https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120250
--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Roger Sayle from comment #7) > I agree with Jakub that my combine.cc change never considered the case where > the constant to be placed in the constant pool is the destination of a > computed jump (and that this might interfere with computed_jump_p). The > suggested "SET_DEST (pat) != pc_rtx" clause looks like a reasonable and safe > fix. And that I think will fix PR 123397 where the ICE is slightly different but we are converting a computed goto into a direct jump to a function.
