IVGCVSW-7400 POW IVGCVSW-7278 SQUARED_DIFFERENCE to CpuAcc and GpuAcc

* Add POW SQUARED_DIFFERENCE and Unit tests for CpuAcc and GpuAcc

Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>

Change-Id: Ifa78af2a2fda2074586d8e4d9a506b1b13fa5755

9 files changed, 213 insertions, 3 deletions

diff --git a/src/backends/cl/ClLayerSupport.cpp b/src/backends/cl/ClLayerSupport.cpp
index 6fa4f3ce51..ff2b576f3d 100644
--- a/src/backends/cl/ClLayerSupport.cpp
+++ b/src/backends/cl/ClLayerSupport.cpp
@@ -37,6 +37,7 @@
 #include "workloads/ClDepthwiseConvolutionWorkload.hpp"
 #include "workloads/ClDequantizeWorkload.hpp"
 #include "workloads/ClDivisionWorkload.hpp"
+#include "workloads/ClElementwiseBinaryWorkload.hpp"
 #include "workloads/ClExpWorkload.hpp"
 #include "workloads/ClFillWorkload.hpp"
 #include "workloads/ClFloorFloatWorkload.hpp"
@@ -390,6 +391,15 @@ bool ClLayerSupport::IsLayerSupported(const LayerType& type,
                                                    infos[1],
                                                    infos[2],
                                                    nullptr);
+                case BinaryOperation::Power:
+                case BinaryOperation::SqDiff:
+                    FORWARD_WORKLOAD_VALIDATE_FUNC(ClElementwiseBinaryValidate,
+                                                   reasonIfUnsupported,
+                                                   infos[0],
+                                                   infos[1],
+                                                   infos[2],
+                                                   desc,
+                                                   nullptr);
                 case BinaryOperation::Sub:
                     FORWARD_WORKLOAD_VALIDATE_FUNC(ClSubtractionValidate,
                                                    reasonIfUnsupported,
diff --git a/src/backends/cl/ClWorkloadFactory.cpp b/src/backends/cl/ClWorkloadFactory.cpp
index 022867710c..493080f7af 100644
--- a/src/backends/cl/ClWorkloadFactory.cpp
+++ b/src/backends/cl/ClWorkloadFactory.cpp
@@ -459,6 +459,13 @@ std::unique_ptr<IWorkload> ClWorkloadFactory::CreateWorkload(LayerType type,
                                                                       info,
                                                                       m_CLCompileContext);
                 }
+                case BinaryOperation::Power:
+                case BinaryOperation::SqDiff:
+                {
+                    return std::make_unique<ClElementwiseBinaryWorkload>(*elementwiseBinaryQueueDescriptor,
+                                                                         info,
+                                                                         m_CLCompileContext);
+                }
                 case BinaryOperation::Sub:
                 {
                     SubtractionQueueDescriptor subtractionQueueDescriptor;
diff --git a/src/backends/cl/backend.mk b/src/backends/cl/backend.mk
index 1f97ae7cc8..03f1a9540d 100644
--- a/src/backends/cl/backend.mk
+++ b/src/backends/cl/backend.mk
@@ -1,5 +1,5 @@
 #
-# Copyright © 2017 ARM Ltd and Contributors. All rights reserved.
+# Copyright © 2017,2023 ARM Ltd and Contributors. All rights reserved.
 # SPDX-License-Identifier: MIT
 #
 
@@ -46,6 +46,7 @@ BACKEND_SOURCES := \
         workloads/ClDepthwiseConvolutionWorkload.cpp \
         workloads/ClDequantizeWorkload.cpp \
         workloads/ClDivisionWorkload.cpp \
+        workloads/ClElementwiseBinaryWorkload.cpp \
         workloads/ClExpWorkload.cpp \
         workloads/ClFillWorkload.cpp \
         workloads/ClFloorFloatWorkload.cpp \
diff --git a/src/backends/cl/test/ClEndToEndTests.cpp b/src/backends/cl/test/ClEndToEndTests.cpp
index a6ddd97ecf..091526fd2b 100644
--- a/src/backends/cl/test/ClEndToEndTests.cpp
+++ b/src/backends/cl/test/ClEndToEndTests.cpp
@@ -13,6 +13,7 @@
 #include <backendsCommon/test/ConcatEndToEndTestImpl.hpp>
 #include <backendsCommon/test/DepthToSpaceEndToEndTestImpl.hpp>
 #include <backendsCommon/test/DequantizeEndToEndTestImpl.hpp>
+#include <backendsCommon/test/ElementwiseBinaryEndToEndTestImpl.hpp>
 #include <backendsCommon/test/ElementwiseUnaryEndToEndTestImpl.hpp>
 #include <backendsCommon/test/FillEndToEndTestImpl.hpp>
 #include <backendsCommon/test/InstanceNormalizationEndToEndTestImpl.hpp>
@@ -46,6 +47,7 @@ TEST_CASE("ClRsqrtEndToEndTestFloat32")
                                                              UnaryOperation::Rsqrt);
 }
 
+// ElementwiseBinary
 // Addition
 TEST_CASE("ClAdditionEndToEndFloat32Test")
 {
@@ -57,6 +59,23 @@ TEST_CASE("ClAdditionEndToEndUint8Test")
     AdditionEndToEnd<armnn::DataType::QAsymmU8>(clDefaultBackends);
 }
 
+// Power
+TEST_CASE("RefPowerEndToEndTestFloat32")
+{
+    ElementwiseBinarySimpleEndToEnd<armnn::DataType::Float32>(clDefaultBackends, BinaryOperation::Power);
+}
+
+// SqDiff
+TEST_CASE("RefSquaredDifferenceEndToEndTestFloat32")
+{
+    ElementwiseBinarySimpleEndToEnd<armnn::DataType::Float32>(clDefaultBackends, BinaryOperation::SqDiff);
+}
+
+TEST_CASE("RefSquaredDifferenceEndToEndTestUint8")
+{
+    ElementwiseBinarySimpleEndToEnd<armnn::DataType::QAsymmU8>(clDefaultBackends, BinaryOperation::SqDiff);
+}
+
 // Batch Mat Mul
 TEST_CASE("ClBatchMatMulEndToEndFloat32Test")
 {
diff --git a/src/backends/cl/test/ClLayerTests.cpp b/src/backends/cl/test/ClLayerTests.cpp
index a84ecc9f9f..03a4d6fc49 100644
--- a/src/backends/cl/test/ClLayerTests.cpp
+++ b/src/backends/cl/test/ClLayerTests.cpp
@@ -913,6 +913,48 @@ ARMNN_AUTO_TEST_FIXTURE_WITH_THF(MultiplicationBroadcast1DVectorUint8,
                                  MultiplicationBroadcast1DVectorUint8Test)
 ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Multiplication5d, ClContextControlFixture, Multiplication5dTest)
 
+// SquaredDifference
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SimpleSquaredDifference, ClContextControlFixture, SquaredDifferenceTest)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcast1Element,
+                                 ClContextControlFixture,
+                                 SquaredDiffBroadcast1ElementTest)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcast, ClContextControlFixture, SquaredDiffBroadcastTest)
+
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDifferenceFloat16, ClContextControlFixture, SquaredDifferenceFloat16Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcast1ElementFloat16,
+                                 ClContextControlFixture,
+                                 SquaredDiffBroadcast1ElementFloat16Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcastFloat16, ClContextControlFixture, SquaredDiffBroadcastFloat16Test)
+
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDifferenceUint8, ClContextControlFixture, SquaredDifferenceUint8Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcastUint8, ClContextControlFixture, SquaredDiffBroadcastUint8Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcast1ElementUint8,
+                                 ClContextControlFixture,
+                                 SquaredDiffBroadcast1ElementUint8Test)
+
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDifferenceInt16, ClContextControlFixture, SquaredDifferenceInt16Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcastInt16, ClContextControlFixture, SquaredDiffBroadcastInt16Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcast1ElementInt16,
+                                 ClContextControlFixture,
+                                 SquaredDiffBroadcast1ElementInt16Test)
+
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDifferenceInt32, ClContextControlFixture, SquaredDifferenceInt32Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcastInt32, ClContextControlFixture, SquaredDiffBroadcastInt32Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SquaredDiffBroadcast1ElementInt32,
+                                 ClContextControlFixture,
+                                 SquaredDiffBroadcast1ElementInt32Test)
+
+// Power
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(SimplePower, ClContextControlFixture, PowerTest)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(PowerBroadcast1Element, ClContextControlFixture, PowerBroadcast1ElementTest)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(PowerBroadcast, ClContextControlFixture, PowerBroadcastTest)
+
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(PowerFloat16, ClContextControlFixture, PowerFloat16Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(PowerBroadcast1ElementFloat16,
+                                 ClContextControlFixture,
+                                 PowerBroadcast1ElementFloat16Test)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(PowerBroadcastFloat16, ClContextControlFixture, PowerBroadcastFloat16Test)
+
 // Batch Norm
 ARMNN_AUTO_TEST_FIXTURE_WITH_THF(BatchNormFloat32, ClContextControlFixture, BatchNormFloat32Test)
 ARMNN_AUTO_TEST_FIXTURE_WITH_THF(BatchNormFloat32Nhwc, ClContextControlFixture, BatchNormFloat32NhwcTest)
diff --git a/src/backends/cl/workloads/CMakeLists.txt b/src/backends/cl/workloads/CMakeLists.txt
index 8616dec078..030d71988f 100644
--- a/src/backends/cl/workloads/CMakeLists.txt
+++ b/src/backends/cl/workloads/CMakeLists.txt
@@ -1,5 +1,5 @@
 #
-# Copyright © 2017 Arm Ltd and Contributors. All rights reserved.
+# Copyright © 2017, 2023 Arm Ltd and Contributors. All rights reserved.
 # SPDX-License-Identifier: MIT
 #
 
@@ -44,6 +44,8 @@ list(APPEND armnnClBackendWorkloads_sources
     ClDequantizeWorkload.hpp
     ClDivisionWorkload.cpp
     ClDivisionWorkload.hpp
+    ClElementwiseBinaryWorkload.cpp
+    ClElementwiseBinaryWorkload.hpp
     ClExpWorkload.cpp
     ClExpWorkload.hpp
     ClFillWorkload.cpp
diff --git a/src/backends/cl/workloads/ClElementwiseBinaryWorkload.cpp b/src/backends/cl/workloads/ClElementwiseBinaryWorkload.cpp
new file mode 100644
index 0000000000..df30feb52a
--- /dev/null
+++ b/src/backends/cl/workloads/ClElementwiseBinaryWorkload.cpp
@@ -0,0 +1,94 @@
+//
+// Copyright © 2023 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "ClElementwiseBinaryWorkload.hpp"
+
+#include <cl/ClTensorHandle.hpp>
+#include <armnn/backends/TensorHandle.hpp>
+#include <aclCommon/ArmComputeTensorUtils.hpp>
+#include <aclCommon/ArmComputeUtils.hpp>
+
+#include "ClWorkloadUtils.hpp"
+
+namespace armnn
+{
+using namespace armcomputetensorutils;
+
+ClElementwiseBinaryWorkload::ClElementwiseBinaryWorkload(const ElementwiseBinaryQueueDescriptor& descriptor,
+                                                         const WorkloadInfo& info,
+                                                         const arm_compute::CLCompileContext& clCompileContext)
+    : ClBaseWorkload<ElementwiseBinaryQueueDescriptor>(descriptor, info)
+{
+    this->m_Data.ValidateInputsOutputs("ClElementwiseBinaryWorkload", 2, 1);
+
+    arm_compute::ICLTensor &input0 = static_cast<IClTensorHandle *>(this->m_Data.m_Inputs[0])->GetTensor();
+    arm_compute::ICLTensor &input1 = static_cast<IClTensorHandle *>(this->m_Data.m_Inputs[1])->GetTensor();
+    arm_compute::ICLTensor &output = static_cast<IClTensorHandle *>(this->m_Data.m_Outputs[0])->GetTensor();
+
+    const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
+    {
+        ARMNN_SCOPED_PROFILING_EVENT(Compute::Undefined, "ClElementwiseBinaryWorkload_configure");
+
+        switch (descriptor.m_Parameters.m_Operation)
+        {
+            case armnn::BinaryOperation::Power:
+            {
+                auto powerLayer = std::make_unique<arm_compute::CLElementwisePower>();
+                powerLayer->configure(clCompileContext, &input0, &input1, &output, activationInfo);
+                m_ElementwiseBinaryLayer.reset(powerLayer.release());
+                break;
+            }
+            case armnn::BinaryOperation::SqDiff:
+            {
+                auto SqDiffLayer = std::make_unique<arm_compute::CLElementwiseSquaredDiff>();
+                SqDiffLayer->configure(clCompileContext, &input0, &input1, &output, activationInfo);
+                m_ElementwiseBinaryLayer.reset(SqDiffLayer.release());
+                break;
+            }
+            default:
+                throw InvalidArgumentException("Unknown binary operator", CHECK_LOCATION());
+        }
+    }
+}
+void ClElementwiseBinaryWorkload::Execute() const
+{
+    if (m_ElementwiseBinaryLayer)
+    {
+        ARMNN_SCOPED_PROFILING_EVENT_CL_GUID("ClElementwiseBinaryWorkload_Execute", this->GetGuid());
+        m_ElementwiseBinaryLayer->run();
+    }
+}
+
+arm_compute::Status ClElementwiseBinaryValidate(const TensorInfo& input0,
+                                                const TensorInfo& input1,
+                                                const TensorInfo& output,
+                                                const ElementwiseBinaryDescriptor& descriptor,
+                                                const ActivationDescriptor* activationDescriptor)
+{
+    const arm_compute::TensorInfo aclInput0Info = BuildArmComputeTensorInfo(input0);
+    const arm_compute::TensorInfo aclInput1Info = BuildArmComputeTensorInfo(input1);
+    const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
+
+    const arm_compute::ActivationLayerInfo activationInfo = ConvertActivationDescriptorToAclActivationLayerInfo(
+            activationDescriptor);
+
+    switch (descriptor.m_Operation)
+    {
+        case armnn::BinaryOperation::Power:
+            return arm_compute::CLElementwisePower::validate(&aclInput0Info,
+                                                             &aclInput1Info,
+                                                             &aclOutputInfo,
+                                                             activationInfo);
+        case armnn::BinaryOperation::SqDiff:
+            return arm_compute::CLElementwiseSquaredDiff::validate(&aclInput0Info,
+                                                                   &aclInput1Info,
+                                                                   &aclOutputInfo,
+                                                                   activationInfo);
+        default:
+            throw InvalidArgumentException("Unknown binary operator", CHECK_LOCATION());
+    }
+}
+
+} //namespace armnn
\ No newline at end of file
diff --git a/src/backends/cl/workloads/ClElementwiseBinaryWorkload.hpp b/src/backends/cl/workloads/ClElementwiseBinaryWorkload.hpp
new file mode 100644
index 0000000000..addd6e6085
--- /dev/null
+++ b/src/backends/cl/workloads/ClElementwiseBinaryWorkload.hpp
@@ -0,0 +1,34 @@
+//
+// Copyright © 2023 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "ClBaseWorkload.hpp"
+
+#include <arm_compute/runtime/CL/functions/CLElementwiseOperations.h>
+
+namespace armnn
+{
+
+class ClElementwiseBinaryWorkload : public ClBaseWorkload<ElementwiseBinaryQueueDescriptor>
+{
+public:
+    ClElementwiseBinaryWorkload(const ElementwiseBinaryQueueDescriptor& descriptor,
+                                const WorkloadInfo& info,
+                                const arm_compute::CLCompileContext& clCompileContext);
+
+    void Execute() const override;
+
+private:
+    std::unique_ptr<arm_compute::IFunction> m_ElementwiseBinaryLayer;
+
+};
+
+arm_compute::Status ClElementwiseBinaryValidate(const TensorInfo& input0,
+                                                const TensorInfo& input1,
+                                                const TensorInfo& output,
+                                                const ElementwiseBinaryDescriptor& descriptor,
+                                                const ActivationDescriptor* activationDescriptor = nullptr);
+} //namespace armnn
\ No newline at end of file
diff --git a/src/backends/cl/workloads/ClWorkloads.hpp b/src/backends/cl/workloads/ClWorkloads.hpp
index 44f3798d7d..d862aab949 100644
--- a/src/backends/cl/workloads/ClWorkloads.hpp
+++ b/src/backends/cl/workloads/ClWorkloads.hpp
@@ -1,5 +1,5 @@
 //
-// Copyright © 2017 Arm Ltd and Contributors. All rights reserved.
+// Copyright © 2017,2023 Arm Ltd and Contributors. All rights reserved.
 // SPDX-License-Identifier: MIT
 //
 
@@ -21,6 +21,7 @@
 #include "ClDepthwiseConvolutionWorkload.hpp"
 #include "ClDequantizeWorkload.hpp"
 #include "ClDivisionWorkload.hpp"
+#include "ClElementwiseBinaryWorkload.hpp"
 #include "ClExpWorkload.hpp"
 #include "ClFillWorkload.hpp"
 #include "ClFloorFloatWorkload.hpp"