Add Auto CPU Scaling Option

[nchapman]

Feature Request: Auto CPU Scaling

Overview

Add a 4th CPU speed option "Auto" that dynamically scales between existing power levels based on CPU load.

Implementation Strategy

Use existing CPU speed levels (no platform-specific frequency tables needed):

• CPU_SPEED_POWERSAVE - Base level when idle
• CPU_SPEED_NORMAL - Standard gaming performance
• CPU_SPEED_PERFORMANCE - Emergency boost when needed

Simple 3-tier auto-scaling algorithm:

if (cpu_usage > 90%) {
    PWR_setCPUSpeed(CPU_SPEED_PERFORMANCE);  // Max boost
}
else if (cpu_usage > 70%) {
    PWR_setCPUSpeed(CPU_SPEED_NORMAL);       // Standard
}
else {
    PWR_setCPUSpeed(CPU_SPEED_POWERSAVE);    // Low power
}

Platform Opt-In Architecture

Platforms can opt into auto-scaling by implementing:

// In platform.h
#define PLAT_SUPPORTS_AUTO_CPU 1

// In platform.c  
void* PLAT_cpu_monitor(void *arg) {
    // Platform-specific monitoring implementation
}

Platforms without PLAT_SUPPORTS_AUTO_CPU will simply not show the "Auto" option in the menu.

Benefits

1. Cross-platform compatible - Uses existing CPU_SPEED_* constants
2. Opt-in - Platforms without support work unchanged
3. Simple - 3-tier scaling vs NextUI's 33-step approach
4. Battery friendly - Stays at low power when possible, boosts only when needed
5. Cool running - Reduces heat during lighter gameplay

Implementation Tasks

• Add volatile int useAutoCpu global flag
• Create reference PLAT_cpu_monitor() implementation for tg5040/miyoomini
• Add CPU usage measurement helpers (CLOCK_PROCESS_CPUTIME_ID)
• Add "Auto" option to minarch CPU Speed menu (when platform supports it)
• Add #ifdef PLAT_SUPPORTS_AUTO_CPU guards
• Test on platforms with auto-scaling support
• Document platform opt-in requirements

Reference Implementation

NextUI's auto-scaling:

• workspace/tg5040/platform/platform.c:2234-2345
• Background pthread monitoring every 20ms
• Uses POSIX clock_gettime(CLOCK_PROCESS_CPUTIME_ID) for accurate CPU time
• Mutex-protected shared state

Questions to Answer

• What threshold percentages work best across platforms? (70%/90% proposed)
• Should we expose thresholds as config options?
• What monitoring interval is optimal? (20ms? 50ms? 100ms?)

---

Related to NextUI investigation in #[TBD]

[nchapman]

Fixed in #56