From 27038b159b50e69ea216a513edbd6ee68cdf89aa Mon Sep 17 00:00:00 2001
From: Mukund Sivaraman <muks@isc.org>
Date: Sun, 10 Jul 2016 19:46:17 +0530
Subject: [PATCH] Use GCC builtin for clz in RPZ lookup code (#42818)

---
 CHANGES       |  3 +++
 config.h.in   |  3 +++
 configure     | 39 +++++++++++++++++++++++++++++++++++++++
 configure.in  | 17 +++++++++++++++++
 lib/dns/rpz.c | 31 +++++++++++++++++++++----------
 5 files changed, 83 insertions(+), 10 deletions(-)

diff --git a/CHANGES b/CHANGES
index 00bf399789..ffc15da81f 100644
--- a/CHANGES
+++ b/CHANGES
@@ -1,3 +1,6 @@
+4407.	[performance]	Use GCC builtin for clz in RPZ lookup code.
+			[RT #42818]
+
 4406.	[bug]		getrrsetbyname with a non absolute name could
 			trigger a infinite recursion bug in lwresd
 			and named with lwres configured if when combined
diff --git a/config.h.in b/config.h.in
index 0f885b1329..f3f947ebc6 100644
--- a/config.h.in
+++ b/config.h.in
@@ -194,6 +194,9 @@ int sigwait(const unsigned int *set, int *sig);
    MSVC and with C++ compilers. */
 #undef FLEXIBLE_ARRAY_MEMBER
 
+/* Define to 1 if the compiler supports __builtin_clz. */
+#undef HAVE_BUILTIN_CLZ
+
 /* Define to 1 if the compiler supports __builtin_expect. */
 #undef HAVE_BUILTIN_EXPECT
 
diff --git a/configure b/configure
index bc9bb81fbf..61222f1806 100755
--- a/configure
+++ b/configure
@@ -20183,6 +20183,45 @@ $as_echo "#define HAVE_BUILTIN_EXPECT 1" >>confdefs.h
 
 fi
 
+#
+# Check for __builtin_clz
+#
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking compiler support for __builtin_clz" >&5
+$as_echo_n "checking compiler support for __builtin_clz... " >&6; }
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+        return (__builtin_clz(0xff) == 24 ? 1 : 0);
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+
+        have_builtin_clz=yes
+        { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+
+else
+
+        have_builtin_clz=no
+        { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+if test "$have_builtin_clz" = "yes"; then
+
+$as_echo "#define HAVE_BUILTIN_CLZ 1" >>confdefs.h
+
+fi
+
 #
 # CPU relax (for spin locks)
 #
diff --git a/configure.in b/configure.in
index 2da78c8f62..9bf4aa2f8d 100644
--- a/configure.in
+++ b/configure.in
@@ -4007,6 +4007,23 @@ if test "$have_builtin_expect" = "yes"; then
         AC_DEFINE(HAVE_BUILTIN_EXPECT, 1, [Define to 1 if the compiler supports __builtin_expect.])
 fi
 
+#
+# Check for __builtin_clz
+#
+AC_MSG_CHECKING([compiler support for __builtin_clz])
+AC_TRY_LINK(, [
+        return (__builtin_clz(0xff) == 24 ? 1 : 0);
+], [
+        have_builtin_clz=yes
+        AC_MSG_RESULT(yes)
+], [
+        have_builtin_clz=no
+        AC_MSG_RESULT(no)
+])
+if test "$have_builtin_clz" = "yes"; then
+        AC_DEFINE(HAVE_BUILTIN_CLZ, 1, [Define to 1 if the compiler supports __builtin_clz.])
+fi
+
 #
 # CPU relax (for spin locks)
 #
diff --git a/lib/dns/rpz.c b/lib/dns/rpz.c
index b90b148982..ddd6eb2e1a 100644
--- a/lib/dns/rpz.c
+++ b/lib/dns/rpz.c
@@ -970,34 +970,43 @@ name2data(dns_rpz_zones_t *rpzs, dns_rpz_num_t rpz_num,
 	(void)dns_name_concatenate(&tmp_name, dns_rootname, trig_name, NULL);
 }
 
-/*
- * Find the first differing bit in a key (IP address) word.
+#ifndef HAVE_BUILTIN_CLZ
+/**
+ * \brief Count Leading Zeros: Find the location of the left-most set
+ * bit.
  */
-static inline int
-ffs_keybit(dns_rpz_cidr_word_t w) {
-	int bit;
+static inline unsigned int
+clz(dns_rpz_cidr_word_t w) {
+	unsigned int bit;
 
 	bit = DNS_RPZ_CIDR_WORD_BITS-1;
+
 	if ((w & 0xffff0000) != 0) {
 		w >>= 16;
 		bit -= 16;
 	}
+
 	if ((w & 0xff00) != 0) {
 		w >>= 8;
 		bit -= 8;
 	}
+
 	if ((w & 0xf0) != 0) {
 		w >>= 4;
 		bit -= 4;
 	}
+
 	if ((w & 0xc) != 0) {
 		w >>= 2;
 		bit -= 2;
 	}
+
 	if ((w & 2) != 0)
 		--bit;
+
 	return (bit);
 }
+#endif
 
 /*
  * Find the first differing bit in two keys (IP addresses).
@@ -1016,12 +1025,14 @@ diff_keys(const dns_rpz_cidr_key_t *key1, dns_rpz_prefix_t prefix1,
 	/*
 	 * find the first differing words
 	 */
-	for (i = 0;
-	     bit < maxbit;
-	     i++, bit += DNS_RPZ_CIDR_WORD_BITS) {
+	for (i = 0; bit < maxbit; i++, bit += DNS_RPZ_CIDR_WORD_BITS) {
 		delta = key1->w[i] ^ key2->w[i];
-		if (delta != 0) {
-			bit += ffs_keybit(delta);
+		if (ISC_UNLIKELY(delta != 0)) {
+#ifdef HAVE_BUILTIN_CLZ
+			bit += __builtin_clz(delta);
+#else
+			bit += clz(delta);
+#endif
 			break;
 		}
 	}