Stochastic superoptmization

  • Loop-free binary super optimization task as a Stochastic search problem
  • MCMC sampler
  • The program optimization task is formulated as a cost minimization problem
  • For instance, rather than considering the trade-offs be- tween register allocation and instruction selection, we simply de- fine a term which reflects the primary consequence of the decision: expected runtime.
  • How is cost-formula related to the decision made during the Markov-process powered sampling? How did we get the distribution in the end?