Go serialization speed test

wordbench_test.go

// Package wordbench measures serialization speed.
package wordbench

import (
	"encoding/binary"
	"fmt"
	"math/rand"
	"testing"
)

func randN64Bit(n int) []uint64 {
	r := rand.New(rand.NewSource(rand.Int63()))
	ints := make([]uint64, n)
	for i := range ints {
		ints[i] = r.Uint64()
	}
	return ints
}

var data = randN64Bit(1024 * 1024)
var buf8 = make([]uint8, len(data)*8)
var buf64 = make([]uint64, len(data))

// BenchmarkShuffle64Bits writes 10 words in arbitrary order.
func BenchmarkShuffle64Bit(b *testing.B) {
	for _, batchSize := range []int{10, 100, 1000, 10000} {
		b.Run(fmt.Sprintf("Batch%d", batchSize), func(b *testing.B) {

			b.Run("Words", func(b *testing.B) {
				for i := 0; i < b.N; i++ {
					src := data[:batchSize]
					dst := buf64[:batchSize]

					for len(src) >= 10 && len(dst) >= 10 {
						dst[0] = src[9]
						dst[1] = src[8]
						dst[2] = src[7]
						dst[3] = src[6]
						dst[4] = src[5]
						dst[5] = src[4]
						dst[6] = src[3]
						dst[7] = src[2]
						dst[8] = src[1]
						dst[9] = src[0]

						src = src[10:]
						dst = dst[10:]
					}
				}
			})

			b.Run("WordsAppend", func(b *testing.B) {
				for i := 0; i < b.N; i++ {
					src := data[:batchSize]
					dst := buf64[:0]

					for len(src) >= 10 {
						dst = append(dst,
							src[9],
							src[8],
							src[7],
							src[6],
							src[5],
							src[4],
							src[3],
							src[2],
							src[1],
							src[0])

						src = src[10:]
					}
				}
			})

			b.Run("Bytes", func(b *testing.B) {
				for i := 0; i < b.N; i++ {
					src := data[:batchSize]
					dst := buf8[:batchSize*8]

					for len(src) >= 10 && len(dst) >= 80 {
						binary.LittleEndian.PutUint64(dst[0*8:], src[9])
						binary.LittleEndian.PutUint64(dst[1*8:], src[8])
						binary.LittleEndian.PutUint64(dst[2*8:], src[7])
						binary.LittleEndian.PutUint64(dst[3*8:], src[6])
						binary.LittleEndian.PutUint64(dst[4*8:], src[5])
						binary.LittleEndian.PutUint64(dst[5*8:], src[4])
						binary.LittleEndian.PutUint64(dst[6*8:], src[3])
						binary.LittleEndian.PutUint64(dst[7*8:], src[2])
						binary.LittleEndian.PutUint64(dst[8*8:], src[1])
						binary.LittleEndian.PutUint64(dst[9*8:], src[0])

						src = src[10:]
						dst = dst[80:]
					}
				}
			})

			b.Run("BytesAppend", func(b *testing.B) {
				for i := 0; i < b.N; i++ {
					src := data[:batchSize]
					dst := buf8[:0]

					for len(src) >= 10 {
						dst = binary.LittleEndian.AppendUint64(dst, src[9])
						dst = binary.LittleEndian.AppendUint64(dst, src[8])
						dst = binary.LittleEndian.AppendUint64(dst, src[7])
						dst = binary.LittleEndian.AppendUint64(dst, src[6])
						dst = binary.LittleEndian.AppendUint64(dst, src[5])
						dst = binary.LittleEndian.AppendUint64(dst, src[4])
						dst = binary.LittleEndian.AppendUint64(dst, src[3])
						dst = binary.LittleEndian.AppendUint64(dst, src[2])
						dst = binary.LittleEndian.AppendUint64(dst, src[1])
						dst = binary.LittleEndian.AppendUint64(dst, src[0])

						src = src[10:]
					}
				}
			})
		})
	}
}

goos: darwin
goarch: arm64
pkg: github.com/ronanh/intcomp/wordbench
BenchmarkShuffle64Bit/Batch10/Words-8   	430637913	         2.559 ns/op
BenchmarkShuffle64Bit/Batch10/WordsAppend-8         	491895675	         2.440 ns/op
BenchmarkShuffle64Bit/Batch10/Bytes-8               	357849769	         3.290 ns/op
BenchmarkShuffle64Bit/Batch10/BytesAppend-8         	289657186	         4.096 ns/op
BenchmarkShuffle64Bit/Batch100/Words-8              	63806452	        18.76 ns/op
BenchmarkShuffle64Bit/Batch100/WordsAppend-8        	66537442	        17.99 ns/op
BenchmarkShuffle64Bit/Batch100/Bytes-8              	46508398	        26.29 ns/op
BenchmarkShuffle64Bit/Batch100/BytesAppend-8        	32336260	        36.99 ns/op
BenchmarkShuffle64Bit/Batch1000/Words-8             	 6497124	       184.9 ns/op
BenchmarkShuffle64Bit/Batch1000/WordsAppend-8       	 6805016	       176.1 ns/op
BenchmarkShuffle64Bit/Batch1000/Bytes-8             	 4687851	       259.6 ns/op
BenchmarkShuffle64Bit/Batch1000/BytesAppend-8       	 3372344	       355.8 ns/op
BenchmarkShuffle64Bit/Batch10000/Words-8            	  484731	      2475 ns/op
BenchmarkShuffle64Bit/Batch10000/WordsAppend-8      	  489068	      2452 ns/op
BenchmarkShuffle64Bit/Batch10000/Bytes-8            	  444733	      2697 ns/op
BenchmarkShuffle64Bit/Batch10000/BytesAppend-8      	  340567	      3522 ns/op