nrf_axon_nn_model_infer_sync()

nrf_axon_result_e nrf_axon_nn_model_infer_sync	(	const nrf_axon_nn_compiled_model_s *	compiled_model,
		const int8_t *	input_vector,
		int8_t *	output_buffer )

Blocking inference function of a compiled model.

• Reserves Axon for its exclusive access (using nrf_axon_platform_reserve_for_user()).
• Copies the input_vector to its location in the interlayer buffer. This input_vector is assumed to be packed in memory.
• Performs the inference.
• copies and packs unpacked output from interlayer buffer to output_buffer (optional)
• Frees Axon for other users.
• Returns to user.

If input_vector is NULL, the user has to copy the model input to the correct location in the interlayer buffer (specified by the compiled_model). In a dynamic system where there are other threads utilizing axon and/or asynchronous inference is occurring, the user must 1st call nrf_axon_platform_reserve_for_user() prior to accessing the interlayer_buffer, then invoke this function.

Similarly, output_buffer cannot be NULL in a dynamic system as there is no way to retain control of axon after inference completes.

Cannot be called from interrupt context.

Parameters

[in]	compiled_model	The compiled model to perform inference on that has been validated by calling nrf_axon_nn_model_validate(compiled_model).
[in]	input_vector	Packed input that is copied to the model's input_ptr prior to inference. Can be NULL (see above).
[out]	output_buffer	Buffer to hold the packed output result. Can be NULL (see above).

Return values

0	on success, are a negative error code.