 const uint8_t *weight_matrix;
233 OPEN_READER(bs, &ctx->gb);
234
235 if (!ctx->is_444) {
236 if (n & 2) {
237 component = 1 + (n & 1);
238 weight_matrix = ctx->cid_table->chroma_weight;
239 } else {
240 component = 0;
241 weight_matrix = ctx->cid_table->luma_weight;
242 }
243 } else {
244 component = (n >> 1) % 3;
245 if (component) {
246 weight_matrix = ctx->cid_table->chroma_weight;
247 } else {
248 weight_matrix = ctx->cid_table->luma_weight;
249 }
250 }
251
252 UPDATE_CACHE(bs, &ctx->gb);
253 GET_VLC(len, bs, &ctx->gb, ctx->dc_vlc.table, DNXHD_DC_VLC_BITS, 1);
254 if (len) {
255 level = GET_CACHE(bs, &ctx->gb);
256 LAST_SKIP_BITS(bs, &ctx->gb, len);
257 sign = ~level >> 31;
258 level = (NEG_USR32(sign ^ level, len) ^ sign) - sign;
259 ctx->last_dc[component] += level;
260 }
261 block[0] = ctx->last_dc[component];
262
263 for (i = 1; ; i++) {
264 UPDATE_CACHE(bs, &ctx->gb);
265 GET_VLC(index1, bs, &ctx->gb, ctx->ac_vlc.table,
266 DNXHD_VLC_BITS, 2);
267 level = ctx->cid_table->ac_level[index1];
268 if (!level) /* EOB */
269 break;
270
271 sign = SHOW_SBITS(bs, &ctx->gb, 1);
272 SKIP_BITS(bs, &ctx->gb, 1);
273
274 if (ctx->cid_table->ac_index_flag[index1]) {
275 level += SHOW_UBITS(bs, &ctx->gb, index_bits) << 6;
276 SKIP_BITS(bs, &ctx->gb, index_bits);
277 }
278
279 if (ctx->cid_table->ac_run_flag[index1]) {
280 UPDATE_CACHE(bs, &ctx->gb);
281 GET_VLC(index2, bs, &ctx->gb, ctx->run_vlc.table,
282 DNXHD_VLC_BITS, 2);
283 i += ctx->cid_table->run[index2];
284 }
285
286 if (i > 63) {
287 av_log(ctx->avctx, AV_LOG_ERROR, "ac tex damaged %d, %d\n", n, i);
288 break;
289 }
290
291 j = ctx->scantable.permutated[i];
292 level = (2 * level + 1) * qscale * weight_matrix[i];
293 if (level_bias < 32 || weight_matrix[i] != level_bias)
294 level += level_bias;
295 level >>= level_shift;
296
297 block[j] = (level ^ sign) - sign;
298 }
299
300 CLOSE_READER(bs, &ctx->gb);
301 }
302
303 static void dnxhd_decode_dct_block_8(DNXHDContext *ctx, int16_t *block,
304 int n, int qscale)
305 {
306 dnxhd_decode_dct_block(ctx, block, n, qscale, 4, 32, 6);
307 }
308
309 static void dnxhd_decode_dct_block_10(DNXHDContext *ctx, int16_t *block,
310 int n, int qscale)
311 {
312 dnxhd_decode_dct_block(ctx, block, n, qscale, 6, 8, 4);
313 }
314
315 static void dnxhd_decode_dct_block_10_444(DNXHDContext *ctx, int16_t *block,
316 int n, int qscale)
317 {
318 dnxhd_decode_dct_block(ctx, block, n, qscale, 6, 32, 6);
319 }
320
321 static int dnxhd_decode_macroblock(DNXHDContext *ctx, AVFrame *frame,
322 int x, int y)
323 {
324 int shift1 = ctx->bit_depth == 10;
325 int dct_linesize_luma = frame->linesize[0];
326 int dct_linesize_chroma = frame->linesize[1];
327 uint8_t *dest_y, *dest_u, *dest_v;
328 int dct_y_offset, dct_x_offset;
329 int qscale, i;
330 int interlaced_mb = 0;
331
332 if (ctx->mbaff) {
333 interlaced_mb = get_bits1(&ctx->gb);
334 qscale = get_bits(&ctx->gb, 10);
335 } else {
336 qscale = get_bits(&ctx->gb, 11);
337 }
338 skip_bits1(&ctx->gb);
339
340 for (i = 0; i < 8; i++) {
341 ctx->bdsp.clear_block(ctx->blocks[i]);
342 ctx->decode_dct_block(ctx, ctx->blocks[i], i, qscale);
343 }
344 if (ctx->is_444) {
345 for (; i < 12; i++) {
346 ctx->bdsp.clear_block(ctx->blocks[i]);
347 ctx->decode_dct_block(ctx, ctx->blocks[i], i, qscale);
348 }
349 }
350
351 if (frame->interlaced_frame) {
352 dct_linesize_luma <<= 1;
353 dct_linesize_chroma <<= 1;
354 }
355
356 dest_y = frame->data[0] + ((y * dct_linesize_luma) << 4) + (x << (4 + shift1));
357 dest_u = frame->data[1] + ((y * dct_linesize_chroma) << 4) + (x << (3 + shift1 + ctx->is_444));
358 dest_v = frame->data[2] + ((y * dct_linesize_chroma) << 4) + (x << (3 + shift1 + ctx->is_444));
359
360 if (ctx->cur_field) {
361 dest_y += frame->linesize[0];
362 dest_u += frame->linesize[1];
363 dest_v += frame->linesize[2];
364 }
365 if (interlaced_mb) {
366 dct_linesize_luma <<= 1;
367 dct_linesize_chroma <<= 1;
368 }
369
370 dct_y_offset = interlaced_mb ? frame->linesize[0] : (dct_linesize_luma << 3);
371 dct_x_offset = 8 << shift1;
372 if (!ctx->is_444) {
373 ctx->idsp.idct_put(dest_y, dct_linesize_luma, ctx->blocks[0]);
374 ctx->idsp.idct_put(dest_y + dct_x_offset, dct_linesize_luma, ctx->blocks[1]);
375 ctx->idsp.idct_put(dest_y + dct_y_offset, dct_linesize_luma, ctx->blocks[4]);
376 ctx->idsp.idct_put(dest_y + dct_y_offset + dct_x_offset, dct_linesize_luma, ctx->blocks[5]);
377
378 if (!(ctx->avctx->flags & AV_CODEC_FLAG_GRAY)) {
379 dct_y_offset = interlaced_mb ? frame->linesize[1] : (dct_linesize_chroma << 3);
380 ctx->idsp.idct_put(dest_u, dct_linesize_chroma, ctx->blocks[2]);
381 ctx->idsp.idct_put(dest_v, dct_linesize_chroma, ctx->blocks[3]);
382 ctx->idsp.idct_put(dest_u + dct_y_offset, dct_linesize_chroma, ctx->blocks[6]);
383 ctx->idsp.idct_put(dest_v + dct_y_offset, dct_linesize_chroma, ctx->blocks[7]);
384 }
385 } else {
386 ctx->idsp.idct_put(dest_y, dct_linesize_luma, ctx->blocks[0]);
387 ctx->idsp.idct_put(dest_y + dct_x_offset, dct_linesize_luma, ctx->blocks[1]);
388 ctx->idsp.idct_put(dest_y + dct_y_offset, dct_linesize_luma, ctx->blocks[6]);
389 ctx->idsp.idct_put(dest_y + dct_y_offset + dct_x_offset, dct_linesize_luma, ctx->blocks[7]);
390
391 if (!(ctx->avctx->flags & AV_CODEC_FLAG_GRAY)) {
392 dct_y_offset = interlaced_mb ? frame->linesize[1] : (dct_linesize_chroma << 3);
393 ctx->idsp.idct_put(dest_u, dct_linesize_chroma, ctx->blocks[2]);
394 ctx->idsp.idct_put(dest_u + dct_x_offset, dct_linesize_chroma, ctx->blocks[3]);
395 ctx->idsp.idct_put(dest_u + dct_y_offset, dct_linesize_chroma, ctx->blocks[8]);
396 ctx->idsp.idct_put(dest_u + dct_y_offset + dct_x_offset, dct_linesize_chroma, ctx->blocks[9]);
397 ctx->idsp.idct_put(dest_v, dct_linesize_chroma, ctx->blocks[4]);
398 ctx->idsp.idct_put(dest_v + dct_x_offset, dct_linesize_chroma, ctx->blocks[5]);
399 ctx->idsp.idct_put(dest_v + dct_y_offset, dct_linesize_chroma, ctx->blocks[10]);
400 ctx->idsp.idct_put(dest_v + dct_y_offset + dct_x_offset, dct_linesize_chroma, ctx->blocks[11]);
401 }
402 }
403
404 return 0;
405 }
406
407 static int dnxhd_decode_macroblocks(DNXHDContext *ctx, AVFrame *frame,
408 const uint8_t *buf, int buf_size)
409 {
410 int x, y;
411 for (y = 0; y < ctx->mb_height; y++) {
412 ctx->last_dc[0] =
413 ctx->last_dc[1] =
414 ctx->last_dc[2] = 1 << (ctx->bit_depth + 2); // for levels +2^(bitdepth-1)
415 init_get_bits(&ctx->gb, buf + ctx->mb_scan_index[y], (buf_size - ctx->mb_scan_index[y]) << 3);
416 for (x = 0; x < ctx->mb_width; x++) {
417 //START_TIMER;
418 dnxhd_decode_macroblock(ctx, frame, x, y);
419 //STOP_TIMER("decode macroblock");
420 }
421 }
422 return 0;
423 }
424
425 static int dnxhd_decode_frame(AVCodecContext *avctx, void *data,
426 int *got_frame, AVPacket *avpkt)
427 {
428 const uint8_t *buf = avpkt->data;
429 int buf_size = avpkt->size;
430 DNXHDContext *ctx = avctx->priv_data;
431 ThreadFrame tf;
432 int first_field = 1;
433 int ret;
434
435 tf.f = data;
436
437 ff_dlog(avctx, "frame size %d\n", buf_size);
438
439 decode_coding_unit:
440 if ((ret = dnxhd_decode_header(ctx, tf.f, buf, buf_size, first_field)) < 0)
441 return ret;
442
443 if ((avctx->width || avctx->height) &&
444 (ctx->width != avctx->width || ctx->height != avctx->height)) {
445 av_log(avctx, AV_LOG_WARNING, "frame size changed: %dx%d -> %ux%u\n",
446 avctx->width, avctx->height, ctx->width, ctx->height);
447 first_field = 1;
448 }
449
450 ret = ff_set_dimensions(avctx, ctx->width, ctx->height);
451 if (ret < 0)
452 return ret;
453
454 if (first_field) {
455 if ((ret = ff_thread_get_buffer(avctx, &tf, 0)) < 0) {
456 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
457 return ret;
458 }
459 tf.f->pict_type = AV_PICTURE_TYPE_I;
460 tf.f->key_frame = 1;
461 }
462
463 dnxhd_decode_macroblocks(ctx, tf.f, buf + 0x280, buf_size - 0x280);
464
465 if (first_field && tf.f->interlaced_frame) {
466 buf += ctx->cid_table->coding_unit_size;
467 buf_size -= ctx->cid_table->coding_unit_size;
468 first_field = 0;
469 goto decode_coding_unit;
470 }
471
472 *got_frame = 1;
473 return buf_size;
474 }
475
476 static av_cold int dnxhd_decode_close(AVCodecContext *avctx)
477 {
478 DNXHDContext *ctx = avctx->priv_data;
479
480 ff_free_vlc(&ctx->ac_vlc);
481 ff_free_vlc(&ctx->dc_vlc);
482 ff_free_vlc(&ctx->run_vlc);
483 return 0;
484 }
485
486 AVCodec ff_dnxhd_decoder = {
487 .name = "dnxhd",
488 .long_name = NULL_IF_CONFIG_SMALL("VC3/DNxHD"),
489 .type = AVMEDIA_TYPE_VIDEO,
490 .id = AV_CODEC_ID_DNXHD,
491 .priv_data_size = sizeof(DNXHDContext),
492 .init = dnxhd_decode_init,
493 .close = dnxhd_decode_close,
494 .decode = dnxhd_decode_frame,
495 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
496 };