Skip to content

【NewIR】add share_data_op #57212

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.

Sign up for GitHub

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

Merged
kangguangli merged 3 commits into from
Sep 15, 2023
Merged

【NewIR】add share_data_op #57212

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
71f930c
add share_data_op
yangguohao Sep 12, 2023
7819e37
fix share_data op
yangguohao Sep 12, 2023
bd4fcb8
Merge branch 'develop' into share_data
yangguohao Sep 14, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions paddle/fluid/pir/dialect/operator/ir/ops.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,6 +163,13 @@
args : (Tensor[] x, bool[] share_dims_and_dtype = {})
output : Tensor[](out){x.size()}, Tensor[](xout){x.size()}

- op : share_data
args : (Tensor x)
output : Tensor(out)
kernel:
func: share_data
param: [x]
Comment on lines +166 to +171
Copy link
Contributor

@zyfncg zyfncg Sep 15, 2023

Choose a reason for hiding this comment

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

share_data 和 share_buffer 的区别是什么？

Copy link
Contributor

@kangguangli kangguangli Sep 15, 2023

Choose a reason for hiding this comment

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

目前看有两个区别吧：

  1. share_data支持DenseTensor和SelectedRows，share_buffer只支持前者
  2. 内部实现有一定区别，这个区别是DenseTensor::ShareDataWith和DenseTensor::ShareBufferWith的区别，从注释上看，这两者的区别是
// NOTE: For historical reasons, this interface has a special behavior,
// sharing other tensor members except lod
DenseTensor& DenseTensor::ShareDataWith(const DenseTensor& src) {

Copy link
Contributor

@zyfncg zyfncg Sep 15, 2023

Choose a reason for hiding this comment

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

这里是不可以考虑下在新IR上能不能顺便把这个历史问题解了？上面这两个算子的差异确实不容易区分，后来的开发者如果不了解背景看到这样逻辑很容易一头雾水，搞不好又会成为历史包袱

Copy link
Contributor

@kangguangli kangguangli Sep 15, 2023

Choose a reason for hiding this comment

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

经讨论决定，保留share_data，删除share_buffer。从新IR下删除share_buffer的工作辛苦 @yangguohao 看下，后面单独提个PR完成下。

Copy link
Contributor Author

@yangguohao yangguohao Sep 15, 2023

Choose a reason for hiding this comment

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

经讨论决定，保留share_data，删除share_buffer。从新IR下删除share_buffer的工作辛苦 @yangguohao 看下，后面单独提个PR完成下。

好的 之后我提个 PR 处理

Copy link
Contributor

@kangguangli kangguangli Sep 18, 2023

Choose a reason for hiding this comment

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

关于share_buffer的删除，只要去掉 paddle/fluid/pir/dialect/operator/ir/ops.yaml 中的定义即可


- op : write_to_array
args : (Tensor i, Tensor x)
output : Tensor[](out)
Expand Down
3 changes: 2 additions & 1 deletion paddle/fluid/pir/dialect/operator/utils/utils.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,8 @@ const std::unordered_set<std::string> LegacyOpList = {
"pd_op.c_reduce_sum_",
"pd_op.c_allreduce_max_",
"pd_op.c_allgather",
"pd_op.seed"};
"pd_op.seed",
"pd_op.share_data"};

enum class AttrType {
UNDEFINED = 0,
Expand Down
6 changes: 6 additions & 0 deletions paddle/phi/api/yaml/op_compat.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3190,6 +3190,12 @@
data_type : int64_t
tensors_name : StepsTensorList

- op: share_data
inputs :
x : X
outputs :
out : Out

- op: sigmoid_cross_entropy_with_logits
backward: sigmoid_cross_entropy_with_logits_grad
inputs :
Expand Down