Difference between revisions of "OpenACC Workshop"
From HPC users
Jump to navigationJump to search
Line 33: | Line 33: | ||
module load CUDA-Toolkit | module load CUDA-Toolkit | ||
./sincos $@ | ./sincos $@ | ||
Start the visual profiler in the background | |||
nvvp & | |||
Find out hostname for allocation | |||
srun hostname |
Latest revision as of 15:31, 29 March 2017
Slides
Agenda
Welcome and Introduction
Day 1 Morning Lecture: Introduction OpenACC I
Day 1 Afternoon Lecture: Introduction OpenACC II
Day 2 Morning Lecture: Advanced OpenACC
Day 2 Afternoon Lecture: GPU-enabled Numerical Library
Quick Guide OpenACC
Modules to load:
module load PGI CUDA-Toolkit
Command to compile:
pgcc -acc -ta=tesla:cc60 -o executable code.c
Command to run:
srun -p mpcg.p --gres=gpu:1 ./executable
Alternatively use partition cfdg.p.
More info in the OpenACC Introduction.
Workshop
To copy data from the course directory:
cp -r /user/gilu2568/<dir> .
For profiling first get allocation:
salloc -p mpcg.p --gres=gpu:1
Create a script that runs an application, e.g. sincos.sh in the sincos example
#!/bin/bash module load hpc-uniol-env module load PGI module load CUDA-Toolkit ./sincos $@
Start the visual profiler in the background
nvvp &
Find out hostname for allocation
srun hostname