Skip to content

NFSU2 Revived v0.15.9

Choose a tag to compare

@github-actions github-actions released this 03 Nov 00:25
· 214 commits to main since this release

What's Changed in v0.15.9

Multi-Thread Performance Cores Affinity & EAS Capacity Discovery

  • Energy Aware Scheduling (EAS) Fallback: On many modern Android kernels (particularly MediaTek Dimensity/Helio and locked Snapdragon devices), /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_max_freq is restricted by SELinux policies, resulting in zeroed frequency readings and defaulting to all cores. Added multi-tier topology discovery in both pure Rust (libperf_cores.so) and C runtime companion (roam_android.c) inspecting scaling_max_freq and ARM EAS cpu_capacity:
    let mut freq = fs::read_to_string(&freq_path)
        .ok()
        .and_then(|s| s.trim().parse::<u64>().ok())
        .unwrap_or(0);
    
    if freq == 0 {
        freq = fs::read_to_string(&scaling_path)
            .ok()
            .and_then(|s| s.trim().parse::<u64>().ok())
            .unwrap_or(0);
    }
    
    // ARM cpu_capacity (EAS): big cores ~1024, little cores ~300-500
    if freq == 0 {
        if let Some(cap) = fs::read_to_string(&capacity_path)
            .ok()
            .and_then(|s| s.trim().parse::<u64>().ok())
        {
            freq = cap * 1000;
        }
    }
  • Whole-Process Task Thread Pinning: Previously, sched_setaffinity only bound the calling thread and the main PID. Asynchronous worker threads spawned by the guest runner, AAudio, and Vulkan driver could still be scheduled on LITTLE efficiency cores. Implemented full thread traversal via /proc/self/task to pin all existing and background threads to Big/Prime performance cores:
    // Bind every existing thread in /proc/self/task (GuestBootWorker, audio, render, etc.)
    if let Ok(entries) = fs::read_dir("/proc/self/task") {
        for entry in entries.flatten() {
            if let Ok(tid) = entry.file_name().to_string_lossy().parse::<i32>() {
                let _ = libc::syscall(
                    libc::SYS_sched_setaffinity,
                    tid as c_int,
                    std::mem::size_of::<c_long>(),
                    &mask_val as *const c_long,
                );
            }
        }
    }

Direct Graphics API Selection & Backend Compatibility

  • Unrestricted Graphics API Selection: Removed dynamic suffix filtering and button hiding logic in GameOptions.java and LauncherActivity.java. All backend choices ("Auto", "Vulkan (DXVK)", and "OpenGL (ES)") are now always displayed and user-selectable:
    static final Option[] ALL = {
            new Option(GRAPHICS_BACKEND, "Graphics API", "auto",
                    new String[] {"auto", "vulkan", "opengl"},
                    new String[] {"Auto", "Vulkan (DXVK)", "OpenGL (ES)"}),
  • Transparent Fallback Behavior: Auto settings now function strictly as intelligent defaults rather than forced overrides, allowing users to test or enforce OpenGL ES or Vulkan directly on any device without the launcher restricting options.

GPU Error Diagnostics & Runtime Message Normalization

  • Descriptive English Device Creation Exceptions: Normalized Spanish GPU error dialog strings in d3d9_bridge.cpp and .rodata. When low-tier GPUs (e.g. Adreno 610) fail Vulkan CreateDevice with 0x8876086A (D3DERR_NOTAVAILABLE), users now receive clear, actionable English instructions:
    The GPU failed to create graphics device (Vulkan, CreateDevice %u). Use "Test driver" in launcher to see what is missing.

Full Changelog: v0.15.8...v0.15.9