Skip to content

Latest commit

 

History

History

c

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
  1. HPC User Wiki
  2. NREL HPC User Community Wiki
  3. Tips and Tricks

HPC User Wiki : C Programs

Created by Southerland, Jennifer on 2018-01-23

Compiling and Running C Programs on the Eagle System

Simple "Hello, world" Program

Follow these instructions to compile and run your first programs written in C on Eagle.

Copy this text to a file named hello_world.c with your favorite text editor (nano, vi, emacs, etc.).

/* C Example */
#include <stdio.h>

int main (argc, argv)
   int argc;
   char *argv[];
{
   printf( "Hello, world \n" );
   return 0;
}

Compile this program using the Intel C compiler, icc:

$ icc -o hello_world hello_world.c

The compiler will create a Linux executable file called hello_world.

To run this program, use your editor to create a job script containing the following text:

#!/bin/bash -l
#PBS -j oe
#PBS -N job_hello_world
#PBS -l walltime=01:00:00
#PBS -l nodes=1

# this ensures your job runs from the directory from which you run the qsub command
cd $PBS_O_WORKDIR

./hello_world

Give the file a name like hello_world.sh.

Submit the script using the qsub command and a valid project handle. Because this is a very quick job, let's submit it to the "short" queue.

$ qsub -q short -A CSC000 hello_world.sh

The output from this job is in a file with the name job_hello_world.o<JID> where JID is the job id. You can see the contents of that file using the "cat" command:

[icarpent@login4 ~]$ cat job_hello_world.o1112070
Warning: no access to tty (Bad file descriptor).
Thus no job control in this shell.
Hello, world

Parallel "Hello, world" Program

For this example, we'll use MPI to write a parallel Hello World program written in C. The Message Passing Interface (MPI)( is one way to create programs that run on more than one processor.

Copy this text to a file named hello_world.c with your favorite text editor (vi, emacs, etc.)  If you are not familiar with text editors, try "nano hello_world.c"

/* C Example */
#include <stdio.h>
#include <mpi.h> 


int main (argc, argv)
   int argc;
   char *argv[];
{
   int rank, size;
   MPI_Init (&argc, &argv); /* starts MPI */
   MPI_Comm_rank (MPI_COMM_WORLD, &rank);     /* get current process id */
   MPI_Comm_size (MPI_COMM_WORLD, &size);     /* get number of processes */

    printf( "Hello, world from process %d of %d\n", rank, size );

    MPI_Finalize();
   return 0;
}

Compile this program using mpicc, which is a C compiler that knows about MPI.

mpicc hello_world.c -o hello_world

The compiler will create a linux executable file called hello_world.

Running your Parallel “Hello, world” program

Eagle is a cluster: a collection of computers connected together with a special network (InfiniBand in this case) and software (Torque) that allows a single program to run across multiple physical computers. Create a submit script which contains options for torque, including the number of nodes you want your parallel program to run on and instructions for how to start the parallel program. For example, create a file named hello.qsub with the contents below.

#!/bin/bash -l
#PBS -j oe
#PBS -N job_hello_world
#PBS -l walltime=01:00:00
#PBS -l nodes=2                 # this asks for 2 nodes 

# this ensures your job runs from the directory from which you run the qsub command
cd $PBS_O_WORKDIR
set -x

mpirun -np 32 ./hello_world >& my_results.out

Because this is an MPI program, we execute it using the mpirun command and we tell it how many processors to run it on with the -np option. In this case we have told the system to run the hello_world program on 32 processors.  The nodes assigned to your job will have either 16 or 24 processors so to use 32 we have asked for 2 nodes. The system will use some processors on the first node and some on the second node. Each processor will run the program and print a line of output.

Once you've saved this file, next create your run directory in /scratch using the mkdir command and copy your program and job script to it.

$ mkdir /scratch/$USER/hello
$ cp hello_world /scratch/$USER/hello
$ cp hello.qsub /scratch/$USER/hello

Go to this directory using the cd command:

$ cd /scratch/$USER/hello

Now, submit your job script using the qsub command with a valid project handle:

$ qsub hello.qsub -A project-handle

The job will run, then you should see an output file named "my_results.out" with one line of output from each MPI process.

[kregimba@login1 hello]$ cat my_results.out
Hello world from process 14 of 32
Hello world from process 30 of 32
...
Hello world from process 26 of 32
[kregimba@login1 hello]$

Document generated by Confluence on 2019-04-03 10:15

Atlassian