[Libre-soc-dev] [RFC] SVP64 Data-dependent fail-first on CR operations (crand, cror, etc)

[Richard Wilbur]

> On Aug 29, 2021, at 07:02, lkcl <luke.leighton at gmail.com> wrote:
>> On August 28, 2021 4:36:02 PM UTC, lkcl <luke.leighton at gmail.com> wrote:
>> https://libre-soc.org/openpower/sv/svp64/appendix/
> 
>> other operations which *actually* operate on entire (complete) CR
>> Fields are usually only 3 bits for specifying the CR Field.  these
>> *would* need the index mode.
> 
> question:
> 
> is it worthwhile to use the elwidth bits of the 24-bit RM, which are meaningless for at least the result if the result is a 4 bit CR Field, to provide additional fields?
> 
> this is already done for SV Branches, the VLi field is for example in elwidth bit zero in SV Branch RM.

I think making meaningful use of resources at hand is a very good thing!  Id est, to take a field which is meaningless in this context and be able to assign a useful meaning is more than worthwhile—it improves the expressiveness and POWER of the instruction set.