https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122280
--- Comment #15 from Benjamin Schulz <schulz.benjamin at googlemail dot com> --- The question is whether this is a gcc problem, or an nvidia-cuda problem or whether my card is failing or two or three of these options... I have no trust on nvidia in this respect. Their latest release notes point to trying to fix matrix multiplication in cuda... https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html If they officially have errors in such a simple computation, then one must assume their code is a mess. And that valgrind output may point to something like this, if its not just the wrong sm file...
