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
|
# 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 Resolve(CMakePackage, CudaPackage, ROCmPackage):
"""ReSolve is a library of GPU-resident sparse linear solvers. It contains iterative and direct
solvers designed to run on NVIDIA and AMD GPUs, as well as CPU devices."""
homepage = "https://github.com/ORNL/ReSolve"
git = "https://github.com/ORNL/ReSolve.git"
maintainers("cameronrutherford", "pelesh", "ryandanehy", "kswirydo")
version(
"0.99.1",
submodules=False,
tag="v0.99.1",
commit="e10dd417e836f47b3dc7c8b123a81bfbb654ee82",
)
version("develop", submodules=False, branch="develop")
variant("klu", default=True, description="Use KLU, AMD and COLAMD Libraries from SuiteSparse")
depends_on("suite-sparse", when="+klu")
with when("+rocm"):
# Need at least 5.6+
depends_on("rocsparse@5.6:")
depends_on("rocblas@5.6:")
depends_on("rocsolver@5.6:")
# Optional profiling dependecies
# Will be controlled by variant in the future
# depends_on("roctracer-dev@5.6:")
# depends_on("roctracer-dev-api@5.6:")
# depends_on("rocprofiler-dev@5.6:")
def cmake_args(self):
args = []
spec = self.spec
args.extend(
[self.define("RESOLVE_USE_KLU", "klu"), self.define("RESOLVE_TEST_WITH_BSUB", False)]
)
if "+cuda" in spec:
cuda_arch_list = spec.variants["cuda_arch"].value
if cuda_arch_list[0] != "none":
args.append(self.define("CMAKE_CUDA_ARCHITECTURES", cuda_arch_list))
else:
args.append(self.define("CMAKE_CUDA_ARCHITECTURES", "70;75;80"))
args.append(self.define("RESOLVE_USE_CUDA", True))
elif "+rocm" in spec:
rocm_arch_list = spec.variants["amdgpu_target"].value
# `+rocm` conflicts with amdgpu_target == "none"...
# if rocm_arch_list[0] == "none":
# rocm_arch_list = "gfx90a"
args.append(self.define("GPU_TARGETS", rocm_arch_list))
args.append(self.define("AMDGPU_TARGETS", rocm_arch_list))
args.append(self.define("RESOLVE_USE_HIP", True))
return args
|
