telsoa01 | 5307bc1 | 2018-03-09 13:51:08 +0000 | [diff] [blame^] | 1 | ------ ArmNN for Android 18.02 Release Notes ------ |
| 2 | |
| 3 | This release of ArmNN for Android supports use as a driver for the Android Neural Networks API. It implements the android.hardware.neuralnetworks@1.0 interface. |
| 4 | |
| 5 | For more information on the Android Neural Networks API, see https://developer.android.com/ndk/guides/neuralnetworks/index.html |
| 6 | |
| 7 | For integration and usage documentation, please see README.md. |
| 8 | |
| 9 | --- Support for Android Neural Networks HAL operations --- |
| 10 | |
| 11 | The following AndroidNN operations are currently supported. |
| 12 | |
| 13 | AndroidNN operator Tensor type supported |
| 14 | ADD (FLOAT32) |
| 15 | AVERAGE_POOL_2D (FLOAT32,QUANT8_ASYMM) |
| 16 | CONCATENATION (FLOAT32) |
| 17 | CONV_2D (FLOAT32,QUANT8_ASYMM**) |
| 18 | DEPTHWISE_CONV_2D*** (FLOAT32,QUANT8_ASYMM) |
| 19 | FLOOR (FLOAT32) |
| 20 | FULLY_CONNECTED (FLOAT32) |
| 21 | L2_NORMALIZATION (FLOAT32) |
| 22 | L2_POOL_2D (FLOAT32) |
| 23 | LOCAL_RESPONSE_NORMALIZATION (FLOAT32) |
| 24 | LOGISTIC (FLOAT32,QUANT8_ASYMM) |
| 25 | MAX_POOL_2D (FLOAT32,QUANT8_ASYMM) |
| 26 | MUL* (FLOAT32) |
| 27 | RELU (FLOAT32,QUANT8_ASYMM) |
| 28 | RELU1 (FLOAT32,QUANT8_ASYMM) |
| 29 | RELU6 (FLOAT32,QUANT8_ASYMM) |
| 30 | RESHAPE (FLOAT32,QUANT8_ASYMM) |
| 31 | RESIZE_BILINEAR (FLOAT32) |
| 32 | SOFTMAX (FLOAT32,QUANT8_ASYMM) |
| 33 | TANH (FLOAT32) |
| 34 | |
| 35 | * MUL currently does not support mixing of different tensor sizes. |
| 36 | |
| 37 | ** QUANT8_ASYMM version does not support asymmetric padding. In addition, only the following configurations are supported: |
| 38 | 1) 1x1 convolution with strides of 1 or 2 or 3 |
| 39 | 2) 3x3 convolution with strides of 1 or 2 |
| 40 | 3) 5x5 convolution with strides of 1 or 2 |
| 41 | |
| 42 | *** Depthwise convolution only supports a value of 1 for the depth multiplier. In addition, the QUANT8_ASYMM version only supports 3x3 kernels. |
| 43 | |
| 44 | |
| 45 | --- Unsupported operators --- |
| 46 | |
| 47 | The following AndroidNN operations are currently not supported. |
| 48 | |
| 49 | DEPTH_TO_SPACE |
| 50 | DEQUANTIZE |
| 51 | EMBEDDING_LOOKUP |
| 52 | HASHTABLE_LOOKUP |
| 53 | LSH_PROJECTION |
| 54 | LSTM |
| 55 | RNN |
| 56 | SPACE_TO_DEPTH |
| 57 | SVDF |
| 58 | |
| 59 | Where operations are not supported by the ArmNN Android NN Driver, the driver indicates this to the framework appropriately and the framework implements those operations using a CPU implementation. |