11 files changed, 224 insertions, 274 deletions

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index efc4a9fe9..3adf13a3f 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -172,6 +172,8 @@ if(ARCHITECTURE_x86_64)
             x64/cpu_wait.h
             x64/native_clock.cpp
             x64/native_clock.h
+            x64/rdtsc.cpp
+            x64/rdtsc.h
             x64/xbyak_abi.h
             x64/xbyak_util.h
     )
diff --git a/src/common/settings.h b/src/common/settings.h
index 9682281b0..3aedf3850 100644
--- a/src/common/settings.h
+++ b/src/common/settings.h
@@ -483,6 +483,7 @@ struct Values {
         AstcRecompression::Uncompressed, AstcRecompression::Uncompressed, AstcRecompression::Bc3,
         "astc_recompression"};
     SwitchableSetting<bool> use_video_framerate{false, "use_video_framerate"};
+    SwitchableSetting<bool> barrier_feedback_loops{true, "barrier_feedback_loops"};
 
     SwitchableSetting<u8> bg_red{0, "bg_red"};
     SwitchableSetting<u8> bg_green{0, "bg_green"};
diff --git a/src/common/steady_clock.cpp b/src/common/steady_clock.cpp
index 782859196..9415eed29 100644
--- a/src/common/steady_clock.cpp
+++ b/src/common/steady_clock.cpp
@@ -28,13 +28,12 @@ static s64 GetSystemTimeNS() {
     // GetSystemTimePreciseAsFileTime returns the file time in 100ns units.
     static constexpr s64 Multiplier = 100;
     // Convert Windows epoch to Unix epoch.
-    static constexpr s64 WindowsEpochToUnixEpochNS = 0x19DB1DED53E8000LL;
+    static constexpr s64 WindowsEpochToUnixEpoch = 0x19DB1DED53E8000LL;
 
     FILETIME filetime;
     GetSystemTimePreciseAsFileTime(&filetime);
     return Multiplier * ((static_cast<s64>(filetime.dwHighDateTime) << 32) +
-                         static_cast<s64>(filetime.dwLowDateTime)) -
-           WindowsEpochToUnixEpochNS;
+                         static_cast<s64>(filetime.dwLowDateTime) - WindowsEpochToUnixEpoch);
 }
 #endif
 
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
index 817e71d52..dc0dcbd68 100644
--- a/src/common/wall_clock.cpp
+++ b/src/common/wall_clock.cpp
@@ -2,88 +2,75 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include "common/steady_clock.h"
-#include "common/uint128.h"
 #include "common/wall_clock.h"
 
 #ifdef ARCHITECTURE_x86_64
 #include "common/x64/cpu_detect.h"
 #include "common/x64/native_clock.h"
+#include "common/x64/rdtsc.h"
 #endif
 
 namespace Common {
 
 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{SteadyClock::Now()} {}
+    explicit StandardWallClock() : start_time{SteadyClock::Now()} {}
 
-    std::chrono::nanoseconds GetTimeNS() override {
+    std::chrono::nanoseconds GetTimeNS() const override {
         return SteadyClock::Now() - start_time;
     }
 
-    std::chrono::microseconds GetTimeUS() override {
-        return std::chrono::duration_cast<std::chrono::microseconds>(GetTimeNS());
+    std::chrono::microseconds GetTimeUS() const override {
+        return static_cast<std::chrono::microseconds>(GetHostTicksElapsed() / NsToUsRatio::den);
     }
 
-    std::chrono::milliseconds GetTimeMS() override {
-        return std::chrono::duration_cast<std::chrono::milliseconds>(GetTimeNS());
+    std::chrono::milliseconds GetTimeMS() const override {
+        return static_cast<std::chrono::milliseconds>(GetHostTicksElapsed() / NsToMsRatio::den);
     }
 
-    u64 GetClockCycles() override {
-        const u128 temp = Common::Multiply64Into128(GetTimeNS().count(), emulated_clock_frequency);
-        return Common::Divide128On32(temp, NS_RATIO).first;
+    u64 GetCNTPCT() const override {
+        return GetHostTicksElapsed() * NsToCNTPCTRatio::num / NsToCNTPCTRatio::den;
     }
 
-    u64 GetCPUCycles() override {
-        const u128 temp = Common::Multiply64Into128(GetTimeNS().count(), emulated_cpu_frequency);
-        return Common::Divide128On32(temp, NS_RATIO).first;
+    u64 GetGPUTick() const override {
+        return GetHostTicksElapsed() * NsToGPUTickRatio::num / NsToGPUTickRatio::den;
     }
 
-    void Pause([[maybe_unused]] bool is_paused) override {
-        // Do nothing in this clock type.
+    u64 GetHostTicksNow() const override {
+        return static_cast<u64>(SteadyClock::Now().time_since_epoch().count());
+    }
+
+    u64 GetHostTicksElapsed() const override {
+        return static_cast<u64>(GetTimeNS().count());
+    }
+
+    bool IsNative() const override {
+        return false;
     }
 
 private:
     SteadyClock::time_point start_time;
 };
 
+std::unique_ptr<WallClock> CreateOptimalClock() {
 #ifdef ARCHITECTURE_x86_64
-
-std::unique_ptr<WallClock> CreateBestMatchingClock(u64 emulated_cpu_frequency,
-                                                   u64 emulated_clock_frequency) {
     const auto& caps = GetCPUCaps();
-    u64 rtsc_frequency = 0;
-    if (caps.invariant_tsc) {
-        rtsc_frequency = caps.tsc_frequency ? caps.tsc_frequency : EstimateRDTSCFrequency();
-    }
 
-    // Fallback to StandardWallClock if the hardware TSC does not have the precision greater than:
-    // - A nanosecond
-    // - The emulated CPU frequency
-    // - The emulated clock counter frequency (CNTFRQ)
-    if (rtsc_frequency <= WallClock::NS_RATIO || rtsc_frequency <= emulated_cpu_frequency ||
-        rtsc_frequency <= emulated_clock_frequency) {
-        return std::make_unique<StandardWallClock>(emulated_cpu_frequency,
-                                                   emulated_clock_frequency);
+    if (caps.invariant_tsc && caps.tsc_frequency >= WallClock::GPUTickFreq) {
+        return std::make_unique<X64::NativeClock>(caps.tsc_frequency);
     } else {
-        return std::make_unique<X64::NativeClock>(emulated_cpu_frequency, emulated_clock_frequency,
-                                                  rtsc_frequency);
+        // Fallback to StandardWallClock if the hardware TSC
+        // - Is not invariant
+        // - Is not more precise than GPUTickFreq
+        return std::make_unique<StandardWallClock>();
     }
-}
-
 #else
-
-std::unique_ptr<WallClock> CreateBestMatchingClock(u64 emulated_cpu_frequency,
-                                                   u64 emulated_clock_frequency) {
-    return std::make_unique<StandardWallClock>(emulated_cpu_frequency, emulated_clock_frequency);
-}
-
+    return std::make_unique<StandardWallClock>();
 #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);
+std::unique_ptr<WallClock> CreateStandardWallClock() {
+    return std::make_unique<StandardWallClock>();
 }
 
 } // namespace Common
diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h
index 157ec5eae..f45d3d8c5 100644
--- a/src/common/wall_clock.h
+++ b/src/common/wall_clock.h
@@ -5,6 +5,7 @@
 
 #include <chrono>
 #include <memory>
+#include <ratio>
 
 #include "common/common_types.h"
 
@@ -12,50 +13,82 @@ namespace Common {
 
 class WallClock {
 public:
-    static constexpr u64 NS_RATIO = 1'000'000'000;
-    static constexpr u64 US_RATIO = 1'000'000;
-    static constexpr u64 MS_RATIO = 1'000;
+    static constexpr u64 CNTFRQ = 19'200'000;         // CNTPCT_EL0 Frequency = 19.2 MHz
+    static constexpr u64 GPUTickFreq = 614'400'000;   // GM20B GPU Tick Frequency = 614.4 MHz
+    static constexpr u64 CPUTickFreq = 1'020'000'000; // T210/4 A57 CPU Tick Frequency = 1020.0 MHz
 
     virtual ~WallClock() = default;
 
-    /// Returns current wall time in nanoseconds
-    [[nodiscard]] virtual std::chrono::nanoseconds GetTimeNS() = 0;
+    /// @returns The time in nanoseconds since the construction of this clock.
+    virtual std::chrono::nanoseconds GetTimeNS() const = 0;
 
-    /// Returns current wall time in microseconds
-    [[nodiscard]] virtual std::chrono::microseconds GetTimeUS() = 0;
+    /// @returns The time in microseconds since the construction of this clock.
+    virtual std::chrono::microseconds GetTimeUS() const = 0;
 
-    /// Returns current wall time in milliseconds
-    [[nodiscard]] virtual std::chrono::milliseconds GetTimeMS() = 0;
+    /// @returns The time in milliseconds since the construction of this clock.
+    virtual std::chrono::milliseconds GetTimeMS() const = 0;
 
-    /// Returns current wall time in emulated clock cycles
-    [[nodiscard]] virtual u64 GetClockCycles() = 0;
+    /// @returns The guest CNTPCT ticks since the construction of this clock.
+    virtual u64 GetCNTPCT() const = 0;
 
-    /// Returns current wall time in emulated cpu cycles
-    [[nodiscard]] virtual u64 GetCPUCycles() = 0;
+    /// @returns The guest GPU ticks since the construction of this clock.
+    virtual u64 GetGPUTick() const = 0;
 
-    virtual void Pause(bool is_paused) = 0;
+    /// @returns The raw host timer ticks since an indeterminate epoch.
+    virtual u64 GetHostTicksNow() const = 0;
 
-    /// Tells if the wall clock, uses the host CPU's hardware clock
-    [[nodiscard]] bool IsNative() const {
-        return is_native;
+    /// @returns The raw host timer ticks since the construction of this clock.
+    virtual u64 GetHostTicksElapsed() const = 0;
+
+    /// @returns Whether the clock directly uses the host's hardware clock.
+    virtual bool IsNative() const = 0;
+
+    static inline u64 NSToCNTPCT(u64 ns) {
+        return ns * NsToCNTPCTRatio::num / NsToCNTPCTRatio::den;
+    }
+
+    static inline u64 NSToGPUTick(u64 ns) {
+        return ns * NsToGPUTickRatio::num / NsToGPUTickRatio::den;
+    }
+
+    // Cycle Timing
+
+    static inline u64 CPUTickToNS(u64 cpu_tick) {
+        return cpu_tick * CPUTickToNsRatio::num / CPUTickToNsRatio::den;
+    }
+
+    static inline u64 CPUTickToUS(u64 cpu_tick) {
+        return cpu_tick * CPUTickToUsRatio::num / CPUTickToUsRatio::den;
+    }
+
+    static inline u64 CPUTickToCNTPCT(u64 cpu_tick) {
+        return cpu_tick * CPUTickToCNTPCTRatio::num / CPUTickToCNTPCTRatio::den;
+    }
+
+    static inline u64 CPUTickToGPUTick(u64 cpu_tick) {
+        return cpu_tick * CPUTickToGPUTickRatio::num / CPUTickToGPUTickRatio::den;
     }
 
 protected:
-    explicit WallClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequency_, bool is_native_)
-        : emulated_cpu_frequency{emulated_cpu_frequency_},
-          emulated_clock_frequency{emulated_clock_frequency_}, is_native{is_native_} {}
+    using NsRatio = std::nano;
+    using UsRatio = std::micro;
+    using MsRatio = std::milli;
+
+    using NsToUsRatio = std::ratio_divide<std::nano, std::micro>;
+    using NsToMsRatio = std::ratio_divide<std::nano, std::milli>;
+    using NsToCNTPCTRatio = std::ratio<CNTFRQ, std::nano::den>;
+    using NsToGPUTickRatio = std::ratio<GPUTickFreq, std::nano::den>;
 
-    u64 emulated_cpu_frequency;
-    u64 emulated_clock_frequency;
+    // Cycle Timing
 
-private:
-    bool is_native;
+    using CPUTickToNsRatio = std::ratio<std::nano::den, CPUTickFreq>;
+    using CPUTickToUsRatio = std::ratio<std::micro::den, CPUTickFreq>;
+    using CPUTickToCNTPCTRatio = std::ratio<CNTFRQ, CPUTickFreq>;
+    using CPUTickToGPUTickRatio = std::ratio<GPUTickFreq, CPUTickFreq>;
 };
 
-[[nodiscard]] std::unique_ptr<WallClock> CreateBestMatchingClock(u64 emulated_cpu_frequency,
-                                                                 u64 emulated_clock_frequency);
+std::unique_ptr<WallClock> CreateOptimalClock();
 
-[[nodiscard]] std::unique_ptr<WallClock> CreateStandardWallClock(u64 emulated_cpu_frequency,
-                                                                 u64 emulated_clock_frequency);
+std::unique_ptr<WallClock> CreateStandardWallClock();
 
 } // namespace Common
diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp
index 72ed6e96c..c998b1197 100644
--- a/src/common/x64/cpu_detect.cpp
+++ b/src/common/x64/cpu_detect.cpp
@@ -14,6 +14,7 @@
 #include "common/common_types.h"
 #include "common/logging/log.h"
 #include "common/x64/cpu_detect.h"
+#include "common/x64/rdtsc.h"
 
 #ifdef _WIN32
 #include <windows.h>
@@ -187,6 +188,8 @@ static CPUCaps Detect() {
             caps.tsc_frequency = static_cast<u64>(caps.crystal_frequency) *
                                  caps.tsc_crystal_ratio_numerator /
                                  caps.tsc_crystal_ratio_denominator;
+        } else {
+            caps.tsc_frequency = X64::EstimateRDTSCFrequency();
         }
     }
 
diff --git a/src/common/x64/cpu_wait.cpp b/src/common/x64/cpu_wait.cpp
index cfeef6a3d..c53dd4945 100644
--- a/src/common/x64/cpu_wait.cpp
+++ b/src/common/x64/cpu_wait.cpp
@@ -9,19 +9,11 @@
 
 #include "common/x64/cpu_detect.h"
 #include "common/x64/cpu_wait.h"
+#include "common/x64/rdtsc.h"
 
 namespace Common::X64 {
 
 #ifdef _MSC_VER
-__forceinline static u64 FencedRDTSC() {
-    _mm_lfence();
-    _ReadWriteBarrier();
-    const u64 result = __rdtsc();
-    _mm_lfence();
-    _ReadWriteBarrier();
-    return result;
-}
-
 __forceinline static void TPAUSE() {
     // 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
     // For reference:
@@ -32,16 +24,6 @@ __forceinline static void TPAUSE() {
     _tpause(0, FencedRDTSC() + PauseCycles);
 }
 #else
-static u64 FencedRDTSC() {
-    u64 eax;
-    u64 edx;
-    asm volatile("lfence\n\t"
-                 "rdtsc\n\t"
-                 "lfence\n\t"
-                 : "=a"(eax), "=d"(edx));
-    return (edx << 32) | eax;
-}
-
 static void TPAUSE() {
     // 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
     // For reference:
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index 277b00662..7d2a26bd9 100644
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -1,164 +1,50 @@
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <array>
-#include <chrono>
-#include <thread>
-
-#include "common/atomic_ops.h"
-#include "common/steady_clock.h"
 #include "common/uint128.h"
 #include "common/x64/native_clock.h"
+#include "common/x64/rdtsc.h"
 
-#ifdef _MSC_VER
-#include <intrin.h>
-#endif
-
-namespace Common {
+namespace Common::X64 {
 
-#ifdef _MSC_VER
-__forceinline static u64 FencedRDTSC() {
-    _mm_lfence();
-    _ReadWriteBarrier();
-    const u64 result = __rdtsc();
-    _mm_lfence();
-    _ReadWriteBarrier();
-    return result;
-}
-#else
-static u64 FencedRDTSC() {
-    u64 eax;
-    u64 edx;
-    asm volatile("lfence\n\t"
-                 "rdtsc\n\t"
-                 "lfence\n\t"
-                 : "=a"(eax), "=d"(edx));
-    return (edx << 32) | eax;
-}
-#endif
+NativeClock::NativeClock(u64 rdtsc_frequency_)
+    : start_ticks{FencedRDTSC()}, rdtsc_frequency{rdtsc_frequency_},
+      ns_rdtsc_factor{GetFixedPoint64Factor(NsRatio::den, rdtsc_frequency)},
+      us_rdtsc_factor{GetFixedPoint64Factor(UsRatio::den, rdtsc_frequency)},
+      ms_rdtsc_factor{GetFixedPoint64Factor(MsRatio::den, rdtsc_frequency)},
+      cntpct_rdtsc_factor{GetFixedPoint64Factor(CNTFRQ, rdtsc_frequency)},
+      gputick_rdtsc_factor{GetFixedPoint64Factor(GPUTickFreq, rdtsc_frequency)} {}
 
-template <u64 Nearest>
-static u64 RoundToNearest(u64 value) {
-    const auto mod = value % Nearest;
-    return mod >= (Nearest / 2) ? (value - mod + Nearest) : (value - mod);
+std::chrono::nanoseconds NativeClock::GetTimeNS() const {
+    return std::chrono::nanoseconds{MultiplyHigh(GetHostTicksElapsed(), ns_rdtsc_factor)};
 }
 
-u64 EstimateRDTSCFrequency() {
-    // Discard the first result measuring the rdtsc.
-    FencedRDTSC();
-    std::this_thread::sleep_for(std::chrono::milliseconds{1});
-    FencedRDTSC();
-
-    // Get the current time.
-    const auto start_time = Common::RealTimeClock::Now();
-    const u64 tsc_start = FencedRDTSC();
-    // Wait for 250 milliseconds.
-    std::this_thread::sleep_for(std::chrono::milliseconds{250});
-    const auto end_time = Common::RealTimeClock::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 RoundToNearest<1000>(tsc_freq);
+std::chrono::microseconds NativeClock::GetTimeUS() const {
+    return std::chrono::microseconds{MultiplyHigh(GetHostTicksElapsed(), us_rdtsc_factor)};
 }
 
-namespace X64 {
-NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequency_,
-                         u64 rtsc_frequency_)
-    : WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, true), rtsc_frequency{
-                                                                               rtsc_frequency_} {
-    // Thread to re-adjust the RDTSC frequency after 10 seconds has elapsed.
-    time_sync_thread = std::jthread{[this](std::stop_token token) {
-        // Get the current time.
-        const auto start_time = Common::RealTimeClock::Now();
-        const u64 tsc_start = FencedRDTSC();
-        // Wait for 10 seconds.
-        if (!Common::StoppableTimedWait(token, std::chrono::seconds{10})) {
-            return;
-        }
-        const auto end_time = Common::RealTimeClock::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);
-        rtsc_frequency = tsc_freq;
-        CalculateAndSetFactors();
-    }};
-
-    time_point.inner.last_measure = FencedRDTSC();
-    time_point.inner.accumulated_ticks = 0U;
-    CalculateAndSetFactors();
+std::chrono::milliseconds NativeClock::GetTimeMS() const {
+    return std::chrono::milliseconds{MultiplyHigh(GetHostTicksElapsed(), ms_rdtsc_factor)};
 }
 
-u64 NativeClock::GetRTSC() {
-    TimePoint new_time_point{};
-    TimePoint current_time_point{};
-
-    current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
-    do {
-        const u64 current_measure = FencedRDTSC();
-        u64 diff = current_measure - current_time_point.inner.last_measure;
-        diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
-        new_time_point.inner.last_measure = current_measure > current_time_point.inner.last_measure
-                                                ? current_measure
-                                                : current_time_point.inner.last_measure;
-        new_time_point.inner.accumulated_ticks = current_time_point.inner.accumulated_ticks + diff;
-    } while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
-                                           current_time_point.pack, current_time_point.pack));
-    return new_time_point.inner.accumulated_ticks;
+u64 NativeClock::GetCNTPCT() const {
+    return MultiplyHigh(GetHostTicksElapsed(), cntpct_rdtsc_factor);
 }
 
-void NativeClock::Pause(bool is_paused) {
-    if (!is_paused) {
-        TimePoint current_time_point{};
-        TimePoint new_time_point{};
-
-        current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
-        do {
-            new_time_point.pack = current_time_point.pack;
-            new_time_point.inner.last_measure = FencedRDTSC();
-        } while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
-                                               current_time_point.pack, current_time_point.pack));
-    }
+u64 NativeClock::GetGPUTick() const {
+    return MultiplyHigh(GetHostTicksElapsed(), gputick_rdtsc_factor);
 }
 
-std::chrono::nanoseconds NativeClock::GetTimeNS() {
-    const u64 rtsc_value = GetRTSC();
-    return std::chrono::nanoseconds{MultiplyHigh(rtsc_value, ns_rtsc_factor)};
+u64 NativeClock::GetHostTicksNow() const {
+    return FencedRDTSC();
 }
 
-std::chrono::microseconds NativeClock::GetTimeUS() {
-    const u64 rtsc_value = GetRTSC();
-    return std::chrono::microseconds{MultiplyHigh(rtsc_value, us_rtsc_factor)};
+u64 NativeClock::GetHostTicksElapsed() const {
+    return FencedRDTSC() - start_ticks;
 }
 
-std::chrono::milliseconds NativeClock::GetTimeMS() {
-    const u64 rtsc_value = GetRTSC();
-    return std::chrono::milliseconds{MultiplyHigh(rtsc_value, ms_rtsc_factor)};
+bool NativeClock::IsNative() const {
+    return true;
 }
 
-u64 NativeClock::GetClockCycles() {
-    const u64 rtsc_value = GetRTSC();
-    return MultiplyHigh(rtsc_value, clock_rtsc_factor);
-}
-
-u64 NativeClock::GetCPUCycles() {
-    const u64 rtsc_value = GetRTSC();
-    return MultiplyHigh(rtsc_value, cpu_rtsc_factor);
-}
-
-void NativeClock::CalculateAndSetFactors() {
-    ns_rtsc_factor = GetFixedPoint64Factor(NS_RATIO, rtsc_frequency);
-    us_rtsc_factor = GetFixedPoint64Factor(US_RATIO, rtsc_frequency);
-    ms_rtsc_factor = GetFixedPoint64Factor(MS_RATIO, rtsc_frequency);
-    clock_rtsc_factor = GetFixedPoint64Factor(emulated_clock_frequency, rtsc_frequency);
-    cpu_rtsc_factor = GetFixedPoint64Factor(emulated_cpu_frequency, rtsc_frequency);
-}
-
-} // namespace X64
-
-} // namespace Common
+} // namespace Common::X64
diff --git a/src/common/x64/native_clock.h b/src/common/x64/native_clock.h
index 03ca291d8..334415eff 100644
--- a/src/common/x64/native_clock.h
+++ b/src/common/x64/native_clock.h
@@ -3,58 +3,39 @@
 
 #pragma once
 
-#include "common/polyfill_thread.h"
 #include "common/wall_clock.h"
 
-namespace Common {
+namespace Common::X64 {
 
-namespace X64 {
 class NativeClock final : public WallClock {
 public:
-    explicit NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequency_,
-                         u64 rtsc_frequency_);
+    explicit NativeClock(u64 rdtsc_frequency_);
 
-    std::chrono::nanoseconds GetTimeNS() override;
+    std::chrono::nanoseconds GetTimeNS() const override;
 
-    std::chrono::microseconds GetTimeUS() override;
+    std::chrono::microseconds GetTimeUS() const override;
 
-    std::chrono::milliseconds GetTimeMS() override;
+    std::chrono::milliseconds GetTimeMS() const override;
 
-    u64 GetClockCycles() override;
+    u64 GetCNTPCT() const override;
 
-    u64 GetCPUCycles() override;
+    u64 GetGPUTick() const override;
 
-    void Pause(bool is_paused) override;
+    u64 GetHostTicksNow() const override;
 
-private:
-    u64 GetRTSC();
-
-    void CalculateAndSetFactors();
-
-    union alignas(16) TimePoint {
-        TimePoint() : pack{} {}
-        u128 pack{};
-        struct Inner {
-            u64 last_measure{};
-            u64 accumulated_ticks{};
-        } inner;
-    };
-
-    TimePoint time_point;
+    u64 GetHostTicksElapsed() const override;
 
-    // factors
-    u64 clock_rtsc_factor{};
-    u64 cpu_rtsc_factor{};
-    u64 ns_rtsc_factor{};
-    u64 us_rtsc_factor{};
-    u64 ms_rtsc_factor{};
+    bool IsNative() const override;
 
-    u64 rtsc_frequency;
-
-    std::jthread time_sync_thread;
+private:
+    u64 start_ticks;
+    u64 rdtsc_frequency;
+
+    u64 ns_rdtsc_factor;
+    u64 us_rdtsc_factor;
+    u64 ms_rdtsc_factor;
+    u64 cntpct_rdtsc_factor;
+    u64 gputick_rdtsc_factor;
 };
-} // namespace X64
-
-u64 EstimateRDTSCFrequency();
 
-} // namespace Common
+} // namespace Common::X64
diff --git a/src/common/x64/rdtsc.cpp b/src/common/x64/rdtsc.cpp
new file mode 100644
index 000000000..9273274a3
--- /dev/null
+++ b/src/common/x64/rdtsc.cpp
@@ -0,0 +1,39 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <thread>
+
+#include "common/steady_clock.h"
+#include "common/uint128.h"
+#include "common/x64/rdtsc.h"
+
+namespace Common::X64 {
+
+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();
+    std::this_thread::sleep_for(std::chrono::milliseconds{1});
+    FencedRDTSC();
+
+    // Get the current time.
+    const auto start_time = RealTimeClock::Now();
+    const u64 tsc_start = FencedRDTSC();
+    // Wait for 100 milliseconds.
+    std::this_thread::sleep_for(std::chrono::milliseconds{100});
+    const auto end_time = RealTimeClock::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 RoundToNearest<100'000>(tsc_freq);
+}
+
+} // namespace Common::X64
diff --git a/src/common/x64/rdtsc.h b/src/common/x64/rdtsc.h
new file mode 100644
index 000000000..0ec4f52f9
--- /dev/null
+++ b/src/common/x64/rdtsc.h
@@ -0,0 +1,37 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
+#include "common/common_types.h"
+
+namespace Common::X64 {
+
+#ifdef _MSC_VER
+__forceinline static u64 FencedRDTSC() {
+    _mm_lfence();
+    _ReadWriteBarrier();
+    const u64 result = __rdtsc();
+    _mm_lfence();
+    _ReadWriteBarrier();
+    return result;
+}
+#else
+static inline u64 FencedRDTSC() {
+    u64 eax;
+    u64 edx;
+    asm volatile("lfence\n\t"
+                 "rdtsc\n\t"
+                 "lfence\n\t"
+                 : "=a"(eax), "=d"(edx));
+    return (edx << 32) | eax;
+}
+#endif
+
+u64 EstimateRDTSCFrequency();
+
+} // namespace Common::X64