Dynamically apply GPU turbo clocks only when GPU submissions are queued

Allows for the GPU to clock down in cases where it's idle for most of the time, while still forcing maximum clocks when we care.

app/src/main/cpp/skyline/common/android_settings.h

@@ -43,6 +43,7 @@ namespace skyline {
             executorSlotCountScale = ktSettings.GetInt<u32>("executorSlotCountScale");
             executorFlushThreshold = ktSettings.GetInt<u32>("executorFlushThreshold");
             useDirectMemoryImport = ktSettings.GetBool("useDirectMemoryImport");
+            forceMaxGpuClocks = ktSettings.GetBool("forceMaxGpuClocks");
             enableFastGpuReadbackHack = ktSettings.GetBool("enableFastGpuReadbackHack");
             isAudioOutputDisabled = ktSettings.GetBool("isAudioOutputDisabled");
             validationLayer = ktSettings.GetBool("validationLayer");

app/src/main/cpp/skyline/common/settings.h

@@ -75,6 +75,7 @@ namespace skyline {
         Setting<u32> executorSlotCountScale; //!< Number of GPU executor slots that can be used concurrently
         Setting<u32> executorFlushThreshold; //!< Number of commands that need to accumulate before they're flushed to the GPU
         Setting<bool> useDirectMemoryImport; //!< If buffer emulation should be done by importing guest buffer mappings
+        Setting<bool> forceMaxGpuClocks; //!< If the GPU should be forced to run at maximum clocks
 
         // Hacks
         Setting<bool> enableFastGpuReadbackHack; //!< If the CPU texture readback skipping hack should be used

app/src/main/cpp/skyline/gpu/interconnect/command_executor.cpp

@@ -2,6 +2,7 @@
 // Copyright © 2021 Skyline Team and Contributors (https://github.com/skyline-emu/)
 
 #include <range/v3/view.hpp>
+#include <adrenotools/driver.h>
 #include <common/settings.h>
 #include <loader/loader.h>
 #include <gpu.h>
@@ -195,13 +196,29 @@ namespace skyline::gpu::interconnect {
     void ExecutionWaiterThread::Run() {
         signal::SetSignalHandler({SIGSEGV}, nce::NCE::HostSignalHandler); // We may access NCE trapped memory
 
+        // Enable turbo clocks to begin with if requested
+        if (*state.settings->forceMaxGpuClocks)
+            adrenotools_set_turbo(true);
+
         while (true) {
             std::pair<std::shared_ptr<FenceCycle>, std::function<void()>> item{};
             {
                 std::unique_lock lock{mutex};
-                idle = true;
-                condition.wait(lock, [this] { return !pendingSignalQueue.empty(); });
-                idle = false;
+                if (pendingSignalQueue.empty()) {
+                    idle = true;
+
+                    // Don't force turbo clocks when the GPU is idle
+                    if (*state.settings->forceMaxGpuClocks)
+                        adrenotools_set_turbo(false);
+
+                    condition.wait(lock, [this] { return !pendingSignalQueue.empty(); });
+
+                    // Once we have work to do, force turbo clocks is enabled
+                    if (*state.settings->forceMaxGpuClocks)
+                        adrenotools_set_turbo(true);
+
+                    idle = false;
+                }
                 item = std::move(pendingSignalQueue.front());
                 pendingSignalQueue.pop();
             }
@@ -216,7 +233,7 @@ namespace skyline::gpu::interconnect {
         }
     }
 
-    ExecutionWaiterThread::ExecutionWaiterThread() : thread{&ExecutionWaiterThread::Run, this} {}
+    ExecutionWaiterThread::ExecutionWaiterThread(const DeviceState &state) : state{state}, thread{&ExecutionWaiterThread::Run, this} {}
 
     bool ExecutionWaiterThread::IsIdle() const {
         return idle;
@@ -232,6 +249,7 @@ namespace skyline::gpu::interconnect {
         : state{state},
           gpu{*state.gpu},
           recordThread{state},
+          waiterThread{state},
           tag{AllocateTag()} {
         RotateRecordSlot();
     }

app/src/main/cpp/skyline/gpu/interconnect/command_executor.h

@@ -97,6 +97,7 @@ namespace skyline::gpu::interconnect {
      */
     class ExecutionWaiterThread {
       private:
+        const DeviceState &state;
         std::thread thread;
         std::mutex mutex;
         std::condition_variable condition;
@@ -106,7 +107,7 @@ namespace skyline::gpu::interconnect {
         void Run();
 
       public:
-        ExecutionWaiterThread();
+        ExecutionWaiterThread(const DeviceState &state);
 
         bool IsIdle() const;
 

app/src/main/java/emu/skyline/EmulationActivity.kt

@@ -256,8 +256,6 @@ class EmulationActivity : AppCompatActivity(), SurfaceHolder.Callback, View.OnTo
 
         force60HzRefreshRate(!preferenceSettings.maxRefreshRate)
         getSystemService<DisplayManager>()?.registerDisplayListener(this, null)
-        if (preferenceSettings.forceMaxGpuClocks)
-            GpuDriverHelper.forceMaxGpuClocks(true)
 
         binding.gameView.setOnTouchListener(this)
 
@@ -291,9 +289,6 @@ class EmulationActivity : AppCompatActivity(), SurfaceHolder.Callback, View.OnTo
     override fun onResume() {
         super.onResume()
 
-        if (preferenceSettings.forceMaxGpuClocks)
-            GpuDriverHelper.forceMaxGpuClocks(true)
-
         changeAudioStatus(true)
 
         if (Build.VERSION.SDK_INT <= Build.VERSION_CODES.R) {

app/src/main/java/emu/skyline/utils/NativeSettings.kt

@@ -29,6 +29,7 @@ class NativeSettings(context : Context, pref : PreferenceSettings) {
     var executorSlotCountScale : Int = pref.executorSlotCountScale
     var executorFlushThreshold : Int = pref.executorFlushThreshold
     var useDirectMemoryImport : Boolean = pref.useDirectMemoryImport
+    var forceMaxGpuClocks : Boolean = pref.forceMaxGpuClocks
 
     // Hacks
     var enableFastGpuReadbackHack : Boolean = pref.enableFastGpuReadbackHack