Hacker News Reader

Comment by donohoe | Hacker News Reader

What surprises me is are some of the results like this:

  IE 7.0.5730.11 / WinXP SP2
  Native Array (length=1000, looped 100 times)

While loop in reverse, simplifying the test condition.

  var i = arr.length; while (i--) {}

Took 0ms to execute... very unexpected

View on HN

It makes sense that a zero comparison is faster, but 0ms does seem strange. It's 31ms for a basic for loop on the same platform, whereas in Safari the times are 2ms and 5ms respectively, which seems a lot more reasonable.

Is it possible that IE 7 is deleting the whole loop as a no-op?

On IA-32, a decrementing-until-zero loop gets its own instruction (called—unsurprisingly—LOOP.) If it optimized in that one case, it would take microseconds to execute the whole test, which would show up as 0ms.

Isn't LOOP one of those microcoded instructions that actually runs slower on modern processors than a RISC-style separate decrement and compare?

It's a slow instruction, but it's a single instruction—in a loop with an empty body, this creates the best possible situation for memory locality: you just keep reading the same memory address/cache line over and over.

Reply on news.ycombinator.com