From: Nicholas Nethercote <njn@valgrind.org>
Date: Sun, 25 Dec 2005 02:47:12 +0000 (+0000)
Subject: A minor performance improvement -- make swizzle() faster.
X-Git-Tag: svn/VALGRIND_3_2_0~448
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=e830c38ce8191762d3c1413bf8864aaaa9908529;p=thirdparty%2Fvalgrind.git

A minor performance improvement -- make swizzle() faster.



git-svn-id: svn://svn.valgrind.org/valgrind/trunk@5431
---

diff --git a/coregrind/m_mallocfree.c b/coregrind/m_mallocfree.c
index 98d03a9a06..faf3e56e6f 100644
--- a/coregrind/m_mallocfree.c
+++ b/coregrind/m_mallocfree.c
@@ -704,7 +704,12 @@ void swizzle ( Arena* a, UInt lno )
    if (p_best == NULL) return;
 
    pn = pp = p_best;
-   for (i = 0; i < 20; i++) {
+
+   // This loop bound was 20 for a long time, but experiments showed that
+   // reducing it to 10 gave the same result in all the tests, and 5 got the
+   // same result in 85--100% of cases.  And it's called often enough to be
+   // noticeable in programs that allocated a lot.
+   for (i = 0; i < 5; i++) {
       pn = get_next_b(pn);
       pp = get_prev_b(pp);
       if (pn < p_best) p_best = pn;