Benchmarking allocation time of Crystal objects with various quantities of instance variables

bench_allocations.cr

require "benchmark"

value = nil
Benchmark.ips do |x|
  x.report "1 ivar" { value = OneIVar.new }
  x.report "2 ivars" { value = TwoIVars.new }
  x.report "4 ivars" { value = FourIVars.new }
  x.report "8 ivars" { value = EightIVars.new }
  x.report "16 ivars" { value = SixteenIVars.new }
end

pp value

class OneIVar
  @var1 = "foo"
end

class TwoIVars
  @var1 = "foo"
  @var2 = "foo"
end

class FourIVars
  @var1 = "foo"
  @var2 = "foo"
  @var3 = "foo"
  @var4 = "foo"
end

class EightIVars
  @var1 = "foo"
  @var2 = "foo"
  @var3 = "foo"
  @var4 = "foo"
  @var5 = "foo"
  @var6 = "foo"
  @var7 = "foo"
  @var8 = "foo"
end

class SixteenIVars
  @var1 = "foo"
  @var2 = "foo"
  @var3 = "foo"
  @var4 = "foo"
  @var5 = "foo"
  @var6 = "foo"
  @var7 = "foo"
  @var8 = "foo"
  @var9 = "foo"
  @var10 = "foo"
  @var11 = "foo"
  @var12 = "foo"
  @var13 = "foo"
  @var14 = "foo"
  @var15 = "foo"
  @var16 = "foo"
end

results.txt

$ crystal run --release bench_allocations.cr
  1 ivar  67.35M ( 14.85ns) (± 1.10%)  32.0B/op        fastest
 2 ivars  66.60M ( 15.01ns) (± 0.99%)  32.0B/op   1.01× slower
 4 ivars  52.89M ( 18.91ns) (± 1.01%)  48.0B/op   1.27× slower
 8 ivars  36.03M ( 27.75ns) (± 1.14%)  80.0B/op   1.87× slower
16 ivars  21.68M ( 46.12ns) (± 1.05%)   144B/op   3.11× slower

s/class/struct/g/ yields equivalent measurements:

 1 ivar 571.66M (  1.75ns) (±14.32%)  0.0B/op        fastest
 2 ivars 505.86M (  1.98ns) (± 9.52%)  0.0B/op   1.13× slower
 4 ivars 500.13M (  2.00ns) (±10.77%)  0.0B/op   1.14× slower
 8 ivars 349.30M (  2.86ns) (±10.59%)  0.0B/op   1.64× slower
16 ivars 192.03M (  5.21ns) (± 9.80%)  0.0B/op   2.98× slower

So this benchmark does not reliably represent a difference based on allocation time.

Oh interesting! 👍🏼