Proposed resolution:

This wording is relative to N5008.

1. Modify [simd.mask.unary] as indicated:

   constexpr basic_simd_mask operator!() const noexcept;
   constexpr basic_simd<integer-from<Bytes>, Abi> operator+() const noexcept;
   constexpr basic_simd<integer-from<Bytes>, Abi> operator-() const noexcept;
   constexpr basic_simd<integer-from<Bytes>, Abi> operator~() const noexcept;
   

   -1- Let op be the operator.

   -2- Returns: A data-parallel object where the i^th element is initialized to the results of applying op to operator[](i) for all i in the range of [0, size()).

   -?- Remarks: If Bytes is greater than 8, `operator+()`, `operator-()`, and `operator~()` are deleted.

Consider:

simd<complex<double>> c = {};
simd_mask<complex<double>> k = (c == c); // sizeof(complex<double>) == 16
auto i = -k; // simd<__int128> !

`basic_simd_mask` unary `+`, `-`, and `~` return basic_simd<integer-from<Bytes, Abi>>. [simd.expos.defn]/2 says:

integer-from<Bytes> is an alias for a signed integer type `T` such that `sizeof(T)` equals `Bytes`.

But `__int128` isn't a vectorizable type. Consequently, simd<__int128> currently is a disabled specialization. So it seems simd<complex<double>> wants `__int128` to be added to the list of vectorizable types.

Options:

1. "Specialize" basic_simd_mask<16, Abi> to return rebind_t<integer-from<8>, basic_simd_mask<integer-from<16>, Abi>> (reduces a 128-bit value of a vector-mask to a 64-bit value in the simd)

2. "Specialize" basic_simd_mask<16, Abi> to return resize_t<size() * 2, rebind_t<integer-from<8>, basic_simd_mask<integer-from<16>, Abi>>> (duplicates a 128-bit value of a vector-mask to two 64-bit values in the simd)

3. delete unary `+`, `-`, and `~` for basic_simd_mask<16, Abi> (closest to the status quo)