[PATCH] Allow FMAs in safe math mode in some cases when one operand of the fmul is either exactly 0.0 or exactly 1.0.

[Stephen Lin]

On Wed, Jul 10, 2013 at 12:37 AM, Stephen Lin <swlin at post.harvard.edu> wrote:
>> What is the behavior of FMA if the intermediate result overflows?
>> E.g. what does 2 * __DBL_MAX__ - __DBL_MAX__ give as FMA?
>>
>> Joerg
>
> I could be wrong but I don't think it even makes sense to ask about
> the intermediate result overflowing; I think the operation is defined
> as given whatever value would be obtained via an infinite-precision
> real number calculation, rounded at the end to the nearest
> representable double at the end, so there's no intermediate register
> to overflow. Presumably the result would be __DBL_MAX__ in the case
> you give.

(nearest representable double assuming the final result does not
overflow, of course...)