Merge pull request #2587 from yuriks/status-bar

Replace built-in Profiler with indicators in status bar

src/core/CMakeLists.txt

@@ -173,6 +173,7 @@ set(SRCS
             loader/smdh.cpp
             tracer/recorder.cpp
             memory.cpp
+            perf_stats.cpp
             settings.cpp
             )
 
@@ -363,6 +364,7 @@ set(HEADERS
             memory.h
             memory_setup.h
             mmio.h
+            perf_stats.h
             settings.h
             )
 

src/core/core.cpp

@@ -109,6 +109,10 @@ void System::PrepareReschedule() {
     reschedule_pending = true;
 }
 
+PerfStats::Results System::GetAndResetPerfStats() {
+    return perf_stats.GetAndResetStats(CoreTiming::GetGlobalTimeUs());
+}
+
 void System::Reschedule() {
     if (!reschedule_pending) {
         return;
@@ -140,6 +144,10 @@ System::ResultStatus System::Init(EmuWindow* emu_window, u32 system_mode) {
 
     LOG_DEBUG(Core, "Initialized OK");
 
+    // Reset counters and set time origin to current frame
+    GetAndResetPerfStats();
+    perf_stats.BeginSystemFrame();
+
     return ResultStatus::Success;
 }
 

src/core/core.h

@@ -6,9 +6,9 @@
 
 #include <memory>
 #include <string>
-
 #include "common/common_types.h"
 #include "core/memory.h"
+#include "core/perf_stats.h"
 
 class EmuWindow;
 class ARM_Interface;
@@ -83,6 +83,8 @@ public:
     /// Prepare the core emulation for a reschedule
     void PrepareReschedule();
 
+    PerfStats::Results GetAndResetPerfStats();
+
     /**
      * Gets a reference to the emulated CPU.
      * @returns A reference to the emulated CPU.
@@ -91,6 +93,9 @@ public:
         return *cpu_core;
     }
 
+    PerfStats perf_stats;
+    FrameLimiter frame_limiter;
+
 private:
     /**
      * Initialize the emulated system.

src/core/frontend/emu_window.cpp

@@ -5,7 +5,7 @@
 #include <algorithm>
 #include <cmath>
 #include "common/assert.h"
-#include "common/profiler_reporting.h"
+#include "core/core.h"
 #include "core/frontend/emu_window.h"
 #include "core/frontend/key_map.h"
 #include "video_core/video_core.h"
@@ -104,8 +104,7 @@ void EmuWindow::AccelerometerChanged(float x, float y, float z) {
 void EmuWindow::GyroscopeChanged(float x, float y, float z) {
     constexpr float FULL_FPS = 60;
     float coef = GetGyroscopeRawToDpsCoefficient();
-    float stretch =
-        FULL_FPS / Common::Profiling::GetTimingResultsAggregator()->GetAggregatedResults().fps;
+    float stretch = Core::System::GetInstance().perf_stats.GetLastFrameTimeScale();
     std::lock_guard<std::mutex> lock(gyro_mutex);
     gyro_x = static_cast<s16>(x * coef * stretch);
     gyro_y = static_cast<s16>(y * coef * stretch);

src/core/hle/kernel/server_session.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <memory>
 #include <string>
 #include "common/assert.h"
 #include "common/common_types.h"

src/core/hle/kernel/thread.h

@@ -11,7 +11,6 @@
 #include <boost/container/flat_set.hpp>
 #include "common/common_types.h"
 #include "core/arm/arm_interface.h"
-#include "core/core.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/result.h"
 

src/core/hle/service/gsp_gpu.cpp

@@ -4,6 +4,7 @@
 
 #include "common/bit_field.h"
 #include "common/microprofile.h"
+#include "core/core.h"
 #include "core/hle/kernel/event.h"
 #include "core/hle/kernel/shared_memory.h"
 #include "core/hle/result.h"
@@ -280,6 +281,7 @@ ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info) {
 
     if (screen_id == 0) {
         MicroProfileFlip();
+        Core::System::GetInstance().perf_stats.EndGameFrame();
     }
 
     return RESULT_SUCCESS;

src/core/hle/service/ldr_ro/ldr_ro.cpp

@@ -6,6 +6,7 @@
 #include "common/common_types.h"
 #include "common/logging/log.h"
 #include "core/arm/arm_interface.h"
+#include "core/core.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/hle/service/ldr_ro/cro_helper.h"

src/core/hw/gpu.cpp

@@ -8,17 +8,13 @@
 #include "common/color.h"
 #include "common/common_types.h"
 #include "common/logging/log.h"
-#include "common/math_util.h"
 #include "common/microprofile.h"
-#include "common/thread.h"
-#include "common/timer.h"
 #include "common/vector_math.h"
 #include "core/core_timing.h"
 #include "core/hle/service/gsp_gpu.h"
 #include "core/hw/gpu.h"
 #include "core/hw/hw.h"
 #include "core/memory.h"
-#include "core/settings.h"
 #include "core/tracer/recorder.h"
 #include "video_core/command_processor.h"
 #include "video_core/debug_utils/debug_utils.h"
@@ -32,19 +28,9 @@ namespace GPU {
 Regs g_regs;
 
 /// 268MHz CPU clocks / 60Hz frames per second
-const u64 frame_ticks = BASE_CLOCK_RATE_ARM11 / 60;
+const u64 frame_ticks = BASE_CLOCK_RATE_ARM11 / SCREEN_REFRESH_RATE;
 /// Event id for CoreTiming
 static int vblank_event;
-/// Total number of frames drawn
-static u64 frame_count;
-/// Start clock for frame limiter
-static u32 time_point;
-/// Total delay caused by slow frames
-static float time_delay;
-constexpr float FIXED_FRAME_TIME = 1000.0f / 60;
-// Max lag caused by slow frames. Can be adjusted to compensate for too many slow frames. Higher
-// values increases time needed to limit frame rate after spikes
-constexpr float MAX_LAG_TIME = 18;
 
 template <typename T>
 inline void Read(T& var, const u32 raw_addr) {
@@ -522,24 +508,8 @@ template void Write<u32>(u32 addr, const u32 data);
 template void Write<u16>(u32 addr, const u16 data);
 template void Write<u8>(u32 addr, const u8 data);
 
-static void FrameLimiter() {
-    time_delay += FIXED_FRAME_TIME;
-    time_delay = MathUtil::Clamp(time_delay, -MAX_LAG_TIME, MAX_LAG_TIME);
-    s32 desired_time = static_cast<s32>(time_delay);
-    s32 elapsed_time = static_cast<s32>(Common::Timer::GetTimeMs() - time_point);
-
-    if (elapsed_time < desired_time) {
-        Common::SleepCurrentThread(desired_time - elapsed_time);
-    }
-
-    u32 frame_time = Common::Timer::GetTimeMs() - time_point;
-
-    time_delay -= frame_time;
-}
-
 /// Update hardware
 static void VBlankCallback(u64 userdata, int cycles_late) {
-    frame_count++;
     VideoCore::g_renderer->SwapBuffers();
 
     // Signal to GSP that GPU interrupt has occurred
@@ -550,12 +520,6 @@ static void VBlankCallback(u64 userdata, int cycles_late) {
     Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC0);
     Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC1);
 
-    if (!Settings::values.use_vsync && Settings::values.toggle_framelimit) {
-        FrameLimiter();
-    }
-
-    time_point = Common::Timer::GetTimeMs();
-
     // Reschedule recurrent event
     CoreTiming::ScheduleEvent(frame_ticks - cycles_late, vblank_event);
 }
@@ -590,9 +554,6 @@ void Init() {
     framebuffer_sub.color_format.Assign(Regs::PixelFormat::RGB8);
     framebuffer_sub.active_fb = 0;
 
-    frame_count = 0;
-    time_point = Common::Timer::GetTimeMs();
-
     vblank_event = CoreTiming::RegisterEvent("GPU::VBlankCallback", VBlankCallback);
     CoreTiming::ScheduleEvent(frame_ticks, vblank_event);
 

src/core/hw/gpu.h

@@ -13,6 +13,8 @@
 
 namespace GPU {
 
+constexpr float SCREEN_REFRESH_RATE = 60;
+
 // Returns index corresponding to the Regs member labeled by field_name
 // TODO: Due to Visual studio bug 209229, offsetof does not return constant expressions
 //       when used with array elements (e.g. GPU_REG_INDEX(memory_fill_config[0])).

src/core/perf_stats.cpp

@@ -0,0 +1,105 @@
+// Copyright 2017 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <chrono>
+#include <mutex>
+#include <thread>
+#include "common/math_util.h"
+#include "core/hw/gpu.h"
+#include "core/perf_stats.h"
+#include "core/settings.h"
+
+using namespace std::chrono_literals;
+using DoubleSecs = std::chrono::duration<double, std::chrono::seconds::period>;
+using std::chrono::duration_cast;
+using std::chrono::microseconds;
+
+namespace Core {
+
+void PerfStats::BeginSystemFrame() {
+    std::lock_guard<std::mutex> lock(object_mutex);
+
+    frame_begin = Clock::now();
+}
+
+void PerfStats::EndSystemFrame() {
+    std::lock_guard<std::mutex> lock(object_mutex);
+
+    auto frame_end = Clock::now();
+    accumulated_frametime += frame_end - frame_begin;
+    system_frames += 1;
+
+    previous_frame_length = frame_end - previous_frame_end;
+    previous_frame_end = frame_end;
+}
+
+void PerfStats::EndGameFrame() {
+    std::lock_guard<std::mutex> lock(object_mutex);
+
+    game_frames += 1;
+}
+
+PerfStats::Results PerfStats::GetAndResetStats(u64 current_system_time_us) {
+    std::lock_guard<std::mutex> lock(object_mutex);
+
+    auto now = Clock::now();
+    // Walltime elapsed since stats were reset
+    auto interval = duration_cast<DoubleSecs>(now - reset_point).count();
+
+    auto system_us_per_second =
+        static_cast<double>(current_system_time_us - reset_point_system_us) / interval;
+
+    Results results{};
+    results.system_fps = static_cast<double>(system_frames) / interval;
+    results.game_fps = static_cast<double>(game_frames) / interval;
+    results.frametime = duration_cast<DoubleSecs>(accumulated_frametime).count() /
+                        static_cast<double>(system_frames);
+    results.emulation_speed = system_us_per_second / 1'000'000.0;
+
+    // Reset counters
+    reset_point = now;
+    reset_point_system_us = current_system_time_us;
+    accumulated_frametime = Clock::duration::zero();
+    system_frames = 0;
+    game_frames = 0;
+
+    return results;
+}
+
+double PerfStats::GetLastFrameTimeScale() {
+    std::lock_guard<std::mutex> lock(object_mutex);
+
+    constexpr double FRAME_LENGTH = 1.0 / GPU::SCREEN_REFRESH_RATE;
+    return duration_cast<DoubleSecs>(previous_frame_length).count() / FRAME_LENGTH;
+}
+
+void FrameLimiter::DoFrameLimiting(u64 current_system_time_us) {
+    // Max lag caused by slow frames. Can be adjusted to compensate for too many slow frames. Higher
+    // values increase the time needed to recover and limit framerate again after spikes.
+    constexpr microseconds MAX_LAG_TIME_US = 25ms;
+
+    if (!Settings::values.toggle_framelimit) {
+        return;
+    }
+
+    auto now = Clock::now();
+
+    frame_limiting_delta_err += microseconds(current_system_time_us - previous_system_time_us);
+    frame_limiting_delta_err -= duration_cast<microseconds>(now - previous_walltime);
+    frame_limiting_delta_err =
+        MathUtil::Clamp(frame_limiting_delta_err, -MAX_LAG_TIME_US, MAX_LAG_TIME_US);
+
+    if (frame_limiting_delta_err > microseconds::zero()) {
+        std::this_thread::sleep_for(frame_limiting_delta_err);
+
+        auto now_after_sleep = Clock::now();
+        frame_limiting_delta_err -= duration_cast<microseconds>(now_after_sleep - now);
+        now = now_after_sleep;
+    }
+
+    previous_system_time_us = current_system_time_us;
+    previous_walltime = now;
+}
+
+} // namespace Core

src/core/perf_stats.h

@@ -0,0 +1,83 @@
+// Copyright 2017 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <chrono>
+#include <mutex>
+#include "common/common_types.h"
+
+namespace Core {
+
+/**
+ * Class to manage and query performance/timing statistics. All public functions of this class are
+ * thread-safe unless stated otherwise.
+ */
+class PerfStats {
+public:
+    using Clock = std::chrono::high_resolution_clock;
+
+    struct Results {
+        /// System FPS (LCD VBlanks) in Hz
+        double system_fps;
+        /// Game FPS (GSP frame submissions) in Hz
+        double game_fps;
+        /// Walltime per system frame, in seconds, excluding any waits
+        double frametime;
+        /// Ratio of walltime / emulated time elapsed
+        double emulation_speed;
+    };
+
+    void BeginSystemFrame();
+    void EndSystemFrame();
+    void EndGameFrame();
+
+    Results GetAndResetStats(u64 current_system_time_us);
+
+    /**
+     * Gets the ratio between walltime and the emulated time of the previous system frame. This is
+     * useful for scaling inputs or outputs moving between the two time domains.
+     */
+    double GetLastFrameTimeScale();
+
+private:
+    std::mutex object_mutex;
+
+    /// Point when the cumulative counters were reset
+    Clock::time_point reset_point = Clock::now();
+    /// System time when the cumulative counters were reset
+    u64 reset_point_system_us = 0;
+
+    /// Cumulative duration (excluding v-sync/frame-limiting) of frames since last reset
+    Clock::duration accumulated_frametime = Clock::duration::zero();
+    /// Cumulative number of system frames (LCD VBlanks) presented since last reset
+    u32 system_frames = 0;
+    /// Cumulative number of game frames (GSP frame submissions) since last reset
+    u32 game_frames = 0;
+
+    /// Point when the previous system frame ended
+    Clock::time_point previous_frame_end = reset_point;
+    /// Point when the current system frame began
+    Clock::time_point frame_begin = reset_point;
+    /// Total visible duration (including frame-limiting, etc.) of the previous system frame
+    Clock::duration previous_frame_length = Clock::duration::zero();
+};
+
+class FrameLimiter {
+public:
+    using Clock = std::chrono::high_resolution_clock;
+
+    void DoFrameLimiting(u64 current_system_time_us);
+
+private:
+    /// Emulated system time (in microseconds) at the last limiter invocation
+    u64 previous_system_time_us = 0;
+    /// Walltime at the last limiter invocation
+    Clock::time_point previous_walltime = Clock::now();
+
+    /// Accumulated difference between walltime and emulated time
+    std::chrono::microseconds frame_limiting_delta_err{0};
+};
+
+} // namespace Core