diff --git a/ssl/test/runner/chacha20_poly1305.go b/ssl/test/runner/chacha20_poly1305.go
new file mode 100644
index 0000000..42911d4
--- /dev/null
+++ b/ssl/test/runner/chacha20_poly1305.go
@@ -0,0 +1,159 @@
+package main
+
+import (
+	"crypto/cipher"
+	"crypto/subtle"
+	"encoding/binary"
+	"errors"
+)
+
+// See draft-agl-tls-chacha20poly1305-04 and
+// draft-irtf-cfrg-chacha20-poly1305-10. Where the two differ, the
+// draft-agl-tls-chacha20poly1305-04 variant is implemented.
+
+func leftRotate(a uint32, n uint) uint32 {
+	return (a << n) | (a >> (32 - n))
+}
+
+func chaChaQuarterRound(state *[16]uint32, a, b, c, d int) {
+	state[a] += state[b]
+	state[d] = leftRotate(state[d]^state[a], 16)
+
+	state[c] += state[d]
+	state[b] = leftRotate(state[b]^state[c], 12)
+
+	state[a] += state[b]
+	state[d] = leftRotate(state[d]^state[a], 8)
+
+	state[c] += state[d]
+	state[b] = leftRotate(state[b]^state[c], 7)
+}
+
+func chaCha20Block(state *[16]uint32, out []byte) {
+	var workingState [16]uint32
+	copy(workingState[:], state[:])
+	for i := 0; i < 10; i++ {
+		chaChaQuarterRound(&workingState, 0, 4, 8, 12)
+		chaChaQuarterRound(&workingState, 1, 5, 9, 13)
+		chaChaQuarterRound(&workingState, 2, 6, 10, 14)
+		chaChaQuarterRound(&workingState, 3, 7, 11, 15)
+		chaChaQuarterRound(&workingState, 0, 5, 10, 15)
+		chaChaQuarterRound(&workingState, 1, 6, 11, 12)
+		chaChaQuarterRound(&workingState, 2, 7, 8, 13)
+		chaChaQuarterRound(&workingState, 3, 4, 9, 14)
+	}
+	for i := 0; i < 16; i++ {
+		binary.LittleEndian.PutUint32(out[i*4:i*4+4], workingState[i]+state[i])
+	}
+}
+
+// sliceForAppend takes a slice and a requested number of bytes. It returns a
+// slice with the contents of the given slice followed by that many bytes and a
+// second slice that aliases into it and contains only the extra bytes. If the
+// original slice has sufficient capacity then no allocation is performed.
+func sliceForAppend(in []byte, n int) (head, tail []byte) {
+	if total := len(in) + n; cap(in) >= total {
+		head = in[:total]
+	} else {
+		head = make([]byte, total)
+		copy(head, in)
+	}
+	tail = head[len(in):]
+	return
+}
+
+type chaCha20Poly1305 struct {
+	key [32]byte
+}
+
+func newChaCha20Poly1305(key []byte) (cipher.AEAD, error) {
+	if len(key) != 32 {
+		return nil, errors.New("bad key length")
+	}
+	aead := new(chaCha20Poly1305)
+	copy(aead.key[:], key)
+	return aead, nil
+}
+
+func (c *chaCha20Poly1305) NonceSize() int { return 8 }
+func (c *chaCha20Poly1305) Overhead() int  { return 16 }
+
+func (c *chaCha20Poly1305) chaCha20(out, in, nonce []byte, counter uint64) {
+	var state [16]uint32
+	state[0] = 0x61707865
+	state[1] = 0x3320646e
+	state[2] = 0x79622d32
+	state[3] = 0x6b206574
+	for i := 0; i < 8; i++ {
+		state[4+i] = binary.LittleEndian.Uint32(c.key[i*4 : i*4+4])
+	}
+	state[14] = binary.LittleEndian.Uint32(nonce[0:4])
+	state[15] = binary.LittleEndian.Uint32(nonce[4:8])
+
+	for i := 0; i < len(in); i += 64 {
+		state[12] = uint32(counter & 0xffffffff)
+		state[13] = uint32(counter >> 32)
+
+		var tmp [64]byte
+		chaCha20Block(&state, tmp[:])
+		count := 64
+		if len(in)-i < count {
+			count = len(in) - i
+		}
+		for j := 0; j < count; j++ {
+			out[i+j] = in[i+j] ^ tmp[j]
+		}
+
+		counter++
+	}
+}
+
+func (c *chaCha20Poly1305) poly1305(tag *[16]byte, nonce, ciphertext, additionalData []byte) {
+	input := make([]byte, 0, len(additionalData)+8+len(ciphertext)+8)
+	input = append(input, additionalData...)
+	input, out := sliceForAppend(input, 8)
+	binary.LittleEndian.PutUint64(out, uint64(len(additionalData)))
+	input = append(input, ciphertext...)
+	input, out = sliceForAppend(input, 8)
+	binary.LittleEndian.PutUint64(out, uint64(len(ciphertext)))
+
+	var poly1305Key [32]byte
+	c.chaCha20(poly1305Key[:], poly1305Key[:], nonce, 0)
+
+	poly1305Sum(tag, input, &poly1305Key)
+}
+
+func (c *chaCha20Poly1305) Seal(dst, nonce, plaintext, additionalData []byte) []byte {
+	if len(nonce) != 8 {
+		panic("Bad nonce length")
+	}
+
+	ret, out := sliceForAppend(dst, len(plaintext)+16)
+	c.chaCha20(out[:len(plaintext)], plaintext, nonce, 1)
+
+	var tag [16]byte
+	c.poly1305(&tag, nonce, out[:len(plaintext)], additionalData)
+	copy(out[len(plaintext):], tag[:])
+
+	return ret
+}
+
+func (c *chaCha20Poly1305) Open(dst, nonce, ciphertext, additionalData []byte) ([]byte, error) {
+	if len(nonce) != 8 {
+		panic("Bad nonce length")
+	}
+	if len(ciphertext) < 16 {
+		return nil, errors.New("chacha20: message authentication failed")
+	}
+	plaintextLen := len(ciphertext) - 16
+
+	var tag [16]byte
+	c.poly1305(&tag, nonce, ciphertext[:plaintextLen], additionalData)
+	if subtle.ConstantTimeCompare(tag[:], ciphertext[plaintextLen:]) != 1 {
+		return nil, errors.New("chacha20: message authentication failed")
+	}
+
+	ret, out := sliceForAppend(dst, plaintextLen)
+	c.chaCha20(out, ciphertext[:plaintextLen], nonce, 1)
+	return ret, nil
+}
diff --git a/ssl/test/runner/chacha20_poly1305_test.go b/ssl/test/runner/chacha20_poly1305_test.go
new file mode 100644
index 0000000..726f482
--- /dev/null
+++ b/ssl/test/runner/chacha20_poly1305_test.go
@@ -0,0 +1,99 @@
+package main
+
+import (
+	"bytes"
+	"encoding/hex"
+	"testing"
+)
+
+// See draft-irtf-cfrg-chacha20-poly1305-10, section 2.1.1.
+func TestChaChaQuarterRound(t *testing.T) {
+	state := [16]uint32{0x11111111, 0x01020304, 0x9b8d6f43, 0x01234567}
+	chaChaQuarterRound(&state, 0, 1, 2, 3)
+
+	a, b, c, d := state[0], state[1], state[2], state[3]
+	if a != 0xea2a92f4 || b != 0xcb1cf8ce || c != 0x4581472e || d != 0x5881c4bb {
+		t.Errorf("Incorrect results: %x", state)
+	}
+}
+
+// See draft-irtf-cfrg-chacha20-poly1305-10, section 2.2.1.
+func TestChaChaQuarterRoundState(t *testing.T) {
+	state := [16]uint32{
+		0x879531e0, 0xc5ecf37d, 0x516461b1, 0xc9a62f8a,
+		0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0x2a5f714c,
+		0x53372767, 0xb00a5631, 0x974c541a, 0x359e9963,
+		0x5c971061, 0x3d631689, 0x2098d9d6, 0x91dbd320,
+	}
+	chaChaQuarterRound(&state, 2, 7, 8, 13)
+
+	expected := [16]uint32{
+		0x879531e0, 0xc5ecf37d, 0xbdb886dc, 0xc9a62f8a,
+		0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0xcfacafd2,
+		0xe46bea80, 0xb00a5631, 0x974c541a, 0x359e9963,
+		0x5c971061, 0xccc07c79, 0x2098d9d6, 0x91dbd320,
+	}
+	for i := range state {
+		if state[i] != expected[i] {
+			t.Errorf("Mismatch at %d: %x vs %x", i, state, expected)
+		}
+	}
+}
+
+// See draft-irtf-cfrg-chacha20-poly1305-10, section 2.3.2.
+func TestChaCha20Block(t *testing.T) {
+	state := [16]uint32{
+		0x61707865, 0x3320646e, 0x79622d32, 0x6b206574,
+		0x03020100, 0x07060504, 0x0b0a0908, 0x0f0e0d0c,
+		0x13121110, 0x17161514, 0x1b1a1918, 0x1f1e1d1c,
+		0x00000001, 0x09000000, 0x4a000000, 0x00000000,
+	}
+	out := make([]byte, 64)
+	chaCha20Block(&state, out)
+
+	expected := []byte{
+		0x10, 0xf1, 0xe7, 0xe4, 0xd1, 0x3b, 0x59, 0x15,
+		0x50, 0x0f, 0xdd, 0x1f, 0xa3, 0x20, 0x71, 0xc4,
+		0xc7, 0xd1, 0xf4, 0xc7, 0x33, 0xc0, 0x68, 0x03,
+		0x04, 0x22, 0xaa, 0x9a, 0xc3, 0xd4, 0x6c, 0x4e,
+		0xd2, 0x82, 0x64, 0x46, 0x07, 0x9f, 0xaa, 0x09,
+		0x14, 0xc2, 0xd7, 0x05, 0xd9, 0x8b, 0x02, 0xa2,
+		0xb5, 0x12, 0x9c, 0xd1, 0xde, 0x16, 0x4e, 0xb9,
+		0xcb, 0xd0, 0x83, 0xe8, 0xa2, 0x50, 0x3c, 0x4e,
+	}
+	if !bytes.Equal(out, expected) {
+		t.Errorf("Got %x, wanted %x", out, expected)
+	}
+}
+
+// See draft-agl-tls-chacha20poly1305-04, section 7.
+func TestChaCha20Poly1305(t *testing.T) {
+	key, _ := hex.DecodeString("4290bcb154173531f314af57f3be3b5006da371ece272afa1b5dbdd1100a1007")
+	input, _ := hex.DecodeString("86d09974840bded2a5ca")
+	nonce, _ := hex.DecodeString("cd7cf67be39c794a")
+	ad, _ := hex.DecodeString("87e229d4500845a079c0")
+	output, _ := hex.DecodeString("e3e446f7ede9a19b62a4677dabf4e3d24b876bb284753896e1d6")
+
+	aead, err := newChaCha20Poly1305(key)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	out, err := aead.Open(nil, nonce, output, ad)
+	if err != nil {
+		t.Errorf("Open failed: %s", err)
+	} else if !bytes.Equal(out, input) {
+		t.Errorf("Open gave %x, wanted %x", out, input)
+	}
+
+	out = aead.Seal(nil, nonce, input, ad)
+	if !bytes.Equal(out, output) {
+		t.Errorf("Open gave %x, wanted %x", out, output)
+	}
+
+	out[0]++
+	_, err = aead.Open(nil, nonce, out, ad)
+	if err == nil {
+		t.Errorf("Open on malformed data unexpectedly succeeded")
+	}
+}
diff --git a/ssl/test/runner/cipher_suites.go b/ssl/test/runner/cipher_suites.go
index 62b8100..162c0c0 100644
--- a/ssl/test/runner/cipher_suites.go
+++ b/ssl/test/runner/cipher_suites.go
@@ -62,6 +62,11 @@
 	suitePSK
 )
 
+type tlsAead struct {
+	cipher.AEAD
+	explicitNonce bool
+}
+
 // A cipherSuite is a specific combination of key agreement, cipher and MAC
 // function. All cipher suites currently assume RSA key agreement.
 type cipherSuite struct {
@@ -75,12 +80,14 @@
 	flags  int
 	cipher func(key, iv []byte, isRead bool) interface{}
 	mac    func(version uint16, macKey []byte) macFunction
-	aead   func(key, fixedNonce []byte) cipher.AEAD
+	aead   func(key, fixedNonce []byte) *tlsAead
 }
 
 var cipherSuites = []*cipherSuite{
 	// Ciphersuite order is chosen so that ECDHE comes before plain RSA
 	// and RC4 comes before AES (because of the Lucky13 attack).
+	{TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256, 32, 0, 0, ecdheECDSAKA, suiteECDHE | suiteECDSA | suiteTLS12, nil, nil, aeadCHACHA20POLY1305},
+	{TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256, 32, 0, 0, ecdheRSAKA, suiteECDHE | suiteTLS12, nil, nil, aeadCHACHA20POLY1305},
 	{TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256, 16, 0, 4, ecdheRSAKA, suiteECDHE | suiteTLS12, nil, nil, aeadAESGCM},
 	{TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256, 16, 0, 4, ecdheECDSAKA, suiteECDHE | suiteECDSA | suiteTLS12, nil, nil, aeadAESGCM},
 	{TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384, 32, 0, 4, ecdheRSAKA, suiteECDHE | suiteTLS12 | suiteSHA384, nil, nil, aeadAESGCM},
@@ -95,6 +102,7 @@
 	{TLS_ECDHE_ECDSA_WITH_AES_256_CBC_SHA384, 32, 48, 16, ecdheECDSAKA, suiteECDHE | suiteECDSA | suiteTLS12 | suiteSHA384, cipherAES, macSHA384, nil},
 	{TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA, 32, 20, 16, ecdheRSAKA, suiteECDHE, cipherAES, macSHA1, nil},
 	{TLS_ECDHE_ECDSA_WITH_AES_256_CBC_SHA, 32, 20, 16, ecdheECDSAKA, suiteECDHE | suiteECDSA, cipherAES, macSHA1, nil},
+	{TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256, 32, 0, 0, dheRSAKA, suiteTLS12, nil, nil, aeadCHACHA20POLY1305},
 	{TLS_DHE_RSA_WITH_AES_128_GCM_SHA256, 16, 0, 4, dheRSAKA, suiteTLS12, nil, nil, aeadAESGCM},
 	{TLS_DHE_RSA_WITH_AES_256_GCM_SHA384, 32, 0, 4, dheRSAKA, suiteTLS12 | suiteSHA384, nil, nil, aeadAESGCM},
 	{TLS_DHE_RSA_WITH_AES_128_CBC_SHA256, 16, 32, 16, dheRSAKA, suiteTLS12, cipherAES, macSHA256, nil},
@@ -216,7 +224,7 @@
 	return f.aead.Open(out, f.openNonce, plaintext, additionalData)
 }
 
-func aeadAESGCM(key, fixedNonce []byte) cipher.AEAD {
+func aeadAESGCM(key, fixedNonce []byte) *tlsAead {
 	aes, err := aes.NewCipher(key)
 	if err != nil {
 		panic(err)
@@ -230,7 +238,15 @@
 	copy(nonce1, fixedNonce)
 	copy(nonce2, fixedNonce)
 
-	return &fixedNonceAEAD{nonce1, nonce2, aead}
+	return &tlsAead{&fixedNonceAEAD{nonce1, nonce2, aead}, true}
+}
+
+func aeadCHACHA20POLY1305(key, fixedNonce []byte) *tlsAead {
+	aead, err := newChaCha20Poly1305(key)
+	if err != nil {
+		panic(err)
+	}
+	return &tlsAead{aead, false}
 }
 
 // ssl30MAC implements the SSLv3 MAC function, as defined in
@@ -391,5 +407,8 @@
 
 // Additional cipher suite IDs, not IANA-assigned.
 const (
-	TLS_ECDHE_PSK_WITH_AES_128_GCM_SHA256 uint16 = 0xcafe
+	TLS_ECDHE_PSK_WITH_AES_128_GCM_SHA256         uint16 = 0xcafe
+	TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256   uint16 = 0xcc13
+	TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256 uint16 = 0xcc14
+	TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256     uint16 = 0xcc15
 )
diff --git a/ssl/test/runner/conn.go b/ssl/test/runner/conn.go
index 0fe34b7..fd198ca 100644
--- a/ssl/test/runner/conn.go
+++ b/ssl/test/runner/conn.go
@@ -338,13 +338,16 @@
 		switch c := hc.cipher.(type) {
 		case cipher.Stream:
 			c.XORKeyStream(payload, payload)
-		case cipher.AEAD:
-			explicitIVLen = 8
-			if len(payload) < explicitIVLen {
-				return false, 0, alertBadRecordMAC
+		case *tlsAead:
+			nonce := seq
+			if c.explicitNonce {
+				explicitIVLen = 8
+				if len(payload) < explicitIVLen {
+					return false, 0, alertBadRecordMAC
+				}
+				nonce = payload[:8]
+				payload = payload[8:]
 			}
-			nonce := payload[:8]
-			payload = payload[8:]
 
 			var additionalData [13]byte
 			copy(additionalData[:], seq)
@@ -468,10 +471,13 @@
 		switch c := hc.cipher.(type) {
 		case cipher.Stream:
 			c.XORKeyStream(payload, payload)
-		case cipher.AEAD:
+		case *tlsAead:
 			payloadLen := len(b.data) - recordHeaderLen - explicitIVLen
 			b.resize(len(b.data) + c.Overhead())
-			nonce := b.data[recordHeaderLen : recordHeaderLen+explicitIVLen]
+			nonce := hc.seq[:]
+			if c.explicitNonce {
+				nonce = b.data[recordHeaderLen : recordHeaderLen+explicitIVLen]
+			}
 			payload := b.data[recordHeaderLen+explicitIVLen:]
 			payload = payload[:payloadLen]
 
@@ -872,7 +878,7 @@
 			}
 		}
 		if explicitIVLen == 0 {
-			if _, ok := c.out.cipher.(cipher.AEAD); ok {
+			if aead, ok := c.out.cipher.(*tlsAead); ok && aead.explicitNonce {
 				explicitIVLen = 8
 				// The AES-GCM construction in TLS has an
 				// explicit nonce so that the nonce can be
diff --git a/ssl/test/runner/dtls.go b/ssl/test/runner/dtls.go
index f9e1148..85c4247 100644
--- a/ssl/test/runner/dtls.go
+++ b/ssl/test/runner/dtls.go
@@ -16,7 +16,6 @@
 
 import (
 	"bytes"
-	"crypto/cipher"
 	"errors"
 	"fmt"
 	"io"
@@ -248,13 +247,15 @@
 	if cbc, ok := c.out.cipher.(cbcMode); ok {
 		// Block cipher modes have an explicit IV.
 		explicitIVLen = cbc.BlockSize()
-	} else if _, ok := c.out.cipher.(cipher.AEAD); ok {
-		explicitIVLen = 8
-		// The AES-GCM construction in TLS has an explicit nonce so that
-		// the nonce can be random. However, the nonce is only 8 bytes
-		// which is too small for a secure, random nonce. Therefore we
-		// use the sequence number as the nonce.
-		explicitIVIsSeq = true
+	} else if aead, ok := c.out.cipher.(*tlsAead); ok {
+		if aead.explicitNonce {
+			explicitIVLen = 8
+			// The AES-GCM construction in TLS has an explicit nonce so that
+			// the nonce can be random. However, the nonce is only 8 bytes
+			// which is too small for a secure, random nonce. Therefore we
+			// use the sequence number as the nonce.
+			explicitIVIsSeq = true
+		}
 	} else if c.out.cipher != nil {
 		panic("Unknown cipher")
 	}
diff --git a/ssl/test/runner/poly1305.go b/ssl/test/runner/poly1305.go
new file mode 100644
index 0000000..51a1009
--- /dev/null
+++ b/ssl/test/runner/poly1305.go
@@ -0,0 +1,1540 @@
+// 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.
+
+package main
+
+// Based on original, public domain implementation from NaCl by D. J.
+// Bernstein.
+
+import (
+	"crypto/subtle"
+	"math"
+)
+
+const (
+	alpham80 = 0.00000000558793544769287109375
+	alpham48 = 24.0
+	alpham16 = 103079215104.0
+	alpha0   = 6755399441055744.0
+	alpha18  = 1770887431076116955136.0
+	alpha32  = 29014219670751100192948224.0
+	alpha50  = 7605903601369376408980219232256.0
+	alpha64  = 124615124604835863084731911901282304.0
+	alpha82  = 32667107224410092492483962313449748299776.0
+	alpha96  = 535217884764734955396857238543560676143529984.0
+	alpha112 = 35076039295941670036888435985190792471742381031424.0
+	alpha130 = 9194973245195333150150082162901855101712434733101613056.0
+	scale    = 0.0000000000000000000000000000000000000036734198463196484624023016788195177431833298649127735047148490821200539357960224151611328125
+	offset0  = 6755408030990331.0
+	offset1  = 29014256564239239022116864.0
+	offset2  = 124615283061160854719918951570079744.0
+	offset3  = 535219245894202480694386063513315216128475136.0
+)
+
+// poly1305Verify returns true if mac is a valid authenticator for m with the
+// given key.
+func poly1305Verify(mac *[16]byte, m []byte, key *[32]byte) bool {
+	var tmp [16]byte
+	poly1305Sum(&tmp, m, key)
+	return subtle.ConstantTimeCompare(tmp[:], mac[:]) == 1
+}
+
+// poly1305Sum generates an authenticator for m using a one-time key and puts
+// the 16-byte result into out. Authenticating two different messages with the
+// same key allows an attacker to forge messages at will.
+func poly1305Sum(out *[16]byte, m []byte, key *[32]byte) {
+	r := key
+	s := key[16:]
+	var (
+		y7        float64
+		y6        float64
+		y1        float64
+		y0        float64
+		y5        float64
+		y4        float64
+		x7        float64
+		x6        float64
+		x1        float64
+		x0        float64
+		y3        float64
+		y2        float64
+		x5        float64
+		r3lowx0   float64
+		x4        float64
+		r0lowx6   float64
+		x3        float64
+		r3highx0  float64
+		x2        float64
+		r0highx6  float64
+		r0lowx0   float64
+		sr1lowx6  float64
+		r0highx0  float64
+		sr1highx6 float64
+		sr3low    float64
+		r1lowx0   float64
+		sr2lowx6  float64
+		r1highx0  float64
+		sr2highx6 float64
+		r2lowx0   float64
+		sr3lowx6  float64
+		r2highx0  float64
+		sr3highx6 float64
+		r1highx4  float64
+		r1lowx4   float64
+		r0highx4  float64
+		r0lowx4   float64
+		sr3highx4 float64
+		sr3lowx4  float64
+		sr2highx4 float64
+		sr2lowx4  float64
+		r0lowx2   float64
+		r0highx2  float64
+		r1lowx2   float64
+		r1highx2  float64
+		r2lowx2   float64
+		r2highx2  float64
+		sr3lowx2  float64
+		sr3highx2 float64
+		z0        float64
+		z1        float64
+		z2        float64
+		z3        float64
+		m0        int64
+		m1        int64
+		m2        int64
+		m3        int64
+		m00       uint32
+		m01       uint32
+		m02       uint32
+		m03       uint32
+		m10       uint32
+		m11       uint32
+		m12       uint32
+		m13       uint32
+		m20       uint32
+		m21       uint32
+		m22       uint32
+		m23       uint32
+		m30       uint32
+		m31       uint32
+		m32       uint32
+		m33       uint64
+		lbelow2   int32
+		lbelow3   int32
+		lbelow4   int32
+		lbelow5   int32
+		lbelow6   int32
+		lbelow7   int32
+		lbelow8   int32
+		lbelow9   int32
+		lbelow10  int32
+		lbelow11  int32
+		lbelow12  int32
+		lbelow13  int32
+		lbelow14  int32
+		lbelow15  int32
+		s00       uint32
+		s01       uint32
+		s02       uint32
+		s03       uint32
+		s10       uint32
+		s11       uint32
+		s12       uint32
+		s13       uint32
+		s20       uint32
+		s21       uint32
+		s22       uint32
+		s23       uint32
+		s30       uint32
+		s31       uint32
+		s32       uint32
+		s33       uint32
+		bits32    uint64
+		f         uint64
+		f0        uint64
+		f1        uint64
+		f2        uint64
+		f3        uint64
+		f4        uint64
+		g         uint64
+		g0        uint64
+		g1        uint64
+		g2        uint64
+		g3        uint64
+		g4        uint64
+	)
+
+	var p int32
+
+	l := int32(len(m))
+
+	r00 := uint32(r[0])
+
+	r01 := uint32(r[1])
+
+	r02 := uint32(r[2])
+	r0 := int64(2151)
+
+	r03 := uint32(r[3])
+	r03 &= 15
+	r0 <<= 51
+
+	r10 := uint32(r[4])
+	r10 &= 252
+	r01 <<= 8
+	r0 += int64(r00)
+
+	r11 := uint32(r[5])
+	r02 <<= 16
+	r0 += int64(r01)
+
+	r12 := uint32(r[6])
+	r03 <<= 24
+	r0 += int64(r02)
+
+	r13 := uint32(r[7])
+	r13 &= 15
+	r1 := int64(2215)
+	r0 += int64(r03)
+
+	d0 := r0
+	r1 <<= 51
+	r2 := int64(2279)
+
+	r20 := uint32(r[8])
+	r20 &= 252
+	r11 <<= 8
+	r1 += int64(r10)
+
+	r21 := uint32(r[9])
+	r12 <<= 16
+	r1 += int64(r11)
+
+	r22 := uint32(r[10])
+	r13 <<= 24
+	r1 += int64(r12)
+
+	r23 := uint32(r[11])
+	r23 &= 15
+	r2 <<= 51
+	r1 += int64(r13)
+
+	d1 := r1
+	r21 <<= 8
+	r2 += int64(r20)
+
+	r30 := uint32(r[12])
+	r30 &= 252
+	r22 <<= 16
+	r2 += int64(r21)
+
+	r31 := uint32(r[13])
+	r23 <<= 24
+	r2 += int64(r22)
+
+	r32 := uint32(r[14])
+	r2 += int64(r23)
+	r3 := int64(2343)
+
+	d2 := r2
+	r3 <<= 51
+
+	r33 := uint32(r[15])
+	r33 &= 15
+	r31 <<= 8
+	r3 += int64(r30)
+
+	r32 <<= 16
+	r3 += int64(r31)
+
+	r33 <<= 24
+	r3 += int64(r32)
+
+	r3 += int64(r33)
+	h0 := alpha32 - alpha32
+
+	d3 := r3
+	h1 := alpha32 - alpha32
+
+	h2 := alpha32 - alpha32
+
+	h3 := alpha32 - alpha32
+
+	h4 := alpha32 - alpha32
+
+	r0low := math.Float64frombits(uint64(d0))
+	h5 := alpha32 - alpha32
+
+	r1low := math.Float64frombits(uint64(d1))
+	h6 := alpha32 - alpha32
+
+	r2low := math.Float64frombits(uint64(d2))
+	h7 := alpha32 - alpha32
+
+	r0low -= alpha0
+
+	r1low -= alpha32
+
+	r2low -= alpha64
+
+	r0high := r0low + alpha18
+
+	r3low := math.Float64frombits(uint64(d3))
+
+	r1high := r1low + alpha50
+	sr1low := scale * r1low
+
+	r2high := r2low + alpha82
+	sr2low := scale * r2low
+
+	r0high -= alpha18
+	r0high_stack := r0high
+
+	r3low -= alpha96
+
+	r1high -= alpha50
+	r1high_stack := r1high
+
+	sr1high := sr1low + alpham80
+
+	r0low -= r0high
+
+	r2high -= alpha82
+	sr3low = scale * r3low
+
+	sr2high := sr2low + alpham48
+
+	r1low -= r1high
+	r1low_stack := r1low
+
+	sr1high -= alpham80
+	sr1high_stack := sr1high
+
+	r2low -= r2high
+	r2low_stack := r2low
+
+	sr2high -= alpham48
+	sr2high_stack := sr2high
+
+	r3high := r3low + alpha112
+	r0low_stack := r0low
+
+	sr1low -= sr1high
+	sr1low_stack := sr1low
+
+	sr3high := sr3low + alpham16
+	r2high_stack := r2high
+
+	sr2low -= sr2high
+	sr2low_stack := sr2low
+
+	r3high -= alpha112
+	r3high_stack := r3high
+
+	sr3high -= alpham16
+	sr3high_stack := sr3high
+
+	r3low -= r3high
+	r3low_stack := r3low
+
+	sr3low -= sr3high
+	sr3low_stack := sr3low
+
+	if l < 16 {
+		goto addatmost15bytes
+	}
+
+	m00 = uint32(m[p+0])
+	m0 = 2151
+
+	m0 <<= 51
+	m1 = 2215
+	m01 = uint32(m[p+1])
+
+	m1 <<= 51
+	m2 = 2279
+	m02 = uint32(m[p+2])
+
+	m2 <<= 51
+	m3 = 2343
+	m03 = uint32(m[p+3])
+
+	m10 = uint32(m[p+4])
+	m01 <<= 8
+	m0 += int64(m00)
+
+	m11 = uint32(m[p+5])
+	m02 <<= 16
+	m0 += int64(m01)
+
+	m12 = uint32(m[p+6])
+	m03 <<= 24
+	m0 += int64(m02)
+
+	m13 = uint32(m[p+7])
+	m3 <<= 51
+	m0 += int64(m03)
+
+	m20 = uint32(m[p+8])
+	m11 <<= 8
+	m1 += int64(m10)
+
+	m21 = uint32(m[p+9])
+	m12 <<= 16
+	m1 += int64(m11)
+
+	m22 = uint32(m[p+10])
+	m13 <<= 24
+	m1 += int64(m12)
+
+	m23 = uint32(m[p+11])
+	m1 += int64(m13)
+
+	m30 = uint32(m[p+12])
+	m21 <<= 8
+	m2 += int64(m20)
+
+	m31 = uint32(m[p+13])
+	m22 <<= 16
+	m2 += int64(m21)
+
+	m32 = uint32(m[p+14])
+	m23 <<= 24
+	m2 += int64(m22)
+
+	m33 = uint64(m[p+15])
+	m2 += int64(m23)
+
+	d0 = m0
+	m31 <<= 8
+	m3 += int64(m30)
+
+	d1 = m1
+	m32 <<= 16
+	m3 += int64(m31)
+
+	d2 = m2
+	m33 += 256
+
+	m33 <<= 24
+	m3 += int64(m32)
+
+	m3 += int64(m33)
+	d3 = m3
+
+	p += 16
+	l -= 16
+
+	z0 = math.Float64frombits(uint64(d0))
+
+	z1 = math.Float64frombits(uint64(d1))
+
+	z2 = math.Float64frombits(uint64(d2))
+
+	z3 = math.Float64frombits(uint64(d3))
+
+	z0 -= alpha0
+
+	z1 -= alpha32
+
+	z2 -= alpha64
+
+	z3 -= alpha96
+
+	h0 += z0
+
+	h1 += z1
+
+	h3 += z2
+
+	h5 += z3
+
+	if l < 16 {
+		goto multiplyaddatmost15bytes
+	}
+
+multiplyaddatleast16bytes:
+
+	m2 = 2279
+	m20 = uint32(m[p+8])
+	y7 = h7 + alpha130
+
+	m2 <<= 51
+	m3 = 2343
+	m21 = uint32(m[p+9])
+	y6 = h6 + alpha130
+
+	m3 <<= 51
+	m0 = 2151
+	m22 = uint32(m[p+10])
+	y1 = h1 + alpha32
+
+	m0 <<= 51
+	m1 = 2215
+	m23 = uint32(m[p+11])
+	y0 = h0 + alpha32
+
+	m1 <<= 51
+	m30 = uint32(m[p+12])
+	y7 -= alpha130
+
+	m21 <<= 8
+	m2 += int64(m20)
+	m31 = uint32(m[p+13])
+	y6 -= alpha130
+
+	m22 <<= 16
+	m2 += int64(m21)
+	m32 = uint32(m[p+14])
+	y1 -= alpha32
+
+	m23 <<= 24
+	m2 += int64(m22)
+	m33 = uint64(m[p+15])
+	y0 -= alpha32
+
+	m2 += int64(m23)
+	m00 = uint32(m[p+0])
+	y5 = h5 + alpha96
+
+	m31 <<= 8
+	m3 += int64(m30)
+	m01 = uint32(m[p+1])
+	y4 = h4 + alpha96
+
+	m32 <<= 16
+	m02 = uint32(m[p+2])
+	x7 = h7 - y7
+	y7 *= scale
+
+	m33 += 256
+	m03 = uint32(m[p+3])
+	x6 = h6 - y6
+	y6 *= scale
+
+	m33 <<= 24
+	m3 += int64(m31)
+	m10 = uint32(m[p+4])
+	x1 = h1 - y1
+
+	m01 <<= 8
+	m3 += int64(m32)
+	m11 = uint32(m[p+5])
+	x0 = h0 - y0
+
+	m3 += int64(m33)
+	m0 += int64(m00)
+	m12 = uint32(m[p+6])
+	y5 -= alpha96
+
+	m02 <<= 16
+	m0 += int64(m01)
+	m13 = uint32(m[p+7])
+	y4 -= alpha96
+
+	m03 <<= 24
+	m0 += int64(m02)
+	d2 = m2
+	x1 += y7
+
+	m0 += int64(m03)
+	d3 = m3
+	x0 += y6
+
+	m11 <<= 8
+	m1 += int64(m10)
+	d0 = m0
+	x7 += y5
+
+	m12 <<= 16
+	m1 += int64(m11)
+	x6 += y4
+
+	m13 <<= 24
+	m1 += int64(m12)
+	y3 = h3 + alpha64
+
+	m1 += int64(m13)
+	d1 = m1
+	y2 = h2 + alpha64
+
+	x0 += x1
+
+	x6 += x7
+
+	y3 -= alpha64
+	r3low = r3low_stack
+
+	y2 -= alpha64
+	r0low = r0low_stack
+
+	x5 = h5 - y5
+	r3lowx0 = r3low * x0
+	r3high = r3high_stack
+
+	x4 = h4 - y4
+	r0lowx6 = r0low * x6
+	r0high = r0high_stack
+
+	x3 = h3 - y3
+	r3highx0 = r3high * x0
+	sr1low = sr1low_stack
+
+	x2 = h2 - y2
+	r0highx6 = r0high * x6
+	sr1high = sr1high_stack
+
+	x5 += y3
+	r0lowx0 = r0low * x0
+	r1low = r1low_stack
+
+	h6 = r3lowx0 + r0lowx6
+	sr1lowx6 = sr1low * x6
+	r1high = r1high_stack
+
+	x4 += y2
+	r0highx0 = r0high * x0
+	sr2low = sr2low_stack
+
+	h7 = r3highx0 + r0highx6
+	sr1highx6 = sr1high * x6
+	sr2high = sr2high_stack
+
+	x3 += y1
+	r1lowx0 = r1low * x0
+	r2low = r2low_stack
+
+	h0 = r0lowx0 + sr1lowx6
+	sr2lowx6 = sr2low * x6
+	r2high = r2high_stack
+
+	x2 += y0
+	r1highx0 = r1high * x0
+	sr3low = sr3low_stack
+
+	h1 = r0highx0 + sr1highx6
+	sr2highx6 = sr2high * x6
+	sr3high = sr3high_stack
+
+	x4 += x5
+	r2lowx0 = r2low * x0
+	z2 = math.Float64frombits(uint64(d2))
+
+	h2 = r1lowx0 + sr2lowx6
+	sr3lowx6 = sr3low * x6
+
+	x2 += x3
+	r2highx0 = r2high * x0
+	z3 = math.Float64frombits(uint64(d3))
+
+	h3 = r1highx0 + sr2highx6
+	sr3highx6 = sr3high * x6
+
+	r1highx4 = r1high * x4
+	z2 -= alpha64
+
+	h4 = r2lowx0 + sr3lowx6
+	r1lowx4 = r1low * x4
+
+	r0highx4 = r0high * x4
+	z3 -= alpha96
+
+	h5 = r2highx0 + sr3highx6
+	r0lowx4 = r0low * x4
+
+	h7 += r1highx4
+	sr3highx4 = sr3high * x4
+
+	h6 += r1lowx4
+	sr3lowx4 = sr3low * x4
+
+	h5 += r0highx4
+	sr2highx4 = sr2high * x4
+
+	h4 += r0lowx4
+	sr2lowx4 = sr2low * x4
+
+	h3 += sr3highx4
+	r0lowx2 = r0low * x2
+
+	h2 += sr3lowx4
+	r0highx2 = r0high * x2
+
+	h1 += sr2highx4
+	r1lowx2 = r1low * x2
+
+	h0 += sr2lowx4
+	r1highx2 = r1high * x2
+
+	h2 += r0lowx2
+	r2lowx2 = r2low * x2
+
+	h3 += r0highx2
+	r2highx2 = r2high * x2
+
+	h4 += r1lowx2
+	sr3lowx2 = sr3low * x2
+
+	h5 += r1highx2
+	sr3highx2 = sr3high * x2
+
+	p += 16
+	l -= 16
+	h6 += r2lowx2
+
+	h7 += r2highx2
+
+	z1 = math.Float64frombits(uint64(d1))
+	h0 += sr3lowx2
+
+	z0 = math.Float64frombits(uint64(d0))
+	h1 += sr3highx2
+
+	z1 -= alpha32
+
+	z0 -= alpha0
+
+	h5 += z3
+
+	h3 += z2
+
+	h1 += z1
+
+	h0 += z0
+
+	if l >= 16 {
+		goto multiplyaddatleast16bytes
+	}
+
+multiplyaddatmost15bytes:
+
+	y7 = h7 + alpha130
+
+	y6 = h6 + alpha130
+
+	y1 = h1 + alpha32
+
+	y0 = h0 + alpha32
+
+	y7 -= alpha130
+
+	y6 -= alpha130
+
+	y1 -= alpha32
+
+	y0 -= alpha32
+
+	y5 = h5 + alpha96
+
+	y4 = h4 + alpha96
+
+	x7 = h7 - y7
+	y7 *= scale
+
+	x6 = h6 - y6
+	y6 *= scale
+
+	x1 = h1 - y1
+
+	x0 = h0 - y0
+
+	y5 -= alpha96
+
+	y4 -= alpha96
+
+	x1 += y7
+
+	x0 += y6
+
+	x7 += y5
+
+	x6 += y4
+
+	y3 = h3 + alpha64
+
+	y2 = h2 + alpha64
+
+	x0 += x1
+
+	x6 += x7
+
+	y3 -= alpha64
+	r3low = r3low_stack
+
+	y2 -= alpha64
+	r0low = r0low_stack
+
+	x5 = h5 - y5
+	r3lowx0 = r3low * x0
+	r3high = r3high_stack
+
+	x4 = h4 - y4
+	r0lowx6 = r0low * x6
+	r0high = r0high_stack
+
+	x3 = h3 - y3
+	r3highx0 = r3high * x0
+	sr1low = sr1low_stack
+
+	x2 = h2 - y2
+	r0highx6 = r0high * x6
+	sr1high = sr1high_stack
+
+	x5 += y3
+	r0lowx0 = r0low * x0
+	r1low = r1low_stack
+
+	h6 = r3lowx0 + r0lowx6
+	sr1lowx6 = sr1low * x6
+	r1high = r1high_stack
+
+	x4 += y2
+	r0highx0 = r0high * x0
+	sr2low = sr2low_stack
+
+	h7 = r3highx0 + r0highx6
+	sr1highx6 = sr1high * x6
+	sr2high = sr2high_stack
+
+	x3 += y1
+	r1lowx0 = r1low * x0
+	r2low = r2low_stack
+
+	h0 = r0lowx0 + sr1lowx6
+	sr2lowx6 = sr2low * x6
+	r2high = r2high_stack
+
+	x2 += y0
+	r1highx0 = r1high * x0
+	sr3low = sr3low_stack
+
+	h1 = r0highx0 + sr1highx6
+	sr2highx6 = sr2high * x6
+	sr3high = sr3high_stack
+
+	x4 += x5
+	r2lowx0 = r2low * x0
+
+	h2 = r1lowx0 + sr2lowx6
+	sr3lowx6 = sr3low * x6
+
+	x2 += x3
+	r2highx0 = r2high * x0
+
+	h3 = r1highx0 + sr2highx6
+	sr3highx6 = sr3high * x6
+
+	r1highx4 = r1high * x4
+
+	h4 = r2lowx0 + sr3lowx6
+	r1lowx4 = r1low * x4
+
+	r0highx4 = r0high * x4
+
+	h5 = r2highx0 + sr3highx6
+	r0lowx4 = r0low * x4
+
+	h7 += r1highx4
+	sr3highx4 = sr3high * x4
+
+	h6 += r1lowx4
+	sr3lowx4 = sr3low * x4
+
+	h5 += r0highx4
+	sr2highx4 = sr2high * x4
+
+	h4 += r0lowx4
+	sr2lowx4 = sr2low * x4
+
+	h3 += sr3highx4
+	r0lowx2 = r0low * x2
+
+	h2 += sr3lowx4
+	r0highx2 = r0high * x2
+
+	h1 += sr2highx4
+	r1lowx2 = r1low * x2
+
+	h0 += sr2lowx4
+	r1highx2 = r1high * x2
+
+	h2 += r0lowx2
+	r2lowx2 = r2low * x2
+
+	h3 += r0highx2
+	r2highx2 = r2high * x2
+
+	h4 += r1lowx2
+	sr3lowx2 = sr3low * x2
+
+	h5 += r1highx2
+	sr3highx2 = sr3high * x2
+
+	h6 += r2lowx2
+
+	h7 += r2highx2
+
+	h0 += sr3lowx2
+
+	h1 += sr3highx2
+
+addatmost15bytes:
+
+	if l == 0 {
+		goto nomorebytes
+	}
+
+	lbelow2 = l - 2
+
+	lbelow3 = l - 3
+
+	lbelow2 >>= 31
+	lbelow4 = l - 4
+
+	m00 = uint32(m[p+0])
+	lbelow3 >>= 31
+	p += lbelow2
+
+	m01 = uint32(m[p+1])
+	lbelow4 >>= 31
+	p += lbelow3
+
+	m02 = uint32(m[p+2])
+	p += lbelow4
+	m0 = 2151
+
+	m03 = uint32(m[p+3])
+	m0 <<= 51
+	m1 = 2215
+
+	m0 += int64(m00)
+	m01 &^= uint32(lbelow2)
+
+	m02 &^= uint32(lbelow3)
+	m01 -= uint32(lbelow2)
+
+	m01 <<= 8
+	m03 &^= uint32(lbelow4)
+
+	m0 += int64(m01)
+	lbelow2 -= lbelow3
+
+	m02 += uint32(lbelow2)
+	lbelow3 -= lbelow4
+
+	m02 <<= 16
+	m03 += uint32(lbelow3)
+
+	m03 <<= 24
+	m0 += int64(m02)
+
+	m0 += int64(m03)
+	lbelow5 = l - 5
+
+	lbelow6 = l - 6
+	lbelow7 = l - 7
+
+	lbelow5 >>= 31
+	lbelow8 = l - 8
+
+	lbelow6 >>= 31
+	p += lbelow5
+
+	m10 = uint32(m[p+4])
+	lbelow7 >>= 31
+	p += lbelow6
+
+	m11 = uint32(m[p+5])
+	lbelow8 >>= 31
+	p += lbelow7
+
+	m12 = uint32(m[p+6])
+	m1 <<= 51
+	p += lbelow8
+
+	m13 = uint32(m[p+7])
+	m10 &^= uint32(lbelow5)
+	lbelow4 -= lbelow5
+
+	m10 += uint32(lbelow4)
+	lbelow5 -= lbelow6
+
+	m11 &^= uint32(lbelow6)
+	m11 += uint32(lbelow5)
+
+	m11 <<= 8
+	m1 += int64(m10)
+
+	m1 += int64(m11)
+	m12 &^= uint32(lbelow7)
+
+	lbelow6 -= lbelow7
+	m13 &^= uint32(lbelow8)
+
+	m12 += uint32(lbelow6)
+	lbelow7 -= lbelow8
+
+	m12 <<= 16
+	m13 += uint32(lbelow7)
+
+	m13 <<= 24
+	m1 += int64(m12)
+
+	m1 += int64(m13)
+	m2 = 2279
+
+	lbelow9 = l - 9
+	m3 = 2343
+
+	lbelow10 = l - 10
+	lbelow11 = l - 11
+
+	lbelow9 >>= 31
+	lbelow12 = l - 12
+
+	lbelow10 >>= 31
+	p += lbelow9
+
+	m20 = uint32(m[p+8])
+	lbelow11 >>= 31
+	p += lbelow10
+
+	m21 = uint32(m[p+9])
+	lbelow12 >>= 31
+	p += lbelow11
+
+	m22 = uint32(m[p+10])
+	m2 <<= 51
+	p += lbelow12
+
+	m23 = uint32(m[p+11])
+	m20 &^= uint32(lbelow9)
+	lbelow8 -= lbelow9
+
+	m20 += uint32(lbelow8)
+	lbelow9 -= lbelow10
+
+	m21 &^= uint32(lbelow10)
+	m21 += uint32(lbelow9)
+
+	m21 <<= 8
+	m2 += int64(m20)
+
+	m2 += int64(m21)
+	m22 &^= uint32(lbelow11)
+
+	lbelow10 -= lbelow11
+	m23 &^= uint32(lbelow12)
+
+	m22 += uint32(lbelow10)
+	lbelow11 -= lbelow12
+
+	m22 <<= 16
+	m23 += uint32(lbelow11)
+
+	m23 <<= 24
+	m2 += int64(m22)
+
+	m3 <<= 51
+	lbelow13 = l - 13
+
+	lbelow13 >>= 31
+	lbelow14 = l - 14
+
+	lbelow14 >>= 31
+	p += lbelow13
+	lbelow15 = l - 15
+
+	m30 = uint32(m[p+12])
+	lbelow15 >>= 31
+	p += lbelow14
+
+	m31 = uint32(m[p+13])
+	p += lbelow15
+	m2 += int64(m23)
+
+	m32 = uint32(m[p+14])
+	m30 &^= uint32(lbelow13)
+	lbelow12 -= lbelow13
+
+	m30 += uint32(lbelow12)
+	lbelow13 -= lbelow14
+
+	m3 += int64(m30)
+	m31 &^= uint32(lbelow14)
+
+	m31 += uint32(lbelow13)
+	m32 &^= uint32(lbelow15)
+
+	m31 <<= 8
+	lbelow14 -= lbelow15
+
+	m3 += int64(m31)
+	m32 += uint32(lbelow14)
+	d0 = m0
+
+	m32 <<= 16
+	m33 = uint64(lbelow15 + 1)
+	d1 = m1
+
+	m33 <<= 24
+	m3 += int64(m32)
+	d2 = m2
+
+	m3 += int64(m33)
+	d3 = m3
+
+	z3 = math.Float64frombits(uint64(d3))
+
+	z2 = math.Float64frombits(uint64(d2))
+
+	z1 = math.Float64frombits(uint64(d1))
+
+	z0 = math.Float64frombits(uint64(d0))
+
+	z3 -= alpha96
+
+	z2 -= alpha64
+
+	z1 -= alpha32
+
+	z0 -= alpha0
+
+	h5 += z3
+
+	h3 += z2
+
+	h1 += z1
+
+	h0 += z0
+
+	y7 = h7 + alpha130
+
+	y6 = h6 + alpha130
+
+	y1 = h1 + alpha32
+
+	y0 = h0 + alpha32
+
+	y7 -= alpha130
+
+	y6 -= alpha130
+
+	y1 -= alpha32
+
+	y0 -= alpha32
+
+	y5 = h5 + alpha96
+
+	y4 = h4 + alpha96
+
+	x7 = h7 - y7
+	y7 *= scale
+
+	x6 = h6 - y6
+	y6 *= scale
+
+	x1 = h1 - y1
+
+	x0 = h0 - y0
+
+	y5 -= alpha96
+
+	y4 -= alpha96
+
+	x1 += y7
+
+	x0 += y6
+
+	x7 += y5
+
+	x6 += y4
+
+	y3 = h3 + alpha64
+
+	y2 = h2 + alpha64
+
+	x0 += x1
+
+	x6 += x7
+
+	y3 -= alpha64
+	r3low = r3low_stack
+
+	y2 -= alpha64
+	r0low = r0low_stack
+
+	x5 = h5 - y5
+	r3lowx0 = r3low * x0
+	r3high = r3high_stack
+
+	x4 = h4 - y4
+	r0lowx6 = r0low * x6
+	r0high = r0high_stack
+
+	x3 = h3 - y3
+	r3highx0 = r3high * x0
+	sr1low = sr1low_stack
+
+	x2 = h2 - y2
+	r0highx6 = r0high * x6
+	sr1high = sr1high_stack
+
+	x5 += y3
+	r0lowx0 = r0low * x0
+	r1low = r1low_stack
+
+	h6 = r3lowx0 + r0lowx6
+	sr1lowx6 = sr1low * x6
+	r1high = r1high_stack
+
+	x4 += y2
+	r0highx0 = r0high * x0
+	sr2low = sr2low_stack
+
+	h7 = r3highx0 + r0highx6
+	sr1highx6 = sr1high * x6
+	sr2high = sr2high_stack
+
+	x3 += y1
+	r1lowx0 = r1low * x0
+	r2low = r2low_stack
+
+	h0 = r0lowx0 + sr1lowx6
+	sr2lowx6 = sr2low * x6
+	r2high = r2high_stack
+
+	x2 += y0
+	r1highx0 = r1high * x0
+	sr3low = sr3low_stack
+
+	h1 = r0highx0 + sr1highx6
+	sr2highx6 = sr2high * x6
+	sr3high = sr3high_stack
+
+	x4 += x5
+	r2lowx0 = r2low * x0
+
+	h2 = r1lowx0 + sr2lowx6
+	sr3lowx6 = sr3low * x6
+
+	x2 += x3
+	r2highx0 = r2high * x0
+
+	h3 = r1highx0 + sr2highx6
+	sr3highx6 = sr3high * x6
+
+	r1highx4 = r1high * x4
+
+	h4 = r2lowx0 + sr3lowx6
+	r1lowx4 = r1low * x4
+
+	r0highx4 = r0high * x4
+
+	h5 = r2highx0 + sr3highx6
+	r0lowx4 = r0low * x4
+
+	h7 += r1highx4
+	sr3highx4 = sr3high * x4
+
+	h6 += r1lowx4
+	sr3lowx4 = sr3low * x4
+
+	h5 += r0highx4
+	sr2highx4 = sr2high * x4
+
+	h4 += r0lowx4
+	sr2lowx4 = sr2low * x4
+
+	h3 += sr3highx4
+	r0lowx2 = r0low * x2
+
+	h2 += sr3lowx4
+	r0highx2 = r0high * x2
+
+	h1 += sr2highx4
+	r1lowx2 = r1low * x2
+
+	h0 += sr2lowx4
+	r1highx2 = r1high * x2
+
+	h2 += r0lowx2
+	r2lowx2 = r2low * x2
+
+	h3 += r0highx2
+	r2highx2 = r2high * x2
+
+	h4 += r1lowx2
+	sr3lowx2 = sr3low * x2
+
+	h5 += r1highx2
+	sr3highx2 = sr3high * x2
+
+	h6 += r2lowx2
+
+	h7 += r2highx2
+
+	h0 += sr3lowx2
+
+	h1 += sr3highx2
+
+nomorebytes:
+
+	y7 = h7 + alpha130
+
+	y0 = h0 + alpha32
+
+	y1 = h1 + alpha32
+
+	y2 = h2 + alpha64
+
+	y7 -= alpha130
+
+	y3 = h3 + alpha64
+
+	y4 = h4 + alpha96
+
+	y5 = h5 + alpha96
+
+	x7 = h7 - y7
+	y7 *= scale
+
+	y0 -= alpha32
+
+	y1 -= alpha32
+
+	y2 -= alpha64
+
+	h6 += x7
+
+	y3 -= alpha64
+
+	y4 -= alpha96
+
+	y5 -= alpha96
+
+	y6 = h6 + alpha130
+
+	x0 = h0 - y0
+
+	x1 = h1 - y1
+
+	x2 = h2 - y2
+
+	y6 -= alpha130
+
+	x0 += y7
+
+	x3 = h3 - y3
+
+	x4 = h4 - y4
+
+	x5 = h5 - y5
+
+	x6 = h6 - y6
+
+	y6 *= scale
+
+	x2 += y0
+
+	x3 += y1
+
+	x4 += y2
+
+	x0 += y6
+
+	x5 += y3
+
+	x6 += y4
+
+	x2 += x3
+
+	x0 += x1
+
+	x4 += x5
+
+	x6 += y5
+
+	x2 += offset1
+	d1 = int64(math.Float64bits(x2))
+
+	x0 += offset0
+	d0 = int64(math.Float64bits(x0))
+
+	x4 += offset2
+	d2 = int64(math.Float64bits(x4))
+
+	x6 += offset3
+	d3 = int64(math.Float64bits(x6))
+
+	f0 = uint64(d0)
+
+	f1 = uint64(d1)
+	bits32 = math.MaxUint64
+
+	f2 = uint64(d2)
+	bits32 >>= 32
+
+	f3 = uint64(d3)
+	f = f0 >> 32
+
+	f0 &= bits32
+	f &= 255
+
+	f1 += f
+	g0 = f0 + 5
+
+	g = g0 >> 32
+	g0 &= bits32
+
+	f = f1 >> 32
+	f1 &= bits32
+
+	f &= 255
+	g1 = f1 + g
+
+	g = g1 >> 32
+	f2 += f
+
+	f = f2 >> 32
+	g1 &= bits32
+
+	f2 &= bits32
+	f &= 255
+
+	f3 += f
+	g2 = f2 + g
+
+	g = g2 >> 32
+	g2 &= bits32
+
+	f4 = f3 >> 32
+	f3 &= bits32
+
+	f4 &= 255
+	g3 = f3 + g
+
+	g = g3 >> 32
+	g3 &= bits32
+
+	g4 = f4 + g
+
+	g4 = g4 - 4
+	s00 = uint32(s[0])
+
+	f = uint64(int64(g4) >> 63)
+	s01 = uint32(s[1])
+
+	f0 &= f
+	g0 &^= f
+	s02 = uint32(s[2])
+
+	f1 &= f
+	f0 |= g0
+	s03 = uint32(s[3])
+
+	g1 &^= f
+	f2 &= f
+	s10 = uint32(s[4])
+
+	f3 &= f
+	g2 &^= f
+	s11 = uint32(s[5])
+
+	g3 &^= f
+	f1 |= g1
+	s12 = uint32(s[6])
+
+	f2 |= g2
+	f3 |= g3
+	s13 = uint32(s[7])
+
+	s01 <<= 8
+	f0 += uint64(s00)
+	s20 = uint32(s[8])
+
+	s02 <<= 16
+	f0 += uint64(s01)
+	s21 = uint32(s[9])
+
+	s03 <<= 24
+	f0 += uint64(s02)
+	s22 = uint32(s[10])
+
+	s11 <<= 8
+	f1 += uint64(s10)
+	s23 = uint32(s[11])
+
+	s12 <<= 16
+	f1 += uint64(s11)
+	s30 = uint32(s[12])
+
+	s13 <<= 24
+	f1 += uint64(s12)
+	s31 = uint32(s[13])
+
+	f0 += uint64(s03)
+	f1 += uint64(s13)
+	s32 = uint32(s[14])
+
+	s21 <<= 8
+	f2 += uint64(s20)
+	s33 = uint32(s[15])
+
+	s22 <<= 16
+	f2 += uint64(s21)
+
+	s23 <<= 24
+	f2 += uint64(s22)
+
+	s31 <<= 8
+	f3 += uint64(s30)
+
+	s32 <<= 16
+	f3 += uint64(s31)
+
+	s33 <<= 24
+	f3 += uint64(s32)
+
+	f2 += uint64(s23)
+	f3 += uint64(s33)
+
+	out[0] = byte(f0)
+	f0 >>= 8
+	out[1] = byte(f0)
+	f0 >>= 8
+	out[2] = byte(f0)
+	f0 >>= 8
+	out[3] = byte(f0)
+	f0 >>= 8
+	f1 += f0
+
+	out[4] = byte(f1)
+	f1 >>= 8
+	out[5] = byte(f1)
+	f1 >>= 8
+	out[6] = byte(f1)
+	f1 >>= 8
+	out[7] = byte(f1)
+	f1 >>= 8
+	f2 += f1
+
+	out[8] = byte(f2)
+	f2 >>= 8
+	out[9] = byte(f2)
+	f2 >>= 8
+	out[10] = byte(f2)
+	f2 >>= 8
+	out[11] = byte(f2)
+	f2 >>= 8
+	f3 += f2
+
+	out[12] = byte(f3)
+	f3 >>= 8
+	out[13] = byte(f3)
+	f3 >>= 8
+	out[14] = byte(f3)
+	f3 >>= 8
+	out[15] = byte(f3)
+}
diff --git a/ssl/test/runner/runner.go b/ssl/test/runner/runner.go
index 3b8aa9e..650c14f 100644
--- a/ssl/test/runner/runner.go
+++ b/ssl/test/runner/runner.go
@@ -1474,12 +1474,14 @@
 	{"DHE-RSA-AES256-GCM", TLS_DHE_RSA_WITH_AES_256_GCM_SHA384},
 	{"DHE-RSA-AES256-SHA", TLS_DHE_RSA_WITH_AES_256_CBC_SHA},
 	{"DHE-RSA-AES256-SHA256", TLS_DHE_RSA_WITH_AES_256_CBC_SHA256},
+	{"DHE-RSA-CHACHA20-POLY1305", TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256},
 	{"ECDHE-ECDSA-AES128-GCM", TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256},
 	{"ECDHE-ECDSA-AES128-SHA", TLS_ECDHE_ECDSA_WITH_AES_128_CBC_SHA},
 	{"ECDHE-ECDSA-AES128-SHA256", TLS_ECDHE_ECDSA_WITH_AES_128_CBC_SHA256},
 	{"ECDHE-ECDSA-AES256-GCM", TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384},
 	{"ECDHE-ECDSA-AES256-SHA", TLS_ECDHE_ECDSA_WITH_AES_256_CBC_SHA},
 	{"ECDHE-ECDSA-AES256-SHA384", TLS_ECDHE_ECDSA_WITH_AES_256_CBC_SHA384},
+	{"ECDHE-ECDSA-CHACHA20-POLY1305", TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256},
 	{"ECDHE-ECDSA-RC4-SHA", TLS_ECDHE_ECDSA_WITH_RC4_128_SHA},
 	{"ECDHE-PSK-WITH-AES-128-GCM-SHA256", TLS_ECDHE_PSK_WITH_AES_128_GCM_SHA256},
 	{"ECDHE-RSA-AES128-GCM", TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256},
@@ -1488,6 +1490,7 @@
 	{"ECDHE-RSA-AES256-GCM", TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384},
 	{"ECDHE-RSA-AES256-SHA", TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA},
 	{"ECDHE-RSA-AES256-SHA384", TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA384},
+	{"ECDHE-RSA-CHACHA20-POLY1305", TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256},
 	{"ECDHE-RSA-RC4-SHA", TLS_ECDHE_RSA_WITH_RC4_128_SHA},
 	{"PSK-AES128-CBC-SHA", TLS_PSK_WITH_AES_128_CBC_SHA},
 	{"PSK-AES256-CBC-SHA", TLS_PSK_WITH_AES_256_CBC_SHA},
@@ -1503,7 +1506,8 @@
 func isTLS12Only(suiteName string) bool {
 	return hasComponent(suiteName, "GCM") ||
 		hasComponent(suiteName, "SHA256") ||
-		hasComponent(suiteName, "SHA384")
+		hasComponent(suiteName, "SHA384") ||
+		hasComponent(suiteName, "POLY1305")
 }
 
 func isDTLSCipher(suiteName string) bool {