[feat] add catcoc mm+ar & ag+mm

[ltcs11]

1. apply 2 stage compile for catcoc (kernel must compile in bisheng)
2. add matmul+allreduce & allgather+matmul ops
3. add unit test of both ops, support fp16/bf16 weightNZ/ND testcases

[gemini-code-assist]

Summary of Changes

Hello @ltcs11, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request integrates a new catcoc module, specifically adding a fused all-gather matrix multiplication kernel. This kernel is designed to optimize performance for distributed computing scenarios on NPU devices by combining computation and communication. The changes include updates to the build system, implementation of the host and device kernels, registration as a PyTorch operator, and a dedicated Python test to ensure correctness.

Highlights

• New catcoc Module Integration: Introduced a new catcoc module into the build system, enabling its compilation and linking with necessary dependencies like shmem.
• Fused AllGatherMatmul Kernel: Added a new fused kernel, catcoc_allgather_matmul, which combines matrix multiplication with an all-gather communication operation. This is designed for optimized distributed computing on NPU devices.
• PyTorch Operator Registration: The new catcoc_allgather_matmul operation is registered as a PyTorch operator (torch.ops.npu.catcoc_allgather_matmul), making it accessible from Python.
• Host and Device Kernel Implementations: Provided both host-side (catcoc_allgather_matmul.cpp) and device-side (catcoc_allgather_matmul_kernel.cpp) implementations for the fused kernel, leveraging Catlass, Catcoc, and shmem libraries.
• Comprehensive Testing: Included a Python test script (test_catcoc_allgather_matmul.py) to validate the functionality of the new fused kernel in a distributed environment, comparing its output against native PyTorch operations.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces a new catcoc_allgather_matmul operation, including its host and kernel implementation, build system integration, and a Python test. The changes are a good starting point for the new feature. My review focuses on improving the build configuration, fixing critical bugs in both the host implementation and the test script, and suggesting some code quality improvements for better maintainability. Key issues include a data type mismatch in the host code that would lead to a crash, incorrect usage of the in-place operation in the Python test, and a typo in a function call. There are also opportunities to reduce code duplication in the CMake files.