https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126696
Merlin Warner-Huish <merlin.warner-huish at embecosm dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |merlin.warner-huish@embecos
| |m.com
--- Comment #3 from Merlin Warner-Huish <merlin.warner-huish at embecosm dot
com> ---
I would like to work on this. My plan is to handle fma (x, 1.0, y) -> x + y,
fma (x, -1.0, y) -> y - x and fma (x, y, -0.0) -> x * y.
For signalling NaNs I think no guard is needed, since unlike x * 1.0 -> x an
arithmetic operation remains that quiets the NaN and raises invalid, but I’m
happy to add !HONOR_SNANS if that’s preferred.
