https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82004
--- Comment #31 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Andrey, can you still reproduce and if yes, see what you get for the sw_absorption.F90 loop after PRE (i.e. if the value of __builtin_pow (10, ...) is precomputed for the first iteration and pow moved to latch or not)? Thanks.