mirror of
https://github.com/firestar5683/StarPilot.git
synced 2026-07-25 03:04:14 +08:00
firestar5683
b945d4c021
Integrate V-ASM from PR #75 with Galaxy-only configuration, stale-state safety, conditional SLV coexistence, and OpenCV inference. Originally contributed by @prabhaavp in #75. Co-authored-by: Prabhaav Pillai <143428353+prabhaavp@users.noreply.github.com>
50 lines
1.6 KiB
Python
50 lines
1.6 KiB
Python
from __future__ import annotations
|
|
|
|
import math
|
|
import time
|
|
|
|
|
|
DEVICE_BUSY_MAX_CPU_USAGE_PERCENT = 89.0
|
|
DEVICE_BUSY_AVG_CPU_USAGE_PERCENT = 74.0
|
|
DEVICE_BUSY_HOT_CORE_COUNT = 4
|
|
|
|
_throttle_state: dict[str, dict[str, float]] = {}
|
|
|
|
|
|
def device_cpu_throttle_factor(cpu_usage, name="vision"):
|
|
"""Return a process-local, low-pass-filtered CPU throttle factor."""
|
|
usage = list(cpu_usage)
|
|
if not usage:
|
|
return 1.0
|
|
|
|
now = time.monotonic()
|
|
state = _throttle_state.setdefault(name, {
|
|
"factor": 1.0,
|
|
"last_time": now,
|
|
"last_logged_factor": 1.0,
|
|
})
|
|
dt = max(0.0, now - state["last_time"])
|
|
state["last_time"] = now
|
|
|
|
average = sum(usage) / len(usage)
|
|
hot_cores = sum(core >= DEVICE_BUSY_MAX_CPU_USAGE_PERCENT for core in usage)
|
|
target_factor = _compute_throttle_factor(average, hot_cores)
|
|
|
|
alpha = min(1.0 - math.exp(-0.8 * dt), 1.0)
|
|
state["factor"] = min(max(target_factor * alpha + state["factor"] * (1.0 - alpha), 1.0), 4.0)
|
|
|
|
if state["factor"] >= state["last_logged_factor"] + 0.6:
|
|
print(f"[{name}] CPU throttle factor={state['factor']:.1f}x (avg={average:.0f}%, hot={hot_cores})")
|
|
state["last_logged_factor"] = state["factor"]
|
|
elif state["factor"] <= 1.05 and state["last_logged_factor"] > 1.05:
|
|
print(f"[{name}] CPU recovered")
|
|
state["last_logged_factor"] = 1.0
|
|
|
|
return state["factor"]
|
|
|
|
|
|
def _compute_throttle_factor(average, hot_cores):
|
|
average_factor = 1.0 if average < DEVICE_BUSY_AVG_CPU_USAGE_PERCENT else 1.0 + (average - DEVICE_BUSY_AVG_CPU_USAGE_PERCENT) / 8.0
|
|
hot_factor = 1.0 + max(0, hot_cores - DEVICE_BUSY_HOT_CORE_COUNT + 1) * 0.5
|
|
return min(max(average_factor, hot_factor), 4.0)
|