Structure Types

Enums and Data Structures used in public API. More...

Data Structures
struct	cmsis_nn_tile
struct	cmsis_nn_context
struct	cmsis_nn_bias_data
struct	cmsis_nn_dims
struct	cmsis_nn_lstm_dims
struct	cmsis_nn_per_channel_quant_params
struct	cmsis_nn_per_tensor_quant_params
struct	cmsis_nn_quant_params
struct	cmsis_nn_activation
struct	cmsis_nn_conv_params
struct	cmsis_nn_transpose_conv_params
struct	cmsis_nn_dw_conv_params
struct	cmsis_nn_pool_params
struct	cmsis_nn_fc_params
struct	cmsis_nn_bmm_params
struct	cmsis_nn_transpose_params
struct	cmsis_nn_svdf_params
struct	cmsis_nn_softmax_lut_s16
struct	cmsis_nn_scaling
struct	cmsis_nn_lstm_gate
struct	cmsis_nn_lstm_params
struct	cmsis_nn_lstm_context
struct	cmsis_nn_activation_f32
	Activation clamp range for floating-point operators. More...
struct	cmsis_nn_conv_params_f32
	Convolution parameters for float32 operators. More...
struct	cmsis_nn_transpose_conv_params_f32
	Transpose convolution parameters for float32 operators. More...
struct	cmsis_nn_dw_conv_params_f32
	Depthwise convolution parameters for float32 operators. More...
struct	cmsis_nn_pool_params_f32
	Pooling parameters for float32 operators. More...
struct	cmsis_nn_fc_params_f32
	Fully connected layer parameters for float32 operators. More...
struct	cmsis_nn_bmm_params_f32
	Batched matrix multiplication parameters for float32 operators. More...
struct	cmsis_nn_ew_params_f32
	Elementwise operator parameters for float32 operators. More...
struct	cmsis_nn_transpose_params_f32
	Transpose parameters for float32 operators. More...
struct	cmsis_nn_svdf_params_f32
	Singular value decomposition filter parameters for float32 operators. More...
struct	cmsis_nn_lstm_gate_f32
	Read-only weights and bias metadata for one float32 LSTM gate. More...
struct	cmsis_nn_lstm_params_f32
	Parameters for a unidirectional float32 LSTM layer. More...
struct	cmsis_nn_lstm_context_f32
	Scratch and mutable state buffers for a float32 LSTM invocation. More...
struct	cmsis_nn_activation_f16
	Activation clamp range for floating-point operators. More...
struct	cmsis_nn_conv_params_f16
	Convolution parameters for float32 operators. More...
struct	cmsis_nn_transpose_conv_params_f16
	Transpose convolution parameters for float32 operators. More...
struct	cmsis_nn_dw_conv_params_f16
	Depthwise convolution parameters for float32 operators. More...
struct	cmsis_nn_pool_params_f16
	Pooling parameters for float32 operators. More...
struct	cmsis_nn_fc_params_f16
	Fully connected layer parameters for float32 operators. More...
struct	cmsis_nn_bmm_params_f16
	Batched matrix multiplication parameters for float32 operators. More...
struct	cmsis_nn_ew_params_f16
	Elementwise operator parameters for float32 operators. More...
struct	cmsis_nn_transpose_params_f16
	Transpose parameters for float32 operators. More...
struct	cmsis_nn_svdf_params_f16
	Singular value decomposition filter parameters for float32 operators. More...
struct	cmsis_nn_lstm_gate_f16
	Read-only weights and bias metadata for one float32 LSTM gate. More...
struct	cmsis_nn_lstm_params_f16
	Parameters for a unidirectional float32 LSTM layer. More...
struct	cmsis_nn_lstm_context_f16
	Scratch and mutable state buffers for a float32 LSTM invocation. More...

Typedefs
typedef char	arm_nn_static_assert___LINE__[(((int) ARM_TANH)<((int) ARM_NN_FLT_ACT_NONE)) ? 1 :-1]
typedef arm_nn_dw_kernel_layout_f32	arm_nn_dw_kernel_layout_f16

Enumerations
enum	arm_nn_activation_type {   ARM_SIGMOID = 0 ,   ARM_TANH = 1 }
enum	arm_cmsis_nn_status {   ARM_CMSIS_NN_SUCCESS = 0 ,   ARM_CMSIS_NN_ARG_ERROR = -1 ,   ARM_CMSIS_NN_NO_IMPL_ERROR = -2 ,   ARM_CMSIS_NN_FAILURE = -3 }
enum	arm_nn_tensor_layout { ARM_NN_LAYOUT_NHWC = 0 }
	Tensor layout selector for floating-point APIs. More...
enum	arm_nn_activation_type_flt {   ARM_NN_FLT_ACT_NONE = 32 ,   ARM_NN_FLT_ACT_SIGMOID = 33 ,   ARM_NN_FLT_ACT_TANH = 34 ,   ARM_NN_FLT_ACT_RELU = 35 ,   ARM_NN_FLT_ACT_RELU6 = 36 ,   ARM_NN_FLT_ACT_HARDSWISH = 37 ,   ARM_NN_FLT_ACT_LEAKY_RELU = 38 }
	Activation selector for floating-point operator APIs. More...
enum	arm_nn_dw_kernel_layout_f32 {   ARM_NN_DW_KERNEL_KC = 0 ,   ARM_NN_DW_KERNEL_CK = 1 }
	Depthwise kernel storage layout selector for floating-point kernels. More...
enum	arm_nn_weight_format_flt {   ARM_NN_WEIGHT_FORMAT_STANDARD = 0 ,   ARM_NN_WEIGHT_FORMAT_NT_N_PACKED = 1 }
	Weight storage format selector for floating-point operators. More...

Description

Enums and Data Structures used in public API.

---

Data Structure Documentation

cmsis_nn_tile

struct cmsis_nn_tile

CMSIS-NN object to contain the width and height of a tile

Data Fields
int32_t	h	Height
int32_t	w	Width

cmsis_nn_context

struct cmsis_nn_context

CMSIS-NN object used for the function context.

Data Fields
void *	buf	Pointer to a buffer needed for the optimization
int32_t	size	Buffer size

cmsis_nn_bias_data

struct cmsis_nn_bias_data

CMSIS-NN object used to hold bias data for int16 variants.

Data Fields
const void *	data	Pointer to bias data
const bool	is_int32_bias	Indicate type of bias data. True means int32 else int64

cmsis_nn_dims

struct cmsis_nn_dims

CMSIS-NN object to contain the dimensions of the tensors

Data Fields
int32_t	c	Input channels
int32_t	h	Height
int32_t	n	Generic dimension to contain either the batch size or output channels. Please refer to the function documentation for more information
int32_t	w	Width

cmsis_nn_lstm_dims

struct cmsis_nn_lstm_dims

CMSIS-NN object to contain LSTM specific input parameters related to dimensions

Data Fields
int32_t	max_time
int32_t	num_batches
int32_t	num_inputs
int32_t	num_outputs

cmsis_nn_per_channel_quant_params

struct cmsis_nn_per_channel_quant_params

CMSIS-NN object for the per-channel quantization parameters

Data Fields
int32_t *	multiplier	Multiplier values
int32_t *	shift	Shift values

cmsis_nn_per_tensor_quant_params

struct cmsis_nn_per_tensor_quant_params

CMSIS-NN object for the per-tensor quantization parameters

Data Fields
int32_t	multiplier	Multiplier value
int32_t	shift	Shift value

cmsis_nn_quant_params

struct cmsis_nn_quant_params

CMSIS-NN object for quantization parameters. This struct supports both per-tensor and per-channels requantization and is recommended for new operators.

Data Fields
int32_t	is_per_channel
int32_t *	multiplier	Multiplier values
int32_t *	shift	Shift values

cmsis_nn_activation

struct cmsis_nn_activation

CMSIS-NN object for the quantized Relu activation

Data Fields
int32_t	max	Max value used to clamp the result
int32_t	min	Min value used to clamp the result

cmsis_nn_conv_params

struct cmsis_nn_conv_params

CMSIS-NN object for the convolution layer parameters

Data Fields
cmsis_nn_activation	activation
cmsis_nn_tile	dilation
int32_t	input_offset	The negative of the zero value for the input tensor
int32_t	output_offset	The negative of the zero value for the output tensor
cmsis_nn_tile	padding
cmsis_nn_tile	stride

cmsis_nn_transpose_conv_params

struct cmsis_nn_transpose_conv_params

CMSIS-NN object for the transpose convolution layer parameters

Data Fields
cmsis_nn_activation	activation
cmsis_nn_tile	dilation
int32_t	input_offset	The negative of the zero value for the input tensor
int32_t	output_offset	The negative of the zero value for the output tensor
cmsis_nn_tile	padding
cmsis_nn_tile	padding_offsets
cmsis_nn_tile	stride

cmsis_nn_dw_conv_params

struct cmsis_nn_dw_conv_params

CMSIS-NN object for the depthwise convolution layer parameters

Data Fields
cmsis_nn_activation	activation
int32_t	ch_mult	Channel Multiplier. ch_mult * in_ch = out_ch
cmsis_nn_tile	dilation
int32_t	input_offset	The negative of the zero value for the input tensor
int32_t	output_offset	The negative of the zero value for the output tensor
cmsis_nn_tile	padding
cmsis_nn_tile	stride

cmsis_nn_pool_params

struct cmsis_nn_pool_params

CMSIS-NN object for pooling layer parameters

Data Fields
cmsis_nn_activation	activation
cmsis_nn_tile	padding
cmsis_nn_tile	stride

cmsis_nn_fc_params

struct cmsis_nn_fc_params

CMSIS-NN object for Fully Connected layer parameters

Data Fields
cmsis_nn_activation	activation
int32_t	filter_offset	The negative of the zero value for the filter tensor
int32_t	input_offset	The negative of the zero value for the input tensor
int32_t	output_offset	The negative of the zero value for the output tensor

cmsis_nn_bmm_params

struct cmsis_nn_bmm_params

CMSIS-NN object for Batch Matmul layer parameters

Data Fields
const bool	adj_x
const bool	adj_y
cmsis_nn_fc_params	fc_params

cmsis_nn_transpose_params

struct cmsis_nn_transpose_params

CMSIS-NN object for Transpose layer parameters

Data Fields
const int32_t	num_dims
const uint32_t *	permutations	The dimensions applied to the input dimensions

cmsis_nn_svdf_params

struct cmsis_nn_svdf_params

CMSIS-NN object for SVDF layer parameters

Data Fields
cmsis_nn_activation	input_activation
int32_t	input_offset	The negative of the zero value for the input tensor
cmsis_nn_activation	output_activation
int32_t	output_offset	The negative of the zero value for the output tensor
int32_t	rank

cmsis_nn_softmax_lut_s16

struct cmsis_nn_softmax_lut_s16

CMSIS-NN object for Softmax s16 layer parameters

Data Fields
const int16_t *	exp_lut
const int16_t *	one_by_one_lut

cmsis_nn_scaling

struct cmsis_nn_scaling

CMSIS-NN object for quantization parameters

Data Fields
int32_t	multiplier	Multiplier value
int32_t	shift	Shift value

cmsis_nn_lstm_gate

struct cmsis_nn_lstm_gate

CMSIS-NN object for LSTM gate parameters

Data Fields
arm_nn_activation_type	activation_type
const void *	bias
const void *	hidden_effective_bias	Precomputed kernel_sum * lhs_offset
int32_t	hidden_multiplier
int32_t	hidden_shift
const void *	hidden_weights
const void *	input_effective_bias	Bias added with precomputed kernel_sum * lhs_offset
int32_t	input_multiplier
int32_t	input_shift
const void *	input_weights

cmsis_nn_lstm_params

struct cmsis_nn_lstm_params

CMSIS-NN object for LSTM parameters

Data Fields
int32_t	batch_size
int32_t	cell_clip	Min/max value of cell output
cmsis_nn_lstm_gate	cell_gate
int32_t	cell_scale_power
cmsis_nn_lstm_gate	forget_gate
int32_t	forget_to_cell_multiplier
int32_t	forget_to_cell_shift
int32_t	hidden_size	Size of output from the LSTM cell, used as output and recursively into the next time step
cmsis_nn_lstm_gate	input_gate
int32_t	input_offset
int32_t	input_size	Size of new data input into the LSTM cell
int32_t	input_to_cell_multiplier
int32_t	input_to_cell_shift
cmsis_nn_lstm_gate	output_gate
int32_t	output_multiplier
int32_t	output_offset
int32_t	output_shift
int32_t	time_major	0 if first dimension is batch, else first dimension is time
int32_t	time_steps

cmsis_nn_lstm_context

struct cmsis_nn_lstm_context

CMSIS-NN object for LSTM scratch buffers

Data Fields
void *	cell_state
void *	hidden_state
void *	temp1
void *	temp2

cmsis_nn_activation_f32

struct cmsis_nn_activation_f32

Activation clamp range for floating-point operators.

Data Fields
float32_t	max	Maximum value used to clamp the result.
float32_t	min	Minimum value used to clamp the result.

cmsis_nn_conv_params_f32

struct cmsis_nn_conv_params_f32

Convolution parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.
cmsis_nn_tile	dilation	Spatial dilation.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	stride	Spatial stride.
arm_nn_weight_format_flt	weight_format	Filter storage format.

cmsis_nn_transpose_conv_params_f32

struct cmsis_nn_transpose_conv_params_f32

Transpose convolution parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.
cmsis_nn_tile	dilation	Spatial dilation.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	padding_offsets	Output padding adjustment for transpose convolution.
cmsis_nn_tile	stride	Spatial stride.

cmsis_nn_dw_conv_params_f32

struct cmsis_nn_dw_conv_params_f32

Depthwise convolution parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.
int32_t	ch_mult	Channel multiplier. ch_mult * in_ch = out_ch.
cmsis_nn_tile	dilation	Spatial dilation.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	stride	Spatial stride.

cmsis_nn_pool_params_f32

struct cmsis_nn_pool_params_f32

Pooling parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	stride	Spatial stride.

cmsis_nn_fc_params_f32

struct cmsis_nn_fc_params_f32

Fully connected layer parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.
arm_nn_weight_format_flt	weight_format	Weight storage format.

cmsis_nn_bmm_params_f32

struct cmsis_nn_bmm_params_f32

Batched matrix multiplication parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.
const bool	adj_x	True when the left-hand-side operand is stored transposed.
const bool	adj_y	True when the right-hand-side operand is stored transposed.
arm_nn_weight_format_flt	rhs_format	Right-hand-side operand storage format. ARM_NN_WEIGHT_FORMAT_NT_N_PACKED is currently supported only when adj_x == false and adj_y == false.

cmsis_nn_ew_params_f32

struct cmsis_nn_ew_params_f32

Elementwise operator parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	activation	Output activation clamp range.

cmsis_nn_transpose_params_f32

struct cmsis_nn_transpose_params_f32

Transpose parameters for float32 operators.

Data Fields
arm_nn_tensor_layout	layout	Layout convention used to interpret tensor dimensions.
int32_t	num_dims	Number of active dimensions in the permutation.
int32_t	perm[4]	Permutation indices.

cmsis_nn_svdf_params_f32

struct cmsis_nn_svdf_params_f32

Singular value decomposition filter parameters for float32 operators.

Data Fields
cmsis_nn_activation_f32	input_activation	Clamp range applied after the input projection.
cmsis_nn_activation_f32	output_activation	Clamp range applied to the final output.
int32_t	rank	SVDF rank.

cmsis_nn_lstm_gate_f32

struct cmsis_nn_lstm_gate_f32

Read-only weights and bias metadata for one float32 LSTM gate.

Data Fields
arm_nn_activation_type_flt	activation_type	Gate activation selector.
const float32_t *	bias	Optional gate bias vector.
const float32_t *	hidden_weights	Hidden-state-to-gate weight matrix.
const float32_t *	input_weights	Input-to-gate weight matrix.

cmsis_nn_lstm_params_f32

struct cmsis_nn_lstm_params_f32

Parameters for a unidirectional float32 LSTM layer.

Data Fields
int32_t	batch_size	Batch size processed per invocation.
float32_t	cell_clip	Optional cell-state clip value.
cmsis_nn_lstm_gate_f32	cell_gate	Cell-update gate weights and activation.
cmsis_nn_lstm_gate_f32	forget_gate	Forget gate weights and activation.
int32_t	hidden_size	Hidden-state size.
cmsis_nn_lstm_gate_f32	input_gate	Input gate weights and activation.
int32_t	input_size	Input feature size per time step.
cmsis_nn_lstm_gate_f32	output_gate	Output gate weights and activation.
int32_t	time_major	Non-zero when input/output tensors are time-major.
int32_t	time_steps	Number of time steps processed per invocation.

cmsis_nn_lstm_context_f32

struct cmsis_nn_lstm_context_f32

Scratch and mutable state buffers for a float32 LSTM invocation.

Data Fields
float32_t *	cell_state	Mutable cell-state buffer.
float32_t *	temp1	Temporary buffer used by matrix and gate computations.
float32_t *	temp2	Temporary buffer used by matrix and gate computations.

cmsis_nn_activation_f16

struct cmsis_nn_activation_f16

Activation clamp range for floating-point operators.

Data Fields
float16_t	max	Maximum value used to clamp the result.
float16_t	min	Minimum value used to clamp the result.

cmsis_nn_conv_params_f16

struct cmsis_nn_conv_params_f16

Convolution parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.
cmsis_nn_tile	dilation	Spatial dilation.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	stride	Spatial stride.
arm_nn_weight_format_flt	weight_format	Filter storage format.

cmsis_nn_transpose_conv_params_f16

struct cmsis_nn_transpose_conv_params_f16

Transpose convolution parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.
cmsis_nn_tile	dilation	Spatial dilation.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	padding_offsets	Output padding adjustment for transpose convolution.
cmsis_nn_tile	stride	Spatial stride.

cmsis_nn_dw_conv_params_f16

struct cmsis_nn_dw_conv_params_f16

Depthwise convolution parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.
int32_t	ch_mult	Channel multiplier. ch_mult * in_ch = out_ch.
cmsis_nn_tile	dilation	Spatial dilation.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	stride	Spatial stride.

cmsis_nn_pool_params_f16

struct cmsis_nn_pool_params_f16

Pooling parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.
cmsis_nn_tile	padding	Spatial zero-padding.
cmsis_nn_tile	stride	Spatial stride.

cmsis_nn_fc_params_f16

struct cmsis_nn_fc_params_f16

Fully connected layer parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.
arm_nn_weight_format_flt	weight_format	Weight storage format.

cmsis_nn_bmm_params_f16

struct cmsis_nn_bmm_params_f16

Batched matrix multiplication parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.
const bool	adj_x	True when the left-hand-side operand is stored transposed.
const bool	adj_y	True when the right-hand-side operand is stored transposed.
arm_nn_weight_format_flt	rhs_format	Right-hand-side operand storage format. ARM_NN_WEIGHT_FORMAT_NT_N_PACKED is currently supported only when adj_x == false and adj_y == false.

cmsis_nn_ew_params_f16

struct cmsis_nn_ew_params_f16

Elementwise operator parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	activation	Output activation clamp range.

cmsis_nn_transpose_params_f16

struct cmsis_nn_transpose_params_f16

Transpose parameters for float32 operators.

Data Fields
arm_nn_tensor_layout	layout	Layout convention used to interpret tensor dimensions.
int32_t	num_dims	Number of active dimensions in the permutation.
int32_t	perm[4]	Permutation indices.

cmsis_nn_svdf_params_f16

struct cmsis_nn_svdf_params_f16

Singular value decomposition filter parameters for float32 operators.

Data Fields
cmsis_nn_activation_f16	input_activation	Clamp range applied after the input projection.
cmsis_nn_activation_f16	output_activation	Clamp range applied to the final output.
int32_t	rank	SVDF rank.

cmsis_nn_lstm_gate_f16

struct cmsis_nn_lstm_gate_f16

Read-only weights and bias metadata for one float32 LSTM gate.

Data Fields
arm_nn_activation_type_flt	activation_type	Gate activation selector.
const float16_t *	bias	Optional gate bias vector.
const float16_t *	hidden_weights	Hidden-state-to-gate weight matrix.
const float16_t *	input_weights	Input-to-gate weight matrix.

cmsis_nn_lstm_params_f16

struct cmsis_nn_lstm_params_f16

Parameters for a unidirectional float32 LSTM layer.

Data Fields
int32_t	batch_size	Batch size processed per invocation.
float16_t	cell_clip	Optional cell-state clip value.
cmsis_nn_lstm_gate_f16	cell_gate	Cell-update gate weights and activation.
cmsis_nn_lstm_gate_f16	forget_gate	Forget gate weights and activation.
int32_t	hidden_size	Hidden-state size.
cmsis_nn_lstm_gate_f16	input_gate	Input gate weights and activation.
int32_t	input_size	Input feature size per time step.
cmsis_nn_lstm_gate_f16	output_gate	Output gate weights and activation.
int32_t	time_major	Non-zero when input/output tensors are time-major.
int32_t	time_steps	Number of time steps processed per invocation.

cmsis_nn_lstm_context_f16

struct cmsis_nn_lstm_context_f16

Scratch and mutable state buffers for a float32 LSTM invocation.

Data Fields
float16_t *	cell_state	Mutable cell-state buffer.
float16_t *	temp1	Temporary buffer used by matrix and gate computations.
float16_t *	temp2	Temporary buffer used by matrix and gate computations.

Typedef Documentation

arm_nn_dw_kernel_layout_f16

typedef arm_nn_dw_kernel_layout_f32 arm_nn_dw_kernel_layout_f16

arm_nn_static_assert___LINE__

typedef char arm_nn_static_assert___LINE__[(((int) ARM_TANH)<((int) ARM_NN_FLT_ACT_NONE)) ? 1 :-1]

Enumeration Type Documentation

arm_cmsis_nn_status

enum arm_cmsis_nn_status

Function return codes

Enumerator
ARM_CMSIS_NN_SUCCESS	No error
ARM_CMSIS_NN_ARG_ERROR	One or more arguments are incorrect
ARM_CMSIS_NN_NO_IMPL_ERROR	No implementation available
ARM_CMSIS_NN_FAILURE	Logical error

arm_nn_activation_type

enum arm_nn_activation_type

Enum for specifying activation function types

Enumerator
ARM_SIGMOID	Sigmoid activation function
ARM_TANH	Tanh activation function

arm_nn_activation_type_flt

enum arm_nn_activation_type_flt

Activation selector for floating-point operator APIs.

Numeric values intentionally live in a dedicated floating-point range to avoid overlap with the legacy integer public activation enum.

Enumerator
ARM_NN_FLT_ACT_NONE	Identity activation function.
ARM_NN_FLT_ACT_SIGMOID	Sigmoid activation function.
ARM_NN_FLT_ACT_TANH	Hyperbolic tangent activation function.
ARM_NN_FLT_ACT_RELU	ReLU activation function.
ARM_NN_FLT_ACT_RELU6	ReLU6 activation function.
ARM_NN_FLT_ACT_HARDSWISH	Hard-swish activation function.
ARM_NN_FLT_ACT_LEAKY_RELU	Leaky ReLU activation function.

arm_nn_dw_kernel_layout_f32

enum arm_nn_dw_kernel_layout_f32

Depthwise kernel storage layout selector for floating-point kernels.

Public float depthwise entry points currently use KC storage ([k][c]).

Enumerator
ARM_NN_DW_KERNEL_KC	Depthwise kernel stored as [kernel][channel].
ARM_NN_DW_KERNEL_CK	Depthwise kernel stored as [channel][kernel].

arm_nn_tensor_layout

enum arm_nn_tensor_layout

Tensor layout selector for floating-point APIs.

Float public APIs currently accept NHWC layout only.

Enumerator
ARM_NN_LAYOUT_NHWC	Tensor dimensions are ordered as [N, H, W, C].

arm_nn_weight_format_flt

enum arm_nn_weight_format_flt

Weight storage format selector for floating-point operators.

This enum allows frameworks to describe whether weights are provided in the standard public operator layout or in a backend-specific packed layout.

ARM_NN_WEIGHT_FORMAT_NT_N_PACKED matches the packed RHS layout consumed by arm_nn_mat_mult_nt_n_packed_f16/f32.

The packed NTxN layout exists because MVE kernels typically perform best when output-channel blocks can be loaded contiguously. With the standard NT x T formulation, vectorizing over output channels tends to require gather-load accesses to the RHS, which is less efficient than a packed non-transposed RHS layout.

For operators that support ARM_NN_WEIGHT_FORMAT_NT_N_PACKED, supplying offline-repacked constant weights in this layout is therefore generally the preferred way to achieve the best MVE performance.

Enumerator
ARM_NN_WEIGHT_FORMAT_STANDARD	Standard public operator layout.
ARM_NN_WEIGHT_FORMAT_NT_N_PACKED	Packed [K][N-block] layout for NTxN matmul helpers.