[XPU][PHI Kernels] add scatter_nd_add_grad kernel & bf16 support for slice OPs #58580
Conversation
This file contains hidden or 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
lj970926
commented
Nov 1, 2023
| int loop_time = static_cast<int>( | ||
| index_dims_size == 0 ? 1 | ||
| : phi::product(phi::slice_ddim( | ||
| index.dims(), 0, index_dims_size - 1))); |
There was a problem hiding this comment.
index tensor最后一维长度为0时，需要按前面所有维度索引updates数组并累加到output中，而不是只累加第一维
lj970926
commented
Nov 1, 2023
| self.index_np = np.array([[[], []], [[], []]]).astype("int32") | ||
| self.updates_np = np.random.random((2, 2, 10, 10)).astype( | ||
| self.dtype | ||
| ) |
There was a problem hiding this comment.
针对scatter_nd_add前向bug的单测修改
RuohengMa
reviewed
Nov 2, 2023
Comment on lines
+66
to
+75
| bool is_out_range = true; | ||
| if (std::isinf(value) || std::isnan(value)) { | ||
| is_out_range = false; | ||
| } | ||
| if ((common_type_value >= | ||
| static_cast<CommonType>(std::numeric_limits<T>::lowest())) && | ||
| (common_type_value <= | ||
| static_cast<CommonType>(std::numeric_limits<T>::max()))) { | ||
| is_out_range = false; | ||
| } |
There was a problem hiding this comment.
这里感觉要修改一下，common_type_value满足条件之后，就算value里有inf和nan，is_out_range也是false
There was a problem hiding this comment.
这个正常行为，用户可以paddle.full_like一个值为Nan的Tensor，这里的逻辑是如果传入的值在数据类型能表示的范围或者是Nan/inf都是合理的。这个参考了GPU实现
RuohengMa
reviewed
Nov 2, 2023
|
|
||
| PD_REGISTER_KERNEL( | ||
| nonzero, XPU, ALL_LAYOUT, phi::NonZeroKernel, int, bool, float) { | ||
| nonzero, XPU, ALL_LAYOUT, phi::NonZeroKernel, int, bool, float, int64_t) { |
There was a problem hiding this comment.
nonzero在Op list里的名字是where_index，这个已经加了
RuohengMa
reviewed
Nov 2, 2023
|
|
||
| PD_REGISTER_KERNEL(prod, XPU, ALL_LAYOUT, phi::ProdKernel, float) {} | ||
| PD_REGISTER_KERNEL( | ||
| prod, XPU, ALL_LAYOUT, phi::ProdKernel, float, int, int64_t) {} |
There was a problem hiding this comment.
prod在Op list里的名字是reduce_prod，这个已经加了
QingshuChen
approved these changes
Nov 2, 2023
zeroRains
pushed a commit
to zeroRains/Paddle
that referenced
this pull request
Nov 8, 2023
…slice OPs (PaddlePaddle#58580) * bevformer and bf16 support * refine format * refine format * refine format * refine format * fix bugs in compilation
danleifeng
pushed a commit
to danleifeng/Paddle
that referenced
this pull request
Nov 14, 2023
…slice OPs (PaddlePaddle#58580) * bevformer and bf16 support * refine format * refine format * refine format * refine format * fix bugs in compilation
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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
Description
index.numel() == 0