nnfw_internal.h File Reference

#include "nnfw.h"

Go to the source code of this file.

Functions
NNFW_STATUS	nnfw_set_config (nnfw_session *session, const char *key, const char *value)
NNFW_STATUS	nnfw_get_config (nnfw_session *session, const char *key, char *value, size_t value_size)
NNFW_STATUS	nnfw_load_circle_from_buffer (nnfw_session *session, uint8_t *buffer, size_t size)
	Load a circle model from buffer.
NNFW_STATUS	nnfw_train_export_circleplus (nnfw_session *session, const char *file_path)
	Export circle+ model.
NNFW_STATUS	nnfw_get_output (nnfw_session *session, uint32_t index, nnfw_tensorinfo *out_info, const void **out_buffer)
	Python-binding-only API to retrieve a read-only output buffer and its tensor info.

Function Documentation

nnfw_get_config()

NNFW_STATUS nnfw_get_config	(	nnfw_session *	session,
		const char *	key,
		char *	value,
		size_t	value_size
	)

Definition at line 439 of file APIImpl.cc.

{
  NNFW_RETURN_ERROR_IF_NULL(session);
  return reinterpret_cast<Session *>(session)->get_config(key, value, value_size);
}

References NNFW_RETURN_ERROR_IF_NULL.

nnfw_get_output()

NNFW_STATUS nnfw_get_output	(	nnfw_session *	session,
		uint32_t	index,
		nnfw_tensorinfo *	out_info,
		const void **	out_buffer
	)

Python-binding-only API to retrieve a read-only output buffer and its tensor info.

After nnfw_run has been called, the session has already resized or allocated the internal output buffer to match the latest output dimensions. This API simply retrieves that internal buffer pointer and the corresponding tensor info without performing any allocation itself.

Note: this function is intended for Python binding only. The buffer is managed internally by the session and must not be modified by the caller. In Python, wrap the pointer as a NumPy array and set array.flags.writeable = False to enforce read-only access.

Important: To use this API, you must call nnfw_set_prepare_config(session, NNFW_ENABLE_INTERNAL_OUTPUT_ALLOC, "true") before calling nnfw_prepare().

Parameters

[in]	session	The session object
[in]	index	Output tensor index
[out]	out_info	nnfw_tensorinfo to be filled with the latest shape/type information
[out]	out_buffer	Pointer to a const buffer managed by the session

Returns

NNFW_STATUS_NO_ERROR on success, otherwise an error code

Definition at line 457 of file APIImpl.cc.

{
  NNFW_RETURN_ERROR_IF_NULL(session);
  return reinterpret_cast<Session *>(session)->get_output(index, out_info, out_buffer);
}

References NNFW_RETURN_ERROR_IF_NULL.

Referenced by onert::api::python::NNFW_SESSION::get_output().

nnfw_load_circle_from_buffer()

NNFW_STATUS nnfw_load_circle_from_buffer	(	nnfw_session *	session,
		uint8_t *	buffer,
		size_t	size
	)

Load a circle model from buffer.

The buffer must outlive the session.

Parameters

[in]	session	session
[in]	buffer	Pointer to the buffer
[in]	size	Buffer size

Returns

NNFW_STATUS

Definition at line 445 of file APIImpl.cc.

{
  NNFW_RETURN_ERROR_IF_NULL(session);
  return reinterpret_cast<Session *>(session)->load_circle_from_buffer(buffer, size);
}

References NNFW_RETURN_ERROR_IF_NULL, and size.

nnfw_set_config()

NNFW_STATUS nnfw_set_config	(	nnfw_session *	session,
		const char *	key,
		const char *	value
	)

Definition at line 433 of file APIImpl.cc.

{
  NNFW_RETURN_ERROR_IF_NULL(session);
  return reinterpret_cast<Session *>(session)->set_config(key, value);
}

References NNFW_RETURN_ERROR_IF_NULL.

nnfw_train_export_circleplus()

NNFW_STATUS nnfw_train_export_circleplus	(	nnfw_session *	session,
		const char *	file_path
	)

Export circle+ model.

Note

This function should be called on training mode This function should be called before or after nnfw_train

Parameters

[in]	session	The session to export training model
[in]	file_path	The path to export training model

Returns

NNFW_STATUS_NO_ERROR if successful

Definition at line 451 of file APIImpl.cc.

{
  NNFW_RETURN_ERROR_IF_NULL(session);
  return reinterpret_cast<Session *>(session)->train_export_circleplus(path);
}

References NNFW_RETURN_ERROR_IF_NULL.