easybuilders · SebastianAchilles · Nov 24, 2024 · Oct 4, 2024 · Oct 4, 2024 · Oct 4, 2024
diff --git a/easybuild/easyconfigs/n/NCCL/NCCL-2.22.3-GCCcore-13.3.0-CUDA-12.6.0.eb b/easybuild/easyconfigs/n/NCCL/NCCL-2.22.3-GCCcore-13.3.0-CUDA-12.6.0.eb
@@ -0,0 +1,26 @@
+name = 'NCCL'
+version = '2.22.3'
+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': 'GCCcore', 'version': '13.3.0'}
+
+github_account = 'NVIDIA'
+source_urls = [GITHUB_SOURCE]
+sources = ['v%(version)s-1.tar.gz']
+checksums = ['45151629a9494460e73375281e8b0fe379141528879301899ece9b776faca024']
+
+builddependencies = [('binutils', '2.42')]
+
+dependencies = [
+    ('CUDA', '12.6.0', '', SYSTEM),
+    ('UCX-CUDA', '1.16.0', versionsuffix),
+]
+
+# default CUDA compute capabilities to use (override via --cuda-compute-capabilities)
+cuda_compute_capabilities = ['5.0', '6.0', '7.0', '7.5', '8.0', '8.6', '9.0']
+
+moduleclass = 'lib'
diff --git a/easybuild/easyconfigs/o/OpenMPI/OpenMPI-5.0.3-NVHPC-24.9-CUDA-12.6.0.eb b/easybuild/easyconfigs/o/OpenMPI/OpenMPI-5.0.3-NVHPC-24.9-CUDA-12.6.0.eb
@@ -0,0 +1,63 @@
+name = 'OpenMPI'
+version = '5.0.3'
+
+homepage = 'https://www.open-mpi.org/'
+description = """The Open MPI Project is an open source MPI-3 implementation."""
+
+toolchain = {'name': 'NVHPC', 'version': '24.9-CUDA-12.6.0'}
+
+source_urls = ['https://www.open-mpi.org/software/ompi/v%(version_major_minor)s/downloads']
+sources = [SOURCELOWER_TAR_BZ2]
+patches = [
+    'OpenMPI-5.0.3_fix_hle_make_errors.patch',
+    'OpenMPI-5.0.3_disable_opal_path_nfs_test.patch',
+    ('OpenMPI-5.0.2_build-with-internal-cuda-header.patch', 1)
+]
+checksums = [
+    {'openmpi-5.0.3.tar.bz2':
+     '990582f206b3ab32e938aa31bbf07c639368e4405dca196fabe7f0f76eeda90b'},
+    {'OpenMPI-5.0.3_fix_hle_make_errors.patch':
+     '881c907a9f5901d5d6af41cd33dffdcecba4a67a9e5123e602542aea57a80895'},
+    {'OpenMPI-5.0.3_disable_opal_path_nfs_test.patch':
+     '75d4417e35252ea3a19b2792f1b06e9aeb408c253aa4921d77226d57b71dee45'},
+    {'OpenMPI-5.0.2_build-with-internal-cuda-header.patch':
+     'f52dc470543f35efef10d651dd159c771ae25f8f76a420d20d87abf4dc769ed7'},
+]
+
+builddependencies = [
+    ('pkgconf', '2.2.0'),
+    ('Perl', '5.38.2'),
+    ('Autotools', '20231222'),
+]
+
+dependencies = [
+    ('zlib', '1.3.1'),
+    ('hwloc', '2.10.0'),
+    ('libevent', '2.1.12'),
+    ('UCX', '1.16.0'),
+    ('UCX-CUDA', '1.16.0', '-CUDA-%(cudaver)s'),
+    ('libfabric', '1.21.0'),
+    ('PMIx', '5.0.2'),
+    ('PRRTE', '3.0.5'),
+    ('UCC', '1.3.0'),
+    ('UCC-CUDA', '1.3.0', '-CUDA-%(cudaver)s'),
+]
+
+# CUDA related patches and custom configure option can be removed if CUDA support isn't wanted.
+preconfigopts = 'nvc -Iopal/mca/cuda/include -shared opal/mca/cuda/lib/cuda.c -o opal/mca/cuda/lib/libcuda.so && '
+# Update configure to include changes from the "disable_opal_path_nfs_test" patch
+preconfigopts += './autogen.pl --force && '
+
+configopts = '--with-cuda=%(start_dir)s/opal/mca/cuda '
+# Required to prevent internal compiler error in opal.
+configopts += '--enable-alt-short-float=no '
+# Set PGI compilers manually, as NVHPC compilers are not correctly detected
+configopts += 'CC=pgcc CXX=pgc++ FC=pgfortran '
+
+# site specific options
+# configopts += '--without-psm2 '
+# configopts += '--disable-oshmem '
+# configopts += '--with-gpfs '
+configopts += '--with-slurm '
+
+moduleclass = 'mpi'
diff --git a/easybuild/easyconfigs/o/OpenMPI/OpenMPI-5.0.3_disable_opal_path_nfs_test.patch b/easybuild/easyconfigs/o/OpenMPI/OpenMPI-5.0.3_disable_opal_path_nfs_test.patch
@@ -0,0 +1,33 @@
+Disable opal_path_nfs test in OpenMPI 5.0.3 as this test can easily fail on some systems,
+when NFS mounts are used. Generally, this test is flaky, which may prevent users
+from installing OpenMPI for no apparent reason.
+
+diff --git a/test/util/Makefile.am b/test/util/Makefile.am
+index e5ad472..33d63c4 100644
+--- a/test/util/Makefile.am
++++ b/test/util/Makefile.am
+@@ -38,7 +38,6 @@ AM_CPPFLAGS = -I$(top_srcdir)/test/support
+
+ check_PROGRAMS = \
+ 	opal_bit_ops \
+-	opal_path_nfs \
+ 	bipartite_graph \
+         opal_sha256
+
+@@ -80,11 +79,11 @@ opal_bit_ops_LDADD = \
+         $(top_builddir)/test/support/libsupport.a
+ opal_bit_ops_DEPENDENCIES = $(opal_path_nfs_LDADD)
+
+-opal_path_nfs_SOURCES = opal_path_nfs.c
+-opal_path_nfs_LDADD = \
+-        $(top_builddir)/opal/lib@[email protected] \
+-        $(top_builddir)/test/support/libsupport.a
+-opal_path_nfs_DEPENDENCIES = $(opal_path_nfs_LDADD)
++# opal_path_nfs_SOURCES = opal_path_nfs.c
++# opal_path_nfs_LDADD = \
++#         $(top_builddir)/opal/lib@[email protected] \
++#         $(top_builddir)/test/support/libsupport.a
++# opal_path_nfs_DEPENDENCIES = $(opal_path_nfs_LDADD)
+
+ #opal_os_path_SOURCES = opal_os_path.c
+ #opal_os_path_LDADD = \
diff --git a/easybuild/easyconfigs/o/OpenMPI/OpenMPI-5.0.3_fix_hle_make_errors.patch b/easybuild/easyconfigs/o/OpenMPI/OpenMPI-5.0.3_fix_hle_make_errors.patch
@@ -0,0 +1,25 @@
+On JUWELS, building OpenMPI 5.0.5 with NVHPC 24.9 fails with errors
+related to `__ATOMIC_HLE_ACQUIRE` and `__ATOMIC_HLE_RELEASE` not being
+defined. Add an additional macro check to let the build succeed.
+
+--- a/opal/include/opal/sys/gcc_builtin/atomic.h	2024-07-23 01:23:20.567556032 +0200
++++ a/opal/include/opal/sys/gcc_builtin/atomic.h	2024-10-02 12:19:53.130698758 +0200
+@@ -187,7 +187,7 @@
+  *
+  *********************************************************************/
+
+-#if defined(__HLE__)
++#if defined(__HLE__) && defined(__ATOMIC_HLE_ACQUIRE) && defined(__ATOMIC_HLE_RELEASE)
+
+ #    include <immintrin.h>
+
+@@ -225,7 +225,7 @@
+                      __ATOMIC_RELEASE | __ATOMIC_HLE_RELEASE);
+ }
+
+-#else  /* #if defined(__HLE__) */
++#else  /* #if defined(__HLE__) && defined(__ATOMIC_HLE_ACQUIRE) && defined(__ATOMIC_HLE_RELEASE) */
+
+ #include "opal/sys/atomic_impl_spinlock.h"
+
+
diff --git a/easybuild/easyconfigs/u/UCC-CUDA/UCC-CUDA-1.3.0-GCCcore-13.3.0-CUDA-12.6.0.eb b/easybuild/easyconfigs/u/UCC-CUDA/UCC-CUDA-1.3.0-GCCcore-13.3.0-CUDA-12.6.0.eb
@@ -0,0 +1,55 @@
+easyblock = 'ConfigureMake'
+
+name = 'UCC-CUDA'
+version = '1.3.0'
+versionsuffix = '-CUDA-%(cudaver)s'
+
+homepage = 'https://www.openucx.org/'
+description = """UCC (Unified Collective Communication) is a collective
+communication operations API and library that is flexible, complete, and 
+feature-rich for current and emerging programming models and runtimes.
+
+This module adds the UCC CUDA support.
+"""
+
+toolchain = {'name': 'GCCcore', 'version': '13.3.0'}
+toolchainopts = {'pic': True}
+
+source_urls = ['https://github.com/openucx/ucc/archive/refs/tags']
+sources = ['v%(version)s.tar.gz']
+patches = [
+    '%(name)s-%(version)s_link_against_existing_UCC_libs.patch',
+]
+checksums = [
+    {'v1.3.0.tar.gz': 'b56379abe5f1c125bfa83be305d78d81a64aa271b7b5fff0ac17b86725ff3acf'},
+    {'UCC-CUDA-1.3.0_link_against_existing_UCC_libs.patch':
+     '758228357ce2a6ae50fb26a0b43e9176feaf379e266365f38205ce679267fc0d'},
+]
+
+builddependencies = [
+    ('binutils', '2.42'),
+    ('Autotools', '20231222'),
+]
+
+dependencies = [
+    ('UCC', version),
+    ('CUDA',  '12.6.0', '', SYSTEM),
+    ('UCX-CUDA', '1.16.0', '-CUDA-%(cudaver)s'),
+    ('NCCL', '2.22.3', '-CUDA-%(cudaver)s'),
+]
+
+preconfigopts = "./autogen.sh && "
+
+buildopts = '-C src/components/mc/cuda V=1 && make -C src/components/tl/nccl V=1'
+installopts = '-C src/components/mc/cuda && make -C src/components/tl/nccl install'
+
+sanity_check_paths = {
+    'files': ['lib/ucc/libucc_mc_cuda.%s' % SHLIB_EXT, 'lib/ucc/libucc_tl_nccl.%s' % SHLIB_EXT],
+    'dirs': ['lib']
+}
+
+sanity_check_commands = ["ucc_info -c"]
+
+modextrapaths = {'EB_UCC_EXTRA_COMPONENT_PATH': 'lib/ucc'}
+
+moduleclass = 'lib'
diff --git a/easybuild/easyconfigs/u/UCC-CUDA/UCC-CUDA-1.3.0_link_against_existing_UCC_libs.patch b/easybuild/easyconfigs/u/UCC-CUDA/UCC-CUDA-1.3.0_link_against_existing_UCC_libs.patch
@@ -0,0 +1,27 @@
+Make CUDA/NCCL plugins link against the existing ucc libraries.
+
+Bart Oldeman, 2022-08-02
+Mikael OEhman, 2023-06-16
+diff -ur ucc-1.3.0.orig/src/components/mc/cuda/Makefile.am ucc-1.3.0/src/components/ec/cuda/Makefile.am
+--- ucc-1.3.0.orig/src/components/mc/cuda/Makefile.am.orig	2023-06-16 12:56:53.205939925 +0200
++++ ucc-1.3.0/src/components/mc/cuda/Makefile.am	2023-06-16 13:02:21.716110609 +0200
+@@ -14,7 +14,7 @@
+ libucc_mc_cuda_la_CFLAGS   = $(BASE_CFLAGS)
+ libucc_mc_cuda_la_LDFLAGS  = -version-info $(SOVERSION) --as-needed $(CUDA_LDFLAGS)
+ libucc_mc_cuda_la_LIBADD   = $(CUDA_LIBS)                      \
+-                             $(UCC_TOP_BUILDDIR)/src/libucc.la
++                             -lucc
+
+ include $(top_srcdir)/config/module.am
+ endif
+diff -ur ucc-1.0.0.orig/src/components/tl/nccl/Makefile.am ucc-1.0.0/src/components/tl/nccl/Makefile.am
+--- ucc-1.0.0.orig/src/components/tl/nccl/Makefile.am	2022-04-15 12:43:33.000000000 +0000
++++ ucc-1.0.0/src/components/tl/nccl/Makefile.am	2022-08-02 12:13:59.334795989 +0000
+@@ -21,6 +21,6 @@
+ libucc_tl_nccl_la_CPPFLAGS = $(AM_CPPFLAGS) $(BASE_CPPFLAGS) $(CUDA_CPPFLAGS) $(NCCL_CPPFLAGS)
+ libucc_tl_nccl_la_CFLAGS   = $(BASE_CFLAGS)
+ libucc_tl_nccl_la_LDFLAGS  = -version-info $(SOVERSION) --as-needed $(CUDA_LDFLAGS) $(NCCL_LDFLAGS)
+-libucc_tl_nccl_la_LIBADD   = $(CUDA_LIBS) $(NCCL_LIBADD) $(UCC_TOP_BUILDDIR)/src/libucc.la
++libucc_tl_nccl_la_LIBADD   = $(CUDA_LIBS) $(NCCL_LIBADD) -lucc
+
+ include $(top_srcdir)/config/module.am