[PyCUDA] Non-contiguous elementwise kernels

[Thomas Unterthiner]

Hi!

Currently pycuda has very little support for non-contiguous arrays. Are
there any plans to extend GPUArray in this direction in the near future?
More concretely: in our application, we would be very desirable to be
able to write elementwise kernels for non-contiguous arrays. Is this
something that can easily done, and I'm just not seeing it?

(I've tried implementing my own version, but was unable to figure out
how to map the thread IDs to valid memory addresses in a general way.
Any pointers/)

Best

Thomas

[Bogdan Opanchuk]

Hi Thomas,

Does PyCUDA have any support for non-contiguous arrays at all?

(I've tried implementing my own version, but was unable to figure out how

to map the thread IDs to valid memory addresses in a general way. Any
pointers/)

I have support for custom strides in my Reikna library, and it just uses
the GPUArray's `strides` attributes to build a flat address out of an index
tuple. Is that what you mean?

On Wed, Oct 28, 2015 at 1:46 AM, Thomas Unterthiner <

Hi!
Currently pycuda has very little support for non-contiguous arrays. Are
there any plans to extend GPUArray in this direction in the near future?
More concretely: in our application, we would be very desirable to be able
to write elementwise kernels for non-contiguous arrays. Is this something
that can easily done, and I'm just not seeing it?
(I've tried implementing my own version, but was unable to figure out how
to map the thread IDs to valid memory addresses in a general way. Any
pointers/)
Best
Thomas
_______________________________________________
PyCUDA mailing list
http://lists.tiker.net/listinfo/pycuda