Re: [RFC] RISC-V: Raise noce if-conversion cost limit

Hi Jeff

> Now we can't really run the scheduler at that point on the fragment to 
> get latency of the sequence and we probably don't want build a full 
> dependency height analysis into the costing hook, but we ought to be 
> able to do significantly better than we are now.    I'm thinking we 
> probably want to attack this problem first since it's pervasive and 
> should help all superscalar designs by bringing significantly more 
> accuracy to the costing model.

Thanks for the suggestion.  I agree that accounting for parallelism is
a better direction than raising a target-specific noce cost limit.

I have dropped the C950-specific multiplier and prepared a draft along
those lines.  It estimates the dependency depth of simple sequences and
uses the target issue rate.  More complex sequences retain the existing
serial cost.

For the MCF sequence with 3 sgt, 6 czero and 3 add, the estimated cost
falls from 48 to 12 on an 8-wide target.

This is still an initial draft.  I would appreciate your feedback on
this approach.

Best regards,
Wang Jue