libport: Added an interlocked_cmpxchg128 function for 64-bit.

include/wine/port.h

@@ -412,6 +412,22 @@ extern inline int interlocked_xchg_add( int *dest, int incr )
     return ret;
 }
 
+#ifdef __x86_64__
+extern inline unsigned char interlocked_cmpxchg128( __int64 *dest, __int64 xchg_high,
+                                                    __int64 xchg_low, __int64 *compare );
+extern inline unsigned char interlocked_cmpxchg128( __int64 *dest, __int64 xchg_high,
+                                                    __int64 xchg_low, __int64 *compare )
+{
+    unsigned char ret;
+    __asm__ __volatile__( "lock cmpxchg16b %0; setz %b2"
+                          : "=m" (dest[0]), "=m" (dest[1]), "=r" (ret),
+                            "=a" (compare[0]), "=d" (compare[1])
+                          : "m" (dest[0]), "m" (dest[1]), "3" (compare[0]), "4" (compare[1]),
+                            "c" (xchg_high), "b" (xchg_low) );
+    return ret;
+}
+#endif
+
 #else  /* __GNUC__ */
 
 extern int interlocked_cmpxchg( int *dest, int xchg, int compare );
@@ -420,6 +436,10 @@ extern __int64 interlocked_cmpxchg64( __int64 *dest, __int64 xchg, __int64 compa
 extern int interlocked_xchg( int *dest, int val );
 extern void *interlocked_xchg_ptr( void **dest, void *val );
 extern int interlocked_xchg_add( int *dest, int incr );
+#ifdef _WIN64
+extern unsigned char interlocked_cmpxchg128( __int64 *dest, __int64 xchg_high,
+                                             __int64 xchg_low, __int64 *compare );
+#endif
 
 #endif  /* __GNUC__ */
 

libs/port/interlocked.c

@@ -139,8 +139,6 @@ __declspec(naked) int interlocked_xchg_add( int *dest, int incr )
 
 #elif defined(__x86_64__)
 
-#ifdef __GNUC__
-
 __ASM_GLOBAL_FUNC(interlocked_cmpxchg,
                   "mov %edx, %eax\n\t"
                   "lock cmpxchgl %esi,(%rdi)\n\t"
@@ -165,10 +163,23 @@ __ASM_GLOBAL_FUNC(interlocked_xchg_add,
                   "mov %esi, %eax\n\t"
                   "lock xaddl %eax, (%rdi)\n\t"
                   "ret")
-
-#else
-# error You must implement the interlocked* functions for your compiler
-#endif
+__ASM_GLOBAL_FUNC(interlocked_cmpxchg128,
+                  "push %rbx\n\t"
+                   ".cfi_adjust_cfa_offset 8\n\t"
+                   ".cfi_rel_offset %rbx,0\n\t"
+                  "mov %rcx,%r8\n\t"  /* compare */
+                  "mov %rdx,%rbx\n\t" /* xchg_low */
+                  "mov %rsi,%rcx\n\t" /* xchg_high */
+                  "mov 0(%r8),%rax\n\t"
+                  "mov 8(%r8),%rdx\n\t"
+                  "lock cmpxchg16b (%rdi)\n\t"
+                  "mov %rax,0(%r8)\n\t"
+                  "mov %rdx,8(%r8)\n\t"
+                  "setz %al\n\t"
+                  "pop %rbx\n\t"
+                   ".cfi_adjust_cfa_offset -8\n\t"
+                   ".cfi_same_value %rbx\n\t"
+                  "ret")
 
 #elif defined(__powerpc__)
 void* interlocked_cmpxchg_ptr( void **dest, void* xchg, void* compare)