SIMD Compilation issues on mac M1

[RagnarGrootKoerkamp]

See RagnarGrootKoerkamp/astar-pairwise-aligner#17 originally:

error[E0425]: cannot find function `_mm_loadu_si128` in this scope
   --> /Users/finn/.cargo/registry/src/index.crates.io-6f17d22bba15001f/editdistancek-1.0.1/src/lib.rs:108:25
    |
108 |                 let x = _mm_loadu_si128(xs.as_ptr() as _);
    |                         ^^^^^^^^^^^^^^^ not found in this scope

error[E0425]: cannot find function `_mm_loadu_si128` in this scope
   --> /Users/finn/.cargo/registry/src/index.crates.io-6f17d22bba15001f/editdistancek-1.0.1/src/lib.rs:109:25
    |
109 |                 let y = _mm_loadu_si128(ys.as_ptr() as _);
    |                         ^^^^^^^^^^^^^^^ not found in this scope

error[E0425]: cannot find function `_mm_cmpeq_epi8` in this scope
   --> /Users/finn/.cargo/registry/src/index.crates.io-6f17d22bba15001f/editdistancek-1.0.1/src/lib.rs:110:25
    |
110 |                 let r = _mm_cmpeq_epi8(x, y);
    |                         ^^^^^^^^^^^^^^ not found in this scope

error[E0425]: cannot find function `_mm_movemask_epi8` in this scope
   --> /Users/finn/.cargo/registry/src/index.crates.io-6f17d22bba15001f/editdistancek-1.0.1/src/lib.rs:111:25
    |
111 |                 let r = _mm_movemask_epi8(r);
    |                         ^^^^^^^^^^^^^^^^^ not found in this scope

For more information about this error, try `rustc --explain E0425`.
error: could not compile `editdistancek` (lib) due to 4 previous errors

you should probably put the SIMD behind a feature flag or so, and/or use neon instructiong (see e.g. https://github.com/Daniel-Liu-c0deb0t/block-aligner/blob/main/src/neon.rs)

[nkkarpov]

Currently, I have simply wrapped the function definition to prevent any compilation errors on M1/M2 architectures. I plan to add support for aarch64 at a later stage.

[RagnarGrootKoerkamp]

ok, fine for me -- I don't actually depend on this anyway. Side note: I will probably make a small benchmark of extend (what you call mismatch) functions somewhere in the pairwise-alignment org, comparing:

• compare per char/u8
• compare per u64
• compare per 128-bit SIMD
• compare per 256-bit SIMD
  I also think there might be value in first checking if the first char matches, and only doing more complicated things if that's the case. Especially in WFA/diagonal transition, most calls are on random sequences, and 75% of time you can stop on first char mismatch already.

[RagnarGrootKoerkamp]

Oh, can you make a PR to rust-bio to update the dependency?

[nkkarpov]

The rust-bio should use the corrected version by default. Yes, it would be nice if you could do some benchmarks that on real-data sequences. Please, share with me your findings.
first char matches: yes, it sounds reasonable.

[RagnarGrootKoerkamp]

thanks for fixing!