macOS MacBook Pro M1 MAX, M1 Ultra, M2, Usage of sdot/udot in ARM

[Appletoni]

Modern ARM processors, including Apple silicon, support sdot/udot instructions which could improve many of your NNUE operations, also overriding Simd::neon_m128_add_dpbusd_epi32x2 could help.

Usage of sdot/udot in ARM · Issue #4193 · official-stockfish/Stockfish

Modern ARM processors, including Apple silicon, support sdot/udot instructions which could improve many of your NNUE operations, also overriding Simd::neon_m128_add_dpbusd_epi32x2 could help. Infor...

github.com

Any ideas how to make it work?

 

[Nygaard]

What do you mean? If the processor supports those instructions, you simply use them. The compiler will automatically use them in obvious situations (assuming it knows about the instructions). Otherwise, you can write assembly/intrinsics and directly use them yourself.