为median添加内核

[cszdrg]

PR Category

Operator Mechanism

PR Types

Bug fixes

Description

pr目的：为median nanmedian反向添加均分

返回索引的情况：

1. 如果索引是一个值且mode为min 则返回中位数索引
2. mode为avg 或者 如果索引是列表、None 则不返回中位数索引

均分的情况：

1. 如果返回索引 不做均分
2. 如果不返回索引 进行均分

---

设计思路

1.仿照nanmeidan为median添加kernel
2. cpu反向添加均分：遍历得到相同值 进行存储 在进行均分 如果是avg 会找不到相同值 则单独处理 赋予梯度
2. Gpu反向添加均分：如果是avg 将选取的两个值都改为中位数的值 再使用ReduceCudaAMaxAMinGrad计算均分
由于增加均分 性能有所下降

---

代码改动

1. 以nanmedian代码为原型 为median增加内核
2. 内核中新增ignore参数 区分nan的处理
3. 内核中新增evenly参数 根据mode和索引值 决定是否进行均分 如果均分 走新代码 否则 走原有代码
4. 反向ymal和函数中增加data 回传中位数值 进行计算
5. 为ReduceCudaAMaxAMinGrad新增参数 来处理为nan进行分配梯度的操作

---

测试结果

PAddleAPITest下所有测试通过

在大tensor下存在索引不同的情况，相关误差来自于topk的索引误差

均分情况

---

另：paddle的topk与torch在nan上的逻辑并未对齐 torch会将nan作为最大值参与排序 但是paddle会出现nan和inf等莫名错误，后续看是否有修复计划

[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 84.93976% with 50 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@4d37570). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
paddle/phi/kernels/cpu/nanmedian_kernel.cc	2.32%	42 Missing ⚠️
paddle/phi/kernels/cpu/median_kernel.cc	97.27%	4 Missing ⚠️
paddle/phi/kernels/cpu/median_grad_kernel.cc	95.23%	3 Missing ⚠️
paddle/phi/kernels/cpu/nanmedian_grad_kernel.cc	96.66%	1 Missing ⚠️

❌ Your patch status has failed because the patch coverage (84.93%) 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   #74767   +/-   ##
==========================================
  Coverage           ?   84.98%           
==========================================
  Files              ?        7           
  Lines              ?      333           
  Branches           ?        0           
==========================================
  Hits               ?      283           
  Misses             ?       50           
  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.