NeuralNetworks

#include <NeuralNetworks.h>

Summary

Enumerations

Anonymous Enum 39{
  ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES = 128
}
enum
For ANeuralNetworksModel_setOperandValue, values with a length smaller or equal to this will be immediately copied into the model.
FuseCode{
  ANEURALNETWORKS_FUSED_NONE = 0,
  ANEURALNETWORKS_FUSED_RELU = 1,
  ANEURALNETWORKS_FUSED_RELU1 = 2,
  ANEURALNETWORKS_FUSED_RELU6 = 3
}
enum
Fused activation function types.
OperandCode{
  ANEURALNETWORKS_FLOAT32 = 0,
  ANEURALNETWORKS_INT32 = 1,
  ANEURALNETWORKS_UINT32 = 2,
  ANEURALNETWORKS_TENSOR_FLOAT32 = 3,
  ANEURALNETWORKS_TENSOR_INT32 = 4,
  ANEURALNETWORKS_TENSOR_QUANT8_ASYMM = 5
}
enum
Operand types.
OperationCode{
  ANEURALNETWORKS_ADD = 0,
  ANEURALNETWORKS_AVERAGE_POOL_2D = 1,
  ANEURALNETWORKS_CONCATENATION = 2,
  ANEURALNETWORKS_CONV_2D = 3,
  ANEURALNETWORKS_DEPTHWISE_CONV_2D = 4,
  ANEURALNETWORKS_DEPTH_TO_SPACE = 5,
  ANEURALNETWORKS_DEQUANTIZE = 6,
  ANEURALNETWORKS_EMBEDDING_LOOKUP = 7,
  ANEURALNETWORKS_FLOOR = 8,
  ANEURALNETWORKS_FULLY_CONNECTED = 9,
  ANEURALNETWORKS_HASHTABLE_LOOKUP = 10,
  ANEURALNETWORKS_L2_NORMALIZATION = 11,
  ANEURALNETWORKS_L2_POOL_2D = 12,
  ANEURALNETWORKS_LOCAL_RESPONSE_NORMALIZATION = 13,
  ANEURALNETWORKS_LOGISTIC = 14,
  ANEURALNETWORKS_LSH_PROJECTION = 15,
  ANEURALNETWORKS_LSTM = 16,
  ANEURALNETWORKS_MAX_POOL_2D = 17,
  ANEURALNETWORKS_MUL = 18,
  ANEURALNETWORKS_RELU = 19,
  ANEURALNETWORKS_RELU1 = 20,
  ANEURALNETWORKS_RELU6 = 21,
  ANEURALNETWORKS_RESHAPE = 22,
  ANEURALNETWORKS_RESIZE_BILINEAR = 23,
  ANEURALNETWORKS_RNN = 24,
  ANEURALNETWORKS_SOFTMAX = 25,
  ANEURALNETWORKS_SPACE_TO_DEPTH = 26,
  ANEURALNETWORKS_SVDF = 27,
  ANEURALNETWORKS_TANH = 28,
  ANEURALNETWORKS_BATCH_TO_SPACE_ND = 29,
  ANEURALNETWORKS_DIV = 30,
  ANEURALNETWORKS_MEAN = 31,
  ANEURALNETWORKS_PAD = 32,
  ANEURALNETWORKS_SPACE_TO_BATCH_ND = 33,
  ANEURALNETWORKS_SQUEEZE = 34,
  ANEURALNETWORKS_STRIDED_SLICE = 35,
  ANEURALNETWORKS_SUB = 36,
  ANEURALNETWORKS_TRANSPOSE = 37
}
enum
Operation types.
PaddingCode{
  ANEURALNETWORKS_PADDING_SAME = 1,
  ANEURALNETWORKS_PADDING_VALID = 2
}
enum
Implicit padding algorithms.
PreferenceCode{
  ANEURALNETWORKS_PREFER_LOW_POWER = 0,
  ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER = 1,
  ANEURALNETWORKS_PREFER_SUSTAINED_SPEED = 2
}
enum
Execution preferences.
ResultCode{
  ANEURALNETWORKS_NO_ERROR = 0,
  ANEURALNETWORKS_OUT_OF_MEMORY = 1,
  ANEURALNETWORKS_INCOMPLETE = 2,
  ANEURALNETWORKS_UNEXPECTED_NULL = 3,
  ANEURALNETWORKS_BAD_DATA = 4,
  ANEURALNETWORKS_OP_FAILED = 5,
  ANEURALNETWORKS_BAD_STATE = 6,
  ANEURALNETWORKS_UNMAPPABLE = 7
}
enum
Result codes.

Typedefs

ANeuralNetworksCompilation typedef
ANeuralNetworksCompilation is an opaque type that can be used to compile a machine learning model.
ANeuralNetworksEvent typedef
ANeuralNetworksEvent is an opaque type that represents an event that will be signaled once an execution completes.
ANeuralNetworksExecution typedef
ANeuralNetworksExecution is an opaque type that can be used to apply a machine learning model to a set of inputs.
ANeuralNetworksMemory typedef
ANeuralNetworksMemory is an opaque type that represents memory.
ANeuralNetworksModel typedef
ANeuralNetworksModel is an opaque type that contains a description of the mathematical operations that constitute the model.
ANeuralNetworksOperandType typedef
ANeuralNetworksOperandType describes the type of an operand.
ANeuralNetworksOperationType typedef
int32_t

Functions

ANeuralNetworksCompilation_create(ANeuralNetworksModel *model, ANeuralNetworksCompilation **compilation)
int
Create a ANeuralNetworksCompilation to compile the given model.
ANeuralNetworksCompilation_finish(ANeuralNetworksCompilation *compilation)
int
Indicate that we have finished modifying a compilation.
ANeuralNetworksCompilation_free(ANeuralNetworksCompilation *compilation)
void
Destroy a compilation.
ANeuralNetworksCompilation_setPreference(ANeuralNetworksCompilation *compilation, int32_t preference)
int
Sets the execution preference.
ANeuralNetworksEvent_free(ANeuralNetworksEvent *event)
void
Destroys the event.
ANeuralNetworksEvent_wait(ANeuralNetworksEvent *event)
int
Waits until the execution completes.
ANeuralNetworksExecution_create(ANeuralNetworksCompilation *compilation, ANeuralNetworksExecution **execution)
int
Create a ANeuralNetworksExecution to apply the given compilation.
ANeuralNetworksExecution_free(ANeuralNetworksExecution *execution)
void
Destroy an execution.
ANeuralNetworksExecution_setInput(ANeuralNetworksExecution *execution, int32_t index, const ANeuralNetworksOperandType *type, const void *buffer, size_t length)
int
Associate a user buffer with an input of the model of the ANeuralNetworksExecution.
ANeuralNetworksExecution_setInputFromMemory(ANeuralNetworksExecution *execution, int32_t index, const ANeuralNetworksOperandType *type, const ANeuralNetworksMemory *memory, size_t offset, size_t length)
int
Associate part of a memory object with an input of the model of the ANeuralNetworksExecution.
ANeuralNetworksExecution_setOutput(ANeuralNetworksExecution *execution, int32_t index, const ANeuralNetworksOperandType *type, void *buffer, size_t length)
int
Associate a user buffer with an output of the model of the ANeuralNetworksExecution.
ANeuralNetworksExecution_setOutputFromMemory(ANeuralNetworksExecution *execution, int32_t index, const ANeuralNetworksOperandType *type, const ANeuralNetworksMemory *memory, size_t offset, size_t length)
int
Associate part of a memory object with an output of the model of the ANeuralNetworksExecution.
ANeuralNetworksExecution_startCompute(ANeuralNetworksExecution *execution, ANeuralNetworksEvent **event)
int
Schedule evaluation of the execution.
ANeuralNetworksMemory_createFromFd(size_t size, int protect, int fd, size_t offset, ANeuralNetworksMemory **memory)
int
Creates a shared memory object from a file descriptor.
ANeuralNetworksMemory_free(ANeuralNetworksMemory *memory)
void
Delete a memory object.
ANeuralNetworksModel_addOperand(ANeuralNetworksModel *model, const ANeuralNetworksOperandType *type)
int
Add an operand to a model.
ANeuralNetworksModel_addOperation(ANeuralNetworksModel *model, ANeuralNetworksOperationType type, uint32_t inputCount, const uint32_t *inputs, uint32_t outputCount, const uint32_t *outputs)
int
Add an operation to a model.
ANeuralNetworksModel_create(ANeuralNetworksModel **model)
int
Create an empty ANeuralNetworksModel.
ANeuralNetworksModel_finish(ANeuralNetworksModel *model)
int
Indicate that we have finished modifying a model.
ANeuralNetworksModel_free(ANeuralNetworksModel *model)
void
Destroy a model.
ANeuralNetworksModel_identifyInputsAndOutputs(ANeuralNetworksModel *model, uint32_t inputCount, const uint32_t *inputs, uint32_t outputCount, const uint32_t *outputs)
int
Specifies which operands will be the model's inputs and outputs.
ANeuralNetworksModel_relaxComputationFloat32toFloat16(ANeuralNetworksModel *model, bool allow)
int
Specifies whether ANEURALNETWORKS_TENSOR_FLOAT32 is allowed to be calculated with range and/or precision as low as that of the IEEE 754 16-bit floating-point format.
ANeuralNetworksModel_setOperandValue(ANeuralNetworksModel *model, int32_t index, const void *buffer, size_t length)
int
Sets an operand to a constant value.
ANeuralNetworksModel_setOperandValueFromMemory(ANeuralNetworksModel *model, int32_t index, const ANeuralNetworksMemory *memory, size_t offset, size_t length)
int
Sets an operand to a value stored in a memory object.

Structs

ANeuralNetworksOperandType

ANeuralNetworksOperandType describes the type of an operand.

Enumerations

Anonymous Enum 39

 Anonymous Enum 39

For ANeuralNetworksModel_setOperandValue, values with a length smaller or equal to this will be immediately copied into the model.

The size is in bytes.

Properties
ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES

FuseCode

 FuseCode

Fused activation function types.

Properties
ANEURALNETWORKS_FUSED_NONE

NO fused activation function.

ANEURALNETWORKS_FUSED_RELU

Fused ReLU activation function.

ANEURALNETWORKS_FUSED_RELU1

Fused ReLU1 activation function.

ANEURALNETWORKS_FUSED_RELU6

Fused ReLU6 activation function.

OperandCode

 OperandCode

Operand types.

The type of operands that can be added to a model.

Although we define many types, most operators accept just a few types. Most used are ANEURALNETWORKS_TENSOR_FLOAT32, ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, and ANEURALNETWORKS_INT32.

Properties
ANEURALNETWORKS_FLOAT32

A 32 bit floating point scalar value.

ANEURALNETWORKS_INT32

A signed 32 bit integer scalar value.

ANEURALNETWORKS_TENSOR_FLOAT32

A tensor of 32 bit floating point values.

ANEURALNETWORKS_TENSOR_INT32

A tensor of 32 bit integer values.

ANEURALNETWORKS_TENSOR_QUANT8_ASYMM

A tensor of 8 bit integers that represent real numbers.

Attached to this tensor are two numbers that can be used to convert the 8 bit integer to the real value and vice versa. These two numbers are:

  • scale: a 32 bit floating point value greater than zero.
  • zeroPoint: a 32 bit integer, in range [0, 255].

The formula is: real_value = (integer_value - zeroPoint) * scale.

ANEURALNETWORKS_UINT32

An unsigned 32 bit integer scalar value.

OperationCode

 OperationCode

Operation types.

The type of operations that can be added to a model.

Properties
ANEURALNETWORKS_ADD

Adds two tensors, element-wise.

Takes two input tensors of identical OperandCode and compatible dimensions. The output is the sum of both input tensors, optionally modified by an activation function.

Two dimensions are compatible when:

  1. they are equal, or
  2. one of them is 1

The size of the output is the maximum size along each dimension of the input operands. It starts with the trailing dimensions, and works its way forward.

Example:

input1.dimension = {4, 1, 2}
input2.dimension = {5, 4, 3, 1}
output.dimension = {5, 4, 3, 2}

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: A tensor.
  • 1: A tensor of the same OperandCode, and compatible dimensions as input0.
  • 2: An ANEURALNETWORKS_INT32 scalar, and has to be one of the FuseCode values. Specifies the activation to invoke on the result.

Outputs:

  • 0: The sum, a tensor of the same OperandCode as input0.

ANEURALNETWORKS_AVERAGE_POOL_2D

Performs a 2-D average pooling operation.

The output dimensions are functions of the filter dimensions, stride, and padding.

The values in the output tensor are computed as:

output[batch, row, col, channel] =
    sum_{i, j}(input[batch, row + i, col + j, channel]) / sum(1)

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" (i.e., Num_samples, Height, Width, and Channels) data layout.

Both explicit padding and implicit padding are supported.

Inputs (explicit padding):

Inputs (implicit padding):

Outputs:

  • 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth].

ANEURALNETWORKS_BATCH_TO_SPACE_ND

BatchToSpace for N-dimensional tensors.

This operation reshapes the batch dimension (dimension 0) into M + 1 dimensions of shape block_shape + [batch], interleaves these blocks back into the grid defined by the spatial dimensions [1, ..., M], to obtain a result with the same rank as the input.

This is the reverse of SpaceToBatch.

Supported tensor OperandCode:

Supported tensor rank: 4

Inputs:

  • 0: An n-D tensor, specifying the tensor to be reshaped
  • 1: A 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the block sizes for each spatial dimension of the input tensor. All values must be >= 1.

Outputs:

ANEURALNETWORKS_CONCATENATION

Concatenates the input tensors along the given dimension.

The input tensors must have identical OperandCode and the same dimensions except the dimension along the concatenation axis.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

Outputs:

  • 0: The output, a tensor of the same OperandCode as the input tensors. The output shape is [D0, D1, ..., sum(Daxis(i)), ..., Dm].

ANEURALNETWORKS_CONV_2D

Performs an 2-D convolution operation.

The CONV_2D op sweeps a 2-D filter that can mix channels together over a batch of images, applying the filter to each window of each image of the appropriate size.

The output dimensions are functions of the filter dimensions, stride, and padding.

The values in the output tensor are computed as:

output[batch, row, col, channel] =
    sum_{i, j} (
        input[batch, row + i, col + j, k] *
        filter[channel, row + i, col + j, k] +
        bias[channel]
    )

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Both explicit padding and implicit padding are supported.

Inputs (explicit padding):

Inputs (implicit padding):

Outputs:

  • 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth_out]. For output tensor of ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, the following condition must be satisfied: output_scale > input_scale * filter_scale.

ANEURALNETWORKS_DEPTHWISE_CONV_2D

Performs a depthwise 2-D convolution operation.

Given an input tensor of shape [batches, height, width, depth_in] and a filter tensor of shape [1, filter_height, filter_width, depth_out] containing depth_out convolutional filters of depth 1, DEPTHWISE_CONV applies a different filter to each input channel (expanding from 1 channel to channel_multiplier channels for each), then concatenates the results together.

The output has depth_out = depth_in * depth_multiplier channels. The output dimensions are functions of the filter dimensions, stride, and padding.

The values in the output tensor are computed as:

output[b, i, j, k * channel_multiplier + q] =
    sum_{di, dj} (
        input[b, strides[1] * i + di, strides[2] * j + dj, k] *
        filter[1, di, dj, k * channel_multiplier + q]
    )

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Both explicit padding and implicit padding are supported.

Inputs (explicit padding):

Inputs (implicit padding):

Outputs:

  • 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth_out]. For output tensor of ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, the following condition must be satisfied: output_scale > input_scale * filter_scale.

ANEURALNETWORKS_DEPTH_TO_SPACE

Rearranges data from depth into blocks of spatial data.

More specifically, this op outputs a copy of the input tensor where values from the depth dimension are moved in spatial blocks to the height and width dimensions. The value block_size indicates the input block size and how the data is moved.

Chunks of data of size block_size * block_size from depth are rearranged into non-overlapping blocks of size block_size x block_size.

The width of the output tensor is input_depth * block_size, whereas the height is input_height * block_size. The depth of the input tensor must be divisible by block_size * block_size

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Inputs:

  • 0: A 4-D tensor, of shape [batches, height, width, depth_in], specifying the input.
  • 1: An ANEURALNETWORKS_INT32 scalar, specifying the block_size. block_size must be >=1 and block_size * block_size must be a divisor of the input depth.

Outputs:

  • 0: The output 4-D tensor, of shape [batch, height*block_size, width*block_size, depth/(block_size*block_size)].

ANEURALNETWORKS_DEQUANTIZE

Dequantizes the input tensor.

The formula is:

output = (input - zeroPoint) * scale.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

Outputs:

ANEURALNETWORKS_DIV

Element-wise division of two tensors.

Takes two input tensors of identical OperandCode and compatible dimensions. The output is the result of dividing the first input tensor by the second, optionally modified by an activation function.

Two dimensions are compatible when:

  1. they are equal, or
  2. one of them is 1

The size of the output is the maximum size along each dimension of the input operands. It starts with the trailing dimensions, and works its way forward.

Example: input1.dimension = {4, 1, 2} input2.dimension = {5, 4, 3, 1} output.dimension = {5, 4, 3, 2}

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: An n-D tensor, specifying the first input.
  • 1: A tensor of the same OperandCode, and compatible dimensions as input0.
  • 2: An ANEURALNETWORKS_INT32 scalar, and has to be one of the FuseCode values. Specifies the activation to invoke on the result.

Outputs:

ANEURALNETWORKS_EMBEDDING_LOOKUP

Looks up sub-tensors in the input tensor.

This operator takes for input a tensor of values (Values) and a one-dimensional tensor of selection indices (Lookups). The output tensor is the concatenation of sub-tensors of Values as selected by Lookups.

Think of Values as being sliced along its first dimension: The entries in Lookups select which slices are concatenated together to create the output tensor.

For example, if Values has shape of [40, 200, 300] and Lookups has shape of [3], all three values found in Lookups are expected to be between 0 and 39. The resulting tensor must have shape of [3, 200, 300].

If a value in Lookups is out of bounds, the operation must fail and an error must be reported.

Inputs:

  • 0: Lookups. A 1-D tensor of ANEURALNETWORKS_TENSOR_INT32. The values are indices into the first dimension of Values.
  • 1: Values. An n-D tensor, where n >= 2, from which sub-tensors are extracted.

Output:

  • 0: A n-D tensor with the same rank and shape as the Values tensor, except for the first dimension which has the same size as Lookups' only dimension.

ANEURALNETWORKS_FLOOR

Computes element-wise floor() on the input tensor.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: A tensor.

Outputs:

  • 0: The output tensor, of the same OperandCode and dimensions as the input tensor.

ANEURALNETWORKS_FULLY_CONNECTED

Denotes a fully (densely) connected layer, which connects all elements in the input tensor with each element in the output tensor.

This layer implements the operation:

outputs = activation(inputs * weights’ + bias)

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor of at least rank 2, specifying the input. If rank is greater than 2, then it gets flattened to a 2-D Tensor. The (flattened) 2-D Tensor is reshaped (if necessary) to [batch_size, input_size], where "input_size" corresponds to the number of inputs to the layer, matching the second dimension of weights, and "batch_size" is calculated by dividing the number of elements by "input_size".
  • 1: A 2-D tensor, specifying the weights, of shape [num_units, input_size], where "num_units" corresponds to the number of output nodes.
  • 2: A 1-D tensor, of shape [num_units], specifying the bias. For input tensor of ANEURALNETWORKS_TENSOR_FLOAT32, the bias should also be of ANEURALNETWORKS_TENSOR_FLOAT32. For input tensor of ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, the bias should be of ANEURALNETWORKS_TENSOR_INT32, with zeroPoint of 0 and bias_scale == input_scale * filter_scale.
  • 3: An ANEURALNETWORKS_INT32 scalar, and has to be one of the FuseCode values. Specifies the activation to invoke on the result.

Outputs:

  • 0: The output tensor, of shape [batch_size, num_units]. For output tensor of ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, the following condition must be satisfied: output_scale > input_scale * filter_scale.

ANEURALNETWORKS_HASHTABLE_LOOKUP

Looks up sub-tensors in the input tensor using a key-value map.

This operator takes for input a tensor of values (Values), a one-dimensional tensor of selection values (Lookups) and a one-dimensional tensor that maps these values to Values indexes. The output tensor is the concatenation of sub-tensors of Values as selected by Lookups via Keys.

Think of Values as being sliced along its outer-most dimension. The output is a concatenation of selected slices, with one slice for each entry of Lookups. The slice selected is the one at the same index as the Maps entry that matches the value in Lookups.

For a hit, the corresponding sub-tensor of Values is included in the Output tensor. For a miss, the corresponding sub-tensor in Output must have zero values.

For example, if Values has shape of [40, 200, 300], Keys should have a shape of [40]. If Lookups tensor has shape of [3], three slices are being concatenated, so the resulting tensor must have the shape of [3, 200, 300]. If the first entry in Lookups has the value 123456, that value must be located in Keys tensor. If the sixth entry of Keys contains 123456, the sixth slice of Values must be selected. If no entry in Keys has 123456, a slice of zeroes must be concatenated.

Inputs:

  • 0: Lookups. A 1-D ANEURALNETWORKS_TENSOR_INT32 tensor with shape [ k ].
  • 1: Keys. A 1-D ANEURALNETWORKS_TENSOR_INT32 tensor with shape [ n ]; Keys and Values pair represent a map, i.e., the ith element in Keys (Keys[i]) is the key to select the ith sub-tensor in Values (Values[i]), where 0 <= i <= n-1. Keys tensor MUST be sorted in ascending order.
  • 2: Values. A tensor with shape of [ n, … ]; i.e., the first dimension must be n.

Outputs:

  • 0: Output. A tensor with shape [ k …].
  • 1: Hits. A boolean tensor with shape [ k ] indicates whether the lookup hits (True) or not (False). Stored as ANEURALNETWORKS_TENSOR_QUANT8_ASYMM with offset 0 and scale 1.0f. A non-zero byte represents True, a hit. A zero indicates otherwise.

ANEURALNETWORKS_L2_NORMALIZATION

Applies L2 normalization along the depth dimension.

The values in the output tensor are computed as:

output[batch, row, col, channel] =
    input[batch, row, col, channel] /
    sqrt(sum_{c} pow(input[batch, row, col, c], 2))

For input tensor with more dimensions, independently normalizes each 1-D slice along dimension dim.

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout (i.e., Num_samples, Height, Width, and Channels).

Inputs:

  • 0: A 4-D tensor, of shape [batches, height, width, depth].

Outputs:

  • 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth].

ANEURALNETWORKS_L2_POOL_2D

Performs an 2-D L2 pooling operation.

The output dimensions are functions of the filter dimensions, stride, and padding.

The values in the output tensor are computed as:

output[batch, row, col, channel] =
    sqrt(sum_{i, j} pow(input[batch, row + i, col + j, channel], 2) /
         sum(1))

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Both explicit padding and implicit padding are supported.

Inputs (explicit padding):

Inputs (implicit padding):

Outputs:

  • 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth].

ANEURALNETWORKS_LOCAL_RESPONSE_NORMALIZATION

Applies Local Response Normalization along the depth dimension.

The 4-D input tensor is treated as a 3-D array of 1-D vectors (along the last dimension), and each vector is normalized independently. Within a given vector, each component is divided by the weighted, squared sum of inputs within depth_radius.

The output is calculated using this formula:

sqr_sum[a, b, c, d] = sum(
    pow(input[a, b, c, d - depth_radius : d + depth_radius + 1], 2))
output = input / pow((bias + alpha * sqr_sum), beta)

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Inputs:

Outputs:

  • 0: The output tensor of same shape as input0.

ANEURALNETWORKS_LOGISTIC

Computes sigmoid activation on the input tensor element-wise.

The output is calculated using this formula:

output = 1 / (1 + exp(-input))

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor, specifying the input.

Outputs:

ANEURALNETWORKS_LSH_PROJECTION

Projects an input to a bit vector via locality senstive hashing.

Inputs:

  • 0: Hash functions. Dim.size == 2, DataType: Float. Tensor[0].Dim[0]: Number of hash functions. Tensor[0].Dim[1]: Number of seeds per hash functions. Tensor[0].Dim[1] <= 32 in sparse case.
  • 1: Input. Dim.size >= 1, no restriction on DataType.
  • 2: Weight. Optional. Dim.size == 1, DataType: Float. If not set, each input element is considered to have the same weight of 1.0. Tensor[1].Dim[0] == Tensor[2].Dim[0]
  • 3: Type: Sparse: Value LSHProjectionType_SPARSE(=1). Computed bit vector is considered to be sparse. Each output element is an int32 made up of multiple bits computed from hash functions.Dense: Value LSHProjectionType_DENSE(=2). Computed bit vector is considered to be dense. Each output element represents a bit and can take the value of either 0 or 1.

Outputs:

  • 0: If the projection type is sparse: Output.Dim == { Tensor[0].Dim[0] } A tensor of int32 that represents hash signatures. If the projection type is Dense: Output.Dim == { Tensor[0].Dim[0] * Tensor[0].Dim[1] } A flattened tensor that represents projected bit vectors.

ANEURALNETWORKS_LSTM

Performs a single time step in a Long Short-Term Memory (LSTM) layer.

The LSTM operation is described by the following equations.

\begin{eqnarray*} i_t =& \sigma(W_{xi}x_t+W_{hi}h_{t-1}+W_{ci}C_{t-1}+b_i) & \\ f_t =& \sigma(W_{xf}x_t+W_{hf}h_{t-1}+W_{cf}C_{t-1}+b_f) & \\ C_t =& clip(f_t \odot C_{t-1} + i_t \odot g(W_{xc}x_t+W_{hc}h_{t-1}+b_c),\ t_{cell}) & \\ o_t =& \sigma(W_{xo}x_t+W_{ho}h_{t-1}+W_{co}C_t+b_o) & \\ & & \\ & clip(W_{proj}(o_t \odot g(C_t))+b_{proj},\ t_{proj}) & if\ there\ is\ a\ projection; \\ h_t =& & \\ & o_t \odot g(C_t) & otherwise. \\ \end{eqnarray*} Where:

  • $x_t$ is the input,
  • $i_t$ is the input gate,
  • $f_t$ is the forget gate,
  • $C_t$ is the cell state,
  • $o_t$ is the output,
  • $h_t$ is the output state,
  • $\sigma$ is the logistic sigmoid function,
  • $g$ is the cell input and cell output activation function, usually $tahn$,
  • $W_{xi}$ is the input-to-input weight matrix,
  • $W_{hi}$ is the recurrent to input weight matrix,
  • $W_{ci}$ is the cell-to-input weight matrix,
  • $b_i$ is the input gate bias,
  • $W_{xf}$ is the input-to-forget weight matrix,
  • $W_{hf}$ is the recurrent-to-forget weight matrix,
  • $W_{cf}$ is the cell-to-forget weight matrix,
  • $b_f$ is the forget gate bias,
  • $W_{xc}$ is the input-to-cell weight matrix,
  • $W_{hc}$ is the recurrent-to-cell weight matrix,
  • $b_c$ is the cell bias,
  • $W_{xo}$ is the input-to-output weight matrix,
  • $W_{ho}$ is the recurrent-to-output weight matrix,
  • $W_{co}$ is the cell-to-output weight matrix,
  • $b_o$ is the output gate bias,
  • $W_{proj}$ is the projection weight matrix,
  • $b_{proj}$ is the projection bias,
  • $t_{cell}$ is the threshold for clipping the cell state, and
  • $t_{proj}$ is the threshold for clipping the projected output.
  • $\odot$ is the Hadamard product that takes two matrices and produces another matrix, each element of which is the product of the corresponding elements of the input matrices.

The operation has the following independently optional inputs:

  • The input-to-input weights ( $W_{xi}$), recurrent-to-input weights ( $W_{hi}$), cell-to-input ( $W_{ci}$) weights, and input gate bias ( $b_i$) either all have values, or none of them have values (i.e., all set to null). If they have no values, coupling of input and forget gates (CIFG) is used, in which case the input gate ( $i_t$) is calculated using the following equation instead. \begin{eqnarray*} i_t = 1 - f_t \end{eqnarray*}
  • The cell-to-forget weights ( $W_{cf}$) and cell-to-output weights ( $W_{co}$) either both have values or neither of them have values. If they have values, the peephole optimization is used. Additionally, if CIFG is not used, cell-to-input weights ( $W_{ci}$) is also required to have values for peephole optimization.
  • The projection weights ( $W_{proj}$) is required only for the recurrent projection layer, and should otherwise have no value.
  • The projection bias ( $b_{proj}$) may (but not required to) have a value if the recurrent projection layer exists, and should otherwise have no value.

References:

The default non-peephole non-CIFG implementation is based on: http://www.bioinf.jku.at/publications/older/2604.pdf S. Hochreiter and J. Schmidhuber. "Long Short-Term Memory". Neural Computation, 9(8):1735-1780, 1997.

The peephole implementation and projection layer is based on: https://research.google.com/pubs/archive/43905.pdf Hasim Sak, Andrew Senior, and Francoise Beaufays. "Long short-term memory recurrent neural network architectures for large scale acoustic modeling." INTERSPEECH, 2014. (However, the concept of peephole optimization was introduced in work prior to this paper.)

The coupling of input and forget gate (CIFG) is based on: http://arxiv.org/pdf/1503.04069.pdf Greff et al. "LSTM: A Search Space Odyssey"

Supported tensor OperandCode:

Inputs:

  • 0: The input ( $x_t$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input.
  • 1: The input-to-input weights ( $W_{xi}$). Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size], where “num_units” corresponds to the number of cell units.
  • 2: The input-to-forget weights ( $W_{xf}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size].
  • 3: The input-to-cell weights ( $W_{xc}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size].
  • 4: The input-to-output weights ( $W_{xo}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size].
  • 5: The recurrent-to-input weights ( $W_{hi}$). Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, output_size], where “output_size” corresponds to either the number of cell units (i.e., “num_units”), or the second dimension of the “projection_weights”, if defined.
  • 6: The recurrent-to-forget weights ( $W_{hf}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, output_size].
  • 7: The recurrent-to-cell weights ( $W_{hc}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, output_size].
  • 8: The recurrent-to-output weights ( $W_{ho}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, output_size].
  • 9: The cell-to-input weights ( $W_{ci}$). Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 10:The cell-to-forget weights ( $W_{cf}$). Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 11:The cell-to-output weights ( $W_{co}$). Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 12:The input gate bias ( $b_i$). Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 13:The forget gate bias ( $b_f$). A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 14:The cell bias ( $b_c$). A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 15:The output gate bias ( $b_o$). A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 16:The projection weights ( $W_{proj}$). Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size, num_units].
  • 17:The projection bias ( $b_{proj}$). Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size].
  • 18:The output state (in) ( $h_{t-1}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size].
  • 19:The cell state (in) ( $C_{t-1}$). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units].
  • 20:The activation function ( $g$). A value indicating the activation function:
    • 0: None;
    • 1: Relu;
    • 3: Relu6;
    • 4: Tanh;
    • 6: Sigmoid.
  • 21:The clipping threshold ( $t_{cell}$) for the cell state, such that values are bound within [-cell_clip, cell_clip]. If set to 0.0 then clipping is disabled.
  • 22:The clipping threshold ( $t_{proj}$) for the output from the projection layer, such that values are bound within [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled.

Outputs:

ANEURALNETWORKS_MAX_POOL_2D

Performs an 2-D max pooling operation.

The output dimensions are functions of the filter dimensions, stride, and padding.

The values in the output tensor are computed as:

output[batch, row, col, channel] =
    max_{i, j} (input[batch, row + i, col + j, channel])

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Both explicit padding and implicit padding are supported.

Inputs (explicit padding):

Inputs (implicit padding):

Outputs:

  • 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth].

ANEURALNETWORKS_MEAN

Computes the mean of elements across dimensions of a tensor.

Reduces the input tensor along the given dimensions to reduce. Unless keep_dims is true, the rank of the tensor is reduced by 1 for each entry in axis. If keep_dims is true, the reduced dimensions are retained with length 1.

If dimensions to reduce have no entries, all dimensions are reduced, and a tensor with a single element is returned.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: A tensor, specifying the input.
  • 1: A 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32. The dimensions to reduce. If None (the default), reduces all dimensions. Must be in the range [-rank(input_tensor), rank(input_tensor)).
  • 2: An ANEURALNETWORKS_INT32 scalar, keep_dims. If positive, retains reduced dimensions with length 1.

Outputs:

ANEURALNETWORKS_MUL

Multiplies two tensors, element-wise.

Takes two input tensors of identical OperandCode and compatible dimensions. The output is the product of both input tensors, optionally modified by an activation function.

Two dimensions are compatible when:

  1. they are equal, or
  2. one of them is 1

The size of the resulting output is the maximum size along each dimension of the input operands. It starts with the trailing dimensions, and works its way forward.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: A tensor.
  • 1: A tensor of the same OperandCode, and compatible dimensions as input0.
  • 2: An ANEURALNETWORKS_INT32 scalar, and has to be one of the FuseCode values. Specifies the activation to invoke on the result.

Outputs:

ANEURALNETWORKS_PAD

Pads a tensor.

This operation pads a tensor according to the specified paddings.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: An n-D tensor, specifying the tensor to be padded.
  • 1: A 2-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the paddings for each spatial dimension of the input tensor. The shape of the tensor must be {rank(input0), 2}. padding[i, 0] specifies the number of element to be padded in the front of dimension i. padding[i, 1] specifies the number of element to be padded after the end of dimension i.

Outputs:

ANEURALNETWORKS_RELU

Computes rectified linear activation on the input tensor element-wise.

The output is calculated using this formula:

output = max(0, input)

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor, specifying the input.

Outputs:

  • 0: The output tensor of same shape as input0.

ANEURALNETWORKS_RELU1

Computes rectified linear 1 activation on the input tensor element-wise.

The output is calculated using this formula:

output = min(1.f, max(-1.f, input))

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor, specifying the input.

Outputs:

  • 0: The output tensor of same shape as input0.

ANEURALNETWORKS_RELU6

Computes rectified linear 6 activation on the input tensor element-wise.

The output is calculated using this formula:

output = min(6, max(0, input))

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor, specifying the input.

Outputs:

  • 0: The output tensor of same shape as input0.

ANEURALNETWORKS_RESHAPE

Reshapes a tensor.

Given tensor, this operation returns a tensor that has the same values as tensor, but with a newly specified shape.

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor, specifying the tensor to be reshaped.
  • 1: A 1-D tensor of ANEURALNETWORKS_TENSOR_INT32, defining the shape of the output tensor. The number of elements implied by shape must be the same as the number of elements in the input tensor.

Outputs:

  • 0: The output tensor, of shape specified by the input shape.

ANEURALNETWORKS_RESIZE_BILINEAR

Resizes images to given size using the bilinear interpretation.

Resized images must be distorted if their output aspect ratio is not the same as input aspect ratio. The corner pixels of output may not be the same as corner pixels of input.

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Inputs:

  • 0: A 4-D tensor, of shape [batches, height, width, depth], specifying the input.
  • 1: An ANEURALNETWORKS_INT32 scalar, specifying the output height of the output tensor.
  • 2: An ANEURALNETWORKS_INT32 scalar, specifying the output width of the output tensor.

Outputs:

  • 0: The output 4-D tensor, of shape [batches, new_height, new_width, depth].

ANEURALNETWORKS_RNN

A basic recurrent neural network layer.

This layer implements the operation: outputs = state = activation(inputs * input_weights + state * recurrent_weights + bias)

Where:

  • “input_weights” is a weight matrix that multiplies the inputs;
  • “recurrent_weights” is a weight matrix that multiplies the current “state” which itself is the output from the previous time step computation;
  • “bias” is a bias vector (added to each output vector in the batch);
  • “activation” is the function passed as the “fused_activation_function” argument (if not “NONE”).

Supported tensor OperandCode:

Inputs:

  • 0: input. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32 of shape [batch_size, input_size], where “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input.
  • 1: weights. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size], where “num_units” corresponds to the number of units.
  • 2: recurrent_weights. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, num_units], with columns corresponding to the weights from each unit.
  • 3: bias. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 4: hidden state (in). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units].
  • 5: fused_activation_function. An optional FuseCode value indicating the activation function. If “NONE” is specified then it results in a linear activation.

Outputs:

ANEURALNETWORKS_SOFTMAX

Computes the softmax activation on the input tensor element-wise, per batch, by normalizing the input vector so the maximum coefficient is zero.

The output is calculated using this formula:

output[batch, i] =
    exp((input[batch, i] - max(input[batch, :])) * beta) /
    sum_{k}{exp((input[batch, k] - max(input[batch, :])) * beta)}

Supported tensor OperandCode:

Supported tensor rank: 2 or 4.

Inputs:

  • 0: A 2-D or 4-D tensor, specifying the tensor to be reshaped.
  • 1: An ANEURALNETWORKS_FLOAT32 scalar, specifying the positive scaling factor for the exponent, beta.

Outputs:

ANEURALNETWORKS_SPACE_TO_BATCH_ND

SpaceToBatch for N-Dimensional tensors.

This operation divides "spatial" dimensions [1, ..., M] of the input into a grid of blocks of shape block_shape, and interleaves these blocks with the "batch" dimension (0) such that in the output, the spatial dimensions [1, ..., M] correspond to the position within the grid, and the batch dimension combines both the position within a spatial block and the original batch position. Prior to division into blocks, the spatial dimensions of the input are optionally zero padded according to paddings.

Supported tensor OperandCode:

Supported tensor rank: 4

Inputs:

  • 0: An n-D tensor, specifying the input.
  • 1: A 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the block sizes for each spatial dimension of the input tensor. All values must be >= 1.
  • 2: A 2-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the paddings for each spatial dimension of the input tensor. All values must be >= 0. The shape of the tensor must be {rank(input0), 2}. padding[i, 0] specifies the number of element to be padded in the front of dimension i. padding[i, 1] specifies the number of element to be padded after the end of dimension i.

Outputs:

ANEURALNETWORKS_SPACE_TO_DEPTH

Rearranges blocks of spatial data, into depth.

More specifically, this op outputs a copy of the input tensor where values from the height and width dimensions are moved to the depth dimension. The value block_size indicates the input block size and how the data is moved.

Chunks of data of size block_size * block_size from depth are rearranged into non-overlapping blocks of size block_size x block_size.

The depth of the output tensor is input_depth * block_size * block_size. The input tensor's height and width must be divisible by block_size.

Supported tensor OperandCode:

Supported tensor rank: 4, with "NHWC" data layout.

Inputs:

  • 0: A 4-D tensor, of shape [batches, height, width, depth_in], specifying the input.
  • 1: An ANEURALNETWORKS_INT32 scalar, specifying the block_size. block_size must be >=1 and block_size must be a divisor of both the input height and width.

Outputs:

  • 0: The output 4-D tensor, of shape [batch, height/block_size, width/block_size, depth*block_size*block_size].

ANEURALNETWORKS_SQUEEZE

Removes dimensions of size 1 from the shape of a tensor.

Given a tensor input, this operation returns a tensor of the same OperandCode with all dimensions of size 1 removed. If you don't want to remove all size 1 dimensions, you can remove specific size 1 dimensions by specifying the axes (input1).

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: An n-D tensor, the tensor to be squeezed.
  • 1: An optional 1-D tensor of ANEURALNETWORKS_TENSOR_INT32. The dimensions to squeeze. If specified only squeezes the dimensions listed. Otherwise, squeezes all dimensions. The dimension index starts at 0. An error must be reported if squeezing a dimension that is not 1.

Outputs:

  • 0: A tensor of the same OperandCode as input0. Contains the same data as input, but has one or more dimensions of size 1 removed.

ANEURALNETWORKS_STRIDED_SLICE

Extracts a strided slice of a tensor.

Roughly speaking, this op extracts a slice of size (end - begin) / stride from the given input tensor. Starting at the location specified by begin the slice continues by adding stride to the index until all dimensions are not less than end. Note that a stride can be negative, which causes a reverse slice.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: An n-D tensor, specifying the tensor to be sliced.
  • 1: A 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the starts of the dimensions of the input tensor to be sliced. The length must be of rank(input0).
  • 2: A 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the ends of the dimensions of the input tensor to be sliced. The length must be of rank(input0).
  • 3: A 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the strides of the dimensions of the input tensor to be sliced. The length must be of rank(input0).
  • 4: An ANEURALNETWORKS_INT32 scalar, begin_mask. If the ith bit of begin_mask is set, begin[i] is ignored and the fullest possible range in that dimension is used instead.
  • 5: An ANEURALNETWORKS_INT32 scalar, end_mask. If the ith bit of end_mask is set, end[i] is ignored and the fullest possible range in that dimension is used instead.
  • 6: An ANEURALNETWORKS_INT32 scalar, shrink_axis_mask. An int32 mask. If the ith bit of shrink_axis_mask is set, it implies that the ith specification shrinks the dimensionality by 1. A slice of size 1 starting from begin[i] in the dimension must be preserved.

Outputs:

ANEURALNETWORKS_SUB

Element-wise subtraction of two tensors.

Takes two input tensors of identical OperandCode and compatible dimensions. The output is the result of subtracting the second input tensor from the first one, optionally modified by an activation function.

Two dimensions are compatible when:

  1. they are equal, or
  2. one of them is 1

The size of the output is the maximum size along each dimension of the input operands. It starts with the trailing dimensions, and works its way forward.

Example: input1.dimension = {4, 1, 2} input2.dimension = {5, 4, 3, 1} output.dimension = {5, 4, 3, 2}

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: An n-D tensor, specifying the first input.
  • 1: A tensor of the same OperandCode, and compatible dimensions as input0.
  • 2: An ANEURALNETWORKS_INT32 scalar, and has to be one of the FuseCode values. Specifies the activation to invoke on the result.

Outputs:

ANEURALNETWORKS_SVDF

SVDF op is a kind of stateful layer derived from the notion that a densely connected layer that's processing a sequence of input frames can be approximated by using a singular value decomposition of each of its nodes.

The implementation is based on:

https://research.google.com/pubs/archive/43813.pdf

P. Nakkiran, R. Alvarez, R. Prabhavalkar, C. Parada. “Compressing Deep Neural Networks using a Rank-Constrained Topology”. INTERSPEECH, 2015.

It processes the incoming input using a 2-stage filtering mechanism:

  • stage 1 performs filtering on the "features" dimension, whose outputs get pushed into a memory of fixed-size memory_size.
  • stage 2 performs filtering on the "time" dimension of the memory_size memoized outputs of stage 1.

Specifically, for rank 1, this layer implements the operation:

memory = push(conv1d(inputs, weights_feature, feature_dim,
                     "ANEURALNETWORKS_PADDING_VALID"));
outputs = activation(memory * weights_time + bias);

Where:

  • “weights_feature” is a weights matrix that processes the inputs (by convolving the input with every “feature filter”), and whose outputs get pushed, stacked in order, into the fixed-size “memory” (the oldest entry gets dropped);
  • “weights_time” is a weights matrix that processes the “memory” (by a batched matrix multiplication on the num_units);
  • “bias” is an optional bias vector (added to each output vector in the batch); and
  • “activation” is the function passed as the “fused_activation_function” argument (if not “NONE”).

Each rank adds a dimension to the weights matrices by means of stacking the filters.

Supported tensor OperandCode:

Inputs:

  • 0: input. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input.
  • 1: weights_feature. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size], where “num_units” corresponds to the number of units.
  • 2: weights_time. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, memory_size], where “memory_size” corresponds to the fixed-size of the memory.
  • 3: bias. An optional 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units].
  • 4: state (in). A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, (memory_size - 1) * num_units * rank].
  • 5: rank. The rank of the SVD approximation.
  • 6: fused_activation_function. An optional FuseCode value indicating the activation function. If “NONE” is specified then it results in a linear activation.

Outputs:

ANEURALNETWORKS_TANH

Computes hyperbolic tangent of input tensor element-wise.

The output is calculated using this formula:

output = tanh(input)

Supported tensor OperandCode:

Supported tensor rank: up to 4.

Inputs:

  • 0: A tensor, specifying the input.

Outputs:

  • 0: The output tensor of same shape as input0.

ANEURALNETWORKS_TRANSPOSE

Transposes the input tensor, permuting the dimensions according to the perm tensor.

The returned tensor's dimension i corresponds to the input dimension perm[i]. If perm is not given, it is set to (n-1...0), where n is the rank of the input tensor. Hence by default, this operation performs a regular matrix transpose on 2-D input Tensors.

Supported tensor OperandCode:

Supported tensor rank: up to 4

Inputs:

  • 0: An n-D tensor, specifying the tensor to be transposed.
  • 1: An optional 1-D Tensor of ANEURALNETWORKS_TENSOR_INT32, the permutation of the dimensions of the input tensor.

Outputs:

PaddingCode

 PaddingCode

Implicit padding algorithms.

Properties
ANEURALNETWORKS_PADDING_SAME

SAME padding.

Padding on both ends are the "same": padding_to_beginning = total_padding / 2 padding_to_end = (total_padding + 1)/2. i.e., for even number of padding, padding to both ends are exactly the same; for odd number of padding, padding to the ending is bigger than the padding to the beginning by 1.

total_padding is a function of input, stride and filter size. It could be computed as follows: out_size = (input + stride - 1) / stride; needed_input = (out_size - 1) * stride + filter_size total_padding = max(0, needed_input - output_size) The computation is the same for the horizontal and vertical directions.

ANEURALNETWORKS_PADDING_VALID

VALID padding.

No padding. When the input size is not evenly divisible by the filter size, the input at the end that could not fill the whole filter tile will simply be ignored.

PreferenceCode

 PreferenceCode

Execution preferences.

Properties
ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER

Prefer returning a single answer as fast as possible, even if this causes more power consumption.

ANEURALNETWORKS_PREFER_LOW_POWER

Prefer executing in a way that minimizes battery drain.

This is desirable for compilations that will be executed often.

ANEURALNETWORKS_PREFER_SUSTAINED_SPEED

Prefer maximizing the throughput of successive frames, for example when processing successive frames coming from the camera.

Typedefs

ANeuralNetworksCompilation

struct ANeuralNetworksCompilation ANeuralNetworksCompilation

ANeuralNetworksCompilation is an opaque type that can be used to compile a machine learning model.

To use:

A compilation is completed by calling ANeuralNetworksCompilation_finish. A compilation is destroyed by calling ANeuralNetworksCompilation_free.

A compilation cannot be modified once ANeuralNetworksCompilation_finish has been called on it.

It is the application's responsibility to make sure that only one thread modifies a compilation at a given time. It is however safe for more than one thread to use the compilation once ANeuralNetworksCompilation_finish has returned.

It is also the application's responsibility to ensure that there are no other uses of the compilation after calling ANeuralNetworksCompilation_free. This includes any execution object created using the compilation.

ANeuralNetworksEvent

struct ANeuralNetworksEvent ANeuralNetworksEvent

ANeuralNetworksEvent is an opaque type that represents an event that will be signaled once an execution completes.

ANeuralNetworksExecution

struct ANeuralNetworksExecution ANeuralNetworksExecution

ANeuralNetworksExecution is an opaque type that can be used to apply a machine learning model to a set of inputs.

To use:

An execution cannot be modified once ANeuralNetworksExecution_startCompute has been called on it.

An execution can be applied to a model with ANeuralNetworksExecution_startCompute only once. Create new executions to do new evaluations of the model.

It is the application's responsibility to make sure that only one thread modifies an execution at a given time. It is however safe for more than one thread to use ANeuralNetworksEvent_wait at the same time.

It is also the application's responsibility to ensure that there are no other uses of the request after calling ANeuralNetworksExecution_free.

ANeuralNetworksMemory

struct ANeuralNetworksMemory ANeuralNetworksMemory

ANeuralNetworksMemory is an opaque type that represents memory.

This type is used to represent shared memory, memory mapped files, and similar memories.

By using shared memory, a program can efficiently communicate to the runtime and drivers the tensors that define a model. See ANeuralNetworksModel_setOperandValueFromMemory. An application should typically create one shared memory object that contains every tensor needed to define a model. ANeuralNetworksMemory_createFromFd can be used to create shared memory from a file handle.

Memory objects can also be used to specify the input and output arguments of an execution. See ANeuralNetworksExecution_setInputFromMemory and ANeuralNetworksExecution_setOutputFromMemory.

ANeuralNetworksModel

struct ANeuralNetworksModel ANeuralNetworksModel

ANeuralNetworksModel is an opaque type that contains a description of the mathematical operations that constitute the model.

Build the model by calling

A model is completed by calling ANeuralNetworksModel_finish. A model is destroyed by calling ANeuralNetworksModel_free.

A model cannot be modified once ANeuralNetworksModel_finish has been called on it.

It is the application's responsibility to make sure that only one thread modifies a model at a given time. It is however safe for more than one thread to use the model once ANeuralNetworksModel_finish has returned.

It is also the application's responsibility to ensure that there are no other uses of the model after calling ANeuralNetworksModel_free. This includes any compilation or execution object created using the model.

ANeuralNetworksOperandType

struct ANeuralNetworksOperandType ANeuralNetworksOperandType

ANeuralNetworksOperandType describes the type of an operand.

This structure is used to describe both scalars and tensors.

A tensor operand type must have a specified rank (number of dimensions) but may have any of its dimensions unspecified.

A tensor operand type with all dimensions specified is "fully specified". Whenever possible (i.e., whenever the dimensions are known at model construction time), a tensor operand type should be (but is not required to be) fully specified, in order to enable the best possible performance.

If a tensor operand's type is not fully specified, the dimensions of the operand are deduced from the operand types and values of the operation for which that operand is an output.

In the following situations, a tensor operand type must be fully specified:

A tensor operand type with some number of unspecified dimensions is represented by setting each unspecified dimension to 0.

ANeuralNetworksOperationType

int32_t ANeuralNetworksOperationType

Functions

ANeuralNetworksCompilation_create

int ANeuralNetworksCompilation_create(
  ANeuralNetworksModel *model,
  ANeuralNetworksCompilation **compilation
)

Create a ANeuralNetworksCompilation to compile the given model.

This only creates the object. Compilation is only performed once ANeuralNetworksCompilation_finish is invoked.

ANeuralNetworksCompilation_finish should be called once all desired properties have been set on the compilation.

ANeuralNetworksModel_free should be called once the compilation is no longer needed.

The provided model must outlive the compilation.

The model must already have been finished by a call to ANeuralNetworksModel_finish.

See ANeuralNetworksCompilation for information on multithreaded usage.

Details
Parameters
model
The ANeuralNetworksModel to be compiled.
compilation
The newly created object or NULL if unsuccessful.
Returns
ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the model is invalid.

ANeuralNetworksCompilation_finish

int ANeuralNetworksCompilation_finish(
  ANeuralNetworksCompilation *compilation
)

Indicate that we have finished modifying a compilation.

Required before calling ANeuralNetworksExecution_create.

An application is responsible to make sure that no other thread uses the compilation at the same time.

This function must only be called once for a given compilation.

See ANeuralNetworksCompilation for information on multithreaded usage.

Details
Parameters
compilation
The compilation to be finished.
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksCompilation_free

void ANeuralNetworksCompilation_free(
  ANeuralNetworksCompilation *compilation
)

Destroy a compilation.

The compilation need not have been finished by a call to ANeuralNetworksModel_finish.

See ANeuralNetworksCompilation for information on multithreaded usage.

Details
Parameters
compilation
The compilation to be destroyed. Passing NULL is acceptable and results in no operation.

ANeuralNetworksCompilation_setPreference

int ANeuralNetworksCompilation_setPreference(
  ANeuralNetworksCompilation *compilation,
  int32_t preference
)

Sets the execution preference.

Provides guidance to the runtime when trade-offs are possible.

See ANeuralNetworksCompilation for information on multithreaded usage.

Details
Parameters
compilation
The compilation to be modified.
preference
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksEvent_free

void ANeuralNetworksEvent_free(
  ANeuralNetworksEvent *event
)

Destroys the event.

See ANeuralNetworksExecution for information on multithreaded usage.

ANeuralNetworksEvent_wait

int ANeuralNetworksEvent_wait(
  ANeuralNetworksEvent *event
)

Waits until the execution completes.

More than one thread can wait on an event. When the execution completes, all threads will be released.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Returns
ANEURALNETWORKS_NO_ERROR if the execution completed normally.

ANeuralNetworksExecution_create

int ANeuralNetworksExecution_create(
  ANeuralNetworksCompilation *compilation,
  ANeuralNetworksExecution **execution
)

Create a ANeuralNetworksExecution to apply the given compilation.

This only creates the object. Computation is only performed once ANeuralNetworksExecution_startCompute is invoked.

The provided compilation must outlive the execution.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
compilation
The ANeuralNetworksCompilation to be evaluated.
execution
The newly created object or NULL if unsuccessful.
Returns
ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the compilation is invalid.

ANeuralNetworksExecution_free

void ANeuralNetworksExecution_free(
  ANeuralNetworksExecution *execution
)

Destroy an execution.

If called on an execution for which ANeuralNetworksExecution_startCompute has been called, the function will return immediately but will mark the execution to be deleted once the computation completes. The related ANeuralNetworksEvent will be signaled and the ANeuralNetworksEvent_wait will return ANEURALNETWORKS_ERROR_DELETED.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
execution
The execution to be destroyed. Passing NULL is acceptable and results in no operation.

ANeuralNetworksExecution_setInput

int ANeuralNetworksExecution_setInput(
  ANeuralNetworksExecution *execution,
  int32_t index,
  const ANeuralNetworksOperandType *type,
  const void *buffer,
  size_t length
)

Associate a user buffer with an input of the model of the ANeuralNetworksExecution.

The provided buffer must outlive the execution.

If the input is optional, you can indicate that it is omitted by passing nullptr for buffer and 0 for length.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
execution
The execution to be modified.
index
The index of the input argument we are setting. It is an index into the lists passed to ANeuralNetworksModel_identifyInputsAndOutputs. It is not the index associated with ANeuralNetworksModel_addOperand.
type
The ANeuralNetworksOperandType of the operand. Unless the input is omitted, this should be used to specify the dimensions that were left unspecified when the operand was added to the model. All other properties of the type must be the same as specified in the model. If the type is the same as specified when the model was built, NULL can be passed.
buffer
The buffer containing the data.
length
The length in bytes of the buffer.
Returns
ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the name is not recognized or the buffer is too small for the input.

ANeuralNetworksExecution_setInputFromMemory

int ANeuralNetworksExecution_setInputFromMemory(
  ANeuralNetworksExecution *execution,
  int32_t index,
  const ANeuralNetworksOperandType *type,
  const ANeuralNetworksMemory *memory,
  size_t offset,
  size_t length
)

Associate part of a memory object with an input of the model of the ANeuralNetworksExecution.

The provided memory must outlive the execution.

If the input is optional, you can indicate that it is omitted by using ANeuralNetworks_setInput instead, passing nullptr for buffer and 0 for length.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
execution
The execution to be modified.
index
The index of the input argument we are setting. It is an index into the lists passed to ANeuralNetworksModel_identifyInputsAndOutputs. It is not the index associated with ANeuralNetworksModel_addOperand.
type
The ANeuralNetworksOperandType of the operand. This should be used to specify the dimensions that were left unspecified when the operand was added to the model. All other properties of the type must be the same as specified in the model. If the type is the same as specified when the model was built, NULL can be passed.
memory
The memory containing the data.
offset
This specifies the location of the data within the memory. The offset is in bytes from the start of memory.
length
The size in bytes of the data value.
Returns
ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the name is not recognized or the buffer is too small for the input.

ANeuralNetworksExecution_setOutput

int ANeuralNetworksExecution_setOutput(
  ANeuralNetworksExecution *execution,
  int32_t index,
  const ANeuralNetworksOperandType *type,
  void *buffer,
  size_t length
)

Associate a user buffer with an output of the model of the ANeuralNetworksExecution.

If the output is optional, you can indicate that it is omitted by passing nullptr for buffer and 0 for length.

The provided buffer must outlive the execution.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
execution
The execution to be modified.
index
The index of the output argument we are setting. It is an index into the lists passed to ANeuralNetworksModel_identifyInputsAndOutputs. It is not the index associated with ANeuralNetworksModel_addOperand.
type
The ANeuralNetworksOperandType of the operand. Unless the output is omitted, this should be used to specify the dimensions that were left unspecified when the operand was added to the model. All other properties of the type must be the same as specified in the model. If the type is the same as specified when the model was built, NULL can be passed.
buffer
The buffer where the data is to be written.
length
The length in bytes of the buffer.
Returns
ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the name is not recognized or the buffer is too small for the output.

ANeuralNetworksExecution_setOutputFromMemory

int ANeuralNetworksExecution_setOutputFromMemory(
  ANeuralNetworksExecution *execution,
  int32_t index,
  const ANeuralNetworksOperandType *type,
  const ANeuralNetworksMemory *memory,
  size_t offset,
  size_t length
)

Associate part of a memory object with an output of the model of the ANeuralNetworksExecution.

If the output is optional, you can indicate that it is omitted by using ANeuralNetworks_setOutput instead, passing nullptr for buffer and 0 for length.

The provided memory must outlive the execution.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
execution
The execution to be modified.
index
The index of the output argument we are setting. It is an index into the lists passed to ANeuralNetworksModel_identifyInputsAndOutputs. It is not the index associated with ANeuralNetworksModel_addOperand.
type
The ANeuralNetworksOperandType of the operand. This should be used to specify the dimensions that were left unspecified when the operand was added to the model. All other properties of the type must be the same as specified in the model. If the type is the same as specified when the model was built, NULL can be passed.
memory
The memory where the data is to be stored.
offset
This specifies the location of the data within the memory. The offset is in bytes from the start of memory.
length
The length in bytes of the data value.
Returns
ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the name is not recognized or the buffer is too small for the output.

ANeuralNetworksExecution_startCompute

int ANeuralNetworksExecution_startCompute(
  ANeuralNetworksExecution *execution,
  ANeuralNetworksEvent **event
)

Schedule evaluation of the execution.

Schedules evaluation of the execution. Once the model has been applied and the outputs are ready to be consumed, the returned event will be signaled. Use ANeuralNetworksEvent_wait to wait for that event.

Multiple executions can be scheduled and evaluated concurrently. The runtime makes no guarantee on the ordering of completion of executions. If it's important to the application, the application should enforce the ordering by using ANeuralNetworksEvent_wait.

ANeuralNetworksEvent_wait must be called to recuperate the resources used by the execution.

See ANeuralNetworksExecution for information on multithreaded usage.

Details
Parameters
execution
The execution to be scheduled and executed.
event
The event that will be signaled on completion. event is set to NULL if there's an error.
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksMemory_createFromFd

int ANeuralNetworksMemory_createFromFd(
  size_t size,
  int protect,
  int fd,
  size_t offset,
  ANeuralNetworksMemory **memory
)

Creates a shared memory object from a file descriptor.

The shared memory is backed by a file descriptor via mmap. See ANeuralNetworksMemory for a description on how to use this shared memory.

Details
Parameters
size
The requested size in bytes. Must not be larger than the file size.
prot
The desired memory protection for the mapping. It is either PROT_NONE or the bitwise OR of one or more of the following flags: PROT_READ, PROT_WRITE.
fd
The requested file descriptor. The file descriptor has to be mmap-able. The file descriptor will be duplicated.
offset
The offset to the beginning of the file of the area to map. The offset has to be aligned to a page size.
memory
The memory object to be created. Set to NULL if unsuccessful.
Returns
ANEURALNETWORKS_NO_ERROR if the request completed normally.

ANeuralNetworksMemory_free

void ANeuralNetworksMemory_free(
  ANeuralNetworksMemory *memory
)

Delete a memory object.

Destroys the object used by the run time to keep track of the memory. This will free the underlying actual memory if no other code has open handles to this memory.

Details
Parameters
memory
The memory object to be freed.

ANeuralNetworksModel_addOperand

int ANeuralNetworksModel_addOperand(
  ANeuralNetworksModel *model,
  const ANeuralNetworksOperandType *type
)

Add an operand to a model.

The order in which the operands are added is important. The first one added to a model will have the index value 0, the second 1, etc. These indexes are used as operand identifiers in ANeuralNetworksModel_addOperation, ANeuralNetworksExecution_setInput, ANeuralNetworksExecution_setInputFromMemory, ANeuralNetworksExecution_setOutput, ANeuralNetworksExecution_setOutputFromMemory and ANeuralNetworksExecution_setOperandValue.

To build a model that can accommodate inputs of various sizes, as you may want to do for a CNN, leave unspecified the dimensions that will vary at run time. If you do so, fully specify dimensions when calling ANeuralNetworksExecution_setInput or ANeuralNetworksExecution_setInputFromMemory.

Attempting to modify a model once ANeuralNetworksModel_finish has been called will return an error.

See ANeuralNetworksModel for information on multithreaded usage.

Details
Parameters
model
The model to be modified.
type
The ANeuralNetworksOperandType that describes the shape of the operand.
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksModel_addOperation

int ANeuralNetworksModel_addOperation(
  ANeuralNetworksModel *model,
  ANeuralNetworksOperationType type,
  uint32_t inputCount,
  const uint32_t *inputs,
  uint32_t outputCount,
  const uint32_t *outputs
)

Add an operation to a model.

The operands specified by inputs and outputs must have been previously added by calls to ANeuralNetworksModel_addOperand.

Details
Parameters
model
The model to be modified.
type
The ANeuralNetworksOperandType of the operation.
inputCount
The number of entries in the inputs array.
inputs
An array of indexes identifying each operand.
outputCount
The number of entries in the outputs array.
outputs
An array of indexes identifying each operand.

Attempting to modify a model once ANeuralNetworksModel_finish has been called will return an error.

See ANeuralNetworksModel for information on multithreaded usage.

Details
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksModel_create

int ANeuralNetworksModel_create(
  ANeuralNetworksModel **model
)

Create an empty ANeuralNetworksModel.

This only creates the object. Computation is performed once ANeuralNetworksExecution_startCompute is invoked.

The model should be constructed with calls to ANeuralNetworksModel_addOperation and ANeuralNetworksModel_addOperand

ANeuralNetworksModel_finish should be called once the model has been fully constructed.

ANeuralNetworksModel_free should be called once the model is no longer needed.

Details
Parameters
model
The ANeuralNetworksModel to be created. Set to NULL if unsuccessful.
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksModel_finish

int ANeuralNetworksModel_finish(
  ANeuralNetworksModel *model
)

Indicate that we have finished modifying a model.

Required before calling ANeuralNetworksCompilation_create.

An application is responsible to make sure that no other thread uses the model at the same time.

This function must only be called once for a given model.

See ANeuralNetworksModel for information on multithreaded usage.

Details
Parameters
model
The model to be finished.
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksModel_free

void ANeuralNetworksModel_free(
  ANeuralNetworksModel *model
)

Destroy a model.

The model need not have been finished by a call to ANeuralNetworksModel_finish.

See ANeuralNetworksModel for information on multithreaded usage.

Details
Parameters
model
The model to be destroyed. Passing NULL is acceptable and results in no operation.

ANeuralNetworksModel_identifyInputsAndOutputs

int ANeuralNetworksModel_identifyInputsAndOutputs(
  ANeuralNetworksModel *model,
  uint32_t inputCount,
  const uint32_t *inputs,
  uint32_t outputCount,
  const uint32_t *outputs
)

Specifies which operands will be the model's inputs and outputs.

An operand cannot be used for both input and output. Doing so will return an error.

The operands specified by inputs and outputs must have been previously added by calls to ANeuralNetworksModel_addOperand.

Details
Parameters
model
The model to be modified.
inputCount
The number of entries in the inputs array.
inputs
An array of indexes identifying the input operands.
outputCount
The number of entries in the outputs array.
outputs
An array of indexes identifying the output operands.

Attempting to modify a model once ANeuralNetworksModel_finish has been called will return an error.

See ANeuralNetworksModel for information on multithreaded usage.

ANeuralNetworksModel_relaxComputationFloat32toFloat16

int ANeuralNetworksModel_relaxComputationFloat32toFloat16(
  ANeuralNetworksModel *model,
  bool allow
)

Specifies whether ANEURALNETWORKS_TENSOR_FLOAT32 is allowed to be calculated with range and/or precision as low as that of the IEEE 754 16-bit floating-point format.

By default, ANEURALNETWORKS_TENSOR_FLOAT32 must be calculated using at least the range and precision of the IEEE 754 32-bit floating-point format.

Attempting to modify a model once ANeuralNetworksModel_finish has been called will return an error.

Details
Parameters
model
The model to be modified.
allow
'true' indicates ANEURALNETWORKS_TENSOR_FLOAT32 may be calculated with range and/or precision as low as that of the IEEE 754 16-bit floating point format. 'false' indicates ANEURALNETWORKS_TENSOR_FLOAT32 must be calculated using at least the range and precision of the IEEE 754 32-bit floating point format.

See ANeuralNetworksModel for information on multithreaded usage.

ANeuralNetworksModel_setOperandValue

int ANeuralNetworksModel_setOperandValue(
  ANeuralNetworksModel *model,
  int32_t index,
  const void *buffer,
  size_t length
)

Sets an operand to a constant value.

Values of length smaller or equal to ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES are immediately copied into the model.

For values of length greater than ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES, a pointer to the buffer is stored within the model. The application is responsible for not changing the content of this region until all executions using this model have completed. As the data may be copied during processing, modifying the data after this call yields undefined results.

For large tensors, using ANeuralNetworksModel_setOperandValueFromMemory is likely to be more efficient.

To indicate that an optional operand should be considered missing, pass nullptr for buffer and 0 for length.

Attempting to modify a model once ANeuralNetworksModel_finish has been called will return an error.

See ANeuralNetworksModel for information on multithreaded usage.

Details
Parameters
model
The model to be modified.
index
The index of the model operand we're setting.
buffer
A pointer to the data to use.
length
The size in bytes of the data value.
Returns
ANEURALNETWORKS_NO_ERROR if successful.

ANeuralNetworksModel_setOperandValueFromMemory

int ANeuralNetworksModel_setOperandValueFromMemory(
  ANeuralNetworksModel *model,
  int32_t index,
  const ANeuralNetworksMemory *memory,
  size_t offset,
  size_t length
)

Sets an operand to a value stored in a memory object.

The content of the memory is not copied. A reference to that memory is stored inside the model. The application is responsible for not changing the content of the memory region until all executions using this model have completed. As the data may be copied during processing, modifying the data after this call yields undefined results.

To indicate that an optional operand should be considered missing, use ANeuralNetworksModel_setOperandValue instead, passing nullptr for buffer.

Attempting to modify a model once ANeuralNetworksModel_finish has been called will return an error.

See ANeuralNetworksModel for information on multithreaded usage.

Details
Parameters
model
The model to be modified.
index
The index of the model operand we're setting.
buffer
A pointer to the data to use.
memory
The memory containing the data.
offset
This specifies the location of the data within the memory. The offset is in bytes from the start of memory.
length
The size in bytes of the data value.
Returns
ANEURALNETWORKS_NO_ERROR if successful.