Hello,

>> FWIW, the IEEE 754-2008 standard specifies 16 bit floating point numbers 
>> (referred to as "binary16").
>> According to <http://en.wikipedia.org/wiki/Half_precision>:
>> 
>> | It is intended for storage (of many floating point values where higher 
>> precision need not be stored),
>> | not for performing arithmetic computations.
> Right. The main problem is how we now define "storage only type" in C / C++ :)

When in an arithmetic context, automatic promotion to some 
implementation-defined floating-point type
that has at least binary16's accuracy, I would propose. Just like integer 
calculations are always performed
on at least int (I sure hope I remember this correctly ;-).

This floating-point type could be whatever is best for the platform.


Jonathan


_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits

Reply via email to