We have been told by PPC architecture experts that on at least some PPC implementations, a cmp/branch forward to unconditional trap is likely to be a better sequence for null checks on PPC (and possibly for array bounds checks as well) because the hardware may do a better job of branch prediction than trap prediction. Implement this alternative and empirically evaluate cmp/branch vs traps. We use a cmp/branch sequence on IA32. Be sure to look at OPT_FinalMIRExpansion on IA32 to see how to handle the GC maps.