https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58517
--- Comment #17 from Oleg Endo <olegendo at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #16) > Now I am curious if T_REG should be BImode rather than SImode ... Then > ifcvt.cc would not have to be modified. I Know BImode is newer than when sh > target was added but maybe if someone cares about the sh target could try to > modify the backend to use BImode for T_REG here. Yeah, I know. Back then I was entertaining the idea. But T-bit being SImode is so intertwined with all other things in the backend, it looks like a big job to undo/redo that. For example there are patterns that use the T bit in other arithmetic insns patterns. Would need to check what would happen to those. Sounds like a can of worms.
