Re: [avr-libc-dev] more accuracy...

[Dmitry K.]

On Monday 24 April 2006 09:25, Björn Haase wrote:
> David Carr wrote on Montag, 24. April 2006 09:37 :
> > Just as a point of reference:
> > Execution times for the aforementioned benchmark
> >
> > 386 (with 387 co-processor) @ 33MHz
> > .67s
> >
> > 486DX2 66MHz
> > .121s
> >
> > Considering that the AVR is 8bit and has no hardware floating point,
> > being only 12x slower than the 386 is impressive especially considering
> > it runs at 1/5 the clock speed.
>
> you are sure that the two PC results were obtained with true float
> calculations? Otherwise the comparison would not be fair :.). I'd assume
> that probably the tan/log/sqrt functions operate on doubles.?
>
> Never the less: the FP emulation for AVR *is* quite efficient.
>
> Bjoern.

avr-libc 1.2.x                     61.1e6 clocks
avr-libc 1.4.2                     46.4e6 clocks
avr-libc with patch to bug #1929   29.8e6 clocks

Dmitry.