[PHI] non-contiguous paddle.nansum accuracy matching

[Enigmatisms]

PR Category

Operator Mechanism

PR Types

Improvements

Description

将组合算子 paddle.nansum 改造为 PHI 算子，以对 PyTorch 做精度对齐。组合算子的 paddle.where 会强制把不连续 tensor 转为连续，导致累加顺序变化 (@baoqiwen)。复用 phi::Reduce 操作以及手动实现的 NanSumOps 可修正此问题。

本 PR 需要考虑 non-contiguous layout，所以在 strided ops 里增加了 nansum。

大 Tensor 精度对齐测试：

Torch 对齐精度测试：

是否引起精度变化

是

Pcard-89620

[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.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 78.00000% with 11 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@cfd0574). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
.../interface/infer_symbolic_shape/unary_infer_sym.cc	0.00%	11 Missing ⚠️

❌ Your patch status has failed because the patch coverage (78.00%) 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   #78219   +/-   ##
==========================================
  Coverage           ?   78.00%           
==========================================
  Files              ?        4           
  Lines              ?       50           
  Branches           ?        0           
==========================================
  Hits               ?       39           
  Misses             ?       11           
  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.

[Enigmatisms]

unary_infer_sym.cc 的 coverage 属于误报，本地实测增加log：

运行 legacy_test/test_nansum_phi_func.py，其中 TestNansumStaticGraph 可以测到（两个log都能打印）：

Coverage 应该能跑到 test_nansum_phi_func，但不知为何覆盖率记录错误。

[wanghuancoder]

LGTM

[Enigmatisms]

本部分的 symbolic shape infer 单测 Coverage 未覆盖问题将 Open 一个新 PR 进行 double check。本地虽然能测过，但 Coverage 则不明原因未覆盖。

[From00]

LGTM

[lugimzzz]

lgtm

[Enigmatisms]

PR #78297 测试了CI反馈未覆盖代码如果增加必定报错的写法（evaluate 为 false 的 paddle_enforce）。发现：Coverage 实际报错了。说明实际能跑到对应的单测，但是 coverage 统计对应测试的覆盖率。