Fix nextpow2 , provide optimized and 64 bit variants

include/ucore/math.h

@@ -19,9 +19,36 @@ uc_phi_32(uint32_t N);
 uint64_t 
 uc_phi_64(uint64_t N);
 
+#if defined(__has_builtin)
+#  if __has_builtin(__builtin_clzll)
+#    define HAVE_BUILTIN_CLZ 1
+#  endif
+#endif
+
+#if HAVE_BUILTIN_CLZ
+
 static inline uint32_t
 uc_nextpow2(uint32_t x)
 {
+    if (x <= 1) return 1;
+    return 1u << (32 - __builtin_clz(x - 1));
+    return x;
+}
+
+static inline uint64_t
+uc_nextpow2_64(uint64_t x)
+{
+    if (x <= 1) return 1;
+    return 1u << (64 - __builtin_clzll(x - 1));
+    return x;
+}
+#else
+static inline uint32_t
+uc_nextpow2(uint32_t x)
+{
+    if (x == 0) {
+        return 1;
+    }
     x--;
     x |= x >> 1;
     x |= x >> 2;
@@ -33,6 +60,26 @@ uc_nextpow2(uint32_t x)
     return x;
 }
 
+static inline uint64_t
+uc_nextpow2(uint64_t x)
+{
+    if (x == 0) {
+        return 1;
+    }
+    x--;
+    x |= x >> 1;
+    x |= x >> 2;
+    x |= x >> 4;
+    x |= x >> 8;
+    x |= x >> 16;
+    x |= x >> 32;
+    x++;
+
+    return x;
+}
+#endif
+
+#undef HAVE_BUILTIN_CLZ
 #ifdef __cplusplus
 }
 #endif