Towards optimizing energy costs of algorithms for shared memory architectures