Skip to content

Commit 2cc5ce9

Browse files
huebner-mhuebner-m
committed
Move (conditional) installation of cuda compat libs to external script
Only install cuda compat libs when either they are not installed yet or they are outdated
1 parent 7d6af69 commit 2cc5ce9

File tree

2 files changed

+87
-60
lines changed

2 files changed

+87
-60
lines changed

‎gpu_support/add_gpu_support.sh‎

Lines changed: 14 additions & 60 deletions
Original file line numberDiff line numberDiff line change
@@ -90,76 +90,30 @@ if [ $ret -ne 0 ]; then
9090
echo $latest_cuda_compat_url
9191
exit 1
9292
fi
93+
latest_driver_version="${latest_cuda_compat_url%-*}"
94+
latest_driver_version="${latest_driver_version##*_}"
9395

94-
# Create a general space for our NVIDIA compat drivers
96+
install_compat_libs=false
9597
host_injections_dir="/cvmfs/pilot.eessi-hpc.org/host_injections/nvidia"
96-
if [ -w /cvmfs/pilot.eessi-hpc.org/host_injections ]; then
97-
mkdir -p ${host_injections_dir}
98-
cd ${host_injections_dir}
98+
# libcuda.so points to actual cuda compat lib with driver version in its name
99+
# if this file exists, cuda compat libs are installed and we can compare the version
100+
if [ -e $host_injections_dir/latest/compat/libcuda.so ]; then
101+
eessi_driver_version=$( realpath $host_injections_dir/latest/compat/libcuda.so)
102+
eessi_driver_version="${eessi_driver_version##*so.}"
99103
else
100-
echo "Cannot write to eessi host_injections space, exiting now..." >&2
101-
exit 1
104+
eessi_driver_version=0
102105
fi
103106

104-
# Check if we have any version installed by checking for the existence of /cvmfs/pilot.eessi-hpc.org/host_injections/nvidia/latest
105-
106-
driver_cuda_version=$(nvidia-smi -q --display=COMPUTE | grep CUDA | awk 'NF>1{print $NF}' | sed s/\\.//)
107-
eessi_cuda_version=$(LD_LIBRARY_PATH=${host_injections_dir}/latest/compat/:$LD_LIBRARY_PATH nvidia-smi -q --display=COMPUTE | grep CUDA | awk 'NF>1{print $NF}' | sed s/\\.//)
108-
if [ "$driver_cuda_version" -gt "$eessi_cuda_version" ]; then echo "You need to update your CUDA compatability libraries"; fi
109-
110-
# Check if our target CUDA is satisfied by what is installed already
111-
# TODO: Find required CUDA version and see if we need an update
112-
113-
# If not, grab the latest compat library RPM or deb
114-
# download and unpack in temporary directory, easier cleanup after installation
115-
tmpdir=$(mktemp -d)
116-
cd $tmpdir
117-
compat_file=${latest_cuda_compat_url##*/}
118-
wget ${latest_cuda_compat_url}
119-
120-
# Unpack it
121-
# (the requirements here are OS dependent, can we get around that?)
122-
# (for rpms looks like we can use https://gitweb.gentoo.org/repo/proj/prefix.git/tree/eclass/rpm.eclass?id=d7fc8cf65c536224bace1d22c0cd85a526490a1e)
123-
# (deb files can be unpacked with ar and tar)
124-
file_extension=${compat_file##*.}
125-
if [[ ${file_extension} == "rpm" ]]; then
126-
rpm2cpio ${compat_file} | cpio -idmv
127-
elif [[ ${file_extension} == "deb" ]]; then
128-
ar x ${compat_file}
129-
tar xf data.tar.*
107+
if [ ${latest_driver_version//./} -gt ${eessi_driver_version//./} ]; then
108+
install_compat_libs=true
130109
else
131-
echo "File extension of cuda compat lib not supported, exiting now..." >&2
132-
exit 1
110+
echo "CUDA compat libs are up-to-date, skip installation."
133111
fi
134-
cd $host_injections_dir
135-
# TODO: This would prevent error messages if folder already exists, but could be problematic if only some files are missing in destination dir
136-
mv -n ${tmpdir}/usr/local/cuda-* .
137-
rm -r ${tmpdir}
138-
139-
# Add a symlink that points to the latest version
140-
latest_cuda_dir=$(find . -maxdepth 1 -type d | grep -i cuda | sort | tail -n1)
141-
ln -sf ${latest_cuda_dir} latest
142112

143-
if [ ! -e latest ] ; then
144-
echo "Symlink to latest cuda compat lib version is broken, exiting now..."
145-
exit 1
146-
fi
147-
148-
# Create the space to host the libraries
149-
host_injection_libs_dir=/cvmfs/pilot.eessi-hpc.org/host_injections/${EESSI_PILOT_VERSION}/compat/${os_family}/${eessi_cpu_family}
150-
mkdir -p ${host_injection_libs_dir}
151-
# Symlink in the path to the latest libraries
152-
if [ ! -d "${host_injection_libs_dir}/lib" ]; then
153-
ln -s ${host_injections_dir}/latest/compat ${host_injection_libs_dir}/lib
154-
elif [ ! "${host_injection_libs_dir}/lib" -ef "${host_injections_dir}/latest/compat" ]; then
155-
echo "CUDA compat libs symlink exists but points to the wrong location, please fix this..."
156-
echo "${host_injection_libs_dir}/lib should point to ${host_injections_dir}/latest/compat"
157-
exit 1
113+
if [ "${install_compat_libs}" == true ]; then
114+
source $(dirname "$BASH_SOURCE")/install_cuda_compatlibs.sh $latest_cuda_compat_url
158115
fi
159116

160-
# return to initial dir
161-
cd $current_dir
162-
163117
###############################################################################################
164118
###############################################################################################
165119
# Install CUDA

‎gpu_support/install_cuda_compatlibs.sh‎

Lines changed: 73 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,73 @@
1+
#!/bin/bash
2+
3+
libs_url=$1
4+
5+
current_dir=$(dirname $(realpath $0))
6+
7+
# Create a general space for our NVIDIA compat drivers
8+
if [ -w /cvmfs/pilot.eessi-hpc.org/host_injections ]; then
9+
mkdir -p ${host_injections_dir}
10+
cd ${host_injections_dir}
11+
else
12+
echo "Cannot write to eessi host_injections space, exiting now..." >&2
13+
exit 1
14+
fi
15+
16+
# Check if we have any version installed by checking for the existence of /cvmfs/pilot.eessi-hpc.org/host_injections/nvidia/latest
17+
18+
driver_cuda_version=$(nvidia-smi -q --display=COMPUTE | grep CUDA | awk 'NF>1{print $NF}' | sed s/\\.//)
19+
eessi_cuda_version=$(LD_LIBRARY_PATH=${host_injections_dir}/latest/compat/:$LD_LIBRARY_PATH nvidia-smi -q --display=COMPUTE | grep CUDA | awk 'NF>1{print $NF}' | sed s/\\.//)
20+
if [ "$driver_cuda_version" -gt "$eessi_cuda_version" ]; then echo "You need to update your CUDA compatability libraries"; fi
21+
22+
# Check if our target CUDA is satisfied by what is installed already
23+
# TODO: Find required CUDA version and see if we need an update
24+
25+
# If not, grab the latest compat library RPM or deb
26+
# download and unpack in temporary directory, easier cleanup after installation
27+
tmpdir=$(mktemp -d)
28+
cd $tmpdir
29+
compat_file=${libs_url##*/}
30+
wget ${libs_url}
31+
32+
# Unpack it
33+
# (the requirements here are OS dependent, can we get around that?)
34+
# (for rpms looks like we can use https://gitweb.gentoo.org/repo/proj/prefix.git/tree/eclass/rpm.eclass?id=d7fc8cf65c536224bace1d22c0cd85a526490a1e)
35+
# (deb files can be unpacked with ar and tar)
36+
file_extension=${compat_file##*.}
37+
if [[ ${file_extension} == "rpm" ]]; then
38+
rpm2cpio ${compat_file} | cpio -idmv
39+
elif [[ ${file_extension} == "deb" ]]; then
40+
ar x ${compat_file}
41+
tar xf data.tar.*
42+
else
43+
echo "File extension of cuda compat lib not supported, exiting now..." >&2
44+
exit 1
45+
fi
46+
cd $host_injections_dir
47+
# TODO: This would prevent error messages if folder already exists, but could be problematic if only some files are missing in destination dir
48+
mv -n ${tmpdir}/usr/local/cuda-* .
49+
rm -r ${tmpdir}
50+
51+
# Add a symlink that points to the latest version
52+
latest_cuda_dir=$(find . -maxdepth 1 -type d | grep -i cuda | sort | tail -n1)
53+
ln -sf ${latest_cuda_dir} latest
54+
55+
if [ ! -e latest ] ; then
56+
echo "Symlink to latest cuda compat lib version is broken, exiting now..."
57+
exit 1
58+
fi
59+
60+
# Create the space to host the libraries
61+
host_injection_libs_dir=/cvmfs/pilot.eessi-hpc.org/host_injections/${EESSI_PILOT_VERSION}/compat/${os_family}/${eessi_cpu_family}
62+
mkdir -p ${host_injection_libs_dir}
63+
# Symlink in the path to the latest libraries
64+
if [ ! -d "${host_injection_libs_dir}/lib" ]; then
65+
ln -s ${host_injections_dir}/latest/compat ${host_injection_libs_dir}/lib
66+
elif [ ! "${host_injection_libs_dir}/lib" -ef "${host_injections_dir}/latest/compat" ]; then
67+
echo "CUDA compat libs symlink exists but points to the wrong location, please fix this..."
68+
echo "${host_injection_libs_dir}/lib should point to ${host_injections_dir}/latest/compat"
69+
exit 1
70+
fi
71+
72+
# return to initial dir
73+
cd $current_dir

0 commit comments

Comments
 (0)