RunAdvancedOptions Struct
Advanced runtime tuning knobs (most users never set these). More...
Declaration
Included Headers
Public Member Attributes Index
| bool | copy_input = ... |
|
Force a copy of every pushed input (useful when the source buffer is short-lived). More... | |
| std::size_t | max_input_bytes = 0 |
|
Reject pushes larger than this many bytes (0 = no cap). More... | |
| int | sync_num_buffers_override = ... |
|
Override the appsrc num-buffers for sync-mode runs (-1 = auto). More... | |
| bool | prepare_output_cpu_visible = false |
|
Prepare zero-copy Gst/SiMa outputs for CPU reads on the InputStream worker. More... | |
Description
Advanced runtime tuning knobs (most users never set these).
Definition at line 95 of file Run.h.
Public Member Attributes
copy_input
|
Force a copy of every pushed input (useful when the source buffer is short-lived).
- Initialiser
-
= false
Definition at line 96 of file Run.h.
max_input_bytes
|
Reject pushes larger than this many bytes (0 = no cap).
Definition at line 98 of file Run.h.
prepare_output_cpu_visible
|
Prepare zero-copy Gst/SiMa outputs for CPU reads on the InputStream worker.
When enabled for a zero-copy output Run, cached SiMa-backed tensors are made CPU-visible before they are placed in the Run output queue. This does not clone or copy tensor payloads; it only performs cache visibility maintenance when producer metadata says a CPU reader could observe stale data.
Keep this disabled for Graph-internal forwarding paths where an appsink is used only as transport to another device-facing Session. Tensor::map(Read) remains the correctness fallback for all outputs.
Definition at line 113 of file Run.h.
sync_num_buffers_override
|
Override the appsrc num-buffers for sync-mode runs (-1 = auto).
- Initialiser
-
= -1
Definition at line 99 of file Run.h.
The documentation for this struct was generated from the following file:
Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.