Merge pull request #305300 from jcaesar/opensplat

opensplat: init at 1.1.2

maintainers/maintainer-list.nix

@@ -9094,6 +9094,12 @@
     github = "joshua-cooper";
     githubId = 35612334;
   };
+  jcaesar = {
+    name = "Julius Michaelis";
+    matrix = "@julius:mtx.liftm.de";
+    github = "jcaesar";
+    githubId = 1753388;
+  };
   jceb = {
     name = "Jan Christoph Ebersbach";
     email = "jceb@e-jc.de";

pkgs/by-name/op/opensplat/package.nix

@@ -0,0 +1,89 @@
+{
+  lib,
+  stdenv,
+  cmake,
+  ninja,
+  fetchFromGitHub,
+  fetchpatch,
+  python3,
+  opencv,
+  nlohmann_json,
+  nanoflann,
+  glm,
+  cxxopts,
+  config,
+  # Upstream has rocm/hip support, too. anyone?
+  cudaSupport ? config.cudaSupport,
+  cudaPackages,
+  autoAddDriverRunpath,
+}:
+let
+  version = "1.1.2";
+  torch = python3.pkgs.torch.override { inherit cudaSupport; };
+  # Using a normal stdenv with cuda torch gives
+  # ld: /nix/store/k1l7y96gv0nc685cg7i3g43i4icmddzk-python3.11-torch-2.2.1-lib/lib/libc10.so: undefined reference to `std::ios_base_library_init()@GLIBCXX_3.4.32'
+  stdenv' = if cudaSupport then cudaPackages.backendStdenv else stdenv;
+in
+stdenv'.mkDerivation {
+  pname = "opensplat";
+  inherit version;
+
+  src = fetchFromGitHub {
+    owner = "pierotofy";
+    repo = "OpenSplat";
+    rev = "refs/tags/v${version}";
+    hash = "sha256-3tk62b5fSf6wzuc5TwkdfAKgUMrw3ZxetCJa2RVMS/s=";
+  };
+
+  patches = [
+    (fetchpatch {
+      name = "install-executables.patch";
+      url = "https://github.com/pierotofy/OpenSplat/commit/b4c4176819b508978583b7ebf66306171807a8e6.patch";
+      hash = "sha256-BUgPMcO3lt3ZEzv24u36k3aTEIoloOhxrCGi1KQ5Epk=";
+    })
+  ];
+
+  postPatch = ''
+    # the two vendored gsplats are so heavily modified they may be considered a fork
+    find vendor ! -name 'gsplat*' -maxdepth 1 -mindepth 1 -exec rm -rf {} +
+    mkdir vendor/{nanoflann,glm}
+    ln -s ${glm}/include/glm vendor/glm/glm
+    ln -s ${nanoflann}/include/nanoflann.hpp vendor/nanoflann/nanoflann.hpp
+    ln -s ${nlohmann_json}/include/nlohmann vendor/json
+    ln -s ${cxxopts}/include/cxxopts.hpp vendor/cxxopts.hpp
+  '';
+
+  nativeBuildInputs = [
+    cmake
+    ninja
+  ] ++ lib.optionals cudaSupport [
+    cudaPackages.cuda_nvcc
+    autoAddDriverRunpath
+  ];
+
+  buildInputs = [
+    nlohmann_json
+    torch.cxxdev
+    torch
+    opencv
+  ] ++ lib.optionals cudaSupport [
+    cudaPackages.cuda_cudart
+  ];
+
+  env.TORCH_CUDA_ARCH_LIST = "${lib.concatStringsSep ";" python3.pkgs.torch.cudaCapabilities}";
+
+  cmakeFlags = [
+    (lib.cmakeBool "CMAKE_SKIP_RPATH" true)
+  ] ++ lib.optionals cudaSupport [
+    (lib.cmakeFeature "GPU_RUNTIME" "CUDA")
+    (lib.cmakeFeature "CUDA_TOOLKIT_ROOT_DIR" "${cudaPackages.cudatoolkit}/")
+  ];
+
+  meta = {
+    description = "Production-grade 3D gaussian splatting";
+    homepage = "https://github.com/pierotofy/OpenSplat/";
+    license = lib.licenses.mit;
+    maintainers = [ lib.maintainers.jcaesar ];
+    platforms = lib.platforms.linux ++ lib.optionals (!cudaSupport) lib.platforms.darwin;
+  };
+}

pkgs/top-level/all-packages.nix

@@ -15324,6 +15324,8 @@ with pkgs;
   colmap = libsForQt5.callPackage ../applications/science/misc/colmap { inherit (config) cudaSupport; };
   colmapWithCuda = colmap.override { cudaSupport = true; };
 
+  opensplatWithCuda = opensplat.override { cudaSupport = true; };
+
   chickenPackages_4 = recurseIntoAttrs (callPackage ../development/compilers/chicken/4 { });
   chickenPackages_5 = recurseIntoAttrs (callPackage ../development/compilers/chicken/5 { });
   chickenPackages = dontRecurseIntoAttrs chickenPackages_5;