Thomas Unterthiner
Thomas Unterthiner
There are no numpy orientation flags. The objects are `pycuda.gpuarray`, not `numpy.ndarray`. ``pycuda.gpuarray.flags` only as flags for `f_contiguous` and `c_contiguous`, and both are false for a strided array (since it...
One option would be the following: first, look if either `f_contiguous` or 'c_contiguous` are True. Only use the strides to determine orientation when both are false. Does that solve the...
Sorry for not having a look at this sooner. I personally think this looks nice! I like the idea of a CSR Matrix a lot, but I'd stick with some...
Having looked through it in more detail, I think this is really nice work, and the approach could be re-used to wrap other things (CULA/Magma/cuDNN/....). But of course, it's up...
Just as a head's up: I ran into a bad memory-leak within cuSOLVER's cholesky factorization, so depending on your needs, you might want to avoid those calls: they leak GPU...
No, IIRC it was fixed in CUDA 7.5
Just food for thought: Since CUDA 7.0 brings its own set of LAPACK-functions ( https://developer.nvidia.com/cusolver ), it probably doesn't pay off to invest a lot of time into improving Magma-wrappers.
yes, I'll get right on it
@davidweichiang I'm a bit weirded out by your last example: ``` # This causes an exception in numpy but not in add_matvec print skcuda.misc.add_matvec(x23, a2) print x23.get() + a2.get() ```...
> which handles 0 with 3+ and 3+ with 0. Sorry, but I do not understand what you mean by that?