bpgg's Introduction

Benchmarking the performance of Go generics

This repository benchmarks the performance of Go generics.

Benchmarks

Boxing

This benchmark demonstrates how Go generics eliminates the need for the boxing that occurs with the use of interface{} by implementing a simple vector:

$ go test -bench . -benchmem -count 5 -memprofile ./boxing/memprofile.out -v ./boxing/
goos: darwin
goarch: arm64
pkg: bpgg/boxing
BenchmarkInterfaceVector
BenchmarkInterfaceVector-8   	18319783	        62.84 ns/op	      96 B/op	       1 allocs/op
BenchmarkInterfaceVector-8   	26550849	        65.01 ns/op	     103 B/op	       1 allocs/op
BenchmarkInterfaceVector-8   	24785962	        58.35 ns/op	      89 B/op	       1 allocs/op
BenchmarkInterfaceVector-8   	23335028	        60.32 ns/op	      94 B/op	       1 allocs/op
BenchmarkInterfaceVector-8   	24852051	        49.56 ns/op	      89 B/op	       1 allocs/op
BenchmarkGenericVector
BenchmarkGenericVector-8     	74692666	        16.99 ns/op	      49 B/op	       0 allocs/op
BenchmarkGenericVector-8     	75874735	        15.93 ns/op	      48 B/op	       0 allocs/op
BenchmarkGenericVector-8     	74793732	        15.93 ns/op	      49 B/op	       0 allocs/op
BenchmarkGenericVector-8     	75018556	        15.83 ns/op	      48 B/op	       0 allocs/op
BenchmarkGenericVector-8     	73958611	        15.85 ns/op	      49 B/op	       0 allocs/op
PASS
ok  	bpgg/boxing	13.495s

A few, key takeaways:

On average the implementation of GVector[T any] was more performant with respect to CPU and memory by 300% and 200%, respectively.
Because no boxing was needed to store an int in GVector[int], there was no need to allocate anything on the heap in order to box the ...

Recommend Projects