[Bug middle-end/61577] [4.9.0 Regression] can't compile on hp-ux v3 ia64

[cameron.heide at betasystems dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61577

--- Comment #109 from C. Heide <cameron.heide at betasystems dot com> ---
_ZN15variable_hasher6removeEP8variable is just this same function, located at
0x6d777a0; I didn't realize at first that the objdump disassembly wasn't
showing me the actual relocation unless I specify '-r', in which case it's:
>   20:   18 00 80 46 90 11       [MMB]       st4 [r35]=r32
>                         22: PCREL21B    .text+0xa780
>   26:   50 02 8c 20 20 00                   ld4 r37=[r35]
>   2c:   08 00 00 50                         br.call.sptk.many b0=20 <_ZN15variable_hasher6removeEP8variable+0x20>
or the relocation section as seen by readelf:
> Relocation section '.rela.gnu.linkonce.t._ZN15variable_hasher6removeEP8variable' at offset 0x1594b8 contains 1 entry:
>  Offset     Info    Type                Sym. Value  Symbol's Name + Addend
> 00000022  00000149 R_IA64_PCREL21B        00000000   .text + a780
and the function at the offset 0xa780 is:
> 0000a780 <_ZL18variable_htab_freePv>:
>     a780:       00 10 21 0a 80 05       [MII]       alloc r34=ar.pfs,8,5,0
>     a786:       30 02 30 00 42 80                   mov r35=r12
>     a78c:       01 64 fc 8c                         adds r12=-64,r12
>     a790:       01 00 00 00 01 00       [MII]       nop.m 0x0
>     a796:       10 02 00 62 00 80                   mov r33=b0
>     a79c:       04 08 00 84                         mov r36=r1;;
which is the static function expected, as seen in the source at
var-tracking.c:507. So it seems fine within the object file itself, but if I
disassemble the final cc1 executable, it's become the invalid address:
> 06d777a0 <_ZN15variable_hasher6removeEP8variable>:
>  6d777a0:       00 10 19 0a 80 05       [MII]       alloc r34=ar.pfs,6,5,0
>  6d777a6:       30 02 30 00 42 80                   mov r35=r12
>  6d777ac:       01 67 fc 8c                         adds r12=-16,r12
>  6d777b0:       01 00 00 00 01 00       [MII]       nop.m 0x0
>  6d777b6:       10 02 00 62 00 80                   mov r33=b0
>  6d777bc:       04 08 00 84                         mov r36=r1;;
>  6d777c0:       18 00 80 46 90 11       [MMB]       st4 [r35]=r32
>  6d777c6:       50 02 8c 20 20 00                   ld4 r37=[r35]
>  6d777cc:       d8 96 cf 50                         br.call.sptk.many b0=7a70e90 <_etext_f+0xc87090>
According to the cc1 symbol table, variable_htab_free should be at 0x5a70e90.
Which I just realized is exactly 0x2000000 off the invalid address...