[Cpp API Compatibility] Add TORCH_WARN macro and fix resize_

[youge325]

PR Category

Execute Infrastructure

PR Types

New features

Description

新增 TORCH_WARN 宏

修复编译 DeepEP 时发现的 resize_ 接口编译错误

/home/may/miniconda3/lib/python3.13/site-packages/paddle/include/paddle/phi/api/include/compat/ATen/ops/resize.h: In member function ‘const at::Tensor& at::Tensor::resize_(at::IntArrayRef, std::optional<c10::MemoryFormat>) const’:
/home/may/miniconda3/lib/python3.13/site-packages/paddle/include/paddle/phi/api/include/compat/ATen/ops/resize.h:80:60: error: ‘using std::__shared_ptr_access<phi::DenseTensor, __gnu_cxx::_S_atomic, false, false>::element_type = class phi::DenseTensor’ {aka ‘class phi::DenseTensor’} has no member named ‘offset’
   80 |           : dense_tensor->Holder()->size() - dense_tensor->offset();
      |                                                            ^~~~~~
/home/may/miniconda3/lib/python3.13/site-packages/paddle/include/paddle/phi/api/include/compat/ATen/ops/resize.h:90:43: error: ‘using std::__shared_ptr_access<phi::DenseTensor, __gnu_cxx::_S_atomic, false, false>::element_type = class phi::DenseTensor’ {aka ‘class phi::DenseTensor’} has no member named ‘offset’
   90 |   const size_t old_offset = dense_tensor->offset();
      |                                           ^~~~~~
/home/may/miniconda3/lib/python3.13/site-packages/paddle/include/paddle/phi/api/include/compat/ATen/ops/resize.h:99:34: error: ‘using std::__shared_ptr_access<phi::DenseTensor, __gnu_cxx::_S_atomic, false, false>::element_type = class phi::DenseTensor’ {aka ‘class phi::DenseTensor’} has no member named ‘mutable_data’
   99 |   void* new_data = dense_tensor->mutable_data(place, dense_tensor->dtype());
      |                                  ^~~~~~~~~~~~
/home/may/DeepEP/csrc/deep_ep.cpp compile failed, command '/usr/bin/g++' failed with exit code 1

是否引起精度变化

否

[paddle-bot]

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

This PR updates the PaddlePaddle C++ API compatibility layer to better match PyTorch behavior by adding TORCH_WARN-family macros and fixing at::Tensor::resize_ compilation issues when interacting with phi::DenseTensor internals.

Changes:

• Add a simplified warning API and TORCH_WARN / TORCH_WARN_DEPRECATION / TORCH_WARN_ONCE macros under c10/util/Exception.h.
• Fix resize_ implementation to use DenseTensorMeta::offset and DenseTensor::ResizeAndAllocate / data() APIs.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
paddle/phi/api/include/compat/c10/util/Exception.h	Introduces TORCH_WARN* macros and minimal warning plumbing for PyTorch compatibility.
paddle/phi/api/include/compat/ATen/ops/resize.h	Fixes resize_ to compile with current phi::DenseTensor by switching from removed APIs (offset(), mutable_data) to supported ones.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

C10_CONCATENATE(_IMPL) is redefined here, but it is already defined in c10/macros/Macros.h (and headers like c10/core/DispatchKeySet.h include both). This can trigger macro-redefinition diagnostics (often treated as errors under -Werror). Prefer including <c10/macros/Macros.h> and removing these duplicate macro definitions (or add #ifndef guards).

Suggested change

	// Helper macro for generating unique variable names
	#define C10_CONCATENATE_IMPL(s1, s2) s1##s2
	#define C10_CONCATENATE(s1, s2) C10_CONCATENATE_IMPL(s1, s2)
	// Helper macro for generating unique variable names
	#ifndef C10_CONCATENATE_IMPL
	#define C10_CONCATENATE_IMPL(s1, s2) s1##s2
	#endif
	#ifndef C10_CONCATENATE
	#define C10_CONCATENATE(s1, s2) C10_CONCATENATE_IMPL(s1, s2)
	#endif

[Copilot]

TORCH_WARN_ONCE expands C10_ANONYMOUS_VARIABLE(...), but this header doesn’t ensure that macro is defined (it’s provided by c10/macros/Macros.h). Using TORCH_WARN_ONCE from a TU that only includes <c10/util/Exception.h> will fail to compile. Include <c10/macros/Macros.h> here (preferred) or add a guarded fallback definition for C10_ANONYMOUS_VARIABLE.

[youge325]

/re-run all-failed