10 files changed, 378 insertions, 40 deletions

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 554d6e253..aacea0ab7 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -167,6 +167,8 @@ add_library(common STATIC
     vector_math.h
     virtual_buffer.cpp
     virtual_buffer.h
+    wall_clock.cpp
+    wall_clock.h
     web_result.h
     zstd_compression.cpp
     zstd_compression.h
@@ -177,6 +179,8 @@ if(ARCHITECTURE_x86_64)
         PRIVATE
             x64/cpu_detect.cpp
             x64/cpu_detect.h
+            x64/native_clock.cpp
+            x64/native_clock.h
             x64/xbyak_abi.h
             x64/xbyak_util.h
     )
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
new file mode 100644
index 000000000..eabbba9da
--- /dev/null
+++ b/src/common/wall_clock.cpp
@@ -0,0 +1,90 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#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"
+#endif
+
+namespace Common {
+
+using base_timer = std::chrono::steady_clock;
+using base_time_point = std::chrono::time_point<base_timer>;
+
+class StandardWallClock : public WallClock {
+public:
+    StandardWallClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency)
+        : WallClock(emulated_cpu_frequency, emulated_clock_frequency, false) {
+        start_time = base_timer::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);
+    }
+
+    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);
+    }
+
+    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);
+    }
+
+    u64 GetClockCycles() override {
+        std::chrono::nanoseconds time_now = GetTimeNS();
+        const u128 temporal = Common::Multiply64Into128(time_now.count(), emulated_clock_frequency);
+        return Common::Divide128On32(temporal, 1000000000).first;
+    }
+
+    u64 GetCPUCycles() override {
+        std::chrono::nanoseconds time_now = GetTimeNS();
+        const u128 temporal = Common::Multiply64Into128(time_now.count(), emulated_cpu_frequency);
+        return Common::Divide128On32(temporal, 1000000000).first;
+    }
+
+private:
+    base_time_point start_time;
+};
+
+#ifdef ARCHITECTURE_x86_64
+
+WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+    const auto& caps = GetCPUCaps();
+    u64 rtsc_frequency = 0;
+    if (caps.invariant_tsc) {
+        if (caps.base_frequency != 0) {
+            rtsc_frequency = static_cast<u64>(caps.base_frequency) * 1000000U;
+        }
+        if (rtsc_frequency == 0) {
+            rtsc_frequency = EstimateRDTSCFrequency();
+        }
+    }
+    if (rtsc_frequency == 0) {
+        return static_cast<WallClock*>(
+            new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency));
+    } else {
+        return static_cast<WallClock*>(
+            new X64::NativeClock(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency));
+    }
+}
+
+#else
+
+WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+    return static_cast<WallClock*>(
+        new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency));
+}
+
+#endif
+
+} // namespace Common
diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h
new file mode 100644
index 000000000..6f763d74b
--- /dev/null
+++ b/src/common/wall_clock.h
@@ -0,0 +1,40 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <chrono>
+
+#include "common/common_types.h"
+
+namespace Common {
+
+class WallClock {
+public:
+    virtual std::chrono::nanoseconds GetTimeNS() = 0;
+    virtual std::chrono::microseconds GetTimeUS() = 0;
+    virtual std::chrono::milliseconds GetTimeMS() = 0;
+    virtual u64 GetClockCycles() = 0;
+    virtual u64 GetCPUCycles() = 0;
+
+    /// Tells if the wall clock, uses the host CPU's hardware clock
+    bool IsNative() const {
+        return is_native;
+    }
+
+protected:
+    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} {}
+
+    u64 emulated_cpu_frequency;
+    u64 emulated_clock_frequency;
+
+private:
+    bool is_native;
+};
+
+WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency);
+
+} // namespace Common
diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp
index c9349a6b4..d767c544c 100644
--- a/src/common/x64/cpu_detect.cpp
+++ b/src/common/x64/cpu_detect.cpp
@@ -62,6 +62,17 @@ static CPUCaps Detect() {
     std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(int));
     std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(int));
     std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(int));
+    if (cpu_id[1] == 0x756e6547 && cpu_id[2] == 0x6c65746e && cpu_id[3] == 0x49656e69)
+        caps.manufacturer = Manufacturer::Intel;
+    else if (cpu_id[1] == 0x68747541 && cpu_id[2] == 0x444d4163 && cpu_id[3] == 0x69746e65)
+        caps.manufacturer = Manufacturer::AMD;
+    else if (cpu_id[1] == 0x6f677948 && cpu_id[2] == 0x656e6975 && cpu_id[3] == 0x6e65476e)
+        caps.manufacturer = Manufacturer::Hygon;
+    else
+        caps.manufacturer = Manufacturer::Unknown;
+
+    u32 family = {};
+    u32 model = {};
 
     __cpuid(cpu_id, 0x80000000);
 
@@ -73,6 +84,14 @@ static CPUCaps Detect() {
     // Detect family and other miscellaneous features
     if (max_std_fn >= 1) {
         __cpuid(cpu_id, 0x00000001);
+        family = (cpu_id[0] >> 8) & 0xf;
+        model = (cpu_id[0] >> 4) & 0xf;
+        if (family == 0xf) {
+            family += (cpu_id[0] >> 20) & 0xff;
+        }
+        if (family >= 6) {
+            model += ((cpu_id[0] >> 16) & 0xf) << 4;
+        }
 
         if ((cpu_id[3] >> 25) & 1)
             caps.sse = true;
@@ -130,6 +149,20 @@ static CPUCaps Detect() {
             caps.fma4 = true;
     }
 
+    if (max_ex_fn >= 0x80000007) {
+        __cpuid(cpu_id, 0x80000007);
+        if (cpu_id[3] & (1 << 8)) {
+            caps.invariant_tsc = true;
+        }
+    }
+
+    if (max_std_fn >= 0x16) {
+        __cpuid(cpu_id, 0x16);
+        caps.base_frequency = cpu_id[0];
+        caps.max_frequency = cpu_id[1];
+        caps.bus_frequency = cpu_id[2];
+    }
+
     return caps;
 }
 
diff --git a/src/common/x64/cpu_detect.h b/src/common/x64/cpu_detect.h
index 20f2ba234..f0676fa5e 100644
--- a/src/common/x64/cpu_detect.h
+++ b/src/common/x64/cpu_detect.h
@@ -6,8 +6,16 @@
 
 namespace Common {
 
+enum class Manufacturer : u32 {
+    Intel = 0,
+    AMD = 1,
+    Hygon = 2,
+    Unknown = 3,
+};
+
 /// x86/x64 CPU capabilities that may be detected by this module
 struct CPUCaps {
+    Manufacturer manufacturer;
     char cpu_string[0x21];
     char brand_string[0x41];
     bool sse;
@@ -24,6 +32,10 @@ struct CPUCaps {
     bool fma;
     bool fma4;
     bool aes;
+    bool invariant_tsc;
+    u32 base_frequency;
+    u32 max_frequency;
+    u32 bus_frequency;
 };
 
 /**
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
new file mode 100644
index 000000000..c799111fd
--- /dev/null
+++ b/src/common/x64/native_clock.cpp
@@ -0,0 +1,128 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <chrono>
+#include <thread>
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#else
+#include <x86intrin.h>
+#endif
+
+#include "common/x64/native_clock.h"
+
+namespace Common {
+
+#ifdef _MSC_VER
+
+namespace {
+
+struct uint128 {
+    u64 low;
+    u64 high;
+};
+
+u64 umuldiv64(u64 a, u64 b, u64 d) {
+    uint128 r{};
+    r.low = _umul128(a, b, &r.high);
+    u64 remainder;
+    return _udiv128(r.high, r.low, d, &remainder);
+}
+
+} // namespace
+
+#else
+
+namespace {
+
+u64 umuldiv64(u64 a, u64 b, u64 d) {
+    const u64 diva = a / d;
+    const u64 moda = a % d;
+    const u64 divb = b / d;
+    const u64 modb = b % d;
+    return diva * b + moda * divb + moda * modb / d;
+}
+
+} // namespace
+
+#endif
+
+u64 EstimateRDTSCFrequency() {
+    const auto milli_10 = std::chrono::milliseconds{10};
+    // get current time
+    _mm_mfence();
+    const u64 tscStart = __rdtsc();
+    const auto startTime = std::chrono::high_resolution_clock::now();
+    // wait roughly 3 seconds
+    while (true) {
+        auto milli = std::chrono::duration_cast<std::chrono::milliseconds>(
+            std::chrono::high_resolution_clock::now() - startTime);
+        if (milli.count() >= 3000)
+            break;
+        std::this_thread::sleep_for(milli_10);
+    }
+    const auto endTime = std::chrono::high_resolution_clock::now();
+    _mm_mfence();
+    const u64 tscEnd = __rdtsc();
+    // calculate difference
+    const u64 timer_diff =
+        std::chrono::duration_cast<std::chrono::nanoseconds>(endTime - startTime).count();
+    const u64 tsc_diff = tscEnd - tscStart;
+    const u64 tsc_freq = umuldiv64(tsc_diff, 1000000000ULL, timer_diff);
+    return tsc_freq;
+}
+
+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} {
+    _mm_mfence();
+    last_measure = __rdtsc();
+    accumulated_ticks = 0U;
+}
+
+u64 NativeClock::GetRTSC() {
+    rtsc_serialize.lock();
+    _mm_mfence();
+    const u64 current_measure = __rdtsc();
+    u64 diff = current_measure - last_measure;
+    diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
+    if (current_measure > last_measure) {
+        last_measure = current_measure;
+    }
+    accumulated_ticks += diff;
+    rtsc_serialize.unlock();
+    return accumulated_ticks;
+}
+
+std::chrono::nanoseconds NativeClock::GetTimeNS() {
+    const u64 rtsc_value = GetRTSC();
+    return std::chrono::nanoseconds{umuldiv64(rtsc_value, 1000000000, rtsc_frequency)};
+}
+
+std::chrono::microseconds NativeClock::GetTimeUS() {
+    const u64 rtsc_value = GetRTSC();
+    return std::chrono::microseconds{umuldiv64(rtsc_value, 1000000, rtsc_frequency)};
+}
+
+std::chrono::milliseconds NativeClock::GetTimeMS() {
+    const u64 rtsc_value = GetRTSC();
+    return std::chrono::milliseconds{umuldiv64(rtsc_value, 1000, rtsc_frequency)};
+}
+
+u64 NativeClock::GetClockCycles() {
+    const u64 rtsc_value = GetRTSC();
+    return umuldiv64(rtsc_value, emulated_clock_frequency, rtsc_frequency);
+}
+
+u64 NativeClock::GetCPUCycles() {
+    const u64 rtsc_value = GetRTSC();
+    return umuldiv64(rtsc_value, emulated_cpu_frequency, rtsc_frequency);
+}
+
+} // namespace X64
+
+} // namespace Common
diff --git a/src/common/x64/native_clock.h b/src/common/x64/native_clock.h
new file mode 100644
index 000000000..b58cf9f5a
--- /dev/null
+++ b/src/common/x64/native_clock.h
@@ -0,0 +1,41 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <optional>
+
+#include "common/spin_lock.h"
+#include "common/wall_clock.h"
+
+namespace Common {
+
+namespace X64 {
+class NativeClock : public WallClock {
+public:
+    NativeClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency, u64 rtsc_frequency);
+
+    std::chrono::nanoseconds GetTimeNS() override;
+
+    std::chrono::microseconds GetTimeUS() override;
+
+    std::chrono::milliseconds GetTimeMS() override;
+
+    u64 GetClockCycles() override;
+
+    u64 GetCPUCycles() override;
+
+private:
+    u64 GetRTSC();
+
+    SpinLock rtsc_serialize{};
+    u64 last_measure{};
+    u64 accumulated_ticks{};
+    u64 rtsc_frequency;
+};
+} // namespace X64
+
+u64 EstimateRDTSCFrequency();
+
+} // namespace Common
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index d9514b2c5..ef9977b76 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -35,7 +35,11 @@ struct CoreTiming::Event {
     }
 };
 
-CoreTiming::CoreTiming() = default;
+CoreTiming::CoreTiming() {
+    Common::WallClock* wall = Common::CreateBestMatchingClock(Core::Timing::BASE_CLOCK_RATE, Core::Timing::CNTFREQ);
+    clock = std::unique_ptr<Common::WallClock>(wall);
+}
+
 CoreTiming::~CoreTiming() = default;
 
 void CoreTiming::ThreadEntry(CoreTiming& instance) {
@@ -46,7 +50,6 @@ void CoreTiming::Initialize() {
     event_fifo_id = 0;
     const auto empty_timed_callback = [](u64, s64) {};
     ev_lost = CreateEvent("_lost_event", empty_timed_callback);
-    start_time = std::chrono::steady_clock::now();
     timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
 }
 
@@ -108,13 +111,11 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, u
 }
 
 u64 CoreTiming::GetCPUTicks() const {
-    std::chrono::nanoseconds time_now = GetGlobalTimeNs();
-    return Core::Timing::nsToCycles(time_now);
+    return clock->GetCPUCycles();
 }
 
 u64 CoreTiming::GetClockTicks() const {
-    std::chrono::nanoseconds time_now = GetGlobalTimeNs();
-    return Core::Timing::nsToClockCycles(time_now);
+    return clock->GetClockCycles();
 }
 
 void CoreTiming::ClearPendingEvents() {
@@ -174,15 +175,11 @@ void CoreTiming::Advance() {
 }
 
 std::chrono::nanoseconds CoreTiming::GetGlobalTimeNs() const {
-    sys_time_point current = std::chrono::steady_clock::now();
-    auto elapsed = current - start_time;
-    return std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed);
+    return clock->GetTimeNS();
 }
 
 std::chrono::microseconds CoreTiming::GetGlobalTimeUs() const {
-    sys_time_point current = std::chrono::steady_clock::now();
-    auto elapsed = current - start_time;
-    return std::chrono::duration_cast<std::chrono::microseconds>(elapsed);
+    return clock->GetTimeUS();
 }
 
 } // namespace Core::Timing
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index 1d053a7fa..f04a150ee 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -17,12 +17,12 @@
 #include "common/spin_lock.h"
 #include "common/thread.h"
 #include "common/threadsafe_queue.h"
+#include "common/wall_clock.h"
 
 namespace Core::HostTiming {
 
 /// A callback that may be scheduled for a particular core timing event.
 using TimedCallback = std::function<void(u64 userdata, s64 cycles_late)>;
-using sys_time_point = std::chrono::time_point<std::chrono::steady_clock>;
 
 /// Contains the characteristics of a particular event.
 struct EventType {
@@ -112,7 +112,7 @@ private:
     static void ThreadEntry(CoreTiming& instance);
     void Advance();
 
-    sys_time_point start_time;
+    std::unique_ptr<Common::WallClock> clock;
 
     u64 global_timer = 0;
 
diff --git a/src/tests/core/host_timing.cpp b/src/tests/core/host_timing.cpp
index ca9c8e50a..3d0532d02 100644
--- a/src/tests/core/host_timing.cpp
+++ b/src/tests/core/host_timing.cpp
@@ -17,7 +17,7 @@
 // Numbers are chosen randomly to make sure the correct one is given.
 static constexpr std::array<u64, 5> CB_IDS{{42, 144, 93, 1026, UINT64_C(0xFFFF7FFFF7FFFF)}};
 static constexpr int MAX_SLICE_LENGTH = 10000; // Copied from CoreTiming internals
-static constexpr std::array<u64, 5> calls_order{{2,0,1,4,3}};
+static constexpr std::array<u64, 5> calls_order{{2, 0, 1, 4, 3}};
 static std::array<s64, 5> delays{};
 
 static std::bitset<CB_IDS.size()> callbacks_ran_flags;
@@ -52,16 +52,11 @@ TEST_CASE("HostTiming[BasicOrder]", "[core]") {
     auto& core_timing = guard.core_timing;
     std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
     events.resize(5);
-    events[0] =
-        Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
-    events[1] =
-        Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
-    events[2] =
-        Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
-    events[3] =
-        Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
-    events[4] =
-        Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+    events[0] = Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
+    events[1] = Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
+    events[2] = Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
+    events[3] = Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
+    events[4] = Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
 
     expected_callback = 0;
 
@@ -70,14 +65,15 @@ TEST_CASE("HostTiming[BasicOrder]", "[core]") {
     u64 one_micro = 1000U;
     for (std::size_t i = 0; i < events.size(); i++) {
         u64 order = calls_order[i];
-        core_timing.ScheduleEvent(i*one_micro + 100U, events[order], CB_IDS[order]);
+        core_timing.ScheduleEvent(i * one_micro + 100U, events[order], CB_IDS[order]);
     }
     /// test pause
     REQUIRE(callbacks_ran_flags.none());
 
     core_timing.Pause(false); // No need to sync
 
-    while (core_timing.HasPendingEvents());
+    while (core_timing.HasPendingEvents())
+        ;
 
     REQUIRE(callbacks_ran_flags.all());
 
@@ -106,16 +102,11 @@ TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
     auto& core_timing = guard.core_timing;
     std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
     events.resize(5);
-    events[0] =
-        Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
-    events[1] =
-        Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
-    events[2] =
-        Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
-    events[3] =
-        Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
-    events[4] =
-        Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+    events[0] = Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
+    events[1] = Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
+    events[2] = Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
+    events[3] = Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
+    events[4] = Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
 
     core_timing.SyncPause(true);
     core_timing.SyncPause(false);
@@ -126,13 +117,14 @@ TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
     u64 one_micro = 1000U;
     for (std::size_t i = 0; i < events.size(); i++) {
         u64 order = calls_order[i];
-        core_timing.ScheduleEvent(i*one_micro + 100U, events[order], CB_IDS[order]);
+        core_timing.ScheduleEvent(i * one_micro + 100U, events[order], CB_IDS[order]);
     }
     u64 end = core_timing.GetGlobalTimeNs().count();
     const double scheduling_time = static_cast<double>(end - start);
     const double timer_time = static_cast<double>(TestTimerSpeed(core_timing));
 
-    while (core_timing.HasPendingEvents());
+    while (core_timing.HasPendingEvents())
+        ;
 
     REQUIRE(callbacks_ran_flags.all());
 
@@ -146,5 +138,6 @@ TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
     const double micro = scheduling_time / 1000.0f;
     const double mili = micro / 1000.0f;
     printf("HostTimer No Pausing Scheduling Time: %.3f %.6f\n", micro, mili);
-    printf("HostTimer No Pausing Timer Time: %.3f %.6f\n", timer_time / 1000.f, timer_time / 1000000.f);
+    printf("HostTimer No Pausing Timer Time: %.3f %.6f\n", timer_time / 1000.f,
+           timer_time / 1000000.f);
 }