Merge all branches

Even if we do an override to papi get the proper configure flags for
cross-compiling, the memory fences are not defined for risc-v:

mb.h:67:2: error: #error Need to define rmb for this architecture!

flake.nix

@@ -2,42 +2,25 @@
   inputs.nixpkgs.url = "nixpkgs";
 
   outputs = { self, nixpkgs, ...}:
-  let
-    system = "x86_64-linux";
-    pkgs = import nixpkgs {
-      inherit system;
+    let
       # For now we only support x86
-      overlays = [ self.overlays.default ];
-    };
-  in
+      system = "x86_64-linux";
+      pkgs = import nixpkgs {
+        inherit system;
+        overlays = [ self.overlays.default ];
+      };
+    in
     {
       bscOverlay = import ./overlay.nix;
       overlays.default = self.bscOverlay;
- 
-      # propagate nixpkgs lib, so we can do bscpkgs.lib
-      inherit (nixpkgs) lib;
-
       # full nixpkgs with our overlay applied
       legacyPackages.${system} = pkgs;
 
-      # packages added by our overlay
-      packages.${system} =
-        let
-          inherit (builtins) attrNames removeAttrs;
-          inherit (nixpkgs.lib) flip filterAttrs getAttrs pipe isDerivation;
-        in
-          # extract the names of the packages from the overlay, then get the
-          # actual packages from the full package set with the overlay applied
-          pipe (self.overlays.default null { inherit (nixpkgs) lib; } ) [
-            attrNames
-            (flip getAttrs pkgs)
-            (flip removeAttrs [ "bsc" ])
-            (filterAttrs (_: isDerivation))
-          ];
-
       hydraJobs = {
-        inherit (self.legacyPackages.x86_64-linux.bsc-ci) test;
-        packages = self.packages.${system};
+        inherit (self.legacyPackages.${system}.bsc-ci) test pkgs cross;
       };
-   };
+
+      # propagate nixpkgs lib, so we can do bscpkgs.lib
+      inherit (nixpkgs) lib;
+    };
 }

overlay.nix

@@ -2,7 +2,7 @@ final: /* Future last stage */
 prev:  /* Previous stage */
 
 let
-  inherit (prev.lib) collect isDerivation mapAttrs;
+  inherit (prev.lib) collect isDerivation mapAttrs filterAttrs;
 
   inherit (final) callPackage;
 
@@ -59,10 +59,21 @@ let
     stdenvClangOmpss2NodesOmpv = final.stdenv.override { cc = final.clangOmpss2NodesOmpv; allowedRequisites = null; };
     tagaspi = callPackage ./pkgs/tagaspi/default.nix { };
     tampi = callPackage ./pkgs/tampi/default.nix { };
+    tacuda = callPackage ./pkgs/tacuda/default.nix { };
     taopencl = callPackage ./pkgs/taopencl/default.nix { };
     tasycl = callPackage ./pkgs/tasycl/default.nix { };
     tasycl-acpp = callPackage ./pkgs/tasycl/default.nix { useIntel = false; };
     wxparaver = callPackage ./pkgs/paraver/default.nix { };
+
+    # Fix for RISC-V cross compilation
+    # TODO: remove once we update to 25.11
+    # https://github.com/NixOS/nixpkgs/commit/d9c222c81435b
+    prrte = prev.prrte.overrideAttrs (prevAttrs: {
+      configureFlags = (prevAttrs.configureFlags or []) ++ [
+        "--with-pmix=${prev.lib.getDev final.pmix}"
+        "--with-pmix-libdir=${prev.lib.getLib final.pmix}/lib"
+      ];
+    });
   };
 
 in bscPkgs // {
@@ -115,15 +126,21 @@ in bscPkgs // {
       };
     };
 
-    pkgs = final.runCommand "ci-pkgs" { }
+    pkgs = filterAttrs (_: isDerivation) bscPkgs;
+
+    pkgsList = final.runCommand "ci-pkgs" { }
       "printf '%s\n' ${toString (collect isDerivation bscPkgs)} > $out";
 
     tests = final.runCommand "ci-tests" { }
       "printf '%s\n' ${toString (collect isDerivation final.bsc-ci.test)} > $out";
 
+    cross = prev.lib.genAttrs [ "riscv64" ] (target:
+      final.pkgsCross.${target}.bsc-ci.pkgs
+    );
+
     all = final.runCommand "ci-all" { }
     ''
-      deps="${toString [ final.bsc-ci.pkgs final.bsc-ci.tests ]}"
+      deps="${toString [ final.bsc-ci.pkgsList final.bsc-ci.tests ]}"
       cat $deps
       printf '%s\n' $deps > $out
     '';

pkgs/extrae/default.nix

@@ -20,6 +20,7 @@
 #, python3Packages
 , installShellFiles
 , symlinkJoin
+, enablePapi ? stdenv.hostPlatform == stdenv.buildPlatform # Disabled when cross-compiling
 }:
 
 let
@@ -87,7 +88,7 @@ stdenv.mkDerivation rec {
       --enable-sampling
       --with-unwind=${libunwind.dev}
       --with-xml-prefix=${libxml2.dev}
-      --with-papi=${papi}
+      ${lib.optionalString enablePapi "--with-papi=${papi}"}
       ${if (mpi != null) then ''--with-mpi=${mpi}''
         else ''--without-mpi''}
       --without-dyninst)

pkgs/lmbench/default.nix

@@ -22,7 +22,11 @@ stdenv.mkDerivation rec {
   '';
 
   buildInputs = [ libtirpc ];
-  patches = [ ./fix-install.patch ./gcc-14.patch ];
+  patches = [
+    ./fix-install.patch
+    ./gcc-14.patch
+    ./fix-cross.patch
+  ];
 
   hardeningDisable = [ "all" ];
 
@@ -35,6 +39,8 @@ stdenv.mkDerivation rec {
       CFLAGS=-Wno-implicit-int
       CPPFLAGS=-I${libtirpc.dev}/include/tirpc
       LDFLAGS=-ltirpc
+      CC=${stdenv.cc.targetPrefix}cc
+      AR=${stdenv.cc.targetPrefix}ar
     )
   '';
 

pkgs/lmbench/fix-cross.patch

@@ -0,0 +1,13 @@
+diff --git a/src/Makefile b/src/Makefile
+index 2555014..356eeb3 100644
+--- a/src/Makefile
++++ b/src/Makefile
+@@ -36,7 +36,7 @@ SHELL=/bin/sh
+ 
+ CC=`../scripts/compiler`
+ MAKE=`../scripts/make`
+-AR=ar
++AR?=ar
+ ARCREATE=cr
+ 
+ # base of installation location

pkgs/nanos6/default.nix

@@ -16,6 +16,7 @@
 , jemallocNanos6 ? null
 , cachelineBytes ? 64
 , enableGlibcxxDebug ? false
+, enablePapi ? stdenv.hostPlatform == stdenv.buildPlatform # Disabled when cross-compiling
 , useGit ? false
 , gitUrl ? "ssh://git@bscpm04.bsc.es/nanos6/nanos6"
 , gitBranch ? "master"
@@ -47,6 +48,8 @@ let
   };
 
   source = if (useGit) then git else release;
+
+  isCross = stdenv.hostPlatform != stdenv.buildPlatform;
 in
   stdenv.mkDerivation (source // {
     pname = "nanos6";
@@ -71,9 +74,11 @@ in
       "--disable-all-instrumentations"
       "--enable-ovni-instrumentation"
       "--with-ovni=${ovni}"
+      "--with-boost=${boost.dev}"
     ] ++
       (optional enableJemalloc "--with-jemalloc=${jemallocNanos6}") ++
-      (optional enableGlibcxxDebug "CXXFLAGS=-D_GLIBCXX_DEBUG");
+      (optional enableGlibcxxDebug "CXXFLAGS=-D_GLIBCXX_DEBUG") ++
+      (optional isCross "--with-symbol-resolution=ifunc");
 
     postConfigure = lib.optionalString (!enableDebug) ''
       # Disable debug
@@ -97,16 +102,14 @@ in
       # TODO: papi_version is needed for configure:
       # ./configure: line 27378: papi_version: command not found
       # This probably breaks cross-compilation
-      papi
-    ];
+    ] ++ lib.optionals enablePapi [ papi ];
 
     buildInputs = [
       boost
       numactl
       hwloc
-      papi
       ovni
-    ];
+    ] ++ lib.optionals enablePapi [ papi ];
 
     # Create a script that sets NANOS6_HOME
     postInstall = ''

pkgs/nosv/default.nix

@@ -7,7 +7,7 @@
 , numactl
 , hwloc
 , papi
-, enablePapi ? true
+, enablePapi ? stdenv.hostPlatform == stdenv.buildPlatform # Disabled when cross-compiling
 , cacheline ? 64 # bits
 , ovni ? null
 , useGit ? false

pkgs/tacuda/default.nix

@@ -1,6 +1,4 @@
 { stdenv
-, lib
-, withCFlags
 , fetchFromGitHub
 , automake
 , autoconf
@@ -8,21 +6,24 @@
 , gnumake
 , autoreconfHook
 , boost
+, cudaPackages
 
 , useGit ? false
 , gitUrl ? "git@gitlab-internal.bsc.es:task-awareness/tacuda/tacuda.git"
 , gitBranch ? "main"
-, gitCommit ? ""
+, gitCommit ? "35234f9445e6149a2bd38d119841e2485d6ee05e"
 }:
 
 let
-  release = rec {
-    version = "2.1.0";
+
+  release_ver = "2.1.0";
+  release = {
+    version = release_ver;
     src = fetchFromGitHub {
       owner = "bsc-pm";
       repo = "tacuda";
-      rev = version;
-      hash = "";
+      rev = release_ver;
+      hash = "sha256-Cj3EiLVJSLvRv0ydeg7Vp4SpkniEqHkcWF+YOJQ8EcM=";
     };
   };
 
@@ -54,8 +55,14 @@ stdenv.mkDerivation {
     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,