compress/bzip2: faster decoding.

benchmark                old ns/op    new ns/op    delta
BenchmarkDecodeDigits     19451173     14347829  -26.24%
BenchmarkDecodeTwain      57516800     42619978  -25.90%

benchmark                 old MB/s     new MB/s  speedup
BenchmarkDecodeDigits         2.22         3.01    1.36x
BenchmarkDecodeTwain          2.17         2.93    1.35x

R=golang-dev, dave, bradfitz, agl
CC=golang-dev
https://golang.org/cl/9915043

src/pkg/compress/bzip2/bit_reader.go

@@ -77,6 +77,14 @@ func (br *bitReader) ReadBit() bool {
 	return n != 0
 }
 
+func (br *bitReader) TryReadBit() (bit byte, ok bool) {
+	if br.bits > 0 {
+		br.bits--
+		return byte(br.n>>br.bits) & 1, true
+	}
+	return 0, false
+}
+
 func (br *bitReader) Err() error {
 	return br.err
 }

src/pkg/compress/bzip2/bzip2_test.go

@@ -155,3 +155,32 @@ const rand2Hex = "92d5652616ac444a4a04af1a8a3964aca0450d43d6cf233bd03233f4ba92f8
 
 const rand3BZ2Hex = "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"
 const rand3Hex = "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"
+
+const (
+	digits = iota
+	twain
+)
+
+var testfiles = []string{
+	// Digits is the digits of the irrational number e. Its decimal representation
+	// does not repeat, but there are only 10 posible digits, so it should be
+	// reasonably compressible.
+	digits: "testdata/e.txt.bz2",
+	// Twain is Project Gutenberg's edition of Mark Twain's classic English novel.
+	twain: "testdata/Mark.Twain-Tom.Sawyer.txt.bz2",
+}
+
+func benchmarkDecode(b *testing.B, testfile int) {
+	compressed, err := ioutil.ReadFile(testfiles[testfile])
+	if err != nil {
+		b.Fatal(err)
+	}
+	b.SetBytes(int64(len(compressed)))
+	for i := 0; i < b.N; i++ {
+		r := bytes.NewBuffer(compressed)
+		io.Copy(ioutil.Discard, NewReader(r))
+	}
+}
+
+func BenchmarkDecodeDigits(b *testing.B) { benchmarkDecode(b, digits) }
+func BenchmarkDecodeTwain(b *testing.B)  { benchmarkDecode(b, twain) }

src/pkg/compress/bzip2/huffman.go

@@ -33,14 +33,17 @@ const invalidNodeValue = 0xffff
 
 // Decode reads bits from the given bitReader and navigates the tree until a
 // symbol is found.
-func (t huffmanTree) Decode(br *bitReader) (v uint16) {
+func (t *huffmanTree) Decode(br *bitReader) (v uint16) {
 	nodeIndex := uint16(0) // node 0 is the root of the tree.
 
 	for {
 		node := &t.nodes[nodeIndex]
-		bit := br.ReadBit()
+		bit, ok := br.TryReadBit()
+		if !ok && br.ReadBit() {
+			bit = 1
+		}
 		// bzip2 encodes left as a true bit.
-		if bit {
+		if bit != 0 {
 			// left
 			if node.left == invalidNodeValue {
 				return node.leftValue

src/pkg/compress/bzip2/move_to_front.go

@@ -15,10 +15,11 @@ type moveToFrontDecoder struct {
 	// Rather than actually keep the list in memory, the symbols are stored
 	// as a circular, double linked list with the symbol indexed by head
 	// at the front of the list.
-	symbols []byte
-	next    []uint8
-	prev    []uint8
+	symbols [256]byte
+	next    [256]uint8
+	prev    [256]uint8
 	head    uint8
+	len     int
 }
 
 // newMTFDecoder creates a move-to-front decoder with an explicit initial list
@@ -28,12 +29,9 @@ func newMTFDecoder(symbols []byte) *moveToFrontDecoder {
 		panic("too many symbols")
 	}
 
-	m := &moveToFrontDecoder{
-		symbols: symbols,
-		next:    make([]uint8, len(symbols)),
-		prev:    make([]uint8, len(symbols)),
-	}
-
+	m := new(moveToFrontDecoder)
+	copy(m.symbols[:], symbols)
+	m.len = len(symbols)
 	m.threadLinkedList()
 	return m
 }
@@ -45,34 +43,29 @@ func newMTFDecoderWithRange(n int) *moveToFrontDecoder {
 		panic("newMTFDecoderWithRange: cannot have > 256 symbols")
 	}
 
-	m := &moveToFrontDecoder{
-		symbols: make([]uint8, n),
-		next:    make([]uint8, n),
-		prev:    make([]uint8, n),
-	}
-
+	m := new(moveToFrontDecoder)
 	for i := 0; i < n; i++ {
-		m.symbols[i] = byte(i)
+		m.symbols[byte(i)] = byte(i)
 	}
-
+	m.len = n
 	m.threadLinkedList()
 	return m
 }
 
 // threadLinkedList creates the initial linked-list pointers.
 func (m *moveToFrontDecoder) threadLinkedList() {
-	if len(m.symbols) == 0 {
+	if m.len == 0 {
 		return
 	}
 
-	m.prev[0] = uint8(len(m.symbols) - 1)
+	m.prev[0] = uint8(m.len - 1)
 
-	for i := 0; i < len(m.symbols)-1; i++ {
+	for i := byte(0); int(i) < m.len-1; i++ {
 		m.next[i] = uint8(i + 1)
 		m.prev[i+1] = uint8(i)
 	}
 
-	m.next[len(m.symbols)-1] = 0
+	m.next[m.len-1] = 0
 }
 
 func (m *moveToFrontDecoder) Decode(n int) (b byte) {