1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
|
# 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 glob
from spack.package import *
class Ffte(Package):
"""Fastest Fourier Transform in the East. Provides FFT for powers of 2, 3,
and 5 lengths in one, two, and three dimensions. Support for vector
hardware, MPI, and CUDA Fortran is also included."""
homepage = 'http://www.ffte.jp/'
url = 'http://www.ffte.jp/ffte-7.0.tgz'
maintainers = ['luszczek']
version('7.0', sha256='078d5f84a5f2479ca5c4a3bd777ad761fe98addf1642a045bac6602a0cae3da0')
version('6.0', sha256='fc82595a8f8323b2796cc5eeb1cc9f7e50ca9e511a14365cc3984da6b7a9b8b4')
version('5.0', sha256='1f46ca16badc3aca0ad13ca91a6a67829a57b403501cdc821b80cfa62b2a89c2')
version('4.0', sha256='61680f73c48659ac45aec60ef5a725547f763bb9017edbd3f44a6a9ad0fda62f')
version('3.0', sha256='dbaab8204a16072c8d572efa3733e9946a9be0d1a051fc19e2d9253be23247ff')
version('2.0', sha256='f5cf1d1f880288e359f4d517191980ffca4420f817ecaa2d754ca5c5421271e3')
version('1.0', sha256='35171e3324019018c25575b2807a6513fa85badad040f30f238fff03d4b4d1ab')
variant('mpi', default=False, description='Build MPI library')
variant('cuda', default=False, description='Use CUDA Fortran')
variant('vector', default=False, description='Use vectorized FFT')
depends_on('mpi', when='+mpi')
conflicts('%cce', when='+cuda', msg='Must use NVHPC compiler')
conflicts('%clang', when='+cuda', msg='Must use NVHPC compiler')
conflicts('%gcc', when='+cuda', msg='Must use NVHPC compiler')
conflicts('%llvm', when='+cuda', msg='Must use NVHPC compiler')
conflicts('%nag', when='+cuda', msg='Must use NVHPC compiler')
conflicts('%intel', when='+cuda', msg='Must use NVHPC compiler')
def edit(self, spec, prefix):
'No make-file, must create one from scratch.'
open('mpi/param.h', 'w').write(open('param.h').read())
for fpfx in ('.', 'mpi'):
vrs = '%sfftever' % ['', '', '', 'p'][len(fpfx)]
s6 = 6 * ' '
fd = open('%s/%s.f' % (fpfx, vrs), 'w')
fd.write('%ssubroutine %s(i,j,k)\n' % (s6, vrs))
fd.write('%sinteger i,j,k\n' % s6)
fd.write('%si=%s\n' % (s6, spec.version[0]))
fd.write('%sj=%s\n' % (s6, spec.version[1]))
fd.write('%sk=0\n%sreturn\n%send\n' % (3 * (s6,)))
fd.close()
ff, cuf, vf = [], [], []
for f in glob.glob('%s/*.f' % fpfx):
nf = f.replace('.f', '.o').replace('%s/' % fpfx, '')
if '/cu' in f or '/pcu' in f:
cuf.append(nf)
elif '/v' in f or '/pv' in f:
vf.append(nf)
else:
ff.append(nf)
if '.' == fpfx:
lbnm = 'libffte.a'
else:
lbnm = 'libfftempi.a'
for spc, vrf, rs, ip in (
('+vector', vf, 'v', 2),
('+cuda', cuf, 'cu', 3),
):
if spc in spec:
for f in vrf: # replace with variant versions
orgf = f[:ip].replace(rs, '') + f[ip:]
if orgf in ff: # already reference implementation?
del ff[ff.index(orgf)]
ff.extend(vrf)
aff = ' '.join(ff)
fd = open('%s/Makefile' % fpfx, 'w')
fd.write('all: %s\n' % lbnm)
if '+mpi' in spec and '.' == fpfx:
fd.write('\tcd mpi ; $(MAKE)\n')
fd.write('\n%s: %s\n' % (lbnm, aff))
fd.write('\tar rc %s %s\n' % (lbnm, aff))
fd.write('\tranlib %s\n' % lbnm)
fd.close()
def install(self, spec, prefix):
self.edit(spec, prefix)
if '+mpi' in spec:
env['CC'] = spec['mpi'].mpicc
env['F77'] = spec['mpi'].mpif77
env['FC'] = spec['mpi'].mpifc
# allow real/complex aliasing in GNU Fortran 10 and up
if spec.satisfies('%gcc@10:'):
env['FFLAGS'] = '-fallow-argument-mismatch'
# enable CUDA Fortran in NVHPC
if spec.satisfies('%nvhpc'):
env['FFLAGS'] = '-Mcuda'
make()
mkdirp(prefix.lib)
install('libffte.a', prefix.lib)
if '+mpi' in spec:
install('mpi/libfftempi.a', prefix.lib)
|
