A partial rollback of the refactoring done in [4032]. It was causing a 4-6%...

A partial rollback of the refactoring done in [4032]. It was causing a 4-6% speed decrease on certain hierarchy selectors ( > + and ~ ) since it did an extra .toUpperCase() which wasn't required.

The part left in was moving one of the .toUpperCase() calls to the var nodeName instead of having it in the loop.  This appears to be giving a speed boost of a couple percent for those same hierarchy selectors.

src/selector.js

@@ -104,7 +104,7 @@ jQuery.extend({
 		context = context || document;
 
 		// Initialize the search
-		var ret = [context], done = [], last;
+		var ret = [context], done = [], last, nodeName;
 
 		// Continue while a selector expression exists, and while
 		// we're no longer looping upon ourselves
@@ -122,12 +122,12 @@ jQuery.extend({
 			var m = re.exec(t);
 
 			if ( m ) {
-				var nodeName = m[1];
+				nodeName = m[1].toUpperCase();
 
 				// Perform our own iteration and filter
 				for ( var i = 0; ret[i]; i++ )
 					for ( var c = ret[i].firstChild; c; c = c.nextSibling )
-						if ( c.nodeType == 1 && (nodeName == "*" || jQuery.nodeName(c, nodeName)) )
+						if ( c.nodeType == 1 && (nodeName == "*" || c.nodeName.toUpperCase() == nodeName) )
 							r.push( c );
 
 				ret = r;
@@ -140,7 +140,7 @@ jQuery.extend({
 				if ( (m = re.exec(t)) != null ) {
 					r = [];
 
-					var nodeName = m[2], merge = {};
+					nodeName = m[2].toUpperCase(), merge = {};
 					m = m[1];
 
 					for ( var j = 0, rl = ret.length; j < rl; j++ ) {
@@ -151,7 +151,7 @@ jQuery.extend({
 
 								if ( m == "~" && merge[id] ) break;
 								
-								if (!nodeName || jQuery.nodeName(n, nodeName)) {
+								if (!nodeName || n.nodeName.toUpperCase() == nodeName ) {
 									if ( m == "~" ) merge[id] = true;
 									r.push( n );
 								}
@@ -436,3 +436,4 @@ jQuery.extend({
 		return r;
 	}
 });
+