Re: Efficient Sequential Access to Arrays

[Keean Schupke <keean.schupke@googlemail.com>]

On Thursday, 19 July 2012 12:18:11 UTC+1, Georg Bauhaus  wrote:
> On 19.07.12 12:41, Keean Schupke wrote:
> > What is more of a problem is that the node structure is 32 bytes long, so two nodes fit in a 64bit cache line, and the index operation is shift left by 5. Adding the extra Index word will probably slow things down significantly, unless we can save some space elsewhere in the structure.
> > 
> > I may be able to replace one Integer with a Short_Int, and use a Short_Int for the index. I think this could be worth trying.
> 
> I'll assume it is a 32 bit program already? (As GNAT uses 32 bis
> for addresses then, not 64.)

Its a 64bit platform, so the pointers are 64bits but the integers 32bit. I think what I was thinking is with the indexed version, I may be able to shrink the whole data structure from 32bytes to 16bytes, and that would fit 4 elements in a cache line instead of two. Maybe this would make up for the slower indexed addressing. It will try and post my results.

Cheers,
Keean.