Add ExecuteDeviceKernel frontend in JAX #32869
Conversation
Summary of ChangesHello @serach24, 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 introduces a powerful new frontend in JAX that allows users to execute custom device kernels, starting with NVIDIA PTX, directly within their JAX programs. This feature significantly extends JAX's capabilities by enabling low-level, device-specific optimizations while retaining compatibility with JAX's high-level transformations like JIT compilation, automatic differentiation, and automatic batching. The implementation includes robust mechanisms for defining kernel parameters, handling data layouts, and ensuring proper integration with JAX's execution model, making it easier to leverage specialized hardware features. Highlights
Using Gemini Code AssistThe 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
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 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
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a valuable new frontend for executing device-specific kernels, starting with PTX for NVIDIA GPUs. The implementation is well-structured, leveraging the existing FFI infrastructure effectively. The API design is clean, and the inclusion of comprehensive tests is commendable. I've identified a few minor issues, primarily concerning error messages, API consistency, and test code clarity, which are detailed in the comments.
| Returns: | ||
| Result array(s) from kernel execution | ||
| """ | ||
| # Validate kernel_type |
There was a problem hiding this comment.
The vectorized argument is deprecated in other JAX APIs like ffi_call. For consistency and to guide users towards the new API, a check should be added to kernel_call to raise an error if vectorized is used.
| # Validate kernel_type | |
| if not isinstance(vectorized, DeprecatedArg): | |
| raise ValueError( | |
| "The 'vectorized' argument of jax.experimental.device_kernels.kernel_call is deprecated. " | |
| "Use 'vmap_method' instead.") | |
| # Validate kernel_type |
| ) | ||
| if not all(isinstance(idx, int) and 0 <= idx < len(args) + expected_num_outputs for idx in output_indices): | ||
| raise ValueError( | ||
| f"Output indices must be integers in range [0, {len(args)}), got {output_indices}" |
There was a problem hiding this comment.
The error message for output_indices validation is incorrect. The upper bound of the valid range should be len(args) + expected_num_outputs, but the error message states len(args). This could be confusing for users.
| f"Output indices must be integers in range [0, {len(args)}), got {output_indices}" | |
| f"Output indices must be integers in range [0, {len(args) + expected_num_outputs}), got {output_indices}" |
| # assert jnp.allclose(b, jnp.array([4.0, 5.0, 6.0], dtype=jnp.float32)) | ||
| # assert jnp.allclose(expected, jnp.array([5.0, 7.0, 9.0], dtype=jnp.float32)) |
There was a problem hiding this comment.
This test contains commented-out assertions. They should be removed to improve test clarity and maintainability. An explicit check should be added to ensure input b is not modified by the kernel call, similar to the existing check for a.
| # assert jnp.allclose(b, jnp.array([4.0, 5.0, 6.0], dtype=jnp.float32)) | |
| # assert jnp.allclose(expected, jnp.array([5.0, 7.0, 9.0], dtype=jnp.float32)) | |
| assert jnp.allclose(b, jnp.array([4.0, 5.0, 6.0], dtype=jnp.float32)) |
1 participant
No description provided.