IVGCVSW-4626 Add Neon Rank Workload

Signed-off-by: David Monahan <david.monahan@arm.com>
Change-Id: I5a85597e75d2b879ae234c6929686fabe99d7bc8

7 files changed, 80 insertions, 0 deletions

diff --git a/src/backends/neon/NeonWorkloadFactory.cpp b/src/backends/neon/NeonWorkloadFactory.cpp
index 72180528f7..0d36110da5 100644
--- a/src/backends/neon/NeonWorkloadFactory.cpp
+++ b/src/backends/neon/NeonWorkloadFactory.cpp
@@ -491,6 +491,12 @@ std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateQuantizedLstm(const Quanti
     return std::make_unique<NeonQuantizedLstmWorkload>(descriptor, info);
 }
 
+std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateRank(const RankQueueDescriptor& descriptor,
+                                                           const WorkloadInfo& info) const
+{
+    return std::make_unique<NeonRankWorkload>(descriptor, info);
+}
+
 std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateReshape(const ReshapeQueueDescriptor& descriptor,
                                                               const WorkloadInfo& info) const
 {
diff --git a/src/backends/neon/NeonWorkloadFactory.hpp b/src/backends/neon/NeonWorkloadFactory.hpp
index 444574ed80..745dece659 100644
--- a/src/backends/neon/NeonWorkloadFactory.hpp
+++ b/src/backends/neon/NeonWorkloadFactory.hpp
@@ -204,6 +204,9 @@ public:
     std::unique_ptr<IWorkload> CreateQuantizedLstm(const QuantizedLstmQueueDescriptor& descriptor,
                                                    const WorkloadInfo& info) const override;
 
+    std::unique_ptr<IWorkload> CreateRank(const RankQueueDescriptor& descriptor,
+                                          const WorkloadInfo& info) const override;
+
     std::unique_ptr<IWorkload> CreateReshape(const ReshapeQueueDescriptor& descriptor,
                                              const WorkloadInfo& info) const override;
 
diff --git a/src/backends/neon/test/NeonLayerTests.cpp b/src/backends/neon/test/NeonLayerTests.cpp
index 8e7742ab55..d351870645 100644
--- a/src/backends/neon/test/NeonLayerTests.cpp
+++ b/src/backends/neon/test/NeonLayerTests.cpp
@@ -610,6 +610,39 @@ ARMNN_AUTO_TEST_CASE_WITH_THF(Multiplication5d, Multiplication5dTest)
 ARMNN_AUTO_TEST_CASE_WITH_THF(BatchNormFloat32, BatchNormFloat32Test)
 ARMNN_AUTO_TEST_CASE_WITH_THF(BatchNormFloat32Nhwc, BatchNormFloat32NhwcTest)
 
+// Rank
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1Float16,  RankDimSize1Test<DataType::Float16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1Float32,  RankDimSize1Test<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1QAsymmU8, RankDimSize1Test<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1Signed32, RankDimSize1Test<DataType::Signed32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1QSymmS16, RankDimSize1Test<DataType::QSymmS16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1QAsymmS8, RankDimSize1Test<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize1BFloat16, RankDimSize1Test<DataType::BFloat16>)
+
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2Float16,  RankDimSize2Test<DataType::Float16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2Float32,  RankDimSize2Test<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2QAsymmU8, RankDimSize2Test<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2Signed32, RankDimSize2Test<DataType::Signed32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2QSymmS16, RankDimSize2Test<DataType::QSymmS16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2QAsymmS8, RankDimSize2Test<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize2BFloat16, RankDimSize2Test<DataType::BFloat16>)
+
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3Float16,  RankDimSize3Test<DataType::Float16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3Float32,  RankDimSize3Test<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3QAsymmU8, RankDimSize3Test<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3Signed32, RankDimSize3Test<DataType::Signed32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3QSymmS16, RankDimSize3Test<DataType::QSymmS16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3QAsymmS8, RankDimSize3Test<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize3BFloat16, RankDimSize3Test<DataType::BFloat16>)
+
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4Float16,  RankDimSize4Test<DataType::Float16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4Float32,  RankDimSize4Test<DataType::Float32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4QAsymmU8, RankDimSize4Test<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4Signed32, RankDimSize4Test<DataType::Signed32>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4QSymmS16, RankDimSize4Test<DataType::QSymmS16>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4QAsymmS8, RankDimSize4Test<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_CASE_WITH_THF(RankDimSize4BFloat16, RankDimSize4Test<DataType::BFloat16>)
+
 // InstanceNormalization
 ARMNN_AUTO_TEST_CASE_WITH_THF(InstanceNormFloat32Nchw, InstanceNormFloat32Test, DataLayout::NCHW);
 ARMNN_AUTO_TEST_CASE_WITH_THF(InstanceNormFloat32Nhwc, InstanceNormFloat32Test, DataLayout::NHWC);
diff --git a/src/backends/neon/workloads/CMakeLists.txt b/src/backends/neon/workloads/CMakeLists.txt
index b03db99989..f1a723b324 100644
--- a/src/backends/neon/workloads/CMakeLists.txt
+++ b/src/backends/neon/workloads/CMakeLists.txt
@@ -92,6 +92,7 @@ list(APPEND armnnNeonBackendWorkloads_sources
     NeonQuantizedLstmWorkload.hpp
     NeonQuantizeWorkload.cpp
     NeonQuantizeWorkload.hpp
+    NeonRankWorkload.hpp
     NeonReshapeWorkload.cpp
     NeonReshapeWorkload.hpp
     NeonResizeWorkload.cpp
diff --git a/src/backends/neon/workloads/NeonRankWorkload.hpp b/src/backends/neon/workloads/NeonRankWorkload.hpp
new file mode 100644
index 0000000000..f4e6959146
--- /dev/null
+++ b/src/backends/neon/workloads/NeonRankWorkload.hpp
@@ -0,0 +1,29 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <backendsCommon/Workload.hpp>
+#include <backendsCommon/WorkloadData.hpp>
+
+#include "NeonWorkloadUtils.hpp"
+
+namespace armnn
+{
+
+struct NeonRankWorkload : public BaseWorkload<RankQueueDescriptor>
+{
+public:
+    using BaseWorkload<RankQueueDescriptor>::BaseWorkload;
+    virtual void Execute() const override
+    {
+        const NeonTensorHandle* neonTensorHandle = PolymorphicDowncast<const NeonTensorHandle*>(m_Data.m_Inputs[0]);
+        const int32_t rank = static_cast<int32_t>(neonTensorHandle->GetShape().GetNumDimensions());
+
+        std::memcpy(GetOutputTensorData<void>(0, m_Data), &rank, sizeof(int32_t));
+    }
+};
+
+} //namespace armnn
diff --git a/src/backends/neon/workloads/NeonWorkloadUtils.hpp b/src/backends/neon/workloads/NeonWorkloadUtils.hpp
index 00ecb13008..589d2ea017 100644
--- a/src/backends/neon/workloads/NeonWorkloadUtils.hpp
+++ b/src/backends/neon/workloads/NeonWorkloadUtils.hpp
@@ -113,4 +113,11 @@ inline auto SetNeonSliceData(const std::vector<unsigned int>& m_begin,
     return std::make_tuple(starts, ends);
 }
 
+template <typename DataType, typename PayloadType>
+DataType* GetOutputTensorData(unsigned int idx, const PayloadType& data)
+{
+    ITensorHandle* tensorHandle = data.m_Outputs[idx];
+    return reinterpret_cast<DataType*>(tensorHandle->Map());
+}
+
 } //namespace armnn
diff --git a/src/backends/neon/workloads/NeonWorkloads.hpp b/src/backends/neon/workloads/NeonWorkloads.hpp
index 1a17b9aea9..949100d50a 100644
--- a/src/backends/neon/workloads/NeonWorkloads.hpp
+++ b/src/backends/neon/workloads/NeonWorkloads.hpp
@@ -48,6 +48,7 @@
 #include "NeonQLstmWorkload.hpp"
 #include "NeonQuantizedLstmWorkload.hpp"
 #include "NeonQuantizeWorkload.hpp"
+#include "NeonRankWorkload.hpp"
 #include "NeonReshapeWorkload.hpp"
 #include "NeonResizeWorkload.hpp"
 #include "NeonRsqrtWorkload.hpp"