Is any plan to publish the manual like nvidia ptx spec?

[alanzhai219]

It is a bit hard to read and understand the syntax.
Do you have any plan to refactor and publish like nvidia PTX virtual instructions, which is easy to understand and invoke?

[ozhdikhin]

Hello @alanzhai219

Thank you for reaching out and for your interest in our project. At the moment, we do not plan to refactor the documentation beyond the information presented in https://github.com/intel/vc-intrinsics/blob/master/GenXIntrinsics/include/llvm/GenXIntrinsics/Intrinsic_definitions.py. If you have specific questions about any of the supported intrinsics, please feel free to send them our way.

Best regards,
Pavel Ozhdikhin

[alanzhai219]

I mean why not publish a manual like https://docs.nvidia.com/cuda/pdf/ptx_isa_8.5.pdf.
A clear manual can help developers to code. It is not a good idea to solve problems through word of mouth; it's too time-consuming. Just maintain a good manual and developers can learn by themselves. I think it is a good way.

[ozhdikhin]

I totally agree that such a manual would be valuable. However, it currently sits low on our priority list. The reason is that most people use mainstream languages like DPC++ (Data Parallel C++) or DPC++ ESIMD (Explicit SIMD) extensions and do not use VC intrinsics directly. Those who use VC intrinsics to implement their programming languages usually know the architecture details very well and do not need detailed manuals. I hope that at some point we'll manage to provide detailed documentation. Until this happens, we can answer questions.

Best regards,
Pavel Ozhdikhin

[alanzhai219]

@ozhdikhin I see. It makes sense for now.

[alanzhai219]

btw, do you guys read the popular paper, deepseek? The kernel is written in ptx and the perf is incredible. That is why I highlight the GPU assembly. Please pay attention to it in the following gens. @ozhdikhin

[alanzhai219]

@ozhdikhin Should provide different API from high level to low level.
Really, really, really trust external programmers ability.
Polish the GPU software stack, tools. Help developers win. Intel will win.
Please.

[ozhdikhin]

@alanzhai219 Your feedback aligns with the strategic direction we are taking in the development of Intel's future GPU software stack. Stay tuned for updates.