[Libre-soc-dev] clamping/saturation semantics

[Lauri Kasanen]

On Sat, 12 Dec 2020 18:41:26 +0200
Lauri Kasanen <cand at gmx.com> wrote:

> On Sat, 12 Dec 2020 18:33:57 +0200
> Lauri Kasanen <cand at gmx.com> wrote:
>
> > All ops should be done "as if" they were done at srcwidth. This means
> > all results will be correct, but the hw can optimize and go faster in
> > some cases.
>
> Addendum: widening multiplies are a special case, for that to be
> correct it needs to happen at dest width.

Sorry for the mail spam. Perhaps the correct formulation is
max(srcwidth, dstwidth) - the larger of the two. Since a widening
addition would also need to happen at the higher precision.

- Lauri