clisp-devel

Re: clisp/doc impbody.xml,1.384,1.385

[Sam S. <sd...@gn...>]

> * Bruno Haible <un...@hf...> [2005-05-30 13:33:31 +0000]:
>
> Update of /cvsroot/clisp/clisp/doc
> In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv20957/doc
>
> Modified Files:
> 	impbody.xml 
> Log Message:
> Document the 48-bit fixnums.
>
>
> Index: impbody.xml
> ===================================================================
> RCS file: /cvsroot/clisp/clisp/doc/impbody.xml,v
> retrieving revision 1.384
> retrieving revision 1.385
> diff -u -d -r1.384 -r1.385
> --- impbody.xml	24 May 2005 21:11:31 -0000	1.384
> +++ impbody.xml	30 May 2005 13:33:21 -0000	1.385
> @@ -1802,10 +1802,12 @@
>   <tgroup cols="3" colsep="1" rowsep="1" align="center">&cpu-type-header;<tbody>
>     <row><entry>&most-positive-fixnum;</entry>
>      <entry>2<superscript>24</superscript>-1 = 16777215</entry>
> -    <entry>2<superscript>32</superscript>-1 = 4294967295</entry></row>
> +    <entry>2<superscript>32</superscript>-1 = 4294967295 or
> +           2<superscript>48</superscript>-1 = 281474976710655</entry></row>
>     <row><entry>&most-negative-fixnum;</entry>
>      <entry>-2<superscript>24</superscript> = -16777216</entry>
> -    <entry>-2<superscript>32</superscript> = -4294967296</entry></row>
> +    <entry>-2<superscript>32</superscript> = -4294967296 or
> +           -2<superscript>48</superscript> = -281474976710656</entry></row>
>  </tbody></tgroup></table></section>
>  
>  <section id="bignum-lim"><title>Bignum Limits
>

this is no good.
if MOST-POSITIVE-FIXNUM is not uniquely determined by the CPU address
size, we need to specify more options.
the user should be able to figure out MOST-POSITIVE-FIXNUM from this
table for his computer.

-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.mideasttruth.com/>
<http://www.jihadwatch.org/> <http://pmw.org.il/> <http://ffii.org/>
All extremists should be taken out and shot.

Re: 48-bit fixnums

[Bruno H. <br...@cl...>]

Sam wrote:
> if MOST-POSITIVE-FIXNUM is not uniquely determined by the CPU address
> size, we need to specify more options.
> the user should be able to figure out MOST-POSITIVE-FIXNUM from this
> table for his computer.

Agreed. Now I had some time to measure clisp's speed with different
oint_data_len values. (On AMD64. "time make check-recompile".)

Result:

oint_data_len=48
user    0m21.145s
user    0m21.186s

oint_data_len=40
user    0m21.138s
user    0m21.190s

oint_data_len=32
user    0m20.808s
user    0m21.026s

As you can see, fixnums larger than 32 bit _do_ cost performance, ca. 1%
slowdown.

But there's no measurable difference between 40 and 48 bit. So I'm setting
oint_data_len to 48 for all 64-bit CPUs.

Bruno