> I think that this is the actual bug. The clobber insn should > depend on the call insn, and not get scheduled before it. Even for call-clobbered registers ? -- Eric Botcazou