GMEX - GFOR Capability and GCOMPILE

[Old posts from the commercial version of ArrayFire] Discussion of ArrayFire using CUDA or OpenCL.

Moderator: pavanky

GMEX - GFOR Capability and GCOMPILE

Postby erbilkonuk » Tue Mar 29, 2011 3:57 pm

Hi everyone...

I wanted to ask the followings..

* What happens if we have CPU compiled mex files deep under the GPU-ran functions? I got no error but garbage output so it is easy to get confused.

* Are gmex and nvcc somehow related considering the GPU build mechanism? Are gmex and GCOMPILE related? And the last combination of course :) Are GCOMPILE and nvcc related?

* Documentation for GMEX says
N/A. GMEX is utilized to construct MEX functions that can be GFOR-capable.

What does it mean? Will MEX functions under GFOR loop run parallel on GPU cores or we have to wrap the MEX function and compile it with GCOMPILE for GFOR capability?

* What is the difference of GMEX and GCOMPILE? Which one are faster? I know that the GMEX produces native API calls but what about GCOMPILE? Are the GCOMPILE outputs run by the Jacket runtime Engine?

Any help and information are greatly appreciated.. Many Thanks..
erbilkonuk
 
Posts: 8
Joined: Fri Nov 20, 2009 5:23 am

Return to [archive-commercial] Programming & Development with ArrayFire

cron