ROCm build broken

[ericcurtin]

The build flags changed in upstream llama.cpp so we were no longer building with ROCm acceleration.

Summary by Sourcery

Fix ROCm builds.

Build:

• Use -DGGML_HIP=ON build flag instead of -DGGML_HIPBLAS=1.
• Remove unnecessary ROCm libraries after build.

[sourcery-ai]

Reviewer's Guide by Sourcery

The build flags for llama.cpp were updated to enable ROCm acceleration. Additionally, the script was updated to remove more ROCm libraries.

Flow diagram for updated ROCm build configuration

graph TD
    A[Start Build] --> B{Container Type?}
    B -->|ROCm| C[Enable ROCm/HIP]
    B -->|CUDA| D[Enable CUDA]
    B -->|CPU| E[CPU Only Build]
    C --> F[Set -DGGML_HIP=ON]
    D --> G[Set -DGGML_CUDA=ON]
    F --> H[Clean Up ROCm Libraries]
    G --> I[Build Complete]
    E --> I
    H --> I
    style C fill:#f9f,stroke:#333
    style F fill:#f9f,stroke:#333

Loading

File-Level Changes

Change	Details	Files
Updated build flags to enable ROCm acceleration.	Changed -DGGML_HIPBLAS=1 to -DGGML_HIP=ON.	container-images/scripts/build_llama_and_whisper.sh
Updated the script to remove more ROCm libraries.	Removed the explicit listing of rocblas and hipblaslt libraries and instead remove all libraries in the /opt/rocm-*/lib/*/library/*gfx9* directory.	container-images/scripts/build_llama_and_whisper.sh

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time. You can also use
  this command to specify where the summary should be inserted.

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sourcery-ai]

Hey @ericcurtin - I've reviewed your changes and they look great!

Here's what I looked at during the review

• 🟢 General issues: all looks good
• 🟢 Security: all looks good
• 🟢 Testing: all looks good
• 🟢 Complexity: all looks good
• 🟢 Documentation: all looks good

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[ericcurtin]

@maxamillion @bmahabirbu upstream llama.cpp changed their build flags for ROCm, as a result we've been building without ROCm here for a little while as the code would still build, just without ROCm.

[ericcurtin]

This was happening basically:

CMake Warning:
  Manually-specified variables were not used by the project:

    GGML_HIPBLAS

[ericcurtin]

@rhatdan we probably need new container images, once we get the green PRs in here. ROCm is broken

[bmahabirbu]

Really good catch! I wonder if there is a way to catch these sort of errors? I can look into editing the make file to look for mismatched arguments

[ericcurtin]

Really good catch! I wonder if there is a way to catch these sort of errors? I can look into editing the make file to look for mismatched arguments

If you find out great, it would be appreciated, I tried briefly to make cmake treat this as an error, but didn't figure it out.

[ericcurtin]

I thought of something to check for cmake warnings @bmahabirbu , I'd prefer something that was built-into cmake, as it would be more elegant than this, but this should work.

[ericcurtin]

This build may actually fail, it might find other hidden warnings around the place

[maxamillion]

LGTM!

This is great, thank you for catching this! This might explain why I wasn't seeing any GPU utilization on my Ryzen APU when watching nvtop ... I had tossed it up to nvtop oddities.

[ericcurtin]

LGTM!

This is great, thank you for catching this! This might explain why I wasn't seeing any GPU utilization on my Ryzen APU when watching nvtop ... I had tossed it up to nvtop oddities.

Could have been! If you want to test this before it gets pushed, there is a flag to manually pass container images to RamaLama (you could build this locally).