Implemented Rtl*ByteSwap() functions, based on a patch by Jon
Griffiths.

dlls/ntdll/ntdll.spec

diff --git a/dlls/ntdll/ntdll.spec b/dlls/ntdll/ntdll.spec
index 3b5a1c2..5ccb7b4 100644
--- a/dlls/ntdll/ntdll.spec
+++ b/dlls/ntdll/ntdll.spec
@@ -530,6 +530,8 @@
 @ stdcall RtlTimeToSecondsSince1970(ptr ptr) RtlTimeToSecondsSince1970
 @ stdcall RtlTimeToSecondsSince1980(ptr ptr) RtlTimeToSecondsSince1980
 @ stdcall RtlTimeToTimeFields (long long) RtlTimeToTimeFields
+@ cdecl -i386 -norelay RtlUlongByteSwap() RtlUlongByteSwap
+@ cdecl -ret64 RtlUlonglongByteSwap(long long) RtlUlonglongByteSwap
 @ stdcall RtlUnicodeStringToAnsiSize(ptr) RtlUnicodeStringToAnsiSize
 @ stdcall RtlUnicodeStringToAnsiString(ptr ptr long) RtlUnicodeStringToAnsiString
 @ stub RtlUnicodeStringToCountedOemString
@@ -554,6 +556,7 @@
 @ stdcall RtlUpperChar(long) RtlUpperChar
 @ stdcall RtlUpperString(ptr ptr) RtlUpperString
 @ stub RtlUsageHeap
+@ cdecl -i386 -norelay RtlUshortByteSwap() RtlUshortByteSwap
 @ stub RtlValidAcl
 @ stdcall RtlValidSecurityDescriptor(ptr) RtlValidSecurityDescriptor
 @ stdcall RtlValidSid(ptr) RtlValidSid

dlls/ntdll/rtl.c

diff --git a/dlls/ntdll/rtl.c b/dlls/ntdll/rtl.c
index 0ca5778..953dd92 100644
--- a/dlls/ntdll/rtl.c
+++ b/dlls/ntdll/rtl.c
@@ -682,3 +682,50 @@
   }
   return ~crc;
 }
+
+
+/*************************************************************************
+ * RtlUlonglongByteSwap    [NTDLL.@]
+ *
+ * Swap the bytes of an unsigned long long value.
+ *
+ * PARAMS
+ *  i [I] Value to swap bytes of
+ *
+ * RETURNS
+ *  The value with its bytes swapped.
+ */
+ULONGLONG __cdecl RtlUlonglongByteSwap(ULONGLONG i)
+{
+  return ((ULONGLONG)RtlUlongByteSwap(i) << 32) | RtlUlongByteSwap(i>>32);
+}
+
+/*************************************************************************
+ * RtlUlongByteSwap    [NTDLL.@]
+ *
+ * Swap the bytes of an unsigned int value.
+ *
+ * NOTES
+ *  ix86 version takes argument in %ecx. Other systems use the inline version.
+ */
+#ifdef __i386__
+__ASM_GLOBAL_FUNC(RtlUlongByteSwap,
+                  "movl %ecx,%eax\n\t"
+                  "bswap %eax\n\t"
+                  "ret");
+#endif
+
+/*************************************************************************
+ * RtlUshortByteSwap    [NTDLL.@]
+ *
+ * Swap the bytes of an unsigned short value.
+ *
+ * NOTES
+ *  i386 version takes argument in %cx. Other systems use the inline version.
+ */
+#ifdef __i386__
+__ASM_GLOBAL_FUNC(RtlUshortByteSwap,
+                  "movb %ch,%al\n\t"
+                  "movb %cl,%ah\n\t"
+                  "ret");
+#endif

include/winternl.h

diff --git a/include/winternl.h b/include/winternl.h
index 809e9fc..b0478d3 100644
--- a/include/winternl.h
+++ b/include/winternl.h
@@ -1050,6 +1050,7 @@
 BOOLEAN   WINAPI RtlTimeToSecondsSince1980(const LARGE_INTEGER *,LPDWORD);
 BOOL      WINAPI RtlTryEnterCriticalSection(RTL_CRITICAL_SECTION *);
 
+ULONGLONG __cdecl RtlUlonglongByteSwap(ULONGLONG);
 DWORD     WINAPI RtlUnicodeStringToAnsiSize(const UNICODE_STRING*);
 NTSTATUS  WINAPI RtlUnicodeStringToAnsiString(PANSI_STRING,PCUNICODE_STRING,BOOLEAN);
 NTSTATUS  WINAPI RtlUnicodeStringToInteger(const UNICODE_STRING *,ULONG,ULONG *);
@@ -1134,6 +1135,22 @@
         memset(_p->BitMapBuffer,0xff,((_p->SizeOfBitMap + 31) & 0xffffffe0) >> 3); \
     } while (0)
 
+/* These are implemented as __fastcall, so we can't let Winelib apps link with them */
+inline static USHORT RtlUshortByteSwap(USHORT s)
+{
+    return (s >> 8) | (s << 8);
+}
+inline static ULONG RtlUlongByteSwap(ULONG i)
+{
+#if defined(__i386__) && defined(__GNUC__)
+    ULONG ret;
+    __asm__("bswap %0" : "=r" (ret) : "0" (i) );
+    return ret;
+#else
+    return ((ULONG)RtlUshortByteSwap(i) << 16) | RtlUshortByteSwap(i >> 16);
+#endif
+}
+
 /*************************************************************************
  * Loader functions and structures.
  *