Skip to content

Conversation

@chenwhql
Copy link
Contributor

@chenwhql chenwhql commented Dec 8, 2021

PR types

Function optimization

PR changes

Others

Describe

[PTen] Refine Kernel Registrar Writing

  • 经过调研与讨论,我们内部kernel的overload情况很多,想要用简单的枚举规范化比较困难,因此overload_name基本属于过度设计了,后续会简化这里的设计,需要overload name的场景仍然直接使用下划连接相应字符串后缀实现,具体的命名根据实际情况决定,表意准确即可
  • 此外,.overload_name机制去除的话,kernel注册写法改回和原fluid注册一致的写法,即直接使用名字,而不是字符串,在编译自动化上更易实现,因此进行了全局修改
@paddle-bot-old
Copy link

paddle-bot-old bot commented Dec 8, 2021

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

* of the kernel, this registration macro will not do automatic template
* instantiation.
*/
#define PT_REGISTER_SINGLE_KERNEL( \
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PT_REGISTER_SINGLE_KERNEL是不是没有使用?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

是的,但是我们需要保留这样一个基础接口,这个是使用一个函数指针进行标准注册的情况,在无法使用PT_REGISTER_KERNEL注册的时候,即使重复写很多遍,也要能注册一个kernel

PT_REGISTER_MODULE(ManipulationCPU);

// TODO(yuanrisheng): "flatten_contiguous_range" is compatible with old kernel
// architecture, kernel_name should be "flatten".
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个todo可以删掉了

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

thx,下一个PR移除

@chenwhql chenwhql merged commit b199ba8 into PaddlePaddle:develop Dec 9, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

4 participants