https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126195
--- Comment #4 from Haochen Jiang <haochen.jiang at intel dot com> --- (In reply to Drea Pinski from comment #3) > Maybe this is not wrong code reading this again. > > But rather just bad luck. > I also slightly lean to this. It is just several 1-2 ULP diff for calculation, but leading to a totally different code path because of the choice of the nodes in mesh, unluckily. That is why I suppose in the original description, one choice is to raise it to SPEC committee to further limit 737.gmsh_r optimization options.
