RE: [fpu] FPU operations

[Jamil Khatib <jamilkhatib75@yahoo.com>]

> >I agree with you on all what you said but do not
> think
> >that the comparison results are used only after
> >compare instruction?
> 
> Well, yes and no. It's true that the comparator may
> run in
> parallel to all other operations. However, none of
> the standard
> instructions will look at those status bits (as far
> as I know).
> The comparator will operate on operands A and B, so
> if I do
> an ADD of A and B, I don't care about 'larger then'
> or 'smaller
> then' or 'equal' flags. I can see that 'unordered'
> and 'infinity'
> might be useful, but even then, I don't see how and
> ADD instruction
> (for example) can use those status flags.
> 
> 
> One thing came to mind thou: A result==zero status
> flag *might* be
> useful, if there are 'increment and jump if zero' or
> 'decrement
> and jump if zero' (and whatever other variations one
> can come up with)
> are. The type of instructions are typical for
> integer operations, I'm
> not sure if they make sense for floating point ...
> 
> >so they will act as register from
> >teh cpu point of view that will be read only after
> the
> >comparison.
> 
> No, again, this is something the CPU MUST do. All
> outputs might
> change every cycle, as the inputs change (however,
> pipelined).
> The CPU must track the results and status
> information. The FPU
> is nothing else as a pipelined calculation engine,
> it has no
> states.
> 

Sorry but still I can not get teh advantage of
calculating teh result signals every clock.
May be we can driver the result comparison in the
result and the compiler has to define the states


Regards
Jamil Khatib

__________________________________________________
Do You Yahoo!?
Kick off your party with Yahoo! Invites.
http://invites.yahoo.com/