neat Namespace

Definition

namespace simaai::neat { ... }

Namespaces Index

namespace	debug
namespace	dlpack
	Minimal DLPack-shaped structs for zero-copy interop with NumPy/PyTorch. More...
namespace	error_codes
namespace	genai
namespace	graph
namespace	internal
namespace	mpk
namespace	nodes
namespace	pipeline_internal
namespace	stages
namespace	validators

Classes Index

struct	ValidationContext
	Context passed to contracts during validation. More...
class	Contract
	A single validation rule applied to a NodeGroup. More...
class	ContractRegistry
	Holds a set of Contracts and runs them to produce a ValidationReport. More...
struct	ContractPortSpec
	Per-port contract: the shape/type/segment requirements for one input or output. More...
struct	ContractFieldSpec
	Per-field contract: where the field value comes from and the override policy. More...
struct	NodeContractDefinition
	Bundle of port and field specs that fully describes a Node's contract. More...
struct	ValidationIssue
	A single reported issue from a Contract. More...
class	ValidationReport
	Report produced by running a ContractRegistry. More...
class	Model
	Loaded form of an MPK; the simplified entry point to run inference on Modalix. More...
struct	ResizeSpec
	Resize / letterbox / crop parameters. More...
struct	ColorConvertSpec
	Color-format conversion parameters. More...
struct	LayoutConvertSpec
	Axis-permutation parameters for HWC↔CHW-style layout conversions. More...
struct	NormalizeSpec
	Mean/stddev normalization parameters. More...
struct	QuantizeSpec
	INT8 (or other low-precision) quantization parameters. More...
struct	TessellateSpec
	MLA tessellation (tile-shuffle) parameters. More...
struct	PreprocessExplicitKnobs
	Per-stage flags recording which preprocess fields the user set explicitly. More...
struct	Transform
	Discriminated union of all per-stage transform specs. More...
struct	PreprocessOptions
	User-facing preprocess intent — what the application asks for. More...
struct	PreprocessContract
	Per-input contract describing what the preprocess stage expects to receive. More...
struct	PreprocessMetaContract
	Metadata-side contract — names and fields the preprocess stage requires. More...
struct	ResolvedPreprocessPlan
	Final resolved plan — what the framework actually compiled and runs. More...
struct	PreparedRuntimeDescriptor
struct	GraphTensorContract
struct	GraphQuantContract
struct	GraphProcessMlaStageRequest
struct	GraphProcessCvuStageRequest
class	FileInput
	Wraps GStreamer's filesrc element — reads encoded media from a path on disk. More...
class	ImageDecode
	Auto-detects and decodes still-image formats (JPEG, PNG, BMP, …) to raw video. More...
class	JpegDecode
	Decodes JPEG-encoded buffers to raw frames. More...
struct	OutputOptions
	Buffering / sync options for the Output Node. More...
class	Output
	Pull-side terminal sink. More...
class	Queue
	Wraps GStreamer's queue element — inserts a thread/buffer boundary. More...
class	VideoConvert
	Wraps GStreamer's videoconvert. More...
class	VideoScale
	Wraps GStreamer's videoscale. More...
class	VideoTrackSelect
	Creates a qtdemux element and selects a specific video pad (e.g., demux.video_0). More...
struct	PreprocessMetaTemplate
	Preprocess metadata template attached to ingress buffers. More...
struct	InputOptions
	Caps, buffering, and pool options for the Input Node. More...
class	Input
	Push-mode source Node. More...
struct	MetadataSenderOptions
class	MetadataSender
class	RTSPInput
	RTSP source Node — pulls a stream from an RTSP URL. More...
class	StillImageInput
	Source Node that emits a single image as a continuous video stream. More...
struct	UdpOutputOptions
	Destination + sync options for the UdpOutput Node. More...
class	UdpOutput
	UDP sink Node — writes packets to a UDP destination. More...
class	H264CapsFixup
	Helper Node that fills in missing H.264 RTP caps fields with fallbacks. More...
class	H264Depacketize
	Depayloads RTP-encapsulated H.264 into a raw H.264 byte stream. More...
struct	CastOptions
	Construction options for a Cast Node. More...
class	Cast
	CVU kernel Node that casts a tensor between FP32 and BF16 (no scale/zero-point). More...
struct	CastTessOptions
	Construction options for a CastTess Node. More...
class	CastTess
	Fused CVU kernel Node: Cast (FP32→BF16) followed by Tess. More...
struct	DequantOptions
	Construction options for a Dequant Node. More...
class	Dequant
	CVU kernel Node that dequantizes INT8 tensors to FP32 using scale + zero-point. More...
struct	DetessOptions
	Construction options for a Detess Node. More...
class	Detess
	CVU kernel Node that detessellates MLA-layout tensors back to natural HWC/CHW order. More...
struct	DetessCastOptions
	Construction options for a DetessCast Node. More...
class	DetessCast
	Fused CVU kernel Node: Detess followed by Cast (BF16→FP32). More...
struct	DetessDequantOptions
	Construction options for a DetessDequant Node. More...
class	DetessDequant
	Fused CVU kernel Node: Detess followed by Dequant (INT8→FP32). More...
class	H264Decode
	Hardware-accelerated H.264 decoder Node. More...
class	H264EncodeSima
	Hardware-accelerated H.264 encoder Node. More...
class	H264Packetize
	RTP H.264 payloader Node — wraps encoded H.264 into RTP packets. More...
struct	H264ParseOptions
	Construction options for an H264Parse Node — config-interval and optional caps lock. More...
class	H264Parse
	H.264 stream parser Node — parses NAL units, finds keyframes, optionally locks caps. More...
struct	PCIeSinkOptions
	Construction options for a PCIeSink Node. More...
class	PCIeSink
	Terminal sink Node that streams samples to a PCIe-connected host. More...
struct	PCIeSrcOptions
	Construction options for a PCIeSrc Node. More...
class	PCIeSrc
	Source-role Node that receives samples from a PCIe-connected host. More...
struct	PreprocOptions
	Construction options for a Preproc Node. More...
class	Preproc
	Fused CVU preprocessing Node — resize + color-convert + normalize (+ optional tess). More...
struct	QuantOptions
	Construction options for a Quant Node. More...
class	Quant
	CVU kernel Node that quantizes FP32 tensors to INT8 using scale + zero-point. More...
struct	QuantTessOptions
	Construction options for a QuantTess Node. More...
class	QuantTess
	Fused CVU kernel Node: Quant (FP32→INT8) followed by Tess. More...
struct	SimaArgMaxOptions
	Construction options for a SimaArgMax Node. More...
class	SimaArgMax
	EV74 postprocess Node that emits the top-class index from a classification tensor. More...
class	SimaBoxDecode
	EV74 postprocess Node that decodes detection-head tensors into object boxes (with NMS). More...
struct	SimaRenderOptions
	Construction options for a SimaRender Node. More...
class	SimaRender
	Node that renders bounding-box overlays onto a video frame. More...
struct	TessOptions
	Construction options for a Tess Node. More...
class	Tess
	CVU kernel Node that tessellates a natural-layout tensor into MLA tile geometry. More...
struct	Box
	One axis-aligned detection produced by BoxDecode. More...
struct	BoxDecodeResult
	Parsed BoxDecode output paired with its raw byte buffer. More...
struct	FormatSpec
	Thin wrapper around FormatTag with implicit string conversions. More...
struct	ProfilerKernelInvocation
	One kernel-invocation telemetry event. More...
struct	ProfilerMemcpySite
	Aggregate counters for one instrumented memcpy site. More...
struct	ProfilerKernelAggregate
	Aggregated timings for one (backend, kernel, stage, slot) tuple. More...
struct	ProfilerReport
	Snapshot bundle returned by LatencyProfiler::finalize(). More...
struct	LatencyProfilerOptions
	Construction options for LatencyProfiler. More...
class	LatencyProfiler
	Per-sample latency tracker; attach to a Run to capture timing telemetry. More...
struct	PowerRailConfig
	Configuration for one PMBus rail exposed by a paged PMIC. More...
struct	PowerFieldReading
	One scalar rail field read from a PMIC. More...
struct	PowerRailReading
	Snapshot for one rail. More...
struct	PowerSnapshot
	Point-in-time board power snapshot. More...
struct	PowerMonitorOptions
	Options for a PowerMonitor. More...
struct	PowerFieldSummary
	Summary statistics for one scalar rail field. More...
struct	PowerRailSummary
	Summary statistics for one rail. More...
struct	PowerSummary
	Aggregated power monitor result. More...
class	PowerMonitor
	Background SOM power sampler. More...
struct	RunAdvancedOptions
	Advanced runtime tuning knobs (most users never set these). More...
struct	RunOptions
	Per-Run runtime options. More...
struct	InputDropInfo
	Diagnostic record for a dropped input frame. More...
struct	InputStreamStats
	Per-Run input-side telemetry: counts, drops, and timing averages. More...
struct	RunStats
	Per-Run end-to-end statistics: counts and latency. More...
struct	RunMeasurementSummary
	One-call runtime measurement summary. More...
struct	RunStageStats
	Per-stage timing telemetry — how long each stage takes per sample. More...
struct	RunElementTimingStats
	Per-element timing — finer-grained than per-stage; one row per GStreamer element. More...
struct	RunElementFlowStats
	Per-element data-flow telemetry — buffer and byte counts, plus caps changes. More...
struct	RunElementPadTimingStats
	Per-pad timing — finest-grained telemetry, one row per (element, pad). More...
struct	RunDiagSnapshot
	Aggregate diagnostic snapshot: stages, boundaries, per-element, per-pad. More...
struct	MeasureOptions
	Options for framework-owned runtime measurement. More...
struct	MeasureLatencyStats
	Percentile summary for a measured latency series. More...
struct	MeasurePluginLatency
	Aggregated per-plugin/kernel timing captured during a measurement window. More...
struct	MeasureReport
	Framework-owned report returned by MeasureScope::stop(). More...
class	MeasureScope
	Observation scope for measuring an application-owned push/pull interval. More...
struct	RunReportOptions
	Toggles for what Run::report() includes in its formatted text output. More...
class	Run
	Live pipeline handle: push inputs in, pull outputs out. More...
struct	RuntimeMetricsOptions
	Options controlling collection and rendering of runtime metrics. More...
struct	RuntimeLatencyMetrics
	Basic latency summary in milliseconds. More...
struct	RuntimeCounters
	Common input/output/drop counters. More...
struct	RuntimeMetricValue
	Generic named scalar metric. More...
struct	RuntimeMetricGroup
	Generic named metric group for adapter-specific diagnostics. More...
struct	RuntimeMetrics
	Unified runtime metrics returned by Run, Model::Runner, GraphRun, and tools. More...
class	RtspServerHandle
	Live handle for a Session running in RTSP server mode. More...
class	Session
	The assembly stage — turns a list of Nodes into a runnable, deterministic pipeline. More...
class	SessionError
	Framework exception type carrying a structured SessionReport. More...
struct	VerboseOptions
	Per-topic framework verbosity controls. More...
struct	RtspServerOptions
	Options for Session::run_rtsp() — controls the RTSP server's mount point and ports. More...
struct	ValidateOptions
	Options for Session::validate(). More...
struct	ProcessCvuOptions
	Simple process-CVU backend placement for model pre/post stages. More...
struct	ProcessMlaOptions
	process-MLA execution options. More...
struct	PreparedRunnerOptions
	Experimental prepared-route runner options. More...
struct	SessionOptions
	Per-Session construction options. More...
struct	OutputTensorOptions
	Options for Session::add_output_tensor() — the tensor-friendly output helper. More...
struct	PullError
	Structured error returned by Run::pull() when status is Error. More...
struct	Sample
	Typed payload returned by Run::pull() and consumed by Run::push(). More...
struct	BusMessage
	A single GStreamer bus message captured during pipeline build/run. More...
struct	BoundaryFlowStats
	Per-boundary flow statistics — buffer counts at identity probes between Nodes. More...
struct	NodeReport
	Per-Node entry in the report, listing what that Node produced in the GStreamer pipeline. More...
struct	BuildAdaptationAction
	One line item from the build-adaptation log. More...
struct	BuildAdaptationSummary
	Snapshot of build-time adaptation state — present for build(input) flows. More...
struct	SessionReport
	Structured pipeline diagnostics — the framework's primary triage record. More...
struct	PcmAudio
struct	ConversionCost
	Cost estimate for a conversion: bytes moved + a coarse compute-class bucket (low/med/high). More...
struct	ConversionTrace
	Single audit-log entry for a conversion that occurred. More...
struct	ConversionTraceCollector
	Collects ConversionTrace entries for post-mortem auditing of a pipeline run. More...
struct	Device
	Device descriptor: type + numeric ID (for multi-device boards). More...
struct	ImageSpec
	Image-tensor metadata: pixel format and (optional) color space. More...
struct	AudioSpec
	Audio-tensor metadata: sample rate, channel count, interleaving. More...
struct	TokensSpec
	Token-tensor metadata for NLP-style tensors. More...
struct	TextSpec
	UTF-8 text tensor metadata. More...
struct	EncodedSpec
	Encoded-stream tensor metadata: which codec the bytes represent. More...
struct	ByteStreamSpec
	Opaque byte-stream tensor metadata. More...
struct	QuantSpec
	Quantization metadata for INT8/INT16 tensors. More...
struct	TessSpec
	Tessellation metadata — tile geometry for the MLA's tile-block layout. More...
struct	PreprocessRuntimeMeta
	Per-buffer preprocessing context — the inverse-transform breadcrumb trail. More...
struct	Semantic
	Discriminated union of "what this tensor represents". More...
struct	Mapping
	Scoped read/write window into a TensorBuffer. More...
struct	Segment
	One named memory segment within a multi-segment tensor buffer (e.g., separate Y / UV planes). More...
struct	TensorBuffer
	Storage handle for a tensor — opaque container for one of four backing memory kinds. More...
struct	Plane
	One plane of a composite (multi-plane) tensor. More...
struct	Nv12View
	Non-owning view into NV12 pixel data: Y plane + interleaved UV plane. More...
struct	Nv12Mapped
	Bundles an NV12 view with the Mapping that keeps its underlying buffer alive. More...
struct	I420View
	Non-owning view into I420 pixel data: separate Y, U, V planes. More...
struct	I420Mapped
	Bundles an I420 view with the Mapping that keeps its underlying buffer alive. More...
struct	TensorRouteMeta
	Routing metadata that travels with a tensor through multi-output pipelines. More...
struct	Tensor
	Universal tensor type — a labeled box of numbers that flows between Nodes. More...
struct	TensorConstraint
	Declarative tensor contract — describes the shape/dtype/device/format a tensor must satisfy. More...

Typedefs Index

using	BusMessageFn = void(*)(const char *type, const char *src, const std::string &line, void *user_data)
	Callback invoked for every bus message drained. More...
using	BusErrorFn = void(*)(const std::string &line, void *user_data)
	Callback invoked when a bus error message is encountered. More...
using	TensorSpec = TensorConstraint
	Tensor specification used by the Model API. More...
using	SampleList = std::vector< Sample >
	Convenience alias for a list of Samples (multi-input/multi-output payloads). More...
using	TensorList = std::vector< Tensor >
	Convenience alias for an ordered list of Tensors (multi-input/multi-output models). More...
using	ByteFormat = ByteStreamSpec::ByteFormat
	Convenience top-level alias for ByteStreamSpec::ByteFormat. More...
using	Storage = TensorBuffer
	Alias for backward-compatibility. New code prefers TensorBuffer directly. More...

Enumerations Index

enum class	MemoryContract { ... }
	How a Node (or Session) wants buffer memory to be sourced/handed back. More...
enum class	CapsMemory { ... }
	Memory class to advertise in GStreamer caps. More...
enum class	ContractFieldSource { ... }
	Where the value of a contract field originates. More...
enum class	ContractOverridePolicy { ... }
	Whether the Builder may override a contract field at build time. More...
enum class	ValidationSeverity { ... }
	Severity level for validation issues. More...
enum class	AutoFlag { ... }
	Tri-state knob: let the planner decide, force on, or force off. More...
enum class	InputKind { ... }
	What kind of input the user is feeding the preprocess pipeline. More...
enum class	ResizeMode { ... }
	How the resize stage maps non-matching input dimensions to the model's input shape. More...
enum class	PreprocessColorFormat { ... }
	Color format hint for the color-convert stage. More...
enum class	NormalizePreset { ... }
	Common normalize presets — convenient shorthand for famous mean/stddev pairs. More...
enum class	PreprocessGraphFamily { ... }
	Which preprocess graph family the planner selected. More...
enum class	TransformType { ... }
	Which transformation a Transform represents — used in user-supplied transform lists. More...
enum class	GraphTensorMaterializationKind : std::uint8_t { ... }
enum class	InputMemoryPolicy { ... }
	Where ingress buffers should be allocated when the application pushes samples. More...
enum class	CastDirection { ... }
	Direction of the BF16/FP32 conversion performed by a Cast Node. More...
enum class	BoxDecodeType : std::int32_t { ... }
	Decode families accepted by the BoxDecode backend. More...
enum class	BoxDecodeTypeOption : std::int32_t { ... }
	Tensor packing/layout option within a decode family. More...
enum class	FormatTag { ... }
	Identifies a media or tensor payload format. More...
enum class	PowerMonitorProfile { ... }
	Named board-level power monitor profiles. More...
enum class	OverflowPolicy { ... }
	What push() does when the input queue is full. More...
enum class	RunPreset { ... }
	Convenience preset bundles for RunOptions. More...
enum class	OutputMemory { ... }
	How output Tensors relate to the underlying GStreamer buffers. More...
enum class	RuntimeMetricsFormat { ... }
	Output format for runtime metrics reports. More...
enum class	VerbosityLevel { ... }
	Coarse-grained framework verbosity selector. More...
enum class	RunMode { ... }
	Timing mode a Run operates in. More...
enum class	SampleKind { ... }
	What kind of payload a Sample carries. More...
enum class	PullStatus { ... }
	Result status of Run::pull(). More...
enum class	ConversionKind { ... }
	What kind of transformation a tensor conversion represents. More...
enum class	ConversionPolicy { ... }
	How strict the framework is about implicit tensor conversions. More...
enum class	DeviceType { ... }
	Where a tensor's data lives — which processor can read it directly. More...
enum class	StorageKind { ... }
	How a TensorBuffer's memory was acquired and how to access it. More...
enum class	TensorMemory { ... }
	Placement for Tensor creators that materialize owned storage. More...
enum class	PlaneRole { ... }
	Role of a plane within a composite (multi-plane) tensor. More...
enum class	MapMode { ... }
	Access mode for TensorBuffer::map(). More...
enum class	TensorDType { ... }
	Element data type of a tensor. More...
enum class	TensorAxisSemantic : uint8_t { ... }
	Per-axis semantic tag — the long-term layout vocabulary. More...
enum class	TensorLayout { ... }
	Transition-only coarse layout token (HWC / CHW / HW). More...

Operators Index

std::ostream &	operator<< (std::ostream &os, const FormatSpec &spec)
	Stream-insert the canonical string form of spec. More...

Functions Index

const char *	to_string (ValidationSeverity s)
	Stable string label for a ValidationSeverity (for reports/logs). More...
std::string	gst_message_to_string (GstMessage *msg)
	Format a single GStreamer bus message into a stable, one-line string. More...
void	drain_bus (GstElement *pipeline, BusMessageFn on_message, void *user_data)
	Drain pending messages from the pipeline's bus, invoking on_message for each. More...
void	throw_if_bus_error (GstElement *pipeline, BusMessageFn on_message, void *user_data, BusErrorFn on_error, void *error_user_data)
	Drain the bus and throw a SessionError if any error message is found. More...
bool	element_exists (const char *factory)
	Returns true if a GStreamer element factory with factory name is registered. More...
bool	element_property_exists (const char *factory, const char *property_name)
	Returns true if factory exists and exposes a property named property_name. More...
std::string	factory_plugin_path (const char *factory)
	Returns the filesystem path of the plugin .so providing factory, or empty if unknown. More...
void	require_element (const char *factory, const char *context)
	Throw a SessionError if factory is not registered. More...
void	require_tensordecoder (const char *context)
	Throw a SessionError if SiMa's tensordecoder element is not available. More...
void	gst_init_once ()
	Initialize GStreamer exactly once across the process. More...
bool	prepare_processmla_runtime_config (ProcessMlaRuntimeConfig *runtime_cfg, std::string *error_message=nullptr)
bool	build_graph_processmla_prepared_stage (const GraphProcessMlaStageRequest &request, simaai::gst::ProcessMlaPreparedStage *out, std::string *error_message=nullptr)
bool	build_graph_processcvu_prepared_stage (const GraphProcessCvuStageRequest &request, simaai::gst::ProcessCvuPreparedStage *out, std::string *error_message=nullptr)
bool	build_prepared_stage_from_manifest_context (const GstContext *static_manifest_context, const char *stage_id_or_name, const char *element_name_fallback, simaai::gst::PreparedStageSpec *out, std::string *error_message=nullptr)
bool	attach_prepared_runtime_context (GstElement *pipeline, PreparedRuntimeDescriptor prepared_runtime, std::string *error_message=nullptr)
constexpr const char *	box_decode_type_token (BoxDecodeType type)
	Stable lower-case token for a decode family (used in caps, manifests, logs). More...
constexpr const char *	box_decode_type_option_token (BoxDecodeTypeOption option)
	Stable lower-case token for a decode-layout option. More...
constexpr bool	box_decode_type_is_yolo_family (BoxDecodeType type)
	True iff type is one of the YOLO-family detection or segmentation variants. More...
constexpr bool	box_decode_type_is_segmentation (BoxDecodeType type)
	True iff type is a segmentation variant (carries a mask head). More...
constexpr const char *	box_decode_type_contract_summary (BoxDecodeType type)
	Human-readable summary of type's tensor contract, used in API/docs/error surfaces. More...
std::vector< Box >	parse_bbox_bytes (const std::vector< uint8_t > &bytes, int img_w, int img_h, int expected_topk, bool strict)
	Parse a packed BBOX byte payload into typed Box records. More...
BoxDecodeResult	decode_bbox_tensor (const simaai::neat::Tensor &tensor, int img_w, int img_h, int expected_topk, bool strict)
	Decode a BBOX-format Tensor into a BoxDecodeResult. More...
simaai::neat::EncodedSpec::Codec	caps_to_codec (const std::string &caps_string)
	Map a GStreamer-style caps string to an EncodedSpec::Codec value. More...
Sample	make_encoded_sample (std::vector< uint8_t > bytes, std::string caps_string, int64_t pts_ns=-1, int64_t dts_ns=-1, int64_t duration_ns=-1)
	Build an encoded Sample from raw bytes and caps metadata. More...
const char *	format_tag_name (FormatTag tag)
	Stable, canonical string token for tag (empty string for Auto). More...
std::string	format_tag_to_string (FormatTag tag)
	std::string form of the canonical format token. More...
std::string	upper_copy_ascii (std::string value)
	ASCII upper-case copy of value (does not touch non-ASCII bytes). More...
bool	tensor_format_is_bf16_alias (std::string value)
	True iff value names BF16 (any of BF16, BFLOAT16, EVXX_BF16, EVXX_BFLOAT16). More...
bool	tensor_format_is_fp32_alias (std::string value)
	True iff value names FP32 (FP32, FLOAT32, EVXX_FLOAT32). More...
bool	tensor_format_is_int8_alias (std::string value)
	True iff value names INT8 (INT8, EVXX_INT8). More...
bool	tensor_format_is_int16_alias (std::string value)
	True iff value names INT16 (INT16, EVXX_INT16). More...
bool	tensor_format_is_int32_alias (std::string value)
	True iff value names INT32 (INT32, EVXX_INT32). More...
std::string	normalize_tensor_caps_format (std::string value)
	Normalize tensor format aliases to the canonical EVXX_* token. More...
std::string	normalize_caps_format_for_media (std::string media_type, std::string format)
	Normalize a caps-format string conditionally on its media type. More...
FormatTag	format_tag_from_string (const std::string &value)
	Parse a string token to a FormatTag; unknown values map to Auto. More...
bool	is_raw_video_format (FormatTag tag)
	True iff tag names a raw (uncompressed) video format. More...
bool	is_tensor_payload_format (FormatTag tag)
	True iff tag names a tensor payload (MLA, BBOX, ARGMAX, dtype kinds, EVXX aliases). More...
std::vector< PowerRailConfig >	default_modalix_som_power_rails ()
	Default Modalix SOM PMIC rail table from the board measurement scripts. More...
std::vector< PowerRailConfig >	default_modalix_dvt_power_rails ()
	Default Modalix DVT PMIC rail table. More...
PowerMonitorProfile	detect_default_power_monitor_profile ()
	Detect the most likely built-in board power profile for the current host. More...
std::string	power_monitor_profile_name (PowerMonitorProfile profile)
	Human-readable name for a power monitor profile. More...
std::vector< PowerRailConfig >	power_rails_for_profile (PowerMonitorProfile profile)
	Resolve a built-in board profile to its rail table. More...
PowerMonitorOptions	board_power_monitor_options (int sample_interval_ms=100, PowerMonitorProfile profile=PowerMonitorProfile::Auto)
	Convenience options enabling board power monitoring with optional auto-detect. More...
PowerMonitorOptions	modalix_som_power_monitor_options (int sample_interval_ms=100)
	Convenience options enabling Modalix SOM power monitoring. More...
PowerMonitorOptions	modalix_dvt_power_monitor_options (int sample_interval_ms=100)
	Convenience options enabling Modalix DVT power monitoring. More...
PowerSnapshot	read_power_snapshot (const PowerMonitorOptions &options)
	Read one immediate power snapshot using native Linux I2C access. More...
std::string	format_power_summary (const PowerSummary &summary)
	Format a compact human-readable power summary. More...
std::string	power_summary_to_json (const PowerSummary &summary, int indent=2)
	Format power summary as a JSON object. More...
std::string	format_runtime_metrics (const RuntimeMetrics &metrics, RuntimeMetricsFormat format=RuntimeMetricsFormat::Text)
	Render runtime metrics in the requested format. More...
std::string	runtime_metrics_to_json (const RuntimeMetrics &metrics, int indent=0)
	Render runtime metrics as JSON. More...
std::string	runtime_metrics_to_text (const RuntimeMetrics &metrics)
	Render runtime metrics as human-readable text. More...
std::string	runtime_metrics_to_compact_text (const RuntimeMetrics &metrics)
	Render runtime metrics as one compact line. More...
Sample	make_tensor_sample (const std::string &port_name, simaai::neat::Tensor tensor)
	Construct a TensorSet-kind Sample wrapping a single Tensor for the named port. More...
Sample	make_bundle_sample (std::initializer_list< Sample > fields)
	Construct a Bundle-kind Sample whose payload is the given list of inner Samples. More...
bool	sample_is_multi_output (const Sample &sample)
	Returns true if the sample carries multiple outputs (Bundle, or TensorSet with size > 1). More...
bool	sample_has_tensor_list (const Sample &sample)
	Returns true if the sample is a TensorSet with at least one tensor. More...
TensorList &	sample_tensor_list (Sample &sample, const char *where=nullptr)
	Mutable reference to the sample's tensor list. Throws SessionError if kind != TensorSet. More...
const TensorList &	sample_tensor_list (const Sample &sample, const char *where=nullptr)
	Const reference to the sample's tensor list. Throws SessionError if kind != TensorSet. More...
Tensor &	require_single_tensor (Sample &sample, const char *where=nullptr)
	Returns the sample's single Tensor; throws if the sample carries 0 or >1 tensors. More...
const Tensor &	require_single_tensor (const Sample &sample, const char *where=nullptr)
	Const variant of require_single_tensor. More...
TensorList	tensors_from_sample (const Sample &sample, bool require_nonempty=true)
	Extract all tensors from any sample shape (Tensor, TensorSet, Bundle); flattens recursively. More...
Sample	sample_from_tensors (const TensorList &tensors)
	Construct a TensorSet sample from a list of tensors. More...
Tensor	from_gst_sample (GstSample *sample)
PcmAudio	tensor_to_pcm_audio (const Tensor &tensor)
ConversionCost	estimate_conversion_cost (ConversionKind kind, std::uint64_t bytes_copied)
	Estimate the cost (bytes + compute class) of a conversion of the given kind moving bytes_copied bytes. More...
bool	conversion_allowed (ConversionPolicy policy, ConversionKind kind)
	Returns true if the policy allows a conversion of this kind (without performing it). More...
std::shared_ptr< TensorBuffer >	make_cpu_owned_storage (std::size_t size_bytes)
	Allocate a CPU-owned heap buffer of size_bytes, initialized to zero. More...
std::shared_ptr< TensorBuffer >	make_cpu_external_storage (void *data, std::size_t size_bytes, std::shared_ptr< void > holder={}, bool read_only=true)
	Wrap a foreign CPU pointer as a TensorBuffer without copying. More...
bool	is_tessellated_int8_format (const std::string &fmt)
	True iff fmt names an INT8 tessellated tensor format string. More...
bool	is_tessellated_bf16_format (const std::string &fmt)
	True iff fmt names a BF16 tessellated tensor format string. More...

Typedefs

BusErrorFn

using simaai::neat::BusErrorFn = typedef void (*)(const std::string& line, void* user_data)

Callback invoked when a bus error message is encountered.

Definition at line 36 of file GstBusWatch.h.

36using BusErrorFn = void (*)(const std::string& line, void* user_data);

BusMessageFn

using simaai::neat::BusMessageFn = typedef void (*)(const char* type, const char* src, const std::string& line, void* user_data)

Callback invoked for every bus message drained.

Parameters

type	GStreamer message type as a stable string (e.g., "warning", "info").
src	Name of the element that posted the message, or empty.
line	Pre-formatted single-line summary of the message.
user_data	Opaque pointer the caller registered.

Definition at line 32 of file GstBusWatch.h.

33 void* user_data);

ByteFormat

using simaai::neat::ByteFormat = typedef ByteStreamSpec::ByteFormat

Convenience top-level alias for ByteStreamSpec::ByteFormat.

Definition at line 216 of file TensorCore.h.

216using ByteFormat = ByteStreamSpec::ByteFormat;

SampleList

using simaai::neat::SampleList = typedef std::vector<Sample>

Convenience alias for a list of Samples (multi-input/multi-output payloads).

Definition at line 463 of file SessionOptions.h.

463using SampleList = std::vector<Sample>;

Storage

using simaai::neat::Storage = typedef TensorBuffer

Alias for backward-compatibility. New code prefers TensorBuffer directly.

Definition at line 461 of file TensorCore.h.

461using Storage = TensorBuffer;

TensorList

using simaai::neat::TensorList = typedef std::vector<Tensor>

Convenience alias for an ordered list of Tensors (multi-input/multi-output models).

Definition at line 23 of file Tensor.h.

23using TensorList = std::vector<Tensor>;

TensorSpec

using simaai::neat::TensorSpec = typedef TensorConstraint

Tensor specification used by the Model API.

Definition at line 47 of file Model.h.

47using TensorSpec = TensorConstraint;

Enumerations

AutoFlag

enum class AutoFlag

strong

Tri-state knob: let the planner decide, force on, or force off.

Enumeration values

Auto	Planner decides based on model contract and inputs (= 0)
On	Force this stage on (= 1)
Off	Force this stage off (= 2)

Used pervasively across PreprocessOptions so the application can express "I don't care, do the right thing" (Auto) without losing the ability to override.

Definition at line 32 of file PreprocessPlan.h.

32enum class AutoFlag {

33 Auto = 0,

34 On = 1,

35 Off = 2,

36};

BoxDecodeType

enum class BoxDecodeType : std::int32_t

strong

Decode families accepted by the BoxDecode backend.

Enumeration values

Unspecified	Sentinel: no decode family selected (fails fast at runtime) (= 0)
Yolo	Generic YOLO family token (= 1)
YoloV5	YOLOv5 detection (= 2)
YoloV5Seg	YOLOv5 segmentation (= 3)
YoloV7	YOLOv7 detection (= 4)
YoloV7Seg	YOLOv7 segmentation (= 5)
YoloV8	YOLOv8 detection (= 6)
YoloV8Seg	YOLOv8 segmentation (= 7)
YoloV8Pose	YOLOv8 pose-estimation heads (= 8)
YoloV9	YOLOv9 detection (= 9)
YoloV9Seg	YOLOv9 segmentation (= 10)
YoloV10	YOLOv10 detection (= 11)
YoloV10Seg	YOLOv10 segmentation (= 12)
Detr	DETR-style transformer detection (= 13)
EffDet	EfficientDet detection (= 14)
RcnnStage1	Region-proposal stage of two-stage R-CNN models (= 15)
Centernet	CenterNet keypoint-style detection (= 16)
YoloV26	YOLO26 detection (raw l/t/r/b distance heads) (= 17)

Unspecified is an internal unset sentinel and must fail fast before runtime decode. Most YOLO-family variants share the same class-inference contract in genericboxdecode_v2:

• decoupled heads: repeated class-depth tensors, class depth > 4
• packed heads: depth = 3 * (num_classes + 5), consistent across heads YoloV26 uses decoupled 4-channel raw l/t/r/b bbox heads paired with class heads.

Definition at line 33 of file BoxDecodeType.h.

33enum class BoxDecodeType : std::int32_t {

34 // Internal sentinel; runtime decode requires a concrete value.

35 Unspecified = 0,

36 // YOLO-family (generic token).

37 Yolo = 1,

38 YoloV5 = 2,

39 YoloV5Seg = 3,

40 YoloV7 = 4,

41 YoloV7Seg = 5,

42 YoloV8 = 6,

43 YoloV8Seg = 7,

44 YoloV8Pose = 8,

45 YoloV9 = 9,

46 YoloV9Seg = 10,

47 YoloV10 = 11,

48 YoloV10Seg = 12,

49 Detr = 13,

50 EffDet = 14,

51 RcnnStage1 = 15,

52 Centernet = 16,

53 YoloV26 = 17,

54};

BoxDecodeTypeOption

enum class BoxDecodeTypeOption : std::int32_t

strong

Tensor packing/layout option within a decode family.

Enumeration values

Auto	Backend infers the layout from tensor shapes (= 0)
PackedPerHead	Each head holds a single packed tensor (box+obj+cls) (= 1)
InterleavedByHead	Heads interleaved within tensors (= 2)
GroupedByRole	Tensors grouped by role (box, score, class) (= 3)
Split3Interleaved	Three split tensors, head-interleaved (= 4)
Split3Grouped	Three split tensors, grouped by role (= 5)
InterleavedByHeadProbability	Interleaved-by-head, class scores as probabilities (= 6)
InterleavedByHeadLogit	Interleaved-by-head, class scores as logits (= 7)
GroupedByRoleProbability	Grouped-by-role, class scores as probabilities (= 8)
GroupedByRoleLogit	Grouped-by-role, class scores as logits (= 9)

Some families admit multiple equivalent head layouts (packed vs interleaved, grouped by role, probability vs logit class scores). Auto lets the planner pick from observed tensor geometry; the explicit values force a particular decoding contract for ambiguous models.

Definition at line 66 of file BoxDecodeType.h.

66enum class BoxDecodeTypeOption : std::int32_t {

67 Auto = 0,

68 PackedPerHead = 1,

69 InterleavedByHead = 2,

70 GroupedByRole = 3,

71 Split3Interleaved = 4,

72 Split3Grouped = 5,

73 InterleavedByHeadProbability = 6,

74 InterleavedByHeadLogit = 7,

75 GroupedByRoleProbability = 8,

76 GroupedByRoleLogit = 9,

77};

CapsMemory

enum class CapsMemory

strong

Memory class to advertise in GStreamer caps.

Enumeration values

Any	Don't constrain memory class in caps (= 0)
SystemMemory	Force caps to advertise memory:SystemMemory

Any means "don't constrain in caps"; SystemMemory forces the caps to carry memory:SystemMemory so upstream/downstream negotiation picks CPU-mappable buffers.

Definition at line 54 of file ContractTypes.h.

54enum class CapsMemory {

55 Any = 0,

56 SystemMemory,

57};

CastDirection

enum class CastDirection

strong

Direction of the BF16/FP32 conversion performed by a Cast Node.

Enumeration values

Bf16ToFp32	Read BF16, emit FP32 (typical post-MLA path) (= 0)
Fp32ToBf16	Read FP32, emit BF16 (typical pre-MLA path) (= 1)

Definition at line 30 of file Cast.h.

30enum class CastDirection {

31 Bf16ToFp32 = 0,

32 Fp32ToBf16 = 1,

33};

ContractFieldSource

enum class ContractFieldSource

strong

Where the value of a contract field originates.

Enumeration values

Fixed	Value is hard-coded in the contract definition
BuilderOption	Value comes from a builder-level option/argument
ModelOnly	Value is supplied by the model file (e.g., MPK manifest)
InputOnly	Value is determined by external input only
UpstreamOnly	Value is derived from the upstream Node's output spec
GraphOwned	Value is owned by the enclosing Graph (cross-Node)

Used by the Builder to decide whether a field is fully bound at definition time, supplied by a builder option, derived from upstream caps, baked into a model file, or owned by the Graph as a whole.

Definition at line 31 of file NodeContractDefinition.h.

31enum class ContractFieldSource {

32 Fixed,

33 BuilderOption,

34 ModelOnly,

35 InputOnly,

36 UpstreamOnly,

37 GraphOwned,

38};

ContractOverridePolicy

enum class ContractOverridePolicy

strong

Whether the Builder may override a contract field at build time.

Enumeration values

Forbidden	Field cannot be overridden; attempts are validation errors
BuilderOnly	Builder code may override; user-facing API may not

Definition at line 44 of file NodeContractDefinition.h.

44enum class ContractOverridePolicy {

45 Forbidden,

46 BuilderOnly,

47};

ConversionKind

enum class ConversionKind

strong

What kind of transformation a tensor conversion represents.

Enumeration values

Reinterpret	Same bytes, different dtype/layout label. No memory traffic (= 0)
View	Pointer-arithmetic view (sub-tensor or strided slice). No copy
Pack	Byte-level rearrangement in place (e.g., layout swap, packing)
Convert	Element-wise math conversion (e.g., FP32 → BF16, INT8 → FP32)
Transfer	Cross-device transfer (e.g., CPU → accelerator scratch memory)

Ordered roughly cheapest-to-most-expensive: a Reinterpret is free (just a type/dtype relabel), a View is free (sub-tensor pointer arithmetic), a Pack rearranges bytes in-place, a Convert does math (e.g., FP32 → BF16), and a Transfer moves data across device boundaries (e.g., CPU → MLA scratchpad).

Definition at line 31 of file TensorConversion.h.

31enum class ConversionKind {

32 Reinterpret = 0,

33 View,

34 Pack,

35 Convert,

36 Transfer,

37};

ConversionPolicy

enum class ConversionPolicy

strong

How strict the framework is about implicit tensor conversions.

Enumeration values

Strict	Refuse implicit conversions; require explicit user code (= 0)
AllowWithTrace	Allow but record each conversion to a trace collector
AllowSilent	Allow without tracking

Strict (default for many internal validations) refuses any implicit conversion; AllowWithTrace permits the conversion but records it to a ConversionTraceCollector for later auditing; AllowSilent permits without recording.

Definition at line 47 of file TensorConversion.h.

47enum class ConversionPolicy {

48 Strict = 0,

49 AllowWithTrace,

50 AllowSilent,

51};

DeviceType

enum class DeviceType

strong

Where a tensor's data lives — which processor can read it directly.

Enumeration values

CPU	Host A65 (general-purpose CPU) (= 0)
SIMA_APU	Audio Processing Unit
SIMA_CVU	Compute/Vector Unit (EV74)
SIMA_MLA	Machine Learning Accelerator
UNKNOWN	Placement not specified

The Modalix's heterogeneous compute is reflected here. CPU is the host A65; SIMA_CVU is the EV74 vector processor's accessible memory; SIMA_MLA is the ML accelerator's scratchpad; SIMA_APU is the audio processing unit. Tensors with the same buffer can sometimes be accessed from multiple devices (zero-copy via unified IOMMU); other times an explicit transfer is required.

Definition at line 57 of file TensorCore.h.

57enum class DeviceType {

58 CPU = 0,

59 SIMA_APU,

60 SIMA_CVU,

61 SIMA_MLA,

62 UNKNOWN,

63};

FormatTag

enum class FormatTag

strong

Identifies a media or tensor payload format.

Enumeration values

Auto	Unset / framework decides (= 0)
RGB	Packed RGB, 8 bits per channel
BGR	Packed BGR, 8 bits per channel (OpenCV default)
GRAY8	Single-plane 8-bit grayscale
NV12	YUV 4:2:0, Y plane + interleaved UV plane
I420	YUV 4:2:0, three planes (Y, U, V)
YUYV	YUV 4:2:2 packed (Y0 U Y1 V)
ENCODED	Generic encoded payload (codec from caps)
H264	H.264 access unit / NAL stream
ByteStream	Opaque byte stream; downstream interprets bytes by contract
MLA	MLA-tessellated tensor payload
BBOX	Decoded bounding-box byte stream
ARGMAX	Argmax/segmentation map
DETESSDEQUANT	Detessellated + dequantized tensor payload
FP32	IEEE-754 32-bit float tensor
INT8	Signed 8-bit integer tensor
UINT8	Unsigned 8-bit integer tensor
BF16	bfloat16 tensor
EVXX_FLOAT32	EV-side alias for FP32
EVXX_INT8	EV-side alias for INT8
EVXX_BFLOAT16	EV-side alias for BF16

Used in caps strings and option fields throughout the pipeline. Auto is the unset sentinel (let the framework pick or sniff). The EVXX_ variants are aliases preferred by the EV74 caps surface.

See Also

FormatSpec

Definition at line 36 of file FormatSpec.h.

36enum class FormatTag {

37 Auto = 0,

38 RGB,

39 BGR,

40 GRAY8,

41 NV12,

42 I420,

43 YUYV,

44 ENCODED,

45 H264,

46 ByteStream,

47 MLA,

48 BBOX,

49 ARGMAX,

50 DETESSDEQUANT,

51 FP32,

52 INT8,

53 UINT8,

54 BF16,

55 EVXX_FLOAT32,

56 EVXX_INT8,

57 EVXX_BFLOAT16,

58};

GraphTensorMaterializationKind

enum class GraphTensorMaterializationKind : std::uint8_t

strong

Enumeration values

Unknown	(= 0)
Direct	(= 1)
OffsetView	(= 2)
Bf16LaneSplitRepack	(= 3)

Definition at line 19 of file PreparedRuntimeBridge.h.

19enum class GraphTensorMaterializationKind : std::uint8_t {

20 Unknown = 0,

21 Direct = 1,

22 OffsetView = 2,

23 Bf16LaneSplitRepack = 3,

24};

InputKind

enum class InputKind

strong

What kind of input the user is feeding the preprocess pipeline.

Enumeration values

Auto	Planner infers from caps / sample type (= 0)
Image	Decoded image (planar/packed pixels) (= 1)
Tensor	Already-shaped tensor; preprocess is mostly a pass-through (= 2)

Definition at line 40 of file PreprocessPlan.h.

40enum class InputKind {

41 Auto = 0,

42 Image = 1,

43 Tensor = 2,

44};

InputMemoryPolicy

enum class InputMemoryPolicy

strong

Where ingress buffers should be allocated when the application pushes samples.

Enumeration values

Auto	Defer to legacy / session-level target selection (= 0)
Ev74	Allocate in EV74-visible memory
Dms0	Allocate in DMS0 memory
SystemMemory	Allocate in plain system memory

Auto defers to legacy / session-level selection; the explicit values force a specific memory target.

Definition at line 62 of file Input.h.

62enum class InputMemoryPolicy {

63 Auto = 0,

64 Ev74,

65 Dms0,

66 SystemMemory,

67};

MapMode

enum class MapMode

strong

Access mode for TensorBuffer::map().

Enumeration values

Read	Read-only map; framework may skip cache invalidation if known clean (= 0)
Write	Write-only map; framework may skip read-back; flushes cache on unmap
ReadWrite	Both directions; framework treats as full ownership transfer for the mapping's lifetime

Affects whether the framework treats the mapping as read or write.

Definition at line 122 of file TensorCore.h.

122enum class MapMode {

123 Read = 0,

124 Write,

125 ReadWrite,

127};

MemoryContract

enum class MemoryContract

strong

How a Node (or Session) wants buffer memory to be sourced/handed back.

Enumeration values

RequireSystemMemoryMappable	Must be CPU-mappable (typically SystemMemory); violations are hard errors (= 0)
PreferDeviceZeroCopy	Prefer device/zero-copy; runner may avoid forcing SystemMemory but still reports mismatches
AllowEitherButReport	Allow either; if non-mappable, return empty payload with explicit reason + location

Drives the runner's behavior at output time and contributes to caps negotiation: requiring CPU-mappable memory forces a (potentially copying) conversion, while PreferDeviceZeroCopy lets the runner keep device memory whenever possible.

Definition at line 30 of file ContractTypes.h.

30enum class MemoryContract {

32 RequireSystemMemoryMappable = 0,

33

35 PreferDeviceZeroCopy,

36

38 AllowEitherButReport,

39};

NormalizePreset

enum class NormalizePreset

strong

Common normalize presets — convenient shorthand for famous mean/stddev pairs.

Enumeration values

None	Use the explicit mean/stddev fields (= 0)
ImageNet	Standard ImageNet stats (= 1)
COCO_YOLO	Stats commonly used by YOLO-family detectors trained on COCO (= 2)

Definition at line 67 of file PreprocessPlan.h.

67enum class NormalizePreset {

68 None = 0,

69 ImageNet = 1,

70 COCO_YOLO = 2,

71};

OutputMemory

enum class OutputMemory

strong

How output Tensors relate to the underlying GStreamer buffers.

Enumeration values

Auto	(= 0)
ZeroCopy
Owned

Auto lets the framework pick based on platform and pipeline shape. ZeroCopy shares storage with GStreamer (faster but lifetime-coupled to the Run). Owned copies into a framework-owned buffer (safer, slightly slower).

Definition at line 88 of file Run.h.

88enum class OutputMemory {

89 Auto = 0,

90 ZeroCopy,

91 Owned,

92};

OverflowPolicy

enum class OverflowPolicy

strong

What push() does when the input queue is full.

Enumeration values

Block	push() blocks until queue space frees up. Lossless. Use for batch processing (= 0)
KeepLatest	Drop the oldest queued frame to make room
DropIncoming	Drop the new frame; keep what's queued

The right choice depends on the input source — file batches want lossless, live cameras want freshness, network feeds with chokepoint pipelines want bounded memory.

Definition at line 59 of file Run.h.

59enum class OverflowPolicy {

60 Block = 0,

61 KeepLatest,

63 DropIncoming,

65};

PlaneRole

enum class PlaneRole

strong

Role of a plane within a composite (multi-plane) tensor.

Enumeration values

Unknown	(= 0)
Y	Luminance plane
U	Chrominance-U plane (I420)
V	Chrominance-V plane (I420)
UV	Interleaved chrominance plane (NV12)

Composite tensors carry their planes as separate Plane records. NV12 has Y + UV; I420 has Y + U + V; single-plane formats use the default Unknown.

Definition at line 112 of file TensorCore.h.

112enum class PlaneRole {

113 Unknown = 0,

114 Y,

115 U,

116 V,

117 UV,

118};

PowerMonitorProfile

enum class PowerMonitorProfile

strong

Named board-level power monitor profiles.

Enumeration values

Auto	Detect from the running board model when possible
ModalixSom	Modalix SOM PMIC table from the board scripts
ModalixDvt	Modalix DVT stable single-PMIC power path
Custom	Caller-provided explicit rail table

Definition at line 20 of file PowerTelemetry.h.

20enum class PowerMonitorProfile {

21 Auto,

22 ModalixSom,

23 ModalixDvt,

24 Custom,

25};

PreprocessColorFormat

enum class PreprocessColorFormat

strong

Color format hint for the color-convert stage.

Enumeration values

Auto	Planner picks based on caps/contract (= 0)
RGB	Packed 8-bit RGB (= 1)
BGR	Packed 8-bit BGR (= 2)
GRAY8	8-bit single-channel grayscale (= 3)
NV12	Y plane + interleaved UV plane (semi-planar 4:2:0) (= 4)
I420	Y, U, V planes (planar 4:2:0) (= 5)

Definition at line 56 of file PreprocessPlan.h.

56enum class PreprocessColorFormat {

57 Auto = 0,

58 RGB = 1,

59 BGR = 2,

60 GRAY8 = 3,

61 NV12 = 4,

62 I420 = 5,

63};

PreprocessGraphFamily

enum class PreprocessGraphFamily

strong

Which preprocess graph family the planner selected.

Enumeration values

Disabled	No preprocess graph (model accepts raw tensors) (= 0)
Preproc	Preprocess only; quant/tess happen inside the MLA (= 1)
Quant	Preprocess + quant; tess happens inside the MLA (= 2)
Tess	Preprocess + tess; quant happens inside the MLA (= 3)
QuantTess	Preprocess + quant + tess; nothing happens inside the MLA (= 4)

The four variants correspond to the four corners of the dtype-contract decision tree (BF16/INT8 × inside/outside MLA tess).

Definition at line 76 of file PreprocessPlan.h.

76enum class PreprocessGraphFamily {

77 Disabled = 0,

78 Preproc = 1,

79 Quant = 2,

80 Tess = 3,

81 QuantTess = 4,

82};

PullStatus

enum class PullStatus

strong

Result status of Run::pull().

Enumeration values

Ok	A sample is available in the output parameter
Timeout	The wait elapsed without a sample arriving
Closed	The pipeline has reached EOS; no more samples will come
Error	A runtime error occurred; check the optional PullError

Definition at line 349 of file SessionOptions.h.

349enum class PullStatus {

350 Ok,

351 Timeout,

352 Closed,

353 Error,

354};

ResizeMode

enum class ResizeMode

strong

How the resize stage maps non-matching input dimensions to the model's input shape.

Enumeration values

Stretch	Anisotropic scale to fit; aspect ratio not preserved (= 0)
Letterbox	Scale to fit while preserving aspect; pad the rest with pad_value (= 1)
Crop	Center-crop to the target shape after isotropic scale (= 2)

Definition at line 48 of file PreprocessPlan.h.

48enum class ResizeMode {

49 Stretch = 0,

50 Letterbox = 1,

51 Crop = 2,

52};

RunMode

enum class RunMode

strong

Timing mode a Run operates in.

Enumeration values

Async	Continuous pipeline; user pushes/pulls asynchronously
Sync	One frame in, one result out, synchronously

Async runs the pipeline continuously with internal worker threads; user code pushes and pulls at its own pace. Sync runs one frame at a time on the calling thread. Choose based on the input source: streaming sources → Async; one-shot/batch → Sync.

See Also

Run

Definition at line 192 of file SessionOptions.h.

192enum class RunMode {

193 Async,

194 Sync,

195};

RunPreset

enum class RunPreset

strong

Convenience preset bundles for RunOptions.

Enumeration values

Realtime	Low-latency; small queues; KeepLatest overflow; metrics off
Balanced	Default; moderate queues; Block overflow; metrics off
Reliable	Lossless; deeper queues; Block overflow; metrics on

Each preset adjusts queue depth, overflow policy, and metrics flags to a profile that's known to work well for one workload class.

Definition at line 74 of file Run.h.

74enum class RunPreset {

75 Realtime,

76 Balanced,

77 Reliable,

78};

RuntimeMetricsFormat

enum class RuntimeMetricsFormat

strong

Output format for runtime metrics reports.

Enumeration values

Text	Human-readable multi-line report
Json	JSON object suitable for tools
CompactText	One-line headline summary

Definition at line 21 of file RuntimeMetrics.h.

21enum class RuntimeMetricsFormat {

22 Text,

23 Json,

24 CompactText

25};

SampleKind

enum class SampleKind

strong

What kind of payload a Sample carries.

Enumeration values

Tensor	Single tensor payload (the tensor field is set)
TensorSet	Multiple tensors at one logical output index (the tensors field is set)
Bundle	Recursive: payload is a vector of Samples (the fields field is set)
Unknown	Default/uninitialized

The framework's outputs come in three shapes depending on the model's output topology: a single Tensor, a flat list of Tensors (TensorSet), or a recursive Bundle of Samples (Bundle, used by multi-logical-output models).

See Also

Sample

Definition at line 338 of file SessionOptions.h.

338enum class SampleKind {

339 Tensor,

340 TensorSet,

341 Bundle,

342 Unknown,

343};

StorageKind

enum class StorageKind

strong

How a TensorBuffer's memory was acquired and how to access it.

Enumeration values

CpuOwned	Framework allocated; freed when the storage refcount hits zero (= 0)
CpuExternal	Wraps a foreign pointer; lifetime tracked via holder
GstSample	References a GStreamer pool sample; refcounted via GStreamer
DeviceHandle	Lives in accelerator memory; CPU access requires an explicit transfer
Unknown

Determines the framework's strategy for map() and lifetime management. CpuOwned is the default — framework allocated, framework frees. CpuExternal wraps a foreign pointer (cv::Mat data, NumPy array, mmap'd file). GstSample carries a reference to a GStreamer buffer pool sample. DeviceHandle lives in accelerator memory (e.g., MLA scratchpad) and requires an explicit transfer for CPU access.

Definition at line 81 of file TensorCore.h.

81enum class StorageKind {

82 CpuOwned = 0,

83 CpuExternal,

84 GstSample,

85 DeviceHandle,

86 Unknown,

87};

TensorAxisSemantic

enum class TensorAxisSemantic : uint8_t

strong

Per-axis semantic tag — the long-term layout vocabulary.

Enumeration values

Unknown	Axis role not annotated (= 0)
N	Batch axis
D	Depth axis (5D tensors: batch + depth + spatial)
H	Height (spatial)
W	Width (spatial)
C	Channel axis

Each axis of a tensor is tagged with a role: batch (N), depth (D), height (H), width (W), or channel (C). Together with shape and strides, this is the authoritative description of tensor layout and supersedes the coarse TensorLayout token for generic 3D/4D/5D tensors.

Definition at line 103 of file TensorTypes.h.

103enum class TensorAxisSemantic : uint8_t {

104 Unknown = 0,

105 N,

106 D,

107 H,

108 W,

109 C,

110};

TensorDType

enum class TensorDType

strong

Element data type of a tensor.

Enumeration values

UInt8	8-bit unsigned integer (typical pixel values)
Int8	8-bit signed integer (typical quantized weights/activations)
UInt16	16-bit unsigned integer
Int16	16-bit signed integer
Int32	32-bit signed integer
BFloat16	16-bit "brain float" (1 sign + 8 exponent + 7 mantissa bits) — MLA's FP path
Float32	IEEE 754 single-precision (the user-facing default)
Float64	IEEE 754 double-precision

Note: floating-point variants include BFloat16 (16-bit "brain float", same dynamic range as Float32), Float32, and Float64. Float16 (IEEE half) is intentionally omitted — the framework targets BF16 for the MLA's floating-point path; FP16 has no first-class path.

Definition at line 83 of file TensorTypes.h.

83enum class TensorDType {

84 UInt8,

85 Int8,

86 UInt16,

87 Int16,

88 Int32,

89 BFloat16,

90 Float32,

91 Float64,

92};

TensorLayout

enum class TensorLayout

strong

Transition-only coarse layout token (HWC / CHW / HW).

Enumeration values

Unknown	(= 0)
HWC	Height × Width × Channels (image-natural layout)
CHW	Channels × Height × Width (PyTorch-natural layout)
HW	Height × Width (single-channel / grayscale)

This is not the long-term semantic truth for generic tensors. Internal tensor semantics should move toward shape + strides + explicit axis_semantics, with layout tokens remaining only at boundary/compatibility surfaces during the migration. Prefer TensorAxisSemantic for new code.

See Also

TensorAxisSemantic

Definition at line 122 of file TensorTypes.h.

122enum class TensorLayout {

123 Unknown = 0,

124 HWC,

125 CHW,

126 HW,

127};

TensorMemory

enum class TensorMemory

strong

Placement for Tensor creators that materialize owned storage.

Enumeration values

EV74	(= 0)
CPU
MLA
A65
Auto

The default for new owning creators is EV74: DDR allocated from the SiMa allocator in the EV74-addressable window (not EV74 VCCM). Use CPU for ordinary malloc-backed host tensors, or MLA for DMS0-backed tensors when the runtime allocator is available.

Definition at line 97 of file TensorCore.h.

97enum class TensorMemory {

98 EV74 = 0,

99 CPU,

100 MLA,

101 A65,

102 Auto,

103};

TransformType

enum class TransformType

strong

Which transformation a Transform represents — used in user-supplied transform lists.

Enumeration values

Resize	Resize / letterbox / crop (= 0)
ColorConvert	Color-format conversion (= 1)
LayoutConvert	Axis permutation (e.g., HWC → CHW) (= 2)
Normalize	Mean/stddev normalization (= 3)
Quantize	INT8 quantization (= 4)
Tessellate	MLA-tile reshuffle (= 5)

Definition at line 86 of file PreprocessPlan.h.

86enum class TransformType {

87 Resize = 0,

88 ColorConvert = 1,

89 LayoutConvert = 2,

90 Normalize = 3,

91 Quantize = 4,

92 Tessellate = 5,

93};

ValidationSeverity

enum class ValidationSeverity

strong

Severity level for validation issues.

Enumeration values

Info	Informational only; never blocks a run (= 0)
Warning	Soft contract violation; pipeline may still run
Error	Hard contract violation; pipeline must not run

Definition at line 29 of file ValidationReport.h.

29enum class ValidationSeverity {

30 Info = 0,

31 Warning,

32 Error,

33};

VerbosityLevel

enum class VerbosityLevel

strong

Coarse-grained framework verbosity selector.

Enumeration values

Quiet	Suppress topic-specific messages; progress is controlled separately
Production	Concise phase updates suitable for end users
Verbose	Rich human-facing diagnostics across all framework topics

Combined with the per-topic flags in VerboseOptions to decide what diagnostic output the framework emits at runtime. Most users pick one of the three preset factories on VerboseOptions (quiet(), production(), debug_all()) rather than setting flags by hand.

Definition at line 49 of file SessionOptions.h.

49enum class VerbosityLevel {

50 Quiet,

51 Production,

52 Verbose,

53};

Operators

operator<<()

std::ostream& simaai::neat::operator<< (std::ostream & os, const FormatSpec & spec)

inline

Stream-insert the canonical string form of spec.

Definition at line 332 of file FormatSpec.h.

332inline std::ostream& operator<<(std::ostream& os, const FormatSpec& spec) {

333 os << spec.str();

334 return os;

335}

Functions

attach_prepared_runtime_context()

bool simaai::neat::attach_prepared_runtime_context (GstElement * pipeline, PreparedRuntimeDescriptor prepared_runtime, std::string * error_message=nullptr)

Definition at line 120 of file PreparedRuntimeBridge.h.

board_power_monitor_options()

PowerMonitorOptions simaai::neat::board_power_monitor_options (int sample_interval_ms=100, PowerMonitorProfile profile=PowerMonitorProfile::Auto)

Convenience options enabling board power monitoring with optional auto-detect.

Definition at line 171 of file PowerTelemetry.h.

box_decode_type_contract_summary()

constexpr const char* simaai::neat::box_decode_type_contract_summary (BoxDecodeType type)

constexpr

Human-readable summary of type's tensor contract, used in API/docs/error surfaces.

Definition at line 205 of file BoxDecodeType.h.

205constexpr const char* box_decode_type_contract_summary(BoxDecodeType type) {

206 switch (type) {

207 case BoxDecodeType::Yolo:

208 case BoxDecodeType::YoloV5:

209 case BoxDecodeType::YoloV7:

210 case BoxDecodeType::YoloV8:

211 case BoxDecodeType::YoloV8Pose:

212 case BoxDecodeType::YoloV9:

213 case BoxDecodeType::YoloV10:

214 return "YOLO tensor contract: decoupled class heads (>4 channels, repeated across heads) or "

215 "packed heads (depth=3*(num_classes+5), consistent across heads).";

216 case BoxDecodeType::YoloV26:

217 return "YOLO26 tensor contract: grouped raw l/t/r/b bbox heads (4 channels) paired "

218 "with repeated class-score heads (>4 channels).";

219 case BoxDecodeType::YoloV5Seg:

220 case BoxDecodeType::YoloV7Seg:

221 case BoxDecodeType::YoloV8Seg:

222 case BoxDecodeType::YoloV9Seg:

223 case BoxDecodeType::YoloV10Seg:

224 return "YOLO-seg contract: same class-depth rules as YOLO plus segmentation decode path.";

225 case BoxDecodeType::Detr:

226 return "DETR contract: class channels inferred from max depth across heads; max depth must be "

227 ">4.";

228 case BoxDecodeType::EffDet:

229 case BoxDecodeType::RcnnStage1:

230 case BoxDecodeType::Centernet:

231 return "Non-YOLO fallback contract: class channels inferred from max depth across heads; "

232 "max depth must be >4.";

233 case BoxDecodeType::Unspecified:

234 default:

235 return "Unset decode type; runtime decode must fail fast until a concrete decode family is "

236 "selected.";

237 }

238}

box_decode_type_is_segmentation()

constexpr bool simaai::neat::box_decode_type_is_segmentation (BoxDecodeType type)

constexpr

True iff type is a segmentation variant (carries a mask head).

Definition at line 178 of file BoxDecodeType.h.

178constexpr bool box_decode_type_is_segmentation(BoxDecodeType type) {

179 switch (type) {

180 case BoxDecodeType::YoloV5Seg:

181 case BoxDecodeType::YoloV7Seg:

182 case BoxDecodeType::YoloV8Seg:

183 case BoxDecodeType::YoloV9Seg:

184 case BoxDecodeType::YoloV10Seg:

185 return true;

186 case BoxDecodeType::Yolo:

187 case BoxDecodeType::YoloV5:

188 case BoxDecodeType::YoloV7:

189 case BoxDecodeType::YoloV8:

190 case BoxDecodeType::YoloV8Pose:

191 case BoxDecodeType::YoloV9:

192 case BoxDecodeType::YoloV10:

193 case BoxDecodeType::YoloV26:

194 case BoxDecodeType::Detr:

195 case BoxDecodeType::EffDet:

196 case BoxDecodeType::RcnnStage1:

197 case BoxDecodeType::Centernet:

198 case BoxDecodeType::Unspecified:

199 default:

200 return false;

201 }

202}

box_decode_type_is_yolo_family()

constexpr bool simaai::neat::box_decode_type_is_yolo_family (BoxDecodeType type)

constexpr

True iff type is one of the YOLO-family detection or segmentation variants.

Definition at line 151 of file BoxDecodeType.h.

151constexpr bool box_decode_type_is_yolo_family(BoxDecodeType type) {

152 switch (type) {

153 case BoxDecodeType::Yolo:

154 case BoxDecodeType::YoloV5:

155 case BoxDecodeType::YoloV5Seg:

156 case BoxDecodeType::YoloV7:

157 case BoxDecodeType::YoloV7Seg:

158 case BoxDecodeType::YoloV8:

159 case BoxDecodeType::YoloV8Seg:

160 case BoxDecodeType::YoloV8Pose:

161 case BoxDecodeType::YoloV9:

162 case BoxDecodeType::YoloV9Seg:

163 case BoxDecodeType::YoloV10:

164 case BoxDecodeType::YoloV10Seg:

165 case BoxDecodeType::YoloV26:

166 return true;

167 case BoxDecodeType::Detr:

168 case BoxDecodeType::EffDet:

169 case BoxDecodeType::RcnnStage1:

170 case BoxDecodeType::Centernet:

171 case BoxDecodeType::Unspecified:

172 default:

173 return false;

174 }

175}

box_decode_type_option_token()

constexpr const char* simaai::neat::box_decode_type_option_token (BoxDecodeTypeOption option)

constexpr

Stable lower-case token for a decode-layout option.

Definition at line 123 of file BoxDecodeType.h.

123constexpr const char* box_decode_type_option_token(BoxDecodeTypeOption option) {

124 switch (option) {

125 case BoxDecodeTypeOption::Auto:

126 return "auto";

127 case BoxDecodeTypeOption::PackedPerHead:

128 return "packed-per-head";

129 case BoxDecodeTypeOption::InterleavedByHead:

130 return "interleaved-by-head";

131 case BoxDecodeTypeOption::GroupedByRole:

132 return "grouped-by-role";

133 case BoxDecodeTypeOption::Split3Interleaved:

134 return "split3-interleaved";

135 case BoxDecodeTypeOption::Split3Grouped:

136 return "split3-grouped";

137 case BoxDecodeTypeOption::InterleavedByHeadProbability:

138 return "interleaved-by-head-probability";

139 case BoxDecodeTypeOption::InterleavedByHeadLogit:

140 return "interleaved-by-head-logit";

141 case BoxDecodeTypeOption::GroupedByRoleProbability:

142 return "grouped-by-role-probability";

143 case BoxDecodeTypeOption::GroupedByRoleLogit:

144 return "grouped-by-role-logit";

145 default:

146 return "auto";

147 }

148}

box_decode_type_token()

constexpr const char* simaai::neat::box_decode_type_token (BoxDecodeType type)

constexpr

Stable lower-case token for a decode family (used in caps, manifests, logs).

Definition at line 80 of file BoxDecodeType.h.

80constexpr const char* box_decode_type_token(BoxDecodeType type) {

81 switch (type) {

82 case BoxDecodeType::Yolo:

83 return "yolo";

84 case BoxDecodeType::YoloV5:

85 return "yolov5";

86 case BoxDecodeType::YoloV5Seg:

87 return "yolov5-seg";

88 case BoxDecodeType::YoloV7:

89 return "yolov7";

90 case BoxDecodeType::YoloV7Seg:

91 return "yolov7-seg";

92 case BoxDecodeType::YoloV8:

93 return "yolov8";

94 case BoxDecodeType::YoloV8Seg:

95 return "yolov8-seg";

96 case BoxDecodeType::YoloV8Pose:

97 return "yolov8-pose";

98 case BoxDecodeType::YoloV9:

99 return "yolov9";

100 case BoxDecodeType::YoloV9Seg:

101 return "yolov9-seg";

102 case BoxDecodeType::YoloV10:

103 return "yolov10";

104 case BoxDecodeType::YoloV10Seg:

105 return "yolov10-seg";

106 case BoxDecodeType::YoloV26:

107 return "yolo26";

108 case BoxDecodeType::Detr:

109 return "detr";

110 case BoxDecodeType::EffDet:

111 return "effdet";

112 case BoxDecodeType::RcnnStage1:

113 return "rcnn-stage1";

114 case BoxDecodeType::Centernet:

115 return "centernet";

116 case BoxDecodeType::Unspecified:

117 default:

118 return "unspecified";

119 }

120}

build_graph_processcvu_prepared_stage()

bool simaai::neat::build_graph_processcvu_prepared_stage (const GraphProcessCvuStageRequest & request, simaai::gst::ProcessCvuPreparedStage * out, std::string * error_message=nullptr)

Definition at line 110 of file PreparedRuntimeBridge.h.

build_graph_processmla_prepared_stage()

bool simaai::neat::build_graph_processmla_prepared_stage (const GraphProcessMlaStageRequest & request, simaai::gst::ProcessMlaPreparedStage * out, std::string * error_message=nullptr)

Definition at line 106 of file PreparedRuntimeBridge.h.

build_prepared_stage_from_manifest_context()

bool simaai::neat::build_prepared_stage_from_manifest_context (const GstContext * static_manifest_context, const char * stage_id_or_name, const char * element_name_fallback, simaai::gst::PreparedStageSpec * out, std::string * error_message=nullptr)

Definition at line 114 of file PreparedRuntimeBridge.h.

caps_to_codec()

simaai::neat::EncodedSpec::Codec simaai::neat::caps_to_codec (const std::string & caps_string)

Map a GStreamer-style caps string to an EncodedSpec::Codec value.

Parameters

caps_string

GStreamer caps string (e.g., "video/x-h264, ...").

Returns

Matching codec enum, or the codec-unspecified sentinel if no match.

Definition at line 31 of file EncodedSampleUtil.h.

conversion_allowed()

bool simaai::neat::conversion_allowed (ConversionPolicy policy, ConversionKind kind)

Returns true if the policy allows a conversion of this kind (without performing it).

Definition at line 95 of file TensorConversion.h.

decode_bbox_tensor()

BoxDecodeResult simaai::neat::decode_bbox_tensor (const simaai::neat::Tensor & tensor, int img_w, int img_h, int expected_topk, bool strict)

Decode a BBOX-format Tensor into a BoxDecodeResult.

Convenience wrapper that maps the tensor, parses bytes via parse_bbox_bytes, and retains the raw byte view alongside the typed boxes.

Parameters

tensor	Tensor whose payload is a BBOX byte stream.
img_w	Original image width used to clamp/scale coordinates.
img_h	Original image height used to clamp/scale coordinates.
expected_topk	Number of detections expected (advisory; see strict).
strict	If true, mismatches against expected_topk throw.

Returns

Parsed detections plus retained raw bytes.

Definition at line 83 of file DetectionTypes.h.

default_modalix_dvt_power_rails()

std::vector<PowerRailConfig> simaai::neat::default_modalix_dvt_power_rails ()

Default Modalix DVT PMIC rail table.

DVT Linux PMBus access is intentionally limited to the verified stable reading: bus 4, PMIC 0x4d, page 0, POUT register 0x96, exponent -2.

Definition at line 146 of file PowerTelemetry.h.

default_modalix_som_power_rails()

std::vector<PowerRailConfig> simaai::neat::default_modalix_som_power_rails ()

Default Modalix SOM PMIC rail table from the board measurement scripts.

Definition at line 137 of file PowerTelemetry.h.

detect_default_power_monitor_profile()

PowerMonitorProfile simaai::neat::detect_default_power_monitor_profile ()

Detect the most likely built-in board power profile for the current host.

Definition at line 152 of file PowerTelemetry.h.

drain_bus()

void simaai::neat::drain_bus (GstElement * pipeline, BusMessageFn on_message, void * user_data)

Drain pending messages from the pipeline's bus, invoking on_message for each.

Non-blocking: pulls all messages currently posted, then returns. Safe to call from the framework's polling loop while the pipeline is running.

Parameters

pipeline	Pipeline whose bus to drain.
on_message	Callback invoked for each message (may be null).
user_data	Opaque pointer forwarded to on_message.

Definition at line 55 of file GstBusWatch.h.

element_exists()

bool simaai::neat::element_exists (const char * factory)

Returns true if a GStreamer element factory with factory name is registered.

Definition at line 20 of file GstHelpers.h.

element_property_exists()

bool simaai::neat::element_property_exists (const char * factory, const char * property_name)

Returns true if factory exists and exposes a property named property_name.

Definition at line 23 of file GstHelpers.h.

estimate_conversion_cost()

ConversionCost simaai::neat::estimate_conversion_cost (ConversionKind kind, std::uint64_t bytes_copied)

Estimate the cost (bytes + compute class) of a conversion of the given kind moving bytes_copied bytes.

Definition at line 93 of file TensorConversion.h.

factory_plugin_path()

std::string simaai::neat::factory_plugin_path (const char * factory)

Returns the filesystem path of the plugin .so providing factory, or empty if unknown.

Definition at line 26 of file GstHelpers.h.

format_power_summary()

std::string simaai::neat::format_power_summary (const PowerSummary & summary)

Format a compact human-readable power summary.

Definition at line 196 of file PowerTelemetry.h.

format_runtime_metrics()

std::string simaai::neat::format_runtime_metrics (const RuntimeMetrics & metrics, RuntimeMetricsFormat format=RuntimeMetricsFormat::Text)

Render runtime metrics in the requested format.

Definition at line 103 of file RuntimeMetrics.h.

format_tag_from_string()

FormatTag simaai::neat::format_tag_from_string (const std::string & value)

inline

Parse a string token to a FormatTag; unknown values map to Auto.

Definition at line 199 of file FormatSpec.h.

199inline FormatTag format_tag_from_string(const std::string& value) {

200 const std::string up = upper_copy_ascii(value);

201 if (up.empty())

202 return FormatTag::Auto;

203 if (up == "RGB")

204 return FormatTag::RGB;

205 if (up == "BGR")

206 return FormatTag::BGR;

207 if (up == "GRAY" || up == "GRAY8")

208 return FormatTag::GRAY8;

209 if (up == "NV12")

210 return FormatTag::NV12;

211 if (up == "I420" || up == "YUV420")

212 return FormatTag::I420;

213 if (up == "YUYV")

214 return FormatTag::YUYV;

215 if (up == "ENCODED")

216 return FormatTag::ENCODED;

217 if (up == "H264")

218 return FormatTag::H264;

219 if (up == "BYTESTREAM" || up == "BYTE_STREAM" || up == "BYTE-STREAM" || up == "RAW_BYTES" ||

220 up == "RAW-BYTES" || up == "OPAQUE_BYTES" || up == "OPAQUE-BYTES" || up == "OCTET_STREAM" ||

221 up == "OCTET-STREAM")

222 return FormatTag::ByteStream;

223 if (up == "MLA")

224 return FormatTag::MLA;

225 if (up == "BBOX")

226 return FormatTag::BBOX;

227 if (up == "ARGMAX")

228 return FormatTag::ARGMAX;

229 if (up == "DETESSDEQUANT")

230 return FormatTag::DETESSDEQUANT;

231 if (up == "FP32")

232 return FormatTag::FP32;

233 if (up == "INT8")

234 return FormatTag::INT8;

235 if (up == "UINT8")

236 return FormatTag::UINT8;

237 if (up == "BF16" || up == "BFLOAT16")

238 return FormatTag::BF16;

239 if (up == "EVXX_BF16")

240 return FormatTag::EVXX_BFLOAT16;

241 if (up == "EVXX_FLOAT32")

242 return FormatTag::EVXX_FLOAT32;

243 if (up == "EVXX_INT8")

244 return FormatTag::EVXX_INT8;

245 if (up == "EVXX_BFLOAT16")

246 return FormatTag::EVXX_BFLOAT16;

247 return FormatTag::Auto;

248}

format_tag_name()

const char* simaai::neat::format_tag_name (FormatTag tag)

inline

Stable, canonical string token for tag (empty string for Auto).

Definition at line 61 of file FormatSpec.h.

61inline const char* format_tag_name(FormatTag tag) {

62 switch (tag) {

63 case FormatTag::Auto:

64 return "";

65 case FormatTag::RGB:

66 return "RGB";

67 case FormatTag::BGR:

68 return "BGR";

69 case FormatTag::GRAY8:

70 return "GRAY8";

71 case FormatTag::NV12:

72 return "NV12";

73 case FormatTag::I420:

74 return "I420";

75 case FormatTag::YUYV:

76 return "YUYV";

77 case FormatTag::ENCODED:

78 return "ENCODED";

79 case FormatTag::H264:

80 return "H264";

81 case FormatTag::ByteStream:

82 return "BYTESTREAM";

83 case FormatTag::MLA:

84 return "MLA";

85 case FormatTag::BBOX:

86 return "BBOX";

87 case FormatTag::ARGMAX:

88 return "ARGMAX";

89 case FormatTag::DETESSDEQUANT:

90 return "DETESSDEQUANT";

91 case FormatTag::FP32:

92 return "FP32";

93 case FormatTag::INT8:

94 return "INT8";

95 case FormatTag::UINT8:

96 return "UINT8";

97 case FormatTag::BF16:

98 return "BF16";

99 case FormatTag::EVXX_FLOAT32:

100 return "EVXX_FLOAT32";

101 case FormatTag::EVXX_INT8:

102 return "EVXX_INT8";

103 case FormatTag::EVXX_BFLOAT16:

104 return "EVXX_BFLOAT16";

105 }

106 return "";

107}

format_tag_to_string()

std::string simaai::neat::format_tag_to_string (FormatTag tag)

inline

std::string form of the canonical format token.

Definition at line 110 of file FormatSpec.h.

110inline std::string format_tag_to_string(FormatTag tag) {

111 return std::string(format_tag_name(tag));

112}

from_gst_sample()

Tensor simaai::neat::from_gst_sample (GstSample * sample)

Definition at line 18 of file TensorAdapters.h.

gst_init_once()

void simaai::neat::gst_init_once ()

Initialize GStreamer exactly once across the process.

Safe to call from multiple threads and multiple times — only the first call performs the actual gst_init(); subsequent calls return immediately.

Definition at line 22 of file GstInit.h.

gst_message_to_string()

std::string simaai::neat::gst_message_to_string (GstMessage * msg)

Format a single GStreamer bus message into a stable, one-line string.

Parameters

msg	The bus message to format. Must be non-null.

Returns

Human-readable summary suitable for logs and SessionReport.

Definition at line 43 of file GstBusWatch.h.

is_raw_video_format()

bool simaai::neat::is_raw_video_format (FormatTag tag)

inline

True iff tag names a raw (uncompressed) video format.

Definition at line 251 of file FormatSpec.h.

251inline bool is_raw_video_format(FormatTag tag) {

252 return tag == FormatTag::RGB || tag == FormatTag::BGR || tag == FormatTag::GRAY8 ||

253 tag == FormatTag::NV12 || tag == FormatTag::I420 || tag == FormatTag::YUYV;

254}

is_tensor_payload_format()

bool simaai::neat::is_tensor_payload_format (FormatTag tag)

inline

True iff tag names a tensor payload (MLA, BBOX, ARGMAX, dtype kinds, EVXX aliases).

Definition at line 257 of file FormatSpec.h.

257inline bool is_tensor_payload_format(FormatTag tag) {

258 return tag == FormatTag::ByteStream || tag == FormatTag::MLA || tag == FormatTag::BBOX ||

259 tag == FormatTag::ARGMAX || tag == FormatTag::DETESSDEQUANT || tag == FormatTag::FP32 ||

260 tag == FormatTag::INT8 || tag == FormatTag::UINT8 || tag == FormatTag::BF16 ||

261 tag == FormatTag::EVXX_FLOAT32 || tag == FormatTag::EVXX_INT8 ||

262 tag == FormatTag::EVXX_BFLOAT16;

263}

is_tessellated_bf16_format()

bool simaai::neat::is_tessellated_bf16_format (const std::string & fmt)

True iff fmt names a BF16 tessellated tensor format string.

Definition at line 20 of file TessellatedTensor.h.

is_tessellated_int8_format()

bool simaai::neat::is_tessellated_int8_format (const std::string & fmt)

True iff fmt names an INT8 tessellated tensor format string.

Definition at line 18 of file TessellatedTensor.h.

make_bundle_sample()

Sample simaai::neat::make_bundle_sample (std::initializer_list< Sample > fields)

inline

Construct a Bundle-kind Sample whose payload is the given list of inner Samples.

Definition at line 484 of file SessionOptions.h.

484inline Sample make_bundle_sample(std::initializer_list<Sample> fields) {

485 Sample out;

486 out.kind = SampleKind::Bundle;

487 out.fields = fields;

488 return out;

489}

make_cpu_external_storage()

std::shared_ptr<TensorBuffer> simaai::neat::make_cpu_external_storage (void * data, std::size_t size_bytes, std::shared_ptr< void > holder={}, bool read_only=true)

Wrap a foreign CPU pointer as a TensorBuffer without copying.

Lifetime of the underlying memory is tracked via the holder shared_ptr — pass a custom deleter or a refcounted handle that keeps the original allocation alive. Set read_only to indicate the framework should not write through this storage.

Definition at line 472 of file TensorCore.h.

make_cpu_owned_storage()

std::shared_ptr<TensorBuffer> simaai::neat::make_cpu_owned_storage (std::size_t size_bytes)

Allocate a CPU-owned heap buffer of size_bytes, initialized to zero.

Definition at line 464 of file TensorCore.h.

make_encoded_sample()

Sample simaai::neat::make_encoded_sample (std::vector< uint8_t > bytes, std::string caps_string, int64_t pts_ns=-1, int64_t dts_ns=-1, int64_t duration_ns=-1)

Build an encoded Sample from raw bytes and caps metadata.

The returned Sample owns bytes (moved in) and carries the supplied caps string plus optional PTS/DTS/duration timestamps. Pass -1 to leave a timestamp unset.

Parameters

bytes	Encoded payload (e.g., H.264 access unit or NAL).
caps_string	Caps string describing the codec/profile.
pts_ns	Presentation timestamp in nanoseconds, or -1 if unknown.
dts_ns	Decode timestamp in nanoseconds, or -1 if unknown.
duration_ns	Frame duration in nanoseconds, or -1 if unknown.

Returns

A Sample wrapping the encoded payload with the supplied caps.

Definition at line 47 of file EncodedSampleUtil.h.

make_tensor_sample()

Sample simaai::neat::make_tensor_sample (const std::string & port_name, simaai::neat::Tensor tensor)

inline

Construct a TensorSet-kind Sample wrapping a single Tensor for the named port.

Definition at line 466 of file SessionOptions.h.

466inline Sample make_tensor_sample(const std::string& port_name, simaai::neat::Tensor tensor) {

467 Sample out;

468 out.kind = SampleKind::TensorSet;

469 out.stream_label = port_name;

470 out.tensors = TensorList{std::move(tensor)};

471 return out;

472}

modalix_dvt_power_monitor_options()

PowerMonitorOptions simaai::neat::modalix_dvt_power_monitor_options (int sample_interval_ms=100)

Convenience options enabling Modalix DVT power monitoring.

Definition at line 184 of file PowerTelemetry.h.

modalix_som_power_monitor_options()

PowerMonitorOptions simaai::neat::modalix_som_power_monitor_options (int sample_interval_ms=100)

Convenience options enabling Modalix SOM power monitoring.

Definition at line 178 of file PowerTelemetry.h.

normalize_caps_format_for_media()

std::string simaai::neat::normalize_caps_format_for_media (std::string media_type, std::string format)

inline

Normalize a caps-format string conditionally on its media type.

Tensor caps (application/vnd.simaai.tensor) are normalized to canonical EVXX_ tokens; other media types (raw video, encoded video) are returned unchanged.

Parameters

media_type	GStreamer-style media type (e.g., application/vnd.simaai.tensor).
format	Format string to potentially normalize.

Returns

Normalized format if applicable, otherwise format unchanged.

Definition at line 187 of file FormatSpec.h.

187inline std::string normalize_caps_format_for_media(std::string media_type, std::string format) {

188 if (format.empty()) {

189 return format;

190 }

191 const std::string media_up = upper_copy_ascii(std::move(media_type));

192 if (media_up == "APPLICATION/VND.SIMAAI.TENSOR") {

193 return normalize_tensor_caps_format(std::move(format));

194 }

195 return format;

196}

normalize_tensor_caps_format()

std::string simaai::neat::normalize_tensor_caps_format (std::string value)

inline

Normalize tensor format aliases to the canonical EVXX_* token.

Maps user-facing or legacy spellings (FP32/FLOAT32, BF16/BFLOAT16, INT8/16/32) onto the EV-side caps form. Unknown values are passed through unchanged.

Definition at line 157 of file FormatSpec.h.

157inline std::string normalize_tensor_caps_format(std::string value) {

158 if (tensor_format_is_fp32_alias(value)) {

159 return "EVXX_FLOAT32";

160 }

161 if (tensor_format_is_bf16_alias(value)) {

162 return "EVXX_BFLOAT16";

163 }

164 if (tensor_format_is_int8_alias(value)) {

165 return "EVXX_INT8";

166 }

167 if (tensor_format_is_int16_alias(value)) {

168 return "EVXX_INT16";

169 }

170 if (tensor_format_is_int32_alias(value)) {

171 return "EVXX_INT32";

172 }

173 return value;

174}

parse_bbox_bytes()

std::vector<Box> simaai::neat::parse_bbox_bytes (const std::vector< uint8_t > & bytes, int img_w, int img_h, int expected_topk, bool strict)

Parse a packed BBOX byte payload into typed Box records.

Parameters

bytes	Raw bytes (the BBOX caps payload).
img_w	Original image width used to clamp/scale coordinates.
img_h	Original image height used to clamp/scale coordinates.
expected_topk	Number of detections expected (advisory; see strict).
strict	If true, mismatches against expected_topk throw; if false, parse best-effort.

Returns

Parsed detections (size may be less than expected_topk).

Definition at line 67 of file DetectionTypes.h.

power_monitor_profile_name()

std::string simaai::neat::power_monitor_profile_name (PowerMonitorProfile profile)

Human-readable name for a power monitor profile.

Definition at line 158 of file PowerTelemetry.h.

power_rails_for_profile()

std::vector<PowerRailConfig> simaai::neat::power_rails_for_profile (PowerMonitorProfile profile)

Resolve a built-in board profile to its rail table.

Definition at line 164 of file PowerTelemetry.h.

power_summary_to_json()

std::string simaai::neat::power_summary_to_json (const PowerSummary & summary, int indent=2)

Format power summary as a JSON object.

Definition at line 202 of file PowerTelemetry.h.

prepare_processmla_runtime_config()

bool simaai::neat::prepare_processmla_runtime_config (ProcessMlaRuntimeConfig * runtime_cfg, std::string * error_message=nullptr)

Definition at line 103 of file PreparedRuntimeBridge.h.

read_power_snapshot()

PowerSnapshot simaai::neat::read_power_snapshot (const PowerMonitorOptions & options)

Read one immediate power snapshot using native Linux I2C access.

Definition at line 190 of file PowerTelemetry.h.

require_element()

void simaai::neat::require_element (const char * factory, const char * context)

Throw a SessionError if factory is not registered.

Parameters

factory	GStreamer element factory name (e.g., "h264parse").
context	Caller context interpolated into the error message (typically the Node kind).

Definition at line 33 of file GstHelpers.h.

require_single_tensor()

Tensor& simaai::neat::require_single_tensor (Sample & sample, const char * where=nullptr)

Returns the sample's single Tensor; throws if the sample carries 0 or >1 tensors.

Definition at line 507 of file SessionOptions.h.

require_single_tensor()

const Tensor& simaai::neat::require_single_tensor (const Sample & sample, const char * where=nullptr)

Const variant of require_single_tensor.

Definition at line 509 of file SessionOptions.h.

require_tensordecoder()

void simaai::neat::require_tensordecoder (const char * context)

Throw a SessionError if SiMa's tensordecoder element is not available.

Parameters

context

Caller context for the error message.

Definition at line 39 of file GstHelpers.h.

runtime_metrics_to_compact_text()

std::string simaai::neat::runtime_metrics_to_compact_text (const RuntimeMetrics & metrics)

Render runtime metrics as one compact line.

Definition at line 122 of file RuntimeMetrics.h.

runtime_metrics_to_json()

std::string simaai::neat::runtime_metrics_to_json (const RuntimeMetrics & metrics, int indent=0)

Render runtime metrics as JSON.

Definition at line 110 of file RuntimeMetrics.h.

runtime_metrics_to_text()

std::string simaai::neat::runtime_metrics_to_text (const RuntimeMetrics & metrics)

Render runtime metrics as human-readable text.

Definition at line 116 of file RuntimeMetrics.h.

sample_from_tensors()

Sample simaai::neat::sample_from_tensors (const TensorList & tensors)

Construct a TensorSet sample from a list of tensors.

Definition at line 513 of file SessionOptions.h.

sample_has_tensor_list()

bool simaai::neat::sample_has_tensor_list (const Sample & sample)

inline

Returns true if the sample is a TensorSet with at least one tensor.

Definition at line 498 of file SessionOptions.h.

498inline bool sample_has_tensor_list(const Sample& sample) {

499 return sample.kind == SampleKind::TensorSet && !sample.tensors.empty();

500}

sample_is_multi_output()

bool simaai::neat::sample_is_multi_output (const Sample & sample)

inline

Returns true if the sample carries multiple outputs (Bundle, or TensorSet with size > 1).

Definition at line 492 of file SessionOptions.h.

492inline bool sample_is_multi_output(const Sample& sample) {

493 return sample.kind == SampleKind::Bundle ||

494 (sample.kind == SampleKind::TensorSet && sample.tensors.size() > 1U);

495}

sample_tensor_list()

TensorList& simaai::neat::sample_tensor_list (Sample & sample, const char * where=nullptr)

Mutable reference to the sample's tensor list. Throws SessionError if kind != TensorSet.

Definition at line 503 of file SessionOptions.h.

sample_tensor_list()

const TensorList& simaai::neat::sample_tensor_list (const Sample & sample, const char * where=nullptr)

Const reference to the sample's tensor list. Throws SessionError if kind != TensorSet.

Definition at line 505 of file SessionOptions.h.

tensor_format_is_bf16_alias()

bool simaai::neat::tensor_format_is_bf16_alias (std::string value)

inline

True iff value names BF16 (any of BF16, BFLOAT16, EVXX_BF16, EVXX_BFLOAT16).

Definition at line 122 of file FormatSpec.h.

122inline bool tensor_format_is_bf16_alias(std::string value) {

123 const std::string up = upper_copy_ascii(std::move(value));

124 return up == "BF16" || up == "BFLOAT16" || up == "EVXX_BF16" || up == "EVXX_BFLOAT16";

125}

tensor_format_is_fp32_alias()

bool simaai::neat::tensor_format_is_fp32_alias (std::string value)

inline

True iff value names FP32 (FP32, FLOAT32, EVXX_FLOAT32).

Definition at line 128 of file FormatSpec.h.

128inline bool tensor_format_is_fp32_alias(std::string value) {

129 const std::string up = upper_copy_ascii(std::move(value));

130 return up == "FP32" || up == "FLOAT32" || up == "EVXX_FLOAT32";

131}

tensor_format_is_int16_alias()

bool simaai::neat::tensor_format_is_int16_alias (std::string value)

inline

True iff value names INT16 (INT16, EVXX_INT16).

Definition at line 140 of file FormatSpec.h.

140inline bool tensor_format_is_int16_alias(std::string value) {

141 const std::string up = upper_copy_ascii(std::move(value));

142 return up == "INT16" || up == "EVXX_INT16";

143}

tensor_format_is_int32_alias()

bool simaai::neat::tensor_format_is_int32_alias (std::string value)

inline

True iff value names INT32 (INT32, EVXX_INT32).

Definition at line 146 of file FormatSpec.h.

146inline bool tensor_format_is_int32_alias(std::string value) {

147 const std::string up = upper_copy_ascii(std::move(value));

148 return up == "INT32" || up == "EVXX_INT32";

149}

tensor_format_is_int8_alias()

bool simaai::neat::tensor_format_is_int8_alias (std::string value)

inline

True iff value names INT8 (INT8, EVXX_INT8).

Definition at line 134 of file FormatSpec.h.

134inline bool tensor_format_is_int8_alias(std::string value) {

135 const std::string up = upper_copy_ascii(std::move(value));

136 return up == "INT8" || up == "EVXX_INT8";

137}

tensor_to_pcm_audio()

PcmAudio simaai::neat::tensor_to_pcm_audio (const Tensor & tensor)

Definition at line 15 of file TensorAudio.h.

tensors_from_sample()

TensorList simaai::neat::tensors_from_sample (const Sample & sample, bool require_nonempty=true)

Extract all tensors from any sample shape (Tensor, TensorSet, Bundle); flattens recursively.

Definition at line 511 of file SessionOptions.h.

throw_if_bus_error()

void simaai::neat::throw_if_bus_error (GstElement * pipeline, BusMessageFn on_message, void * user_data, BusErrorFn on_error, void * error_user_data)

Drain the bus and throw a SessionError if any error message is found.

Drains exactly like drain_bus but additionally invokes on_error and raises a SessionError if any GST_MESSAGE_ERROR is observed.

Parameters

pipeline	Pipeline whose bus to drain.
on_message	Callback invoked for every message (may be null).
user_data	Opaque pointer forwarded to on_message.
on_error	Callback invoked once for the error line before throwing (may be null).
error_user_data	Opaque pointer forwarded to on_error.

Definition at line 69 of file GstBusWatch.h.

to_string()

const char* simaai::neat::to_string (ValidationSeverity s)

inline

Stable string label for a ValidationSeverity (for reports/logs).

Definition at line 36 of file ValidationReport.h.

36inline const char* to_string(ValidationSeverity s) {

37 switch (s) {

38 case ValidationSeverity::Info:

39 return "INFO";

40 case ValidationSeverity::Warning:

41 return "WARNING";

42 case ValidationSeverity::Error:

43 return "ERROR";

44 }

45 return "UNKNOWN";

46}

upper_copy_ascii()

std::string simaai::neat::upper_copy_ascii (std::string value)

inline

ASCII upper-case copy of value (does not touch non-ASCII bytes).

Definition at line 115 of file FormatSpec.h.

115inline std::string upper_copy_ascii(std::string value) {

116 std::transform(value.begin(), value.end(), value.begin(),

117 [](unsigned char c) { return static_cast<char>(std::toupper(c)); });

118 return value;

119}

---

The documentation for this namespace was generated from the following files:

• ContractTypes.h
• NodeContractDefinition.h
• ValidationReport.h
• GstBusWatch.h
• GstHelpers.h
• GstInit.h
• Model.h
• PreprocessPlan.h
• PreparedRuntimeBridge.h
• Input.h
• Cast.h
• BoxDecodeType.h
• DetectionTypes.h
• EncodedSampleUtil.h
• FormatSpec.h
• PowerTelemetry.h
• Run.h
• RuntimeMetrics.h
• SessionOptions.h
• Tensor.h
• TensorAdapters.h
• TensorAudio.h
• TensorConversion.h
• TensorCore.h
• TensorTypes.h
• TessellatedTensor.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.1.