https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121617
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|--- |MOVED
Status|UNCONFIRMED |RESOLVED
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
All mdaz-ftz does is set the bits that changes how x87 and sse handle
denormals. That is they are handled as zero inputs.
So what most likely is happening is fma/fmaf is doing is using the x87
registers to do the the multiply to in 80bit precision rather than infinite
ones.
Anways NOT a GCC bug. If it is a bug is in where fma/fmaf are implemented which
is most likely glibc in your case.
