[Libre-soc-isa] [Bug 1092] OPF RFC ISA WG questions feedback on ls002 float-load-immediate

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Wed May 31 23:40:16 BST 2023


--- Comment #9 from Paul Mackerras <paulus at ozlabs.org> ---
(In reply to Jacob Lifshay from comment #6)
> (In reply to Konstantinos Margaritis (markos) from comment #4)
> > > Are you OK with fishmv generating some result which is very different from the input value
> > > in this kind of case? (If so, a programming note warning about the possibility is probably
> > > appropriate.)
> yes, fishmv is intended to be used only with f32 values (or more
> specifically, all possible results of fmvis which is a subset of f32
> values), if it isn't a valid f32-in-f64 then i'm fine with values changing
> because behaving like a bfp32 store/write 16-bits/bfp32 load is exactly
> what's intended.

Right. Perhaps a programming note warning about the possibility could be added.
There is nothing to stop people doing fishmv on any arbitrary double-precision
value (nothing says you can only use it on the result of fmvis), and getting a
value very different from the input value might be surprising.

> > IMHO, the result should not be different from the input value.
> i explained the misunderstanding in the meeting today, konstantinos was
> worried it would give different values than encoded in the immediates of a
> fmvis/fishmv pair, but it doesn't do that, it correctly handles all possible
> f32 bitpatterns.


You are receiving this mail because:
You are on the CC list for the bug.

More information about the Libre-SOC-ISA mailing list