> This seems like a trick question… just an add right? This sort of thing is part of why you'd often like to target an intermediate vector representation and l

> This seems like a trick question… just an add right? This sort of thing is part of why you'd often like to target an intermediate vector representation and let the compiler figure out the details. E.g., on Haswell chips you had multiple floating point execution units per core, and the CPU could definitely execute more than one pipelined FP operation simultaneously, but only one of those could be an `add` instruction. If you had a bunch of additions to do which didn't rely on the previously computed results (avoiding stalls), you could double your addition throughput by also dispatching a fused-multiply-add instruction where the multiplicand was just one. That _could_ execute at the same time as a normal vector FP addition.