Hi,

this patch fixes PR66444, a problem with -fipa-ra in reload_combine.

The problem is that for the test-case, reload_combine combines these two 
insns:
...
(insn 13 12 14 2 (parallel [
             (set (reg/v/f:DI 37 r8 [orig:92 xD.1858 ] [92])
		(plus:DI (reg:DI 37 r8 [96])
                     (reg:DI 0 ax [orig:95 D.1884 ] [95])))
             (clobber (reg:CC 17 flags))
         ])
      (expr_list:REG_EQUAL (plus:DI (reg:DI 0 ax [orig:95 D.1884 ] [95])
             (const_int 962072674304 [0xe000000000]))
         (nil)))
(insn 14 13 15 2 (set (reg:DI 5 di)
         (reg/v/f:DI 37 r8 [orig:92 xD.1858 ] [92]))
      (nil))
...

into this insn:
...
(insn 14 12 15 2 (set (reg:DI 5 di)
	(plus:DI (reg:DI 37 r8 [96])
             (reg:DI 0 ax [orig:95 D.1884 ] [95])))
      (nil))
...

That removes the set of r8 by insn 13. And that set of r8 is used by 
insn 16:
....
(call_insn 15 ...)
(insn 16 15 17 2 (set (reg:DI 5 di)
	(reg/v/f:DI 37 r8 [orig:92 x ] [92])) test.c:33 85 {*movdi_internal}
      (nil))
...

But reload_combine doesn't acknowledge that use, because it considers 
that r8 is killed by call_insn 15.

The patch fixes the problem by using get_call_reg_set_usage to find out 
that r8 is actually not killed by the call_insn.

Bootstrapped and reg-tested on x86_64 on top of trunk.

OK for trunk and gcc-5-branch?

Thanks,
- Tom