RunOptions Struct
Per-Run runtime options. More...
Declaration
Included Headers
Public Member Functions Index
| RunOptions & | enable_board_power (int sample_interval_ms=100) |
|
Enable board power monitoring, read from Sentinel. More... | |
| RunOptions & | enable_modalix_som_power (int sample_interval_ms=100) |
|
Same as enable_board_power(). More... | |
| RunOptions & | enable_modalix_dvt_power (int sample_interval_ms=100) |
|
Same as enable_board_power(). 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 |
|
Output-wait timeout for input-mode Graph::run() and supported seeded build preflight. 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 power telemetry from Sentinel. 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 154 of file Run.h.
Public Member Functions
disable_power_monitor()
| inline |
Disable power monitoring after it was enabled/configured.
- Deprecated
-
Use TelemetryMonitor.
Definition at line 233 of file Run.h.
enable_board_power()
| inline |
Enable board power monitoring, read from Sentinel.
sample_interval_ms is deprecated and ignored.
- Deprecated
-
Use TelemetryMonitor around the measured work; it also records temperatures, CPU usage, and memory.
Definition at line 210 of file Run.h.
enable_modalix_dvt_power()
| inline |
Same as enable_board_power().
Definition at line 224 of file Run.h.
enable_modalix_som_power()
| inline |
Same as enable_board_power().
Definition at line 217 of file Run.h.
Public Member Attributes
advanced
|
Advanced tuning (rarely needed).
Definition at line 194 of file Run.h.
graph_run_export
|
Backward-compatible spelling for the previous graph-run export field.
Definition at line 200 of file Run.h.
input_timeout_ms
|
Output-wait timeout for input-mode Graph::run() and supported seeded build preflight.
In milliseconds; -1 selects framework defaults (and legacy env-var fallback). Run::run() and Model::run() use their per-call timeout_ms instead; -1 means unlimited, as in Run::pull(-1).
Definition at line 179 of file Run.h.
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 244 of file Run.h.
output_memory
|
overflow_policy
|
Overflow policy for ingress and general runtime queues.
A public Output node uses its OutputOptions::drop policy instead.
Definition at line 169 of file Run.h.
power_monitor
|
Optional board power telemetry from Sentinel.
- Deprecated
-
Use TelemetryMonitor around the measured work.
Definition at line 197 of file Run.h.
preset
|
Convenience preset that tunes the other fields.
Definition at line 155 of file Run.h.
queue_depth
|
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 163 of file Run.h.
run_export
|
Optional build-time Run/graph JSON export.
Definition at line 198 of file Run.h.
startup_preflight
|
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 193 of file Run.h.
The documentation for this struct was generated from the following file:
Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.15.0.