nixpkgs/pkgs/tools/system/nvtop/default.nix
Josef Kemetmüller 0996ac6b0e nvtop: Make independent of Nvidia driver version
For packages relying on Nvidia drivers we don't want to hard-code the
location of a particular driver version in the RPATH. Otherwise the
software might fail with errors such as:

> Impossible to initialize nvidia nvml : Driver/library version mismatch

To fix this the usual pattern in nixpkgs is to add `/run/opengl-driver/lib`
to the RPATH of these binaries.
This is the global location where Nvidia libraries are available on
NixOS and so this makes sure that the nvtop binary can be run with all
versions of the driver.

At build time however these libraries are not available at that
particular location. For precisely this use case, Nvidia ships stub
versions of the driver libraries with the cudatoolkit. So we just need
to point CMake to that directory.

Fixes: #98328
2020-09-21 08:43:43 +02:00

34 lines
908 B
Nix

{ stdenv, fetchFromGitHub, cmake, cudatoolkit, ncurses, addOpenGLRunpath }:
stdenv.mkDerivation rec {
pname = "nvtop";
version = "1.0.0";
src = fetchFromGitHub {
owner = "Syllo";
repo = "nvtop";
rev = version;
sha256 = "1b6yz54xddip1r0k8cbqg41dpyhds18fj29bj3yf40xvysklb0f4";
};
cmakeFlags = [
"-DNVML_INCLUDE_DIRS=${cudatoolkit}/include"
"-DNVML_LIBRARIES=${cudatoolkit}/targets/x86_64-linux/lib/stubs/libnvidia-ml.so"
"-DCMAKE_BUILD_TYPE=Release"
];
nativeBuildInputs = [ cmake addOpenGLRunpath ];
buildInputs = [ ncurses cudatoolkit ];
postFixup = ''
addOpenGLRunpath $out/bin/nvtop
'';
meta = with stdenv.lib; {
description = "A (h)top like task monitor for NVIDIA GPUs";
homepage = "https://github.com/Syllo/nvtop";
license = licenses.gpl3;
platforms = platforms.linux;
maintainers = with maintainers; [ willibutz ];
};
}