C/C++ API Reference (Doxygen)¶

The authoritative reference for the C/C++ core is generated by Doxygen from the SNEPPX_* Doxygen comments in include/neural_core/ (Layers 1 & 4 of the commenting standard).

Regenerating locally¶

No CI is involved (see GitHub Pages deployment). From the repository root, on any platform with Doxygen 1.9+ installed:

# Use the root Doxyfile (output directory = docs/api/doxygen)
doxygen Doxyfile

# (or on POSIX)
doxygen Doxyfile

The generated site is written to docs/api/doxygen/html/index.html.

If you get doxygen: command not found, install it: - Windows (vcpkg): vcpkg install doxygen then run from the vcpkg toolchain prompt. - Linux (apt): sudo apt-get install doxygen graphviz - macOS: brew install doxygen graphviz

graphviz (dot) is optional but enables caller/callee graphs.

Configuration¶

The active configuration is the repo-root Doxyfile, which:

sets SEARCHENGINE = YES (client-side JS search is embedded in the output),
extracts from include/neural_core/kernel, …/security, …/architecture, plus kernel/ and security/ implementation directories,
emits HTML only (GENERATE_LATEX = NO),
defines __cplusplus and SNEPPX_STATIC_INLINE=static inline for clean macro expansion.

An alias lives at docs/doxygen/Doxyfile (outputs to docs/doxygen/output/html); the root file is canonical.

Browsing the reference¶

The rendered Doxygen tree is embedded below. Use the search box in the navigation sidebar to jump to any SNEPPX_* symbol.

If the frame does not render, open the reference directly: doxygen/html/index.html.

Key headers to start from¶

Header	Surface
`neural_core/security/cryptographic_primitives_bundle.h`	Umbrella header for S0 crypto
`neural_core/kernel/tensor.h`	`SNEPPXTensor`, `SNEPPXDtype`, device enums
`neural_core/kernel/autodiff.h`	`SNEPPXTape`, `SNEPPXVariable`
`neural_core/kernel/optimizer.h`	`SNEPPXOptimizer` (SGD/AdamW/Lion/LAMB)
`neural_core/kernel/distributed.h`	`SNEPPXDistributedConfig`, collectives
`neural_core/kernel/checkpoint.h`	`SNEPPX_CheckpointCoord`, fault tolerance
`neural_core/kernel/profiler.h`	`SNEPPX_Profiler`, `SNEPPX_KernelTimer`
`neural_core/architecture/neural_programming_engine.h`	NPE VM + 32-opcode ISA
`neural_core/architecture/adversarial_robustness_certification.h`	ARC config + guards
`neural_core/architecture/hierarchical_state_space.h`	HSS config + forward
`neural_core/architecture/sparse_expert_routing.h`	SER config + router
`neural_core/architecture/federated_memory.h`	FM controller + sync

Python ↔ C mapping¶

Python (`SneppX_ALG`)	C core
`Tensor`	`SNEPPXTensor` (`kernel/tensor/`)
`Module`, `Linear`	`SNEPPXModule` (`kernel/arch/`)
`Optimizer`, `AdamW`	`SNEPPXOptimizer` (`kernel/optimizer/`)
`HSSModel`	`SNEPPXHSSModel` (`algorithms/hss/core/`)
`SERModel`	`SNEPPXSERModel` (`algorithms/ser/core/`)
`ARCModel`	`SNEPPXARCLayer` (`algorithms/arc/core/`)
`NPEModel`	`SNEPPXNPEVM` (`algorithms/npe/core/`)
`FMModel`	`FMController` (`algorithms/fm/core/`)
`QuantizedLinear`	`SNEPPXQuantizedLinear` (`kernel/quantization/`)
`CheckpointWriter`	`SNEPPX_CheckpointWriter` (`fs/format/`)
`Profiler`	`SNEPPX_Profiler` (`kernel/profiler.c`)

Every Python wrapper checks _HAS_C_BACKEND and either delegates to the _neural_engine_bridge (C) module or raises / falls back to a pure-NumPy stub. See the Python API for the full class reference.