3 * Copyright (c) 2016 Timo Rothenpieler <timo@rothenpieler.org>
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
22 #include "libavutil/buffer.h"
23 #include "libavutil/mathematics.h"
24 #include "libavutil/hwcontext.h"
25 #include "libavutil/hwcontext_cuda.h"
26 #include "libavutil/fifo.h"
27 #include "libavutil/log.h"
28 #include "libavutil/opt.h"
33 #include "compat/cuda/nvcuvid.h"
35 #define MAX_FRAME_COUNT 25
37 typedef struct CuvidContext
41 CUvideodecoder cudecoder;
42 CUvideoparser cuparser;
46 AVBufferRef *hwdevice;
51 AVFifoBuffer *frame_queue;
59 cudaVideoCodec codec_type;
60 cudaVideoChromaFormat chroma_format;
62 CUVIDPARSERPARAMS cuparseinfo;
63 CUVIDEOFORMATEX cuparse_ext;
66 typedef struct CuvidParsedFrame
68 CUVIDPARSERDISPINFO dispinfo;
73 static int check_cu(AVCodecContext *avctx, CUresult err, const char *func)
76 const char *err_string;
78 av_log(avctx, AV_LOG_TRACE, "Calling %s\n", func);
80 if (err == CUDA_SUCCESS)
83 cuGetErrorName(err, &err_name);
84 cuGetErrorString(err, &err_string);
86 av_log(avctx, AV_LOG_ERROR, "%s failed", func);
87 if (err_name && err_string)
88 av_log(avctx, AV_LOG_ERROR, " -> %s: %s", err_name, err_string);
89 av_log(avctx, AV_LOG_ERROR, "\n");
91 return AVERROR_EXTERNAL;
94 #define CHECK_CU(x) check_cu(avctx, (x), #x)
96 static int CUDAAPI cuvid_handle_video_sequence(void *opaque, CUVIDEOFORMAT* format)
98 AVCodecContext *avctx = opaque;
99 CuvidContext *ctx = avctx->priv_data;
100 AVHWFramesContext *hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
101 CUVIDDECODECREATEINFO cuinfo;
103 av_log(avctx, AV_LOG_TRACE, "pfnSequenceCallback, progressive_sequence=%d\n", format->progressive_sequence);
105 ctx->internal_error = 0;
107 avctx->width = format->display_area.right;
108 avctx->height = format->display_area.bottom;
110 ff_set_sar(avctx, av_div_q(
111 (AVRational){ format->display_aspect_ratio.x, format->display_aspect_ratio.y },
112 (AVRational){ avctx->width, avctx->height }));
114 if (!format->progressive_sequence && ctx->deint_mode == cudaVideoDeinterlaceMode_Weave)
115 avctx->flags |= AV_CODEC_FLAG_INTERLACED_DCT;
117 avctx->flags &= ~AV_CODEC_FLAG_INTERLACED_DCT;
119 if (format->video_signal_description.video_full_range_flag)
120 avctx->color_range = AVCOL_RANGE_JPEG;
122 avctx->color_range = AVCOL_RANGE_MPEG;
124 avctx->color_primaries = format->video_signal_description.color_primaries;
125 avctx->color_trc = format->video_signal_description.transfer_characteristics;
126 avctx->colorspace = format->video_signal_description.matrix_coefficients;
129 avctx->bit_rate = format->bitrate;
131 if (format->frame_rate.numerator && format->frame_rate.denominator) {
132 avctx->framerate.num = format->frame_rate.numerator;
133 avctx->framerate.den = format->frame_rate.denominator;
137 && avctx->coded_width == format->coded_width
138 && avctx->coded_height == format->coded_height
139 && ctx->chroma_format == format->chroma_format
140 && ctx->codec_type == format->codec)
143 if (ctx->cudecoder) {
144 av_log(avctx, AV_LOG_ERROR, "re-initializing decoder is not supported\n");
145 ctx->internal_error = AVERROR(EINVAL);
149 if (hwframe_ctx->pool && (
150 hwframe_ctx->width < avctx->width ||
151 hwframe_ctx->height < avctx->height ||
152 hwframe_ctx->format != AV_PIX_FMT_CUDA ||
153 hwframe_ctx->sw_format != AV_PIX_FMT_NV12)) {
154 av_log(avctx, AV_LOG_ERROR, "AVHWFramesContext is already initialized with incompatible parameters\n");
155 ctx->internal_error = AVERROR(EINVAL);
159 if (format->chroma_format != cudaVideoChromaFormat_420) {
160 av_log(avctx, AV_LOG_ERROR, "Chroma formats other than 420 are not supported\n");
161 ctx->internal_error = AVERROR(EINVAL);
165 avctx->coded_width = format->coded_width;
166 avctx->coded_height = format->coded_height;
168 ctx->chroma_format = format->chroma_format;
170 memset(&cuinfo, 0, sizeof(cuinfo));
172 cuinfo.CodecType = ctx->codec_type = format->codec;
173 cuinfo.ChromaFormat = format->chroma_format;
174 cuinfo.OutputFormat = cudaVideoSurfaceFormat_NV12;
176 cuinfo.ulWidth = avctx->coded_width;
177 cuinfo.ulHeight = avctx->coded_height;
178 cuinfo.ulTargetWidth = cuinfo.ulWidth;
179 cuinfo.ulTargetHeight = cuinfo.ulHeight;
181 cuinfo.target_rect.left = 0;
182 cuinfo.target_rect.top = 0;
183 cuinfo.target_rect.right = cuinfo.ulWidth;
184 cuinfo.target_rect.bottom = cuinfo.ulHeight;
186 cuinfo.ulNumDecodeSurfaces = MAX_FRAME_COUNT;
187 cuinfo.ulNumOutputSurfaces = 1;
188 cuinfo.ulCreationFlags = cudaVideoCreate_PreferCUVID;
189 cuinfo.bitDepthMinus8 = format->bit_depth_luma_minus8;
191 if (format->progressive_sequence) {
192 ctx->deint_mode = cuinfo.DeinterlaceMode = cudaVideoDeinterlaceMode_Weave;
194 cuinfo.DeinterlaceMode = ctx->deint_mode;
197 if (ctx->deint_mode != cudaVideoDeinterlaceMode_Weave)
198 avctx->framerate = av_mul_q(avctx->framerate, (AVRational){2, 1});
200 ctx->internal_error = CHECK_CU(cuvidCreateDecoder(&ctx->cudecoder, &cuinfo));
201 if (ctx->internal_error < 0)
204 if (!hwframe_ctx->pool) {
205 hwframe_ctx->format = AV_PIX_FMT_CUDA;
206 hwframe_ctx->sw_format = AV_PIX_FMT_NV12;
207 hwframe_ctx->width = FFALIGN(avctx->width, 32);
208 hwframe_ctx->height = FFALIGN(avctx->height, 32);
210 if ((ctx->internal_error = av_hwframe_ctx_init(ctx->hwframe)) < 0) {
211 av_log(avctx, AV_LOG_ERROR, "av_hwframe_ctx_init failed\n");
219 static int CUDAAPI cuvid_handle_picture_decode(void *opaque, CUVIDPICPARAMS* picparams)
221 AVCodecContext *avctx = opaque;
222 CuvidContext *ctx = avctx->priv_data;
224 av_log(avctx, AV_LOG_TRACE, "pfnDecodePicture\n");
226 ctx->internal_error = CHECK_CU(cuvidDecodePicture(ctx->cudecoder, picparams));
227 if (ctx->internal_error < 0)
233 static int CUDAAPI cuvid_handle_picture_display(void *opaque, CUVIDPARSERDISPINFO* dispinfo)
235 AVCodecContext *avctx = opaque;
236 CuvidContext *ctx = avctx->priv_data;
237 CuvidParsedFrame parsed_frame = { *dispinfo, 0, 0 };
239 ctx->internal_error = 0;
241 if (ctx->deint_mode == cudaVideoDeinterlaceMode_Weave) {
242 av_fifo_generic_write(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
244 parsed_frame.is_deinterlacing = 1;
245 av_fifo_generic_write(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
246 parsed_frame.second_field = 1;
247 av_fifo_generic_write(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
253 static int cuvid_decode_packet(AVCodecContext *avctx, const AVPacket *avpkt)
255 CuvidContext *ctx = avctx->priv_data;
256 AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)ctx->hwdevice->data;
257 AVCUDADeviceContext *device_hwctx = device_ctx->hwctx;
258 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
259 CUVIDSOURCEDATAPACKET cupkt;
260 AVPacket filter_packet = { 0 };
261 AVPacket filtered_packet = { 0 };
262 int ret = 0, eret = 0, is_flush = ctx->decoder_flushing;
264 av_log(avctx, AV_LOG_TRACE, "cuvid_decode_packet\n");
266 if (is_flush && avpkt && avpkt->size)
269 if (av_fifo_size(ctx->frame_queue) / sizeof(CuvidParsedFrame) > MAX_FRAME_COUNT - 2 && avpkt && avpkt->size)
270 return AVERROR(EAGAIN);
272 if (ctx->bsf && avpkt && avpkt->size) {
273 if ((ret = av_packet_ref(&filter_packet, avpkt)) < 0) {
274 av_log(avctx, AV_LOG_ERROR, "av_packet_ref failed\n");
278 if ((ret = av_bsf_send_packet(ctx->bsf, &filter_packet)) < 0) {
279 av_log(avctx, AV_LOG_ERROR, "av_bsf_send_packet failed\n");
280 av_packet_unref(&filter_packet);
284 if ((ret = av_bsf_receive_packet(ctx->bsf, &filtered_packet)) < 0) {
285 av_log(avctx, AV_LOG_ERROR, "av_bsf_receive_packet failed\n");
289 avpkt = &filtered_packet;
292 ret = CHECK_CU(cuCtxPushCurrent(cuda_ctx));
294 av_packet_unref(&filtered_packet);
298 memset(&cupkt, 0, sizeof(cupkt));
300 if (avpkt && avpkt->size) {
301 cupkt.payload_size = avpkt->size;
302 cupkt.payload = avpkt->data;
304 if (avpkt->pts != AV_NOPTS_VALUE) {
305 cupkt.flags = CUVID_PKT_TIMESTAMP;
306 if (avctx->pkt_timebase.num && avctx->pkt_timebase.den)
307 cupkt.timestamp = av_rescale_q(avpkt->pts, avctx->pkt_timebase, (AVRational){1, 10000000});
309 cupkt.timestamp = avpkt->pts;
312 cupkt.flags = CUVID_PKT_ENDOFSTREAM;
313 ctx->decoder_flushing = 1;
316 ret = CHECK_CU(cuvidParseVideoData(ctx->cuparser, &cupkt));
318 av_packet_unref(&filtered_packet);
323 // cuvidParseVideoData doesn't return an error just because stuff failed...
324 if (ctx->internal_error) {
325 av_log(avctx, AV_LOG_ERROR, "cuvid decode callback error\n");
326 ret = ctx->internal_error;
331 eret = CHECK_CU(cuCtxPopCurrent(&dummy));
343 static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
345 CuvidContext *ctx = avctx->priv_data;
346 AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)ctx->hwdevice->data;
347 AVCUDADeviceContext *device_hwctx = device_ctx->hwctx;
348 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
349 CUdeviceptr mapped_frame = 0;
350 int ret = 0, eret = 0;
352 av_log(avctx, AV_LOG_TRACE, "cuvid_output_frame\n");
354 if (ctx->decoder_flushing) {
355 ret = cuvid_decode_packet(avctx, NULL);
356 if (ret < 0 && ret != AVERROR_EOF)
360 ret = CHECK_CU(cuCtxPushCurrent(cuda_ctx));
364 if (av_fifo_size(ctx->frame_queue)) {
365 CuvidParsedFrame parsed_frame;
366 CUVIDPROCPARAMS params;
367 unsigned int pitch = 0;
371 av_fifo_generic_read(ctx->frame_queue, &parsed_frame, sizeof(CuvidParsedFrame), NULL);
373 memset(¶ms, 0, sizeof(params));
374 params.progressive_frame = parsed_frame.dispinfo.progressive_frame;
375 params.second_field = parsed_frame.second_field;
376 params.top_field_first = parsed_frame.dispinfo.top_field_first;
378 ret = CHECK_CU(cuvidMapVideoFrame(ctx->cudecoder, parsed_frame.dispinfo.picture_index, &mapped_frame, &pitch, ¶ms));
382 if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
383 ret = av_hwframe_get_buffer(ctx->hwframe, frame, 0);
385 av_log(avctx, AV_LOG_ERROR, "av_hwframe_get_buffer failed\n");
389 ret = ff_decode_frame_props(avctx, frame);
391 av_log(avctx, AV_LOG_ERROR, "ff_decode_frame_props failed\n");
395 for (i = 0; i < 2; i++) {
396 CUDA_MEMCPY2D cpy = {
397 .srcMemoryType = CU_MEMORYTYPE_DEVICE,
398 .dstMemoryType = CU_MEMORYTYPE_DEVICE,
399 .srcDevice = mapped_frame,
400 .dstDevice = (CUdeviceptr)frame->data[i],
402 .dstPitch = frame->linesize[i],
404 .WidthInBytes = FFMIN(pitch, frame->linesize[i]),
405 .Height = avctx->height >> (i ? 1 : 0),
408 ret = CHECK_CU(cuMemcpy2D(&cpy));
412 offset += avctx->coded_height;
414 } else if (avctx->pix_fmt == AV_PIX_FMT_NV12) {
415 AVFrame *tmp_frame = av_frame_alloc();
417 av_log(avctx, AV_LOG_ERROR, "av_frame_alloc failed\n");
418 ret = AVERROR(ENOMEM);
422 tmp_frame->format = AV_PIX_FMT_CUDA;
423 tmp_frame->hw_frames_ctx = av_buffer_ref(ctx->hwframe);
424 tmp_frame->data[0] = (uint8_t*)mapped_frame;
425 tmp_frame->linesize[0] = pitch;
426 tmp_frame->data[1] = (uint8_t*)(mapped_frame + avctx->coded_height * pitch);
427 tmp_frame->linesize[1] = pitch;
428 tmp_frame->width = avctx->width;
429 tmp_frame->height = avctx->height;
431 ret = ff_get_buffer(avctx, frame, 0);
433 av_log(avctx, AV_LOG_ERROR, "ff_get_buffer failed\n");
434 av_frame_free(&tmp_frame);
438 ret = av_hwframe_transfer_data(frame, tmp_frame, 0);
440 av_log(avctx, AV_LOG_ERROR, "av_hwframe_transfer_data failed\n");
441 av_frame_free(&tmp_frame);
445 av_frame_free(&tmp_frame);
451 frame->width = avctx->width;
452 frame->height = avctx->height;
453 if (avctx->pkt_timebase.num && avctx->pkt_timebase.den)
454 frame->pts = av_rescale_q(parsed_frame.dispinfo.timestamp, (AVRational){1, 10000000}, avctx->pkt_timebase);
456 frame->pts = parsed_frame.dispinfo.timestamp;
458 if (parsed_frame.second_field) {
459 if (ctx->prev_pts == INT64_MIN) {
460 ctx->prev_pts = frame->pts;
461 frame->pts += (avctx->pkt_timebase.den * avctx->framerate.den) / (avctx->pkt_timebase.num * avctx->framerate.num);
463 int pts_diff = (frame->pts - ctx->prev_pts) / 2;
464 ctx->prev_pts = frame->pts;
465 frame->pts += pts_diff;
469 /* CUVIDs opaque reordering breaks the internal pkt logic.
470 * So set pkt_pts and clear all the other pkt_ fields.
472 frame->pkt_pts = frame->pts;
473 av_frame_set_pkt_pos(frame, -1);
474 av_frame_set_pkt_duration(frame, 0);
475 av_frame_set_pkt_size(frame, -1);
477 frame->interlaced_frame = !parsed_frame.is_deinterlacing && !parsed_frame.dispinfo.progressive_frame;
479 if (frame->interlaced_frame)
480 frame->top_field_first = parsed_frame.dispinfo.top_field_first;
481 } else if (ctx->decoder_flushing) {
484 ret = AVERROR(EAGAIN);
489 eret = CHECK_CU(cuvidUnmapVideoFrame(ctx->cudecoder, mapped_frame));
491 eret = CHECK_CU(cuCtxPopCurrent(&dummy));
499 static int cuvid_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *avpkt)
501 CuvidContext *ctx = avctx->priv_data;
502 AVFrame *frame = data;
505 av_log(avctx, AV_LOG_TRACE, "cuvid_decode_frame\n");
507 if (ctx->deint_mode != cudaVideoDeinterlaceMode_Weave) {
508 av_log(avctx, AV_LOG_ERROR, "Deinterlacing is not supported via the old API\n");
509 return AVERROR(EINVAL);
512 if (!ctx->decoder_flushing) {
513 ret = cuvid_decode_packet(avctx, avpkt);
518 ret = cuvid_output_frame(avctx, frame);
519 if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF) {
521 } else if (ret < 0) {
530 static av_cold int cuvid_decode_end(AVCodecContext *avctx)
532 CuvidContext *ctx = avctx->priv_data;
534 av_fifo_freep(&ctx->frame_queue);
537 av_bsf_free(&ctx->bsf);
540 cuvidDestroyVideoParser(ctx->cuparser);
543 cuvidDestroyDecoder(ctx->cudecoder);
545 av_buffer_unref(&ctx->hwframe);
546 av_buffer_unref(&ctx->hwdevice);
551 static int cuvid_test_dummy_decoder(AVCodecContext *avctx, CUVIDPARSERPARAMS *cuparseinfo)
553 CUVIDDECODECREATEINFO cuinfo;
554 CUvideodecoder cudec = 0;
557 memset(&cuinfo, 0, sizeof(cuinfo));
559 cuinfo.CodecType = cuparseinfo->CodecType;
560 cuinfo.ChromaFormat = cudaVideoChromaFormat_420;
561 cuinfo.OutputFormat = cudaVideoSurfaceFormat_NV12;
563 cuinfo.ulWidth = 1280;
564 cuinfo.ulHeight = 720;
565 cuinfo.ulTargetWidth = cuinfo.ulWidth;
566 cuinfo.ulTargetHeight = cuinfo.ulHeight;
568 cuinfo.target_rect.left = 0;
569 cuinfo.target_rect.top = 0;
570 cuinfo.target_rect.right = cuinfo.ulWidth;
571 cuinfo.target_rect.bottom = cuinfo.ulHeight;
573 cuinfo.ulNumDecodeSurfaces = MAX_FRAME_COUNT;
574 cuinfo.ulNumOutputSurfaces = 1;
575 cuinfo.ulCreationFlags = cudaVideoCreate_PreferCUVID;
576 cuinfo.bitDepthMinus8 = 0;
578 cuinfo.DeinterlaceMode = cudaVideoDeinterlaceMode_Weave;
580 ret = CHECK_CU(cuvidCreateDecoder(&cudec, &cuinfo));
584 ret = CHECK_CU(cuvidDestroyDecoder(cudec));
591 static av_cold int cuvid_decode_init(AVCodecContext *avctx)
593 CuvidContext *ctx = avctx->priv_data;
594 AVCUDADeviceContext *device_hwctx;
595 AVHWDeviceContext *device_ctx;
596 AVHWFramesContext *hwframe_ctx;
597 CUVIDSOURCEDATAPACKET seq_pkt;
598 CUcontext cuda_ctx = NULL;
600 const AVBitStreamFilter *bsf;
603 enum AVPixelFormat pix_fmts[3] = { AV_PIX_FMT_CUDA,
607 ret = ff_get_format(avctx, pix_fmts);
609 av_log(avctx, AV_LOG_ERROR, "ff_get_format failed: %d\n", ret);
613 ctx->frame_queue = av_fifo_alloc(MAX_FRAME_COUNT * sizeof(CuvidParsedFrame));
614 if (!ctx->frame_queue) {
615 ret = AVERROR(ENOMEM);
619 avctx->pix_fmt = ret;
621 if (avctx->hw_frames_ctx) {
622 ctx->hwframe = av_buffer_ref(avctx->hw_frames_ctx);
624 ret = AVERROR(ENOMEM);
628 hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
630 ctx->hwdevice = av_buffer_ref(hwframe_ctx->device_ref);
631 if (!ctx->hwdevice) {
632 ret = AVERROR(ENOMEM);
636 ret = av_hwdevice_ctx_create(&ctx->hwdevice, AV_HWDEVICE_TYPE_CUDA, ctx->cu_gpu, NULL, 0);
640 ctx->hwframe = av_hwframe_ctx_alloc(ctx->hwdevice);
642 av_log(avctx, AV_LOG_ERROR, "av_hwframe_ctx_alloc failed\n");
643 ret = AVERROR(ENOMEM);
647 hwframe_ctx = (AVHWFramesContext*)ctx->hwframe->data;
650 device_ctx = hwframe_ctx->device_ctx;
651 device_hwctx = device_ctx->hwctx;
652 cuda_ctx = device_hwctx->cuda_ctx;
654 memset(&ctx->cuparseinfo, 0, sizeof(ctx->cuparseinfo));
655 memset(&ctx->cuparse_ext, 0, sizeof(ctx->cuparse_ext));
656 memset(&seq_pkt, 0, sizeof(seq_pkt));
658 ctx->cuparseinfo.pExtVideoInfo = &ctx->cuparse_ext;
660 switch (avctx->codec->id) {
661 #if CONFIG_H263_CUVID_DECODER
662 case AV_CODEC_ID_H263:
663 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG4;
666 #if CONFIG_H264_CUVID_DECODER
667 case AV_CODEC_ID_H264:
668 ctx->cuparseinfo.CodecType = cudaVideoCodec_H264;
671 #if CONFIG_HEVC_CUVID_DECODER
672 case AV_CODEC_ID_HEVC:
673 ctx->cuparseinfo.CodecType = cudaVideoCodec_HEVC;
676 #if CONFIG_MJPEG_CUVID_DECODER
677 case AV_CODEC_ID_MJPEG:
678 ctx->cuparseinfo.CodecType = cudaVideoCodec_JPEG;
681 #if CONFIG_MPEG1_CUVID_DECODER
682 case AV_CODEC_ID_MPEG1VIDEO:
683 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG1;
686 #if CONFIG_MPEG2_CUVID_DECODER
687 case AV_CODEC_ID_MPEG2VIDEO:
688 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG2;
691 #if CONFIG_MPEG4_CUVID_DECODER
692 case AV_CODEC_ID_MPEG4:
693 ctx->cuparseinfo.CodecType = cudaVideoCodec_MPEG4;
696 #if CONFIG_VP8_CUVID_DECODER
697 case AV_CODEC_ID_VP8:
698 ctx->cuparseinfo.CodecType = cudaVideoCodec_VP8;
701 #if CONFIG_VP9_CUVID_DECODER
702 case AV_CODEC_ID_VP9:
703 ctx->cuparseinfo.CodecType = cudaVideoCodec_VP9;
706 #if CONFIG_VC1_CUVID_DECODER
707 case AV_CODEC_ID_VC1:
708 ctx->cuparseinfo.CodecType = cudaVideoCodec_VC1;
712 av_log(avctx, AV_LOG_ERROR, "Invalid CUVID codec!\n");
716 if (avctx->codec->id == AV_CODEC_ID_H264 || avctx->codec->id == AV_CODEC_ID_HEVC) {
717 if (avctx->codec->id == AV_CODEC_ID_H264)
718 bsf = av_bsf_get_by_name("h264_mp4toannexb");
720 bsf = av_bsf_get_by_name("hevc_mp4toannexb");
723 ret = AVERROR_BSF_NOT_FOUND;
726 if (ret = av_bsf_alloc(bsf, &ctx->bsf)) {
729 if (((ret = avcodec_parameters_from_context(ctx->bsf->par_in, avctx)) < 0) || ((ret = av_bsf_init(ctx->bsf)) < 0)) {
730 av_bsf_free(&ctx->bsf);
734 ctx->cuparse_ext.format.seqhdr_data_length = ctx->bsf->par_out->extradata_size;
735 memcpy(ctx->cuparse_ext.raw_seqhdr_data,
736 ctx->bsf->par_out->extradata,
737 FFMIN(sizeof(ctx->cuparse_ext.raw_seqhdr_data), ctx->bsf->par_out->extradata_size));
738 } else if (avctx->extradata_size > 0) {
739 ctx->cuparse_ext.format.seqhdr_data_length = avctx->extradata_size;
740 memcpy(ctx->cuparse_ext.raw_seqhdr_data,
742 FFMIN(sizeof(ctx->cuparse_ext.raw_seqhdr_data), avctx->extradata_size));
745 ctx->cuparseinfo.ulMaxNumDecodeSurfaces = MAX_FRAME_COUNT;
746 ctx->cuparseinfo.ulMaxDisplayDelay = 4;
747 ctx->cuparseinfo.pUserData = avctx;
748 ctx->cuparseinfo.pfnSequenceCallback = cuvid_handle_video_sequence;
749 ctx->cuparseinfo.pfnDecodePicture = cuvid_handle_picture_decode;
750 ctx->cuparseinfo.pfnDisplayPicture = cuvid_handle_picture_display;
752 ret = CHECK_CU(cuCtxPushCurrent(cuda_ctx));
756 ret = cuvid_test_dummy_decoder(avctx, &ctx->cuparseinfo);
760 ret = CHECK_CU(cuvidCreateVideoParser(&ctx->cuparser, &ctx->cuparseinfo));
764 seq_pkt.payload = ctx->cuparse_ext.raw_seqhdr_data;
765 seq_pkt.payload_size = ctx->cuparse_ext.format.seqhdr_data_length;
767 if (seq_pkt.payload && seq_pkt.payload_size) {
768 ret = CHECK_CU(cuvidParseVideoData(ctx->cuparser, &seq_pkt));
773 ret = CHECK_CU(cuCtxPopCurrent(&dummy));
777 ctx->prev_pts = INT64_MIN;
779 if (!avctx->pkt_timebase.num || !avctx->pkt_timebase.den)
780 av_log(avctx, AV_LOG_WARNING, "Invalid pkt_timebase, passing timestamps as-is.\n");
785 cuvid_decode_end(avctx);
789 static void cuvid_flush(AVCodecContext *avctx)
791 CuvidContext *ctx = avctx->priv_data;
792 AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)ctx->hwdevice->data;
793 AVCUDADeviceContext *device_hwctx = device_ctx->hwctx;
794 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
795 CUVIDSOURCEDATAPACKET seq_pkt = { 0 };
798 ret = CHECK_CU(cuCtxPushCurrent(cuda_ctx));
802 av_fifo_freep(&ctx->frame_queue);
804 ctx->frame_queue = av_fifo_alloc(MAX_FRAME_COUNT * sizeof(CuvidParsedFrame));
805 if (!ctx->frame_queue) {
806 av_log(avctx, AV_LOG_ERROR, "Failed to recreate frame queue on flush\n");
810 if (ctx->cudecoder) {
811 cuvidDestroyDecoder(ctx->cudecoder);
812 ctx->cudecoder = NULL;
816 cuvidDestroyVideoParser(ctx->cuparser);
817 ctx->cuparser = NULL;
820 ret = CHECK_CU(cuvidCreateVideoParser(&ctx->cuparser, &ctx->cuparseinfo));
824 seq_pkt.payload = ctx->cuparse_ext.raw_seqhdr_data;
825 seq_pkt.payload_size = ctx->cuparse_ext.format.seqhdr_data_length;
827 if (seq_pkt.payload && seq_pkt.payload_size) {
828 ret = CHECK_CU(cuvidParseVideoData(ctx->cuparser, &seq_pkt));
833 ret = CHECK_CU(cuCtxPopCurrent(&dummy));
837 ctx->prev_pts = INT64_MIN;
838 ctx->decoder_flushing = 0;
842 av_log(avctx, AV_LOG_ERROR, "CUDA reinit on flush failed\n");
845 #define OFFSET(x) offsetof(CuvidContext, x)
846 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
847 static const AVOption options[] = {
848 { "deint", "Set deinterlacing mode", OFFSET(deint_mode), AV_OPT_TYPE_INT, { .i64 = cudaVideoDeinterlaceMode_Weave }, cudaVideoDeinterlaceMode_Weave, cudaVideoDeinterlaceMode_Adaptive, VD, "deint" },
849 { "weave", "Weave deinterlacing (do nothing)", 0, AV_OPT_TYPE_CONST, { .i64 = cudaVideoDeinterlaceMode_Weave }, 0, 0, VD, "deint" },
850 { "bob", "Bob deinterlacing", 0, AV_OPT_TYPE_CONST, { .i64 = cudaVideoDeinterlaceMode_Bob }, 0, 0, VD, "deint" },
851 { "adaptive", "Adaptive deinterlacing", 0, AV_OPT_TYPE_CONST, { .i64 = cudaVideoDeinterlaceMode_Adaptive }, 0, 0, VD, "deint" },
852 { "gpu", "GPU to be used for decoding", OFFSET(cu_gpu), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, VD },
856 #define DEFINE_CUVID_CODEC(x, X) \
857 static const AVClass x##_cuvid_class = { \
858 .class_name = #x "_cuvid", \
859 .item_name = av_default_item_name, \
861 .version = LIBAVUTIL_VERSION_INT, \
863 AVHWAccel ff_##x##_cuvid_hwaccel = { \
864 .name = #x "_cuvid", \
865 .type = AVMEDIA_TYPE_VIDEO, \
866 .id = AV_CODEC_ID_##X, \
867 .pix_fmt = AV_PIX_FMT_CUDA, \
869 AVCodec ff_##x##_cuvid_decoder = { \
870 .name = #x "_cuvid", \
871 .long_name = NULL_IF_CONFIG_SMALL("Nvidia CUVID " #X " decoder"), \
872 .type = AVMEDIA_TYPE_VIDEO, \
873 .id = AV_CODEC_ID_##X, \
874 .priv_data_size = sizeof(CuvidContext), \
875 .priv_class = &x##_cuvid_class, \
876 .init = cuvid_decode_init, \
877 .close = cuvid_decode_end, \
878 .decode = cuvid_decode_frame, \
879 .send_packet = cuvid_decode_packet, \
880 .receive_frame = cuvid_output_frame, \
881 .flush = cuvid_flush, \
882 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_AVOID_PROBING, \
883 .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_CUDA, \
888 #if CONFIG_HEVC_CUVID_DECODER
889 DEFINE_CUVID_CODEC(hevc, HEVC)
892 #if CONFIG_H263_CUVID_DECODER
893 DEFINE_CUVID_CODEC(h263, H263)
896 #if CONFIG_H264_CUVID_DECODER
897 DEFINE_CUVID_CODEC(h264, H264)
900 #if CONFIG_MJPEG_CUVID_DECODER
901 DEFINE_CUVID_CODEC(mjpeg, MJPEG)
904 #if CONFIG_MPEG1_CUVID_DECODER
905 DEFINE_CUVID_CODEC(mpeg1, MPEG1VIDEO)
908 #if CONFIG_MPEG2_CUVID_DECODER
909 DEFINE_CUVID_CODEC(mpeg2, MPEG2VIDEO)
912 #if CONFIG_MPEG4_CUVID_DECODER
913 DEFINE_CUVID_CODEC(mpeg4, MPEG4)
916 #if CONFIG_VP8_CUVID_DECODER
917 DEFINE_CUVID_CODEC(vp8, VP8)
920 #if CONFIG_VP9_CUVID_DECODER
921 DEFINE_CUVID_CODEC(vp9, VP9)
924 #if CONFIG_VC1_CUVID_DECODER
925 DEFINE_CUVID_CODEC(vc1, VC1)