Reapply "Fix some more core aten ops (#6342)" (#6377) (#6387)

pytorch · Jan 26, 2024 · 9e4db96 · 9e4db96
1 parent c70e4cc
commit 9e4db96
Show file tree

Hide file tree

Showing 6 changed files with 44 additions and 14 deletions.
diff --git a/codegen/xla_native_functions.yaml b/codegen/xla_native_functions.yaml
@@ -77,6 +77,7 @@ full_codegen:
   - rsqrt
   - selu
   - sgn
+  - sigmoid
   - sign
   - silu
   - silu_backward
@@ -304,7 +305,6 @@ supported:
   - select_scatter
   - selu_
   - set_.source_Tensor
-  - sigmoid
   - sigmoid_backward
   - slice_copy.Tensor
   - slice_scatter

diff --git a/test/test_core_aten_ops.py b/test/test_core_aten_ops.py
@@ -1904,11 +1904,17 @@ def test_aten_gelu_0(self):
     kwargs = dict()
     run_export_and_compare(self, torch.ops.aten.gelu, args, kwargs)
 
-  @unittest.skip
   def test_aten_gelu_1(self):
     args = (torch.randn((10, 10)).to(torch.float16),)
     kwargs = dict()
-    run_export_and_compare(self, torch.ops.aten.gelu, args, kwargs)
+    run_export_and_compare(
+        self,
+        torch.ops.aten.gelu,
+        args,
+        kwargs,
+        rtol=0.001,
+        atol=0.01,
+    )
 
   def test_aten_glu_0(self):
     args = (
@@ -3082,7 +3088,6 @@ def test_aten_native_group_norm_0(self):
     kwargs = dict()
     run_export_and_compare(self, torch.ops.aten.native_group_norm, args, kwargs)
 
-  @unittest.skip
   def test_aten_native_group_norm_1(self):
     args = (
         torch.randn((1, 3, 2, 10)).to(torch.float16),
@@ -3095,7 +3100,14 @@ def test_aten_native_group_norm_1(self):
         0.0,
     )
     kwargs = dict()
-    run_export_and_compare(self, torch.ops.aten.native_group_norm, args, kwargs)
+    run_export_and_compare(
+        self,
+        torch.ops.aten.native_group_norm,
+        args,
+        kwargs,
+        rtol=0.001,
+        atol=0.01,
+    )
 
   def test_aten_native_layer_norm_0(self):
     args = (
@@ -3411,7 +3423,6 @@ def test_aten_reciprocal_1(self):
     kwargs = dict()
     run_export_and_compare(self, torch.ops.aten.reciprocal, args, kwargs)
 
-  @unittest.skip
   def test_aten_reciprocal_2(self):
     args = (torch.randint(0, 10, (10, 10)).to(torch.int32),)
     kwargs = dict()
@@ -4009,7 +4020,6 @@ def test_aten_sigmoid_1(self):
     kwargs = dict()
     run_export_and_compare(self, torch.ops.aten.sigmoid, args, kwargs)
 
-  @unittest.skip
   def test_aten_sigmoid_2(self):
     args = (torch.randint(0, 10, (10, 10)).to(torch.int32),)
     kwargs = dict()

diff --git a/torch_xla/csrc/aten_xla_type.cpp b/torch_xla/csrc/aten_xla_type.cpp
@@ -2791,12 +2791,6 @@ at::Tensor& XLANativeFunctions::set_(at::Tensor& self,
   return self;
 }
 
-at::Tensor XLANativeFunctions::sigmoid(const at::Tensor& self) {
-  TORCH_LAZY_FN_COUNTER_TIMED_TRACING("xla::");
-  return bridge::AtenFromXlaTensor(
-      tensor_methods::sigmoid(bridge::GetXlaTensor(self)));
-}
-
 at::Tensor XLANativeFunctions::sigmoid_backward(const at::Tensor& grad_output,
                                                 const at::Tensor& output) {
   TORCH_LAZY_FN_COUNTER_TIMED_TRACING("xla::");

diff --git a/torch_xla/csrc/ops/ops_lower_fn.cpp b/torch_xla/csrc/ops/ops_lower_fn.cpp
@@ -684,6 +684,10 @@ torch_xla::XlaOpVector NeTensor::Lower(LoweringContext* loctx) const {
 
 torch_xla::XlaOpVector Reciprocal::Lower(LoweringContext* loctx) const {
   xla::XlaOp xla_input = loctx->GetOutputOp(operand(0));
+  if (xla::primitive_util::IsIntegralType(XlaHelpers::TypeOfXlaOp(xla_input))) {
+    xla::PrimitiveType input_type = XlaHelpers::TypeOfXlaOp(xla_input);
+    xla_input = ConvertTo(xla_input, input_type, xla::PrimitiveType::F32);
+  }
   return ReturnOp(BuildReciprocal(xla_input), loctx);
 }
 
@@ -726,6 +730,14 @@ torch_xla::XlaOpVector Sgn::Lower(LoweringContext* loctx) const {
   return ReturnOp(BuildSgn(xla_input), loctx);
 }
 
+torch_xla::XlaOpVector Sigmoid::Lower(LoweringContext* loctx) const {
+  xla::XlaOp xla_input = loctx->GetOutputOp(operand(0));
+  if (xla::primitive_util::IsIntegralType(XlaHelpers::TypeOfXlaOp(xla_input))) {
+    xla_input = xla::ConvertElementType(xla_input, xla::PrimitiveType::F32);
+  }
+  return ReturnOp(xla::Logistic(xla_input), loctx);
+}
+
 torch_xla::XlaOpVector Sign::Lower(LoweringContext* loctx) const {
   xla::XlaOp xla_input = loctx->GetOutputOp(operand(0));
   return ReturnOp(BuildSign(xla_input), loctx);

diff --git a/torch_xla/csrc/ops/ops_xla_shape_fn.cpp b/torch_xla/csrc/ops/ops_xla_shape_fn.cpp
@@ -762,7 +762,11 @@ xla::Shape NeTensorOutputShape(const torch::lazy::Value& self,
 }
 
 xla::Shape ReciprocalOutputShape(const torch::lazy::Value& input) {
-  return GetXlaShape(input);
+  xla::Shape result_shape = GetXlaShape(input);
+  if (xla::primitive_util::IsIntegralType(result_shape.element_type())) {
+    result_shape.set_element_type(xla::PrimitiveType::F32);
+  }
+  return result_shape;
 }
 
 xla::Shape ReluOutputShape(const torch::lazy::Value& input) {
@@ -804,6 +808,14 @@ xla::Shape SgnOutputShape(const torch::lazy::Value& input) {
   return GetXlaShape(input);
 }
 
+xla::Shape SigmoidOutputShape(const torch::lazy::Value& input) {
+  xla::Shape result_shape = GetXlaShape(input);
+  if (xla::primitive_util::IsIntegralType(result_shape.element_type())) {
+    result_shape.set_element_type(xla::PrimitiveType::F32);
+  }
+  return result_shape;
+}
+
 xla::Shape SignOutputShape(const torch::lazy::Value& input) {
   return GetXlaShape(input);
 }

diff --git a/torch_xla/csrc/ops/ops_xla_shape_fn.h b/torch_xla/csrc/ops/ops_xla_shape_fn.h
@@ -248,6 +248,8 @@ xla::Shape SeluOutputShape(const torch::lazy::Value& input);
 
 xla::Shape SgnOutputShape(const torch::lazy::Value& input);
 
+xla::Shape SigmoidOutputShape(const torch::lazy::Value& input);
+
 xla::Shape SignOutputShape(const torch::lazy::Value& input);
 
 xla::Shape SiluOutputShape(const torch::lazy::Value& input);