diff --git a/src/libsodium/crypto_pwhash/scryptsalsa208sha256/crypto_scrypt.h b/src/libsodium/crypto_pwhash/scryptsalsa208sha256/crypto_scrypt.h
index 5f1bb8a4..47a30f78 100644
--- a/src/libsodium/crypto_pwhash/scryptsalsa208sha256/crypto_scrypt.h
+++ b/src/libsodium/crypto_pwhash/scryptsalsa208sha256/crypto_scrypt.h
@@ -32,6 +32,12 @@
 
 #include <stdint.h>
 
+#if SIZE_MAX > 0xffffffffUL
+# define ARCH_BITS 64
+#else
+# define ARCH_BITS 32
+#endif
+
 #define crypto_pwhash_scryptsalsa208sha256_STRPREFIXBYTES 14
 #define crypto_pwhash_scryptsalsa208sha256_STRSETTINGBYTES 57
 #define crypto_pwhash_scryptsalsa208sha256_STRSALTBYTES 32
@@ -46,6 +52,11 @@ typedef struct {
 	size_t size;
 } escrypt_region_t;
 
+typedef union {
+	uint64_t d[8];
+	uint32_t w[16];
+} escrypt_block_t;
+
 typedef escrypt_region_t escrypt_local_t;
 
 extern int escrypt_init_local(escrypt_local_t * __local);
diff --git a/src/libsodium/crypto_pwhash/scryptsalsa208sha256/nosse/pwhash_scryptsalsa208sha256_nosse.c b/src/libsodium/crypto_pwhash/scryptsalsa208sha256/nosse/pwhash_scryptsalsa208sha256_nosse.c
index a9ab966e..97d3701a 100644
--- a/src/libsodium/crypto_pwhash/scryptsalsa208sha256/nosse/pwhash_scryptsalsa208sha256_nosse.c
+++ b/src/libsodium/crypto_pwhash/scryptsalsa208sha256/nosse/pwhash_scryptsalsa208sha256_nosse.c
@@ -39,27 +39,61 @@
 #include "../crypto_scrypt.h"
 
 static inline void
-blkcpy(void * dest, const void * src, size_t len)
+blkcpy_64(escrypt_block_t *dest, const escrypt_block_t *src)
 {
-	size_t * D = (size_t *) dest;
-	const size_t * S = (const size_t *) src;
-	size_t L = len / sizeof(size_t);
-	size_t i;
+	int i;
 
-	for (i = 0; i < L; i++)
-		D[i] = S[i];
+#if (ARCH_BITS==32)
+	for (i = 0; i < 16; ++i)
+		dest->w[i] = src->w[i];
+#else
+	for (i = 0; i < 8; ++i)
+		dest->d[i] = src->d[i];
+#endif
 }
 
 static inline void
-blkxor(void * dest, const void * src, size_t len)
+blkxor_64(escrypt_block_t *dest, const escrypt_block_t *src)
 {
-	size_t * D = (size_t *) dest;
-	const size_t * S = (const size_t *) src;
-	size_t L = len / sizeof(size_t);
-	size_t i;
+	int i;
 
-	for (i = 0; i < L; i++)
-		D[i] ^= S[i];
+#if (ARCH_BITS==32)
+	for (i = 0; i < 16; ++i)
+		dest->w[i] ^= src->w[i];
+#else
+	for (i = 0; i < 8; ++i)
+		dest->d[i] ^= src->d[i];
+#endif
+}
+
+static inline void
+blkcpy(escrypt_block_t *dest, const escrypt_block_t *src, int len)
+{
+	int i, L;
+#if (ARCH_BITS==32)
+	L = (len>>2);
+	for (i = 0; i < L; ++i)
+		dest->w[i] = src->w[i];
+#else
+	L = (len>>3);
+	for (i = 0; i < L; ++i)
+		dest->d[i] = src->d[i];
+#endif
+}
+
+static inline void
+blkxor(escrypt_block_t *dest, const escrypt_block_t *src, int len)
+{
+	int i, L;
+#if (ARCH_BITS==32)
+	L = (len>>2);
+	for (i = 0; i < L; ++i)
+		dest->w[i] ^= src->w[i];
+#else
+	L = (len>>3);
+	for (i = 0; i < L; ++i)
+		dest->d[i] ^= src->d[i];
+#endif
 }
 
 /**
@@ -69,10 +103,11 @@ blkxor(void * dest, const void * src, size_t len)
 static void
 salsa20_8(uint32_t B[16])
 {
-	uint32_t x[16];
+	escrypt_block_t X;
+	uint32_t *x = X.w;
 	size_t i;
 
-	blkcpy(x, B, 64);
+	blkcpy_64(&X, (escrypt_block_t*)B);
 	for (i = 0; i < 8; i += 2) {
 #define R(a,b) (((a) << (b)) | ((a) >> (32 - (b))))
 		/* Operate on columns. */
@@ -118,25 +153,25 @@ blockmix_salsa8(const uint32_t * Bin, uint32_t * Bout, uint32_t * X, size_t r)
 	size_t i;
 
 	/* 1: X <-- B_{2r - 1} */
-	blkcpy(X, &Bin[(2 * r - 1) * 16], 64);
+	blkcpy_64((escrypt_block_t*)X, (escrypt_block_t*)&Bin[(2 * r - 1) * 16]);
 
 	/* 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_64((escrypt_block_t*)X, (escrypt_block_t*)&Bin[i * 16]);
 		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);
+		blkcpy_64((escrypt_block_t*)&Bout[i * 8], (escrypt_block_t*)X);
 
 		/* 3: X <-- H(X \xor B_i) */
-		blkxor(X, &Bin[i * 16 + 16], 64);
+		blkxor_64((escrypt_block_t*)X, (escrypt_block_t*)&Bin[i * 16 + 16]);
 		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 + r * 16], X, 64);
+		blkcpy_64((escrypt_block_t*)&Bout[i * 8 + r * 16], (escrypt_block_t*)X);
 	}
 }
 
@@ -177,13 +212,13 @@ smix(uint8_t * B, size_t r, uint64_t N, uint32_t * V, uint32_t * XY)
 	/* 2: for i = 0 to N - 1 do */
 	for (i = 0; i < N; i += 2) {
 		/* 3: V_i <-- X */
-		blkcpy(&V[i * (32 * r)], X, 128 * r);
+		blkcpy((escrypt_block_t*)&V[i * (32 * r)], (escrypt_block_t*)X, 128 * r);
 
 		/* 4: X <-- H(X) */
 		blockmix_salsa8(X, Y, Z, r);
 
 		/* 3: V_i <-- X */
-		blkcpy(&V[(i + 1) * (32 * r)], Y, 128 * r);
+		blkcpy((escrypt_block_t*)&V[(i + 1) * (32 * r)], (escrypt_block_t*)Y, 128 * r);
 
 		/* 4: X <-- H(X) */
 		blockmix_salsa8(Y, X, Z, r);
@@ -195,14 +230,14 @@ smix(uint8_t * B, size_t r, uint64_t N, uint32_t * V, uint32_t * XY)
 		j = integerify(X, r) & (N - 1);
 
 		/* 8: X <-- H(X \xor V_j) */
-		blkxor(X, &V[j * (32 * r)], 128 * r);
+		blkxor((escrypt_block_t*)X, (escrypt_block_t*)&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((escrypt_block_t*)Y, (escrypt_block_t*)&V[j * (32 * r)], 128 * r);
 		blockmix_salsa8(Y, X, Z, r);
 	}
 	/* 10: B' <-- X */