64-bit PowerPC code quality regression

[David Edelsohn]

>>>>> Richard Henderson writes:

Richard> You mean an insn match?  Because they're completely separate 
Richard> operations.  Because in addition to splitting "invalid" insns,
Richard> define_split is used for lowering code by splitting compound
Richard> patterns into their constituant parts.

	I clearly didn't not understand the semantics of define_split
properly :^(.  I didn't expect the splitter to be applied when a
define_insn preceded it and matched with a match that was not "#".  Thanks
for clarifying this.

Thanks, David