@@ -25,9 +25,7 @@ pinned to specific CPU cores, to get reliable and deterministic results. There
2525are several ways to achieve this:
2626
2727* Through environment variables. All MPI/OpenMP distributions provide a set of
28- environment variables to control process/thread pinning. Devito also
29- supplies the ` set_omp_pinning.sh ` program (under ` /scripts ` ), which helps
30- with thread pinning (though, currently, only limited to Intel architectures).
28+ environment variables to control process/thread pinning.
3129* Through a program such as ` numactl ` or ` taskset ` .
3230
3331If running on a NUMA system, where multiple nodes of CPU cores ("sockets") and
@@ -46,7 +44,7 @@ working set). There are multiple scenarios that are worth considering:
4644 logical cores are created. This can be changed by setting the OpenMP-standard
4745 ` OMP_NUM_THREADS ` environment variable to a different value. When might we
4846 want to do this?
49- - Unless on a hyperthreads-centerd system, such as an Intel Knights Landing,
47+ - Unless on a hyperthreads-centered system, such as an Intel Knights Landing,
5048 spawning only as many threads as * physical* cores usually results in
5149 slightly better performance due to less contention for hardware resources.
5250 - Since, here, we are merely interested in benchmarking, when running on a
0 commit comments