[Libre-soc-dev] LD/ST Elwidth Overrides

Jacob Lifshay programmerjake at gmail.com
Thu Sep 2 19:25:25 BST 2021

On Thu, Sep 2, 2021, 11:03 lkcl <luke.leighton at gmail.com> wrote:

> On September 2, 2021 5:56:32 PM UTC, Jacob Lifshay <
> programmerjake at gmail.com> wrote:
> >ok, in that case I think we should re-allow ffirst for all modes and
> >have a
> >note stating that hw implementers *should* limit it to avoid massive
> >page
> >probing, suggesting the above algorithm.
> agreed...except the notes (and spec, and bugreport) says that the mode
> bits on Vector Indexed instructions are to be used for an alternative
> (secondary) purpose.

oops, I forgot about that...

> spotted this last night and it iiiis... bug #561
> https://bugs.libre-soc.org/show_bug.cgi?id=561
> i meant to ask you about that, what strided would mean (pseudocode
> preferred, at the bugreport)
> i *think* i managed to get both unit and elstrided LDST shoehorned in.  if
> this is covered then closing 591 as invalid can be done, and the bits
> reallocated to ffirst, but it has to be checked.

unit strided can be encoded as plain strided if we need a spare mode


More information about the Libre-soc-dev mailing list