Remove Ger lowering as it is not needed (#3855)

pytorch · Aug 10, 2022 · 2d880d1 · 2d880d1
1 parent 45689b9
commit 2d880d1
Show file tree

Hide file tree

Showing 9 changed files with 0 additions and 48 deletions.
diff --git a/test/cpp/test_aten_xla_tensor.cpp b/test/cpp/test_aten_xla_tensor.cpp
@@ -3604,7 +3604,6 @@ TEST_F(AtenXlaTensorTest, TestGer) {
   });
 
   ExpectCounterNotChanged("aten::.*", cpp_test::GetIgnoredCounters());
-  ExpectCounterChanged("xla::ger", cpp_test::GetIgnoredCounters());
 }
 
 TEST_F(AtenXlaTensorTest, TestMv) {

diff --git a/torch_xla/csrc/aten_xla_type.cpp b/torch_xla/csrc/aten_xla_type.cpp
@@ -1451,13 +1451,6 @@ at::Tensor XLANativeFunctions::gelu_backward(const at::Tensor& grad,
       bridge::GetXlaTensor(grad), bridge::GetXlaTensor(self), approximate));
 }
 
-at::Tensor XLANativeFunctions::ger(const at::Tensor& self,
-                                   const at::Tensor& vec2) {
-  XLA_FN_COUNTER("xla::");
-  return bridge::AtenFromXlaTensor(
-      XLATensor::ger(bridge::GetXlaTensor(self), bridge::GetXlaTensor(vec2)));
-}
-
 at::Tensor XLANativeFunctions::gt(const at::Tensor& self,
                                   const at::Scalar& other) {
   XLA_FN_COUNTER("xla::");

diff --git a/torch_xla/csrc/ops/ops.cpp b/torch_xla/csrc/ops/ops.cpp
@@ -238,27 +238,6 @@ torch::lazy::NodePtr Celu(const torch::lazy::Value& input,
                    GetXlaShape(input), std::move(lower_fn));
 }
 
-torch::lazy::NodePtr Ger(const torch::lazy::Value& input,
-                         const torch::lazy::Value& other) {
-  auto lower_fn = [](const XlaNode& node,
-                     LoweringContext* loctx) -> XlaOpVector {
-    xla::XlaOp xla_input = loctx->GetOutputOp(node.operand(0));
-    xla::XlaOp xla_other = loctx->GetOutputOp(node.operand(1));
-    return node.ReturnOp(BuildGer(xla_input, xla_other), loctx);
-  };
-  auto lower_for_shape_fn =
-      [](absl::Span<const xla::XlaOp> operands) -> xla::XlaOp {
-    return BuildGer(operands[0], operands[1]);
-  };
-  return GenericOp(torch::lazy::OpKind(at::aten::ger), {input, other},
-                   [&]() {
-                     return InferOutputShape(
-                         {GetXlaShape(input), GetXlaShape(other)},
-                         lower_for_shape_fn);
-                   },
-                   std::move(lower_fn));
-}
-
 torch::lazy::NodePtr AddMatMulOp(const torch::lazy::Value& input,
                                  const torch::lazy::Value& weight,
                                  const torch::lazy::Value& bias) {

diff --git a/torch_xla/csrc/ops/ops.h b/torch_xla/csrc/ops/ops.h
@@ -140,9 +140,6 @@ torch::lazy::NodePtr Celu(const torch::lazy::Value& input,
 
 torch::lazy::NodePtr FracOp(const torch::lazy::Value& input);
 
-torch::lazy::NodePtr Ger(const torch::lazy::Value& input,
-                         const torch::lazy::Value& other);
-
 torch::lazy::NodePtr AddMatMulOp(const torch::lazy::Value& input,
                                  const torch::lazy::Value& weight,
                                  const torch::lazy::Value& bias);

diff --git a/torch_xla/csrc/tensor.h b/torch_xla/csrc/tensor.h
@@ -624,8 +624,6 @@ class XLATensor : public c10::intrusive_ptr_target {
                                     const XLATensorPtr& input,
                                     const c10::string_view approximate);
 
-  static XLATensorPtr ger(const XLATensorPtr& input, const XLATensorPtr& vec2);
-
   static XLATensorPtr gt(const XLATensorPtr& input, const at::Scalar& other);
 
   static XLATensorPtr gt(const XLATensorPtr& input, const XLATensorPtr& other);

diff --git a/torch_xla/csrc/tensor_methods.cpp b/torch_xla/csrc/tensor_methods.cpp
@@ -1398,11 +1398,6 @@ XLATensorPtr XLATensor::gelu_backward(const XLATensorPtr& grad,
   }
 }
 
-XLATensorPtr XLATensor::ger(const XLATensorPtr& input,
-                            const XLATensorPtr& vec2) {
-  return input->CreateFrom(Ger(input->GetIrValue(), vec2->GetIrValue()));
-}
-
 XLATensorPtr XLATensor::gt(const XLATensorPtr& input, const at::Scalar& other) {
   return DispatchComparisonOp(at::aten::gt, input, other);
 }

diff --git a/torch_xla/csrc/xla_lower_util.cpp b/torch_xla/csrc/xla_lower_util.cpp
@@ -430,12 +430,6 @@ xla::XlaOp CreateMatMul(xla::XlaOp lhs, xla::XlaOp rhs) {
               << rhs_shape << ")";
 }
 
-xla::XlaOp BuildGer(xla::XlaOp lhs, xla::XlaOp rhs) {
-  xla::XlaOp lhs_reshaped = BuildUnsqueeze(lhs, 1);
-  xla::XlaOp rhs_reshaped = BuildUnsqueeze(rhs, 0);
-  return BuildDot(lhs_reshaped, rhs_reshaped);
-}
-
 xla::XlaOp BuildMatMul(xla::XlaOp lhs, xla::XlaOp rhs, xla::XlaOp bias) {
   xla::XlaOp dot = BuildDot(lhs, rhs);
   const xla::Shape& dot_shape = XlaHelpers::ShapeOfXlaOp(dot);

diff --git a/torch_xla/csrc/xla_lower_util.h b/torch_xla/csrc/xla_lower_util.h
@@ -20,8 +20,6 @@ std::vector<xla::XlaOp> CreateTopK(xla::XlaOp input, int64_t k, int64_t dim,
 
 xla::XlaOp CreateMatMul(xla::XlaOp lhs, xla::XlaOp rhs);
 
-xla::XlaOp BuildGer(xla::XlaOp lhs, xla::XlaOp rhs);
-
 xla::XlaOp BuildMatMul(xla::XlaOp lhs, xla::XlaOp rhs, xla::XlaOp bias);
 
 xla::XlaOp BuildMatMulWithMultiplier(xla::XlaOp lhs, xla::XlaOp rhs,

diff --git a/xla_native_functions.yaml b/xla_native_functions.yaml
@@ -168,7 +168,6 @@ supported:
   - ge.Tensor
   - gelu
   - gelu_backward
-  - ger
   - gt.Scalar
   - gt.Tensor
   - hardshrink
-Original file line number
+Diff line change
@@ Expand Up / @@ -3604,7 +3604,6 @@ TEST_F(AtenXlaTensorTest, TestGer) { @@
       });
       ExpectCounterNotChanged("aten::.*", cpp_test::GetIgnoredCounters());
-      ExpectCounterChanged("xla::ger", cpp_test::GetIgnoredCounters());
     }
     TEST_F(AtenXlaTensorTest, TestMv) {
@@ Expand Down @@