IVGCVSW-7835 Add ReverseV2 CL and Neon Workloads

 * Added ReverseV2 to CL and Neon backends
 * Added Cl and Neon ReverseV2 Layer unit tests

Signed-off-by: Tianle Cheng <tianle.cheng@arm.com>
Change-Id: I646275c629caf17dac1950b0cd7083f23f87f387

10 files changed, 143 insertions, 1 deletions

diff --git a/src/backends/neon/NeonLayerSupport.cpp b/src/backends/neon/NeonLayerSupport.cpp
index ef1d21835a..4be5b7cb4e 100644
--- a/src/backends/neon/NeonLayerSupport.cpp
+++ b/src/backends/neon/NeonLayerSupport.cpp
@@ -69,6 +69,7 @@
 #include "workloads/NeonReduceWorkload.hpp"
 #include "workloads/NeonReshapeWorkload.hpp"
 #include "workloads/NeonResizeWorkload.hpp"
+#include "workloads/NeonReverseV2Workload.hpp"
 #include "workloads/NeonRsqrtWorkload.hpp"
 #include "workloads/NeonSinWorkload.hpp"
 #include "workloads/NeonSliceWorkload.hpp"
@@ -582,6 +583,11 @@ bool IsLayerTypeSupported(const LayerType& type,
                                              infos[1],
                                              *(PolymorphicDowncast<const ReduceDescriptor*>(&descriptor)),
                                              reasonIfUnsupported);
+        case LayerType::ReverseV2:
+            return support.IsReverseV2Supported(infos[0],
+                                                infos[1],
+                                                infos[2],
+                                                reasonIfUnsupported);
         case LayerType::Shape:
             return support.IsShapeSupported(infos[0],
                                             infos[1],
@@ -1515,6 +1521,18 @@ bool NeonLayerSupport::IsResizeSupported(const TensorInfo& input,
                                    descriptor);
 }
 
+bool NeonLayerSupport::IsReverseV2Supported(const armnn::TensorInfo &input,
+                                            const armnn::TensorInfo &axis,
+                                            const armnn::TensorInfo &output,
+                                            Optional<std::string &> reasonIfUnsupported) const
+{
+    FORWARD_WORKLOAD_VALIDATE_FUNC(NeonReverseV2WorkloadValidate,
+                                   reasonIfUnsupported,
+                                   input,
+                                   axis,
+                                   output);
+}
+
 bool NeonLayerSupport::IsSliceSupported(const TensorInfo& input,
                                         const TensorInfo& output,
                                         const SliceDescriptor& descriptor,
diff --git a/src/backends/neon/NeonLayerSupport.hpp b/src/backends/neon/NeonLayerSupport.hpp
index 0295c2b3e2..8372df3e08 100644
--- a/src/backends/neon/NeonLayerSupport.hpp
+++ b/src/backends/neon/NeonLayerSupport.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
 //
 #pragma once
@@ -295,6 +295,11 @@ public:
                            const ResizeDescriptor& descriptor,
                            Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const;
 
+    bool IsReverseV2Supported(const TensorInfo& input,
+                              const TensorInfo& axis,
+                              const TensorInfo& output,
+                              Optional<std::string&> reasonIfUnsupported) const;
+
     bool IsSliceSupported(const TensorInfo& input,
                           const TensorInfo& output,
                           const SliceDescriptor& descriptor,
diff --git a/src/backends/neon/NeonWorkloadFactory.cpp b/src/backends/neon/NeonWorkloadFactory.cpp
index 4f131ac575..9b74f750f9 100644
--- a/src/backends/neon/NeonWorkloadFactory.cpp
+++ b/src/backends/neon/NeonWorkloadFactory.cpp
@@ -572,6 +572,11 @@ std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateWorkload(LayerType type,
             auto resizeQueueDescriptor = PolymorphicDowncast<const ResizeQueueDescriptor*>(&descriptor);
             return std::make_unique<NeonResizeWorkload>(*resizeQueueDescriptor, info);
         }
+        case LayerType::ReverseV2 :
+        {
+            auto reverseV2QueueDescriptor = PolymorphicDowncast<const ReverseV2QueueDescriptor*>(&descriptor);
+            return std::make_unique<NeonReverseV2Workload>(*reverseV2QueueDescriptor, info);
+        }
         case LayerType::Slice :
         {
             auto sliceQueueDescriptor = PolymorphicDowncast<const SliceQueueDescriptor*>(&descriptor);
diff --git a/src/backends/neon/backend.mk b/src/backends/neon/backend.mk
index 3961ed1e34..3ab07a9431 100644
--- a/src/backends/neon/backend.mk
+++ b/src/backends/neon/backend.mk
@@ -77,6 +77,7 @@ BACKEND_SOURCES := \
         workloads/NeonReduceWorkload.cpp \
         workloads/NeonReshapeWorkload.cpp \
         workloads/NeonResizeWorkload.cpp \
+        workloads/NeonReverseV2Workload.cpp \
         workloads/NeonRsqrtWorkload.cpp \
         workloads/NeonSinWorkload.cpp \
         workloads/NeonSliceWorkload.cpp \
diff --git a/src/backends/neon/test/NeonEndToEndTests.cpp b/src/backends/neon/test/NeonEndToEndTests.cpp
index e582fadf11..1e2636bf6f 100644
--- a/src/backends/neon/test/NeonEndToEndTests.cpp
+++ b/src/backends/neon/test/NeonEndToEndTests.cpp
@@ -23,6 +23,7 @@
 #include <backendsCommon/test/QuantizedLstmEndToEndTestImpl.hpp>
 #include <backendsCommon/test/ReduceEndToEndTestImpl.hpp>
 #include <backendsCommon/test/ReshapeEndToEndTestImpl.hpp>
+#include <backendsCommon/test/ReverseV2EndToEndTestImpl.hpp>
 #include <backendsCommon/test/SliceEndToEndTestImpl.hpp>
 #include <backendsCommon/test/SpaceToDepthEndToEndTestImpl.hpp>
 #include <backendsCommon/test/SplitterEndToEndTestImpl.hpp>
@@ -671,6 +672,12 @@ TEST_CASE("NeonReshapeEndToEndTestFloat16")
     ReshapeEndToEndFloat16<armnn::DataType::Float16>(neonDefaultBackends);
 }
 
+// ReverseV2
+TEST_CASE("NeonReverseV2EndToEndTest")
+{
+    ReverseV2EndToEnd<armnn::DataType::Float32>(neonDefaultBackends);
+}
+
 // Slice
 TEST_CASE("NeonSliceEndtoEndTestFloat32")
 {
diff --git a/src/backends/neon/test/NeonLayerTests.cpp b/src/backends/neon/test/NeonLayerTests.cpp
index a938ceb9c3..9db6d2c2e6 100644
--- a/src/backends/neon/test/NeonLayerTests.cpp
+++ b/src/backends/neon/test/NeonLayerTests.cpp
@@ -1352,6 +1352,36 @@ ARMNN_AUTO_TEST_CASE_WITH_THF(AlignCornersResizeNearestNeighbourUint8Nhwc,
                               AlignCornersResizeNearestNeighbourTest<DataType::QAsymmU8>,
                               DataLayout::NHWC)
 
+// ReverseV2
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2SimpleEmptyAxisFloat32, ReverseV2SimpleTestEmptyAxis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple1DimFloat32, ReverseV2SimpleTest1Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple2Dim1AxisFloat32, ReverseV2SimpleTest2Dim1Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple2Dim2AxisFloat32, ReverseV2SimpleTest2Dim2Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple3Dim1AxisFloat32, ReverseV2SimpleTest3Dim1Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple3Dim2AxisFloat32, ReverseV2SimpleTest3Dim2Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple3Dim3AxisFloat32, ReverseV2SimpleTest3Dim3Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple4Dim1AxisFloat32, ReverseV2SimpleTest4Dim1Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple4Dim2AxisFloat32, ReverseV2SimpleTest4Dim2Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple4Dim3AxisFloat32, ReverseV2SimpleTest4Dim3Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple4Dim4AxisFloat32, ReverseV2SimpleTest4Dim4Axis<DataType::Float32>)
+
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2EvenRowOddCol2DimFloat32, ReverseV2EvenRowOddColTest2Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2EvenRowOddCol3DimFloat32, ReverseV2EvenRowOddColTest3Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2EvenRowEvenCol2DimFloat32, ReverseV2EvenRowEvenColTest2Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2EvenRowEvenCol3DimFloat32, ReverseV2EvenRowEvenColTest3Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2OddRowOddCol2DimFloat32, ReverseV2OddRowOddColTest2Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2OddRowOddCol3DimFloat32, ReverseV2OddRowOddColTest3Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2OddRowEvenCol2DimFloat32, ReverseV2OddRowEvenColTest2Dim<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2OddRowEvenCol3DimFloat32, ReverseV2OddRowEvenColTest3Dim<DataType::Float32>)
+
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2NegAxis2Dim1AxisFloat32, ReverseV2NegAxisTest2Dim1Axis<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2NegAxis3Dim2AxisFloat32, ReverseV2NegAxisTest3Dim2Axis<DataType::Float32>)
+
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple2Dim2AxisFloat16, ReverseV2SimpleTest2Dim2Axis<DataType::Float16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple2Dim2AxisQAsymmS8, ReverseV2SimpleTest2Dim2Axis<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple2Dim2AxisQAsymmU8, ReverseV2SimpleTest2Dim2Axis<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(ReverseV2Simple2Dim2AxisQSymmS16, ReverseV2SimpleTest2Dim2Axis<DataType::QSymmS16>)
+
 // Slice
 ARMNN_AUTO_TEST_CASE(Slice4dFloat32, Slice4dFloat32Test)
 ARMNN_AUTO_TEST_CASE(Slice3dFloat32, Slice3dFloat32Test)
diff --git a/src/backends/neon/workloads/CMakeLists.txt b/src/backends/neon/workloads/CMakeLists.txt
index f4438e4baa..357fdcdf91 100644
--- a/src/backends/neon/workloads/CMakeLists.txt
+++ b/src/backends/neon/workloads/CMakeLists.txt
@@ -113,6 +113,8 @@ list(APPEND armnnNeonBackendWorkloads_sources
     NeonReshapeWorkload.hpp
     NeonResizeWorkload.cpp
     NeonResizeWorkload.hpp
+    NeonReverseV2Workload.cpp
+    NeonReverseV2Workload.hpp
     NeonRsqrtWorkload.cpp
     NeonRsqrtWorkload.hpp
     NeonSinWorkload.cpp
diff --git a/src/backends/neon/workloads/NeonReverseV2Workload.cpp b/src/backends/neon/workloads/NeonReverseV2Workload.cpp
new file mode 100644
index 0000000000..c7d9fa1245
--- /dev/null
+++ b/src/backends/neon/workloads/NeonReverseV2Workload.cpp
@@ -0,0 +1,44 @@
+//
+// Copyright © 2023 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "NeonReverseV2Workload.hpp"
+#include "NeonWorkloadUtils.hpp"
+#include <armnn/utility/PolymorphicDowncast.hpp>
+#include <aclCommon/ArmComputeUtils.hpp>
+#include <backendsCommon/WorkloadUtils.hpp>
+
+namespace armnn
+{
+arm_compute::Status NeonReverseV2WorkloadValidate(const TensorInfo& input,
+                                                  const TensorInfo& axis,
+                                                  const TensorInfo& output)
+{
+    const arm_compute::TensorInfo aclInput = BuildArmComputeTensorInfo(input);
+    const arm_compute::TensorInfo aclAxis = BuildArmComputeTensorInfo(axis);
+    const arm_compute::TensorInfo aclOutput = BuildArmComputeTensorInfo(output);
+
+    return arm_compute::NEReverse::validate(&aclInput, &aclOutput, &aclAxis, true);
+}
+
+NeonReverseV2Workload::NeonReverseV2Workload(const ReverseV2QueueDescriptor& descriptor,
+                                             const WorkloadInfo& info)
+        : BaseWorkload<ReverseV2QueueDescriptor>(descriptor, info)
+{
+    m_Data.ValidateInputsOutputs("NeonReverseV2Workload", 2, 1);
+
+    arm_compute::ITensor& input = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
+    arm_compute::ITensor& axis = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
+    arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
+
+    m_Layer.configure(&input, &output, &axis, true);
+}
+
+void NeonReverseV2Workload::Execute() const
+{
+    ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID("NeonReverseV2Workload_Execute");
+    m_Layer.run();
+}
+
+} // namespace armnn
\ No newline at end of file
diff --git a/src/backends/neon/workloads/NeonReverseV2Workload.hpp b/src/backends/neon/workloads/NeonReverseV2Workload.hpp
new file mode 100644
index 0000000000..fa6a71d392
--- /dev/null
+++ b/src/backends/neon/workloads/NeonReverseV2Workload.hpp
@@ -0,0 +1,29 @@
+//
+// Copyright © 2023 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "NeonBaseWorkload.hpp"
+
+#include <arm_compute/runtime/NEON/functions/NEReverse.h>
+#include <arm_compute/runtime/Tensor.h>
+
+namespace armnn
+{
+arm_compute::Status NeonReverseV2WorkloadValidate(const TensorInfo& input,
+                                                  const TensorInfo& axis,
+                                                  const TensorInfo& output);
+
+class NeonReverseV2Workload : public BaseWorkload<ReverseV2QueueDescriptor>
+{
+public:
+    NeonReverseV2Workload(const ReverseV2QueueDescriptor& descriptor, const WorkloadInfo& info);
+    void Execute() const override;
+
+private:
+    mutable arm_compute::NEReverse m_Layer;
+};
+
+} // namespace armnn
\ No newline at end of file
diff --git a/src/backends/neon/workloads/NeonWorkloads.hpp b/src/backends/neon/workloads/NeonWorkloads.hpp
index 615e5d87c8..c14b9b44af 100644
--- a/src/backends/neon/workloads/NeonWorkloads.hpp
+++ b/src/backends/neon/workloads/NeonWorkloads.hpp
@@ -59,6 +59,7 @@
 #include "NeonReduceWorkload.hpp"
 #include "NeonReshapeWorkload.hpp"
 #include "NeonResizeWorkload.hpp"
+#include "NeonReverseV2Workload.hpp"
 #include "NeonRsqrtWorkload.hpp"
 #include "NeonSinWorkload.hpp"
 #include "NeonSliceWorkload.hpp"