For loop in model causes extra CUDA memcpy HtoD

ethancaballero opened this issue 5 years ago · 0 comments

ethancaballero commented 5 years ago

Probably has something to do with arithmetic op in this line
https://github.com/mila-iqia/myia/blob/master/myia/composite.py#L363

Here is example that causes extra "CUDA memcpy HtoD":
https://github.com/mila-iqia/myia/blob/master/examples/rnn.py#L118-L119