Optimizing Burst matrix multiplications

A deep dive into optimizing matrix multiplications. A story of multiple optimization attempts, moving from basic implementation to hand-tuned intrinsics.
Key takeaways from the optimization journey:
* Switching to Unity’s Mathematics package (SIMD-friendly) immediately gave a massive boost over standard C#.
* The Low-Hanging Fruit: Using FloatMode.Fast allows the compiler to merge instructions into Fused Multiply-Add (FMLA).
* Reinterpreting NativeArray<Matrix4x4> as float4x4 helps the compiler see the data layout more clearly, reducing instruction count.
* Utilizing Affine Matrices (since most game objects don't use perspective projection) allows you to skip calculating the 0,0,0,1 row, saving significant cycles.
* Using Intrinsics and float4x3 to pack data tightly into vector registers.
This is a must-read if you are working on high-performance physics, skinning, or animation systems in Unity.

https://medium.com/toca-boca-tech-blog/optimizing-burst-matrix-multiplications-bfb301de80ce

#burst #optimiziation