As you mentioned, functions of core.matrix are well-defined.
So openCL may be suitable.
(call native functions by JNA)
(ereduce some-reduce-function (emap some-mapping-function some-gpu-array))
This is wonderful.
Since GPU is appropriate for "map/reduce operations", I think there is a
room for i
Hello Mikera,
Thanks!
I'll check the numerical Clojure group.
2014年8月7日木曜日 2時46分18秒 UTC+9 Mikera:
>
> Sounds like a great project, would be really great to have a core.matrix
> implementation running on the GPU!
>
> I'm not too familiar with aparapi. Since core.matrix uses fairly
> well-defined
Sounds like a great project, would be really great to have a core.matrix
implementation running on the GPU!
I'm not too familiar with aparapi. Since core.matrix uses fairly
well-defined operations, it might be simpler to code the key ones directly
in OpenCL rather than going via Java bytecode t
