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
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
|
# Copyright 2013-2024 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 os
import os.path
import llnl.util.lang
from spack.package import *
class FftwBase(AutotoolsPackage):
"""Base class for building Fftw, shared with the AMD optimized version
of the library in the 'amdfftw' package.
"""
variant(
"precision",
values=any_combination_of("float", "double", "long_double", "quad")
.prohibit_empty_set()
.with_default("float,double"),
description="Build the selected floating-point precision libraries",
)
variant("openmp", default=False, description="Enable OpenMP support.")
variant("mpi", default=True, description="Activate MPI support")
depends_on("mpi", when="+mpi")
depends_on("llvm-openmp", when="%apple-clang +openmp")
# https://github.com/FFTW/fftw3/commit/902d0982522cdf6f0acd60f01f59203824e8e6f3
conflicts("%gcc@8.0:8", when="@3.3.7")
conflicts(
"precision=long_double",
when="@2.1.5",
msg="Long double precision is not supported in FFTW 2",
)
conflicts("precision=quad", when="@2.1.5", msg="Quad precision is not supported in FFTW 2")
conflicts("precision=quad", when="+mpi", msg="Quad precision is not supported in MPI")
@property
def libs(self):
# Reduce repetitions of entries
query_parameters = list(llnl.util.lang.dedupe(self.spec.last_query.extra_parameters))
# List of all the suffixes associated with float precisions
precisions = [("float", "f"), ("double", ""), ("long_double", "l"), ("quad", "q")]
# Retrieve the correct suffixes, or use double as a default
suffixes = [v for k, v in precisions if k in query_parameters] or [""]
# Construct the list of libraries that needs to be found
libraries = []
for sfx in suffixes:
if "mpi" in query_parameters and "+mpi" in self.spec:
libraries.append("libfftw3" + sfx + "_mpi")
if "openmp" in query_parameters and "+openmp" in self.spec:
libraries.append("libfftw3" + sfx + "_omp")
libraries.append("libfftw3" + sfx)
return find_libraries(libraries, root=self.prefix, recursive=True)
def flag_handler(self, name, flags):
if name == "cflags":
if self.spec.satisfies("%clang@15:"):
flags.append("-Wno-error=int-conversion")
return flags, None, None
def patch(self):
# If fftw/config.h exists in the source tree, it will take precedence
# over the copy in build dir. As only the latter has proper config
# for our build, this is a problem. See e.g. issue #7372 on github
if os.path.isfile("fftw/config.h"):
os.rename("fftw/config.h", "fftw/config.h.SPACK_RENAMED")
def autoreconf(self, spec, prefix):
if "+pfft_patches" in spec:
autoreconf = which("autoreconf")
autoreconf("-ifv")
@property
def selected_precisions(self):
"""Precisions that have been selected in this build"""
return self.spec.variants["precision"].value
def setup_build_environment(self, env):
if self.spec.satisfies("%apple-clang +openmp"):
env.append_flags("CPPFLAGS", self.compiler.openmp_flag)
env.append_flags("CFLAGS", self.spec["llvm-openmp"].headers.include_flags)
env.append_flags("CXXFLAGS", self.spec["llvm-openmp"].headers.include_flags)
env.append_flags("LDFLAGS", self.spec["llvm-openmp"].libs.ld_flags)
# FFTW first configures libtool without MPI, and later uses it with
# MPI. libtool then calls wrong linker to create shared libraries
# (it calls `$CC` instead of `$MPICC`), and MPI symbols
# remain undefined because `-lmpi` is not passed to the linker.
# https://github.com/FFTW/fftw3/issues/274
# https://github.com/spack/spack/issues/29224
if self.spec.satisfies("+mpi") and self.spec.satisfies("platform=darwin"):
env.append_flags("LIBS", self.spec["mpi"].libs.ld_flags)
def configure(self, spec, prefix):
# Base options
options = ["--prefix={0}".format(prefix), "--enable-shared", "--enable-threads"]
if not self.compiler.f77 or not self.compiler.fc:
options.append("--disable-fortran")
if spec.satisfies("@:2"):
options.append("--enable-type-prefix")
# Variants that affect every precision
if "+openmp" in spec:
options.append("--enable-openmp")
if spec.satisfies("@:2"):
# TODO: libtool strips CFLAGS, so 2.x libxfftw_threads
# isn't linked to the openmp library. Patch Makefile?
options.insert(0, "CFLAGS=" + self.compiler.openmp_flag)
if "+mpi" in spec:
options.append("--enable-mpi")
# Specific SIMD support.
# all precisions
simd_features = ["sse2", "avx", "avx2", "avx512", "avx-128-fma", "kcvi", "vsx"]
# float only
float_simd_features = ["altivec", "sse", "neon"]
# Workaround PGI compiler bug when avx2 is enabled
if spec.satisfies("%pgi") and "avx2" in simd_features:
simd_features.remove("avx2")
# Workaround NVIDIA/PGI compiler bug when avx512 is enabled
if spec.satisfies("%nvhpc") or spec.satisfies("%pgi"):
if "avx512" in simd_features:
simd_features.remove("avx512")
# NVIDIA compiler does not support Altivec intrinsics
if spec.satisfies("%nvhpc") and "vsx" in simd_features:
simd_features.remove("vsx")
if spec.satisfies("%nvhpc") and "altivec" in float_simd_features:
float_simd_features.remove("altivec")
# NVIDIA compiler does not support Neon intrinsics
if spec.satisfies("%nvhpc") and "neon" in simd_features:
simd_features.remove("neon")
simd_options = []
for feature in simd_features:
msg = "--enable-{0}" if feature in spec.target else "--disable-{0}"
simd_options.append(msg.format(feature))
# If no features are found, enable the generic ones
if not any(f in spec.target for f in simd_features + float_simd_features):
# Workaround NVIDIA compiler bug
if not spec.satisfies("%nvhpc"):
simd_options += ["--enable-generic-simd128", "--enable-generic-simd256"]
simd_options += ["--enable-fma" if "fma" in spec.target else "--disable-fma"]
# Double is the default precision, for all the others we need
# to enable the corresponding option.
enable_precision = {
"float": ["--enable-float"],
"double": None,
"long_double": ["--enable-long-double"],
"quad": ["--enable-quad-precision"],
}
# Different precisions must be configured and compiled one at a time
configure = Executable("../configure")
for precision in self.selected_precisions:
opts = (enable_precision[precision] or []) + options[:]
# SIMD optimizations are available only for float and double
# starting from FFTW 3
if precision in ("float", "double") and spec.satisfies("@3:"):
opts += simd_options
# float-only acceleration
if precision == "float":
for feature in float_simd_features:
if feature in spec.target:
msg = "--enable-{0}"
else:
msg = "--disable-{0}"
opts.append(msg.format(feature))
with working_dir(precision, create=True):
configure(*opts)
def for_each_precision_make(self, *targets):
for precision in self.selected_precisions:
with working_dir(precision):
make(*targets)
def build(self, spec, prefix):
self.for_each_precision_make()
def check(self):
self.for_each_precision_make("check")
def install(self, spec, prefix):
self.for_each_precision_make("install")
class Fftw(FftwBase):
"""FFTW is a C subroutine library for computing the discrete Fourier
transform (DFT) in one or more dimensions, of arbitrary input
size, and of both real and complex data (as well as of even/odd
data, i.e. the discrete cosine/sine transforms or DCT/DST). We
believe that FFTW, which is free software, should become the FFT
library of choice for most applications."""
homepage = "https://www.fftw.org"
url = "https://www.fftw.org/fftw-3.3.4.tar.gz"
list_url = "https://www.fftw.org/download.html"
license("GPL-2.0-or-later")
version("3.3.10", sha256="56c932549852cddcfafdab3820b0200c7742675be92179e59e6215b340e26467")
version("3.3.9", sha256="bf2c7ce40b04ae811af714deb512510cc2c17b9ab9d6ddcf49fe4487eea7af3d")
version("3.3.8", sha256="6113262f6e92c5bd474f2875fa1b01054c4ad5040f6b0da7c03c98821d9ae303")
version("3.3.7", sha256="3b609b7feba5230e8f6dd8d245ddbefac324c5a6ae4186947670d9ac2cd25573")
version("3.3.6-pl2", sha256="a5de35c5c824a78a058ca54278c706cdf3d4abba1c56b63531c2cb05f5d57da2")
version("3.3.5", sha256="8ecfe1b04732ec3f5b7d279fdb8efcad536d555f9d1e8fabd027037d45ea8bcf")
version("3.3.4", sha256="8f0cde90929bc05587c3368d2f15cd0530a60b8a9912a8e2979a72dbe5af0982")
version("2.1.5", sha256="f8057fae1c7df8b99116783ef3e94a6a44518d49c72e2e630c24b689c6022630")
variant(
"pfft_patches",
default=False,
description="Add extra transpose functions for PFFT compatibility",
)
depends_on("automake", type="build", when="+pfft_patches")
depends_on("autoconf", type="build", when="+pfft_patches")
depends_on("libtool", type="build", when="+pfft_patches")
provides("fftw-api@2", when="@2.1.5")
provides("fftw-api@3", when="@3:")
patch("pfft-3.3.9.patch", when="@3.3.9:+pfft_patches", level=0)
patch("pfft-3.3.5.patch", when="@3.3.5:3.3.8+pfft_patches", level=0)
patch("pfft-3.3.4.patch", when="@3.3.4+pfft_patches", level=0)
patch("pgi-3.3.6-pl2.patch", when="@3.3.6-pl2%pgi", level=0)
patch("intel-configure.patch", when="@3:3.3.8%intel", level=0)
|
