Merge pull request #16 from ipfs/feat/buzhash

Implement buzhash

benchmark_test.go

@@ -0,0 +1,59 @@
+package chunk
+
+import (
+	"bytes"
+	"io"
+	"math/rand"
+	"testing"
+)
+
+type newSplitter func(io.Reader) Splitter
+
+type bencSpec struct {
+	size int
+	name string
+}
+
+var bSizes = []bencSpec{
+	{1 << 10, "1K"},
+	{1 << 20, "1M"},
+	{16 << 20, "16M"},
+	{100 << 20, "100M"},
+}
+
+func benchmarkChunker(b *testing.B, ns newSplitter) {
+	for _, s := range bSizes {
+		s := s
+		b.Run(s.name, func(b *testing.B) {
+			benchmarkChunkerSize(b, ns, s.size)
+		})
+	}
+}
+
+func benchmarkChunkerSize(b *testing.B, ns newSplitter, size int) {
+	rng := rand.New(rand.NewSource(1))
+	data := make([]byte, size)
+	rng.Read(data)
+
+	b.SetBytes(int64(size))
+	b.ReportAllocs()
+	b.ResetTimer()
+
+	var res uint64
+
+	for i := 0; i < b.N; i++ {
+		r := ns(bytes.NewReader(data))
+
+		for {
+			chunk, err := r.NextBytes()
+			if err != nil {
+				if err == io.EOF {
+					break
+				}
+				b.Fatal(err)
+			}
+			res = res + uint64(len(chunk))
+		}
+	}
+	Res = Res + res
+}

buzhash.go

@@ -0,0 +1,128 @@
+package chunk
+
+import (
+	"io"
+	"math/bits"
+
+	pool "github.com/libp2p/go-buffer-pool"
+)
+
+const (
+	buzMin  = 128 << 10
+	buzMax  = 512 << 10
+	buzMask = 1<<17 - 1
+)
+
+type Buzhash struct {
+	r   io.Reader
+	buf []byte
+	n   int
+
+	err error
+}
+
+func NewBuzhash(r io.Reader) *Buzhash {
+	return &Buzhash{
+		r:   r,
+		buf: pool.Get(buzMax),
+	}
+}
+
+func (b *Buzhash) Reader() io.Reader {
+	return b.r
+}
+
+func (b *Buzhash) NextBytes() ([]byte, error) {
+	if b.err != nil {
+		return nil, b.err
+	}
+
+	n, err := io.ReadFull(b.r, b.buf[b.n:])
+	if err != nil {
+		if err == io.ErrUnexpectedEOF || err == io.EOF {
+			if b.n+n < buzMin {
+				b.err = io.EOF
+				res := make([]byte, b.n+n)
+				copy(res, b.buf)
+
+				pool.Put(b.buf)
+				b.buf = nil
+				return res, nil
+			}
+		} else {
+			b.err = err
+			pool.Put(b.buf)
+			b.buf = nil
+			return nil, err
+		}
+	}
+
+	i := buzMin - 32
+
+	var state uint32 = 0
+
+	for ; i < buzMin; i++ {
+		state = bits.RotateLeft32(state, 1)
+		state = state ^ bytehash[b.buf[i]]
+	}
+
+	if b.n+n > len(b.buf) {
+		panic("this is impossible, but gives +9 to performance")
+	}
+
+	for ; state&buzMask != 0 && i < b.n+n; i++ {
+		state = bits.RotateLeft32(state, 1) ^ bytehash[b.buf[i-32]] ^ bytehash[b.buf[i]]
+	}
+
+	res := make([]byte, i)
+	copy(res, b.buf)
+
+	b.n = copy(b.buf, b.buf[i:b.n+n])
+
+	return res, nil
+}
+
+var bytehash = [256]uint32{
+	0x6236e7d5, 0x10279b0b, 0x72818182, 0xdc526514, 0x2fd41e3d, 0x777ef8c8,
+	0x83ee5285, 0x2c8f3637, 0x2f049c1a, 0x57df9791, 0x9207151f, 0x9b544818,
+	0x74eef658, 0x2028ca60, 0x0271d91a, 0x27ae587e, 0xecf9fa5f, 0x236e71cd,
+	0xf43a8a2e, 0xbb13380, 0x9e57912c, 0x89a26cdb, 0x9fcf3d71, 0xa86da6f1,
+	0x9c49f376, 0x346aecc7, 0xf094a9ee, 0xea99e9cb, 0xb01713c6, 0x88acffb,
+	0x2960a0fb, 0x344a626c, 0x7ff22a46, 0x6d7a1aa5, 0x6a714916, 0x41d454ca,
+	0x8325b830, 0xb65f563, 0x447fecca, 0xf9d0ea5e, 0xc1d9d3d4, 0xcb5ec574,
+	0x55aae902, 0x86edc0e7, 0xd3a9e33, 0xe70dc1e1, 0xe3c5f639, 0x9b43140a,
+	0xc6490ac5, 0x5e4030fb, 0x8e976dd5, 0xa87468ea, 0xf830ef6f, 0xcc1ed5a5,
+	0x611f4e78, 0xddd11905, 0xf2613904, 0x566c67b9, 0x905a5ccc, 0x7b37b3a4,
+	0x4b53898a, 0x6b8fd29d, 0xaad81575, 0x511be414, 0x3cfac1e7, 0x8029a179,
+	0xd40efeda, 0x7380e02, 0xdc9beffd, 0x2d049082, 0x99bc7831, 0xff5002a8,
+	0x21ce7646, 0x1cd049b, 0xf43994f, 0xc3c6c5a5, 0xbbda5f50, 0xec15ec7,
+	0x9adb19b6, 0xc1e80b9, 0xb9b52968, 0xae162419, 0x2542b405, 0x91a42e9d,
+	0x6be0f668, 0x6ed7a6b9, 0xbc2777b4, 0xe162ce56, 0x4266aad5, 0x60fdb704,
+	0x66f832a5, 0x9595f6ca, 0xfee83ced, 0x55228d99, 0x12bf0e28, 0x66896459,
+	0x789afda, 0x282baa8, 0x2367a343, 0x591491b0, 0x2ff1a4b1, 0x410739b6,
+	0x9b7055a0, 0x2e0eb229, 0x24fc8252, 0x3327d3df, 0xb0782669, 0x1c62e069,
+	0x7f503101, 0xf50593ae, 0xd9eb275d, 0xe00eb678, 0x5917ccde, 0x97b9660a,
+	0xdd06202d, 0xed229e22, 0xa9c735bf, 0xd6316fe6, 0x6fc72e4c, 0x206dfa2,
+	0xd6b15c5a, 0x69d87b49, 0x9c97745, 0x13445d61, 0x35a975aa, 0x859aa9b9,
+	0x65380013, 0xd1fb6391, 0xc29255fd, 0x784a3b91, 0xb9e74c26, 0x63ce4d40,
+	0xc07cbe9e, 0xe6e4529e, 0xfb3632f, 0x9438d9c9, 0x682f94a8, 0xf8fd4611,
+	0x257ec1ed, 0x475ce3d6, 0x60ee2db1, 0x2afab002, 0x2b9e4878, 0x86b340de,
+	0x1482fdca, 0xfe41b3bf, 0xd4a412b0, 0xe09db98c, 0xc1af5d53, 0x7e55e25f,
+	0xd3346b38, 0xb7a12cbd, 0x9c6827ba, 0x71f78bee, 0x8c3a0f52, 0x150491b0,
+	0xf26de912, 0x233e3a4e, 0xd309ebba, 0xa0a9e0ff, 0xca2b5921, 0xeeb9893c,
+	0x33829e88, 0x9870cc2a, 0x23c4b9d0, 0xeba32ea3, 0xbdac4d22, 0x3bc8c44c,
+	0x1e8d0397, 0xf9327735, 0x783b009f, 0xeb83742, 0x2621dc71, 0xed017d03,
+	0x5c760aa1, 0x5a69814b, 0x96e3047f, 0xa93c9cde, 0x615c86f5, 0xb4322aa5,
+	0x4225534d, 0xd2e2de3, 0xccfccc4b, 0xbac2a57, 0xf0a06d04, 0xbc78d737,
+	0xf2d1f766, 0xf5a7953c, 0xbcdfda85, 0x5213b7d5, 0xbce8a328, 0xd38f5f18,
+	0xdb094244, 0xfe571253, 0x317fa7ee, 0x4a324f43, 0x3ffc39d9, 0x51b3fa8e,
+	0x7a4bee9f, 0x78bbc682, 0x9f5c0350, 0x2fe286c, 0x245ab686, 0xed6bf7d7,
+	0xac4988a, 0x3fe010fa, 0xc65fe369, 0xa45749cb, 0x2b84e537, 0xde9ff363,
+	0x20540f9a, 0xaa8c9b34, 0x5bc476b3, 0x1d574bd7, 0x929100ad, 0x4721de4d,
+	0x27df1b05, 0x58b18546, 0xb7e76764, 0xdf904e58, 0x97af57a1, 0xbd4dc433,
+	0xa6256dfd, 0xf63998f3, 0xf1e05833, 0xe20acf26, 0xf57fd9d6, 0x90300b4d,
+	0x89df4290, 0x68d01cbc, 0xcf893ee3, 0xcc42a046, 0x778e181b, 0x67265c76,
+	0xe981a4c4, 0x82991da1, 0x708f7294, 0xe6e2ae62, 0xfc441870, 0x95e1b0b6,
+	0x445f825, 0x5a93b47f, 0x5e9cf4be, 0x84da71e7, 0x9d9582b0, 0x9bf835ef,
+	0x591f61e2, 0x43325985, 0x5d2de32e, 0x8d8fbf0f, 0x95b30f38, 0x7ad5b6e,
+	0x4e934edf, 0x3cd4990e, 0x9053e259, 0x5c41857d}

buzhash_test.go

@@ -0,0 +1,71 @@
+package chunk
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"testing"
+
+	util "github.com/ipfs/go-ipfs-util"
+)
+
+func TestBuzhashChunking(t *testing.T) {
+	data := make([]byte, 1024*1024*16)
+	util.NewTimeSeededRand().Read(data)
+
+	r := NewBuzhash(bytes.NewReader(data))
+
+	var chunks [][]byte
+
+	for {
+		chunk, err := r.NextBytes()
+		if err != nil {
+			if err == io.EOF {
+				break
+			}
+			t.Fatal(err)
+		}
+
+		chunks = append(chunks, chunk)
+	}
+
+	t.Logf("average block size: %d\n", len(data)/len(chunks))
+
+	unchunked := bytes.Join(chunks, nil)
+	if !bytes.Equal(unchunked, data) {
+		fmt.Printf("%d %d\n", len(unchunked), len(data))
+		//ioutil.WriteFile("./incorrect", unchunked, 0777)
+		//ioutil.WriteFile("./correct", data, 0777)
+		t.Fatal("data was chunked incorrectly")
+	}
+}
+
+func TestBuzhashChunkReuse(t *testing.T) {
+	newBuzhash := func(r io.Reader) Splitter {
+		return NewBuzhash(r)
+	}
+	testReuse(t, newBuzhash)
+}
+
+func BenchmarkBuzhash2(b *testing.B) {
+	benchmarkChunker(b, func(r io.Reader) Splitter {
+		return NewBuzhash(r)
+	})
+}
+
+func TestBuzhashBitsHashBias(t *testing.T) {
+	counts := make([]byte, 32)
+	for _, h := range bytehash {
+		for i := 0; i < 32; i++ {
+			if h&1 == 1 {
+				counts[i]++
+			}
+			h = h >> 1
+		}
+	}
+	for i, c := range counts {
+		if c != 128 {
+			t.Errorf("Bit balance in position %d broken, %d ones", i, c)
+		}
+	}
+}

gen/main.go

@@ -0,0 +1,33 @@
+// This file generates bytehash LUT
+package main
+
+import (
+	"fmt"
+	"math/rand"
+)
+
+const nRounds = 200
+
+func main() {
+	rnd := rand.New(rand.NewSource(0))
+
+	lut := make([]uint32, 256)
+	for i := 0; i < 256/2; i++ {
+		lut[i] = 1<<32 - 1
+	}
+
+	for r := 0; r < nRounds; r++ {
+		for b := uint32(0); b < 32; b++ {
+			mask := uint32(1) << b
+			nmask := ^mask
+			for i, j := range rnd.Perm(256) {
+				li := lut[i]
+				lj := lut[j]
+				lut[i] = li&nmask | (lj & mask)
+				lut[j] = lj&nmask | (li & mask)
+			}
+		}
+	}
+
+	fmt.Printf("%#v", lut)
+}

parse.go

@@ -14,8 +14,8 @@ var (
 )
 
 // FromString returns a Splitter depending on the given string:
-// it supports "default" (""), "size-{size}", "rabin", "rabin-{blocksize}" and
-// "rabin-{min}-{avg}-{max}".
+// it supports "default" (""), "size-{size}", "rabin", "rabin-{blocksize}",
+// "rabin-{min}-{avg}-{max}" and "buzhash".
 func FromString(r io.Reader, chunker string) (Splitter, error) {
 	switch {
 	case chunker == "" || chunker == "default":
@@ -34,6 +34,9 @@ func FromString(r io.Reader, chunker string) (Splitter, error) {
 	case strings.HasPrefix(chunker, "rabin"):
 		return parseRabinString(r, chunker)
 
+	case chunker == "buzhash":
+		return NewBuzhash(r), nil
+
 	default:
 		return nil, fmt.Errorf("unrecognized chunker option: %s", chunker)
 	}

rabin_test.go

@@ -39,8 +39,8 @@ func TestRabinChunking(t *testing.T) {
 	}
 }
 
-func chunkData(t *testing.T, data []byte) map[string]blocks.Block {
-	r := NewRabin(bytes.NewReader(data), 1024*256)
+func chunkData(t *testing.T, newC newSplitter, data []byte) map[string]blocks.Block {
+	r := newC(bytes.NewReader(data))
 
 	blkmap := make(map[string]blocks.Block)
 
@@ -60,12 +60,12 @@ func chunkData(t *testing.T, data []byte) map[string]blocks.Block {
 	return blkmap
 }
 
-func TestRabinChunkReuse(t *testing.T) {
+func testReuse(t *testing.T, cr newSplitter) {
 	data := make([]byte, 1024*1024*16)
 	util.NewTimeSeededRand().Read(data)
 
-	ch1 := chunkData(t, data[1000:])
-	ch2 := chunkData(t, data)
+	ch1 := chunkData(t, cr, data[1000:])
+	ch2 := chunkData(t, cr, data)
 
 	var extra int
 	for k := range ch2 {
@@ -76,35 +76,21 @@ func TestRabinChunkReuse(t *testing.T) {
 	}
 
 	if extra > 2 {
-		t.Log("too many spare chunks made")
+		t.Logf("too many spare chunks made: %d", extra)
 	}
 }
 
+func TestRabinChunkReuse(t *testing.T) {
+	newRabin := func(r io.Reader) Splitter {
+		return NewRabin(r, 256*1024)
+	}
+	testReuse(t, newRabin)
+}
+
 var Res uint64
 
 func BenchmarkRabin(b *testing.B) {
-	data := make([]byte, 16<<20)
-	util.NewTimeSeededRand().Read(data)
-
-	b.SetBytes(16 << 20)
-	b.ReportAllocs()
-	b.ResetTimer()
-
-	var res uint64
-
-	for i := 0; i < b.N; i++ {
-		r := NewRabin(bytes.NewReader(data), 1024*256)
-
-		for {
-			chunk, err := r.NextBytes()
-			if err != nil {
-				if err == io.EOF {
-					break
-				}
-				b.Fatal(err)
-			}
-			res = res + uint64(len(chunk))
-		}
-	}
-	Res = Res + res
+	benchmarkChunker(b, func(r io.Reader) Splitter {
+		return NewRabin(r, 256<<10)
+	})
 }