ucc: CUDA fixups (#369956)

This commit is contained in:
Markus Kowalewski
2025-01-19 23:18:09 +01:00
committed by GitHub
+73 -35
View File
@@ -1,28 +1,52 @@
{
stdenv,
lib,
fetchFromGitHub,
libtool,
automake,
inputs@{
autoconf,
ucx,
automake,
config,
enableCuda ? config.cudaSupport,
cudaPackages,
fetchFromGitHub,
lib,
libtool,
stdenv,
ucx,
# Configuration options
enableAvx ? stdenv.hostPlatform.avxSupport,
enableCuda ? config.cudaSupport,
enableSse41 ? stdenv.hostPlatform.sse4_1Support,
enableSse42 ? stdenv.hostPlatform.sse4_2Support,
}:
let
inherit (lib.attrsets) getLib;
inherit (lib.lists) optionals;
inherit (lib.strings) concatStringsSep;
inherit (cudaPackages)
cuda_cccl
cuda_cudart
cuda_nvcc
cuda_nvml_dev
cudaFlags
nccl
;
stdenv = throw "Use effectiveStdenv instead";
effectiveStdenv = if enableCuda then cudaPackages.backendStdenv else inputs.stdenv;
in
effectiveStdenv.mkDerivation (finalAttrs: {
__structuredAttrs = true;
# TODO(@connorbaker):
# When strictDeps is enabled, `cuda_nvcc` is required as the argument to `--with-cuda` in `configureFlags` or else
# configurePhase fails with `checking for cuda_runtime.h... no`.
# This is odd, especially given `cuda_runtime.h` is provided by `cuda_cudart.dev`, which is already in `buildInputs`.
strictDeps = true;
stdenv.mkDerivation rec {
pname = "ucc";
version = "1.3.0";
src = fetchFromGitHub {
owner = "openucx";
repo = "ucc";
rev = "v${version}";
sha256 = "sha256-xcJLYktkxNK2ewWRgm8zH/dMaIoI+9JexuswXi7MpAU=";
tag = "v${finalAttrs.version}";
hash = "sha256-xcJLYktkxNK2ewWRgm8zH/dMaIoI+9JexuswXi7MpAU=";
};
outputs = [
@@ -32,44 +56,58 @@ stdenv.mkDerivation rec {
enableParallelBuilding = true;
# NOTE: We use --replace-quiet because not all Makefile.am files contain /bin/bash.
postPatch = ''
for comp in $(find src/components -name Makefile.am); do
substituteInPlace $comp \
--replace "/bin/bash" "${stdenv.shell}"
substituteInPlace "$comp" \
--replace-quiet \
"/bin/bash" \
"${effectiveStdenv.shell}"
done
'';
nativeBuildInputs = [
libtool
automake
autoconf
] ++ lib.optionals enableCuda [ cudaPackages.cuda_nvcc ];
automake
libtool
] ++ optionals enableCuda [ cuda_nvcc ];
buildInputs =
[ ucx ]
++ lib.optionals enableCuda [
cudaPackages.cuda_cccl
cudaPackages.cuda_cudart
++ optionals enableCuda [
cuda_cccl
cuda_cudart
cuda_nvml_dev
nccl
];
preConfigure =
''
./autogen.sh
''
+ lib.optionalString enableCuda ''
configureFlagsArray+=( "--with-nvcc-gencode=${builtins.concatStringsSep " " cudaPackages.cudaFlags.gencode}" )
'';
# NOTE: With `__structuredAttrs` enabled, `LDFLAGS` must be set under `env` so it is assured to be a string;
# otherwise, we might have forgotten to convert it to a string and Nix would make LDFLAGS a shell variable
# referring to an array!
env.LDFLAGS = builtins.toString (
optionals enableCuda [
# Fake libnvidia-ml.so (the real one is deployed impurely)
"-L${getLib cuda_nvml_dev}/lib/stubs"
]
);
preConfigure = ''
./autogen.sh
'';
configureFlags =
[ ]
++ lib.optional enableSse41 "--with-sse41"
++ lib.optional enableSse42 "--with-sse42"
++ lib.optional enableAvx "--with-avx"
++ lib.optional enableCuda "--with-cuda=${cudaPackages.cuda_cudart}";
optionals enableSse41 [ "--with-sse41" ]
++ optionals enableSse42 [ "--with-sse42" ]
++ optionals enableAvx [ "--with-avx" ]
++ optionals enableCuda [
"--with-cuda=${cuda_nvcc}"
"--with-nvcc-gencode=${concatStringsSep " " cudaFlags.gencode}"
];
postInstall = ''
find $out/lib/ -name "*.la" -exec rm -f \{} \;
find "$out/lib/" -name "*.la" -exec rm -f \{} \;
moveToOutput bin/ucc_info $dev
moveToOutput bin/ucc_info "$dev"
'';
meta = with lib; {
@@ -79,4 +117,4 @@ stdenv.mkDerivation rec {
maintainers = [ maintainers.markuskowa ];
platforms = platforms.linux;
};
}
})