Add TACUDA through cudaPackages extensions

See: https://lkml.org/lkml/2025/4/9/1709

flake.lock

@@ -2,16 +2,16 @@
   "nodes": {
     "nixpkgs": {
       "locked": {
-        "lastModified": 1752436162,
-        "narHash": "sha256-Kt1UIPi7kZqkSc5HVj6UY5YLHHEzPBkgpNUByuyxtlw=",
+        "lastModified": 1764522689,
+        "narHash": "sha256-SqUuBFjhl/kpDiVaKLQBoD8TLD+/cTUzzgVFoaHrkqY=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "dfcd5b901dbab46c9c6e80b265648481aafb01f8",
+        "rev": "8bb5646e0bed5dbd3ab08c7a7cc15b75ab4e1d0f",
         "type": "github"
       },
       "original": {
         "owner": "NixOS",
-        "ref": "nixos-25.05",
+        "ref": "nixos-25.11",
         "repo": "nixpkgs",
         "type": "github"
       }

flake.nix

@@ -1,6 +1,6 @@
 {
   inputs = {
-    nixpkgs.url = "github:NixOS/nixpkgs/nixos-25.05";
+    nixpkgs.url = "github:NixOS/nixpkgs/nixos-25.11";
   };
 
   outputs = { self, nixpkgs, ... }:

m/common/base/env.nix

@@ -1,10 +1,10 @@
-{ pkgs, config, ... }:
+{ pkgs, ... }:
 
 {
   environment.systemPackages = with pkgs; [
     vim wget git htop tmux pciutils tcpdump ripgrep nix-index nixos-option
     nix-diff ipmitool freeipmi ethtool lm_sensors cmake gnumake file tree
-    ncdu config.boot.kernelPackages.perf ldns pv
+    ncdu perf ldns pv
     # From jungle overlay
     osumb nixgen
   ];

m/common/base/watchdog.nix

@@ -5,5 +5,5 @@
   boot.kernelModules = [ "ipmi_watchdog" ];
 
   # Enable systemd watchdog with 30 s interval
-  systemd.watchdog.runtimeTime = "30s";
+  systemd.settings.Manager.RuntimeWatchdogSec = 30;
 }

m/module/debuginfod.nix

@@ -1,3 +1,10 @@
 {
-  services.nixseparatedebuginfod.enable = true;
+  services.nixseparatedebuginfod2 = {
+    enable = true;
+    substituters = [
+      "local:"
+      "https://cache.nixos.org"
+      "http://hut/cache"
+    ];
+  };
 }

overlay.nix

@@ -30,7 +30,8 @@ let
       amd-uprof-driver = _prev.callPackage ./pkgs/amd-uprof/driver.nix { };
     });
     lmbench = callPackage ./pkgs/lmbench/default.nix { };
-    mcxx = callPackage ./pkgs/mcxx/default.nix { };
+    # Broken and unmantained
+    # mcxx = callPackage ./pkgs/mcxx/default.nix { };
     meteocat-exporter = prev.callPackage ./pkgs/meteocat-exporter/default.nix { };
     mpi = final.mpich; # Set MPICH as default
     mpich = callPackage ./pkgs/mpich/default.nix { mpich = prev.mpich; };
@@ -49,6 +50,8 @@ let
     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
@@ -61,7 +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 {
@@ -95,6 +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

pkgs/amd-uprof/driver.nix

@@ -19,7 +19,7 @@ in stdenv.mkDerivation {
   '';
   hardeningDisable = [ "pic" "format" ];
   nativeBuildInputs = kernel.moduleBuildDependencies;
-  patches = [ ./makefile.patch ./hrtimer.patch ];
+  patches = [ ./makefile.patch ./hrtimer.patch ./remove-wr-rdmsrq.patch ];
   makeFlags = [
     "KERNEL_VERSION=${kernel.modDirVersion}"
     "KERNEL_DIR=${kernel.dev}/lib/modules/${kernel.modDirVersion}/build"

pkgs/amd-uprof/remove-wr-rdmsrq.patch

@@ -0,0 +1,20 @@
+diff --git a/inc/PwrProfAsm.h b/inc/PwrProfAsm.h
+index d77770a..c93a0e9 100644
+--- a/inc/PwrProfAsm.h
++++ b/inc/PwrProfAsm.h
+@@ -347,6 +347,7 @@
+ 
+ #endif
+ 
++/*
+ #define rdmsrq(msr,val1,val2,val3,val4) ({ \
+         __asm__ __volatile__( \
+                               "rdmsr\n" \
+@@ -362,6 +363,7 @@
+                               :"c"(msr), "a"(val1), "d"(val2), "S"(val3), "D"(val4) \
+                             ); \
+     })
++*/
+ 
+ #define rdmsrpw(msr,val1,val2,val3,val4) ({ \
+         __asm__ __volatile__( \

pkgs/cudainfo/default.nix

@@ -12,7 +12,7 @@ stdenv.mkDerivation (finalAttrs: {
   src = ./.;
   buildInputs = [
     cudatoolkit # Required for nvcc
-    cudaPackages.cuda_cudart.static # Required for -lcudart_static
+    (lib.getOutput "static" cudaPackages.cuda_cudart) # Required for -lcudart_static
     autoAddDriverRunpath
   ];
   installPhase = ''

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/intel-oneapi/2023.nix

@@ -10,7 +10,7 @@
 , zlib
 , autoPatchelfHook
 , libfabric
-, gcc13
+, gcc
 , wrapCCWith
 }:
 
@@ -33,8 +33,6 @@ let
     maintainers = with lib.maintainers.bsc; [ abonerib ];
   };
 
-  gcc = gcc13;
-
   v = {
     hpckit   = "2023.1.0";
     compiler = "2023.1.0";

pkgs/llvm-ompss2/default.nix

@@ -27,10 +27,10 @@ let
   # We need to replace the lld linker from bintools with our linker just built,
   # otherwise we run into incompatibility issues when mixing compiler and linker
   # versions.
-  bintools-unwrapped = llvmPackages_latest.tools.bintools-unwrapped.override {
+  bintools-unwrapped = llvmPackages_latest.bintools-unwrapped.override {
     lld = clangOmpss2Unwrapped;
   };
-  bintools = llvmPackages_latest.tools.bintools.override {
+  bintools = llvmPackages_latest.bintools.override {
     bintools = bintools-unwrapped;
   };
   targetConfig = stdenv.targetPlatform.config;

pkgs/mcxx/default.nix

@@ -65,6 +65,7 @@ stdenv.mkDerivation rec {
   ];
 
   meta = {
+    broken = true;
     homepage = "https://github.com/bsc-pm/mcxx";
     description = "C/C++/Fortran source-to-source compilation infrastructure aimed at fast prototyping";
     maintainers = with lib.maintainers.bsc; [ rpenacob ];

pkgs/meteocat-exporter/default.nix

@@ -1,9 +1,11 @@
 { python3Packages, lib }:
 
-python3Packages.buildPythonApplication rec {
+python3Packages.buildPythonApplication {
   pname = "meteocat-exporter";
   version = "1.0";
 
+  pyproject = true;
+
   src = ./.;
 
   doCheck = false;

pkgs/paraver/default.nix

@@ -12,7 +12,7 @@
 , paraverKernel
 , openssl
 , glibcLocales
-, wrapGAppsHook
+, wrapGAppsHook3
 }:
 
 let
@@ -64,7 +64,7 @@ stdenv.mkDerivation rec {
     autoconf
     automake
     autoreconfHook
-    wrapGAppsHook
+    wrapGAppsHook3
   ];
 
   buildInputs = [

pkgs/pocl/0001-cmake-do-not-use-suffix.patch

@@ -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
+

pkgs/pocl/0001-cmake-native-build-tools.patch

@@ -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
+

pkgs/pocl/0001-cmake-use-clang-from-cmake-variable.patch

@@ -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
+

pkgs/pocl/default.nix

@@ -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;
+  };
+})

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" ];
+})

pkgs/upc-qaire-exporter/default.nix

@@ -1,9 +1,11 @@
 { python3Packages, lib }:
 
-python3Packages.buildPythonApplication rec {
+python3Packages.buildPythonApplication {
   pname = "upc-qaire-exporter";
   version = "1.0";
 
+  pyproject = true;
+
   src = ./.;
 
   doCheck = false;

test/compilers/opencl.c

@@ -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;
+}

test/compilers/opencl.nix

@@ -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
+        '';
+  };
+})

test/compilers/pocl.nix

@@ -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
+  ''