https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107160
Kewen Lin <linkw at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Last reconfirmed| |2022-10-12 Ever confirmed|0 |1 --- Comment #6 from Kewen Lin <linkw at gcc dot gnu.org> --- Confirmed. The option can be reduced to: -m64 -O3 -mcpu=power8 -ffast-math The culprit loop is the one in function l1_norm of file vector.cc (actually it's from the header file). The resulting gimple after loop vect pass looks weird to me, it seems not to expose one issue related to math but one actual vect issue. I'll check it further and post one reduced test case.