Compiling parallel scientific codes on parallella

[gmogni]

Hello,

First of all I have to admit that I have little experience running parallel codes outside of normal SMP desktop machines, so therefore please forgive my ignorance about the way Parallella works. Before I invest any money into buying Parallella boards, I would like to know: is it possible to simply compile pre-existing scientific codes such as the VASP Quantum chemistry code on the parallella board in order to take advantage of the 16-cores Epiphany CPU, or does the code necessarily have to support GPGPU protocols such as CUDA or OpenCL to be run on the RISC cores of the Epiphany? The code I would like to use (VASP) provides Makefiles for the following architectures: Altrix, Cray, DEC, HPUX Itanium, IBM SPX, Linux Absoft compiler, Intel Fortran compiler for ITANIUM 2 systems, Linux Gfortran, Intel Fortran compiler for Pentium/Athlon/Opteron, Portland Group F90/HPF compiler, NEC SX-5, RS 6000, FUJITSU VPP, SUN, SGI and HP PA-RISC..... I normally use Linux Gfortran on Ubuntu and it works just fine. Would any of these makefiles be suitable to compile VASP on the Epiphany architecture?

Many thanks,

Gabriele

[sebraa]

[gmogni]

OK I understand, so I suppose there’s little chance that any of those makefiles will ever work on Epiphany, including the HP PA-RISC one which I suppose was developed for a similar HP RISC architecture?

[gmogni]

Another question: is there any easy way to port a Fortran code that is already compatible with OpenMPI to OpenCL, or does it require re-writing the code from scratch? I’m not much of a MPI developer myself, more of a MPI utilizer

[gmogni]

Actually upon re-reading your reply you mentioned that Epiphany is also MPI compatible. Does that mean that VASP will be Epiphany-compatible if it is already designed to support OpenMPI or MPICH?

[sebraa]

[gmogni]

OK, I still believe that compiling VASP with the gfortran makefile and OpenMPI support through mpif90 might have half a chance to work, as this particular makefile is compiler-dependent as opposed to architecture-dependent.... I will give it a go as soon as I get hold of a parallella board and report on any success!

[gmogni]

One last question I have is: is there an easy way to check the CPU usage on the Epiphany to make sure applications executed there are running correctly?

[sebraa]

[gmogni]

Do you agree that this is the correct way of executing MPI programs on the Epiphany?

http://stackoverflow.com/questions/2551 ... permission