Potential bug in stmt.c (expand_value_return)

[Joern Rennecke]

> No, it wasn't accidental.  The information is available instead
> from EXIT_BLOCK_PTR->global_live_at_start.

That's not a conventient place when you are looking at a sucesion of insns
with a recursive function that gets passed just the piece of rtl.

> What optimizations need the actual use insn?

I can't search through the whole compiler now, but I know that
reload_combine needs it, and the register scavenging for out-of
range branchs on the sh (config/sh/sh.c:gen_block_redirect).

USEs are also much easier to digest for peepholes than looking into
a global array when you hit a return insn.