https://bugs.kde.org/show_bug.cgi?id=484426
--- Comment #2 from Paul Floyd <pjfl...@wanadoo.fr> ---
Comment from code:
* FRINTA, FRINTN are kludged .. they just round to nearest. No special
handling for the "ties" case. FRINTX might be dubious too.
* Ditto FCVTXN. No idea what "round to odd" means. This implementation
just rounds to nearest.
"round to odd" does what the name says and rounds to odd integers. It's one of
several rounding methods with the advantage that (like round to even) there is
no bias, half of ties round up ("even point 5") and half round down ("odd point
five"). So 0.5 rounds to 1.0.
https://developer.arm.com/documentation/dui0801/g/A64-Floating-point-Instructions/FCVT?lang=en
says that depends on the FPCR
gdb says
fpcr 0x0 [ Len=0 Stride=0 RMode=0 ]
RMode 0 means round to zero, which is probably just about the only thing ever
used.
For float to double there should be no rounding (which I guess only applies to
the ULP for floating point conversions).
--
You are receiving this mail because:
You are watching all bug changes.
