Skip to content

Multiple dispatch compatibility for copy_ in XLA. #1031

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 6 commits into from
Sep 17, 2019
Merged

Multiple dispatch compatibility for copy_ in XLA. #1031

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
3ead02d
Multiple dispatch compatibility for copy_ in XLA.
ezyang Sep 13, 2019
25cc271
Add upstream patch.
ezyang Sep 13, 2019
3997b6e
bugfix
ezyang Sep 13, 2019
6599c9c
bugfixes
ezyang Sep 13, 2019
aefaf81
Remove patch
ezyang Sep 13, 2019
037aad7
Merge branch 'master' of https://github.com/pytorch/xla into pr/multi…
Sep 17, 2019
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
19 changes: 15 additions & 4 deletions torch_xla/csrc/aten_xla_type.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -866,12 +866,23 @@ AtenXlaType::convolution_backward_overrideable(

at::Tensor& AtenXlaType::copy_(at::Tensor& self, const at::Tensor& src,
bool non_blocking) {
XLATensor self_tensor = bridge::GetXlaTensor(self);
c10::optional<XLATensor> self_tensor = bridge::TryGetXlaTensor(self);
c10::optional<XLATensor> src_tensor = bridge::TryGetXlaTensor(src);
if (src_tensor) {
XLATensor::copy_(self_tensor, *src_tensor);

if (!src_tensor) {
XLA_CHECK(self_tensor);
self_tensor->SetTensor(CopyTensor(src, self.scalar_type()));
} else if (!self_tensor) {
// TODO: Is self_tensor good enough? I don't think so... therefore
// the hack below:
std::vector<at::Tensor> tensors = {src};
auto xla_tensors = bridge::XlaCreateTensorList(tensors);
// Hack in an overwrite of a const tensor.
at::Tensor t = CopyTensor(xla_tensors.front(), self.scalar_type());
const_cast<at::Tensor&>(self).unsafeGetTensorImpl()->shallow_copy_from(
t.getIntrusivePtr());
} else {
self_tensor.SetTensor(CopyTensor(src, self.scalar_type()));
XLATensor::copy_(*self_tensor, *src_tensor);
}
return self;
}
Expand Down