Implement chunks_exact and chunks_exact_mut

[zesterer]

Follow-up PR from #510. chunks_exact (and its mutable counterpart) are not yet implemented on ParallelIterator, only ParallelSlice.

[seijikun]

Hey guys, is there any news on this?
I'm currently depending on @zesterer's branch with my crate (kmeans), because its already delivering a speedup by a factor of ~4x in its current form for me (as compared to par_bridge()).

I tried having a look at the produced asm code, but couldn't figure out if the chunk-size-check was still in there.

[cuviper]

@seijikun I'm not surprised that it would perform better than par_bridge, but have you compared to the inexact par_chunks / par_chunks_mut?

[seijikun]

@cuviper Ah, totally forgot about that. I wasn't able to use it when I decided to use this branch (because of the remainder). Refactored the code to manually pad the data (so that there is no other-sized remainder).
I now barely see a difference between par_chunks() and par_chunks_exact().

[cuviper]

That's a good point -- the remainder API has value regardless of size-check optimizations.

[Coder-256]

Is this still being worked on?

[cuviper]

I think this is ready now -- thanks for the PR, @zesterer!

bors r+

[bors]

Build succeeded

• Test (windows-latest, beta)
• Format
• Test (macos-latest, stable)
• Test (windows-latest, stable)
• Test (windows-latest, nightly)
• Demo (stable)
• Test (ubuntu-latest, beta)
• Test (ubuntu-latest, stable)
• Check (1.31.0)
• Test (macos-latest, beta)
• Test (ubuntu-latest, nightly)
• Test (macos-latest, nightly)
• Demo (nightly)
• WebAssembly