RE: [fpu] FPU operations

["Rudolf Usselmann" <russelmann@hotmail.com>]

Again, the standard has very strong wording for the comparison
operation. Here is the summary:

1) Comparison has to be exact (no overflow/underflow)
2) Four mutually exclusive relations shall be generated:
	a) equal
	b) less than
	c) greater than
	d) unordered
3) Any comparison involving a NaN shall produce 'unordered' only.
4) Comparison shall ignore the sign of zero '0'.


Regarding a MAC instruction, we should see if we can design a separate
block fro it that is faster. For example the rounding and normalization
has to be done only once. Again, this has to investigated. I a separate
block can be made run a lot faster, then we should probably use that,
if not, just use the existing mul and add blocks.
You have to ask yourself why people came up with a MAC instruction, then
I think, the choice will be clear.


rudi

>-----Original Message-----
>From: owner-fpu@opencores.org [mailto:owner-fpu@opencores.org]On Behalf
>Of Damjan Lampret
>
>
> > > - SF compare
> > compare for equality only?
> > Depends. You want one insn that will set status bits (eq, lt, gt) or do
>you want separate compare instructions. GCC really doesn't care.
>
> > > If we are going to define MAC we should have a
> > seperate block for it
>
>Why? With one block for Mul and the other for Add you get MAC if they
>work as a pipeline (though you need accumulator register)
>
>regards,
>Damjan
________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com