[Cpp API Compatibility] Add CUDABlas related APIs

[youge325]

PR Category

Execute Infrastructure

PR Types

New features

Description

新增 at::cuda::getCurrentCUDABlasHandle at::cuda::blas::gemm<T> 接口

新增 Allocator 结构体

文档详见 PFCCLab/PaddleCppAPITest#46

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 extends Paddle’s LibTorch/ATen compatibility layer by introducing a lightweight CUDA context interface and adding missing CUDA/CUBLAS handle accessors needed for C++ API compatibility (notably at::cuda::getCurrentCUDABlasHandle), along with a compat c10::Allocator abstraction.

Changes:

• Add c10::cuda::device_count() API to the compat CUDA functions header.
• Introduce compat c10::Allocator (plus DataPtr::release_context() support) for raw allocation APIs.
• Add ATen/cuda/CUDAContextLight.{h,cpp} and switch ATen/cuda/CUDAContext.h to include the new light header; wire the new .cpp into the compat build.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
paddle/phi/api/include/compat/c10/cuda/CUDAFunctions.h	Adds device_count() to the compat c10 CUDA API surface.
paddle/phi/api/include/compat/c10/core/Allocator.h	Adds a compat c10::Allocator interface and DataPtr::release_context().
paddle/phi/api/include/compat/CMakeLists.txt	Adds the new CUDAContextLight.cpp to the compat compilation sources.
paddle/phi/api/include/compat/ATen/cuda/CUDAContextLight.h	Declares lightweight at::cuda CUDA context APIs, including cuBLAS handle getters.
paddle/phi/api/include/compat/ATen/cuda/CUDAContextLight.cpp	Implements the lightweight CUDA context APIs via phi::GPUContext.
paddle/phi/api/include/compat/ATen/cuda/CUDAContext.h	Redirects CUDAContext to the new lightweight header.

---

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

[codecov-commenter]

Codecov Report

❌ Patch coverage is 95.18072% with 24 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@362b943). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
.../api/include/compat/ATen/cuda/CUDAContextLight.cpp	89.74%	8 Missing ⚠️
paddle/phi/api/include/compat/c10/core/Storage.h	93.79%	8 Missing ⚠️
...ddle/phi/api/include/compat/ATen/core/TensorBase.h	92.72%	4 Missing ⚠️
...addle/phi/api/include/compat/c10/core/DeviceType.h	90.90%	2 Missing ⚠️
...ddle/phi/api/include/compat/ATen/core/TensorBody.h	0.00%	1 Missing ⚠️
paddle/phi/api/include/compat/c10/core/Allocator.h	97.61%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop   #78060   +/-   ##
==========================================
  Coverage           ?   95.18%           
==========================================
  Files              ?       16           
  Lines              ?      498           
  Branches           ?        0           
==========================================
  Hits               ?      474           
  Misses             ?       24           
  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.

[youge325]

/re-run windows-gpu

[SigureMo]

咦，这行改动的原因是？没必要是么？

[youge325]

这行不去掉的话，在编译的时候 Linux-DCU build会遇到redefinition错误，和 #include <cublas_v2.h> 冲突了，本来是要换一下 include 顺序，但是cpplint会直接换回来，后来发现直接去掉也没问题，这里暂时是用不到 cublasXt.h 的，后面需要用到的时候再 include 也没问题，具体可以看涉及到 fix dcu 的 commit

[SigureMo]

这轮正式结论调整为 APPROVE。

太不容易了啊这个 PR

[youge325]

/re-run all-failed

[youge325]

/re-run all-failed

[SigureMo]

LGTMeow