Emerging llvm-roc with an AMD/Nvidia dual gpu system causes access violations during configure: F: open_wr S: deny P: /dev/nvidiactl A: /dev/nvidiactl R: /dev/nvidiactl C: /tmp/portage/sys-devel/llvm-roc-4.1.1/work/llvm-roc-4.1.1_build/CMakeFiles/CMakeTmp/cmTC_d396e Reproducible: Always Steps to Reproduce: 1. Have nvidiactl available 2. emerge llvm-roc
Created attachment 707604 [details] emerge --info
Created attachment 707607 [details] build.log
Looks like same clang had to deal with (bug #760645) I only have 1 gpu (nvidia), but this seems to work for me (otherwise I get access violations as well). MYCMAKEARGS=-DCMAKE_DISABLE_FIND_PACKAGE_CUDA=ON emerge -1 llvm-roc
(In reply to Ionen Wolkens from comment #3) > Looks like same clang had to deal with (bug #760645) > > I only have 1 gpu (nvidia), but this seems to work for me (otherwise I get > access violations as well). > > MYCMAKEARGS=-DCMAKE_DISABLE_FIND_PACKAGE_CUDA=ON emerge -1 llvm-roc configure suceeds with that option. Don't know if everything works as it should, though.
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=f32ed0e9c59158047fb10f8e8c0d27d232f015b8 commit f32ed0e9c59158047fb10f8e8c0d27d232f015b8 Author: Craig Andrews <candrews@gentoo.org> AuthorDate: 2021-05-18 15:05:45 +0000 Commit: Craig Andrews <candrews@gentoo.org> CommitDate: 2021-05-18 15:06:17 +0000 sys-devel/llvm-roc: Disable CUDA target autodetection This package is for AMD ROC, so Nvidia CUDA detection is irrelevant. Closes: https://bugs.gentoo.org/789693 Package-Manager: Portage-3.0.18, Repoman-3.0.3 Signed-off-by: Craig Andrews <candrews@gentoo.org> sys-devel/llvm-roc/llvm-roc-4.1.1.ebuild | 1 + 1 file changed, 1 insertion(+)