Neural Network - Neural network library

Small feed-forward neural-network library shared by Windows and otStudio. More...

Modules
	Version History
Data Structures
struct	otNNTopology
struct	otTinnQ15MemoryInfo
	Stable snapshot of Q15 runtime memory placement. More...
struct	otTinnWarmStartReport
class	otTinn
	Small feed-forward neural network with generic float inference. More...
class	otNeuralFramework
	Dataset and training support for otTinn. More...
class	otTinn
	Small feed-forward neural network with generic float inference. More...
class	otTinn
	Small feed-forward neural network with generic float inference. More...
class	otNeuralFramework
	Dataset and training support for otTinn. More...
class	otNeuralFramework
	Dataset and training support for otTinn. More...
Defines
#define	OTNN_WINDOWS   0
#define	OTNN_EMBEDDED   1
#define	OTNN_USE_FAST_SIGMOID   1
#define	OTNN_API
#define	OTNN_MAX_HIDDEN_LAYERS   4
Enumerations
enum	otTinnMemoryPolicy { OTNN_MEMORY_AUTO = 0, OTNN_MEMORY_PREFER_FAST, OTNN_MEMORY_FAST_ONLY, OTNN_MEMORY_NORMAL }
	Q15 runtime memory placement policy. More...
enum	otTinnMemoryMode { OTNN_Q15_MEMORY_NONE = 0, OTNN_Q15_MEMORY_NORMAL, OTNN_Q15_MEMORY_FAST, OTNN_Q15_MEMORY_MIXED }
	Effective Q15 runtime memory mode after prepareQ15(). More...
enum	otTinnWarmStartMode { OTNN_WARMSTART_NONE = 0, OTNN_WARMSTART_EXACT_CLONE = 1, OTNN_WARMSTART_FUNCTION_PRESERVING_WIDEN = 2, OTNN_WARMSTART_STRUCTURED_PRUNE_COPY = 3 }
	Result of knowledge-preserving Candidate initialization. More...

---

Detailed Description

The master model and all training remain float. An optional Q15 inference runtime can be generated after training/load with prepareQ15(); it keeps the historical .tinn format unchanged. The raw Q15 path uses integer MACs, Q8 pre-activations and an integer-interpolated sigmoid table, so the BF518 hot inference core contains no floating-point operations.

Embedded dependencies: FILE - Class for file management, Malloc - Memory management, Printf - Print formatted data to stdout, Math Library, String library.

Note:

The Blackfin build deliberately uses the default constructor followed by init()/close(). Do not replace this lifecycle with a parameterized global/static constructor: the GCC toolchain used by otStudio requires this form for reliable startup. \

---

Define Documentation

#define OTNN_API

#define OTNN_EMBEDDED   1

#define OTNN_MAX_HIDDEN_LAYERS   4

#define OTNN_USE_FAST_SIGMOID   1

#define OTNN_WINDOWS   0

---

Enumeration Type Documentation

enum otTinnMemoryMode

Enumeration values: OTNN_Q15_MEMORY_NONE  OTNN_Q15_MEMORY_NORMAL  OTNN_Q15_MEMORY_FAST  OTNN_Q15_MEMORY_MIXED

enum otTinnMemoryPolicy

Enumeration values: OTNN_MEMORY_AUTO  Preferred bank only, then normal-heap fallback. OTNN_MEMORY_PREFER_FAST  Preferred bank, other fast bank, then normal heap. OTNN_MEMORY_FAST_ONLY  Use fast banks only; prepareQ15() fails if they cannot fit. OTNN_MEMORY_NORMAL  Keep the complete Q15 runtime in the normal heap.

enum otTinnWarmStartMode

Enumeration values: OTNN_WARMSTART_NONE  OTNN_WARMSTART_EXACT_CLONE  OTNN_WARMSTART_FUNCTION_PRESERVING_WIDEN  OTNN_WARMSTART_STRUCTURED_PRUNE_COPY

otStudio - Library Reference - (C) 2020-25 Officina Turini, All Rights Reserved
Document built with Doxygen 1.4.0