AMX support for tiled matrix multiplications

[ashvardanian]

Both Intel and Apple now have specialized AMX tiled matrix multiplication extensions. Both are tricky to use, but may result in substantial performance improvements. Potentially even for single vector dot-products and cosine distances.

Resources:

• Intel: https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#techs=AMX
• Apple: https://github.com/corsix/amx

[MarkReedZ]

I'll try doing the BF16 dot product with this. Shall we add a new function matmul as well for matrix multiplication?

[ashvardanian]

I think it may be a new file called "dots.h", that implements matrix multiplications 🤗