CUDA-related updates and cleanups

pkgs/applications/misc/blender/default.nix

@@ -1,4 +1,4 @@
-{ stdenv, stdenv_gcc5, lib, fetchurl, boost, cmake, ffmpeg, gettext, glew
+{ stdenv, lib, fetchurl, boost, cmake, ffmpeg, gettext, glew
 , ilmbase, libXi, libX11, libXext, libXrender
 , libjpeg, libpng, libsamplerate, libsndfile
 , libtiff, mesa, openal, opencolorio, openexr, openimageio, openjpeg_1, python
@@ -10,7 +10,7 @@
 
 with lib;
 
-(if cudaSupport then stdenv_gcc5 else stdenv).mkDerivation rec {
+stdenv.mkDerivation rec {
   name = "blender-2.79";
 
   src = fetchurl {
@@ -57,9 +57,8 @@ with lib;
     ++ optional jackaudioSupport "-DWITH_JACK=ON"
     ++ optionals cudaSupport
       [ "-DWITH_CYCLES_CUDA_BINARIES=ON"
-        # Disable the sm_20 architecture to work around a segfault in
-        # ptxas, as suggested on #blendercoders.
-        "-DCYCLES_CUDA_BINARIES_ARCH=sm_21;sm_30;sm_35;sm_37;sm_50;sm_52;sm_60;sm_61"
+        # Disable architectures before sm_30 to support new CUDA toolkits.
+        "-DCYCLES_CUDA_BINARIES_ARCH=sm_30;sm_35;sm_37;sm_50;sm_52;sm_60;sm_61"
       ]
     ++ optional colladaSupport "-DWITH_OPENCOLLADA=ON";
 

pkgs/applications/science/math/caffe/default.nix

@@ -22,22 +22,23 @@ assert pythonSupport -> (python != null && numpy != null);
 
 stdenv.mkDerivation rec {
   name = "caffe-${version}";
-  version = "1.0-rc5";
+  version = "1.0";
 
   src = fetchFromGitHub {
     owner = "BVLC";
     repo = "caffe";
-    rev = "rc5";
-    sha256 = "0lfmmc0n6xvkpygvxclzrvd0zigb4yfc5612anv2ahlxpfi9031c";
+    rev = version;
+    sha256 = "104jp3cm823i3cdph7hgsnj6l77ygbwsy35mdmzhmsi4jxprd9j3";
   };
 
   enableParallelBuilding = true;
 
   nativeBuildInputs = [ cmake doxygen ];
 
-  cmakeFlags = [ "-DCUDA_ARCH_NAME=All" ]
-               ++ lib.optional (!cudaSupport) "-DCPU_ONLY=ON"
-               ++ lib.optional (!pythonSupport) "-DBUILD_python=OFF";
+  cmakeFlags = [
+    "-DCUDA_ARCH_NAME=All"
+    (if pythonSupport then "-Dpython_version=${python.version}" else "-DBUILD_python=OFF")
+  ] ++ lib.optional (!cudaSupport) "-DCPU_ONLY=ON";
 
   buildInputs = [ boost google-gflags glog protobuf hdf5-cpp lmdb leveldb snappy opencv atlas ]
                 ++ lib.optional cudaSupport cudatoolkit
@@ -49,6 +50,16 @@ stdenv.mkDerivation rec {
   outputs = [ "bin" "out"];
   propagatedBuildOutputs = []; # otherwise propagates out -> bin cycle
 
+  preConfigure = lib.optionalString (cudaSupport && lib.versionAtLeast cudatoolkit.version "9.0") ''
+    # CUDA 9.0 doesn't support sm_20
+    sed -i 's,20 21(20) ,,' cmake/Cuda.cmake
+  '' + lib.optionalString (python.isPy3 or false) ''
+    sed -i \
+      -e 's,"python-py''${boost_py_version}",python3,g' \
+      -e 's,''${Boost_PYTHON-PY''${boost_py_version}_FOUND},''${Boost_PYTHON3_FOUND},g' \
+      cmake/Dependencies.cmake
+  '';
+
   postInstall = ''
     # Internal static library.
     rm $out/lib/libproto.a

pkgs/development/compilers/cudatoolkit/default.nix

@@ -1,30 +1,44 @@
-{ lib, stdenv, fetchurl, patchelf, perl, ncurses, expat, python27, zlib
+{ lib, stdenv, makeWrapper, fetchurl, requireFile, patchelf, perl, ncurses, expat, python27, zlib
+, gcc48, gcc49, gcc5, gcc6
 , xorg, gtk2, glib, fontconfig, freetype, unixODBC, alsaLib, glibc
 }:
 
 let
 
   common =
-    { version, url, sha256
+    args@{ gcc, version, sha256
+    , url ? ""
+    , name ? ""
+    , developerProgram ? false
     , python ? python27
     }:
 
     stdenv.mkDerivation rec {
       name = "cudatoolkit-${version}";
+      inherit (args) version;
 
       dontPatchELF = true;
       dontStrip = true;
 
       src =
-        if stdenv.system == "x86_64-linux" then
-          fetchurl {
-            inherit url sha256;
+        if developerProgram then
+          requireFile {
+            message = ''
+              This nix expression requires that ${args.name} is already part of the store.
+              Register yourself to NVIDIA Accelerated Computing Developer Program, retrieve the CUDA toolkit
+              at https://developer.nvidia.com/cuda-toolkit, and run the following command in the download directory:
+              nix-prefetch-url file://${args.name}
+            '';
+            inherit (args) name sha256;
           }
-        else throw "cudatoolkit does not support platform ${stdenv.system}";
+        else
+          fetchurl {
+            inherit (args) url sha256;
+          };
 
       outputs = [ "out" "lib" "doc" ];
 
-      buildInputs = [ perl ];
+      nativeBuildInputs = [ perl makeWrapper ];
 
       runtimeDependencies = [
         ncurses expat python zlib glibc
@@ -37,8 +51,8 @@ let
       unpackPhase = ''
         sh $src --keep --noexec
         cd pkg/run_files
-        sh cuda-linux64-rel-${version}-*.run --keep --noexec
-        sh cuda-samples-linux-${version}-*.run --keep --noexec
+        sh cuda-linux*.run --keep --noexec
+        sh cuda-samples*.run --keep --noexec
         cd pkg
       '';
 
@@ -92,15 +106,25 @@ let
         # Remove OpenCL libraries as they are provided by ocl-icd and driver.
         rm -f $out/lib64/libOpenCL*
 
+        # Set compiler for NVCC.
+        wrapProgram $out/bin/nvcc \
+          --prefix PATH : ${gcc}/bin
       '' + lib.optionalString (lib.versionOlder version "8.0") ''
         # Hack to fix building against recent Glibc/GCC.
         echo "NIX_CFLAGS_COMPILE+=' -D_FORCE_INLINES'" >> $out/nix-support/setup-hook
       '';
 
+      passthru = {
+        cc = gcc;
+        majorVersion =
+          let versionParts = lib.splitString "." version;
+          in "${lib.elemAt versionParts 0}.${lib.elemAt versionParts 1}";
+      };
+
       meta = with stdenv.lib; {
         description = "A compiler for NVIDIA GPUs, math libraries, and tools";
-        homepage = https://developer.nvidia.com/cuda-toolkit;
-        platforms = platforms.linux;
+        homepage = "https://developer.nvidia.com/cuda-toolkit";
+        platforms = [ "x86_64-linux" ];
         license = licenses.unfree;
       };
     };
@@ -109,32 +133,44 @@ in {
 
   cudatoolkit6 = common {
     version = "6.0.37";
-    url = http://developer.download.nvidia.com/compute/cuda/6_0/rel/installers/cuda_6.0.37_linux_64.run;
+    url = "http://developer.download.nvidia.com/compute/cuda/6_0/rel/installers/cuda_6.0.37_linux_64.run";
     sha256 = "991e436c7a6c94ec67cf44204d136adfef87baa3ded270544fa211179779bc40";
+    gcc = gcc48;
   };
 
   cudatoolkit65 = common {
     version = "6.5.19";
-    url = http://developer.download.nvidia.com/compute/cuda/6_5/rel/installers/cuda_6.5.19_linux_64.run;
+    url = "http://developer.download.nvidia.com/compute/cuda/6_5/rel/installers/cuda_6.5.19_linux_64.run";
     sha256 = "1x9zdmk8z784d3d35vr2ak1l4h5v4jfjhpxfi9fl9dvjkcavqyaj";
+    gcc = gcc48;
   };
 
   cudatoolkit7 = common {
     version = "7.0.28";
-    url = http://developer.download.nvidia.com/compute/cuda/7_0/Prod/local_installers/cuda_7.0.28_linux.run;
+    url = "http://developer.download.nvidia.com/compute/cuda/7_0/Prod/local_installers/cuda_7.0.28_linux.run";
     sha256 = "1km5hpiimx11jcazg0h3mjzk220klwahs2vfqhjavpds5ff2wafi";
+    gcc = gcc49;
   };
 
   cudatoolkit75 = common {
     version = "7.5.18";
-    url = http://developer.download.nvidia.com/compute/cuda/7.5/Prod/local_installers/cuda_7.5.18_linux.run;
+    url = "http://developer.download.nvidia.com/compute/cuda/7.5/Prod/local_installers/cuda_7.5.18_linux.run";
     sha256 = "1v2ylzp34ijyhcxyh5p6i0cwawwbbdhni2l5l4qm21s1cx9ish88";
+    gcc = gcc49;
   };
 
   cudatoolkit8 = common {
     version = "8.0.61";
-    url = https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda_8.0.61_375.26_linux-run;
+    url = "https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda_8.0.61_375.26_linux-run";
     sha256 = "1i4xrsqbad283qffvysn88w2pmxzxbbby41lw0j1113z771akv4w";
+    gcc = gcc5;
+  };
+
+  cudatoolkit9 = common {
+    version = "9.0.176";
+    url = "https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda_9.0.176_384.81_linux-run";
+    sha256 = "0308rmmychxfa4inb1ird9bpgfppgr9yrfg1qp0val5azqik91ln";
+    gcc = gcc6;
   };
 
 }

pkgs/development/libraries/opencv/3.x.nix

@@ -16,7 +16,7 @@
 , enableGStreamer ? false, gst_all_1
 , enableEigen     ? true, eigen
 , enableOpenblas  ? true, openblas
-, enableCuda      ? false, cudatoolkit, gcc5
+, enableCuda      ? false, cudatoolkit
 , enableTesseract ? false, tesseract, leptonica
 , AVFoundation, Cocoa, QTKit
 }:
@@ -145,7 +145,7 @@ stdenv.mkDerivation rec {
     # simply enabled automatically if contrib is built, and it detects
     # tesseract & leptonica.
     ++ lib.optionals enableTesseract [ tesseract leptonica ]
-    ++ lib.optionals enableCuda [ cudatoolkit gcc5 ]
+    ++ lib.optional enableCuda cudatoolkit
     ++ lib.optional buildContrib protobuf
     ++ lib.optionals stdenv.isDarwin [ AVFoundation Cocoa QTKit ];
 
@@ -165,8 +165,10 @@ stdenv.mkDerivation rec {
     (opencvFlag "OPENEXR" enableEXR)
     (opencvFlag "CUDA" enableCuda)
     (opencvFlag "CUBLAS" enableCuda)
-  ] ++ lib.optionals enableCuda [ "-DCUDA_FAST_MATH=ON" ]
-    ++ lib.optional buildContrib "-DBUILD_PROTOBUF=off"
+  ] ++ lib.optionals enableCuda [
+    "-DCUDA_FAST_MATH=ON"
+    "-DCUDA_HOST_COMPILER=${cudatoolkit.cc}/bin/gcc"
+  ] ++ lib.optional buildContrib "-DBUILD_PROTOBUF=off"
     ++ lib.optionals stdenv.isDarwin ["-DWITH_OPENCL=OFF" "-DWITH_LAPACK=OFF"];
 
   enableParallelBuilding = true;

pkgs/development/libraries/opensubdiv/default.nix

@@ -1,17 +1,17 @@
-{ lib, stdenv, stdenv_gcc5, fetchurl, fetchFromGitHub, cmake, pkgconfig, xorg, mesa_glu
+{ lib, stdenv, fetchurl, fetchFromGitHub, cmake, pkgconfig, xorg, mesa_glu
 , mesa_noglu, glew, ocl-icd, python3
 , cudaSupport ? false, cudatoolkit
 }:
 
-(if cudaSupport then stdenv_gcc5 else stdenv).mkDerivation rec {
+stdenv.mkDerivation rec {
   name = "opensubdiv-${version}";
-  version = "3.2.0";
+  version = "3.3.0";
 
   src = fetchFromGitHub {
     owner = "PixarAnimationStudios";
     repo = "OpenSubdiv";
     rev = "v${lib.replaceChars ["."] ["_"] version}";
-    sha256 = "0wk12n1s8za3sz8d6bmfm3rfjyx20j48gy1xp57dvbnjvlvzqy3w";
+    sha256 = "0wpjwfik4q9s4r30hndhzmfyzv968mmg5lgng0123l07mn47d2yl";
   };
 
   outputs = [ "out" "dev" ];
@@ -30,7 +30,7 @@
       "-DNO_EXAMPLES=1"
       "-DGLEW_INCLUDE_DIR=${glew.dev}/include"
       "-DGLEW_LIBRARY=${glew.dev}/lib"
-    ];
+    ] ++ lib.optional cudaSupport "-DOSD_CUDA_NVCC_FLAGS=--gpu-architecture=compute_30";
 
   enableParallelBuilding = true;