Skip to content

fix cuda kernel test for windows on cuda13#75684

Closed
youge325 wants to merge 2 commits intoPaddlePaddle:developfrom
youge325:testKernel
Closed

fix cuda kernel test for windows on cuda13#75684
youge325 wants to merge 2 commits intoPaddlePaddle:developfrom
youge325:testKernel

Conversation

@youge325
Copy link
Copy Markdown
Contributor

@youge325 youge325 commented Oct 8, 2025

PR Category

Environment Adaptation

PR Types

Bug fixes

Description

修改内容

  • isfinite_kernel.h:新增 INSTANTIATE_ISFINITE_KERNEL_Isfinite 宏,补全 Windows 显式模板实例化。
  • isfinite_kernel.cc 与 isfinite_kernel.cu:为 CPU/GPU 上的 float、double、整型及半精度类型补上 Isfinite 实例化。
  • enforce_test.cc:对 CUFFT_INCOMPLETE_PARAMETER_LISTCUFFT_PARSE_ERRORCUFFT_LICENSE_ERROR#ifdef 防护。
  • test_tensorrt_engine_instruction.cc:改用 Shuffle + MatrixMultiply + ElementWise 来构建等价的全连接层,并根据 TensorRT 版本选择 getNbIOTensorsgetNbBindings

覆盖情况

  • 修复 tensor_util_test 中 phi::IsfiniteKernel 未解析符号。
  • 修复 enforce_test 中缺失 CUFFT 枚举导致的编译错误。
  • 修复 TensorRT 单测对已移除 API 的调用。

@paddle-bot
Copy link
Copy Markdown

paddle-bot Bot commented Oct 8, 2025

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.

@paddle-bot paddle-bot Bot added the contributor External developers label Oct 8, 2025
@luotao1 luotao1 added the HappyOpenSource 快乐开源活动issue与PR label Oct 9, 2025
@luotao1 luotao1 assigned luotao1 and YqGe585 and unassigned YqGe585 Oct 9, 2025
@youge325
Copy link
Copy Markdown
Contributor Author

youge325 commented Oct 9, 2025

/re-run all-failed

4 similar comments
@youge325
Copy link
Copy Markdown
Contributor Author

youge325 commented Oct 9, 2025

/re-run all-failed

@youge325
Copy link
Copy Markdown
Contributor Author

youge325 commented Oct 9, 2025

/re-run all-failed

@youge325
Copy link
Copy Markdown
Contributor Author

youge325 commented Oct 9, 2025

/re-run all-failed

@youge325
Copy link
Copy Markdown
Contributor Author

youge325 commented Oct 9, 2025

/re-run all-failed

swgu98
swgu98 previously approved these changes Oct 11, 2025
@youge325
Copy link
Copy Markdown
Contributor Author

/re-run all-failed

1 similar comment
@youge325
Copy link
Copy Markdown
Contributor Author

/re-run all-failed

@luotao1
Copy link
Copy Markdown
Contributor

luotao1 commented Oct 13, 2025

  1. 建议分3个PR:isfinite_kernel、enforce_test、test_tensorrt_engine_instruction,便于不同方向的研发review和快速合入。
  2. 分开后的PR描述里,建议贴一下支持cuda13前后的before和after

@youge325
Copy link
Copy Markdown
Contributor Author

  1. 建议分3个PR:isfinite_kernel、enforce_test、test_tensorrt_engine_instruction,便于不同方向的研发review和快速合入。
  2. 分开后的PR描述里,建议贴一下支持cuda13前后的before和after

好的

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

contributor External developers HappyOpenSource 快乐开源活动issue与PR

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants