このページについて

go でよく話題に上がるsliceで事前に容量を渡す場合とそうでない場合で
どの程度パフォーマンスに差分があるか簡単に見れるように試してみました。

内容

サンプルコード

main.go

package slices

func AllocationSlice(from []string, to []string) {
	for _, item := range from {
		to = append(to, item)
	}
}

ベンチマークコード

main_test.go

package slices

import "testing"

func StrSlice(num int) []string {
	re := make([]string, num, num)
	for idx, _ := range re {
		re[idx] = "a"
	}
	return re
}

func BenchmarkAllocationSlice(b *testing.B) {
	type args struct {
		from []string
		to   []string
	}
	tests := []struct {
		name string
		args args
	}{
		{
			name: "pre_alloc_100",
			args: args{
				from: StrSlice(100),
				to:   make([]string, 0, 100),
			},
		},
		{
			name: "no_alloc_100",
			args: args{
				from: StrSlice(100),
				to:   []string{},
			},
		},
		{
			name: "pre_alloc_500",
			args: args{
				from: StrSlice(500),
				to:   make([]string, 0, 500),
			},
		},
		{
			name: "no_alloc_500",
			args: args{
				from: StrSlice(500),
				to:   []string{},
			},
		},
		{
			name: "pre_alloc_1000",
			args: args{
				from: StrSlice(1000),
				to:   make([]string, 0, 1000),
			},
		},
		{
			name: "no_alloc_1000",
			args: args{
				from: StrSlice(1000),
				to:   []string{},
			},
		},
		{
			name: "pre_alloc_3000",
			args: args{
				from: StrSlice(3000),
				to:   make([]string, 0, 3000),
			},
		},
		{
			name: "no_alloc_3000",
			args: args{
				from: StrSlice(3000),
				to:   []string{},
			},
		},
	}
	for _, tt := range tests {
		b.Run(tt.name, func(b *testing.B) {
			b.ResetTimer()
			for i := 0; i < b.N; i++ {
				AllocationSlice(tt.args.from, tt.args.to)
			}
		})
	}
}

結果

goos: darwin
goarch: amd64
BenchmarkAllocationSlice/pre_alloc_100-8                10442223               114 ns/op               0 B/op          0 allocs/op
BenchmarkAllocationSlice/no_alloc_100-8                   933322              1261 ns/op            4080 B/op          8 allocs/op
BenchmarkAllocationSlice/pre_alloc_500-8                 2157058               547 ns/op               0 B/op          0 allocs/op
BenchmarkAllocationSlice/no_alloc_500-8                   304494              4070 ns/op           16368 B/op         10 allocs/op
BenchmarkAllocationSlice/pre_alloc_1000-8                1000000              1076 ns/op               0 B/op          0 allocs/op
BenchmarkAllocationSlice/no_alloc_1000-8                  168874              7100 ns/op           32752 B/op         11 allocs/op
BenchmarkAllocationSlice/pre_alloc_3000-8                 362504              3205 ns/op               0 B/op          0 allocs/op
BenchmarkAllocationSlice/no_alloc_3000-8                   33088             35524 ns/op          178801 B/op         15 allocs/op
PASS
ok      github.com/smith-30/gopg/slices 13.624s

まとめ

場合によっては、速度に10倍以上差が出るので細かいところだけど気を使っていきたい