Public bug reported:
The stock nvidia-cuda-toolkit package installs nvcc 12.0.140 which is
supposed to be compatible with GCC 13, but GCC 13.2.1 introduced a
feature (an Intel AMX intrinsic) that breaks many codes, including
something as simple as C++17 parallel STL. The exact mechanism is
detailed here: https://github.com/ggml-
org/llama.cpp/issues/9473#issuecomment-2350668752
Any CUDA code where the compiler just sees #include <execution> fails to
compile with error: identifier "__builtin_ia32_ldtilecfg" is undefined.
Description: Ubuntu 24.04.2 LTS
Release: 24.04
nvidia-cuda-toolkit:
Telepítve: 12.0.140~12.0.1-4build4
Jelölt: 12.0.140~12.0.1-4build4
Repro:
apt install nvidia-cuda-toolkit g++
main.cu:
#include <execution>
int main() {}
Issue nvcc ./main.cu
Marvel at compiler error.
** Affects: nvidia-cuda-toolkit (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2112548
Title:
stock cuda-toolkit with nvcc 12.0.140 incompatible with stock gcc 13.3
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/nvidia-cuda-toolkit/+bug/2112548/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
