https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116109
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2024-07-30
Target| |x86_64-*-*
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
There is also --param avoid-fma-max-bits which can be used to avoid creating
loop-around FMAs. This is enabled for some AMD tunings.
Note the vectorizer would create two independent FMA chains iff the target
would tell it to unroll.
I think -fsplit-ivs-in-unroller isn't what you seek,
-fvariable-expansion-in-unroller might be but I'm not sure either works for
vector float reductions.
