10 files changed, 290 insertions, 79 deletions

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 850ce8006..1e8e1b215 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -91,6 +91,8 @@ add_library(common STATIC
     logging/log.h
     logging/text_formatter.cpp
     logging/text_formatter.h
+    lz4_compression.cpp
+    lz4_compression.h
     math_util.h
     memory_hook.cpp
     memory_hook.h
@@ -123,6 +125,8 @@ add_library(common STATIC
     uint128.h
     vector_math.h
     web_result.h
+    zstd_compression.cpp
+    zstd_compression.h
 )
 
 if(ARCHITECTURE_x86_64)
@@ -136,3 +140,4 @@ endif()
 create_target_directory_groups(common)
 
 target_link_libraries(common PUBLIC Boost::boost fmt microprofile)
+target_link_libraries(common PRIVATE lz4_static libzstd_static)
diff --git a/src/common/assert.h b/src/common/assert.h
index 6002f7ab1..4b0e3f64e 100644
--- a/src/common/assert.h
+++ b/src/common/assert.h
@@ -57,3 +57,21 @@ __declspec(noinline, noreturn)
 
 #define UNIMPLEMENTED_IF(cond) ASSERT_MSG(!(cond), "Unimplemented code!")
 #define UNIMPLEMENTED_IF_MSG(cond, ...) ASSERT_MSG(!(cond), __VA_ARGS__)
+
+// If the assert is ignored, execute _b_
+#define ASSERT_OR_EXECUTE(_a_, _b_)                                                                \
+    do {                                                                                           \
+        ASSERT(_a_);                                                                               \
+        if (!(_a_)) {                                                                              \
+            _b_                                                                                    \
+        }                                                                                          \
+    } while (0)
+
+// If the assert is ignored, execute _b_
+#define ASSERT_OR_EXECUTE_MSG(_a_, _b_, ...)                                                       \
+    do {                                                                                           \
+        ASSERT_MSG(_a_, __VA_ARGS__);                                                              \
+        if (!(_a_)) {                                                                              \
+            _b_                                                                                    \
+        }                                                                                          \
+    } while (0)
diff --git a/src/common/bit_util.h b/src/common/bit_util.h
index a4f9ed4aa..d032df413 100644
--- a/src/common/bit_util.h
+++ b/src/common/bit_util.h
@@ -32,7 +32,7 @@ inline u32 CountLeadingZeroes32(u32 value) {
     return 32;
 }
 
-inline u64 CountLeadingZeroes64(u64 value) {
+inline u32 CountLeadingZeroes64(u64 value) {
     unsigned long leading_zero = 0;
 
     if (_BitScanReverse64(&leading_zero, value) != 0) {
@@ -47,15 +47,15 @@ inline u32 CountLeadingZeroes32(u32 value) {
         return 32;
     }
 
-    return __builtin_clz(value);
+    return static_cast<u32>(__builtin_clz(value));
 }
 
-inline u64 CountLeadingZeroes64(u64 value) {
+inline u32 CountLeadingZeroes64(u64 value) {
     if (value == 0) {
         return 64;
     }
 
-    return __builtin_clzll(value);
+    return static_cast<u32>(__builtin_clzll(value));
 }
 #endif
 
@@ -70,7 +70,7 @@ inline u32 CountTrailingZeroes32(u32 value) {
     return 32;
 }
 
-inline u64 CountTrailingZeroes64(u64 value) {
+inline u32 CountTrailingZeroes64(u64 value) {
     unsigned long trailing_zero = 0;
 
     if (_BitScanForward64(&trailing_zero, value) != 0) {
@@ -85,15 +85,15 @@ inline u32 CountTrailingZeroes32(u32 value) {
         return 32;
     }
 
-    return __builtin_ctz(value);
+    return static_cast<u32>(__builtin_ctz(value));
 }
 
-inline u64 CountTrailingZeroes64(u64 value) {
+inline u32 CountTrailingZeroes64(u64 value) {
     if (value == 0) {
         return 64;
     }
 
-    return __builtin_ctzll(value);
+    return static_cast<u32>(__builtin_ctzll(value));
 }
 #endif
 
diff --git a/src/common/lz4_compression.cpp b/src/common/lz4_compression.cpp
new file mode 100644
index 000000000..ade6759bb
--- /dev/null
+++ b/src/common/lz4_compression.cpp
@@ -0,0 +1,76 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <lz4hc.h>
+
+#include "common/assert.h"
+#include "common/lz4_compression.h"
+
+namespace Common::Compression {
+
+std::vector<u8> CompressDataLZ4(const u8* source, std::size_t source_size) {
+    ASSERT_MSG(source_size <= LZ4_MAX_INPUT_SIZE, "Source size exceeds LZ4 maximum input size");
+
+    const auto source_size_int = static_cast<int>(source_size);
+    const int max_compressed_size = LZ4_compressBound(source_size_int);
+    std::vector<u8> compressed(max_compressed_size);
+
+    const int compressed_size = LZ4_compress_default(reinterpret_cast<const char*>(source),
+                                                     reinterpret_cast<char*>(compressed.data()),
+                                                     source_size_int, max_compressed_size);
+
+    if (compressed_size <= 0) {
+        // Compression failed
+        return {};
+    }
+
+    compressed.resize(compressed_size);
+
+    return compressed;
+}
+
+std::vector<u8> CompressDataLZ4HC(const u8* source, std::size_t source_size,
+                                  s32 compression_level) {
+    ASSERT_MSG(source_size <= LZ4_MAX_INPUT_SIZE, "Source size exceeds LZ4 maximum input size");
+
+    compression_level = std::clamp(compression_level, LZ4HC_CLEVEL_MIN, LZ4HC_CLEVEL_MAX);
+
+    const auto source_size_int = static_cast<int>(source_size);
+    const int max_compressed_size = LZ4_compressBound(source_size_int);
+    std::vector<u8> compressed(max_compressed_size);
+
+    const int compressed_size = LZ4_compress_HC(
+        reinterpret_cast<const char*>(source), reinterpret_cast<char*>(compressed.data()),
+        source_size_int, max_compressed_size, compression_level);
+
+    if (compressed_size <= 0) {
+        // Compression failed
+        return {};
+    }
+
+    compressed.resize(compressed_size);
+
+    return compressed;
+}
+
+std::vector<u8> CompressDataLZ4HCMax(const u8* source, std::size_t source_size) {
+    return CompressDataLZ4HC(source, source_size, LZ4HC_CLEVEL_MAX);
+}
+
+std::vector<u8> DecompressDataLZ4(const std::vector<u8>& compressed,
+                                  std::size_t uncompressed_size) {
+    std::vector<u8> uncompressed(uncompressed_size);
+    const int size_check = LZ4_decompress_safe(reinterpret_cast<const char*>(compressed.data()),
+                                               reinterpret_cast<char*>(uncompressed.data()),
+                                               static_cast<int>(compressed.size()),
+                                               static_cast<int>(uncompressed.size()));
+    if (static_cast<int>(uncompressed_size) != size_check) {
+        // Decompression failed
+        return {};
+    }
+    return uncompressed;
+}
+
+} // namespace Common::Compression
diff --git a/src/common/lz4_compression.h b/src/common/lz4_compression.h
new file mode 100644
index 000000000..4c16f6e03
--- /dev/null
+++ b/src/common/lz4_compression.h
@@ -0,0 +1,57 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+
+#include "common/common_types.h"
+
+namespace Common::Compression {
+
+/**
+ * Compresses a source memory region with LZ4 and returns the compressed data in a vector.
+ *
+ * @param source the uncompressed source memory region.
+ * @param source_size the size in bytes of the uncompressed source memory region.
+ *
+ * @return the compressed data.
+ */
+std::vector<u8> CompressDataLZ4(const u8* source, std::size_t source_size);
+
+/**
+ * Utilizes the LZ4 subalgorithm LZ4HC with the specified compression level. Higher compression
+ * levels result in a smaller compressed size, but require more CPU time for compression. The
+ * compression level has almost no impact on decompression speed. Data compressed with LZ4HC can
+ * also be decompressed with the default LZ4 decompression.
+ *
+ * @param source the uncompressed source memory region.
+ * @param source_size the size in bytes of the uncompressed source memory region.
+ * @param compression_level the used compression level. Should be between 3 and 12.
+ *
+ * @return the compressed data.
+ */
+std::vector<u8> CompressDataLZ4HC(const u8* source, std::size_t source_size, s32 compression_level);
+
+/**
+ * Utilizes the LZ4 subalgorithm LZ4HC with the highest possible compression level.
+ *
+ * @param source the uncompressed source memory region.
+ * @param source_size the size in bytes of the uncompressed source memory region.
+ *
+ * @return the compressed data.
+ */
+std::vector<u8> CompressDataLZ4HCMax(const u8* source, std::size_t source_size);
+
+/**
+ * Decompresses a source memory region with LZ4 and returns the uncompressed data in a vector.
+ *
+ * @param compressed the compressed source memory region.
+ * @param uncompressed_size the size in bytes of the uncompressed data.
+ *
+ * @return the decompressed data.
+ */
+std::vector<u8> DecompressDataLZ4(const std::vector<u8>& compressed, std::size_t uncompressed_size);
+
+} // namespace Common::Compression
\ No newline at end of file
diff --git a/src/common/multi_level_queue.h b/src/common/multi_level_queue.h
index 2b61b91e0..9cb448f56 100644
--- a/src/common/multi_level_queue.h
+++ b/src/common/multi_level_queue.h
@@ -72,7 +72,7 @@ public:
                 u64 prios = mlq.used_priorities;
                 prios &= ~((1ULL << (current_priority + 1)) - 1);
                 if (prios == 0) {
-                    current_priority = mlq.depth();
+                    current_priority = static_cast<u32>(mlq.depth());
                 } else {
                     current_priority = CountTrailingZeroes64(prios);
                     it = GetBeginItForPrio();
diff --git a/src/common/scope_exit.h b/src/common/scope_exit.h
index baf1f1c9e..1176a72b1 100644
--- a/src/common/scope_exit.h
+++ b/src/common/scope_exit.h
@@ -20,7 +20,7 @@ struct ScopeExitHelper {
 
 template <typename Func>
 ScopeExitHelper<Func> ScopeExit(Func&& func) {
-    return ScopeExitHelper<Func>(std::move(func));
+    return ScopeExitHelper<Func>(std::forward<Func>(func));
 }
 } // namespace detail
 
diff --git a/src/common/swap.h b/src/common/swap.h
index b3eab1324..71932c2bb 100644
--- a/src/common/swap.h
+++ b/src/common/swap.h
@@ -21,11 +21,6 @@
 
 #if defined(_MSC_VER)
 #include <cstdlib>
-#elif defined(__linux__)
-#include <byteswap.h>
-#elif defined(__Bitrig__) || defined(__DragonFly__) || defined(__FreeBSD__) ||                     \
-    defined(__NetBSD__) || defined(__OpenBSD__)
-#include <sys/endian.h>
 #endif
 #include <cstring>
 #include "common/common_types.h"
@@ -62,86 +57,49 @@
 namespace Common {
 
 #ifdef _MSC_VER
-inline u16 swap16(u16 _data) {
-    return _byteswap_ushort(_data);
+[[nodiscard]] inline u16 swap16(u16 data) noexcept {
+    return _byteswap_ushort(data);
 }
-inline u32 swap32(u32 _data) {
-    return _byteswap_ulong(_data);
+[[nodiscard]] inline u32 swap32(u32 data) noexcept {
+    return _byteswap_ulong(data);
 }
-inline u64 swap64(u64 _data) {
-    return _byteswap_uint64(_data);
+[[nodiscard]] inline u64 swap64(u64 data) noexcept {
+    return _byteswap_uint64(data);
 }
-#elif defined(ARCHITECTURE_ARM) && (__ARM_ARCH >= 6)
-inline u16 swap16(u16 _data) {
-    u32 data = _data;
-    __asm__("rev16 %0, %1\n" : "=l"(data) : "l"(data));
-    return (u16)data;
-}
-inline u32 swap32(u32 _data) {
-    __asm__("rev %0, %1\n" : "=l"(_data) : "l"(_data));
-    return _data;
-}
-inline u64 swap64(u64 _data) {
-    return ((u64)swap32(_data) << 32) | swap32(_data >> 32);
-}
-#elif __linux__
-inline u16 swap16(u16 _data) {
-    return bswap_16(_data);
-}
-inline u32 swap32(u32 _data) {
-    return bswap_32(_data);
-}
-inline u64 swap64(u64 _data) {
-    return bswap_64(_data);
-}
-#elif __APPLE__
-inline __attribute__((always_inline)) u16 swap16(u16 _data) {
-    return (_data >> 8) | (_data << 8);
-}
-inline __attribute__((always_inline)) u32 swap32(u32 _data) {
-    return __builtin_bswap32(_data);
-}
-inline __attribute__((always_inline)) u64 swap64(u64 _data) {
-    return __builtin_bswap64(_data);
-}
-#elif defined(__Bitrig__) || defined(__OpenBSD__)
+#elif defined(__clang__) || defined(__GNUC__)
+#if defined(__Bitrig__) || defined(__OpenBSD__)
 // redefine swap16, swap32, swap64 as inline functions
 #undef swap16
 #undef swap32
 #undef swap64
-inline u16 swap16(u16 _data) {
-    return __swap16(_data);
-}
-inline u32 swap32(u32 _data) {
-    return __swap32(_data);
-}
-inline u64 swap64(u64 _data) {
-    return __swap64(_data);
-}
-#elif defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__)
-inline u16 swap16(u16 _data) {
-    return bswap16(_data);
+#endif
+[[nodiscard]] inline u16 swap16(u16 data) noexcept {
+    return __builtin_bswap16(data);
 }
-inline u32 swap32(u32 _data) {
-    return bswap32(_data);
+[[nodiscard]] inline u32 swap32(u32 data) noexcept {
+    return __builtin_bswap32(data);
 }
-inline u64 swap64(u64 _data) {
-    return bswap64(_data);
+[[nodiscard]] inline u64 swap64(u64 data) noexcept {
+    return __builtin_bswap64(data);
 }
 #else
-// Slow generic implementation.
-inline u16 swap16(u16 data) {
+// Generic implementation.
+[[nodiscard]] inline u16 swap16(u16 data) noexcept {
     return (data >> 8) | (data << 8);
 }
-inline u32 swap32(u32 data) {
-    return (swap16(data) << 16) | swap16(data >> 16);
+[[nodiscard]] inline u32 swap32(u32 data) noexcept {
+    return ((data & 0xFF000000U) >> 24) | ((data & 0x00FF0000U) >> 8) |
+           ((data & 0x0000FF00U) << 8) | ((data & 0x000000FFU) << 24);
 }
-inline u64 swap64(u64 data) {
-    return ((u64)swap32(data) << 32) | swap32(data >> 32);
+[[nodiscard]] inline u64 swap64(u64 data) noexcept {
+    return ((data & 0xFF00000000000000ULL) >> 56) | ((data & 0x00FF000000000000ULL) >> 40) |
+           ((data & 0x0000FF0000000000ULL) >> 24) | ((data & 0x000000FF00000000ULL) >> 8) |
+           ((data & 0x00000000FF000000ULL) << 8) | ((data & 0x0000000000FF0000ULL) << 24) |
+           ((data & 0x000000000000FF00ULL) << 40) | ((data & 0x00000000000000FFULL) << 56);
 }
 #endif
 
-inline float swapf(float f) {
+[[nodiscard]] inline float swapf(float f) noexcept {
     static_assert(sizeof(u32) == sizeof(float), "float must be the same size as uint32_t.");
 
     u32 value;
@@ -153,7 +111,7 @@ inline float swapf(float f) {
     return f;
 }
 
-inline double swapd(double f) {
+[[nodiscard]] inline double swapd(double f) noexcept {
     static_assert(sizeof(u64) == sizeof(double), "double must be the same size as uint64_t.");
 
     u64 value;
diff --git a/src/common/zstd_compression.cpp b/src/common/zstd_compression.cpp
new file mode 100644
index 000000000..60a35c67c
--- /dev/null
+++ b/src/common/zstd_compression.cpp
@@ -0,0 +1,53 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <zstd.h>
+
+#include "common/assert.h"
+#include "common/zstd_compression.h"
+
+namespace Common::Compression {
+
+std::vector<u8> CompressDataZSTD(const u8* source, std::size_t source_size, s32 compression_level) {
+    compression_level = std::clamp(compression_level, 1, ZSTD_maxCLevel());
+
+    const std::size_t max_compressed_size = ZSTD_compressBound(source_size);
+    std::vector<u8> compressed(max_compressed_size);
+
+    const std::size_t compressed_size =
+        ZSTD_compress(compressed.data(), compressed.size(), source, source_size, compression_level);
+
+    if (ZSTD_isError(compressed_size)) {
+        // Compression failed
+        return {};
+    }
+
+    compressed.resize(compressed_size);
+
+    return compressed;
+}
+
+std::vector<u8> CompressDataZSTDDefault(const u8* source, std::size_t source_size) {
+    return CompressDataZSTD(source, source_size, ZSTD_CLEVEL_DEFAULT);
+}
+
+std::vector<u8> DecompressDataZSTD(const std::vector<u8>& compressed) {
+    const std::size_t decompressed_size =
+        ZSTD_getDecompressedSize(compressed.data(), compressed.size());
+    std::vector<u8> decompressed(decompressed_size);
+
+    const std::size_t uncompressed_result_size = ZSTD_decompress(
+        decompressed.data(), decompressed.size(), compressed.data(), compressed.size());
+
+    if (decompressed_size != uncompressed_result_size || ZSTD_isError(uncompressed_result_size)) {
+        // Decompression failed
+        return {};
+    }
+    return decompressed;
+}
+
+} // namespace Common::Compression
diff --git a/src/common/zstd_compression.h b/src/common/zstd_compression.h
new file mode 100644
index 000000000..e9de941c8
--- /dev/null
+++ b/src/common/zstd_compression.h
@@ -0,0 +1,44 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+
+#include "common/common_types.h"
+
+namespace Common::Compression {
+
+/**
+ * Compresses a source memory region with Zstandard and returns the compressed data in a vector.
+ *
+ * @param source the uncompressed source memory region.
+ * @param source_size the size in bytes of the uncompressed source memory region.
+ * @param compression_level the used compression level. Should be between 1 and 22.
+ *
+ * @return the compressed data.
+ */
+std::vector<u8> CompressDataZSTD(const u8* source, std::size_t source_size, s32 compression_level);
+
+/**
+ * Compresses a source memory region with Zstandard with the default compression level and returns
+ * the compressed data in a vector.
+ *
+ * @param source the uncompressed source memory region.
+ * @param source_size the size in bytes of the uncompressed source memory region.
+ *
+ * @return the compressed data.
+ */
+std::vector<u8> CompressDataZSTDDefault(const u8* source, std::size_t source_size);
+
+/**
+ * Decompresses a source memory region with Zstandard and returns the uncompressed data in a vector.
+ *
+ * @param compressed the compressed source memory region.
+ *
+ * @return the decompressed data.
+ */
+std::vector<u8> DecompressDataZSTD(const std::vector<u8>& compressed);
+
+} // namespace Common::Compression
\ No newline at end of file