IVGCVSW-3341 Add Neon backend support for Quantized_LSTM (16bit cell state)

 * Add Neon Workload
 * Update NeonWorkloads.hpp
 * Update NeonWorkloadFactory
 * Update NeonLayerSupport
 * Update backends.mk and CMakeLists.txt
 * Add NeonCreateWorkload test
 * Enable LayerTest

!android-nn-driver:1685

Change-Id: Idd799bbf039acf0d59084d02c3b57766ce3691b5
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
Signed-off-by: Matthew Bentham <Matthew.Bentham@arm.com>

1 files changed, 52 insertions, 0 deletions

diff --git a/src/backends/neon/workloads/NeonQuantizedLstmWorkload.hpp b/src/backends/neon/workloads/NeonQuantizedLstmWorkload.hpp
new file mode 100644
index 0000000000..ab8ea71437
--- /dev/null
+++ b/src/backends/neon/workloads/NeonQuantizedLstmWorkload.hpp
@@ -0,0 +1,52 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <backendsCommon/Workload.hpp>
+#include <backendsCommon/WorkloadData.hpp>
+
+#include <arm_compute/graph/Tensor.h>
+#include <arm_compute/runtime/NEON/functions/NELSTMLayerQuantized.h>
+
+namespace armnn
+{
+
+class NeonQuantizedLstmWorkload : public BaseWorkload<QuantizedLstmQueueDescriptor>
+{
+public:
+    NeonQuantizedLstmWorkload(const QuantizedLstmQueueDescriptor& descriptor, const WorkloadInfo& info);
+    virtual void Execute() const override;
+
+private:
+    mutable arm_compute::NELSTMLayerQuantized m_QuantizedLstmLayer;
+
+    std::unique_ptr<arm_compute::Tensor> m_InputToInputWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_InputToForgetWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_InputToCellWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_InputToOutputWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_RecurrentToInputWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_RecurrentToForgetWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_RecurrentToCellWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_RecurrentToOutputWeightsTensor;
+    std::unique_ptr<arm_compute::Tensor> m_InputGateBiasTensor;
+    std::unique_ptr<arm_compute::Tensor> m_ForgetGateBiasTensor;
+    std::unique_ptr<arm_compute::Tensor> m_CellBiasTensor;
+    std::unique_ptr<arm_compute::Tensor> m_OutputGateBiasTensor;
+    std::unique_ptr<arm_compute::Tensor> m_CellStateInTensor;
+    std::unique_ptr<arm_compute::Tensor> m_OutputStateInTensor;
+    std::unique_ptr<arm_compute::Tensor> m_CellStateOutTensor;
+
+    void FreeUnusedTensors();
+};
+
+arm_compute::Status NeonQuantizedLstmWorkloadValidate(const TensorInfo& input,
+                                                      const TensorInfo& outputStateIn,
+                                                      const TensorInfo& cellStateIn,
+                                                      const TensorInfo& outputStateOut,
+                                                      const TensorInfo& cellStateOut,
+                                                      const QuantizedLstmInputParamsInfo& paramsInfo);
+
+} //namespace armnn