Add AVX-512 support#231

Open

Shnatsel wants to merge 40 commits into

linebender:mainfrom

Shnatsel:avx512-yes-really

Commits on May 24, 2026

Initial AVX-512 support. Adds level detection, mask operations, and dedicated AVX-512 implementations for complex int/float vector operations that benefit the most.
Shnatsel
committed
Add checked_transmute_copy and ban transmute_copy to statically prevent the spooky bug I almost introduced
Shnatsel
committed
Expand native type conversion test coverage
Shnatsel
committed
Rename test: mask_methods.rs -> mask_roundtrip.rs
Shnatsel
committed
Check in the new generated AVX-512 file
Shnatsel
committed
Fix build after file rename
Shnatsel
committed
Use AVX-512 instructions for f32 -> u32 conversions. Expand test coverage for these ops.
Shnatsel
committed
Optimize load_array/as_array on AVX-512 masks; the initial impl was scalar, now we use the dedicated intrinsics.
Shnatsel
committed
Split set_mask into a backend method so it could be specialized per backend, and specialize it for AVX-512. Add test coverage that sets every single bit and verifies it was set correctly.
Shnatsel
committed
Optimize load_interleaved/store_interleaved for AVX-512. Add one more test to exercise it.
Shnatsel
committed
Optimize variable shifts for 8-bit and 16-bit types. Expand test coverage. Only for 8-bit left shift LLVM autovectorizes the scalar fallback into GFNI instructions on 256-bit halves which emits mor…
Shnatsel
committed
Optimize floor/ceil/round_ties_even/trunc/approximate_recip for 512-bit vectors on AVX-512; expand test coverage
Shnatsel
committed
Use AVX-512 rcp14 for smaller vector sizes too; improves precision at no cost to throughput
Shnatsel
committed
Optimize slide_within_blocks for AVX-512; verified with exhaustive slide test
Shnatsel
committed
Remove stale tests for mask slide APIs; they were under #[cfg(false)] so they didn't show up earlier when I removed those methods.
Shnatsel
committed
consistent clippy error messages
Shnatsel
committed
satisfy Clippy
Shnatsel
committed
get rid of useless extra braces
Shnatsel
committed
KISS the native type mask roundtrip tests
Shnatsel
committed
cargo fmt
Shnatsel
committed
Satisfy clippy some more. Hoisted by my own restriction lint.
Shnatsel
committed
Satisfy the toml formatting check
Shnatsel
committed
Stick an #[expect] onto checked_transmute_copy on wasm32, otherwise we get dead code warnings
Shnatsel
committed
Suppress an apparently buggy Clippy lint; surfaced only in `cargo clippy --tests` without a reported location, I've failed to isolate it to a specific crate and suppress it there
Shnatsel
committed
Satisfy the toml formatter again
Shnatsel
committed
Add miri out-outs for extra slow tests
Shnatsel
committed

Commits on May 26, 2026

Commits on May 27, 2026