I’m having a little difficulty with type inference in real and imag. I have the following expression:
creal(t1[par__j // 3, par__j + (-3)*(par__j // 3), 0])
t1: type: np:dtype('complex128'), shape: (1, 3, 2), dim_tags: (N2:stride:6, N1:stride:2, N0:stride:1) scope:auto
and par__j and par__i are iname implementation tags.
This causes type inference to fail:
I presume that this is because I haven’t told loopy that creal has a return type of double. Where should I be putting this information?
Dr David Ham
Department of Mathematics
Imperial College London
I would like to use loo.py to make multi GPU with my python software. It
already works using one GPU.
Did you try to use loo.py with multi GPU (and NVLink infrastructure) ?
If it's possible, what do I have to know ?
If not, is it implemented by using pycuda or pyOpenCL ?