Optimize swap for larger swaps

1 files changed, 14 insertions, 0 deletions

diff --git a/common.h b/common.h
index d7e1e47..c1bfb6c 100644
--- a/common.h
+++ b/common.h
@@ -7,6 +7,20 @@ size_t binary_search(const char *needle, char *haystack, size_t nmel, size_t wid
 
 static void swap(char *a, char *b, size_t width)
 {
+#ifdef __GNUC__
+	typedef uint32_t __attribute__((__may_alias__)) u32;
+
+	if ((uintptr_t)a % 4 == 0 && (uintptr_t)b % 4 == 0) {
+		for (; width >= 4; width -= 4) {
+			uint32_t tmp = *((u32*)a);
+			*((u32*)a) = *((u32*)b);
+			*((u32*)b) = tmp;
+			a += 4;
+			b += 4;
+		}
+	}
+#endif
+
 	while (width--) {
 		char tmp = *a;
 		*a++ = *b;