| Jan Eilers | 31a7c89 | 2021-01-29 14:18:13 +0000 | [diff] [blame^] | 1 | /// Copyright (c) 2021 ARM Limited and Contributors. All rights reserved. |
| 2 | /// |
| 3 | /// SPDX-License-Identifier: MIT |
| 4 | /// |
| 5 | |
| 6 | namespace armnn |
| 7 | { |
| 8 | /** |
| 9 | @page delegate TfLite Delegate |
| 10 | @tableofcontents |
| 11 | |
| 12 | |
| 13 | @section delegateintro About the delegate |
| 14 | 'armnnDelegate' is a library for accelerating certain TensorFlow Lite (TfLite) operators on Arm hardware. It can be |
| 15 | integrated in TfLite using its delegation mechanism. TfLite will then delegate the execution of operators supported by |
| 16 | Arm NN to Arm NN. |
| 17 | |
| 18 | The main difference to our @ref S6_tf_lite_parser is the amount of operators you can run with it. If none of the active |
| 19 | backends support an operation in your model you won't be able to execute it with our parser. In contrast to that, TfLite |
| 20 | only delegates operations to the armnnDelegate if it does support them and otherwise executes them itself. In other |
| 21 | words, every TfLite model can be executed and every operation in your model that we can accelerate will be accelerated. |
| 22 | That is the reason why the armnnDelegate is our recommended way to accelerate TfLite models. |
| 23 | |
| 24 | If you need help building the armnnDelegate, please take a look at our [build guide](delegate/BuildGuideNative.md). |
| 25 | An example how to setup TfLite to integrate the armnnDelegate can be found in this |
| 26 | guide: [Integrate the delegate into python](delegate/IntegrateDelegateIntoPython.md) |
| 27 | |
| 28 | |
| 29 | @section delegatesupport Supported Operators |
| 30 | This reference guide provides a list of TensorFlow Lite operators the Arm NN SDK currently supports. |
| 31 | |
| 32 | @subsection delegatefullysupported Fully supported |
| 33 | |
| 34 | The Arm NN SDK TensorFlow Lite delegate currently supports the following operators: |
| 35 | |
| 36 | - ABS |
| 37 | |
| 38 | - ADD |
| 39 | |
| 40 | - ARGMAX |
| 41 | |
| 42 | - ARGMIN |
| 43 | |
| 44 | - AVERAGE_POOL_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE |
| 45 | |
| 46 | - CONCATENATION, Supported Fused Activation: RELU , RELU6 , TANH, NONE |
| 47 | |
| 48 | - CONV_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE |
| 49 | |
| 50 | - DEPTH_TO_SPACE |
| 51 | |
| 52 | - DEPTHWISE_CONV_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE |
| 53 | |
| 54 | - DEQUANTIZE |
| 55 | |
| 56 | - DIV |
| 57 | |
| 58 | - EQUAL |
| 59 | |
| 60 | - ELU |
| 61 | |
| 62 | - EXP |
| 63 | |
| 64 | - FULLY_CONNECTED, Supported Fused Activation: RELU , RELU6 , TANH, NONE |
| 65 | |
| 66 | - GATHER |
| 67 | |
| 68 | - GREATER |
| 69 | |
| 70 | - GREATER_OR_EQUAL |
| 71 | |
| 72 | - HARD_SWISH |
| 73 | |
| 74 | - LESS |
| 75 | |
| 76 | - LESS_OR_EQUAL |
| 77 | |
| 78 | - LOCAL_RESPONSE_NORMALIZATION |
| 79 | |
| 80 | - LOGICAL_AND |
| 81 | - |
| 82 | - LOGICAL_NOT |
| 83 | - |
| 84 | - LOGICAL_OR |
| 85 | |
| 86 | - LOGISTIC |
| 87 | |
| 88 | - LOG_SOFTMAX |
| 89 | |
| 90 | - L2_NORMALIZATION |
| 91 | |
| 92 | - L2_POOL_2D |
| 93 | |
| 94 | - MAXIMUM |
| 95 | |
| 96 | - MAX_POOL_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE |
| 97 | |
| 98 | - MEAN |
| 99 | |
| 100 | - MINIMUM |
| 101 | |
| 102 | - MUL |
| 103 | |
| 104 | - NEG |
| 105 | |
| 106 | - NOT_EQUAL |
| 107 | |
| 108 | - PAD |
| 109 | |
| 110 | - QUANTIZE |
| 111 | |
| 112 | - RESHAPE |
| 113 | |
| 114 | - RESIZE_BILINEAR |
| 115 | |
| 116 | - RESIZE_NEAREST_NEIGHBOR |
| 117 | |
| 118 | - RELU |
| 119 | |
| 120 | - RELU6 |
| 121 | |
| 122 | - RSQRT |
| 123 | |
| 124 | - SOFTMAX |
| 125 | |
| 126 | - SPACE_TO_DEPTH |
| 127 | |
| 128 | - SPLIT |
| 129 | |
| 130 | - SPLIT_V |
| 131 | |
| 132 | - SQRT |
| 133 | |
| 134 | - SUB |
| 135 | |
| 136 | - TANH |
| 137 | |
| 138 | - TRANSPOSE |
| 139 | |
| 140 | - TRANSPOSE_CONV |
| 141 | |
| 142 | More machine learning operators will be supported in future releases. |
| 143 | **/ |
| 144 | } |