API

Functions
int	Cy_ML_Model_Inference (void *modelPt, void *input, void *output, int *in_out_q)
	: Cy_ML_Model_Inference() is the API function to perform NN inference. More...
int	Cy_ML_Model_Parse (char *fn_prms, cy_stc_ml_model_info_t *mdl_infoPt)
	: Cy_ML_Model_Parse() is the API function to parse NN model parameters to get basic info. More...
int	Cy_ML_Model_Init (void **dPt_container, char *fn_prms, char *fn_ptr, char *persistent_mem, char *scratch_mem, cy_stc_ml_model_info_t *mdl_infoPt)
	: Cy_ML_Model_Init() is the API function to parse NN model and initialize CY data container before start inference. More...
int	Cy_ML_Rnn_State_Control (void *modelPt, int rnn_status, int window_size)
	: Cy_ML_Rnn_State_Control() is the API function to rest recurrent NN state, inference engine will reset Rnn state when this API is called. More...
int	Cy_ML_Profile_Get_Tsc (uint32_t *val)
	: Cy_ML_Profile_Get_Tsc() is an API function to read time stamp counter (TSC) . More...
int	Cy_ML_Profile_Init (void *modelPt, cy_en_ml_profile_config_t config, cy_ml_cb_fun cb_func, void *cb_arg)
	: Initialize profile configuration. More...
int	Cy_ML_Profile_Control (void *modelPt, cy_en_ml_profile_config_t config)
	: Update profile configuration. More...
int	Cy_ML_Profile_Print (void *modelPt)
	: Print profile log. More...

Detailed Description

Function Documentation

Cy_ML_Model_Inference()

int Cy_ML_Model_Inference	(	void *	modelPt,
		void *	input,
		void *	output,
		int *	in_out_q
	)

: Cy_ML_Model_Inference() is the API function to perform NN inference.

CY NN inference API function targeted for PSoC6 embedded ML application. CY NN inference supports floating-point and fixed-point data format. When doing floating-point inference, input data and output data are in floating-point format. When doing fixed-point inference, input data can be either 16-bit or 8-bit fixed-point. Its inference output will be 8-bit fixed-point when both input data and NN weight are 8-bit, otherwise it will be 16-bit fixed-point. The parameter in_ou_q shall be set to input data fixed-point Q factor at the start of function call, and it will be equal to output data fixed-point Q factor at the end of the function call.

Parameters

[in]	modelPt	: Pointer to CY parsed NN model data container
[in]	input	: Input data pointer
[out]	output	: Inference output data pointer
[in,out]	in_out_q	: Pointer to input data and output data fixed-point Q factor; not applicable (i.e. not used) to floating-point

Returns

: Return 0 when success, otherwise return following error code CY_ML_ERR_INVALID_ARGUMENT if input or output argument is invalid CY_ML_ERR_OTHER_INPUT_MISSING or error code from specific inference engine module. Please note error code is 8bit LSB, line number where the error happened in code is in 16bit MSB, and its layer index if applicable will be at bit 8 to 15 in the combined 32bit return value.

Cy_ML_Model_Init()

int Cy_ML_Model_Init	(	void **	dPt_container,
		char *	fn_prms,
		char *	fn_ptr,
		char *	persistent_mem,
		char *	scratch_mem,
		cy_stc_ml_model_info_t *	mdl_infoPt
	)

: Cy_ML_Model_Init() is the API function to parse NN model and initialize CY data container before start inference.

From the API inputs, it parses NN model and initializes CY data container. For floating-point NN model, its weights and biases shall be in floating-point stored in buffer pointer by parameter fn_ptr. For fixed-point NN mode, its weights and biases can be either 16-bit or 8-bit format stored in buffer pointer by parameter fn_ptr.

Parameters

[out]	dPt_container	: Pointer to CY parsed NN model data container pointer
[in]	fn_prms	: CY parsed NN model parameter buffer pointer
[in]	fn_ptr	: NN model weights and biases buffer pointer
[in]	persistent_mem	: Pointer to allocated persistent memory
[in]	scratch_mem	: Pointer to allocated scratch memory
[in]	mdl_infoPt	: Pointer to cy_stc_ml_model_info_t structure

Returns

: Return 0 when success, otherwise return error code Return CY_ML_ERR_INVALID_ARGUMENT if input or output argument is invalid, Otherwise return other errors: e.g. CY_ML_ERR_OTHER_INPUT_MISSING if the other input is missing in ADD layer Please note error code is 8bit LSB, line number where the error happened in code is in 16bit MSB, and its layer index if applicable will be at bit 8 to 15 in the combined 32bit return value.

Cy_ML_Model_Parse()

int Cy_ML_Model_Parse	(	char *	fn_prms,
		cy_stc_ml_model_info_t *	mdl_infoPt
	)

: Cy_ML_Model_Parse() is the API function to parse NN model parameters to get basic info.

It will parse info from the NN model parameter buffer to get required basic info such as persistent and scratch memory sizes, input data size, output classification size and stored them in to cy_stc_ml_model_info_t structure.

Parameters

[in]	fn_prms	: CY parsed NN model parameter buffer pointer
[out]	mdl_infoPt	: Pointer to cy_stc_ml_model_info_t structure

Returns

: Return 0 0 when success, otherwise return following error code CY_ML_ERR_INVALID_ARGUMENT if input or output argument is invalid CY_ML_ERR_LAYER_NOT_SUPPORTED if layer is not supported CY_ML_ERR_ACT_NOT_SUPPORTED if activation is not supported Please note error code is 8bit LSB, line number where the error happened in code is in 16bit MSB, and its layer index if applicable will be at bit 8 to 15 in the combined 32bit return value.

Cy_ML_Profile_Control()

int Cy_ML_Profile_Control	(	void *	modelPt,
		cy_en_ml_profile_config_t	config
	)

: Update profile configuration.

This API is used to update profile configuration.

Parameters

[in]	modelPt	: Pointer to CY parsed NN model data container pointer
[in]	config	: Profile setting

Returns

: Return 0 when success, otherwise return error code Return CY_ML_ERR_INVALID_ARGUMENT if input parameter is invalid.

Cy_ML_Profile_Get_Tsc()

int Cy_ML_Profile_Get_Tsc

(

uint32_t *

val

)

: Cy_ML_Profile_Get_Tsc() is an API function to read time stamp counter (TSC) .

Platform specific function to read HW time stamp counter or OS tick timer counter for profiling. The application program developer should provide this function if profiling is enabled.

Parameters

[out]

val

: Pointer to time stamp counter return value

Returns

: Return 0 when success, otherwise return error code

Cy_ML_Profile_Init()

int Cy_ML_Profile_Init	(	void *	modelPt,
		cy_en_ml_profile_config_t	config,
		cy_ml_cb_fun	cb_func,
		void *	cb_arg
	)

: Initialize profile configuration.

This API is used to setup profile configuration and specify the callback function to handle the profile log. If no callback function is specified, the profile log will be printed out on console.

Parameters

[in]	modelPt	: Pointer to CY parsed NN model data container pointer
[in]	config	: Profile setting
[in]	cb_func	: Callback function to handle the profile result.
[in]	cb_arg	: Callback function argument

Returns

: Return 0 when success, otherwise return error code Return CY_ML_ERR_INVALID_ARGUMENT if input parameter is invalid.

Cy_ML_Profile_Print()

int Cy_ML_Profile_Print

(

void *

modelPt

)

: Print profile log.

Parameters

[in]

modelPt

: Pointer to CY parsed NN model data container pointer

Returns

: Return 0 when success, otherwise return error code Return CY_ML_ERR_INVALID_ARGUMENT if input parameter is invalid.

Cy_ML_Rnn_State_Control()

int Cy_ML_Rnn_State_Control	(	void *	modelPt,
		int	rnn_status,
		int	window_size
	)

: Cy_ML_Rnn_State_Control() is the API function to rest recurrent NN state, inference engine will reset Rnn state when this API is called.

This API is used to control RNN state. This API can be called after initilization. if one only time reset is desired, use this API to 1) reset the network, and then 2) clear the reset with this API. If the NN model is not recurrent type network. This API will do nothing.

Parameters

[out]	modelPt	: Pointer to CY parsed NN model data container pointer
[in]	rnn_status	: Recurrent NN reset (1) or clear (0)
[in]	window_size	: Recurrent NN reset window size

Returns

: Return 0 when success, otherwise return error code Return CY_ML_ERR_INVALID_ARGUMENT if input argument is invalid.