8 files changed, 267 insertions, 28 deletions

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 56b247ac4..58ff5f2f3 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -113,6 +113,8 @@ add_library(common STATIC
     socket_types.h
     spin_lock.cpp
     spin_lock.h
+    steady_clock.cpp
+    steady_clock.h
     stream.cpp
     stream.h
     string_util.cpp
@@ -142,6 +144,14 @@ add_library(common STATIC
     zstd_compression.h
 )
 
+if (WIN32)
+  target_sources(common PRIVATE
+    windows/timer_resolution.cpp
+    windows/timer_resolution.h
+  )
+  target_link_libraries(common PRIVATE ntdll)
+endif()
+
 if(ARCHITECTURE_x86_64)
     target_sources(common
         PRIVATE
diff --git a/src/common/steady_clock.cpp b/src/common/steady_clock.cpp
new file mode 100644
index 000000000..0d5908aa7
--- /dev/null
+++ b/src/common/steady_clock.cpp
@@ -0,0 +1,56 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#if defined(_WIN32)
+#include <windows.h>
+#else
+#include <time.h>
+#endif
+
+#include "common/steady_clock.h"
+
+namespace Common {
+
+#ifdef _WIN32
+static s64 WindowsQueryPerformanceFrequency() {
+    LARGE_INTEGER frequency;
+    QueryPerformanceFrequency(&frequency);
+    return frequency.QuadPart;
+}
+
+static s64 WindowsQueryPerformanceCounter() {
+    LARGE_INTEGER counter;
+    QueryPerformanceCounter(&counter);
+    return counter.QuadPart;
+}
+#endif
+
+SteadyClock::time_point SteadyClock::Now() noexcept {
+#if defined(_WIN32)
+    static const auto freq = WindowsQueryPerformanceFrequency();
+    const auto counter = WindowsQueryPerformanceCounter();
+
+    // 10 MHz is a very common QPC frequency on modern PCs.
+    // Optimizing for this specific frequency can double the performance of
+    // this function by avoiding the expensive frequency conversion path.
+    static constexpr s64 TenMHz = 10'000'000;
+
+    if (freq == TenMHz) [[likely]] {
+        static_assert(period::den % TenMHz == 0);
+        static constexpr s64 Multiplier = period::den / TenMHz;
+        return time_point{duration{counter * Multiplier}};
+    }
+
+    const auto whole = (counter / freq) * period::den;
+    const auto part = (counter % freq) * period::den / freq;
+    return time_point{duration{whole + part}};
+#elif defined(__APPLE__)
+    return time_point{duration{clock_gettime_nsec_np(CLOCK_MONOTONIC_RAW)}};
+#else
+    timespec ts;
+    clock_gettime(CLOCK_MONOTONIC, &ts);
+    return time_point{std::chrono::seconds{ts.tv_sec} + std::chrono::nanoseconds{ts.tv_nsec}};
+#endif
+}
+
+}; // namespace Common
diff --git a/src/common/steady_clock.h b/src/common/steady_clock.h
new file mode 100644
index 000000000..9497cf865
--- /dev/null
+++ b/src/common/steady_clock.h
@@ -0,0 +1,23 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <chrono>
+
+#include "common/common_types.h"
+
+namespace Common {
+
+struct SteadyClock {
+    using rep = s64;
+    using period = std::nano;
+    using duration = std::chrono::nanoseconds;
+    using time_point = std::chrono::time_point<SteadyClock>;
+
+    static constexpr bool is_steady = true;
+
+    [[nodiscard]] static time_point Now() noexcept;
+};
+
+} // namespace Common
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
index ae07f2811..817e71d52 100644
--- a/src/common/wall_clock.cpp
+++ b/src/common/wall_clock.cpp
@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include "common/steady_clock.h"
 #include "common/uint128.h"
 #include "common/wall_clock.h"
 
@@ -11,45 +12,32 @@
 
 namespace Common {
 
-using base_timer = std::chrono::steady_clock;
-using base_time_point = std::chrono::time_point<base_timer>;
-
 class StandardWallClock final : public WallClock {
 public:
     explicit StandardWallClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequency_)
-        : WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, false) {
-        start_time = base_timer::now();
-    }
+        : WallClock{emulated_cpu_frequency_, emulated_clock_frequency_, false},
+          start_time{SteadyClock::Now()} {}
 
     std::chrono::nanoseconds GetTimeNS() override {
-        base_time_point current = base_timer::now();
-        auto elapsed = current - start_time;
-        return std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed);
+        return SteadyClock::Now() - start_time;
     }
 
     std::chrono::microseconds GetTimeUS() override {
-        base_time_point current = base_timer::now();
-        auto elapsed = current - start_time;
-        return std::chrono::duration_cast<std::chrono::microseconds>(elapsed);
+        return std::chrono::duration_cast<std::chrono::microseconds>(GetTimeNS());
     }
 
     std::chrono::milliseconds GetTimeMS() override {
-        base_time_point current = base_timer::now();
-        auto elapsed = current - start_time;
-        return std::chrono::duration_cast<std::chrono::milliseconds>(elapsed);
+        return std::chrono::duration_cast<std::chrono::milliseconds>(GetTimeNS());
     }
 
     u64 GetClockCycles() override {
-        std::chrono::nanoseconds time_now = GetTimeNS();
-        const u128 temporary =
-            Common::Multiply64Into128(time_now.count(), emulated_clock_frequency);
-        return Common::Divide128On32(temporary, 1000000000).first;
+        const u128 temp = Common::Multiply64Into128(GetTimeNS().count(), emulated_clock_frequency);
+        return Common::Divide128On32(temp, NS_RATIO).first;
     }
 
     u64 GetCPUCycles() override {
-        std::chrono::nanoseconds time_now = GetTimeNS();
-        const u128 temporary = Common::Multiply64Into128(time_now.count(), emulated_cpu_frequency);
-        return Common::Divide128On32(temporary, 1000000000).first;
+        const u128 temp = Common::Multiply64Into128(GetTimeNS().count(), emulated_cpu_frequency);
+        return Common::Divide128On32(temp, NS_RATIO).first;
     }
 
     void Pause([[maybe_unused]] bool is_paused) override {
@@ -57,7 +45,7 @@ public:
     }
 
 private:
-    base_time_point start_time;
+    SteadyClock::time_point start_time;
 };
 
 #ifdef ARCHITECTURE_x86_64
@@ -93,4 +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
diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h
index 828a523a8..157ec5eae 100644
--- a/src/common/wall_clock.h
+++ b/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
diff --git a/src/common/windows/timer_resolution.cpp b/src/common/windows/timer_resolution.cpp
new file mode 100644
index 000000000..29c6e5c7e
--- /dev/null
+++ b/src/common/windows/timer_resolution.cpp
@@ -0,0 +1,109 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <windows.h>
+
+#include "common/windows/timer_resolution.h"
+
+extern "C" {
+// http://undocumented.ntinternals.net/index.html?page=UserMode%2FUndocumented%20Functions%2FTime%2FNtQueryTimerResolution.html
+NTSYSAPI LONG NTAPI NtQueryTimerResolution(PULONG MinimumResolution, PULONG MaximumResolution,
+                                           PULONG CurrentResolution);
+
+// http://undocumented.ntinternals.net/index.html?page=UserMode%2FUndocumented%20Functions%2FTime%2FNtSetTimerResolution.html
+NTSYSAPI LONG NTAPI NtSetTimerResolution(ULONG DesiredResolution, BOOLEAN SetResolution,
+                                         PULONG CurrentResolution);
+
+// http://undocumented.ntinternals.net/index.html?page=UserMode%2FUndocumented%20Functions%2FNT%20Objects%2FThread%2FNtDelayExecution.html
+NTSYSAPI LONG NTAPI NtDelayExecution(BOOLEAN Alertable, PLARGE_INTEGER DelayInterval);
+}
+
+// Defines for compatibility with older Windows 10 SDKs.
+
+#ifndef PROCESS_POWER_THROTTLING_EXECUTION_SPEED
+#define PROCESS_POWER_THROTTLING_EXECUTION_SPEED 0x1
+#endif
+#ifndef PROCESS_POWER_THROTTLING_IGNORE_TIMER_RESOLUTION
+#define PROCESS_POWER_THROTTLING_IGNORE_TIMER_RESOLUTION 0x4
+#endif
+
+namespace Common::Windows {
+
+namespace {
+
+using namespace std::chrono;
+
+constexpr nanoseconds ToNS(ULONG hundred_ns) {
+    return nanoseconds{hundred_ns * 100};
+}
+
+constexpr ULONG ToHundredNS(nanoseconds ns) {
+    return static_cast<ULONG>(ns.count()) / 100;
+}
+
+struct TimerResolution {
+    std::chrono::nanoseconds minimum;
+    std::chrono::nanoseconds maximum;
+    std::chrono::nanoseconds current;
+};
+
+TimerResolution GetTimerResolution() {
+    ULONG MinimumTimerResolution;
+    ULONG MaximumTimerResolution;
+    ULONG CurrentTimerResolution;
+    NtQueryTimerResolution(&MinimumTimerResolution, &MaximumTimerResolution,
+                           &CurrentTimerResolution);
+    return {
+        .minimum{ToNS(MinimumTimerResolution)},
+        .maximum{ToNS(MaximumTimerResolution)},
+        .current{ToNS(CurrentTimerResolution)},
+    };
+}
+
+void SetHighQoS() {
+    // https://learn.microsoft.com/en-us/windows/win32/procthread/quality-of-service
+    PROCESS_POWER_THROTTLING_STATE PowerThrottling{
+        .Version{PROCESS_POWER_THROTTLING_CURRENT_VERSION},
+        .ControlMask{PROCESS_POWER_THROTTLING_EXECUTION_SPEED |
+                     PROCESS_POWER_THROTTLING_IGNORE_TIMER_RESOLUTION},
+        .StateMask{},
+    };
+    SetProcessInformation(GetCurrentProcess(), ProcessPowerThrottling, &PowerThrottling,
+                          sizeof(PROCESS_POWER_THROTTLING_STATE));
+}
+
+} // Anonymous namespace
+
+nanoseconds GetMinimumTimerResolution() {
+    return GetTimerResolution().minimum;
+}
+
+nanoseconds GetMaximumTimerResolution() {
+    return GetTimerResolution().maximum;
+}
+
+nanoseconds GetCurrentTimerResolution() {
+    return GetTimerResolution().current;
+}
+
+nanoseconds SetCurrentTimerResolution(nanoseconds timer_resolution) {
+    // Set the timer resolution, and return the current timer resolution.
+    const auto DesiredTimerResolution = ToHundredNS(timer_resolution);
+    ULONG CurrentTimerResolution;
+    NtSetTimerResolution(DesiredTimerResolution, TRUE, &CurrentTimerResolution);
+    return ToNS(CurrentTimerResolution);
+}
+
+nanoseconds SetCurrentTimerResolutionToMaximum() {
+    SetHighQoS();
+    return SetCurrentTimerResolution(GetMaximumTimerResolution());
+}
+
+void SleepForOneTick() {
+    LARGE_INTEGER DelayInterval{
+        .QuadPart{-1},
+    };
+    NtDelayExecution(FALSE, &DelayInterval);
+}
+
+} // namespace Common::Windows
diff --git a/src/common/windows/timer_resolution.h b/src/common/windows/timer_resolution.h
new file mode 100644
index 000000000..e1e50a62d
--- /dev/null
+++ b/src/common/windows/timer_resolution.h
@@ -0,0 +1,38 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <chrono>
+
+namespace Common::Windows {
+
+/// Returns the minimum (least precise) supported timer resolution in nanoseconds.
+std::chrono::nanoseconds GetMinimumTimerResolution();
+
+/// Returns the maximum (most precise) supported timer resolution in nanoseconds.
+std::chrono::nanoseconds GetMaximumTimerResolution();
+
+/// Returns the current timer resolution in nanoseconds.
+std::chrono::nanoseconds GetCurrentTimerResolution();
+
+/**
+ * Sets the current timer resolution.
+ *
+ * @param timer_resolution Timer resolution in nanoseconds.
+ *
+ * @returns The current timer resolution.
+ */
+std::chrono::nanoseconds SetCurrentTimerResolution(std::chrono::nanoseconds timer_resolution);
+
+/**
+ * Sets the current timer resolution to the maximum supported timer resolution.
+ *
+ * @returns The current timer resolution.
+ */
+std::chrono::nanoseconds SetCurrentTimerResolutionToMaximum();
+
+/// Sleep for one tick of the current timer resolution.
+void SleepForOneTick();
+
+} // namespace Common::Windows
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index 8b08332ab..bc1a973b0 100644
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -6,6 +6,7 @@
 #include <thread>
 
 #include "common/atomic_ops.h"
+#include "common/steady_clock.h"
 #include "common/uint128.h"
 #include "common/x64/native_clock.h"
 
@@ -39,6 +40,12 @@ static u64 FencedRDTSC() {
 }
 #endif
 
+template <u64 Nearest>
+static u64 RoundToNearest(u64 value) {
+    const auto mod = value % Nearest;
+    return mod >= (Nearest / 2) ? (value - mod + Nearest) : (value - mod);
+}
+
 u64 EstimateRDTSCFrequency() {
     // Discard the first result measuring the rdtsc.
     FencedRDTSC();
@@ -46,18 +53,18 @@ u64 EstimateRDTSCFrequency() {
     FencedRDTSC();
 
     // Get the current time.
-    const auto start_time = std::chrono::steady_clock::now();
+    const auto start_time = Common::SteadyClock::Now();
     const u64 tsc_start = FencedRDTSC();
-    // Wait for 200 milliseconds.
-    std::this_thread::sleep_for(std::chrono::milliseconds{200});
-    const auto end_time = std::chrono::steady_clock::now();
+    // Wait for 250 milliseconds.
+    std::this_thread::sleep_for(std::chrono::milliseconds{250});
+    const auto end_time = Common::SteadyClock::Now();
     const u64 tsc_end = FencedRDTSC();
     // Calculate differences.
     const u64 timer_diff = static_cast<u64>(
         std::chrono::duration_cast<std::chrono::nanoseconds>(end_time - start_time).count());
     const u64 tsc_diff = tsc_end - tsc_start;
     const u64 tsc_freq = MultiplyAndDivide64(tsc_diff, 1000000000ULL, timer_diff);
-    return tsc_freq;
+    return RoundToNearest<1000>(tsc_freq);
 }
 
 namespace X64 {