Difficult to get started

[cqh963852]

Hi everyone, I've recently been learning NPU programming and encountered many problems. Let me introduce my situation. I majored in Computer Science in college and am currently a full-stack engineer.
I've used WebGPU and learned GPU compute, and written some shaders.
I know there are some differences between NPUs and GPUs. So I wanted to try it out on consumer hardware, so I bought an HX370 mini PC.

However, when I searched for "AMD NPU program," various promotional pages popped up. I finally found this repository after searching through Gaia and other resources.

This process took me about two days, a whole weekend, and I found the information gathering process extremely painful. Compared to searching for "NVIDIA" using the keyword "CUDA," which returns the NVIDIA/CUDA-tile repository, "AMD NPU program" returns the amd/gaia repository.

After finding this repository, I spent about a day figuring out the functions of Python and C++ respectively. I can generate MLIR using Python and then execute it using C++ (please correct me if my understanding is incorrect).

During this process, I noticed some duplicate documentation; for example, there are four README.md files explaining how to install the environment, with only minor differences. I've previously looked at document systems and found that documentation based on them is generally more readable and has a gentler learning curve. Could some improvements be made to the documentation?

Although there are still some issues preventing me from getting it running on the NPU, I plan to read some relevant documentation and try running the code before asking questions here again.

[hunhoffe]

Hello! Certainly, improvements to documentation could be made and we welcome contributions to documentation (and to other areas of this code). If you continue to have issues, please do feel free to open issues as you encounter them!