Reinstate simd_cast but implement using rebind

[danieltowner]

Real-world use of rebind_simd_t suggests that simd_cast is a concise and readable way of indiciating what you mean, while rebind_simd_t is a bit verbose.

[mattkretz]

I'd like to see a paper to SG6 proposing a general value_type / representation type cast.

• std::value_cast<double>(1): int to double
• std::value_cast<int>(simd): simd<T> to rebind_simd_t<int, simd<T>>
• maybe std::value_cast<float>(5min): duration<int, std::ratio<60>> to duration<float, std::ratio<60>>
• std::value_cast<double>(1.f + 1.if): complex<float> to complex<double>
• ... esp. calling out to papers on numeric library types

[mattkretz]

@mpusz FYI. I guess such a cast would be interesting for mp-units as well. It would be equivalent to your quantity_cast overload for casting the representation type.

[mpusz]

Yes, we call it a representation_cast in mp-units V2 framework for now but value_cast is also a good name. However, in our case it will cast not only an underlying representation type but also will be used to change the unit as such conversion may also change/truncate the value. To make it even more entertaining, Unit is provided by value (NTTP) and representation type as a type parameter kind.

For example:

auto s1 = representation_cast<int>(speed);
auto s2 = representation_cast<km / h>(speed);

[mattkretz]

Assuming WG21 adds a std::value_cast with a specific goal to allow UDTs to reuse it. Then there would be some mechanism to call a function defined by mp-units that implements the cast.

I wrote a quick prototype: https://godbolt.org/z/M6e5q9ao7.

Who wants to run with it and write a paper? :wink: