ATPESC18

GPU programming module for Argonne Training Program in Extreme Scale Computing 2018

Instructions for Accelerator Presentation

Slides: www.math.vt.edu/people/tcew/ATPESC18

OCCA repo (version 0.2 branch): git clone https://github.com/libocca/occa -b 0.2

Paranumal website: http://www.paranumal.com

Paranumal blog including tips on GPU performance: https://www.paranumal.com/blog

Ray Loy's on-boarding ALCF slides: https://extremecomputingtraining.anl.gov/sessions/presentation-quick-start-on-atpesc-resources/

Examples:

To access the demos from the GPU programming lecture:

git clone https://github.com/tcew/ATPESC18

Note: You may need to use ssh-keygen on cooley and update the github ssh public keys associated with the ssh public key from your cooley account)

Flow animation hands-on:

Follow these instructions to get access to a cooley GPU node and configure for CUDA

1. ssh to cooley

ssh username@cooley.alcf.anl.gov

2. Make sure these options or similar appear in your .soft.cooley file:

+mvapich2
+cuda-7.5.18
+ffmpeg-1.0.1
@default

3. make sure packages are loaded

resoft

CUDA version of flow simulation

A. get files for GPU tutorial

git clone https://github.com/tcew/ATPESC18

B. build Lattice Boltzmann flow solver

cd ATPESC18/handsOn
cd lbm
make

C. submit job request

qsub -A ATPESC2018 -I -n 1 -t 120 -q training

D. change path/name of your png file to your image

./cudaLBM ./images/fsm.png 400

E. create movie

ffmpeg -start_number 0 -r 24 -i bah%06d.png -c:v mpeg4 test.mp4

OCCA version of flow simulation

A. retrieve OCCA

git clone https://github.com/libocca/occa -b 0.2

B. retrieve ATPESC18

git clone https://github.com/tcew/ATPESC18

C. submit job request

qsub -A ATPESC2018 -I -n 1 -t 120 -q training

D. build OCCA

cd occa
export OCCA_DIR=`pwd` export LD_LIBRARY_PATH=$OCCA_DIR/lib:$LD_LIBRARY_PATH:$OCCA_DIR/lib
make -j
./bin/occainfo

E. build Lattice Boltzmann flow solver

cd ../ATPESC18/handsOn
cd lbm
make -f makefile.occa

F. run the lbm code with an example png image (using 400 as a flow volume threshold)

./occaLBM ./images/fsm.png 400

G. on the compute node convert the output bah png files to foo.mp4

ffmpeg -r 24 -i bah%06d.png -b:v 16384k -vf scale=1024:-1 foo.mp4
ffmpeg -r 24 -i bah%06d.png -b:v 16384k -vf scale=768:-1 -vcodec mpeg4 foo.mp4