https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82725
--- Comment #2 from Uroš Bizjak <ubizjak at gmail dot com> ---
(In reply to Martin Liška from comment #1)
> Looks it's related just to i386 target. Can't reproduce with
> x86_64-linux-gnu and -m32.
I can trigger the failure with cc1plus on x86_64-linux-gnu, without -m32:
(gdb) down
#2 0x0000000000ce9cac in change_address_1 (memref=0x2aaaae9238a0,
mode=E_DImode, addr=0x2aaaae90bf00, validate=1, inplace=false)
at ../../git/gcc/gcc/emit-rtl.c:2162
2162 gcc_assert (memory_address_addr_space_p (mode, addr, as));
(gdb) p mode
$1 = E_DImode
(gdb) p as
$2 = 1 '\001'
(gdb) p debug_rtx (addr)
(const:DI (plus:DI (unspec:DI [
(symbol_ref:DI ("_ZZ7tempDirvE5cache") [flags 0x2a] <var_decl
0x2aaaae785d80 cache>)
] UNSPEC_NTPOFF)
(const_int 8 [0x8])))
$3 = void
