mirror of
https://github.com/NixOS/nixpkgs.git
synced 2024-11-15 14:26:33 +01:00
e3244e9ff0
The xxxFlagsArray variants were never meant to be used in nix code, at least they can't be used properly without __structuredAttrs turned on. If no spaces are passed in the argument the xxxFlagsArray can be replaced as-is with the non-Array variant. When whitespace needs to be passed, the derivation is additionally changed to enable __structuredAttrs.
115 lines
2.9 KiB
Nix
115 lines
2.9 KiB
Nix
# NOTE: Though NCCL is called within the cudaPackages package set, we avoid passing in
|
|
# the names of dependencies from that package set directly to avoid evaluation errors
|
|
# in the case redistributable packages are not available.
|
|
{
|
|
lib,
|
|
fetchFromGitHub,
|
|
python3,
|
|
which,
|
|
autoAddDriverRunpath,
|
|
cudaPackages,
|
|
# passthru.updateScript
|
|
gitUpdater,
|
|
}:
|
|
let
|
|
inherit (cudaPackages)
|
|
backendStdenv
|
|
cuda_cccl
|
|
cuda_cudart
|
|
cuda_nvcc
|
|
cudaAtLeast
|
|
cudaFlags
|
|
cudaOlder
|
|
cudatoolkit
|
|
;
|
|
in
|
|
backendStdenv.mkDerivation (finalAttrs: {
|
|
pname = "nccl";
|
|
version = "2.21.5-1";
|
|
|
|
src = fetchFromGitHub {
|
|
owner = "NVIDIA";
|
|
repo = "nccl";
|
|
rev = "v${finalAttrs.version}";
|
|
hash = "sha256-IF2tILwW8XnzSmfn7N1CO7jXL95gUp02guIW5n1eaig=";
|
|
};
|
|
|
|
__structuredAttrs = true;
|
|
strictDeps = true;
|
|
|
|
outputs = [
|
|
"out"
|
|
"dev"
|
|
];
|
|
|
|
nativeBuildInputs =
|
|
[
|
|
which
|
|
autoAddDriverRunpath
|
|
python3
|
|
]
|
|
++ lib.optionals (cudaOlder "11.4") [ cudatoolkit ]
|
|
++ lib.optionals (cudaAtLeast "11.4") [ cuda_nvcc ];
|
|
|
|
buildInputs =
|
|
lib.optionals (cudaOlder "11.4") [ cudatoolkit ]
|
|
++ lib.optionals (cudaAtLeast "11.4") [
|
|
cuda_nvcc # crt/host_config.h
|
|
cuda_cudart
|
|
]
|
|
# NOTE: CUDA versions in Nixpkgs only use a major and minor version. When we do comparisons
|
|
# against other version, like below, it's important that we use the same format. Otherwise,
|
|
# we'll get incorrect results.
|
|
# For example, lib.versionAtLeast "12.0" "12.0.0" == false.
|
|
++ lib.optionals (cudaAtLeast "12.0") [ cuda_cccl ];
|
|
|
|
env.NIX_CFLAGS_COMPILE = toString [ "-Wno-unused-function" ];
|
|
|
|
postPatch = ''
|
|
patchShebangs ./src/device/generate.py
|
|
'';
|
|
|
|
makeFlags =
|
|
[
|
|
"PREFIX=$(out)"
|
|
"NVCC_GENCODE=${cudaFlags.gencodeString}"
|
|
]
|
|
++ lib.optionals (cudaOlder "11.4") [
|
|
"CUDA_HOME=${cudatoolkit}"
|
|
"CUDA_LIB=${lib.getLib cudatoolkit}/lib"
|
|
"CUDA_INC=${lib.getDev cudatoolkit}/include"
|
|
]
|
|
++ lib.optionals (cudaAtLeast "11.4") [
|
|
"CUDA_HOME=${cuda_nvcc}"
|
|
"CUDA_LIB=${lib.getLib cuda_cudart}/lib"
|
|
"CUDA_INC=${lib.getDev cuda_cudart}/include"
|
|
];
|
|
|
|
enableParallelBuilding = true;
|
|
|
|
postFixup = ''
|
|
moveToOutput lib/libnccl_static.a $dev
|
|
'';
|
|
|
|
passthru.updateScript = gitUpdater {
|
|
inherit (finalAttrs) pname version;
|
|
rev-prefix = "v";
|
|
};
|
|
|
|
meta = with lib; {
|
|
description = "Multi-GPU and multi-node collective communication primitives for NVIDIA GPUs";
|
|
homepage = "https://developer.nvidia.com/nccl";
|
|
license = licenses.bsd3;
|
|
platforms = platforms.linux;
|
|
# NCCL is not supported on Jetson, because it does not use NVLink or PCI-e for inter-GPU communication.
|
|
# https://forums.developer.nvidia.com/t/can-jetson-orin-support-nccl/232845/9
|
|
badPlatforms = lib.optionals cudaFlags.isJetsonBuild [ "aarch64-linux" ];
|
|
maintainers =
|
|
with maintainers;
|
|
[
|
|
mdaiter
|
|
orivej
|
|
]
|
|
++ teams.cuda.members;
|
|
};
|
|
})
|