[Ffmpeg-devel] ac3enc.c modifications

[Rich Felker]

On Thu, May 19, 2005 at 06:15:08PM -0600, Mike Melanson wrote:
> 	How about this scenario: Both the floating point op and integer op 
> 	take the same amount of time, say, 1 clock cycle. Where is the advantage 
> for integer in this case?

Well-defined results that don't vary from one cpu to the next. :)

> Especially since the code probably has to do a 
> few more maintenance operations on the integer after the operation 
> (precision scaling, whatever).

I agree, and still maintain that this is a bad cpu design. If you can
do any floating point op in one cycle, you should be able to do 2
integer ops per cycle (or many more :)).

Rich