Another performance question

[Michel Bardiaux]

Kevin Ryde wrote:

> Michel Bardiaux <mbardiaux at peaktime.be> writes:
> 
>>What would be the impact,
>>risks, etc of having flags (1 in mpz, 2 in mpq) telling that a limb is
>>*not* a pointer but a number making by itself a 1-integer limb?
> 
> 
> I suppose that could work, though every function reading an mpz would
> need to check this flag and change the way it accesses the data.

I dont see why, only the allocation/deallocation would be impacted; data
access would still have a limb pointer and count.

> 
> But we're probably inclined to leave this to an application.  If
> you've got a lot of small numbers you probably don't even want to be
> making function calls to handle them, instead establish some fixnum
> versus bignum like various language interpreters do (lisps, smalltalk,
> etc).

Well, I tried that route, and my smalles common denominators overwhelmed
32-bits almost at once, and 64-bits after a few days of develeopment
work, which is why I went to GMP.

HaND,
-- 
Michel Bardiaux
Peaktime Belgium S.A.  Bd. du Souverain, 191  B-1160 Bruxelles
Tel : +32 2 790.29.41