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

FFTW: expose optimization handles via variants (SIMD + FMA) (#7038)

Browse Source
This commit is contained in:
Massimiliano Culpo 2018-02-13 11:06:58 +01:00 committed by Todd Gamblin
parent b98cdf098a
commit 4d97b540a8
1 changed files with 20 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
var/spack/repos/builtin/packages/fftw/package.py
View File

@ -66,6 +66,21 @@ class Fftw(AutotoolsPackage):
'pfft_patches', default=False, 'pfft_patches', default=False,
description='Add extra transpose functions for PFFT compatibility') description='Add extra transpose functions for PFFT compatibility')
variant(
'simd',
default='sse2,avx,avx2',
values=(
'sse', 'sse2', 'avx', 'avx2', 'avx512', # Intel
'avx-128-fma', 'kcvi', # Intel
'altivec', 'vsx', # IBM
'neon', # ARM
'generic-simd128', 'generic-simd256' # Generic
),
description='Optimizations that are enabled in this build',
multi=True
)
variant('fma', default=False, description='Activate support for fma')
depends_on('mpi', when='+mpi') depends_on('mpi', when='+mpi')
depends_on('automake', type='build', when='+pfft_patches') depends_on('automake', type='build', when='+pfft_patches')
depends_on('autoconf', type='build', when='+pfft_patches') depends_on('autoconf', type='build', when='+pfft_patches')
@ -109,12 +124,11 @@ def configure(self, spec, prefix):
options.append('--enable-mpi') options.append('--enable-mpi')
# SIMD support # SIMD support
# TODO: add support for more architectures float_options, double_options = [], []
float_options = [] if spec.satisfies('@3:', strict=True):
double_options = [] for opts in (float_options, double_options):
if 'x86_64' in spec.architecture and spec.satisfies('@3:'): opts += self.enable_or_disable('simd')
float_options.append('--enable-sse2') opts += self.enable_or_disable('fma')
double_options.append('--enable-sse2')
configure = Executable('../configure') configure = Executable('../configure')