[bluss]

The new version of matrixmultiply supports AVX-512 (by the usual target feature detection), as a default feature (but ndarray needs to turn it back on).

• Changelog link
• PR with some benchmarks: Add AVX-512 kernel bluss/matrixmultiply#100
• Better performance on the upside and slightly longer compile / larger code size as a downside (x86-64 only).

[bluss]

Thanks! (All good for merge?)

[akern40]

Merged!