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
|
# Copyright 2013-2023 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)
from spack.package import *
class Nsimd(CMakePackage):
"""NSIMD is a vectorization library that abstracts SIMD programming.
It was designed to exploit the maximum power of processors
at a low development cost.
"""
homepage = "https://agenium-scale.github.io/nsimd/"
url = "https://github.com/agenium-scale/nsimd/archive/v1.0.tar.gz"
maintainers("eschnett")
license("MIT")
version("3.0.1", sha256="6a90d7ce5f9da5cfac872463951f3374bb0e0824d92f714db0fd4901b32497fd")
version("3.0", sha256="5cab09020ce3a6819ddb3b3b8cafa6bc1377821b596c0f2954f52c852d092d5c")
version("2.2", sha256="7916bec6c8ea9ddc690a5bfc80fb1b9402f9e1b2a4b4bb6b6bb8eb5a07eb018e")
version("2.1", sha256="3274f1061d1fac170130b8c75378a6b94580629b3dc1d53db244b51500ee4695")
# Version 2.0 is disabled since it does not support cmake
# version('2.0', sha256='b239e98316f93257161b25c8232634884edcee358982a74742981cc9b68da642')
version("1.0", sha256="523dae83f1d93eab30114321f1c9a67e2006a52595da4c51f121ca139abe0857")
variant(
"simd",
default="auto",
description="SIMD instruction set",
values=(
"auto",
"CPU",
"SSE2",
"SSE42",
"AVX",
"AVX2",
"AVX512_KNL",
"AVX512_SKYLAKE",
"NEON128",
"AARCH64",
"SVE",
conditional("SVE128", "SVE256", "SVE512", "SVE1024", "SVE2048", when="@2:"),
conditional("VMX", "VSX", when="@3:"),
conditional("CUDA", "ROCM", when="@2:"),
),
multi=False,
)
variant(
"optionals",
values=any_combination_of("FMA", "FP16"),
when="@:1",
description="Optional SIMD features",
)
# Requires a C++14 compiler for building.
# The C++ interface requires a C++11 compiler to use.
depends_on("cmake@2.8.7:", type="build")
depends_on("cmake@3.0.2:", type="build", when="@2:")
# We need `collections.MutableMapping`, which was removed in
# Python 3.10; see
# <https://docs.python.org/3.9/library/collections.html>
depends_on("python@3.0:3.9", type="build")
depends_on("py-chardet", type="build", when="@3:")
depends_on("py-requests", type="build", when="@3:")
depends_on("py-urllib3", type="build", when="@3:")
@run_before("cmake")
def generate_code(self):
"""Auto-generates code in the build directory"""
spec = self.spec
if self.spec.satisfies("@:1"):
options = ["egg/hatch.py", "--all", "--force"]
python = spec["python"].command
python(*options)
def cmake_args(self):
# Required SIMD argument
simd = self.spec.variants["simd"].value
if simd == "auto":
# x86
if "avx512" in self.spec.target:
simd = "AVX512_SKYLAKE"
elif self.spec.satisfies("target=mic_knl"):
simd = "AVX512_KNL"
elif "avx2" in self.spec.target:
simd = "AVX2"
elif "avx" in self.spec.target:
simd = "AVX"
elif "sse4_2" in self.spec.target:
simd = "SSE42"
elif "sse2" in self.spec.target:
simd = "SSE2"
# ARM
elif "sve" in self.spec.target:
# We require an explicit choice for particluar bit widths
simd = "SVE"
elif self.spec.satisfies("target=aarch64:"):
simd = "AARCH64"
elif "neon" in self.spec.target:
simd = "NEON128"
# POWER
elif "vsx" in self.spec.target:
simd = "VSX"
elif self.spec.satisfies("target=ppc64:") or self.spec.satisfies("target=ppc64le:"):
simd = "VMX"
# Unknown CPU architecture
else:
simd = "CPU"
if self.spec.satisfies("@:1"):
cmake_args = ["-DSIMD={0}".format(simd)]
else:
cmake_args = ["-Dsimd={0}".format(simd)]
if self.spec.satisfies("@:1"):
# Optional SIMD instructions to be turned on explicitly
optionals_value = self.spec.variants["optionals"].value
if optionals_value != "none":
optionals_arg = ";".join(optionals_value)
cmake_args.append("-DSIMD_OPTIONALS={0}".format(optionals_arg))
return cmake_args
|
