GFNI as microkernel

[Malkovsky]

Few years ago Intel released Galois Field New Instructions mostly for something like Reed-Solomon codes and Rijndael encryption. Besides multiplication in GF(256) it also provides a GF2P8AFFINEQB instruction which technically computes A=BX+C in chunks of 2/4/8 for 128/256/512 registers where A, B, X being 8x8 binary matrices, C is binary vector of 8. This instruction can entirely replace M4R precomputed tables.

Here's an example
https://github.com/Malkovsky/galois/blob/main/src/field.cc#L187

I can contribute if you help me with the entry point.

[malb]

That would be quite interesting and I'm happy to help. Note that I haven't worked on this code actively in like a decade, though :)