A statistics-driven micro-benchmarking framework heavily inspired by the wonderful criterion library for Haskell; originally created by LemonBoy.

Works, API is still not 100% stable yet.

import criterion
var cfg = newDefaultConfig()
benchmark cfg:
  func fib(n: int): int =
    case n
    of 0: 1
    of 1: 1
    else: fib(n-1) + fib(n-2)
  # on nim-1.0 you have to use {.measure: [].} instead
  proc fib5() {.measure.} =
    var n = 5
    blackBox fib(n)
  # ... equivalent to ...
  iterator argFactory(): int =
    for x in [5]:
      yield x
  proc fibN(x: int) {.measure: argFactory.} =
    blackBox fib(x)
  # ... equivalent to ...
  proc fibN1(x: int) {.measure: [5].} =
    blackBox fib(x)

Gives the following output:

A bit too much info? Just set cfg.brief = true and the results will be output in a condensed format:

Much easier to parse, isn't it?

If you need to pass more than a single argument to your benchmark fixture just use a tuple: they are automagically unpacked at compile-time.

import criterion
let cfg = newDefaultConfig()
benchmark cfg:
  proc foo(x: int, y: float) {.measure: [(1,1.0),(2,2.0)].} =
    discard x.float + y

If you need the measurement data in order to compare different benchmarks, to plot the results or to post-process them you can do so by adding a single line to your benchmark setup:

let cfg = newDefaultConfig()
# Your usual config goes here...
cfg.outputPath = "my_benchmark.json"
benchmark(cfg):
  # Your benchmark code goes here...

The data will be dumped once the block has been completed into a json file that's ready for consumption by other tools.

See the documentation for the criterion module as generated directly from the source.

The source of the many test.

MIT