IVGCVSW-1900 : CL backend folder structure
[platform/upstream/armnn.git] / src / backends / cl / workloads / ClPermuteWorkload.cpp
1 //
2 // Copyright © 2017 Arm Ltd. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5
6 #include "ClPermuteWorkload.hpp"
7 #include <backends/cl/ClTensorHandle.hpp>
8 #include <backends/aclCommon/ArmComputeTensorUtils.hpp>
9
10 #include <arm_compute/core/Error.h>
11
12 #include "ClWorkloadUtils.hpp"
13
14 namespace armnn
15 {
16
17 arm_compute::Status ClPermuteWorkloadValidate(const PermuteDescriptor& descriptor)
18 {
19     const armnn::PermutationVector& perm = descriptor.m_DimMappings;
20
21     ARM_COMPUTE_RETURN_ERROR_ON_MSG(!perm.IsEqual({ 0U, 3U, 1U, 2U })
22                                     && !perm.IsEqual({ 0U, 2U, 3U, 1U })
23                                     && !perm.IsEqual({ 3U, 2U, 0U, 1U }),
24     "Only [0, 3, 1, 2], [0, 2, 3, 1] and [3, 2, 0, 1] permutations are supported");
25
26     return arm_compute::Status{};
27 }
28
29 template <armnn::DataType... DataTypes>
30 ClPermuteWorkload<DataTypes...>::ClPermuteWorkload(const PermuteQueueDescriptor& descriptor,
31                                                const WorkloadInfo& info)
32     : TypedWorkload<PermuteQueueDescriptor, DataTypes...>(descriptor, info)
33 {
34     using armcomputetensorutils::BuildArmComputePermutationVector;
35
36     m_Data.ValidateInputsOutputs(GetName(), 1, 1);
37
38     const arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
39     arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
40     const armnn::PermutationVector& mappings = m_Data.m_Parameters.m_DimMappings;
41
42     // Run the layer.
43     m_PermuteFunction.configure(&input, &output, BuildArmComputePermutationVector(mappings));
44 }
45
46 template <armnn::DataType... DataTypes>
47 void ClPermuteWorkload<DataTypes...>::Execute() const
48 {
49     ARMNN_SCOPED_PROFILING_EVENT_CL( GetName() + "_Execute");
50     m_PermuteFunction.run();
51 }
52
53 template class ClPermuteWorkload<DataType::Float16, DataType::Float32>;
54 template class ClPermuteWorkload<DataType::QuantisedAsymm8>;
55
56 } // namespace armnn