--- libecb/ecb.h	2011/08/04 14:47:53	1.64
+++ libecb/ecb.h	2011/09/25 21:27:02	1.69
@@ -73,8 +73,8 @@
 #endif
 
 #ifndef ECB_MEMORY_FENCE
-  #if ECB_GCC_VERSION(2,5)
-    #if __x86
+  #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__)
+    #if __i386__
       #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
       #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
       #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
@@ -86,16 +86,16 @@
       #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("sync" : : : "memory")
     #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \
        || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__)
-      #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,4" : : "r" (0) : "memory")
+      #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
     #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \
        || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ )
-      #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("dsb" : : : "memory")
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("dmb" : : : "memory")
     #endif
   #endif
 #endif
 
 #ifndef ECB_MEMORY_FENCE
-  #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER)
+  #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__)
     #define ECB_MEMORY_FENCE         __sync_synchronize ()
     /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
     /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release      (&dummy   ); }) */
@@ -375,6 +375,22 @@
   #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
 #endif
 
+#if __cplusplus
+  template<typename T>
+  static inline T ecb_div_rd (T val, T div)
+  {
+    return val < 0 ? - ((-val + div - 1) / div) : (val          ) / div;
+  }
+  template<typename T>
+  static inline T ecb_div_ru (T val, T div)
+  {
+    return val < 0 ? - ((-val          ) / div) : (val + div - 1) / div;
+  }
+#else
+  #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val)            ) / (div))
+  #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val)            ) / (div)) : ((val) + (div) - 1) / (div))
+#endif
+
 #if ecb_cplusplus_does_not_suck
   /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
   template<typename T, int N>