1 //
2 // Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5
6 #include "ConvertFp32ToBf16Layer.hpp"
7 #include "LayerCloneBase.hpp"
8
9 #include <armnn/TypesUtils.hpp>
10
11 #include <backendsCommon/WorkloadData.hpp>
12 #include <backendsCommon/WorkloadFactory.hpp>
13
14 namespace armnn
15 {
16
ConvertFp32ToBf16Layer(const char * name)17 ConvertFp32ToBf16Layer::ConvertFp32ToBf16Layer(const char* name)
18 : Layer(1, 1, LayerType::ConvertFp32ToBf16, name)
19 {
20 }
21
CreateWorkload(const IWorkloadFactory & factory) const22 std::unique_ptr<IWorkload> ConvertFp32ToBf16Layer::CreateWorkload(const IWorkloadFactory& factory) const
23 {
24 ConvertFp32ToBf16QueueDescriptor descriptor;
25 SetAdditionalInfo(descriptor);
26
27 return factory.CreateConvertFp32ToBf16(descriptor, PrepInfoAndDesc(descriptor));
28 }
29
Clone(Graph & graph) const30 ConvertFp32ToBf16Layer* ConvertFp32ToBf16Layer::Clone(Graph& graph) const
31 {
32 return CloneBase<ConvertFp32ToBf16Layer>(graph, GetName());
33 }
34
ValidateTensorShapesFromInputs()35 void ConvertFp32ToBf16Layer::ValidateTensorShapesFromInputs()
36 {
37
38 VerifyLayerConnections(1, CHECK_LOCATION());
39
40 const TensorShape& outputShape = GetOutputSlot(0).GetTensorInfo().GetShape();
41
42 VerifyShapeInferenceType(outputShape, m_ShapeInferenceMethod);
43
44 auto inferredShapes = InferOutputShapes({ GetInputSlot(0).GetConnection()->GetTensorInfo().GetShape() });
45
46 ARMNN_ASSERT(inferredShapes.size() == 1);
47
48 ValidateAndCopyShape(outputShape, inferredShapes[0], m_ShapeInferenceMethod, "LayerName");
49 }
50
Accept(ILayerVisitor & visitor) const51 void ConvertFp32ToBf16Layer::Accept(ILayerVisitor& visitor) const
52 {
53 // these conversion layers are only inserted by the
54 // optimizer and so will never be in an input graph.
55 IgnoreUnused(visitor);
56 throw armnn::Exception("ConvertFp32ToBf16Layer should never appear in an input graph");
57 }
58
59 } // namespace armnn
60