zhangyiss/spack
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

heffte: add v2.1.0 (#24599)

Browse Source
This commit is contained in:
Miroslav Stoyanov 2021-07-09 11:11:13 -04:00 committed by GitHub
parent a53f4c36c6
commit b3bdc2ef38
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 28 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
var/spack/repos/builtin/packages/heffte/package.py
View File

@ -6,7 +6,7 @@
from spack import *
class Heffte(CMakePackage):
class Heffte(CMakePackage, CudaPackage, ROCmPackage):
"""Highly Efficient FFT for Exascale"""
homepage = "https://bitbucket.org/icl/heffte"
@ -18,6 +18,7 @@ class Heffte(CMakePackage):
test_requires_compiler = True
version('develop', branch='master')
version('2.1.0', sha256='527a3e21115231715a0342afdfaf6a8878d2dd0f02f03c92b53692340fd940b9')
version('2.0.0', sha256='12f2b49a1a36c416eac174cf0cc50e729d56d68a9f68886d8c34bd45a0be26b6')
version('1.0', sha256='0902479fb5b1bad01438ca0a72efd577a3529c3d8bad0028f3c18d3a4935ca74')
version('0.2', sha256='4e76ae60982b316c2e873b2e5735669b22620fefa1fc82f325cdb6989bec78d1')
@ -29,7 +30,6 @@ class Heffte(CMakePackage):
variant('shared', default=True, description='Builds with shared libraries')
variant('fftw', default=False, description='Builds with support for FFTW backend')
variant('mkl', default=False, description='Builds with support for MKL backend')
variant('cuda', default=False, description='Builds with support for GPUs via CUDA')
variant('magma', default=False, description='Use helper methods from the UTK MAGMA library')
variant('python', default=False, description='Install the Python bindings')
variant('fortran', default=False, description='Install the Fortran modules')
@ -40,10 +40,11 @@ class Heffte(CMakePackage):
depends_on('py-numba', when='+python+cuda', type=('build', 'run'))
extends('python', when='+python', type=('build', 'run'))
conflicts('~fftw', when='~mkl~cuda') # requires at least one backend
conflicts('~fftw', when='@:2.1.0~mkl~cuda') # requires at least one backend
conflicts('+fftw', when='+mkl@:1.0') # old API supports at most one CPU backend
conflicts('^openmpi~cuda', when='+cuda') # +cuda requires CUDA enabled OpenMPI
conflicts('~cuda', when='+magma') # magma requires CUDA or HIP
conflicts('+rocm', when='@:2.1.0') # heffte+rocm is in in development in spack
conflicts('+python', when="@:1.0") # python support was added post v1.0
conflicts('+fortran', when="@:1.0") # fortran support was added post v1.0
conflicts('+magma', when="@:1.0") # magma support was added post v1.0
@ -51,14 +52,19 @@ class Heffte(CMakePackage):
depends_on('mpi', type=('build', 'run'))
depends_on('fftw@3.3.8:', when="+fftw", type=('build', 'run'))
depends_on('intel@16.0:', when="+mkl", type=('build', 'run'))
depends_on('intel-mkl@2018.0.128:', when="+mkl", type=('build', 'run'))
depends_on('cuda@8.0:', when="+cuda", type=('build', 'run'))
depends_on('hip@3.8.0:', when='+rocm')
depends_on('rocfft@3.8.0:', when='+rocm')
depends_on('magma@2.5.3:', when="+cuda+magma", type=('build', 'run'))
examples_src_dir = 'examples'
def cmake_args(self):
return [
args = [
self.define_from_variant('BUILD_SHARED_LIBS', 'shared'),
self.define_from_variant('Heffte_ENABLE_CUDA', 'cuda'),
self.define_from_variant('Heffte_ENABLE_ROCM', 'rocm'),
self.define_from_variant('Heffte_ENABLE_FFTW', 'fftw'),
self.define_from_variant('Heffte_ENABLE_MKL', 'mkl'),
self.define_from_variant('Heffte_ENABLE_MAGMA', 'magma'),
@ -68,7 +74,23 @@ def cmake_args(self):
'ON' if ('+cuda' in self.spec and
'+fftw' in self.spec) else 'OFF'), ]
examples_src_dir = 'examples'
if '+cuda' in self.spec:
cuda_arch = self.spec.variants['cuda_arch'].value
if len(cuda_arch) > 0 or cuda_arch[0] != 'none':
nvcc_flags = ""
for nvflag in self.cuda_flags(cuda_arch):
nvcc_flags += "{0};".format(nvflag)
args.append('-DCUDA_NVCC_FLAGS={0}'.format(nvcc_flags))
if '+rocm' in self.spec:
args.append('-DCMAKE_CXX_COMPILER={0}'.format(self.spec['hip'].hipcc))
rocm_arch = self.spec.variants['amdgpu_target'].value
if 'none' not in rocm_arch:
args.append('-DCMAKE_CXX_FLAGS={0}'.format(self.hip_flags(rocm_arch)))
return args
@run_after('install')
def cache_test_sources(self):