Puzzle 16: Attempt to speed up the "reverse" function

Even though we change from 3 small memcpy()s per 2 items, to 1 small
memcpy() per item plus 1 big memcpy(), it goes slower?

(On part 1 of the puzzle with a 40 minute limit, the timing goes from
6.6s to 6.8s, average of 5 runs.)

Leave commented out for now.

1 files changed, 10 insertions, 0 deletions

diff --git a/16.c b/16.c
index 15e9671..bb565bf 100644
--- a/16.c
+++ b/16.c
@@ -34,6 +34,16 @@ qreverse(void * base, size_t nmemb, size_t size)
 	if (nmemb < 2)
 		return;
 
+#if 0	// This goes slower?!?
+	if (nmemb < BUFSIZ / size) {
+		char tmp[nmemb * size];
+		memcpy(tmp, base, nmemb * size);
+		for (i = 0; i < nmemb; ++i)
+			memcpy(base + i * size, tmp + (nmemb - 1 - i) * size, size);
+		return;
+	}
+#endif
+
 	for (i = 0, j = nmemb - 1; i < j; ++i, --j)
 		qswap(base, size, i, j);
 }