[Aldor-l] Re: Aldor program: Assistance requested.

[Ralf Hemmecke]

> I guess it is best to discuss ALL aldor related problems in public. It  
> might save time for others.

Agreed.

>> The program is attached (matex.as).
...
>> So we see some rounding error which I find quite annoying, but it comes
>> from the output routine not the internal representation.

Why I did not have an error like Constantine Frangos was that I compiled 
the libraries locally myself. The aldor.org distribution had a bug with 
regards to the object files. That is resolved now and silently 1.0.3 
slipped out.

> The output is faulty -- I admit that. The answer however probably  
> misleading.

> Take for example 20.9 (first vector, second slot)
> Its internal representation is 109576619*2^(-19) in IEEE 32-bit float.  
> This evaluates to
> 20.8999996185302734375.
> 
> So I cannot see how this can be rounded to
> 2.08999991e+1
> which is Aldor's output.

It seems that the output routine shows one digit too much. Anyway, the 
output routine does not look quite correct to me. There is a line	

-- b = x 2^l  1 <= x < 2   =   y 10^k  1 <= y < 10
z := (x-1.5)*0.289529654602168 + 0.1760912590558
	+ l*0.301029995663981;

where a convertion between base 2 and base 10 takes place, this line 
does not completely make sense to me even if I figure out what these 
constants are.

> However, I am curious which number format Aldor is using internally. I  
> suppose IEEE 32bit floats. However, then the output routine is a bit fishy.

Sure without knowing the internal representation of SFlo, one cannot say 
much. Documentation is, unfortunately, also non-existing.

Ralf