Which I think is the argument behind declarative languages. If we can encode the intent precisely, then at least in theory the compiler can find the optimal sol

Which I think is the argument behind declarative languages. If we can encode the intent precisely, then at least in theory the compiler can find the optimal solution for the given architecture and other constraints. Of course in practice that probably requires strong AI to accomplish.