Activation Functions

Functions
void	arm_nn_activations_direct_q15 (q15_t *data, uint16_t size, uint16_t int_width, arm_nn_activation_type type)
	neural network activation function using direct table look-up More...
void	arm_nn_activations_direct_q7 (q7_t *data, uint16_t size, uint16_t int_width, arm_nn_activation_type type)
	Q7 neural network activation function using direct table look-up. More...
void	arm_relu6_s8 (q7_t *data, uint16_t size)
	s8 ReLU6 function More...
void	arm_relu_q15 (q15_t *data, uint16_t size)
	Q15 RELU function. More...
void	arm_relu_q7 (q7_t *data, uint16_t size)
	Q7 RELU function. More...

Description

Perform activation layers, including ReLU (Rectified Linear Unit), sigmoid and tanh

Function Documentation

void arm_nn_activations_direct_q15	(	q15_t *	data,
		uint16_t	size,
		uint16_t	int_width,
		arm_nn_activation_type	type
	)

Q15 neural network activation function using direct table look-up.

Note

Refer header file for details.

References ARM_SIGMOID, ARM_TANH, sigmoidTable_q15, and tanhTable_q15.

void arm_nn_activations_direct_q7	(	q7_t *	data,
		uint16_t	size,
		uint16_t	int_width,
		arm_nn_activation_type	type
	)

Parameters

[in,out]	data	pointer to input
[in]	size	number of elements
[in]	int_width	bit-width of the integer part, assume to be smaller than 3
[in]	type	type of activation functions

This is the direct table look-up approach.

Assume here the integer part of the fixed-point is <= 3. More than 3 just not making much sense, makes no difference with saturation followed by any of these activation functions.

References ARM_SIGMOID, ARM_TANH, sigmoidTable_q7, and tanhTable_q7.

void arm_relu6_s8	(	q7_t *	data,
		uint16_t	size
	)

Parameters

[in,out]	data	pointer to input
[in]	size	number of elements

References MAX, and MIN.

void arm_relu_q15	(	q15_t *	data,
		uint16_t	size
	)

Parameters

[in,out]	data	pointer to input
[in]	size	number of elements

Optimized relu with QSUB instructions.

References arm_nn_read_q15x2_ia(), and arm_nn_write_q15x2_ia().

void arm_relu_q7	(	q7_t *	data,
		uint16_t	size
	)

Parameters

[in,out]	data	pointer to input
[in]	size	number of elements

Optimized relu with QSUB instructions.

References arm_nn_read_q7x4_ia(), and arm_nn_write_q7x4_ia().