This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug rtl-optimization/56151] [4.8 Regression] Performance degradation after r194054 on x86 Atom.
- From: "jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Mon, 04 Feb 2013 12:03:52 +0000
- Subject: [Bug rtl-optimization/56151] [4.8 Regression] Performance degradation after r194054 on x86 Atom.
- Auto-submitted: auto-generated
- References: <bug-56151-4@http.gcc.gnu.org/bugzilla/>
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56151
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org,
| |uros at gcc dot gnu.org
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-02-04 12:03:52 UTC ---
Well, we already have peephole2s for MEM op= ARG (the TARGET_READ_MODIFY_WRITE
guarded peephole2s), the reason why it doesn't trigger in this case is that ARG
in this case ends up being a MEM, and
MEM op= MEM2 isn't a valid insn:
(insn 95 94 96 11 (set (reg:SI 3 bx [orig:135 D.1801 ] [135])
(mem:SI (reg/f:SI 4 si [orig:86 D.1803 ] [86]) [2 *_41+0 S4 A32]))
pr56151.c:32 89 {*movsi_internal}
(nil))
(insn 96 95 97 11 (parallel [
(set (reg:SI 3 bx [orig:135 D.1801 ] [135])
(ior:SI (reg:SI 3 bx [orig:135 D.1801 ] [135])
(mem:SI (plus:SI (mult:SI (reg:SI 6 bp [orig:132 D.1801 ]
[132])
(const_int 4 [0x4]))
(symbol_ref:SI ("setmask") <var_decl
0x7f587927a390 setmask>)) [2 setmask S4 A32])))
(clobber (reg:CC 17 flags))
]) pr56151.c:32 421 {*iorsi_1}
(expr_list:REG_DEAD (reg:SI 6 bp [orig:132 D.1801 ] [132])
(nil)))
(insn 97 96 98 11 (set (mem:SI (reg/f:SI 4 si [orig:86 D.1803 ] [86]) [2 *_41+0
S4 A32])
(reg:SI 3 bx [orig:135 D.1801 ] [135])) pr56151.c:32 89
{*movsi_internal}
(expr_list:REG_DEAD (reg:SI 3 bx [orig:135 D.1801 ] [135])
(expr_list:REG_DEAD (reg/f:SI 4 si [orig:86 D.1803 ] [86])
(nil))))
I guess for temp = MEM; temp op= MEM2; MEM = temp; for temp dead afterwards
(this case), we could as well just add another set of the (up to 3?)
peephole2's, that would transform that into
temp = MEM2;
MEM op= temp;
instead.