Skip to content

Check XLATensorImpl is always contiguous #256

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
merged 1 commit into from
Feb 14, 2019
Merged

Check XLATensorImpl is always contiguous #256

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 10 additions & 0 deletions test/cpp/test_aten_xla_tensor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -910,6 +910,16 @@ TEST_F(AtenXlaTensorTest, TestDim) {
});
}

TEST_F(AtenXlaTensorTest, TestContiguous) {
at::Tensor input = GetTestTensor({2, 3});
at::Tensor output = at::native::contiguous(input);
ForEachDevice([&](const Device& device) {
at::Tensor xla_input = bridge::CreateXlaTensor(input, device);
at::Tensor xla_output = at::native::contiguous(xla_input);
AllClose(output, xla_output);
});
}

TEST_F(AtenXlaTensorTest, TestAvgPool2DBackward) {
int kernel_size = 2;
for (int stride = 1; stride <= 2; ++stride) {
Expand Down
6 changes: 6 additions & 0 deletions torch_xla/csrc/tensor_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,6 +74,12 @@ c10::intrusive_ptr<c10::TensorImpl> XLATensorImpl::shallow_copy_and_detach()
return impl;
}

bool XLATensorImpl::is_contiguous() const {
Copy link
Collaborator

@dlibenzi dlibenzi Feb 14, 2019

Choose a reason for hiding this comment

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

When this trigger?
I mean, in which occasions we have seen.
Asking this because their TensorImpl is a moving target, so the less we fiddle with it, the lower the chances we break their CI.

Copy link
Contributor Author

@asuhan asuhan Feb 14, 2019

Choose a reason for hiding this comment

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

Here: https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/TensorProperties.cpp#L56. If the tensor is contiguous (always the case for us), the operation is no-op. I doubt this is the kind of method which would see API changes.

// Only check that the storage is already contiguous.
XLA_CHECK(is_contiguous_) << "Non-contiguous storage for XLA tensor";
return true;
}

void XLATensorImpl::SetupSizeProperties() {
// Fill up the basic dimension data members which the base class
// implementation uses in its APIs.
Expand Down
2 changes: 2 additions & 0 deletions torch_xla/csrc/tensor_impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,8 @@ class XLATensorImpl : public c10::TensorImpl {

c10::intrusive_ptr<c10::TensorImpl> shallow_copy_and_detach() const override;

bool is_contiguous() const override;

private:
void SetupSizeProperties();

Expand Down