[Dy2St] Support pass cuda graph state and dispatch key in run program op and move get value name to python side #73417
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
你的PR提交成功，感谢你对开源项目的贡献! |
…spatch-key-in-run-program-op
SigureMo
commented
Jun 19, 2025
There was a problem hiding this comment.
Pull Request Overview
This PR introduces new support for passing cuda graph state and dispatch key parameters to the run program op. Key changes include:
- Adding a new IntEnum (CUDAGraphState) to represent the different cuda graph states.
- Updating attribute preparation and caching logic to include cuda graph state and dispatch key.
- Propagating the new parameters in various op function bindings and interpreter interfaces.
Reviewed Changes
Copilot reviewed 13 out of 13 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| python/paddle/jit/dy2static/utils.py | Added CUDAGraphState IntEnum for cuda graph states. |
| python/paddle/jit/dy2static/pir_partial_program.py | Refactored attribute preparation and run implementation to include cuda graph state and dispatch key. |
| paddle/fluid/pybind/pir.cc | Bound a new helper to retrieve fake value names. |
| paddle/fluid/pybind/op_function_common.cc | Added new attribute definitions for cuda graph state and dispatch key. |
| paddle/fluid/framework/{new_executor/,executor_cache.} | Updated interpreter and caching logic to handle new cuda graph parameters. |
| paddle/fluid/eager/to_static/{run_program_op_node.h,run_program_op_func.h} | Integrated cuda graph state support through new attribute extraction and usage. |
Comments suppressed due to low confidence (1)
paddle/fluid/pybind/op_function_common.cc:1257
- [nitpick] Consider adding inline comments for the new 'cuda_graph_state' and 'cuda_graph_dispatch_key' attributes to document their expected values in accordance with the CUDAGraphState enum, which will help future maintainability.
{"cuda_graph_state", CastPyArg2AttrLong},
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
SigureMo
commented
Jun 19, 2025
zyfncg
previously approved these changes
Jun 20, 2025
zhangbo9674
previously approved these changes
Jun 20, 2025
…spatch-key-in-run-program-op
…spatch-key-in-run-program-op
|
skip-reason: 执行器已经 review 过，后续更新无相关变动 |
gouzil
approved these changes
Jun 21, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR Category
Execute Infrastructure
PR Types
New features
Description
动转静 run program op 支持传递
cuda_graph_state和cuda_graph_dispatch_key，以确保CudaGraphInstruction能够感知cuda_graph_state的变化本 PR 会和 #73393 联调
另外降低 value 获取 name 开销，将所有 name 获取统一移到 python 端，移除 value 传递到 cpp 端的操作（目前
fx还没有删掉）