diff --git a/n/NCCL/NCCL-2.18.3-CUDA-12.2.0.eb b/n/NCCL/NCCL-2.18.3-CUDA-12.2.0.eb index 7472b3fd..d93f250c 100644 --- a/n/NCCL/NCCL-2.18.3-CUDA-12.2.0.eb +++ b/n/NCCL/NCCL-2.18.3-CUDA-12.2.0.eb @@ -21,7 +21,10 @@ dependencies = [ # ('UCX-CUDA', '1.14.1', versionsuffix, ('GCCcore', '12.2.0')), ] -# default CUDA compute capabilities to use (override via --cuda-compute-capabilities) -cuda_compute_capabilities = ['8.0'] +import os +if os.environ.get("CLUSTERNAME") in ["KAROLINA"]: + cuda_compute_capabilities = ['8.0'] # A100 +else: + cuda_compute_capabilities = ['7.0'] # V100 moduleclass = 'lib' diff --git a/n/NCCL/NCCL-2.18.3-NVHPC-23.7-CUDA-12.2.0.eb b/n/NCCL/NCCL-2.18.3-NVHPC-23.7-CUDA-12.2.0.eb new file mode 100644 index 00000000..063f52a5 --- /dev/null +++ b/n/NCCL/NCCL-2.18.3-NVHPC-23.7-CUDA-12.2.0.eb @@ -0,0 +1,30 @@ +# IT4Innovations +# LK 2023 + +name = 'NCCL' +version = '2.18.3' +versionsuffix = '-CUDA-%(cudaver)s' + +homepage = 'https://developer.nvidia.com/nccl' +description = """The NVIDIA Collective Communications Library (NCCL) implements multi-GPU and multi-node collective +communication primitives that are performance optimized for NVIDIA GPUs.""" + +toolchain = {'name': 'NVHPC', 'version': '23.7'} + +github_account = 'NVIDIA' +source_urls = [GITHUB_SOURCE] +sources = ['v%(version)s-1.tar.gz'] +checksums = ['b4f5d7d9eea2c12e32e7a06fe138b2cfc75969c6d5c473aa6f819a792db2fc96'] + +dependencies = [ + ('CUDA', '12.2.0', '', SYSTEM), +# ('UCX-CUDA', '1.14.1', versionsuffix, ('GCCcore', '12.2.0')), +] + +import os +if os.environ.get("CLUSTERNAME") in ["KAROLINA"]: + cuda_compute_capabilities = ['8.0'] # A100 +else: + cuda_compute_capabilities = ['7.0'] # V100 + +moduleclass = 'lib' diff --git a/n/NVHPC/NVHPC-23.7.eb b/n/NVHPC/NVHPC-23.7.eb new file mode 100644 index 00000000..7b617a92 --- /dev/null +++ b/n/NVHPC/NVHPC-23.7.eb @@ -0,0 +1,31 @@ +# IT4Innovations +# LK 2023 + +name = 'NVHPC' +version = '23.7' + +homepage = 'https://developer.nvidia.com/hpc-sdk/' +description = """C, C++ and Fortran compilers included with the NVIDIA HPC SDK (previously: PGI)""" + +toolchain = SYSTEM + +# By downloading, you accept the HPC SDK Software License Agreement (https://docs.nvidia.com/hpc-sdk/eula/index.html) +accept_eula = True +source_urls = ['https://developer.download.nvidia.com/hpc-sdk/%(version)s/'] +local_tarball_tmpl = 'nvhpc_2023_%%(version_major)s%%(version_minor)s_Linux_%s_cuda_multi.tar.gz' +sources = [local_tarball_tmpl % '%(arch)s'] +checksums = ['fea91d95ff18bca1ce7afde50371caa02001ade8bed6ddfc5ff70862ccbebece'] + +dependencies = [ + ('numactl', '2.0.16') +] + +default_cuda_version = '12.2' + +import os +if os.environ.get("CLUSTERNAME") in ["KAROLINA"]: + cuda_compute_capabilities = "8.0" +else: + cuda_compute_capabilities = "7.0" + +moduleclass = 'compiler' diff --git a/n/numactl/numactl-2.0.16.eb b/n/numactl/numactl-2.0.16.eb new file mode 100644 index 00000000..d4ceb8dc --- /dev/null +++ b/n/numactl/numactl-2.0.16.eb @@ -0,0 +1,36 @@ +# IT4Innovations +# LK 2023 + +easyblock = 'ConfigureMake' + +name = 'numactl' +version = '2.0.16' + +homepage = 'https://github.com/numactl/numactl' + +description = """ + The numactl program allows you to run your application program on specific + cpu's and memory nodes. It does this by supplying a NUMA memory policy to + the operating system before running your program. The libnuma library provides + convenient ways for you to add NUMA memory policies into your own program. +""" + +toolchain = SYSTEM +toolchainopts = {'pic': True} + +source_urls = ['https://github.com/numactl/numactl/archive/'] +sources = ['v%(version)s.tar.gz'] +checksums = ['a35c3bdb3efab5c65927e0de5703227760b1101f5e27ab741d8f32b3d5f0a44c'] + +builddependencies = [ + ('Autotools', '20220317'), +] + +preconfigopts = "./autogen.sh && " + +sanity_check_paths = { + 'files': ['bin/numactl', 'bin/numastat', 'lib/libnuma.%s' % SHLIB_EXT, 'lib/libnuma.a'], + 'dirs': ['share/man', 'include'] +} + +moduleclass = 'tools' diff --git a/o/OpenMPI/OpenMPI-4.1.5-NVHPC-23.5-CUDA-12.2.0-test.eb b/o/OpenMPI/OpenMPI-4.1.5-NVHPC-23.5-CUDA-12.2.0-test.eb deleted file mode 100644 index e08b8f47..00000000 --- a/o/OpenMPI/OpenMPI-4.1.5-NVHPC-23.5-CUDA-12.2.0-test.eb +++ /dev/null @@ -1,117 +0,0 @@ -# IT4Innovations -# LK 2023 - -name = 'OpenMPI' -version = '4.1.5' -versionsuffix= '-test' - -homepage = 'https://www.open-mpi.org/' -description = """The Open MPI Project is an open source MPI-3 implementation.""" - -toolchain = {'name': 'NVHPC', 'version': '23.5'} - -source_urls = ['https://www.open-mpi.org/software/ompi/v%(version_major_minor)s/downloads'] -sources = [SOURCELOWER_TAR_BZ2] -patches = [ - 'OpenMPI-4.1.1_build-with-internal-cuda-header.patch', - 'OpenMPI-4.1.1_opal-datatype-cuda-performance.patch', -] -checksums = [ - 'a640986bc257389dd379886fdae6264c8cfa56bc98b71ce3ae3dfbd8ce61dbe3', # openmpi-4.1.5.tar.bz2 - # OpenMPI-4.1.1_build-with-internal-cuda-header.patch - '63eac52736bdf7644c480362440a7f1f0ae7c7cae47b7565f5635c41793f8c83', - # OpenMPI-4.1.1_opal-datatype-cuda-performance.patch - 'b767c7166cf0b32906132d58de5439c735193c9fd09ec3c5c11db8d5fa68750e', -] - -builddependencies = [ - ('pkgconf', '1.9.3'), - ('Perl', '5.36.0'), - ('Autotools', '20220317'), -] - -dependencies = [ - ('zlib', '1.2.12'), - ('CUDA', '12.2.0', '', True), - ('hwloc', '2.8.0'), - ('libevent', '2.1.12'), - ('UCX', '1.14.1'), - ('UCX-CUDA', '1.14.1', '-CUDA-%(cudaver)s'), - ('libfabric', '1.16.1'), - ('PMIx', '4.2.2'), - ('UCC', '1.1.0'), - ('UCC-CUDA', '1.1.0', '-CUDA-%(cudaver)s'), -] - -# Update configure to include changes from the "internal-cuda" patch -# by running a subset of autogen.pl sufficient to achieve this -# without doing the full, long-running regeneration. -preconfigopts = ' && '.join([ - 'cd config', - 'autom4te --language=m4sh opal_get_version.m4sh -o opal_get_version.sh', - 'cd ..', - 'autoconf', - 'autoheader', - 'aclocal', - 'automake', - '' -]) - -# CUDA related patches and custom configure option can be removed if CUDA support isn't wanted. -configopts = ' --with-cuda=internal' -configopts += ' CC=pgcc CXX=pgc++ FC=pgfortran' -configopts += ' CXXFLAGS="-fPIC"' - -# IT4I-specific settings - -#configopts += '--enable-shared ' -configopts += ' --enable-mpi-thread-multiple' -configopts += ' --with-verbs' -configopts += ' --enable-mpirun-prefix-by-default' -configopts += ' --with-hwloc=$EBROOTHWLOC' # hwloc support -#configopts += ' --with-tm=/opt/pbs ' # Enable PBS -configopts += ' --with-slurm ' # Enable slurm -configopts += ' --enable-mpi-cxx' # Enable building the C++ MPI bindings -configopts += ' --with-ucx=$EBROOTUCX' -configopts += ' --with-knem=/opt/knem-1.1.4.90mlnx1' -#configopts += ' --with-pmix=/opt/it4i-libs/PMIx/4.2.2:/opt/it4i-libs/PMIx/3.2.3:/opt/slurm/lib64/' - -osdependencies = [('libibverbs-dev', 'libibverbs-devel', 'rdma-core-devel')] - -postinstallcmds = [ - 'echo "# By default, for Open MPI 4.0 and later, infiniband ports on a device are not used by default." >> %(installdir)s/etc/openmpi-mca-params.conf', - 'echo "btl_openib_allow_ib = true" >> %(installdir)s/etc/openmpi-mca-params.conf', -] - -local_libs = ["mpi_mpifh", "mpi", "ompitrace", "open-pal", "open-rte"] -sanity_check_paths = { - 'files': [ - "bin/%s" % - binfile for binfile in [ - "ompi_info", "opal_wrapper", "orterun"]] + [ - "lib/lib%s.%s" % - (libfile, SHLIB_EXT) for libfile in local_libs] + [ - "include/%s.h" % - x for x in [ - "mpi-ext", "mpif-config", "mpif", "mpi", "mpi_portable_platform"]], 'dirs': [], } - -import os -if os.environ.get("CLUSTERNAME") in ["BARBORA"]: - modextravars = {'OMPI_MCA_btl_openib_if_include': 'mlx5_0', - 'OMPI_MCA_btl_tcp_if_include': '10.33.4.0/24', - 'OMPI_MCA_orte_base_help_aggregate': '0', - 'SLURM_MPI_TYPE': 'pmix_v4', - } -elif os.environ.get("CLUSTERNAME") in ["KAROLINA"]: - modextravars = {'OMPI_MCA_btl_openib_if_include': 'mlx5_0', - 'OMPI_MCA_orte_base_help_aggregate': '0', - 'SLURM_MPI_TYPE': 'pmix_v4', - } -else: - modextravars = {'OMPI_MCA_btl_openib_if_include': 'mlx4_0', - 'OMPI_MCA_oob_tcp_if_include': '10.0.0.0/8', - 'SLURM_MPI_TYPE': 'pmix_v4', - } - - -moduleclass = 'mpi'