Compare commits

..

10 Commits

28 changed files with 43920 additions and 951 deletions

View File

@@ -19,11 +19,7 @@ let
cudainfo = prev.callPackage ./pkgs/cudainfo/default.nix { };
#extrae = callPackage ./pkgs/extrae/default.nix { }; # Broken and outdated
gpi-2 = callPackage ./pkgs/gpi-2/default.nix { };
intel-apt = callPackage ./pkgs/intel-oneapi/packages.nix { };
intelPackages_2023 = callPackage ./pkgs/intel-oneapi/2023.nix { };
intelPackages_2024 = final.intel-apt.hpckit_2024;
intelPackages_2025 = final.intel-apt.hpckit_2025;
intelPackages = final.intelPackages_2025;
jemallocNanos6 = callPackage ./pkgs/nanos6/jemalloc.nix { };
# FIXME: Extend this to all linuxPackages variants. Open problem, see:
# https://discourse.nixos.org/t/whats-the-right-way-to-make-a-custom-kernel-module-available/4636
@@ -48,13 +44,14 @@ let
#nix-wrap = callPackage ./pkgs/nix-wrap/default.nix { };
nodes = callPackage ./pkgs/nodes/default.nix { };
nosv = callPackage ./pkgs/nosv/default.nix { };
oneMath = callPackage ./pkgs/onemath/default.nix { };
openmp = callPackage ./pkgs/llvm-ompss2/openmp.nix { monorepoSrc = final.clangOmpss2Unwrapped.src; version = final.clangOmpss2Unwrapped.version; };
openmpv = final.openmp.override { enableNosv = true; enableOvni = true; };
osumb = callPackage ./pkgs/osu/default.nix { };
ovni = callPackage ./pkgs/ovni/default.nix { };
ovniGit = final.ovni.override { useGit = true; };
paraverKernel = callPackage ./pkgs/paraver/kernel.nix { };
pocl = callPackage ./pkgs/pocl/default.nix { };
pocl-noicd = callPackage ./pkgs/pocl/default.nix { enableICD = false; };
prometheus-slurm-exporter = prev.callPackage ./pkgs/slurm-exporter/default.nix { };
#pscom = callPackage ./pkgs/parastation/pscom.nix { }; # Unmaintaned
#psmpi = callPackage ./pkgs/parastation/psmpi.nix { }; # Unmaintaned
@@ -66,10 +63,17 @@ let
stdenvClangOmpss2NodesOmpv = final.stdenv.override { cc = final.clangOmpss2NodesOmpv; allowedRequisites = null; };
tagaspi = callPackage ./pkgs/tagaspi/default.nix { };
tampi = callPackage ./pkgs/tampi/default.nix { };
tasycl = callPackage ./pkgs/tasycl/default.nix { };
tasycl-acpp = callPackage ./pkgs/tasycl/default.nix { useIntel = false; };
upc-qaire-exporter = prev.callPackage ./pkgs/upc-qaire-exporter/default.nix { };
taopencl = callPackage ./pkgs/taopencl/default.nix { };
wxparaver = callPackage ./pkgs/paraver/default.nix { };
gromacs = callPackage ./pkgs/gromacs/default.nix { enableSYCL = true; };
_cuda = prev._cuda.extend (_: _prev: final.lib.recursiveUpdate _prev {
extensions = _prev.extensions ++ [(finalAttrs: _: {
tacuda = finalAttrs.callPackage ./pkgs/tacuda/default.nix { };
})];
});
};
tests = rec {
@@ -77,8 +81,6 @@ let
#sigsegv = callPackage ./test/reproducers/sigsegv.nix { };
hello-c = callPackage ./test/compilers/hello-c.nix { };
hello-cpp = callPackage ./test/compilers/hello-cpp.nix { };
hello-sycl = callPackage ./test/compilers/hello-sycl.nix { };
hello-syclompss = callPackage ./test/compilers/icpx-ompss2.nix { };
lto = callPackage ./test/compilers/lto.nix { };
asan = callPackage ./test/compilers/asan.nix { };
intel2023-icx-c = hello-c.override { stdenv = final.intelPackages_2023.stdenv; };
@@ -88,13 +90,6 @@ let
intel2023-ifort = callPackage ./test/compilers/hello-f.nix {
stdenv = final.intelPackages_2023.stdenv-ifort;
};
intel2024-icx-c = hello-c.override { stdenv = final.intelPackages_2024.stdenv; };
intel2025-icx-c = hello-c.override { stdenv = final.intelPackages_2025.stdenv; };
intel2024-icx-cpp = hello-cpp.override { stdenv = final.intelPackages_2024.stdenv; };
intel2025-icx-cpp = hello-cpp.override { stdenv = final.intelPackages_2025.stdenv; };
# intel2023-sycl = hello-sycl.override { intelPackages = final.intelPackages_2023; }; # broken
intel2024-sycl = hello-sycl.override { intelPackages = final.intelPackages_2024; };
intel2025-sycl = hello-sycl.override { intelPackages = final.intelPackages_2025; };
clangOmpss2-lto = lto.override { stdenv = final.stdenvClangOmpss2Nanos6; };
clangOmpss2-asan = asan.override { stdenv = final.stdenvClangOmpss2Nanos6; };
clangOmpss2-task = callPackage ./test/compilers/ompss2.nix {
@@ -112,6 +107,13 @@ let
clangNosvOmpv-ld = callPackage ./test/compilers/clang-openmp-ld.nix {
stdenv = final.stdenvClangOmpss2NodesOmpv;
};
ocl-build = callPackage ./test/compilers/opencl.nix { };
ocl-build-noicd = callPackage ./test/compilers/opencl.nix { ocl-icd = final.pocl-noicd; };
pocl = callPackage ./test/compilers/pocl.nix { };
ocl-run-pocl = ocl-build.test-icd;
ocl-run-pocl-noicd = ocl-build-noicd.test-noicd;
};
# For now, only build toplevel packages in CI/Hydra

190
pkgs/gromacs/default.nix Normal file
View File

@@ -0,0 +1,190 @@
{
lib,
stdenv,
fetchurl,
cmake,
hwloc,
fftw,
perl,
blas,
lapack,
llvmPackages,
intelPackages ? null,
mpi,
cudaPackages,
plumed,
singlePrec ? true,
config,
enableSYCL ? false,
enableCuda ? config.cudaSupport,
enableMpi ? false,
enablePlumed ? false,
cpuAcceleration ? null,
}:
# CUDA is only implemented for single precission
assert enableCuda -> singlePrec;
assert !(enableSYCL && enableCuda);
assert enableSYCL -> intelPackages != null;
let
inherit (cudaPackages.flags) cmakeCudaArchitecturesString;
# Select reasonable defaults for all major platforms
# The possible values are defined in CMakeLists.txt:
# AUTO None SSE2 SSE4.1 AVX_128_FMA AVX_256 AVX2_256
# AVX2_128 AVX_512 AVX_512_KNL MIC ARM_NEON ARM_NEON_ASIMD
SIMD =
x:
if (cpuAcceleration != null) then
x
else if stdenv.hostPlatform.system == "i686-linux" then
"SSE2"
else if stdenv.hostPlatform.system == "x86_64-linux" then
"SSE4.1"
else if stdenv.hostPlatform.system == "x86_64-darwin" then
"SSE4.1"
else if stdenv.hostPlatform.system == "aarch64-linux" then
"ARM_NEON_ASIMD"
else
"None";
source =
if enablePlumed then
{
version = "2024.2";
hash = "sha256-gCp+M18uiVdw9XsVnk7DaOuw/yzm2sz3BsboAlw2hSs=";
}
else
{
version = "2025.3";
hash = "sha256-i9/KAmjz8Qp8o8BuWbYvc+oCQgxnIRwP85EvMteDPGU=";
};
stdenv' = if enableSYCL then intelPackages.stdenv else stdenv;
in
stdenv'.mkDerivation rec {
pname = "gromacs";
version = source.version;
src = fetchurl {
url = "ftp://ftp.gromacs.org/pub/gromacs/gromacs-${version}.tar.gz";
inherit (source) hash;
};
patches = [ (if enablePlumed then ./pkgconfig-2024.patch else ./pkgconfig-2025.patch) ];
postPatch = lib.optionalString enablePlumed ''
plumed patch -p -e gromacs-${source.version}
'';
outputs = [
"out"
"dev"
"man"
];
nativeBuildInputs = [
cmake
]
++ lib.optional enablePlumed plumed
++ lib.optionals enableCuda [ cudaPackages.cuda_nvcc ];
env.MKLROOT = intelPackages.mkl;
buildInputs = [
fftw
perl
hwloc
blas
lapack
]
++ lib.optional enableMpi mpi
++ lib.optionals enableCuda [
cudaPackages.cuda_cccl
cudaPackages.cuda_cudart
cudaPackages.libcufft
cudaPackages.cuda_profiler_api
]
++ lib.optional stdenv.hostPlatform.isDarwin llvmPackages.openmp;
propagatedBuildInputs = lib.optional enableMpi mpi;
propagatedUserEnvPkgs = lib.optional enableMpi mpi;
cmakeFlags = [
(lib.cmakeBool "GMX_HWLOC" true)
"-DGMX_SIMD:STRING=${SIMD cpuAcceleration}"
"-DGMX_OPENMP:BOOL=TRUE"
"-DBUILD_SHARED_LIBS=ON"
]
++ (
if singlePrec then
[
"-DGMX_DOUBLE=OFF"
]
else
[
"-DGMX_DOUBLE=ON"
"-DGMX_DEFAULT_SUFFIX=OFF"
]
)
++ (
if enableMpi then
[
"-DGMX_MPI:BOOL=TRUE"
"-DGMX_THREAD_MPI:BOOL=FALSE"
]
else
[
"-DGMX_MPI:BOOL=FALSE"
]
)
++ (lib.optionals enableSYCL [
"-DGMX_GPU=SYCL"
"-DGMX_OPENMP=OFF" # TODO: enable OpenMP with SYCL?
])
++ lib.optionals enableCuda [
"-DGMX_GPU=CUDA"
(lib.cmakeFeature "CMAKE_CUDA_ARCHITECTURES" cmakeCudaArchitecturesString)
# Gromacs seems to ignore and override the normal variables, so we add this ad hoc:
(lib.cmakeFeature "GMX_CUDA_TARGET_COMPUTE" cmakeCudaArchitecturesString)
];
postInstall = ''
moveToOutput share/cmake $dev
'';
meta = with lib; {
homepage = "https://www.gromacs.org";
license = licenses.lgpl21Plus;
description = "Molecular dynamics software package";
longDescription = ''
GROMACS is a versatile package to perform molecular dynamics,
i.e. simulate the Newtonian equations of motion for systems
with hundreds to millions of particles.
It is primarily designed for biochemical molecules like
proteins, lipids and nucleic acids that have a lot of
complicated bonded interactions, but since GROMACS is
extremely fast at calculating the nonbonded interactions (that
usually dominate simulations) many groups are also using it
for research on non-biological systems, e.g. polymers.
GROMACS supports all the usual algorithms you expect from a
modern molecular dynamics implementation, (check the online
reference or manual for details), but there are also quite a
few features that make it stand out from the competition.
See: https://www.gromacs.org/about.html for details.
'';
platforms = platforms.unix;
maintainers = with maintainers; [
sheepforce
markuskowa
];
};
}

View File

@@ -0,0 +1,24 @@
diff --git a/src/external/muparser/muparser.pc.in b/src/external/muparser/muparser.pc.in
index 646787cb53..9b97ad57f7 100644
--- a/src/external/muparser/muparser.pc.in
+++ b/src/external/muparser/muparser.pc.in
@@ -1,7 +1,5 @@
-prefix=@CMAKE_INSTALL_PREFIX@
-exec_prefix=${prefix}
-libdir=${prefix}/@CMAKE_INSTALL_LIBDIR@
-includedir=${prefix}/@CMAKE_INSTALL_INCLUDEDIR@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
Name: @PACKAGE_NAME@
Description: Mathematical expressions parser library
diff --git a/src/gromacs/libgromacs.pc.cmakein b/src/gromacs/libgromacs.pc.cmakein
index ec1ed6684e..ca1105474a 100644
--- a/src/gromacs/libgromacs.pc.cmakein
+++ b/src/gromacs/libgromacs.pc.cmakein
@@ -1,4 +1,4 @@
-libdir=@CMAKE_INSTALL_PREFIX@/@CMAKE_INSTALL_LIBDIR@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
Name: libgromacs@GMX_LIBS_SUFFIX@
Description: Gromacs library

View File

@@ -0,0 +1,36 @@
diff --git a/src/external/muparser/muparser.pc.in b/src/external/muparser/muparser.pc.in
index 646787cb53..d26e84de8f 100644
--- a/src/external/muparser/muparser.pc.in
+++ b/src/external/muparser/muparser.pc.in
@@ -1,11 +1,9 @@
-prefix=@CMAKE_INSTALL_PREFIX@
-exec_prefix=${prefix}
-libdir=${prefix}/@CMAKE_INSTALL_LIBDIR@
-includedir=${prefix}/@CMAKE_INSTALL_INCLUDEDIR@
-
-Name: @PACKAGE_NAME@
-Description: Mathematical expressions parser library
-Version: @MUPARSER_VERSION@
-Requires:
-Libs: -L${libdir} -lmuparser
-Cflags: -I${includedir} @PKG_CONFIG_FLAGS@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
+
+Name: @PACKAGE_NAME@
+Description: Mathematical expressions parser library
+Version: @MUPARSER_VERSION@
+Requires:
+Libs: -L${libdir} -lmuparser
+Cflags: -I${includedir} @PKG_CONFIG_FLAGS@
diff --git a/src/gromacs/libgromacs.pc.cmakein b/src/gromacs/libgromacs.pc.cmakein
index af9b5a6dc0..5f58d549bf 100644
--- a/src/gromacs/libgromacs.pc.cmakein
+++ b/src/gromacs/libgromacs.pc.cmakein
@@ -1,5 +1,4 @@
-prefix=@CMAKE_INSTALL_PREFIX@
-libdir=${prefix}/@CMAKE_INSTALL_LIBDIR@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
Name: libgromacs@GMX_LIBS_SUFFIX@
Description: Gromacs library

View File

@@ -40,19 +40,45 @@ let
mpi = "2021.9.0";
};
findMatch = name:
let
aptPackages = builtins.fromJSON (builtins.readFile ./packages.json);
matches = lib.filter (x: name == x.pname) aptPackages;
n = lib.length matches;
match = builtins.traceVerbose (name + " -- ${builtins.toString n}") (builtins.head matches);
aptPackageIndex = stdenv.mkDerivation {
name = "intel-oneapi-packages";
srcs = [
# Run update.sh to update the package lists
./amd64-packages ./all-packages
];
phases = [ "installPhase" ];
installPhase = ''
awk -F': ' '\
BEGIN { print "[ {" } \
NR>1 && /^Package: / { print "} {"; } \
/: / { printf "%s = \"%s\";\n", $1, $2 } \
END { print "} ]" }' $srcs > $out
'';
};
apthost = "https://apt.repos.intel.com/oneapi/";
aptPackages = import aptPackageIndex;
apthost = "https://apt.repos.intel.com/oneapi/";
getSum = pkgList: name:
let
matches = lib.filter (x: name == x.Package) pkgList;
#n = lib.length matches;
#match = builtins.trace (name + " -- ${builtins.toString n}") (lib.elemAt matches 0);
match = lib.elemAt matches 0;
in
{
url = apthost + match.filename;
sha256 = match.sha256;
};
match.SHA256;
getUrl = pkgList: name:
let
matches = lib.filter (x: name == x.Package) pkgList;
#match = assert lib.length matches == 1; lib.elemAt matches 0;
n = lib.length matches;
match =
#builtins.trace (name + " -- n=${builtins.toString n}")
(lib.elemAt matches 0);
in
apthost + match.Filename;
uncompressDebs = debs: name: stdenv.mkDerivation {
name = name;
@@ -72,7 +98,10 @@ let
joinDebs = name: names:
let
debs = builtins.map (x: builtins.fetchurl (findMatch x)) names;
urls = builtins.map (x: getUrl aptPackages x) names;
sums = builtins.map (x: getSum aptPackages x) names;
getsrc = url: sha256: builtins.fetchurl { inherit url sha256; };
debs = lib.zipListsWith getsrc urls sums;
in
uncompressDebs debs "${name}-source";
@@ -441,7 +470,7 @@ let
'';
};
ifort-wrapper = wrapIntel {
ifort-wrapper = wrapIntel rec {
cc = intel-compiler-fortran;
mygcc = gcc;
extraBuild = ''

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

View File

@@ -1,297 +0,0 @@
{
lib,
stdenv,
callPackage,
dpkg,
fetchurl,
sqlite,
elfutils,
}:
let
inherit (builtins)
attrNames
attrValues
concatMap
elem
filter
fromJSON
getAttr
groupBy
head
isNull
listToAttrs
map
mapAttrs
readFile
replaceStrings
splitVersion
;
inherit (lib)
converge
findFirst
groupBy'
hasPrefix
optional
pipe
take
toInt
toList
versionAtLeast
versionOlder
;
aptData = fromJSON (readFile ./packages.json);
# Compare versions in debian control file syntax
# See: https://www.debian.org/doc/debian-policy/ch-relationships.html#syntax-of-relationship-fields
#
# NOTE: this is not a proper version comparison
#
# A proper version solver, should aggregate dependencies with the same name
# and compute the constraint (e.g. a (>= 2) a (<< 5) -> 2 <= a << 5)
#
# But in the intel repo, there are no such "duplicated" dependencies to specify
# upper limits, which leads to issues when intel-hpckit-2021 depends on things
# like intel-basekit >= 2021.1.0-2403 and we end up installing the newest
# basekit instead of the one from 2021.
#
# To mitigate this, >= is set to take the latest version with matching major
# and minor (only revision and patch are allowed to change)
compareVersions =
got: kind: want:
let
g0 = take 2 (splitVersion got);
w0 = take 2 (splitVersion want);
in
if isNull want then
true
else if kind == "=" then
got == want
else if kind == "<<" then
versionOlder got want
else if kind == "<=" then
versionAtLeast want got
else if kind == ">>" then
versionOlder want got
else if kind == ">=" then
(g0 == w0) && versionAtLeast got want # always match major version
else
throw "unknown operation: ${kind}";
findMatching =
{
pname,
kind,
version,
}:
findFirst (x: pname == x.pname && compareVersions x.version kind version) null aptData;
isIntel = pkg: (hasPrefix "intel-" pkg.pname);
expandDeps =
pkg: (map findMatching (filter isIntel pkg.dependencies)) ++ (optional (pkg.size != 0) pkg);
# get the oldest by major version. If they have the same major version, take
# the newest. This prevents most issues with resolutions
# versionOlder b a -> true if b is older than a (b `older` a)
getNewerInMajor =
a: b:
let
va = a.version;
vb = b.version;
va0 = head (splitVersion va);
vb0 = head (splitVersion vb);
in
if isNull a then
b
else if va0 != vb0 then
if va0 > vb0 then b else a
else if versionOlder vb va then
a
else
b;
removeDups = l: attrValues (groupBy' getNewerInMajor null (getAttr "provides") l);
_resolveDeps = converge (l: removeDups (concatMap expandDeps l));
resolveDeps =
pkg:
let
deps = _resolveDeps (toList pkg);
namedDeps = (map (x: "${x.pname}-${x.version}") deps);
in
builtins.traceVerbose (builtins.deepSeq namedDeps namedDeps) deps;
blacklist = [
"intel-basekit-env"
"intel-basekit-getting-started"
"intel-hpckit-env"
"intel-hpckit-getting-started"
"intel-oneapi-advisor"
"intel-oneapi-common-licensing"
"intel-oneapi-common-oneapi-vars"
"intel-oneapi-common-vars"
"intel-oneapi-compiler-cpp-eclipse-cfg"
"intel-oneapi-compiler-dpcpp-eclipse-cfg"
"intel-oneapi-condaindex"
"intel-oneapi-dev-utilities-eclipse-cfg"
"intel-oneapi-dpcpp-ct-eclipse-cfg"
"intel-oneapi-eclipse-ide"
"intel-oneapi-hpc-toolkit-getting-started"
"intel-oneapi-icc-eclipse-plugin-cpp"
"intel-oneapi-vtune"
"intel-oneapi-vtune-eclipse-plugin-vtune"
];
isInBlacklist = pkg: elem pkg.provides blacklist;
removeBlacklist = filter (e: !(isInBlacklist e));
dpkgExtractAll =
pname: version:
{ srcs, deps }:
stdenv.mkDerivation {
inherit pname version srcs;
nativeBuildInputs = [ dpkg ];
phases = [ "installPhase" ];
passthru = { inherit deps; };
installPhase = ''
mkdir -p $out
for src in $srcs; do
echo "Unpacking $src"
dpkg -x $src $out
done
'';
};
apthost = "https://apt.repos.intel.com/oneapi/";
fetchDeb =
p:
fetchurl {
url = apthost + p.filename;
inherit (p) sha256;
};
buildIntel =
pkg:
pipe pkg [
resolveDeps
removeBlacklist
(l: {
srcs = map fetchDeb l;
deps = l;
})
(dpkgExtractAll "${pkg.provides}-extracted" pkg.version)
];
findHpcKit =
year:
findMatching {
pname = "intel-hpckit";
kind = "<<";
version = toString (year + 1);
};
years = map toInt (attrNames components);
patchIntel = callPackage ./patch_intel.nix { };
# Version information for each hpckit. This is used to normalize the paths
# so that files are in $out/{bin,lib,include...} instead of all over the place
# in $out/opt/intel/oneapi/*/*/{...}.
#
# The most important is the compiler component, which is used to build the
# stdenv for the hpckit.
#
# NOTE: this have to be manually specified, so we can avoid IFD. To add a
# new version, add a new field with an empty attrset, (e.g. "2026" = {}; ),
# build hpckit_2026.unpatched and use the values from
# result/opt/intel/oneapi/* to populate the attrset.
#
# WARN: if there are more than one version in the folders of the unpatched
# components, our dependency resolution hacks have probably failed and the
# package set may be broken.
components = {
"2025" = {
ishmem = "1.4";
pti = "0.13";
tcm = "1.4";
umf = "0.11";
ccl = "2021.16";
compiler = "2025.2";
dal = "2025.8";
debugger = "2025.2";
dev-utilities = "2025.2";
dnnl = "2025.2";
dpcpp-ct = "2025.2";
dpl = "2022.9";
ipp = "2022.2";
ippcp = "2025.2";
mkl = "2025.2";
mpi = "2021.16";
tbb = "2022.2";
};
"2024" = {
tcm = "1.1";
ccl = "2021.13";
compiler = "2024.2";
dal = "2024.6";
debugger = "2024.2";
dev-utilities = "2024.2";
diagnostics = "2024.2";
dnnl = "2024.2";
dpcpp-ct = "2024.2";
dpl = "2022.6";
ipp = "2021.12";
ippcp = "2021.12";
mkl = "2024.2";
mpi = "2021.13";
tbb = "2021.13";
extraPackages = [
sqlite
elfutils
];
};
};
replaceDots = replaceStrings [ "." ] [ "_" ];
in
lib.recurseIntoAttrs (
listToAttrs (
map (
year:
let
year_str = toString year;
in
{
name = "hpckit_${year_str}";
value = patchIntel {
unpatched = buildIntel (findHpcKit year);
components = components.${year_str};
};
}
) years
)
)
// {
apt = pipe aptData [
(groupBy (p: replaceDots p.provides))
(mapAttrs (
_: l:
listToAttrs (
map (pkg: {
name = replaceDots ("v" + pkg.version);
value = pkg;
}) l
)
))
];
inherit resolveDeps patchIntel buildIntel;
}

View File

@@ -1,189 +0,0 @@
{
stdenv,
stdenvNoCC,
lib,
symlinkJoin,
autoPatchelfHook,
wrapCCWith,
overrideCC,
gcc,
hwloc,
libelf,
libffi_3_3,
libpsm2,
libuuid,
libxml2,
numactl,
ocl-icd,
openssl,
python3,
rdma-core,
ucx,
zlib,
}:
lib.makeOverridable (
{
unpatched,
components ? { },
extraPackages ? components.extraPackages or [ ],
}:
let
inherit (builtins)
attrValues
filter
mapAttrs
removeAttrs
;
__components = removeAttrs components [ "extraPackages" ];
_components = __components;
# _components = lib.traceSeqN 2 {
# inherit unpatched __components;
# deps = builtins.map (x: "${x.pname}-${x.version}") unpatched.deps;
# } __components;
wrapIntel =
cc:
let
targetConfig = stdenv.targetPlatform.config;
in
(wrapCCWith {
inherit cc;
nixSupport = {
cc-ldflags = [
"-L${gcc.cc}/lib/gcc/${targetConfig}/${gcc.version}"
"-L${gcc.cc.lib}/lib"
"-L${cc}/lib"
];
cc-cflags = [
"--gcc-toolchain=${gcc.cc}"
"-isystem \"${cc.original}/lib/clang/*/include\""
"-isystem ${cc}/include"
"-isystem ${cc}/include/intel64"
"-isystem ${gcc.cc}/lib/gcc/${targetConfig}/${gcc.version}/include"
];
libcxx-cxxflags = [
# "--gcc-toolchain=${gcc.cc}"
"-isystem ${gcc.cc}/include/c++/${gcc.version}"
"-isystem ${gcc.cc}/include/c++/${gcc.version}/${targetConfig}"
];
};
extraBuildCommands = ''
# FIXME: We should find a better way to modify the PATH instead of using
# this ugly hack. See https://jungle.bsc.es/git/rarias/bscpkgs/issues/9
echo 'path_backup="${gcc.cc}/bin:$path_backup"' >>$out/nix-support/cc-wrapper-hook
# Disable hardening by default
echo "" > $out/nix-support/add-hardening.sh
wrap icx $wrapper $ccPath/icx
wrap icpx $wrapper $ccPath/icpx
wrap ifx $wrapper $ccPath/ifx
ln -s $out/bin/icpx $out/bin/c++
ln -s $out/bin/icx $out/bin/cc
sed -i 's/.*isCxx=0/isCxx=1/' $out/bin/icpx
# Use this to detect when a compiler subprocess is called
# from icpx (--fsycl-host-compiler)
echo 'export NIX_CC_WRAPPER_INTEL=1' >>$out/nix-support/cc-wrapper-hook
# oneMath looks for sycl libraries in bin/../lib
ln -s ${cc}/lib $out/lib
ln -s ${cc}/include $out/include
'';
}).overrideAttrs
(old: {
installPhase = old.installPhase + ''
export named_cc="icx"
export named_cxx="icpx"
export named_fc="ifx"
'';
});
in
stdenvNoCC.mkDerivation (finalAttrs: {
pname = lib.removeSuffix "-extracted" unpatched.pname;
inherit (unpatched) version;
src = unpatched;
phases = [
"installPhase"
"fixupPhase"
];
buildInputs = [
libffi_3_3
libelf
libxml2
hwloc
numactl
libuuid
libpsm2
zlib
ocl-icd
rdma-core
ucx
openssl
python3
stdenv.cc.cc.lib
]
++ extraPackages;
autoPatchelfIgnoreMissingDeps = [
"libhwloc.so.5"
"libcuda.so.1"
"libze_loader.so.1"
];
# There are broken symlinks that go outside packages, ignore them
dontCheckForBrokenSymlinks = true;
nativeBuildInputs = [ autoPatchelfHook ];
installPhase = ''
cp -r $src/opt/intel/oneapi/ $out
'';
passthru =
let
pkgs = mapAttrs (
folder: version:
let
original = "${finalAttrs.finalPackage}/${folder}/${version}";
in
(symlinkJoin {
pname = "intel-${folder}";
inherit version;
paths = [ original ];
}).overrideAttrs
{ passthru = { inherit original; }; }
) _components;
in
pkgs
// {
inherit unpatched;
pkgs = lib.recurseIntoAttrs pkgs;
components = _components;
# This contains all packages properly symlinked into toplevel directories
# in $out.
#
# NOTE: there are clashes with packages that have symlinks outside their
# scope (libtcm and env/vars.sh)
all = symlinkJoin {
pname = finalAttrs.finalPackage + "-symlinked";
inherit (finalAttrs.finalPackage) version;
paths = filter lib.isDerivation (attrValues finalAttrs.finalPackage.pkgs);
};
stdenv = overrideCC stdenv finalAttrs.finalPackage.cc;
cc = wrapIntel finalAttrs.finalPackage.pkgs.compiler;
};
})
)

View File

@@ -1,29 +0,0 @@
#!/usr/bin/env -S jq -f
def extract_fields: {
pname : .Package,
version : .Version,
provides : .Package | sub("[0-9.-]*$"; ""),
filename : .Filename,
size : ."Installed-Size" | tonumber,
sha256 : .SHA256,
dependencies : .Depends,
} ;
# parses dependencies into a list of [{.pname, .kind, .version}]
# some dependencies do not have a version specified, in which case, kind = version = null
#
# example dependencies:
# intel-oneapi-common-vars (>= 2023.0.0-25325), intel-oneapi-common-licensing-2023.0.0
def split_dependencies : map(try(.dependencies |= split(",\\s?"; "")) // .dependencies |= []) ;
def match_version : capture("(?<pname>[a-zA-Z0-9_\\-.]*) *(\\((?<kind>[<>=]*) *(?<version>.*)\\))?"; "") ;
def parse_dependencies : map_values(.dependencies.[] |= match_version) ;
def sort_version_decreasing : sort_by(.version | split("[-.]"; "") | map(tonumber)) | reverse ;
map(extract_fields) | split_dependencies | parse_dependencies | sort_version_decreasing
# [.[] | select(.pname == "intel-hpckit") | .version]

View File

@@ -1,29 +0,0 @@
#!/usr/bin/env -S awk -f
BEGIN {
FS=": "
prev_empty=1
t=" "
print "[ {"
}
!NF { # empty line, update separator so next non empty line closes the dict
prev_empty=1
t="},\n{ "
next # skip line (we won't match anything else)
}
{
printf t "\"%s\" : \"%s\"\n", $1, $2
if (prev_empty) {
# we were the first after a group of empty lines, following ones have to
# have a comma
prev_empty=0
t=", "
}
}
END { print "} ]" }

View File

@@ -1,11 +1,4 @@
#!/bin/sh
out_64=$(mktemp intel-api.64.XXXXXX)
out_all=$(mktemp intel-api.all.XXXXXX)
trap 'rm -f "$out_64" "$out_all"' EXIT INT HUP
curl https://apt.repos.intel.com/oneapi/dists/all/main/binary-amd64/Packages -o "$out_64"
curl https://apt.repos.intel.com/oneapi/dists/all/main/binary-all/Packages -o "$out_all"
# NOTE: we use `jq -r tostring` to minify the json (3.2Mb -> 2.3Mb)
cat "$out_64" "$out_all" | ./toJson.awk | ./process.jq | jq -r tostring >packages.json
curl https://apt.repos.intel.com/oneapi/dists/all/main/binary-amd64/Packages -o amd64-packages
curl https://apt.repos.intel.com/oneapi/dists/all/main/binary-all/Packages -o all-packages

View File

@@ -3,7 +3,6 @@
, lib
, gcc
, clangOmpss2Unwrapped
, writeShellScript
, openmp ? null
, wrapCCWith
, llvmPackages_latest
@@ -38,7 +37,10 @@ let
inherit gcc;
cc = clangOmpss2Unwrapped;
gccVersion = with versions; let v = gcc.version; in concatStringsSep "." [(major v) (minor v) (patch v)];
in wrapCCWith {
inherit cc bintools;
# extraPackages adds packages to depsTargetTargetPropagated
extraPackages = optional (openmp != null) openmp;
extraBuildCommands = ''
echo "-target ${targetConfig}" >> $out/nix-support/cc-cflags
echo "-B${gcc.cc}/lib/gcc/${targetConfig}/${gccVersion}" >> $out/nix-support/cc-cflags
@@ -55,50 +57,14 @@ let
echo "--gcc-toolchain=${gcc}" >> $out/nix-support/cc-cflags
wrap clang++ $wrapper $ccPath/clang++
'' + optionalString (openmp != null) ''
echo "export OPENMP_RUNTIME=${ompname}" >> $out/nix-support/cc-wrapper-hook
'' + optionalString (ompss2rt != null) ''
echo "export OMPSS2_RUNTIME=${rtname}" >> $out/nix-support/cc-wrapper-hook
echo "export ${homevar}=${ompss2rt}" >> $out/nix-support/cc-wrapper-hook
'' + optionalString (ompss2rt != null && ompss2rt.pname == "nodes") ''
echo "export NOSV_HOME=${ompss2rt.nosv}" >> $out/nix-support/cc-wrapper-hook
'';
}
envExports = lib.optionalString (openmp != null) ''
echo "export OPENMP_RUNTIME=${ompname}" >> $out/nix-support/cc-wrapper-hook
'' + optionalString (ompss2rt != null) ''
echo "export OMPSS2_RUNTIME=${rtname}" >> $out/nix-support/cc-wrapper-hook
echo "export ${homevar}=${ompss2rt}" >> $out/nix-support/cc-wrapper-hook
'' + optionalString (ompss2rt != null && ompss2rt.pname == "nodes") ''
echo "export NOSV_HOME=${ompss2rt.nosv}" >> $out/nix-support/cc-wrapper-hook
'';
extraPackages = optional (openmp != null) openmp;
wrappedCC = wrapCCWith {
# extraPackages adds packages to depsTargetTargetPropagated
inherit cc bintools extraPackages;
extraBuildCommands = extraBuildCommands + envExports;
};
resetIntelCCFlags = let tconf = builtins.replaceStrings ["-"] ["_"] targetConfig;
in writeShellScript "remove-intel.sh" ''
if [ "''${NIX_CC_WRAPPER_INTEL:-0}" = 1 ]; then
unset NIX_CFLAGS_COMPILE_${tconf}
unset NIX_CC_WRAPPER_FLAGS_SET_${tconf}
if (( "''${NIX_DEBUG:-0}" >= 1 )); then
echo "ompss2: cleaned NIX_CFLAGS_COMPILE_${tconf} (invokation from intel compiler detected)"
fi
fi
'';
intelExtraBuildCommands = ''
sed -i 's|# Flirting.*|source ${resetIntelCCFlags}\n\n&|' $out/bin/clang
sed -i 's|# Flirting.*|source ${resetIntelCCFlags}\n\n&|' $out/bin/clang++
'';
wrappedCCIntel = wrapCCWith {
inherit cc bintools extraPackages;
# extraPackages adds packages to depsTargetTargetPropagated
extraBuildCommands = intelExtraBuildCommands + envExports;
};
in wrappedCC.overrideAttrs (oldAttrs: {
passthru = oldAttrs.passthru // {
forIcpx = wrappedCCIntel;
};
})

View File

@@ -1,91 +0,0 @@
{
lib,
fetchFromGitHub,
cmake,
withCFlags,
intelPackages,
mklSupport ? true,
config,
cudaSupport ? config.cudaSupport,
cudaPackages ? { },
rocmSupport ? config.rocmSupport,
hipTargets ? null, # only one target at a time supported
rocmPackages ? { },
}:
let
# rocmSupport is not enough, we need a specific target
enableHip = rocmSupport && hipTargets != null;
stdenv = withCFlags (lib.optionals cudaSupport [
"--cuda-path=${cudaPackages.cudatoolkit}"
]) intelPackages.stdenv;
in
# at least one backend has to be enabled
assert mklSupport || cudaSupport || enableHip;
stdenv.mkDerivation rec {
pname = "oneMath";
version = "0.8";
src = fetchFromGitHub {
owner = "uxlfoundation";
repo = "oneMath";
rev = "v${version}";
sha256 = "sha256-xK8lKI3oqKlx3xtvdScpMq+HXAuoYCP0BZdkEqnJP5o=";
};
cmakeFlags = [
(lib.cmakeBool "ENABLE_MKLCPU_BACKEND" mklSupport)
(lib.cmakeBool "ENABLE_MKLGPU_BACKEND" mklSupport)
(lib.cmakeBool "ENABLE_CUBLAS_BACKEND" cudaSupport)
(lib.cmakeBool "ENABLE_CUFFT_BACKEND" cudaSupport)
(lib.cmakeBool "ENABLE_CURAND_BACKEND" cudaSupport)
(lib.cmakeBool "ENABLE_CUSOLVER_BACKEND" cudaSupport)
(lib.cmakeBool "ENABLE_CUSPARSE_BACKEND" cudaSupport)
(lib.cmakeBool "ENABLE_ROCBLAS_BACKEND" enableHip)
(lib.cmakeBool "ENABLE_ROCFFT_BACKEND" enableHip)
(lib.cmakeBool "ENABLE_ROCSOLVER_BACKEND" enableHip)
(lib.cmakeBool "ENABLE_ROCRAND_BACKEND" enableHip)
(lib.cmakeBool "ENABLE_ROCSPARSE_BACKEND" enableHip)
(lib.cmakeBool "BUILD_FUNCTIONAL_TESTS" false)
(lib.cmakeBool "BUILD_EXAMPLES" false)
]
++ lib.optionals enableHip [
(lib.cmakeFeature "HIP_TARGETS" hipTargets)
];
nativeBuildInputs = [ cmake ];
buildInputs =
lib.optionals (mklSupport) [
intelPackages.mkl
intelPackages.tbb
]
++ lib.optionals (enableHip) [
rocmPackages.rocmPath
rocmPackages.rocblas
rocmPackages.rocfft
rocmPackages.rocsolver
rocmPackages.rocrand
rocmPackages.rocsparse
]
++ lib.optionals (cudaSupport) [
(lib.getDev cudaPackages.cuda_cudart)
cudaPackages.cudatoolkit
cudaPackages.libcublas
cudaPackages.libcurand
cudaPackages.libcufft
cudaPackages.libcusparse
cudaPackages.libcusolver
];
}

View File

@@ -0,0 +1,26 @@
From ccf301659caac9b5e973ba1f2d32352acf617a98 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aleix=20Bon=C3=A9?= <aleix.boneribo@bsc.es>
Date: Tue, 2 Jul 2024 16:45:05 +0200
Subject: [PATCH] cmake: do not use suffix
---
cmake/LLVM.cmake | 3 ++
1 files changed, 3 insertions(+)
diff --git a/cmake/LLVM.cmake b/cmake/LLVM.cmake
index f4dbda065..e29144dce 100644
--- a/cmake/LLVM.cmake
+++ b/cmake/LLVM.cmake
@@ -65,6 +65,9 @@ else()
message(WARNING "Cannot determine llvm binary suffix from ${LLVM_CONFIG}")
endif()
message(STATUS "LLVM binaries suffix : ${LLVM_BINARY_SUFFIX}")
+
+ # We don't want suffixes in nix
+ set(LLVM_BINARY_SUFFIX "")
endif()
get_filename_component(LLVM_CONFIG_LOCATION "${LLVM_CONFIG}" DIRECTORY)
--
2.45.1

View File

@@ -0,0 +1,33 @@
From f24b456c50f587b05cc8f2699c94d8cdefc5b13e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aleix=20Bon=C3=A9?= <aleix.boneribo@bsc.es>
Date: Tue, 2 Jul 2024 17:11:11 +0200
Subject: [PATCH] cmake: native build tools
---
cmake/LLVM.cmake | 7 +-
1 files changed, 6 insertions(+), 1 deletion(-)
diff --git a/cmake/LLVM.cmake b/cmake/LLVM.cmake
index e29144dce..b9f14ce6a 100644
--- a/cmake/LLVM.cmake
+++ b/cmake/LLVM.cmake
@@ -269,10 +269,15 @@ endforeach()
####################################################################
+if(DEFINED LLVM_NATIVE_TOOL_DIR)
+ set(TOOL_DIR "${LLVM_NATIVE_TOOL_DIR}")
+ message(STATUS "TOOL_DIR=${TOOL_DIR}")
+endif()
+
macro(find_program_or_die OUTPUT_VAR PROG_NAME DOCSTRING)
find_program(${OUTPUT_VAR}
NAMES "${PROG_NAME}${LLVM_BINARY_SUFFIX}${CMAKE_EXECUTABLE_SUFFIX}"
- HINTS "${LLVM_BINDIR}" "${LLVM_CONFIG_LOCATION}"
+ HINTS "${TOOL_DIR}" "${LLVM_BINDIR}" "${LLVM_CONFIG_LOCATION}"
DOC "${DOCSTRING}"
NO_CMAKE_PATH
NO_CMAKE_ENVIRONMENT_PATH
--
2.45.1

View File

@@ -0,0 +1,48 @@
From 4688b5ce895761c884ae15fc0234ed8d623b988b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aleix=20Bon=C3=A9?= <aleix.boneribo@bsc.es>
Date: Tue, 2 Jul 2024 15:41:47 +0200
Subject: [PATCH] cmake: use clang from cmake variable
---
cmake/LLVM.cmake | 22 ++++++++++++++++++++--
1 file changed, 20 insertions(+), 2 deletions(-)
diff --git a/cmake/LLVM.cmake b/cmake/LLVM.cmake
index 71b786dc7..f4dbda065 100644
--- a/cmake/LLVM.cmake
+++ b/cmake/LLVM.cmake
@@ -281,11 +281,29 @@ macro(find_program_or_die OUTPUT_VAR PROG_NAME DOCSTRING)
endif()
endmacro()
-find_program_or_die( CLANG "clang" "clang binary")
+# -DCMAKE_C_COMPILER=riscv64-unknown-linux-gnu-clang -DCMAKE_CXX_COMPILER=riscv64-unknown-linux-gnu-clang++
+
+# if(DEFINED CLANG_EXECUTABLE)
+if(DEFINED CMAKE_C_COMPILER)
+ # set(CLANG "${CLANG_EXECUTABLE}")
+ set(CLANG "${CMAKE_C_COMPILER}")
+ message(STATUS "Using CLANG_EXECUTABLE: ${CLANG}")
+else()
+ find_program_or_die( CLANG "clang" "clang binary")
+endif()
execute_process(COMMAND "${CLANG}" "--version" OUTPUT_VARIABLE LLVM_CLANG_VERSION RESULT_VARIABLE CLANG_RES)
+
# TODO this should be optional
-find_program_or_die( CLANGXX "clang++" "clang++ binary")
+# if(DEFINED CLANGXX_EXECUTABLE)
+if(DEFINED CMAKE_CXX_COMPILER)
+ # set(CLANGXX "${CLANGXX_EXECUTABLE}")
+ set(CLANGXX "${CMAKE_CXX_COMPILER}")
+ message(STATUS "Using CLANGXX_EXECUTABLE: ${CLANGXX}")
+else()
+ find_program_or_die( CLANGXX "clang++" "clang++ binary")
+endif()
execute_process(COMMAND "${CLANGXX}" "--version" OUTPUT_VARIABLE LLVM_CLANGXX_VERSION RESULT_VARIABLE CLANGXX_RES)
+
if(CLANGXX_RES OR CLANG_RES)
message(FATAL_ERROR "Failed running clang/clang++ --version")
endif()
--
2.45.1

255
pkgs/pocl/default.nix Normal file
View File

@@ -0,0 +1,255 @@
{
lib,
llvmPackages,
ninja,
cmake,
hwloc,
ocl-icd,
opencl-headers,
getconf,
pkg-config,
spirv-llvm-translator,
spirv-tools,
libxml2, # required for statically linked llvm
python3,
writableTmpDirAsHomeHook,
writeText,
runCommand,
opencl-cts,
fetchFromGitHub,
useGit ? false,
gitUrl ? "git@gitlab-internal.bsc.es:task-awareness/pocl/pocl.git",
gitBranch ? "new-device",
gitCommit ? "dd10c2221b31223cbb796182df6a07f11c7541f5",
staticLLVM ? true,
enableICD ? true,
enableSPIRV ? true,
enableHWLOC ? true,
enableRemoteServer ? false,
enableRemoteClient ? false,
enableDistroVariants ? false,
lttng-ust,
enableLTTNG ? false,
onetbb,
enableTBB ? false,
nosv ? null,
ovni ? null,
enableNOSV ? useGit,
enableOVNI ? useGit,
buildPackages,
targetPackages,
nix-update-script,
}:
assert (enableNOSV || enableOVNI) -> useGit;
let
release = rec {
version = "7.1";
src = fetchFromGitHub {
owner = "pocl";
repo = "pocl";
tag = "v${version}";
hash = "sha256-bS6vTIjLO7YLs7qYLKW0cYYbEJ/hRS/+IjjAKbkj8ac=";
};
};
git = rec {
version = src.shortRev;
src = builtins.fetchGit {
url = gitUrl;
ref = gitBranch;
rev = gitCommit;
};
};
source = if (useGit) then git else release;
stdenv = llvmPackages.stdenv;
buildLlvmPackages = buildPackages.llvmPackages;
isCross = stdenv.hostPlatform != stdenv.buildPlatform;
featureList = lib.concatStringsSep "+" (
(lib.optionals enableICD [ "icd" ])
++ (lib.optionals enableNOSV [ "nosv" ])
++ (lib.optionals enableOVNI [ "ovni" ])
++ (lib.optionals enableSPIRV [ "spirv" ])
++ (lib.optionals enableTBB [ "tbb" ])
++ (lib.optionals enableRemoteClient [ "server" ])
++ (lib.optionals enableRemoteServer [ "client" ])
);
in
stdenv.mkDerivation (finalAttrs: {
pname = "pocl";
version = featureList + "-" + source.version;
inherit (source) src;
patches = lib.optionals useGit [
./0001-cmake-use-clang-from-cmake-variable.patch
./0001-cmake-do-not-use-suffix.patch
./0001-cmake-native-build-tools.patch
];
cmakeFlags = [
# TODO: all these are broken when cross compiling. Upstream has refactored
# all the cmake infra for cross compilation, but it's not in a release yet
(lib.cmakeOptionType "filepath" "WITH_LLVM_CONFIG" (
lib.getExe' buildLlvmPackages.llvm.dev "llvm-config"
))
(lib.cmakeOptionType "filepath" "CLANG" (lib.getExe' buildLlvmPackages.clangUseLLVM "clang"))
(lib.cmakeOptionType "filepath" "CLANGXX" (lib.getExe' buildLlvmPackages.clangUseLLVM "clang++"))
(lib.cmakeOptionType "path" "CLANG_RESOURCE_DIR" "${llvmPackages.stdenv.cc}/resource-root")
(lib.cmakeBool "ENABLE_TESTS" finalAttrs.finalPackage.doCheck)
(lib.cmakeBool "ENABLE_RELOCATION" true)
(lib.cmakeBool "ENABLE_POCL_BUILDING" true)
(lib.cmakeBool "POCL_ICD_ABSOLUTE_PATH" true)
(lib.cmakeBool "ENABLE_TBB_DEVICE" enableTBB)
(lib.cmakeBool "ENABLE_ICD" enableICD)
(lib.cmakeBool "ENABLE_HWLOC" enableHWLOC)
(lib.cmakeBool "ENABLE_LTTNG" enableLTTNG)
(lib.cmakeBool "ENABLE_REMOTE_CLIENT" enableRemoteClient)
(lib.cmakeBool "ENABLE_REMOTE_SERVER" enableRemoteServer)
# avoid the runtime linker pulling in a different llvm e.g. from graphics drivers
(lib.cmakeBool "STATIC_LLVM" staticLLVM)
]
++ lib.optionals (enableNOSV || enableOVNI) [
(lib.cmakeBool "ENABLE_NOSV" enableNOSV)
(lib.cmakeBool "ENABLE_OVNI" enableOVNI)
]
++ lib.optionals (!isCross && enableDistroVariants) [
(lib.cmakeFeature "KERNELLIB_HOST_CPU_VARIANTS" "distro") # TODO: check if we can do it when cross compiling
]
++ lib.optionals isCross [
(lib.cmakeFeature "CMAKE_SYSTEM_NAME" "Linux")
(lib.cmakeFeature "CMAKE_FIND_ROOT_PATH_MODE_PROGRAM" "NEVER")
(lib.cmakeFeature "CMAKE_FIND_ROOT_PATH_MODE_LIBRARY" "ONLY")
(lib.cmakeFeature "CMAKE_FIND_ROOT_PATH_MODE_INCLUDE" "ONLY")
(lib.cmakeFeature "CMAKE_FIND_ROOT_PATH_MODE_PACKAGE" "ONLY")
]
++ lib.optionals stdenv.hostPlatform.isRiscV [
(lib.cmakeFeature "LLC_TRIPLE" "riscv64-unknown-linux-gnu")
(lib.cmakeFeature "LLC_HOST_CPU" "rv64g")
(lib.cmakeFeature "CLANG_MARCH_FLAG" "-march=")
];
# Fixes error: ld: cannot find -lgcc / -lgcc_s when linking from libpocl on
# host
preConfigure = ''
cmakeFlagsArray+=(
-DEXTRA_HOST_LD_FLAGS="${lib.escapeShellArg (builtins.readFile "${targetPackages.stdenv.cc}/nix-support/cc-ldflags")}"
-DEXTRA_KERNEL_FLAGS="${lib.escapeShellArg (builtins.readFile "${targetPackages.stdenv.cc}/nix-support/cc-ldflags")}"
)
'';
nativeBuildInputs = [
cmake
getconf
ninja
pkg-config
python3
];
buildInputs = [
opencl-headers
llvmPackages.libclang
llvmPackages.llvm
]
++ lib.optionals staticLLVM [ libxml2 ]
++ lib.optionals enableNOSV [ nosv ]
++ lib.optionals enableOVNI [ ovni ]
++ lib.optionals enableHWLOC [ hwloc ]
++ lib.optionals enableTBB [ onetbb ]
++ lib.optionals enableICD [ ocl-icd ]
++ lib.optionals enableLTTNG [ lttng-ust ]
++ lib.optionals enableSPIRV [
(spirv-llvm-translator.override { inherit (llvmPackages) llvm; })
spirv-tools
];
nativeInstallCheckInputs = [
writableTmpDirAsHomeHook # needed for POCL_CACHE_DIR
];
doInstallCheck = true;
installCheckPhase = lib.optionalString enableICD ''
runHook preInstallCheck
export OCL_ICD_VENDORS=$out/etc/OpenCL/vendors
$out/bin/poclcc -o poclcc.cl.pocl $src/examples/poclcc/poclcc.cl
runHook postInstallCheck
'';
setupHook = lib.optionalDrvAttr enableICD (
writeText "setup-hook" ''
addToSearchPath OCL_ICD_VENDORS @out@/etc/OpenCL/vendors
''
);
passthru =
let
self = finalAttrs.finalPackage;
in
{
updateScript = nix-update-script { };
git = self.override { useGit = true; };
withICD = enableICD;
test = {
# Run builtin tests (causes a rebuild)
self = self.overrideAttrs { doCheck = true; };
# Run basic test from Khronos OpenCL Conformance Test Suite
# WARN: despite its name, test_basic is very exhaustive, and can take more
# than 1 hour in a marenostrum5 node.
cts =
assert enableICD;
lib.genAttrs [ "api" "basic" "c11_atomics" "printf" "svm" ] (
name:
runCommand "pocl-cts-test_${name}" {
nativeBuildInputs = [ self ];
requiredSystemFeatures = [ "sys-devices" ];
env = {
POCL_DEBUG = "error,warn";
POCL_CACHE_DIR = "/build/pocl_cache";
};
} "${opencl-cts}/bin/Release/test_${name} | tee $out"
);
};
};
env.HWLOC_SYNTHETIC = lib.optionalDrvAttr enableHWLOC "node:1 core:1 pu:1";
propagatedBuildInputs = [ stdenv.cc.cc ];
meta = with lib; {
description = "A portable open source (MIT-licensed) implementation of the OpenCL standard";
homepage = "http://portablecl.org";
license = licenses.mit;
maintainers = with maintainers; [
leixb
jansol
xddxdd
bsc.maintainers.leixb
];
platforms = platforms.linux ++ platforms.darwin;
};
})

72
pkgs/tacuda/default.nix Normal file
View File

@@ -0,0 +1,72 @@
{
backendStdenv,
fetchFromGitHub,
automake,
autoconf,
libtool,
gnumake,
autoreconfHook,
boost,
cudatoolkit,
libcublas,
cuda_cudart,
useGit ? false,
gitUrl ? "git@gitlab-internal.bsc.es:task-awareness/tacuda/tacuda.git",
gitBranch ? "main",
gitCommit ? "35234f9445e6149a2bd38d119841e2485d6ee05e",
}:
let
release_ver = "2.1.0";
release = {
version = release_ver;
src = fetchFromGitHub {
owner = "bsc-pm";
repo = "tacuda";
rev = release_ver;
hash = "sha256-Cj3EiLVJSLvRv0ydeg7Vp4SpkniEqHkcWF+YOJQ8EcM=";
};
};
git = rec {
version = src.shortRev;
src = builtins.fetchGit {
url = gitUrl;
ref = gitBranch;
rev = gitCommit;
};
};
source = if (useGit) then git else release;
in
backendStdenv.mkDerivation {
pname = "tacuda";
inherit (source) src version;
enableParallelBuilding = true;
separateDebugInfo = true;
strictDeps = true;
nativeBuildInputs = [
autoreconfHook
automake
autoconf
libtool
gnumake
];
patches = [ ./fix_config.patch ];
configureFlags = [ "--with-cuda-include=${cudatoolkit}/include" ];
buildInputs = [
boost
libcublas
cuda_cudart
];
}

View File

@@ -0,0 +1,13 @@
diff --git a/m4/cuda.m4 b/m4/cuda.m4
index 23f5c94..8f9b534 100644
--- a/m4/cuda.m4
+++ b/m4/cuda.m4
@@ -40,7 +40,7 @@ search_libs="cuda cublas cudart"
required_libs=""
m4_foreach([function],
- [cuInit,
+ [
cublasSgemm,
cudaStreamCreate,
cudaLaunchKernel,

46
pkgs/taopencl/default.nix Normal file
View File

@@ -0,0 +1,46 @@
{
stdenv,
automake,
autoconf,
libtool,
gnumake,
boost,
mpi,
gcc,
autoreconfHook,
ocl-icd,
opencl-headers,
}:
stdenv.mkDerivation (finalAttrs: {
pname = "taopencl";
version = finalAttrs.src.shortRev;
src = builtins.fetchGit {
url = "git@gitlab-internal.bsc.es:task-awareness/taopencl/taopencl.git";
ref = "master";
rev = "c3b7b32ae8fa4af7ceff598532a881f8f1490aaf";
};
enableParallelBuilding = true;
separateDebugInfo = true;
configureFlags = [
"--with-opencl-lib=${ocl-icd}/lib"
"--with-opencl-include=${opencl-headers}/include"
];
buildInputs = [
autoreconfHook
automake
autoconf
libtool
gnumake
boost
mpi
gcc
];
dontDisableStatic = true;
hardeningDisable = [ "all" ];
})

View File

@@ -1,87 +0,0 @@
{
lib,
stdenv,
autoconf,
automake,
autoreconfHook,
boost,
fetchFromGitHub,
gnumake,
libtool,
withCFlags,
useIntel ? true,
adaptivecpp ? null,
intelPackages ? null,
useGit ? false,
gitUrl ? "git@gitlab-internal.bsc.es:task-awareness/tasycl/tasycl.git",
gitBranch ? "main",
gitCommit ? "78f98dcf60a66e0eaa3b4ebcf55be076bec64825",
}:
assert !useIntel -> adaptivecpp != null;
assert useIntel -> intelPackages != null;
let
variant = if useIntel then "intel" else "acpp";
syclStdenv = withCFlags [ "-O3" ] (if useIntel then intelPackages.stdenv else stdenv);
release = rec {
version = "2.1.0";
src = fetchFromGitHub {
owner = "bsc-pm";
repo = "tasycl";
rev = version;
hash = "sha256-0kXnb0lHeQNHR27GTLbJ8xbiICLU8k2+FqEnnFSrzzo=";
};
};
git = rec {
version = src.shortRev;
src = builtins.fetchGit {
url = gitUrl;
ref = gitBranch;
rev = gitCommit;
};
};
source = if (useGit) then git else release;
in
syclStdenv.mkDerivation {
pname = "tasycl-${variant}";
inherit (source) src version;
enableParallelBuilding = true;
separateDebugInfo = true;
nativeBuildInputs = [
autoreconfHook
automake
autoconf
libtool
gnumake
];
buildInputs = [
boost
];
configureFlags = lib.optionals (!useIntel) [
"CXX=${lib.getExe adaptivecpp}"
];
# add symlinks so we can explicitly link with tasycl-intel / tasycl-acpp
postInstall = ''
pushd $out/lib
for i in libtasycl* ; do
ln -s "$i" "''\${i/tasycl/tasycl-${variant}}"
done
popd
'';
hardeningDisable = [ "all" ];
}

View File

@@ -1,62 +0,0 @@
{
intelPackages,
writeText,
strace,
}:
let
stdenv = intelPackages.stdenv;
hello_sycl = writeText "hello.cpp" ''
#include <sycl/sycl.hpp>
class hello_world;
int main(int argc, char** argv) try {
auto device_selector = sycl::default_selector_v;
sycl::queue queue(device_selector);
std::cout << "Running on: "
<< queue.get_device().get_info<sycl::info::device::name>()
<< std::endl;
queue.submit([&] (sycl::handler& cgh) {
auto os = sycl::stream{128, 128, cgh};
cgh.single_task<hello_world>([=]() {
os << "Hello World! (on device)\n";
});
});
return 0;
} catch (sycl::exception &e) {
std::cout << "SYCL exception: " << e.what() << std::endl;
return 0; // we excpect to fail since no devices should be available;
}
'';
in
stdenv.mkDerivation {
version = "0.0.1";
name = "hello-sycl";
buildInputs = [
stdenv
strace
];
src = hello_sycl;
dontUnpack = true;
dontConfigure = true;
NIX_DEBUG = 0;
buildPhase = ''
cp $src hello.cpp
set -x
echo CXX=$CXX
command -v $CXX
$CXX -fsycl hello.cpp -o hello
./hello
set +x
'';
installPhase = ''
touch $out
'';
}

View File

@@ -1,81 +0,0 @@
{
writeText,
intelPackages,
nodes,
nosv,
clangOmpss2Nodes,
strace,
}:
let
hello_cpp = writeText "hello.cpp" ''
#include <cstdio>
#include <sycl/sycl.hpp>
int main(int argc, char** argv) try {
sycl::queue queue;
std::cout << "Running on: "
<< queue.get_device().get_info<sycl::info::device::name>()
<< std::endl;
#pragma oss task
queue.submit([&] (sycl::handler& cgh) {
auto os = sycl::stream{128, 128, cgh};
cgh.single_task<class hello_world>([=]() {
os << "Hello World! (on device)\n";
});
}).wait();
return 0;
} catch (sycl::exception &e) {
std::cout << "SYCL exception: " << e.what() << std::endl;
return 0; // we expect to fail since no devices should be available;
}
'';
in
intelPackages.stdenv.mkDerivation {
version = "0.0.1";
name = "hello-syclompss";
src = hello_cpp;
nativeBuildInputs = [
strace
nodes
nosv
];
dontUnpack = true;
dontConfigure = true;
# NODES requires access to /sys/devices to request NUMA information
requiredSystemFeatures = [ "sys-devices" ];
env.NODES_HOME = nodes;
NIX_DEBUG = 1;
buildPhase = ''
cp $src hello.cpp
set -x
echo CXX=$CXX
echo NODES_HOME=$NODES_HOME
command -v $CXX
icpx -Wno-deprecated-declarations -fsycl \
-fsycl-host-compiler=${clangOmpss2Nodes.forIcpx}/bin/clang++ \
-fsycl-host-compiler-options='-Wno-deprecated-declarations -fompss-2=libnodes' \
-lnodes -lnosv \
$NODES_HOME/lib/nodes-main-wrapper.o \
hello.cpp -o hello
./hello
set +x
'';
installPhase = ''
touch $out
'';
}

127
test/compilers/opencl.c Normal file
View File

@@ -0,0 +1,127 @@
#include <CL/cl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#define BUF_LEN 128
cl_int exit_err = CL_SUCCESS;
#define CHECK(cmd) \
do { \
cl_int err = cmd; \
if (err != CL_SUCCESS) { \
printf("[ERROR] " #cmd " (Error code: %d)\n" \
"@ "__FILE__ \
":%d", \
err, __LINE__); \
exit_err = err; \
goto cleanup; \
} \
} while (0)
const char *kernel_source = "__kernel void vector_add(__global const float *a, "
"__global const float *b, __global float *c) {\n"
" int gid = get_global_id(0);\n"
" c[gid] = a[gid] + b[gid];\n"
"}\n";
cl_int test_kernel_compilation(cl_platform_id platform) {
cl_device_id device = NULL;
cl_context context = NULL;
cl_program program = NULL;
cl_int err = CL_SUCCESS;
char device_name[BUF_LEN];
// Get first device for this platform
CHECK(clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 1, &device, NULL));
// Get device name
CHECK(clGetDeviceInfo(device, CL_DEVICE_NAME, sizeof(device_name),
device_name, NULL));
printf(" Device: %s\n", device_name);
context = clCreateContext(NULL, 1, &device, NULL, NULL, &err);
CHECK(err);
program = clCreateProgramWithSource(context, 1, &kernel_source, NULL, &err);
CHECK(err);
err = clBuildProgram(program, 1, &device, NULL, NULL, NULL);
if (err != CL_SUCCESS) {
printf(" [ERROR] Kernel compilation failed (Error code: %d)\n", err);
// Build log
size_t log_size;
CHECK(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, NULL,
&log_size));
char *log = (char *)malloc(log_size);
CHECK(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, log_size,
log, NULL));
printf(" Build log:\n%s\n", log);
free(log);
goto cleanup;
}
printf(" [OK] Kernel compiled successfully!\n");
cleanup:
if (program)
clReleaseProgram(program);
if (context)
clReleaseContext(context);
printf("\n");
return exit_err;
}
int main() {
cl_uint num_platforms;
cl_platform_id *platforms = NULL;
// Get number of platforms
CHECK(clGetPlatformIDs(0, NULL, &num_platforms));
printf("Found %d OpenCL platform(s)\n\n", num_platforms);
if (num_platforms == 0) {
printf("No OpenCL platforms found!\n");
return EXIT_FAILURE;
}
// Allocate memory for platforms
platforms = (cl_platform_id *)malloc(sizeof(cl_platform_id) * num_platforms);
// Get platform IDs
CHECK(clGetPlatformIDs(num_platforms, platforms, NULL));
// Query each platform and test kernel compilation
for (cl_uint i = 0; i < num_platforms; i++) {
char platform_name[BUF_LEN];
char platform_vendor[BUF_LEN];
char platform_version[BUF_LEN];
CHECK(clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME,
sizeof(platform_name), platform_name, NULL));
CHECK(clGetPlatformInfo(platforms[i], CL_PLATFORM_VENDOR,
sizeof(platform_vendor), platform_vendor, NULL));
CHECK(clGetPlatformInfo(platforms[i], CL_PLATFORM_VERSION,
sizeof(platform_version), platform_version, NULL));
printf("Platform %d: %s\n", i, platform_name);
printf(" Vendor: %s\n", platform_vendor);
printf(" Version: %s\n", platform_version);
// Test kernel compilation
CHECK(test_kernel_compilation(platforms[i]));
}
cleanup:
if (platforms)
free(platforms);
if (exit_err == CL_SUCCESS)
return EXIT_SUCCESS;
return EXIT_FAILURE;
}

104
test/compilers/opencl.nix Normal file
View File

@@ -0,0 +1,104 @@
{
stdenv,
ocl-icd,
opencl-headers,
pocl,
clinfo,
runCommand,
lib,
}:
stdenv.mkDerivation (finalAttrs: {
pname = "opencl-pocl-test";
version = "1.0.0";
src = ./opencl.c;
dontUnpack = true;
dontConfigure = true;
buildInputs = [
ocl-icd
opencl-headers
];
buildPhase = ''
runHook preBuild
$CC -Wall -Wextra -D CL_TARGET_OPENCL_VERSION=100 $src -o ocl-hello -lOpenCL
runHook postBuild
'';
installPhase = ''
runHook preInstall
mkdir -p $out/bin
cp ocl-hello $out/bin
runHook postInstall
'';
passthru = {
test-icd =
runCommand "custom-clinfo"
{
nativeBuildInputs =
assert !(lib.hasPrefix "pocl" ocl-icd.pname);
[
clinfo
finalAttrs.finalPackage
pocl
];
env = {
POCL_DEBUG = "error,warn";
POCL_CACHE_DIR = "/build/pocl_cache";
};
requiredSystemFeatures = [ "sys-devices" ];
}
''
set -x
mkdir $out
clinfo -l >$out/clinfo
clinfo >$out/clinfo-full
ocl-hello >$out/hello
for i in $out/*; do
grep "Portable Computing Language" $i
done
echo $OCL_ICD_VENDORS >$out/env
grep "pocl.*/etc/OpenCL/vendors" $out/env
set +x
'';
# this needs: ocl-icd = pocl-noicd
test-noicd =
runCommand "custom-clinfo"
{
nativeBuildInputs = [
finalAttrs.finalPackage
];
env = {
POCL_DEBUG = "error,warn";
POCL_CACHE_DIR = "/build/pocl_cache";
};
requiredSystemFeatures = [ "sys-devices" ];
}
''
ocl-hello >$out
grep "Portable Computing Language" $out
'';
};
})

27
test/compilers/pocl.nix Normal file
View File

@@ -0,0 +1,27 @@
{
runCommand,
pocl,
clinfo,
}:
runCommand "clinfo-pocl"
{
nativeBuildInputs = [
clinfo
pocl
];
requiredSystemFeatures = [ "sys-devices" ];
env = {
POCL_DEBUG = "error,warn";
POCL_CACHE_DIR = "/build/pocl_cache";
};
}
''
echo $OCL_ICD_VENDORS
clinfo >$out
# check that we have PoCL:
grep "Portable Computing Language" $out
''