[DLPack][TVM FFI] Return PyCapsule in __c_dlpack_exchange_api__ in tvm ffi 0.1.4+

[SigureMo]

PR Category

Execute Infrastructure

PR Types

Improvements

Description

在 TVM FFI 0.1.4+ 或者说只要用户没有装 TVM FFI 0.1.0/0.1.1/0.1.2/0.1.3，__c_dlpack_exchange_api__ 就返回按照新的协议的指针包装后的 PyCapsule，以匹配最新协议（这里更多是实现而非原本协议中的内容，见 pytorch/pytorch#162845、pytorch/pytorch#165483）

虽然 TVM FFI 0.1.4 已经做了兼容性处理（apache/tvm-ffi#288），但我们不敢保证这一兼容性处理具体会持续多久，我们希望即将发布的 3.3.0 能够与下游生态有更松的耦合（FlashInfer、TileLang 等），确保这一版本尽可能长期可用，因此在 3.3.0 会添加这一兼容性逻辑

此外 CI 只监控 tvm-ffi 0.1.4 行为，0.1.3- 行为已经在 548666a 验证过

[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 adds compatibility for TVM FFI 0.1.4+ by conditionally returning either a raw pointer or a PyCapsule from the __c_dlpack_exchange_api__ interface. The change ensures backward compatibility with TVM FFI versions 0.1.0-0.1.3 while supporting the new protocol in 0.1.4+.

Key Changes:

• Introduced version detection logic to determine which DLPack exchange API format to return
• Added new C++ binding dlpack_exchange_api_pycapsule() that returns a PyCapsule
• Modified tensor patch method to call version-aware _get_c_dlpack_exchange_api() function instead of directly using the raw pointer

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
python/paddle/base/dygraph/tensor_patch_methods.py	Implements version detection logic and conditional return of pointer vs PyCapsule for DLPack exchange API
paddle/fluid/pybind/pybind.cc	Adds new Python binding function to return DLPack exchange API as a PyCapsule

---

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

[codecov-commenter]

Codecov Report

❌ Patch coverage is 66.66667% with 3 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@584c692). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
python/paddle/base/dygraph/tensor_patch_methods.py	66.66%	3 Missing ⚠️

❌ Your patch status has failed because the patch coverage (66.66%) is below the target coverage (90.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop   #76673   +/-   ##
==========================================
  Coverage           ?   66.66%           
==========================================
  Files              ?        1           
  Lines              ?        9           
  Branches           ?        0           
==========================================
  Hits               ?        6           
  Misses             ?        3           
  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.

[SigureMo]

tvm_ffi 0.1.0-0.1.3 正确性覆盖已经在 548666a 验证

[SigureMo]

如上所述，最终未覆盖部分在之前 commit 验证过，最终只监控最新的 tvm ffi 0.1.4，不监控 0.1.3，故豁免

[ooooo-create]

/skip-ci coverage