Crystal: Cost too much time to run .cr file

[Freakwill]

It cost too much time to run a crystal file
I write a crystal file to compare with other languages in finding primes
It show that it only costed 04.03ms, but I wait several seconds to see the ouput.

[~] crystal run --release Programming/Benchmarks/prime.cr             20:12:44
primes   9.61  (104.03ms) (± 2.92%)  66292569 B/op  fastest

[j8r]

Crystal is a compiled language, compile once, then run this binary how many times you want. That's what crystal run does.

[Freakwill]

Crystal is a compiled language, compile once, then run this binary how many times you want. That's what crystal run does.

Yes, I know, but It is slower then I thought. It spent several seconds to show the report, though it said that the process only costed 104.04ms.

[~] crystal build --release Programming/Benchmarks/prime.cr 
[~] ./prime
primes   9.61  (104.04ms) (± 5.91%)  66292494 B/op  fastest

[jhass]

This looks like the benchmark/ips output. The 104ms is the average runtime, the library runs it in a loop until it gets a somewhat stable value, including a couple of times before even measuring to warm up any caches.

[vladfaust]

To be exact, Benchmark#ips does 2 seconds warmup + 5 seconds calculation by default.

[asterite]

It would also help if you provide some code that we can run, and what times you expect.

[Freakwill]

It would also help if you provide some code that we can run, and what times you expect.

@asterite Thank you. I would be pleased that you can review https://github.com/Freakwill/benchmarks

[asterite]

@Freakwill Replace the last 3 lines of primes with this (translated from the Go code):

  p = [2]

  i = 0
  while i < (n - 1) / 2
    x = 3 + 2 * i
    p << x if f[x] == 1
    i += 1
  end

  p

In my case it goes from 113.37ms per call to 98.22ms per call.

But yeah, you get a slow execution because Benchmark.ips will run your code many times to benchmark it, as others said. If you don't want to do that, just do:

n = 10000000

time = Time.measure do
  primes(n)
end

puts time

(instead of using benchmark)