Chapel

Chapel

Chapel is a general-purpose, compiled, high-level parallel programming language with built-in abstractions for shared- and distributed-memory parallelism. There are two styles of parallel programming in Chapel: (1) task parallelism, in which parallelism is driven by programmer-specified tasks, and (2) data parallelism, in which parallelism is driven by computations over collections of data elements or their indices sitting in shared memory on one node or distributed among multiple nodes.

These high-level abstractions make Chapel ideal for learning parallel programming for a novice HPC user. Chapel is incredibly intuitive, striving to merge the ease-of-use of Python and the performance of traditional compiled languages such as C and Fortran. Parallel constructs that typically take tens of lines of MPI code can be expressed in only a few lines of Chapel code. Chapel is open source and can run on any Unix-like operating system, with hardware support from laptops to large HPC systems.

Chapel has a relatively small user base, so many libraries that exist for C, C++, Fortran have not yet been implemented in Chapel. Hopefully, that will change in coming years, if Chapel adoption continues to gain momentum in the HPC community.

For more information, please watch our three-part Chapel webinar.

Single-locale Chapel

On Cedar and Graham, single-locale (shared-memory) Chapel is implemented as a module. For testing, you can use salloc to run Chapel codes in serial:

[name@server ~]$ module load gcc chapel-single/1.15.0
[name@server ~]$ chpl test.chpl -o test
[name@server ~]$ ./test

or on multiple cores on the same node:

[name@server ~]$ module load gcc chapel-single/1.15.0
[name@server ~]$ chpl test.chpl -o test
[name@server ~]$ ./test

For production jobs, please write a job submission script and submit it with sbatch.

Multi-locale Chapel

Installing multi-locale (distributed-memory) Chapel requires fine-tuning its launcher for the specific physical interconnect on the cluster. Since Compute Canada clusters Cedar and Graham employ two different physical interconnects, we do not have one multi-locale Chapel module for both machines. Instead, multi-locale Chapel has been compiled in a separate directory as an experimental setup on each system. You can test this setup with the following Chapel code printing basic information about the nodes available inside your job:

File : probeLocales.chpl

use Memory;
for loc in Locales do
  on loc {
    writeln("locale #", here.id, "...");
    writeln("  ...is named: ", here.name);
    writeln("  ...has ", here.numPUs(), " processor cores");
    writeln("  ...has ", here.physicalMemory(unit=MemUnits.GB, retType=real), " GB of memory");
    writeln("  ...has ", here.maxTaskPar, " maximum parallelism");
  }

Load Chapel and start an interactive job requesting four nodes and three cores on each node:

module unload chapel-single . /home/razoumov/startMultiLocale.sh salloc --time=0:30:0 --nodes=4 --cpus-per-task=3 --mem-per-cpu=3500 --account=def-someprof

Once the interactive job starts, you can compile and run your code from the prompt on the first allocated compute node:

chpl probeLocales.chpl -o probeLocales ./probeLocales -nl 4 For production jobs, please write a Slurm submission script and submit your job with sbatch instead.

Message Passing Interface