Add sparse conv3d kernel #39879
Merged
Add sparse conv3d kernel #39879
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
added 13 commits
February 17, 2022 09:24
|
Thanks for your contribution! |
| @@ -40,6 +40,19 @@ inline const DDim InferDenseDims(const DDim& x_dims, | |||
| return values_dims; | |||
| } | |||
|
|
|||
| template <typename Context> | |||
| inline void GetGpuLaunchConfig1D(const Context& dev_ctx, | |||
There was a problem hiding this comment.
好的，这个函数之前还没迁移过来临时写的，后面PR统一改掉。
| DenseTensor* rulebook) { | ||
| // update padding and dilation | ||
| // Currently, only support x.layout is NDHWC, groups = 1 | ||
| // if x.layout != NDHWC then transpose(x), transpose(weight) |
There was a problem hiding this comment.
如果layout是channel first，是用户需要自己去transpose？
There was a problem hiding this comment.
目前transpose的kernel还没迁移过来，所以准备先在前端借口处理好layout再传下来。
xingfeng01
reviewed
Feb 25, 2022
| HOSTDEVICE const int& operator[](int i) const { return dims[i]; } | ||
| }; | ||
|
|
||
| inline HOSTDEVICE bool Check(const int& x, |
| return (lower >= 0 && lower % stride == 0 && uper < xdim); | ||
| } | ||
|
|
||
| inline HOSTDEVICE bool Check(const Dims4D& dims, |
chenwhql
reviewed
Feb 25, 2022
| DDim* out_dims) { | ||
| PADDLE_ENFORCE_EQ(x_dims.size(), | ||
| 5, | ||
| paddle::platform::errors::InvalidArgument( |
There was a problem hiding this comment.
paddle::platform::的前缀建议去掉，phi下面有errors的实现
| "the shape of x should be (N, D, H, W, C)")); | ||
| PADDLE_ENFORCE_EQ(kernel_dims.size(), | ||
| 5, | ||
| paddle::platform::errors::InvalidArgument( |
| } // namespace phi | ||
|
|
||
| PD_REGISTER_KERNEL( | ||
| conv, CPU, ALL_LAYOUT, phi::sparse::Conv3dKernel, float, double) {} |
There was a problem hiding this comment.
这里是不是应该是sparse_conv？如果这里占了kernelname，原先的conv不能迁移了
There was a problem hiding this comment.
或者这个kenrel是不是只能给SPARSE的layout用
There was a problem hiding this comment.
PD_REGISTER_KERNEL(
conv, CPU, SPARSE_COO, phi::sparse::Conv3dKernel, float, double) {
kernel->InputAt(1).SetLayout(phi::DataLayout::NCHW);
}
There was a problem hiding this comment.
如沟通，当前先使用ALL_LAYOUT，等cmake中获取layout后在改成SPARSE_COO：
PD_REGISTER_KERNEL(
sparse_conv3d, CPU, ALL_LAYOUT, phi::sparse::Conv3dKernel, float, double) {
kernel->InputAt(0).SetDataLayout(phi::DataLayout::SPARSE_COO);
}
added 4 commits
February 25, 2022 12:07
xingfeng01
approved these changes
Feb 28, 2022
AnnaTrainingG
approved these changes
Feb 28, 2022
chenwhql
approved these changes
Feb 28, 2022
This was referenced Mar 8, 2022
This was referenced Mar 10, 2022
Merged
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
OPs
Describe
本PR是实现phi::sparse::Conv3dKernel 前向代码的CPU串行实现，GPU代码在PR
sparse conv3d采Second算法，算法大体思路：
因为输入数据是稀疏数据，存储在SparseCooTensor中，即只有非0元素列表，其中每个channel数据作为一个基础非0元素，这种情况下很难取一个窗口（比如3x3）的数据，因此该算法通过构造一个输入输出映射表rulebook，然后根据rulebook进行gather、gemm和scatter
构造rulebook：
gather：对kernel每一个位置，根据rulebook中记录的输入数据索引，把对应输入数据收集到一个临时tensor tmp_in(n, in_channels)，n表示当前这个位置参与的输入数据个数
gemm：将tmp_in和当前这个位置的kernel数据tmp_kernel(in_channels, out_channels)进行矩阵乘法，得到tmp_out(n, out_channels)
scatter：将所有的tmp_out进行scatter add，即把同一个输出位置的数据进行累加得到最终输出结果。