-
Notifications
You must be signed in to change notification settings - Fork 5.9k
【Hackathon 9th Sprint No.06】NO.06 paddle.nn.MaxPool2D兼容性增强 #76457
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
【Hackathon 9th Sprint No.06】NO.06 paddle.nn.MaxPool2D兼容性增强 #76457
Conversation
|
你的PR提交成功,感谢你对开源项目的贡献! |
|
@zhwesky2010 老师,根据https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/phi/kernels/xpu/pool_kernel.cc#L106,要想 XPU 支持 dilation 参数,需要底层 xpudnn::max_pool2d 支持 dilation 参数,这个是不是不属于 Paddle 仓库?xpudnn::max_pool2d 支持 dilation 参数吗? |
这个是xpu的sdk,不是在paddle的仓库。目前xpu的sdk都不支持dilation吗 |
| kernel_size: Size2 | ||
| stride: Size2 | None | ||
| padding: _PaddingSizeMode | Size2 | Size4 | ||
| dilation: Size2 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
单测没有加,API单测、OP单测需要好好测试一下前方向。
动态图、PIR、动转静CINN模式 都需要测。
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
好的好的
| const std::vector<int64_t>& ksize, | ||
| const std::vector<int64_t>& strides, | ||
| const std::vector<int64_t>& paddings, | ||
| const std::vector<int64_t>& dilations, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
infermeta、infersy_mbolic 没有添加。单测应该是过不了的。
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
研发大大您好,新增的 dilations 不影响 shape,infermeta、infer_symbolic 用之前就可以了
xpu那边我不太了解,但是除此之外还有很多不支持dilation,比如gpudnn |
|
@zhwesky2010 研发大大,我感觉在 pool_2d 上加参数,兼容性会出问题,打算换一种思路,参照 max_pool2d_with_index 的实现思路,新增一个kernel:max_pool2d_with_dilation,可以吗? |

PR Category
Operator Mechanism
PR Types
New features
Description
paddle.nn.MaxPool2D兼容性增强