This repository supports readers of our book:
The OpenMP Common Core: making OpenMP Simple Again
We encourage readers to explore the material actively; mixing time spent reading with time spent coding. To that end, we provide lots of code for our readers to explore examples organized into the following directories.
Book: Code used in the book. This includes any
code that appears in the figures as well as code
fragments embedded in the text of the book.
We provide these in both C and Fortran.
ChallengeProblems: A set of more complex problems to
support advanced studies in OpenMP. In each case,
we include a serial code you can start with and a
directory of solutions. Currently, these are
only provided in C.
Exercises: code and solutions for the running
examples used in the book. We recommend pausing
as you read the book and experiment with these
exercises ... after all, we all learn most "by doing"
We provide these in both C and Fortran.
MakeDefs: Files included by our makefiles to support
C and Fortran on various platforms. You should copy
one that is close to what you need, and modify it
to reference the compilers you wish to use. Put
this copy in the MakeDefs directory and call it make.def
(for Perlmutter, copy perlmutter_gnu.def to be make.def)
You can use any system with a working OpenMP compiler for the hands-on exercises in the "Exercises" directory. The templates for "make.def" to be used for different architectures are available in the "MakeDefs" directory.
Training accounts on the NERSC Perlmutter system for SC23 are provided from Nov 11 to Nov 17, 2023:
-- Get an account on Perlmutter
-- Apply for a NERSC training account: https://iris.nersc.gov/train use 4-letter code: epg6
-- Obtain the exercises on Perlmutter
% ssh trainxxx@perlmutter.nersc.gov
% cd $SCRATCH
% cp -r /global/cfs/cdirs/training/2023/OpenMP_SC23 . (notice the “dot” at the end)
% cd OpenMP_SC23/Exercises/C (or % cd OpenMP_SC23/Exercises/Fortran)
% make (for all codes)
-- Compile on Perlmutter
The default compiler is GNU. Use compiler wrappers (ftn, cc, and CC)
and the OpenMP compiler flag to build an individual code, such as:
% cc -fopenmp -O0 mycode.c
% ftn -fopenmp -O0 mycode.f90
To use another compiler, such as Nvidia:
% module swap PrgEnv-gnu PrgEnv-nvidia
% cc -mp -O0 mycode.c
Here -O0 is used on purpose to prevent compiler optimizations from
"fixing" certain OpenMP issues (such as false sharing) that we would
like to expose. You can set to a higher optimization level such as
-O3 to improve performance.
-- Run on Perlmutter
Get on a compute node with an interactive batch session first
With a node reservation during the SC23 tutorial:
– to get a dedicated node:
% salloc -N 1 -C cpu -q regular -t 1:00:00 --reservation=omp_sc23 -A ntrain2
- to run on a shared node:
% salloc -N 1 -C cpu -q shared -c 16 -t 1:00:00 --reservation=omp_sc23 -A ntrain2
Without a node reservation outside the SC23 tutorial:
– to get a dedicated node:
% salloc -N 1 -C cpu -q regular -t 1:00:00 -A ntrain2
- to run on a shared node:
% salloc -N 1 -C cpu -q shared -c 16 -t 1:00:00 -A ntrain2
To run on the allocated compute node:
-- Set number of threads, such as:
% export OMP_NUM_THREADS=4
-- Then run: % ./a.out