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
|
# 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)
from spack.package import *
class Nvshmem(MakefilePackage, CudaPackage):
"""NVSHMEM is a parallel programming interface based on OpenSHMEM that
provides efficient and scalable communication for NVIDIA GPU
clusters. NVSHMEM creates a global address space for data that spans
the memory of multiple GPUs and can be accessed with fine-grained
GPU-initiated operations, CPU-initiated operations, and operations on
CUDA streams."""
homepage = "https://developer.nvidia.com/nvshmem"
maintainers("bvanessen")
license("BSD-3-Clause-Open-MPI")
version("2.7.0-6", sha256="23ed9b0187104dc87d5d2bc1394b6f5ff29e8c19138dc019d940b109ede699df")
version("2.6.0-1", sha256="fc0e8de61b034f3a079dc231b1d0955e665a9f57b5013ee98b6743647bd60417")
version("2.5.0-19", sha256="dd800b40f1d296e1d3ed2a9885adcfe745c3e57582bc809860e87bd32abcdc60")
version("2.4.1-3", sha256="8b6c0eab321b6352911e470f9e81a777a49e58148ec3728453b9522446dba178")
version("2.2.1-0", sha256="c8efc6cd560e0ed66d5fe4c5837c650247bec7b0dc65b5089deb8ab49658e1c3")
version("2.1.2-0", sha256="367211808df99b4575fb901977d9f4347065c61a26642d65887f24d60342a4ec")
version("2.0.3-0", sha256="20da93e8508511e21aaab1863cb4c372a3bec02307b932144a7d757ea5a1bad2")
variant("cuda", default=True, description="Build with CUDA")
variant("ucx", default=True, description="Build with UCX support")
variant("nccl", default=True, description="Build with NCCL support")
variant("gdrcopy", default=True, description="Build with gdrcopy support")
variant("mpi", default=True, description="Build with MPI support")
variant("shmem", default=False, description="Build with shmem support")
variant(
"gpu_initiated_support",
default=False,
when="@2.6:",
description="Build with support for GPU initiated communication",
)
conflicts("~cuda")
def url_for_version(self, version):
ver_str = "{0}".format(version)
directory = ver_str.split("-")[0]
url_fmt = "https://developer.download.nvidia.com/compute/redist/nvshmem/{0}/source/nvshmem_src_{1}.txz"
return url_fmt.format(directory, version)
depends_on("mpi", when="+mpi")
depends_on("ucx", when="+ucx")
depends_on("gdrcopy", when="+gdrcopy")
depends_on("nccl", when="+nccl")
def setup_build_environment(self, env):
env.set("CUDA_HOME", self.spec["cuda"].prefix)
env.set("NVSHMEM_PREFIX", self.prefix)
if "+ucx" in self.spec:
env.set("NVSHMEM_UCX_SUPPORT", "1")
env.set("UCX_HOME", self.spec["ucx"].prefix)
if "+gdrcopy" in self.spec:
env.set("NVSHMEM_USE_GDRCOPY", "1")
env.set("GDRCOPY_HOME", self.spec["gdrcopy"].prefix)
if "+nccl" in self.spec:
env.set("NVSHMEM_USE_NCCL", "1")
env.set("NCCL_HOME", self.spec["nccl"].prefix)
if "+mpi" in self.spec:
env.set("NVSHMEM_MPI_SUPPORT", "1")
env.set("MPI_HOME", self.spec["mpi"].prefix)
if self.spec.satisfies("^spectrum-mpi") or self.spec.satisfies("^openmpi"):
env.set("NVSHMEM_MPI_IS_OMPI", "1")
else:
env.set("NVSHMEM_MPI_IS_OMPI", "0")
if "+shmem" in self.spec:
env.set("NVSHMEM_SHMEM_SUPPORT", "1")
env.set("SHMEM_HOME", self.spec["mpi"].prefix)
if "+gpu_initiated_support" in self.spec:
env.set("NVSHMEM_GPUINITIATED_SUPPORT", "1")
|
