core_timing: Use higher precision sleeps on Windows

The precision of sleep_for and wait_for is limited to 1-1.5ms on Windows.
Using SleepForOneTick() allows us to sleep for exactly one interval of the current timer resolution.
This allows us to take advantage of systems that have a timer resolution of 0.5ms to reduce CPU overhead in the event loop.

src/common/wall_clock.cpp

@@ -81,4 +81,9 @@ std::unique_ptr<WallClock> CreateBestMatchingClock(u64 emulated_cpu_frequency,
 
 #endif
 
+std::unique_ptr<WallClock> CreateStandardWallClock(u64 emulated_cpu_frequency,
+                                                   u64 emulated_clock_frequency) {
+    return std::make_unique<StandardWallClock>(emulated_cpu_frequency, emulated_clock_frequency);
+}
+
 } // namespace Common

src/common/wall_clock.h

@@ -55,4 +55,7 @@ private:
 [[nodiscard]] std::unique_ptr<WallClock> CreateBestMatchingClock(u64 emulated_cpu_frequency,
                                                                  u64 emulated_clock_frequency);
 
+[[nodiscard]] std::unique_ptr<WallClock> CreateStandardWallClock(u64 emulated_cpu_frequency,
+                                                                 u64 emulated_clock_frequency);
+
 } // namespace Common

src/core/core_timing.cpp

@@ -6,6 +6,10 @@
 #include <string>
 #include <tuple>
 
+#ifdef _WIN32
+#include "common/windows/timer_resolution.h"
+#endif
+
 #include "common/microprofile.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
@@ -38,7 +42,8 @@ struct CoreTiming::Event {
 };
 
 CoreTiming::CoreTiming()
-    : clock{Common::CreateBestMatchingClock(Hardware::BASE_CLOCK_RATE, Hardware::CNTFREQ)} {}
+    : cpu_clock{Common::CreateBestMatchingClock(Hardware::BASE_CLOCK_RATE, Hardware::CNTFREQ)},
+      event_clock{Common::CreateStandardWallClock(Hardware::BASE_CLOCK_RATE, Hardware::CNTFREQ)} {}
 
 CoreTiming::~CoreTiming() {
     Reset();
@@ -185,15 +190,15 @@ void CoreTiming::ResetTicks() {
 }
 
 u64 CoreTiming::GetCPUTicks() const {
-    if (is_multicore) {
+    if (is_multicore) [[likely]] {
-        return clock->GetCPUCycles();
+        return cpu_clock->GetCPUCycles();
     }
     return ticks;
 }
 
 u64 CoreTiming::GetClockTicks() const {
-    if (is_multicore) {
+    if (is_multicore) [[likely]] {
-        return clock->GetClockCycles();
+        return cpu_clock->GetClockCycles();
     }
     return CpuCyclesToClockCycles(ticks);
 }
@@ -252,22 +257,21 @@ void CoreTiming::ThreadLoop() {
             const auto next_time = Advance();
             if (next_time) {
                 // There are more events left in the queue, wait until the next event.
-                const auto wait_time = *next_time - GetGlobalTimeNs().count();
+                auto wait_time = *next_time - GetGlobalTimeNs().count();
                 if (wait_time > 0) {
 #ifdef _WIN32
-                    // Assume a timer resolution of 1ms.
+                    const auto timer_resolution_ns =
-                    static constexpr s64 TimerResolutionNS = 1000000;
+                        Common::Windows::GetCurrentTimerResolution().count();
 
-                    // Sleep in discrete intervals of the timer resolution, and spin the rest.
+                    while (!paused && !event.IsSet() && wait_time > 0) {
-                    const auto sleep_time = wait_time - (wait_time % TimerResolutionNS);
+                        wait_time = *next_time - GetGlobalTimeNs().count();
-                    if (sleep_time > 0) {
-                        event.WaitFor(std::chrono::nanoseconds(sleep_time));
-                    }
 
-                    while (!paused && !event.IsSet() && GetGlobalTimeNs().count() < *next_time) {
+                        if (wait_time >= timer_resolution_ns) {
-                        // Yield to reduce thread starvation.
+                            Common::Windows::SleepForOneTick();
+                        } else {
                             std::this_thread::yield();
                         }
+                    }
 
                     if (event.IsSet()) {
                         event.Reset();
@@ -285,9 +289,9 @@ void CoreTiming::ThreadLoop() {
         }
 
         paused_set = true;
-        clock->Pause(true);
+        event_clock->Pause(true);
         pause_event.Wait();
-        clock->Pause(false);
+        event_clock->Pause(false);
     }
 }
 
@@ -303,16 +307,23 @@ void CoreTiming::Reset() {
     has_started = false;
 }
 
+std::chrono::nanoseconds CoreTiming::GetCPUTimeNs() const {
+    if (is_multicore) [[likely]] {
+        return cpu_clock->GetTimeNS();
+    }
+    return CyclesToNs(ticks);
+}
+
 std::chrono::nanoseconds CoreTiming::GetGlobalTimeNs() const {
-    if (is_multicore) {
+    if (is_multicore) [[likely]] {
-        return clock->GetTimeNS();
+        return event_clock->GetTimeNS();
     }
     return CyclesToNs(ticks);
 }
 
 std::chrono::microseconds CoreTiming::GetGlobalTimeUs() const {
-    if (is_multicore) {
+    if (is_multicore) [[likely]] {
-        return clock->GetTimeUS();
+        return event_clock->GetTimeUS();
     }
     return CyclesToUs(ticks);
 }

src/core/core_timing.h

@@ -122,6 +122,9 @@ public:
     /// Returns current time in emulated in Clock cycles
     u64 GetClockTicks() const;
 
+    /// Returns current time in nanoseconds.
+    std::chrono::nanoseconds GetCPUTimeNs() const;
+
     /// Returns current time in microseconds.
     std::chrono::microseconds GetGlobalTimeUs() const;
 
@@ -139,7 +142,8 @@ private:
 
     void Reset();
 
-    std::unique_ptr<Common::WallClock> clock;
+    std::unique_ptr<Common::WallClock> cpu_clock;
+    std::unique_ptr<Common::WallClock> event_clock;
 
     s64 global_timer = 0;
 

src/video_core/gpu.cpp

@@ -197,7 +197,7 @@ struct GPU::Impl {
         constexpr u64 gpu_ticks_num = 384;
         constexpr u64 gpu_ticks_den = 625;
 
-        u64 nanoseconds = system.CoreTiming().GetGlobalTimeNs().count();
+        u64 nanoseconds = system.CoreTiming().GetCPUTimeNs().count();
         if (Settings::values.use_fast_gpu_time.GetValue()) {
             nanoseconds /= 256;
         }