nvdec: Don't use default CUDA stream

NVDEC launches CUDA kernel function (ConvertNV12BLtoNV12 or so)
when CuvidMapVideoFrame() is called. Which seems to be
NVDEC's internal post-processing kernel function, maybe
to convert tiled YUV to linear YUV format or something similar.

A problem if we don't pass CUDA stream to the CuvidMapVideoFrame()
call is that the NVDEC's internel kernel function will use default CUDA stream.
Then lots of the other CUDA API calls will be blocked/serialized.

To avoid the unnecessary blocking, we should pass our own
CUDA stream object to the CuvidMapVideoFrame() call

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3605>

diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdec.c b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdec.c
index ff73773..8c259cf 100644 (file)
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdec.c
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdec.c
@@ -1167,6 +1167,7 @@ copy_video_frame_to_gl_textures (GstGLContext * context,
   proc_params.progressive_frame = dispinfo->progressive_frame;
   proc_params.top_field_first = dispinfo->top_field_first;
   proc_params.unpaired_field = dispinfo->repeat_first_field == -1;
+  proc_params.output_stream = nvdec->cuda_stream;
 
   data->ret = TRUE;
 
@@ -1309,6 +1310,7 @@ gst_nvdec_copy_device_to_memory (GstNvDec * nvdec,
   params.second_field = dispinfo->repeat_first_field + 1;
   params.top_field_first = dispinfo->top_field_first;
   params.unpaired_field = dispinfo->repeat_first_field < 0;
+  params.output_stream = nvdec->cuda_stream;
 
   if (!gst_cuda_result (CuvidMapVideoFrame (nvdec->decoder,
               dispinfo->picture_index, &dptr, &pitch, &params))) {

diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.c b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.c
index d39938c..e9c7baf 100644 (file)
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.c
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.c
@@ -397,6 +397,7 @@ gst_nv_decoder_frame_map (GstNvDecoderFrame * frame)
 
   /* TODO: check interlaced */
   params.progressive_frame = 1;
+  params.output_stream = self->cuda_stream;
 
   if (frame->mapped) {
     GST_WARNING_OBJECT (self, "Frame %p is mapped already", frame);