benchmark passing structs vs pointers vs mixed (first level struct, second level pointer)

interval_test.go

package smap

import (
        "bytes"
        "testing"
)

type skey string

func (s skey) Cmp(other Key) int {
        return bytes.Compare([]byte(string(s)), []byte(string(other.(skey))))
}

type PInterval struct {
        From, To *Edge
}

func valueReceiver(i Interval) Interval {
        i.From.Key.Cmp(i.To.Key)
        return i
}
func mixedReceiver(i PInterval) PInterval {
        i.From.Key.Cmp(i.To.Key)
        return i
}
func pointerReceiver(i *PInterval) *PInterval {
        i.From.Key.Cmp(i.To.Key)
        return i
}

func BenchmarkByValue(b *testing.B) {
        I := Interval{Edge{skey("hello"), false}, Edge{skey("world"), false}}
        var got Interval
        for i := 0; i < b.N; i++ {
                got = valueReceiver(I)
        }
        var _ = got
}

func BenchmarkPointer(b *testing.B) {
        I := &PInterval{&Edge{skey("hello"), false}, &Edge{skey("world"), false}}
        var got *PInterval
        for i := 0; i < b.N; i++ {
                got = pointerReceiver(I)
        }
        var _ = got
}
func BenchmarkMixed(b *testing.B) {
        I := PInterval{&Edge{skey("hello"), false}, &Edge{skey("world"), false}}
        var got PInterval
        for i := 0; i < b.N; i++ {
                got = mixedReceiver(I)
        }
        var _ = got
}

benchmarks show little difference when there's some work to be done on the arguments. in this case it's a simple Cmp. Usually the work per Interval argument is >> 1
$ go test -bench . -benchmem
testing: warning: no tests to run
PASS
BenchmarkByValue 10000000 187 ns/op 16 B/op 2 allocs/op
BenchmarkPointer 10000000 154 ns/op 16 B/op 2 allocs/op
BenchmarkMixed 10000000 158 ns/op 16 B/op 2 allocs/op