DO NOT MERGE Faster blkxor in plain smix

We decided not to merge this because:

    Now that we autodetect SSE2 support (which the original code didn't
    do) approximately 0% of x86 users should be running the non-SSE2
    code, and we would need to benchmark a range of different non-x86
    systems (and compilers) to get a sense of whether a change would
    help there.
    #334 (comment)

lib/crypto/crypto_scrypt_smix.c

@@ -34,7 +34,7 @@
 #include "crypto_scrypt_smix.h"
 
 static void blkcpy(uint32_t *, const uint32_t *, size_t);
-static void blkxor(uint32_t *, const uint32_t *, size_t);
+static void blkxor(unsigned char *, const unsigned char *, size_t);
 static void salsa20_8(uint32_t[16]);
 static void blockmix_salsa8(const uint32_t *, uint32_t *, uint32_t *, size_t);
 static uint64_t integerify(const uint32_t *, size_t);
@@ -47,11 +47,11 @@ blkcpy(uint32_t * dest, const uint32_t * src, size_t len)
 }
 
 static void
-blkxor(uint32_t * dest, const uint32_t * src, size_t len)
+blkxor(unsigned char * dest, const unsigned char * src, size_t len)
 {
 	size_t i;
 
-	for (i = 0; i < len / 4; i++)
+	for (i = 0; i < len; i++)
 		dest[i] ^= src[i];
 }
 
@@ -116,15 +116,17 @@ blockmix_salsa8(const uint32_t * Bin, uint32_t * Bout, uint32_t * X, size_t r)
 	/* 2: for i = 0 to 2r - 1 do */
 	for (i = 0; i < 2 * r; i += 2) {
 		/* 3: X <-- H(X \xor B_i) */
-		blkxor(X, &Bin[i * 16], 64);
+		blkxor((unsigned char *)X,
+		    (const unsigned char *)&Bin[i * 16], 64);
 		salsa20_8(X);
 
 		/* 4: Y_i <-- X */
 		/* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
 		blkcpy(&Bout[i * 8], X, 64);
 
 		/* 3: X <-- H(X \xor B_i) */
-		blkxor(X, &Bin[i * 16 + 16], 64);
+		blkxor((unsigned char *)X,
+		    (const unsigned char *)&Bin[i * 16 + 16], 64);
 		salsa20_8(X);
 
 		/* 4: Y_i <-- X */
@@ -189,14 +191,16 @@ crypto_scrypt_smix(uint8_t * B, size_t r, uint64_t N, void * _v, void * XY)
 		j = integerify(X, r) & (N - 1);
 
 		/* 8: X <-- H(X \xor V_j) */
-		blkxor(X, &V[j * (32 * r)], 128 * r);
+		blkxor((unsigned char *)X,
+		    (const unsigned char *)&V[j * (32 * r)], 128 * r);
 		blockmix_salsa8(X, Y, Z, r);
 
 		/* 7: j <-- Integerify(X) mod N */
 		j = integerify(Y, r) & (N - 1);
 
 		/* 8: X <-- H(X \xor V_j) */
-		blkxor(Y, &V[j * (32 * r)], 128 * r);
+		blkxor((unsigned char *)Y,
+		    (const unsigned char *)&V[j * (32 * r)], 128 * r);
 		blockmix_salsa8(Y, X, Z, r);
 	}