4 files changed, 108 insertions, 0 deletions
diff --git a/src/backends/neon/workloads/CMakeLists.txt b/src/backends/neon/workloads/CMakeLists.txt
index 7bde80859e..dea0228377 100644
--- a/src/backends/neon/workloads/CMakeLists.txt
+++ b/src/backends/neon/workloads/CMakeLists.txt
@@ -68,6 +68,8 @@ list(APPEND armnnNeonBackendWorkloads_sources
     NeonSpaceToDepthWorkload.hpp
     NeonSplitterWorkload.cpp
     NeonSplitterWorkload.hpp
+    NeonStackWorkload.cpp
+    NeonStackWorkload.hpp
     NeonSubtractionWorkload.cpp
     NeonSubtractionWorkload.hpp
     NeonWorkloads.hpp
diff --git a/src/backends/neon/workloads/NeonStackWorkload.cpp b/src/backends/neon/workloads/NeonStackWorkload.cpp
new file mode 100644
index 0000000000..b21494397d
--- /dev/null
+++ b/src/backends/neon/workloads/NeonStackWorkload.cpp
@@ -0,0 +1,76 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+#include "NeonStackWorkload.hpp"
+#include "NeonWorkloadUtils.hpp"
+
+#include <aclCommon/ArmComputeTensorUtils.hpp>
+#include <backendsCommon/CpuTensorHandle.hpp>
+#include <neon/NeonTensorHandle.hpp>
+
+#include <boost/numeric/conversion/cast.hpp>
+#include <boost/polymorphic_pointer_cast.hpp>
+
+namespace armnn
+{
+using namespace armcomputetensorutils;
+
+namespace
+{
+int CalcAxis(const unsigned int axis, const unsigned int inputDimensions)
+{
+    const int intAxis = boost::numeric_cast<int>(axis);
+    return boost::numeric_cast<int>(inputDimensions) - intAxis;
+}
+} //namespace
+
+arm_compute::Status NeonStackWorkloadValidate(const std::vector<const TensorInfo*>& inputs,
+                                            const TensorInfo& output,
+                                            const StackDescriptor& descriptor)
+{
+    std::vector<arm_compute::TensorInfo> aclInputs;
+    for (const TensorInfo* input : inputs)
+    {
+        arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(*input, armnn::DataLayout::NCHW);
+        aclInputs.emplace_back(aclInputInfo);
+    }
+
+    std::vector<arm_compute::ITensorInfo*> aclInputPtrs;
+    for (arm_compute::ITensorInfo& input : aclInputs)
+    {
+        aclInputPtrs.emplace_back(&input);
+    }
+
+    const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
+    int aclAxis = CalcAxis(descriptor.m_Axis, descriptor.m_InputShape.GetNumDimensions());
+    return arm_compute::NEStackLayer::validate(aclInputPtrs, aclAxis, &aclOutputInfo);
+}
+
+NeonStackWorkload::NeonStackWorkload(const StackQueueDescriptor& descriptor, const WorkloadInfo& info)
+: BaseWorkload<StackQueueDescriptor>(descriptor, info)
+{
+    std::vector<arm_compute::ITensor*> aclInputs;
+    for (auto input : m_Data.m_Inputs)
+    {
+        arm_compute::ITensor& aclInput = boost::polymorphic_pointer_downcast<IAclTensorHandle>(input)->GetTensor();
+        aclInputs.emplace_back(&aclInput);
+    }
+    arm_compute::ITensor& output = boost::polymorphic_pointer_downcast<IAclTensorHandle>(
+        m_Data.m_Outputs[0])->GetTensor();
+
+    m_Layer.reset(new arm_compute::NEStackLayer());
+    int aclAxis = CalcAxis(descriptor.m_Parameters.m_Axis, descriptor.m_Parameters.m_InputShape.GetNumDimensions());
+    m_Layer->configure(aclInputs, aclAxis, &output);
+}
+
+void NeonStackWorkload::Execute() const
+{
+    if (m_Layer)
+    {
+        ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonStackWorkload_Execute");
+        m_Layer->run();
+    }
+}
+
+} //namespace armnn
diff --git a/src/backends/neon/workloads/NeonStackWorkload.hpp b/src/backends/neon/workloads/NeonStackWorkload.hpp
new file mode 100644
index 0000000000..24808bfe2d
--- /dev/null
+++ b/src/backends/neon/workloads/NeonStackWorkload.hpp
@@ -0,0 +1,29 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <backendsCommon/Workload.hpp>
+
+#include <arm_compute/runtime/NEON/functions/NEStackLayer.h>
+
+namespace armnn
+{
+arm_compute::Status NeonStackWorkloadValidate(const std::vector<const TensorInfo*>& inputs,
+                                              const TensorInfo& output,
+                                              const StackDescriptor& descriptor);
+
+class NeonStackWorkload : public BaseWorkload<StackQueueDescriptor>
+{
+public:
+    NeonStackWorkload(const StackQueueDescriptor& descriptor, const WorkloadInfo& info);
+
+    void Execute() const override;
+
+private:
+    mutable std::unique_ptr<arm_compute::NEStackLayer> m_Layer;
+};
+
+} //namespace armnn
diff --git a/src/backends/neon/workloads/NeonWorkloads.hpp b/src/backends/neon/workloads/NeonWorkloads.hpp
index c5f4a54b77..7cb6c3b7b1 100644
--- a/src/backends/neon/workloads/NeonWorkloads.hpp
+++ b/src/backends/neon/workloads/NeonWorkloads.hpp
@@ -35,4 +35,5 @@
 #include "NeonSoftmaxUint8Workload.hpp"
 #include "NeonSpaceToDepthWorkload.hpp"
 #include "NeonSplitterWorkload.hpp"
+#include "NeonStackWorkload.hpp"
 #include "NeonSubtractionWorkload.hpp"