[RFC,4/3] Fix previous patch and add header.

diff --git a/string/common.h b/string/common.h
new file mode 100644
index 0000000..84f9767
--- /dev/null
+++ b/string/common.h
@@ -0,0 +1,35 @@ 
+#include <stdint.h>
+
+static const unsigned long int ones = (~0UL / 255); /* 0x0101...*/
+static const unsigned long int add = 127 * (~0UL / 255);
+static const unsigned long int high_bits = 128 * (~0UL / 255);
+
+/* Use vector arithmetic tricks. Idea is to take expression works on
+   unsigned byte and evaluates 0 for nozero byte and nonzero on zero byte.
+   Our expression is ((s + 127) ^ (~s)) & 128  
+   Now we evaluate this expression on each byte in parallel and on first 
+   nonzero byte our expression will have nonzero value. */
+
+static __always_inline
+unsigned long int 
+contains_zero (unsigned long int s)
+{
+  return ((s + add) ^ ~s) & high_bits;
+}
+
+#define LSIZE sizeof (unsigned long int)
+#define CROSS_PAGE(x, n) (((uintptr_t)x) % 4096 >= 4096 - n)
+
+static __always_inline
+size_t
+first_nonzero_byte (unsigned long int u)
+{
+#ifdef FAST_FFS
+  return ffsl (u) / 8 - 1;
+#else
+  u = u = u ^ (u - 1);
+  u = u & ones;
+  u = u * ones;
+  return (u >> (8 * LSIZE - 8)) - 1;
+#endif
+}
diff --git a/string/strchr.c b/string/strchr.c
index 17473b8..ca543f4 100644
--- a/string/strchr.c
+++ b/string/strchr.c
@@ -25,9 +25,8 @@ 
 
 #undef strchr
 
-
 /* Include strchrnul and let gcc inline it.  */
-#define AS_STRCHRNUL
+#define AS_STRCHR
 #define STRCHRNUL strchrnul_i
 #include "string/strchrnul.c"
 
diff --git a/string/strchrnul.c b/string/strchrnul.c
index ea91195..5be9da9 100644
--- a/string/strchrnul.c
+++ b/string/strchrnul.c
@@ -23,6 +23,7 @@ 
 #include <string.h>
 #include <libc-internal.h>
 #include <stdint.h>
+#include "string/common.h"
 #undef __strchrnul
 #undef strchrnul
 
@@ -30,24 +31,8 @@ 
 # define STRCHRNUL __strchrnul
 #endif
 
-static const unsigned long int ones = (~0UL / 255); /* 0x0101...*/
-static const unsigned long int add = 127 * (~0UL / 255);
-static const unsigned long int high_bits = 128 * (~0UL / 255);
 
-/* Use vector arithmetic tricks. Idea is to take expression works on
-   unsigned byte and evaluates 0 for nozero byte and nonzero on zero byte.
-   Our expression is ((s + 127) ^ (~s)) & 128  
-   Now we evaluate this expression on each byte in parallel and on first 
-   nonzero byte our expression will have nonzero value. */
-
-static always_inline
-unsigned long int 
-contains_zero (unsigned long int s)
-{
-  return ((s + add) ^ ~s) & high_bits;
-}
-
-static always_inline
+static __always_inline
 int
 found_in_long_bytes(char *s, unsigned long int cmask, char **result)
 {
@@ -55,19 +40,16 @@  found_in_long_bytes(char *s, unsigned long int cmask, char **result)
   unsigned long int mask = contains_zero (*lptr) | contains_zero (*lptr ^ cmask);
   if (mask)
     {
-      *result = s + ffsl (mask) / 8 - 1;
+      *result = s + first_nonzero_byte (mask);
       return 1;
     }
   else
     return 0;
 }
 
-#define LSIZE sizeof (unsigned long int)
-#define CROSS_PAGE(x, n) (((uintptr_t)x)%4096 >= 4096 - n)
-
 /* Find the first occurrence of C in S or the final NUL byte.  */
 #ifdef AS_STRCHR
-static always_inline
+static __always_inline
 #endif
 char *
 STRCHRNUL (const char *s_in, int c_in)
@@ -122,7 +104,7 @@  STRCHRNUL (const char *s_in, int c_in)
              >> (8 * (s_aligned - s));
 
       if (mask)
-        return s + ffsl (mask) / 8 - 1;
+        return s + first_nonzero_byte (mask);
 
       if (found_in_long_bytes (s + 1 * LSIZE, cmask, &r))
         return r;
@@ -160,14 +142,6 @@  STRCHRNUL (const char *s_in, int c_in)
  return r;
 }
 
-
-char *strchr(char *s, int c)
-{
-  char *r = __strchrnule(s,c);
-  return *r ? r : 0;
-
-}
-
-#ifndef AS_STRCHR
+#ifdef AS_STRCHR
 weak_alias (__strchrnul, strchrnul)
 #endif