[SOT][FasterGuard] add `IsDenseTensorHoldAllocationMatchGuard` #72823

gouzil · 2025-05-20T14:10:17Z

PR Category

Performance Optimization

PR Types

Performance

Description

添加 IsDenseTensorHoldAllocationMatchGuard 用于检查 tensor 是否分配

paddle-bot · 2025-05-20T14:10:23Z

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

Copilot

Pull Request Overview

Adds a new guard to the JIT executor that checks whether a dense tensor holds an allocation before proceeding.

Uses IsDenseTensorHoldAllocationMatchGuard in make_faster_guard when encountering a null/uninitialized tensor.
Declares and implements the new guard in C++ (guards.h and guards.cc).
Exposes the guard to Python via pybind11 binding in jit.cc.

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
python/paddle/jit/sot/opcode_translator/executor/variables/basic.py	Insert `IsDenseTensorHoldAllocationMatchGuard` in `make_faster_guard`
paddle/fluid/pybind/sot/guards.h	Declare `IsDenseTensorHoldAllocationMatchGuard`
paddle/fluid/pybind/sot/guards.cc	Implement `IsDenseTensorHoldAllocationMatchGuard::check`
paddle/fluid/pybind/jit.cc	Bind the new guard to Python

Comments suppressed due to low confidence (1)

python/paddle/jit/sot/opcode_translator/executor/variables/basic.py:529

[nitpick] This new guard path is introduced without accompanying unit tests to verify behavior when tensors do or do not hold allocation; consider adding tests to cover both scenarios.

paddle.framework.core.GuardNode(

paddle/fluid/pybind/sot/guards.cc

Copilot · 2025-05-20T14:13:03Z

paddle/fluid/pybind/sot/guards.cc

+    return false;
+  }
+
+  PyObject* result = PyObject_CallOneArg(method, value);


The result object returned by PyObject_CallOneArg is never decref'd, causing a memory leak. Add Py_DECREF(result) after extracting its truth value.

Copilot · 2025-05-20T14:13:04Z

paddle/fluid/pybind/jit.cc

+             std::shared_ptr<IsDenseTensorHoldAllocationMatchGuard>>(
+      *m,
+      "IsDenseTensorHoldAllocationMatchGuard",
+      R"DOC(IsDenseTensorHoldAllocationMatchGuard Class.)DOC")


[nitpick] The pybind11 binding for IsDenseTensorHoldAllocationMatchGuard lacks a descriptive docstring. Including its purpose and expected behavior would help maintainers.

Suggested change

R"DOC(IsDenseTensorHoldAllocationMatchGuard Class.)DOC")

R"DOC(

IsDenseTensorHoldAllocationMatchGuard Class.

This guard checks whether a DenseTensor instance holds a valid allocation.

It is used to ensure that the memory allocation associated with a DenseTensor

is consistent and valid during runtime. This is particularly useful in scenarios

where memory integrity and allocation state need to be verified.

Example:

guard = IsDenseTensorHoldAllocationMatchGuard()

# Use the guard in conjunction with other runtime checks.

)DOC")

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull Request Overview

This PR introduces a new guard IsDenseTensorHoldAllocationMatchGuard to optimize tensor allocation checks in the SOT executor.

Defines and implements IsDenseTensorHoldAllocationMatchGuard in C++ (guards.h / guards.cc)
Exposes the new guard to Python via pybind11 in jit.cc
Applies the guard in the Python executor when meta.is_null()

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
python/paddle/jit/sot/opcode_translator/executor/variables/basic.py	Use the new guard in the null-meta branch
paddle/fluid/pybind/sot/guards.h	Declare the `IsDenseTensorHoldAllocationMatchGuard` class
paddle/fluid/pybind/sot/guards.cc	Implement the guard’s `check` method
paddle/fluid/pybind/jit.cc	Bind `IsDenseTensorHoldAllocationMatchGuard` to Python

Comments suppressed due to low confidence (2)

paddle/fluid/pybind/sot/guards.cc:240

No tests have been added to verify the behavior of IsDenseTensorHoldAllocationMatchGuard::check. Consider adding unit tests for both allocated and unallocated tensor cases to ensure correctness.

bool IsDenseTensorHoldAllocationMatchGuard::check(PyObject* value) {

python/paddle/jit/sot/opcode_translator/executor/variables/basic.py:530

The code references IsDenseTensorHoldAllocationMatchGuard under paddle.framework.core, but the pybind binding is in the JIT module namespace. Ensure the class is exported in paddle.framework.core or adjust the import path to match where it's bound.

paddle.framework.core.IsDenseTensorHoldAllocationMatchGuard(),

Copilot · 2025-05-20T15:17:00Z

paddle/fluid/pybind/jit.cc

+             std::shared_ptr<IsDenseTensorHoldAllocationMatchGuard>>(
+      *m,
+      "IsDenseTensorHoldAllocationMatchGuard",
+      R"DOC(IsDenseTensorHoldAllocationMatchGuard Class.)DOC")


[nitpick] The binding for IsDenseTensorHoldAllocationMatchGuard lacks a descriptive docstring. Consider enhancing the R"DOC(...)" block to explain the guard’s purpose and when it should be used.

Suggested change

R"DOC(IsDenseTensorHoldAllocationMatchGuard Class.)DOC")

R"DOC(

IsDenseTensorHoldAllocationMatchGuard Class.

This guard is used to check whether a DenseTensor object holds a specific

memory allocation. It is primarily used in scenarios where the memory

allocation of a DenseTensor needs to be validated for correctness or

consistency during runtime.

Use this guard when you need to ensure that a DenseTensor's memory

allocation matches expected criteria, such as during debugging or

when implementing custom tensor operations.

Example:

guard = IsDenseTensorHoldAllocationMatchGuard()

# Use the guard in your logic to validate DenseTensor allocations.

)DOC")

SigureMo

LGTMeow

[SOT][FasterGuard] add IsDenseTensorHoldAllocationMatchGuard

49a1b82

gouzil requested a review from Copilot May 20, 2025 14:10

paddle-bot bot added the contributor External developers label May 20, 2025

Copilot AI reviewed May 20, 2025

View reviewed changes

gouzil and others added 2 commits May 20, 2025 22:46

Update paddle/fluid/pybind/sot/guards.cc

093067b

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

fix res

b343e0c

gouzil requested a review from Copilot May 20, 2025 15:15

Copilot AI reviewed May 20, 2025

View reviewed changes

add test

6d9c0fb

gouzil requested review from SigureMo and zrr1999 as code owners May 21, 2025 08:36

SigureMo approved these changes May 21, 2025

View reviewed changes

gouzil changed the title ~~[WIP][SOT][FasterGuard] add IsDenseTensorHoldAllocationMatchGuard~~ [SOT][FasterGuard] add IsDenseTensorHoldAllocationMatchGuard May 21, 2025

SigureMo merged commit 05a7baa into PaddlePaddle:develop May 22, 2025
55 of 56 checks passed

SigureMo deleted the sot/FasterGuard/add_IsDenseTensorHoldAllocationMatchGuard branch May 22, 2025 01:12

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[SOT][FasterGuard] add `IsDenseTensorHoldAllocationMatchGuard` #72823

[SOT][FasterGuard] add `IsDenseTensorHoldAllocationMatchGuard` #72823

Uh oh!

gouzil commented May 20, 2025 •

edited

Loading

Uh oh!

paddle-bot bot commented May 20, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Copilot AI May 20, 2025

Uh oh!

Copilot AI May 20, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI May 20, 2025

Uh oh!

SigureMo left a comment

Uh oh!

Uh oh!

Uh oh!

-      R"DOC(IsDenseTensorHoldAllocationMatchGuard Class.)DOC")
+      R"DOC(
+          IsDenseTensorHoldAllocationMatchGuard Class.
+          This guard checks whether a DenseTensor instance holds a valid allocation.
+          It is used to ensure that the memory allocation associated with a DenseTensor
+          is consistent and valid during runtime. This is particularly useful in scenarios
+          where memory integrity and allocation state need to be verified.
+          Example:
+              guard = IsDenseTensorHoldAllocationMatchGuard()
+              # Use the guard in conjunction with other runtime checks.
+      )DOC")

-      R"DOC(IsDenseTensorHoldAllocationMatchGuard Class.)DOC")
+      R"DOC(
+IsDenseTensorHoldAllocationMatchGuard Class.
+This guard is used to check whether a DenseTensor object holds a specific
+memory allocation. It is primarily used in scenarios where the memory
+allocation of a DenseTensor needs to be validated for correctness or
+consistency during runtime.
+Use this guard when you need to ensure that a DenseTensor's memory
+allocation matches expected criteria, such as during debugging or
+when implementing custom tensor operations.
+Example:
+    guard = IsDenseTensorHoldAllocationMatchGuard()
+    # Use the guard in your logic to validate DenseTensor allocations.
+)DOC")

[SOT][FasterGuard] add IsDenseTensorHoldAllocationMatchGuard #72823

[SOT][FasterGuard] add IsDenseTensorHoldAllocationMatchGuard #72823

Uh oh!

Conversation

gouzil commented May 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Category

PR Types

Description

Uh oh!

paddle-bot bot commented May 20, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

Copilot AI May 20, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI May 20, 2025

Choose a reason for hiding this comment

Uh oh!

SigureMo left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

[SOT][FasterGuard] add `IsDenseTensorHoldAllocationMatchGuard` #72823

[SOT][FasterGuard] add `IsDenseTensorHoldAllocationMatchGuard` #72823

gouzil commented May 20, 2025 •

edited

Loading