PR 12828

[Dale Johannesen]

ppc has some patterns that look like this:

(insn 67 66 68 0x0 (parallel [
             (set (reg:DF 129)  (float:DF (reg:SI 128)))
              ......
             (clobber (mem:DF (plus:SI (reg/f:SI 114 virtual-stack-vars)
                         (const_int 262144 [0x40000])) [0 S8 A8]))
             ......

where the stack temp it needs is recorded only inside the clobber
(the splitter creates insns from it later).
This particular one has a large offset due to large stack arrays 
elsewhere.
That offset is too big to fit in the immediate field, so
instantiate_virtual_regs() changes this to

(insn 228 66 67 6 0x0 (set (reg:SI 194)
         (const_int 262176 [0x40020])) -1 (nil)
     (nil))

(insn 67 228 68 6 0x40586630 (parallel [
             (set (reg:DF 129)  (float:DF (reg:SI 128)))
             ......
             (clobber (mem:DF (plus:SI (reg/f:SI 30 r30)
                         (reg:SI 194)) [0 S8 A8]))
            ........

This confuses the loop optimizer: when pulling the set of R129 out of a 
loop,
it does not know there's a dependency on R194.  It looks like there 
ought to
be a USE of R194 inside the parallel, but I see no reason that 
instantiate_virtual_regs
can assume such a pattern exists (it doesn't now).  I could change the 
loop
optimizer to look inside CLOBBER(MEM, but I'm concerned that similar 
problems
could exist elsewhere.   What's the right approach here?