http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49798
Uros Bizjak <ubizjak at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rguenth at gcc dot gnu.org
--- Comment #12 from Uros Bizjak <ubizjak at gmail dot com> 2011-07-21 16:46:36
UTC ---
(In reply to comment #9)
> Please compare optimized tree dumps from i686 (a) compilation vs x32 (b):
>
> (a)
>
> foo (union U u)
> {
> union U v;
> _Bool D.2000;
> double D.1999;
> double D.1998;
> int D.1997;
>
> <bb 2>:
> v = {};
> v.m = &xxxx;
> D.1998_1 = u.d;
> D.1999_2 = v.d;
> D.2000_3 = D.1998_1 == D.1999_2;
> D.1997_4 = (int) D.2000_3;
> return D.1997_4;
>
> }
>
> (b)
>
> foo (union U u)
> {
> double D.2709;
> _Bool D.2704;
> double D.2702;
> int D.2701;
>
> <bb 2>:
> D.2709_8 = VIEW_CONVERT_EXPR<double>(&xxxx);
> D.2702_1 = u.d;
> D.2704_3 = D.2702_1 == D.2709_8;
> D.2701_4 = (int) D.2704_3;
> return D.2701_4;
>
> }
>
> We can't directly move &xxxx (32bit value) to double (64bit value).
>
> However, we expand to:
>
> (insn 6 5 11 (set (reg/f:DI 66)
> (symbol_ref:DI ("xxxx") [flags 0x40] <var_decl 0x7f71c5925140 xxxx>))
> pr49798.c:12 -1
> (nil))
>
> ...
>
> (insn 13 12 14 (set (reg:CCFPU 17 flags)
> (compare:CCFPU (reg:DF 73)
> (subreg:DF (reg/f:DI 66) 0))) pr49798.c:13 -1
> (nil))
>
> Does this looks OK?
IMO, VIEW_CONVERT_EXPR is already wrong, let's CC richi for opinion.
