Index: configure.ac
==================================================================
--- configure.ac
+++ configure.ac
@@ -583,37 +583,38 @@
 		])
 	])
 
 	atomic_ops="none"
 
-	AC_MSG_CHECKING(whether we have an atomic ops assembly implementation)
-	AC_EGREP_CPP(yes, [
-		#if defined(__GNUC__) && (defined(__i386__) || \
-			defined(__x86_64__) || defined(__amd64__))
-		yes
-		#endif
-	], [
-		AC_MSG_RESULT(yes)
-		atomic_ops="assembly implementation"
-	], [
-		AC_MSG_RESULT(no)
-	])
-
 	AC_MSG_CHECKING(whether __sync_* works)
 	AC_TRY_LINK([#include <stdint.h>], [
 		int32_t i, j;
 		if (__sync_add_and_fetch(&i, 1))
 			j = __sync_sub_and_fetch(&i, 1);
 		while (!__sync_bool_compare_and_swap(&i, 0, 1));
 	], [
 		AC_MSG_RESULT(yes)
-		test x"$atomic_ops" = x"none" && atomic_ops="gcc builtins"
+		atomic_ops="gcc builtins"
 		AC_DEFINE(OF_HAVE_GCC_ATOMIC_OPS, 1,
 			[Whether gcc atomic operations are available])
 	], [
 		AC_MSG_RESULT(no)
 	])
+
+	AC_MSG_CHECKING(whether we have an atomic ops assembly implementation)
+	AC_EGREP_CPP(yes, [
+		#if defined(__GNUC__) && (defined(__i386__) || \
+			defined(__x86_64__) || defined(__amd64__))
+		yes
+		#endif
+	], [
+		AC_MSG_RESULT(yes)
+		test x"$atomic_ops" = x"none" && \
+			atomic_ops="assembly implementation"
+	], [
+		AC_MSG_RESULT(no)
+	])
 
 	AC_CHECK_HEADER(libkern/OSAtomic.h, [
 		test x"$atomic_ops" = x"none" && atomic_ops="libkern/OSAtomic.h"
 		AC_DEFINE(OF_HAVE_OSATOMIC, 1,
 			[Whether we have libkern/OSAtomic.h])

Index: src/atomic.h
==================================================================
--- src/atomic.h
+++ src/atomic.h
@@ -31,10 +31,12 @@
 static OF_INLINE int
 of_atomic_int_add(volatile int *p, int i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p += i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_add_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
 		    "lock\n\t"
 		    "xaddl	%0, %2\n\t"
@@ -54,12 +56,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_add_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicAdd32Barrier(i, p);
 #else
 # error of_atomic_int_add not implemented!
 #endif
@@ -68,10 +68,12 @@
 static OF_INLINE int32_t
 of_atomic_int32_add(volatile int32_t *p, int32_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p += i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_add_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	__asm__ __volatile__ (
 	    "lock\n\t"
 	    "xaddl	%0, %2\n\t"
 	    "addl	%1, %0"
@@ -78,12 +80,10 @@
 	    : "+&r"(i)
 	    : "r"(i), "m"(*p)
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_add_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicAdd32Barrier(i, p);
 #else
 # error of_atomic_int32_add not implemented!
 #endif
@@ -92,10 +92,12 @@
 static OF_INLINE void*
 of_atomic_ptr_add(void* volatile *p, intptr_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*(char* volatile*)p += i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_add_and_fetch(p, (void*)i);
 #elif defined(OF_X86_64_ASM)
 	__asm__ __volatile__ (
 	    "lock\n\t"
 	    "xaddq	%0, %2\n\t"
 	    "addq	%1, %0"
@@ -112,12 +114,10 @@
 	    : "+&r"(i)
 	    : "r"(i), "m"(*p)
 	);
 
 	return (void*)i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_add_and_fetch(p, (void*)i);
 #elif defined(OF_HAVE_OSATOMIC)
 # ifdef __LP64__
 	return (void*)OSAtomicAdd64Barrier(i, (int64_t*)p);
 # else
 	return (void*)OSAtomicAdd32Barrier(i, (int32_t*)p);
@@ -130,10 +130,12 @@
 static OF_INLINE int
 of_atomic_int_sub(volatile int *p, int i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p -= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_sub_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
 		    "negl	%0\n\t"
 		    "lock\n\t"
@@ -155,12 +157,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_sub_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicAdd32Barrier(-i, p);
 #else
 # error of_atomic_int_sub not implemented!
 #endif
@@ -169,10 +169,12 @@
 static OF_INLINE int32_t
 of_atomic_int32_sub(volatile int32_t *p, int32_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p -= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_sub_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	__asm__ __volatile__ (
 	    "negl	%0\n\t"
 	    "lock\n\t"
 	    "xaddl	%0, %2\n\t"
@@ -180,12 +182,10 @@
 	    : "+&r"(i)
 	    : "r"(i), "m"(*p)
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_sub_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicAdd32Barrier(-i, p);
 #else
 # error of_atomic_int32_sub not implemented!
 #endif
@@ -194,10 +194,12 @@
 static OF_INLINE void*
 of_atomic_ptr_sub(void* volatile *p, intptr_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*(char* volatile*)p -= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_sub_and_fetch(p, (void*)i);
 #elif defined(OF_X86_64_ASM)
 	__asm__ __volatile__ (
 	    "negq	%0\n\t"
 	    "lock\n\t"
 	    "xaddq	%0, %2\n\t"
@@ -216,12 +218,10 @@
 	    : "+&r"(i)
 	    : "r"(i), "m"(*p)
 	);
 
 	return (void*)i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_sub_and_fetch(p, (void*)i);
 #elif defined(OF_HAVE_OSATOMIC)
 # ifdef __LP64__
 	return (void*)OSAtomicAdd64Barrier(-i, (int64_t*)p);
 # else
 	return (void*)OSAtomicAdd32Barrier(-i, (int32_t*)p);
@@ -234,10 +234,12 @@
 static OF_INLINE int
 of_atomic_int_inc(volatile int *p)
 {
 #if !defined(OF_HAVE_THREADS)
 	return ++*p;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_add_and_fetch(p, 1);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	int i;
 
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
@@ -263,12 +265,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_add_and_fetch(p, 1);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicIncrement32Barrier(p);
 #else
 # error of_atomic_int_inc not implemented!
 #endif
@@ -277,10 +277,12 @@
 static OF_INLINE int32_t
 of_atomic_int32_inc(volatile int32_t *p)
 {
 #if !defined(OF_HAVE_THREADS)
 	return ++*p;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_add_and_fetch(p, 1);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	uint32_t i;
 
 	__asm__ __volatile__ (
 	    "xorl	%0, %0\n\t"
@@ -291,12 +293,10 @@
 	    : "=&r"(i)
 	    : "m"(*p)
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_add_and_fetch(p, 1);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicIncrement32Barrier(p);
 #else
 # error of_atomic_int32_inc not implemented!
 #endif
@@ -305,10 +305,12 @@
 static OF_INLINE int
 of_atomic_int_dec(volatile int *p)
 {
 #if !defined(OF_HAVE_THREADS)
 	return --*p;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_sub_and_fetch(p, 1);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	int i;
 
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
@@ -334,12 +336,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_sub_and_fetch(p, 1);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicDecrement32Barrier(p);
 #else
 # error of_atomic_int_dec not implemented!
 #endif
@@ -348,10 +348,12 @@
 static OF_INLINE int32_t
 of_atomic_int32_dec(volatile int32_t *p)
 {
 #if !defined(OF_HAVE_THREADS)
 	return --*p;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_sub_and_fetch(p, 1);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	uint32_t i;
 
 	__asm__ __volatile__ (
 	    "xorl	%0, %0\n\t"
@@ -362,12 +364,10 @@
 	    : "=&r"(i)
 	    : "m"(*p)
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_sub_and_fetch(p, 1);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicDecrement32Barrier(p);
 #else
 # error of_atomic_int32_dec not implemented!
 #endif
@@ -376,10 +376,12 @@
 static OF_INLINE unsigned int
 of_atomic_int_or(volatile unsigned int *p, unsigned int i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p |= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_or_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
 		    "0:\n\t"
 		    "movl	%2, %0\n\t"
@@ -409,12 +411,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_or_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicOr32Barrier(i, p);
 #else
 # error of_atomic_int_or not implemented!
 #endif
@@ -423,10 +423,12 @@
 static OF_INLINE uint32_t
 of_atomic_int32_or(volatile uint32_t *p, uint32_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p |= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_or_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	__asm__ __volatile__ (
 	    "0:\n\t"
 	    "movl	%2, %0\n\t"
 	    "movl	%0, %%eax\n\t"
@@ -438,12 +440,10 @@
 	    : "r"(i), "m"(*p)
 	    : "eax", "cc"
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_or_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicOr32Barrier(i, p);
 #else
 # error of_atomic_int32_or not implemented!
 #endif
@@ -452,10 +452,12 @@
 static OF_INLINE unsigned int
 of_atomic_int_and(volatile unsigned int *p, unsigned int i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p &= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_and_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
 		    "0:\n\t"
 		    "movl	%2, %0\n\t"
@@ -485,12 +487,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_and_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicAnd32Barrier(i, p);
 #else
 # error of_atomic_int_and not implemented!
 #endif
@@ -499,10 +499,12 @@
 static OF_INLINE uint32_t
 of_atomic_int32_and(volatile uint32_t *p, uint32_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p &= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_and_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	__asm__ __volatile__ (
 	    "0:\n\t"
 	    "movl	%2, %0\n\t"
 	    "movl	%0, %%eax\n\t"
@@ -514,12 +516,10 @@
 	    : "r"(i), "m"(*p)
 	    : "eax", "cc"
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_and_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicAnd32Barrier(i, p);
 #else
 # error of_atomic_int32_and not implemented!
 #endif
@@ -528,10 +528,12 @@
 static OF_INLINE unsigned int
 of_atomic_int_xor(volatile unsigned int *p, unsigned int i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p ^= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_xor_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	if (sizeof(int) == 4)
 		__asm__ __volatile__ (
 		    "0:\n\t"
 		    "movl	%2, %0\n\t"
@@ -561,12 +563,10 @@
 # endif
 	else
 		abort();
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_xor_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicXor32Barrier(i, p);
 #else
 # error of_atomic_int_xor not implemented!
 #endif
@@ -575,10 +575,12 @@
 static OF_INLINE uint32_t
 of_atomic_int32_xor(volatile uint32_t *p, uint32_t i)
 {
 #if !defined(OF_HAVE_THREADS)
 	return (*p ^= i);
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_xor_and_fetch(p, i);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	__asm__ __volatile__ (
 	    "0:\n\t"
 	    "movl	%2, %0\n\t"
 	    "movl	%0, %%eax\n\t"
@@ -590,12 +592,10 @@
 	    : "r"(i), "m"(*p)
 	    : "eax", "cc"
 	);
 
 	return i;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_xor_and_fetch(p, i);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicXor32Barrier(i, p);
 #else
 # error of_atomic_int32_xor not implemented!
 #endif
@@ -609,10 +609,12 @@
 		*p = n;
 		return true;
 	}
 
 	return false;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_bool_compare_and_swap(p, o, n);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	int r;
 
 	__asm__ __volatile__ (
 	    "lock\n\t"
@@ -623,12 +625,10 @@
 	    : "r"(n), "m"(*p)
 	    : "cc"
 	);
 
 	return r;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_bool_compare_and_swap(p, o, n);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicCompareAndSwapIntBarrier(o, n, p);
 #else
 # error of_atomic_int_cmpswap not implemented!
 #endif
@@ -642,10 +642,12 @@
 		*p = n;
 		return true;
 	}
 
 	return false;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_bool_compare_and_swap(p, o, n);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	int r;
 
 	__asm__ __volatile__ (
 	    "lock\n\t"
@@ -656,12 +658,10 @@
 	    : "r"(n), "m"(*p)
 	    : "cc"
 	);
 
 	return r;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_bool_compare_and_swap(p, o, n);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicCompareAndSwap32Barrier(o, n, p);
 #else
 # error of_atomic_int32_cmpswap not implemented!
 #endif
@@ -675,10 +675,12 @@
 		*p = n;
 		return true;
 	}
 
 	return false;
+#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
+	return __sync_bool_compare_and_swap(p, o, n);
 #elif defined(OF_X86_64_ASM) || defined(OF_X86_ASM)
 	int r;
 
 	__asm__ __volatile__ (
 	    "lock\n\t"
@@ -689,12 +691,10 @@
 	    : "r"(n), "m"(*p)
 	    : "cc"
 	);
 
 	return r;
-#elif defined(OF_HAVE_GCC_ATOMIC_OPS)
-	return __sync_bool_compare_and_swap(p, o, n);
 #elif defined(OF_HAVE_OSATOMIC)
 	return OSAtomicCompareAndSwapPtrBarrier(o, n, p);
 #else
 # error of_atomic_ptr_cmpswap not implemented!
 #endif