Add TACUDA package

overlay.nix

@@ -8,9 +8,6 @@ let
 
   bscPkgs = {
     amd-uprof = prev.callPackage ./pkgs/amd-uprof/default.nix { };
-    aoccUnwrapped = callPackage ./pkgs/aocc/unwrapped.nix { };
-    aocc = callPackage ./pkgs/aocc/default.nix { };
-    stdenvAocc = final.overrideCC final.stdenv final.aocc;
     bench6 = callPackage ./pkgs/bench6/default.nix { };
     bigotes = callPackage ./pkgs/bigotes/default.nix { };
     clangOmpss2 = callPackage ./pkgs/llvm-ompss2/default.nix { };
@@ -58,6 +55,7 @@ let
     stdenvClangOmpss2Nanos6 = final.stdenv.override { cc = final.clangOmpss2Nanos6; allowedRequisites = null; };
     stdenvClangOmpss2Nodes = final.stdenv.override { cc = final.clangOmpss2Nodes; allowedRequisites = null; };
     stdenvClangOmpss2NodesOmpv = final.stdenv.override { cc = final.clangOmpss2NodesOmpv; allowedRequisites = null; };
+    tacuda = callPackage ./pkgs/tacuda/default.nix { };
     tagaspi = callPackage ./pkgs/tagaspi/default.nix { };
     tampi = callPackage ./pkgs/tampi/default.nix { };
     upc-qaire-exporter = prev.callPackage ./pkgs/upc-qaire-exporter/default.nix { };

pkgs/aocc/default.nix

@@ -1,16 +0,0 @@
-{ wrapCCWith
-, aoccUnwrapped
-}:
-
-let
-  cc = aoccUnwrapped;
-in wrapCCWith {
-  inherit cc;
-
-  extraBuildCommands = ''
-    echo "-isystem ${cc}/include" >> $out/nix-support/cc-cflags
-
-    wrap aocc   $wrapper $ccPath/clang
-    wrap aocc++ $wrapper $ccPath/clang++
-  '';
-}

pkgs/aocc/unwrapped.nix

@@ -1,51 +0,0 @@
-{ lib
-, fetchurl
-, stdenv
-, autoPatchelfHook
-, rocmPackages
-, zlib
-, libffi
-, elfutils
-}:
-
-let
-  # in newer nixpkgs the runtime is hsakmt
-  rocmRuntime =
-    if rocmPackages ? hsakmt then
-      rocmPackages.hsakmt
-    else
-      rocmPackages.rocm-runtime;
-in
-stdenv.mkDerivation {
-  pname = "aocc-compiler";
-  version = "5.0.0";
-
-  src = fetchurl {
-    url = "https://download.amd.com/developer/eula/aocc/aocc-5-0/aocc-compiler-5.0.0.tar";
-    sha256 = "sha256-lm+sLSx1np3m6WnBCtp6ezBsET9/HgfqN2gp7IY4Dao=";
-  };
-
-  nativeBuildInputs = [ autoPatchelfHook ];
-  buildInputs = [
-    elfutils
-    zlib
-    rocmRuntime
-    stdenv.cc.cc.lib
-  ];
-  phases = [ "unpackPhase" "installPhase" "fixupPhase" ];
-  dontStrip = true;
-
-  installPhase = ''
-    mkdir -p $out
-    cp -a . $out/
-
-    ln -s ${lib.getLib libffi}/lib/libffi.so $out/lib/libffi.so.6
-  '';
-
-  passthru = {
-    inherit rocmRuntime;
-    isClang = true; # Needed for wrapCCWith
-  };
-
-  meta.mainProgram = "clang";
-}

pkgs/tacuda/default.nix

@@ -0,0 +1,68 @@
+{ stdenv
+, fetchFromGitHub
+, automake
+, autoconf
+, libtool
+, gnumake
+, autoreconfHook
+, boost
+, cudaPackages
+
+, 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
+
+stdenv.mkDerivation {
+  pname = "tacuda";
+  inherit (source) src version;
+
+  enableParallelBuilding = true;
+  separateDebugInfo = true;
+
+  nativeBuildInputs = [
+    autoreconfHook
+    automake
+    autoconf
+    libtool
+    gnumake
+  ];
+
+  patches = [ ./fix_config.patch ];
+
+  configureFlags = [ "--with-cuda-include=${cudaPackages.cudatoolkit}/include" ];
+
+  buildInputs = [
+    boost
+    cudaPackages.libcublas
+    cudaPackages.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,