Add mean.dtype_out op for Ads model (pytorch#7404)

Summary:

title

Reviewed By: manuelcandales

Differential Revision: D67453766

Author: David Lin

kernels/aten/functions.yaml

@@ -257,6 +257,8 @@
 
 - op: mean.out
 
+- op: mean.dtype_out
+
 - op: min.dim_min
 
 - op: min.unary_out

kernels/portable/cpu/op_mean.cpp

@@ -66,6 +66,14 @@ Tensor& mean_dim_out(
   return out;
 }
 
+Tensor& mean_dtype_out(
+    KernelRuntimeContext& ctx,
+    const Tensor& in,
+    optional<ScalarType> dtype,
+    Tensor& out) {
+  return mean_dim_out(ctx, in, ArrayRef<int64_t>(), false, dtype, out);
+}
+
 } // namespace native
 } // namespace executor
 } // namespace torch

kernels/portable/cpu/util/reduce_util.cpp

@@ -386,6 +386,7 @@ bool check_mean_dim_args(
       check_reduction_args(in, dim_list, keepdim, dtype, out));
 
   if (dtype) {
+    ET_LOG(Info, "dtype is %hhd", static_cast<int8_t>(dtype.value()));
     ET_LOG_AND_RETURN_IF_FALSE(torch::executor::isFloatingType(dtype.value()));
     ET_LOG_AND_RETURN_IF_FALSE(out.scalar_type() == dtype.value());
   } else {

kernels/portable/functions.yaml

@@ -577,6 +577,11 @@
     - arg_meta: null
       kernel_name: torch::executor::mean_dim_out
 
+- op: mean.dtype_out
+  kernels:
+    - arg_meta: null
+      kernel_name: torch::executor::mean_dtype_out
+
 - op: min.dim_min
   kernels:
     - arg_meta: null

kernels/test/op_mean_test.cpp

@@ -10,6 +10,7 @@
 #include <executorch/kernels/test/TestUtil.h>
 #include <executorch/kernels/test/supported_features.h>
 #include <executorch/runtime/core/exec_aten/exec_aten.h>
+#include <executorch/runtime/core/error.h>
 #include <executorch/runtime/core/exec_aten/testing_util/tensor_factory.h>
 #include <executorch/runtime/core/exec_aten/testing_util/tensor_util.h>
 #include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>
@@ -23,6 +24,7 @@ using exec_aten::optional;
 using exec_aten::ScalarType;
 using exec_aten::Tensor;
 using torch::executor::testing::TensorFactory;
+using executorch::runtime::Error;
 
 class OpMeanOutTest : public OperatorTest {
  protected:
@@ -36,6 +38,14 @@ class OpMeanOutTest : public OperatorTest {
         context_, self, dim, keepdim, dtype, out);
   }
 
+  Tensor& op_mean_dtype_out(
+      const Tensor& self,
+      optional<ScalarType> dtype,
+      Tensor& out) {
+    return torch::executor::aten::mean_outf(
+        context_, self, dtype, out);
+  }
+
   template <ScalarType IN_DTYPE, ScalarType OUT_DTYPE>
   void test_mean_dim_out_invalid_dimensions() {
     TensorFactory<IN_DTYPE> tf_in;
@@ -466,3 +476,80 @@ TEST_F(OpMeanOutTest, DynamicShapeUnbound) {
       op_mean_out(x, ArrayRef<int64_t>{1}, false, ScalarType::Float, out);
   EXPECT_TENSOR_CLOSE(out, expected_result);
 }
+
+TEST_F(OpMeanOutTest, DTypeOutFloatValid) {
+  TensorFactory<ScalarType::Float> tf;
+
+  Tensor x = tf.make(
+      {10, 10},
+      {1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0});
+  Tensor expected_result =
+      tf.make({}, {1.0});
+
+  Tensor out = tf.zeros({});
+  Tensor ret =
+      op_mean_dtype_out(x, ScalarType::Float, out);
+  EXPECT_TENSOR_CLOSE(out, expected_result);
+}
+
+TEST_F(OpMeanOutTest, DTypeOutFloatToBoolInvalid) {
+  TensorFactory<ScalarType::Float> tf;
+
+  Tensor x = tf.make(
+      {10, 10},
+      {1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
+       1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0});
+  Tensor expected_result =
+      tf.make({}, {1.0});
+
+  Tensor out = tf.zeros({});
+   
+  ET_EXPECT_KERNEL_FAILURE(
+      context_,
+      op_mean_dtype_out(x, ScalarType::Bool, out));
+}
+
+TEST_F(OpMeanOutTest, DTypeOutFloatInfinity) {
+  TensorFactory<ScalarType::Float> tf;
+
+  Tensor x = tf.make(
+      {2, 1},
+      {INFINITY, INFINITY});
+  Tensor expected_result =
+      tf.make({}, {INFINITY});
+
+  Tensor out = tf.zeros({});
+   
+  Tensor ret =
+      op_mean_dtype_out(x, ScalarType::Float, out);
+  EXPECT_TENSOR_CLOSE(out, expected_result);
+}
+
+TEST_F(OpMeanOutTest, DTypeOutFloatNAN) {
+  TensorFactory<ScalarType::Float> tf;
+
+  Tensor x = tf.make(
+      {2, 1},
+      {NAN, INFINITY});
+  Tensor expected_result =
+      tf.make({}, {NAN});
+
+  Tensor out = tf.zeros({});
+   
+  Tensor ret =
+      op_mean_dtype_out(x, ScalarType::Float, out);
+  EXPECT_TENSOR_CLOSE(out, expected_result);
+}