Adding upstream version 1.20.14.upstream/1.20.14upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

13 files changed, 2143 insertions, 0 deletions

diff --git a/src/crypto/md5/example_test.go b/src/crypto/md5/example_test.go
new file mode 100644
index 0000000..af8c1bf
--- /dev/null
+++ b/src/crypto/md5/example_test.go
@@ -0,0 +1,42 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package md5_test
+
+import (
+	"crypto/md5"
+	"fmt"
+	"io"
+	"log"
+	"os"
+)
+
+func ExampleNew() {
+	h := md5.New()
+	io.WriteString(h, "The fog is getting thicker!")
+	io.WriteString(h, "And Leon's getting laaarger!")
+	fmt.Printf("%x", h.Sum(nil))
+	// Output: e2c569be17396eca2a2e3c11578123ed
+}
+
+func ExampleSum() {
+	data := []byte("These pretzels are making me thirsty.")
+	fmt.Printf("%x", md5.Sum(data))
+	// Output: b0804ec967f48520697662a204f5fe72
+}
+
+func ExampleNew_file() {
+	f, err := os.Open("file.txt")
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer f.Close()
+
+	h := md5.New()
+	if _, err := io.Copy(h, f); err != nil {
+		log.Fatal(err)
+	}
+
+	fmt.Printf("%x", h.Sum(nil))
+}
diff --git a/src/crypto/md5/gen.go b/src/crypto/md5/gen.go
new file mode 100644
index 0000000..cd2700a
--- /dev/null
+++ b/src/crypto/md5/gen.go
@@ -0,0 +1,259 @@
+// Copyright 2012 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build ignore
+
+// This program generates md5block.go
+// Invoke as
+//
+//	go run gen.go -output md5block.go
+
+package main
+
+import (
+	"bytes"
+	"flag"
+	"go/format"
+	"log"
+	"os"
+	"strings"
+	"text/template"
+)
+
+var filename = flag.String("output", "md5block.go", "output file name")
+
+func main() {
+	flag.Parse()
+
+	var buf bytes.Buffer
+
+	t := template.Must(template.New("main").Funcs(funcs).Parse(program))
+	if err := t.Execute(&buf, data); err != nil {
+		log.Fatal(err)
+	}
+
+	data, err := format.Source(buf.Bytes())
+	if err != nil {
+		log.Fatal(err)
+	}
+	err = os.WriteFile(*filename, data, 0644)
+	if err != nil {
+		log.Fatal(err)
+	}
+}
+
+type Data struct {
+	a, b, c, d string
+	Shift1     []int
+	Shift2     []int
+	Shift3     []int
+	Shift4     []int
+	Table1     []uint32
+	Table2     []uint32
+	Table3     []uint32
+	Table4     []uint32
+}
+
+var funcs = template.FuncMap{
+	"dup":     dup,
+	"relabel": relabel,
+	"rotate":  rotate,
+	"idx":     idx,
+	"seq":     seq,
+}
+
+func dup(count int, x []int) []int {
+	var out []int
+	for i := 0; i < count; i++ {
+		out = append(out, x...)
+	}
+	return out
+}
+
+func relabel(s string) string {
+	return strings.NewReplacer("arg0", data.a, "arg1", data.b, "arg2", data.c, "arg3", data.d).Replace(s)
+}
+
+func rotate() string {
+	data.a, data.b, data.c, data.d = data.d, data.a, data.b, data.c
+	return "" // no output
+}
+
+func idx(round, index int) int {
+	v := 0
+	switch round {
+	case 1:
+		v = index
+	case 2:
+		v = (1 + 5*index) & 15
+	case 3:
+		v = (5 + 3*index) & 15
+	case 4:
+		v = (7 * index) & 15
+	}
+	return v
+}
+
+func seq(i int) []int {
+	s := make([]int, i)
+	for i := range s {
+		s[i] = i
+	}
+	return s
+}
+
+var data = Data{
+	a:      "a",
+	b:      "b",
+	c:      "c",
+	d:      "d",
+	Shift1: []int{7, 12, 17, 22},
+	Shift2: []int{5, 9, 14, 20},
+	Shift3: []int{4, 11, 16, 23},
+	Shift4: []int{6, 10, 15, 21},
+
+	// table[i] = int((1<<32) * abs(sin(i+1 radians))).
+	Table1: []uint32{
+		// round 1
+		0xd76aa478,
+		0xe8c7b756,
+		0x242070db,
+		0xc1bdceee,
+		0xf57c0faf,
+		0x4787c62a,
+		0xa8304613,
+		0xfd469501,
+		0x698098d8,
+		0x8b44f7af,
+		0xffff5bb1,
+		0x895cd7be,
+		0x6b901122,
+		0xfd987193,
+		0xa679438e,
+		0x49b40821,
+	},
+	Table2: []uint32{
+		// round 2
+		0xf61e2562,
+		0xc040b340,
+		0x265e5a51,
+		0xe9b6c7aa,
+		0xd62f105d,
+		0x2441453,
+		0xd8a1e681,
+		0xe7d3fbc8,
+		0x21e1cde6,
+		0xc33707d6,
+		0xf4d50d87,
+		0x455a14ed,
+		0xa9e3e905,
+		0xfcefa3f8,
+		0x676f02d9,
+		0x8d2a4c8a,
+	},
+	Table3: []uint32{
+		// round3
+		0xfffa3942,
+		0x8771f681,
+		0x6d9d6122,
+		0xfde5380c,
+		0xa4beea44,
+		0x4bdecfa9,
+		0xf6bb4b60,
+		0xbebfbc70,
+		0x289b7ec6,
+		0xeaa127fa,
+		0xd4ef3085,
+		0x4881d05,
+		0xd9d4d039,
+		0xe6db99e5,
+		0x1fa27cf8,
+		0xc4ac5665,
+	},
+	Table4: []uint32{
+		// round 4
+		0xf4292244,
+		0x432aff97,
+		0xab9423a7,
+		0xfc93a039,
+		0x655b59c3,
+		0x8f0ccc92,
+		0xffeff47d,
+		0x85845dd1,
+		0x6fa87e4f,
+		0xfe2ce6e0,
+		0xa3014314,
+		0x4e0811a1,
+		0xf7537e82,
+		0xbd3af235,
+		0x2ad7d2bb,
+		0xeb86d391,
+	},
+}
+
+var program = `// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Code generated by go run gen.go -output md5block.go; DO NOT EDIT.
+
+package md5
+
+import (
+	"encoding/binary"
+	"math/bits"
+)
+
+func blockGeneric(dig *digest, p []byte) {
+	// load state
+	a, b, c, d := dig.s[0], dig.s[1], dig.s[2], dig.s[3]
+
+	for i := 0; i <= len(p)-BlockSize; i += BlockSize {
+		// eliminate bounds checks on p
+		q := p[i:]
+		q = q[:BlockSize:BlockSize]
+
+		// save current state
+		aa, bb, cc, dd := a, b, c, d
+
+		// load input block
+		{{range $i := seq 16 -}}
+			{{printf "x%x := binary.LittleEndian.Uint32(q[4*%#x:])" $i $i}}
+		{{end}}
+
+		// round 1
+		{{range $i, $s := dup 4 .Shift1 -}}
+			{{printf "arg0 = arg1 + bits.RotateLeft32((((arg2^arg3)&arg1)^arg3)+arg0+x%x+%#08x, %d)" (idx 1 $i) (index $.Table1 $i) $s | relabel}}
+			{{rotate -}}
+		{{end}}
+	
+		// round 2
+		{{range $i, $s := dup 4 .Shift2 -}}
+			{{printf "arg0 = arg1 + bits.RotateLeft32((((arg1^arg2)&arg3)^arg2)+arg0+x%x+%#08x, %d)" (idx 2 $i) (index $.Table2 $i) $s | relabel}}
+			{{rotate -}}
+		{{end}}
+	
+		// round 3
+		{{range $i, $s := dup 4 .Shift3 -}}
+			{{printf "arg0 = arg1 + bits.RotateLeft32((arg1^arg2^arg3)+arg0+x%x+%#08x, %d)" (idx 3 $i) (index $.Table3 $i) $s | relabel}}
+			{{rotate -}}
+		{{end}}
+	
+		// round 4
+		{{range $i, $s := dup 4 .Shift4 -}}
+			{{printf "arg0 = arg1 + bits.RotateLeft32((arg2^(arg1|^arg3))+arg0+x%x+%#08x, %d)" (idx 4 $i) (index $.Table4 $i) $s | relabel}}
+			{{rotate -}}
+		{{end}}
+
+		// add saved state
+		a += aa
+		b += bb
+		c += cc
+		d += dd
+	}
+
+	// save state
+	dig.s[0], dig.s[1], dig.s[2], dig.s[3] = a, b, c, d
+}
+`
diff --git a/src/crypto/md5/md5.go b/src/crypto/md5/md5.go
new file mode 100644
index 0000000..ccee4ea
--- /dev/null
+++ b/src/crypto/md5/md5.go
@@ -0,0 +1,183 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:generate go run gen.go -output md5block.go
+
+// Package md5 implements the MD5 hash algorithm as defined in RFC 1321.
+//
+// MD5 is cryptographically broken and should not be used for secure
+// applications.
+package md5
+
+import (
+	"crypto"
+	"encoding/binary"
+	"errors"
+	"hash"
+)
+
+func init() {
+	crypto.RegisterHash(crypto.MD5, New)
+}
+
+// The size of an MD5 checksum in bytes.
+const Size = 16
+
+// The blocksize of MD5 in bytes.
+const BlockSize = 64
+
+const (
+	init0 = 0x67452301
+	init1 = 0xEFCDAB89
+	init2 = 0x98BADCFE
+	init3 = 0x10325476
+)
+
+// digest represents the partial evaluation of a checksum.
+type digest struct {
+	s   [4]uint32
+	x   [BlockSize]byte
+	nx  int
+	len uint64
+}
+
+func (d *digest) Reset() {
+	d.s[0] = init0
+	d.s[1] = init1
+	d.s[2] = init2
+	d.s[3] = init3
+	d.nx = 0
+	d.len = 0
+}
+
+const (
+	magic         = "md5\x01"
+	marshaledSize = len(magic) + 4*4 + BlockSize + 8
+)
+
+func (d *digest) MarshalBinary() ([]byte, error) {
+	b := make([]byte, 0, marshaledSize)
+	b = append(b, magic...)
+	b = binary.BigEndian.AppendUint32(b, d.s[0])
+	b = binary.BigEndian.AppendUint32(b, d.s[1])
+	b = binary.BigEndian.AppendUint32(b, d.s[2])
+	b = binary.BigEndian.AppendUint32(b, d.s[3])
+	b = append(b, d.x[:d.nx]...)
+	b = b[:len(b)+len(d.x)-d.nx] // already zero
+	b = binary.BigEndian.AppendUint64(b, d.len)
+	return b, nil
+}
+
+func (d *digest) UnmarshalBinary(b []byte) error {
+	if len(b) < len(magic) || string(b[:len(magic)]) != magic {
+		return errors.New("crypto/md5: invalid hash state identifier")
+	}
+	if len(b) != marshaledSize {
+		return errors.New("crypto/md5: invalid hash state size")
+	}
+	b = b[len(magic):]
+	b, d.s[0] = consumeUint32(b)
+	b, d.s[1] = consumeUint32(b)
+	b, d.s[2] = consumeUint32(b)
+	b, d.s[3] = consumeUint32(b)
+	b = b[copy(d.x[:], b):]
+	b, d.len = consumeUint64(b)
+	d.nx = int(d.len % BlockSize)
+	return nil
+}
+
+func consumeUint64(b []byte) ([]byte, uint64) {
+	return b[8:], binary.BigEndian.Uint64(b[0:8])
+}
+
+func consumeUint32(b []byte) ([]byte, uint32) {
+	return b[4:], binary.BigEndian.Uint32(b[0:4])
+}
+
+// New returns a new hash.Hash computing the MD5 checksum. The Hash also
+// implements encoding.BinaryMarshaler and encoding.BinaryUnmarshaler to
+// marshal and unmarshal the internal state of the hash.
+func New() hash.Hash {
+	d := new(digest)
+	d.Reset()
+	return d
+}
+
+func (d *digest) Size() int { return Size }
+
+func (d *digest) BlockSize() int { return BlockSize }
+
+func (d *digest) Write(p []byte) (nn int, err error) {
+	// Note that we currently call block or blockGeneric
+	// directly (guarded using haveAsm) because this allows
+	// escape analysis to see that p and d don't escape.
+	nn = len(p)
+	d.len += uint64(nn)
+	if d.nx > 0 {
+		n := copy(d.x[d.nx:], p)
+		d.nx += n
+		if d.nx == BlockSize {
+			if haveAsm {
+				block(d, d.x[:])
+			} else {
+				blockGeneric(d, d.x[:])
+			}
+			d.nx = 0
+		}
+		p = p[n:]
+	}
+	if len(p) >= BlockSize {
+		n := len(p) &^ (BlockSize - 1)
+		if haveAsm {
+			block(d, p[:n])
+		} else {
+			blockGeneric(d, p[:n])
+		}
+		p = p[n:]
+	}
+	if len(p) > 0 {
+		d.nx = copy(d.x[:], p)
+	}
+	return
+}
+
+func (d *digest) Sum(in []byte) []byte {
+	// Make a copy of d so that caller can keep writing and summing.
+	d0 := *d
+	hash := d0.checkSum()
+	return append(in, hash[:]...)
+}
+
+func (d *digest) checkSum() [Size]byte {
+	// Append 0x80 to the end of the message and then append zeros
+	// until the length is a multiple of 56 bytes. Finally append
+	// 8 bytes representing the message length in bits.
+	//
+	// 1 byte end marker :: 0-63 padding bytes :: 8 byte length
+	tmp := [1 + 63 + 8]byte{0x80}
+	pad := (55 - d.len) % 64                             // calculate number of padding bytes
+	binary.LittleEndian.PutUint64(tmp[1+pad:], d.len<<3) // append length in bits
+	d.Write(tmp[:1+pad+8])
+
+	// The previous write ensures that a whole number of
+	// blocks (i.e. a multiple of 64 bytes) have been hashed.
+	if d.nx != 0 {
+		panic("d.nx != 0")
+	}
+
+	var digest [Size]byte
+	binary.LittleEndian.PutUint32(digest[0:], d.s[0])
+	binary.LittleEndian.PutUint32(digest[4:], d.s[1])
+	binary.LittleEndian.PutUint32(digest[8:], d.s[2])
+	binary.LittleEndian.PutUint32(digest[12:], d.s[3])
+	return digest
+}
+
+// Sum returns the MD5 checksum of the data.
+func Sum(data []byte) [Size]byte {
+	var d digest
+	d.Reset()
+	d.Write(data)
+	return d.checkSum()
+}
diff --git a/src/crypto/md5/md5_test.go b/src/crypto/md5/md5_test.go
new file mode 100644
index 0000000..851e7fb
--- /dev/null
+++ b/src/crypto/md5/md5_test.go
@@ -0,0 +1,294 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package md5
+
+import (
+	"bytes"
+	"crypto/rand"
+	"encoding"
+	"fmt"
+	"hash"
+	"io"
+	"testing"
+	"unsafe"
+)
+
+type md5Test struct {
+	out       string
+	in        string
+	halfState string // marshaled hash state after first half of in written, used by TestGoldenMarshal
+}
+
+var golden = []md5Test{
+	{"d41d8cd98f00b204e9800998ecf8427e", "", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tv\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"},
+	{"0cc175b9c0f1b6a831c399e269772661", "a", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tv\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"},
+	{"187ef4436122d1cc2f40dc2b92f0eba0", "ab", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tva\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"},
+	{"900150983cd24fb0d6963f7d28e17f72", "abc", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tva\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"},
+	{"e2fc714c4727ee9395f324cd2e7f331f", "abcd", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"},
+	{"ab56b4d92b40713acc5af89985d4b786", "abcde", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"},
+	{"e80b5017098950fc58aad83c8c14978e", "abcdef", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvabc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"},
+	{"7ac66c0f148de9519b8bd264312c4d64", "abcdefg", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvabc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"},
+	{"e8dc4081b13434b45189a720b77b6818", "abcdefgh", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvabcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04"},
+	{"8aa99b1f439ff71293e95357bac6fd94", "abcdefghi", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvabcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04"},
+	{"a925576942e94b2ef57a066101b48876", "abcdefghij", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvabcde\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x05"},
+	{"d747fc1719c7eacb84058196cfe56d57", "Discard medicine more than two years old.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvDiscard medicine mor\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x14"},
+	{"bff2dcb37ef3a44ba43ab144768ca837", "He who has a shady past knows that nice guys finish last.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvHe who has a shady past know\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"},
+	{"0441015ecb54a7342d017ed1bcfdbea5", "I wouldn't marry him with a ten foot pole.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvI wouldn't marry him \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x15"},
+	{"9e3cac8e9e9757a60c3ea391130d3689", "Free! Free!/A trip/to Mars/for 900/empty jars/Burma Shave", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvFree! Free!/A trip/to Mars/f\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"},
+	{"a0f04459b031f916a59a35cc482dc039", "The days of the digital watch are numbered.  -Tom Stoppard", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvThe days of the digital watch\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1d"},
+	{"e7a48e0fe884faf31475d2a04b1362cc", "Nepal premier won't resign.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvNepal premier\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\r"},
+	{"637d2fe925c07c113800509964fb0e06", "For every action there is an equal and opposite government program.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvFor every action there is an equa\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00!"},
+	{"834a8d18d5c6562119cf4c7f5086cb71", "His money is twice tainted: 'taint yours and 'taint mine.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvHis money is twice tainted: \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"},
+	{"de3a4d2fd6c73ec2db2abad23b444281", "There is no reason for any individual to have a computer in their home. -Ken Olsen, 1977", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvThere is no reason for any individual to hav\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00,"},
+	{"acf203f997e2cf74ea3aff86985aefaf", "It's a tiny change to the code and not completely disgusting. - Bob Manchek", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvIt's a tiny change to the code and no\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00%"},
+	{"e1c1384cb4d2221dfdd7c795a4222c9a", "size:  a.out:  bad magic", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102Tvsize:  a.out\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\f"},
+	{"c90f3ddecc54f34228c063d7525bf644", "The major problem is with sendmail.  -Mark Horton", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvThe major problem is wit\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x18"},
+	{"cdf7ab6c1fd49bd9933c43f3ea5af185", "Give me a rock, paper and scissors and I will move the world.  CCFestoon", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvGive me a rock, paper and scissors a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00$"},
+	{"83bc85234942fc883c063cbd7f0ad5d0", "If the enemy is within range, then so are you.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvIf the enemy is within \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x17"},
+	{"277cbe255686b48dd7e8f389394d9299", "It's well we cannot hear the screams/That we create in others' dreams.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvIt's well we cannot hear the scream\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00#"},
+	{"fd3fb0a7ffb8af16603f3d3af98f8e1f", "You remind me of a TV show, but that's all right: I watch it anyway.", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvYou remind me of a TV show, but th\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\""},
+	{"469b13a78ebf297ecda64d4723655154", "C is as portable as Stonehedge!!", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvC is as portable\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x10"},
+	{"63eb3a2f466410104731c4b037600110", "Even if I could be Shakespeare, I think I should still choose to be Faraday. - A. Huxley", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvEven if I could be Shakespeare, I think I sh\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00,"},
+	{"72c2ed7592debca1c90fc0100f931a2f", "The fugacity of a constituent in a mixture of gases at a given temperature is proportional to its mole fraction.  Lewis-Randall Rule", "md5\x01\xa7\xc9\x18\x9b\xc3E\x18\xf2\x82\xfd\xf3$\x9d_\v\nem\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00B"},
+	{"132f7619d33b523b1d9e5bd8e0928355", "How can you write a big system without C++?  -Paul Glick", "md5\x01gE#\x01\xefͫ\x89\x98\xba\xdc\xfe\x102TvHow can you write a big syst\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"},
+}
+
+func TestGolden(t *testing.T) {
+	for i := 0; i < len(golden); i++ {
+		g := golden[i]
+		s := fmt.Sprintf("%x", Sum([]byte(g.in)))
+		if s != g.out {
+			t.Fatalf("Sum function: md5(%s) = %s want %s", g.in, s, g.out)
+		}
+		c := New()
+		buf := make([]byte, len(g.in)+4)
+		for j := 0; j < 3+4; j++ {
+			if j < 2 {
+				io.WriteString(c, g.in)
+			} else if j == 2 {
+				io.WriteString(c, g.in[0:len(g.in)/2])
+				c.Sum(nil)
+				io.WriteString(c, g.in[len(g.in)/2:])
+			} else if j > 2 {
+				// test unaligned write
+				buf = buf[1:]
+				copy(buf, g.in)
+				c.Write(buf[:len(g.in)])
+			}
+			s := fmt.Sprintf("%x", c.Sum(nil))
+			if s != g.out {
+				t.Fatalf("md5[%d](%s) = %s want %s", j, g.in, s, g.out)
+			}
+			c.Reset()
+		}
+	}
+}
+
+func TestGoldenMarshal(t *testing.T) {
+	for _, g := range golden {
+		h := New()
+		h2 := New()
+
+		io.WriteString(h, g.in[:len(g.in)/2])
+
+		state, err := h.(encoding.BinaryMarshaler).MarshalBinary()
+		if err != nil {
+			t.Errorf("could not marshal: %v", err)
+			continue
+		}
+
+		if string(state) != g.halfState {
+			t.Errorf("md5(%q) state = %q, want %q", g.in, state, g.halfState)
+			continue
+		}
+
+		if err := h2.(encoding.BinaryUnmarshaler).UnmarshalBinary(state); err != nil {
+			t.Errorf("could not unmarshal: %v", err)
+			continue
+		}
+
+		io.WriteString(h, g.in[len(g.in)/2:])
+		io.WriteString(h2, g.in[len(g.in)/2:])
+
+		if actual, actual2 := h.Sum(nil), h2.Sum(nil); !bytes.Equal(actual, actual2) {
+			t.Errorf("md5(%q) = 0x%x != marshaled 0x%x", g.in, actual, actual2)
+		}
+	}
+}
+
+func TestLarge(t *testing.T) {
+	const N = 10000
+	ok := "2bb571599a4180e1d542f76904adc3df" // md5sum of "0123456789" * 1000
+	block := make([]byte, 10004)
+	c := New()
+	for offset := 0; offset < 4; offset++ {
+		for i := 0; i < N; i++ {
+			block[offset+i] = '0' + byte(i%10)
+		}
+		for blockSize := 10; blockSize <= N; blockSize *= 10 {
+			blocks := N / blockSize
+			b := block[offset : offset+blockSize]
+			c.Reset()
+			for i := 0; i < blocks; i++ {
+				c.Write(b)
+			}
+			s := fmt.Sprintf("%x", c.Sum(nil))
+			if s != ok {
+				t.Fatalf("md5 TestLarge offset=%d, blockSize=%d = %s want %s", offset, blockSize, s, ok)
+			}
+		}
+	}
+}
+
+// Tests that blockGeneric (pure Go) and block (in assembly for amd64, 386, arm) match.
+func TestBlockGeneric(t *testing.T) {
+	gen, asm := New().(*digest), New().(*digest)
+	buf := make([]byte, BlockSize*20) // arbitrary factor
+	rand.Read(buf)
+	blockGeneric(gen, buf)
+	block(asm, buf)
+	if *gen != *asm {
+		t.Error("block and blockGeneric resulted in different states")
+	}
+}
+
+// Tests for unmarshaling hashes that have hashed a large amount of data
+// The initial hash generation is omitted from the test, because it takes a long time.
+// The test contains some already-generated states, and their expected sums
+// Tests a problem that is outlined in GitHub issue #29541
+// The problem is triggered when an amount of data has been hashed for which
+// the data length has a 1 in the 32nd bit. When casted to int, this changes
+// the sign of the value, and causes the modulus operation to return a
+// different result.
+type unmarshalTest struct {
+	state string
+	sum   string
+}
+
+var largeUnmarshalTests = []unmarshalTest{
+	// Data length: 7_102_415_735
+	{
+		state: "md5\x01\xa5\xf7\xf0=\xd6S\x85\xd9M\n}\xc3\u0601\x89\xe7@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuv\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xa7VCw",
+		sum:   "cddefcf74ffec709a0b45a6a987564d5",
+	},
+	// Data length: 6_565_544_823
+	{
+		state: "md5\x01{\xda\x1a\xc7\xc9'?\x83EX\xe0\x88q\xfeG\x18@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuv\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x87VCw",
+		sum:   "fd9f41874ab240698e7bc9c3ae70c8e4",
+	},
+}
+
+func safeSum(h hash.Hash) (sum []byte, err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			err = fmt.Errorf("sum panic: %v", r)
+		}
+	}()
+
+	return h.Sum(nil), nil
+}
+
+func TestLargeHashes(t *testing.T) {
+	for i, test := range largeUnmarshalTests {
+
+		h := New()
+		if err := h.(encoding.BinaryUnmarshaler).UnmarshalBinary([]byte(test.state)); err != nil {
+			t.Errorf("test %d could not unmarshal: %v", i, err)
+			continue
+		}
+
+		sum, err := safeSum(h)
+		if err != nil {
+			t.Errorf("test %d could not sum: %v", i, err)
+			continue
+		}
+
+		if fmt.Sprintf("%x", sum) != test.sum {
+			t.Errorf("test %d sum mismatch: expect %s got %x", i, test.sum, sum)
+		}
+	}
+}
+
+func TestAllocations(t *testing.T) {
+	in := []byte("hello, world!")
+	out := make([]byte, 0, Size)
+	h := New()
+	n := int(testing.AllocsPerRun(10, func() {
+		h.Reset()
+		h.Write(in)
+		out = h.Sum(out[:0])
+	}))
+	if n > 0 {
+		t.Errorf("allocs = %d, want 0", n)
+	}
+}
+
+var bench = New()
+var buf = make([]byte, 1024*1024*8+1)
+var sum = make([]byte, bench.Size())
+
+func benchmarkSize(b *testing.B, size int, unaligned bool) {
+	b.SetBytes(int64(size))
+	buf := buf
+	if unaligned {
+		if uintptr(unsafe.Pointer(&buf[0]))&(unsafe.Alignof(uint32(0))-1) == 0 {
+			buf = buf[1:]
+		}
+	}
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		bench.Reset()
+		bench.Write(buf[:size])
+		bench.Sum(sum[:0])
+	}
+}
+
+func BenchmarkHash8Bytes(b *testing.B) {
+	benchmarkSize(b, 8, false)
+}
+
+func BenchmarkHash64(b *testing.B) {
+	benchmarkSize(b, 64, false)
+}
+
+func BenchmarkHash128(b *testing.B) {
+	benchmarkSize(b, 128, false)
+}
+
+func BenchmarkHash256(b *testing.B) {
+	benchmarkSize(b, 256, false)
+}
+
+func BenchmarkHash512(b *testing.B) {
+	benchmarkSize(b, 512, false)
+}
+
+func BenchmarkHash1K(b *testing.B) {
+	benchmarkSize(b, 1024, false)
+}
+
+func BenchmarkHash8K(b *testing.B) {
+	benchmarkSize(b, 8192, false)
+}
+
+func BenchmarkHash1M(b *testing.B) {
+	benchmarkSize(b, 1024*1024, false)
+}
+
+func BenchmarkHash8M(b *testing.B) {
+	benchmarkSize(b, 8*1024*1024, false)
+}
+
+func BenchmarkHash8BytesUnaligned(b *testing.B) {
+	benchmarkSize(b, 8, true)
+}
+
+func BenchmarkHash1KUnaligned(b *testing.B) {
+	benchmarkSize(b, 1024, true)
+}
+
+func BenchmarkHash8KUnaligned(b *testing.B) {
+	benchmarkSize(b, 8192, true)
+}
diff --git a/src/crypto/md5/md5block.go b/src/crypto/md5/md5block.go
new file mode 100644
index 0000000..4ff289e
--- /dev/null
+++ b/src/crypto/md5/md5block.go
@@ -0,0 +1,125 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Code generated by go run gen.go -output md5block.go; DO NOT EDIT.
+
+package md5
+
+import (
+	"encoding/binary"
+	"math/bits"
+)
+
+func blockGeneric(dig *digest, p []byte) {
+	// load state
+	a, b, c, d := dig.s[0], dig.s[1], dig.s[2], dig.s[3]
+
+	for i := 0; i <= len(p)-BlockSize; i += BlockSize {
+		// eliminate bounds checks on p
+		q := p[i:]
+		q = q[:BlockSize:BlockSize]
+
+		// save current state
+		aa, bb, cc, dd := a, b, c, d
+
+		// load input block
+		x0 := binary.LittleEndian.Uint32(q[4*0x0:])
+		x1 := binary.LittleEndian.Uint32(q[4*0x1:])
+		x2 := binary.LittleEndian.Uint32(q[4*0x2:])
+		x3 := binary.LittleEndian.Uint32(q[4*0x3:])
+		x4 := binary.LittleEndian.Uint32(q[4*0x4:])
+		x5 := binary.LittleEndian.Uint32(q[4*0x5:])
+		x6 := binary.LittleEndian.Uint32(q[4*0x6:])
+		x7 := binary.LittleEndian.Uint32(q[4*0x7:])
+		x8 := binary.LittleEndian.Uint32(q[4*0x8:])
+		x9 := binary.LittleEndian.Uint32(q[4*0x9:])
+		xa := binary.LittleEndian.Uint32(q[4*0xa:])
+		xb := binary.LittleEndian.Uint32(q[4*0xb:])
+		xc := binary.LittleEndian.Uint32(q[4*0xc:])
+		xd := binary.LittleEndian.Uint32(q[4*0xd:])
+		xe := binary.LittleEndian.Uint32(q[4*0xe:])
+		xf := binary.LittleEndian.Uint32(q[4*0xf:])
+
+		// round 1
+		a = b + bits.RotateLeft32((((c^d)&b)^d)+a+x0+0xd76aa478, 7)
+		d = a + bits.RotateLeft32((((b^c)&a)^c)+d+x1+0xe8c7b756, 12)
+		c = d + bits.RotateLeft32((((a^b)&d)^b)+c+x2+0x242070db, 17)
+		b = c + bits.RotateLeft32((((d^a)&c)^a)+b+x3+0xc1bdceee, 22)
+		a = b + bits.RotateLeft32((((c^d)&b)^d)+a+x4+0xf57c0faf, 7)
+		d = a + bits.RotateLeft32((((b^c)&a)^c)+d+x5+0x4787c62a, 12)
+		c = d + bits.RotateLeft32((((a^b)&d)^b)+c+x6+0xa8304613, 17)
+		b = c + bits.RotateLeft32((((d^a)&c)^a)+b+x7+0xfd469501, 22)
+		a = b + bits.RotateLeft32((((c^d)&b)^d)+a+x8+0x698098d8, 7)
+		d = a + bits.RotateLeft32((((b^c)&a)^c)+d+x9+0x8b44f7af, 12)
+		c = d + bits.RotateLeft32((((a^b)&d)^b)+c+xa+0xffff5bb1, 17)
+		b = c + bits.RotateLeft32((((d^a)&c)^a)+b+xb+0x895cd7be, 22)
+		a = b + bits.RotateLeft32((((c^d)&b)^d)+a+xc+0x6b901122, 7)
+		d = a + bits.RotateLeft32((((b^c)&a)^c)+d+xd+0xfd987193, 12)
+		c = d + bits.RotateLeft32((((a^b)&d)^b)+c+xe+0xa679438e, 17)
+		b = c + bits.RotateLeft32((((d^a)&c)^a)+b+xf+0x49b40821, 22)
+
+		// round 2
+		a = b + bits.RotateLeft32((((b^c)&d)^c)+a+x1+0xf61e2562, 5)
+		d = a + bits.RotateLeft32((((a^b)&c)^b)+d+x6+0xc040b340, 9)
+		c = d + bits.RotateLeft32((((d^a)&b)^a)+c+xb+0x265e5a51, 14)
+		b = c + bits.RotateLeft32((((c^d)&a)^d)+b+x0+0xe9b6c7aa, 20)
+		a = b + bits.RotateLeft32((((b^c)&d)^c)+a+x5+0xd62f105d, 5)
+		d = a + bits.RotateLeft32((((a^b)&c)^b)+d+xa+0x02441453, 9)
+		c = d + bits.RotateLeft32((((d^a)&b)^a)+c+xf+0xd8a1e681, 14)
+		b = c + bits.RotateLeft32((((c^d)&a)^d)+b+x4+0xe7d3fbc8, 20)
+		a = b + bits.RotateLeft32((((b^c)&d)^c)+a+x9+0x21e1cde6, 5)
+		d = a + bits.RotateLeft32((((a^b)&c)^b)+d+xe+0xc33707d6, 9)
+		c = d + bits.RotateLeft32((((d^a)&b)^a)+c+x3+0xf4d50d87, 14)
+		b = c + bits.RotateLeft32((((c^d)&a)^d)+b+x8+0x455a14ed, 20)
+		a = b + bits.RotateLeft32((((b^c)&d)^c)+a+xd+0xa9e3e905, 5)
+		d = a + bits.RotateLeft32((((a^b)&c)^b)+d+x2+0xfcefa3f8, 9)
+		c = d + bits.RotateLeft32((((d^a)&b)^a)+c+x7+0x676f02d9, 14)
+		b = c + bits.RotateLeft32((((c^d)&a)^d)+b+xc+0x8d2a4c8a, 20)
+
+		// round 3
+		a = b + bits.RotateLeft32((b^c^d)+a+x5+0xfffa3942, 4)
+		d = a + bits.RotateLeft32((a^b^c)+d+x8+0x8771f681, 11)
+		c = d + bits.RotateLeft32((d^a^b)+c+xb+0x6d9d6122, 16)
+		b = c + bits.RotateLeft32((c^d^a)+b+xe+0xfde5380c, 23)
+		a = b + bits.RotateLeft32((b^c^d)+a+x1+0xa4beea44, 4)
+		d = a + bits.RotateLeft32((a^b^c)+d+x4+0x4bdecfa9, 11)
+		c = d + bits.RotateLeft32((d^a^b)+c+x7+0xf6bb4b60, 16)
+		b = c + bits.RotateLeft32((c^d^a)+b+xa+0xbebfbc70, 23)
+		a = b + bits.RotateLeft32((b^c^d)+a+xd+0x289b7ec6, 4)
+		d = a + bits.RotateLeft32((a^b^c)+d+x0+0xeaa127fa, 11)
+		c = d + bits.RotateLeft32((d^a^b)+c+x3+0xd4ef3085, 16)
+		b = c + bits.RotateLeft32((c^d^a)+b+x6+0x04881d05, 23)
+		a = b + bits.RotateLeft32((b^c^d)+a+x9+0xd9d4d039, 4)
+		d = a + bits.RotateLeft32((a^b^c)+d+xc+0xe6db99e5, 11)
+		c = d + bits.RotateLeft32((d^a^b)+c+xf+0x1fa27cf8, 16)
+		b = c + bits.RotateLeft32((c^d^a)+b+x2+0xc4ac5665, 23)
+
+		// round 4
+		a = b + bits.RotateLeft32((c^(b|^d))+a+x0+0xf4292244, 6)
+		d = a + bits.RotateLeft32((b^(a|^c))+d+x7+0x432aff97, 10)
+		c = d + bits.RotateLeft32((a^(d|^b))+c+xe+0xab9423a7, 15)
+		b = c + bits.RotateLeft32((d^(c|^a))+b+x5+0xfc93a039, 21)
+		a = b + bits.RotateLeft32((c^(b|^d))+a+xc+0x655b59c3, 6)
+		d = a + bits.RotateLeft32((b^(a|^c))+d+x3+0x8f0ccc92, 10)
+		c = d + bits.RotateLeft32((a^(d|^b))+c+xa+0xffeff47d, 15)
+		b = c + bits.RotateLeft32((d^(c|^a))+b+x1+0x85845dd1, 21)
+		a = b + bits.RotateLeft32((c^(b|^d))+a+x8+0x6fa87e4f, 6)
+		d = a + bits.RotateLeft32((b^(a|^c))+d+xf+0xfe2ce6e0, 10)
+		c = d + bits.RotateLeft32((a^(d|^b))+c+x6+0xa3014314, 15)
+		b = c + bits.RotateLeft32((d^(c|^a))+b+xd+0x4e0811a1, 21)
+		a = b + bits.RotateLeft32((c^(b|^d))+a+x4+0xf7537e82, 6)
+		d = a + bits.RotateLeft32((b^(a|^c))+d+xb+0xbd3af235, 10)
+		c = d + bits.RotateLeft32((a^(d|^b))+c+x2+0x2ad7d2bb, 15)
+		b = c + bits.RotateLeft32((d^(c|^a))+b+x9+0xeb86d391, 21)
+
+		// add saved state
+		a += aa
+		b += bb
+		c += cc
+		d += dd
+	}
+
+	// save state
+	dig.s[0], dig.s[1], dig.s[2], dig.s[3] = a, b, c, d
+}
diff --git a/src/crypto/md5/md5block_386.s b/src/crypto/md5/md5block_386.s
new file mode 100644
index 0000000..30d4209
--- /dev/null
+++ b/src/crypto/md5/md5block_386.s
@@ -0,0 +1,182 @@
+// Original source:
+//	http://www.zorinaq.com/papers/md5-amd64.html
+//	http://www.zorinaq.com/papers/md5-amd64.tar.bz2
+//
+// Translated from Perl generating GNU assembly into
+// #defines generating 8a assembly, and adjusted for 386,
+// by the Go Authors.
+
+#include "textflag.h"
+
+// MD5 optimized for AMD64.
+//
+// Author: Marc Bevand <bevand_m (at) epita.fr>
+// Licence: I hereby disclaim the copyright on this code and place it
+// in the public domain.
+
+#define ROUND1(a, b, c, d, index, const, shift) \
+	XORL	c, BP; \
+	LEAL	const(a)(DI*1), a; \
+	ANDL	b, BP; \
+	XORL d, BP; \
+	MOVL (index*4)(SI), DI; \
+	ADDL BP, a; \
+	ROLL $shift, a; \
+	MOVL c, BP; \
+	ADDL b, a
+
+#define ROUND2(a, b, c, d, index, const, shift) \
+	LEAL	const(a)(DI*1),a; \
+	MOVL	d,		DI; \
+	ANDL	b,		DI; \
+	MOVL	d,		BP; \
+	NOTL	BP; \
+	ANDL	c,		BP; \
+	ORL	DI,		BP; \
+	MOVL	(index*4)(SI),DI; \
+	ADDL	BP,		a; \
+	ROLL	$shift,	a; \
+	ADDL	b,		a
+
+#define ROUND3(a, b, c, d, index, const, shift) \
+	LEAL	const(a)(DI*1),a; \
+	MOVL	(index*4)(SI),DI; \
+	XORL	d,		BP; \
+	XORL	b,		BP; \
+	ADDL	BP,		a; \
+	ROLL	$shift,		a; \
+	MOVL	b,		BP; \
+	ADDL	b,		a
+
+#define ROUND4(a, b, c, d, index, const, shift) \
+	LEAL	const(a)(DI*1),a; \
+	ORL	b,		BP; \
+	XORL	c,		BP; \
+	ADDL	BP,		a; \
+	MOVL	(index*4)(SI),DI; \
+	MOVL	$0xffffffff,	BP; \
+	ROLL	$shift,		a; \
+	XORL	c,		BP; \
+	ADDL	b,		a
+
+TEXT	·block(SB),NOSPLIT,$24-16
+	MOVL	dig+0(FP),	BP
+	MOVL	p+4(FP),	SI
+	MOVL	p_len+8(FP), DX
+	SHRL	$6,		DX
+	SHLL	$6,		DX
+
+	LEAL	(SI)(DX*1),	DI
+	MOVL	(0*4)(BP),	AX
+	MOVL	(1*4)(BP),	BX
+	MOVL	(2*4)(BP),	CX
+	MOVL	(3*4)(BP),	DX
+
+	CMPL	SI,		DI
+	JEQ	end
+
+	MOVL	DI,		16(SP)
+
+loop:
+	MOVL	AX,		0(SP)
+	MOVL	BX,		4(SP)
+	MOVL	CX,		8(SP)
+	MOVL	DX,		12(SP)
+
+	MOVL	(0*4)(SI),	DI
+	MOVL	DX,		BP
+
+	ROUND1(AX,BX,CX,DX, 1,0xd76aa478, 7);
+	ROUND1(DX,AX,BX,CX, 2,0xe8c7b756,12);
+	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
+	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
+	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
+	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
+	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
+	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
+	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
+	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);
+	ROUND1(CX,DX,AX,BX,11,0xffff5bb1,17);
+	ROUND1(BX,CX,DX,AX,12,0x895cd7be,22);
+	ROUND1(AX,BX,CX,DX,13,0x6b901122, 7);
+	ROUND1(DX,AX,BX,CX,14,0xfd987193,12);
+	ROUND1(CX,DX,AX,BX,15,0xa679438e,17);
+	ROUND1(BX,CX,DX,AX, 0,0x49b40821,22);
+
+	MOVL	(1*4)(SI),	DI
+	MOVL	DX,		BP
+
+	ROUND2(AX,BX,CX,DX, 6,0xf61e2562, 5);
+	ROUND2(DX,AX,BX,CX,11,0xc040b340, 9);
+	ROUND2(CX,DX,AX,BX, 0,0x265e5a51,14);
+	ROUND2(BX,CX,DX,AX, 5,0xe9b6c7aa,20);
+	ROUND2(AX,BX,CX,DX,10,0xd62f105d, 5);
+	ROUND2(DX,AX,BX,CX,15, 0x2441453, 9);
+	ROUND2(CX,DX,AX,BX, 4,0xd8a1e681,14);
+	ROUND2(BX,CX,DX,AX, 9,0xe7d3fbc8,20);
+	ROUND2(AX,BX,CX,DX,14,0x21e1cde6, 5);
+	ROUND2(DX,AX,BX,CX, 3,0xc33707d6, 9);
+	ROUND2(CX,DX,AX,BX, 8,0xf4d50d87,14);
+	ROUND2(BX,CX,DX,AX,13,0x455a14ed,20);
+	ROUND2(AX,BX,CX,DX, 2,0xa9e3e905, 5);
+	ROUND2(DX,AX,BX,CX, 7,0xfcefa3f8, 9);
+	ROUND2(CX,DX,AX,BX,12,0x676f02d9,14);
+	ROUND2(BX,CX,DX,AX, 0,0x8d2a4c8a,20);
+
+	MOVL	(5*4)(SI),	DI
+	MOVL	CX,		BP
+
+	ROUND3(AX,BX,CX,DX, 8,0xfffa3942, 4);
+	ROUND3(DX,AX,BX,CX,11,0x8771f681,11);
+	ROUND3(CX,DX,AX,BX,14,0x6d9d6122,16);
+	ROUND3(BX,CX,DX,AX, 1,0xfde5380c,23);
+	ROUND3(AX,BX,CX,DX, 4,0xa4beea44, 4);
+	ROUND3(DX,AX,BX,CX, 7,0x4bdecfa9,11);
+	ROUND3(CX,DX,AX,BX,10,0xf6bb4b60,16);
+	ROUND3(BX,CX,DX,AX,13,0xbebfbc70,23);
+	ROUND3(AX,BX,CX,DX, 0,0x289b7ec6, 4);
+	ROUND3(DX,AX,BX,CX, 3,0xeaa127fa,11);
+	ROUND3(CX,DX,AX,BX, 6,0xd4ef3085,16);
+	ROUND3(BX,CX,DX,AX, 9, 0x4881d05,23);
+	ROUND3(AX,BX,CX,DX,12,0xd9d4d039, 4);
+	ROUND3(DX,AX,BX,CX,15,0xe6db99e5,11);
+	ROUND3(CX,DX,AX,BX, 2,0x1fa27cf8,16);
+	ROUND3(BX,CX,DX,AX, 0,0xc4ac5665,23);
+
+	MOVL	(0*4)(SI),	DI
+	MOVL	$0xffffffff,	BP
+	XORL	DX,		BP
+
+	ROUND4(AX,BX,CX,DX, 7,0xf4292244, 6);
+	ROUND4(DX,AX,BX,CX,14,0x432aff97,10);
+	ROUND4(CX,DX,AX,BX, 5,0xab9423a7,15);
+	ROUND4(BX,CX,DX,AX,12,0xfc93a039,21);
+	ROUND4(AX,BX,CX,DX, 3,0x655b59c3, 6);
+	ROUND4(DX,AX,BX,CX,10,0x8f0ccc92,10);
+	ROUND4(CX,DX,AX,BX, 1,0xffeff47d,15);
+	ROUND4(BX,CX,DX,AX, 8,0x85845dd1,21);
+	ROUND4(AX,BX,CX,DX,15,0x6fa87e4f, 6);
+	ROUND4(DX,AX,BX,CX, 6,0xfe2ce6e0,10);
+	ROUND4(CX,DX,AX,BX,13,0xa3014314,15);
+	ROUND4(BX,CX,DX,AX, 4,0x4e0811a1,21);
+	ROUND4(AX,BX,CX,DX,11,0xf7537e82, 6);
+	ROUND4(DX,AX,BX,CX, 2,0xbd3af235,10);
+	ROUND4(CX,DX,AX,BX, 9,0x2ad7d2bb,15);
+	ROUND4(BX,CX,DX,AX, 0,0xeb86d391,21);
+
+	ADDL	0(SP),	AX
+	ADDL	4(SP),	BX
+	ADDL	8(SP),	CX
+	ADDL	12(SP),	DX
+
+	ADDL	$64,		SI
+	CMPL	SI,		16(SP)
+	JB	loop
+
+end:
+	MOVL	dig+0(FP),	BP
+	MOVL	AX,		(0*4)(BP)
+	MOVL	BX,		(1*4)(BP)
+	MOVL	CX,		(2*4)(BP)
+	MOVL	DX,		(3*4)(BP)
+	RET
diff --git a/src/crypto/md5/md5block_amd64.s b/src/crypto/md5/md5block_amd64.s
new file mode 100644
index 0000000..7c7d92d
--- /dev/null
+++ b/src/crypto/md5/md5block_amd64.s
@@ -0,0 +1,179 @@
+// Original source:
+//	http://www.zorinaq.com/papers/md5-amd64.html
+//	http://www.zorinaq.com/papers/md5-amd64.tar.bz2
+//
+// Translated from Perl generating GNU assembly into
+// #defines generating 6a assembly by the Go Authors.
+
+#include "textflag.h"
+
+// MD5 optimized for AMD64.
+//
+// Author: Marc Bevand <bevand_m (at) epita.fr>
+// Licence: I hereby disclaim the copyright on this code and place it
+// in the public domain.
+
+TEXT	·block(SB),NOSPLIT,$8-32
+	MOVQ	dig+0(FP),	BP
+	MOVQ	p+8(FP),	SI
+	MOVQ	p_len+16(FP), DX
+	SHRQ	$6,		DX
+	SHLQ	$6,		DX
+
+	LEAQ	(SI)(DX*1),	DI
+	MOVL	(0*4)(BP),	AX
+	MOVL	(1*4)(BP),	BX
+	MOVL	(2*4)(BP),	CX
+	MOVL	(3*4)(BP),	DX
+
+	CMPQ	SI,		DI
+	JEQ	end
+
+loop:
+	MOVL	AX,		R12
+	MOVL	BX,		R13
+	MOVL	CX,		R14
+	MOVL	DX,		R15
+
+	MOVL	(0*4)(SI),	R8
+	MOVL	DX,		R9
+
+#define ROUND1(a, b, c, d, index, const, shift) \
+	XORL	c, R9; \
+	LEAL	const(a)(R8*1), a; \
+	ANDL	b, R9; \
+	XORL d, R9; \
+	MOVL (index*4)(SI), R8; \
+	ADDL R9, a; \
+	ROLL $shift, a; \
+	MOVL c, R9; \
+	ADDL b, a
+
+	ROUND1(AX,BX,CX,DX, 1,0xd76aa478, 7);
+	ROUND1(DX,AX,BX,CX, 2,0xe8c7b756,12);
+	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
+	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
+	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
+	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
+	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
+	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
+	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
+	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);
+	ROUND1(CX,DX,AX,BX,11,0xffff5bb1,17);
+	ROUND1(BX,CX,DX,AX,12,0x895cd7be,22);
+	ROUND1(AX,BX,CX,DX,13,0x6b901122, 7);
+	ROUND1(DX,AX,BX,CX,14,0xfd987193,12);
+	ROUND1(CX,DX,AX,BX,15,0xa679438e,17);
+	ROUND1(BX,CX,DX,AX, 0,0x49b40821,22);
+
+	MOVL	(1*4)(SI),	R8
+	MOVL	DX,		R9
+	MOVL	DX,		R10
+
+#define ROUND2(a, b, c, d, index, const, shift) \
+	NOTL	R9; \
+	LEAL	const(a)(R8*1),a; \
+	ANDL	b,		R10; \
+	ANDL	c,		R9; \
+	MOVL	(index*4)(SI),R8; \
+	ORL	R9,		R10; \
+	MOVL	c,		R9; \
+	ADDL	R10,		a; \
+	MOVL	c,		R10; \
+	ROLL	$shift,	a; \
+	ADDL	b,		a
+
+	ROUND2(AX,BX,CX,DX, 6,0xf61e2562, 5);
+	ROUND2(DX,AX,BX,CX,11,0xc040b340, 9);
+	ROUND2(CX,DX,AX,BX, 0,0x265e5a51,14);
+	ROUND2(BX,CX,DX,AX, 5,0xe9b6c7aa,20);
+	ROUND2(AX,BX,CX,DX,10,0xd62f105d, 5);
+	ROUND2(DX,AX,BX,CX,15, 0x2441453, 9);
+	ROUND2(CX,DX,AX,BX, 4,0xd8a1e681,14);
+	ROUND2(BX,CX,DX,AX, 9,0xe7d3fbc8,20);
+	ROUND2(AX,BX,CX,DX,14,0x21e1cde6, 5);
+	ROUND2(DX,AX,BX,CX, 3,0xc33707d6, 9);
+	ROUND2(CX,DX,AX,BX, 8,0xf4d50d87,14);
+	ROUND2(BX,CX,DX,AX,13,0x455a14ed,20);
+	ROUND2(AX,BX,CX,DX, 2,0xa9e3e905, 5);
+	ROUND2(DX,AX,BX,CX, 7,0xfcefa3f8, 9);
+	ROUND2(CX,DX,AX,BX,12,0x676f02d9,14);
+	ROUND2(BX,CX,DX,AX, 0,0x8d2a4c8a,20);
+
+	MOVL	(5*4)(SI),	R8
+	MOVL	CX,		R9
+
+#define ROUND3(a, b, c, d, index, const, shift) \
+	LEAL	const(a)(R8*1),a; \
+	MOVL	(index*4)(SI),R8; \
+	XORL	d,		R9; \
+	XORL	b,		R9; \
+	ADDL	R9,		a; \
+	ROLL	$shift,		a; \
+	MOVL	b,		R9; \
+	ADDL	b,		a
+
+	ROUND3(AX,BX,CX,DX, 8,0xfffa3942, 4);
+	ROUND3(DX,AX,BX,CX,11,0x8771f681,11);
+	ROUND3(CX,DX,AX,BX,14,0x6d9d6122,16);
+	ROUND3(BX,CX,DX,AX, 1,0xfde5380c,23);
+	ROUND3(AX,BX,CX,DX, 4,0xa4beea44, 4);
+	ROUND3(DX,AX,BX,CX, 7,0x4bdecfa9,11);
+	ROUND3(CX,DX,AX,BX,10,0xf6bb4b60,16);
+	ROUND3(BX,CX,DX,AX,13,0xbebfbc70,23);
+	ROUND3(AX,BX,CX,DX, 0,0x289b7ec6, 4);
+	ROUND3(DX,AX,BX,CX, 3,0xeaa127fa,11);
+	ROUND3(CX,DX,AX,BX, 6,0xd4ef3085,16);
+	ROUND3(BX,CX,DX,AX, 9, 0x4881d05,23);
+	ROUND3(AX,BX,CX,DX,12,0xd9d4d039, 4);
+	ROUND3(DX,AX,BX,CX,15,0xe6db99e5,11);
+	ROUND3(CX,DX,AX,BX, 2,0x1fa27cf8,16);
+	ROUND3(BX,CX,DX,AX, 0,0xc4ac5665,23);
+
+	MOVL	(0*4)(SI),	R8
+	MOVL	$0xffffffff,	R9
+	XORL	DX,		R9
+
+#define ROUND4(a, b, c, d, index, const, shift) \
+	LEAL	const(a)(R8*1),a; \
+	ORL	b,		R9; \
+	XORL	c,		R9; \
+	ADDL	R9,		a; \
+	MOVL	(index*4)(SI),R8; \
+	MOVL	$0xffffffff,	R9; \
+	ROLL	$shift,		a; \
+	XORL	c,		R9; \
+	ADDL	b,		a
+
+	ROUND4(AX,BX,CX,DX, 7,0xf4292244, 6);
+	ROUND4(DX,AX,BX,CX,14,0x432aff97,10);
+	ROUND4(CX,DX,AX,BX, 5,0xab9423a7,15);
+	ROUND4(BX,CX,DX,AX,12,0xfc93a039,21);
+	ROUND4(AX,BX,CX,DX, 3,0x655b59c3, 6);
+	ROUND4(DX,AX,BX,CX,10,0x8f0ccc92,10);
+	ROUND4(CX,DX,AX,BX, 1,0xffeff47d,15);
+	ROUND4(BX,CX,DX,AX, 8,0x85845dd1,21);
+	ROUND4(AX,BX,CX,DX,15,0x6fa87e4f, 6);
+	ROUND4(DX,AX,BX,CX, 6,0xfe2ce6e0,10);
+	ROUND4(CX,DX,AX,BX,13,0xa3014314,15);
+	ROUND4(BX,CX,DX,AX, 4,0x4e0811a1,21);
+	ROUND4(AX,BX,CX,DX,11,0xf7537e82, 6);
+	ROUND4(DX,AX,BX,CX, 2,0xbd3af235,10);
+	ROUND4(CX,DX,AX,BX, 9,0x2ad7d2bb,15);
+	ROUND4(BX,CX,DX,AX, 0,0xeb86d391,21);
+
+	ADDL	R12,	AX
+	ADDL	R13,	BX
+	ADDL	R14,	CX
+	ADDL	R15,	DX
+
+	ADDQ	$64,		SI
+	CMPQ	SI,		DI
+	JB	loop
+
+end:
+	MOVL	AX,		(0*4)(BP)
+	MOVL	BX,		(1*4)(BP)
+	MOVL	CX,		(2*4)(BP)
+	MOVL	DX,		(3*4)(BP)
+	RET
diff --git a/src/crypto/md5/md5block_arm.s b/src/crypto/md5/md5block_arm.s
new file mode 100644
index 0000000..54d02b7
--- /dev/null
+++ b/src/crypto/md5/md5block_arm.s
@@ -0,0 +1,299 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+//
+// ARM version of md5block.go
+
+#include "textflag.h"
+
+// Register definitions
+#define Rtable	R0	// Pointer to MD5 constants table
+#define Rdata	R1	// Pointer to data to hash
+#define Ra	R2	// MD5 accumulator
+#define Rb	R3	// MD5 accumulator
+#define Rc	R4	// MD5 accumulator
+#define Rd	R5	// MD5 accumulator
+#define Rc0	R6	// MD5 constant
+#define Rc1	R7	// MD5 constant
+#define Rc2	R8	// MD5 constant
+// r9, r10 are forbidden
+// r11 is OK provided you check the assembler that no synthetic instructions use it
+#define Rc3	R11	// MD5 constant
+#define Rt0	R12	// temporary
+#define Rt1	R14	// temporary
+
+// func block(dig *digest, p []byte)
+// 0(FP) is *digest
+// 4(FP) is p.array (struct Slice)
+// 8(FP) is p.len
+//12(FP) is p.cap
+//
+// Stack frame
+#define p_end	end-4(SP)	// pointer to the end of data
+#define p_data	data-8(SP)	// current data pointer
+#define buf	buffer-(8+4*16)(SP)	//16 words temporary buffer
+		// 3 words at 4..12(R13) for called routine parameters
+
+TEXT	·block(SB), NOSPLIT, $84-16
+	MOVW	p+4(FP), Rdata	// pointer to the data
+	MOVW	p_len+8(FP), Rt0	// number of bytes
+	ADD	Rdata, Rt0
+	MOVW	Rt0, p_end	// pointer to end of data
+
+loop:
+	MOVW	Rdata, p_data	// Save Rdata
+	AND.S	$3, Rdata, Rt0	// TST $3, Rdata not working see issue 5921
+	BEQ	aligned			// aligned detected - skip copy
+
+	// Copy the unaligned source data into the aligned temporary buffer
+	// memmove(to=4(R13), from=8(R13), n=12(R13)) - Corrupts all registers
+	MOVW	$buf, Rtable	// to
+	MOVW	$64, Rc0		// n
+	MOVM.IB	[Rtable,Rdata,Rc0], (R13)
+	BL	runtime·memmove(SB)
+
+	// Point to the local aligned copy of the data
+	MOVW	$buf, Rdata
+
+aligned:
+	// Point to the table of constants
+	// A PC relative add would be cheaper than this
+	MOVW	$·table(SB), Rtable
+
+	// Load up initial MD5 accumulator
+	MOVW	dig+0(FP), Rc0
+	MOVM.IA (Rc0), [Ra,Rb,Rc,Rd]
+
+// a += (((c^d)&b)^d) + X[index] + const
+// a = a<<shift | a>>(32-shift) + b
+#define ROUND1(Ra, Rb, Rc, Rd, index, shift, Rconst) \
+	EOR	Rc, Rd, Rt0		; \
+	AND	Rb, Rt0			; \
+	EOR	Rd, Rt0			; \
+	MOVW	(index<<2)(Rdata), Rt1	; \
+	ADD	Rt1, Rt0			; \
+	ADD	Rconst, Rt0			; \
+	ADD	Rt0, Ra			; \
+	ADD	Ra@>(32-shift), Rb, Ra	;
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND1(Ra, Rb, Rc, Rd,  0,	7, Rc0)
+	ROUND1(Rd, Ra, Rb, Rc,  1, 12, Rc1)
+	ROUND1(Rc, Rd, Ra, Rb,  2, 17, Rc2)
+	ROUND1(Rb, Rc, Rd, Ra,  3, 22, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND1(Ra, Rb, Rc, Rd,  4,	7, Rc0)
+	ROUND1(Rd, Ra, Rb, Rc,  5, 12, Rc1)
+	ROUND1(Rc, Rd, Ra, Rb,  6, 17, Rc2)
+	ROUND1(Rb, Rc, Rd, Ra,  7, 22, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND1(Ra, Rb, Rc, Rd,  8,	7, Rc0)
+	ROUND1(Rd, Ra, Rb, Rc,  9, 12, Rc1)
+	ROUND1(Rc, Rd, Ra, Rb, 10, 17, Rc2)
+	ROUND1(Rb, Rc, Rd, Ra, 11, 22, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND1(Ra, Rb, Rc, Rd, 12,	7, Rc0)
+	ROUND1(Rd, Ra, Rb, Rc, 13, 12, Rc1)
+	ROUND1(Rc, Rd, Ra, Rb, 14, 17, Rc2)
+	ROUND1(Rb, Rc, Rd, Ra, 15, 22, Rc3)
+
+// a += (((b^c)&d)^c) + X[index] + const
+// a = a<<shift | a>>(32-shift) + b
+#define ROUND2(Ra, Rb, Rc, Rd, index, shift, Rconst) \
+	EOR	Rb, Rc, Rt0		; \
+	AND	Rd, Rt0			; \
+	EOR	Rc, Rt0			; \
+	MOVW	(index<<2)(Rdata), Rt1	; \
+	ADD	Rt1, Rt0			; \
+	ADD	Rconst, Rt0			; \
+	ADD	Rt0, Ra			; \
+	ADD	Ra@>(32-shift), Rb, Ra	;
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND2(Ra, Rb, Rc, Rd,  1,	5, Rc0)
+	ROUND2(Rd, Ra, Rb, Rc,  6,	9, Rc1)
+	ROUND2(Rc, Rd, Ra, Rb, 11, 14, Rc2)
+	ROUND2(Rb, Rc, Rd, Ra,  0, 20, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND2(Ra, Rb, Rc, Rd,  5,	5, Rc0)
+	ROUND2(Rd, Ra, Rb, Rc, 10,	9, Rc1)
+	ROUND2(Rc, Rd, Ra, Rb, 15, 14, Rc2)
+	ROUND2(Rb, Rc, Rd, Ra,  4, 20, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND2(Ra, Rb, Rc, Rd,  9,	5, Rc0)
+	ROUND2(Rd, Ra, Rb, Rc, 14,	9, Rc1)
+	ROUND2(Rc, Rd, Ra, Rb,  3, 14, Rc2)
+	ROUND2(Rb, Rc, Rd, Ra,  8, 20, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND2(Ra, Rb, Rc, Rd, 13,	5, Rc0)
+	ROUND2(Rd, Ra, Rb, Rc,  2,	9, Rc1)
+	ROUND2(Rc, Rd, Ra, Rb,  7, 14, Rc2)
+	ROUND2(Rb, Rc, Rd, Ra, 12, 20, Rc3)
+
+// a += (b^c^d) + X[index] + const
+// a = a<<shift | a>>(32-shift) + b
+#define ROUND3(Ra, Rb, Rc, Rd, index, shift, Rconst) \
+	EOR	Rb, Rc, Rt0		; \
+	EOR	Rd, Rt0			; \
+	MOVW	(index<<2)(Rdata), Rt1	; \
+	ADD	Rt1, Rt0			; \
+	ADD	Rconst, Rt0			; \
+	ADD	Rt0, Ra			; \
+	ADD	Ra@>(32-shift), Rb, Ra	;
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND3(Ra, Rb, Rc, Rd,  5,	4, Rc0)
+	ROUND3(Rd, Ra, Rb, Rc,  8, 11, Rc1)
+	ROUND3(Rc, Rd, Ra, Rb, 11, 16, Rc2)
+	ROUND3(Rb, Rc, Rd, Ra, 14, 23, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND3(Ra, Rb, Rc, Rd,  1,	4, Rc0)
+	ROUND3(Rd, Ra, Rb, Rc,  4, 11, Rc1)
+	ROUND3(Rc, Rd, Ra, Rb,  7, 16, Rc2)
+	ROUND3(Rb, Rc, Rd, Ra, 10, 23, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND3(Ra, Rb, Rc, Rd, 13,	4, Rc0)
+	ROUND3(Rd, Ra, Rb, Rc,  0, 11, Rc1)
+	ROUND3(Rc, Rd, Ra, Rb,  3, 16, Rc2)
+	ROUND3(Rb, Rc, Rd, Ra,  6, 23, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND3(Ra, Rb, Rc, Rd,  9,	4, Rc0)
+	ROUND3(Rd, Ra, Rb, Rc, 12, 11, Rc1)
+	ROUND3(Rc, Rd, Ra, Rb, 15, 16, Rc2)
+	ROUND3(Rb, Rc, Rd, Ra,  2, 23, Rc3)
+
+// a += (c^(b|^d)) + X[index] + const
+// a = a<<shift | a>>(32-shift) + b
+#define ROUND4(Ra, Rb, Rc, Rd, index, shift, Rconst) \
+	MVN	Rd, Rt0			; \
+	ORR	Rb, Rt0			; \
+	EOR	Rc, Rt0			; \
+	MOVW	(index<<2)(Rdata), Rt1	; \
+	ADD	Rt1, Rt0			; \
+	ADD	Rconst, Rt0			; \
+	ADD	Rt0, Ra			; \
+	ADD	Ra@>(32-shift), Rb, Ra	;
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND4(Ra, Rb, Rc, Rd,  0,	6, Rc0)
+	ROUND4(Rd, Ra, Rb, Rc,  7, 10, Rc1)
+	ROUND4(Rc, Rd, Ra, Rb, 14, 15, Rc2)
+	ROUND4(Rb, Rc, Rd, Ra,  5, 21, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND4(Ra, Rb, Rc, Rd, 12,	6, Rc0)
+	ROUND4(Rd, Ra, Rb, Rc,  3, 10, Rc1)
+	ROUND4(Rc, Rd, Ra, Rb, 10, 15, Rc2)
+	ROUND4(Rb, Rc, Rd, Ra,  1, 21, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND4(Ra, Rb, Rc, Rd,  8,	6, Rc0)
+	ROUND4(Rd, Ra, Rb, Rc, 15, 10, Rc1)
+	ROUND4(Rc, Rd, Ra, Rb,  6, 15, Rc2)
+	ROUND4(Rb, Rc, Rd, Ra, 13, 21, Rc3)
+
+	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
+	ROUND4(Ra, Rb, Rc, Rd,  4,	6, Rc0)
+	ROUND4(Rd, Ra, Rb, Rc, 11, 10, Rc1)
+	ROUND4(Rc, Rd, Ra, Rb,  2, 15, Rc2)
+	ROUND4(Rb, Rc, Rd, Ra,  9, 21, Rc3)
+
+	MOVW	dig+0(FP), Rt0
+	MOVM.IA (Rt0), [Rc0,Rc1,Rc2,Rc3]
+
+	ADD	Rc0, Ra
+	ADD	Rc1, Rb
+	ADD	Rc2, Rc
+	ADD	Rc3, Rd
+
+	MOVM.IA [Ra,Rb,Rc,Rd], (Rt0)
+
+	MOVW	p_data, Rdata
+	MOVW	p_end, Rt0
+	ADD	$64, Rdata
+	CMP	Rt0, Rdata
+	BLO	loop
+
+	RET
+
+// MD5 constants table
+
+	// Round 1
+	DATA	·table+0x00(SB)/4, $0xd76aa478
+	DATA	·table+0x04(SB)/4, $0xe8c7b756
+	DATA	·table+0x08(SB)/4, $0x242070db
+	DATA	·table+0x0c(SB)/4, $0xc1bdceee
+	DATA	·table+0x10(SB)/4, $0xf57c0faf
+	DATA	·table+0x14(SB)/4, $0x4787c62a
+	DATA	·table+0x18(SB)/4, $0xa8304613
+	DATA	·table+0x1c(SB)/4, $0xfd469501
+	DATA	·table+0x20(SB)/4, $0x698098d8
+	DATA	·table+0x24(SB)/4, $0x8b44f7af
+	DATA	·table+0x28(SB)/4, $0xffff5bb1
+	DATA	·table+0x2c(SB)/4, $0x895cd7be
+	DATA	·table+0x30(SB)/4, $0x6b901122
+	DATA	·table+0x34(SB)/4, $0xfd987193
+	DATA	·table+0x38(SB)/4, $0xa679438e
+	DATA	·table+0x3c(SB)/4, $0x49b40821
+	// Round 2
+	DATA	·table+0x40(SB)/4, $0xf61e2562
+	DATA	·table+0x44(SB)/4, $0xc040b340
+	DATA	·table+0x48(SB)/4, $0x265e5a51
+	DATA	·table+0x4c(SB)/4, $0xe9b6c7aa
+	DATA	·table+0x50(SB)/4, $0xd62f105d
+	DATA	·table+0x54(SB)/4, $0x02441453
+	DATA	·table+0x58(SB)/4, $0xd8a1e681
+	DATA	·table+0x5c(SB)/4, $0xe7d3fbc8
+	DATA	·table+0x60(SB)/4, $0x21e1cde6
+	DATA	·table+0x64(SB)/4, $0xc33707d6
+	DATA	·table+0x68(SB)/4, $0xf4d50d87
+	DATA	·table+0x6c(SB)/4, $0x455a14ed
+	DATA	·table+0x70(SB)/4, $0xa9e3e905
+	DATA	·table+0x74(SB)/4, $0xfcefa3f8
+	DATA	·table+0x78(SB)/4, $0x676f02d9
+	DATA	·table+0x7c(SB)/4, $0x8d2a4c8a
+	// Round 3
+	DATA	·table+0x80(SB)/4, $0xfffa3942
+	DATA	·table+0x84(SB)/4, $0x8771f681
+	DATA	·table+0x88(SB)/4, $0x6d9d6122
+	DATA	·table+0x8c(SB)/4, $0xfde5380c
+	DATA	·table+0x90(SB)/4, $0xa4beea44
+	DATA	·table+0x94(SB)/4, $0x4bdecfa9
+	DATA	·table+0x98(SB)/4, $0xf6bb4b60
+	DATA	·table+0x9c(SB)/4, $0xbebfbc70
+	DATA	·table+0xa0(SB)/4, $0x289b7ec6
+	DATA	·table+0xa4(SB)/4, $0xeaa127fa
+	DATA	·table+0xa8(SB)/4, $0xd4ef3085
+	DATA	·table+0xac(SB)/4, $0x04881d05
+	DATA	·table+0xb0(SB)/4, $0xd9d4d039
+	DATA	·table+0xb4(SB)/4, $0xe6db99e5
+	DATA	·table+0xb8(SB)/4, $0x1fa27cf8
+	DATA	·table+0xbc(SB)/4, $0xc4ac5665
+	// Round 4
+	DATA	·table+0xc0(SB)/4, $0xf4292244
+	DATA	·table+0xc4(SB)/4, $0x432aff97
+	DATA	·table+0xc8(SB)/4, $0xab9423a7
+	DATA	·table+0xcc(SB)/4, $0xfc93a039
+	DATA	·table+0xd0(SB)/4, $0x655b59c3
+	DATA	·table+0xd4(SB)/4, $0x8f0ccc92
+	DATA	·table+0xd8(SB)/4, $0xffeff47d
+	DATA	·table+0xdc(SB)/4, $0x85845dd1
+	DATA	·table+0xe0(SB)/4, $0x6fa87e4f
+	DATA	·table+0xe4(SB)/4, $0xfe2ce6e0
+	DATA	·table+0xe8(SB)/4, $0xa3014314
+	DATA	·table+0xec(SB)/4, $0x4e0811a1
+	DATA	·table+0xf0(SB)/4, $0xf7537e82
+	DATA	·table+0xf4(SB)/4, $0xbd3af235
+	DATA	·table+0xf8(SB)/4, $0x2ad7d2bb
+	DATA	·table+0xfc(SB)/4, $0xeb86d391
+	// Global definition
+	GLOBL	·table(SB),8,$256
diff --git a/src/crypto/md5/md5block_arm64.s b/src/crypto/md5/md5block_arm64.s
new file mode 100644
index 0000000..39b9851
--- /dev/null
+++ b/src/crypto/md5/md5block_arm64.s
@@ -0,0 +1,167 @@
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+//
+// ARM64 version of md5block.go
+// derived from crypto/md5/md5block_amd64.s
+
+#include "textflag.h"
+
+TEXT	·block(SB),NOSPLIT,$0-32
+	MOVD	dig+0(FP), R0
+	MOVD	p+8(FP), R1
+	MOVD	p_len+16(FP), R2
+	AND	$~63, R2
+	CBZ	R2, zero
+
+	ADD	R1, R2, R21
+	LDPW	(0*8)(R0), (R4, R5)
+	LDPW	(1*8)(R0), (R6, R7)
+
+loop:
+	MOVW	R4, R12
+	MOVW	R5, R13
+	MOVW	R6, R14
+	MOVW	R7, R15
+
+	MOVW	(0*4)(R1), R8
+	MOVW	R7, R9
+
+#define ROUND1(a, b, c, d, index, const, shift) \
+	ADDW	$const, a; \
+	ADDW	R8, a; \
+	MOVW	(index*4)(R1), R8; \
+	EORW	c, R9; \
+	ANDW	b, R9; \
+	EORW	d, R9; \
+	ADDW	R9, a; \
+	RORW	$(32-shift), a; \
+	MOVW	c, R9; \
+	ADDW	b, a
+
+	ROUND1(R4,R5,R6,R7, 1,0xd76aa478, 7);
+	ROUND1(R7,R4,R5,R6, 2,0xe8c7b756,12);
+	ROUND1(R6,R7,R4,R5, 3,0x242070db,17);
+	ROUND1(R5,R6,R7,R4, 4,0xc1bdceee,22);
+	ROUND1(R4,R5,R6,R7, 5,0xf57c0faf, 7);
+	ROUND1(R7,R4,R5,R6, 6,0x4787c62a,12);
+	ROUND1(R6,R7,R4,R5, 7,0xa8304613,17);
+	ROUND1(R5,R6,R7,R4, 8,0xfd469501,22);
+	ROUND1(R4,R5,R6,R7, 9,0x698098d8, 7);
+	ROUND1(R7,R4,R5,R6,10,0x8b44f7af,12);
+	ROUND1(R6,R7,R4,R5,11,0xffff5bb1,17);
+	ROUND1(R5,R6,R7,R4,12,0x895cd7be,22);
+	ROUND1(R4,R5,R6,R7,13,0x6b901122, 7);
+	ROUND1(R7,R4,R5,R6,14,0xfd987193,12);
+	ROUND1(R6,R7,R4,R5,15,0xa679438e,17);
+	ROUND1(R5,R6,R7,R4, 0,0x49b40821,22);
+
+	MOVW	(1*4)(R1), R8
+	MOVW	R7, R9
+	MOVW	R7, R10
+
+#define ROUND2(a, b, c, d, index, const, shift) \
+	ADDW	$const, a; \
+	ADDW	R8, a; \
+	MOVW	(index*4)(R1), R8; \
+	ANDW	b, R10; \
+	BICW	R9, c, R9; \
+	ORRW	R9, R10; \
+	MOVW	c, R9; \
+	ADDW	R10, a; \
+	MOVW	c, R10; \
+	RORW	$(32-shift), a; \
+	ADDW	b, a
+
+	ROUND2(R4,R5,R6,R7, 6,0xf61e2562, 5);
+	ROUND2(R7,R4,R5,R6,11,0xc040b340, 9);
+	ROUND2(R6,R7,R4,R5, 0,0x265e5a51,14);
+	ROUND2(R5,R6,R7,R4, 5,0xe9b6c7aa,20);
+	ROUND2(R4,R5,R6,R7,10,0xd62f105d, 5);
+	ROUND2(R7,R4,R5,R6,15, 0x2441453, 9);
+	ROUND2(R6,R7,R4,R5, 4,0xd8a1e681,14);
+	ROUND2(R5,R6,R7,R4, 9,0xe7d3fbc8,20);
+	ROUND2(R4,R5,R6,R7,14,0x21e1cde6, 5);
+	ROUND2(R7,R4,R5,R6, 3,0xc33707d6, 9);
+	ROUND2(R6,R7,R4,R5, 8,0xf4d50d87,14);
+	ROUND2(R5,R6,R7,R4,13,0x455a14ed,20);
+	ROUND2(R4,R5,R6,R7, 2,0xa9e3e905, 5);
+	ROUND2(R7,R4,R5,R6, 7,0xfcefa3f8, 9);
+	ROUND2(R6,R7,R4,R5,12,0x676f02d9,14);
+	ROUND2(R5,R6,R7,R4, 0,0x8d2a4c8a,20);
+
+	MOVW	(5*4)(R1), R8
+	MOVW	R6, R9
+
+#define ROUND3(a, b, c, d, index, const, shift) \
+	ADDW	$const, a; \
+	ADDW	R8, a; \
+	MOVW	(index*4)(R1), R8; \
+	EORW	d, R9; \
+	EORW	b, R9; \
+	ADDW	R9, a; \
+	RORW	$(32-shift), a; \
+	MOVW	b, R9; \
+	ADDW	b, a
+
+	ROUND3(R4,R5,R6,R7, 8,0xfffa3942, 4);
+	ROUND3(R7,R4,R5,R6,11,0x8771f681,11);
+	ROUND3(R6,R7,R4,R5,14,0x6d9d6122,16);
+	ROUND3(R5,R6,R7,R4, 1,0xfde5380c,23);
+	ROUND3(R4,R5,R6,R7, 4,0xa4beea44, 4);
+	ROUND3(R7,R4,R5,R6, 7,0x4bdecfa9,11);
+	ROUND3(R6,R7,R4,R5,10,0xf6bb4b60,16);
+	ROUND3(R5,R6,R7,R4,13,0xbebfbc70,23);
+	ROUND3(R4,R5,R6,R7, 0,0x289b7ec6, 4);
+	ROUND3(R7,R4,R5,R6, 3,0xeaa127fa,11);
+	ROUND3(R6,R7,R4,R5, 6,0xd4ef3085,16);
+	ROUND3(R5,R6,R7,R4, 9, 0x4881d05,23);
+	ROUND3(R4,R5,R6,R7,12,0xd9d4d039, 4);
+	ROUND3(R7,R4,R5,R6,15,0xe6db99e5,11);
+	ROUND3(R6,R7,R4,R5, 2,0x1fa27cf8,16);
+	ROUND3(R5,R6,R7,R4, 0,0xc4ac5665,23);
+
+	MOVW	(0*4)(R1), R8
+	MVNW	R7, R9
+
+#define ROUND4(a, b, c, d, index, const, shift) \
+	ADDW	$const, a; \
+	ADDW	R8, a; \
+	MOVW	(index*4)(R1), R8; \
+	ORRW	b, R9; \
+	EORW	c, R9; \
+	ADDW	R9, a; \
+	RORW	$(32-shift), a; \
+	MVNW	c, R9; \
+	ADDW	b, a
+
+	ROUND4(R4,R5,R6,R7, 7,0xf4292244, 6);
+	ROUND4(R7,R4,R5,R6,14,0x432aff97,10);
+	ROUND4(R6,R7,R4,R5, 5,0xab9423a7,15);
+	ROUND4(R5,R6,R7,R4,12,0xfc93a039,21);
+	ROUND4(R4,R5,R6,R7, 3,0x655b59c3, 6);
+	ROUND4(R7,R4,R5,R6,10,0x8f0ccc92,10);
+	ROUND4(R6,R7,R4,R5, 1,0xffeff47d,15);
+	ROUND4(R5,R6,R7,R4, 8,0x85845dd1,21);
+	ROUND4(R4,R5,R6,R7,15,0x6fa87e4f, 6);
+	ROUND4(R7,R4,R5,R6, 6,0xfe2ce6e0,10);
+	ROUND4(R6,R7,R4,R5,13,0xa3014314,15);
+	ROUND4(R5,R6,R7,R4, 4,0x4e0811a1,21);
+	ROUND4(R4,R5,R6,R7,11,0xf7537e82, 6);
+	ROUND4(R7,R4,R5,R6, 2,0xbd3af235,10);
+	ROUND4(R6,R7,R4,R5, 9,0x2ad7d2bb,15);
+	ROUND4(R5,R6,R7,R4, 0,0xeb86d391,21);
+
+	ADDW	R12, R4
+	ADDW	R13, R5
+	ADDW	R14, R6
+	ADDW	R15, R7
+
+	ADD	$64, R1
+	CMP	R1, R21
+	BNE	loop
+
+	STPW	(R4, R5), (0*8)(R0)
+	STPW	(R6, R7), (1*8)(R0)
+zero:
+	RET
diff --git a/src/crypto/md5/md5block_decl.go b/src/crypto/md5/md5block_decl.go
new file mode 100644
index 0000000..6716a0c
--- /dev/null
+++ b/src/crypto/md5/md5block_decl.go
@@ -0,0 +1,13 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build amd64 || 386 || arm || ppc64le || ppc64 || s390x || arm64
+
+package md5
+
+const haveAsm = true
+
+//go:noescape
+
+func block(dig *digest, p []byte)
diff --git a/src/crypto/md5/md5block_generic.go b/src/crypto/md5/md5block_generic.go
new file mode 100644
index 0000000..c929c2b
--- /dev/null
+++ b/src/crypto/md5/md5block_generic.go
@@ -0,0 +1,13 @@
+// Copyright 2014 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !amd64 && !386 && !arm && !ppc64le && !ppc64 && !s390x && !arm64
+
+package md5
+
+const haveAsm = false
+
+func block(dig *digest, p []byte) {
+	blockGeneric(dig, p)
+}
diff --git a/src/crypto/md5/md5block_ppc64x.s b/src/crypto/md5/md5block_ppc64x.s
new file mode 100644
index 0000000..69a20e7
--- /dev/null
+++ b/src/crypto/md5/md5block_ppc64x.s
@@ -0,0 +1,212 @@
+// Original source:
+//	http://www.zorinaq.com/papers/md5-amd64.html
+//	http://www.zorinaq.com/papers/md5-amd64.tar.bz2
+//
+// MD5 optimized for ppc64le using Go's assembler for
+// ppc64le, based on md5block_amd64.s implementation by
+// the Go authors.
+//
+// Author: Marc Bevand <bevand_m (at) epita.fr>
+// Licence: I hereby disclaim the copyright on this code and place it
+// in the public domain.
+
+//go:build ppc64 || ppc64le
+
+#include "textflag.h"
+
+// ENDIAN_MOVE generates the appropriate
+// 4 byte load for big or little endian.
+// The 4 bytes at ptr+off is loaded into dst.
+// The idx reg is only needed for big endian
+// and is clobbered when used.
+#ifdef GOARCH_ppc64le
+#define ENDIAN_MOVE(off, ptr, dst, idx) \
+	MOVWZ	off(ptr),dst
+#else
+#define ENDIAN_MOVE(off, ptr, dst, idx) \
+	MOVD	$off,idx; \
+	MOVWBR	(idx)(ptr), dst
+#endif
+
+#define M00 R18
+#define M01 R19
+#define M02 R20
+#define M03 R24
+#define M04 R25
+#define M05 R26
+#define M06 R27
+#define M07 R28
+#define M08 R29
+#define M09 R21
+#define M10 R11
+#define M11 R8
+#define M12 R7
+#define M13 R12
+#define M14 R23
+#define M15 R10
+
+#define ROUND1(a, b, c, d, index, const, shift) \
+	ADD	$const, index, R9; \
+	ADD	R9, a; \
+	AND     b, c, R9; \
+	ANDN    b, d, R31; \
+	OR	R9, R31, R9; \
+	ADD	R9, a; \
+	ROTLW	$shift, a; \
+	ADD	b, a;
+
+#define ROUND2(a, b, c, d, index, const, shift) \
+	ADD	$const, index, R9; \
+	ADD	R9, a; \
+	AND	b, d, R31; \
+	ANDN	d, c, R9; \
+	OR	R9, R31; \
+	ADD	R31, a; \
+	ROTLW	$shift, a; \
+	ADD	b, a;
+
+#define ROUND3(a, b, c, d, index, const, shift) \
+	ADD	$const, index, R9; \
+	ADD	R9, a; \
+	XOR	d, c, R31; \
+	XOR	b, R31; \
+	ADD	R31, a; \
+	ROTLW	$shift, a; \
+	ADD	b, a;
+
+#define ROUND4(a, b, c, d, index, const, shift) \
+	ADD	$const, index, R9; \
+	ADD	R9, a; \
+	ORN     d, b, R31; \
+	XOR	c, R31; \
+	ADD	R31, a; \
+	ROTLW	$shift, a; \
+	ADD	b, a;
+
+
+TEXT ·block(SB),NOSPLIT,$0-32
+	MOVD	dig+0(FP), R10
+	MOVD	p+8(FP), R6
+	MOVD	p_len+16(FP), R5
+
+	// We assume p_len >= 64
+	SRD 	$6, R5
+	MOVD	R5, CTR
+
+	MOVWZ	0(R10), R22
+	MOVWZ	4(R10), R3
+	MOVWZ	8(R10), R4
+	MOVWZ	12(R10), R5
+
+loop:
+	MOVD	R22, R14
+	MOVD	R3, R15
+	MOVD	R4, R16
+	MOVD	R5, R17
+
+	ENDIAN_MOVE( 0,R6,M00,M15)
+	ENDIAN_MOVE( 4,R6,M01,M15)
+	ENDIAN_MOVE( 8,R6,M02,M15)
+	ENDIAN_MOVE(12,R6,M03,M15)
+
+	ROUND1(R22,R3,R4,R5,M00,0xd76aa478, 7);
+	ROUND1(R5,R22,R3,R4,M01,0xe8c7b756,12);
+	ROUND1(R4,R5,R22,R3,M02,0x242070db,17);
+	ROUND1(R3,R4,R5,R22,M03,0xc1bdceee,22);
+
+	ENDIAN_MOVE(16,R6,M04,M15)
+	ENDIAN_MOVE(20,R6,M05,M15)
+	ENDIAN_MOVE(24,R6,M06,M15)
+	ENDIAN_MOVE(28,R6,M07,M15)
+
+	ROUND1(R22,R3,R4,R5,M04,0xf57c0faf, 7);
+	ROUND1(R5,R22,R3,R4,M05,0x4787c62a,12);
+	ROUND1(R4,R5,R22,R3,M06,0xa8304613,17);
+	ROUND1(R3,R4,R5,R22,M07,0xfd469501,22);
+
+	ENDIAN_MOVE(32,R6,M08,M15)
+	ENDIAN_MOVE(36,R6,M09,M15)
+	ENDIAN_MOVE(40,R6,M10,M15)
+	ENDIAN_MOVE(44,R6,M11,M15)
+
+	ROUND1(R22,R3,R4,R5,M08,0x698098d8, 7);
+	ROUND1(R5,R22,R3,R4,M09,0x8b44f7af,12);
+	ROUND1(R4,R5,R22,R3,M10,0xffff5bb1,17);
+	ROUND1(R3,R4,R5,R22,M11,0x895cd7be,22);
+
+	ENDIAN_MOVE(48,R6,M12,M15)
+	ENDIAN_MOVE(52,R6,M13,M15)
+	ENDIAN_MOVE(56,R6,M14,M15)
+	ENDIAN_MOVE(60,R6,M15,M15)
+
+	ROUND1(R22,R3,R4,R5,M12,0x6b901122, 7);
+	ROUND1(R5,R22,R3,R4,M13,0xfd987193,12);
+	ROUND1(R4,R5,R22,R3,M14,0xa679438e,17);
+	ROUND1(R3,R4,R5,R22,M15,0x49b40821,22);
+
+	ROUND2(R22,R3,R4,R5,M01,0xf61e2562, 5);
+	ROUND2(R5,R22,R3,R4,M06,0xc040b340, 9);
+	ROUND2(R4,R5,R22,R3,M11,0x265e5a51,14);
+	ROUND2(R3,R4,R5,R22,M00,0xe9b6c7aa,20);
+	ROUND2(R22,R3,R4,R5,M05,0xd62f105d, 5);
+	ROUND2(R5,R22,R3,R4,M10, 0x2441453, 9);
+	ROUND2(R4,R5,R22,R3,M15,0xd8a1e681,14);
+	ROUND2(R3,R4,R5,R22,M04,0xe7d3fbc8,20);
+	ROUND2(R22,R3,R4,R5,M09,0x21e1cde6, 5);
+	ROUND2(R5,R22,R3,R4,M14,0xc33707d6, 9);
+	ROUND2(R4,R5,R22,R3,M03,0xf4d50d87,14);
+	ROUND2(R3,R4,R5,R22,M08,0x455a14ed,20);
+	ROUND2(R22,R3,R4,R5,M13,0xa9e3e905, 5);
+	ROUND2(R5,R22,R3,R4,M02,0xfcefa3f8, 9);
+	ROUND2(R4,R5,R22,R3,M07,0x676f02d9,14);
+	ROUND2(R3,R4,R5,R22,M12,0x8d2a4c8a,20);
+
+	ROUND3(R22,R3,R4,R5,M05,0xfffa3942, 4);
+	ROUND3(R5,R22,R3,R4,M08,0x8771f681,11);
+	ROUND3(R4,R5,R22,R3,M11,0x6d9d6122,16);
+	ROUND3(R3,R4,R5,R22,M14,0xfde5380c,23);
+	ROUND3(R22,R3,R4,R5,M01,0xa4beea44, 4);
+	ROUND3(R5,R22,R3,R4,M04,0x4bdecfa9,11);
+	ROUND3(R4,R5,R22,R3,M07,0xf6bb4b60,16);
+	ROUND3(R3,R4,R5,R22,M10,0xbebfbc70,23);
+	ROUND3(R22,R3,R4,R5,M13,0x289b7ec6, 4);
+	ROUND3(R5,R22,R3,R4,M00,0xeaa127fa,11);
+	ROUND3(R4,R5,R22,R3,M03,0xd4ef3085,16);
+	ROUND3(R3,R4,R5,R22,M06, 0x4881d05,23);
+	ROUND3(R22,R3,R4,R5,M09,0xd9d4d039, 4);
+	ROUND3(R5,R22,R3,R4,M12,0xe6db99e5,11);
+	ROUND3(R4,R5,R22,R3,M15,0x1fa27cf8,16);
+	ROUND3(R3,R4,R5,R22,M02,0xc4ac5665,23);
+
+	ROUND4(R22,R3,R4,R5,M00,0xf4292244, 6);
+	ROUND4(R5,R22,R3,R4,M07,0x432aff97,10);
+	ROUND4(R4,R5,R22,R3,M14,0xab9423a7,15);
+	ROUND4(R3,R4,R5,R22,M05,0xfc93a039,21);
+	ROUND4(R22,R3,R4,R5,M12,0x655b59c3, 6);
+	ROUND4(R5,R22,R3,R4,M03,0x8f0ccc92,10);
+	ROUND4(R4,R5,R22,R3,M10,0xffeff47d,15);
+	ROUND4(R3,R4,R5,R22,M01,0x85845dd1,21);
+	ROUND4(R22,R3,R4,R5,M08,0x6fa87e4f, 6);
+	ROUND4(R5,R22,R3,R4,M15,0xfe2ce6e0,10);
+	ROUND4(R4,R5,R22,R3,M06,0xa3014314,15);
+	ROUND4(R3,R4,R5,R22,M13,0x4e0811a1,21);
+	ROUND4(R22,R3,R4,R5,M04,0xf7537e82, 6);
+	ROUND4(R5,R22,R3,R4,M11,0xbd3af235,10);
+	ROUND4(R4,R5,R22,R3,M02,0x2ad7d2bb,15);
+	ROUND4(R3,R4,R5,R22,M09,0xeb86d391,21);
+
+	ADD	R14, R22
+	ADD	R15, R3
+	ADD	R16, R4
+	ADD	R17, R5
+	ADD	$64, R6
+	BC	16, 0, loop // bdnz
+
+end:
+	MOVD	dig+0(FP), R10
+	MOVWZ	R22, 0(R10)
+	MOVWZ	R3, 4(R10)
+	MOVWZ	R4, 8(R10)
+	MOVWZ	R5, 12(R10)
+
+	RET
diff --git a/src/crypto/md5/md5block_s390x.s b/src/crypto/md5/md5block_s390x.s
new file mode 100644
index 0000000..68f501c
--- /dev/null
+++ b/src/crypto/md5/md5block_s390x.s
@@ -0,0 +1,175 @@
+// Original source:
+//	http://www.zorinaq.com/papers/md5-amd64.html
+//	http://www.zorinaq.com/papers/md5-amd64.tar.bz2
+//
+// MD5 adapted for s390x using Go's assembler for
+// s390x, based on md5block_amd64.s implementation by
+// the Go authors.
+//
+// Author: Marc Bevand <bevand_m (at) epita.fr>
+// Licence: I hereby disclaim the copyright on this code and place it
+// in the public domain.
+
+#include "textflag.h"
+
+// func block(dig *digest, p []byte)
+TEXT ·block(SB),NOSPLIT,$16-32
+	MOVD	dig+0(FP), R1
+	MOVD	p+8(FP), R6
+	MOVD	p_len+16(FP), R5
+	AND	$-64, R5
+	LAY	(R6)(R5*1), R7
+
+	LMY	0(R1), R2, R5
+	CMPBEQ	R6, R7, end
+
+loop:
+	STMY	R2, R5, tmp-16(SP)
+
+	MOVWBR	0(R6), R8
+	MOVWZ	R5, R9
+
+#define ROUND1(a, b, c, d, index, const, shift) \
+	XOR	c, R9; \
+	ADD	$const, a; \
+	ADD	R8, a; \
+	MOVWBR	(index*4)(R6), R8; \
+	AND	b, R9; \
+	XOR	d, R9; \
+	ADD	R9, a; \
+	RLL	$shift, a; \
+	MOVWZ	c, R9; \
+	ADD	b, a
+
+	ROUND1(R2,R3,R4,R5, 1,0xd76aa478, 7);
+	ROUND1(R5,R2,R3,R4, 2,0xe8c7b756,12);
+	ROUND1(R4,R5,R2,R3, 3,0x242070db,17);
+	ROUND1(R3,R4,R5,R2, 4,0xc1bdceee,22);
+	ROUND1(R2,R3,R4,R5, 5,0xf57c0faf, 7);
+	ROUND1(R5,R2,R3,R4, 6,0x4787c62a,12);
+	ROUND1(R4,R5,R2,R3, 7,0xa8304613,17);
+	ROUND1(R3,R4,R5,R2, 8,0xfd469501,22);
+	ROUND1(R2,R3,R4,R5, 9,0x698098d8, 7);
+	ROUND1(R5,R2,R3,R4,10,0x8b44f7af,12);
+	ROUND1(R4,R5,R2,R3,11,0xffff5bb1,17);
+	ROUND1(R3,R4,R5,R2,12,0x895cd7be,22);
+	ROUND1(R2,R3,R4,R5,13,0x6b901122, 7);
+	ROUND1(R5,R2,R3,R4,14,0xfd987193,12);
+	ROUND1(R4,R5,R2,R3,15,0xa679438e,17);
+	ROUND1(R3,R4,R5,R2, 0,0x49b40821,22);
+
+	MOVWBR	(1*4)(R6), R8
+	MOVWZ	R5, R9
+	MOVWZ	R5, R1
+
+#define ROUND2(a, b, c, d, index, const, shift) \
+	XOR	$0xffffffff, R9; \ // NOTW R9
+	ADD	$const, a; \
+	ADD	R8, a; \
+	MOVWBR	(index*4)(R6), R8; \
+	AND	b, R1; \
+	AND	c, R9; \
+	OR	R9, R1; \
+	MOVWZ	c, R9; \
+	ADD	R1, a; \
+	MOVWZ	c, R1; \
+	RLL	$shift,	a; \
+	ADD	b, a
+
+	ROUND2(R2,R3,R4,R5, 6,0xf61e2562, 5);
+	ROUND2(R5,R2,R3,R4,11,0xc040b340, 9);
+	ROUND2(R4,R5,R2,R3, 0,0x265e5a51,14);
+	ROUND2(R3,R4,R5,R2, 5,0xe9b6c7aa,20);
+	ROUND2(R2,R3,R4,R5,10,0xd62f105d, 5);
+	ROUND2(R5,R2,R3,R4,15, 0x2441453, 9);
+	ROUND2(R4,R5,R2,R3, 4,0xd8a1e681,14);
+	ROUND2(R3,R4,R5,R2, 9,0xe7d3fbc8,20);
+	ROUND2(R2,R3,R4,R5,14,0x21e1cde6, 5);
+	ROUND2(R5,R2,R3,R4, 3,0xc33707d6, 9);
+	ROUND2(R4,R5,R2,R3, 8,0xf4d50d87,14);
+	ROUND2(R3,R4,R5,R2,13,0x455a14ed,20);
+	ROUND2(R2,R3,R4,R5, 2,0xa9e3e905, 5);
+	ROUND2(R5,R2,R3,R4, 7,0xfcefa3f8, 9);
+	ROUND2(R4,R5,R2,R3,12,0x676f02d9,14);
+	ROUND2(R3,R4,R5,R2, 0,0x8d2a4c8a,20);
+
+	MOVWBR	(5*4)(R6), R8
+	MOVWZ	R4, R9
+
+#define ROUND3(a, b, c, d, index, const, shift) \
+	ADD	$const, a; \
+	ADD	R8, a; \
+	MOVWBR	(index*4)(R6), R8; \
+	XOR	d, R9; \
+	XOR	b, R9; \
+	ADD	R9, a; \
+	RLL	$shift, a; \
+	MOVWZ	b, R9; \
+	ADD	b, a
+
+	ROUND3(R2,R3,R4,R5, 8,0xfffa3942, 4);
+	ROUND3(R5,R2,R3,R4,11,0x8771f681,11);
+	ROUND3(R4,R5,R2,R3,14,0x6d9d6122,16);
+	ROUND3(R3,R4,R5,R2, 1,0xfde5380c,23);
+	ROUND3(R2,R3,R4,R5, 4,0xa4beea44, 4);
+	ROUND3(R5,R2,R3,R4, 7,0x4bdecfa9,11);
+	ROUND3(R4,R5,R2,R3,10,0xf6bb4b60,16);
+	ROUND3(R3,R4,R5,R2,13,0xbebfbc70,23);
+	ROUND3(R2,R3,R4,R5, 0,0x289b7ec6, 4);
+	ROUND3(R5,R2,R3,R4, 3,0xeaa127fa,11);
+	ROUND3(R4,R5,R2,R3, 6,0xd4ef3085,16);
+	ROUND3(R3,R4,R5,R2, 9, 0x4881d05,23);
+	ROUND3(R2,R3,R4,R5,12,0xd9d4d039, 4);
+	ROUND3(R5,R2,R3,R4,15,0xe6db99e5,11);
+	ROUND3(R4,R5,R2,R3, 2,0x1fa27cf8,16);
+	ROUND3(R3,R4,R5,R2, 0,0xc4ac5665,23);
+
+	MOVWBR	(0*4)(R6), R8
+	MOVWZ	$0xffffffff, R9
+	XOR	R5, R9
+
+#define ROUND4(a, b, c, d, index, const, shift) \
+	ADD	$const, a; \
+	ADD	R8, a; \
+	MOVWBR	(index*4)(R6), R8; \
+	OR	b, R9; \
+	XOR	c, R9; \
+	ADD	R9, a; \
+	MOVWZ	$0xffffffff, R9; \
+	RLL	$shift,	a; \
+	XOR	c, R9; \
+	ADD	b, a
+
+	ROUND4(R2,R3,R4,R5, 7,0xf4292244, 6);
+	ROUND4(R5,R2,R3,R4,14,0x432aff97,10);
+	ROUND4(R4,R5,R2,R3, 5,0xab9423a7,15);
+	ROUND4(R3,R4,R5,R2,12,0xfc93a039,21);
+	ROUND4(R2,R3,R4,R5, 3,0x655b59c3, 6);
+	ROUND4(R5,R2,R3,R4,10,0x8f0ccc92,10);
+	ROUND4(R4,R5,R2,R3, 1,0xffeff47d,15);
+	ROUND4(R3,R4,R5,R2, 8,0x85845dd1,21);
+	ROUND4(R2,R3,R4,R5,15,0x6fa87e4f, 6);
+	ROUND4(R5,R2,R3,R4, 6,0xfe2ce6e0,10);
+	ROUND4(R4,R5,R2,R3,13,0xa3014314,15);
+	ROUND4(R3,R4,R5,R2, 4,0x4e0811a1,21);
+	ROUND4(R2,R3,R4,R5,11,0xf7537e82, 6);
+	ROUND4(R5,R2,R3,R4, 2,0xbd3af235,10);
+	ROUND4(R4,R5,R2,R3, 9,0x2ad7d2bb,15);
+	ROUND4(R3,R4,R5,R2, 0,0xeb86d391,21);
+
+	MOVWZ	tmp-16(SP), R1
+	ADD	R1, R2
+	MOVWZ	tmp-12(SP), R1
+	ADD	R1, R3
+	MOVWZ	tmp-8(SP), R1
+	ADD	R1, R4
+	MOVWZ	tmp-4(SP), R1
+	ADD	R1, R5
+
+	LA	64(R6), R6
+	CMPBLT	R6, R7, loop
+
+end:
+	MOVD	dig+0(FP), R1
+	STMY	R2, R5, 0(R1)
+	RET