Optional four-message batch extension of Hashable. More...

Concept definition

template<typename Hash>
concept HasHash4 =  Hashable<Hash> &&
    requires(Hash hash, cuda::std::span<const cuda::std::array<int4, 4>, 4> msgs) {
  { hash.Hash4(msgs) } -> std::same_as<cuda::std::array<cuda::std::array<int4, 2>, 4>>;
}

Detailed Description

Optional four-message batch extension of Hashable.

Hash4() borrows four independent 64B messages through cuda::std::span<const cuda::std::array<int4, 4>, 4> and maps them to four 32B digests, indexed as [message_index][output_index]. Custom hash backends opt in by providing Hash4(msgs); no registration, inheritance, or specialization is required. Host evaluators use this method when available and otherwise call Hash() four times.

For every j in [0, 4), Hash4(msgs)[j] must equal Hash(msgs[j]) word for word, evaluated with the same object and configuration. Hash4() must be callable on the host. It borrows the messages only for the duration of the call, must not modify them, and must not retain pointers or references to them; the returned digests are owned by the caller. Hash4() must not change the backend hash configuration, and scalar Hash() calls made afterwards with the same object remain valid. Concurrent-call requirements are the same as for the existing scalar Hash(); this concept adds no thread-safety promise.

This concept checks only the expression shape and cannot verify these semantic or host-callability requirements.

Template Parameters

Hash	User-provided or library-provided hash type.