Full error output: cuda-glibc-2.41-cmake-errors.txt · GitHub These trigonometric functions are new additions to Glibc’s math headers: https://www.phoronix.com/news/Glibc-2.41-More-C23

For the record, this naive patch fixes my problem on Gentoo ~amd64 (where downgrading glibc is not possible): diff '--color=auto' -ur builds.orig/cuda_nvcc/targets/x86_64-linux/include/crt/math_functions.h builds/cuda_nvcc/targets/x86_64-linux/include/crt/math_functions.h --- a/builds.orig/cuda_nvc…

The patch from stefan works for me. Looks like CUDAs headers have a literal bug, where the extern definition doesn’t match the actual function definition. His patch just makes them match. NVIDIA should probably make this change official in every architecture variant of math_functions.h.

Hello, I confirm the problem with Ubuntu 25.04! The compilation of my program fails as soon as the cmake. After modifying /usr/local/cuda-12.8/targets/x86_64-linux/include/crt/math_functions.h everything works again!

I found this post right after after asking the same question on StackOverflow (and right after hitting the same problem on my machine). I adapted the patch slightly - changing the a/ and b/ parts not to assume the same directory structure. Also gave you credit. NVCC compilation error: exception spe…

I work with Fedora 42, clang 18.1.8, CUDA 12.8. For me, I needed to add some noexcepts to math_functions.hpp as well for CUDA compilation to work.

This worked for me too on Fedora 42 , nvcc version Build cuda_12.9.r12.9/compiler.35813241_0 The workarounds involved Installing the Fedora 41 version of the CUDA toolkit (this mostly worked) Install the gcc14 “sudo dnf5 install gcc14 gcc14-c++” Edit the file /usr/local/cuda-12.9/targets…

i ran this: sudo cmake .. -DGMX_GPU=CUDA -DCMAKE_C_COMPILER=gcc-11 -DCMAKE_CXX_COMPILER=g+±11 -DGMX_BUILD_OWN=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda – The C compiler identification is GNU 11.5.0 – The CXX compiler identification is GNU 11.5.0 – Detecting C compiler ABI info – Detecting C com…

@stefantalpalaru Could you help me the issue occurred when I try to build JAX with local CUDA and local Clang? According to this comment , it’s a compatibility problem between glibc 2.42 and CUDNN 9.14.0 although that Arch maintainers use this patch to make CUDA 13 compatible with glibc 2.42.

"error: exception specification is incompatible" for cospi/sinpi/cospif/sinpif with glibc-2.41

Accelerated Computing CUDA CUDA NVCC Compiler

Robert_Crovella February 13, 2025, 3:40am 2

glibc 2.41 is not supported for any version of CUDA at the moment. (e.g. CUDA 12.8)

Topic		Replies	Views
Cuda headers in crt/math_functions.h still broken in debian-13 repo CUDA NVCC Compiler linux , llama	1	173	March 9, 2026
CUDA 12.x crt/math_functions.h: noexcept redeclaration mismatch with modern glibc (2.38+) in C++17 NVIDIA Nemotron cuda , kernel , linux , fedora	0	124	May 30, 2026
Nvcc cannot compile the math_functions.h shipped along with it CUDA Programming and Performance	1	1044	March 11, 2025
CUDA headers in crt/math_functions.h still broken in debian-13 repo Linux llama	1	144	March 9, 2026
Fedora 43 and NVCC / Cuda13.1 error "exception specification is incompatible" rsqrt / rsqrtf CUDA Setup and Installation	9	2384	March 13, 2026
Nvc/nvc++ miscompiles if cosf/sinf is called nvc, nvc++ and nvfortran	3	594	October 12, 2022
sinpi and cospi in global namespace CUDA Programming and Performance	1	1178	June 11, 2011
[Request] Add NVCC compatibility with glibc 2.26 CUDA Programming and Performance	16	8240	November 12, 2019
Nvcc 12.3 with gcc 13.2 not working CUDA NVCC Compiler	10	11696	March 11, 2024
NVCC fails if C++11 activated and certain headers included CUDA Programming and Performance	6	7710	November 12, 2015

"error: exception specification is incompatible" for cospi/sinpi/cospif/sinpif with glibc-2.41

Related topics