# Copyright Spack Project Developers. See COPYRIGHT file for details.
#
# SPDX-License-Identifier: (Apache-2.0 OR MIT)
import re
from typing import Iterable, List
from spack.package import PackageBase, any_combination_of, conflicts, depends_on, variant, when
[docs]
class CudaPackage(PackageBase):
"""Auxiliary class which contains CUDA variant, dependencies and conflicts
and is meant to unify and facilitate its usage.
Maintainers: ax3l, Rombur, davidbeckingsale, pauleonix
"""
# https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#gpu-feature-list
# https://developer.nvidia.com/cuda-gpus
# https://en.wikipedia.org/wiki/CUDA#GPUs_supported
cuda_arch_values = (
"10",
"11",
"12",
"13",
"20",
"21",
"30",
"32",
"35",
"37",
"50",
"52",
"53",
"60",
"61",
"62",
"70",
"72",
"75",
"80",
"86",
"87",
"89",
"90",
"90a",
"100",
"100a",
"100f",
"101",
"101a",
"101f",
"103",
"103a",
"103f",
"110",
"110a",
"110f",
"120",
"120a",
"120f",
"121",
"121a",
"121f",
)
# FIXME: keep cuda and cuda_arch separate to make usage easier until
# Spack has depends_on(cuda, when='cuda_arch!=None') or alike
variant("cuda", default=False, description="Build with CUDA")
variant(
"cuda_arch",
description="CUDA architecture",
values=any_combination_of(*cuda_arch_values),
sticky=True,
when="+cuda",
)
# https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#nvcc-examples
# https://llvm.org/docs/CompileCudaWithLLVM.html#compiling-cuda-code
[docs]
@staticmethod
def cuda_flags(arch_list):
return [
(
"--generate-code arch=compute_{0},code=sm_{0} "
"--generate-code arch=compute_{0},code=compute_{0}"
).format(s)
for s in arch_list
]
[docs]
@staticmethod
def compute_capabilities(arch_list: Iterable[str]) -> List[str]:
"""Adds a decimal place to each CUDA arch.
>>> compute_capabilities(['90', '90a', '100f'])
['9.0', '9.0a', '10.0f']
Args:
arch_list: A list of integer strings, optionally followed by a suffix.
Returns:
A list of float strings, optionally followed by a suffix
"""
pattern = re.compile(r"(\d+)")
capabilities = []
for arch in arch_list:
_, number, letter = re.split(pattern, arch)
number = "{0:.1f}".format(float(number) / 10.0)
capabilities.append(number + letter)
return capabilities
depends_on("cuda", when="+cuda")
# CUDA version vs Architecture
# https://en.wikipedia.org/wiki/CUDA#GPUs_supported
# https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#deprecated-features
# Tesla support:
depends_on("cuda@:6.0", when="cuda_arch=10")
depends_on("cuda@:6.5", when="cuda_arch=11")
depends_on("cuda@2.1:6.5", when="cuda_arch=12")
depends_on("cuda@2.1:6.5", when="cuda_arch=13")
# Fermi support:
depends_on("cuda@3.0:8.0", when="cuda_arch=20")
depends_on("cuda@3.2:8.0", when="cuda_arch=21")
# Kepler support:
depends_on("cuda@5.0:10.2", when="cuda_arch=30")
depends_on("cuda@5.0:10.2", when="cuda_arch=32")
depends_on("cuda@5.0:11.8", when="cuda_arch=35")
depends_on("cuda@6.5:11.8", when="cuda_arch=37")
# Maxwell support:
depends_on("cuda@6.0:12.9", when="cuda_arch=50")
depends_on("cuda@6.5:12.9", when="cuda_arch=52")
depends_on("cuda@6.5:12.9", when="cuda_arch=53")
# Pascal support:
depends_on("cuda@8.0:12.9", when="cuda_arch=60")
depends_on("cuda@8.0:12.9", when="cuda_arch=61")
depends_on("cuda@8.0:12.9", when="cuda_arch=62")
# Volta support:
depends_on("cuda@9.0:12.9", when="cuda_arch=70")
# Turing support:
depends_on("cuda@9.0:", when="cuda_arch=72")
depends_on("cuda@10.0:", when="cuda_arch=75")
# Ampere support:
depends_on("cuda@11.0:", when="cuda_arch=80")
depends_on("cuda@11.1:", when="cuda_arch=86")
depends_on("cuda@11.4:", when="cuda_arch=87")
# Ada support:
depends_on("cuda@11.8:", when="cuda_arch=89")
# Hopper support:
depends_on("cuda@12.0:", when="cuda_arch=90")
depends_on("cuda@12.0:", when="cuda_arch=90a")
# Blackwell support:
depends_on("cuda@12.8:", when="cuda_arch=100")
depends_on("cuda@12.8:", when="cuda_arch=100a")
depends_on("cuda@12.9:", when="cuda_arch=100f")
depends_on("cuda@12.9:", when="cuda_arch=103")
depends_on("cuda@12.9:", when="cuda_arch=103a")
depends_on("cuda@12.9:", when="cuda_arch=103f")
# Compute Capability 101 was renamed to 110 in CUDA 13
depends_on("cuda@12.8:12.9", when="cuda_arch=101")
depends_on("cuda@12.8:12.9", when="cuda_arch=101a")
depends_on("cuda@12.9", when="cuda_arch=101f")
depends_on("cuda@13.0:", when="cuda_arch=110")
depends_on("cuda@13.0:", when="cuda_arch=110a")
depends_on("cuda@13.0:", when="cuda_arch=110f")
depends_on("cuda@12.8:", when="cuda_arch=120")
depends_on("cuda@12.8:", when="cuda_arch=120a")
depends_on("cuda@12.9:", when="cuda_arch=120f")
depends_on("cuda@12.9:", when="cuda_arch=121")
depends_on("cuda@12.9:", when="cuda_arch=121a")
depends_on("cuda@12.9:", when="cuda_arch=121f")
# From the NVIDIA install guide we know of conflicts for particular
# platforms (linux, darwin), architectures (x86, powerpc) and compilers
# (gcc, clang). We don't restrict %gcc and %clang conflicts to
# platform=linux, since they may apply to platform=darwin. We currently
# do not provide conflicts for platform=darwin with %apple-clang.
# Linux x86_64 compiler conflicts from here:
# https://gist.github.com/ax3l/9489132
with when("+cuda %cuda~allow-unsupported-compilers"):
# GCC
# According to
# https://github.com/spack/spack/pull/25054#issuecomment-886531664
# these conflicts are valid independently from the architecture
# minimum supported versions
conflicts("%gcc@:4", when="%cuda@11.0:")
conflicts("%gcc@:5", when="%cuda@11.4:")
conflicts("%clang@:6", when="%cuda@12.2:")
# maximum supported version
# NOTE:
# in order to not constrain future cuda version to old gcc versions,
# it has been decided to use an upper bound for the latest version.
# This implies that the last one in the list has to be updated at
# each release of a new cuda minor version.
conflicts("%gcc@10:", when="%cuda@:11.0")
conflicts("%gcc@11:", when="%cuda@:11.4.0")
conflicts("%gcc@11.2:", when="%cuda@:11.5")
conflicts("%gcc@12:", when="%cuda@:11.8")
conflicts("%gcc@13:", when="%cuda@:12.3")
conflicts("%gcc@14:", when="%cuda@:12.6")
conflicts("%gcc@15:", when="%cuda@:12.9")
conflicts("%gcc@16:", when="%cuda@:13.1")
conflicts("%gcc@15:", when="%cuda@13.1:")
conflicts("%clang@12:", when="%cuda@:11.4.0")
conflicts("%clang@13:", when="%cuda@:11.5")
conflicts("%clang@14:", when="%cuda@:11.7")
conflicts("%clang@15:", when="%cuda@:12.0")
conflicts("%clang@16:", when="%cuda@:12.1")
conflicts("%clang@17:", when="%cuda@:12.3")
conflicts("%clang@18:", when="%cuda@:12.5")
conflicts("%clang@19:", when="%cuda@:12.6")
conflicts("%clang@20:", when="%cuda@:12.9")
conflicts("%clang@21:", when="%cuda@:13.1")
conflicts("%clang@20:", when="%cuda@13.1:")
# https://gist.github.com/ax3l/9489132#gistcomment-3860114
conflicts("%gcc@10", when="%cuda@:11.4.0")
conflicts("%gcc@5:", when="%cuda@:7.5 target=x86_64:")
conflicts("%gcc@6:", when="%cuda@:8 target=x86_64:")
conflicts("%gcc@7:", when="%cuda@:9.1 target=x86_64:")
conflicts("%gcc@8:", when="%cuda@:10.0.130 target=x86_64:")
conflicts("%gcc@9:", when="%cuda@:10.2.89 target=x86_64:")
conflicts("%clang@:3.4", when="%cuda@:7.5 target=x86_64:")
conflicts("%clang@:3.7,4:", when="%cuda@8.0:9.0 target=x86_64:")
conflicts("%clang@:3.7,4.1:", when="%cuda@9.1 target=x86_64:")
conflicts("%clang@:3.7,5.1:", when="%cuda@9.2 target=x86_64:")
conflicts("%clang@:3.7,6.1:", when="%cuda@10.0.130 target=x86_64:")
conflicts("%clang@:3.7,7.1:", when="%cuda@10.1.105 target=x86_64:")
conflicts("%clang@:3.7,8.1:", when="%cuda@10.1.105:10.1.243 target=x86_64:")
conflicts("%clang@:3.2,9:", when="%cuda@10.2.89 target=x86_64:")
conflicts("%clang@:5", when="%cuda@11.0.2: target=x86_64:")
conflicts("%clang@10:", when="%cuda@:11.0.3 target=x86_64:")
conflicts("%clang@11:", when="%cuda@:11.1.0 target=x86_64:")
# x86_64 vs. ppc64le differ according to NVidia docs
# Linux ppc64le compiler conflicts from Table from the docs below:
# https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html
# https://docs.nvidia.com/cuda/archive/9.2/cuda-installation-guide-linux/index.html
# https://docs.nvidia.com/cuda/archive/9.1/cuda-installation-guide-linux/index.html
# https://docs.nvidia.com/cuda/archive/9.0/cuda-installation-guide-linux/index.html
# https://docs.nvidia.com/cuda/archive/8.0/cuda-installation-guide-linux/index.html
# information prior to CUDA 9 difficult to find
conflicts("%gcc@6:", when="%cuda@:9 target=ppc64le:")
conflicts("%gcc@8:", when="%cuda@:10.0.130 target=ppc64le:")
conflicts("%gcc@9:", when="%cuda@:10.1.243 target=ppc64le:")
# officially, CUDA 11.0.2 only supports the system GCC 8.3 on ppc64le
conflicts("%clang@4:", when="%cuda@:9.0.176 target=ppc64le:")
conflicts("%clang@5:", when="%cuda@:9.1 target=ppc64le:")
conflicts("%clang@6:", when="%cuda@:9.2 target=ppc64le:")
conflicts("%clang@7:", when="%cuda@10.0.130 target=ppc64le:")
conflicts("%clang@7.1:", when="%cuda@:10.1.105 target=ppc64le:")
conflicts("%clang@8.1:", when="%cuda@:10.2.89 target=ppc64le:")
conflicts("%clang@:5", when="%cuda@11.0.2: target=ppc64le:")
conflicts("%clang@10:", when="%cuda@:11.0.2 target=ppc64le:")
conflicts("%clang@11:", when="%cuda@:11.1.0 target=ppc64le:")
# Intel is mostly relevant for x86_64 Linux, even though it also
# exists for Mac OS X. No information prior to CUDA 3.2 or Intel 11.1
conflicts("%intel@:11.0", when="%cuda@:3.1")
conflicts("%intel@:12.0", when="%cuda@5.5:")
conflicts("%intel@:13.0", when="%cuda@6.0:")
conflicts("%intel@:13.2", when="%cuda@6.5:")
conflicts("%intel@:14.9", when="%cuda@7:")
# Intel 15.x is compatible with CUDA 7 thru current CUDA
conflicts("%intel@16.0:", when="%cuda@:8.0.43")
conflicts("%intel@17.0:", when="%cuda@:8.0.60")
conflicts("%intel@18.0:", when="%cuda@:9.9")
conflicts("%intel@19.0:", when="%cuda@:10.0")
conflicts("%intel@19.1:", when="%cuda@:10.1")
conflicts("%intel@19.2:", when="%cuda@:11.1.0")
conflicts("%intel@2021:", when="%cuda@:11.4.0")
conflicts("%intel", when="%cuda@13.0:")
# ARM
# https://github.com/spack/spack/pull/39666#issuecomment-2377609263
# Might need to be expanded to other gcc versions
conflicts("%gcc@13.2.0", when="%cuda@:12.4 target=aarch64:")
# XL is mostly relevant for ppc64le Linux
conflicts("%xl@:12,14:", when="%cuda@:9.1")
conflicts("%xl@:12,14:15,17:", when="%cuda@9.2")
conflicts("%xl@:12,17:", when="%cuda@:11.1.0")
# PowerPC.
conflicts("target=ppc64le", when="%cuda@12.5:")
# Darwin.
# TODO: add missing conflicts for %apple-clang cuda@:10
conflicts("platform=darwin", when="%cuda@11.0.2:")