/** @brief returns whether or not the profile is likely to be slow on a particular device
  *  @param dev the given device*/
 bool is_slow(viennacl::ocl::device const & dev) const{
   bool res = false;
   if(dev.type()==CL_DEVICE_TYPE_GPU){
     std::size_t warp_size = 32;
     if(dev.vendor_id()==4098)
       warp_size = 64;
     res = static_cast<bool>(((local_size_1_*local_size_2_)%warp_size)>0);
   }
   return res || is_slow_impl(dev);
 }