Skip to content

Different benchmarks for Go generics starting from 1.18

License

Notifications You must be signed in to change notification settings

GRbit/go-generics-benchmark

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

go-generics-benchmark

Different benchmarks for Go generics starting from 1.18

How to use

First of all, before Go 1.18 release you have to build Go 1.18 by yourself or use gotip.

$ go install golang.org/dl/gotip@latest
$ gotip download
$ gotip version
go version devel go1.18-a5c0b190 Sun Jan 30 02:29:51 2022 +0000 linux/amd64

While containts_test.go and fib_test.go have simple go benchmarks which could be run as:

# After installing gotip, you can use it instead of 'go' command
$ gotip test -bench=.

Benchmarks in the generated folder are harder to run. They are here to test compilation speed with generics and with native static-typed code.

To run it, you need to generate new files for every test with a new number of functions (nn constat in generate.go). In other cases go will save the cache of existing files and compile time will be decreased.

$ sed -i 's/nn *=.*/nn = 10004/' generate.go
$ gotip run generate.go
$ gotip clean --cache 
$ time gotip build generated/generics.go generated/dummy_main.go

real	0m0.520s
user	0m0.864s
sys	0m0.102s

You can test different cases by choosing different files in the following combinations:

$ gotip build generated/[generics|static].go generated/[dummy|calls]_main.go

The first file choice will determine which version of functions will be used: with static types or with generics. The second choice will determine will all the functions be called inside of the func main()

Results

Simple tests

BTW: I don't know why the second iteration of the same tests takes less time. BenchmarkGenericC1 is equal to BenchmarkGenericC2. If someone can explain it to me (in a more detailed way than "it's some kind of optimizations"), please make a PR.

$ gotip test -bench=.
goos: linux
goarch: amd64
pkg: tst
cpu: Intel(R) Core(TM) i5-8365U CPU @ 1.60GHz
BenchmarkReflectC1-8     	   20428	     62570 ns/op
BenchmarkGenericC1-8     	 3996031	       300.4 ns/op
BenchmarkNativeC1-8      	 3765502	       300.8 ns/op
BenchmarkReflectC2-8     	   18903	     58921 ns/op
BenchmarkGenericC2-8     	 2190943	       545.4 ns/op
BenchmarkNativeC2-8      	 2074528	       537.8 ns/op
BenchmarkInterfaceF1-8   	   13393	     88923 ns/op
BenchmarkGenericF1-8     	   41120	     30936 ns/op
BenchmarkNativeF1-8      	   37862	     31916 ns/op
BenchmarkInterfaceF2-8   	   13342	     92812 ns/op
BenchmarkGenericF2-8     	   37729	     32275 ns/op
BenchmarkNativeF2-8      	   39980	     32544 ns/op
PASS
ok  	tst	20.541s

Compile time tests

# Static
[0] $ time gotip build static.go dummy_main.go 
real	0m1.347s
user	0m3.566s
sys     0m0.192s
[0] $ time gotip build static.go calls_main.go 
real    0m5.614s
user    0m9.924s
sys     0m0.423s

# Generics
[0] $ time gotip build generics.go dummy_main.go 
real   0m0.499s
user   0m0.843s
sys    0m0.097s
[0] $ time gotip build generics.go calls_main.go 
real   0m5.419s
user   0m10.395s
sys    0m0.409s

As you can see, if you don't call generic functions it will not be compiled at all.

About

Different benchmarks for Go generics starting from 1.18

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published