onetbb: always use hwloc built without CUDA support (#480524)
This commit is contained in:
@@ -1,15 +1,20 @@
|
||||
{
|
||||
args@{
|
||||
lib,
|
||||
stdenv,
|
||||
fetchFromGitHub,
|
||||
fetchpatch,
|
||||
cmake,
|
||||
hwloc,
|
||||
hwloc, # Purposefully shadowed below
|
||||
ninja,
|
||||
pkg-config,
|
||||
ctestCheckHook,
|
||||
}:
|
||||
|
||||
let
|
||||
# The behavior of OneTBB does not change if it is built with hwloc with support for CUDA.
|
||||
# However, the derivation *does* change, causing rebuilds of packages like Nix.
|
||||
# To avoid these pointless rebuilds, we make sure to always use a version of hwloc with CUDA
|
||||
# support disabled.
|
||||
hwloc = args.hwloc.override { enableCuda = false; };
|
||||
in
|
||||
stdenv.mkDerivation (finalAttrs: {
|
||||
pname = "onetbb";
|
||||
version = "2022.3.0";
|
||||
|
||||
Reference in New Issue
Block a user