[PATCH] netfilter: unfold two critical loops in ip_packet_match()

Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]
From: Eric Dumazet
Date: Thursday, January 29, 2009 - 8:31 am

While doing oprofile tests I noticed two loops are not properly unrolled by gcc

Using a hand coded unrolled loop provides nice speedup : ipt_do_table
credited of 2.52 % of cpu instead of 3.29 % in tbench.


Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
---
 net/ipv4/netfilter/ip_tables.c |   32 +++++++++++++++++--------------
 1 files changed, 18 insertions(+), 14 deletions(-)

diff --git a/net/ipv4/netfilter/ip_tables.c b/net/ipv4/netfilter/ip_tables.c
index ef8b6ca..7bdb2cf 100644
--- a/net/ipv4/netfilter/ip_tables.c
+++ b/net/ipv4/netfilter/ip_tables.c
@@ -65,6 +65,22 @@ do {								\
 #define inline
 #endif
 
+static unsigned long ifname_compare(const void *_a, const void *_b, const void *_mask)
+{
+	const unsigned long *a = (const unsigned long *)_a;
+	const unsigned long *b = (const unsigned long *)_b;
+	const unsigned long *mask = (const unsigned long *)_mask;
+	unsigned long ret;
+
+	ret = (a[0] ^ b[0]) & mask[0];
+	ret |= (a[1] ^ b[1]) & mask[1];
+	if (IFNAMSIZ > 2 * sizeof(unsigned long))
+		ret |= (a[2] ^ b[2]) & mask[2];
+	if (IFNAMSIZ > 3 * sizeof(unsigned long))
+		ret |= (a[3] ^ b[3]) & mask[3];
+	return ret;
+}
+
 /*
    We keep a set of rules for each CPU, so we can avoid write-locking
    them in the softirq when updating the counters and therefore
@@ -83,7 +99,6 @@ ip_packet_match(const struct iphdr *ip,
 		const struct ipt_ip *ipinfo,
 		int isfrag)
 {
-	size_t i;
 	unsigned long ret;
 
 #define FWINV(bool, invflg) ((bool) ^ !!(ipinfo->invflags & (invflg)))
@@ -103,13 +118,7 @@ ip_packet_match(const struct iphdr *ip,
 		return false;
 	}
 
-	/* Look for ifname matches; this should unroll nicely. */
-	for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
-		ret |= (((const unsigned long *)indev)[i]
-			^ ((const unsigned long *)ipinfo->iniface)[i])
-			& ((const unsigned long *)ipinfo->iniface_mask)[i];
-	}
-
+	ret = ifname_compare(indev, ipinfo->iniface, ipinfo->iniface_mask);
 	if (FWINV(ret != 0, IPT_INV_VIA_IN)) {
 		dprintf("VIA in mismatch (%s vs %s).%s\n",
 			indev, ipinfo->iniface,
@@ -117,12 +126,7 @@ ip_packet_match(const struct iphdr *ip,
 		return false;
 	}
 
-	for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
-		ret |= (((const unsigned long *)outdev)[i]
-			^ ((const unsigned long *)ipinfo->outiface)[i])
-			& ((const unsigned long *)ipinfo->outiface_mask)[i];
-	}
-
+	ret = ifname_compare(outdev, ipinfo->outiface, ipinfo->outiface_mask);
 	if (FWINV(ret != 0, IPT_INV_VIA_OUT)) {
 		dprintf("VIA out mismatch (%s vs %s).%s\n",
 			outdev, ipinfo->outiface,
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]

Messages in current thread:
32 core net-next stack/netfilter &quot;scaling&quot;, Rick Jones, (Mon Jan 26, 3:15 pm)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Stephen Hemminger, (Mon Jan 26, 4:14 pm)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Patrick McHardy, (Tue Jan 27, 2:15 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Patrick McHardy, (Tue Jan 27, 4:37 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Patrick McHardy, (Tue Jan 27, 10:33 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Patrick McHardy, (Wed Jan 28, 9:25 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Eric Dumazet, (Wed Jan 28, 10:07 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Eric Dumazet, (Wed Jan 28, 10:34 am)
[PATCH] netfilter: unfold two critical loops in ip_packet_ ..., Eric Dumazet, (Thu Jan 29, 8:31 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Patrick McHardy, (Mon Feb 9, 7:57 am)
Re: 32 core net-next stack/netfilter &quot;scaling&quot;, Stephen Hemminger, (Tue Feb 10, 11:44 am)
[PATCH] netfilter: xt_physdev fixes, Eric Dumazet, (Wed Feb 18, 10:36 am)
Re: [PATCH] netfilter: xt_physdev fixes, Patrick McHardy, (Wed Feb 18, 11:14 am)
[PATCH] netfilter: unfold two loops in physdev_mt(), Eric Dumazet, (Thu Feb 19, 1:00 am)
Re: [PATCH] netfilter: unfold two loops in physdev_mt(), Patrick McHardy, (Thu Feb 19, 3:17 am)