slice_test.go

package main

import (
        "testing"
)

var data = [64]int{}
var data2 = [64]int{}

func BenchmarkOffset2(b *testing.B) {
        pos := 32
        s := data[:]
        for i := 0; i < b.N; i++ {
        k := pos+32
                for i := 0; i < k; i++ {
                        s[i] = i
                }
        }
}
func BenchmarkOffset(b *testing.B) {
        pos := 32
        s := data[:]
        for i := 0; i < b.N; i++ {
                for i := 0; i < pos+32; i++ {
                        s[i] = i
                }
        }
}

func BenchmarkRange(b *testing.B) {
        pos := 32
        s := data[:]
        for i := 0; i < b.N; i++ {
                a := s[pos : pos+32]
                for i, _ := range a {
                        a[i] = i
                }
        }
}

func BenchmarkUnrolled(b *testing.B) {
        pos := 32
        s := data[:]
        for i := 0; i < b.N; i++ {
        s[pos+0] = 0
        s[pos+1] = 1
        s[pos+2] = 2
        s[pos+3] = 3
        s[pos+4] = 4
        s[pos+5] = 5
        s[pos+6] = 6
        s[pos+7] = 7
        s[pos+8] = 8
        s[pos+9] = 9
        s[pos+10] = 10
        s[pos+11] = 11
        s[pos+12] = 12
        s[pos+13] = 13
        s[pos+14] = 14
        s[pos+15] = 15
        s[pos+16] = 16
        s[pos+17] = 17
        s[pos+18] = 18
        s[pos+19] = 19
        s[pos+20] = 20
        s[pos+21] = 21
        s[pos+22] = 22
        s[pos+23] = 23
        s[pos+24] = 24
        s[pos+25] = 25
        s[pos+26] = 26
        s[pos+27] = 27
        s[pos+28] = 28
        s[pos+29] = 29
        s[pos+30] = 30
        s[pos+31] = 31
        }
}

func BenchmarkCopy(b *testing.B) {
    pos := 32
    s := data[:]
    for i := 0; i < b.N; i++ {
        copy(data2[pos:pos+32], s[pos:pos+32])
    }
}

func BenchmarkRangeCopy(b *testing.B) {
    pos := 32
    s := data[:]
    for i := 0; i < b.N; i++ {
        a := s[pos:pos+32]
        for i, _ := range a {
            a[i] = data2[i]
        }
    }
}

func BenchmarkCopyUnrolled(b *testing.B) {
    pos := 32
    s := data[:]
    for i := 0; i < b.N; i++ {
        s[pos+0] = data2[pos+0]
        s[pos+1] = data2[pos+1]
        s[pos+2] = data2[pos+2]
        s[pos+3] = data2[pos+3]
        s[pos+4] = data2[pos+4]
        s[pos+5] = data2[pos+5]
        s[pos+6] = data2[pos+6]
        s[pos+7] = data2[pos+7]
        s[pos+8] = data2[pos+8]
        s[pos+9] = data2[pos+9]
        s[pos+10] = data2[pos+10]
        s[pos+11] = data2[pos+11]
        s[pos+12] = data2[pos+12]
        s[pos+13] = data2[pos+13]
        s[pos+14] = data2[pos+14]
        s[pos+15] = data2[pos+15]
        s[pos+16] = data2[pos+16]
        s[pos+17] = data2[pos+17]
        s[pos+18] = data2[pos+18]
        s[pos+19] = data2[pos+19]
        s[pos+20] = data2[pos+20]
        s[pos+21] = data2[pos+21]
        s[pos+22] = data2[pos+22]
        s[pos+23] = data2[pos+23]
        s[pos+24] = data2[pos+24]
        s[pos+25] = data2[pos+25]
        s[pos+26] = data2[pos+26]
        s[pos+27] = data2[pos+27]
        s[pos+28] = data2[pos+28]
        s[pos+29] = data2[pos+29]
        s[pos+30] = data2[pos+30]
        s[pos+31] = data2[pos+31]
    }
}

Testing Different Ways of Setting and Copying a Slice

--- Output of using **gccgo -O3** ---
BenchmarkOffset2        50000000                27.6 ns/op
BenchmarkOffset 50000000                32.0 ns/op
BenchmarkRange  500000000                4.46 ns/op
BenchmarkUnrolled       200000000                8.92 ns/op
BenchmarkCopy   500000000                7.52 ns/op
BenchmarkRangeCopy      100000000               13.9 ns/op
BenchmarkCopyUnrolled   200000000                8.91 ns/op

--- Output of using standard go compiler ---
BenchmarkOffset2        20000000               137 ns/op
BenchmarkOffset 20000000               137 ns/op
BenchmarkRange  50000000                31.3 ns/op
BenchmarkUnrolled       100000000               13.9 ns/op
BenchmarkCopy   100000000               10.5 ns/op
BenchmarkRangeCopy      50000000                44.6 ns/op
BenchmarkCopyUnrolled   50000000                30.4 ns/op