Decode in parallel

[eigmax]

Test if we can decode in parallel, then perform it. https://github.com/zkMIPS/zkm/blob/main/prover/src/witness/transition.rs#L531

[VanhGer]

In the simulate_cpu function, we call transition(state, kernel) sequentially to update the state. Inside the transition function, we use try_perform_instruction once, where we decode the state.registers once. So the simplest solution to parallelize this computation is by parallelizing the matching pattern inside the decode function.

To do that, we need to create a vector consists of all possible patterns and their corresponding results, allowing us to use iter_par for parallel processing. However, this approach takes a lot of memory, and slows down the decode because of the time required to construct the vector. So it does not bring any tangible benefit.

We think about changing the whole simulator process for optimization.