Connecting and running Xeon Phi Accelerators

In order to connect to Xeon Phi co-processor, compile or run software directly on the co-processor (not via offload), the following steps are required:

1. From the main machine (rclsms, 10.10.10.74) connect via ssh to xphi1.

2. Update the environment variables according to the Intel Parallel Studio environment, using the command:

source /opt/sw/intel/parallel_studio_xe_2017/psxevars.sh

3. The program can be compiled using icc for c / cpp code, or using ifort for fortran code. Use the following flags:

-mmic - performs compilation for the co-processor.

-O3 - optimization.

-qopenmp - parallelize using OpenMP .

-o {name} - name of the executable file.

For example:

ifort -O3 -mmic -qopenmp -o prog mycode.f90

4. Update the environment variable:

export SINK_LD_LIBRARY_PATH=/opt/sw/intel/compilers_and_libraries/linux/lib/mic/

This environment variable defines the micnativeloadex tool where to search the co-processor libraries.

5. To run the program use the command:

micnativeloadex {program_name}

The micnativeloadex tool loads and runs the program in the co-processor, saving the need for connection to ssh and setting environment variables in the mic.