Rework Performance Statistics

We used instantaneous values for FPS previously which led to a lot of variation in it and the inability to determine a proper FPS value due to constant fluctuations. All FPS values are now averaged to allow reading out a stable value and a deviation statistic has been added for the frame-time to judge judder and frame-pacing which allows for a significantly better measure of overall performance. The formatting for all the floating-point numbers is now fixed-point to prevent shifting of position due to decimal digits becoming 0.

.gitignore

@@ -51,6 +51,7 @@ build/
 .idea/caches
 .idea/assetWizardSettings.xml
 .idea/runConfigurations.xml
+.idea/deploymentTargetDropDown.xml
 
 # Mongo Explorer plugin
 .idea/**/mongoSettings.xml

app/src/main/cpp/emu_jni.cpp

@@ -19,15 +19,17 @@
 #include "skyline/input.h"
 #include "skyline/kernel/types/KProcess.h"
 
-skyline::i32 Fps;
+jint Fps; //!< An approximation of the amount of frames being submitted every second
-skyline::i32 FrameTime;
+jfloat AverageFrametimeMs; //!< The average time it takes for a frame to be rendered and presented in milliseconds
+jfloat AverageFrametimeDeviationMs; //!< The average deviation of the average frametimes in milliseconds
+
 std::weak_ptr<skyline::kernel::OS> OsWeak;
 std::weak_ptr<skyline::gpu::GPU> GpuWeak;
 std::weak_ptr<skyline::input::Input> InputWeak;
 
 // https://cs.android.com/android/platform/superproject/+/master:bionic/libc/tzcode/bionic.cpp;l=43;drc=master;bpv=1;bpt=1
 static std::string GetTimeZoneName() {
-    const char* nameEnv = getenv("TZ");
+    const char *nameEnv = getenv("TZ");
     if (nameEnv)
         return std::string(nameEnv);
 
@@ -52,7 +54,7 @@ static std::string GetTimeZoneName() {
 
 extern "C" JNIEXPORT void Java_emu_skyline_EmulationActivity_executeApplication(JNIEnv *env, jobject instance, jstring romUriJstring, jint romType, jint romFd, jint preferenceFd, jstring appFilesPathJstring, jobject assetManager) {
     skyline::signal::ScopedStackBlocker stackBlocker; // We do not want anything to unwind past JNI code as there are invalid stack frames which can lead to a segmentation fault
-    Fps = FrameTime = 0;
+    Fps = AverageFrametimeMs = AverageFrametimeDeviationMs = 0;
 
     pthread_setname_np(pthread_self(), "EmuMain");
 
@@ -133,10 +135,15 @@ extern "C" JNIEXPORT void Java_emu_skyline_EmulationActivity_updatePerformanceSt
         fpsField = env->GetFieldID(clazz, "fps", "I");
     env->SetIntField(thiz, fpsField, Fps);
 
-    static jfieldID frametimeField{};
+    static jfieldID averageFrametimeField{};
-    if (!frametimeField)
+    if (!averageFrametimeField)
-        frametimeField = env->GetFieldID(clazz, "frametime", "F");
+        averageFrametimeField = env->GetFieldID(clazz, "averageFrametime", "F");
-    env->SetFloatField(thiz, frametimeField, static_cast<float>(FrameTime) / 100);
+    env->SetFloatField(thiz, averageFrametimeField, AverageFrametimeMs);
+
+    static jfieldID averageFrametimeDeviationField{};
+    if (!averageFrametimeDeviationField)
+        averageFrametimeDeviationField = env->GetFieldID(clazz, "averageFrametimeDeviation", "F");
+    env->SetFloatField(thiz, averageFrametimeDeviationField, AverageFrametimeDeviationMs);
 }
 
 extern "C" JNIEXPORT void JNICALL Java_emu_skyline_EmulationActivity_setController(JNIEnv *, jobject, jint index, jint type, jint partnerIndex) {

app/src/main/cpp/skyline/gpu/presentation_engine.cpp

@@ -13,8 +13,9 @@
 #include "native_window.h"
 #include "texture/format.h"
 
-extern skyline::i32 Fps;
+extern jint Fps;
-extern skyline::i32 FrameTime;
+extern jfloat AverageFrametimeMs;
+extern jfloat AverageFrametimeDeviationMs;
 
 namespace skyline::gpu {
     using namespace service::hosbinder;
@@ -39,11 +40,12 @@ namespace skyline::gpu {
 
     void PresentationEngine::ChoreographerCallback(long frameTimeNanos, PresentationEngine *engine) {
         u64 cycleLength{frameTimeNanos - engine->lastChoreographerTime};
-        if (std::abs(static_cast<i64>(cycleLength - engine->refreshCycleDuration)) > (constant::NsInMillisecond / 2))
+        if (std::abs(static_cast<i64>(cycleLength - engine->refreshCycleDuration)) > (constant::NsInMillisecond / 2)) {
             if (engine->window)
                 engine->window->perform(engine->window, NATIVE_WINDOW_GET_REFRESH_CYCLE_DURATION, &engine->refreshCycleDuration);
             else
                 engine->refreshCycleDuration = cycleLength;
+        }
 
         engine->lastChoreographerTime = frameTimeNanos;
         engine->vsyncEvent->Signal();
@@ -271,8 +273,17 @@ namespace skyline::gpu {
 
         if (frameTimestamp) {
             auto now{util::GetTimeNs()};
-            FrameTime = static_cast<u32>((now - frameTimestamp) / 10000); // frametime / 100 is the real ms value, this is to retain the first two decimals
+            auto sampleWeight{swapInterval ? constant::NsInSecond / (refreshCycleDuration * swapInterval) : 10}; //!< The weight of each sample in calculating the average, we arbitrarily average 10 samples for unlocked FPS
-            Fps = static_cast<u16>(constant::NsInSecond / (now - frameTimestamp));
+
+            u64 currentFrametime{now - frameTimestamp};
+            averageFrametimeNs = (((sampleWeight - 1) * averageFrametimeNs) / sampleWeight) + (currentFrametime / sampleWeight);
+            AverageFrametimeMs = static_cast<jfloat>(averageFrametimeNs) / constant::NsInMillisecond;
+
+            i64 currentFrametimeDeviation{std::abs(static_cast<i64>(averageFrametimeNs - currentFrametime))};
+            averageFrametimeDeviationNs = (((sampleWeight - 1) * averageFrametimeDeviationNs) / sampleWeight) + (currentFrametimeDeviation / sampleWeight);
+            AverageFrametimeDeviationMs = static_cast<jfloat>(averageFrametimeDeviationNs) / constant::NsInMillisecond;
+
+            Fps = std::round(static_cast<float>(constant::NsInSecond) / averageFrametimeNs);
 
             TRACE_EVENT_INSTANT("gpu", "Present", presentationTrack, "FrameTimeNs", now - frameTimestamp, "Fps", Fps);
 

app/src/main/cpp/skyline/gpu/presentation_engine.h

@@ -40,7 +40,9 @@ namespace skyline::gpu {
         static constexpr size_t MaxSwapchainImageCount{6}; //!< The maximum amount of swapchain textures, this affects the amount of images that can be in the swapchain
         std::array<std::shared_ptr<Texture>, MaxSwapchainImageCount> images; //!< All the swapchain textures in the same order as supplied by the host swapchain
 
-        u64 frameTimestamp{}; //!< The timestamp of the last frame being shown
+        u64 frameTimestamp{}; //!< The timestamp of the last frame being shown in nanoseconds
+        u64 averageFrametimeNs{}; //!< The average time between frames in nanoseconds
+        u64 averageFrametimeDeviationNs{}; //!< The average deviation of frametimes in nanoseconds
         perfetto::Track presentationTrack; //!< Perfetto track used for presentation events
 
         std::thread choreographerThread; //!< A thread for signalling the V-Sync event using AChoreographer

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

@@ -86,7 +86,8 @@ class EmulationActivity : AppCompatActivity(), SurfaceHolder.Callback, View.OnTo
     private external fun setSurface(surface : Surface?) : Boolean
 
     var fps : Int = 0
-    var frametime : Float = 0.0f
+    var averageFrametime : Float = 0.0f
+    var averageFrametimeDeviation : Float = 0.0f
 
     /**
      * Writes the current performance statistics into [fps] and [frametime] fields
@@ -204,7 +205,7 @@ class EmulationActivity : AppCompatActivity(), SurfaceHolder.Callback, View.OnTo
                 postDelayed(object : Runnable {
                     override fun run() {
                         updatePerformanceStatistics()
-                        text = "$fps FPS\n${frametime}ms"
+                        text = "$fps FPS\n${"%.1f".format(averageFrametime)}±${"%.2f".format(averageFrametimeDeviation)}ms"
                         postDelayed(this, 250)
                     }
                 }, 250)