Re: [Qemu-arm] [RFC PATCH for 2.11 00/23] Implementing FP16 for ARMv8.2 using SoftFloat2a and 3c

[Peter Maydell]

On 20 July 2017 at 16:04, Alex Bennée <address@hidden> wrote:
> So while any review is gratefully accepted my main interest is if this
> sort of duel approach is worth it?
>
> I think the long term aim would be to move fully over to the newer
> library if we can but I appreciate a not inconsiderable amount of work
> has gone into the existing 2a code. Having said that as long as the
> testing is good we are likely to pick up bugs in both implementations.
> While its nice to have a more modern SoftFloat library it still lags
> somewhat on QEMUs general requirements claiming only direct support
> for IEEE754 1985 whereas most modern CPUs are closer to the 2008
> revision of the specification.

I'm definitely not a fan of having two different implementations
of softfloat in the codebase, except perhaps for some well defined
*short* transition period. Otherwise we'll end up carrying both
of them around forever, given our past history at success in
transitioning things to new APIs.

I don't have a strong view about whether we're better off
moving to softfloat3c and forward porting any bugfixes and
improvements, or backporting 16-bit fp code from 3c to our
current codebase.

thanks
-- PMM