19 files changed, 146 insertions, 389 deletions

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index c839ce173..aa6eee2a3 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -47,7 +47,6 @@ set(HEADERS
             microprofile.h
             microprofileui.h
             platform.h
-            profiler.h
             profiler_reporting.h
             scm_rev.h
             scope_exit.h
diff --git a/src/common/assert.h b/src/common/assert.h
index 6849778b7..cd9b819a9 100644
--- a/src/common/assert.h
+++ b/src/common/assert.h
@@ -39,6 +39,7 @@ static void assert_noinline_call(const Fn& fn) {
     }); } while (0)
 
 #define UNREACHABLE() ASSERT_MSG(false, "Unreachable code!")
+#define UNREACHABLE_MSG(...) ASSERT_MSG(false, __VA_ARGS__)
 
 #ifdef _DEBUG
 #define DEBUG_ASSERT(_a_) ASSERT(_a_)
@@ -49,3 +50,4 @@ static void assert_noinline_call(const Fn& fn) {
 #endif
 
 #define UNIMPLEMENTED() DEBUG_ASSERT_MSG(false, "Unimplemented code!")
+#define UNIMPLEMENTED_MSG(_a_, ...) ASSERT_MSG(false, _a_, __VA_ARGS__)
\ No newline at end of file
diff --git a/src/common/bit_field.h b/src/common/bit_field.h
index 371eb17a1..4748999ed 100644
--- a/src/common/bit_field.h
+++ b/src/common/bit_field.h
@@ -186,5 +186,5 @@ private:
 #pragma pack()
 
 #if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
-static_assert(std::is_trivially_copyable<BitField<0, 1, u32>>::value, "BitField must be trivially copyable");
+static_assert(std::is_trivially_copyable<BitField<0, 1, unsigned>>::value, "BitField must be trivially copyable");
 #endif
diff --git a/src/common/bit_set.h b/src/common/bit_set.h
index 85f91e786..7f5de8df2 100644
--- a/src/common/bit_set.h
+++ b/src/common/bit_set.h
@@ -7,6 +7,7 @@
 #include <intrin.h>
 #endif
 #include <initializer_list>
+#include <new>
 #include <type_traits>
 #include "common/common_types.h"
 
@@ -186,4 +187,4 @@ public:
 typedef Common::BitSet<u8> BitSet8;
 typedef Common::BitSet<u16> BitSet16;
 typedef Common::BitSet<u32> BitSet32;
-typedef Common::BitSet<u64> BitSet64;
\ No newline at end of file
+typedef Common::BitSet<u64> BitSet64;
diff --git a/src/common/code_block.h b/src/common/code_block.h
index 9ef7296d3..2fa4a0090 100644
--- a/src/common/code_block.h
+++ b/src/common/code_block.h
@@ -4,8 +4,10 @@
 
 #pragma once
 
-#include "common_types.h"
-#include "memory_util.h"
+#include <cstddef>
+
+#include "common/common_types.h"
+#include "common/memory_util.h"
 
 // Everything that needs to generate code should inherit from this.
 // You get memory management for free, plus, you can use all emitter functions without
diff --git a/src/common/common_funcs.h b/src/common/common_funcs.h
index aa6aff7b9..ab3515683 100644
--- a/src/common/common_funcs.h
+++ b/src/common/common_funcs.h
@@ -4,6 +4,10 @@
 
 #pragma once
 
+#if !defined(ARCHITECTURE_x86_64) && !defined(_M_ARM)
+#include <cstdlib> // for exit
+#endif
+
 #include "common_types.h"
 
 #define ARRAY_SIZE(a) (sizeof(a) / sizeof(a[0]))
diff --git a/src/common/file_util.cpp b/src/common/file_util.cpp
index 9ada09f8a..6e2867658 100644
--- a/src/common/file_util.cpp
+++ b/src/common/file_util.cpp
@@ -69,9 +69,10 @@ static void StripTailDirSlashes(std::string &fname)
 {
     if (fname.length() > 1)
     {
-        size_t i = fname.length() - 1;
-        while (fname[i] == DIR_SEP_CHR)
-            fname[i--] = '\0';
+        size_t i = fname.length();
+        while (i > 0 && fname[i - 1] == DIR_SEP_CHR)
+            --i;
+        fname.resize(i);
     }
     return;
 }
@@ -85,6 +86,10 @@ bool Exists(const std::string &filename)
     StripTailDirSlashes(copy);
 
 #ifdef _WIN32
+    // Windows needs a slash to identify a driver root
+    if (copy.size() != 0 && copy.back() == ':')
+        copy += DIR_SEP_CHR;
+
     int result = _wstat64(Common::UTF8ToUTF16W(copy).c_str(), &file_info);
 #else
     int result = stat64(copy.c_str(), &file_info);
@@ -102,6 +107,10 @@ bool IsDirectory(const std::string &filename)
     StripTailDirSlashes(copy);
 
 #ifdef _WIN32
+    // Windows needs a slash to identify a driver root
+    if (copy.size() != 0 && copy.back() == ':')
+        copy += DIR_SEP_CHR;
+
     int result = _wstat64(Common::UTF8ToUTF16W(copy).c_str(), &file_info);
 #else
     int result = stat64(copy.c_str(), &file_info);
@@ -824,13 +833,12 @@ size_t WriteStringToFile(bool text_file, const std::string &str, const char *fil
 
 size_t ReadFileToString(bool text_file, const char *filename, std::string &str)
 {
-    FileUtil::IOFile file(filename, text_file ? "r" : "rb");
-    auto const f = file.GetHandle();
+    IOFile file(filename, text_file ? "r" : "rb");
 
-    if (!f)
+    if (!file)
         return false;
 
-    str.resize(static_cast<u32>(GetSize(f)));
+    str.resize(static_cast<u32>(file.GetSize()));
     return file.ReadArray(&str[0], str.size());
 }
 
@@ -877,15 +885,10 @@ void SplitFilename83(const std::string& filename, std::array<char, 9>& short_nam
 }
 
 IOFile::IOFile()
-    : m_file(nullptr), m_good(true)
-{}
-
-IOFile::IOFile(std::FILE* file)
-    : m_file(file), m_good(true)
-{}
+{
+}
 
 IOFile::IOFile(const std::string& filename, const char openmode[])
-    : m_file(nullptr), m_good(true)
 {
     Open(filename, openmode);
 }
@@ -896,7 +899,6 @@ IOFile::~IOFile()
 }
 
 IOFile::IOFile(IOFile&& other)
-    : m_file(nullptr), m_good(true)
 {
     Swap(other);
 }
@@ -935,26 +937,12 @@ bool IOFile::Close()
     return m_good;
 }
 
-std::FILE* IOFile::ReleaseHandle()
-{
-    std::FILE* const ret = m_file;
-    m_file = nullptr;
-    return ret;
-}
-
-void IOFile::SetHandle(std::FILE* file)
-{
-    Close();
-    Clear();
-    m_file = file;
-}
-
-u64 IOFile::GetSize()
+u64 IOFile::GetSize() const
 {
     if (IsOpen())
         return FileUtil::GetSize(m_file);
-    else
-        return 0;
+
+    return 0;
 }
 
 bool IOFile::Seek(s64 off, int origin)
@@ -965,12 +953,12 @@ bool IOFile::Seek(s64 off, int origin)
     return m_good;
 }
 
-u64 IOFile::Tell()
+u64 IOFile::Tell() const
 {
     if (IsOpen())
         return ftello(m_file);
-    else
-        return -1;
+
+    return -1;
 }
 
 bool IOFile::Flush()
diff --git a/src/common/file_util.h b/src/common/file_util.h
index a85121aa6..c6a8694ce 100644
--- a/src/common/file_util.h
+++ b/src/common/file_util.h
@@ -7,13 +7,17 @@
 #include <array>
 #include <fstream>
 #include <functional>
-#include <cstddef>
 #include <cstdio>
 #include <string>
+#include <type_traits>
 #include <vector>
 
 #include "common/common_types.h"
 
+#ifdef _MSC_VER
+#include "common/string_util.h"
+#endif
+
 // User directory indices for GetUserPath
 enum {
     D_USER_IDX,
@@ -172,7 +176,6 @@ class IOFile : public NonCopyable
 {
 public:
     IOFile();
-    IOFile(std::FILE* file);
     IOFile(const std::string& filename, const char openmode[]);
 
     ~IOFile();
@@ -188,6 +191,11 @@ public:
     template <typename T>
     size_t ReadArray(T* data, size_t length)
     {
+        static_assert(std::is_standard_layout<T>(), "Given array does not consist of standard layout objects");
+#if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
+        static_assert(std::is_trivially_copyable<T>(), "Given array does not consist of trivially copyable objects");
+#endif
+
         if (!IsOpen()) {
             m_good = false;
             return -1;
@@ -203,9 +211,10 @@ public:
     template <typename T>
     size_t WriteArray(const T* data, size_t length)
     {
-        static_assert(std::is_standard_layout<T>::value, "Given array does not consist of standard layout objects");
-        // TODO: gcc 4.8 does not support is_trivially_copyable, but we really should check for it here.
-        //static_assert(std::is_trivially_copyable<T>::value, "Given array does not consist of trivially copyable objects");
+        static_assert(std::is_standard_layout<T>(), "Given array does not consist of standard layout objects");
+#if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
+        static_assert(std::is_trivially_copyable<T>(), "Given array does not consist of trivially copyable objects");
+#endif
 
         if (!IsOpen()) {
             m_good = false;
@@ -235,32 +244,24 @@ public:
         return WriteArray(&object, 1);
     }
 
-    bool IsOpen() { return nullptr != m_file; }
+    bool IsOpen() const { return nullptr != m_file; }
 
     // m_good is set to false when a read, write or other function fails
-    bool IsGood() {    return m_good; }
-    operator void*() { return m_good ? m_file : nullptr; }
-
-    std::FILE* ReleaseHandle();
-
-    std::FILE* GetHandle() { return m_file; }
-
-    void SetHandle(std::FILE* file);
+    bool IsGood() const { return m_good; }
+    explicit operator bool() const { return IsGood(); }
 
     bool Seek(s64 off, int origin);
-    u64 Tell();
-    u64 GetSize();
+    u64 Tell() const;
+    u64 GetSize() const;
     bool Resize(u64 size);
     bool Flush();
 
     // clear error state
     void Clear() { m_good = true; std::clearerr(m_file); }
 
-    std::FILE* m_file;
-    bool m_good;
 private:
-    IOFile(IOFile&);
-    IOFile& operator=(IOFile& other);
+    std::FILE* m_file = nullptr;
+    bool m_good = true;
 };
 
 }  // namespace
diff --git a/src/common/logging/backend.cpp b/src/common/logging/backend.cpp
index 3d39f94d5..d7008fc66 100644
--- a/src/common/logging/backend.cpp
+++ b/src/common/logging/backend.cpp
@@ -65,6 +65,7 @@ namespace Log {
         SUB(Render, OpenGL) \
         CLS(Audio) \
         SUB(Audio, DSP) \
+        SUB(Audio, Sink) \
         CLS(Loader)
 
 // GetClassName is a macro defined by Windows.h, grrr...
diff --git a/src/common/logging/log.h b/src/common/logging/log.h
index 521362317..c6910b1c7 100644
--- a/src/common/logging/log.h
+++ b/src/common/logging/log.h
@@ -78,8 +78,9 @@ enum class Class : ClassType {
     Render,                     ///< Emulator video output and hardware acceleration
     Render_Software,            ///< Software renderer backend
     Render_OpenGL,              ///< OpenGL backend
-    Audio,                      ///< Emulator audio output
+    Audio,                      ///< Audio emulation
     Audio_DSP,                  ///< The HLE implementation of the DSP
+    Audio_Sink,                 ///< Emulator audio output backend
     Loader,                     ///< ROM loader
 
     Count ///< Total number of logging classes
diff --git a/src/common/microprofile.h b/src/common/microprofile.h
index d3b6cb97c..ef312c6e1 100644
--- a/src/common/microprofile.h
+++ b/src/common/microprofile.h
@@ -4,6 +4,10 @@
 
 #pragma once
 
+// Uncomment this to disable microprofile. This will get you cleaner profiles when using
+// external sampling profilers like "Very Sleepy", and will improve performance somewhat.
+// #define MICROPROFILE_ENABLED 0
+
 // Customized Citra settings.
 // This file wraps the MicroProfile header so that these are consistent everywhere.
 #define MICROPROFILE_WEBSERVER 0
diff --git a/src/common/microprofileui.h b/src/common/microprofileui.h
index 97c369bd9..41abe6b75 100644
--- a/src/common/microprofileui.h
+++ b/src/common/microprofileui.h
@@ -13,4 +13,7 @@
 #define MICROPROFILE_HELP_ALT "Right-Click"
 #define MICROPROFILE_HELP_MOD "Ctrl"
 
+// This isn't included by microprofileui.h :(
+#include <cstdlib> // For std::abs
+
 #include <microprofileui.h>
diff --git a/src/common/profiler.cpp b/src/common/profiler.cpp
index 7792edd2f..49eb3f40c 100644
--- a/src/common/profiler.cpp
+++ b/src/common/profiler.cpp
@@ -7,71 +7,16 @@
 #include <vector>
 
 #include "common/assert.h"
-#include "common/profiler.h"
 #include "common/profiler_reporting.h"
 #include "common/synchronized_wrapper.h"
 
-#if defined(_MSC_VER) && _MSC_VER <= 1800 // MSVC 2013.
-    #define WIN32_LEAN_AND_MEAN
-    #include <Windows.h> // For QueryPerformanceCounter/Frequency
-#endif
-
 namespace Common {
 namespace Profiling {
 
-#if ENABLE_PROFILING
-thread_local Timer* Timer::current_timer = nullptr;
-#endif
-
-#if defined(_MSC_VER) && _MSC_VER <= 1800 // MSVC 2013
-QPCClock::time_point QPCClock::now() {
-    static LARGE_INTEGER freq;
-    // Use this dummy local static to ensure this gets initialized once.
-    static BOOL dummy = QueryPerformanceFrequency(&freq);
-
-    LARGE_INTEGER ticks;
-    QueryPerformanceCounter(&ticks);
-
-    // This is prone to overflow when multiplying, which is why I'm using micro instead of nano. The
-    // correct way to approach this would be to just return ticks as a time_point and then subtract
-    // and do this conversion when creating a duration from two time_points, however, as far as I
-    // could tell the C++ requirements for these types are incompatible with this approach.
-    return time_point(duration(ticks.QuadPart * std::micro::den / freq.QuadPart));
-}
-#endif
-
-TimingCategory::TimingCategory(const char* name, TimingCategory* parent)
-        : accumulated_duration(0) {
-
-    ProfilingManager& manager = GetProfilingManager();
-    category_id = manager.RegisterTimingCategory(this, name);
-    if (parent != nullptr)
-        manager.SetTimingCategoryParent(category_id, parent->category_id);
-}
-
 ProfilingManager::ProfilingManager()
         : last_frame_end(Clock::now()), this_frame_start(Clock::now()) {
 }
 
-unsigned int ProfilingManager::RegisterTimingCategory(TimingCategory* category, const char* name) {
-    TimingCategoryInfo info;
-    info.category = category;
-    info.name = name;
-    info.parent = TimingCategoryInfo::NO_PARENT;
-
-    unsigned int id = (unsigned int)timing_categories.size();
-    timing_categories.push_back(std::move(info));
-
-    return id;
-}
-
-void ProfilingManager::SetTimingCategoryParent(unsigned int category, unsigned int parent) {
-    ASSERT(category < timing_categories.size());
-    ASSERT(parent < timing_categories.size());
-
-    timing_categories[category].parent = parent;
-}
-
 void ProfilingManager::BeginFrame() {
     this_frame_start = Clock::now();
 }
@@ -82,11 +27,6 @@ void ProfilingManager::FinishFrame() {
     results.interframe_time = now - last_frame_end;
     results.frame_time = now - this_frame_start;
 
-    results.time_per_category.resize(timing_categories.size());
-    for (size_t i = 0; i < timing_categories.size(); ++i) {
-        results.time_per_category[i] = timing_categories[i].category->GetAccumulatedTime();
-    }
-
     last_frame_end = now;
 }
 
@@ -100,26 +40,9 @@ void TimingResultsAggregator::Clear() {
     window_size = cursor = 0;
 }
 
-void TimingResultsAggregator::SetNumberOfCategories(size_t n) {
-    size_t old_size = times_per_category.size();
-    if (n == old_size)
-        return;
-
-    times_per_category.resize(n);
-
-    for (size_t i = old_size; i < n; ++i) {
-        times_per_category[i].resize(max_window_size, Duration::zero());
-    }
-}
-
 void TimingResultsAggregator::AddFrame(const ProfilingFrameResult& frame_result) {
-    SetNumberOfCategories(frame_result.time_per_category.size());
-
     interframe_times[cursor] = frame_result.interframe_time;
     frame_times[cursor] = frame_result.frame_time;
-    for (size_t i = 0; i < frame_result.time_per_category.size(); ++i) {
-        times_per_category[i][cursor] = frame_result.time_per_category[i];
-    }
 
     ++cursor;
     if (cursor == max_window_size)
@@ -162,11 +85,6 @@ AggregatedFrameResult TimingResultsAggregator::GetAggregatedResults() const {
         result.fps = 0.0f;
     }
 
-    result.time_per_category.resize(times_per_category.size());
-    for (size_t i = 0; i < times_per_category.size(); ++i) {
-        result.time_per_category[i] = AggregateField(times_per_category[i], window_size);
-    }
-
     return result;
 }
 
diff --git a/src/common/profiler.h b/src/common/profiler.h
deleted file mode 100644
index 3e967b4bc..000000000
--- a/src/common/profiler.h
+++ /dev/null
@@ -1,152 +0,0 @@
-// Copyright 2015 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <atomic>
-#include <chrono>
-
-#include "common/assert.h"
-#include "common/thread.h"
-
-namespace Common {
-namespace Profiling {
-
-// If this is defined to 0, it turns all Timers into no-ops.
-#ifndef ENABLE_PROFILING
-#define ENABLE_PROFILING 1
-#endif
-
-#if defined(_MSC_VER) && _MSC_VER <= 1800 // MSVC 2013
-// MSVC up to 2013 doesn't use QueryPerformanceCounter for high_resolution_clock, so it has bad
-// precision. We manually implement a clock based on QPC to get good results.
-
-struct QPCClock {
-    using duration = std::chrono::microseconds;
-    using time_point = std::chrono::time_point<QPCClock>;
-    using rep = duration::rep;
-    using period = duration::period;
-    static const bool is_steady = false;
-
-    static time_point now();
-};
-
-using Clock = QPCClock;
-#else
-using Clock = std::chrono::high_resolution_clock;
-#endif
-
-using Duration = Clock::duration;
-
-/**
- * Represents a timing category that measured time can be accounted towards. Should be declared as a
- * global variable and passed to Timers.
- */
-class TimingCategory final {
-public:
-    TimingCategory(const char* name, TimingCategory* parent = nullptr);
-
-    unsigned int GetCategoryId() const {
-        return category_id;
-    }
-
-    /// Adds some time to this category. Can safely be called from multiple threads at the same time.
-    void AddTime(Duration amount) {
-        std::atomic_fetch_add_explicit(
-                &accumulated_duration, amount.count(),
-                std::memory_order_relaxed);
-    }
-
-    /**
-     * Atomically retrieves the accumulated measured time for this category and resets the counter
-     * to zero. Can be safely called concurrently with AddTime.
-     */
-    Duration GetAccumulatedTime() {
-        return Duration(std::atomic_exchange_explicit(
-                &accumulated_duration, (Duration::rep)0,
-                std::memory_order_relaxed));
-    }
-
-private:
-    unsigned int category_id;
-    std::atomic<Duration::rep> accumulated_duration;
-};
-
-/**
- * Measures time elapsed between a call to Start and a call to Stop and attributes it to the given
- * TimingCategory. Start/Stop can be called multiple times on the same timer, but each call must be
- * appropriately paired.
- *
- * When a Timer is started, it automatically pauses a previously running timer on the same thread,
- * which is resumed when it is stopped. As such, no special action needs to be taken to avoid
- * double-accounting of time on two categories.
- */
-class Timer {
-public:
-    Timer(TimingCategory& category) : category(category) {
-    }
-
-    void Start() {
-#if ENABLE_PROFILING
-        ASSERT(!running);
-        previous_timer = current_timer;
-        current_timer = this;
-        if (previous_timer != nullptr)
-            previous_timer->StopTiming();
-
-        StartTiming();
-#endif
-    }
-
-    void Stop() {
-#if ENABLE_PROFILING
-        ASSERT(running);
-        StopTiming();
-
-        if (previous_timer != nullptr)
-            previous_timer->StartTiming();
-        current_timer = previous_timer;
-#endif
-    }
-
-private:
-#if ENABLE_PROFILING
-    void StartTiming() {
-        start = Clock::now();
-        running = true;
-    }
-
-    void StopTiming() {
-        auto duration = Clock::now() - start;
-        running = false;
-        category.AddTime(std::chrono::duration_cast<Duration>(duration));
-    }
-
-    Clock::time_point start;
-    bool running = false;
-
-    Timer* previous_timer;
-    static thread_local Timer* current_timer;
-#endif
-
-    TimingCategory& category;
-};
-
-/**
- * A Timer that automatically starts timing when created and stops at the end of the scope. Should
- * be used in the majority of cases.
- */
-class ScopeTimer : public Timer {
-public:
-    ScopeTimer(TimingCategory& category) : Timer(category) {
-        Start();
-    }
-
-    ~ScopeTimer() {
-        Stop();
-    }
-};
-
-} // namespace Profiling
-} // namespace Common
diff --git a/src/common/profiler_reporting.h b/src/common/profiler_reporting.h
index df98e05b7..fa1ac883f 100644
--- a/src/common/profiler_reporting.h
+++ b/src/common/profiler_reporting.h
@@ -4,22 +4,17 @@
 
 #pragma once
 
+#include <chrono>
 #include <cstddef>
 #include <vector>
 
-#include "common/profiler.h"
 #include "common/synchronized_wrapper.h"
 
 namespace Common {
 namespace Profiling {
 
-struct TimingCategoryInfo {
-    static const unsigned int NO_PARENT = -1;
-
-    TimingCategory* category;
-    const char* name;
-    unsigned int parent;
-};
+using Clock = std::chrono::high_resolution_clock;
+using Duration = Clock::duration;
 
 struct ProfilingFrameResult {
     /// Time since the last delivered frame
@@ -27,22 +22,12 @@ struct ProfilingFrameResult {
 
     /// Time spent processing a frame, excluding VSync
     Duration frame_time;
-
-    /// Total amount of time spent inside each category in this frame. Indexed by the category id
-    std::vector<Duration> time_per_category;
 };
 
 class ProfilingManager final {
 public:
     ProfilingManager();
 
-    unsigned int RegisterTimingCategory(TimingCategory* category, const char* name);
-    void SetTimingCategoryParent(unsigned int category, unsigned int parent);
-
-    const std::vector<TimingCategoryInfo>& GetTimingCategoriesInfo() const {
-        return timing_categories;
-    }
-
     /// This should be called after swapping screen buffers.
     void BeginFrame();
     /// This should be called before swapping screen buffers.
@@ -54,7 +39,6 @@ public:
     }
 
 private:
-    std::vector<TimingCategoryInfo> timing_categories;
     Clock::time_point last_frame_end;
     Clock::time_point this_frame_start;
 
@@ -73,9 +57,6 @@ struct AggregatedFrameResult {
     AggregatedDuration frame_time;
 
     float fps;
-
-    /// Total amount of time spent inside each category in this frame. Indexed by the category id
-    std::vector<AggregatedDuration> time_per_category;
 };
 
 class TimingResultsAggregator final {
@@ -83,7 +64,6 @@ public:
     TimingResultsAggregator(size_t window_size);
 
     void Clear();
-    void SetNumberOfCategories(size_t n);
 
     void AddFrame(const ProfilingFrameResult& frame_result);
 
@@ -95,7 +75,6 @@ public:
 
     std::vector<Duration> interframe_times;
     std::vector<Duration> frame_times;
-    std::vector<std::vector<Duration>> times_per_category;
 };
 
 ProfilingManager& GetProfilingManager();
diff --git a/src/common/swap.h b/src/common/swap.h
index a7c37bc44..1749bd7a4 100644
--- a/src/common/swap.h
+++ b/src/common/swap.h
@@ -25,6 +25,8 @@
     #include <sys/endian.h>
 #endif
 
+#include <cstring>
+
 #include "common/common_types.h"
 
 // GCC 4.6+
@@ -58,9 +60,6 @@
 
 namespace Common {
 
-inline u8 swap8(u8 _data) {return _data;}
-inline u32 swap24(const u8* _data) {return (_data[0] << 16) | (_data[1] << 8) | _data[2];}
-
 #ifdef _MSC_VER
 inline u16 swap16(u16 _data) {return _byteswap_ushort(_data);}
 inline u32 swap32(u32 _data) {return _byteswap_ulong (_data);}
@@ -92,52 +91,29 @@ inline u64 swap64(u64 data) {return ((u64)swap32(data) << 32) | swap32(data >> 3
 #endif
 
 inline float swapf(float f) {
-    union {
-        float f;
-        unsigned int u32;
-    } dat1, dat2;
-
-    dat1.f = f;
-    dat2.u32 = swap32(dat1.u32);
+    static_assert(sizeof(u32) == sizeof(float),
+                  "float must be the same size as uint32_t.");
 
-    return dat2.f;
-}
-
-inline double swapd(double f) {
-    union  {
-        double f;
-        unsigned long long u64;
-    } dat1, dat2;
+    u32 value;
+    std::memcpy(&value, &f, sizeof(u32));
 
-    dat1.f = f;
-    dat2.u64 = swap64(dat1.u64);
+    value = swap32(value);
+    std::memcpy(&f, &value, sizeof(u32));
 
-    return dat2.f;
+    return f;
 }
 
-inline u16 swap16(const u8* _pData) {return swap16(*(const u16*)_pData);}
-inline u32 swap32(const u8* _pData) {return swap32(*(const u32*)_pData);}
-inline u64 swap64(const u8* _pData) {return swap64(*(const u64*)_pData);}
-
-template <int count>
-void swap(u8*);
+inline double swapd(double f) {
+    static_assert(sizeof(u64) == sizeof(double),
+                  "double must be the same size as uint64_t.");
 
-template <>
-inline void swap<1>(u8* data) { }
+    u64 value;
+    std::memcpy(&value, &f, sizeof(u64));
 
-template <>
-inline void swap<2>(u8* data) {
-    *reinterpret_cast<u16*>(data) = swap16(data);
-}
-
-template <>
-inline void swap<4>(u8* data) {
-    *reinterpret_cast<u32*>(data) = swap32(data);
-}
+    value = swap64(value);
+    std::memcpy(&f, &value, sizeof(u64));
 
-template <>
-inline void swap<8>(u8* data) {
-    *reinterpret_cast<u64*>(data) = swap64(data);
+    return f;
 }
 
 }  // Namespace Common
@@ -534,35 +510,35 @@ bool operator==(const S &p, const swap_struct_t<T, F> v) {
 template <typename T>
 struct swap_64_t {
     static T swap(T x) {
-        return (T)Common::swap64(*(u64 *)&x);
+        return static_cast<T>(Common::swap64(x));
     }
 };
 
 template <typename T>
 struct swap_32_t {
     static T swap(T x) {
-        return (T)Common::swap32(*(u32 *)&x);
+        return static_cast<T>(Common::swap32(x));
     }
 };
 
 template <typename T>
 struct swap_16_t {
     static T swap(T x) {
-        return (T)Common::swap16(*(u16 *)&x);
+        return static_cast<T>(Common::swap16(x));
     }
 };
 
 template <typename T>
 struct swap_float_t {
     static T swap(T x) {
-        return (T)Common::swapf(*(float *)&x);
+        return static_cast<T>(Common::swapf(x));
     }
 };
 
 template <typename T>
 struct swap_double_t {
     static T swap(T x) {
-        return (T)Common::swapd(*(double *)&x);
+        return static_cast<T>(Common::swapd(x));
     }
 };
 
diff --git a/src/common/thread.h b/src/common/thread.h
index 8255ee6d3..bbfa8befa 100644
--- a/src/common/thread.h
+++ b/src/common/thread.h
@@ -30,8 +30,7 @@
 #   endif
 #endif
 
-namespace Common
-{
+namespace Common {
 
 int CurrentThreadId();
 
@@ -43,55 +42,55 @@ public:
     Event() : is_set(false) {}
 
     void Set() {
-        std::lock_guard<std::mutex> lk(m_mutex);
+        std::lock_guard<std::mutex> lk(mutex);
         if (!is_set) {
             is_set = true;
-            m_condvar.notify_one();
+            condvar.notify_one();
         }
     }
 
     void Wait() {
-        std::unique_lock<std::mutex> lk(m_mutex);
-        m_condvar.wait(lk, [&]{ return is_set; });
+        std::unique_lock<std::mutex> lk(mutex);
+        condvar.wait(lk, [&]{ return is_set; });
         is_set = false;
     }
 
     void Reset() {
-        std::unique_lock<std::mutex> lk(m_mutex);
+        std::unique_lock<std::mutex> lk(mutex);
         // no other action required, since wait loops on the predicate and any lingering signal will get cleared on the first iteration
         is_set = false;
     }
 
 private:
     bool is_set;
-    std::condition_variable m_condvar;
-    std::mutex m_mutex;
+    std::condition_variable condvar;
+    std::mutex mutex;
 };
 
 class Barrier {
 public:
-    Barrier(size_t count) : m_count(count), m_waiting(0) {}
+    explicit Barrier(size_t count_) : count(count_), waiting(0), generation(0) {}
 
     /// Blocks until all "count" threads have called Sync()
     void Sync() {
-        std::unique_lock<std::mutex> lk(m_mutex);
+        std::unique_lock<std::mutex> lk(mutex);
+        const size_t current_generation = generation;
 
-        // TODO: broken when next round of Sync()s
-        // is entered before all waiting threads return from the notify_all
-
-        if (++m_waiting == m_count) {
-            m_waiting = 0;
-            m_condvar.notify_all();
+        if (++waiting == count) {
+            generation++;
+            waiting = 0;
+            condvar.notify_all();
         } else {
-            m_condvar.wait(lk, [&]{ return m_waiting == 0; });
+            condvar.wait(lk, [this, current_generation]{ return current_generation != generation; });
         }
     }
 
 private:
-    std::condition_variable m_condvar;
-    std::mutex m_mutex;
-    const size_t m_count;
-    size_t m_waiting;
+    std::condition_variable condvar;
+    std::mutex mutex;
+    const size_t count;
+    size_t waiting;
+    size_t generation; // Incremented once each time the barrier is used
 };
 
 void SleepCurrentThread(int ms);
@@ -100,8 +99,7 @@ void SwitchCurrentThread();    // On Linux, this is equal to sleep 1ms
 // Use this function during a spin-wait to make the current thread
 // relax while another thread is working. This may be more efficient
 // than using events because event functions use kernel calls.
-inline void YieldCPU()
-{
+inline void YieldCPU() {
     std::this_thread::yield();
 }
 
diff --git a/src/common/x64/emitter.cpp b/src/common/x64/emitter.cpp
index 1dcf2416c..5662f7f86 100644
--- a/src/common/x64/emitter.cpp
+++ b/src/common/x64/emitter.cpp
@@ -455,6 +455,18 @@ void XEmitter::CALL(const void* fnptr)
     Write32(u32(distance));
 }
 
+FixupBranch XEmitter::CALL()
+{
+    FixupBranch branch;
+    branch.type = 1;
+    branch.ptr = code + 5;
+
+    Write8(0xE8);
+    Write32(0);
+
+    return branch;
+}
+
 FixupBranch XEmitter::J(bool force5bytes)
 {
     FixupBranch branch;
@@ -531,6 +543,22 @@ void XEmitter::SetJumpTarget(const FixupBranch& branch)
     }
 }
 
+void XEmitter::SetJumpTarget(const FixupBranch& branch, const u8* target)
+{
+    if (branch.type == 0)
+    {
+        s64 distance = (s64)(target - branch.ptr);
+        ASSERT_MSG(distance >= -0x80 && distance < 0x80, "Jump target too far away, needs force5Bytes = true");
+        branch.ptr[-1] = (u8)(s8)distance;
+    }
+    else if (branch.type == 1)
+    {
+        s64 distance = (s64)(target - branch.ptr);
+        ASSERT_MSG(distance >= -0x80000000LL && distance < 0x80000000LL, "Jump target too far away, needs indirect register");
+        ((s32*)branch.ptr)[-1] = (s32)distance;
+    }
+}
+
 //Single byte opcodes
 //There is no PUSHAD/POPAD in 64-bit mode.
 void XEmitter::INT3() {Write8(0xCC);}
diff --git a/src/common/x64/emitter.h b/src/common/x64/emitter.h
index 7c6548fb5..60a77dfe1 100644
--- a/src/common/x64/emitter.h
+++ b/src/common/x64/emitter.h
@@ -17,6 +17,8 @@
 
 #pragma once
 
+#include <cstddef>
+
 #include "common/assert.h"
 #include "common/bit_set.h"
 #include "common/common_types.h"
@@ -425,12 +427,14 @@ public:
 #undef CALL
 #endif
     void CALL(const void* fnptr);
+    FixupBranch CALL();
     void CALLptr(OpArg arg);
 
     FixupBranch J_CC(CCFlags conditionCode, bool force5bytes = false);
     void J_CC(CCFlags conditionCode, const u8* addr, bool force5Bytes = false);
 
     void SetJumpTarget(const FixupBranch& branch);
+    void SetJumpTarget(const FixupBranch& branch, const u8* target);
 
     void SETcc(CCFlags flag, OpArg dest);
     // Note: CMOV brings small if any benefit on current cpus.