nixpkgs/pkgs/development/libraries/science/math/magma/generic.nix

226 lines
8.4 KiB
Nix

# Type aliases
# Release = {
# version: String
# hash: String
# supportedGpuTargets: List String
# }
{ autoPatchelfHook
, blas
, cmake
, cudaPackages_11 ? null
, cudaPackages
, cudaSupport ? config.cudaSupport
, fetchurl
, gfortran
, gpuTargets ? [ ] # Non-CUDA targets, that is HIP
, rocmPackages_5
, lapack
, lib
, libpthreadstubs
, magmaRelease
, ninja
, python3
, config
# At least one back-end has to be enabled,
# and we can't default to CUDA since it's unfree
, rocmSupport ? !cudaSupport
, static ? stdenv.hostPlatform.isStatic
, stdenv
}:
let
inherit (lib) getLib lists strings trivial;
inherit (magmaRelease) version hash supportedGpuTargets;
# Per https://icl.utk.edu/magma/downloads, support for CUDA 12 wasn't added until 2.7.1.
# If we're building a version prior to that, use the latest release of the 11.x series.
effectiveCudaPackages =
if strings.versionOlder version "2.7.1"
then cudaPackages_11
else cudaPackages;
inherit (effectiveCudaPackages) cudaAtLeast flags cudaOlder;
# move to newer ROCm version once supported
rocmPackages = rocmPackages_5;
# NOTE: The lists.subtractLists function is perhaps a bit unintuitive. It subtracts the elements
# of the first list *from* the second list. That means:
# lists.subtractLists a b = b - a
# For ROCm
# NOTE: The hip.gpuTargets are prefixed with "gfx" instead of "sm" like flags.realArches.
# For some reason, Magma's CMakeLists.txt file does not handle the "gfx" prefix, so we must
# remove it.
rocmArches = lists.map (x: strings.removePrefix "gfx" x) rocmPackages.clr.gpuTargets;
supportedRocmArches = lists.intersectLists rocmArches supportedGpuTargets;
unsupportedRocmArches = lists.subtractLists supportedRocmArches rocmArches;
supportedCustomGpuTargets = lists.intersectLists gpuTargets supportedGpuTargets;
unsupportedCustomGpuTargets = lists.subtractLists supportedCustomGpuTargets gpuTargets;
# Use trivial.warnIf to print a warning if any unsupported GPU targets are specified.
gpuArchWarner = supported: unsupported:
trivial.throwIf (supported == [ ])
(
"No supported GPU targets specified. Requested GPU targets: "
+ strings.concatStringsSep ", " unsupported
)
supported;
gpuTargetString = strings.concatStringsSep "," (
if gpuTargets != [ ] then
# If gpuTargets is specified, it always takes priority.
gpuArchWarner supportedCustomGpuTargets unsupportedCustomGpuTargets
else if rocmSupport then
gpuArchWarner supportedRocmArches unsupportedRocmArches
else if cudaSupport then
[ ] # It's important we pass explicit -DGPU_TARGET to reset magma's defaults
else
throw "No GPU targets specified"
);
cudaArchitecturesString = flags.cmakeCudaArchitecturesString;
minArch =
let
# E.g. [ "80" "86" "90" ]
cudaArchitectures = (builtins.map flags.dropDot flags.cudaCapabilities);
minArch' = builtins.head (builtins.sort strings.versionOlder cudaArchitectures);
in
# "75" -> "750" Cf. https://bitbucket.org/icl/magma/src/f4ec79e2c13a2347eff8a77a3be6f83bc2daec20/CMakeLists.txt#lines-273
"${minArch'}0";
in
assert (builtins.match "[^[:space:]]*" gpuTargetString) != null;
stdenv.mkDerivation {
pname = "magma";
inherit version;
src = fetchurl {
name = "magma-${version}.tar.gz";
url = "https://icl.cs.utk.edu/projectsfiles/magma/downloads/magma-${version}.tar.gz";
inherit hash;
};
# Magma doesn't have anything which could be run under doCheck, but it does build test suite executables.
# These are moved to $test/bin/ and $test/lib/ in postInstall.
outputs = ["out" "test"];
# Fixup for the python test runners
postPatch = ''
patchShebangs ./testing/run_{tests,summarize}.py
substituteInPlace ./testing/run_tests.py \
--replace-fail \
"print >>sys.stderr, cmdp, \"doesn't exist (original name: \" + cmd + \", precision: \" + precision + \")\"" \
"print(f\"{cmdp} doesn't exist (original name: {cmd}, precision: {precision})\", file=sys.stderr)"
'';
nativeBuildInputs = [
autoPatchelfHook
cmake
ninja
gfortran
] ++ lists.optionals cudaSupport [
effectiveCudaPackages.cuda_nvcc
];
buildInputs = [
libpthreadstubs
lapack
blas
python3
(getLib gfortran.cc) # libgfortran.so
] ++ lists.optionals cudaSupport (with effectiveCudaPackages; [
cuda_cudart # cuda_runtime.h
libcublas # cublas_v2.h
libcusparse # cusparse.h
] ++ lists.optionals (cudaOlder "11.8") [
cuda_nvprof # <cuda_profiler_api.h>
] ++ lists.optionals (cudaAtLeast "11.8") [
cuda_profiler_api # <cuda_profiler_api.h>
] ++ lists.optionals (cudaAtLeast "12.0") [
cuda_cccl # <nv/target>
]) ++ lists.optionals rocmSupport [
rocmPackages.clr
rocmPackages.hipblas
rocmPackages.hipsparse
rocmPackages.llvm.openmp
];
cmakeFlags = [
(strings.cmakeFeature "GPU_TARGET" gpuTargetString)
(strings.cmakeBool "MAGMA_ENABLE_CUDA" cudaSupport)
(strings.cmakeBool "MAGMA_ENABLE_HIP" rocmSupport)
(strings.cmakeBool "BUILD_SHARED_LIBS" (!static))
# Set the Fortran name mangling scheme explicitly. We must set FORTRAN_CONVENTION manually because it will
# otherwise not be set in NVCC_FLAGS or DEVCCFLAGS (which we cannot modify).
# See https://github.com/NixOS/nixpkgs/issues/281656#issuecomment-1902931289
(strings.cmakeBool "USE_FORTRAN" true)
(strings.cmakeFeature "CMAKE_C_FLAGS" "-DADD_")
(strings.cmakeFeature "CMAKE_CXX_FLAGS" "-DADD_")
(strings.cmakeFeature "FORTRAN_CONVENTION" "-DADD_")
] ++ lists.optionals cudaSupport [
(strings.cmakeFeature "CMAKE_CUDA_ARCHITECTURES" cudaArchitecturesString)
(strings.cmakeFeature "MIN_ARCH" minArch) # Disarms magma's asserts
] ++ lists.optionals rocmSupport [
(strings.cmakeFeature "CMAKE_C_COMPILER" "${rocmPackages.clr}/bin/hipcc")
(strings.cmakeFeature "CMAKE_CXX_COMPILER" "${rocmPackages.clr}/bin/hipcc")
];
# Magma doesn't have a test suite we can easily run, just loose executables, all of which require a GPU.
doCheck = false;
# Copy the files to the test output and fix the RPATHs.
postInstall =
# NOTE: The python scripts aren't copied by CMake into the build directory, so we must copy them from the source.
# TODO(@connorbaker): This should be handled by having CMakeLists.txt install them, but such a patch is
# out of the scope of the PR which introduces the `test` output: https://github.com/NixOS/nixpkgs/pull/283777.
# See https://github.com/NixOS/nixpkgs/pull/283777#discussion_r1482125034 for more information.
# Such work is tracked by https://github.com/NixOS/nixpkgs/issues/296286.
''
install -Dm755 ../testing/run_{tests,summarize}.py -t "$test/bin/"
''
# Copy core test executables and libraries over to the test output.
# NOTE: Magma doesn't provide tests for sparse solvers for ROCm, but it does for CUDA -- we put them both in the same
# install command to avoid the case where a glob would fail to find any files and cause the install command to fail
# because it has no files to install.
+ ''
install -Dm755 ./testing/testing_* ./sparse/testing/testing_* -t "$test/bin/"
install -Dm755 ./lib/lib*test*.* -t "$test/lib/"
''
# All of the test executables and libraries will have a reference to the build directory in their RPATH, which we
# must remove. We do this by shrinking the RPATH to only include the Nix store. The autoPatchelfHook will take care
# of supplying the correct RPATH for needed libraries (like `libtester.so`).
+ ''
find "$test" -type f -exec \
patchelf \
--shrink-rpath \
--allowed-rpath-prefixes "$NIX_STORE" \
{} \;
'';
passthru = {
inherit cudaSupport rocmSupport gpuTargets;
cudaPackages = effectiveCudaPackages;
};
meta = with lib; {
description = "Matrix Algebra on GPU and Multicore Architectures";
license = licenses.bsd3;
homepage = "http://icl.cs.utk.edu/magma/index.html";
platforms = platforms.linux;
maintainers = with maintainers; [ connorbaker ];
# Cf. https://bitbucket.org/icl/magma/src/fcfe5aa61c1a4c664b36a73ebabbdbab82765e9f/CMakeLists.txt#lines-20
broken =
!(cudaSupport || rocmSupport) # At least one back-end enabled
|| (cudaSupport && rocmSupport) # Mutually exclusive
|| (cudaSupport && cudaOlder "9.0")
|| (cudaSupport && strings.versionOlder version "2.7.1" && cudaPackages_11 == null);
};
}