Linux not recognizing all of the cores?

[baseballguy]

Hi everyone,

I was running some benchmarks on the Parallella 16 core board and I was getting some weird results. I used the "nproc" command (described here: http://www.cyberciti.biz/faq/linux-get- ... e-command/) to see how many processing units are available. I was expecting to see 16, but instead 2 was outputted. I attached a picture of the rest of the details. I was wondering if anyone came across this problem or knows what to do to fix it?

Thanks

[toralf]

Shouldn't that output just show the ARM cores ?

[tincman]

[baseballguy]

Sorry, I'm pretty new to parallel programming and stuff, but is it possible to run these benchmarks (PARSEC) on the Epiphany cores instead of the ARM cores?

[baseballguy]

Basically from what I understand, the CPU usage will indicate how many cores are running. For example, if 8 cores really are running, then the CPU usage would be 800%. If 4 cores are running , it would be at 400%. However, all of the PARSEC programs maxed out at 200% showing that only 2 cores(the ARM) are running.

Is my understanding correct or am I missing something?

Thanks

[9600]

[baseballguy]

Thanks for the reply. So my next step would be to try and port the PARSEC programs so that they have host and Epiphany codes?

Also, is there any way to know if the Epiphany cores are working? (any library call?)

Thanks for the help

[9600]

[baseballguy]

Right now, I'm trying to run the Parsec benchmarks on the Epiphany cores. The programs on Parsec have this option where you get to choose your own build configuration out of these options:

- 'gcc' Build parallel version of suite with gcc
- 'gcc-serial' Build serial version of suite with gcc
- 'gcc-hooks' Build parallel version of suite with PARSEC hooks
enabled with gcc
- 'icc' Build parallel version of suite with Intel compiler
- 'gcc-pthreads' Build with pthreads parallelization (if supported)
- 'gcc-openmp' Build with OpenMP parallelization (if supported)
- 'gcc-tbb' Build with TBB parallelization (if supported)

Let's take the pthreads one for example. Let's say that I build the program using the command with the pthreads. What would be the next step to making sure that the program runs on 16 cores?

P.S (This is what I tried to do): I tried to say "parsecmgmt -a build -p fluidanimate -c gcc-pthreads" and then I said "./run.sh". The program worked fine. However, I used the erm program(https://github.com/adapteva/epiphany-ex ... r/apps/erm) to track the cores' activity and there were no changes(all the statistics stayed at 0)

Any help you guys could give me would be appreciated.

[sebraa]

If you select gcc-pthreads, your benchmark is compiled to use POSIX threads using the ARM cores. So it will parallelize, just not on the Epiphany.

The host application is responsible for loading the programs into the Epiphany cores, communicate with them and make use of the data it receives. It is compiled with the host compiler (gcc) and runs on the host (ARM).
Then, you have the Epiphany applications, which are doing the heavy lifting. They are compiled with the Epiphany compiler (e-gcc), which is part of the SDK.

You cannot take a ready-made program, compile it with the defaults and expect it to use your Epiphany (unless it is written for or specifically supports it). It might work with OpenCL, but I haven't looked at it yet. So unless your application supports Epiphany or OpenCL, you need to port it yourself first.