-
Notifications
You must be signed in to change notification settings - Fork 0
Building and Running Devil Ray on LLNL HPC Systems
Cyrus Harrison edited this page Nov 8, 2019
·
8 revisions
Run these on Pascal with OpenMP:
git clone --recursive ssh://[email protected]:7999/vis/devil_ray.git
# if this fails, you may need to setup ssh keys -- ask us for help on process
cd devil_ray
# make a directory to build in
mkdir build && cd build
# get newer cmake
module load cmake
# ask for a batch node
salloc -N 1
# (wait for job to be allocated)
# configure our build with pre-built dependencies openmp variant
cmake -C /usr/workspace/wsb/hmc_19/tpls/pascal_openmp/[email protected] ../src
# build
make
# run tests (some may fail!)
make test
To enable logging and stats use the following during the cmake configure:
cmake -DENABLE_LOGGING=ON -DENABLE_STATS=ON -C /usr/workspace/wsb/hmc_19/tpls/pascal_openmp/[email protected] ../src
Interactive allocation using vis partition on Pascal:
salloc -N 1 -t 200 -ppvis
Run these on Pascal with CUDA:
cd devil_ray
# make a directory to build in
mkdir build-cuda && cd build-cuda
# get newer cmake
module load cmake
# get cuda tools in env
module load cuda
# ask for a batch node
salloc -N 1
# (wait for job to be allocated)
# configure our build with pre-built dependencies cuda variant
cmake -C /usr/workspace/wsb/hmc_19/tpls/pascal_cuda/[email protected] ../src
# build
make
# run tests (some may fail!)
make test