Add TAOpenCL

overlay.nix

@@ -62,7 +62,14 @@ 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 { };
+
+    _cuda = prev._cuda.extend (_: _prev: final.lib.recursiveUpdate _prev {
+      extensions = _prev.extensions ++ [(finalAttrs: _: {
+        tacuda = finalAttrs.callPackage ./pkgs/tacuda/default.nix { };
+      })];
+    });
   };
 
   tests = rec {

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 +1,48 @@
+{
+  stdenv,
+  automake,
+  autoconf,
+  libtool,
+  gnumake,
+  boost,
+  mpi,
+  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;
+  strictDeps = true;
+
+  configureFlags = [
+    "--with-opencl-lib=${ocl-icd}/lib"
+    "--with-opencl-include=${opencl-headers}/include"
+  ];
+
+  nativeBuildInputs = [
+    autoreconfHook
+    automake
+    autoconf
+    libtool
+    gnumake
+  ];
+
+  buildInputs = [
+    boost
+    mpi
+  ];
+
+  dontDisableStatic = true;
+  hardeningDisable = [ "all" ];
+})