[Cfp-interest 1526] AI about INFINITY and NAN macros and _Float16

[Jim Thomas]

Action item:
>     Jim: Write a note for _Float16 getting infinity as a _Float16 type in Part 3 as an annex.


Here’s a draft:

In the subclause for <math.h> macros, add:

NOTE	Because the INFINITY and NAN macros have type float, they can be used in a type-generic way and are converted, without raising a floating-point exception, to an infinity and a quiet NaN in the floating type determined by the context, provided that type is at least as wide as float. However, combined with operands only of type _Float16, the macros cause the semantic type of the expression to be float instead of _Float16. For example, if x has type _Float16, the expression x + INFINITY has semantic type float. If the narrower _Float16 type is desired, the macro should be cast to _Float_16: x + (_Float16)INFINITY.


- JIm Thomas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20200304/0f0109cd/attachment-0001.html