Skip to content
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

Kernel hint design #6790

Merged
merged 4 commits into from
Dec 21, 2017
Merged

Conversation

jacquesqiao
Copy link
Member

No description provided.

const std::string kUseCUDNN = "use_cudnn";
const std::string kUseMKLDNN = "use_mkldnn";

KernelType GetExpectedKernelTyp() {
Copy link
Collaborator

Choose a reason for hiding this comment

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

Typo

Copy link
Member Author

Choose a reason for hiding this comment

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

fixed

LayoutType layout_;
};
```
`place_` `data_type_` and `layout_` can come from the input tensor of the operator, `GetActualKernelType(inputs)` use inputs to infer the proper kernel key that fit the incoming data, user can not config it.
Copy link
Collaborator

Choose a reason for hiding this comment

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

Users cannot configure it.

Actually, users can configure input data and influence the actual kernel type.

Copy link
Member Author

Choose a reason for hiding this comment

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

done

## Solution

### potential choice
1, Do nothing, let the user add the information they want to operator‘s attribute and get them inside `GetExpectedKernelType`, this can work right. But there is a little problem that users may define many kinds of hints for the same purpose, such as `force_cpu`, `use_cpu`, `CPU` for CPU kernel, and `use_cudnn`, `force_cudnn`, `cudnn_kernel` for use of CUDNN kernel.
Copy link
Collaborator

Choose a reason for hiding this comment

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

1, --> 1.

Copy link
Member Author

Choose a reason for hiding this comment

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

done

## Solution

### potential choice
1, Do nothing, let the user add the information they want to operator‘s attribute and get them inside `GetExpectedKernelType`, this can work right. But there is a little problem that users may define many kinds of hints for the same purpose, such as `force_cpu`, `use_cpu`, `CPU` for CPU kernel, and `use_cudnn`, `force_cudnn`, `cudnn_kernel` for use of CUDNN kernel.
Copy link
Collaborator

Choose a reason for hiding this comment

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

We can define some global constant for these attribute name.

Copy link
Member Author

Choose a reason for hiding this comment

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

done


### final choice
To provide enough flexibility while avoiding confusion definition, we can predefine some options, such as `force_cpu`, `use_cudnn`, `use_mkldnn` for a user to choose.

Copy link
Collaborator

Choose a reason for hiding this comment

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

Using string to represent force_cpu, use_cudnn, use_mkldnn is not extendable. How to make an operator use CPU forcibly and use MKLDNN?

const std::string kUseMKLDNN = "use_mkldnn";

KernelType GetExpectedKernelTyp() {
// "kernel_hint" is a user defined attribute name
Copy link
Collaborator

Choose a reason for hiding this comment

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

I do not think kernel_hint is a formal name. It is confusing for users.

Copy link
Member Author

Choose a reason for hiding this comment

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

done, use force_cpu as key and bool as value.


KernelType GetExpectedKernelTyp() {
// "kernel_hint" is a user defined attribute name
if (Attr<std::string>("kernel_hint") == kForceCPU) {
Copy link
Collaborator

Choose a reason for hiding this comment

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

Why not use Attr<bool>(kForceCPU) ?

Copy link
Member Author

@jacquesqiao jacquesqiao Dec 20, 2017

Choose a reason for hiding this comment

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

good idea, done

reyoung
reyoung previously approved these changes Dec 20, 2017
Copy link
Collaborator

@reyoung reyoung left a comment

Choose a reason for hiding this comment

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

Basically LGTM.

const std::string kUseMKLDNN = "use_mkldnn";

KernelType GetExpectedKernelType() {
if (Attr<bool>(kForceCPU)) {
Copy link
Collaborator

Choose a reason for hiding this comment

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

Need a better Indent

Copy link
Member Author

Choose a reason for hiding this comment

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

change tab to space

FORCE_CPU = core.kForceCPU()

def xx_layer(..., force_cpu=false):
layer_helper = LayerHelper(...)
Copy link
Collaborator

Choose a reason for hiding this comment

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

need a better indent

Copy link
Member Author

Choose a reason for hiding this comment

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

change tab to space

Copy link
Collaborator

@reyoung reyoung left a comment

Choose a reason for hiding this comment

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

LGTM.

@jacquesqiao jacquesqiao merged commit f734d0c into PaddlePaddle:develop Dec 21, 2017
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants