Hand unrolled version of dslash in a separate class.

Useful to compare; raises Intel compiler from 9GFlop/s to 17.5 Gflops.
                   on ivybridge core. Raises Clang form 14.5 to 17.5

lib/Grid_init.cc

@@ -142,6 +142,9 @@ void Grid_init(int *argc,char ***argv)
   if( !GridCmdOptionExists(*argv,*argv+*argc,"--debug-stdout") ){
     Grid_quiesce_nodes();
   }
+  if( GridCmdOptionExists(*argv,*argv+*argc,"--dslash-opt") ){
+    WilsonMatrix::HandOptDslash=1;
+  }
   GridParseLayout(*argv,*argc,
 		  Grid_default_latt,
 		  Grid_default_mpi);