IVGCVSW-6174 Add Cl Pooling3d Workload

 * Add IsSupported for Pooling3d
 * Add CreateWorkload case for Pooling3d
 * Create new ClPooling3dWorkload header and source files
 * Add Pooling3d workload to ClWorkloads.hpp
 * Add tests for Pooling3d workload
 * Add Pooling3d build function to ArmComputeTensorUtils

Change-Id: Ia270b0fe809a171ed73af14376de8708b346d500
Signed-off-by: Ryan OShea <ryan.oshea3@arm.com>

4 files changed, 110 insertions, 0 deletions

diff --git a/src/backends/cl/workloads/CMakeLists.txt b/src/backends/cl/workloads/CMakeLists.txt
index 6bb8d68d07..52326ae4a9 100644
--- a/src/backends/cl/workloads/CMakeLists.txt
+++ b/src/backends/cl/workloads/CMakeLists.txt
@@ -86,6 +86,8 @@ list(APPEND armnnClBackendWorkloads_sources
     ClPermuteWorkload.hpp
     ClPooling2dWorkload.cpp
     ClPooling2dWorkload.hpp
+    ClPooling3dWorkload.cpp
+    ClPooling3dWorkload.hpp
     ClPreluWorkload.cpp
     ClPreluWorkload.hpp
     ClQLstmWorkload.cpp
diff --git a/src/backends/cl/workloads/ClPooling3dWorkload.cpp b/src/backends/cl/workloads/ClPooling3dWorkload.cpp
new file mode 100644
index 0000000000..a896110a2e
--- /dev/null
+++ b/src/backends/cl/workloads/ClPooling3dWorkload.cpp
@@ -0,0 +1,73 @@
+//
+// Copyright © 2022 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "ClPooling3dWorkload.hpp"
+#include <cl/ClLayerSupport.hpp>
+#include <cl/ClTensorHandle.hpp>
+#include <aclCommon/ArmComputeUtils.hpp>
+#include <aclCommon/ArmComputeTensorUtils.hpp>
+
+#include "ClWorkloadUtils.hpp"
+
+namespace armnn
+{
+    using namespace armcomputetensorutils;
+
+    arm_compute::Status ClPooling3dWorkloadValidate(const TensorInfo& input,
+                                                    const TensorInfo& output,
+                                                    const Pooling3dDescriptor& descriptor)
+    {
+        const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input, descriptor.m_DataLayout);
+        const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output, descriptor.m_DataLayout);
+
+        arm_compute::Pooling3dLayerInfo layerInfo = BuildArmComputePooling3dLayerInfo(descriptor);
+
+        return arm_compute::CLPooling3dLayer::validate(&aclInputInfo, &aclOutputInfo, layerInfo);
+    }
+
+    ClPooling3dWorkload::ClPooling3dWorkload( const Pooling3dQueueDescriptor& descriptor,
+                                              const WorkloadInfo& info,
+                                              const arm_compute::CLCompileContext& clCompileContext)
+                                              : ClBaseWorkload<Pooling3dQueueDescriptor>(descriptor, info)
+    {
+        // Report Profiling Details
+        ARMNN_REPORT_PROFILING_WORKLOAD_DESC("ClPooling3dWorkload_Construct",
+                                             descriptor.m_Parameters,
+                                             info,
+                                             this->GetGuid());
+
+        m_Data.ValidateInputsOutputs("ClPooling3dWorkload", 1, 1);
+
+        arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
+        arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
+
+        arm_compute::DataLayout aclDataLayout = ConvertDataLayout(m_Data.m_Parameters.m_DataLayout);
+        input.info()->set_data_layout(aclDataLayout);
+        output.info()->set_data_layout(aclDataLayout);
+
+        // flag to use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy
+        // enable fp_mixed_precision for the the FP16 cases that
+        // accumulation reaches a limit beyond which there is no more increment of the value
+        bool fpMixedPrecision = false;
+
+        arm_compute::Pooling3dLayerInfo layerInfo = BuildArmComputePooling3dLayerInfo(m_Data.m_Parameters,
+                                                                                      fpMixedPrecision);
+
+        {
+            ARMNN_SCOPED_PROFILING_EVENT(Compute::Undefined, "ClPooling3dWorkload_configure");
+            // Run the layer.
+            m_PoolingLayer.configure(clCompileContext, &input, &output, layerInfo);
+        }
+    }
+
+    void ClPooling3dWorkload::Execute() const
+    {
+        ARMNN_SCOPED_PROFILING_EVENT_CL_GUID("ClPooling3dWorkload_Execute", this->GetGuid());
+        RunClFunction(m_PoolingLayer, CHECK_LOCATION());
+    }
+
+}
+
+
diff --git a/src/backends/cl/workloads/ClPooling3dWorkload.hpp b/src/backends/cl/workloads/ClPooling3dWorkload.hpp
new file mode 100644
index 0000000000..75777db188
--- /dev/null
+++ b/src/backends/cl/workloads/ClPooling3dWorkload.hpp
@@ -0,0 +1,34 @@
+//
+// Copyright © 2022 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "ClBaseWorkload.hpp"
+
+#include <arm_compute/runtime/CL/functions/CLPooling3dLayer.h>
+
+namespace armnn
+{
+
+    arm_compute::Status ClPooling3dWorkloadValidate(const TensorInfo& input,
+                                                    const TensorInfo& output,
+                                                    const Pooling3dDescriptor& descriptor);
+
+    class ClPooling3dWorkload : public ClBaseWorkload<Pooling3dQueueDescriptor>
+    {
+    public:
+        using BaseWorkload<Pooling3dQueueDescriptor>::m_Data;
+
+        ClPooling3dWorkload(const Pooling3dQueueDescriptor& descriptor,
+                            const WorkloadInfo& info,
+                            const arm_compute::CLCompileContext& clCompileContext);
+
+        void Execute() const override;
+
+    private:
+        mutable arm_compute::CLPooling3dLayer m_PoolingLayer;
+    };
+
+} //namespace armnn
diff --git a/src/backends/cl/workloads/ClWorkloads.hpp b/src/backends/cl/workloads/ClWorkloads.hpp
index 3558da341e..27119bb2dc 100644
--- a/src/backends/cl/workloads/ClWorkloads.hpp
+++ b/src/backends/cl/workloads/ClWorkloads.hpp
@@ -43,6 +43,7 @@
 #include "ClPermuteWorkload.hpp"
 #include "ClPadWorkload.hpp"
 #include "ClPooling2dWorkload.hpp"
+#include "ClPooling3dWorkload.hpp"
 #include "ClPreluWorkload.hpp"
 #include "ClQLstmWorkload.hpp"
 #include "ClQuantizeWorkload.hpp"