How is repa implemented?

75 views
Skip to first unread message

alan hawkins

unread,
Jun 9, 2016, 8:11:22 PM6/9/16
to Haskell Repa
Hi,

I'm curious as to how this is implemented under the hood.

Is this just a multi threaded CPU thing thing or is there SIMD and/or GPU involved as well?

Regards,

Alan Hawkins

Ben Lippmeier

unread,
Jun 9, 2016, 10:17:02 PM6/9/16
to alan hawkins, Haskell Repa
On 10 Jun 2016, at 10:01 am, alan hawkins <hawk...@gmail.com> wrote:
I'm curious as to how this is implemented under the hood.

Is this just a multi threaded CPU thing thing or is there SIMD and/or GPU involved as well?


Multithreaded CPU thing.

Repa doesn’t do any SIMD vectorisation itself. In some cases LLVM could perhaps be provoked applying its own vectorisation pass, but it would be very fiddly due to aliasing issues, and I haven’t tried it.

If you want to run something on the GPU then try Accelerate.


Ben.


Reply all
Reply to author
Forward
0 new messages