https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121488
--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #6) > (In reply to Richard Biener from comment #5) > > Fixed on trunk. > > Thanks for the fix, the testcase from PR 92645 looks much improved now. > > There is still one case which seems to be missed: > Let me file this as a seperate bug. I see what happens now, it is basically PR 121405 but the need to take 2 fre to do the optimization. We only get that form of what PR 121405 looks like originally at fre5 and that produces the BIT_FIELD_REF but we don't have another fre afterwards to fix comment #5.