Constexpr next #789
Constexpr next #789
Conversation
[ci skip]
[ci skip]
[ci skip]
[ci skip]
[ci skip]
[ci skip]
[ci skip]
[ci skip]
|
@NAThompson Can you review this? It is almost a 1:1 copy of the existing nextafter implementation with standard library cmath functions being replaced by our ccmath functions. The results match exactly which makes me feel good about the other functions. |
|
@mborland : Looks good; make sure to squash merge with a nice commit message. |
[ci skip]
| constexpr result_type float_next(const T& val); | ||
|
|
||
| template <typename T> | ||
| struct has_hidden_guard_digits; |
There was a problem hiding this comment.
In the runtime version of this code, these traits classes were introduced as a point of specialization for multiprecision types that have guard digits (cpp_dec_float, gmp_float). I think it's probably irrelevant for this use case were we'll be dealing with builtin types only.
| // Special version for some base other than 2: | ||
| // | ||
| template <typename T> | ||
| constexpr T float_next_imp(const T& val, const std::false_type&) |
There was a problem hiding this comment.
Do we actually need to support bases other than 2 in the constexpr case? I would be inclined to remove this and static_assert that the radix is 2.
| // Special version for bases other than 2: | ||
| // | ||
| template <typename T> | ||
| constexpr T float_prior_imp(const T& val, const std::false_type&) |
There was a problem hiding this comment.
as above, possibly delete all this?
Implements constexpr:
nextafter,nextafterf,nextafterl,nexttoward,nexttowardf, andnexttowardl. Once this and #734 are complete all functions pending standardization will be constexpr.