Named permutation functions

[danieltowner]

There is a proposal to add a generated permutation function which can be used like this:

auto evenValues = permute<_OutputSize>(simd_value, [](size_t idx) -> size_t { return idx % 2; };

This allows an arbitrarily sized output simd to be created, where the mapping of input index to output index is specified through a supplied generator function. It is inevitable that certain permutations are common, and it would be desirable to have pre-defined functions for freeing users from having to reinvent them. By providing named functions it also becomes possible for the library implementation to provide fast alternatives where the target supports them natively.

A few common functions could include:

reverse(simd_value);   // Reverse the order of the elements in the simd.
                       // Could be given a [_Begin, _End] range sub-region to reverse.
shift_left(simd_value, offset);
slice(simd_value, start, size, stride); // Like valarray::slice
rotate(simd_value, offset);
interleave(value0, value1); // Interleave values from each argument in turn (equivalent to matrix transpose by simd rows)

Like issue #38, should these allow dynamic or static behaviour?

Where these functions could operate on subsets of a simd, should it be up to the user to handle the subset explicitly? For example, to reverse a subset of elements in a simd:

auto e = extract<_Begin, _End>(source);
auto r = reverse(e);
output = insert<_Begin>(source, r);

Note that insert, extract and resize are already pre-defined permutations which can be implemented using then generic permute-with-generator.