diff --git a/src/core/bitmath_func.hpp b/src/core/bitmath_func.hpp
index 31e679b005..a40c7dba6e 100644
--- a/src/core/bitmath_func.hpp
+++ b/src/core/bitmath_func.hpp
@@ -365,13 +365,32 @@ static inline T ROR(const T x, const uint8 n)
 	 * (since it will use hardware swapping if available).
 	 * Even though they should return uint16 and uint32, we get
 	 * warnings if we don't cast those (why?) */
+	#define BSWAP64(x) ((uint64)CFSwapInt64(x))
 	#define BSWAP32(x) ((uint32)CFSwapInt32(x))
 	#define BSWAP16(x) ((uint16)CFSwapInt16(x))
 #elif defined(_MSC_VER)
 	/* MSVC has intrinsics for swapping, resulting in faster code */
+	#define BSWAP64(x) (_byteswap_uint64(x))
 	#define BSWAP32(x) (_byteswap_ulong(x))
 	#define BSWAP16(x) (_byteswap_ushort(x))
 #else
+	/**
+	 * Perform a 64 bits endianness bitswap on x.
+	 * @param x the variable to bitswap
+	 * @return the bitswapped value.
+	 */
+	static inline uint64 BSWAP64(uint64 x)
+	{
+#if !defined(__ICC) && (defined(__GNUC__) || defined(__clang__))
+		/* GCC >= 4.3 provides a builtin, resulting in faster code */
+		return (uint64)__builtin_bswap64((uint64)x);
+#else
+		return ((x >> 56) & 0xFFULL) | ((x >> 40) & 0xFF00ULL) | ((x >> 24) & 0xFF0000ULL) | ((x >> 8) & 0xFF000000ULL) |
+				((x << 8) & 0xFF00000000ULL) | ((x << 24) & 0xFF0000000000ULL) | ((x << 40) & 0xFF000000000000ULL) | ((x << 56) & 0xFF000000000000ULL);
+				;
+#endif /* __GNUC__ || __clang__ */
+	}
+
 	/**
 	 * Perform a 32 bits endianness bitswap on x.
 	 * @param x the variable to bitswap
diff --git a/src/core/endian_func.hpp b/src/core/endian_func.hpp
index ab5b181500..48060bd1ad 100644
--- a/src/core/endian_func.hpp
+++ b/src/core/endian_func.hpp
@@ -19,25 +19,33 @@
 #if TTD_ENDIAN == TTD_BIG_ENDIAN
 	#define FROM_BE16(x) (x)
 	#define FROM_BE32(x) (x)
+	#define FROM_BE64(x) (x)
 	#define TO_BE16(x)   (x)
 	#define TO_BE32(x)   (x)
 	#define TO_BE32X(x)  (x)
+	#define TO_BE64(x)   (x)
 	#define FROM_LE16(x) BSWAP16(x)
 	#define FROM_LE32(x) BSWAP32(x)
+	#define FROM_LE64(x) BSWAP64(x)
 	#define TO_LE16(x)   BSWAP16(x)
 	#define TO_LE32(x)   BSWAP32(x)
 	#define TO_LE32X(x)  BSWAP32(x)
+	#define TO_LE64(x)   BSWAP64(x)
 #else
 	#define FROM_BE16(x) BSWAP16(x)
 	#define FROM_BE32(x) BSWAP32(x)
+	#define FROM_BE64(x) BSWAP64(x)
 	#define TO_BE16(x)   BSWAP16(x)
 	#define TO_BE32(x)   BSWAP32(x)
 	#define TO_BE32X(x)  BSWAP32(x)
+	#define TO_BE64(x)   BSWAP64(x)
 	#define FROM_LE16(x) (x)
 	#define FROM_LE32(x) (x)
+	#define FROM_LE64(x) (x)
 	#define TO_LE16(x)   (x)
 	#define TO_LE32(x)   (x)
 	#define TO_LE32X(x)  (x)
+	#define TO_LE64(x)   (x)
 #endif /* TTD_ENDIAN == TTD_BIG_ENDIAN */
 
 static inline uint16 ReadLE16Aligned(const void *x)
diff --git a/src/scope.h b/src/scope.h
index d341b50b1f..b546813ac3 100644
--- a/src/scope.h
+++ b/src/scope.h
@@ -12,6 +12,8 @@
 #ifndef SCOPE_H
 #define SCOPE_H
 
+#include <utility>
+
 template <typename T>
 class scope_exit_obj {
 	T f;
diff --git a/src/stdafx.h b/src/stdafx.h
index 46028d7a70..be682396b2 100644
--- a/src/stdafx.h
+++ b/src/stdafx.h
@@ -426,6 +426,16 @@ assert_compile(SIZE_MAX >= UINT32_MAX);
 #define unlikely(x)     (x)
 #endif /* __GNUC__ || __clang__ */
 
+#if defined(__GNUC__) || defined(__clang__)
+__attribute__((aligned(1))) typedef uint16 unaligned_uint16;
+__attribute__((aligned(1))) typedef uint32 unaligned_uint32;
+__attribute__((aligned(1))) typedef uint64 unaligned_uint64;
+#else
+typedef uint16 unaligned_uint16;
+typedef uint32 unaligned_uint32;
+typedef uint64 unaligned_uint64;
+#endif /* __GNUC__ || __clang__ */
+
 void NORETURN CDECL usererror(const char *str, ...) WARN_FORMAT(1, 2);
 void NORETURN CDECL error(const char *str, ...) WARN_FORMAT(1, 2);
 void NORETURN CDECL assert_msg_error(int line, const char *file, const char *expr, const char *extra, const char *str, ...) WARN_FORMAT(5, 6);