On Nov 29, 1:28 am, "Bill Hart" <[EMAIL PROTECTED]> wrote:
> It's true these are not used much. But at this point we definitely
> don't want to break strict compatibility with GMP.
>
> I don't have an opinion on the other options though. Perhaps someone
> else has something meaningful...
>
> Bill.
>
> 2008/11/29 <[EMAIL PROTECTED]>:
>
>
>
> > Now I notice that some of the existing mpn fn's have alternate entry points
> > eg mpn_add_n and mpn_add_nc where you have an extra parameter to specify a
> > carry in .
> > I can do this to all the new K8 asm fn's , it pretty trivial. It just a
> > question of how to call them , for example
>
> > .align 16
> > mpn_add_n:
> > mov $0,carryreg
> > .align 16
> > mpn_add_nc:
> > here is the usual stuff
>
> > Here we either align it and have to skip over 16 bytes (which is at least 1
> > extra cycle) or we dont bother with the alignment for the _nc version , and
> > have a penalty for an unaligned fn call (which costs what?) .
>
> > The other way is to just have mpn_add_nc , and make mpn_add_n a macro with a
> > carry in of zero.This would break strict compatibility with gmp.
>
> > Or perhaps not bother , as their not used much.
>
> > Any thoughts? on this , preferances?
I agree with Bill - we should keep strict compatibility with GMP
because it is important that MPIR is a 'drop-in' replacement for it.
As long as the most used one is the fastest and the hit on the least
used one is small, I don't think the performance hit is worth worrying
about.
If the hit was large we could give the library user the choice of
duplicating the code using a macro of the form :
macro madd_n
align 16
here is the usual stuff
endm
mpn_add_n :
ifdef MAXIMISE_SPEED
madd_n 0
elif REDUCE_CODE_SIZE
mov 0, carry_reg
endif
mpn_add_nc :
madd_n carry_reg
or something similar.
Brian
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"mpir-devel" group.
To post to this group, send email to mpir-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at
http://groups.google.com/group/mpir-devel?hl=en
-~----------~----~----~----~------~----~------~--~---
