Running MAGMA (GPU) code with MAGMA MIC

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
Post Reply
jrm4412
Posts: 5
Joined: Wed Nov 05, 2014 11:29 pm

Running MAGMA (GPU) code with MAGMA MIC

Post by jrm4412 » Fri Nov 14, 2014 7:28 pm

So, I've finally got all my code up and running using MAGMA and an Nvidia C2050 GPU. I'm calling the MAGMA routines from Fortran using a combination of the provided Fortran wrappers (e.g. calling magmaf_[myfunction]) and wrappers that I have written to allocate device memory and send the provided matrices/vectors to the device, etc. This wrapper is similar to Cublas' fortran_thunking, but all variables and calls use magma routines (magma_zmalloc, malloc_zsetmatrix, etc).

Soon, I'll be installing an Intel Phi on my workstation and will have a very short timeframe for getting some testing done using MAGMA.

My question is: when I install MAGMA MIC and start the server on the MIC, can I simply simply recycle my code with the correct flags to run it on the Phi?

luszczek
Posts: 3
Joined: Fri Jul 17, 2015 10:25 am

Re: Running MAGMA (GPU) code with MAGMA MIC

Post by luszczek » Fri Jul 17, 2015 10:30 am

Yes, this is the intended behavior. Please use the latest MAGMA release for Phi: 1.4.0

Post Reply