mirrors/gstreamer
mirrors/gstreamer
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gstreamer.git synced 2024-12-23 00:36:51 +00:00
Code Issues 1 Projects Releases Wiki Activity

nvdecoder: Skip synchronization if downstream buffer holds CUDA stream

Browse source 
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3629>
This commit is contained in:
Seungha Yang 2022-12-21 00:13:56 +09:00 committed by GStreamer Marge Bot
parent ddc5f1d425
commit aabcba16db
2 changed files with 38 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
subprojects/gst-plugins-bad/sys/nvcodec/gstnvdec.c
View file

@ -1272,14 +1272,23 @@ gst_nvdec_copy_device_to_memory (GstNvDec * nvdec,
gint i;
GstMemory *mem;
gboolean use_device_copy = FALSE;
gboolean need_sync = TRUE;
GstMapFlags map_flags = GST_MAP_WRITE;
CUstream stream = gst_cuda_stream_get_handle (nvdec->stream);
if (nvdec->mem_type == GST_NVDEC_MEM_TYPE_CUDA &&
(mem = gst_buffer_peek_memory (output_buffer, 0)) &&
gst_is_cuda_memory (mem)) {
GstCudaStream *mem_stream;
map_flags |= GST_MAP_CUDA;
use_device_copy = TRUE;
mem_stream = gst_cuda_memory_get_stream (GST_CUDA_MEMORY_CAST (mem));
if (mem_stream) {
stream = gst_cuda_stream_get_handle (mem_stream);
need_sync = FALSE;
}
}
if (!gst_video_frame_map (&video_frame, info, output_buffer, map_flags)) {
@ -1333,7 +1342,8 @@ gst_nvdec_copy_device_to_memory (GstNvDec * nvdec,
}
}
gst_cuda_result (CuStreamSynchronize (stream));
if (need_sync)
gst_cuda_result (CuStreamSynchronize (stream));
gst_video_frame_unmap (&video_frame);

35
subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.c
View file

@ -372,7 +372,7 @@ gst_nv_decoder_new_frame (GstNvDecoder * decoder)
/* must be called with gst_cuda_context_push */
static gboolean
gst_nv_decoder_frame_map (GstNvDecoderFrame * frame)
gst_nv_decoder_frame_map (GstNvDecoderFrame * frame, GstCudaStream * stream)
{
GstNvDecoder *self;
CUVIDPROCPARAMS params = { 0 };
@ -385,7 +385,7 @@ gst_nv_decoder_frame_map (GstNvDecoderFrame * frame)
/* TODO: check interlaced */
params.progressive_frame = 1;
params.output_stream = gst_cuda_stream_get_handle (self->stream);
params.output_stream = gst_cuda_stream_get_handle (stream);
if (frame->mapped) {
GST_WARNING_OBJECT (self, "Frame %p is mapped already", frame);
@ -749,14 +749,15 @@ done:
static gboolean
gst_nv_decoder_copy_frame_to_cuda (GstNvDecoder * decoder,
GstNvDecoderFrame * frame, GstBuffer * buffer)
GstNvDecoderFrame * frame, GstBuffer * buffer, GstCudaStream * stream,
gboolean need_sync)
{
CUDA_MEMCPY2D copy_params = { 0, };
GstMemory *mem;
gint i;
gboolean ret = FALSE;
GstVideoFrame video_frame;
CUstream stream = gst_cuda_stream_get_handle (decoder->stream);
CUstream stream_handle = gst_cuda_stream_get_handle (stream);
mem = gst_buffer_peek_memory (buffer, 0);
if (!gst_is_cuda_memory (mem)) {
@ -790,13 +791,15 @@ gst_nv_decoder_copy_frame_to_cuda (GstNvDecoder * decoder,
* GST_VIDEO_INFO_COMP_PSTRIDE (&decoder->info, 0);
copy_params.Height = GST_VIDEO_INFO_COMP_HEIGHT (&decoder->info, i);
if (!gst_cuda_result (CuMemcpy2DAsync (&copy_params, stream))) {
if (!gst_cuda_result (CuMemcpy2DAsync (&copy_params, stream_handle))) {
GST_ERROR_OBJECT (decoder, "failed to copy %dth plane", i);
goto done;
}
}
gst_cuda_result (CuStreamSynchronize (stream));
/* Don't sync if we are using downstream memory's stream */
if (need_sync)
gst_cuda_result (CuStreamSynchronize (stream_handle));
ret = TRUE;
@ -816,6 +819,9 @@ gst_nv_decoder_finish_frame (GstNvDecoder * decoder, GstVideoDecoder * videodec,
{
GstBuffer *outbuf = NULL;
gboolean ret = FALSE;
GstCudaStream *stream;
GstCudaStream *mem_stream = NULL;
gboolean need_sync = TRUE;
g_return_val_if_fail (GST_IS_NV_DECODER (decoder), GST_FLOW_ERROR);
g_return_val_if_fail (GST_IS_VIDEO_DECODER (videodec), GST_FLOW_ERROR);
@ -835,12 +841,24 @@ gst_nv_decoder_finish_frame (GstNvDecoder * decoder, GstVideoDecoder * videodec,
return FALSE;
}
stream = decoder->stream;
if (decoder->output_type == GST_NV_DECODER_OUTPUT_TYPE_CUDA) {
GstCudaMemory *cmem = (GstCudaMemory *) gst_buffer_peek_memory (outbuf, 0);
/* Use downstream CUDA stream if available */
mem_stream = gst_cuda_memory_get_stream (cmem);
if (mem_stream) {
need_sync = FALSE;
stream = mem_stream;
}
}
if (!gst_cuda_context_push (decoder->context)) {
GST_ERROR_OBJECT (decoder, "Failed to push CUDA context");
goto error;
}
if (!gst_nv_decoder_frame_map (frame)) {
if (!gst_nv_decoder_frame_map (frame, stream)) {
GST_ERROR_OBJECT (decoder, "Couldn't map frame");
gst_cuda_context_pop (NULL);
goto error;
@ -861,7 +879,8 @@ gst_nv_decoder_finish_frame (GstNvDecoder * decoder, GstVideoDecoder * videodec,
break;
#endif
case GST_NV_DECODER_OUTPUT_TYPE_CUDA:
ret = gst_nv_decoder_copy_frame_to_cuda (decoder, frame, outbuf);
ret = gst_nv_decoder_copy_frame_to_cuda (decoder,
frame, outbuf, stream, need_sync);
break;
default:
g_assert_not_reached ();