Convert CPU caps into a global instead of MT safe global

src/common/thread.cpp

@@ -4,6 +4,7 @@
 // SPDX-FileCopyrightText: 2014 Citra Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <chrono>
 #include <string>
 #include <thread>
 
@@ -168,17 +169,19 @@ __attribute__((target("waitpkg,mwaitx")))
 #pragma GCC target("waitpkg")
 #pragma GCC target("mwaitx")
 #endif
-bool Event::WaitFor(const std::chrono::nanoseconds& time) {
+bool Event::WaitFor(const std::chrono::nanoseconds time) {
-    auto const& caps = Common::GetCPUCaps();
+    auto const start = Common::X64::FencedRDTSC();
+
+    auto const& caps = Common::g_cpu_caps;
     auto const ns_ratio = std::max<u64>(1, caps.base_frequency / 1'000);
-    auto const target_tsc = Common::X64::FencedRDTSC() + time.count() * ns_ratio;
+    auto const end = start + time.count() * ns_ratio;
     if (caps.monitorx) {
         while (true) {
             _mm_monitorx(reinterpret_cast<u64*>(std::addressof(is_set)), 0, 0);
-            if (!IsSet()) {
+            if (!is_set.load()) {
                 constexpr auto EnableWaitTimeFlag = 1U << 1;
                 constexpr auto RequestC1State = 0U;
-                _mm_mwaitx(EnableWaitTimeFlag, RequestC1State, target_tsc);
+                _mm_mwaitx(EnableWaitTimeFlag, RequestC1State, end);
                 if (!is_set.load())
                     return false;
             }
@@ -190,7 +193,7 @@ bool Event::WaitFor(const std::chrono::nanoseconds& time) {
         // #UD If CPUID.7.0:ECX.WAITPKG[bit 5]=0.
         while (true) {
             _umonitor(std::addressof(is_set));
-            if (!IsSet() && !_umwait(1, target_tsc))
+            if (!is_set.load() && !_umwait(1, end))
                 return false;
             bool expected = true;
             if (is_set.compare_exchange_weak(expected, false, std::memory_order_release))
@@ -198,9 +201,9 @@ bool Event::WaitFor(const std::chrono::nanoseconds& time) {
         }
     } else {
 #ifdef _WIN32
-        while (!IsSet() && _rdtsc() < target_tsc)
+        while (!is_set.load() && _rdtsc() < end)
             Common::Windows::SleepForOneTick();
-        if (IsSet())
+        if (is_set.load())
             Reset();
         return true;
 #else
@@ -213,11 +216,14 @@ bool Event::WaitFor(const std::chrono::nanoseconds& time) {
     }
 }
 #else
-bool Event::WaitFor(const std::chrono::nanoseconds& time) {
+bool Event::WaitFor(const std::chrono::nanoseconds time) {
 #ifdef _WIN32
-    while (!IsSet() && _rdtsc() < target_tsc)
+    s64 rem = s64(time.count()); //98 years
+    while (!is_set.load() && rem > 0) {
         Common::Windows::SleepForOneTick();
-    if (IsSet())
+        rem = s64(GetGlobalTimeNs().count()) - s64(time.count());
+    }
+    if (is_set.load())
         Reset();
     return true;
 #else

src/common/thread.h

@@ -35,10 +35,10 @@ public:
         is_set = false;
     }
 
-    bool WaitFor(const std::chrono::nanoseconds& time);
+    bool WaitFor(const std::chrono::nanoseconds time);
 
     template<class Clock, class Duration>
-    bool WaitUntil(const std::chrono::time_point<Clock, Duration>& time) {
+    bool WaitUntil(const std::chrono::time_point<Clock, Duration> time) {
         std::unique_lock lk{mutex};
         if (!condvar.wait_until(lk, time, [this] { return is_set.load(); }))
             return false;

src/common/wall_clock.cpp

@@ -183,7 +183,7 @@ bool WallClock::IsNative() const {
 
 WallClock CreateOptimalClock() noexcept {
 #if defined(ARCHITECTURE_x86_64)
-    auto const& caps = GetCPUCaps();
+    auto const& caps = Common::g_cpu_caps;
     return WallClock(caps.invariant_tsc && caps.tsc_frequency >= std::nano::den, caps.tsc_frequency);
 #elif defined(HAS_NCE)
     return WallClock(false, 1);

src/common/x64/cpu_detect.cpp

@@ -76,13 +76,12 @@ CPUCaps::Manufacturer CPUCaps::ParseManufacturer(std::string_view brand_string)
     return Manufacturer::Unknown;
 }
 
-// Detects the various CPU features
+/// @brief Detects the various CPU features
-static CPUCaps Detect() {
+const CPUCaps g_cpu_caps = [] {
     CPUCaps caps = {};
 
     // Assumes the CPU supports the CPUID instruction. Those that don't would likely not support
     // yuzu at all anyway
-
     int cpu_id[4];
 
     // Detect CPU's CPUID capabilities and grab manufacturer string
@@ -197,14 +196,8 @@ static CPUCaps Detect() {
         caps.max_frequency = cpu_id[1];
         caps.bus_frequency = cpu_id[2];
     }
-
     return caps;
-}
+}();
-
-const CPUCaps& GetCPUCaps() {
-    static CPUCaps caps = Detect();
-    return caps;
-}
 
 std::optional<int> GetProcessorCount() {
 #if defined(_WIN32)

src/common/x64/cpu_detect.h

@@ -70,11 +70,8 @@ struct CPUCaps {
     bool waitpkg : 1;
 };
 
-/**
+/// @brief Global cpu caps
- * Gets the supported capabilities of the host CPU
+extern const CPUCaps g_cpu_caps;
- * @return Reference to a CPUCaps struct with the detected host CPU capabilities
- */
-const CPUCaps& GetCPUCaps();
 
 /// Detects CPU core count
 std::optional<int> GetProcessorCount();

src/qt_common/qt_common.cpp

@@ -214,7 +214,7 @@ void Init(QWidget* root) {
     LOG_INFO(Frontend, "Eden Version: {}", yuzu_build_version);
     LogRuntimes();
 #ifdef ARCHITECTURE_x86_64
-    const auto& caps = Common::GetCPUCaps();
+    const auto& caps = Common::g_cpu_caps;
     std::string cpu_string = caps.cpu_string;
     if (caps.avx || caps.avx2 || caps.avx512f) {
         cpu_string += " | AVX";

src/video_core/host1x/vic.cpp

@@ -55,9 +55,8 @@ namespace Tegra::Host1x {
 namespace {
 
 static bool HasSSE41() {
-#if defined(ARCHITECTURE_x86_64)
+#ifdef ARCHITECTURE_x86_64
-    static bool has_sse41 = Common::GetCPUCaps().sse4_1;
+    return Common::g_cpu_caps.sse4_1;
-    return has_sse41;
 #else
     return false;
 #endif

src/video_core/shader_environment.cpp

@@ -254,11 +254,12 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
     static constexpr u64 SELF_BRANCH_A = 0xE2400FFFFF87000FULL;
     static constexpr u64 SELF_BRANCH_B = 0xE2400FFFFF07000FULL;
 
+    code.resize(MAXIMUM_SIZE / INST_SIZE);
+
     GPUVAddr guest_addr{program_base + start_address};
     size_t offset{0};
     size_t size{BLOCK_SIZE};
     while (size <= MAXIMUM_SIZE) {
-        code.resize(size / INST_SIZE);
         u64* const data = code.data() + offset / INST_SIZE;
         gpu_memory->ReadBlock(guest_addr, data, BLOCK_SIZE);
         for (size_t index = 0; index < BLOCK_SIZE; index += INST_SIZE) {

src/yuzu/main_window.cpp

@@ -3119,7 +3119,7 @@ void MainWindow::OnMenuReportCompatibility() {
         tr("Compatibility list reporting is currently disabled. Check back later!"));
 
     // #if defined(ARCHITECTURE_x86_64) && !defined(__APPLE__)
-    //     const auto& caps = Common::GetCPUCaps();
+    //     const auto& caps = g_cpu_caps;
     //     const bool has_fma = caps.fma;
     //     const auto processor_count = std::thread::hardware_concurrency();
     //     const bool has_4threads = processor_count == 0 || processor_count >= 4;