7 files changed, 165 insertions, 211 deletions

diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp
index 72ed6e96c..780120a5b 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>
@@ -167,6 +168,7 @@ static CPUCaps Detect() {
         __cpuid(cpu_id, 0x80000001);
         caps.lzcnt = Common::Bit<5>(cpu_id[2]);
         caps.fma4 = Common::Bit<16>(cpu_id[2]);
+        caps.monitorx = Common::Bit<29>(cpu_id[2]);
     }
 
     if (max_ex_fn >= 0x80000007) {
@@ -187,6 +189,8 @@ 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_detect.h b/src/common/x64/cpu_detect.h
index 8253944d6..756459417 100644
--- a/src/common/x64/cpu_detect.h
+++ b/src/common/x64/cpu_detect.h
@@ -63,6 +63,7 @@ struct CPUCaps {
     bool gfni : 1;
     bool invariant_tsc : 1;
     bool lzcnt : 1;
+    bool monitorx : 1;
     bool movbe : 1;
     bool pclmulqdq : 1;
     bool popcnt : 1;
diff --git a/src/common/x64/cpu_wait.cpp b/src/common/x64/cpu_wait.cpp
index cfeef6a3d..41d385f59 100644
--- a/src/common/x64/cpu_wait.cpp
+++ b/src/common/x64/cpu_wait.cpp
@@ -9,58 +9,64 @@
 
 #include "common/x64/cpu_detect.h"
 #include "common/x64/cpu_wait.h"
+#include "common/x64/rdtsc.h"
 
 namespace Common::X64 {
 
+namespace {
+
+// 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
+// For reference:
+// At 1 GHz, 100K cycles is 100us
+// At 2 GHz, 100K cycles is 50us
+// At 4 GHz, 100K cycles is 25us
+constexpr auto PauseCycles = 100'000U;
+
+} // Anonymous namespace
+
 #ifdef _MSC_VER
-__forceinline static u64 FencedRDTSC() {
-    _mm_lfence();
-    _ReadWriteBarrier();
-    const u64 result = __rdtsc();
-    _mm_lfence();
-    _ReadWriteBarrier();
-    return result;
+__forceinline static void TPAUSE() {
+    static constexpr auto RequestC02State = 0U;
+    _tpause(RequestC02State, FencedRDTSC() + PauseCycles);
 }
 
-__forceinline static void TPAUSE() {
-    // 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
-    // For reference:
-    // At 1 GHz, 100K cycles is 100us
-    // At 2 GHz, 100K cycles is 50us
-    // At 4 GHz, 100K cycles is 25us
-    static constexpr auto PauseCycles = 100'000;
-    _tpause(0, FencedRDTSC() + PauseCycles);
+__forceinline static void MWAITX() {
+    static constexpr auto EnableWaitTimeFlag = 1U << 1;
+    static constexpr auto RequestC1State = 0U;
+
+    // monitor_var should be aligned to a cache line.
+    alignas(64) u64 monitor_var{};
+    _mm_monitorx(&monitor_var, 0, 0);
+    _mm_mwaitx(EnableWaitTimeFlag, RequestC1State, 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:
-    // At 1 GHz, 100K cycles is 100us
-    // At 2 GHz, 100K cycles is 50us
-    // At 4 GHz, 100K cycles is 25us
-    static constexpr auto PauseCycles = 100'000;
+    static constexpr auto RequestC02State = 0U;
     const auto tsc = FencedRDTSC() + PauseCycles;
     const auto eax = static_cast<u32>(tsc & 0xFFFFFFFF);
     const auto edx = static_cast<u32>(tsc >> 32);
-    asm volatile("tpause %0" : : "r"(0), "d"(edx), "a"(eax));
+    asm volatile("tpause %0" : : "r"(RequestC02State), "d"(edx), "a"(eax));
+}
+
+static void MWAITX() {
+    static constexpr auto EnableWaitTimeFlag = 1U << 1;
+    static constexpr auto RequestC1State = 0U;
+
+    // monitor_var should be aligned to a cache line.
+    alignas(64) u64 monitor_var{};
+    asm volatile("monitorx" : : "a"(&monitor_var), "c"(0), "d"(0));
+    asm volatile("mwaitx" : : "a"(RequestC1State), "b"(PauseCycles), "c"(EnableWaitTimeFlag));
 }
 #endif
 
 void MicroSleep() {
     static const bool has_waitpkg = GetCPUCaps().waitpkg;
+    static const bool has_monitorx = GetCPUCaps().monitorx;
 
     if (has_waitpkg) {
         TPAUSE();
+    } else if (has_monitorx) {
+        MWAITX();
     } else {
         std::this_thread::yield();
     }
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