nixpkgs/pkgs/development/libraries/openimagedenoise/default.nix
Andrew Marshall 333e7f2841 openimagedenoise: Support building with CUDA
It also has HIP, Metal, and SYCL support, but I do not have hardware to
test any of those actually work. Further, Blender 4.1.0 does not enable
GPU OIDN on AMD (HIP), but may in 4.2.0 it seems.
2024-03-26 20:02:21 -04:00

54 lines
1.3 KiB
Nix

{
cmake,
config,
cudaPackages,
cudaSupport ? config.cudaSupport,
fetchzip,
ispc,
lib,
python3,
stdenv,
tbb,
}:
stdenv.mkDerivation rec {
pname = "openimagedenoise";
version = "2.2.2";
# The release tarballs include pretrained weights, which would otherwise need to be fetched with git-lfs
src = fetchzip {
url = "https://github.com/OpenImageDenoise/oidn/releases/download/v${version}/oidn-${version}.src.tar.gz";
sha256 = "sha256-ZIrs4oEb+PzdMh2x2BUFXKyu/HBlFb3CJX24ciEHy3Q=";
};
patches = lib.optional cudaSupport ./cuda.patch;
nativeBuildInputs = [
cmake
python3
ispc
] ++ lib.optional cudaSupport cudaPackages.cuda_nvcc;
buildInputs =
[ tbb ]
++ lib.optionals cudaSupport [
cudaPackages.cuda_cudart
cudaPackages.cuda_cccl
];
cmakeFlags = [
(lib.cmakeBool "OIDN_DEVICE_CUDA" cudaSupport)
(lib.cmakeFeature "TBB_INCLUDE_DIR" "${tbb.dev}/include")
(lib.cmakeFeature "TBB_ROOT" "${tbb}")
];
meta = with lib; {
homepage = "https://openimagedenoise.github.io";
description = "High-Performance Denoising Library for Ray Tracing";
license = licenses.asl20;
maintainers = [ maintainers.leshainc ];
platforms = platforms.unix;
changelog = "https://github.com/OpenImageDenoise/oidn/blob/v${version}/CHANGELOG.md";
};
}