friendlyelec
/
rk35xx-android14
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from 'a14ee00cc7'
${ noResults }
rk35xx-android14/external/armnn/docs/02_operator_list.dox

3415 lines
59 KiB
Raw Blame History

/// Copyright (c) 2021 ARM Limited and Contributors. All rights reserved.
///
/// SPDX-License-Identifier: MIT
///
namespace armnn
{
/**
@page operator_list Arm NN Operators
@tableofcontents
@section S5_1_operator_list Arm NN Operators
Arm NN supports operators that are listed in below table.
Arm NN supports a wide list of data-types.
The main data-types that the Machine Learning functions support are the following:
<ul>
<li><b>BFLOAT16:</b> 16-bit non-standard brain floating point
<li><b>QASYMMU8:</b> 8-bit unsigned asymmetric quantized
<li><b>QASYMMS8:</b> 8-bit signed asymmetric quantized
<li><b>QUANTIZEDSYMM8PERAXIS:</b> 8-bit signed symmetric quantized
<li><b>QSYMMS8:</b> 8-bit signed symmetric quantized
<li><b>QSYMMS16:</b> 16-bit signed symmetric quantized
<li><b>FLOAT32:</b> 32-bit single precision floating point
<li><b>FLOAT16:</b> 16-bit half precision floating point
<li><b>SIGNED32:</b> 32-bit signed integer
<li><b>BOOLEAN:</b> 8-bit unsigned char
<li><b>All:</b> Agnostic to any specific data type
</ul>
Arm NN supports the following data layouts (fast changing dimension from right to left):
<ul>
<li><b>NHWC:</b> Layout where channels are in the fastest changing dimension
<li><b>NCHW:</b> Layout where width is in the fastest changing dimension
<li><b>All:</b> Agnostic to any specific data layout
</ul>
where N = batches, C = channels, H = height, W = width
<table>
<caption id="multi_row"></caption>
<tr>
<th>Operator
<th>Description
<th>Equivalent Android NNAPI Operator
<th>Backends
<th>Data Layouts
<th>Data Types
<tr>
<td rowspan="3">AbsLayer
<td rowspan="3"> Layer to perform absolute operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_ABS
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ActivationLayer
<td rowspan="3" style="width:200px;"> Layer to simulate an activation layer with the specified activation function.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_ABS
<li>ANEURALNETWORKS_ELU
<li>ANEURALNETWORKS_HARD_SWISH
<li>ANEURALNETWORKS_LOGISTIC
<li>ANEURALNETWORKS_PRELU
<li>ANEURALNETWORKS_RELU
<li>ANEURALNETWORKS_RELU1
<li>ANEURALNETWORKS_RELU6
<li>ANEURALNETWORKS_SQRT
<li>ANEURALNETWORKS_TANH
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">AdditionLayer
<td rowspan="3" style="width:200px;"> Layer to add 2 tensors.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_ADD
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ArgMinMaxLayer
<td rowspan="3" style="width:200px;"> Layer to calculate the index of the minimum or maximum values in a tensor
based on an axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_ARGMAX
<li>ANEURALNETWORKS_ARGMIN
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>SIGNED64
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">BatchMatMulLayer
<td rowspan="3" style="width:200px;"> Layer to perform batch matrix multiplication.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">BatchNormalizationLayer
<td rowspan="3" style="width:200px;"> Layer to perform batch normalization.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
<tr><td>FLOAT16
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
<tr><td>FLOAT16
</table>
<tr>
<td rowspan="3">BatchToSpaceNdLayer
<td rowspan="3" style="width:200px;"> Layer to perform a batch to space transformation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_BATCH_TO_SPACE_ND
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">CastLayer
<td rowspan="3" style="width:200px;"> Layer to cast a tensor to a type.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_CAST
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QSYMMS8
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>FLOAT16
<tr><td>SIGNED32
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ChannelShuffleLayer
<td rowspan="3" style="width:200px;"> Layer to reorganize the channels of a tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_CHANNEL_SHUFFLE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QSYMMS8
<tr><td>QASYMMS8
<tr><td>QASYMMU8
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ComparisonLayer
<td rowspan="3" style="width:200px;"> Layer to compare 2 tensors.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_EQUAL
<li>ANEURALNETWORKS_GREATER
<li>ANEURALNETWORKS_GREATER_EQUAL
<li>ANEURALNETWORKS_LESS
<li>ANEURALNETWORKS_LESS_EQUAL
<li>ANEURALNETWORKS_NOT_EQUAL
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>BOOLEAN
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">ConcatLayer
<td rowspan="3" style="width:200px;"> Layer to concatenate tensors along a given axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_CONCATENATION
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ConstantLayer
<td rowspan="3" style="width:200px;"> Layer to provide a constant tensor.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">ConvertFp16ToFp32Layer
<td rowspan="3" style="width:200px;"> Layer to convert Float16 tensor to Float32 tensor.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ConvertFp32ToFp16Layer
<td rowspan="3" style="width:200px;"> Layer to convert Float32 tensor to Float16 tensor.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">Convolution2dLayer
<td rowspan="3" style="width:200px;"> Layer to compute a convolution operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_CONV_2D
<li>ANEURALNETWORKS_GROUPED_CONV_2D
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
</table>
<tr>
<td rowspan="3">Convolution3dLayer
<td rowspan="3" style="width:200px;"> Layer to compute a 3D convolution operation.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>NDHWC
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>N/A
</ul>
<td>
<ul>
<li>N/A
</ul>
<tr>
<td>GpuAcc
<td>
<ul>
<li>N/A
</ul>
<td>
<ul>
<li>N/A
</ul>
<tr>
<td rowspan="1">DebugLayer
<td rowspan="1" style="width:200px;"> Layer to print out inter layer tensor information.
<td rowspan="1">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td rowspan="3">DepthToSpaceLayer
<td rowspan="3" style="width:200px;"> Layer to perform Depth to Space transformation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_DEPTH_TO_SPACE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">DepthwiseConvolution2dLayer
<td rowspan="3" style="width:200px;"> Layer to compute a deconvolution or transpose convolution.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_DEPTHWISE_CONV_2D
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
</table>
<tr>
<td rowspan="3">DequantizeLayer
<td rowspan="3" style="width:200px;"> Layer to dequantize the values in a tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_DEQUANTIZE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td rowspan="2">DetectionPostProcessLayer
<td rowspan="2" style="width:200px;"> Layer to generate the detection output based on center size encoded boxes, class prediction and anchors by doing non maximum suppression (NMS).
<td rowspan="2">
<ul>
<li>ANEURALNETWORKS_DETECTION_POSTPROCESSING
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">DivisionLayer
<td rowspan="3" style="width:200px;"> Layer to divide 2 tensors.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_DIV
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ElementwiseBaseLayer
<td rowspan="3" style="width:200px;"> Layer to perform Add - Div - Max - Min - Mul operations.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_ADD
<li>ANEURALNETWORKS_DIV
<li>ANEURALNETWORKS_MAXIMUM
<li>ANEURALNETWORKS_MINIMUM
<li>ANEURALNETWORKS_MUL
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ElementwiseUnaryLayer
<td rowspan="3" style="width:200px;"> Layer to perform Rsqrt - Exp - Neg - Log - Abs - Sin - Sqrt operations.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_ABS
<li>ANEURALNETWORKS_EXP
<li>ANEURALNETWORKS_LOG
<li>ANEURALNETWORKS_NEG
<li>ANEURALNETWORKS_RSQRT
<li>ANEURALNETWORKS_SIN
<li>ANEURALNETWORKS_SQRT
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="1">FakeQuantizationLayer
<td rowspan="1" style="width:200px;"> Layer to quantize float values and dequantize afterwards. The current implementation does not dequantize the values.
<td rowspan="1">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">FillLayer
<td rowspan="3" style="width:200px;"> Layer to set the values of a tensor with a given value.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_FILL
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">FloorLayer
<td rowspan="3" style="width:200px;"> Layer to round the value to the lowest whole number.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_FLOOR
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
<tr><td>FLOAT16
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
<tr><td>FLOAT16
</table>
<tr>
<td rowspan="3">FullyConnectedLayer
<td rowspan="3" style="width:200px;"> Layer to perform a fully connected / dense operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_FULLY_CONNECTED
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
</table>
<tr>
<td rowspan="3">GatherLayer
<td rowspan="3" style="width:200px;"> Layer to perform the gather operation along the chosen axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_GATHER
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">GatherNdLayer
<td rowspan="3" style="width:200px;"> Layer to perform the gatherNd operation.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td rowspan="1">InputLayer
<td rowspan="1" style="width:200px;"> Special layer used to provide input data to the computational network.
<td rowspan="1">
<ul>
<li>N/A
</ul>
<td>All
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">InstanceNormalizationLayer
<td rowspan="3" style="width:200px;"> Layer to perform an instance normalization on a given axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_INSTANCE_NORMALIZATION
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">L2NormalizationLayer
<td rowspan="3" style="width:200px;"> Layer to perform an L2 normalization on a given axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_L2_NORMALIZATION
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">LogSoftmaxLayer
<td rowspan="3" style="width:200px;"> Layer to perform the log softmax activations given logits.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">LogicalBinaryLayer
<td rowspan="3" style="width:200px;"> Layer to perform Logical AND - Logical NOT - Logical OR operations.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_LOGICAL_AND
<li>ANEURALNETWORKS_LOGICAL_NOT
<li>ANEURALNETWORKS_LOGICAL_OR
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BOOLEAN
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BOOLEAN
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BOOLEAN
</table>
<tr>
<td rowspan="3">LstmLayer
<td rowspan="3" style="width:200px;"> Layer to perform a single time step in a Long Short-Term Memory (LSTM) operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_LSTM
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">MapLayer
<td rowspan="3" style="width:200px;"> Layer to perform map operation on tensor.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">MaximumLayer
<td rowspan="3" style="width:200px;"> Layer to perform an elementwise maximum of two tensors.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td rowspan="3">MeanLayer
<td rowspan="3" style="width:200px;"> Layer to perform reduce mean operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_MEAN
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">MemCopyLayer
<td rowspan="3" style="width:200px;"> Layer to perform memory copy operation.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>BOOLEAN
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">MemImportLayer
<td rowspan="3" style="width:200px;"> Layer to perform memory import operation.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">MergeLayer
<td rowspan="3" style="width:200px;"> Layer to concatenate tensors along a given axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_CONCATENATION
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">MinimumLayer
<td rowspan="3" style="width:200px;"> Layer to perform an elementwise minimum of two tensors.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_MINIMUM
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td rowspan="3">MultiplicationLayer
<td rowspan="3" style="width:200px;"> Layer to perform an elementwise multiplication of two tensors.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_MUL
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td rowspan="3">NormalizationLayer
<td rowspan="3" style="width:200px;"> Layer to compute normalization operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_LOCAL_RESPONSE_NORMALIZATION
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
<tr><td>FLOAT16
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT32
<tr><td>FLOAT16
</table>
<tr>
<td rowspan="1">OutputLayer
<td rowspan="1" style="width:200px;"> A special layer providing access to a user supplied buffer into which the output of a network can be written.
<td rowspan="1">
<ul>
<li>N/A
</ul>
<td>All
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">PadLayer
<td rowspan="3" style="width:200px;"> Layer to pad a tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_PAD
<li>ANEURALNETWORKS_PAD_V2
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">PermuteLayer
<td rowspan="3" style="width:200px;"> Layer to transpose an ND tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_TRANSPOSE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">Pooling2dLayer
<td rowspan="3" style="width:200px;"> Layer to perform 2D pooling with the specified pooling operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_AVERAGE_POOL_2D
<li>ANEURALNETWORKS_L2_POOL_2D
<li>ANEURALNETWORKS_MAX_POOL_2D
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">Pooling3dLayer
<td rowspan="3" style="width:200px;"> Layer to perform 3D pooling with the specified pooling operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_AVERAGE_POOL_3D
<li>ANEURALNETWORKS_L2_POOL_3D
<li>ANEURALNETWORKS_MAX_POOL_3D
</ul>
<td>CpuRef
<td>
<ul>
<li>NDHWC
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NA
</ul>
<td>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NDHWC
</ul>
<tr>
<td rowspan="1">PreCompiledLayer
<td rowspan="1" style="width:200px;"> Opaque layer provided by a backend which provides an executable representation of a subgraph from the original network.
<td rowspan="1">
<ul>
<li>N/A
</ul>
<td>N/A
<td>N/A
<td>N/A
<tr>
<td rowspan="3">PreluLayer
<td rowspan="3" style="width:200px;"> Layer to compute the activation layer with the PRELU activation function.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_PRELU
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">QLstmLayer
<td rowspan="3" style="width:200px;"> Layer to perform quantized LSTM (Long Short-Term Memory) operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_QUANTIZED_LSTM
<li>ANEURALNETWORKS_QUANTIZED_16BIT_LSTM
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>SIGNED32
<tr><td>QSYMMS16
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>SIGNED32
<tr><td>QSYMMS16
</table>
<tr>
<td rowspan="3">QuantizeLayer
<td rowspan="3" style="width:200px;"> Layer to perform quantization operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_QUANTIZE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QASYMM16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QASYMM16
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">QuantizedLstmLayer
<td rowspan="3" style="width:200px;"> Layer to perform quantized LSTM (Long Short-Term Memory) operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_QUANTIZED_LSTM
<li>ANEURALNETWORKS_QUANTIZED_16BIT_LSTM
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td rowspan="3">RankLayer
<td rowspan="3" style="width:200px;"> Layer to perform a rank operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_RANK
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">ReduceLayer
<td rowspan="3" style="width:200px;"> Layer to perform reduce with the following operations - ARG_IDX_MAX: Index of the max value - ARG_IDX_MIN: Index of the min value - MEAN_SUM: Mean of sum - PROD: Product - SUM_SQUARE: Sum of squares - SUM: Sum - MIN: Min - MAX: Max
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_REDUCE_MAX
<li>ANEURALNETWORKS_REDUCE_MIN
<li>ANEURALNETWORKS_REDUCE_SUM
<li>ANEURALNETWORKS_REDUCE_PROD
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td rowspan="3">ReshapeLayer
<td rowspan="3" style="width:200px;"> Layer to reshape a tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_RESHAPE
<li>ANEURALNETWORKS_SQUEEZE
<li>ANEURALNETWORKS_EXPAND_DIMS
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>BOOLEAN
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">ResizeLayer
<td rowspan="3" style="width:200px;"> Layer to perform resize of a tensor using one of the interpolation methods: - Bilinear - Nearest Neighbor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_RESIZE_BILINEAR
<li>ANEURALNETWORKS_RESIZE_NEAREST_NEIGHBOR
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">RsqrtLayer
<td rowspan="3" style="width:200px;"> Layer to perform Rsqrt operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_RSQRT
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>SIGNED32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">ShapeLayer
<td rowspan="3" style="width:200px;"> Layer to return the shape of the input tensor.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">SliceLayer
<td rowspan="3" style="width:200px;"> Layer to perform tensor slicing.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_SLICE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">SoftmaxLayer
<td rowspan="3" style="width:200px;"> Layer to perform softmax, log-softmax operation over the specified axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_LOG_SOFTMAX
<li>ANEURALNETWORKS_SOFTMAX
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">SpaceToBatchNdLayer
<td rowspan="3" style="width:200px;"> Layer to divide spatial dimensions of the tensor into a grid of blocks and interleaves these blocks with the batch dimension.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_SPACE_TO_BATCH_ND
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">SpaceToDepthLayer
<td rowspan="3" style="width:200px;"> Layer to rearrange blocks of spatial data into depth.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_SPACE_TO_DEPTH
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">SplitterLayer
<td rowspan="3" style="width:200px;"> Layer to split a tensor along a given axis.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_SPLIT
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">StackLayer
<td rowspan="3" style="width:200px;"> Layer to stack tensors along an axis.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="1">StandInLayer
<td rowspan="1" style="width:200px;"> A layer to represent "unknown" or "unsupported" operations in the input graph. It has a configurable number of input and output slots and an optional name.
<td rowspan="1">
<ul>
<li>N/A
</ul>
<td>N/A
<td>N/A
<td>N/A
<tr>
<td rowspan="3">StridedSliceLayer
<td rowspan="3" style="width:200px;"> Layer to extract a strided slice of a tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_STRIDED_SLICE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">SubtractionLayer
<td rowspan="3" style="width:200px;"> Layer to perform an elementwise subtract of 2 tensors.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_SUB
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
<tr><td>SIGNED32
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QSYMMS16
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">TransposeConvolution2dLayer
<td rowspan="3" style="width:200px;"> Layer to perform 2D transpose convolution (deconvolution) operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_TRANSPOSE_CONV_2D
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>SIGNED32
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMU8
<tr><td>QASYMMS8
<tr><td>QUANTIZEDSYMM8PERAXIS
</table>
<tr>
<td rowspan="3">TransposeLayer
<td rowspan="3" style="width:200px;"> Layer to transpose a tensor.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_TRANSPOSE
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>BFLOAT16
<tr><td>FLOAT16
<tr><td>FLOAT32
<tr><td>QASYMMS8
<tr><td>QASYMMU8
<tr><td>QSYMMS16
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td rowspan="3">UnidirectionalSquenceLstmLayer
<td rowspan="3" style="width:200px;"> Layer to perform unidirectional sequence LSTM operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_LSTM
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>Input Types
<tr><td>FLOAT32
</table>
<table>
<tr><th>Weight Types
<tr><td>FLOAT32
<tr><td>QASYMMS8
</table>
<td>CpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>Input Types
<tr><td>FLOAT32
</table>
<table>
<tr><th>Weight Types
<tr><td>FLOAT32
</table>
<td>GpuAcc
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>Input Types
<tr><td>FLOAT32
</table>
<table>
<tr><th>Weight Types
<tr><td>FLOAT32
</table>
<tr>
<td rowspan="3">UnmapLayer
<td rowspan="3" style="width:200px;"> Layer to perform unmap operation on tensor.
<td rowspan="3">
<ul>
<li>N/A
</ul>
<td>CpuRef
<td>
<ul>
<li>All
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>CpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
<tr>
<td>GpuAcc
<td>
<ul>
<li>NHWC
<li>NCHW
</ul>
<td>
<table>
<tr><th>
<tr><td>All
</table>
</table>
*/
} // namespace
Reference in new issue View Git Blame Copy Permalink