[REFACTOR] Further cleanup node redirections#18844
[REFACTOR] Further cleanup node redirections#18844tqchen wants to merge 6 commits intoapache:mainfrom
Conversation
The header `include/tvm/runtime/packed_func.h` only re-exports `ffi::Any`, `ffi::AnyView` into `tvm::runtime` and aliases `TVM_DLL_EXPORT_TYPED_FUNC`. None of the 5 files that include it use any of those symbols, so the includes are dead. Remove all 5 dead includes and delete the now-unused header.
The AttrRegistry template is used by Op, TargetKind, and TargetTag, which all live under src/ir/ and src/target/. Moving the header to src/ir/ better reflects its logical home. Update all 5 include sites and update the include guard and file doc comment accordingly.
…on.h The headers `include/tvm/ir/serialization.h` and `include/tvm/node/serialization.h` declared `tvm::SaveJSON` and `tvm::LoadJSON` as thin wrappers over `ffi::ToJSONGraph` and `ffi::FromJSONGraph`. Remove these headers and replace all call sites with direct calls to the ffi functions. The `node.SaveJSON` and `node.LoadJSON` FFI registrations are preserved in `src/ir/serialization.cc` so the Python API (`tvm.ir.save_json` / `tvm.ir.load_json`) is unchanged.
… ffi The C++ `node.MakeNode` function in `src/node/reflection.cc` handled `DictAttrsNode` specially (using `InitByPackedArgs`) while delegating all other types to `ffi.MakeObjectFromPackedArgs`. Move the DictAttrs special case to the Python `make_node` wrapper in `python/tvm/ir/attrs.py` by passing the kwargs as the `__dict__` field of `DictAttrs`, and use `tvm_ffi._ffi_api.MakeObjectFromPackedArgs` directly for all other types. Remove `src/node/reflection.cc` which contained only the now-deleted `MakeNode` function and its FFI registration.
Summary of ChangesHello, 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 significantly refactors the TVM codebase by eliminating several layers of legacy indirection within the node and runtime layers. The changes focus on removing outdated header files, reorganizing core components for better logical placement, and modernizing serialization processes to directly utilize the Foreign Function Interface (FFI). This cleanup simplifies the architecture, reduces redundant abstractions, and aligns the system with more direct and efficient FFI interactions, ultimately improving maintainability and clarity. Highlights
🧠 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. Changelog
Activity
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 is a great cleanup of node and serialization indirections. The changes are well-contained and follow the summary in the description. I've found one recurring issue across several files where replacing tvm::SaveJSON with direct calls to ffi::ToJSONGraph drops the tvm_version metadata from the serialized JSON. This could lead to compatibility issues if these JSONs are persisted. I've added comments with suggestions to restore this metadata.
Restore the `tvm_version` metadata parameter to all `ffi::ToJSONGraph()` call sites that replaced the old `tvm::SaveJSON()`, which always included this metadata. Add an explicit `<tvm/runtime/base.h>` include in `src/script/printer/utils.h` for the `TVM_VERSION` macro.
1 participant
Summary
Remove node/serialization indirection headers and redirect to direct ffi API calls.
Changes