--- libecb/ecb.h	2011/08/04 14:47:53	1.64
+++ libecb/ecb.h	2012/01/18 12:44:23	1.73
@@ -1,7 +1,7 @@
 /*
  * libecb - http://software.schmorp.de/pkg/libecb
  *
- * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de>
+ * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de>
  * Copyright (©) 2011 Emanuele Giaquinta
  * All rights reserved.
  *
@@ -73,12 +73,12 @@
 #endif
 
 #ifndef ECB_MEMORY_FENCE
-  #if ECB_GCC_VERSION(2,5)
-    #if __x86
+  #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
+    #if __i386 || __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 */
-    #elif __amd64
+    #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
       #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("mfence" : : : "memory")
       #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
       #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
@@ -86,16 +86,20 @@
       #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")
+    #elif __sparc || __sparc__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("membar #LoadStore | #StoreLoad | #LoadLoad | #StoreStore" : : : "memory")
+      #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadLoad" : : : "memory")
+      #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #StoreStore")
     #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   ); }) */
@@ -107,6 +111,11 @@
   #elif defined(_WIN32)
     #include <WinNT.h>
     #define ECB_MEMORY_FENCE         MemoryBarrier () /* actually just xchg on x86... scary */
+  #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
+    #include <mbarrier.h>
+    #define ECB_MEMORY_FENCE         __machine_rw_barrier ()
+    #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier  ()
+    #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier  ()
   #endif
 #endif
 
@@ -291,6 +300,36 @@
   }
 #endif
 
+ecb_function_ uint8_t  ecb_bitrev8  (uint8_t  x) ecb_const;
+ecb_function_ uint8_t  ecb_bitrev8  (uint8_t  x)
+{
+  return (  (x * 0x0802U & 0x22110U)
+          | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 
+}
+
+ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
+ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
+{
+  x = ((x >>  1) &     0x5555) | ((x &     0x5555) <<  1);
+  x = ((x >>  2) &     0x3333) | ((x &     0x3333) <<  2);
+  x = ((x >>  4) &     0x0f0f) | ((x &     0x0f0f) <<  4);
+  x = ( x >>  8              ) | ( x               <<  8);
+
+  return x;
+}
+
+ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
+ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
+{
+  x = ((x >>  1) & 0x55555555) | ((x & 0x55555555) <<  1);
+  x = ((x >>  2) & 0x33333333) | ((x & 0x33333333) <<  2);
+  x = ((x >>  4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) <<  4);
+  x = ((x >>  8) & 0x00ff00ff) | ((x & 0x00ff00ff) <<  8);
+  x = ( x >> 16              ) | ( x               << 16);
+
+  return x;
+}
+
 /* popcount64 is only available on 64 bit cpus as gcc builtin */
 /* so for this version we are lazy */
 ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
@@ -375,6 +414,22 @@
   #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>