From 581742d8b6c25bf0917d6bdcd5f9b8a5e72b370d Mon Sep 17 00:00:00 2001
From: Sadik Armagan <sadik.armagan@arm.com>
Date: Mon, 12 Aug 2019 14:11:37 +0100
Subject: IVGCVSW-3163 Support Neon TransposeConv2d

* NEON support for TransposeConolution2d
* Indentation fix on SpaceToDepth LayerSupport method

Signed-off-by: Keith Davis <keith.davis@arm.com>
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Signed-off-by: Aron Virginas-Tar <aron.virginas-tar@arm.com>

Change-Id: Ie3a01a22e05f62f475f768dfe05ce561d2f3b0ad
---
 .../NeonTransposeConvolution2dWorkload.hpp         | 45 ++++++++++++++++++++++
 1 file changed, 45 insertions(+)
 create mode 100644 src/backends/neon/workloads/NeonTransposeConvolution2dWorkload.hpp

(limited to 'src/backends/neon/workloads/NeonTransposeConvolution2dWorkload.hpp')

diff --git a/src/backends/neon/workloads/NeonTransposeConvolution2dWorkload.hpp b/src/backends/neon/workloads/NeonTransposeConvolution2dWorkload.hpp
new file mode 100644
index 0000000000..995495e2ab
--- /dev/null
+++ b/src/backends/neon/workloads/NeonTransposeConvolution2dWorkload.hpp
@@ -0,0 +1,45 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <backendsCommon/Workload.hpp>
+
+#include <arm_compute/runtime/IFunction.h>
+#include <arm_compute/runtime/Tensor.h>
+#include <arm_compute/runtime/MemoryManagerOnDemand.h>
+
+#include <arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h>
+
+#include <memory>
+
+namespace armnn
+{
+
+arm_compute::Status NeonTransposeConvolution2dWorkloadValidate(const TensorInfo& input,
+                                                               const TensorInfo& output,
+                                                               const TransposeConvolution2dDescriptor& descriptor,
+                                                               const TensorInfo& weights,
+                                                               const Optional<TensorInfo>& biases);
+
+class NeonTransposeConvolution2dWorkload : public BaseWorkload<TransposeConvolution2dQueueDescriptor>
+{
+public:
+    NeonTransposeConvolution2dWorkload(const TransposeConvolution2dQueueDescriptor& descriptor,
+                                       const WorkloadInfo& info,
+                                       std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager);
+
+    void Execute() const override;
+
+private:
+    std::unique_ptr<arm_compute::NEDeconvolutionLayer> m_Layer;
+
+    std::unique_ptr<arm_compute::Tensor> m_KernelTensor;
+    std::unique_ptr<arm_compute::Tensor> m_BiasTensor;
+
+    void FreeUnusedTensors();
+};
+
+} //namespace armnn
-- 
cgit v1.2.1