[libre-riscv-dev] formal proof of the conditions where smaller fp ops can be emulated by larger fp ops

James Cloos cloos at jhcloos.com
Thu Jan 30 13:45:12 GMT 2020

there was apost on the 754 list several months ago which said that one
of the origina 754 authors proved that after -- i think it was -- three
ops on the wider type the results of narrowing will only work if the
wide ops used the opposite rounding than desired for the narrow type.

ie, if even is the target use odd for all wide ops and even only for the
final narrowing.

-JimC (appologies for any typos or terseness; only right side working...)
James Cloos <cloos at jhcloos.com>         OpenPGP: 0x997A9F17ED7DAEA6

More information about the libre-riscv-dev mailing list