optimization bug

[Jim Wilson]

I think combine never should have created the USE insn in the first place.

The USE insn should only be created when combine can't the insn setting and/or
using a reg in the current block, which implies that the reg must be live
in multiple blocks, and the insn using it in the current block has been
optimized away.  Neither is true in this case, the reg is live in a single
block, and the insn using it is still there.

Since the local-alloc optimization only applies to variables that are set
and used in a single block, then we should never have any combiner USE insns
emitted for variables which are eligible for the update_equiv_regs
optimization.

I need to look at this some more though.  I built egcs on an alpha-dec-osf4.0
system here, but gdb isn't working right, so I started another build on a
different system which I know has a working gdb.

Jim