Skip to content

Latest commit

 

History

History
750 lines (495 loc) · 24.3 KB

README.md

File metadata and controls

750 lines (495 loc) · 24.3 KB
Raw

fast-crystal

Benchmarks of common idioms in Crystal, to help write more performant code.

This project is a port to Crystal of fast-ruby project by JuanitoFatas.

I did this while learning Crystal, as an exercise to dive deeper into various aspects of the superb Crystal language. I also added some Crystal-specific benchmarks, such as comparing NamedTuple vs Hash. ( Spoiler: NamedTuples are much faster!), and removed some that don't make sense for Crystal (there is only Array#size in Crystal, no Array#length and Array#count is provided by Enumerable).

This is also a useful tool for those coming to Crystal from Ruby and other dynamic languages. Some of the assumptions I had about certain idioms in Ruby, are no longer true and maybe reverse in Crystal. For example in Crystal Array#find is faster on a sorted Array than Array#bsearch, but in Ruby it's the reverse

I'm planning to add some benchmarks for Sets, Tuples & NamedTuples as well as showcase some usa-cases vs Arrays & Hashes. If you have any intersting benchmarks, idioms, tips - please submit them!

How to contribute new benchmarks

  • Fork & clone .
  • Add a benchmark.
  • Update README.md with the results.
  • Make a Pull Request with a short explanation / rationale.
  • Kick-back and have a drink.
  • If you are not sure if the benchmark is warranted, open an issue - let's discuss it.

Template for the benchmark

require "benchmark"
puts Crystal::DESCRIPTION # Always include this, so that we know which OS & version of Crystal, you used.

def fast
end

def slow
end

Benchmark.ips do |x|
  x.report("fast code description") { fast }
  x.report("slow code description") { slow }
end

When you run your code don't forget to use --release flag, or you are not going to get accurate benchmarks (you don't need to build first), i.e.:

crystal code/array/array-first-vs-index.cr --release

Copy and paste output from console into README.md under appropriate section.

Results

Idioms

Index

Array

Array#[](0) vs Array#first code
crystal array-first-vs-index.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

  Array#[0] 405.11M (  2.47ns) (± 3.92%)  0 B/op        fastest
Array#first 358.67M (  2.79ns) (± 3.73%)  0 B/op   1.13× slower
Array#[](-1) vs Array#last code
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Array#[-1] 404.99M (  2.47ns) (± 3.10%)  0 B/op        fastest
Array#last 358.27M (  2.79ns) (± 4.23%)  0 B/op   1.13× slower
Array#bsearch vs Array#find code

WARNING: bsearch ONLY works on sorted array. bsearch is a bit faster on smaller Arrays, but still lags behind.

 crystal code/array/bsearch-vs-find.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

   find  396.3M (  2.52ns) (± 3.52%)  0 B/op        fastest
bsearch  12.11M ( 82.61ns) (± 2.32%)  0 B/op  32.74× slower
Array#shuffle.first vs Array#sample code

Array#shuffle allocates an extra array.
Array#sample indexes into the array without allocating an extra array.

crystal code/array/shuffle-first-vs-sample.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

       Array#sample  76.52M ( 13.07ns) (± 3.39%)    0 B/op         fastest
Array#shuffle.first  535.8k (  1.87µs) (± 2.43%)  832 B/op  142.81× slower
Array#insert vs Array#unshift code
crystal code/array/insert-vs-unshift.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Array#unshift   1.35  (742.07ms) (± 1.42%)  1573696 B/op        fastest
 Array#insert   1.34  (746.59ms) (± 1.20%)  1574201 B/op   1.01× slower

Range

covers? vs includes? vs plain comparison code

covers? is essentially an alias for includes?, so unlike in Ruby the performance is identical, however they maybe be faster than using > and < depending on the type of data you are dealing with. For Int32 there is no difference, but for Time(Dates) includes? is 1.5x faster.

crystal code/range/cover-vs-include.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

  range#covers? 461.38M (  2.17ns) (± 4.44%)  0 B/op   1.00× slower
range#includes? 463.13M (  2.16ns) (± 3.97%)  0 B/op        fastest
  plain compare 292.48M (  3.42ns) (± 4.50%)  0 B/op   1.58× slowe

Time (Date)

Time.iso8601 vs Time.parse code

If you know incoming datetime format it's better to use specific function. Unlike in Ruby there is no separate basic type - Date & Time. Everything is just Time. So there is no reason for a separate "Date" benchmark

crystal code/time/iso8601-vs-parse.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Time.parse_iso8601   9.12M (109.63ns) (± 1.95%)    0 B/op         fastest
        Time.parse  49.88k ( 20.05µs) (± 2.94%)  360 B/op  182.88× slower

Proc & Block

Block vs Symbol#to_proc code

There is a small performance penalty in Ruby for using &:shortcut , but not in crystal, so use away!

crystal code/proc-and-block/block-vs-to_proc.cr   --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
         Block  283.4k (  3.53µs) (± 3.31%)  2656 B/op        fastest
Symbol#to_proc  281.9k (  3.55µs) (± 3.10%)  2656 B/op   1.01× slower
Proc#call and block arguments vs yieldcode

In Ruby, method with yield can be much faster, but in Crystal, it's almost identical to it's counterparts. This benchmark has more context in Ruby, because block arguments were passed differently in different versions, incurring heap allocation and implementing lazy Proc allocation and conversion. This benchmark was ported to Crystal, just to see if there is any penalty (none!) and to see how it works in Crystal. Pay attention to how what you do with block arguments and methods.

def slow(&block)
  block.call
end

slow {1+1}
# In Crystal returns
nil

# In Ruby returns
2

Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

   block.call 531.71M (  1.88ns) (± 5.89%)  0 B/op   1.00× slower
block + yield 518.12M (  1.93ns) (± 7.24%)  0 B/op   1.03× slower
 unused block 533.64M (  1.87ns) (± 5.46%)  0 B/op        fastest
        yield 525.73M (   1.9ns) (± 6.76%)  0 B/op   1.02× slower

String

String#dup vs String#+ code

In Crystal, dup is much more effecient at creating shallow copies. No reason to use "string" + "" trick anymore ( if you did before)

crystal code/string/dup-vs-unary-plus.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

 String#+@ 471.96M (  2.12ns) (± 9.13%)  0 B/op   1.07× slower
String#dup 502.68M (  1.99ns) (± 5.32%)  0 B/op        fastest
String#compare vs String#downcase + == code
crystal code/string/compare-vs-downcase-==.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

String#downcase + ==  26.74M (  37.4ns) (± 9.13%)  32 B/op        fastest
      String#compare  21.83M (  45.8ns) (± 3.37%)   0 B/op   1.22× slower
String Concatenation code

In Crystal there is no <<, String#concat or String#append. You really only have 2 choices and one is almost 20x faster than the other. So it's not even really a choice: use interpolation whenever possible.

Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

      String#+  27.71M ( 36.09ns) (± 4.35%)  32 B/op  19.28× slower
{"foo"}{"bar"} 534.16M (  1.87ns) (± 4.21%)   0 B/op        fastest
String#match vs String.=~ vs String#starts_with?/String#ends_with? code (start) code (end)

If you have limited number of possibilities to match against, prefer to use starts_with/ends_with whenever possible, because it's orders of magnitude faster.

 crystal code/string/end-string-checking-match-vs-end_with.cr  --release
\Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

        String#=~   2.93M (341.26ns) (± 5.67%)  32 B/op  133.53× slower
     String#match   3.07M (325.78ns) (± 2.29%)  32 B/op  127.47× slower
String#ends_with? 391.28M (  2.56ns) (± 3.56%)   0 B/op         fastest
Regexp#=== vs String#match vs String#=~ code

While there is speed advantage to using =~ in Ruby, there is no difference in Crystal in equivalent situations. Keep in mind that === returns true/false while =~ returns start position of the match and .match returns MatchData object which ok if you are using to check for falsy/thruthy type values. Also keep in mind that these expressions are not 100% equivalent. ie.:

"boo".match(/boo/)`  #=> truthy
/boo/ === "boo"      #=> true
"book".match(/boo/)  #=> truthy
/book/ === "boo"     #=> false!!! So be very careful!
crystal code/string/===-vs-=\~-vs-match.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

   String#=~  13.76M ( 72.66ns) (± 1.00%)  16 B/op   1.00× slower
  Regexp#===  13.81M (  72.4ns) (± 2.75%)  16 B/op        fastest
String#match  13.77M ( 72.63ns) (± 2.24%)  16 B/op   1.00× slower
String#gsub vs String#sub code
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

 String#sub   2.21M (451.77ns) (± 4.22%)  1249 B/op        fastest
String#gsub   1.15M (869.95ns) (± 2.92%)  1249 B/op   1.93× slower
String#gsub vs String#tr code

In Ruby .tr is a few times faster, but in Crystal you are slightly faster using gsub.

Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

String#gsub  24.88M ( 40.19ns) (± 2.33%)  32 B/op        fastest
  String#tr  22.68M ( 44.09ns) (± 2.88%)  32 B/op   1.10× slower
String#sub! vs String#gsub! vs String#[]= code

Whenever possible sub will provide fastest performance. (Keep in mind unlike Ruby, Crystal has IMMUTABLE strings, so string[2] = "Ruby", doesn't work!)

crystal code/string/sub-vs-gsub_with_regex.cr   --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

 String#sub(string)   2.66M (375.53ns) (± 6.95%)  384 B/op        fastest
String#gsub(string)   1.22M (817.47ns) (± 5.45%)  384 B/op   2.18× slower
 String#sub/regexp/    1.9M (526.88ns) (± 8.28%)  432 B/op   1.40× slower
String#gsub/regexp/   1.13M (884.29ns) (± 6.07%)  449 B/op   2.35× slower
String#sub vs String#lchop code

Keep in mind that String#delete is faster than Sting#sub, but it's greedy! ( Thus remove from this test)

crystal code/string/sub-vs-lchop.cr   --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
String#lchop   29.8M ( 33.56ns) (± 9.20%)   32 B/op        fastest
  String#sub    3.2M (312.77ns) (± 5.80%)  176 B/op   9.32× slower
String#sub vs String#chomp code
crystal code/string/sub-vs-chomp.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

  String#sub   3.16M ( 316.3ns) (± 2.58%)  176 B/op   8.60× slower
String#chomp   27.2M ( 36.76ns) (± 5.01%)   32 B/op        fastest
Remove extra spaces (or other contiguous characters) code

The code is tested against contiguous spaces but should work for other chars too.

crystal code/string/remove-extra-spaces-or-other-chars.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

String#gsub/regex+/  87.68k ( 11.41µs) (± 1.44%)  2976 B/op   1.55× slower
     String#squeeze  135.8k (  7.36µs) (± 1.26%)   608 B/op        fastest

Hash

Hash#[] vs Hash#fetch vs Hash#dig code

Using symbols as keys is generally faster, and using .dig while slightly slower, is definitely more convenient (then fetch) and safer(!), especially on a deeply nested hash. Using Union type for keys Hash(String | Symbol, String slows down access on hash. Use Hash#dig?(:a, :b, :c) if you are not sure if :b or :c are "diggable"

crystal code/hash/bracket-vs-fetch-vs-dig.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
   Hash#[], symbol  63.59M ( 15.73ns) (± 4.75%)  0 B/op   1.10× slower
Hash#fetch, symbol  69.67M ( 14.35ns) (± 2.91%)  0 B/op        fastest
  Hash#dig, symbol  66.98M ( 14.93ns) (± 2.39%)  0 B/op   1.04× slower
   Hash#[], string  56.32M ( 17.75ns) (± 2.92%)  0 B/op   1.24× slower
Hash#fetch, string  59.74M ( 16.74ns) (± 2.93%)  0 B/op   1.17× slower
  Hash#dig, string  56.82M (  17.6ns) (± 2.68%)  0 B/op   1.23× slower
Hash#dig vs Hash#[] vs Hash#fetch code

fetch seems to break on deeply nested hash. Using dig? is the preferred method as the one being safer

crystal code/hash/bracket-vs-fetch-vs-dig-nested-hash.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
 Hash#dig?  12.78M ( 78.23ns) (± 1.99%)  0 B/op   1.01× slower
  Hash#dig  12.72M ( 78.64ns) (± 2.79%)  0 B/op   1.01× slower
   Hash#[]  12.85M ( 77.81ns) (± 1.95%)  0 B/op        fastest
Hash#[] &&   4.54M (220.02ns) (± 2.08%)  0 B/op   2.83× slower
Hash#fetch with argument vs Hash#fetch + block code

There is no difference in Crystal, while in Ruby fetch with non-constant argument is 30% slower

crystal code/hash/fetch-vs-fetch-with-block.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
Hash#fetch + const  73.21M ( 13.66ns) (± 3.06%)  0 B/op        fastest
Hash#fetch + block  72.93M ( 13.71ns) (± 2.67%)  0 B/op   1.00× slower
  Hash#fetch + arg  72.17M ( 13.86ns) (± 3.11%)  0 B/op   1.01× slower
Hash#each_key instead of Hash#keys.each code
crystal code/hash/keys-each-vs-each_key.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Hash#keys.each   1.79M (559.32ns) (± 4.86%)  416 B/op   1.07× slower
 Hash#each_key   1.92M (521.93ns) (± 7.75%)  338 B/op        fastest

Hash#key? instead of Hash#keys.include? code

Hash#keys.include? allocates an array of keys and performs an O(n) search;
Hash#has_key? performs an O(1) hash lookup without allocating a new array.

 crystal code/hash/keys-includes-vs-has_key.cr   --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
Hash#keys.includes?   5.87k ( 170.3µs) (± 3.24%)  146272 B/op  6472.68× slower
      Hash#has_key?  38.01M ( 26.31ns) (± 4.81%)       0 B/op          fastest
Hash#value? instead of Hash#values.include? code

Hash#values.includes? allocates an array of values and performs an O(n) search;
Hash#has_value? performs an O(n) search without allocating a new array.

crystal code/hash/values-includes-vs-has_value.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
Hash#values.includes?   5.73k (174.43µs) (± 3.79%)  146272 B/op  90033.84× slower
      Hash#has_value? 516.16M (  1.94ns) (± 6.66%)       0 B/op           fastest
Hash#merge! vs Hash#[]= code
crystal code/hash/merge-bang-vs-brackets.cr    --release               Mon Feb 25 22:20:32 2019
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
Hash#merge!  33.69k ( 29.68µs) (± 4.23%)  26415 B/op   3.74× slower
   Hash#[]= 125.96k (  7.94µs) (± 3.21%)   5538 B/op        fastest
Hash#merge vs Hash#merge! code
crystal code/hash/merge-vs-merge-bang.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

 Hash#merge   2.38k (420.05µs) (± 3.01%)  304740 B/op  13.86× slower
Hash#merge!  32.99k ( 30.32µs) (± 1.83%)   26415 B/op        fastest
{}#merge!(Hash) vs Hash#merge({}) vs Hash#dup#merge!({}) code
crystal code/hash/merge-bang-vs-merge-vs-dup-merge-bang.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
{}#merge!(Hash) do end  24.35k ( 41.06µs) (± 6.10%)  35758 B/op        fastest
        Hash#merge({})  15.08k ( 66.32µs) (± 8.26%)  59754 B/op   1.61× slower
   Hash#dup#merge!({})  15.11k ( 66.17µs) (± 5.16%)  59763 B/op   1.61× slower
Hash#sort_by vs Hash#sort code

To sort hash by key. Keep in mind that if you need to do this, is a code smell of a wrong data structure. Even in Ruby, sort & sort_by make implicit calls .to_a first. But remember, in Crystal we have Tuples!, so consider using: Array(Tuple(Symbol, String)) instead? Having said that, this is probably not a bottleneck in your app :)

crystal code/hash/hash-key-sort_by-vs-sort.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
sort_by + to_h 156.46k (  6.39µs) (± 2.96%)  5159 B/op   1.19× slower
   sort + to_h  185.7k (  5.39µs) (± 4.60%)  4330 B/op        fastest

Enumerable

Enumerable#each + push vs Enumerable#map code
crystal code/enumerable/each-push-vs-map.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Array#each + push 849.93k (  1.18µs) (± 0.75%)  1712 B/op   2.65× slower
        Array#map   2.25M (443.66ns) (± 7.10%)   480 B/op        fastest
Enumerable#each vs loop vs while vs step vs upto vs times vs downto code

each is the fastest way to iterate over an array. But if another approach makes your code clearer - use it.

crystal code/enumerable/each-vs-loop-vs-while-vs-step-vs-upto-vs-downto-vs-times.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

  #step   6.07M (164.82ns) (± 2.22%)  0 B/op  65.18× slower
  #upto   8.95M ( 111.7ns) (± 1.91%)  0 B/op  44.17× slower
#downto   8.49M (117.72ns) (± 1.85%)  0 B/op  46.55× slower
 #times  12.92M ( 77.42ns) (± 2.37%)  0 B/op  30.62× slower
  #each 395.48M (  2.53ns) (± 4.25%)  0 B/op        fastest
  while   7.57M (132.03ns) (± 2.05%)  0 B/op  52.21× slower
   loop   6.03M (165.79ns) (± 2.12%)  0 B/op  65.56× slower
Enumerable#each_with_index vs while loop code
crystal code/enumerable/each_with_index-vs-while-loop.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

     While Loop   9.54M (104.83ns) (± 2.32%)  0 B/op  41.38× slower
each_with_index 394.74M (  2.53ns) (± 4.30%)  0 B/op        fastest
Enumerable#map. vs Enumerable#flat_map code
crystal code/enumerable/map-flatten-vs-flat_map.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Array#map.flatten  99.74k ( 10.03µs) (± 4.40%)  9646 B/op   1.30× slower
   Array#flat_map 129.47k (  7.72µs) (± 3.14%)  7366 B/op        fastest
Enumerable#reverse.each vs Enumerable#reverse_each code

Enumerable#reverse allocates an extra array. Enumerable#reverse_each yields each value without allocating an extra array. 

crystal code/enumerable/reverse-each-vs-reverse_each.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
Array#reverse.each   2.16M ( 463.7ns) (± 5.72%)  480 B/op  175.64× slower
Array#reverse_each 378.79M (  2.64ns) (± 5.22%)    0 B/op         fastes
Enumerable#sort_by.first vs Enumerable#min_by code

Enumerable#sort_by performs a sort of the enumerable and allocates a new array the size of the enumerable. Enumerable#min_by doesn't perform a sort or allocate an array the size of the enumerable. Similar comparisons hold for Enumerable#sort_by.last vs Enumerable#max_by, Enumerable#sort.first vs Enumerable#min, and Enumerable#sort.last vs Enumerable#max.

crystal code/enumerable/sort_by-first-vs-min_by.cr --release        Tue Feb 26 00:57:16 2019
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

         Enumerable#min_by 404.49M (  2.47ns) (± 3.90%)     0 B/op          fastest
Enumerable#sort_by...first 360.16k (  2.78µs) (± 4.20%)  1888 B/op  1123.08× slower
Enumerable#find vs Enumerable#select.first code

Keeping with Crystal's convention of having more or less 1 obvious way of doing stuff, for example there is no Enumerable#detect, like in Ruby.

crystal code/enumerable/select-first-vs-find.cr --release    1.3m  Tue Feb 26 00:47:29 2019
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Enumerable#select.first   6.79M (147.18ns) (± 3.54%)  48 B/op  66.37× slower
        Enumerable#find 450.97M (  2.22ns) (± 5.47%)   0 B/op        fastest
Enumerable#select.last vs Enumerable#reverse.detect code
crystal code/enumerable/select-last-vs-reverse-find.cr  --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Enumerable#reverse.find   2.05M (487.74ns) (± 3.14%)  480 B/op   1.44× slower
 Enumerable#select.last   2.95M (338.92ns) (± 3.59%)  145 B/op        fastest
Enumerable#sort vs Enumerable#sort_by code
crystal code/enumerable/sort-vs-sort_by.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

Enumerable#sort_by (Symbol#to_proc)  104.9k (  9.53µs) (± 2.24%)  3136 B/op   1.58× slower
                 Enumerable#sort_by 104.28k (  9.59µs) (± 2.73%)  3136 B/op   1.59× slower
                    Enumerable#sort 165.91k (  6.03µs) (± 2.02%)  1056 B/op        fastest
Enumerable#reduce Block vs Enumerable#reduce Proc code
crystal code/enumerable/reduce-proc-vs-block.cr --release
Crystal 0.27.2 (2019-02-05)

LLVM: 6.0.1
Default target: x86_64-apple-macosx

reduce to_proc 389.91M (  2.56ns) (± 5.00%)  0 B/op   1.03× slower
  reduce block  401.7M (  2.49ns) (± 4.65%)  0 B/op        fastest

License

CC-BY-SA

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Code License

CC0 1.0 Universal

To the extent possible under law, @konung has waived all copyright and related or neighboring rights to "fast-crystal".

This work belongs to the community.