[Perf][CustomOp] Optimize custom operator dispatch overhead

[DongBaiYue]

PR Category

Performance Optimization

PR Types

Performance

Description

Paddle 自定义算子（Custom Operator）通过 _C_ops._run_custom_op 调用时存在显著的 CPU 调度开销。原始实现的热路径问题：

1. OpMetaInfo 查找：每次调用都需要字符串 hash 查找
2. 属性解析：运行时字符串解析 + if-else 链类型匹配
3. Output Tensor 创建：生成唯一 tensor 名称，涉及原子计数器和字符串分配

优化内容：

• Add ParsedOpMeta global cache to avoid repeated parsing at plugin load time
• Add MinimalEmptyTensor() to skip unnecessary name generation and grad node setup
• Use enum switch instead of string comparison for attribute parsing

性能对比：

基准测试（空 kernel，16x16 tensor，100 次迭代）：

操作	优化前	优化后	提升
纯调度（1in_0out）	3.10 us	2.27 us	-27%
+ 输出构建	6.04 us	3.49 us	-42%
+ 内存分配	8.74 us	6.02 us	-31%

真实场景：XPU BF16 FC 算子 (xpu_fc_bias_bf16, shape [8192, 1536] x [1536, 8192])：

阶段	优化前	优化后	提升
C++ dispatch	~75us	~35us	-53%
总计	~275us	~235us	-15%

兼容性：

• 注册机制不变：PD_BUILD_OP API 保持不变
• 调用方式不变：_C_ops._run_custom_op(op_name, ...) 完全兼容
• 插件无需修改：现有 paddle_xpu 等插件无需改动即可受益

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.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 66.66667% with 4 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@dff7725). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
paddle/fluid/framework/custom_operator.cc	63.63%	4 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   #78540   +/-   ##
==========================================
  Coverage           ?   66.66%           
==========================================
  Files              ?        2           
  Lines              ?       12           
  Branches           ?        0           
==========================================
  Hits               ?        8           
  Misses             ?        4           
  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.

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ DongBaiYue
❌ root

---

root seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}