Skip to content

Add arg min / max and dot wrappers #424

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 2 commits into from
Mar 10, 2019
Merged

Add arg min / max and dot wrappers #424

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1bb53cb
Add _argmin and _argmax to ATen XLA tensor
asuhan Mar 9, 2019
f5bf077
Add dot to ATen XLA tensor
asuhan Mar 9, 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
36 changes: 36 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 @@ -1099,6 +1099,18 @@ TEST_F(AtenXlaTensorTest, TestArgMinSameValue) {
});
}

TEST_F(AtenXlaTensorTest, TestArgMinWrapper) {
at::Tensor a = at::rand({4, 4, 4}, at::TensorOptions(at::kFloat));
for (int dim : {1, -2}) {
at::Tensor b = at::_argmin(a, dim, /*keepdim=*/false);
ForEachDevice([&](const Device& device) {
at::Tensor xla_a = bridge::CreateXlaTensor(a, device);
at::Tensor xla_b = at::_argmin(xla_a, dim, /*keepdim=*/false);
AllClose(b, xla_b);
});
}
}

TEST_F(AtenXlaTensorTest, TestArgMax) {
at::Tensor a = at::rand({4, 4, 4}, at::TensorOptions(at::kFloat));
at::Tensor b = at::argmax(a);
Expand Down Expand Up @@ -1143,6 +1155,18 @@ TEST_F(AtenXlaTensorTest, TestArgMaxSameValue) {
});
}

TEST_F(AtenXlaTensorTest, TestArgMaxWrapper) {
at::Tensor a = at::rand({4, 4, 4}, at::TensorOptions(at::kFloat));
for (int dim : {1, -2}) {
at::Tensor b = at::_argmax(a, dim, /*keepdim=*/false);
ForEachDevice([&](const Device& device) {
at::Tensor xla_a = bridge::CreateXlaTensor(a, device);
at::Tensor xla_b = at::_argmax(xla_a, dim, /*keepdim=*/false);
AllClose(b, xla_b);
});
}
}

TEST_F(AtenXlaTensorTest, TestAsin) {
at::Tensor a = at::rand({2, 2}, at::TensorOptions(at::kFloat));
at::Tensor b = at::asin(a);
Expand Down Expand Up @@ -1497,6 +1521,18 @@ TEST_F(AtenXlaTensorTest, TestMatmulBcast) {
});
}

TEST_F(AtenXlaTensorTest, TestDot) {
at::Tensor a = at::rand({4}, at::TensorOptions(at::kFloat));
at::Tensor b = at::rand({4}, at::TensorOptions(at::kFloat));
at::Tensor c = at::dot(a, b);
ForEachDevice([&](const Device& device) {
at::Tensor xla_a = bridge::CreateXlaTensor(a, device);
at::Tensor xla_b = bridge::CreateXlaTensor(b, device);
at::Tensor xla_c = at::dot(xla_a, xla_b);
AllClose(c, xla_c);
});
}

TEST_F(AtenXlaTensorTest, TestBatchMatMul) {
at::Tensor a = at::rand({3, 6, 4}, at::TensorOptions(at::kFloat));
at::Tensor b = at::rand({3, 4, 5}, at::TensorOptions(at::kFloat));
Expand Down
19 changes: 19 additions & 0 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 @@ -1401,6 +1401,15 @@ at::Tensor AtenXlaType::bmm(const at::Tensor& self,
XLATensor::bmm(bridge::GetXlaTensor(self), bridge::GetXlaTensor(mat2)));
}

at::Tensor AtenXlaType::dot(const at::Tensor& self,
const at::Tensor& tensor) const {
XLA_CHECK_EQ(self.dim(), 1)
<< "dot: Expected 1-D argument self, but got " << self.dim() << "-D";
XLA_CHECK_EQ(tensor.dim(), 1)
<< "dot: Expected 1-D argument tensor, but got " << tensor.dim() << "-D";
return matmul(self, tensor);
}

std::vector<at::Tensor> AtenXlaType::broadcast_tensors(
at::TensorList tensors) const {
return bridge::AtenFromXlaTensors(
Expand Down Expand Up @@ -1787,6 +1796,11 @@ at::Tensor AtenXlaType::argmax(const at::Tensor& self, int64_t dim,
XLATensor::argmax(bridge::GetXlaTensor(self), dim, keepdim));
}

at::Tensor AtenXlaType::_argmax(const at::Tensor& self, int64_t dim,
bool keepdim) const {
return at::native::_argmax(self, dim, keepdim);
}

at::Tensor AtenXlaType::argmax(const at::Tensor& self) const {
return bridge::AtenFromXlaTensor(
XLATensor::argmax(bridge::GetXlaTensor(self)));
Expand All @@ -1798,6 +1812,11 @@ at::Tensor AtenXlaType::argmin(const at::Tensor& self, int64_t dim,
XLATensor::argmin(bridge::GetXlaTensor(self), dim, keepdim));
}

at::Tensor AtenXlaType::_argmin(const at::Tensor& self, int64_t dim,
bool keepdim) const {
return at::native::_argmin(self, dim, keepdim);
}

at::Tensor AtenXlaType::argmin(const at::Tensor& self) const {
return bridge::AtenFromXlaTensor(
XLATensor::argmin(bridge::GetXlaTensor(self)));
Expand Down
7 changes: 7 additions & 0 deletions torch_xla/csrc/aten_xla_type.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -429,6 +429,9 @@ class AtenXlaType : public AtenXlaTypeBase {

at::Tensor bmm(const at::Tensor& self, const at::Tensor& mat2) const override;

at::Tensor dot(const at::Tensor& self,
const at::Tensor& tensor) const override;

std::vector<at::Tensor> broadcast_tensors(
at::TensorList tensors) const override;

Expand Down Expand Up @@ -562,10 +565,14 @@ class AtenXlaType : public AtenXlaTypeBase {
at::Tensor argmax(const at::Tensor& self, int64_t dim,
bool keepdim) const override;
at::Tensor argmax(const at::Tensor& self) const override;
at::Tensor _argmax(const at::Tensor& self, int64_t dim,
bool keepdim) const override;

at::Tensor argmin(const at::Tensor& self, int64_t dim,
bool keepdim) const override;
at::Tensor argmin(const at::Tensor& self) const override;
at::Tensor _argmin(const at::Tensor& self, int64_t dim,
bool keepdim) const override;

std::tuple<at::Tensor, at::Tensor, at::Tensor> native_batch_norm_backward(
const at::Tensor& grad_out, const at::Tensor& input,
Expand Down