CUDA hello world in C

< Examples

Load modules

[jarunanp@eu-login-10 ~]$ env2lmod
[jarunanp@eu-login-10 ~]$ module load gcc/6.3.0 cuda/11.0.3

The following have been reloaded with a version change:
  1) gcc/4.8.5 => gcc/6.3.0

[jarunanp@eu-login-10 ~]$ which nvcc
/cluster/apps/gcc-6.3.0/cuda-11.0.3-qdlibd2luz2fy7izfefao4c5yitxwjus/bin/nvcc

CUDA Hello World

• Go to $SCRATCH and create a work directory

[jarunanp@eu-login-10 ~]$ cd $SCRATCH
[jarunanp@eu-login-10 jarunanp]$ pwd
/cluster/scratch/jarunanp
[jarunanp@eu-login-10 jarunanp]$ mkdir test_cuda
[jarunanp@eu-login-10 jarunanp]$ cd test_cuda
[jarunanp@eu-login-10 test_cuda]$ 

• Download a CUDA Hello World example

[jarunanp@eu-login-10 test_cuda]$ wget -c https://gitlab.ethz.ch/jarunanp/hpc-examples/-/raw/main/cuda/cuda_hello.cu?inline=false -O cuda_hello.c

• Compile the code

[jarunanp@eu-login-10 test_cuda]$ nvcc cuda_hello.c -o cuda_hello 

• Testing the executable

[jarunanp@eu-login-10 test_cuda]$ bsub -R "rusage[ngpus_excl_p=1]" -I "./cuda_hello"
Generic job.
Job <195522896> is submitted to queue <gpu.4h>.
<<Waiting for dispatch ...>>
<<Starting on eu-g3-045>>
Hello World from GPU!
[jarunanp@eu-login-10 test_cuda]$

Using CUDA built-in variables

We have provided codes here which use the CUDA built-in variables threadIdx.x and blockIdx.x. These examples were taken from this CUDA tutorial.

• Compile the code

[jarunanp@eu-login-10 test_cuda]$ module load gcc/6.3.0 cuda/11.0.3
[jarunanp@eu-login-10 test_cuda]$ nvcc vector_add.cu -o vector_add_cu

• Request an interactive session on a compute node

[jarunanp@eu-login-10 test_cuda]$ bsub -R "rusage[ngpus_excl_p=1]" -Is bash
Generic job.
Job <195523378> is submitted to queue <gpu.4h>.
<<Waiting for dispatch ...>>
<<Starting on eu-g3-039>>
FILE: /sys/fs/cgroup/cpuset/lsf/euler/job.195523378.50598.1638799736/tasks
[jarunanp@eu-g3-039 test_cuda]$

• Profile the CUDA executable

[jarunanp@eu-g3-039 test_cuda]$ nvprof ./vector_add_cu
==112917== NVPROF is profiling process 112917, command: ./vector_add_cu
out[0] = 3.000000
PASSED
==112917== Profiling application: ./vector_add_cu
==112917== Profiling result:
            Type  Time(%)      Time     Calls       Avg       Min       Max  Name
 GPU activities:   92.57%  524.00ms         1  524.00ms  524.00ms  524.00ms  vector_add(float*, float*, float*, int)
                    4.63%  26.209ms         1  26.209ms  26.209ms  26.209ms  [CUDA memcpy DtoH]
                    2.80%  15.860ms         2  7.9298ms  7.9215ms  7.9381ms  [CUDA memcpy HtoD]

< Examples