Add TACUDA through cudaPackages extensions

overlay.nix

@@ -64,7 +64,16 @@ let
     tagaspi = callPackage ./pkgs/tagaspi/default.nix { };
     tampi = callPackage ./pkgs/tampi/default.nix { };
     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 {

pkgs/gromacs/default.nix

@@ -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
+    ];
+  };
+}

pkgs/gromacs/pkgconfig-2024.patch

@@ -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

pkgs/gromacs/pkgconfig-2025.patch

@@ -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

pkgs/pocl/default.nix

@@ -14,6 +14,8 @@
   python3,
   writableTmpDirAsHomeHook,
   writeText,
+  runCommand,
+  opencl-cts,
 
   fetchFromGitHub,
 
@@ -201,12 +203,38 @@ stdenv.mkDerivation (finalAttrs: {
     ''
   );
 
-  passthru = {
+  passthru =
-    updateScript = nix-update-script { };
+    let
-    git = finalAttrs.finalPackage.override { useGit = true; };
+      self = finalAttrs.finalPackage;
-    test = finalAttrs.finalPackage.overrideAttrs { doCheck = true; };
+    in
-    withICD = enableICD;
+    {
-  };
+      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";
 

pkgs/tacuda/default.nix

@@ -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
+  ];
+
+}

pkgs/tacuda/fix_config.patch

@@ -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,

pkgs/taopencl/default.nix

@@ -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" ];
+})