add --enable-bound-port-transfer config option

This uses gcc extensions to remove one level of indirection from the
virtual function calls done by push/pull.  I have not actually seen a
measurable performance improvement from this (in my very limited
testing).  I am curious if anyone else sees any performance win from
it though.

This work was sponsored by Meraki, Inc.

Signed-off-by: Cliff Frey <cliff@meraki.com>

configure

@@ -773,6 +773,7 @@ enable_experimental
 enable_skip_elements
 with_freebsd
 enable_int64
+enable_bound_port_transfer
 enable_nanotimestamp
 enable_tools
 enable_dynamic_linking
@@ -1450,6 +1451,7 @@ Optional Features:
   --enable-experimental   enable experimental elements in normal groups
   --enable-skip-elements=ELTS disable comma-separated elements
   --disable-int64         disable 64-bit integer support
+  --enable-bound-port-transfer  enable port transfer function ptr optimization
   --enable-nanotimestamp  enable nanosecond timestamps
   --enable-tools=WHERE    enable tools (host/build/mixed/no) [mixed]
   --disable-dynamic-linking disable dynamic linking
@@ -7014,7 +7016,7 @@ else
 
 	    	    ac_cv_endian=0
 	    cat > conftest.$ac_ext <<EOF
-#line 7017 "configure"
+#line 7019 "configure"
 #include "confdefs.h"
 #include <$endian_hdr>
 #ifdef __BYTE_ORDER
@@ -7638,6 +7640,23 @@ $as_echo "#define HAVE_CLOCK_GETTIME 1" >>confdefs.h
     fi
 
 
+
+
+# Check whether --enable-bound-port-transfer was given.
+if test "${enable_bound_port_transfer+set}" = set; then :
+  enableval=$enable_bound_port_transfer; :
+else
+  enable_bound_port_transfer=no
+fi
+
+
+if test "$enable_bound_port_transfer" = yes; then
+
+$as_echo "#define HAVE_BOUND_PORT_TRANSFER 1" >>confdefs.h
+
+  CXXFLAGS="$CXXFLAGS -Wno-pmf-conversions"
+fi
+
 # Check whether --enable-nanotimestamp was given.
 if test "${enable_nanotimestamp+set}" = set; then :
   enableval=$enable_nanotimestamp;

configure.in

@@ -445,6 +445,20 @@ if test "x$have_timespec" = xyes; then
 fi
 CLICK_CHECK_POSIX_CLOCKS
 
+
+dnl
+dnl check forms of port transfer
+dnl
+
+AC_ARG_ENABLE(bound-port-transfer,
+  [[  --enable-bound-port-transfer  enable port transfer function ptr optimization]],
+  :, enable_bound_port_transfer=no)
+
+if test "$enable_bound_port_transfer" = yes; then
+  AC_DEFINE([HAVE_BOUND_PORT_TRANSFER], [1], [Define if Port::push/Port::pull should use bound function pointers.])
+  CXXFLAGS="$CXXFLAGS -Wno-pmf-conversions"
+fi
+
 AC_ARG_ENABLE(nanotimestamp, [  --enable-nanotimestamp  enable nanosecond timestamps])
 # Default to nanosecond-precision timestamps if clock_gettime is available.
 if test "x$enable_nanotimestamp" = x -a "x$have_clock_gettime" = xyes; then

include/click/element.hh

@@ -213,6 +213,12 @@ class Element { public:
 
 	Element* _e;
 	int _port;
+#if HAVE_BOUND_PORT_TRANSFER
+	union {
+	    void (*push)(Element *e, int port, Packet *p);
+	    Packet *(*pull)(Element *e, int port);
+	} _bound;
+#endif
 
 #if CLICK_STATS >= 1
 	mutable unsigned _packets;	// How many packets have we moved?
@@ -509,6 +515,17 @@ Element::Port::assign(Element *owner, Element *e, int port, bool isoutput)
     _e = e;
     _port = port;
     (void) isoutput;
+#if HAVE_BOUND_PORT_TRANSFER
+    if (e) {
+	if (isoutput) {
+	    void (Element::*pusher)(int, Packet *) = &Element::push;
+	    _bound.push = (void (*)(Element *, int, Packet *)) (e->*pusher);
+	} else {
+	    Packet *(Element::*puller)(int) = &Element::pull;
+	    _bound.pull = (Packet *(*)(Element *, int)) (e->*puller);
+	}
+    }
+#endif
 }
 
 /** @brief Returns whether this port is active (a push output or a pull input).
@@ -570,13 +587,21 @@ Element::Port::push(Packet* p) const
 #if CLICK_STATS >= 2
     ++_e->input(_port)._packets;
     click_cycles_t c0 = click_get_cycles();
+# if HAVE_BOUND_PORT_TRANSFER
+    _bound.push(_e, _port, p);
+# else
     _e->push(_port, p);
+# endif
     click_cycles_t x = click_get_cycles() - c0;
     ++_e->_calls;
     _e->_self_cycles += x;
     _owner->_child_cycles += x;
 #else
+# if HAVE_BOUND_PORT_TRANSFER
+    _bound.push(_e, _port, p);
+# else
     _e->push(_port, p);
+# endif
 #endif
 }
 
@@ -602,15 +627,23 @@ Element::Port::pull() const
     assert(_e);
 #if CLICK_STATS >= 2
     click_cycles_t c0 = click_get_cycles();
+# if HAVE_BOUND_PORT_TRANSFER
+    Packet *p = _bound.pull(_e, _port);
+# else
     Packet *p = _e->pull(_port);
+# endif
     click_cycles_t x = click_get_cycles() - c0;
     ++_e->_calls;
     _e->_self_cycles += x;
     _owner->_child_cycles += x;
     if (p)
 	++_e->output(_port)._packets;
 #else
+# if HAVE_BOUND_PORT_TRANSFER
+    Packet *p = _bound.pull(_e, _port);
+# else
     Packet *p = _e->pull(_port);
+# endif
 #endif
 #if CLICK_STATS >= 1
     if (p)