csarofeen / pytorch

Tensors and Dynamic neural networks in Python with strong GPU acceleration
http://pytorch.org
Other
26 stars 7 forks source link

Adding current stream to evt record which is reqd. #2521

Closed mmigdal-nv closed 1 year ago

mmigdal-nv commented 1 year ago

Kernels are not necessarily launched on the default stream. For the benchmarks to be correct, we need to record the start/end events on the stream where kernels are launched.