[Libre-soc-dev] [RFC] SVP64 on branch instructions
luke.leighton at gmail.com
Wed Aug 4 18:45:19 BST 2021
On August 4, 2021 1:14:40 PM UTC, lkcl <luke.leighton at gmail.com> wrote:
>regarding overwrite and use of AA for alternative purposes, i realised
>after some thought that actually, combining the predicate mask with the
>Vector-Branch-CR test is not appropriate to do inside Branch itself.
... but _is_ appropriate for svstep mode, to allow some
situations where you want to know what a REMAP schedule might look like (and to obtain all the endpoints of all loops in one hit), yet in others you don't care, you just want to branch/loop.
i've therefore put re-purposing of AA as Rc back.in, sigh.
implementations of this are going to be... tricky. although, just thinking about it: hypothetically, and just like LD/ST, it may still be possible to use the existing scalar v3.0B instruction, but "fake" what data it receives.
(for Vector LDST in ISACaller i actually changed the immediate D to contain D*srcstep and other modes. something similar might be possible with branches. have to see)
More information about the Libre-soc-dev