bytes: make bytes.Buffer cache-friendly

During benchmark of an internal tool we found out that (*Buffer).Reset() was
surprisingly showing up in CPU profiles.

This CL contains two related changes aimed at speeding up Reset():
1. Create a fast path for Truncate(0) by moving the logic to Reset()
   (this makes Reset() a simple leaf func that gets inlined since it
   gets compiled to 3 MOVx instructions). Accordingly change calls in
   the rest of the Buffer methods to call Reset() instead of Truncate(0).
2. Reorder the fields in the Buffer struct so that frequently accessed
   fields are packed together (buf, off, lastRead). This also make them
   likely to be in the same cacheline.

Ideally it would be advisable to have Buffer{} cacheline-aligned, but I
couldn't find a way to do this without changing the size of the bootstrap
array (but this will cause some regressions, because it will make duffcopy
show up in CPU profiles where it wasn't showing up before).

go1 benchmarks are not really affected, but some other benchmarks that
exercise Buffer more show improvements:

name                     old time/op    new time/op    delta
BinaryTree17-4              2.46s ± 9%     2.43s ± 3%    ~     (p=0.982 n=14+14)
Fannkuch11-4                2.98s ± 1%     2.90s ± 1%  -2.58%  (p=0.000 n=15+14)
FmtFprintfEmpty-4          45.2ns ± 1%    45.2ns ± 1%    ~     (p=0.494 n=14+15)
FmtFprintfString-4         76.8ns ± 1%    83.1ns ± 2%  +8.23%  (p=0.000 n=10+15)
FmtFprintfInt-4            78.0ns ± 2%    74.6ns ± 1%  -4.46%  (p=0.000 n=15+15)
FmtFprintfIntInt-4          113ns ± 1%     109ns ± 2%  -2.91%  (p=0.000 n=13+15)
FmtFprintfPrefixedInt-4     152ns ± 2%     143ns ± 2%  -6.04%  (p=0.000 n=15+14)
FmtFprintfFloat-4           224ns ± 1%     222ns ± 2%  -1.08%  (p=0.001 n=15+14)
FmtManyArgs-4               464ns ± 2%     463ns ± 2%    ~     (p=0.303 n=14+15)
GobDecode-4                6.25ms ± 2%    6.32ms ± 3%  +1.20%  (p=0.002 n=14+14)
GobEncode-4                5.41ms ± 2%    5.41ms ± 2%    ~     (p=0.967 n=15+15)
Gzip-4                      215ms ± 2%     218ms ± 2%  +1.35%  (p=0.002 n=15+15)
Gunzip-4                   34.3ms ± 2%    34.2ms ± 2%    ~     (p=0.539 n=15+15)
HTTPClientServer-4         76.4µs ± 2%    75.4µs ± 1%  -1.31%  (p=0.000 n=15+15)
JSONEncode-4               14.7ms ± 2%    14.6ms ± 3%    ~     (p=0.094 n=14+14)
JSONDecode-4               48.0ms ± 1%    48.5ms ± 1%  +0.92%  (p=0.001 n=14+12)
Mandelbrot200-4            4.04ms ± 2%    4.06ms ± 1%    ~     (p=0.108 n=15+13)
GoParse-4                  2.99ms ± 2%    3.00ms ± 1%    ~     (p=0.130 n=15+13)
RegexpMatchEasy0_32-4      78.3ns ± 1%    79.5ns ± 1%  +1.51%  (p=0.000 n=15+14)
RegexpMatchEasy0_1K-4       185ns ± 1%     186ns ± 1%  +0.76%  (p=0.005 n=15+15)
RegexpMatchEasy1_32-4      79.0ns ± 2%    76.7ns ± 1%  -2.87%  (p=0.000 n=14+15)

name                     old speed      new speed      delta
GobDecode-4               123MB/s ± 2%   121MB/s ± 3%  -1.18%  (p=0.002 n=14+14)
GobEncode-4               142MB/s ± 2%   142MB/s ± 1%    ~     (p=0.959 n=15+15)
Gzip-4                   90.3MB/s ± 2%  89.1MB/s ± 2%  -1.34%  (p=0.002 n=15+15)
Gunzip-4                  565MB/s ± 2%   567MB/s ± 2%    ~     (p=0.539 n=15+15)
JSONEncode-4              132MB/s ± 2%   133MB/s ± 3%    ~     (p=0.091 n=14+14)
JSONDecode-4             40.4MB/s ± 1%  40.0MB/s ± 1%  -0.92%  (p=0.001 n=14+12)
GoParse-4                19.4MB/s ± 2%  19.3MB/s ± 1%    ~     (p=0.121 n=15+13)
RegexpMatchEasy0_32-4     409MB/s ± 1%   403MB/s ± 1%  -1.47%  (p=0.000 n=15+14)
RegexpMatchEasy0_1K-4    5.53GB/s ± 1%  5.49GB/s ± 1%  -0.86%  (p=0.002 n=15+15)
RegexpMatchEasy1_32-4     405MB/s ± 2%   417MB/s ± 1%  +2.94%  (p=0.000 n=14+15)

name                old time/op  new time/op  delta
PoolsSingle1K-4     34.9ns ± 2%  30.4ns ± 4%  -12.80%  (p=0.000 n=15+15)
PoolsSingle64K-4    36.9ns ± 1%  34.4ns ± 4%   -6.72%  (p=0.000 n=14+15)
PoolsRandomSmall-4  34.8ns ± 3%  29.5ns ± 1%  -15.19%  (p=0.000 n=15+14)
PoolsRandomLarge-4  38.6ns ± 1%  34.3ns ± 3%  -11.17%  (p=0.000 n=14+15)
PoolSingle1K-4      26.1ns ± 1%  21.2ns ± 2%  -18.59%  (p=0.000 n=15+14)
PoolSingle64K-4     26.7ns ± 2%  21.5ns ± 2%  -19.72%  (p=0.000 n=15+15)
MakeSingle1K-4      24.2ns ± 2%  24.3ns ± 3%     ~     (p=0.132 n=13+15)
MakeSingle64K-4     6.76µs ± 1%  6.96µs ± 5%   +2.94%  (p=0.002 n=13+13)
MakeRandomSmall-4    531ns ± 4%   538ns ± 5%     ~     (p=0.066 n=14+15)
MakeRandomLarge-4    152µs ± 0%   152µs ± 1%   -0.31%  (p=0.001 n=14+13)

Change-Id: I86d7d9d2cac65335baf62214fbb35ba0fd8f9528
Reviewed-on: https://go-review.googlesource.com/37416
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

src/bytes/buffer.go

@@ -15,10 +15,15 @@ import (
 // A Buffer is a variable-sized buffer of bytes with Read and Write methods.
 // The zero value for Buffer is an empty buffer ready to use.
 type Buffer struct {
-	buf       []byte   // contents are the bytes buf[off : len(buf)]
-	off       int      // read at &buf[off], write at &buf[len(buf)]
-	bootstrap [64]byte // memory to hold first slice; helps small buffers avoid allocation.
-	lastRead  readOp   // last read operation, so that Unread* can work correctly.
+	buf      []byte // contents are the bytes buf[off : len(buf)]
+	off      int    // read at &buf[off], write at &buf[len(buf)]
+	lastRead readOp // last read operation, so that Unread* can work correctly.
+	// FIXME: lastRead can fit in a single byte
+
+	// memory to hold first slice; helps small buffers avoid allocation.
+	// FIXME: it would be advisable to align Buffer to cachelines to avoid false
+	// sharing.
+	bootstrap [64]byte
 }
 
 // The readOp constants describe the last action performed on
@@ -68,13 +73,13 @@ func (b *Buffer) Cap() int { return cap(b.buf) }
 // but continues to use the same allocated storage.
 // It panics if n is negative or greater than the length of the buffer.
 func (b *Buffer) Truncate(n int) {
+	if n == 0 {
+		b.Reset()
+		return
+	}
 	b.lastRead = opInvalid
-	switch {
-	case n < 0 || n > b.Len():
+	if n < 0 || n > b.Len() {
 		panic("bytes.Buffer: truncation out of range")
-	case n == 0:
-		// Reuse buffer space.
-		b.off = 0
 	}
 	b.buf = b.buf[0 : b.off+n]
 }
@@ -82,7 +87,11 @@ func (b *Buffer) Truncate(n int) {
 // Reset resets the buffer to be empty,
 // but it retains the underlying storage for use by future writes.
 // Reset is the same as Truncate(0).
-func (b *Buffer) Reset() { b.Truncate(0) }
+func (b *Buffer) Reset() {
+	b.buf = b.buf[:0]
+	b.off = 0
+	b.lastRead = opInvalid
+}
 
 // grow grows the buffer to guarantee space for n more bytes.
 // It returns the index where bytes should be written.
@@ -91,7 +100,7 @@ func (b *Buffer) grow(n int) int {
 	m := b.Len()
 	// If buffer is empty, reset to recover space.
 	if m == 0 && b.off != 0 {
-		b.Truncate(0)
+		b.Reset()
 	}
 	if len(b.buf)+n > cap(b.buf) {
 		var buf []byte
@@ -161,7 +170,7 @@ func (b *Buffer) ReadFrom(r io.Reader) (n int64, err error) {
 	b.lastRead = opInvalid
 	// If buffer is empty, reset to recover space.
 	if b.off >= len(b.buf) {
-		b.Truncate(0)
+		b.Reset()
 	}
 	for {
 		if free := cap(b.buf) - len(b.buf); free < MinRead {
@@ -225,7 +234,7 @@ func (b *Buffer) WriteTo(w io.Writer) (n int64, err error) {
 		}
 	}
 	// Buffer is now empty; reset.
-	b.Truncate(0)
+	b.Reset()
 	return
 }
 
@@ -264,7 +273,7 @@ func (b *Buffer) Read(p []byte) (n int, err error) {
 	b.lastRead = opInvalid
 	if b.off >= len(b.buf) {
 		// Buffer is empty, reset to recover space.
-		b.Truncate(0)
+		b.Reset()
 		if len(p) == 0 {
 			return
 		}
@@ -302,7 +311,7 @@ func (b *Buffer) ReadByte() (byte, error) {
 	b.lastRead = opInvalid
 	if b.off >= len(b.buf) {
 		// Buffer is empty, reset to recover space.
-		b.Truncate(0)
+		b.Reset()
 		return 0, io.EOF
 	}
 	c := b.buf[b.off]
@@ -320,7 +329,7 @@ func (b *Buffer) ReadRune() (r rune, size int, err error) {
 	b.lastRead = opInvalid
 	if b.off >= len(b.buf) {
 		// Buffer is empty, reset to recover space.
-		b.Truncate(0)
+		b.Reset()
 		return 0, 0, io.EOF
 	}
 	c := b.buf[b.off]