Add possibility to use different implementations in cli

Also make API smaller and more consistent and fix typos in documentation.

README.md

@@ -13,16 +13,16 @@ make auto
 
 # Homomorphic hashing in golang
 
-Package **tz** containts pure-Go implementation of hashing function described by Tillich and Źemor in [1] .
+Package `tz` containts pure-Go implementation of hashing function described by Tillich and Źemor in [1] .
 
-There are existing implementations already (e.g. [2]), however it is written in C.
+There are existing implementations already (e.g. [2]), however they are written in C.
 
-Package **gf127** contains arithmetic in GF(2^127) with _x^127+x^63+1_ as reduction polynomial.
+Package `gf127` contains arithmetic in `GF(2^127)` with `x^127+x^63+1` as reduction polynomial.
 
 # Description
 
 It can be used instead of Merkle-tree for data-validation, because homomorphic hashes
-are concatable: hash sum of data can be calculated based on hashes of chunks.
+are concatenable: hash sum of data can be calculated based on hashes of chunks.
 
 The example of how it works can be seen in tests.
 

benchmark

@@ -6,8 +6,8 @@
 tmpfile=$1
 
 go build ./cmd/tzsum && \
-for c in tzC tz tzbits; do
+for impl in avx avx2 avx2inline; do
-	./tzsum -cpuprofile cpu.prof -name $tmpfile -hash $c
+	echo $impl implementation:
-	echo top | go tool pprof cpu.prof
+	/usr/bin/env time -f "time: %e seconds" ./tzsum -name $tmpfile -impl $impl
 	echo
 done

cmd/tzsum/main.go

@@ -3,6 +3,7 @@ package main
 import (
 	"flag"
 	"fmt"
+	"hash"
 	"io"
 	"log"
 	"os"
@@ -16,13 +17,13 @@ var (
 	cpuprofile = flag.String("cpuprofile", "", "write cpu profile to `file`")
 	memprofile = flag.String("memprofile", "", "write memory profile to `file`")
 	filename   = flag.String("name", "-", "file to use")
+	hashimpl   = flag.String("impl", "avx2inline", "implementation to use")
 )
 
 func main() {
 	var (
 		f   io.Reader
 		err error
-		h   = tz.New()
 	)
 
 	flag.Parse()
@@ -45,6 +46,18 @@ func main() {
 		f = os.Stdin
 	}
 
+	var h hash.Hash
+	switch *hashimpl {
+	case "avx":
+		h = tz.NewWith(tz.AVX)
+	case "avx2":
+		h = tz.NewWith(tz.AVX2)
+	case "avx2inline":
+		h = tz.NewWith(tz.AVX2Inline)
+	default:
+		h = tz.New()
+	}
+
 	if _, err := io.Copy(h, f); err != nil {
 		log.Fatal("error while reading file: ", err)
 	}

tz/avx.go

@@ -0,0 +1,80 @@
+// Copyright 2018 (c) NSPCC
+//
+// This file contains AVX implementation.
+package tz
+
+import (
+	"hash"
+	"math"
+
+	"github.com/nspcc-dev/tzhash/gf127"
+)
+
+type digest struct {
+	x [4]gf127.GF127
+}
+
+// type assertion
+var _ hash.Hash = (*digest)(nil)
+
+var (
+	minmax  = [2]gf127.GF127{{0, 0}, {math.MaxUint64, math.MaxUint64}}
+	x127x63 = gf127.GF127{1 << 63, 1 << 63}
+)
+
+func newAVX() *digest {
+	d := new(digest)
+	d.Reset()
+	return d
+}
+
+func (d *digest) Sum(in []byte) []byte {
+	// Make a copy of d so that caller can keep writing and summing.
+	d0 := *d
+	h := d0.checkSum()
+	return append(in, h[:]...)
+}
+
+func (d *digest) checkSum() [hashSize]byte {
+	return d.byteArray()
+}
+
+func (d *digest) byteArray() (b [hashSize]byte) {
+	copy(b[:], d.x[0].ByteArray())
+	copy(b[16:], d.x[1].ByteArray())
+	copy(b[32:], d.x[2].ByteArray())
+	copy(b[48:], d.x[3].ByteArray())
+	return
+}
+
+func (d *digest) Reset() {
+	d.x[0] = gf127.GF127{1, 0}
+	d.x[1] = gf127.GF127{0, 0}
+	d.x[2] = gf127.GF127{0, 0}
+	d.x[3] = gf127.GF127{1, 0}
+}
+
+func (d *digest) Write(data []byte) (n int, err error) {
+	n = len(data)
+	for _, b := range data {
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>7)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>6)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>5)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>4)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>3)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>2)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>1)&1])
+		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>0)&1])
+	}
+	return
+}
+
+func (d *digest) Size() int {
+	return hashSize
+}
+
+func (d *digest) BlockSize() int {
+	return hashBlockSize
+}
+
+func mulBitRight(c00, c01, c10, c11, e *gf127.GF127)

tz/avx2.go

@@ -1,3 +1,6 @@
+// Copyright 2019 (c) NSPCC
+//
+// This file contains AVX2 implementation.
 package tz
 
 import (
@@ -10,9 +13,10 @@ type digest2 struct {
 	x [2]gf127.GF127x2
 }
 
+// type assertion
 var _ hash.Hash = (*digest2)(nil)
 
-func NewAVX2() hash.Hash {
+func newAVX2() *digest2 {
 	d := new(digest2)
 	d.Reset()
 	return d

tz/avx2_inline.go

@@ -1,3 +1,7 @@
+// Copyright 2019 (c) NSPCC
+//
+// This file contains AVX2 implementation with inlined
+// assembly calls.
 package tz
 
 import (
@@ -10,9 +14,10 @@ type digest3 struct {
 	x [2]gf127.GF127x2
 }
 
+// type assertion
 var _ hash.Hash = (*digest3)(nil)
 
-func NewAVX2Inline() hash.Hash {
+func newAVX2Inline() *digest3 {
 	d := new(digest3)
 	d.Reset()
 	return d

tz/hash.go

@@ -6,97 +6,54 @@ package tz
 import (
 	"errors"
 	"hash"
-	"math"
-
-	"github.com/nspcc-dev/tzhash/gf127"
 )
 
+type Implementation int
+
 const (
 	hashSize      = 64
 	hashBlockSize = 128
+
+	_ Implementation = iota
+	AVX
+	AVX2
+	AVX2Inline
 )
 
-type digest struct {
+func (impl Implementation) String() string {
-	x [4]gf127.GF127
+	switch impl {
+	case AVX:
+		return "AVX"
+	case AVX2:
+		return "AVX2"
+	case AVX2Inline:
+		return "AVX2Inline"
+	default:
+		return "UNKNOWN"
+	}
 }
 
-// type assertion
+func NewWith(impl Implementation) hash.Hash {
-var _ hash.Hash = (*digest)(nil)
+	switch impl {
-
+	case AVX:
-var (
+		return newAVX()
-	minmax  = [2]gf127.GF127{{0, 0}, {math.MaxUint64, math.MaxUint64}}
+	case AVX2:
-	x127x63 = gf127.GF127{1 << 63, 1 << 63}
+		return newAVX2()
-)
+	case AVX2Inline:
+		return newAVX2Inline()
+	default:
+		return New()
+	}
+}
 
 // New returns a new hash.Hash computing the Tillich-Zémor checksum.
 func New() hash.Hash {
-	d := new(digest)
+	return newAVX2Inline()
-	d.Reset()
-	return d
-}
-
-func (d *digest) Sum(in []byte) []byte {
-	// Make a copy of d so that caller can keep writing and summing.
-	d0 := *d
-	h := d0.checkSum()
-	return append(in, h[:]...)
-}
-
-func (d *digest) checkSum() [hashSize]byte {
-	return d.byteArray()
-}
-
-func (d *digest) byteArray() (b [hashSize]byte) {
-	copy(b[:], d.x[0].ByteArray())
-	copy(b[16:], d.x[1].ByteArray())
-	copy(b[32:], d.x[2].ByteArray())
-	copy(b[48:], d.x[3].ByteArray())
-	return
-}
-
-func (d *digest) Reset() {
-	d.x[0] = gf127.GF127{1, 0}
-	d.x[1] = gf127.GF127{0, 0}
-	d.x[2] = gf127.GF127{0, 0}
-	d.x[3] = gf127.GF127{1, 0}
-}
-
-func (d *digest) Write(data []byte) (n int, err error) {
-	n = len(data)
-	for _, b := range data {
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>7)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>6)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>5)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>4)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>3)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>2)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>1)&1])
-		mulBitRight(&d.x[0], &d.x[1], &d.x[2], &d.x[3], &minmax[(b>>0)&1])
-	}
-	return
-}
-
-func (d *digest) Size() int {
-	return hashSize
-}
-
-func (d *digest) BlockSize() int {
-	return hashBlockSize
-}
-
-// Sum returnz Tillich-Zémor checksum of data.
-// It uses only AVX instructions (no AVX2).
-func SumAVX(data []byte) [hashSize]byte {
-	d := new(digest)
-	d.Reset()
-	_, _ = d.Write(data) // no errors
-	return d.checkSum()
 }
 
 // Sum returns Tillich-Zémor checksum of data.
 func Sum(data []byte) [hashSize]byte {
-	d := new(digest2)
+	d := newAVX2Inline()
-	d.Reset()
 	_, _ = d.Write(data) // no errors
 	return d.checkSum()
 }
@@ -178,5 +135,3 @@ func SubtractL(c, a []byte) (b []byte, err error) {
 
 	return p2.MarshalBinary()
 }
-
-func mulBitRight(c00, c01, c10, c11, e *gf127.GF127)

tz/hash_test.go

@@ -11,6 +11,23 @@ import (
 
 const benchDataSize = 100000
 
+var providers = []Implementation{
+	AVX,
+	AVX2,
+	AVX2Inline,
+}
+
+func TestNewWith(t *testing.T) {
+	d := NewWith(AVX)
+	require.IsType(t, (*digest)(nil), d)
+
+	d = NewWith(AVX2)
+	require.IsType(t, (*digest2)(nil), d)
+
+	d = NewWith(AVX2Inline)
+	require.IsType(t, (*digest3)(nil), d)
+}
+
 var testCases = []struct {
 	input []byte
 	hash  string
@@ -30,38 +47,18 @@ var testCases = []struct {
 }
 
 func TestHash(t *testing.T) {
-	t.Run("test AVX digest", func(t *testing.T) {
+	for i := range providers {
-		d := new(digest)
+		p := providers[i]
+		t.Run("test "+p.String()+" digest", func(t *testing.T) {
+			d := NewWith(p)
 			for _, tc := range testCases {
 				d.Reset()
 				_, _ = d.Write(tc.input)
-			sum := d.checkSum()
+				sum := d.Sum(nil)
-
 				require.Equal(t, tc.hash, hex.EncodeToString(sum[:]))
 			}
 		})
-
-	t.Run("test AVX2 digest", func(t *testing.T) {
-		d := new(digest2)
-		for _, tc := range testCases {
-			d.Reset()
-			_, _ = d.Write(tc.input)
-			sum := d.checkSum()
-
-			require.Equal(t, tc.hash, hex.EncodeToString(sum[:]))
 	}
-	})
-
-	t.Run("test AVX2 digest with inline asm function", func(t *testing.T) {
-		d := new(digest3)
-		for _, tc := range testCases {
-			d.Reset()
-			_, _ = d.Write(tc.input)
-			sum := d.checkSum()
-
-			require.Equal(t, tc.hash, hex.EncodeToString(sum[:]))
-		}
-	})
 }
 
 func newBuffer() (data []byte) {
@@ -75,46 +72,24 @@ func newBuffer() (data []byte) {
 	return
 }
 
-func BenchmarkAVX(b *testing.B) {
+func BenchmarkSum(b *testing.B) {
 	data := newBuffer()
+	size := int64(len(data))
 
+	for i := range providers {
+		p := providers[i]
+		b.Run("bench"+p.String()+"digest", func(b *testing.B) {
 			b.ResetTimer()
 			b.ReportAllocs()
-	d := new(digest)
+			d := NewWith(p)
 			for i := 0; i < b.N; i++ {
 				d.Reset()
 				_, _ = d.Write(data)
-		d.checkSum()
+				d.Sum(nil)
 			}
-	b.SetBytes(int64(len(data)))
+			b.SetBytes(size)
+		})
 	}
-
-func BenchmarkAVX2(b *testing.B) {
-	data := newBuffer()
-
-	b.ResetTimer()
-	b.ReportAllocs()
-	d := new(digest2)
-	for i := 0; i < b.N; i++ {
-		d.Reset()
-		_, _ = d.Write(data)
-		d.checkSum()
-	}
-	b.SetBytes(int64(len(data)))
-}
-
-func BenchmarkAVX2Inline(b *testing.B) {
-	data := newBuffer()
-
-	b.ResetTimer()
-	b.ReportAllocs()
-	d := new(digest3)
-	for i := 0; i < b.N; i++ {
-		d.Reset()
-		_, _ = d.Write(data)
-		d.checkSum()
-	}
-	b.SetBytes(int64(len(data)))
 }
 
 func TestHomomorphism(t *testing.T) {
@@ -189,14 +164,14 @@ func TestConcat(t *testing.T) {
 
 func TestValidate(t *testing.T) {
 	var (
-		hash []byte
+		h   []byte
 		ps  [][]byte
 		got bool
 		err error
 	)
 
 	for _, tc := range testCasesConcat {
-		hash, _ = hex.DecodeString(tc.Hash)
+		h, _ = hex.DecodeString(tc.Hash)
 		require.NoError(t, err)
 
 		ps = make([][]byte, len(tc.Parts))
@@ -205,7 +180,7 @@ func TestValidate(t *testing.T) {
 			require.NoError(t, err)
 		}
 
-		got, err = Validate(hash, ps)
+		got, err = Validate(h, ps)
 		require.NoError(t, err)
 		require.True(t, got)
 	}