https://gcc.gnu.org/bugzilla/show_bug.cgi?id=33780
--- Comment #7 from Andrew Pinski ---
I can only reproduce this these days on 32bit x86 when x87 is in use ...
--- Comment #6 from jv244 at cam dot ac dot uk 2008-02-11 11:25 ---
(In reply to comment #5)
> Or are you complaining that we constant fold integer powers as if they were
> non-integer powers
Richard,
I think I'm suggesting more a enhancement than a bug fix. Some good quality
compiler
--- Comment #5 from rguenth at gcc dot gnu dot org 2008-02-05 11:08 ---
The difference with your last testcase is indeed that we unroll the loop
setting c[] and constant-fold the result by computing the integral powers
with 0.5ulp precision (and a single rounding(!)). But all your other
--- Comment #4 from dominiq at lps dot ens dot fr 2007-10-18 15:23 ---
> while we expand a constant integral power to an optimal (as of the count of
> multiplications / additions) series of multiplications and additions.
It seems the difference is coming from something else. I'll bet th
--- Comment #3 from dominiq at lps dot ens dot fr 2007-10-15 15:51 ---
> that causes numerical results with CP2K to change going from -O0 to -O3.
If you do expect that optimization optimizes your computation, you should
expect some change of the numerical results, so put some tolerance
--- Comment #2 from jv244 at cam dot ac dot uk 2007-10-15 15:30 ---
(In reply to comment #1)
> As Fortran allows open-coding of integral powers this is not a bug. But
> I'll leave final closing as INVALID to more fortran-knowing people.
I agree that Fortran-wise this is not a bug. Howe
--- Comment #1 from rguenth at gcc dot gnu dot org 2007-10-15 15:01 ---
Confirmed. It changes for me -O2 vs. -O2 -fpeel-loops which causes the loop
in f to be completely unrolled (which at the first sight doesn't look wrong).
And we expand __powidf2 inline because the exponent is now a
