Resources for understanding and developing assembly implementation

[Greenie0701]

Hi @martin-frbg

First of all, thank you for maintaining OpenBLAS, the repo is very well-structured and easy to follow.
I’m interested in contributing to ARMv8 assembly math kernels, but as a beginner it’s difficult to understand where to start. Could we have some documentation or pointers on:

• How the ARMv8 kernels are organized in the repo,
• Common patterns used in the assembly implementations,
• Learning resources or references for beginners.

Even a high-level overview would be a big help for new contributors who want to get started with ARMv8 kernel development.
Thanks!

[Mousius]

Hiya @Greenie0701,

Glad to hear you're keen to help out 😸

Most of the kernels live under kernel/arm64, and they are mapped to targets via the KERNEL. Makefile's like KERNEL.ARMV8SVE or KERNEL.NEOVERSEV1 - which extends ARMV8SVE.

For assembly implementations, there's inlined assembly like dot_kernel_asimd.c and then there's full assembly kernels like sgemm_kernel_sve_v2x8.S. There isn't a strict convention as far as I know.

[Greenie0701]

Hey @Mousius
Thanks for your kind words, this is really helpful, but if you can recommend some docs that could guide beginners to learn assembly implementation from scratch that could be a additional help😊

[Mousius]

We have a an introduction to assembly in general on developer.arm.com: https://developer.arm.com/documentation/107829/0201?lang=en

I found these helpful for learning about SVE specifically:
https://developer.arm.com/-/media/developer/products/software-tools/hpc/White%20papers/arm-scalable-vector-extensions-and-application-to-machine-learning.pdf?revision=510ee340-fce1-4fd8-bad6-bade674620a5
https://www.stonybrook.edu/commcms/ookami/support/_docs/5%20-%20Advanced%20SVE.pdf

And one just about GEMM 😸
https://gist.github.com/nadavrot/5b35d44e8ba3dd718e595e40184d03f0

[Greenie0701]

Thanks @Mousius for providing docs and references :)) this is really wonderful