optimization-manual VS portable-simd

There's two ways I vectorized linear and binary search (in practice you often want a combination, always benchmark on your real-world datasets!) - Do N binary searches simultaneously, each lane is essentially doing one bsearch. Obviously, this only works if you are doing multiple searches. - use the VPCONFLICT instruction for the linear search parts, there's even code from the Intel SDM doing it: https://github.com/intel/optimization-manual/blob/main/chap18/ex20/avx512_vector_dp.asm

The Intel optimization manual has a fun example where they use vpconflict for vectorizing sparse dot products: https://github.com/intel/optimization-manual/blob/main/chap1...

I benchmarked it on Intel, and it was indeed quite fast/a good improvement over the scalar version. Will be interesting to try that on AMD.

Jubilee and Caleb Zulawski for their tireless work on the portable SIMD project. It will land, some day, and when it does it's going to be an amazing boon for the project.

The portable SIMD project implements an as_simd() function for slices. The basics are that you get 3 slices and the middle one is a SIMD slice. It allows for fast aligned loads of the data, which could matter if your algorithm is becoming memory bound; it is also a convenient and safe abstraction. In my case, I actually have 2 vectors (say, x and y). I can take them apart using as_simd() like so:

Rust has generics and monomorphization. You can write the algorithm once and compile for multiple targets. rust-lang/portable-simd

This Rust issue [0] was the best short summary of what an SIMD Shuffle is I could find:

„A "shuffle", in SIMD terms, takes a SIMD vector (or possibly two vectors) and a pattern of source lane indexes (usually as an immediate), and then produces a new SIMD vector where the output is the source lane values in the pattern given.“

[0] https://github.com/rust-lang/portable-simd/issues/11

Yet by default it generates code which is only compatible with Pentium4 or newer. In fact lots of serious issues relate to older CPUs and rustc developers plan is to declare them closed when they would be able to drop i686 support (all AMD CPUs which support SSE2 support x86-64, too while Intel situation is mess).

The standard API crate for portable simd is at https://github.com/rust-lang/portable-simd, but using this requires nightly, which I don't want to do. I'd like to use a crate for simd that works on both x86_64 and wasm in stable rust. wide looks fine for this purpose. Are there any potentially better choices?

See: https://github.com/rust-lang/portable-simd

As for portable SIMD, there's relatively recent activity (last commit 20 days ago) on this repository: https://github.com/rust-lang/portable-simd