[Openmcl-devel] roundings...

Gary Byers gb at clozure.com
Sat Sep 8 20:02:48 UTC 2012


I'd parsed the original question as asking how to change the rounding
mode used by the hardware, in which case section 4.11 of the CCL manual
would be suggested reading.

CCL defaults to using "round to nearest [even number]" and IIRC C defaults
to "round towards 0", so things like (ROUND 1.5) will yield different 
results than the C equivalent.

? (get-fpu-mode :rounding-mode)
:NEAREST
? (round 1.5)
2
-0.5
? (set-fpu-mode :rounding-mode :zero)
14592
? (round 1.5)
1
0.5
?


On Fri, 7 Sep 2012, Bill St. Clair wrote:

> Joakim,
>
> Please send an example of a .5 case that doesn't round to the even number.
>
> Your example is running into the limit of resolution of single floats.
> 5.4999999 IS 5.5, so it rounds to 6:
>
> CL-USER> 5.4999999
> 5.5
> CL-USER> (round *)
> 6
> -0.5
> CL-USER> (= 5.4999999 (+ 5 1/2))
> T
> CL-USER> 5.4999999d0
> 5.4999999D0
> CL-USER> (round *)
> 5
> 0.4999998999999997D0
>
> Note the rounding error in the final result. Almost, but not quite
> 0.4999999 . So it is with floats. If you need precise math, use
> RATIONALs (INTEGERs and RATIOs).
>
> If your C code works differently, are you sure you're using single
> floats there? May I see it?
>
> -Bill
>
> On Fri, Sep 7, 2012 at 4:30 AM, Joakim Sandgren
> <joakim at joakimsandgren.com> wrote:
>> Hi,
>>
>> how can I do a more 'c-like' rounding in lisp ?
>> that is in the .5 cases not always go to the even number.
>> i would like to have it to round up
>> and perhaps have better float resolution ? in lisp: (round 5.499999) => 5 (round 5.4999999) => 6
>>
>>
>>
>> are there some way around this?
>>
>> sincerely
>> joakim
>>
>>
>>
>>
>>
>> _______________________________________________
>> Openmcl-devel mailing list
>> Openmcl-devel at clozure.com
>> http://clozure.com/mailman/listinfo/openmcl-devel
> _______________________________________________
> Openmcl-devel mailing list
> Openmcl-devel at clozure.com
> http://clozure.com/mailman/listinfo/openmcl-devel
>
>



More information about the Openmcl-devel mailing list