https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61139
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Status|UNCONFIRMED |NEW
Last reconfirmed| |2014-05-13
Ever confirmed|0 |1
Severity|normal |enhancement
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is that reassoc doesn't fold stmts but we end up with
<bb 2>:
_3 = *x_2(D);
- _5 = w_4(D) - _3;
- _6 = _5 + w_4(D);
+ _8 = -_3;
+ _9 = w_4(D) + w_4(D);
+ _6 = _9 - _3;
instead of
_9 = w_4(D) * 2.;
of course FMA recognition could also be enhanced to handle this case
(of course all the issue re-appears for w + w + w then, which should
be w * 3.). So probably reassocs job in the end anyway.
Confirmed.