common: Rework timekeeping with native RDTSC and port to linux

.gitmodules

@@ -33,11 +33,10 @@
 	path = third-party/toml11
 	url = https://github.com/ToruNiina/toml11
 	branch = master
-[submodule "third-party/vulkan"]
-	path = third-party/vulkan
-	url = https://github.com/shadps4/vulkan.git
-	branch = main
 [submodule "third-party/xxHash"]
 	path = third-party/xxHash
 	url = https://github.com/Cyan4973/xxHash.git
 	branch = dev
+[submodule "third-party/vulkan"]
+	path = third-party/vulkan
+	url = https://github.com/GPUCode/vulkan

CMakeLists.txt

@@ -30,8 +30,8 @@ endfunction()
 add_subdirectory(third-party)
 include_directories(src)
 
-set(LIBC_SOURCES src/core/hle/libraries/libc/Libc.cpp
+set(LIBC_SOURCES src/core/hle/libraries/libc/libc.cpp
-                 src/core/hle/libraries/libc/Libc.h
+                 src/core/hle/libraries/libc/libc.h
                  src/core/hle/libraries/libc/printf.h
                  src/core/hle/libraries/libc/va_ctx.h
                  src/core/hle/libraries/libc/libc_cxa.cpp
@@ -77,12 +77,15 @@ add_executable(shadps4
     src/common/fs_file.h
     src/common/log.cpp
     src/common/log.h
+    src/common/native_clock.cpp
+    src/common/native_clock.h
+    src/common/rdtsc.cpp
+    src/common/rdtsc.h
     src/common/singleton.h
     src/common/string_util.cpp
     src/common/string_util.h
-    src/common/timer.cpp
-    src/common/timer.h
     src/common/types.h
+    src/common/uint128.h
     src/common/version.h
     ${LIBC_SOURCES}
     ${USERSERVICE_SOURCES}
@@ -143,8 +146,6 @@ add_executable(shadps4
     src/core/PS4/HLE/Graphics/graphics_render.h
     src/core/PS4/GPU/tile_manager.cpp
     src/core/PS4/GPU/tile_manager.h
-    src/emuTimer.cpp
-    src/emuTimer.h
     src/core/hle/libraries/libkernel/time_management.cpp
     src/core/hle/libraries/libkernel/time_management.h
  "src/common/io_file.cpp" "src/common/io_file.h")
@@ -161,6 +162,8 @@ add_custom_command(TARGET shadps4 POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_if_different
      $<TARGET_FILE:SDL3-shared>
      $<TARGET_FILE_DIR:shadps4>)
-add_custom_command(TARGET shadps4 POST_BUILD
+if (WIN32)
+ add_custom_command(TARGET shadps4 POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_if_different
      "${PROJECT_SOURCE_DIR}/third-party/winpthread/bin/libwinpthread-1.dll" $<TARGET_FILE_DIR:shadps4>)
+endif()

src/common/log.cpp

@@ -109,7 +109,11 @@ int Init(bool use_stdout) {
     if (use_stdout) {
         sinks.push_back(std::make_shared<spdlog::sinks::stdout_color_sink_mt>());
     }
+#ifdef _WIN64
     sinks.push_back(std::make_shared<spdlog::sinks::basic_file_sink_mt>(L"shadps4.txt", true));
+#else
+    sinks.push_back(std::make_shared<spdlog::sinks::basic_file_sink_mt>("shadps4.txt", true));
+#endif
     spdlog::set_default_logger(std::make_shared<spdlog::logger>("shadps4 logger", begin(sinks), end(sinks)));
     auto f = std::make_unique<spdlog::pattern_formatter>("%^|%L|: %v%$", spdlog::pattern_time_type::local, std::string(""));  // disable eol
     spdlog::set_formatter(std::move(f));

src/common/native_clock.cpp

@@ -0,0 +1,43 @@
+// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/uint128.h"
+#include "common/native_clock.h"
+#include "common/rdtsc.h"
+#ifdef _WIN64
+#include <pthread_time.h>
+#else
+#include <time.h>
+#endif
+
+namespace Common {
+
+NativeClock::NativeClock()
+    : rdtsc_frequency{EstimateRDTSCFrequency()}, ns_rdtsc_factor{GetFixedPoint64Factor(std::nano::den,
+                                                                               rdtsc_frequency)},
+      us_rdtsc_factor{GetFixedPoint64Factor(std::micro::den, rdtsc_frequency)},
+      ms_rdtsc_factor{GetFixedPoint64Factor(std::milli::den, rdtsc_frequency)} {}
+
+u64 NativeClock::GetTimeNS() const {
+    return MultiplyHigh(GetUptime(), ns_rdtsc_factor);
+}
+
+u64 NativeClock::GetTimeUS() const {
+    return MultiplyHigh(GetUptime(), us_rdtsc_factor);
+}
+
+u64 NativeClock::GetTimeMS() const {
+    return MultiplyHigh(GetUptime(), ms_rdtsc_factor);
+}
+
+u64 NativeClock::GetUptime() const {
+    return FencedRDTSC();
+}
+
+u64 NativeClock::GetProcessTimeUS() const {
+    timespec ret;
+    clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ret);
+    return ret.tv_nsec / 1000 + ret.tv_sec * 1000000;
+}
+
+} // namespace Common::X64

src/common/native_clock.h

@@ -0,0 +1,32 @@
+// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <chrono>
+#include "common/types.h"
+
+namespace Common {
+
+class NativeClock final {
+public:
+    explicit NativeClock();
+
+    u64 GetTscFrequency() const {
+        return rdtsc_frequency;
+    }
+
+    u64 GetTimeNS() const;
+    u64 GetTimeUS() const;
+    u64 GetTimeMS() const;
+    u64 GetUptime() const;
+    u64 GetProcessTimeUS() const;
+
+private:
+    u64 rdtsc_frequency;
+    u64 ns_rdtsc_factor;
+    u64 us_rdtsc_factor;
+    u64 ms_rdtsc_factor;
+};
+
+} // namespace Common

src/common/rdtsc.cpp

@@ -0,0 +1,60 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <thread>
+#include "common/rdtsc.h"
+#include "common/uint128.h"
+
+#ifdef _WIN64
+#include <windows.h>
+#endif
+
+namespace Common {
+
+static constexpr size_t SecondToNanoseconds = 1000000000ULL;
+
+template <u64 Nearest>
+static u64 RoundToNearest(u64 value) {
+    const auto mod = value % Nearest;
+    return mod >= (Nearest / 2) ? (value - mod + Nearest) : (value - mod);
+}
+
+static u64 GetTimeNs() {
+#ifdef _WIN64
+    // GetSystemTimePreciseAsFileTime returns the file time in 100ns units.
+    static constexpr u64 Multiplier = 100;
+    // Convert Windows epoch to Unix epoch.
+    static constexpr u64 WindowsEpochToUnixEpoch = 0x19DB1DED53E8000LL;
+    FILETIME filetime;
+    GetSystemTimePreciseAsFileTime(&filetime);
+    return Multiplier * ((static_cast<u64>(filetime.dwHighDateTime) << 32) +
+                          static_cast<u64>(filetime.dwLowDateTime) - WindowsEpochToUnixEpoch);
+#elif defined(__APPLE__)
+    return clock_gettime_nsec_np(CLOCK_REALTIME);
+#else
+    timespec ts;
+    clock_gettime(CLOCK_REALTIME, &ts);
+    return ts.tv_sec * SecondToNanoseconds + ts.tv_nsec;
+#endif
+}
+
+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 = GetTimeNs();
+    const u64 tsc_start = FencedRDTSC();
+    // Wait for 100 milliseconds.
+    std::this_thread::sleep_for(std::chrono::milliseconds{100});
+    const auto end_time = GetTimeNs();
+    const u64 tsc_end = FencedRDTSC();
+    // Calculate differences.
+    const u64 tsc_diff = tsc_end - tsc_start;
+    const u64 tsc_freq = MultiplyAndDivide64(tsc_diff, 1000000000ULL, end_time - start_time);
+    return RoundToNearest<100'000>(tsc_freq);
+}
+
+} // namespace Common

src/common/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/types.h"
+
+namespace Common {
+
+#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

src/common/timer.cpp

@@ -1,108 +0,0 @@
-#include "common/timer.h"
-
-#ifdef _WIN64
-#include <windows.h>
-#endif
-
-namespace Common {
-
-Timer::Timer() {
-#ifdef _WIN64
-    LARGE_INTEGER f;
-    QueryPerformanceFrequency(&f);
-    m_Frequency = f.QuadPart;
-#else
-#error Unimplemented Timer constructor
-#endif
-}
-
-void Timer::Start() {
-#ifdef _WIN64
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    m_StartTime = c.QuadPart;
-#else
-#error Unimplemented Timer::Start()
-#endif
-    m_is_timer_paused = false;
-}
-
-void Timer::Pause() {
-#ifdef _WIN64
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    m_PauseTime = c.QuadPart;
-#else
-#error Unimplemented Timer::Pause()
-#endif
-    m_is_timer_paused = true;
-}
-
-void Timer::Resume() {
-    u64 current_time = 0;
-#ifdef _WIN64
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    current_time = c.QuadPart;
-#else
-#error Unimplemented Timer::Resume()
-#endif
-    m_StartTime += current_time - m_PauseTime;
-    m_is_timer_paused = false;
-}
-
-double Timer::GetTimeMsec() const {
-    if (m_is_timer_paused) {
-        return 1000.0 * (static_cast<double>(m_PauseTime - m_StartTime)) / static_cast<double>(m_Frequency);
-    }
-
-    u64 current_time = 0;
-#ifdef _WIN64
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    current_time = c.QuadPart;
-#else
-#error Unimplemented Timer::GetTimeMsec()
-#endif
-    return 1000.0 * (static_cast<double>(current_time - m_StartTime)) / static_cast<double>(m_Frequency);
-}
-
-double Timer::GetTimeSec() const {
-    if (m_is_timer_paused) {
-        return (static_cast<double>(m_PauseTime - m_StartTime)) / static_cast<double>(m_Frequency);
-    }
-
-    u64 current_time = 0;
-#ifdef _WIN64
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    current_time = c.QuadPart;
-#else
-#error Unimplemented Timer::GetTimeSec()
-#endif
-    return (static_cast<double>(current_time - m_StartTime)) / static_cast<double>(m_Frequency);
-}
-
-u64 Timer::GetTicks() const {
-    if (m_is_timer_paused) {
-        return (m_PauseTime - m_StartTime);
-    }
-
-    u64 current_time = 0;
-#ifdef _WIN64
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    current_time = c.QuadPart;
-#else
-#error Unimplemented Timer::GetTicks()
-#endif
-    return (current_time - m_StartTime);
-}
-
-u64 Timer::getQueryPerformanceCounter() {
-    LARGE_INTEGER c;
-    QueryPerformanceCounter(&c);
-    return c.QuadPart;
-}
-
-} // namespace Common

src/common/timer.h

@@ -1,43 +0,0 @@
-#pragma once
-
-#include "common/types.h"
-
-namespace Common {
-
-class Timer final {
-public:
-    Timer();
-    ~Timer() = default;
-
-    void Start();
-    void Pause();
-    void Resume();
-
-    bool IsPaused() const {
-        return m_is_timer_paused;
-    }
-
-    u64 GetFrequency() const {
-        return m_Frequency;
-    }
-
-    double GetTimeMsec() const;
-    double GetTimeSec() const;
-    u64 GetTicks() const;
-
-    [[nodiscard]] static u64 getQueryPerformanceCounter();
-
-public:
-    Timer(const Timer&) = delete;
-    Timer& operator=(const Timer&) = delete;
-    Timer(Timer&&) = delete;
-    Timer& operator=(Timer&&) = delete;
-
-private:
-    bool m_is_timer_paused = true;
-    u64 m_Frequency{};
-    u64 m_StartTime{};
-    u64 m_PauseTime{};
-};
-
-} // namespace Common

src/common/types.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <array>
 #include <cstdint>
 
 using s08 = std::int8_t;
@@ -15,9 +16,12 @@ using u64 = std::uint64_t;
 using f32 = float;
 using f64 = double;
 
+using u128 = std::array<std::uint64_t, 2>;
+static_assert(sizeof(u128) == 16, "u128 must be 128 bits wide");
+
 #define PS4_SYSV_ABI __attribute__((sysv_abi))
 
 // UDLs for memory size values
-constexpr u64 operator""_KB(u64 x) { return 1024ULL * x; }
+constexpr unsigned long long operator""_KB(unsigned long long x) { return 1024ULL * x; }
-constexpr u64 operator""_MB(u64 x) { return 1024_KB * x; }
+constexpr unsigned long long operator""_MB(unsigned long long x) { return 1024_KB * x; }
-constexpr u64 operator""_GB(u64 x) { return 1024_MB * x; }
+constexpr unsigned long long operator""_GB(unsigned long long x) { return 1024_MB * x; }

src/common/uint128.h

@@ -0,0 +1,115 @@
+// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <utility>
+
+#ifdef _MSC_VER
+#ifndef __clang__
+#define HAS_INTRINSICS
+#include <intrin.h>
+#pragma intrinsic(__umulh)
+#pragma intrinsic(_umul128)
+#pragma intrinsic(_udiv128)
+#else
+#endif
+#else
+#include <cstring>
+#endif
+
+#include "common/types.h"
+
+namespace Common {
+
+// This function multiplies 2 u64 values and divides it by a u64 value.
+[[nodiscard]] static inline u64 MultiplyAndDivide64(u64 a, u64 b, u64 d) {
+#ifdef HAS_INTRINSICS
+    u128 r{};
+    r[0] = _umul128(a, b, &r[1]);
+    u64 remainder;
+    return _udiv128(r[1], r[0], d, &remainder);
+#else
+    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;
+#endif
+}
+
+// This function multiplies 2 u64 values and produces a u128 value;
+[[nodiscard]] static inline u128 Multiply64Into128(u64 a, u64 b) {
+    u128 result;
+#ifdef HAS_INTRINSICS
+    result[0] = _umul128(a, b, &result[1]);
+#else
+    unsigned __int128 tmp = a;
+    tmp *= b;
+    std::memcpy(&result, &tmp, sizeof(u128));
+#endif
+    return result;
+}
+
+[[nodiscard]] static inline u64 GetFixedPoint64Factor(u64 numerator, u64 divisor) {
+#ifdef __SIZEOF_INT128__
+    const auto base = static_cast<unsigned __int128>(numerator) << 64ULL;
+    return static_cast<u64>(base / divisor);
+#elif defined(_M_X64) || defined(_M_ARM64)
+    std::array<u64, 2> r = {0, numerator};
+    u64 remainder;
+    return _udiv128(r[1], r[0], divisor, &remainder);
+#else
+    // This one is bit more inaccurate.
+    return MultiplyAndDivide64(std::numeric_limits<u64>::max(), numerator, divisor);
+#endif
+}
+
+[[nodiscard]] static inline u64 MultiplyHigh(u64 a, u64 b) {
+#ifdef __SIZEOF_INT128__
+    return (static_cast<unsigned __int128>(a) * static_cast<unsigned __int128>(b)) >> 64;
+#elif defined(_M_X64) || defined(_M_ARM64)
+    return __umulh(a, b); // MSVC
+#else
+    // Generic fallback
+    const u64 a_lo = u32(a);
+    const u64 a_hi = a >> 32;
+    const u64 b_lo = u32(b);
+    const u64 b_hi = b >> 32;
+
+    const u64 a_x_b_hi = a_hi * b_hi;
+    const u64 a_x_b_mid = a_hi * b_lo;
+    const u64 b_x_a_mid = b_hi * a_lo;
+    const u64 a_x_b_lo = a_lo * b_lo;
+
+    const u64 carry_bit = (static_cast<u64>(static_cast<u32>(a_x_b_mid)) +
+                           static_cast<u64>(static_cast<u32>(b_x_a_mid)) + (a_x_b_lo >> 32)) >>
+                          32;
+
+    const u64 multhi = a_x_b_hi + (a_x_b_mid >> 32) + (b_x_a_mid >> 32) + carry_bit;
+
+    return multhi;
+#endif
+}
+
+// This function divides a u128 by a u32 value and produces two u64 values:
+// the result of division and the remainder
+[[nodiscard]] static inline std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor) {
+    u64 remainder = dividend[0] % divisor;
+    u64 accum = dividend[0] / divisor;
+    if (dividend[1] == 0)
+        return {accum, remainder};
+    // We ignore dividend[1] / divisor as that overflows
+    const u64 first_segment = (dividend[1] % divisor) << 32;
+    accum += (first_segment / divisor) << 32;
+    const u64 second_segment = (first_segment % divisor) << 32;
+    accum += (second_segment / divisor);
+    remainder += second_segment % divisor;
+    if (remainder >= divisor) {
+        accum++;
+        remainder -= divisor;
+    }
+    return {accum, remainder};
+}
+
+} // namespace Common

src/core/PS4/GPU/gpu_memory.cpp

@@ -1,5 +1,5 @@
 #include "gpu_memory.h"
-
+#include <atomic>
 #include <xxh3.h>
 
 #include "common/singleton.h"

src/core/PS4/HLE/Graphics/graphics_ctx.h

@@ -2,8 +2,9 @@
 
 #include "common/types.h"
 #include <vector>
-#include <vulkan/vulkan_core.h>
+#include <vulkan/vulkan.h>
 #include <mutex>
+#include <memory>
 
 namespace HLE::Libs::Graphics {
 

src/core/file_sys/fs.h

@@ -1,4 +1,6 @@
 #pragma once
+
+#include <atomic>
 #include <mutex>
 #include <string>
 #include <vector>
@@ -52,4 +54,4 @@ class HandleTable {
     std::mutex m_mutex;
 };
 
-}  // namespace Core::FileSys
+}  // namespace Core::FileSys

src/core/hle/kernel/Objects/event_queue.cpp

@@ -1,6 +1,5 @@
 #include "common/debug.h"
-#include "common/timer.h"
+#include "core/hle/kernel/Objects/event_queue.h"
-#include "core/hle/kernel/objects/event_queue.h"
 
 namespace Core::Kernel {
 
@@ -24,28 +23,19 @@ int EqueueInternal::addEvent(const EqueueEvent& event) {
 
 int EqueueInternal::waitForEvents(SceKernelEvent* ev, int num, u32 micros) {
     std::unique_lock lock{m_mutex};
+    int ret = 0;
 
-    u32 timeElapsed = 0;
+    const auto predicate = [&] {
-    Common::Timer t;
+        ret = getTriggeredEvents(ev, num);
-    t.Start();
+        return ret > 0;
+    };
 
-    for (;;) {
+    if (micros == 0) {
-        int ret = getTriggeredEvents(ev, num);
+        m_cond.wait(lock, predicate);
-
+    } else {
-        if (ret > 0 || (timeElapsed >= micros && micros != 0)) {
+        m_cond.wait_for(lock, std::chrono::microseconds(micros), predicate);
-            return ret;
-        }
-
-        if (micros == 0) {
-            m_cond.wait(lock);
-        } else {
-            m_cond.wait_for(lock, std::chrono::microseconds(micros - timeElapsed));
-        }
-
-        timeElapsed = static_cast<uint32_t>(t.GetTimeSec() * 1000000.0);
     }
-
+    return ret;
-    return 0;
 }
 
 bool EqueueInternal::triggerEvent(u64 ident, s16 filter, void* trigger_data) {

src/core/hle/kernel/Objects/event_queue.h

@@ -3,6 +3,7 @@
 #include <mutex>
 #include <string>
 #include <vector>
+#include <condition_variable>
 #include "common/types.h"
 
 namespace Core::Kernel {

src/core/hle/kernel/Objects/physical_memory.cpp

@@ -1,4 +1,4 @@
-#include "core/hle/kernel/objects/physical_memory.h"
+#include "core/hle/kernel/Objects/physical_memory.h"
 
 namespace Core::Kernel {
 

src/core/hle/kernel/event_queues.h

@@ -1,6 +1,6 @@
 #pragma once
 
-#include "core/hle/kernel/objects/event_queue.h"
+#include "core/hle/kernel/Objects/event_queue.h"
 
 namespace Core::Kernel {
 

src/core/hle/libraries/libc/libc_math.cpp

@@ -4,35 +4,35 @@
 namespace Core::Libraries::LibC {
 
 float PS4_SYSV_ABI ps4_atan2f(float y, float x) {
-    return std::atan2f(y, x);
+    return atan2f(y, x);
 }
 
 float PS4_SYSV_ABI ps4_acosf(float num) {
-    return std::acosf(num);
+    return acosf(num);
 }
 
 float PS4_SYSV_ABI ps4_tanf(float num) {
-    return std::tanf(num);
+    return tanf(num);
 }
 
 float PS4_SYSV_ABI ps4_asinf(float num) {
-    return std::asinf(num);
+    return asinf(num);
 }
 
 double PS4_SYSV_ABI ps4_pow(double base, double exponent) {
-    return std::pow(base, exponent);
+    return pow(base, exponent);
 }
 
 double PS4_SYSV_ABI ps4__Sin(double x) {
-    return std::sin(x);
+    return sin(x);
 }
 
 float PS4_SYSV_ABI ps4__Fsin(float arg) {
-    return std::sinf(arg);
+    return sinf(arg);
 }
 
 double PS4_SYSV_ABI ps4_exp2(double arg) {
-    return std::exp2(arg);
+    return exp2(arg);
 }
 
 } // namespace Core::Libraries::LibC

src/core/hle/libraries/libc/libc_stdio.cpp

@@ -12,7 +12,7 @@ int PS4_SYSV_ABI ps4_printf(VA_ARGS) {
 }
 
 int PS4_SYSV_ABI ps4_fprintf(FILE* file, VA_ARGS) {
-    int fd = _fileno(file);
+    int fd = fileno(file);
     if (fd == 1 || fd == 2) {  // output stdout and stderr to console
         VA_CTX(ctx);
         return printf_ctx(&ctx);

src/core/hle/libraries/libkernel/libkernel.cpp

@@ -15,6 +15,8 @@
 #ifdef _WIN64
 #include <windows.h>
 #include <io.h>
+#else
+#include <sys/mman.h>
 #endif
 #include "thread_management.h"
 
@@ -56,6 +58,7 @@ int* PS4_SYSV_ABI __Error() { return &libc_error; }
 #define PROT_WRITE 0x2
 
 int PS4_SYSV_ABI sceKernelMmap(void* addr, u64 len, int prot, int flags, int fd, off_t offset, void** res) {
+#ifdef _WIN64
     PRINT_FUNCTION_NAME();
     if (prot > 3)  // READ,WRITE or bitwise READ | WRITE supported
     {
@@ -86,6 +89,14 @@ int PS4_SYSV_ABI sceKernelMmap(void* addr, u64 len, int prot, int flags, int fd,
     }
     *res = ret;
     return 0;
+#else
+    void* result = mmap(addr, len, prot, flags, fd, offset);
+    if (result != MAP_FAILED) {
+        *res = result;
+        return 0;
+    }
+    std::abort();
+#endif
 }
 
 PS4_SYSV_ABI void* posix_mmap(void* addr, u64 len, int prot, int flags, int fd, u64 offset) {

src/core/hle/libraries/libkernel/time_management.cpp

@@ -1,27 +1,31 @@
-#include "common/timer.h"
+#include "common/native_clock.h"
 #include "core/hle/libraries/libkernel/time_management.h"
 #include "core/hle/libraries/libs.h"
-#include "emuTimer.h"
 
 namespace Core::Libraries::LibKernel {
 
+static u64 initial_ptc;
+static std::unique_ptr<Common::NativeClock> clock;
+
 u64 PS4_SYSV_ABI sceKernelGetProcessTime() {
-    return static_cast<u64>(Emulator::emuTimer::getTimeMsec() * 1000.0);  // return time in microseconds
+    return clock->GetProcessTimeUS();
 }
 
 u64 PS4_SYSV_ABI sceKernelGetProcessTimeCounter() {
-    return Emulator::emuTimer::getTimeCounter();
+    return clock->GetUptime() - initial_ptc;
 }
 
 u64 PS4_SYSV_ABI sceKernelGetProcessTimeCounterFrequency() {
-    return Emulator::emuTimer::getTimeFrequency();
+    return clock->GetTscFrequency();
 }
 
 u64 PS4_SYSV_ABI sceKernelReadTsc() {
-    return Common::Timer::getQueryPerformanceCounter();
+    return clock->GetUptime();
 }
 
 void timeSymbolsRegister(Loader::SymbolsResolver* sym) {
+    clock = std::make_unique<Common::NativeClock>();
+    initial_ptc = clock->GetUptime();
     LIB_FUNCTION("4J2sUJmuHZQ", "libkernel", 1, "libkernel", 1, 1, sceKernelGetProcessTime); 
     LIB_FUNCTION("fgxnMeTNUtY", "libkernel", 1, "libkernel", 1, 1, sceKernelGetProcessTimeCounter);
     LIB_FUNCTION("BNowx2l588E", "libkernel", 1, "libkernel", 1, 1, sceKernelGetProcessTimeCounterFrequency);

src/core/linker.cpp

@@ -658,12 +658,12 @@ void Linker::Resolve(const std::string& name, int Symtype, Module* m, Loader::Sy
 		}
 		else
 		{
-            __debugbreak();//den tha prepei na ftasoume edo
+            //__debugbreak();//den tha prepei na ftasoume edo
 		}
 	}
 	else
 	{
-        __debugbreak();//oute edo mallon
+        //__debugbreak();//oute edo mallon
 	}
 
 }

src/core/loader/elf.cpp

@@ -5,7 +5,7 @@
 
 namespace Core::Loader {
 
-constexpr bool log_file_loader = true;  // disable it to disable logging
+constexpr bool log_file_loader = false;  // disable it to disable logging
 
 static std::string_view getProgramTypeName(program_type_es type) {
     switch (type) {

src/core/virtual_memory.cpp

@@ -79,7 +79,8 @@ bool memory_protect(u64 address, u64 size, MemoryMode mode, MemoryMode* old_mode
     }
     return true;
 #else
-#error Unimplement memory_protect function
+    int ret = mprotect(reinterpret_cast<void*>(address), size, convertMemoryMode(mode));
+    return true;
 #endif
 }
 
@@ -117,6 +118,7 @@ bool memory_patch(u64 vaddr, u64 value) {
 static u64 AlignUp(u64 pos, u64 align) { return (align != 0 ? (pos + (align - 1)) & ~(align - 1) : pos); }
 
 u64 memory_alloc_aligned(u64 address, u64 size, MemoryMode mode, u64 alignment) {
+#ifdef _WIN64
     // try allocate aligned address inside user area
     MEM_ADDRESS_REQUIREMENTS req{};
     MEM_EXTENDED_PARAMETER param{};
@@ -134,5 +136,13 @@ u64 memory_alloc_aligned(u64 address, u64 size, MemoryMode mode, u64 alignment)
         LOG_ERROR_IF(true, "VirtualAlloc2() failed: 0x{:X}\n", err);
     }
     return ptr;
+#else
+    void* hint_address = reinterpret_cast<void*>(AlignUp(address, alignment));
+    void* ptr = mmap(hint_address, size, convertMemoryMode(mode), MAP_ANON | MAP_PRIVATE, -1, 0);
+    if (ptr == MAP_FAILED) {
+        std::abort();
+    }
+    return reinterpret_cast<u64>(ptr);
+#endif
 }
 }  // namespace VirtualMemory

src/emuTimer.cpp

@@ -1,23 +0,0 @@
-#include "common/timer.h"
-
-namespace Emulator::emuTimer {
-
-static Common::Timer timer;
-
-void start() {
-    timer.Start();
-}
-
-double getTimeMsec() {
-    return timer.GetTimeMsec();
-}
-
-u64 getTimeCounter() {
-    return timer.GetTicks();
-}
-
-u64 getTimeFrequency() {
-    return timer.GetFrequency();
-}
-
-}  // namespace Emulator::emuTimer

src/emuTimer.h

@@ -1,10 +0,0 @@
-#pragma once
-
-#include "common/types.h"
-
-namespace Emulator::emuTimer {
-void start();
-double getTimeMsec();
-u64 getTimeCounter();
-u64 getTimeFrequency();
-}  // namespace Emulator::emuTimer

src/emulator.cpp

@@ -1,6 +1,5 @@
 #include <fmt/core.h>
 #include <vulkan_util.h>
-#include "common/timer.h"
 #include "common/singleton.h"
 #include "common/version.h"
 #include "emulator.h"
@@ -90,8 +89,6 @@ static void calculateFps(double game_time_s) {
     }
 }
 void emuRun() {
-    Common::Timer timer;
-    timer.Start();
     auto window_ctx = Common::Singleton<Emu::WindowCtx>::Instance();
     {
         // init window and wait until init finishes
@@ -100,7 +97,7 @@ void emuRun() {
         Graphics::Vulkan::vulkanCreate(window_ctx);
         window_ctx->m_is_graphic_initialized = true;
         window_ctx->m_graphic_initialized_cond.notify_one();
-        calculateFps(timer.GetTimeSec());
+        calculateFps(0); // TODO: Proper fps
     }
 
     bool exit_loop = false;
@@ -138,10 +135,6 @@ void emuRun() {
             continue;
         }
         if (m_game_is_paused) {
-            if (!timer.IsPaused()) {
-                timer.Pause();
-            }
-
             SDL_WaitEvent(&event);
 
             switch (event.type) {
@@ -171,21 +164,13 @@ void emuRun() {
             continue;
         }
         exit_loop = m_emu_needs_exit;
-        if (m_game_is_paused) {
+        if (!m_game_is_paused) {
-            if (!timer.IsPaused()) {
-                timer.Pause();
-            }
-        } else {
-            if (timer.IsPaused()) {
-                timer.Resume();
-            }
-
             if (!exit_loop) {
                 update();
             }
             if (!exit_loop) {
                 if (HLE::Libs::Graphics::VideoOut::videoOutFlip(100000)) {  // flip every 0.1 sec
-                    calculateFps(timer.GetTimeSec());
+                    calculateFps(0); // TODO: Proper fps
                 }
             }
         }
@@ -215,7 +200,7 @@ void DrawBuffer(HLE::Libs::Graphics::VideoOutVulkanImage* image) {
     window_ctx->swapchain.current_index = static_cast<u32>(-1);
 
     auto result = vkAcquireNextImageKHR(window_ctx->m_graphic_ctx.m_device, window_ctx->swapchain.swapchain, UINT64_MAX, nullptr,
-                                        window_ctx->swapchain.present_complete_fence, &window_ctx->swapchain.current_index);
+                                        VK_NULL_HANDLE, &window_ctx->swapchain.current_index);
 
     if (result != VK_SUCCESS) {
         fmt::print("Can't aquireNextImage\n");
@@ -226,16 +211,6 @@ void DrawBuffer(HLE::Libs::Graphics::VideoOutVulkanImage* image) {
         std::exit(0);
     }
 
-    do {
-        result = vkWaitForFences(window_ctx->m_graphic_ctx.m_device, 1, &window_ctx->swapchain.present_complete_fence, VK_TRUE, 100000000);
-    } while (result == VK_TIMEOUT);
-    if (result != VK_SUCCESS) {
-        fmt::print("vkWaitForFences is not success\n");
-        std::exit(0);
-    }
-
-    vkResetFences(window_ctx->m_graphic_ctx.m_device, 1, &window_ctx->swapchain.present_complete_fence);
-
     auto blt_src_image = image;
     auto blt_dst_image = window_ctx->swapchain;
 
@@ -272,6 +247,7 @@ void DrawBuffer(HLE::Libs::Graphics::VideoOutVulkanImage* image) {
 
     buffer.end();
     buffer.executeWithSemaphore();
+    buffer.waitForFence(); // HACK: The whole vulkan backend needs a rewrite
 
     VkPresentInfoKHR present{};
     present.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;

src/main.cpp

@@ -14,7 +14,6 @@
 #include "core/PS4/HLE/Graphics/video_out.h"
 #include "core/hle/libraries/libs.h"
 #include "core/linker.h"
-#include "emuTimer.h"
 #include "emulator.h"
 #include <core/hle/libraries/libkernel/thread_management.h>
 #include "core/file_sys/fs.h"
@@ -31,7 +30,6 @@ int main(int argc, char* argv[]) {
     auto height = Config::getScreenHeight();
     Emu::emuInit(width, height);
     HLE::Libs::Graphics::VideoOut::videoOutInit(width, height);
-    Emulator::emuTimer::start();
 
     // Argument 1 is the path of self file to boot
     const char* const path = argv[1];

src/video_core/gpu_memory.cpp

@@ -0,0 +1,188 @@
+#include "gpu_memory.h"
+#include <atomic>
+#include <xxh3.h>
+
+#include "common/singleton.h"
+
+void* GPU::memoryCreateObj(u64 submit_id, HLE::Libs::Graphics::GraphicCtx* ctx, void* todo /*CommandBuffer?*/, u64 virtual_addr, u64 size,
+                           const GPUObject& info) {
+    auto* gpumemory = Common::Singleton<GPUMemory>::Instance();
+
+    return gpumemory->memoryCreateObj(submit_id, ctx, nullptr, &virtual_addr, &size, 1, info);
+}
+
+void GPU::memorySetAllocArea(u64 virtual_addr, u64 size) {
+    auto* gpumemory = Common::Singleton<GPUMemory>::Instance();
+
+    std::scoped_lock lock{gpumemory->m_mutex};
+
+    MemoryHeap h;
+    h.allocated_virtual_addr = virtual_addr;
+    h.allocated_size = size;
+
+    gpumemory->m_heaps.push_back(h);
+}
+
+u64 GPU::calculate_hash(const u08* buf, u64 size) { return (size > 0 && buf != nullptr ? XXH3_64bits(buf, size) : 0); }
+
+bool GPU::vulkanAllocateMemory(HLE::Libs::Graphics::GraphicCtx* ctx, HLE::Libs::Graphics::VulkanMemory* mem) {
+    static std::atomic_uint64_t unique_id = 0;
+
+    VkPhysicalDeviceMemoryProperties memory_properties{};
+    vkGetPhysicalDeviceMemoryProperties(ctx->m_physical_device, &memory_properties);
+
+    u32 index = 0;
+    for (; index < memory_properties.memoryTypeCount; index++) {
+        if ((mem->requirements.memoryTypeBits & (static_cast<uint32_t>(1) << index)) != 0 &&
+            (memory_properties.memoryTypes[index].propertyFlags & mem->property) == mem->property) {
+            break;
+        }
+    }
+
+    mem->type = index;
+    mem->offset = 0;
+
+    VkMemoryAllocateInfo alloc_info{};
+    alloc_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
+    alloc_info.pNext = nullptr;
+    alloc_info.allocationSize = mem->requirements.size;
+    alloc_info.memoryTypeIndex = index;
+
+    mem->unique_id = ++unique_id;
+
+    auto result = vkAllocateMemory(ctx->m_device, &alloc_info, nullptr, &mem->memory);
+
+    if (result == VK_SUCCESS) {
+        return true;
+    }
+    return false;
+}
+
+void GPU::flushGarlic(HLE::Libs::Graphics::GraphicCtx* ctx) {
+    auto* gpumemory = Common::Singleton<GPUMemory>::Instance();
+    gpumemory->flushAllHeaps(ctx);
+}
+
+int GPU::GPUMemory::getHeapId(u64 virtual_addr, u64 size) {
+    int index = 0;
+    for (const auto& heap : m_heaps) {
+        if ((virtual_addr >= heap.allocated_virtual_addr && virtual_addr < heap.allocated_virtual_addr + heap.allocated_size) ||
+            ((virtual_addr + size - 1) >= heap.allocated_virtual_addr &&
+             (virtual_addr + size - 1) < heap.allocated_virtual_addr + heap.allocated_size)) {
+            return index;
+        }
+        index++;
+    }
+    return -1;
+}
+
+void* GPU::GPUMemory::memoryCreateObj(u64 submit_id, HLE::Libs::Graphics::GraphicCtx* ctx, void* todo, const u64* virtual_addr, const u64* size,
+                                      int virtual_addr_num, const GPUObject& info) {
+    auto* gpumemory = Common::Singleton<GPUMemory>::Instance();
+
+    std::scoped_lock lock{gpumemory->m_mutex};
+
+    int heap_id = gpumemory->getHeapId(virtual_addr[0], size[0]);
+
+    if (heap_id < 0) {
+        return nullptr;
+    }
+    auto& heap = m_heaps[heap_id];
+
+    ObjInfo objInfo = {};
+
+    // Copy parameters from info to obj
+    for (int i = 0; i < 8; i++) {
+        objInfo.obj_params[i] = info.obj_params[i];
+    }
+
+    objInfo.gpu_object.objectType = info.objectType;
+    objInfo.gpu_object.obj = nullptr;
+
+    for (int h = 0; h < virtual_addr_num; h++) {
+        if (info.check_hash) {
+            objInfo.hash[h] = GPU::calculate_hash(reinterpret_cast<const u08*>(virtual_addr[h]), size[h]);
+        } else {
+            objInfo.hash[h] = 0;
+        }
+    }
+    objInfo.submit_id = submit_id;
+    objInfo.check_hash = info.check_hash;
+
+    objInfo.gpu_object.obj = info.getCreateFunc()(ctx, objInfo.obj_params, virtual_addr, size, virtual_addr_num, &objInfo.mem);
+
+    objInfo.update_func = info.getUpdateFunc();
+    int index = static_cast<int>(heap.objects.size());
+
+    HeapObject hobj{};
+    hobj.block = createHeapBlock(virtual_addr, size, virtual_addr_num, heap_id, index);
+    hobj.info = objInfo;
+    hobj.free = false;
+    heap.objects.push_back(hobj);
+
+    return objInfo.gpu_object.obj;
+}
+
+GPU::HeapBlock GPU::GPUMemory::createHeapBlock(const u64* virtual_addr, const u64* size, int virtual_addr_num, int heap_id, int obj_id) {
+    auto& heap = m_heaps[heap_id];
+
+    GPU::HeapBlock heapBlock{};
+    heapBlock.virtual_addr_num = virtual_addr_num;
+    for (int vi = 0; vi < virtual_addr_num; vi++) {
+        heapBlock.virtual_addr[vi] = virtual_addr[vi];
+        heapBlock.size[vi] = size[vi];
+    }
+    return heapBlock;
+}
+
+void GPU::GPUMemory::update(u64 submit_id, HLE::Libs::Graphics::GraphicCtx* ctx, int heap_id, int obj_id) {
+    auto& heap = m_heaps[heap_id];
+
+    auto& heapObj = heap.objects[obj_id];
+    auto& objInfo = heapObj.info;
+    bool need_update = false;
+
+    if (submit_id > objInfo.submit_id) {
+        uint64_t hash[3] = {};
+
+        for (int i = 0; i < heapObj.block.virtual_addr_num; i++) {
+            if (objInfo.check_hash) {
+                hash[i] = GPU::calculate_hash(reinterpret_cast<const uint8_t*>(heapObj.block.virtual_addr[i]), heapObj.block.size[i]);
+            } else {
+                hash[i] = 0;
+            }
+        }
+
+        for (int i = 0; i < heapObj.block.virtual_addr_num; i++) {
+            if (objInfo.hash[i] != hash[i]) {
+                need_update = true;
+                objInfo.hash[i] = hash[i];
+            }
+        }
+
+        if (submit_id != UINT64_MAX) {
+            objInfo.submit_id = submit_id;
+        }
+    }
+
+    if (need_update) {
+        objInfo.update_func(ctx, objInfo.obj_params, objInfo.gpu_object.obj, heapObj.block.virtual_addr, heapObj.block.size,
+                            heapObj.block.virtual_addr_num);
+    }
+}
+
+void GPU::GPUMemory::flushAllHeaps(HLE::Libs::Graphics::GraphicCtx* ctx) {
+    std::scoped_lock lock{m_mutex};
+
+    int heap_id = 0;
+    for (auto& heap : m_heaps) {
+        int index = 0;
+        for (auto& heapObj : heap.objects) {
+            if (!heapObj.free) {
+                update(UINT64_MAX, ctx, heap_id, index);
+            }
+            index++;
+        }
+        heap_id++;
+    }
+}

src/video_core/gpu_memory.h

@@ -0,0 +1,86 @@
+#pragma once
+
+#include "common/types.h"
+#include <mutex>
+#include <vector>
+
+namespace VideoCore {
+
+class GPUObject;
+
+enum class MemoryMode : u32 {
+    NoAccess = 0,
+    Read = 1,
+    Write = 2,
+    ReadWrite = 3,
+};
+
+enum class MemoryObjectType : u64 {
+    Invalid,
+    VideoOutBuffer,
+};
+
+struct GpuMemoryObject {
+    MemoryObjectType object_type = MemoryObjectType::Invalid;
+    void* obj = nullptr;
+};
+
+struct HeapBlock {
+    std::array<u64, 3> virtual_address{};
+    std::array<u64, 3> size{};
+    u32 virtual_addr_num = 0;
+};
+
+class GPUObject {
+public:
+    GPUObject() = default;
+    virtual ~GPUObject() = default;
+    u64 obj_params[8] = {};
+    bool check_hash = false;
+    bool isReadOnly = false;
+    MemoryObjectType objectType = MemoryObjectType::Invalid;
+};
+
+struct ObjInfo {
+    std::array<u64, 8> obj_params{};
+    GpuMemoryObject gpu_object;
+    std::array<u64, 3> hash{};
+    u64 submit_id = 0;
+    bool check_hash = false;
+};
+
+struct HeapObject {
+    HeapBlock block;
+    ObjInfo info;
+    bool free = true;
+};
+struct MemoryHeap {
+    u64 allocated_virtual_addr = 0;
+    u64 allocated_size = 0;
+    std::vector<HeapObject> objects;
+};
+
+class GPUMemory {
+public:
+    GPUMemory() {}
+    virtual ~GPUMemory() {}
+    int getHeapId(u64 vaddr, u64 size);
+    void* memoryCreateObj(u64 submit_id, HLE::Libs::Graphics::GraphicCtx* ctx, /*CommandBuffer* buffer*/ void* todo, const u64* virtual_addr,
+                          const u64* size, int virtual_addr_num, const GPUObject& info);
+    HeapBlock createHeapBlock(const u64* virtual_addr, const u64* size, int virtual_addr_num, int heap_id, int obj_id);
+    void update(u64 submit_id, HLE::Libs::Graphics::GraphicCtx* ctx, int heap_id, int obj_id);
+    void flushAllHeaps(HLE::Libs::Graphics::GraphicCtx* ctx);
+
+private:
+    std::mutex m_mutex;
+    std::vector<MemoryHeap> m_heaps;
+};
+
+void memorySetAllocArea(u64 virtual_addr, u64 size);
+void* memoryCreateObj(u64 submit_id, HLE::Libs::Graphics::GraphicCtx* ctx, /*CommandBuffer* buffer*/ void* todo, u64 virtual_addr, u64 size,
+                      const GPUObject& info);
+u64 calculate_hash(const u08* buf, u64 size);
+bool vulkanAllocateMemory(HLE::Libs::Graphics::GraphicCtx* ctx, HLE::Libs::Graphics::VulkanMemory* mem);
+void flushGarlic(HLE::Libs::Graphics::GraphicCtx* ctx);
+
+}  // namespace VideoCore

src/video_core/tile_manager.cpp

@@ -0,0 +1,151 @@
+#include <bit>
+#include <cstring>
+#include "video_core/tile_manager.h"
+
+namespace VideoCore {
+
+class TileManager32 {
+  public:
+    u32 m_macro_tile_height = 0;
+    u32 m_bank_height = 0;
+    u32 m_num_banks = 0;
+    u32 m_num_pipes = 0;
+    u32 m_padded_width = 0;
+    u32 m_padded_height = 0;
+    u32 m_pipe_bits = 0;
+    u32 m_bank_bits = 0;
+
+    TileManager32(u32 width, u32 height, bool is_neo) {
+        m_macro_tile_height = (is_neo ? 128 : 64);
+        m_bank_height = is_neo ? 2 : 1;
+        m_num_banks = is_neo ? 8 : 16;
+        m_num_pipes = is_neo ? 16 : 8;
+        m_padded_width = width;
+        if (height == 1080) {
+            m_padded_height = is_neo ? 1152 : 1088;
+        }
+        if (height == 720) {
+            m_padded_height = 768;
+        }
+        m_pipe_bits = is_neo ? 4 : 3;
+        m_bank_bits = is_neo ? 3 : 4;
+    }
+
+    static u32 GetElementIndex(u32 x, u32 y) {
+        u32 elem = 0;
+        elem |= ((x >> 0u) & 0x1u) << 0u;
+        elem |= ((x >> 1u) & 0x1u) << 1u;
+        elem |= ((y >> 0u) & 0x1u) << 2u;
+        elem |= ((x >> 2u) & 0x1u) << 3u;
+        elem |= ((y >> 1u) & 0x1u) << 4u;
+        elem |= ((y >> 2u) & 0x1u) << 5u;
+
+        return elem;
+    }
+
+    static u32 GetPipeIndex(u32 x, u32 y, bool is_neo) {
+        u32 pipe = 0;
+
+        if (!is_neo) {
+            pipe |= (((x >> 3u) ^ (y >> 3u) ^ (x >> 4u)) & 0x1u) << 0u;
+            pipe |= (((x >> 4u) ^ (y >> 4u)) & 0x1u) << 1u;
+            pipe |= (((x >> 5u) ^ (y >> 5u)) & 0x1u) << 2u;
+        } else {
+            pipe |= (((x >> 3u) ^ (y >> 3u) ^ (x >> 4u)) & 0x1u) << 0u;
+            pipe |= (((x >> 4u) ^ (y >> 4u)) & 0x1u) << 1u;
+            pipe |= (((x >> 5u) ^ (y >> 5u)) & 0x1u) << 2u;
+            pipe |= (((x >> 6u) ^ (y >> 5u)) & 0x1u) << 3u;
+        }
+
+        return pipe;
+    }
+
+    static u32 GetBankIndex(u32 x, u32 y, u32 bank_width, u32 bank_height, u32 num_banks, u32 num_pipes) {
+        const u32 x_shift_offset = std::bit_width(bank_width * num_pipes);
+        const u32 y_shift_offset = std::bit_width(bank_height);
+        const u32 xs = x >> x_shift_offset;
+        const u32 ys = y >> y_shift_offset;
+        u32 bank = 0;
+        switch (num_banks) {
+            case 8:
+                bank |= (((xs >> 3u) ^ (ys >> 5u)) & 0x1u) << 0u;
+                bank |= (((xs >> 4u) ^ (ys >> 4u) ^ (ys >> 5u)) & 0x1u) << 1u;
+                bank |= (((xs >> 5u) ^ (ys >> 3u)) & 0x1u) << 2u;
+                break;
+            case 16:
+                bank |= (((xs >> 3u) ^ (ys >> 6u)) & 0x1u) << 0u;
+                bank |= (((xs >> 4u) ^ (ys >> 5u) ^ (ys >> 6u)) & 0x1u) << 1u;
+                bank |= (((xs >> 5u) ^ (ys >> 4u)) & 0x1u) << 2u;
+                bank |= (((xs >> 6u) ^ (ys >> 3u)) & 0x1u) << 3u;
+                break;
+            default:;
+        }
+
+        return bank;
+    }
+
+    u64 GetTiledOffset(u32 x, u32 y, bool is_neo) const {
+        u64 element_index = GetElementIndex(x, y);
+
+        u32 xh = x;
+        u32 yh = y;
+        u64 pipe = GetPipeIndex(xh, yh, is_neo);
+        u64 bank = GetBankIndex(xh, yh, 1, m_bank_height, m_num_banks, m_num_pipes);
+        u32 tile_bytes = (8 * 8 * 32 + 7) / 8;
+        u64 element_offset = (element_index * 32);
+        u64 tile_split_slice = 0;
+
+        if (tile_bytes > 512) {
+            tile_split_slice = element_offset / (static_cast<u64>(512) * 8);
+            element_offset %= (static_cast<u64>(512) * 8);
+            tile_bytes = 512;
+        }
+
+        u64 macro_tile_bytes = (128 / 8) * (m_macro_tile_height / 8) * tile_bytes / (m_num_pipes * m_num_banks);
+        u64 macro_tiles_per_row = m_padded_width / 128;
+        u64 macro_tile_row_index = y / m_macro_tile_height;
+        u64 macro_tile_column_index = x / 128;
+        u64 macro_tile_index = (macro_tile_row_index * macro_tiles_per_row) + macro_tile_column_index;
+        u64 macro_tile_offset = macro_tile_index * macro_tile_bytes;
+        u64 macro_tiles_per_slice = macro_tiles_per_row * (m_padded_height / m_macro_tile_height);
+        u64 slice_bytes = macro_tiles_per_slice * macro_tile_bytes;
+        u64 slice_offset = tile_split_slice * slice_bytes;
+        u64 tile_row_index = (y / 8) % m_bank_height;
+        u64 tile_index = tile_row_index;
+        u64 tile_offset = tile_index * tile_bytes;
+
+        u64 tile_split_slice_rotation = ((m_num_banks / 2) + 1) * tile_split_slice;
+        bank ^= tile_split_slice_rotation;
+        bank &= (m_num_banks - 1);
+
+        u64 total_offset = (slice_offset + macro_tile_offset + tile_offset) * 8 + element_offset;
+        u64 bit_offset = total_offset & 0x7u;
+        total_offset /= 8;
+
+        u64 pipe_interleave_offset = total_offset & 0xffu;
+        u64 offset = total_offset >> 8u;
+        u64 byte_offset = pipe_interleave_offset | (pipe << (8u)) | (bank << (8u + m_pipe_bits)) | (offset << (8u + m_pipe_bits + m_bank_bits));
+
+        return ((byte_offset << 3u) | bit_offset) / 8;
+    }
+};
+
+void ConvertTileToLinear(u08* dst, const u08* src,u32 width, u32 height, bool is_neo) {
+    const TileManager32 t{width, height, is_neo};
+    for (u32 y = 0; y < height; y++) {
+        u32 x = 0;
+        u64 linear_offset = y * width * 4;
+
+        for (; x + 1 < width; x += 2) {
+            auto tiled_offset = t.GetTiledOffset(x, y, is_neo);
+            std::memcpy(dst + linear_offset, src + tiled_offset, sizeof(u64));
+            linear_offset += sizeof(u64);
+        }
+        if (x < width) {
+            auto tiled_offset = t.GetTiledOffset(x, y, is_neo);
+            std::memcpy(dst + linear_offset, src + tiled_offset, sizeof(u32));
+        }
+    }
+}
+
+} // namespace VideoCore

src/video_core/tile_manager.h

@@ -0,0 +1,9 @@
+#pragma once
+
+#include "common/types.h"
+
+namespace VideoCore {
+
+void ConvertTileToLinear(void* dst, const void* src, u32 width, u32 height, bool neo);
+
+} // namespace VideoCore

third-party/CMakeLists.txt

@@ -30,7 +30,6 @@ add_subdirectory(toml11 EXCLUDE_FROM_ALL)
 
 # Vulkan
 add_subdirectory(vulkan EXCLUDE_FROM_ALL)
-target_include_directories(vulkan-1 INTERFACE vulkan/include)
 
 # Winpthreads
 if (WIN32)
@@ -66,6 +65,6 @@ target_include_directories(imgui PUBLIC
     imgui/include
 )
 
-target_link_libraries(imgui PRIVATE SDL3-shared ${CMAKE_DL_LIBS} Zydis winpthread discord-rpc)
+target_link_libraries(imgui PRIVATE SDL3-shared ${CMAKE_DL_LIBS} Zydis discord-rpc)
 
 

third-party/fmt

@@ -1 +1 @@
-Subproject commit 661b23edeb52d400cf5812e7330f14f05c072fab
+Subproject commit 8e42eef4950feb5d2b76574a9cd2591dfaae2449

third-party/spdlog

@@ -1 +1 @@
-Subproject commit 76dfc7e7c0d3c69d3cdaa3399b63545235ccbb02
+Subproject commit 134f9194bb93072b72b8cfa27ac3bb30a0fb5b57

third-party/vulkan

@@ -1 +1 @@
-Subproject commit 8c9feb4f480b32f7c7421af546aa6ffb558bdd5e
+Subproject commit 72b2e740754bc6b86b724fa5b2c90dca6f69462e