Merge staging-next into staging

pkgs/development/libraries/science/math/cudnn/generic.nix

@@ -8,6 +8,14 @@
 , cudatoolkit
 , fetchurl
 , addOpenGLRunpath
+, # The distributed version of CUDNN includes both dynamically liked .so files,
+  # as well as statically linked .a files.  However, CUDNN is quite large
+  # (multiple gigabytes), so you can save some space in your nix store by
+  # removing the statically linked libraries if you are not using them.
+  #
+  # Setting this to true removes the statically linked .a files.
+  # Setting this to false keeps these statically linked .a files.
+  removeStatic ? false
 }:
 
 stdenv.mkDerivation {
@@ -23,6 +31,8 @@ stdenv.mkDerivation {
   nativeBuildInputs = [ addOpenGLRunpath ];
 
   installPhase = ''
+    runHook preInstall
+
     function fixRunPath {
       p=$(patchelf --print-rpath $1)
       patchelf --set-rpath "''${p:+$p:}${lib.makeLibraryPath [ stdenv.cc.cc ]}:\$ORIGIN/" $1
@@ -35,6 +45,10 @@ stdenv.mkDerivation {
     mkdir -p $out
     cp -a include $out/include
     cp -a lib64 $out/lib64
+  '' + lib.optionalString removeStatic ''
+    rm -f $out/lib64/*.a
+  '' + ''
+    runHook postInstall
   '';
 
   # Set RUNPATH so that libcuda in /run/opengl-driver(-32)/lib can be found.

pkgs/development/libraries/science/math/cutensor/default.nix

@@ -0,0 +1,37 @@
+{ callPackage
+, cudatoolkit_10_1, cudatoolkit_10_2
+, cudatoolkit_11_0, cudatoolkit_11_1, cudatoolkit_11_2
+}:
+
+rec {
+  cutensor_cudatoolkit_10_1 = callPackage ./generic.nix rec {
+    version = "1.2.2.5";
+    libPath = "lib/10.1";
+    cudatoolkit = cudatoolkit_10_1;
+    # 1.2.2 is compatible with CUDA 11.0, 11.1, and 11.2:
+    # ephemeral doc at https://developer.nvidia.com/cutensor/downloads
+    sha256 = "1dl9bd71frhac9cb8lvnh71zfsnqxbxbfhndvva2zf6nh0my4klm";
+  };
+
+  cutensor_cudatoolkit_10_2 = cutensor_cudatoolkit_10_1.override {
+    libPath = "lib/10.2";
+    cudatoolkit = cudatoolkit_10_2;
+  };
+
+  cutensor_cudatoolkit_10 = cutensor_cudatoolkit_10_2;
+
+  cutensor_cudatoolkit_11_0 = cutensor_cudatoolkit_10_2.override {
+    libPath = "lib/11";
+    cudatoolkit = cudatoolkit_11_0;
+  };
+
+  cutensor_cudatoolkit_11_1 = cutensor_cudatoolkit_11_0.override {
+    cudatoolkit = cudatoolkit_11_1;
+  };
+
+  cutensor_cudatoolkit_11_2 = cutensor_cudatoolkit_11_0.override {
+    cudatoolkit = cudatoolkit_11_2;
+  };
+
+  cutensor_cudatoolkit_11 = cutensor_cudatoolkit_11_2;
+}

pkgs/development/libraries/science/math/cutensor/generic.nix

@@ -0,0 +1,69 @@
+{ stdenv
+, lib
+, libPath
+, cudatoolkit
+, fetchurl
+, autoPatchelfHook
+, addOpenGLRunpath
+
+, version
+, sha256
+}:
+
+let
+  mostOfVersion = builtins.concatStringsSep "."
+    (lib.take 3 (lib.versions.splitVersion version));
+in
+
+stdenv.mkDerivation {
+  pname = "cudatoolkit-${cudatoolkit.majorVersion}-cutensor";
+  inherit version;
+
+  src = fetchurl {
+    url = "https://developer.download.nvidia.com/compute/cutensor/${mostOfVersion}/local_installers/libcutensor-${stdenv.hostPlatform.parsed.kernel.name}-${stdenv.hostPlatform.parsed.cpu.name}-${version}.tar.gz";
+    inherit sha256;
+  };
+
+  outputs = [ "out" "dev" ];
+
+  nativeBuildInputs = [
+    autoPatchelfHook
+    addOpenGLRunpath
+  ];
+
+  buildInputs = [
+    stdenv.cc.cc.lib
+  ];
+
+  propagatedBuildInputs = [
+    cudatoolkit
+  ];
+
+  # Set RUNPATH so that libcuda in /run/opengl-driver(-32)/lib can be found.
+  # See the explanation in addOpenGLRunpath.
+  installPhase = ''
+    mkdir -p "$out" "$dev"
+    mv include "$dev"
+    mv ${libPath} "$out/lib"
+
+    function finalRPathFixups {
+      for lib in $out/lib/lib*.so; do
+        addOpenGLRunpath $lib
+      done
+    }
+    postFixupHooks+=(finalRPathFixups)
+  '';
+
+  passthru = {
+    inherit cudatoolkit;
+    majorVersion = lib.versions.major version;
+  };
+
+  meta = with lib; {
+    description = "cuTENSOR: A High-Performance CUDA Library For Tensor Primitives";
+    homepage = "https://developer.nvidia.com/cutensor";
+    license = licenses.unfree;
+    platforms = [ "x86_64-linux" ];
+    maintainers = with maintainers; [ obsidian-systems-maintenance ];
+  };
+}

pkgs/development/libraries/science/math/liblapack/default.nix

@@ -10,7 +10,7 @@ let
   version = "3.9.0";
 in
 
-stdenv.mkDerivation {
+stdenv.mkDerivation rec {
   pname = "liblapack";
   inherit version;
 
@@ -23,6 +23,9 @@ stdenv.mkDerivation {
 
   nativeBuildInputs = [ gfortran cmake ];
 
+  # Configure stage fails on aarch64-darwin otherwise, due to either clang 11 or gfortran 10.
+  hardeningDisable = lib.optionals (stdenv.isDarwin && stdenv.isAarch64) [ "stackprotector" ];
+
   cmakeFlags = [
     "-DCMAKE_Fortran_FLAGS=-fPIC"
     "-DLAPACKE=ON"
@@ -33,6 +36,26 @@ stdenv.mkDerivation {
 
   doCheck = true;
 
+  # Some CBLAS related tests fail on Darwin:
+  #  14 - CBLAS-xscblat2 (Failed)
+  #  15 - CBLAS-xscblat3 (Failed)
+  #  17 - CBLAS-xdcblat2 (Failed)
+  #  18 - CBLAS-xdcblat3 (Failed)
+  #  20 - CBLAS-xccblat2 (Failed)
+  #  21 - CBLAS-xccblat3 (Failed)
+  #  23 - CBLAS-xzcblat2 (Failed)
+  #  24 - CBLAS-xzcblat3 (Failed)
+  #
+  # Upstream issue to track:
+  # * https://github.com/Reference-LAPACK/lapack/issues/440
+  ctestArgs = lib.optionalString stdenv.isDarwin "-E '^(CBLAS-(x[sdcz]cblat[23]))$'";
+
+  checkPhase = ''
+    runHook preCheck
+    ctest ${ctestArgs}
+    runHook postCheck
+  '';
+
   meta = with lib; {
     inherit version;
     description = "Linear Algebra PACKage";

pkgs/development/libraries/science/math/libtorch/bin.nix

@@ -8,8 +8,8 @@
 , fixDarwinDylibNames
 
 , cudaSupport
-, cudatoolkit_10_2
-, cudnn_cudatoolkit_10_2
+, cudatoolkit_11_1
+, cudnn_cudatoolkit_11_1
 }:
 
 let
@@ -38,7 +38,7 @@ in stdenv.mkDerivation {
 
   installPhase = ''
     # Copy headers and CMake files.
-    install -Dm755 -t $dev/lib lib/*.a
+    mkdir -p $dev
     cp -r include $dev
     cp -r share $dev
 
@@ -109,8 +109,8 @@ in stdenv.mkDerivation {
 
   passthru.tests.cmake = callPackage ./test {
     inherit cudaSupport;
-    cudatoolkit = cudatoolkit_10_2;
-    cudnn = cudnn_cudatoolkit_10_2;
+    cudatoolkit = cudatoolkit_11_1;
+    cudnn = cudnn_cudatoolkit_11_1;
   };
 
   meta = with lib; {

pkgs/development/libraries/science/math/libtorch/binary-hashes.nix

@@ -8,7 +8,7 @@ version: {
     hash = "sha256-xBaNyI7eiQnSArHMITonrQQLZnZCZK/SWKOTWnxzdpc=";
   };
   x86_64-linux-cuda = {
-    url = "https://download.pytorch.org/libtorch/cu102/libtorch-cxx11-abi-shared-with-deps-${version}.zip";
-    hash = "sha256-rNEyE4+jfeX7cU0aNYd5b0pZGYT0PNPnDnS1PIsrMeM=";
+    url = "https://download.pytorch.org/libtorch/cu111/libtorch-cxx11-abi-shared-with-deps-${version}%2Bcu111.zip";
+    hash = "sha256-uQ7ptOuzowJ0JSPIvJHyNotBfpsqAnxpMDLq7Vl6L00=";
   };
 }

pkgs/development/libraries/science/math/primesieve/default.nix

@@ -2,13 +2,13 @@
 
 stdenv.mkDerivation rec {
   pname = "primesieve";
-  version = "7.5";
+  version = "7.6";
 
   nativeBuildInputs = [cmake];
 
   src = fetchurl {
     url = "https://github.com/kimwalisch/primesieve/archive/v${version}.tar.gz";
-    sha256 = "0g60br3p8di92jx3pr2bb51xh15gg57l7qvwzwn7xf7l585hgi7v";
+    sha256 = "sha256-SFZp6Pmmx05SiUfSdN9wXxPKrydtRg0PA3uNvAycCpk=";
   };
 
   meta = with lib; {