Since programs use virtual addresses and the cache operates on only physical addresses, does the a TLB(and potential page table access) have to happen before accessing the physical cache? If not, how would it be possible to optimize data/instruction accesses to access both in parallel?