Resources

Some interesting resources related to optimizing for M series.

• Comparing Apple’s M1 matmul performance – AMX2 vs NEON https://nod.ai/comparing-apple-m1-with-amx2-m1-with-neon/

• Exploring the Neural Engine https://github.com/hollance/neural-engine

• Reverse Engineering https://gist.github.com/dougallj/7a75a3be1ec69ca550e7c36dc75e0d6f

• Apple Developer Forum (sizeof(long double) differs in Intel vs. ARM64 (ARM64 long doubles are just doubles). How to get extended precision in Apple Silicon?) https://developer.apple.com/forums/thread/673482

• Apple Developer (FloatingPoint) https://developer.apple.com/documentation/swift/floatingpoint

• Apple Developer (Floating Point Types) https://developer.apple.com/library/archive/documentation/System/Conceptual/ManPages_iPhoneOS/man3/float.3.html

• Can you trust floating-point arithmetic on Apple Silicon? https://eclecticlight.co/2021/04/22/can-you-trust-floating-point-arithmetic-on-apple-silicon/

• Accelerating the M1 Mac: an introduction to SIMD https://eclecticlight.co/2021/08/06/accelerating-the-m1-mac-an-introduction-to-simd/

• Apple Silicon https://developer.apple.com/documentation/apple-silicon

• What’s in an M1 chip, and what does it do differently? https://eclecticlight.co/2021/08/24/whats-in-an-m1-chip-and-what-does-it-do-differently/

• When idiomatic code is slower, and how to Accelerate When idiomatic code is slower, and how to Accelerate

• Reddit (Apple Silicon) https://www.reddit.com/r/Applesilicon/