Re: [avr-libc-dev] Re: Add items on TODO

[E. Weddington]

On 30 Oct 2002 at 13:19, Joerg Wunsch wrote:

> As Theodore A. Roth wrote:
> 
> [the asin() problem]
> 
> > My investigation seemed lead to there not being a clear
> > understanding of the format of the floats. gcc uses ieee format
> > floats so 1.0 looks like 0x3f800000, while the asin() implementation
> > seems to think 1.0 is 0x7f000000.
> 
> Nope.  asin() uses 0x3f800001 to denote a ``number slightly larger
> than 1.0''.  This is correct.  The 0x7f you're seeing atop is masking
> out the sign of the operand.  The 0x3f800001 is micro-optimized and
> not displayed completely, i. e. the third byte is taken from
> __zero_reg__ instead (obfuscated as "rT1c" in the source).
> 
> So the problem is indeed that the implemented algorithm becomes fairly
> inaccurate for small numbers.

Joerg,

Any thoughts then on re-implementing asin() as an atan2() function? 
The biggest issue that I can see is that atan2 is huge in comparison. 
But I'd rather have a (more) accurate function than a small one 
that's incorrect.

Eric