【Paddle Tensor 规范化第二期】pow support complex #71230
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! |
|
|
Comment on lines
+207
to
+211
| funcs::CudaPowFunctor<T> functor; | ||
| auto attrs = functor.GetAttrs(); | ||
| *(attrs[0].second) = factor.to<float>(); | ||
| ActivationGPUImpl<T, Context, funcs::CudaPowFunctor<T>>( | ||
| dev_ctx, x, out, functor); |
There was a problem hiding this comment.
这里可以特判一下factor为0和1情况，0的话直接构造 实部全1，虚部全0 的张量作为结果，1的话直接拷贝x即可
Comment on lines
251
to
271
| template <typename T, typename Context> | ||
| void PowGradKernel(const Context& dev_ctx, | ||
| const DenseTensor& x, | ||
| const DenseTensor& dout, | ||
| const Scalar& factor, | ||
| DenseTensor* dx) { | ||
| PADDLE_ENFORCE_NOT_NULL( | ||
| dx, errors::NotFound("The output DenseTensor dX can not be nullptr")); | ||
| dev_ctx.template Alloc<T>(dx); | ||
| auto dout_flatten = EigenVector<T>::Flatten( | ||
| GET_DATA_SAFELY(&dout, "Input", "Out@GRAD", "PowGrad")); | ||
| auto dx_flatten = EigenVector<T>::Flatten( | ||
| GET_DATA_SAFELY(dx, "Output", "X@GRAD", "PowGrad")); | ||
| auto x_flatten = | ||
| EigenVector<T>::Flatten(GET_DATA_SAFELY(&x, "Input", "X", "PowGrad")); | ||
| auto* place = dev_ctx.eigen_device(); | ||
| phi::funcs::PowGradFunctor<T> functor; | ||
| auto attrs = functor.GetAttrs(); | ||
| *(attrs[0].second) = factor.to<float>(); | ||
| functor(*place, x_flatten, nullptr, dout_flatten, dx_flatten); | ||
| } |
There was a problem hiding this comment.
同理，特判一下factor=0的情况，1反向稍微麻烦点，可以正常调functor计算
| + 1j * np.random.uniform(0.1, 1, [1, 3, 6]) | ||
| ).astype(np.complex64), | ||
| } | ||
| self.attrs = {"factor": 2.0} |
There was a problem hiding this comment.
factor建议多测几组，比如：[-3.4, -3, -2, -1, -0.1, 0, 0.7, 1, 1.3, 2, 3, 5]
Comment on lines
264
to
265
| PADDLE_ENFORCE_NOT_NULL( | ||
| dx, errors::NotFound("The output DenseTensor dX can not be nullptr")); |
There was a problem hiding this comment.
这里不应该用NotFound，
，而是InvalidArgument
| const Scalar& factor, | ||
| DenseTensor* out) { | ||
| PADDLE_ENFORCE_NOT_NULL(out, | ||
| errors::NotFound("Output Out should not be nullptr")); |
| return; | ||
| } | ||
| PADDLE_ENFORCE_NOT_NULL( | ||
| dx, errors::NotFound("The output DenseTensor dX can not be nullptr")); |
zyfncg
approved these changes
Mar 4, 2025
Enigmatisms
pushed a commit
to Enigmatisms/Paddle
that referenced
this pull request
Mar 6, 2025
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 Category
User Experience
PR Types
Bug fixes
Description
Pcard-75624
为pd_op.pow增加了复数支持与cuda实现，将pd_op.elementwise_powGPU上的实现由std改为cuda实现
通过nsys分析，powkernel执行时间从平均7.364μs降到6.986μs，elementwise_powkernel从平均9.631μs降到6.744μs