Improve read speed

Utilize philhofer/fwd#32 to peek bytes

Eliminate bounds checks, mostly on reading sizes, but also integers.

```
benchmark                            old MB/s     new MB/s     speedup
BenchmarkReadMapHeaderBytes-32       1136.75      1136.89      1.00x
BenchmarkReadArrayHeaderBytes-32     1133.52      1149.34      1.01x
BenchmarkTestReadBytesHeader-32      1200.18      1171.21      0.98x
BenchmarkReadNilByte-32              2907.50      2176.63      0.75x
BenchmarkReadFloat64Bytes-32         4308.86      3869.64      0.90x
BenchmarkReadFloat32Bytes-32         2354.55      2332.21      0.99x
BenchmarkReadBoolBytes-32            863.66       1124.07      1.30x
BenchmarkReadTimeBytes-32            3710.27      3668.07      0.99x
BenchmarkReadMapHeader-32            228.74       391.76       1.71x
BenchmarkReadArrayHeader-32          243.90       454.31       1.86x
BenchmarkReadNil-32                  127.42       191.55       1.50x
BenchmarkReadFloat64-32              1046.08      1142.49      1.09x
BenchmarkReadFloat32-32              595.67       669.65       1.12x
BenchmarkReadInt64-32                422.29       798.35       1.89x
BenchmarkReadUintWithInt64-32        186.59       371.28       1.99x
BenchmarkReadUint64-32               282.87       377.76       1.34x
BenchmarkReadIntWithUint64-32        465.30       684.49       1.47x
BenchmarkRead16Bytes-32              1021.15      1021.38      1.00x
BenchmarkRead256Bytes-32             5324.42      6166.06      1.16x
BenchmarkRead2048Bytes-32            8405.41      8382.61      1.00x
BenchmarkRead16StringAsBytes-32      1102.72      1321.71      1.20x
BenchmarkRead256StringAsBytes-32     6959.78      6627.49      0.95x
BenchmarkRead16String-32             320.75       378.46       1.18x
BenchmarkRead256String-32            2066.40      2242.78      1.09x
BenchmarkReadComplex64-32            900.13       1064.21      1.18x
BenchmarkReadComplex128-32           1809.88      2028.11      1.12x
BenchmarkReadTime-32                 1362.46      1425.79      1.05x
```

Apologies for the noisy benchmark. `BenchmarkReadNilByte` and `BenchmarkReadFloat64Byte` does not touch new code, so it seems to be "micro-bench-itis", where random deltas show up. Bench takes ~30m to run for whatever reason.

The compiler is not clever enough to track back to `l := len(b)`, so it inserts a bounds check. Also `len(b) < 3` and `big.Uint16(b[1:])` causes a bounds check. So we rewrite those to avoid it.

This should cover the lowest hanging fruits.

go.mod

@@ -3,7 +3,7 @@ module github.com/tinylib/msgp
 go 1.20
 
 require (
-	github.com/philhofer/fwd v1.1.3-0.20240612014219-fbbf4953d986
+	github.com/philhofer/fwd v1.1.3-0.20240916144458-20a13a1f6b7c
 	golang.org/x/tools v0.22.0
 )
 

go.sum

@@ -1,5 +1,5 @@
-github.com/philhofer/fwd v1.1.3-0.20240612014219-fbbf4953d986 h1:jYi87L8j62qkXzaYHAQAhEapgukhenIMZRBKTNRLHJ4=
-github.com/philhofer/fwd v1.1.3-0.20240612014219-fbbf4953d986/go.mod h1:RqIHx9QI14HlwKwm98g9Re5prTQ6LdeRQn+gXJFxsJM=
+github.com/philhofer/fwd v1.1.3-0.20240916144458-20a13a1f6b7c h1:dAMKvw0MlJT1GshSTtih8C2gDs04w8dReiOGXrGLNoY=
+github.com/philhofer/fwd v1.1.3-0.20240916144458-20a13a1f6b7c/go.mod h1:RqIHx9QI14HlwKwm98g9Re5prTQ6LdeRQn+gXJFxsJM=
 golang.org/x/mod v0.18.0 h1:5+9lSbEzPSdWkH32vYPBwEpX8KwDbM52Ud9xBUvNlb0=
 golang.org/x/mod v0.18.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
 golang.org/x/sync v0.7.0 h1:YsImfSBoP9QPYL0xyKJPq0gcaJdG3rInoqxTWbfQu9M=

msgp/json.go

@@ -346,14 +346,12 @@ func rwExtension(dst jsWriter, src *Reader) (n int, err error) {
 }
 
 func rwString(dst jsWriter, src *Reader) (n int, err error) {
-	var p []byte
-	p, err = src.R.Peek(1)
+	lead, err := src.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead := p[0]
 	var read int
-
+	var p []byte
 	if isfixstr(lead) {
 		read = int(rfixstr(lead))
 		src.R.Skip(1)

msgp/read.go

@@ -236,13 +236,13 @@ func (m *Reader) BufferSize() int { return m.R.BufferSize() }
 
 // NextType returns the next object type to be decoded.
 func (m *Reader) NextType() (Type, error) {
-	p, err := m.R.Peek(1)
+	next, err := m.R.PeekByte()
 	if err != nil {
 		return InvalidType, err
 	}
-	t := getType(p[0])
+	t := getType(next)
 	if t == InvalidType {
-		return t, InvalidPrefixError(p[0])
+		return t, InvalidPrefixError(next)
 	}
 	if t == ExtensionType {
 		v, err := m.peekExtensionType()
@@ -264,8 +264,8 @@ func (m *Reader) NextType() (Type, error) {
 // IsNil returns whether or not
 // the next byte is a null messagepack byte
 func (m *Reader) IsNil() bool {
-	p, err := m.R.Peek(1)
-	return err == nil && p[0] == mnil
+	p, err := m.R.PeekByte()
+	return err == nil && p == mnil
 }
 
 // getNextSize returns the size of the next object on the wire.
@@ -276,11 +276,10 @@ func (m *Reader) IsNil() bool {
 // use uintptr b/c it's guaranteed to be large enough
 // to hold whatever we can fit in memory.
 func getNextSize(r *fwd.Reader) (uintptr, uintptr, error) {
-	b, err := r.Peek(1)
+	lead, err := r.PeekByte()
 	if err != nil {
 		return 0, 0, err
 	}
-	lead := b[0]
 	spec := getBytespec(lead)
 	size, mode := spec.size, spec.extra
 	if size == 0 {
@@ -289,7 +288,7 @@ func getNextSize(r *fwd.Reader) (uintptr, uintptr, error) {
 	if mode >= 0 {
 		return uintptr(size), uintptr(mode), nil
 	}
-	b, err = r.Peek(int(size))
+	b, err := r.Peek(int(size))
 	if err != nil {
 		return 0, 0, err
 	}
@@ -373,11 +372,10 @@ func (m *Reader) Skip() error {
 func (m *Reader) ReadMapHeader() (sz uint32, err error) {
 	var p []byte
 	var lead byte
-	p, err = m.R.Peek(1)
+	lead, err = m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead = p[0]
 	if isfixmap(lead) {
 		sz = uint32(rfixmap(lead))
 		_, err = m.R.Skip(1)
@@ -427,12 +425,12 @@ func (m *Reader) ReadMapKey(scratch []byte) ([]byte, error) {
 // method; writing into the returned slice may
 // corrupt future reads.
 func (m *Reader) ReadMapKeyPtr() ([]byte, error) {
-	p, err := m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return nil, err
 	}
-	lead := p[0]
 	var read int
+	var p []byte
 	if isfixstr(lead) {
 		read = int(rfixstr(lead))
 		m.R.Skip(1)
@@ -471,18 +469,16 @@ fill:
 // array header and returns the size of the array
 // and the number of bytes read.
 func (m *Reader) ReadArrayHeader() (sz uint32, err error) {
-	var lead byte
-	var p []byte
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead = p[0]
 	if isfixarray(lead) {
 		sz = uint32(rfixarray(lead))
 		_, err = m.R.Skip(1)
 		return
 	}
+	var p []byte
 	switch lead {
 	case marray16:
 		p, err = m.R.Next(3)
@@ -508,12 +504,12 @@ func (m *Reader) ReadArrayHeader() (sz uint32, err error) {
 
 // ReadNil reads a 'nil' MessagePack byte from the reader
 func (m *Reader) ReadNil() error {
-	p, err := m.R.Peek(1)
+	p, err := m.R.PeekByte()
 	if err != nil {
 		return err
 	}
-	if p[0] != mnil {
-		return badPrefix(NilType, p[0])
+	if p != mnil {
+		return badPrefix(NilType, p)
 	}
 	_, err = m.R.Skip(1)
 	return err
@@ -566,17 +562,17 @@ func (m *Reader) ReadFloat32() (f float32, err error) {
 
 // ReadBool reads a bool from the reader
 func (m *Reader) ReadBool() (b bool, err error) {
-	var p []byte
-	p, err = m.R.Peek(1)
+	var p byte
+	p, err = m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	switch p[0] {
+	switch p {
 	case mtrue:
 		b = true
 	case mfalse:
 	default:
-		err = badPrefix(BoolType, p[0])
+		err = badPrefix(BoolType, p)
 		return
 	}
 	_, err = m.R.Skip(1)
@@ -592,12 +588,10 @@ func (m *Reader) ReadDuration() (d time.Duration, err error) {
 // ReadInt64 reads an int64 from the reader
 func (m *Reader) ReadInt64() (i int64, err error) {
 	var p []byte
-	var lead byte
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead = p[0]
 
 	if isfixint(lead) {
 		i = int64(rfixint(lead))
@@ -738,12 +732,10 @@ func (m *Reader) ReadInt() (i int, err error) {
 // ReadUint64 reads a uint64 from the reader
 func (m *Reader) ReadUint64() (u uint64, err error) {
 	var p []byte
-	var lead byte
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead = p[0]
 	if isfixint(lead) {
 		u = uint64(rfixint(lead))
 		_, err = m.R.Skip(1)
@@ -958,11 +950,11 @@ func (m *Reader) ReadBytes(scratch []byte) (b []byte, err error) {
 // way.
 func (m *Reader) ReadBytesHeader() (sz uint32, err error) {
 	var p []byte
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	switch p[0] {
+	switch lead {
 	case mbin8:
 		p, err = m.R.Next(2)
 		if err != nil {
@@ -1036,12 +1028,10 @@ func (m *Reader) ReadExactBytes(into []byte) error {
 // if it is non-nil.
 func (m *Reader) ReadStringAsBytes(scratch []byte) (b []byte, err error) {
 	var p []byte
-	var lead byte
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead = p[0]
 	var read int64
 
 	if isfixstr(lead) {
@@ -1088,17 +1078,16 @@ fill:
 // for dealing with the next 'sz' bytes from
 // the reader in an application-specific manner.
 func (m *Reader) ReadStringHeader() (sz uint32, err error) {
-	var p []byte
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead := p[0]
 	if isfixstr(lead) {
 		sz = uint32(rfixstr(lead))
 		m.R.Skip(1)
 		return
 	}
+	var p []byte
 	switch lead {
 	case mstr8:
 		p, err = m.R.Next(2)
@@ -1129,15 +1118,13 @@ func (m *Reader) ReadStringHeader() (sz uint32, err error) {
 
 // ReadString reads a utf-8 string from the reader
 func (m *Reader) ReadString() (s string, err error) {
-	var p []byte
-	var lead byte
 	var read int64
-	p, err = m.R.Peek(1)
+	lead, err := m.R.PeekByte()
 	if err != nil {
 		return
 	}
-	lead = p[0]
 
+	var p []byte
 	if isfixstr(lead) {
 		read = int64(rfixstr(lead))
 		m.R.Skip(1)