RunOptions Struct

Per-Run runtime options. More...

Declaration

struct simaai::neat::RunOptions { ... }

Included Headers

#include <Run.h>

Public Member Functions Index

RunOptions &	enable_board_power (int sample_interval_ms=100)
	Enable board power monitoring using built-in auto-detect. More...
RunOptions &	enable_modalix_som_power (int sample_interval_ms=100)
	Enable default Modalix SOM PMIC rail power monitoring for this Run. More...
RunOptions &	enable_modalix_dvt_power (int sample_interval_ms=100)
	Enable the built-in Modalix DVT PMIC table for this Run. More...
RunOptions &	disable_power_monitor ()
	Disable power monitoring after it was enabled/configured. More...

Public Member Attributes Index

RunPreset	preset = RunPreset::Balanced
	Convenience preset that tunes the other fields. More...
int	queue_depth = 4
	Capacity of ingress and internal runtime queues. More...
OverflowPolicy	overflow_policy = OverflowPolicy::Block
	Overflow policy for ingress and general runtime queues. More...
OutputMemory	output_memory = ...
	Whether output tensors are zero-copy or owned. More...
int	input_timeout_ms = -1
	Default pull timeout for build()/run() input-mode paths, in milliseconds. More...
bool	startup_preflight = true
	Validate seeded build(input, ...) calls by pushing/pulling the seed once. More...
RunAdvancedOptions	advanced {}
	Advanced tuning (rarely needed). More...
PowerMonitorOptions	power_monitor {}
	Optional board rail power telemetry. More...
RunAutoExportOptions	run_export {}
	Optional build-time Run/graph JSON export. More...
RunAutoExportOptions	graph_run_export {}
	Backward-compatible spelling for the previous graph-run export field. More...
std::function< void(const struct InputDropInfo &)>	on_input_drop
	Optional callback invoked when an input frame is dropped by the overflow policy. More...

Description

Per-Run runtime options.

Passed to Graph::build() and Model::run().

Most fields default to sensible values. The most commonly customized are queue_depth (deeper for jittery sources, shallower for low latency) and overflow_policy (driven by the input source's behavior — see the OverflowPolicy enum docs).

Definition at line 164 of file Run.h.

Public Member Functions

disable_power_monitor()

RunOptions & simaai::neat::RunOptions::disable_power_monitor ()

inline

Disable power monitoring after it was enabled/configured.

Definition at line 250 of file Run.h.

enable_board_power()

RunOptions & simaai::neat::RunOptions::enable_board_power (int sample_interval_ms=100)

inline

Enable board power monitoring using built-in auto-detect.

This is the preferred no-env path for capturing latency, throughput, and power together from any Run.

Definition at line 215 of file Run.h.

enable_modalix_dvt_power()

RunOptions & simaai::neat::RunOptions::enable_modalix_dvt_power (int sample_interval_ms=100)

inline

Enable the built-in Modalix DVT PMIC table for this Run.

Definition at line 242 of file Run.h.

enable_modalix_som_power()

RunOptions & simaai::neat::RunOptions::enable_modalix_som_power (int sample_interval_ms=100)

inline

Enable default Modalix SOM PMIC rail power monitoring for this Run.

Prefer enable_board_power() unless you need to force the SOM profile:

 RunOptions opt;

 opt.enable_board_power();

 auto run = graph.build(inputs, opt);

 auto scope = run.start_measurement();

 // push/pull workload

 auto report = scope.stop();

Definition at line 234 of file Run.h.

Public Member Attributes

advanced

RunAdvancedOptions simaai::neat::RunOptions::advanced {}

Advanced tuning (rarely needed).

Definition at line 203 of file Run.h.

graph_run_export

RunAutoExportOptions simaai::neat::RunOptions::graph_run_export {}

Backward-compatible spelling for the previous graph-run export field.

Definition at line 207 of file Run.h.

input_timeout_ms

int simaai::neat::RunOptions::input_timeout_ms = -1

Default pull timeout for build()/run() input-mode paths, in milliseconds.

-1 keeps framework defaults (and legacy env-var fallback). Override per-call by passing an explicit timeout_ms to the relevant pull() or run() overload.

Definition at line 188 of file Run.h.

on_input_drop

std::function<void(const struct InputDropInfo&)> simaai::neat::RunOptions::on_input_drop

Optional callback invoked when an input frame is dropped by the overflow policy.

Useful for telemetry — log the drop reason, update a counter, etc. Callback runs on the pushing thread; keep it short.

Definition at line 261 of file Run.h.

output_memory

OutputMemory simaai::neat::RunOptions::output_memory

Whether output tensors are zero-copy or owned.

Initialiser

= OutputMemory::Auto

Definition at line 180 of file Run.h.

overflow_policy

OverflowPolicy simaai::neat::RunOptions::overflow_policy = OverflowPolicy::Block

Overflow policy for ingress and general runtime queues.

A public Output node uses its OutputOptions::drop policy instead.

Definition at line 179 of file Run.h.

power_monitor

PowerMonitorOptions simaai::neat::RunOptions::power_monitor {}

Optional board rail power telemetry.

Definition at line 204 of file Run.h.

preset

RunPreset simaai::neat::RunOptions::preset = RunPreset::Balanced

Convenience preset that tunes the other fields.

Definition at line 165 of file Run.h.

queue_depth

int simaai::neat::RunOptions::queue_depth = 4

Capacity of ingress and internal runtime queues.

Framework-created output queues also use this default. A public Output node owns its terminal queue contract through OutputOptions::max_buffers, which takes precedence over this value.

Definition at line 173 of file Run.h.

run_export

RunAutoExportOptions simaai::neat::RunOptions::run_export {}

Optional build-time Run/graph JSON export.

Definition at line 205 of file Run.h.

startup_preflight

bool simaai::neat::RunOptions::startup_preflight = true

Validate seeded build(input, ...) calls by pushing/pulling the seed once.

This applies only to build overloads that receive an initial cv::Mat, TensorList, or Sample. It catches payload-level failures (for example invalid encoded bitstreams or caps that only the GStreamer element can reject) at build time instead of letting the first failure surface asynchronously after a Run has already been returned. Seedless/source builds ignore this option.

Set to false for latency-sensitive callers that only want the seed for caps/shape adaptation and are prepared to observe first-sample errors through Run::last_error() or pull().

Definition at line 202 of file Run.h.

---

The documentation for this struct was generated from the following file:

• Run.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.