Revert "blkcpy -> memcpy"

This reverts commit e39be36cc0744d084653264191490fb135e81d3d.

Causes performance regression for MIPS and PPC (even though it's
faster on x86).

1 files changed, 18 insertions, 6 deletions

diff --git a/scrypt.c b/scrypt.c
index 2504b7b..5dda34a 100644
--- a/scrypt.c
+++ b/scrypt.c
@@ -473,6 +473,18 @@ static uint64_t integerify(void *, size_t);
 static void smix(uint8_t *, size_t, uint64_t, uint32_t *, uint32_t *);
 
 static void
+blkcpy(void * dest, void * src, size_t len)
+{
+	size_t * D = dest;
+	size_t * S = src;
+	size_t L = len / sizeof(size_t);
+	size_t i;
+
+	for (i = 0; i < L; i++)
+		D[i] = S[i];
+}
+
+static void
 blkxor(void * dest, void * src, size_t len)
 {
 	size_t * D = dest;
@@ -494,7 +506,7 @@ salsa20_8(uint32_t B[16])
 	uint32_t x[16];
 	size_t i;
 
-	memcpy(x, B, 64);
+	blkcpy(x, B, 64);
 	for (i = 0; i < 8; i += 2) {
 #define R(a,b) (((a) << (b)) | ((a) >> (32 - (b))))
 		/* Operate on columns. */
@@ -540,7 +552,7 @@ blockmix_salsa8(uint32_t * Bin, uint32_t * Bout, uint32_t * X, size_t r)
 	size_t i;
 
 	/* 1: X <-- B_{2r - 1} */
-	memcpy(X, &Bin[(2 * r - 1) * 16], 64);
+	blkcpy(X, &Bin[(2 * r - 1) * 16], 64);
 
 	/* 2: for i = 0 to 2r - 1 do */
 	for (i = 0; i < 2 * r; i += 2) {
@@ -550,7 +562,7 @@ blockmix_salsa8(uint32_t * Bin, uint32_t * Bout, uint32_t * X, size_t r)
 
 		/* 4: Y_i <-- X */
 		/* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
-		memcpy(&Bout[i * 8], X, 64);
+		blkcpy(&Bout[i * 8], X, 64);
 
 		/* 3: X <-- H(X \xor B_i) */
 		blkxor(X, &Bin[i * 16 + 16], 64);
@@ -558,7 +570,7 @@ blockmix_salsa8(uint32_t * Bin, uint32_t * Bout, uint32_t * X, size_t r)
 
 		/* 4: Y_i <-- X */
 		/* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
-		memcpy(&Bout[i * 8 + r * 16], X, 64);
+		blkcpy(&Bout[i * 8 + r * 16], X, 64);
 	}
 }
 
@@ -599,13 +611,13 @@ 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 */
-		memcpy(&V[i * (32 * r)], X, 128 * r);
+		blkcpy(&V[i * (32 * r)], X, 128 * r);
 
 		/* 4: X <-- H(X) */
 		blockmix_salsa8(X, Y, Z, r);
 
 		/* 3: V_i <-- X */
-		memcpy(&V[(i + 1) * (32 * r)], Y, 128 * r);
+		blkcpy(&V[(i + 1) * (32 * r)], Y, 128 * r);
 
 		/* 4: X <-- H(X) */
 		blockmix_salsa8(Y, X, Z, r);