[MPlayer-cvslog] CVS: main/libaf af_format.c,1.26,1.27

[Alexander Strasser]

Alex Beregszaszi wrote ( On Tue, Feb 22, 2005 at 07:24:11PM +0100 ):
> > > > finally the dreaded white-noise-with-floats bug is fixed!!!!
> > > >
> > > > the problem is that lrintf was not prototyped on some systems, but
> > > > it's easier and faster just not to use it at all. looks like the
> > > > cola
> > >
> > > did you benchmark this change, AFAIK lrintf() should be MUCH faster
> > > then casting to int, as the later needs to change the FPU rounding
> > > mode several times
> > 
> > Benchmarked, for 5 million samples of ram garbage, lrintf took 0.5
> > seconds, cast took 0.9 seconds. lrintf proved itself to be risky, so
> > for an optimization of this size, imho it's not worth it.
> 
> That's a slowdown by almost two times... what do you say now, Rich? :)

@ods15: what do you mean with risky?

In general lrintf should work out fine, as long as it is properly coded
and inlined.

I think fixing up configure is the way to go.
Maybe test with -Werror or so and make two tests.
If the first one fails, 
we define the strange symbols and test again.

  Alex (beastd)