COMPMID-1764 NEON: Implement ArgMax/ArgMin

Change-Id: Ibe23aa90b36ffd8553d1d1c35fada5d300fab829
Reviewed-on: https://review.mlplatform.org/475
Reviewed-by: Isabella Gottardi <isabella.gottardi@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>

7 files changed, 239 insertions, 67 deletions

diff --git a/tests/validation/NEON/ArgMinMax.cpp b/tests/validation/NEON/ArgMinMax.cpp
new file mode 100644
index 0000000000..611495a41d
--- /dev/null
+++ b/tests/validation/NEON/ArgMinMax.cpp
@@ -0,0 +1,167 @@
+/*
+ * Copyright (c) 2018-2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEArgMinMaxLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+
+#include "tests/NEON/Accessor.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/datasets/SplitDataset.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/ArgMinMaxFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+TEST_SUITE(NEON)
+TEST_SUITE(ArgMinMax)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
+        framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 3U, 16U, 2U), 1, DataType::F32), // Invalid axis
+                                                TensorInfo(TensorShape(27U, 3U, 16U, 2U), 1, DataType::F32), // Invalid output shape
+                                                TensorInfo(TensorShape(32U, 16U, 16U, 2U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(32U, 16U, 16U, 2U), 1, DataType::F32) // Invalid operation
+        }),
+        framework::dataset::make("OutputInfo", { TensorInfo(TensorShape(27U, 3U, 1U, 2U), 1, DataType::F32),
+                                                 TensorInfo(TensorShape(27U, 3U, 1U, 2U), 1, DataType::F32),
+                                                 TensorInfo(TensorShape(32U, 16U, 1U, 2U), 1, DataType::U32),
+                                                 TensorInfo(TensorShape(32U, 16U, 1U, 2U), 1, DataType::F32)
+        })),
+        framework::dataset::make("Axis", { 4, 0, 2, 0 })),
+        framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MAX, ReductionOperation::ARG_IDX_MAX, ReductionOperation::ARG_IDX_MAX, ReductionOperation::MEAN_SUM })),
+        framework::dataset::make("Expected", { false, false, true, false })),
+        input_info, output_info, axis, operation, expected)
+{
+    const Status status = NEArgMinMaxLayer::validate(&input_info.clone()->set_is_resizable(false), axis, &output_info.clone()->set_is_resizable(false), operation);
+    ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+DATA_TEST_CASE(Configuration,
+               framework::DatasetMode::ALL,
+               combine(datasets::SmallShapes(), framework::dataset::make("DataType", { DataType::F16, DataType::F32 })),
+               shape, data_type)
+{
+    // Create tensors
+    Tensor ref_src = create_tensor<Tensor>(shape, data_type);
+    Tensor dst;
+
+    // Create and Configure function
+    NEArgMinMaxLayer arg_min_max_layer;
+    arg_min_max_layer.configure(&ref_src, 1, &dst, ReductionOperation::ARG_IDX_MAX);
+
+    // Validate valid region
+    TensorShape output_shape = shape;
+    output_shape.set(1, 1);
+    const ValidRegion valid_region = shape_to_valid_region(output_shape);
+    validate(dst.info()->valid_region(), valid_region);
+}
+
+template <typename T>
+using NEArgMinMaxValidationFixture = ArgMinMaxValidationFixture<Tensor, Accessor, NEArgMinMaxLayer, T>;
+
+TEST_SUITE(Float)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEArgMinMaxValidationFixture<half>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(datasets::Small4DShapes(), framework::dataset::make("DataType", DataType::F16)), framework::dataset::make("Axis", { 0, 1, 2, 3 })), framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MIN, ReductionOperation::ARG_IDX_MAX })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEArgMinMaxValidationFixture<half>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(datasets::Large4DShapes(), framework::dataset::make("DataType", DataType::F16)), framework::dataset::make("Axis", { 0, 1, 2, 3 })), framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MIN, ReductionOperation::ARG_IDX_MAX })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP16
+#endif           // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+template <typename T>
+using NEArgMinMaxQuantizedValidationFixture = ArgMinMaxValidationQuantizedFixture<Tensor, Accessor, NEArgMinMaxLayer, T>;
+
+TEST_SUITE(QASYMM8)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEArgMinMaxQuantizedValidationFixture<uint8_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(datasets::Small4DShapes(), framework::dataset::make("DataType", DataType::QASYMM8)), framework::dataset::make("Axis", { 0, 1, 2, 3 })),
+                                       framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MIN, ReductionOperation::ARG_IDX_MAX })),
+                               framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEArgMinMaxQuantizedValidationFixture<uint8_t>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(datasets::Large4DShapes(), framework::dataset::make("DataType", DataType::QASYMM8)), framework::dataset::make("Axis", { 0, 1, 2, 3 })),
+                                       framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MIN, ReductionOperation::ARG_IDX_MAX })),
+                               framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // QASYMM8
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       NEArgMinMaxValidationFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(datasets::Small4DShapes(), framework::dataset::make("DataType", DataType::F32)), framework::dataset::make("Axis", { 0, 1, 2, 3 })), framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MIN, ReductionOperation::ARG_IDX_MAX })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       NEArgMinMaxValidationFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(datasets::Large4DShapes(), framework::dataset::make("DataType", DataType::F32)), framework::dataset::make("Axis", { 0, 1, 2, 3 })), framework::dataset::make("Operation", { ReductionOperation::ARG_IDX_MIN, ReductionOperation::ARG_IDX_MAX })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // FP32
+TEST_SUITE_END() // Float
+TEST_SUITE_END() // ArgMinMax
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/ArgMinMaxFixture.h b/tests/validation/fixtures/ArgMinMaxFixture.h
index 5f5f85c104..e263b25bf2 100644
--- a/tests/validation/fixtures/ArgMinMaxFixture.h
+++ b/tests/validation/fixtures/ArgMinMaxFixture.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,28 +42,38 @@ namespace test
 namespace validation
 {
 template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
-class ArgMinMaxValidationFixture : public framework::Fixture
+class ArgMinMaxValidationBaseFixture : public framework::Fixture
 {
 public:
     template <typename...>
-    void setup(TensorShape shape, DataType data_type, int axis, ReductionOperation op)
+    void setup(TensorShape shape, DataType data_type, int axis, ReductionOperation op, QuantizationInfo q_info)
     {
-        _target    = compute_target(shape, data_type, axis, op);
-        _reference = compute_reference(shape, data_type, axis, op);
+        _target    = compute_target(shape, data_type, axis, op, q_info);
+        _reference = compute_reference(shape, data_type, axis, op, q_info);
     }
 
 protected:
     template <typename U>
     void fill(U &&tensor)
     {
-        std::uniform_real_distribution<> distribution(-1.0f, 1.0f);
-        library->fill(tensor, distribution, 0);
+        if(!is_data_type_quantized(tensor.data_type()))
+        {
+            std::uniform_real_distribution<> distribution(-1.0f, 1.0f);
+            library->fill(tensor, distribution, 0);
+        }
+        else
+        {
+            std::pair<int, int> bounds = get_quantized_bounds(tensor.quantization_info(), -1.0f, 1.0f);
+            std::uniform_int_distribution<uint8_t> distribution(bounds.first, bounds.second);
+
+            library->fill(tensor, distribution, 0);
+        }
     }
 
-    TensorType compute_target(TensorShape &src_shape, DataType data_type, int axis, ReductionOperation op)
+    TensorType compute_target(TensorShape &src_shape, DataType data_type, int axis, ReductionOperation op, QuantizationInfo q_info)
     {
         // Create tensors
-        TensorType src = create_tensor<TensorType>(src_shape, data_type, 1);
+        TensorType src = create_tensor<TensorType>(src_shape, data_type, 1, q_info);
         TensorType dst;
 
         // Create and configure function
@@ -89,21 +99,43 @@ protected:
         return dst;
     }
 
-    SimpleTensor<T> compute_reference(TensorShape &src_shape, DataType data_type, int axis, ReductionOperation op)
+    SimpleTensor<uint32_t> compute_reference(TensorShape &src_shape, DataType data_type, int axis, ReductionOperation op, QuantizationInfo q_info)
     {
         // Create reference
-        SimpleTensor<T> src{ src_shape, data_type, 1 };
+        SimpleTensor<T> src{ src_shape, data_type, 1, q_info };
 
         // Fill reference
         fill(src);
 
         TensorShape output_shape = src_shape;
         output_shape.set(axis, 1);
-        return reference::reduction_operation<T>(src, output_shape, axis, op);
+        return reference::reduction_operation<T, uint32_t>(src, output_shape, axis, op);
     }
 
-    TensorType      _target{};
-    SimpleTensor<T> _reference{};
+    TensorType             _target{};
+    SimpleTensor<uint32_t> _reference{};
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class ArgMinMaxValidationQuantizedFixture : public ArgMinMaxValidationBaseFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(const TensorShape &shape, DataType data_type, int axis, ReductionOperation op, QuantizationInfo quantization_info)
+    {
+        ArgMinMaxValidationBaseFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, data_type, axis, op, quantization_info);
+    }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class ArgMinMaxValidationFixture : public ArgMinMaxValidationBaseFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+    template <typename...>
+    void setup(const TensorShape &shape, DataType data_type, int axis, ReductionOperation op)
+    {
+        ArgMinMaxValidationBaseFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, data_type, axis, op, QuantizationInfo());
+    }
 };
 } // namespace validation
 } // namespace test
diff --git a/tests/validation/fixtures/ReduceMeanFixture.h b/tests/validation/fixtures/ReduceMeanFixture.h
index 769d7f674f..44bb9fca6a 100644
--- a/tests/validation/fixtures/ReduceMeanFixture.h
+++ b/tests/validation/fixtures/ReduceMeanFixture.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -113,7 +113,7 @@ protected:
         {
             TensorShape output_shape = i == 0 ? src_shape : out.shape();
             output_shape.set(axis[i], 1);
-            out = reference::reduction_operation<T>(i == 0 ? src : out, output_shape, axis[i], ReductionOperation::MEAN_SUM);
+            out = reference::reduction_operation<T, T>(i == 0 ? src : out, output_shape, axis[i], ReductionOperation::MEAN_SUM);
         }
 
         if(!keep_dims)
diff --git a/tests/validation/fixtures/ReductionOperationFixture.h b/tests/validation/fixtures/ReductionOperationFixture.h
index 9079b47cbb..d01f41abf0 100644
--- a/tests/validation/fixtures/ReductionOperationFixture.h
+++ b/tests/validation/fixtures/ReductionOperationFixture.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -107,7 +107,7 @@ protected:
         // Fill reference
         fill(src);
 
-        return reference::reduction_operation<T>(src, dst_shape, axis, op);
+        return reference::reduction_operation<T, T>(src, dst_shape, axis, op);
     }
 
     TensorType      _target{};
diff --git a/tests/validation/reference/L2NormalizeLayer.cpp b/tests/validation/reference/L2NormalizeLayer.cpp
index fcd6226f07..43885b29e2 100644
--- a/tests/validation/reference/L2NormalizeLayer.cpp
+++ b/tests/validation/reference/L2NormalizeLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -54,7 +54,7 @@ SimpleTensor<T> l2_normalize(const SimpleTensor<T> &src, unsigned int axis, floa
     SimpleTensor<T> dst{ src.shape(), src.data_type() };
 
     // Reduce across given axis
-    SimpleTensor<T> sum = reduction_operation(src, get_output_shape(src.shape(), axis), axis, ReductionOperation::SUM_SQUARE);
+    SimpleTensor<T> sum = reduction_operation<T, T>(src, get_output_shape(src.shape(), axis), axis, ReductionOperation::SUM_SQUARE);
 
     // Compute reference
     const int upper_dims     = src.shape().total_size_upper(axis + 1);
diff --git a/tests/validation/reference/ReductionOperation.cpp b/tests/validation/reference/ReductionOperation.cpp
index 37a9be86c0..fc12e31d75 100644
--- a/tests/validation/reference/ReductionOperation.cpp
+++ b/tests/validation/reference/ReductionOperation.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -49,20 +49,20 @@ OT reduce_operation(const T *ptr, int reduce_elements, ReductionOperation op, in
         uint32_t int_res = 0;
         for(int i = 0; i < reduce_elements; ++i)
         {
-            auto elem = static_cast<uint32_t>(*(ptr + stride * i));
+            auto elem = *(ptr + stride * i);
 
             switch(op)
             {
                 case ReductionOperation::ARG_IDX_MIN:
-                    if(static_cast<uint32_t>(*(ptr + stride * static_cast<uint32_t>(res))) > elem)
+                    if(*(ptr + stride * static_cast<uint32_t>(int_res)) > elem)
                     {
-                        res = static_cast<uint32_t>(i);
+                        int_res = static_cast<uint32_t>(i);
                     }
                     break;
                 case ReductionOperation::ARG_IDX_MAX:
-                    if(static_cast<uint32_t>(*(ptr + stride * static_cast<uint32_t>(res))) < elem)
+                    if(*(ptr + stride * static_cast<uint32_t>(int_res)) < elem)
                     {
-                        res = static_cast<uint32_t>(i);
+                        int_res = static_cast<uint32_t>(i);
                     }
                     break;
                 case ReductionOperation::SUM_SQUARE:
@@ -122,13 +122,13 @@ OT reduce_operation(const T *ptr, int reduce_elements, ReductionOperation op, in
 }
 } // namespace
 
-template <typename T>
-SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op)
+template <typename T, typename OT>
+SimpleTensor<OT> reduction_operation(const SimpleTensor<T> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op)
 {
     // Create reference
     const bool         is_arg_min_max   = (op == ReductionOperation::ARG_IDX_MIN || op == ReductionOperation::ARG_IDX_MAX);
     DataType           output_data_type = is_arg_min_max ? DataType::U32 : src.data_type();
-    SimpleTensor<T>    dst{ dst_shape, output_data_type, 1, src.quantization_info() };
+    SimpleTensor<OT>   dst{ dst_shape, output_data_type, 1, src.quantization_info() };
     const unsigned int src_width    = src.shape().x();
     const unsigned int src_height   = src.shape().y();
     const unsigned int src_depth    = src.shape().z();
@@ -143,14 +143,7 @@ SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShap
             for(unsigned int du = 0; du < upper_dims; ++du)
             {
                 const T *src_row_ptr = src.data() + du * reduce_elems;
-                if(is_arg_min_max)
-                {
-                    dst[du] = reduce_operation<T, uint32_t>(src_row_ptr, reduce_elems, op, 1);
-                }
-                else
-                {
-                    dst[du] = reduce_operation<T, T>(src_row_ptr, reduce_elems, op, 1);
-                }
+                dst[du]              = reduce_operation<T, OT>(src_row_ptr, reduce_elems, op, 1);
             }
         }
         break;
@@ -164,15 +157,7 @@ SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShap
                     const int in_offset   = du * src_height * src_width + x;
                     const int out_offset  = du * src_width + x;
                     const T *src_row_ptr = src.data() + in_offset;
-
-                    if(is_arg_min_max)
-                    {
-                        dst[out_offset] = reduce_operation<T, uint32_t>(src_row_ptr, reduce_elems, op, src_width);
-                    }
-                    else
-                    {
-                        dst[out_offset] = reduce_operation<T, T>(src_row_ptr, reduce_elems, op, src_width);
-                    }
+                    dst[out_offset]       = reduce_operation<T, OT>(src_row_ptr, reduce_elems, op, src_width);
                 }
             }
         }
@@ -189,15 +174,7 @@ SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShap
                         const int in_offset   = du * src_depth * src_height * src_width + y * src_width + x;
                         const int out_offset  = du * src_width * src_height + y * src_width + x;
                         const T *src_row_ptr = src.data() + in_offset;
-
-                        if(is_arg_min_max)
-                        {
-                            dst[out_offset] = reduce_operation<T, uint32_t>(src_row_ptr, reduce_elems, op, src_height * src_width);
-                        }
-                        else
-                        {
-                            dst[out_offset] = reduce_operation<T, T>(src_row_ptr, reduce_elems, op, src_height * src_width);
-                        }
+                        dst[out_offset]       = reduce_operation<T, OT>(src_row_ptr, reduce_elems, op, src_height * src_width);
                     }
                 }
             }
@@ -217,14 +194,7 @@ SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShap
                             const int in_offset   = du * src_batch * src_depth * src_height * src_width + z * src_width * src_height + y * src_width + x;
                             const int out_offset  = du * src_depth * src_height * src_width + z * src_width * src_height + y * src_width + x;
                             const T *src_row_ptr = src.data() + in_offset;
-                            if(is_arg_min_max)
-                            {
-                                dst[out_offset] = reduce_operation<T, uint32_t>(src_row_ptr, reduce_elems, op, src_width * src_height * src_depth);
-                            }
-                            else
-                            {
-                                dst[out_offset] = reduce_operation<T, T>(src_row_ptr, reduce_elems, op, src_width * src_height * src_depth);
-                            }
+                            dst[out_offset]       = reduce_operation<T, OT>(src_row_ptr, reduce_elems, op, src_width * src_height * src_depth);
                         }
                     }
                 }
@@ -238,6 +208,9 @@ SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShap
     return dst;
 }
 
+template SimpleTensor<uint32_t> reduction_operation(const SimpleTensor<float> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
+template SimpleTensor<uint32_t> reduction_operation(const SimpleTensor<half> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
+template SimpleTensor<uint32_t> reduction_operation(const SimpleTensor<uint8_t> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
 template SimpleTensor<float> reduction_operation(const SimpleTensor<float> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
 template SimpleTensor<half> reduction_operation(const SimpleTensor<half> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
 template SimpleTensor<uint8_t> reduction_operation(const SimpleTensor<uint8_t> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
diff --git a/tests/validation/reference/ReductionOperation.h b/tests/validation/reference/ReductionOperation.h
index 859b57aa7b..9f7050f551 100644
--- a/tests/validation/reference/ReductionOperation.h
+++ b/tests/validation/reference/ReductionOperation.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -35,10 +35,10 @@ namespace validation
 {
 namespace reference
 {
-template <typename T>
-SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
+template <typename T, typename OT>
+SimpleTensor<OT> reduction_operation(const SimpleTensor<T> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
 } // namespace reference
 } // namespace validation
 } // namespace test
 } // namespace arm_compute
-#endif /* __ARM_COMPUTE_TEST_FLOOR_H__ */
+#endif /* __ARM_COMPUTE_TEST_REDUCTION_OPERATION_H__ */