Some optimisations

Signed-off-by: Mattias Andrée <maandree@kth.se>

1 files changed, 32 insertions, 16 deletions

diff --git a/src/zxor.c b/src/zxor.c
index f973774..601a947 100644
--- a/src/zxor.c
+++ b/src/zxor.c
@@ -2,18 +2,33 @@
 #include "internals.h"
 
 
-static inline void
-zxor_impl(zahl_char_t *restrict a, const zahl_char_t *restrict b, size_t n)
+O2 static inline void
+zxor_impl_3(register zahl_char_t *restrict a,
+	    register const zahl_char_t *restrict b, size_t n)
 {
 	size_t i;
 	for (i = 0; i < n; i++)
 		a[i] ^= b[i];
 }
 
+static inline void
+zxor_impl_5(register zahl_char_t *restrict a,
+	    register const zahl_char_t *restrict b, size_t n,
+	    register const zahl_char_t *restrict c, size_t m)
+{
+	size_t i;
+	for (i = 0; i < n; i++)
+		a[i] = b[i] ^ c[i];
+	for (; i < m; i++)
+		a[i] = c[i];
+}
+
 void
 zxor(z_t a, z_t b, z_t c)
 {
-	size_t n, m;
+	size_t n, m, bn, cn;
+	const zahl_char_t *restrict bc;
+	const zahl_char_t *restrict cc;
 
 	if (unlikely(zzero(b))) {
 		SET(a, c);
@@ -23,25 +38,26 @@ zxor(z_t a, z_t b, z_t c)
 		return;
 	}
 
-	m = MAX(b->used, c->used);
-	n = b->used + c->used - m;
+	bn = b->used;
+	bc = b->chars;
+	cn = c->used;
+	cc = c->chars;
 
+	MIN_MAX_1(n, m, bn, cn);
 	ENSURE_SIZE(a, m);
 
 	if (a == b) {
-		if (a->used < c->used)
-			zmemcpy(a->chars + n, c->chars + n, m - n);
-		zxor_impl(a->chars, c->chars, n);
+		zxor_impl_3(a->chars, cc, n);
+		if (a->used < cn)
+			zmemcpy_range(a->chars, cc, n, m);
 	} else if (unlikely(a == c)) {
-		if (a->used < b->used)
-			zmemcpy(a->chars + n, b->chars + n, m - n);
-		zxor_impl(a->chars, b->chars, n);
-	} else if (m == b->used) {
-		zmemcpy(a->chars, b->chars, m);
-		zxor_impl(a->chars, c->chars, n);
+		zxor_impl_3(a->chars, bc, n);
+		if (a->used < bn)
+			zmemcpy_range(a->chars, bc, n, m);
+	} else if (m == bn) {
+		zxor_impl_5(a->chars, cc, n, bc, m);
 	} else {
-		zmemcpy(a->chars, c->chars, m);
-		zxor_impl(a->chars, b->chars, n);
+		zxor_impl_5(a->chars, bc, n, cc, m);
 	}
 
 	a->used = m;