Re: [Caml-list] Fixed-point arithmetic on 32-bit architectures

[Niki Yoshiuchi <aplusbi@gmail.com>]

What range of numbers are you hoping to represent and what operations
are you commonly performing on them?  If it's possible for you to
represent your numbers as 24.8 fixed-point (for example) and you are
mostly adding and subtracting, then I would recommend using 32 bit
integers and casting to 64 bit for multiplication and division.
Alternatively you can try and be clever about your shifting instead of
casting (shifting the multiplicand instead of the product, or doing a
half shift on each multiplicand, etc).

On Wed, Jan 5, 2011 at 11:22 AM, Eray Ozkural <examachine@gmail.com> wrote:
> Hello there,
> I'm using fixed-point arithmetic in an algorithm. I am troubled by the
> inefficiency of the fixed-point multiplication and division operations on
> 32-bit architectures. On the Intel 64-bit architecture, I can use the
> Nativeint module and it's quite fast, on 32-bit I had to use the Int64
> module (for the necessary shifts and mul/div's) and I encountered a
> significant slowdown, naturally. is there a preferred way of performing
> fixed point arithmetic with ocaml on 32-bit architectures that I might be
> overlooking?
> Best,
> --
> Eray Ozkural, PhD candidate.  Comp. Sci. Dept., Bilkent University, Ankara
> http://groups.yahoo.com/group/ai-philosophy
>
>