Finn Williams | b454c5c | 2021-02-09 15:56:23 +0000 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2021 Arm Ltd and Contributors. All rights reserved. |
| 3 | // SPDX-License-Identifier: MIT |
| 4 | // |
| 5 | |
| 6 | #include "Network.hpp" |
| 7 | #include "NetworkQuantizerUtils.hpp" |
| 8 | #include "StaticRangeStrategy.hpp" |
| 9 | |
| 10 | #include <armnn/utility/NumericCast.hpp> |
| 11 | #include <armnn/utility/PolymorphicDowncast.hpp> |
| 12 | |
| 13 | namespace armnn |
| 14 | { |
| 15 | class QuantizerStrategy : public IStrategy |
| 16 | { |
| 17 | public : |
| 18 | QuantizerStrategy(const RangeTracker& rangeTracker, |
| 19 | const IQuantizationScheme* quantizationScheme, |
| 20 | bool preserveType); |
| 21 | |
| 22 | ~QuantizerStrategy() = default; |
| 23 | |
| 24 | void ExecuteStrategy(const armnn::IConnectableLayer* layer, |
| 25 | const BaseDescriptor& descriptor, |
| 26 | const std::vector<armnn::ConstTensor>& constants, |
| 27 | const char* name, |
| 28 | const armnn::LayerBindingId id) override; |
| 29 | |
| 30 | /// Extract the quantized network |
| 31 | INetworkPtr RetrieveFinalNetwork() { return std::move(m_QuantizedNetwork); } |
| 32 | |
| 33 | private: |
| 34 | /// Connects the layer to preceeding layers and sets the quantization parameters based on recorded ranges |
| 35 | void SetQuantizedInputConnections(const IConnectableLayer* srcLayer, IConnectableLayer* quantizedLayer); |
| 36 | |
| 37 | /// Record the guids so we can easily find the layers later |
| 38 | void RecordLayer(const IConnectableLayer* srcLayer, IConnectableLayer* qLayer); |
| 39 | |
| 40 | /// Sets the bias quantization scale based on input and weight scales |
| 41 | ConstTensor CreateQuantizedBias(const IConnectableLayer* srcLayer, |
| 42 | const ConstTensor& weights, |
| 43 | const Optional<ConstTensor>& biases, |
| 44 | std::vector<int32_t>& weightsBacking); |
| 45 | |
| 46 | /// Reference to the static range visitor used to retrieve the quantization ranges |
| 47 | const RangeTracker& m_Ranges; |
| 48 | |
| 49 | /// Quantized version of the model we are building up |
| 50 | INetworkPtr m_QuantizedNetwork; |
| 51 | |
| 52 | /// Mapping from input network guids to quantized network guids |
| 53 | std::unordered_map<LayerGuid, LayerGuid> m_OriginalToQuantizedGuidMap; |
| 54 | |
| 55 | /// Mapping from guid to layer in quantized network |
| 56 | std::unordered_map<LayerGuid, IConnectableLayer*> m_QuantizedGuidToLayerMap; |
| 57 | |
| 58 | const IQuantizationScheme* m_QuantizationScheme; |
| 59 | |
| 60 | const bool m_PreserveType; |
| 61 | }; |
| 62 | |
| 63 | } //namespace armnn |