Fix rotation macros to portably support zero rotation count.

ChangeLog

 2013-05-17  Niels Möller  <nisse@lysator.liu.se>
 
+	* macros.h (ROTL32, ROTL64): Avoid undefined behaviour for zero
+	rotation count. Unfortunately makes CAST128 a bit slower with
+	gcc-4.6.3.
+
 	* ecc-j-to-a.c (ecc_j_to_a): Fixed ecc_modp_mul call, to avoid
 	invalid overlap of arguments to mpn_mul_n. Problem tracked down by
 	Magnus Holmgren.

macros.h

@@ -141,9 +141,11 @@ do {						\
 		  (dst) += (blocksize),		\
 		  (src) += (blocksize)) )
 
-#define ROTL32(n,x) (((x)<<(n)) | ((x)>>(32-(n))))
+/* The masking of the right shift is needed to allow n == 0 (using
+   just 32 - n and 64 - n results in undefined behaviour). */
+#define ROTL32(n,x) (((x)<<(n)) | ((x)>>((-(n)&31))))
   
-#define ROTL64(n,x) (((x)<<(n)) | ((x)>>(64-(n))))
+#define ROTL64(n,x) (((x)<<(n)) | ((x)>>((-(n))&63)))
 
 /* Requires that size > 0 */
 #define INCREMENT(size, ctr)			\