If you're preparing for systems or performance-engineering roles, this repo shows how a simple matmul evolves into a high-performance kernel.

It demonstrates:

why loop order matters

how cache locality dominates performance

how tiling + registers change everything

how multithreading scales

You can run all benchmarks with one script and see ~100× speedup from naive → optimized.

Good practice for:

low-level optimization

ML systems

HPC

performance interviews

Repo: https://github.com/arun-reddy-a/matmul-cpu