IVGCVSW-4622 Add NEON FILL Workload

* Added Neon workload for Fill Operator
* Enabled Fill operator tests on Neon
* NEFill function does not have validate() function yet
  IsLayerSupported() function return true at the moment
* Added INT32 supported type for CpuRef

Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I17bf5ec13750f46322a30653e15ba2a514f61f08

12 files changed, 98 insertions, 2 deletions

diff --git a/src/backends/aclCommon/ArmComputeTensorUtils.cpp b/src/backends/aclCommon/ArmComputeTensorUtils.cpp
index 7fbaa303db..f9335058c2 100644
--- a/src/backends/aclCommon/ArmComputeTensorUtils.cpp
+++ b/src/backends/aclCommon/ArmComputeTensorUtils.cpp
@@ -267,6 +267,8 @@ arm_compute::PixelValue GetPixelValue(arm_compute::ITensor& input, float pixelVa
         case arm_compute::DataType::QASYMM8_SIGNED:
         case arm_compute::DataType::QSYMM8_PER_CHANNEL:
             return arm_compute::PixelValue(static_cast<int8_t>(pixelValue));
+        case arm_compute::DataType::S32:
+            return arm_compute::PixelValue(static_cast<int32_t>(pixelValue));
         default:
             throw InvalidArgumentException("Unsupported DataType: [" +
                                            std::to_string(static_cast<int>(input.info()->data_type())) + "]");
diff --git a/src/backends/neon/NeonLayerSupport.cpp b/src/backends/neon/NeonLayerSupport.cpp
index a514c8fae2..b611bf45f9 100644
--- a/src/backends/neon/NeonLayerSupport.cpp
+++ b/src/backends/neon/NeonLayerSupport.cpp
@@ -405,6 +405,14 @@ bool NeonLayerSupport::IsElementwiseUnarySupported(const TensorInfo& input,
     }
 }
 
+bool NeonLayerSupport::IsFillSupported(const TensorInfo&,
+                                       const TensorInfo&,
+                                       const FillDescriptor&,
+                                       Optional<std::string&>) const
+{
+    return true;
+}
+
 bool NeonLayerSupport::IsFloorSupported(const TensorInfo& input,
                                         const TensorInfo& output,
                                         Optional<std::string&> reasonIfUnsupported) const
diff --git a/src/backends/neon/NeonLayerSupport.hpp b/src/backends/neon/NeonLayerSupport.hpp
index 25817822fd..7217fc8971 100644
--- a/src/backends/neon/NeonLayerSupport.hpp
+++ b/src/backends/neon/NeonLayerSupport.hpp
@@ -112,6 +112,11 @@ public:
                                const ElementwiseUnaryDescriptor& descriptor,
                                Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const override;
 
+    bool IsFillSupported(const TensorInfo& input,
+                         const TensorInfo& output,
+                         const FillDescriptor& descriptor,
+                         Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const override;
+
     bool IsFloorSupported(const TensorInfo& input,
                           const TensorInfo& output,
                           Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const override;
diff --git a/src/backends/neon/NeonWorkloadFactory.cpp b/src/backends/neon/NeonWorkloadFactory.cpp
index 82934c3b25..34f8445f95 100644
--- a/src/backends/neon/NeonWorkloadFactory.cpp
+++ b/src/backends/neon/NeonWorkloadFactory.cpp
@@ -265,6 +265,12 @@ std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateEqual(const EqualQueueDesc
     return CreateComparison(comparisonDescriptor, info);
 }
 
+std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFill(const FillQueueDescriptor& descriptor,
+                                                           const WorkloadInfo& info) const
+{
+    return std::make_unique<NeonFillWorkload>(descriptor, info);
+}
+
 std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFloor(const FloorQueueDescriptor& descriptor,
                                                             const WorkloadInfo& info) const
 {
diff --git a/src/backends/neon/NeonWorkloadFactory.hpp b/src/backends/neon/NeonWorkloadFactory.hpp
index d551f22345..de49146f10 100644
--- a/src/backends/neon/NeonWorkloadFactory.hpp
+++ b/src/backends/neon/NeonWorkloadFactory.hpp
@@ -106,6 +106,9 @@ public:
     std::unique_ptr<IWorkload> CreateEqual(const EqualQueueDescriptor& descriptor,
                                            const WorkloadInfo& info) const override;
 
+    std::unique_ptr<IWorkload> CreateFill(const FillQueueDescriptor& descriptor,
+                                          const WorkloadInfo& info) const override;
+
     std::unique_ptr<IWorkload> CreateFloor(const FloorQueueDescriptor& descriptor,
                                            const WorkloadInfo& info) const override;
 
diff --git a/src/backends/neon/backend.mk b/src/backends/neon/backend.mk
index 9e9c38cdec..72d1ab3a92 100644
--- a/src/backends/neon/backend.mk
+++ b/src/backends/neon/backend.mk
@@ -40,6 +40,7 @@ BACKEND_SOURCES := \
         workloads/NeonDequantizeWorkload.cpp \
         workloads/NeonDetectionPostProcessWorkload.cpp \
         workloads/NeonExpWorkload.cpp \
+        workloads/NeonFillWorkload.cpp \
         workloads/NeonFloorFloatWorkload.cpp \
         workloads/NeonFullyConnectedWorkload.cpp \
         workloads/NeonGatherWorkload.cpp \
diff --git a/src/backends/neon/test/NeonLayerTests.cpp b/src/backends/neon/test/NeonLayerTests.cpp
index e56c5ddb2e..c4b5d18f08 100644
--- a/src/backends/neon/test/NeonLayerTests.cpp
+++ b/src/backends/neon/test/NeonLayerTests.cpp
@@ -1221,6 +1221,10 @@ ARMNN_AUTO_TEST_CASE(Exo3d, Exp3dTest<DataType::Float32>)
 ARMNN_AUTO_TEST_CASE(ExpZero, ExpZeroTest<DataType::Float32>)
 ARMNN_AUTO_TEST_CASE(ExpNegative, ExpNegativeTest<DataType::Float32>)
 
+// Fill
+ARMNN_AUTO_TEST_CASE(SimpleFill, SimpleFillTest<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE(SimpleFillF16, SimpleFillTest<DataType::Float16>)
+
 #if defined(ARMNNREF_ENABLED)
 
 // The ARMNN_COMPARE_REF_AUTO_TEST_CASE and the ARMNN_COMPARE_REF_FIXTURE_TEST_CASE test units are not available
diff --git a/src/backends/neon/workloads/CMakeLists.txt b/src/backends/neon/workloads/CMakeLists.txt
index 685d75be12..e3b74f665f 100644
--- a/src/backends/neon/workloads/CMakeLists.txt
+++ b/src/backends/neon/workloads/CMakeLists.txt
@@ -42,6 +42,8 @@ list(APPEND armnnNeonBackendWorkloads_sources
     NeonDetectionPostProcessWorkload.hpp
     NeonExpWorkload.cpp
     NeonExpWorkload.hpp
+    NeonFillWorkload.cpp
+    NeonFillWorkload.hpp
     NeonFloorFloatWorkload.cpp
     NeonFloorFloatWorkload.hpp
     NeonFullyConnectedWorkload.cpp
diff --git a/src/backends/neon/workloads/NeonFillWorkload.cpp b/src/backends/neon/workloads/NeonFillWorkload.cpp
new file mode 100644
index 0000000000..5965d2000c
--- /dev/null
+++ b/src/backends/neon/workloads/NeonFillWorkload.cpp
@@ -0,0 +1,38 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "NeonFillWorkload.hpp"
+
+#include <neon/NeonTensorHandle.hpp>
+#include <aclCommon/ArmComputeTensorUtils.hpp>
+#include <arm_compute/core/Types.h>
+#include <arm_compute/runtime/NEON/functions/NEFill.h>
+
+#include "NeonWorkloadUtils.hpp"
+
+namespace armnn
+{
+using namespace armcomputetensorutils;
+
+NeonFillWorkload::NeonFillWorkload(const FillQueueDescriptor& descriptor, const WorkloadInfo& info)
+    : BaseWorkload<FillQueueDescriptor>(descriptor, info)
+{
+    m_Data.ValidateInputsOutputs("NeonFillWorkload", 1, 1);
+
+    arm_compute::ITensor& output = static_cast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
+    arm_compute::PixelValue pixelValue = GetPixelValue(output, descriptor.m_Parameters.m_Value);
+
+    auto layer = std::make_unique<arm_compute::NEFill>();
+    layer->configure(&output, pixelValue);
+    m_Layer.reset(layer.release());
+}
+
+void NeonFillWorkload::Execute() const
+{
+    ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonFillWorkload_Execute");
+    m_Layer->run();
+}
+
+} // namespace armnn
diff --git a/src/backends/neon/workloads/NeonFillWorkload.hpp b/src/backends/neon/workloads/NeonFillWorkload.hpp
new file mode 100644
index 0000000000..40e4aa68c4
--- /dev/null
+++ b/src/backends/neon/workloads/NeonFillWorkload.hpp
@@ -0,0 +1,25 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <backendsCommon/WorkloadData.hpp>
+#include <backendsCommon/Workload.hpp>
+#include <arm_compute/runtime/IFunction.h>
+#include <arm_compute/core/Error.h>
+
+namespace armnn {
+
+class NeonFillWorkload : public BaseWorkload<FillQueueDescriptor>
+{
+public:
+    NeonFillWorkload(const FillQueueDescriptor& descriptor, const WorkloadInfo& info);
+    void Execute() const override;
+
+private:
+    std::unique_ptr<arm_compute::IFunction> m_Layer;
+};
+
+} //namespace armnn
diff --git a/src/backends/neon/workloads/NeonWorkloads.hpp b/src/backends/neon/workloads/NeonWorkloads.hpp
index 243f5a46ee..c1bc4cca29 100644
--- a/src/backends/neon/workloads/NeonWorkloads.hpp
+++ b/src/backends/neon/workloads/NeonWorkloads.hpp
@@ -24,6 +24,7 @@
 #include "NeonDetectionPostProcessWorkload.hpp"
 #include "NeonDivisionWorkload.hpp"
 #include "NeonExpWorkload.hpp"
+#include "NeonFillWorkload.hpp"
 #include "NeonFloorFloatWorkload.hpp"
 #include "NeonFullyConnectedWorkload.hpp"
 #include "NeonGatherWorkload.hpp"
diff --git a/src/backends/reference/RefLayerSupport.cpp b/src/backends/reference/RefLayerSupport.cpp
index e13a2e23c0..18b36a5fa8 100644
--- a/src/backends/reference/RefLayerSupport.cpp
+++ b/src/backends/reference/RefLayerSupport.cpp
@@ -867,10 +867,11 @@ bool RefLayerSupport::IsFillSupported(const TensorInfo& input,
 
     bool supported = true;
 
-    std::array<DataType,2> supportedTypes =
+    std::array<DataType,3> supportedTypes =
     {
         DataType::Float32,
-        DataType::Float16
+        DataType::Float16,
+        DataType::Signed32
     };
 
     supported &= CheckSupportRule(TypeAnyOf(input, supportedTypes), reasonIfUnsupported,