Re: [Boost-users] [numeric_conversion] rounding

From: Peter (p.schregle_at_[hidden])
Date: 2008-04-28 11:14:30

Rainer,

this works, thanks a lot.

However, as I could see by single stepping with the debugger, this seems to 
do much more than what I want, so I suspect it has efficiency issues. I will 
be doing this for pixels of images, so it better be as fast as possible.

It seems to do overflow checking, which I don't need (I know it won't 
overflow). How can I disable this? I tried silent_overflow_handler, but this 
seems still to do the checking.

Also I'm not sure if the rounding couldn't be more efficient (I only need to 
add 0.5, no need to be more elaborate). Is there a way to plug in my own 
rounding?

Thanks
Peter

"Rainer Thaden" <RThaden_at_[hidden]> schrieb im Newsbeitrag 
news:4815DAFE.8020005_at_web.de...
> Hi Peter,
>
>
> Peter wrote:
>> I'm wondering how to use the numeric_conversion library to perform 
>> rounding
>> based on types.
>>
>> Assume I have a double value of 3.7. If I convert it to integer, the 
>> result
>> should be 4. If I convert it to float, the result should be 3.7. The 
>> types I
>> want to convert from and to are both specified in a template, i.e. they 
>> form
>> an open set of types.
>>
>>>From reading the numeric_conversion docs I tend to think that the library
>> can do what I want, but I'm unable to find out how.
>>
>> Can anybody help?
>
> the default behaviour of numeric_cast seems to be to truncate so 3.7
> will be casted to 3.
> I asked nearly the same question some time ago and got a useful answer
> from Fernanco Cacciola. Look at his post from 23.04.08 (don't know how
> to link it here)
>
> Maybe I should just paste it, so here it comes:
>
>> numeric_cast<> uses the default policies, but you an easily create you 
>> own
>> version with any other policy, like the one which rounds floats to the
>> nearest integer: RoundEven<>
>>
>> #include <boost/numeric/conversion/converter.hpp>
>>
>> template<typename Target, typename Source>
>> inline
>> Target my_numeric_cast ( Source arg )
>> {
>>   typedef boost::numeric::conversion_traits<Target,Source> Traits ;
>>
>>   typedef
>> boost::numeric::converter<Target,Source,Traits,boost::numeric::def_overflow_handler,boost::numeric::RoundEven<Source>
>>  > Converter ;
>>
>>   return Converter::convert(arg);
>> }
>>
>>
>> HTH
>>
>> -- 
> >Fernando Cacciola
>
>
> Regards,
>
> Rainer