--- deliantra/server/include/util.h	2008/12/19 22:47:29	1.79
+++ deliantra/server/include/util.h	2009/01/03 01:04:19	1.86
@@ -30,10 +30,12 @@
 # define is_constant(c)             __builtin_constant_p (c)
 # define expect(expr,value)         __builtin_expect ((expr),(value))
 # define prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
+# define noinline                   __attribute__((__noinline__))
 #else
 # define is_constant(c)             0
 # define expect(expr,value)         (expr)
 # define prefetch(addr,rw,locality)
+# define noinline
 #endif
 
 #if __GNUC__ < 4 || (__GNUC__ == 4 || __GNUC_MINOR__ < 4)
@@ -43,8 +45,8 @@
 // put into ifs if you are very sure that the expression
 // is mostly true or mosty false. note that these return
 // booleans, not the expression.
-#define expect_false(expr) expect ((expr) != 0, 0)
-#define expect_true(expr)  expect ((expr) != 0, 1)
+#define expect_false(expr) expect ((expr) ? 1 : 0, 0)
+#define expect_true(expr)  expect ((expr) ? 1 : 0, 1)
 
 #include <pthread.h>
 
@@ -74,7 +76,7 @@
 // use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
 #define auto(var,expr) decltype(expr) var = (expr)
 
-// very ugly macro that basicaly declares and initialises a variable
+// very ugly macro that basically declares and initialises a variable
 // that is in scope for the next statement only
 // works only for stuff that can be assigned 0 and converts to false
 // (note: works great for pointers)
@@ -98,9 +100,9 @@
 template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
 template<typename T, typename U, typename V> static inline T clamp (T v, U a, V b) { return v < (T)a ? (T)a : v >(T)b ? (T)b : v; }
 
-template<typename T> static inline void min_it (T &v, T m) { v = min (v, m); }
-template<typename T> static inline void max_it (T &v, T m) { v = max (v, m); }
-template<typename T> static inline void clamp_it (T &v, T a, T b) { v = clamp (v, a, b); }
+template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); }
+template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); }
+template<typename T, typename U, typename V> static inline void clamp_it (T &v, U a, V b) { v = clamp (v, (T)a, (T)b); }
 
 template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; }
 
@@ -391,7 +393,6 @@
 // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
 struct tausworthe_random_generator
 {
-  // generator
   uint32_t state [4];
 
   void operator =(const tausworthe_random_generator &src)
@@ -404,13 +405,47 @@
 
   void seed (uint32_t seed);
   uint32_t next ();
+};
+
+// Xorshift RNGs, George Marsaglia
+// http://www.jstatsoft.org/v08/i14/paper
+// this one is about 40% faster than the tausworthe one above (i.e. not much),
+// despite the inlining, and has the issue of only creating 2**32-1 numbers.
+// see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf
+struct xorshift_random_generator
+{
+  uint32_t x, y;
+
+  void operator =(const xorshift_random_generator &src)
+  {
+    x = src.x;
+    y = src.y;
+  }
+
+  void seed (uint32_t seed)
+  {
+    x = seed;
+    y = seed * 69069U;
+  }
 
+  uint32_t next ()
+  {
+    uint32_t t = x ^ (x << 10);
+    x = y;
+    y = y ^ (y >> 13) ^ t ^ (t >> 10);
+    return y;
+  }
+};
+
+template<class generator>
+struct random_number_generator : generator
+{
   // uniform distribution, 0 .. max (0, num - 1)
   uint32_t operator ()(uint32_t num)
   {
-    return is_constant (num)
-             ? (next () * (uint64_t)num) >> 32U
-             : get_range (num);
+    return !is_constant (num) ? get_range (num)                        // non-constant
+         : num & (num - 1)    ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two
+         :                      this->next () & (num - 1);             // constant, power-of-two
   }
 
   // return a number within (min .. max)
@@ -431,7 +466,7 @@
   int get_range (int r_min, int r_max);
 };
 
-typedef tausworthe_random_generator rand_gen;
+typedef random_number_generator<tausworthe_random_generator> rand_gen;
 
 extern rand_gen rndm, rmg_rndm;
 
@@ -510,7 +545,8 @@
 {
   std::size_t operator ()(const char *s) const
   {
-    unsigned long hash = 0;
+#if 0
+    uint32_t hash = 0;
 
     /* use the one-at-a-time hash function, which supposedly is
      * better than the djb2-like one used by perl5.005, but
@@ -527,6 +563,17 @@
     hash += hash <<  3;
     hash ^= hash >> 11;
     hash += hash << 15;
+#else
+    // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/)
+    // it is about twice as fast as the one-at-a-time one,
+    // with good distribution.
+    // FNV-1a is faster on many cpus because the multiplication
+    // runs concurrent with the looping logic.
+    uint32_t hash = 2166136261;
+    
+    while (*s)
+      hash = (hash ^ *s++) * 16777619;
+#endif
 
     return hash;
   }