# Reason About CPU Microarchitecture and Simulator Design Answer the following systems questions for a modern out-of-order CPU. State architectural assumptions when behavior can vary by implementation. ### Part 1: Trace an Instruction Given a short assembly sequence containing a conditional branch, a load, and a dependent arithmetic instruction, explain how it moves through fetch, prediction, decode, rename, scheduling, execution, completion, and retirement. Describe what happens on a branch misprediction and on a cache miss. #### What This Part Should Cover - Branch prediction and speculative fetch - Register renaming and reorder-buffer state - Dependency tracking and out-of-order issue - Precise recovery and in-order retirement ### Part 2: Explain the Memory Hierarchy Explain the roles and interactions of caches, MOESI coherence states, miss-status holding registers, virtually indexed physically tagged caches, the TLB, and a hardware page walk. #### What This Part Should Cover - Coherence permissions and dirty-data ownership - Tracking several outstanding misses - VIPT index and tag timing constraints - Translation hits, misses, page walks, and resulting cache accesses ### Part 3: Design a Branch-Predictor Simulator Module Design a C++ module for simulating a branch predictor. Define the interface, state, update timing, statistics, and tests. Explain how speculative updates are recovered if the chosen model includes them. #### What This Part Should Cover - Predict and update APIs with stable branch identity - Separation of predictor policy from simulator plumbing - Deterministic state and reproducible traces - Accuracy, aliasing, warm-up, and rollback tests ### Part 4: Choose a Simulation Model Compare cycle-based, event-based, cycle-accurate, and cycle-approximate simulation. Explain which model you would choose for correctness exploration, detailed performance work, and fast design-space search, and how you would optimize it. #### What This Part Should Cover - Fidelity versus speed trade-offs - Event scheduling and idle-period skipping - Modeling of contention and backpressure - Profiling, batching, parallelism limits, and validation against a reference ### What a Strong Answer Covers A strong answer connects each named structure to the instruction's progress, separates architectural correctness from timing fidelity, and designs simulator interfaces that are testable, deterministic, and explicit about approximation. ### Follow-up Questions - How can a VIPT L1 cache avoid synonym problems? - When is an Owned coherence state useful? - Which predictor updates must be undone after a squash? - How would you validate a fast approximate simulator?