Re: Numerics
josé bollo <[email protected]>
| Newsgroups | gmane.comp.lang.eiffel.smalleiffel |
|---|---|
| Message-ID | <[email protected]> |
Dominique Colnet a écrit :
> josé bollo wrote:
>
>> hi all,
>
> Hi,
>
>>
>> i have looked at the generated C code!
>
> Good ;-)
>
>>
>> supposing that x is a REAL_32, x+1 is a valid expression and its C
>> form is something like x+(int8)(1)
>>
>> when i write x+{REAL_32 1.0} then the C from is somthing like that
>> x+(float)(1.0F)
>>
>> so the way it is written influencies the performances
>
> Yes. And it is a good news.
not really because it means that the compiler don't know that 1.0F is better
>
>>
>> but one can not write x+1.0
>
> I think you can. Here the traditional balancing rule apply and this
> expression is promoted to REAL_64.
but i want a REAL_32 as result
>
>>
>> it is very complicate
>
> Yes and we are here to find the best solution.
> Best regards as usual,
regards