[geometric]Add paddle.geometric.send_uv API #44848
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
你的PR提交成功，感谢你对开源项目的贡献! |
DesmonDay
changed the title
5 tasks
wawltor
reviewed
Aug 14, 2022
| @@ -0,0 +1,134 @@ | |||
| # Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved. | |||
There was a problem hiding this comment.
比较好奇这里的文件名为什么不在send_recv里面
| Args: | ||
| x (Tensor): The source node feature tensor, and the available data type is float32, float64, int32, int64. And we support float16 in gpu version. | ||
| y (Tensor): The destination node feature tensor, and the available data type is float32, float64, int32, int64. And we support float16 in gpu version. | ||
| src_index (Tensor): An 1-D tensor, and the available data type is int32, int64. |
| from paddle import _C_ops | ||
|
|
||
| from .utils import reshape_lhs_rhs | ||
|
|
| @@ -14,3 +14,4 @@ | |||
|
|
|||
| from .send_recv import send_u_recv # noqa: F401 | |||
| from .send_recv import send_ue_recv # noqa: F401 | |||
| from .send import send_uv # noqa: F401 | |||
wawltor
reviewed
Aug 14, 2022
| } | ||
| ctx.template Alloc<T>(out); | ||
| T* out_data = out->data<T>(); | ||
| if (index_size == 0) return; |
There was a problem hiding this comment.
这里index_size为0，直接return 是否合适了，要么直接报错？直接return 看起来是一个未定义的行为
| memset(x_grad_data, 0, memset_bytes_x); | ||
| memset(y_grad_data, 0, memset_bytes_y); | ||
|
|
||
| if (index_size == 0) return; |
wawltor
reviewed
Aug 14, 2022
| T* x_grad_off = x_grad + dst * slice_size; | ||
| const T* out_grad_off = out_grad + i * slice_size; | ||
| for (int64_t j = 0; j < slice_size; j++) { | ||
| if (out_grad_off[j] != 0) { |
There was a problem hiding this comment.
这里有个讨论，这里是不是也可以直接加上out_grad_off，不管梯度是否为0；因为在多线程计算里面，看起来直接加法会比有分支更好
There was a problem hiding this comment.
主要是因为这里会涉及到一个原子操作，所以加上分支判断可以减少原子操作。
| auto out_grad_dims_1 = phi::vectorize<int>(out_grad_dims); | ||
| std::vector<int> out_grad_dims_2(out_grad_dims_1.begin() + 1, | ||
| out_grad_dims_1.end()); | ||
| out_grad_dims_2.insert(out_grad_dims_2.begin(), x_grad_dims[0]); |
There was a problem hiding this comment.
虽然这里的计算开销相对较小，不过建议是使用emplace
wawltor
reviewed
Aug 14, 2022
| } | ||
| ctx.template Alloc<T>(out); | ||
| T* out_data = out->data<T>(); | ||
| if (index_size == 0) return; |
| cudaMemset(y_grad_data, 0, memset_bytes_y); | ||
| #endif | ||
|
|
||
| if (index_size == 0) return; |
wawltor
reviewed
Aug 14, 2022
| bcast_info.use_bcast, | ||
| mul_functor, | ||
| sum_functor); | ||
| } else { |
There was a problem hiding this comment.
这里有个疑问，为什么MUL ADD两种聚合的方式的代码为什么不写在一起，在有些分支上可以ADD MUL来区分
There was a problem hiding this comment.
这么写逻辑更加清晰，而且看代码只有在 reduce_sum的地方有点重复，其他部分是不太一样的。
DesmonDay
added a commit
to DesmonDay/Paddle
that referenced
this pull request
Oct 13, 2022
* initial commit * fix op maker bug * fix mul grad bug * add unittest * fix add grad bug, add cpu kernel * add paddle.geometric.message_passing * add paddle.geometric.send_uv api, add unittest * add fp16 judgement * fix file typo, move compute_type to message_op * add impl file * fix unittest timeout time * add review revise
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
New features
PR changes
APIs
Describe
Add paddle.geometric.send_uv API, mainly used in graph learning domain.