CUDA tutorial: Difference between revisions

Jump to navigation Jump to search

CUDA tutorial (view source)

Revision as of 16:28, 7 March 2018

2,552 bytes removed ,  6 years ago
removed quickstart material to new CUDA page
m (Rdickson moved page CUDA to CUDA tutorial without leaving a redirect: Part of translatable page "CUDA")
(removed quickstart material to new CUDA page)
Line 2: Line 2:
<translate>
<translate>
[[Category:Software]]
[[Category:Software]]
=Quick start guide= <!--T:32-->
To begin working with CUDA load a CUDA module.
<source lang="console">
$ module purge
$ module load cuda
</source>
As a first step we will add two numbers together on a GPU. Save the below file as <code>add.cu</code>. '''The <code>cu</code> file extension is important!'''.
{{File 
  |name=add.cu
  |lang="c++"
  |contents=
#include <iostream>
<!--T:33-->
__global__ void add (int *a, int *b, int *c){
  *c = *a + *b;
}
<!--T:34-->
int main(void){
  int a, b, c;
  int *dev_a, *dev_b, *dev_c;
  int size = sizeof(int);
 
  //  allocate device copies of a,b, c
  cudaMalloc ( (void**) &dev_a, size);
  cudaMalloc ( (void**) &dev_b, size);
  cudaMalloc ( (void**) &dev_c, size);
 
  a=2; b=7;
  //  copy inputs to device
  cudaMemcpy (dev_a, &a, size, cudaMemcpyHostToDevice);
  cudaMemcpy (dev_b, &b, size, cudaMemcpyHostToDevice);
 
  // launch add() kernel on GPU, passing parameters
  add <<< 1, 1 >>> (dev_a, dev_b, dev_c);
 
  // copy device result back to host
  cudaMemcpy (&c, dev_c, size, cudaMemcpyDeviceToHost);
  std::cout<<a<<"+"<<b<<"="<<c<<std::endl;
 
  cudaFree ( dev_a ); cudaFree ( dev_b ); cudaFree ( dev_c );
}
}}
To build the program use the command below, which will create an executable named <code>add</code>.
<source lang="console">
$ nvcc add.cu -o add
</source>
To run the program first create a Slurm job script called gpu_job.sh. Be sure to replace <code>def-someuser</code> with your specific account (see [[Running_jobs#Accounts_and_projects|accounts and projects]]). For various ways to schedule jobs with GPUs see [[Using GPUs with Slurm|using GPUs with Slurm]].
{{File
  |name=gpu_job.sh
  |lang="sh"
  |contents=
#!/bin/bash
#SBATCH --account=def-someuser
#SBATCH --gres=gpu:1              # Number of GPUs (per node)
#SBATCH --mem=400M                # memory (per node)
#SBATCH --time=0-00:10            # time (DD-HH:MM)
./add #name of your program
}}
<!--T:35-->
Submit your GPU job to the scheduler with this command.
<source lang="console">
$ sbatch gpu_job.sh
Submitted batch job 3127733
</source>For information about the <code>sbatch</code> command and running and monitoring jobs see the [[Running jobs|running jobs]] page.
<!--T:36-->
Once your job has finished you should see an output file similar to this.
<source lang="console">
$ cat slurm-3127733.out
2+7=9
</source>
If you run this without a GPU present you might see output like <code>2+7=0</code>. To learn more about how the above program works and how to make the use of a GPUs parallelism keep reading.


=Introduction= <!--T:1-->
=Introduction= <!--T:1-->
This tutorial introduces the graphics processing unit (GPU) as a massively parallel computing device; the CUDA parallel programming language; and some of the CUDA numerical libraries for high performance computing.
This tutorial introduces the graphics processing unit (GPU) as a massively parallel computing device; the [[CUDA]] parallel programming language; and some of the CUDA numerical libraries for high performance computing.
{{Prerequisites
{{Prerequisites
|title=Prerequisites
|title=Prerequisites
Rdickson
Bureaucrats, cc_docs_admin, cc_staff
2,879

edits

Retrieved from "https://docs.alliancecan.ca/wiki/Special:MobileDiff/47995"

Navigation menu