[Ffmpeg-devel] ac3enc.c modifications

[Rich Felker]

On Fri, May 20, 2005 at 01:38:43AM +0200, M?ns Rullg?rd wrote:
> Rich Felker <dalias at aerifal.cx> writes:
> 
> > On Fri, May 20, 2005 at 12:56:07AM +0200, M?ns Rullg?rd wrote:
> >> Rich Felker <dalias at aerifal.cx> writes:
> >> 
> >> > On Thu, May 19, 2005 at 09:53:33PM +0200, Michael Niedermayer wrote:
> >> >> both float and int should be supported, there are people with cpus
> >> >> where float IS faster then int
> >> >
> >> > This is only possible if the cpu has a horribly botched integer
> >> > implementation.
> >> 
> >> Or a damn good floating point unit.  Have you ever used an IBM Power
> >> or Digital Alpha CPU?  A 300 MHz IBM Power3 (Power5 is the latest) I
> >> ran some tests on beat a 1 GHz Pentium3 at floating point math, as did
> >> a 500 MHz Alpha a few generations older than the P3.
> >
> > but will a 300 mhz ibm power3 using floats beat a 1ghz athlon using
> > integers? i think not!
> 
> Probably not, but the point was that there are machines where floating
> point is significantly faster than on x86.  On these machines, the
> difference between floating point and integer math is rather small,
> which is not equivalent to having a "horribly botched" integer unit,
> as you asserted.

Yes it is. Performing an operation on an integer is inherently less
work than performing the same operation on a floating point number,
since you have the added exponent to deal with. If you make a cpu
where the floating point op is just as fast or faster, it means you
misdesigned your integer unit horribly, since a simplified version of
the floating point unit should be able to do the analogous integer
operation in less time.

Rich