IVGCVSW-2443 Add Cl support for Greater operator

Change-Id: I7eb3b2e712dbaa5c13f6cf3a1da63df4d2bd4de5

11 files changed, 176 insertions, 2 deletions

diff --git a/src/backends/backendsCommon/WorkloadFactory.cpp b/src/backends/backendsCommon/WorkloadFactory.cpp
index 6cca174141..72d0b19d04 100644
--- a/src/backends/backendsCommon/WorkloadFactory.cpp
+++ b/src/backends/backendsCommon/WorkloadFactory.cpp
@@ -729,7 +729,10 @@ bool IWorkloadFactory::IsLayerSupported(const BackendId& backendId,
             const TensorInfo& input0 = layer.GetInputSlot(0).GetConnection()->GetTensorInfo();
             const TensorInfo& input1 = layer.GetInputSlot(1).GetConnection()->GetTensorInfo();
             const TensorInfo& output = layer.GetOutputSlot(0).GetTensorInfo();
-            result = layerSupportObject->IsGreaterSupported(input0, input1, output, reason);
+            result = layerSupportObject->IsGreaterSupported(OverrideDataType(input0, dataType),
+                                                            OverrideDataType(input1, dataType),
+                                                            OverrideDataType(output, DataType::Boolean),
+                                                            reason);
             break;
         }
         default:
diff --git a/src/backends/cl/ClLayerSupport.cpp b/src/backends/cl/ClLayerSupport.cpp
index 410cb04288..a7c1358ae7 100644
--- a/src/backends/cl/ClLayerSupport.cpp
+++ b/src/backends/cl/ClLayerSupport.cpp
@@ -25,6 +25,7 @@
 #include "workloads/ClDepthwiseConvolutionWorkload.hpp"
 #include "workloads/ClDivisionFloatWorkload.hpp"
 #include "workloads/ClFullyConnectedWorkload.hpp"
+#include "workloads/ClGreaterWorkload.hpp"
 #include "workloads/ClL2NormalizationFloatWorkload.hpp"
 #include "workloads/ClLstmFloatWorkload.hpp"
 #include "workloads/ClMaximumWorkload.hpp"
@@ -288,6 +289,18 @@ bool ClLayerSupport::IsFullyConnectedSupported(const TensorInfo& input,
                                    descriptor);
 }
 
+bool ClLayerSupport::IsGreaterSupported(const TensorInfo& input0,
+                                        const TensorInfo& input1,
+                                        const TensorInfo& output,
+                                        Optional<std::string&> reasonIfUnsupported) const
+{
+    FORWARD_WORKLOAD_VALIDATE_FUNC(ClGreaterWorkloadValidate,
+                                   reasonIfUnsupported,
+                                   input0,
+                                   input1,
+                                   output);
+}
+
 bool ClLayerSupport::IsInputSupported(const TensorInfo& input,
                                       Optional<std::string&> reasonIfUnsupported) const
 {
diff --git a/src/backends/cl/ClLayerSupport.hpp b/src/backends/cl/ClLayerSupport.hpp
index 07e24335dc..6393a11e84 100644
--- a/src/backends/cl/ClLayerSupport.hpp
+++ b/src/backends/cl/ClLayerSupport.hpp
@@ -77,6 +77,11 @@ public:
                                    const FullyConnectedDescriptor& descriptor,
                                    Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const override;
 
+    bool IsGreaterSupported(const TensorInfo& input0,
+                            const TensorInfo& input1,
+                            const TensorInfo& ouput,
+                            Optional<std::string&> reasonIfUnsupported) const override;
+
     bool IsInputSupported(const TensorInfo& input,
                           Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const override;
 
diff --git a/src/backends/cl/ClWorkloadFactory.cpp b/src/backends/cl/ClWorkloadFactory.cpp
index 6cd9c6dfa5..d41a7e5997 100644
--- a/src/backends/cl/ClWorkloadFactory.cpp
+++ b/src/backends/cl/ClWorkloadFactory.cpp
@@ -348,7 +348,7 @@ std::unique_ptr<IWorkload> ClWorkloadFactory::CreateMinimum(const MinimumQueueDe
 std::unique_ptr<IWorkload> ClWorkloadFactory::CreateGreater(const GreaterQueueDescriptor& descriptor,
                                                             const WorkloadInfo& info) const
 {
-    return MakeWorkload<NullWorkload, NullWorkload>(descriptor, info);
+    return MakeWorkload<ClGreaterFloat32Workload, ClGreaterUint8Workload>(descriptor, info);
 }
 
 std::unique_ptr<IWorkload> ClWorkloadFactory::CreateDebug(const DebugQueueDescriptor& descriptor,
diff --git a/src/backends/cl/backend.mk b/src/backends/cl/backend.mk
index cb3163a561..a6d6e002c4 100644
--- a/src/backends/cl/backend.mk
+++ b/src/backends/cl/backend.mk
@@ -26,6 +26,7 @@ BACKEND_SOURCES := \
         workloads/ClDivisionFloatWorkload.cpp \
         workloads/ClFloorFloatWorkload.cpp \
         workloads/ClFullyConnectedWorkload.cpp \
+        workloads/ClGreaterWorkload.cpp \
         workloads/ClL2NormalizationFloatWorkload.cpp \
         workloads/ClLstmFloatWorkload.cpp \
         workloads/ClMaximumWorkload.cpp \
diff --git a/src/backends/cl/test/ClEndToEndTests.cpp b/src/backends/cl/test/ClEndToEndTests.cpp
index ba94353049..bc0bad2fa6 100644
--- a/src/backends/cl/test/ClEndToEndTests.cpp
+++ b/src/backends/cl/test/ClEndToEndTests.cpp
@@ -5,6 +5,7 @@
 
 #include <backendsCommon/test/EndToEndTestImpl.hpp>
 #include <backendsCommon/test/MergerTestImpl.hpp>
+#include <backendsCommon/test/ArithmeticTestImpl.hpp>
 
 #include <boost/test/unit_test.hpp>
 
@@ -47,4 +48,45 @@ BOOST_AUTO_TEST_CASE(ClMergerEndToEndDim3Uint8Test)
     MergerDim3EndToEnd<armnn::DataType::QuantisedAsymm8>(defaultBackends);
 }
 
+BOOST_AUTO_TEST_CASE(ClGreaterSimpleEndToEndTest)
+{
+    const std::vector<uint8_t> expectedOutput({ 0, 0, 0, 0,  1, 1, 1, 1,
+                                                0, 0, 0, 0,  0, 0, 0, 0 });
+
+    ArithmeticSimpleEndToEnd<armnn::DataType::Float32, armnn::DataType::Boolean>(defaultBackends,
+                                                                                 LayerType::Greater,
+                                                                                 expectedOutput);
+}
+
+BOOST_AUTO_TEST_CASE(ClGreaterSimpleEndToEndUint8Test)
+{
+    const std::vector<uint8_t> expectedOutput({ 0, 0, 0, 0,  1, 1, 1, 1,
+                                                0, 0, 0, 0,  0, 0, 0, 0 });
+
+    ArithmeticSimpleEndToEnd<armnn::DataType::QuantisedAsymm8, armnn::DataType::Boolean>(defaultBackends,
+                                                                                         LayerType::Greater,
+                                                                                         expectedOutput);
+}
+
+BOOST_AUTO_TEST_CASE(ClGreaterBroadcastEndToEndTest)
+{
+    const std::vector<uint8_t> expectedOutput({ 0, 1, 0, 0, 0, 1,
+                                                1, 1, 1, 1, 1, 1 });
+
+    ArithmeticBroadcastEndToEnd<armnn::DataType::Float32, armnn::DataType::Boolean>(defaultBackends,
+                                                                                    LayerType::Greater,
+                                                                                    expectedOutput);
+}
+
+BOOST_AUTO_TEST_CASE(ClGreaterBroadcastEndToEndUint8Test)
+{
+    const std::vector<uint8_t> expectedOutput({ 0, 1, 0, 0, 0, 1,
+                                                1, 1, 1, 1, 1, 1 });
+
+    ArithmeticBroadcastEndToEnd<armnn::DataType::QuantisedAsymm8, armnn::DataType::Boolean>(defaultBackends,
+                                                                                            LayerType::Greater,
+                                                                                            expectedOutput);
+}
+
+
 BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/backends/cl/test/ClLayerTests.cpp b/src/backends/cl/test/ClLayerTests.cpp
index 2fe03abb3c..641e964470 100644
--- a/src/backends/cl/test/ClLayerTests.cpp
+++ b/src/backends/cl/test/ClLayerTests.cpp
@@ -366,6 +366,14 @@ ARMNN_AUTO_TEST_CASE(MinimumBroadcast1Element1, MinimumBroadcast1ElementTest1)
 ARMNN_AUTO_TEST_CASE(MinimumBroadcast1Element2, MinimumBroadcast1ElementTest2)
 ARMNN_AUTO_TEST_CASE(MinimumBroadcast1DVectorUint8, MinimumBroadcast1DVectorUint8Test)
 
+// Greater
+ARMNN_AUTO_TEST_CASE(SimpleGreater, GreaterSimpleTest)
+ARMNN_AUTO_TEST_CASE(GreaterBroadcast1Element, GreaterBroadcast1ElementTest)
+ARMNN_AUTO_TEST_CASE(GreaterBroadcast1DVector, GreaterBroadcast1DVectorTest)
+ARMNN_AUTO_TEST_CASE(GreaterUint8, GreaterUint8Test)
+ARMNN_AUTO_TEST_CASE(GreaterBroadcast1ElementUint8, GreaterBroadcast1ElementUint8Test)
+ARMNN_AUTO_TEST_CASE(GreaterBroadcast1DVectorUint8, GreaterBroadcast1DVectorUint8Test)
+
 // Space To Batch Nd
 ARMNN_AUTO_TEST_CASE(SpaceToBatchNdSimpleFloat32, SpaceToBatchNdSimpleFloat32Test)
 ARMNN_AUTO_TEST_CASE(SpaceToBatchNdMultiChannelsFloat32, SpaceToBatchNdMultiChannelsFloat32Test)
diff --git a/src/backends/cl/workloads/CMakeLists.txt b/src/backends/cl/workloads/CMakeLists.txt
index 4dfce43679..d5dccb607c 100644
--- a/src/backends/cl/workloads/CMakeLists.txt
+++ b/src/backends/cl/workloads/CMakeLists.txt
@@ -28,6 +28,8 @@ list(APPEND armnnClBackendWorkloads_sources
     ClFloorFloatWorkload.hpp
     ClFullyConnectedWorkload.cpp
     ClFullyConnectedWorkload.hpp
+    ClGreaterWorkload.cpp
+    ClGreaterWorkload.hpp
     ClL2NormalizationFloatWorkload.cpp
     ClL2NormalizationFloatWorkload.hpp
     ClLstmFloatWorkload.cpp
diff --git a/src/backends/cl/workloads/ClGreaterWorkload.cpp b/src/backends/cl/workloads/ClGreaterWorkload.cpp
new file mode 100644
index 0000000000..435ead853f
--- /dev/null
+++ b/src/backends/cl/workloads/ClGreaterWorkload.cpp
@@ -0,0 +1,65 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "ClGreaterWorkload.hpp"
+
+#include "ClWorkloadUtils.hpp"
+
+#include <aclCommon/ArmComputeUtils.hpp>
+#include <aclCommon/ArmComputeTensorUtils.hpp>
+
+#include <backendsCommon/CpuTensorHandle.hpp>
+
+#include <cl/ClLayerSupport.hpp>
+#include <cl/ClTensorHandle.hpp>
+#include <cl/ClLayerSupport.hpp>
+
+namespace armnn
+{
+
+using namespace armcomputetensorutils;
+
+arm_compute::Status ClGreaterWorkloadValidate(const TensorInfo& input0,
+                                              const TensorInfo& input1,
+                                              const TensorInfo& output)
+{
+    const arm_compute::TensorInfo aclInput0Info = BuildArmComputeTensorInfo(input0);
+    const arm_compute::TensorInfo aclInput1Info = BuildArmComputeTensorInfo(input1);
+    const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
+
+    const arm_compute::Status aclStatus = arm_compute::CLComparison::validate(
+        &aclInput0Info,
+        &aclInput1Info,
+        &aclOutputInfo,
+        arm_compute::ComparisonOperation::Greater);
+
+    return aclStatus;
+}
+
+template<DataType T>
+ClGreaterWorkload<T>::ClGreaterWorkload(const GreaterQueueDescriptor& descriptor,
+                                        const WorkloadInfo& info)
+    : MultiTypedWorkload<GreaterQueueDescriptor, T, DataType::Boolean>(descriptor, info)
+{
+    m_Data.ValidateInputsOutputs("ClGreaterWorkload", 2, 1);
+
+    arm_compute::ICLTensor& input0 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
+    arm_compute::ICLTensor& input1 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
+    arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
+
+    m_GreaterLayer.configure(&input0, &input1, &output, arm_compute::ComparisonOperation::Greater);
+}
+
+template<DataType T>
+void ClGreaterWorkload<T>::Execute() const
+{
+    ARMNN_SCOPED_PROFILING_EVENT_CL("ClGreaterWorkload_Execute");
+    RunClFunction(m_GreaterLayer, CHECK_LOCATION());
+}
+
+template class ClGreaterWorkload<DataType::Float32>;
+template class ClGreaterWorkload<DataType::QuantisedAsymm8>;
+
+} //namespace armnn
diff --git a/src/backends/cl/workloads/ClGreaterWorkload.hpp b/src/backends/cl/workloads/ClGreaterWorkload.hpp
new file mode 100644
index 0000000000..69d017e6c6
--- /dev/null
+++ b/src/backends/cl/workloads/ClGreaterWorkload.hpp
@@ -0,0 +1,34 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <backendsCommon/Workload.hpp>
+
+#include <arm_compute/runtime/CL/CLFunctions.h>
+
+namespace armnn
+{
+
+arm_compute::Status ClGreaterWorkloadValidate(const TensorInfo& input0,
+                                              const TensorInfo& input1,
+                                              const TensorInfo& output);
+
+template<DataType T>
+class ClGreaterWorkload : public MultiTypedWorkload<GreaterQueueDescriptor, T, DataType::Boolean>
+{
+public:
+    ClGreaterWorkload(const GreaterQueueDescriptor& descriptor, const WorkloadInfo& info);
+    void Execute() const override;
+
+private:
+    using MultiTypedWorkload<GreaterQueueDescriptor, T, DataType::Boolean>::m_Data;
+    mutable arm_compute::CLComparison m_GreaterLayer;
+};
+
+using ClGreaterFloat32Workload = ClGreaterWorkload<DataType::Float32>;
+using ClGreaterUint8Workload = ClGreaterWorkload<DataType::QuantisedAsymm8>;
+
+} //namespace armnn
diff --git a/src/backends/cl/workloads/ClWorkloads.hpp b/src/backends/cl/workloads/ClWorkloads.hpp
index be223921d0..36f2b90609 100644
--- a/src/backends/cl/workloads/ClWorkloads.hpp
+++ b/src/backends/cl/workloads/ClWorkloads.hpp
@@ -14,6 +14,7 @@
 #include "ClDivisionFloatWorkload.hpp"
 #include "ClFloorFloatWorkload.hpp"
 #include "ClFullyConnectedWorkload.hpp"
+#include "ClGreaterWorkload.hpp"
 #include "ClL2NormalizationFloatWorkload.hpp"
 #include "ClLstmFloatWorkload.hpp"
 #include "ClMergerWorkload.hpp"