docs: prefer ptoas autosync in user-facing samples#255
docs: prefer ptoas autosync in user-facing samples#255zhoubot wants to merge 1 commit intozhangstevenunity:mainfrom
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7768da7397
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| # This sample intentionally contains NO explicit sync ops. | ||
| # | ||
| # Use it with: | ||
| # ptoas --enable-insert-sync |
There was a problem hiding this comment.
This update makes the sample depend on ptoas --enable-insert-sync, but first-party build/test paths still invoke ptoas without that flag (README.md lines 228-231, docker/Dockerfile lines 94-95 and 137-138, docker/test_ptoas_cli.sh lines 46-47). Because the CLI option defaults to off (tools/ptoas/ptoas.cpp lines 106-108), those existing flows now compile this sample with no explicit record_event/wait_event and no auto-inserted sync, which can leave TMOV/TMATMUL/TSTORE ordering unsafeguarded at runtime.
Useful? React with 👍 / 👎.
| %sub = pto.subset %src[%i, %j] sizes [32, 32] : !pto.tile_buf<loc=vec, dtype=f16, rows=64, cols=64, v_row=64, v_col=64, blayout=row_major, slayout=none_box, fractal=512, pad=0> | ||
| ``` | ||
|
|
||
| ##### `pto.set_validshape` - Rebind Tile Valid Row/Col |
There was a problem hiding this comment.
The manual now documents pto.set_validshape as a concrete PTO IR op, but there is no corresponding op definition or binding in the repository (e.g., include/PTO/IR/PTOOps.td has no "set_validshape" op, and repo-wide search in include/, lib/, python/, and test/ finds no implementation). Publishing a non-existent op in the primary IR reference will cause users to write IR that current tooling cannot parse or lower.
Useful? React with 👍 / 👎.
1 participant
Summary
ptoas --enable-insert-syncValidation
python3 test/samples/MatMul/tmatmulk.pyContext
This keeps PTOAS aligned with the current PTODSL authoring contract: kernel source should stay explicit-sync-free and PTOAS should insert the needed sync during lowering.