[LANG] Added __debug_barrier() call to force insertion of a CUDA

__syncthreads

lib/lang/code_gen.cc

@@ -348,6 +348,10 @@ void Generator::VisitFuncCall(FuncCall* funcCall) {
     ir::value* false_val = ret_;
     return set_ret(bld_->create_select(cond, true_val, false_val));
   }
+  if(name == "__debug_barrier"){
+    bld_->create_barrier();
+    return;
+  }
   return error_not_implemented("function calls not implemented");
 }
 

lib/runtime/function.cc

@@ -102,6 +102,7 @@ extern void atomic_add_float_1x1(float*, float, bool);
 extern int atomic_cas(int*, int, int);
 extern int atomic_xchg(int*, int);
 extern int get_program_id(int);
+extern void __debug_barrier();
 extern int get_num_programs(int);
 extern int select(bool, int, int);
 extern char __constant__ * calloc(int);