https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116891
--- Comment #2 from Walter Mascarenhas <walter.mascarenhas at gmail dot com> ---
Hi,
You will find two files attached to this message:
1) the cpp file contains the C++ code, with comments describing
the exact options used and the g++ version. The bug is
described in detail in a comment in this file.
2) the .s file was obtained with g++'s -S flag.
In a quick look at it you will notice that it contains only
vfmadd132sd instructions and no changes of signs,
that is, g++ is "optimizing"
- ( fma( -a, x, - y) )
to fma( a, x, y ), and this is not supposed to happen
when -frounding-math is used.
walter.
On Sun, Sep 29, 2024 at 11:10 PM pinskia at gcc dot gnu.org <
[email protected]> wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116891
>
> Andrew Pinski <pinskia at gcc dot gnu.org> changed:
>
> What |Removed |Added
>
> ----------------------------------------------------------------------------
> Status|UNCONFIRMED |WAITING
> Component|c++ |middle-end
> Last reconfirmed| |2024-09-30
> Ever confirmed|0 |1
>
> --- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
> I tried:
> ```
> #include <cmath>
>
> double bad_fma( double y, double z, double w) {
> return std::fma( -y, z, -w );
> }
> ```
>
> and it produces with `-O3 -frounding-math -mfma`:
> ```
> _Z7bad_fmaddd:
> .cfi_startproc
> vfnmsub132sd %xmm1, %xmm2, %xmm0 # tmp104, tmp105, <retval>
> ret
> ```
>
> https://godbolt.org/z/fzs1cso3P
>
>
> Can you provide the preprocessed source you are using and/or even the full
> testcase including what headers you are using and what exact options you
> used?
>
> --
> You are receiving this mail because:
> You reported the bug.
