Medium
A 20-cycle divide ahead of two independent single-cycle adds
An out-of-order superscalar core issues three independent instructions in program order: I1 (a 20-cycle-latency integer divide), I2 and I3 (single-cycle adds with no dependency on I1 or each other). The scheduler dispatches all three as soon as their operands are ready, so I2 and I3 finish execution long before I1 does. What governs when I2's and I3's results become visible to the rest of the machine (committed to architectural state)?
6 of 47 in Computer Architecture