Parallella Community

Is is possible to use the current MPI libraries (v1.6.x) with stdcl only without using coprthr?

I'm trying to get a performance test running and I always get JIT compiler errors when I run my test system even though the kernel is linked. The examples nbody_mpi and mpi_fft2d use coprthr and I'm wondering if that is a requirement.

nick

I think it's probably possible, but not very easy. You could look at the compilation steps of clcc with cldebug and replicate the important parts of the compilation while still linking with the MPI library. I haven't tried this, though.

Thanks JAR,

I'll give it a go. There's a lot of stuff in the cldebug output so I'll make up my mind pretty quickly to either pursue it or just bite the bullet and go with coprthr.

nick

Hi, I read your ANN blogs and executed the nnP code. There is a question that I want to discuss.
I defined a 48,48,34 network and it worked well, the predict process will cost ~0.15s(forka executing time), but in ARM a same topology network based on opencv only need ~0.02s. I changed the cl code to make the "forwardPass" run 2 to 100 times in "k_forward", the forka time only increased ~0.005s*n. Is that means the forward predict process only cost ~0.005s in Epiphany, and the data movement cost most of the other time? If that is true, do you think there is any way to reduce the time cost?

This is really a new discussion. I'm copying your post to a new thread.