i386 ESP adjustments opt.

[Jan Hubicka]

> > + ;; Convert epilogue deallocator to pop.
> > + (define_peephole2
> > +   [(match_scratch:SI 0 "r")
> > +    (parallel [(set (reg:SI 7) (plus:SI (reg:SI 7) (const_int 4)))
> > + 	      (set (reg:SI 6) (reg:SI 6))
> > + 	      (clobber (reg:CC 17))])]
> > +   "optimize_size || !TARGET_ADD_ESP_4"
> > +   [(parallel [(set (match_dup 0) (mem:SI (reg:SI 7)))
> > + 	      (set (reg:SI 7) (plus:SI (reg:SI 7) (const_int 4)))])]
> > +   "")
> 
> Um, this seems distinctly non-safe.  You've lost the tie between
> esp and ebp.  Were you intending a blockage as you used for the
> two pop cases?
Huh, you are right. I've probably took some older version of my patch, since I
remember, I've solved the problem originaly by adding new push and pop patterns
with the (set (reg:SI 6) (reg:SI 6)) and use them for the expanders.

Perhaps this is more sensible solution, how many optimizations we may
miss by the blockage appearing in the rtl chain?

Honza