Re: benchmarks - sort

[David Bateman]

Thomas,

I suspect I see where you get your speed-up. Check the line

>    return *p1<*p2;                              //*double less than

that should read

     return (xisnan(*p2) || *p1<*p2);

overwise NaN's aren't sorted correctly, or at ll. I'm sure Octave's
sorting code can be made about twice as fast if it didn't have to
check for NaN's... As a test try

octave:1> a = rand(1,10); a(2) = NaN; b = mysort2(a)

where is the NaN in b? It should be in b(10), but with your comparison
test it'll end up in b(2). This is something that caused me a lot of
pain, and has a lot of ugly additional code to treat correctly in
merge.cc for the case of IEEE 754 integer compares..

Cheers
David

-- 
David Bateman                                address@hidden
Motorola CRM                                 +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 1 69 35 77 01 (Fax) 
91193 Gif-Sur-Yvette FRANCE

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary