RE: [Axiom-developer] Compiler coercion and DoubleFloat

[Bill Page]

Grég,

On February 27, 2006 6:05 PM you wrote:
> ... 
> In fact, the parser, when it encounters a real number,
> returns integer part, fractional part, length of fractional part
> and exponent and call MAKE-FLOAT. MAKE-FLOAT returns mantissa,
> exponent (in base 10) and a "property" attached |:BF:| 
> (for big float) if $useBFasDefault is true (its default value).
> postBigFloat will return, later, 
> ((|elt| (|Float|) |float|) 23538526683701998541 -2 10)
> 
> But if $useBFasDefault is false MAKE-FLOAT returns a lisp 
> double-float.
> 
> So i add, at the beginning of the code
> 
> )boot $useBFasDefault:=false
> 
> the code is:
>          two4: R := 24.0
>          fourty: R := 40.0
>          exp40: R := 235385266837019985.41
> 
> and the compiler output is:
> 
> (SEQ (LET (|:| |two4| (|DoubleFloat|)) 24.0)
>      (LET (|:| |fourty| (|DoubleFloat|)) 40.0)
>      (LET (|:| |exp40| (|DoubleFloat|)) 2.3538526683702E17)
> 

That's cool.

Does '$useBFasDefault:=false' globally change the representation
of DoubleFloat to lisp double-float?

What is the advantage of lisp double-float over Axiom's
DoubleFloat? Higher precision?

(1) -> precision()$Float

   (1)  68
   Type: PositiveInteger

Precisioni of Float is adjustable but the following are fixed.

(2) -> precision()$DoubleFloat

   (2)  53
   Type: PositiveInteger

(3) -> precision()$MachineFloat

   (3)  16
   Type: PositiveInteger

-------

If I set $useBFasDefault in the Axiom interpreter I get the
following error:

(1) -> )boot $useBFasDefault:=false
(SPADLET |$useBFasDefault| NIL)
Value = NIL
(1) -> two4: DoubleFloat := 24.0

   >> System error:
   Caught fatal error [memory may be damaged]

protected-symbol-warn called with (NIL)

-------

So I suppose you have some specific application of this in the
SPAD compiler?

Regards,
Bill Page.