https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115740
--- Comment #6 from Lockal <lockalsash at gmail dot com> --- Oops, did not expect that someone already reported this to llvm/llvm-project. This was noticed as an issue of ROCm, not CUDA. Did not try CUDA, probably won't work with gcc-14 due to multiple reasons (starting with version check in preprocessor). But I think it has nothing to do with LLVM: 1) I did not enable -D_GLIBCXX_ASSERTIONS - I don't want see calls to __glibcxx_assert_fail (device code can't comprehend this) 2) Specifically clang dislikes calls to host code from device code