• Frederic's avatar
    Use the async gpu kernel call by default. · 9cd61627
    Frederic 提交于
    Our transfer call are synchronized version. So no problem there. The
    problem that we need to work around is that the Theano gc could free
    the output var before we finish with it. cudaFree is instananeous, it
    don't get in the stream of command to execute.
    9cd61627
cuda_ndarray.cuh 10.3 KB