On Thu, 18 Apr 2002, Jesse Welton wrote:

> Bijan Parsia wrote:
> > 
> > Summary: "An even simpler possibility is to use floating point operations
> > for semi-long integers."
> > 
> > It made me wonder want the impact of a scheme like that would be on
> > Squeak. I.e., would some loops benefit? How about all over memory
> > consumption? etc.
> 
> Unless this has changed recently, Squeak already uses floats for the
> integer arithmetic in its pseudorandom number generation. 

Interesting! I wouldn't have guessed! Thanks for the pointer.

> When I
> played with this a couple of years ago, it was definitely much faster
> than using ints which overflowed into long ints.  Surprizingly, it was
> even faster than ints constrained to a short int range by arbitrarily
> reducing the modulus of the arithmetic.

Wow. That's a bit surprising. But indeed:

	"This Random Number Generator graciously contributed by David
N. Smith.  It is an adaptation of the Park-Miller RNG which uses Floats to
avoid the need for LargeInteger arithmetic."

I wonder if this is still faster than the current LargeInteger support...

>  I never fully understood why.

Perhaps the cost of the integer calculations outweighed the cost of
unboxing?

I wonder if this would speed up SecureHasAlgorithm...and on which
architecture. Hmm. Probably kill on the iPaq with that FPless Arm...

Cheers,
Bijan Parsia.