1 /*
2 * Nvidia CUVID decoder
3 * Copyright (c) 2016 Timo Rothenpieler <timo@rothenpieler.org>
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 #include "compat/cuda/dynlink_loader.h"
23
24 #include "libavutil/buffer.h"
25 #include "libavutil/mathematics.h"
26 #include "libavutil/hwcontext.h"
27 #include "libavutil/hwcontext_cuda_internal.h"
28 #include "libavutil/cuda_check.h"
29 #include "libavutil/fifo.h"
30 #include "libavutil/log.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/pixdesc.h"
33
34 #include "avcodec.h"
35 #include "decode.h"
36 #include "hwconfig.h"
37 #include "nvdec.h"
38 #include "internal.h"
39
40 #if !NVDECAPI_CHECK_VERSION(9, 0)
41 #define cudaVideoSurfaceFormat_YUV444 2
42 #define cudaVideoSurfaceFormat_YUV444_16Bit 3
43 #endif
44
45 typedef struct CuvidContext
46 {
47 AVClass *avclass;
48
49 CUvideodecoder cudecoder;
50 CUvideoparser cuparser;
51
52 char *cu_gpu;
53 int nb_surfaces;
54 int drop_second_field;
55 char *crop_expr;
56 char *resize_expr;
57
58 struct {
59 int left;
60 int top;
61 int right;
62 int bottom;
63 } crop;
64
65 struct {
66 int width;
67 int height;
68 } resize;
69
70 AVBufferRef *hwdevice;
71 AVBufferRef *hwframe;
72
73 AVFifoBuffer *frame_queue;
74
75 int deint_mode;
76 int deint_mode_current;
77 int64_t prev_pts;
78 int progressive_sequence;
79
80 int internal_error;
81 int decoder_flushing;
82
83 int *key_frame;
84
85 cudaVideoCodec codec_type;
86 cudaVideoChromaFormat chroma_format;
87
88 CUVIDDECODECAPS caps8, caps10, caps12;
89
90 CUVIDPARSERPARAMS cuparseinfo;
91 CUVIDEOFORMATEX *cuparse_ext;
92
93 CudaFunctions *cudl;
94 CuvidFunctions *cvdl;
95 } CuvidContext;
96
97 typedef struct CuvidParsedFrame
98 {
99 CUVIDPARSERDISPINFO dispinfo;
100 int second_field;
101 int is_deinterlacing;
102 } CuvidParsedFrame;
103
104 #define CHECK_CU(x) FF_CUDA_CHECK_DL(avctx, ctx->cudl, x)
105
cuvid_handle_video_sequence(void * opaque,CUVIDEOFORMAT * format)106 static int CUDAAPI cuvid_handle_video_sequence(void *opaque, CUVIDEOFORMAT* format)
107 {
108 AVCodecContext *avctx = opaque;
109 CuvidContext *ctx = avctx->priv_data;
110 AVHWFramesContext *hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
111 CUVIDDECODECAPS *caps = NULL;
112 CUVIDDECODECREATEINFO cuinfo;
113 int surface_fmt;
114 int chroma_444;
115
116 int old_width = avctx->width;
117 int old_height = avctx->height;
118
119 enum AVPixelFormat pix_fmts[3] = { AV_PIX_FMT_CUDA,
120 AV_PIX_FMT_NONE, // Will be updated below
121 AV_PIX_FMT_NONE };
122
123 av_log(avctx, AV_LOG_TRACE, "pfnSequenceCallback, progressive_sequence=%d\n", format->progressive_sequence);
124
125 memset(&cuinfo, 0, sizeof(cuinfo));
126
127 ctx->internal_error = 0;
128
129 avctx->coded_width = cuinfo.ulWidth = format->coded_width;
130 avctx->coded_height = cuinfo.ulHeight = format->coded_height;
131
132 // apply cropping
133 cuinfo.display_area.left = format->display_area.left + ctx->crop.left;
134 cuinfo.display_area.top = format->display_area.top + ctx->crop.top;
135 cuinfo.display_area.right = format->display_area.right - ctx->crop.right;
136 cuinfo.display_area.bottom = format->display_area.bottom - ctx->crop.bottom;
137
138 // width and height need to be set before calling ff_get_format
139 if (ctx->resize_expr) {
140 avctx->width = ctx->resize.width;
141 avctx->height = ctx->resize.height;
142 } else {
143 avctx->width = cuinfo.display_area.right - cuinfo.display_area.left;
144 avctx->height = cuinfo.display_area.bottom - cuinfo.display_area.top;
145 }
146
147 // target width/height need to be multiples of two
148 cuinfo.ulTargetWidth = avctx->width = (avctx->width + 1) & ~1;
149 cuinfo.ulTargetHeight = avctx->height = (avctx->height + 1) & ~1;
150
151 // aspect ratio conversion, 1:1, depends on scaled resolution
152 cuinfo.target_rect.left = 0;
153 cuinfo.target_rect.top = 0;
154 cuinfo.target_rect.right = cuinfo.ulTargetWidth;
155 cuinfo.target_rect.bottom = cuinfo.ulTargetHeight;
156
157 chroma_444 = format->chroma_format == cudaVideoChromaFormat_444;
158
159 switch (format->bit_depth_luma_minus8) {
160 case 0: // 8-bit
161 pix_fmts[1] = chroma_444 ? AV_PIX_FMT_YUV444P : AV_PIX_FMT_NV12;
162 caps = &ctx->caps8;
163 break;
164 case 2: // 10-bit
165 pix_fmts[1] = chroma_444 ? AV_PIX_FMT_YUV444P16 : AV_PIX_FMT_P010;
166 caps = &ctx->caps10;
167 break;
168 case 4: // 12-bit
169 pix_fmts[1] = chroma_444 ? AV_PIX_FMT_YUV444P16 : AV_PIX_FMT_P016;
170 caps = &ctx->caps12;
171 break;
172 default:
173 break;
174 }
175
176 if (!caps || !caps->bIsSupported) {
177 av_log(avctx, AV_LOG_ERROR, "unsupported bit depth: %d\n",
178 format->bit_depth_luma_minus8 + 8);
179 ctx->internal_error = AVERROR(EINVAL);
180 return 0;
181 }
182
183 surface_fmt = ff_get_format(avctx, pix_fmts);
184 if (surface_fmt < 0) {
185 av_log(avctx, AV_LOG_ERROR, "ff_get_format failed: %d\n", surface_fmt);
186 ctx->internal_error = AVERROR(EINVAL);
187 return 0;
188 }
189
190 av_log(avctx, AV_LOG_VERBOSE, "Formats: Original: %s | HW: %s | SW: %s\n",
191 av_get_pix_fmt_name(avctx->pix_fmt),
192 av_get_pix_fmt_name(surface_fmt),
193 av_get_pix_fmt_name(avctx->sw_pix_fmt));
194
195 avctx->pix_fmt = surface_fmt;
196
197 // Update our hwframe ctx, as the get_format callback might have refreshed it!
198 if (avctx->hw_frames_ctx) {
199 av_buffer_unref(&ctx->hwframe);
200
201 ctx->hwframe = av_buffer_ref(avctx->hw_frames_ctx);
202 if (!ctx->hwframe) {
203 ctx->internal_error = AVERROR(ENOMEM);
204 return 0;
205 }
206
207 hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
208 }
209
210 ff_set_sar(avctx, av_div_q(
211 (AVRational){ format->display_aspect_ratio.x, format->display_aspect_ratio.y },
212 (AVRational){ avctx->width, avctx->height }));
213
214 ctx->deint_mode_current = format->progressive_sequence
215 ? cudaVideoDeinterlaceMode_Weave
216 : ctx->deint_mode;
217
218 ctx->progressive_sequence = format->progressive_sequence;
219
220 if (!format->progressive_sequence && ctx->deint_mode_current == cudaVideoDeinterlaceMode_Weave)
221 avctx->flags |= AV_CODEC_FLAG_INTERLACED_DCT;
222 else
223 avctx->flags &= ~AV_CODEC_FLAG_INTERLACED_DCT;
224
225 if (format->video_signal_description.video_full_range_flag)
226 avctx->color_range = AVCOL_RANGE_JPEG;
227 else
228 avctx->color_range = AVCOL_RANGE_MPEG;
229
230 avctx->color_primaries = format->video_signal_description.color_primaries;
231 avctx->color_trc = format->video_signal_description.transfer_characteristics;
232 avctx->colorspace = format->video_signal_description.matrix_coefficients;
233
234 if (format->bitrate)
235 avctx->bit_rate = format->bitrate;
236
237 if (format->frame_rate.numerator && format->frame_rate.denominator) {
238 avctx->framerate.num = format->frame_rate.numerator;
239 avctx->framerate.den = format->frame_rate.denominator;
240 }
241
242 if (ctx->cudecoder
243 && avctx->coded_width == format->coded_width
244 && avctx->coded_height == format->coded_height
245 && avctx->width == old_width
246 && avctx->height == old_height
247 && ctx->chroma_format == format->chroma_format
248 && ctx->codec_type == format->codec)
249 return 1;
250
251 if (ctx->cudecoder) {
252 av_log(avctx, AV_LOG_TRACE, "Re-initializing decoder\n");
253 ctx->internal_error = CHECK_CU(ctx->cvdl->cuvidDestroyDecoder(ctx->cudecoder));
254 if (ctx->internal_error < 0)
255 return 0;
256 ctx->cudecoder = NULL;
257 }
258
259 if (hwframe_ctx->pool && (
260 hwframe_ctx->width < avctx->width ||
261 hwframe_ctx->height < avctx->height ||
262 hwframe_ctx->format != AV_PIX_FMT_CUDA ||
263 hwframe_ctx->sw_format != avctx->sw_pix_fmt)) {
264 av_log(avctx, AV_LOG_ERROR, "AVHWFramesContext is already initialized with incompatible parameters\n");
265 av_log(avctx, AV_LOG_DEBUG, "width: %d <-> %d\n", hwframe_ctx->width, avctx->width);
266 av_log(avctx, AV_LOG_DEBUG, "height: %d <-> %d\n", hwframe_ctx->height, avctx->height);
267 av_log(avctx, AV_LOG_DEBUG, "format: %s <-> cuda\n", av_get_pix_fmt_name(hwframe_ctx->format));
268 av_log(avctx, AV_LOG_DEBUG, "sw_format: %s <-> %s\n",
269 av_get_pix_fmt_name(hwframe_ctx->sw_format), av_get_pix_fmt_name(avctx->sw_pix_fmt));
270 ctx->internal_error = AVERROR(EINVAL);
271 return 0;
272 }
273
274 ctx->chroma_format = format->chroma_format;
275
276 cuinfo.CodecType = ctx->codec_type = format->codec;
277 cuinfo.ChromaFormat = format->chroma_format;
278
279 switch (avctx->sw_pix_fmt) {
280 case AV_PIX_FMT_NV12:
281 cuinfo.OutputFormat = cudaVideoSurfaceFormat_NV12;
282 break;
283 case AV_PIX_FMT_P010:
284 case AV_PIX_FMT_P016:
285 cuinfo.OutputFormat = cudaVideoSurfaceFormat_P016;
286 break;
287 case AV_PIX_FMT_YUV444P:
288 cuinfo.OutputFormat = cudaVideoSurfaceFormat_YUV444;
289 break;
290 case AV_PIX_FMT_YUV444P16:
291 cuinfo.OutputFormat = cudaVideoSurfaceFormat_YUV444_16Bit;
292 break;
293 default:
294 av_log(avctx, AV_LOG_ERROR, "Unsupported output format: %s\n",
295 av_get_pix_fmt_name(avctx->sw_pix_fmt));
296 ctx->internal_error = AVERROR(EINVAL);
297 return 0;
298 }
299
300 cuinfo.ulNumDecodeSurfaces = ctx->nb_surfaces;
301 cuinfo.ulNumOutputSurfaces = 1;
302 cuinfo.ulCreationFlags = cudaVideoCreate_PreferCUVID;
303 cuinfo.bitDepthMinus8 = format->bit_depth_luma_minus8;
304 cuinfo.DeinterlaceMode = ctx->deint_mode_current;
305
306 if (ctx->deint_mode_current != cudaVideoDeinterlaceMode_Weave && !ctx->drop_second_field)
307 avctx->framerate = av_mul_q(avctx->framerate, (AVRational){2, 1});
308
309 ctx->internal_error = CHECK_CU(ctx->cvdl->cuvidCreateDecoder(&ctx->cudecoder, &cuinfo));
310 if (ctx->internal_error < 0)
311 return 0;
312
313 if (!hwframe_ctx->pool) {
314 hwframe_ctx->format = AV_PIX_FMT_CUDA;
315 hwframe_ctx->sw_format = avctx->sw_pix_fmt;
316 hwframe_ctx->width = avctx->width;
317 hwframe_ctx->height = avctx->height;
318
319 if ((ctx->internal_error = av_hwframe_ctx_init(ctx->hwframe)) < 0) {
320 av_log(avctx, AV_LOG_ERROR, "av_hwframe_ctx_init failed\n");
321 return 0;
322 }
323 }
324
325 return 1;
326 }
327
cuvid_handle_picture_decode(void * opaque,CUVIDPICPARAMS * picparams)328 static int CUDAAPI cuvid_handle_picture_decode(void *opaque, CUVIDPICPARAMS* picparams)
329 {
330 AVCodecContext *avctx = opaque;
331 CuvidContext *ctx = avctx->priv_data;
332
333 av_log(avctx, AV_LOG_TRACE, "pfnDecodePicture\n");
334
335 ctx->key_frame[picparams->CurrPicIdx] = picparams->intra_pic_flag;
336
337 ctx->internal_error = CHECK_CU(ctx->cvdl->cuvidDecodePicture(ctx->cudecoder, picparams));
338 if (ctx->internal_error < 0)
339 return 0;
340
341 return 1;
342 }
343
cuvid_handle_picture_display(void * opaque,CUVIDPARSERDISPINFO * dispinfo)344 static int CUDAAPI cuvid_handle_picture_display(void *opaque, CUVIDPARSERDISPINFO* dispinfo)
345 {
346 AVCodecContext *avctx = opaque;
347 CuvidContext *ctx = avctx->priv_data;
348 CuvidParsedFrame parsed_frame = { { 0 } };
349
350 parsed_frame.dispinfo = *dispinfo;
351 ctx->internal_error = 0;
352
353 // For some reason, dispinfo->progressive_frame is sometimes wrong.
354 parsed_frame.dispinfo.progressive_frame = ctx->progressive_sequence;
355
356 if (ctx->deint_mode_current == cudaVideoDeinterlaceMode_Weave) {
357 av_fifo_generic_write(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
358 } else {
359 parsed_frame.is_deinterlacing = 1;
360 av_fifo_generic_write(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
361 if (!ctx->drop_second_field) {
362 parsed_frame.second_field = 1;
363 av_fifo_generic_write(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
364 }
365 }
366
367 return 1;
368 }
369
cuvid_is_buffer_full(AVCodecContext * avctx)370 static int cuvid_is_buffer_full(AVCodecContext *avctx)
371 {
372 CuvidContext *ctx = avctx->priv_data;
373
374 int delay = ctx->cuparseinfo.ulMaxDisplayDelay;
375 if (ctx->deint_mode != cudaVideoDeinterlaceMode_Weave && !ctx->drop_second_field)
376 delay *= 2;
377
378 return (av_fifo_size(ctx->frame_queue) / sizeof(CuvidParsedFrame)) + delay >= ctx->nb_surfaces;
379 }
380
cuvid_decode_packet(AVCodecContext * avctx,const AVPacket * avpkt)381 static int cuvid_decode_packet(AVCodecContext *avctx, const AVPacket *avpkt)
382 {
383 CuvidContext *ctx = avctx->priv_data;
384 AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)ctx->hwdevice->data;
385 AVCUDADeviceContext *device_hwctx = device_ctx->hwctx;
386 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
387 CUVIDSOURCEDATAPACKET cupkt;
388 int ret = 0, eret = 0, is_flush = ctx->decoder_flushing;
389
390 av_log(avctx, AV_LOG_TRACE, "cuvid_decode_packet\n");
391
392 if (is_flush && avpkt && avpkt->size)
393 return AVERROR_EOF;
394
395 if (cuvid_is_buffer_full(avctx) && avpkt && avpkt->size)
396 return AVERROR(EAGAIN);
397
398 ret = CHECK_CU(ctx->cudl->cuCtxPushCurrent(cuda_ctx));
399 if (ret < 0) {
400 return ret;
401 }
402
403 memset(&cupkt, 0, sizeof(cupkt));
404
405 if (avpkt && avpkt->size) {
406 cupkt.payload_size = avpkt->size;
407 cupkt.payload = avpkt->data;
408
409 if (avpkt->pts != AV_NOPTS_VALUE) {
410 cupkt.flags = CUVID_PKT_TIMESTAMP;
411 if (avctx->pkt_timebase.num && avctx->pkt_timebase.den)
412 cupkt.timestamp = av_rescale_q(avpkt->pts, avctx->pkt_timebase, (AVRational){1, 10000000});
413 else
414 cupkt.timestamp = avpkt->pts;
415 }
416 } else {
417 cupkt.flags = CUVID_PKT_ENDOFSTREAM;
418 ctx->decoder_flushing = 1;
419 }
420
421 ret = CHECK_CU(ctx->cvdl->cuvidParseVideoData(ctx->cuparser, &cupkt));
422
423 if (ret < 0)
424 goto error;
425
426 // cuvidParseVideoData doesn't return an error just because stuff failed...
427 if (ctx->internal_error) {
428 av_log(avctx, AV_LOG_ERROR, "cuvid decode callback error\n");
429 ret = ctx->internal_error;
430 goto error;
431 }
432
433 error:
434 eret = CHECK_CU(ctx->cudl->cuCtxPopCurrent(&dummy));
435
436 if (eret < 0)
437 return eret;
438 else if (ret < 0)
439 return ret;
440 else if (is_flush)
441 return AVERROR_EOF;
442 else
443 return 0;
444 }
445
cuvid_output_frame(AVCodecContext * avctx,AVFrame * frame)446 static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
447 {
448 CuvidContext *ctx = avctx->priv_data;
449 AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)ctx->hwdevice->data;
450 AVCUDADeviceContext *device_hwctx = device_ctx->hwctx;
451 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
452 CUdeviceptr mapped_frame = 0;
453 int ret = 0, eret = 0;
454
455 av_log(avctx, AV_LOG_TRACE, "cuvid_output_frame\n");
456
457 if (ctx->decoder_flushing) {
458 ret = cuvid_decode_packet(avctx, NULL);
459 if (ret < 0 && ret != AVERROR_EOF)
460 return ret;
461 }
462
463 if (!cuvid_is_buffer_full(avctx)) {
464 AVPacket pkt = {0};
465 ret = ff_decode_get_packet(avctx, &pkt);
466 if (ret < 0 && ret != AVERROR_EOF)
467 return ret;
468 ret = cuvid_decode_packet(avctx, &pkt);
469 av_packet_unref(&pkt);
470 // cuvid_is_buffer_full() should avoid this.
471 if (ret == AVERROR(EAGAIN))
472 ret = AVERROR_EXTERNAL;
473 if (ret < 0 && ret != AVERROR_EOF)
474 return ret;
475 }
476
477 ret = CHECK_CU(ctx->cudl->cuCtxPushCurrent(cuda_ctx));
478 if (ret < 0)
479 return ret;
480
481 if (av_fifo_size(ctx->frame_queue)) {
482 const AVPixFmtDescriptor *pixdesc;
483 CuvidParsedFrame parsed_frame;
484 CUVIDPROCPARAMS params;
485 unsigned int pitch = 0;
486 int offset = 0;
487 int i;
488
489 av_fifo_generic_read(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
490
491 memset(¶ms, 0, sizeof(params));
492 params.progressive_frame = parsed_frame.dispinfo.progressive_frame;
493 params.second_field = parsed_frame.second_field;
494 params.top_field_first = parsed_frame.dispinfo.top_field_first;
495
496 ret = CHECK_CU(ctx->cvdl->cuvidMapVideoFrame(ctx->cudecoder, parsed_frame.dispinfo.picture_index, &mapped_frame, &pitch, ¶ms));
497 if (ret < 0)
498 goto error;
499
500 if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
501 ret = av_hwframe_get_buffer(ctx->hwframe, frame, 0);
502 if (ret < 0) {
503 av_log(avctx, AV_LOG_ERROR, "av_hwframe_get_buffer failed\n");
504 goto error;
505 }
506
507 ret = ff_decode_frame_props(avctx, frame);
508 if (ret < 0) {
509 av_log(avctx, AV_LOG_ERROR, "ff_decode_frame_props failed\n");
510 goto error;
511 }
512
513 pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
514
515 for (i = 0; i < pixdesc->nb_components; i++) {
516 int height = avctx->height >> (i ? pixdesc->log2_chroma_h : 0);
517 CUDA_MEMCPY2D cpy = {
518 .srcMemoryType = CU_MEMORYTYPE_DEVICE,
519 .dstMemoryType = CU_MEMORYTYPE_DEVICE,
520 .srcDevice = mapped_frame,
521 .dstDevice = (CUdeviceptr)frame->data[i],
522 .srcPitch = pitch,
523 .dstPitch = frame->linesize[i],
524 .srcY = offset,
525 .WidthInBytes = FFMIN(pitch, frame->linesize[i]),
526 .Height = height,
527 };
528
529 ret = CHECK_CU(ctx->cudl->cuMemcpy2DAsync(&cpy, device_hwctx->stream));
530 if (ret < 0)
531 goto error;
532
533 offset += height;
534 }
535 } else if (avctx->pix_fmt == AV_PIX_FMT_NV12 ||
536 avctx->pix_fmt == AV_PIX_FMT_P010 ||
537 avctx->pix_fmt == AV_PIX_FMT_P016 ||
538 avctx->pix_fmt == AV_PIX_FMT_YUV444P ||
539 avctx->pix_fmt == AV_PIX_FMT_YUV444P16) {
540 unsigned int offset = 0;
541 AVFrame *tmp_frame = av_frame_alloc();
542 if (!tmp_frame) {
543 av_log(avctx, AV_LOG_ERROR, "av_frame_alloc failed\n");
544 ret = AVERROR(ENOMEM);
545 goto error;
546 }
547
548 pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
549
550 tmp_frame->format = AV_PIX_FMT_CUDA;
551 tmp_frame->hw_frames_ctx = av_buffer_ref(ctx->hwframe);
552 tmp_frame->width = avctx->width;
553 tmp_frame->height = avctx->height;
554
555 /*
556 * Note that the following logic would not work for three plane
557 * YUV420 because the pitch value is different for the chroma
558 * planes.
559 */
560 for (i = 0; i < pixdesc->nb_components; i++) {
561 tmp_frame->data[i] = (uint8_t*)mapped_frame + offset;
562 tmp_frame->linesize[i] = pitch;
563 offset += pitch * (avctx->height >> (i ? pixdesc->log2_chroma_h : 0));
564 }
565
566 ret = ff_get_buffer(avctx, frame, 0);
567 if (ret < 0) {
568 av_log(avctx, AV_LOG_ERROR, "ff_get_buffer failed\n");
569 av_frame_free(&tmp_frame);
570 goto error;
571 }
572
573 ret = av_hwframe_transfer_data(frame, tmp_frame, 0);
574 if (ret) {
575 av_log(avctx, AV_LOG_ERROR, "av_hwframe_transfer_data failed\n");
576 av_frame_free(&tmp_frame);
577 goto error;
578 }
579 av_frame_free(&tmp_frame);
580 } else {
581 ret = AVERROR_BUG;
582 goto error;
583 }
584
585 frame->key_frame = ctx->key_frame[parsed_frame.dispinfo.picture_index];
586 frame->width = avctx->width;
587 frame->height = avctx->height;
588 if (avctx->pkt_timebase.num && avctx->pkt_timebase.den)
589 frame->pts = av_rescale_q(parsed_frame.dispinfo.timestamp, (AVRational){1, 10000000}, avctx->pkt_timebase);
590 else
591 frame->pts = parsed_frame.dispinfo.timestamp;
592
593 if (parsed_frame.second_field) {
594 if (ctx->prev_pts == INT64_MIN) {
595 ctx->prev_pts = frame->pts;
596 frame->pts += (avctx->pkt_timebase.den * avctx->framerate.den) / (avctx->pkt_timebase.num * avctx->framerate.num);
597 } else {
598 int pts_diff = (frame->pts - ctx->prev_pts) / 2;
599 ctx->prev_pts = frame->pts;
600 frame->pts += pts_diff;
601 }
602 }
603
604 /* CUVIDs opaque reordering breaks the internal pkt logic.
605 * So set pkt_pts and clear all the other pkt_ fields.
606 */
607 #if FF_API_PKT_PTS
608 FF_DISABLE_DEPRECATION_WARNINGS
609 frame->pkt_pts = frame->pts;
610 FF_ENABLE_DEPRECATION_WARNINGS
611 #endif
612 frame->pkt_pos = -1;
613 frame->pkt_duration = 0;
614 frame->pkt_size = -1;
615
616 frame->interlaced_frame = !parsed_frame.is_deinterlacing && !parsed_frame.dispinfo.progressive_frame;
617
618 if (frame->interlaced_frame)
619 frame->top_field_first = parsed_frame.dispinfo.top_field_first;
620 } else if (ctx->decoder_flushing) {
621 ret = AVERROR_EOF;
622 } else {
623 ret = AVERROR(EAGAIN);
624 }
625
626 error:
627 if (mapped_frame)
628 eret = CHECK_CU(ctx->cvdl->cuvidUnmapVideoFrame(ctx->cudecoder, mapped_frame));
629
630 eret = CHECK_CU(ctx->cudl->cuCtxPopCurrent(&dummy));
631
632 if (eret < 0)
633 return eret;
634 else
635 return ret;
636 }
637
cuvid_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * avpkt)638 static int cuvid_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *avpkt)
639 {
640 CuvidContext *ctx = avctx->priv_data;
641 AVFrame *frame = data;
642 int ret = 0;
643
644 av_log(avctx, AV_LOG_TRACE, "cuvid_decode_frame\n");
645
646 if (ctx->deint_mode_current != cudaVideoDeinterlaceMode_Weave) {
647 av_log(avctx, AV_LOG_ERROR, "Deinterlacing is not supported via the old API\n");
648 return AVERROR(EINVAL);
649 }
650
651 if (!ctx->decoder_flushing) {
652 ret = cuvid_decode_packet(avctx, avpkt);
653 if (ret < 0)
654 return ret;
655 }
656
657 ret = cuvid_output_frame(avctx, frame);
658 if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF) {
659 *got_frame = 0;
660 } else if (ret < 0) {
661 return ret;
662 } else {
663 *got_frame = 1;
664 }
665
666 return 0;
667 }
668
cuvid_decode_end(AVCodecContext * avctx)669 static av_cold int cuvid_decode_end(AVCodecContext *avctx)
670 {
671 CuvidContext *ctx = avctx->priv_data;
672
673 av_fifo_freep(&ctx->frame_queue);
674
675 if (ctx->cuparser)
676 ctx->cvdl->cuvidDestroyVideoParser(ctx->cuparser);
677
678 if (ctx->cudecoder)
679 ctx->cvdl->cuvidDestroyDecoder(ctx->cudecoder);
680
681 ctx->cudl = NULL;
682
683 av_buffer_unref(&ctx->hwframe);
684 av_buffer_unref(&ctx->hwdevice);
685
686 av_freep(&ctx->key_frame);
687 av_freep(&ctx->cuparse_ext);
688
689 cuvid_free_functions(&ctx->cvdl);
690
691 return 0;
692 }
693
cuvid_test_capabilities(AVCodecContext * avctx,const CUVIDPARSERPARAMS * cuparseinfo,int probed_width,int probed_height,int bit_depth)694 static int cuvid_test_capabilities(AVCodecContext *avctx,
695 const CUVIDPARSERPARAMS *cuparseinfo,
696 int probed_width,
697 int probed_height,
698 int bit_depth)
699 {
700 CuvidContext *ctx = avctx->priv_data;
701 CUVIDDECODECAPS *caps;
702 int res8 = 0, res10 = 0, res12 = 0;
703
704 if (!ctx->cvdl->cuvidGetDecoderCaps) {
705 av_log(avctx, AV_LOG_WARNING, "Used Nvidia driver is too old to perform a capability check.\n");
706 av_log(avctx, AV_LOG_WARNING, "The minimum required version is "
707 #if defined(_WIN32) || defined(__CYGWIN__)
708 "378.66"
709 #else
710 "378.13"
711 #endif
712 ". Continuing blind.\n");
713 ctx->caps8.bIsSupported = ctx->caps10.bIsSupported = 1;
714 // 12 bit was not supported before the capability check was introduced, so disable it.
715 ctx->caps12.bIsSupported = 0;
716 return 0;
717 }
718
719 ctx->caps8.eCodecType = ctx->caps10.eCodecType = ctx->caps12.eCodecType
720 = cuparseinfo->CodecType;
721 ctx->caps8.eChromaFormat = ctx->caps10.eChromaFormat = ctx->caps12.eChromaFormat
722 = cudaVideoChromaFormat_420;
723
724 ctx->caps8.nBitDepthMinus8 = 0;
725 ctx->caps10.nBitDepthMinus8 = 2;
726 ctx->caps12.nBitDepthMinus8 = 4;
727
728 res8 = CHECK_CU(ctx->cvdl->cuvidGetDecoderCaps(&ctx->caps8));
729 res10 = CHECK_CU(ctx->cvdl->cuvidGetDecoderCaps(&ctx->caps10));
730 res12 = CHECK_CU(ctx->cvdl->cuvidGetDecoderCaps(&ctx->caps12));
731
732 av_log(avctx, AV_LOG_VERBOSE, "CUVID capabilities for %s:\n", avctx->codec->name);
733 av_log(avctx, AV_LOG_VERBOSE, "8 bit: supported: %d, min_width: %d, max_width: %d, min_height: %d, max_height: %d\n",
734 ctx->caps8.bIsSupported, ctx->caps8.nMinWidth, ctx->caps8.nMaxWidth, ctx->caps8.nMinHeight, ctx->caps8.nMaxHeight);
735 av_log(avctx, AV_LOG_VERBOSE, "10 bit: supported: %d, min_width: %d, max_width: %d, min_height: %d, max_height: %d\n",
736 ctx->caps10.bIsSupported, ctx->caps10.nMinWidth, ctx->caps10.nMaxWidth, ctx->caps10.nMinHeight, ctx->caps10.nMaxHeight);
737 av_log(avctx, AV_LOG_VERBOSE, "12 bit: supported: %d, min_width: %d, max_width: %d, min_height: %d, max_height: %d\n",
738 ctx->caps12.bIsSupported, ctx->caps12.nMinWidth, ctx->caps12.nMaxWidth, ctx->caps12.nMinHeight, ctx->caps12.nMaxHeight);
739
740 switch (bit_depth) {
741 case 10:
742 caps = &ctx->caps10;
743 if (res10 < 0)
744 return res10;
745 break;
746 case 12:
747 caps = &ctx->caps12;
748 if (res12 < 0)
749 return res12;
750 break;
751 default:
752 caps = &ctx->caps8;
753 if (res8 < 0)
754 return res8;
755 }
756
757 if (!ctx->caps8.bIsSupported) {
758 av_log(avctx, AV_LOG_ERROR, "Codec %s is not supported.\n", avctx->codec->name);
759 return AVERROR(EINVAL);
760 }
761
762 if (!caps->bIsSupported) {
763 av_log(avctx, AV_LOG_ERROR, "Bit depth %d is not supported.\n", bit_depth);
764 return AVERROR(EINVAL);
765 }
766
767 if (probed_width > caps->nMaxWidth || probed_width < caps->nMinWidth) {
768 av_log(avctx, AV_LOG_ERROR, "Video width %d not within range from %d to %d\n",
769 probed_width, caps->nMinWidth, caps->nMaxWidth);
770 return AVERROR(EINVAL);
771 }
772
773 if (probed_height > caps->nMaxHeight || probed_height < caps->nMinHeight) {
774 av_log(avctx, AV_LOG_ERROR, "Video height %d not within range from %d to %d\n",
775 probed_height, caps->nMinHeight, caps->nMaxHeight);
776 return AVERROR(EINVAL);
777 }
778
779 if ((probed_width * probed_height) / 256 > caps->nMaxMBCount) {
780 av_log(avctx, AV_LOG_ERROR, "Video macroblock count %d exceeds maximum of %d\n",
781 (int)(probed_width * probed_height) / 256, caps->nMaxMBCount);
782 return AVERROR(EINVAL);
783 }
784
785 return 0;
786 }
787
cuvid_decode_init(AVCodecContext * avctx)788 static av_cold int cuvid_decode_init(AVCodecContext *avctx)
789 {
790 CuvidContext *ctx = avctx->priv_data;
791 AVCUDADeviceContext *device_hwctx;
792 AVHWDeviceContext *device_ctx;
793 AVHWFramesContext *hwframe_ctx;
794 CUVIDSOURCEDATAPACKET seq_pkt;
795 CUcontext cuda_ctx = NULL;
796 CUcontext dummy;
797 uint8_t *extradata;
798 int extradata_size;
799 int ret = 0;
800
801 enum AVPixelFormat pix_fmts[3] = { AV_PIX_FMT_CUDA,
802 AV_PIX_FMT_NV12,
803 AV_PIX_FMT_NONE };
804
805 int probed_width = avctx->coded_width ? avctx->coded_width : 1280;
806 int probed_height = avctx->coded_height ? avctx->coded_height : 720;
807 int probed_bit_depth = 8;
808
809 const AVPixFmtDescriptor *probe_desc = av_pix_fmt_desc_get(avctx->pix_fmt);
810 if (probe_desc && probe_desc->nb_components)
811 probed_bit_depth = probe_desc->comp[0].depth;
812
813 // Accelerated transcoding scenarios with 'ffmpeg' require that the
814 // pix_fmt be set to AV_PIX_FMT_CUDA early. The sw_pix_fmt, and the
815 // pix_fmt for non-accelerated transcoding, do not need to be correct
816 // but need to be set to something. We arbitrarily pick NV12.
817 ret = ff_get_format(avctx, pix_fmts);
818 if (ret < 0) {
819 av_log(avctx, AV_LOG_ERROR, "ff_get_format failed: %d\n", ret);
820 return ret;
821 }
822 avctx->pix_fmt = ret;
823
824 if (ctx->resize_expr && sscanf(ctx->resize_expr, "%dx%d",
825 &ctx->resize.width, &ctx->resize.height) != 2) {
826 av_log(avctx, AV_LOG_ERROR, "Invalid resize expressions\n");
827 ret = AVERROR(EINVAL);
828 goto error;
829 }
830
831 if (ctx->crop_expr && sscanf(ctx->crop_expr, "%dx%dx%dx%d",
832 &ctx->crop.top, &ctx->crop.bottom,
833 &ctx->crop.left, &ctx->crop.right) != 4) {
834 av_log(avctx, AV_LOG_ERROR, "Invalid cropping expressions\n");
835 ret = AVERROR(EINVAL);
836 goto error;
837 }
838
839 ret = cuvid_load_functions(&ctx->cvdl, avctx);
840 if (ret < 0) {
841 av_log(avctx, AV_LOG_ERROR, "Failed loading nvcuvid.\n");
842 goto error;
843 }
844
845 ctx->frame_queue = av_fifo_alloc(ctx->nb_surfaces * sizeof(CuvidParsedFrame));
846 if (!ctx->frame_queue) {
847 ret = AVERROR(ENOMEM);
848 goto error;
849 }
850
851 if (avctx->hw_frames_ctx) {
852 ctx->hwframe = av_buffer_ref(avctx->hw_frames_ctx);
853 if (!ctx->hwframe) {
854 ret = AVERROR(ENOMEM);
855 goto error;
856 }
857
858 hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
859
860 ctx->hwdevice = av_buffer_ref(hwframe_ctx->device_ref);
861 if (!ctx->hwdevice) {
862 ret = AVERROR(ENOMEM);
863 goto error;
864 }
865 } else {
866 if (avctx->hw_device_ctx) {
867 ctx->hwdevice = av_buffer_ref(avctx->hw_device_ctx);
868 if (!ctx->hwdevice) {
869 ret = AVERROR(ENOMEM);
870 goto error;
871 }
872 } else {
873 ret = av_hwdevice_ctx_create(&ctx->hwdevice, AV_HWDEVICE_TYPE_CUDA, ctx->cu_gpu, NULL, 0);
874 if (ret < 0)
875 goto error;
876 }
877
878 ctx->hwframe = av_hwframe_ctx_alloc(ctx->hwdevice);
879 if (!ctx->hwframe) {
880 av_log(avctx, AV_LOG_ERROR, "av_hwframe_ctx_alloc failed\n");
881 ret = AVERROR(ENOMEM);
882 goto error;
883 }
884
885 hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
886 }
887
888 device_ctx = hwframe_ctx->device_ctx;
889 device_hwctx = device_ctx->hwctx;
890
891 cuda_ctx = device_hwctx->cuda_ctx;
892 ctx->cudl = device_hwctx->internal->cuda_dl;
893
894 memset(&ctx->cuparseinfo, 0, sizeof(ctx->cuparseinfo));
895 memset(&seq_pkt, 0, sizeof(seq_pkt));
896
897 switch (avctx->codec->id) {
898 #if CONFIG_H264_CUVID_DECODER
899 case AV_CODEC_ID_H264:
900 ctx->cuparseinfo.CodecType = cudaVideoCodec_H264;
901 break;
902 #endif
903 #if CONFIG_HEVC_CUVID_DECODER
904 case AV_CODEC_ID_HEVC:
905 ctx->cuparseinfo.CodecType = cudaVideoCodec_HEVC;
906 break;
907 #endif
908 #if CONFIG_MJPEG_CUVID_DECODER
909 case AV_CODEC_ID_MJPEG:
910 ctx->cuparseinfo.CodecType = cudaVideoCodec_JPEG;
911 break;
912 #endif
913 #if CONFIG_MPEG1_CUVID_DECODER
914 case AV_CODEC_ID_MPEG1VIDEO:
915 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG1;
916 break;
917 #endif
918 #if CONFIG_MPEG2_CUVID_DECODER
919 case AV_CODEC_ID_MPEG2VIDEO:
920 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG2;
921 break;
922 #endif
923 #if CONFIG_MPEG4_CUVID_DECODER
924 case AV_CODEC_ID_MPEG4:
925 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG4;
926 break;
927 #endif
928 #if CONFIG_VP8_CUVID_DECODER
929 case AV_CODEC_ID_VP8:
930 ctx->cuparseinfo.CodecType = cudaVideoCodec_VP8;
931 break;
932 #endif
933 #if CONFIG_VP9_CUVID_DECODER
934 case AV_CODEC_ID_VP9:
935 ctx->cuparseinfo.CodecType = cudaVideoCodec_VP9;
936 break;
937 #endif
938 #if CONFIG_VC1_CUVID_DECODER
939 case AV_CODEC_ID_VC1:
940 ctx->cuparseinfo.CodecType = cudaVideoCodec_VC1;
941 break;
942 #endif
943 default:
944 av_log(avctx, AV_LOG_ERROR, "Invalid CUVID codec!\n");
945 return AVERROR_BUG;
946 }
947
948 if (avctx->codec->bsfs) {
949 const AVCodecParameters *par = avctx->internal->bsf->par_out;
950 extradata = par->extradata;
951 extradata_size = par->extradata_size;
952 } else {
953 extradata = avctx->extradata;
954 extradata_size = avctx->extradata_size;
955 }
956
957 ctx->cuparse_ext = av_mallocz(sizeof(*ctx->cuparse_ext)
958 + FFMAX(extradata_size - (int)sizeof(ctx->cuparse_ext->raw_seqhdr_data), 0));
959 if (!ctx->cuparse_ext) {
960 ret = AVERROR(ENOMEM);
961 goto error;
962 }
963
964 if (extradata_size > 0)
965 memcpy(ctx->cuparse_ext->raw_seqhdr_data, extradata, extradata_size);
966 ctx->cuparse_ext->format.seqhdr_data_length = extradata_size;
967
968 ctx->cuparseinfo.pExtVideoInfo = ctx->cuparse_ext;
969
970 ctx->key_frame = av_mallocz(ctx->nb_surfaces * sizeof(int));
971 if (!ctx->key_frame) {
972 ret = AVERROR(ENOMEM);
973 goto error;
974 }
975
976 ctx->cuparseinfo.ulMaxNumDecodeSurfaces = ctx->nb_surfaces;
977 ctx->cuparseinfo.ulMaxDisplayDelay = 4;
978 ctx->cuparseinfo.pUserData = avctx;
979 ctx->cuparseinfo.pfnSequenceCallback = cuvid_handle_video_sequence;
980 ctx->cuparseinfo.pfnDecodePicture = cuvid_handle_picture_decode;
981 ctx->cuparseinfo.pfnDisplayPicture = cuvid_handle_picture_display;
982
983 ret = CHECK_CU(ctx->cudl->cuCtxPushCurrent(cuda_ctx));
984 if (ret < 0)
985 goto error;
986
987 ret = cuvid_test_capabilities(avctx, &ctx->cuparseinfo,
988 probed_width,
989 probed_height,
990 probed_bit_depth);
991 if (ret < 0)
992 goto error;
993
994 ret = CHECK_CU(ctx->cvdl->cuvidCreateVideoParser(&ctx->cuparser, &ctx->cuparseinfo));
995 if (ret < 0)
996 goto error;
997
998 seq_pkt.payload = ctx->cuparse_ext->raw_seqhdr_data;
999 seq_pkt.payload_size = ctx->cuparse_ext->format.seqhdr_data_length;
1000
1001 if (seq_pkt.payload && seq_pkt.payload_size) {
1002 ret = CHECK_CU(ctx->cvdl->cuvidParseVideoData(ctx->cuparser, &seq_pkt));
1003 if (ret < 0)
1004 goto error;
1005 }
1006
1007 ret = CHECK_CU(ctx->cudl->cuCtxPopCurrent(&dummy));
1008 if (ret < 0)
1009 goto error;
1010
1011 ctx->prev_pts = INT64_MIN;
1012
1013 if (!avctx->pkt_timebase.num || !avctx->pkt_timebase.den)
1014 av_log(avctx, AV_LOG_WARNING, "Invalid pkt_timebase, passing timestamps as-is.\n");
1015
1016 return 0;
1017
1018 error:
1019 cuvid_decode_end(avctx);
1020 return ret;
1021 }
1022
cuvid_flush(AVCodecContext * avctx)1023 static void cuvid_flush(AVCodecContext *avctx)
1024 {
1025 CuvidContext *ctx = avctx->priv_data;
1026 AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)ctx->hwdevice->data;
1027 AVCUDADeviceContext *device_hwctx = device_ctx->hwctx;
1028 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
1029 CUVIDSOURCEDATAPACKET seq_pkt = { 0 };
1030 int ret;
1031
1032 ret = CHECK_CU(ctx->cudl->cuCtxPushCurrent(cuda_ctx));
1033 if (ret < 0)
1034 goto error;
1035
1036 av_fifo_freep(&ctx->frame_queue);
1037
1038 ctx->frame_queue = av_fifo_alloc(ctx->nb_surfaces * sizeof(CuvidParsedFrame));
1039 if (!ctx->frame_queue) {
1040 av_log(avctx, AV_LOG_ERROR, "Failed to recreate frame queue on flush\n");
1041 return;
1042 }
1043
1044 if (ctx->cudecoder) {
1045 ctx->cvdl->cuvidDestroyDecoder(ctx->cudecoder);
1046 ctx->cudecoder = NULL;
1047 }
1048
1049 if (ctx->cuparser) {
1050 ctx->cvdl->cuvidDestroyVideoParser(ctx->cuparser);
1051 ctx->cuparser = NULL;
1052 }
1053
1054 ret = CHECK_CU(ctx->cvdl->cuvidCreateVideoParser(&ctx->cuparser, &ctx->cuparseinfo));
1055 if (ret < 0)
1056 goto error;
1057
1058 seq_pkt.payload = ctx->cuparse_ext->raw_seqhdr_data;
1059 seq_pkt.payload_size = ctx->cuparse_ext->format.seqhdr_data_length;
1060
1061 if (seq_pkt.payload && seq_pkt.payload_size) {
1062 ret = CHECK_CU(ctx->cvdl->cuvidParseVideoData(ctx->cuparser, &seq_pkt));
1063 if (ret < 0)
1064 goto error;
1065 }
1066
1067 ret = CHECK_CU(ctx->cudl->cuCtxPopCurrent(&dummy));
1068 if (ret < 0)
1069 goto error;
1070
1071 ctx->prev_pts = INT64_MIN;
1072 ctx->decoder_flushing = 0;
1073
1074 return;
1075 error:
1076 av_log(avctx, AV_LOG_ERROR, "CUDA reinit on flush failed\n");
1077 }
1078
1079 #define OFFSET(x) offsetof(CuvidContext, x)
1080 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1081 static const AVOption options[] = {
1082 { "deint", "Set deinterlacing mode", OFFSET(deint_mode), AV_OPT_TYPE_INT, { .i64 = cudaVideoDeinterlaceMode_Weave }, cudaVideoDeinterlaceMode_Weave, cudaVideoDeinterlaceMode_Adaptive, VD, "deint" },
1083 { "weave", "Weave deinterlacing (do nothing)", 0, AV_OPT_TYPE_CONST, { .i64 = cudaVideoDeinterlaceMode_Weave }, 0, 0, VD, "deint" },
1084 { "bob", "Bob deinterlacing", 0, AV_OPT_TYPE_CONST, { .i64 = cudaVideoDeinterlaceMode_Bob }, 0, 0, VD, "deint" },
1085 { "adaptive", "Adaptive deinterlacing", 0, AV_OPT_TYPE_CONST, { .i64 = cudaVideoDeinterlaceMode_Adaptive }, 0, 0, VD, "deint" },
1086 { "gpu", "GPU to be used for decoding", OFFSET(cu_gpu), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, VD },
1087 { "surfaces", "Maximum surfaces to be used for decoding", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 25 }, 0, INT_MAX, VD },
1088 { "drop_second_field", "Drop second field when deinterlacing", OFFSET(drop_second_field), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VD },
1089 { "crop", "Crop (top)x(bottom)x(left)x(right)", OFFSET(crop_expr), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, VD },
1090 { "resize", "Resize (width)x(height)", OFFSET(resize_expr), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, VD },
1091 { NULL }
1092 };
1093
1094 static const AVCodecHWConfigInternal *cuvid_hw_configs[] = {
1095 &(const AVCodecHWConfigInternal) {
1096 .public = {
1097 .pix_fmt = AV_PIX_FMT_CUDA,
1098 .methods = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX |
1099 AV_CODEC_HW_CONFIG_METHOD_INTERNAL,
1100 .device_type = AV_HWDEVICE_TYPE_CUDA
1101 },
1102 .hwaccel = NULL,
1103 },
1104 NULL
1105 };
1106
1107 #define DEFINE_CUVID_CODEC(x, X, bsf_name) \
1108 static const AVClass x##_cuvid_class = { \
1109 .class_name = #x "_cuvid", \
1110 .item_name = av_default_item_name, \
1111 .option = options, \
1112 .version = LIBAVUTIL_VERSION_INT, \
1113 }; \
1114 AVCodec ff_##x##_cuvid_decoder = { \
1115 .name = #x "_cuvid", \
1116 .long_name = NULL_IF_CONFIG_SMALL("Nvidia CUVID " #X " decoder"), \
1117 .type = AVMEDIA_TYPE_VIDEO, \
1118 .id = AV_CODEC_ID_##X, \
1119 .priv_data_size = sizeof(CuvidContext), \
1120 .priv_class = &x##_cuvid_class, \
1121 .init = cuvid_decode_init, \
1122 .close = cuvid_decode_end, \
1123 .decode = cuvid_decode_frame, \
1124 .receive_frame = cuvid_output_frame, \
1125 .flush = cuvid_flush, \
1126 .bsfs = bsf_name, \
1127 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AVOID_PROBING | AV_CODEC_CAP_HARDWARE, \
1128 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_CUDA, \
1129 AV_PIX_FMT_NV12, \
1130 AV_PIX_FMT_P010, \
1131 AV_PIX_FMT_P016, \
1132 AV_PIX_FMT_NONE }, \
1133 .hw_configs = cuvid_hw_configs, \
1134 .wrapper_name = "cuvid", \
1135 };
1136
1137 #if CONFIG_HEVC_CUVID_DECODER
1138 DEFINE_CUVID_CODEC(hevc, HEVC, "hevc_mp4toannexb")
1139 #endif
1140
1141 #if CONFIG_H264_CUVID_DECODER
1142 DEFINE_CUVID_CODEC(h264, H264, "h264_mp4toannexb")
1143 #endif
1144
1145 #if CONFIG_MJPEG_CUVID_DECODER
1146 DEFINE_CUVID_CODEC(mjpeg, MJPEG, NULL)
1147 #endif
1148
1149 #if CONFIG_MPEG1_CUVID_DECODER
1150 DEFINE_CUVID_CODEC(mpeg1, MPEG1VIDEO, NULL)
1151 #endif
1152
1153 #if CONFIG_MPEG2_CUVID_DECODER
1154 DEFINE_CUVID_CODEC(mpeg2, MPEG2VIDEO, NULL)
1155 #endif
1156
1157 #if CONFIG_MPEG4_CUVID_DECODER
1158 DEFINE_CUVID_CODEC(mpeg4, MPEG4, NULL)
1159 #endif
1160
1161 #if CONFIG_VP8_CUVID_DECODER
1162 DEFINE_CUVID_CODEC(vp8, VP8, NULL)
1163 #endif
1164
1165 #if CONFIG_VP9_CUVID_DECODER
1166 DEFINE_CUVID_CODEC(vp9, VP9, NULL)
1167 #endif
1168
1169 #if CONFIG_VC1_CUVID_DECODER
1170 DEFINE_CUVID_CODEC(vc1, VC1, NULL)
1171 #endif
1172