Replies: 1 comment
-
Most likely you did not specify a default TARGET to use for all the common code in OpenBLAS, so the base set of compiler options already contained something like |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Version 0.3.27.
I compiled with dynamic_arch true flag on an Intel machine with AVX-512 support. My app statically linked in openblas.a. When running on an AMD zen3 machine, it got coredump/SIGILL at an AVX-512 instruction like vmovdqa32 or vmovdaq64 inside sgemm_incopy_ZEN() called from sgemm_tn(). Why the dynamic CPU detection did not choose correct kernel to use on AMD zen3? I verified that at initialization step, support_avx512() correctly returned 0 on zen3 and OPENBLAS_VERBOSE shows core detected is "Zen".
Thanks
Beta Was this translation helpful? Give feedback.
All reactions