[API Compatibility No.6] Sink bitwise_xor_ to cpp -part by Manfredss · Pull Request #76341 · PaddlePaddle/Paddle

Manfredss · 2025-11-11T02:56:42Z

PR Category

User Experience

PR Types

New Features

Description

任务：No. 6 bitwise_xor_ , No.42 bitwise_xor

修改内容

下沉 bitwise_xor_ 到 cpp
支持参数别名:
- input 作为 x 的别名
- other 作为 y 的别名
在 test/legacy_test/test_inplace.py 中添加参数别名测试用例

paddle-bot · 2025-11-11T02:56:48Z

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.

build/python/paddle/tensor/logic.py

Manfredss · 2025-11-12T08:04:31Z

/re-run all-failed

Manfredss · 2025-11-12T20:51:31Z

/re-run all-failed

codecov-commenter · 2025-11-13T14:10:32Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@4df7c64). Learn more about missing BASE report.

Additional details and impacted files

@@             Coverage Diff             @@
##             develop    #76341   +/-   ##
===========================================
  Coverage           ?   100.00%           
===========================================
  Files              ?         2           
  Lines              ?         3           
  Branches           ?         0           
===========================================
  Hits               ?         3           
  Misses             ?         0           
  Partials           ?         0

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

Manfredss · 2025-11-14T00:00:28Z

@luotao1 @zhwesky2010

Hi, 我注意到有两个 CI 失败：

CI / Distribute-stable-formers / formers-Test
CI-Build / Slice / Slice test

经过分析，这两个失败应该与本 PR 的改动无关，原因如下：

本 PR 仅修改了 bitwise_xor_ API 的参数别名支持，不涉及：
- 分布式训练相关功能
- 张量切片相关功能
- 任何模型训练逻辑
改动范围极小且独立：只在 logic.py 添加了装饰器，以及对应的单元测试
从其他类似的 API 兼容性 PR（如 bitwise_or_）来看，应该不会影响这些无关的测试

luotao1 · 2025-11-14T03:22:21Z

CI / Distribute-stable-formers / formers-Test

非required流水线不用管，目前还在调试中

CI-Build / Slice / Slice test

等研发review过后，我会进行豁免

python/paddle/tensor/logic.py

Manfredss · 2025-11-17T04:50:41Z

/re-run all-failed

zhwesky2010 · 2025-11-17T09:17:28Z

paddle/phi/ops/yaml/python_api_info.yaml

+- op : bitwise_xor
+  name : [paddle.bitwise_xor, paddle.Tensor.bitwise_xor]
+  args_alias :
+    x : [input]


这个用use_default_mapping : True 就可以

zhwesky2010 · 2025-11-17T09:18:33Z

python/paddle/_paddle_docs.py


 # other
+add_doc_and_signature(
+    "bitwise_xor_",


这个是bitwise_xor还是bitwise_xor_？

前者是非inplace的，后者是inplace的

修复列表里写的是 bitwise_xor_，所以这里写的 bitwise_xor_。这个 API 只能通过装饰器来修改

但 bitwise_xor 是能够 cpp 下沉的，所以有些疑惑 @zhwesky2010

zhwesky2010 · 2025-11-17T09:18:40Z

python/paddle/tensor/logic.py

 if TYPE_CHECKING:
    from paddle import Tensor

+    # @overload


这些不需要

zhwesky2010 · 2025-11-17T09:19:06Z

python/paddle/tensor/logic.py

@@ -1226,7 +1245,9 @@ def bitwise_xor(



下沉后原来的bitwise_xor需要删除掉，从_C_ops中导入

Manfredss · 2025-11-20T07:40:34Z

我觉得不能删去 logic.py 中 bitwise_xor 的定义，因为测试代码 test_bitwise_op.py 中

self.python_api = paddle.tensor.logic.bitwise_xor

框架会使用 inspect.getfullargspec() 来获取函数签名信息，如果直接导入会报错

ValueError: no signature found for builtin <built-in function bitwise_xor>
TypeError: unsupported callable

bitwise_xor 需要支持参数别名 (input, other)，这要使用 @param_two_alias；要支持 out 参数，这需要 python 层的处理器；要支持静态图模式，这需要调用 _bitwise_op 辅助函数。以上

如果直接从 _C_op 导入，这些功能会无法实现。

最后需要确认一下，API 兼容性增强 No.6 针对的是 bitwise_xor 还是 bitwise_xor_？多谢解答！ @luotao1 @zhwesky2010

Manfredss · 2025-11-22T01:34:24Z

/re-run all-failed

zhwesky2010 · 2025-11-24T12:27:11Z

@Manfredss 以表格中的字符串为准。与表格一一对应就行。

zhwesky2010 · 2025-11-24T12:29:13Z

python/paddle/tensor/logic.py

        return _C_ops.bitwise_or_(x, y)


+@param_two_alias(["x", "input"], ["y", "other"])


写了下沉就不要写装饰器了。

两者选其一就可，另外下沉inplace API(bitwise_xor_)之前需要先把非inplace原版(bitwise_xor)也下沉了，因为两者是复用的一个kernel。

Manfredss · 2025-11-27T07:00:52Z

@zhwesky2010 已根据意见完成修改，烦请再 review 一下，谢谢！

zhwesky2010 · 2025-12-02T08:18:21Z

python/paddle/tensor/logic.py



 @inplace_apis_in_dygraph_only
+@ParamAliasDecorator({"x": ["input"], "y": ["other"]})


这个也下沉吧？

zhwesky2010 · 2025-12-02T08:18:51Z

paddle/phi/ops/yaml/python_api_info.yaml

  args_alias:
   use_default_mapping : True

+- op : bitwise_xor


下沉后原来的python 层API需要删掉，只有c++接口了

#76341 (comment) 主要是这个问题，所以我还留着

把两个一起下沉了就行了

zhwesky2010 · 2025-12-03T11:19:59Z

python/paddle/tensor/logic.py

 if TYPE_CHECKING:
    from paddle import Tensor

+


下沉后，需要移除之前的python api，就没有这一层逻辑了。直接走c++

zhwesky2010 · 2025-12-11T03:28:57Z

python/paddle/_paddle_docs.py

+    x: Tensor,
+    y: Tensor,
+    out: Tensor | None = None,
+    name: str | None = None,


签名统一参照其他的：

(x, y, name, *， out)

zhwesky2010 · 2025-12-11T03:30:02Z

test/legacy_test/test_zero_dim_binary_api.py

            x_np = np.random.randint(-10, 10, [])
            y_np = np.random.randint(-10, 10, [])
-            out_np = eval(f'np.{api.__name__}(x_np, y_np)')
+            out_np = eval(f'np.{api.__name__.lstrip("_")}(x_np, y_np)')


zero_dim这里为啥需要改

这里的改动是因为测试会有报错

====================================================================== ERROR: test_dygraph_binary (__main__.TestBinaryAPI.test_dygraph_binary) ---------------------------------------------------------------------- Traceback (most recent call last): File "D:\Xue\ML\PaddleDebug\test\legacy_test\test_zero_dim_binary_api.py", line 170, in test_dygraph_binary out_np = eval(f'np.{api.__name__}(x_np, y_np)') ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "<string>", line 1, in <module> File "C:\Users\tzy12\anaconda3\envs\paddle\Lib\site-packages\numpy\__init__.py", line 808, in __getattr__ raise AttributeError(f"module {__name__!r} has no attribute {attr!r}") AttributeError: module 'numpy' has no attribute '_bitwise_xor'. Did you mean: 'bitwise_xor'? ---------------------------------------------------------------------- Ran 6 tests in 0.508s

zhwesky2010 · 2025-12-11T03:30:40Z

python/paddle/tensor/logic.py

-    Inplace version of ``bitwise_xor`` API, the output Tensor will be inplaced with input ``x``.
-    Please refer to :ref:`api_paddle_bitwise_xor`.
-    """
-    out_shape = broadcast_shape(x.shape, y.shape)


下沉后，这个逻辑可能会丢掉了，需要测一下

zhwesky2010 · 2025-12-11T03:31:21Z

test/legacy_test/test_inplace.py

        return paddle.bitwise_xor(var, self.y)


+class TestDygraphInplacBitwiseXorAlias1(TestDygraphInplacBitwiseAnd):


paddle.bitwise_xor测一下compat情况下的用法

bitwise_xor_ 需要吗？好像 bitwise_xor_ 用了 compat 会有 segfault 的情况，我认为可能是 bitwise_xor_ 下沉了但是 ops.yaml 里没有 bitwise_xor_ 的定义。需要添加吗？

bitwise_xor_ 需要吗？好像 bitwise_xor_ 用了 compat 会有 segfault 的情况，我认为可能是 bitwise_xor_ 下沉了但是 ops.yaml 里没有 bitwise_xor_ 的定义。需要添加吗？

ops.yaml里不需要bitwise_xor_定义，会根据bitwise_xor的inplace字段，来生成bitwise_xor_

两者本质是共享一个kernel的，只在输出out处理这里有些区别。

bitwise_xor_ 需要吗？好像 bitwise_xor_ 用了 compat 会有 segfault 的情况，我认为可能是 bitwise_xor_ 下沉了但是 ops.yaml 里没有 bitwise_xor_ 的定义。需要添加吗？

先按标准来测试，不要绕过，先提前暴露问题。

如果inplace的API下沉不了就先用装饰器吧，底层问题先暴露出来。因为这也是第1个inplace的API下沉。

好像是 Paddle 的 eager execution engine 执行 inplace 函数的问题，第一个参数（即被修改的参数）必须作为位置参数传递，而不是 kwarg。因为 C++ 返回逻辑会专门在位置元组 args 中查找输入张量并将其作为结果返回。如果将其作为 kwarg 传递，则 args 为空，导致 segfault。

Manfredss · 2025-12-16T07:01:16Z

/re-run all-failed

zhwesky2010

看下CI问题

zhwesky2010 · 2026-01-13T09:27:29Z

#77294

这个PR修复了一些inplace api下沉的机制问题，bitwise_xor已经下沉了，这个PR下沉bitwise_xor_就可以

paddle-bot bot added the contributor External developers label Nov 11, 2025

luotao1 added the PaddlePaddle Hackathon label Nov 11, 2025

luotao1 assigned luotao1 and zhwesky2010 Nov 11, 2025

luotao1 mentioned this pull request Nov 11, 2025

【启航计划】PaddlePaddle API兼容性增强 #76301

Open

zhwesky2010 added the API Compatibility label Nov 11, 2025

zhwesky2010 reviewed Nov 11, 2025

View reviewed changes

build/python/paddle/tensor/logic.py Outdated Show resolved Hide resolved

zhwesky2010 changed the title ~~[API Compatibility No.6] Add parameter alias support for bitwise_xor_ -part~~ [API Compatibility No.6] Add parameter alias support for bitwise_xor_ Nov 11, 2025

Manfredss force-pushed the ApiEnhance6 branch from 1ab24a5 to ccab584 Compare November 11, 2025 21:59

luotao1 changed the title ~~[API Compatibility No.6] Add parameter alias support for bitwise_xor_~~ [API Compatibility No.6] Add parameter alias support for bitwise_xor_ -part Nov 13, 2025

Manfredss closed this Nov 13, 2025

Manfredss force-pushed the ApiEnhance6 branch from faa6f44 to 17d6201 Compare November 13, 2025 05:08

Manfredss reopened this Nov 13, 2025

zhwesky2010 reviewed Nov 14, 2025

View reviewed changes

python/paddle/tensor/logic.py Outdated Show resolved Hide resolved

zhwesky2010 reviewed Nov 17, 2025

View reviewed changes

Manfredss force-pushed the ApiEnhance6 branch 2 times, most recently from eb48a1f to 0f7b938 Compare November 21, 2025 06:13

zhwesky2010 reviewed Nov 24, 2025

View reviewed changes

zhwesky2010 reviewed Dec 2, 2025

View reviewed changes

Manfredss force-pushed the ApiEnhance6 branch 2 times, most recently from 7bad604 to d5240bf Compare December 3, 2025 06:54

zhwesky2010 reviewed Dec 3, 2025

View reviewed changes

Manfredss closed this Dec 4, 2025

Manfredss reopened this Dec 4, 2025

zhwesky2010 reviewed Dec 11, 2025

View reviewed changes

Manfredss changed the title ~~[API Compatibility No.6] Add parameter alias support for bitwise_xor_ -part~~ [API Compatibility No.6] Sink bitwise_xor and bitwise_xor_ to cpp -part Dec 16, 2025

zhwesky2010 reviewed Dec 19, 2025

View reviewed changes

luotao1 added HappyOpenSource 快乐开源活动issue与PR and removed PaddlePaddle Hackathon labels Dec 22, 2025

zhwesky2010 changed the title ~~[API Compatibility No.6] Sink bitwise_xor and bitwise_xor_ to cpp -part~~ [API Compatibility No.6] Sink bitwise_xor_ to cpp -part Jan 13, 2026

cpp sink of bitwise_xor_

084b5d8

Manfredss force-pushed the ApiEnhance6 branch from 9ef2b07 to 084b5d8 Compare January 26, 2026 03:26

zhwesky2010 approved these changes Jan 26, 2026

View reviewed changes

zhwesky2010 closed this Jan 30, 2026

		return _C_ops.bitwise_or_(x, y)


		@param_two_alias(["x", "input"], ["y", "other"])



		@inplace_apis_in_dygraph_only
		@ParamAliasDecorator({"x": ["input"], "y": ["other"]})

		return paddle.bitwise_xor(var, self.y)


		class TestDygraphInplacBitwiseXorAlias1(TestDygraphInplacBitwiseAnd):

Conversation

Manfredss commented Nov 11, 2025 • edited by zhwesky2010 Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Category

PR Types

Description

修改内容

Uh oh!

paddle-bot bot commented Nov 11, 2025

Uh oh!

Uh oh!

Manfredss commented Nov 12, 2025

Uh oh!

Manfredss commented Nov 12, 2025

Uh oh!

codecov-commenter commented Nov 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Manfredss commented Nov 14, 2025

Uh oh!

luotao1 commented Nov 14, 2025

Uh oh!

Uh oh!

Manfredss commented Nov 17, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manfredss Nov 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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

Uh oh!

Uh oh!

Manfredss commented Nov 22, 2025

Uh oh!

zhwesky2010 commented Nov 24, 2025

Uh oh!

zhwesky2010 Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manfredss commented Nov 27, 2025

Uh oh!

zhwesky2010 Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zhwesky2010 Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manfredss Dec 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manfredss Dec 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Manfredss commented Nov 11, 2025 •

edited by zhwesky2010

Loading

codecov-commenter commented Nov 13, 2025 •

edited

Loading

Manfredss Nov 17, 2025 •

edited

Loading

Manfredss commented Nov 20, 2025 •

edited

Loading

zhwesky2010 Nov 24, 2025 •

edited

Loading

zhwesky2010 Dec 2, 2025 •

edited

Loading

zhwesky2010 Dec 2, 2025 •

edited

Loading

Manfredss Dec 12, 2025 •

edited

Loading

Manfredss Dec 12, 2025 •

edited

Loading

zhwesky2010 Dec 17, 2025 •

edited

Loading