This release is an alpha. It is expected to have bugs and has been tested on a limited number of platforms compared to the main distribution. The supported platform is CentOS 6.2 \ Intel Compiler v13.0.1
Please contact shoc-help@elist.ornl.gov with any questions concerning this release.
Please see the release notes below for other limitations and known issues.
Type make
inside this directory. This assumes the intel toolchain
(a recent version of icc
) is in your current path. Your paths
should also include MKL, which is used in some of the benchmarks.
How to run all the benchmarks:
1) Use the provided run script in /bin
How to run an individual benchmark:
2) Set the appropriate environment variables (see run scripts)
For example, to run MD workload
$ export MIC_ENV_PREFIX=MIC
$ export MIC_USE_2MB_BUFFERS=32K
$ export MIC_KMP_AFFINITY=granularity=fine,balanced
$ export MIC_BUFFERSIZE=128M
$ export nc=61
$ export MIC_OMP_NUM_THREADS=4*$nc-4
$ ./MD -s 4
For another example, to run FFT workload
$ export MIC_ENV_PREFIX=MIC
$ export MIC_USE_2MB_BUFFERS=32K
$ export MIC_KMP_AFFINITY=granularity=fine,balanced
$ export MIC_BUFFERSIZE=128M
$ export MIC_MKL_DYNAMIC=false
$ nc=61
$ export MIC_OMP_NUM_THREADS=4*$nc-4
$ ./FFT -s 4
The value for -s
is a number between 1 and four that corresponds to
problem size from smallest to largest (consistent with the CUDA and
OpenCL versions).