[Libre-soc-dev] data-dependent fail first writing outputs of failing element

lkcl luke.leighton at gmail.com
Thu Oct 6 22:18:29 BST 2022


On Thursday, October 6, 2022, Jacob Lifshay <programmerjake at gmail.com>
wrote:
> On Thu, Oct 6, 2022, 12:30 lkcl via Libre-soc-dev <
libre-soc-dev at lists.libre-soc.org> wrote:
>>
>> * have to sort out sv.add/ff because regfile write terminates
>>    at VL and so should only write to the penultimate element
>>    when VLI is also set.
>
> uuh, actually iirc we agreed that the failing element should always write
all of its outputs.

good catch, i forgot.  spec needed updating. funfunfun.

implementation-wise it's one hell of a lot easier as there is
no need to try to unwind CR reg changes or try to cancel things.

which bodes well ha ha for predicate-result mode which *does*
interact with the dest-predicate, post-analysis side, and can
stop regfile writing.

it's odd and slightly annoying because a case can be made
either way.

actually... ahh... subvl. vec2/3/4.  oh dear.  holding all
of a subvector back before deciding to write? not happening.
write it is.

l.


More information about the Libre-soc-dev mailing list