rs6000 64-bit combine.c:subst() bug

[David Edelsohn <dje at watson dot ibm dot com>]

	I'm running into a testsuite failure (execute/950607-2.c) when
using my 64-bit PowerPC work that I have tracked down to the following
event in combine.c:subst().  If I step over the SUBST macro with the state
shown by the following GDB output, the resulting RTL gains a circular
reference which eventually crashes GCC when it blows out the stack trying
to break it apart.
	I don't know whether subst() has a bug or this RTL never should
have been created by combine.

Thanks, David


Breakpoint 6, subst (x=0x2003ab8c, from=0x2003aa90, to=0x2003aa98, in_dest=0,
    unique_copy=0) at combine.c:3129
3129                  new = (unique_copy && n_occurrences ? copy_rtx (to) : to);
(gdb) next
3130                  n_occurrences++;
(gdb)
3131                }
(gdb)
3154              if (GET_CODE (new) == CLOBBER && XEXP (new, 0) == const0_rtx)
(gdb)
3157              SUBST (XEXP (x, i), new);
(gdb) print x
$31 = 0x2003ab8c
(gdb) pr
(ior:SI (reg:SI 102)
    (reg:SI 103))
(gdb) print new
$32 = 0x2003aa98
(gdb) pr

(and:SI (subreg:SI (subreg:DI (ior:SI (reg:SI 102)
                (reg:SI 103)) 0) 0)
    (const_int 1))
(gdb) print i
$11 = 0
(gdb) where
#0  subst (x=0x2003ab8c, from=0x2003aa90, to=0x2003aa98, in_dest=0,
    unique_copy=0) at combine.c:3129
#1  0x10148178 in subst (x=0x200708e4, from=0x2003aa90, to=0x2003aa98,
    in_dest=0, unique_copy=0) at combine.c:3141
#2  0x10148178 in subst (x=0x2003aa84, from=0x2003aa90, to=0x2003aa98,
    in_dest=0, unique_copy=0) at combine.c:3141
#3  0x10148178 in subst (x=0x2003aafc, from=0x2003aa90, to=0x2003aa98,
    in_dest=0, unique_copy=0) at combine.c:3141
#4  0x10148178 in subst (x=0x2003ab04, from=0x2003aa90, to=0x2003aa98,
    in_dest=0, unique_copy=0) at combine.c:3141
#5  0x10144818 in try_combine (i3=0x2003ab10, i2=0x2003aad4, i1=0x0)
    at combine.c:1725
#6  0x10142450 in combine_instructions (f=0x2003c844, nregs=537111312)
    at combine.c:593
#7  0x10004978 in rest_of_compilation (decl=0x2006c7c8) at toplev.c:3391
#8  0x10084a38 in finish_function (nested=0) at c-decl.c:7124
#9  0x1018d87c in yyparse () at c-parse.y:130
#10 0x10003110 in compile_file (name=0x38270 <Address 0x38270 out of bounds>)
    at toplev.c:2528
#11 0x100061ac in main (argc=4, argv=0x2ff2286c, envp=0x2003aa98)
    at toplev.c:4353
#12 0x100001f4 in __start ()

===============================================================================
David Edelsohn                                      T.J. Watson Research Center
dje@watson.ibm.com                                  P.O. Box 218
+1 914 945 4364 (TL 862)                            Yorktown Heights, NY 10598