# Copyright 2013-2022 Lawrence Livermore National Security, LLC and other # Spack Project Developers. See the top-level COPYRIGHT file for details. # # SPDX-License-Identifier: (Apache-2.0 OR MIT) import sys from spack import * from spack.pkg.builtin.boost import Boost class Hpx(CMakePackage, CudaPackage, ROCmPackage): """C++ runtime system for parallel and distributed applications.""" homepage = "https://hpx.stellar-group.org/" url = "https://github.com/STEllAR-GROUP/hpx/archive/1.2.1.tar.gz" git = "https://github.com/STEllAR-GROUP/hpx.git" maintainers = ['msimberg', 'albestro', 'teonnik'] tags = ['e4s'] version('master', branch='master') version('stable', tag='stable') version('1.8.0', sha256='93f147ab7cf0ab4161f37680ea720d3baeb86540a95382f2fb591645b2a9b135') version('1.7.1', sha256='008a0335def3c551cba31452eda035d7e914e3e4f77eec679eea070ac71bd83b') version('1.7.0', sha256='05099b860410aa5d8a10d6915b1a8818733aa1aa2d5f2b9774730ca7e6de5fac') version('1.6.0', sha256='4ab715613c1e1808edc93451781cc9bc98feec4e422ccd4322858a680f6d9017') version('1.5.1', sha256='b2f9358ce2a9446b9d8fb1998c30913e7199b007aa82e46d0aa05c763331c635') version('1.5.0', sha256='de2901d8ae017592c513e0af9cf58de295abc9802e55ece00424cbd8a3801920') version('1.4.1', sha256='965dabe44d17480e326d92da4eec56722d98b33943c53d2b0f8f4655cb208023') version('1.4.0', sha256='241a1c47fafba751848fac12446e7bf4ad3d342d5eb2fa1ef94dd904acc329ed') version('1.3.0', sha256='cd34da674064c4cc4a331402edbd65c5a1f8058fb46003314ca18fa08423c5ad') version('1.2.1', sha256='8cba9b48e919035176d3b7bbfc2c110df6f07803256626f1dad8d9dde16ab77a') version('1.2.0', sha256='20942314bd90064d9775f63b0e58a8ea146af5260a4c84d0854f9f968077c170') version('1.1.0', sha256='1f28bbe58d8f0da600d60c3a74a644d75ac777b20a018a5c1c6030a470e8a1c9') generator = 'Ninja' map_cxxstd = lambda cxxstd: '2a' if cxxstd == '20' else cxxstd cxxstds = ('11', '14', '17', '20') variant('cxxstd', default='17', values=cxxstds, description='Use the specified C++ standard when building.') variant( 'malloc', default='tcmalloc', description='Define which allocator will be linked in', values=('system', 'jemalloc', 'mimalloc', 'tbbmalloc', 'tcmalloc') ) variant('max_cpu_count', default='64', description='Max number of OS-threads for HPX applications', values=lambda x: isinstance(x, str) and x.isdigit()) instrumentation_values = ('apex', 'google_perftools', 'papi', 'valgrind') variant('instrumentation', values=any_combination_of(*instrumentation_values), description='Add support for various kind of instrumentation') variant( "networking", values=any_combination_of("tcp", "mpi").with_default("tcp"), description="Support for networking through parcelports", ) default_generic_coroutines = True if sys.platform.startswith('linux') or sys.platform == 'win32': default_generic_coroutines = False variant( "generic_coroutines", default=default_generic_coroutines, description='Use Boost.Context as the underlying coroutines' ' context switch implementation.') variant('tools', default=False, description='Build HPX tools') variant('examples', default=False, description='Build examples') variant('async_mpi', default=False, description='Enable MPI Futures.') variant('async_cuda', default=False, description='Enable CUDA Futures.') # Build dependencies depends_on('python', type=('build', 'test', 'run')) depends_on('ninja', type='build') depends_on('pkgconfig', type='build') depends_on('git', type='build') depends_on('cmake', type='build') # Other dependecies depends_on('hwloc') depends_on(Boost.with_default_variants) depends_on('boost +context', when='+generic_coroutines') for cxxstd in cxxstds: depends_on( "boost cxxstd={0}".format(map_cxxstd(cxxstd)), when="cxxstd={0}".format(cxxstd) ) depends_on('asio', when='@1.7:') for cxxstd in cxxstds: depends_on( "asio cxxstd={0}".format(map_cxxstd(cxxstd)), when="cxxstd={0} ^asio".format(cxxstd), ) depends_on('gperftools', when='malloc=tcmalloc') depends_on('jemalloc', when='malloc=jemalloc') depends_on('mimalloc@1', when='malloc=mimalloc') depends_on('tbb', when='malloc=tbbmalloc') depends_on('mpi', when='networking=mpi') depends_on('mpi', when='+async_mpi') depends_on('cuda', when='+async_cuda') depends_on('otf2', when='instrumentation=apex') depends_on('gperftools', when='instrumentation=google_perftools') depends_on('papi', when='instrumentation=papi') depends_on('valgrind', when='instrumentation=valgrind') # Restrictions for stable/master with when("@master"): conflicts("cxxstd=14") depends_on("cuda@11:", when="+cuda") with when("@stable"): conflicts("cxxstd=14") depends_on("cuda@11:", when="+cuda") # Restrictions for 1.7.X with when('@1.7:'): depends_on('cmake@3.18.0:', type='build') depends_on('boost@1.71.0:') depends_on('asio@1.12.0:') conflicts('%gcc@:6') conflicts('%clang@:6') # Restrictions for 1.6.X conflicts('+rocm', when='@:1.5') # Restrictions for 1.5.x conflicts('cxxstd=11', when='@1.5:') # Restrictions for 1.2.X with when('@:1.2.1'): depends_on('cmake@3.9.0:', type='build') depends_on('boost@1.62.0:') depends_on('hwloc@1.11:') # Restrictions before 1.2 with when('@:1.1.0'): depends_on('boost@1.55.0:') depends_on('hwloc@1.6:') # Patches and one-off conflicts # Certain Asio headers don't compile with nvcc from 1.17.0 onwards with # C++17. Starting with CUDA 11.3 they compile again. conflicts("asio@1.17.0:", when="+cuda cxxstd=17 ^cuda@:11.2") # Starting from ROCm 5.0.0 hipcc miscompiles asio 1.17.0 and newer conflicts("asio@1.17.0:", when="+rocm ^hip@5:") # Boost and HIP don't work together in certain versions: # https://github.com/boostorg/config/issues/392. Boost 1.78.0 and HPX 1.8.0 # both include a fix. conflicts("boost@:1.77.0", when="@:1.7 +rocm") # boost 1.73.0 build problem with HPX 1.4.0 and 1.4.1 # https://github.com/STEllAR-GROUP/hpx/issues/4728#issuecomment-640685308 depends_on('boost@:1.72.0', when='@:1.4') # COROUTINES # ~generic_coroutines conflict is not fully implemented # for additional information see: # https://github.com/spack/spack/pull/17654 # https://github.com/STEllAR-GROUP/hpx/issues/4829 depends_on('boost+context', when='+generic_coroutines') _msg_generic_coroutines = 'This platform requires +generic_coroutines' conflicts('~generic_coroutines', when='platform=darwin', msg=_msg_generic_coroutines) # Patches APEX patch('git_external.patch', when='@1.3.0 instrumentation=apex') def instrumentation_args(self): args = [] for value in self.instrumentation_values: condition = 'instrumentation={0}'.format(value) args.append(self.define( 'HPX_WITH_{0}'.format(value.upper()), condition in self.spec )) return args def cmake_args(self): spec, args = self.spec, [] args += [ self.define( 'HPX_WITH_CXX{0}'.format(spec.variants['cxxstd'].value), True), self.define_from_variant('HPX_WITH_MALLOC', 'malloc'), self.define_from_variant('HPX_WITH_CUDA', 'cuda'), self.define_from_variant('HPX_WITH_HIP', 'rocm'), self.define_from_variant('HPX_WITH_TOOLS', 'tools'), self.define_from_variant('HPX_WITH_EXAMPLES', 'examples'), self.define_from_variant('HPX_WITH_ASYNC_MPI', 'async_mpi'), self.define_from_variant('HPX_WITH_ASYNC_CUDA', 'async_cuda'), self.define('HPX_WITH_TESTS', self.run_tests), self.define('HPX_WITH_NETWORKING', 'networking=none' not in spec), self.define('HPX_WITH_PARCELPORT_TCP', 'networking=tcp' in spec), self.define('HPX_WITH_PARCELPORT_MPI', 'networking=mpi' in spec), self.define_from_variant( 'HPX_WITH_MAX_CPU_COUNT', 'max_cpu_count'), self.define_from_variant( 'HPX_WITH_GENERIC_CONTEXT_COROUTINES', 'generic_coroutines'), self.define('BOOST_ROOT', spec['boost'].prefix), self.define('HWLOC_ROOT', spec['hwloc'].prefix), self.define('HPX_WITH_BOOST_ALL_DYNAMIC_LINK', True), self.define('BUILD_SHARED_LIBS', True), self.define('HPX_DATASTRUCTURES_WITH_ADAPT_STD_TUPLE', False), ] # Enable unity builds when available if spec.satisfies("@1.7:"): args += [self.define('HPX_WITH_UNITY_BUILD', True)] # HIP support requires compiling with hipcc if '+rocm' in self.spec: args += [self.define('CMAKE_CXX_COMPILER', self.spec['hip'].hipcc)] if self.spec.satisfies('^cmake@3.21.0:3.21.2'): args += [self.define('__skip_rocmclang', True)] # Instrumentation args += self.instrumentation_args() if 'instrumentation=apex' in spec: args += [ self.define('APEX_WITH_OTF2', True), self.define('OTF2_ROOT', spec['otf2'].prefix), ] # it seems like there was a bug in the default version of APEX in 1.5.x if spec.satisfies("@1.5"): args += [self.define('HPX_WITH_APEX_TAG', "v2.3.0")] return args