I didn't check if numpy have that sort of code. I have implemented it in Theano.
Your proposal is good for me. I keep that solution in mind when I'm there. For the new name of .gpudata, what about using .data as numpy? The only problem that I see is that people use the new gpu array with numpy code and that the numpy code use the .data without checking if its an numpy.ndarray first. I don't think this is going to be a real problem. What do you think? thanks Fred 2011/5/3 Andreas Kloeckner <[email protected]>: > On Tue, 3 May 2011 14:54:49 -0400, Frédéric Bastien <[email protected]> wrote: >> I was just tring to see the potential problem that we could have. When >> I redo those optimization, I will try in python first and check if it >> is a problem or not. At least, I know I will need to use something >> else then pycuda in that case. > > I think this can be done once, generically (i.e. for arbitrary number of > in/outputs) in C/C++, and I'd be perfectly OK having such a thing in > PyCUDA's compiled part. Presumably, this code could even be stolen > one-to-one from numpy. > > Andreas > _______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
