Skip to content

aligned (and faster) uniq impls + less wrapping#5830

Merged
kares merged 7 commits intojruby:masterfrom
kares:fast-uniq
Aug 14, 2019
Merged

aligned (and faster) uniq impls + less wrapping#5830
kares merged 7 commits intojruby:masterfrom
kares:fast-uniq

Conversation

@kares
Copy link
Member

@kares kares commented Aug 13, 2019

noticed a lot of unnecessary single element array wrapping at Enumerable's impls
... this could be aided with a simple override on the (native) BlockCallback

the most interesting piece is doing less work on Array#uniq (and also Enumerable#uniq)
also the Enumerable is still faster to have on the native end, numbers :

MRI 2.5.7

kares@clevo:~/workspace/oss/jruby$ ruby bench/core/array/unique_bench.rb
Rehearsal --------------------------------------------------------------------------------
Array#uniq (size: 8) [5000000x]                2.207314   0.000000   2.207314 (  2.207925)
Array#uniq (size: 32) [5000000x]               5.405314   0.000000   5.405314 (  5.406322)
Array#uniq(&block) (size: 8) [5000000x]        3.594613   0.000000   3.594613 (  3.595398)
Enumerable#uniq (size: 8) [5000000x]           4.275704   0.000000   4.275704 (  4.276550)
Enumerable#uniq (size: 32) [5000000x]          9.692536   0.000000   9.692536 (  9.693569)
Enumerable#uniq(&block) (size: 8) [5000000x]   5.502709   0.000000   5.502709 (  5.503421)
---------------------------------------------------------------------- total: 30.678190sec

                                                   user     system      total        real
Array#uniq (size: 8) [5000000x]                2.138970   0.000000   2.138970 (  2.139309)
Array#uniq (size: 32) [5000000x]               5.358952   0.000000   5.358952 (  5.359636)
Array#uniq(&block) (size: 8) [5000000x]        3.634595   0.000000   3.634595 (  3.635386)
Enumerable#uniq (size: 8) [5000000x]           4.194265   0.000000   4.194265 (  4.195012)
Enumerable#uniq (size: 32) [5000000x]          9.900353   0.000000   9.900353 (  9.902616)
Enumerable#uniq(&block) (size: 8) [5000000x]   5.433238   0.000000   5.433238 (  5.434582)

current JRuby (9.2.8.0)

Rehearsal --------------------------------------------------------------------------------
Array#uniq (size: 8) [5000000x]                1.980000   0.140000   2.120000 (  1.537581)
Array#uniq (size: 32) [5000000x]               4.430000   0.130000   4.560000 (  4.355980)
Array#uniq(&block) (size: 8) [5000000x]        3.390000   0.030000   3.420000 (  2.969637)
Enumerable#uniq (size: 8) [5000000x]           5.790000   0.020000   5.810000 (  5.115501)
Enumerable#uniq (size: 32) [5000000x]         15.500000   0.030000  15.530000 ( 15.426172)
Enumerable#uniq(&block) (size: 8) [5000000x]   7.700000   0.010000   7.710000 (  7.280119)
---------------------------------------------------------------------- total: 39.150000sec

                                                   user     system      total        real
Array#uniq (size: 8) [5000000x]                1.360000   0.000000   1.360000 (  1.331043)
Array#uniq (size: 32) [5000000x]               4.130000   0.000000   4.130000 (  4.100323)
Array#uniq(&block) (size: 8) [5000000x]        2.920000   0.010000   2.930000 (  2.867646)
Enumerable#uniq (size: 8) [5000000x]           5.080000   0.010000   5.090000 (  5.031277)
Enumerable#uniq (size: 32) [5000000x]         15.420000   0.040000  15.460000 ( 15.222436)
Enumerable#uniq(&block) (size: 8) [5000000x]   7.330000   0.030000   7.360000 (  7.235881)

this PR

Rehearsal --------------------------------------------------------------------------------
Array#uniq (size: 8) [5000000x]                1.940000   0.130000   2.070000 (  1.376955)
Array#uniq (size: 32) [5000000x]               3.470000   0.090000   3.560000 (  3.473800)
Array#uniq(&block) (size: 8) [5000000x]        2.950000   0.060000   3.010000 (  2.650662)
Enumerable#uniq (size: 8) [5000000x]           2.210000   0.020000   2.230000 (  1.840881)
Enumerable#uniq (size: 32) [5000000x]          5.040000   0.020000   5.060000 (  4.998871)
Enumerable#uniq(&block) (size: 8) [5000000x]   3.990000   0.010000   4.000000 (  3.533291)
---------------------------------------------------------------------- total: 19.930000sec

                                                   user     system      total        real
Array#uniq (size: 8) [5000000x]                1.160000   0.020000   1.180000 (  1.161547)
Array#uniq (size: 32) [5000000x]               3.330000   0.010000   3.340000 (  3.314024)
Array#uniq(&block) (size: 8) [5000000x]        2.600000   0.000000   2.600000 (  2.587071)
Enumerable#uniq (size: 8) [5000000x]           1.910000   0.000000   1.910000 (  1.844716)
Enumerable#uniq (size: 32) [5000000x]          5.090000   0.010000   5.100000 (  5.077093)
Enumerable#uniq(&block) (size: 8) [5000000x]   3.380000   0.000000   3.380000 (  3.301146)

kares added 6 commits August 13, 2019 13:56
@kares
[refactor] faster (native) Enumerable#uniq impl
6282d17 
re-using RubyHash internals (wout double hashing) seems nice
@kares
adust RubyHash put (needs to be overriden in Map proxy)
a4c8f02
@kares
[refactor] Array#uniq to do less work (writes)
0ed427d 
we do not need to fill a hash and than go on deleting entries
rely on Hash ordering + make Hash always map to ary's values
@kares
[refactor] use the internal blank array factory
a0703f0
@kares
[refactor] avoid args pack + remove duplicate method
eb6d7f8
@kares
a simple benchmark for Array/Enumerable#uniq
4030727 
... confirming previous work made sense (at least for now):

```
----------------------------------------------------------------------
total: 30.678190sec

user     system
total        real
Array#uniq (size: 8) [5000000x]                2.138970   0.000000
2.138970 (  2.139309)
Array#uniq (size: 32) [5000000x]               5.358952   0.000000
5.358952 (  5.359636)
Array#uniq(&block) (size: 8) [5000000x]        3.634595   0.000000
3.634595 (  3.635386)
Enumerable#uniq (size: 8) [5000000x]           4.194265   0.000000
4.194265 (  4.195012)
Enumerable#uniq (size: 32) [5000000x]          9.900353   0.000000
9.900353 (  9.902616)
Enumerable#uniq(&block) (size: 8) [5000000x]   5.433238   0.000000
5.433238 (  5.434582)
```

```
----------------------------------------------------------------------
total: 39.150000sec

user     system
total        real
Array#uniq (size: 8) [5000000x]                1.360000   0.000000
1.360000 (  1.331043)
Array#uniq (size: 32) [5000000x]               4.130000   0.000000
4.130000 (  4.100323)
Array#uniq(&block) (size: 8) [5000000x]        2.920000   0.010000
2.930000 (  2.867646)
Enumerable#uniq (size: 8) [5000000x]           5.080000   0.010000
5.090000 (  5.031277)
Enumerable#uniq (size: 32) [5000000x]         15.420000   0.040000
15.460000 ( 15.222436)
Enumerable#uniq(&block) (size: 8) [5000000x]   7.330000   0.030000
7.360000 (  7.235881)
```

```
----------------------------------------------------------------------
total: 19.930000sec

user     system
total        real
Array#uniq (size: 8) [5000000x]                1.160000   0.020000
1.180000 (  1.161547)
Array#uniq (size: 32) [5000000x]               3.330000   0.010000
3.340000 (  3.314024)
Array#uniq(&block) (size: 8) [5000000x]        2.600000   0.000000
2.600000 (  2.587071)
Enumerable#uniq (size: 8) [5000000x]           1.910000   0.000000
1.910000 (  1.844716)
Enumerable#uniq (size: 32) [5000000x]          5.090000   0.010000
5.100000 (  5.077093)
Enumerable#uniq(&block) (size: 8) [5000000x]   3.380000   0.000000
3.380000 (  3.301146)
```
@kares kares added this to the JRuby 9.2.9.0 milestone Aug 13, 2019
@kares kares merged commit 9fb38a7 into jruby:master Aug 14, 2019
@kares kares deleted the fast-uniq branch October 31, 2019 08:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

JRuby 9.2.9.0

Development

Successfully merging this pull request may close these issues.

1 participant

@kares