src/vendor,crypto/tls: update to latest x/crypto and use new X25519 API

Change-Id: Icd5006e37861d892a5f3d4397c3826179c1b12ad
Reviewed-on: https://go-review.googlesource.com/c/go/+/206657
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Katie Hockman <katie@golang.org>

src/crypto/tls/key_schedule.go

@@ -7,14 +7,14 @@ package tls
 import (
 	"crypto/elliptic"
 	"crypto/hmac"
-	"crypto/subtle"
 	"errors"
-	"golang.org/x/crypto/cryptobyte"
-	"golang.org/x/crypto/curve25519"
-	"golang.org/x/crypto/hkdf"
 	"hash"
 	"io"
 	"math/big"
+
+	"golang.org/x/crypto/cryptobyte"
+	"golang.org/x/crypto/curve25519"
+	"golang.org/x/crypto/hkdf"
 )
 
 // This file contains the functions necessary to compute the TLS 1.3 key
@@ -111,12 +111,15 @@ type ecdheParameters interface {
 
 func generateECDHEParameters(rand io.Reader, curveID CurveID) (ecdheParameters, error) {
 	if curveID == X25519 {
-		p := &x25519Parameters{}
-		if _, err := io.ReadFull(rand, p.privateKey[:]); err != nil {
+		privateKey := make([]byte, curve25519.ScalarSize)
+		if _, err := io.ReadFull(rand, privateKey); err != nil {
 			return nil, err
 		}
-		curve25519.ScalarBaseMult(&p.publicKey, &p.privateKey)
-		return p, nil
+		publicKey, err := curve25519.X25519(privateKey, curve25519.Basepoint)
+		if err != nil {
+			return nil, err
+		}
+		return &x25519Parameters{privateKey: privateKey, publicKey: publicKey}, nil
 	}
 
 	curve, ok := curveForCurveID(curveID)
@@ -178,8 +181,8 @@ func (p *nistParameters) SharedKey(peerPublicKey []byte) []byte {
 }
 
 type x25519Parameters struct {
-	privateKey [32]byte
-	publicKey  [32]byte
+	privateKey []byte
+	publicKey  []byte
 }
 
 func (p *x25519Parameters) CurveID() CurveID {
@@ -191,19 +194,9 @@ func (p *x25519Parameters) PublicKey() []byte {
 }
 
 func (p *x25519Parameters) SharedKey(peerPublicKey []byte) []byte {
-	if len(peerPublicKey) != 32 {
-		return nil
-	}
-
-	var theirPublicKey, sharedKey [32]byte
-	copy(theirPublicKey[:], peerPublicKey)
-	curve25519.ScalarMult(&sharedKey, &p.privateKey, &theirPublicKey)
-
-	// Check for low-order inputs. See RFC 8422, Section 5.11.
-	var allZeroes [32]byte
-	if subtle.ConstantTimeCompare(allZeroes[:], sharedKey[:]) == 1 {
+	sharedKey, err := curve25519.X25519(p.privateKey, peerPublicKey)
+	if err != nil {
 		return nil
 	}
-
-	return sharedKey[:]
+	return sharedKey
 }

src/go.mod

@@ -3,7 +3,7 @@ module std
 go 1.14
 
 require (
-	golang.org/x/crypto v0.0.0-20190611184440-5c40567a22f8
+	golang.org/x/crypto v0.0.0-20191111213947-16651526fdb4
 	golang.org/x/net v0.0.0-20191105084925-a882066a44e0
 	golang.org/x/sys v0.0.0-20190529130038-5219a1e1c5f8 // indirect
 	golang.org/x/text v0.3.3-0.20191031172631-4b67af870c6f // indirect

src/go.sum

 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.0.0-20190611184440-5c40567a22f8 h1:1wopBVtVdWnn03fZelqdXTqk7U7zPQCb+T4rbU9ZEoU=
-golang.org/x/crypto v0.0.0-20190611184440-5c40567a22f8/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
+golang.org/x/crypto v0.0.0-20191111213947-16651526fdb4 h1:AGVXd+IAyeAb3FuQvYDYQ9+WR2JHm0+C0oYJaU1C4rs=
+golang.org/x/crypto v0.0.0-20191111213947-16651526fdb4/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20191105084925-a882066a44e0 h1:QPlSTtPE2k6PZPasQUbzuK3p9JbS+vMXYVto8g/yrsg=
 golang.org/x/net v0.0.0-20191105084925-a882066a44e0/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=

src/vendor/golang.org/x/crypto/internal/chacha20/chacha_arm64.go → src/vendor/golang.org/x/crypto/chacha20/chacha_arm64.go

@@ -3,29 +3,15 @@
 // license that can be found in the LICENSE file.
 
 // +build go1.11
-// +build !gccgo
+// +build !gccgo,!appengine
 
 package chacha20
 
-const (
-	haveAsm = true
-	bufSize = 256
-)
+const bufSize = 256
 
 //go:noescape
 func xorKeyStreamVX(dst, src []byte, key *[8]uint32, nonce *[3]uint32, counter *uint32)
 
-func (c *Cipher) xorKeyStreamAsm(dst, src []byte) {
-
-	if len(src) >= bufSize {
+func (c *Cipher) xorKeyStreamBlocks(dst, src []byte) {
 	xorKeyStreamVX(dst, src, &c.key, &c.nonce, &c.counter)
-	}
-
-	if len(src)%bufSize != 0 {
-		i := len(src) - len(src)%bufSize
-		c.buf = [bufSize]byte{}
-		copy(c.buf[:], src[i:])
-		xorKeyStreamVX(c.buf[:], c.buf[:], &c.key, &c.nonce, &c.counter)
-		c.len = bufSize - copy(dst[i:], c.buf[:len(src)%bufSize])
-	}
 }

src/vendor/golang.org/x/crypto/chacha20/chacha_generic.go

+// Copyright 2016 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 chacha20 implements the ChaCha20 and XChaCha20 encryption algorithms
+// as specified in RFC 8439 and draft-irtf-cfrg-xchacha-01.
+package chacha20
+
+import (
+	"crypto/cipher"
+	"encoding/binary"
+	"errors"
+	"math/bits"
+
+	"golang.org/x/crypto/internal/subtle"
+)
+
+const (
+	// KeySize is the size of the key used by this cipher, in bytes.
+	KeySize = 32
+
+	// NonceSize is the size of the nonce used with the standard variant of this
+	// cipher, in bytes.
+	//
+	// Note that this is too short to be safely generated at random if the same
+	// key is reused more than 2³² times.
+	NonceSize = 12
+
+	// NonceSizeX is the size of the nonce used with the XChaCha20 variant of
+	// this cipher, in bytes.
+	NonceSizeX = 24
+)
+
+// Cipher is a stateful instance of ChaCha20 or XChaCha20 using a particular key
+// and nonce. A *Cipher implements the cipher.Stream interface.
+type Cipher struct {
+	// The ChaCha20 state is 16 words: 4 constant, 8 of key, 1 of counter
+	// (incremented after each block), and 3 of nonce.
+	key     [8]uint32
+	counter uint32
+	nonce   [3]uint32
+
+	// The last len bytes of buf are leftover key stream bytes from the previous
+	// XORKeyStream invocation. The size of buf depends on how many blocks are
+	// computed at a time.
+	buf [bufSize]byte
+	len int
+
+	// The counter-independent results of the first round are cached after they
+	// are computed the first time.
+	precompDone      bool
+	p1, p5, p9, p13  uint32
+	p2, p6, p10, p14 uint32
+	p3, p7, p11, p15 uint32
+}
+
+var _ cipher.Stream = (*Cipher)(nil)
+
+// NewUnauthenticatedCipher creates a new ChaCha20 stream cipher with the given
+// 32 bytes key and a 12 or 24 bytes nonce. If a nonce of 24 bytes is provided,
+// the XChaCha20 construction will be used. It returns an error if key or nonce
+// have any other length.
+//
+// Note that ChaCha20, like all stream ciphers, is not authenticated and allows
+// attackers to silently tamper with the plaintext. For this reason, it is more
+// appropriate as a building block than as a standalone encryption mechanism.
+// Instead, consider using package golang.org/x/crypto/chacha20poly1305.
+func NewUnauthenticatedCipher(key, nonce []byte) (*Cipher, error) {
+	// This function is split into a wrapper so that the Cipher allocation will
+	// be inlined, and depending on how the caller uses the return value, won't
+	// escape to the heap.
+	c := &Cipher{}
+	return newUnauthenticatedCipher(c, key, nonce)
+}
+
+func newUnauthenticatedCipher(c *Cipher, key, nonce []byte) (*Cipher, error) {
+	if len(key) != KeySize {
+		return nil, errors.New("chacha20: wrong key size")
+	}
+	if len(nonce) == NonceSizeX {
+		// XChaCha20 uses the ChaCha20 core to mix 16 bytes of the nonce into a
+		// derived key, allowing it to operate on a nonce of 24 bytes. See
+		// draft-irtf-cfrg-xchacha-01, Section 2.3.
+		key, _ = HChaCha20(key, nonce[0:16])
+		cNonce := make([]byte, NonceSize)
+		copy(cNonce[4:12], nonce[16:24])
+		nonce = cNonce
+	} else if len(nonce) != NonceSize {
+		return nil, errors.New("chacha20: wrong nonce size")
+	}
+
+	c.key = [8]uint32{
+		binary.LittleEndian.Uint32(key[0:4]),
+		binary.LittleEndian.Uint32(key[4:8]),
+		binary.LittleEndian.Uint32(key[8:12]),
+		binary.LittleEndian.Uint32(key[12:16]),
+		binary.LittleEndian.Uint32(key[16:20]),
+		binary.LittleEndian.Uint32(key[20:24]),
+		binary.LittleEndian.Uint32(key[24:28]),
+		binary.LittleEndian.Uint32(key[28:32]),
+	}
+	c.nonce = [3]uint32{
+		binary.LittleEndian.Uint32(nonce[0:4]),
+		binary.LittleEndian.Uint32(nonce[4:8]),
+		binary.LittleEndian.Uint32(nonce[8:12]),
+	}
+	return c, nil
+}
+
+// The constant first 4 words of the ChaCha20 state.
+const (
+	j0 uint32 = 0x61707865 // expa
+	j1 uint32 = 0x3320646e // nd 3
+	j2 uint32 = 0x79622d32 // 2-by
+	j3 uint32 = 0x6b206574 // te k
+)
+
+const blockSize = 64
+
+// quarterRound is the core of ChaCha20. It shuffles the bits of 4 state words.
+// It's executed 4 times for each of the 20 ChaCha20 rounds, operating on all 16
+// words each round, in columnar or diagonal groups of 4 at a time.
+func quarterRound(a, b, c, d uint32) (uint32, uint32, uint32, uint32) {
+	a += b
+	d ^= a
+	d = bits.RotateLeft32(d, 16)
+	c += d
+	b ^= c
+	b = bits.RotateLeft32(b, 12)
+	a += b
+	d ^= a
+	d = bits.RotateLeft32(d, 8)
+	c += d
+	b ^= c
+	b = bits.RotateLeft32(b, 7)
+	return a, b, c, d
+}
+
+// XORKeyStream XORs each byte in the given slice with a byte from the
+// cipher's key stream. Dst and src must overlap entirely or not at all.
+//
+// If len(dst) < len(src), XORKeyStream will panic. It is acceptable
+// to pass a dst bigger than src, and in that case, XORKeyStream will
+// only update dst[:len(src)] and will not touch the rest of dst.
+//
+// Multiple calls to XORKeyStream behave as if the concatenation of
+// the src buffers was passed in a single run. That is, Cipher
+// maintains state and does not reset at each XORKeyStream call.
+func (s *Cipher) XORKeyStream(dst, src []byte) {
+	if len(src) == 0 {
+		return
+	}
+	if len(dst) < len(src) {
+		panic("chacha20: output smaller than input")
+	}
+	dst = dst[:len(src)]
+	if subtle.InexactOverlap(dst, src) {
+		panic("chacha20: invalid buffer overlap")
+	}
+
+	// First, drain any remaining key stream from a previous XORKeyStream.
+	if s.len != 0 {
+		keyStream := s.buf[bufSize-s.len:]
+		if len(src) < len(keyStream) {
+			keyStream = keyStream[:len(src)]
+		}
+		_ = src[len(keyStream)-1] // bounds check elimination hint
+		for i, b := range keyStream {
+			dst[i] = src[i] ^ b
+		}
+		s.len -= len(keyStream)
+		src = src[len(keyStream):]
+		dst = dst[len(keyStream):]
+	}
+
+	const blocksPerBuf = bufSize / blockSize
+	numBufs := (uint64(len(src)) + bufSize - 1) / bufSize
+	if uint64(s.counter)+numBufs*blocksPerBuf >= 1<<32 {
+		panic("chacha20: counter overflow")
+	}
+
+	// xorKeyStreamBlocks implementations expect input lengths that are a
+	// multiple of bufSize. Platform-specific ones process multiple blocks at a
+	// time, so have bufSizes that are a multiple of blockSize.
+
+	rem := len(src) % bufSize
+	full := len(src) - rem
+
+	if full > 0 {
+		s.xorKeyStreamBlocks(dst[:full], src[:full])
+	}
+
+	// If we have a partial (multi-)block, pad it for xorKeyStreamBlocks, and
+	// keep the leftover keystream for the next XORKeyStream invocation.
+	if rem > 0 {
+		s.buf = [bufSize]byte{}
+		copy(s.buf[:], src[full:])
+		s.xorKeyStreamBlocks(s.buf[:], s.buf[:])
+		s.len = bufSize - copy(dst[full:], s.buf[:])
+	}
+}
+
+func (s *Cipher) xorKeyStreamBlocksGeneric(dst, src []byte) {
+	if len(dst) != len(src) || len(dst)%blockSize != 0 {
+		panic("chacha20: internal error: wrong dst and/or src length")
+	}
+
+	// To generate each block of key stream, the initial cipher state
+	// (represented below) is passed through 20 rounds of shuffling,
+	// alternatively applying quarterRounds by columns (like 1, 5, 9, 13)
+	// or by diagonals (like 1, 6, 11, 12).
+	//
+	//      0:cccccccc   1:cccccccc   2:cccccccc   3:cccccccc
+	//      4:kkkkkkkk   5:kkkkkkkk   6:kkkkkkkk   7:kkkkkkkk
+	//      8:kkkkkkkk   9:kkkkkkkk  10:kkkkkkkk  11:kkkkkkkk
+	//     12:bbbbbbbb  13:nnnnnnnn  14:nnnnnnnn  15:nnnnnnnn
+	//
+	//            c=constant k=key b=blockcount n=nonce
+	var (
+		c0, c1, c2, c3   = j0, j1, j2, j3
+		c4, c5, c6, c7   = s.key[0], s.key[1], s.key[2], s.key[3]
+		c8, c9, c10, c11 = s.key[4], s.key[5], s.key[6], s.key[7]
+		_, c13, c14, c15 = s.counter, s.nonce[0], s.nonce[1], s.nonce[2]
+	)
+
+	// Three quarters of the first round don't depend on the counter, so we can
+	// calculate them here, and reuse them for multiple blocks in the loop, and
+	// for future XORKeyStream invocations.
+	if !s.precompDone {
+		s.p1, s.p5, s.p9, s.p13 = quarterRound(c1, c5, c9, c13)
+		s.p2, s.p6, s.p10, s.p14 = quarterRound(c2, c6, c10, c14)
+		s.p3, s.p7, s.p11, s.p15 = quarterRound(c3, c7, c11, c15)
+		s.precompDone = true
+	}
+
+	for i := 0; i < len(src); i += blockSize {
+		// The remainder of the first column round.
+		fcr0, fcr4, fcr8, fcr12 := quarterRound(c0, c4, c8, s.counter)
+
+		// The second diagonal round.
+		x0, x5, x10, x15 := quarterRound(fcr0, s.p5, s.p10, s.p15)
+		x1, x6, x11, x12 := quarterRound(s.p1, s.p6, s.p11, fcr12)
+		x2, x7, x8, x13 := quarterRound(s.p2, s.p7, fcr8, s.p13)
+		x3, x4, x9, x14 := quarterRound(s.p3, fcr4, s.p9, s.p14)
+
+		// The remaining 18 rounds.
+		for i := 0; i < 9; i++ {
+			// Column round.
+			x0, x4, x8, x12 = quarterRound(x0, x4, x8, x12)
+			x1, x5, x9, x13 = quarterRound(x1, x5, x9, x13)
+			x2, x6, x10, x14 = quarterRound(x2, x6, x10, x14)
+			x3, x7, x11, x15 = quarterRound(x3, x7, x11, x15)
+
+			// Diagonal round.
+			x0, x5, x10, x15 = quarterRound(x0, x5, x10, x15)
+			x1, x6, x11, x12 = quarterRound(x1, x6, x11, x12)
+			x2, x7, x8, x13 = quarterRound(x2, x7, x8, x13)
+			x3, x4, x9, x14 = quarterRound(x3, x4, x9, x14)
+		}
+
+		// Finally, add back the initial state to generate the key stream.
+		x0 += c0
+		x1 += c1
+		x2 += c2
+		x3 += c3
+		x4 += c4
+		x5 += c5
+		x6 += c6
+		x7 += c7
+		x8 += c8
+		x9 += c9
+		x10 += c10
+		x11 += c11
+		x12 += s.counter
+		x13 += c13
+		x14 += c14
+		x15 += c15
+
+		s.counter += 1
+		if s.counter == 0 {
+			panic("chacha20: internal error: counter overflow")
+		}
+
+		in, out := src[i:], dst[i:]
+		in, out = in[:blockSize], out[:blockSize] // bounds check elimination hint
+
+		// XOR the key stream with the source and write out the result.
+		xor(out[0:], in[0:], x0)
+		xor(out[4:], in[4:], x1)
+		xor(out[8:], in[8:], x2)
+		xor(out[12:], in[12:], x3)
+		xor(out[16:], in[16:], x4)
+		xor(out[20:], in[20:], x5)
+		xor(out[24:], in[24:], x6)
+		xor(out[28:], in[28:], x7)
+		xor(out[32:], in[32:], x8)
+		xor(out[36:], in[36:], x9)
+		xor(out[40:], in[40:], x10)
+		xor(out[44:], in[44:], x11)
+		xor(out[48:], in[48:], x12)
+		xor(out[52:], in[52:], x13)
+		xor(out[56:], in[56:], x14)
+		xor(out[60:], in[60:], x15)
+	}
+}
+
+// HChaCha20 uses the ChaCha20 core to generate a derived key from a 32 bytes
+// key and a 16 bytes nonce. It returns an error if key or nonce have any other
+// length. It is used as part of the XChaCha20 construction.
+func HChaCha20(key, nonce []byte) ([]byte, error) {
+	// This function is split into a wrapper so that the slice allocation will
+	// be inlined, and depending on how the caller uses the return value, won't
+	// escape to the heap.
+	out := make([]byte, 32)
+	return hChaCha20(out, key, nonce)
+}
+
+func hChaCha20(out, key, nonce []byte) ([]byte, error) {
+	if len(key) != KeySize {
+		return nil, errors.New("chacha20: wrong HChaCha20 key size")
+	}
+	if len(nonce) != 16 {
+		return nil, errors.New("chacha20: wrong HChaCha20 nonce size")
+	}
+
+	x0, x1, x2, x3 := j0, j1, j2, j3
+	x4 := binary.LittleEndian.Uint32(key[0:4])
+	x5 := binary.LittleEndian.Uint32(key[4:8])
+	x6 := binary.LittleEndian.Uint32(key[8:12])
+	x7 := binary.LittleEndian.Uint32(key[12:16])
+	x8 := binary.LittleEndian.Uint32(key[16:20])
+	x9 := binary.LittleEndian.Uint32(key[20:24])
+	x10 := binary.LittleEndian.Uint32(key[24:28])
+	x11 := binary.LittleEndian.Uint32(key[28:32])
+	x12 := binary.LittleEndian.Uint32(nonce[0:4])
+	x13 := binary.LittleEndian.Uint32(nonce[4:8])
+	x14 := binary.LittleEndian.Uint32(nonce[8:12])
+	x15 := binary.LittleEndian.Uint32(nonce[12:16])
+
+	for i := 0; i < 10; i++ {
+		// Diagonal round.
+		x0, x4, x8, x12 = quarterRound(x0, x4, x8, x12)
+		x1, x5, x9, x13 = quarterRound(x1, x5, x9, x13)
+		x2, x6, x10, x14 = quarterRound(x2, x6, x10, x14)
+		x3, x7, x11, x15 = quarterRound(x3, x7, x11, x15)
+
+		// Column round.
+		x0, x5, x10, x15 = quarterRound(x0, x5, x10, x15)
+		x1, x6, x11, x12 = quarterRound(x1, x6, x11, x12)
+		x2, x7, x8, x13 = quarterRound(x2, x7, x8, x13)
+		x3, x4, x9, x14 = quarterRound(x3, x4, x9, x14)
+	}
+
+	_ = out[31] // bounds check elimination hint
+	binary.LittleEndian.PutUint32(out[0:4], x0)
+	binary.LittleEndian.PutUint32(out[4:8], x1)
+	binary.LittleEndian.PutUint32(out[8:12], x2)
+	binary.LittleEndian.PutUint32(out[12:16], x3)
+	binary.LittleEndian.PutUint32(out[16:20], x12)
+	binary.LittleEndian.PutUint32(out[20:24], x13)
+	binary.LittleEndian.PutUint32(out[24:28], x14)
+	binary.LittleEndian.PutUint32(out[28:32], x15)
+	return out, nil
+}

src/vendor/golang.org/x/crypto/internal/chacha20/chacha_noasm.go → src/vendor/golang.org/x/crypto/chacha20/chacha_noasm.go

@@ -2,15 +2,12 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !ppc64le,!arm64,!s390x arm64,!go1.11 gccgo appengine
+// +build !arm64,!s390x,!ppc64le arm64,!go1.11 gccgo appengine
 
 package chacha20
 
-const (
-	bufSize = 64
-	haveAsm = false
-)
+const bufSize = blockSize
 
-func (*Cipher) xorKeyStreamAsm(dst, src []byte) {
-	panic("not implemented")
+func (s *Cipher) xorKeyStreamBlocks(dst, src []byte) {
+	s.xorKeyStreamBlocksGeneric(dst, src)
 }

src/vendor/golang.org/x/crypto/chacha20/chacha_ppc64le.go

+// Copyright 2019 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.
+
+// +build !gccgo,!appengine
+
+package chacha20
+
+const bufSize = 256
+
+//go:noescape
+func chaCha20_ctr32_vsx(out, inp *byte, len int, key *[8]uint32, counter *uint32)
+
+func (c *Cipher) xorKeyStreamBlocks(dst, src []byte) {
+	chaCha20_ctr32_vsx(&dst[0], &src[0], len(src), &c.key, &c.counter)
+}

src/vendor/golang.org/x/crypto/chacha20/chacha_ppc64le.s

+// Copyright 2019 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.
+
+// Based on CRYPTOGAMS code with the following comment:
+// # ====================================================================
+// # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+// # project. The module is, however, dual licensed under OpenSSL and
+// # CRYPTOGAMS licenses depending on where you obtain it. For further
+// # details see http://www.openssl.org/~appro/cryptogams/.
+// # ====================================================================
+
+// Code for the perl script that generates the ppc64 assembler
+// can be found in the cryptogams repository at the link below. It is based on
+// the original from openssl.
+
+// https://github.com/dot-asm/cryptogams/commit/a60f5b50ed908e91
+
+// The differences in this and the original implementation are
+// due to the calling conventions and initialization of constants.
+
+// +build !gccgo,!appengine
+
+#include "textflag.h"
+
+#define OUT  R3
+#define INP  R4
+#define LEN  R5
+#define KEY  R6
+#define CNT  R7
+#define TMP  R15
+
+#define CONSTBASE  R16
+#define BLOCKS R17
+
+DATA consts<>+0x00(SB)/8, $0x3320646e61707865
+DATA consts<>+0x08(SB)/8, $0x6b20657479622d32
+DATA consts<>+0x10(SB)/8, $0x0000000000000001
+DATA consts<>+0x18(SB)/8, $0x0000000000000000
+DATA consts<>+0x20(SB)/8, $0x0000000000000004
+DATA consts<>+0x28(SB)/8, $0x0000000000000000
+DATA consts<>+0x30(SB)/8, $0x0a0b08090e0f0c0d
+DATA consts<>+0x38(SB)/8, $0x0203000106070405
+DATA consts<>+0x40(SB)/8, $0x090a0b080d0e0f0c
+DATA consts<>+0x48(SB)/8, $0x0102030005060704
+DATA consts<>+0x50(SB)/8, $0x6170786561707865
+DATA consts<>+0x58(SB)/8, $0x6170786561707865
+DATA consts<>+0x60(SB)/8, $0x3320646e3320646e
+DATA consts<>+0x68(SB)/8, $0x3320646e3320646e
+DATA consts<>+0x70(SB)/8, $0x79622d3279622d32
+DATA consts<>+0x78(SB)/8, $0x79622d3279622d32
+DATA consts<>+0x80(SB)/8, $0x6b2065746b206574
+DATA consts<>+0x88(SB)/8, $0x6b2065746b206574
+DATA consts<>+0x90(SB)/8, $0x0000000100000000
+DATA consts<>+0x98(SB)/8, $0x0000000300000002
+GLOBL consts<>(SB), RODATA, $0xa0
+
+//func chaCha20_ctr32_vsx(out, inp *byte, len int, key *[8]uint32, counter *uint32)
+TEXT ·chaCha20_ctr32_vsx(SB),NOSPLIT,$64-40
+	MOVD out+0(FP), OUT
+	MOVD inp+8(FP), INP
+	MOVD len+16(FP), LEN
+	MOVD key+24(FP), KEY
+	MOVD counter+32(FP), CNT
+
+	// Addressing for constants
+	MOVD $consts<>+0x00(SB), CONSTBASE
+	MOVD $16, R8
+	MOVD $32, R9
+	MOVD $48, R10
+	MOVD $64, R11
+	SRD $6, LEN, BLOCKS
+	// V16
+	LXVW4X (CONSTBASE)(R0), VS48
+	ADD $80,CONSTBASE
+
+	// Load key into V17,V18
+	LXVW4X (KEY)(R0), VS49
+	LXVW4X (KEY)(R8), VS50
+
+	// Load CNT, NONCE into V19
+	LXVW4X (CNT)(R0), VS51
+
+	// Clear V27
+	VXOR V27, V27, V27
+
+	// V28
+	LXVW4X (CONSTBASE)(R11), VS60
+
+	// splat slot from V19 -> V26
+	VSPLTW $0, V19, V26
+
+	VSLDOI $4, V19, V27, V19
+	VSLDOI $12, V27, V19, V19
+
+	VADDUWM V26, V28, V26
+
+	MOVD $10, R14
+	MOVD R14, CTR
+
+loop_outer_vsx:
+	// V0, V1, V2, V3
+	LXVW4X (R0)(CONSTBASE), VS32
+	LXVW4X (R8)(CONSTBASE), VS33
+	LXVW4X (R9)(CONSTBASE), VS34
+	LXVW4X (R10)(CONSTBASE), VS35
+
+	// splat values from V17, V18 into V4-V11
+	VSPLTW $0, V17, V4
+	VSPLTW $1, V17, V5
+	VSPLTW $2, V17, V6
+	VSPLTW $3, V17, V7
+	VSPLTW $0, V18, V8
+	VSPLTW $1, V18, V9
+	VSPLTW $2, V18, V10
+	VSPLTW $3, V18, V11
+
+	// VOR
+	VOR V26, V26, V12
+
+	// splat values from V19 -> V13, V14, V15
+	VSPLTW $1, V19, V13
+	VSPLTW $2, V19, V14
+	VSPLTW $3, V19, V15
+
+	// splat   const values
+	VSPLTISW $-16, V27
+	VSPLTISW $12, V28
+	VSPLTISW $8, V29
+	VSPLTISW $7, V30
+
+loop_vsx:
+	VADDUWM V0, V4, V0
+	VADDUWM V1, V5, V1
+	VADDUWM V2, V6, V2
+	VADDUWM V3, V7, V3
+
+	VXOR V12, V0, V12
+	VXOR V13, V1, V13
+	VXOR V14, V2, V14
+	VXOR V15, V3, V15
+
+	VRLW V12, V27, V12
+	VRLW V13, V27, V13
+	VRLW V14, V27, V14
+	VRLW V15, V27, V15
+
+	VADDUWM V8, V12, V8
+	VADDUWM V9, V13, V9
+	VADDUWM V10, V14, V10
+	VADDUWM V11, V15, V11
+
+	VXOR V4, V8, V4
+	VXOR V5, V9, V5
+	VXOR V6, V10, V6
+	VXOR V7, V11, V7
+
+	VRLW V4, V28, V4
+	VRLW V5, V28, V5
+	VRLW V6, V28, V6
+	VRLW V7, V28, V7
+
+	VADDUWM V0, V4, V0
+	VADDUWM V1, V5, V1
+	VADDUWM V2, V6, V2
+	VADDUWM V3, V7, V3
+
+	VXOR V12, V0, V12
+	VXOR V13, V1, V13
+	VXOR V14, V2, V14
+	VXOR V15, V3, V15
+
+	VRLW V12, V29, V12
+	VRLW V13, V29, V13
+	VRLW V14, V29, V14
+	VRLW V15, V29, V15
+
+	VADDUWM V8, V12, V8
+	VADDUWM V9, V13, V9
+	VADDUWM V10, V14, V10
+	VADDUWM V11, V15, V11
+
+	VXOR V4, V8, V4
+	VXOR V5, V9, V5
+	VXOR V6, V10, V6
+	VXOR V7, V11, V7
+
+	VRLW V4, V30, V4
+	VRLW V5, V30, V5
+	VRLW V6, V30, V6
+	VRLW V7, V30, V7
+
+	VADDUWM V0, V5, V0
+	VADDUWM V1, V6, V1
+	VADDUWM V2, V7, V2
+	VADDUWM V3, V4, V3
+
+	VXOR V15, V0, V15
+	VXOR V12, V1, V12
+	VXOR V13, V2, V13
+	VXOR V14, V3, V14
+
+	VRLW V15, V27, V15
+	VRLW V12, V27, V12
+	VRLW V13, V27, V13
+	VRLW V14, V27, V14
+
+	VADDUWM V10, V15, V10
+	VADDUWM V11, V12, V11
+	VADDUWM V8, V13, V8
+	VADDUWM V9, V14, V9
+
+	VXOR V5, V10, V5
+	VXOR V6, V11, V6
+	VXOR V7, V8, V7
+	VXOR V4, V9, V4
+
+	VRLW V5, V28, V5
+	VRLW V6, V28, V6
+	VRLW V7, V28, V7
+	VRLW V4, V28, V4
+
+	VADDUWM V0, V5, V0
+	VADDUWM V1, V6, V1
+	VADDUWM V2, V7, V2
+	VADDUWM V3, V4, V3
+
+	VXOR V15, V0, V15
+	VXOR V12, V1, V12
+	VXOR V13, V2, V13
+	VXOR V14, V3, V14
+
+	VRLW V15, V29, V15
+	VRLW V12, V29, V12
+	VRLW V13, V29, V13
+	VRLW V14, V29, V14
+
+	VADDUWM V10, V15, V10
+	VADDUWM V11, V12, V11
+	VADDUWM V8, V13, V8
+	VADDUWM V9, V14, V9
+
+	VXOR V5, V10, V5
+	VXOR V6, V11, V6
+	VXOR V7, V8, V7
+	VXOR V4, V9, V4
+
+	VRLW V5, V30, V5
+	VRLW V6, V30, V6
+	VRLW V7, V30, V7
+	VRLW V4, V30, V4
+	BC   16, LT, loop_vsx
+
+	VADDUWM V12, V26, V12
+
+	WORD $0x13600F8C		// VMRGEW V0, V1, V27
+	WORD $0x13821F8C		// VMRGEW V2, V3, V28
+
+	WORD $0x10000E8C		// VMRGOW V0, V1, V0
+	WORD $0x10421E8C		// VMRGOW V2, V3, V2
+
+	WORD $0x13A42F8C		// VMRGEW V4, V5, V29
+	WORD $0x13C63F8C		// VMRGEW V6, V7, V30
+
+	XXPERMDI VS32, VS34, $0, VS33
+	XXPERMDI VS32, VS34, $3, VS35
+	XXPERMDI VS59, VS60, $0, VS32
+	XXPERMDI VS59, VS60, $3, VS34
+
+	WORD $0x10842E8C		// VMRGOW V4, V5, V4
+	WORD $0x10C63E8C		// VMRGOW V6, V7, V6
+
+	WORD $0x13684F8C		// VMRGEW V8, V9, V27
+	WORD $0x138A5F8C		// VMRGEW V10, V11, V28
+
+	XXPERMDI VS36, VS38, $0, VS37
+	XXPERMDI VS36, VS38, $3, VS39
+	XXPERMDI VS61, VS62, $0, VS36
+	XXPERMDI VS61, VS62, $3, VS38
+
+	WORD $0x11084E8C		// VMRGOW V8, V9, V8
+	WORD $0x114A5E8C		// VMRGOW V10, V11, V10
+
+	WORD $0x13AC6F8C		// VMRGEW V12, V13, V29
+	WORD $0x13CE7F8C		// VMRGEW V14, V15, V30
+
+	XXPERMDI VS40, VS42, $0, VS41
+	XXPERMDI VS40, VS42, $3, VS43
+	XXPERMDI VS59, VS60, $0, VS40
+	XXPERMDI VS59, VS60, $3, VS42
+
+	WORD $0x118C6E8C		// VMRGOW V12, V13, V12
+	WORD $0x11CE7E8C		// VMRGOW V14, V15, V14
+
+	VSPLTISW $4, V27
+	VADDUWM V26, V27, V26
+
+	XXPERMDI VS44, VS46, $0, VS45
+	XXPERMDI VS44, VS46, $3, VS47
+	XXPERMDI VS61, VS62, $0, VS44
+	XXPERMDI VS61, VS62, $3, VS46
+
+	VADDUWM V0, V16, V0
+	VADDUWM V4, V17, V4
+	VADDUWM V8, V18, V8
+	VADDUWM V12, V19, V12
+
+	CMPU LEN, $64
+	BLT tail_vsx
+
+	// Bottom of loop
+	LXVW4X (INP)(R0), VS59
+	LXVW4X (INP)(R8), VS60
+	LXVW4X (INP)(R9), VS61
+	LXVW4X (INP)(R10), VS62
+
+	VXOR V27, V0, V27
+	VXOR V28, V4, V28
+	VXOR V29, V8, V29
+	VXOR V30, V12, V30
+
+	STXVW4X VS59, (OUT)(R0)
+	STXVW4X VS60, (OUT)(R8)
+	ADD     $64, INP
+	STXVW4X VS61, (OUT)(R9)
+	ADD     $-64, LEN
+	STXVW4X VS62, (OUT)(R10)
+	ADD     $64, OUT
+	BEQ     done_vsx
+
+	VADDUWM V1, V16, V0
+	VADDUWM V5, V17, V4
+	VADDUWM V9, V18, V8
+	VADDUWM V13, V19, V12
+
+	CMPU  LEN, $64
+	BLT   tail_vsx
+
+	LXVW4X (INP)(R0), VS59
+	LXVW4X (INP)(R8), VS60
+	LXVW4X (INP)(R9), VS61
+	LXVW4X (INP)(R10), VS62
+	VXOR   V27, V0, V27
+
+	VXOR V28, V4, V28
+	VXOR V29, V8, V29
+	VXOR V30, V12, V30
+
+	STXVW4X VS59, (OUT)(R0)
+	STXVW4X VS60, (OUT)(R8)
+	ADD     $64, INP
+	STXVW4X VS61, (OUT)(R9)
+	ADD     $-64, LEN
+	STXVW4X VS62, (OUT)(V10)
+	ADD     $64, OUT
+	BEQ     done_vsx
+
+	VADDUWM V2, V16, V0
+	VADDUWM V6, V17, V4
+	VADDUWM V10, V18, V8
+	VADDUWM V14, V19, V12
+
+	CMPU LEN, $64
+	BLT  tail_vsx
+
+	LXVW4X (INP)(R0), VS59
+	LXVW4X (INP)(R8), VS60
+	LXVW4X (INP)(R9), VS61
+	LXVW4X (INP)(R10), VS62
+
+	VXOR V27, V0, V27
+	VXOR V28, V4, V28
+	VXOR V29, V8, V29
+	VXOR V30, V12, V30
+
+	STXVW4X VS59, (OUT)(R0)
+	STXVW4X VS60, (OUT)(R8)
+	ADD     $64, INP
+	STXVW4X VS61, (OUT)(R9)
+	ADD     $-64, LEN
+	STXVW4X VS62, (OUT)(R10)
+	ADD     $64, OUT
+	BEQ     done_vsx
+
+	VADDUWM V3, V16, V0
+	VADDUWM V7, V17, V4
+	VADDUWM V11, V18, V8
+	VADDUWM V15, V19, V12
+
+	CMPU  LEN, $64
+	BLT   tail_vsx
+
+	LXVW4X (INP)(R0), VS59
+	LXVW4X (INP)(R8), VS60
+	LXVW4X (INP)(R9), VS61
+	LXVW4X (INP)(R10), VS62
+
+	VXOR V27, V0, V27
+	VXOR V28, V4, V28
+	VXOR V29, V8, V29
+	VXOR V30, V12, V30
+
+	STXVW4X VS59, (OUT)(R0)
+	STXVW4X VS60, (OUT)(R8)
+	ADD     $64, INP
+	STXVW4X VS61, (OUT)(R9)
+	ADD     $-64, LEN
+	STXVW4X VS62, (OUT)(R10)
+	ADD     $64, OUT
+
+	MOVD $10, R14
+	MOVD R14, CTR
+	BNE  loop_outer_vsx
+
+done_vsx:
+	// Increment counter by number of 64 byte blocks
+	MOVD (CNT), R14
+	ADD  BLOCKS, R14
+	MOVD R14, (CNT)
+	RET
+
+tail_vsx:
+	ADD  $32, R1, R11
+	MOVD LEN, CTR
+
+	// Save values on stack to copy from
+	STXVW4X VS32, (R11)(R0)
+	STXVW4X VS36, (R11)(R8)
+	STXVW4X VS40, (R11)(R9)
+	STXVW4X VS44, (R11)(R10)
+	ADD $-1, R11, R12
+	ADD $-1, INP
+	ADD $-1, OUT
+
+looptail_vsx:
+	// Copying the result to OUT
+	// in bytes.
+	MOVBZU 1(R12), KEY
+	MOVBZU 1(INP), TMP
+	XOR    KEY, TMP, KEY
+	MOVBU  KEY, 1(OUT)
+	BC     16, LT, looptail_vsx
+
+	// Clear the stack values
+	STXVW4X VS48, (R11)(R0)
+	STXVW4X VS48, (R11)(R8)
+	STXVW4X VS48, (R11)(R9)
+	STXVW4X VS48, (R11)(R10)
+	BR      done_vsx

src/vendor/golang.org/x/crypto/internal/chacha20/chacha_s390x.go → src/vendor/golang.org/x/crypto/chacha20/chacha_s390x.go

@@ -2,28 +2,25 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build s390x,!gccgo,!appengine
+// +build !gccgo,!appengine
 
 package chacha20
 
-import (
-	"golang.org/x/sys/cpu"
-)
+import "golang.org/x/sys/cpu"
 
 var haveAsm = cpu.S390X.HasVX
 
 const bufSize = 256
 
 // xorKeyStreamVX is an assembly implementation of XORKeyStream. It must only
-// be called when the vector facility is available.
-// Implementation in asm_s390x.s.
+// be called when the vector facility is available. Implementation in asm_s390x.s.
 //go:noescape
-func xorKeyStreamVX(dst, src []byte, key *[8]uint32, nonce *[3]uint32, counter *uint32, buf *[256]byte, len *int)
-
-func (c *Cipher) xorKeyStreamAsm(dst, src []byte) {
-	xorKeyStreamVX(dst, src, &c.key, &c.nonce, &c.counter, &c.buf, &c.len)
+func xorKeyStreamVX(dst, src []byte, key *[8]uint32, nonce *[3]uint32, counter *uint32)
+
+func (c *Cipher) xorKeyStreamBlocks(dst, src []byte) {
+	if cpu.S390X.HasVX {
+		xorKeyStreamVX(dst, src, &c.key, &c.nonce, &c.counter)
+	} else {
+		c.xorKeyStreamBlocksGeneric(dst, src)
+	}
 }
-
-// EXRL targets, DO NOT CALL!
-func mvcSrcToBuf()
-func mvcBufToDst()

src/vendor/golang.org/x/crypto/internal/chacha20/chacha_s390x.s → src/vendor/golang.org/x/crypto/chacha20/chacha_s390x.s

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build s390x,!gccgo,!appengine
+// +build !gccgo,!appengine
 
 #include "go_asm.h"
 #include "textflag.h"
@@ -24,15 +24,6 @@ DATA ·constants<>+0x14(SB)/4, $0x3320646e
 DATA ·constants<>+0x18(SB)/4, $0x79622d32
 DATA ·constants<>+0x1c(SB)/4, $0x6b206574
 
-// EXRL targets:
-TEXT ·mvcSrcToBuf(SB), NOFRAME|NOSPLIT, $0
-	MVC $1, (R1), (R8)
-	RET
-
-TEXT ·mvcBufToDst(SB), NOFRAME|NOSPLIT, $0
-	MVC $1, (R8), (R9)
-	RET
-
 #define BSWAP V5
 #define J0    V6
 #define KEY0  V7
@@ -144,7 +135,7 @@ TEXT ·mvcBufToDst(SB), NOFRAME|NOSPLIT, $0
 	VMRHF v, w, c \ // c = {a[2], b[2], c[2], d[2]}
 	VMRLF v, w, d // d = {a[3], b[3], c[3], d[3]}
 
-// func xorKeyStreamVX(dst, src []byte, key *[8]uint32, nonce *[3]uint32, counter *uint32, buf *[256]byte, len *int)
+// func xorKeyStreamVX(dst, src []byte, key *[8]uint32, nonce *[3]uint32, counter *uint32)
 TEXT ·xorKeyStreamVX(SB), NOSPLIT, $0
 	MOVD $·constants<>(SB), R1
 	MOVD dst+0(FP), R2         // R2=&dst[0]
@@ -152,25 +143,10 @@ TEXT ·xorKeyStreamVX(SB), NOSPLIT, $0
 	MOVD key+48(FP), R5        // R5=key
 	MOVD nonce+56(FP), R6      // R6=nonce
 	MOVD counter+64(FP), R7    // R7=counter
-	MOVD buf+72(FP), R8        // R8=buf
-	MOVD len+80(FP), R9        // R9=len
 
 	// load BSWAP and J0
 	VLM (R1), BSWAP, J0
 
-	// set up tail buffer
-	ADD     $-1, R4, R12
-	MOVBZ   R12, R12
-	CMPUBEQ R12, $255, aligned
-	MOVD    R4, R1
-	AND     $~255, R1
-	MOVD    $(R3)(R1*1), R1
-	EXRL    $·mvcSrcToBuf(SB), R12
-	MOVD    $255, R0
-	SUB     R12, R0
-	MOVD    R0, (R9)               // update len
-
-aligned:
 	// setup
 	MOVD  $95, R0
 	VLM   (R5), KEY0, KEY1
@@ -217,9 +193,7 @@ loop:
 
 	// decrement length
 	ADD $-256, R4
-	BLT tail
 
-continue:
 	// rearrange vectors
 	SHUFFLE(X0, X1, X2, X3, M0, M1, M2, M3)
 	ADDV(J0, X0, X1, X2, X3)
@@ -245,16 +219,6 @@ continue:
 	MOVD $256(R3), R3
 
 	CMPBNE  R4, $0, chacha
-	CMPUBEQ R12, $255, return
-	EXRL    $·mvcBufToDst(SB), R12 // len was updated during setup
 
-return:
 	VSTEF $0, CTR, (R7)
 	RET
-
-tail:
-	MOVD R2, R9
-	MOVD R8, R2
-	MOVD R8, R3
-	MOVD $0, R4
-	JMP  continue

src/vendor/golang.org/x/crypto/internal/chacha20/xor.go → src/vendor/golang.org/x/crypto/chacha20/xor.go

@@ -4,9 +4,7 @@
 
 package chacha20
 
-import (
-	"runtime"
-)
+import "runtime"
 
 // Platforms that have fast unaligned 32-bit little endian accesses.
 const unaligned = runtime.GOARCH == "386" ||

src/vendor/golang.org/x/crypto/chacha20poly1305/chacha20poly1305.go

@@ -2,13 +2,13 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// Package chacha20poly1305 implements the ChaCha20-Poly1305 AEAD as specified in RFC 7539,
-// and its extended nonce variant XChaCha20-Poly1305.
+// Package chacha20poly1305 implements the ChaCha20-Poly1305 AEAD and its
+// extended nonce variant XChaCha20-Poly1305, as specified in RFC 8439 and
+// draft-irtf-cfrg-xchacha-01.
 package chacha20poly1305 // import "golang.org/x/crypto/chacha20poly1305"
 
 import (
 	"crypto/cipher"
-	"encoding/binary"
 	"errors"
 )
 
@@ -29,7 +29,7 @@ const (
 )
 
 type chacha20poly1305 struct {
-	key [8]uint32
+	key [KeySize]byte
 }
 
 // New returns a ChaCha20-Poly1305 AEAD that uses the given 256-bit key.
@@ -38,14 +38,7 @@ func New(key []byte) (cipher.AEAD, error) {
 		return nil, errors.New("chacha20poly1305: bad key length")
 	}
 	ret := new(chacha20poly1305)
-	ret.key[0] = binary.LittleEndian.Uint32(key[0:4])
-	ret.key[1] = binary.LittleEndian.Uint32(key[4:8])
-	ret.key[2] = binary.LittleEndian.Uint32(key[8:12])
-	ret.key[3] = binary.LittleEndian.Uint32(key[12:16])
-	ret.key[4] = binary.LittleEndian.Uint32(key[16:20])
-	ret.key[5] = binary.LittleEndian.Uint32(key[20:24])
-	ret.key[6] = binary.LittleEndian.Uint32(key[24:28])
-	ret.key[7] = binary.LittleEndian.Uint32(key[28:32])
+	copy(ret.key[:], key)
 	return ret, nil
 }
 

src/vendor/golang.org/x/crypto/chacha20poly1305/chacha20poly1305_amd64.go

@@ -25,23 +25,23 @@ var (
 
 // setupState writes a ChaCha20 input matrix to state. See
 // https://tools.ietf.org/html/rfc7539#section-2.3.
-func setupState(state *[16]uint32, key *[8]uint32, nonce []byte) {
+func setupState(state *[16]uint32, key *[32]byte, nonce []byte) {
 	state[0] = 0x61707865
 	state[1] = 0x3320646e
 	state[2] = 0x79622d32
 	state[3] = 0x6b206574
 
-	state[4] = key[0]
-	state[5] = key[1]
-	state[6] = key[2]
-	state[7] = key[3]
-	state[8] = key[4]
-	state[9] = key[5]
-	state[10] = key[6]
-	state[11] = key[7]
+	state[4] = binary.LittleEndian.Uint32(key[0:4])
+	state[5] = binary.LittleEndian.Uint32(key[4:8])
+	state[6] = binary.LittleEndian.Uint32(key[8:12])
+	state[7] = binary.LittleEndian.Uint32(key[12:16])
+	state[8] = binary.LittleEndian.Uint32(key[16:20])
+	state[9] = binary.LittleEndian.Uint32(key[20:24])
+	state[10] = binary.LittleEndian.Uint32(key[24:28])
+	state[11] = binary.LittleEndian.Uint32(key[28:32])
 
 	state[12] = 0
-	state[13] = binary.LittleEndian.Uint32(nonce[:4])
+	state[13] = binary.LittleEndian.Uint32(nonce[0:4])
 	state[14] = binary.LittleEndian.Uint32(nonce[4:8])
 	state[15] = binary.LittleEndian.Uint32(nonce[8:12])
 }

src/vendor/golang.org/x/crypto/chacha20poly1305/chacha20poly1305_generic.go

@@ -7,7 +7,7 @@ package chacha20poly1305
 import (
 	"encoding/binary"
 
-	"golang.org/x/crypto/internal/chacha20"
+	"golang.org/x/crypto/chacha20"
 	"golang.org/x/crypto/internal/subtle"
 	"golang.org/x/crypto/poly1305"
 )
@@ -22,14 +22,10 @@ func (c *chacha20poly1305) sealGeneric(dst, nonce, plaintext, additionalData []b
 		panic("chacha20poly1305: invalid buffer overlap")
 	}
 
-	var polyKey [32]byte
-	s := chacha20.New(c.key, [3]uint32{
-		binary.LittleEndian.Uint32(nonce[0:4]),
-		binary.LittleEndian.Uint32(nonce[4:8]),
-		binary.LittleEndian.Uint32(nonce[8:12]),
-	})
+	var polyKey, discardBuf [32]byte
+	s, _ := chacha20.NewUnauthenticatedCipher(c.key[:], nonce)
 	s.XORKeyStream(polyKey[:], polyKey[:])
-	s.Advance() // skip the next 32 bytes
+	s.XORKeyStream(discardBuf[:], discardBuf[:]) // skip the next 32 bytes
 	s.XORKeyStream(out, plaintext)
 
 	polyInput := make([]byte, roundTo16(len(additionalData))+roundTo16(len(plaintext))+8+8)
@@ -50,14 +46,10 @@ func (c *chacha20poly1305) openGeneric(dst, nonce, ciphertext, additionalData []
 	copy(tag[:], ciphertext[len(ciphertext)-16:])
 	ciphertext = ciphertext[:len(ciphertext)-16]
 
-	var polyKey [32]byte
-	s := chacha20.New(c.key, [3]uint32{
-		binary.LittleEndian.Uint32(nonce[0:4]),
-		binary.LittleEndian.Uint32(nonce[4:8]),
-		binary.LittleEndian.Uint32(nonce[8:12]),
-	})
+	var polyKey, discardBuf [32]byte
+	s, _ := chacha20.NewUnauthenticatedCipher(c.key[:], nonce)
 	s.XORKeyStream(polyKey[:], polyKey[:])
-	s.Advance() // skip the next 32 bytes
+	s.XORKeyStream(discardBuf[:], discardBuf[:]) // skip the next 32 bytes
 
 	polyInput := make([]byte, roundTo16(len(additionalData))+roundTo16(len(ciphertext))+8+8)
 	copy(polyInput, additionalData)

src/vendor/golang.org/x/crypto/chacha20poly1305/xchacha20poly1305.go

@@ -6,14 +6,13 @@ package chacha20poly1305
 
 import (
 	"crypto/cipher"
-	"encoding/binary"
 	"errors"
 
-	"golang.org/x/crypto/internal/chacha20"
+	"golang.org/x/crypto/chacha20"
 )
 
 type xchacha20poly1305 struct {
-	key [8]uint32
+	key [KeySize]byte
 }
 
 // NewX returns a XChaCha20-Poly1305 AEAD that uses the given 256-bit key.
@@ -27,14 +26,7 @@ func NewX(key []byte) (cipher.AEAD, error) {
 		return nil, errors.New("chacha20poly1305: bad key length")
 	}
 	ret := new(xchacha20poly1305)
-	ret.key[0] = binary.LittleEndian.Uint32(key[0:4])
-	ret.key[1] = binary.LittleEndian.Uint32(key[4:8])
-	ret.key[2] = binary.LittleEndian.Uint32(key[8:12])
-	ret.key[3] = binary.LittleEndian.Uint32(key[12:16])
-	ret.key[4] = binary.LittleEndian.Uint32(key[16:20])
-	ret.key[5] = binary.LittleEndian.Uint32(key[20:24])
-	ret.key[6] = binary.LittleEndian.Uint32(key[24:28])
-	ret.key[7] = binary.LittleEndian.Uint32(key[28:32])
+	copy(ret.key[:], key)
 	return ret, nil
 }
 
@@ -60,15 +52,10 @@ func (x *xchacha20poly1305) Seal(dst, nonce, plaintext, additionalData []byte) [
 		panic("chacha20poly1305: plaintext too large")
 	}
 
-	hNonce := [4]uint32{
-		binary.LittleEndian.Uint32(nonce[0:4]),
-		binary.LittleEndian.Uint32(nonce[4:8]),
-		binary.LittleEndian.Uint32(nonce[8:12]),
-		binary.LittleEndian.Uint32(nonce[12:16]),
-	}
-	c := &chacha20poly1305{
-		key: chacha20.HChaCha20(&x.key, &hNonce),
-	}
+	c := new(chacha20poly1305)
+	hKey, _ := chacha20.HChaCha20(x.key[:], nonce[0:16])
+	copy(c.key[:], hKey)
+
 	// The first 4 bytes of the final nonce are unused counter space.
 	cNonce := make([]byte, NonceSize)
 	copy(cNonce[4:12], nonce[16:24])
@@ -87,15 +74,10 @@ func (x *xchacha20poly1305) Open(dst, nonce, ciphertext, additionalData []byte)
 		panic("chacha20poly1305: ciphertext too large")
 	}
 
-	hNonce := [4]uint32{
-		binary.LittleEndian.Uint32(nonce[0:4]),
-		binary.LittleEndian.Uint32(nonce[4:8]),
-		binary.LittleEndian.Uint32(nonce[8:12]),
-		binary.LittleEndian.Uint32(nonce[12:16]),
-	}
-	c := &chacha20poly1305{
-		key: chacha20.HChaCha20(&x.key, &hNonce),
-	}
+	c := new(chacha20poly1305)
+	hKey, _ := chacha20.HChaCha20(x.key[:], nonce[0:16])
+	copy(c.key[:], hKey)
+
 	// The first 4 bytes of the final nonce are unused counter space.
 	cNonce := make([]byte, NonceSize)
 	copy(cNonce[4:12], nonce[16:24])

src/vendor/golang.org/x/crypto/curve25519/const_amd64.s

-// 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.
-
-// This code was translated into a form compatible with 6a from the public
-// domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
-
-// +build amd64,!gccgo,!appengine
-
-// These constants cannot be encoded in non-MOVQ immediates.
-// We access them directly from memory instead.
-
-DATA ·_121666_213(SB)/8, $996687872
-GLOBL ·_121666_213(SB), 8, $8
-
-DATA ·_2P0(SB)/8, $0xFFFFFFFFFFFDA
-GLOBL ·_2P0(SB), 8, $8
-
-DATA ·_2P1234(SB)/8, $0xFFFFFFFFFFFFE
-GLOBL ·_2P1234(SB), 8, $8

src/vendor/golang.org/x/crypto/curve25519/cswap_amd64.s

-// 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.
-
-// +build amd64,!gccgo,!appengine
-
-// func cswap(inout *[4][5]uint64, v uint64)
-TEXT ·cswap(SB),7,$0
-	MOVQ inout+0(FP),DI
-	MOVQ v+8(FP),SI
-
-	SUBQ $1, SI
-	NOTQ SI
-	MOVQ SI, X15
-	PSHUFD $0x44, X15, X15
-
-	MOVOU 0(DI), X0
-	MOVOU 16(DI), X2
-	MOVOU 32(DI), X4
-	MOVOU 48(DI), X6
-	MOVOU 64(DI), X8
-	MOVOU 80(DI), X1
-	MOVOU 96(DI), X3
-	MOVOU 112(DI), X5
-	MOVOU 128(DI), X7
-	MOVOU 144(DI), X9
-
-	MOVO X1, X10
-	MOVO X3, X11
-	MOVO X5, X12
-	MOVO X7, X13
-	MOVO X9, X14
-
-	PXOR X0, X10
-	PXOR X2, X11
-	PXOR X4, X12
-	PXOR X6, X13
-	PXOR X8, X14
-	PAND X15, X10
-	PAND X15, X11
-	PAND X15, X12
-	PAND X15, X13
-	PAND X15, X14
-	PXOR X10, X0
-	PXOR X10, X1
-	PXOR X11, X2
-	PXOR X11, X3
-	PXOR X12, X4
-	PXOR X12, X5
-	PXOR X13, X6
-	PXOR X13, X7
-	PXOR X14, X8
-	PXOR X14, X9
-
-	MOVOU X0, 0(DI)
-	MOVOU X2, 16(DI)
-	MOVOU X4, 32(DI)
-	MOVOU X6, 48(DI)
-	MOVOU X8, 64(DI)
-	MOVOU X1, 80(DI)
-	MOVOU X3, 96(DI)
-	MOVOU X5, 112(DI)
-	MOVOU X7, 128(DI)
-	MOVOU X9, 144(DI)
-	RET

src/vendor/golang.org/x/crypto/curve25519/curve25519.go

-// Copyright 2013 The Go Authors. All rights reserved.
+// Copyright 2019 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.
 
-// We have an implementation in amd64 assembly so this code is only run on
-// non-amd64 platforms. The amd64 assembly does not support gccgo.
-// +build !amd64 gccgo appengine
-
-package curve25519
+// Package curve25519 provides an implementation of the X25519 function, which
+// performs scalar multiplication on the elliptic curve known as Curve25519.
+// See RFC 7748.
+package curve25519 // import "golang.org/x/crypto/curve25519"
 
 import (
-	"encoding/binary"
+	"crypto/subtle"
+	"fmt"
 )
 
-// This code is a port of the public domain, "ref10" implementation of
-// curve25519 from SUPERCOP 20130419 by D. J. Bernstein.
-
-// fieldElement represents an element of the field GF(2^255 - 19). An element
-// t, entries t[0]...t[9], represents the integer t[0]+2^26 t[1]+2^51 t[2]+2^77
-// t[3]+2^102 t[4]+...+2^230 t[9]. Bounds on each t[i] vary depending on
-// context.
-type fieldElement [10]int32
-
-func feZero(fe *fieldElement) {
-	for i := range fe {
-		fe[i] = 0
-	}
-}
-
-func feOne(fe *fieldElement) {
-	feZero(fe)
-	fe[0] = 1
-}
-
-func feAdd(dst, a, b *fieldElement) {
-	for i := range dst {
-		dst[i] = a[i] + b[i]
-	}
-}
-
-func feSub(dst, a, b *fieldElement) {
-	for i := range dst {
-		dst[i] = a[i] - b[i]
-	}
-}
-
-func feCopy(dst, src *fieldElement) {
-	for i := range dst {
-		dst[i] = src[i]
-	}
-}
-
-// feCSwap replaces (f,g) with (g,f) if b == 1; replaces (f,g) with (f,g) if b == 0.
-//
-// Preconditions: b in {0,1}.
-func feCSwap(f, g *fieldElement, b int32) {
-	b = -b
-	for i := range f {
-		t := b & (f[i] ^ g[i])
-		f[i] ^= t
-		g[i] ^= t
-	}
-}
-
-// load3 reads a 24-bit, little-endian value from in.
-func load3(in []byte) int64 {
-	var r int64
-	r = int64(in[0])
-	r |= int64(in[1]) << 8
-	r |= int64(in[2]) << 16
-	return r
-}
-
-// load4 reads a 32-bit, little-endian value from in.
-func load4(in []byte) int64 {
-	return int64(binary.LittleEndian.Uint32(in))
-}
-
-func feFromBytes(dst *fieldElement, src *[32]byte) {
-	h0 := load4(src[:])
-	h1 := load3(src[4:]) << 6
-	h2 := load3(src[7:]) << 5
-	h3 := load3(src[10:]) << 3
-	h4 := load3(src[13:]) << 2
-	h5 := load4(src[16:])
-	h6 := load3(src[20:]) << 7
-	h7 := load3(src[23:]) << 5
-	h8 := load3(src[26:]) << 4
-	h9 := (load3(src[29:]) & 0x7fffff) << 2
-
-	var carry [10]int64
-	carry[9] = (h9 + 1<<24) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-	carry[1] = (h1 + 1<<24) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[3] = (h3 + 1<<24) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[5] = (h5 + 1<<24) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-	carry[7] = (h7 + 1<<24) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-
-	carry[0] = (h0 + 1<<25) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[2] = (h2 + 1<<25) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[4] = (h4 + 1<<25) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[6] = (h6 + 1<<25) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-	carry[8] = (h8 + 1<<25) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-
-	dst[0] = int32(h0)
-	dst[1] = int32(h1)
-	dst[2] = int32(h2)
-	dst[3] = int32(h3)
-	dst[4] = int32(h4)
-	dst[5] = int32(h5)
-	dst[6] = int32(h6)
-	dst[7] = int32(h7)
-	dst[8] = int32(h8)
-	dst[9] = int32(h9)
-}
-
-// feToBytes marshals h to s.
-// Preconditions:
-//   |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-//
-// Write p=2^255-19; q=floor(h/p).
-// Basic claim: q = floor(2^(-255)(h + 19 2^(-25)h9 + 2^(-1))).
-//
-// Proof:
-//   Have |h|<=p so |q|<=1 so |19^2 2^(-255) q|<1/4.
-//   Also have |h-2^230 h9|<2^230 so |19 2^(-255)(h-2^230 h9)|<1/4.
+// ScalarMult sets dst to the product scalar * point.
 //
-//   Write y=2^(-1)-19^2 2^(-255)q-19 2^(-255)(h-2^230 h9).
-//   Then 0<y<1.
-//
-//   Write r=h-pq.
-//   Have 0<=r<=p-1=2^255-20.
-//   Thus 0<=r+19(2^-255)r<r+19(2^-255)2^255<=2^255-1.
-//
-//   Write x=r+19(2^-255)r+y.
-//   Then 0<x<2^255 so floor(2^(-255)x) = 0 so floor(q+2^(-255)x) = q.
-//
-//   Have q+2^(-255)x = 2^(-255)(h + 19 2^(-25) h9 + 2^(-1))
-//   so floor(2^(-255)(h + 19 2^(-25) h9 + 2^(-1))) = q.
-func feToBytes(s *[32]byte, h *fieldElement) {
-	var carry [10]int32
-
-	q := (19*h[9] + (1 << 24)) >> 25
-	q = (h[0] + q) >> 26
-	q = (h[1] + q) >> 25
-	q = (h[2] + q) >> 26
-	q = (h[3] + q) >> 25
-	q = (h[4] + q) >> 26
-	q = (h[5] + q) >> 25
-	q = (h[6] + q) >> 26
-	q = (h[7] + q) >> 25
-	q = (h[8] + q) >> 26
-	q = (h[9] + q) >> 25
-
-	// Goal: Output h-(2^255-19)q, which is between 0 and 2^255-20.
-	h[0] += 19 * q
-	// Goal: Output h-2^255 q, which is between 0 and 2^255-20.
-
-	carry[0] = h[0] >> 26
-	h[1] += carry[0]
-	h[0] -= carry[0] << 26
-	carry[1] = h[1] >> 25
-	h[2] += carry[1]
-	h[1] -= carry[1] << 25
-	carry[2] = h[2] >> 26
-	h[3] += carry[2]
-	h[2] -= carry[2] << 26
-	carry[3] = h[3] >> 25
-	h[4] += carry[3]
-	h[3] -= carry[3] << 25
-	carry[4] = h[4] >> 26
-	h[5] += carry[4]
-	h[4] -= carry[4] << 26
-	carry[5] = h[5] >> 25
-	h[6] += carry[5]
-	h[5] -= carry[5] << 25
-	carry[6] = h[6] >> 26
-	h[7] += carry[6]
-	h[6] -= carry[6] << 26
-	carry[7] = h[7] >> 25
-	h[8] += carry[7]
-	h[7] -= carry[7] << 25
-	carry[8] = h[8] >> 26
-	h[9] += carry[8]
-	h[8] -= carry[8] << 26
-	carry[9] = h[9] >> 25
-	h[9] -= carry[9] << 25
-	// h10 = carry9
-
-	// Goal: Output h[0]+...+2^255 h10-2^255 q, which is between 0 and 2^255-20.
-	// Have h[0]+...+2^230 h[9] between 0 and 2^255-1;
-	// evidently 2^255 h10-2^255 q = 0.
-	// Goal: Output h[0]+...+2^230 h[9].
-
-	s[0] = byte(h[0] >> 0)
-	s[1] = byte(h[0] >> 8)
-	s[2] = byte(h[0] >> 16)
-	s[3] = byte((h[0] >> 24) | (h[1] << 2))
-	s[4] = byte(h[1] >> 6)
-	s[5] = byte(h[1] >> 14)
-	s[6] = byte((h[1] >> 22) | (h[2] << 3))
-	s[7] = byte(h[2] >> 5)
-	s[8] = byte(h[2] >> 13)
-	s[9] = byte((h[2] >> 21) | (h[3] << 5))
-	s[10] = byte(h[3] >> 3)
-	s[11] = byte(h[3] >> 11)
-	s[12] = byte((h[3] >> 19) | (h[4] << 6))
-	s[13] = byte(h[4] >> 2)
-	s[14] = byte(h[4] >> 10)
-	s[15] = byte(h[4] >> 18)
-	s[16] = byte(h[5] >> 0)
-	s[17] = byte(h[5] >> 8)
-	s[18] = byte(h[5] >> 16)
-	s[19] = byte((h[5] >> 24) | (h[6] << 1))
-	s[20] = byte(h[6] >> 7)
-	s[21] = byte(h[6] >> 15)
-	s[22] = byte((h[6] >> 23) | (h[7] << 3))
-	s[23] = byte(h[7] >> 5)
-	s[24] = byte(h[7] >> 13)
-	s[25] = byte((h[7] >> 21) | (h[8] << 4))
-	s[26] = byte(h[8] >> 4)
-	s[27] = byte(h[8] >> 12)
-	s[28] = byte((h[8] >> 20) | (h[9] << 6))
-	s[29] = byte(h[9] >> 2)
-	s[30] = byte(h[9] >> 10)
-	s[31] = byte(h[9] >> 18)
+// Deprecated: when provided a low-order point, ScalarMult will set dst to all
+// zeroes, irrespective of the scalar. Instead, use the X25519 function, which
+// will return an error.
+func ScalarMult(dst, scalar, point *[32]byte) {
+	scalarMult(dst, scalar, point)
 }
 
-// feMul calculates h = f * g
-// Can overlap h with f or g.
-//
-// Preconditions:
-//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//    |g| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//
-// Postconditions:
-//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+// ScalarBaseMult sets dst to the product scalar * base where base is the
+// standard generator.
 //
-// Notes on implementation strategy:
-//
-// Using schoolbook multiplication.
-// Karatsuba would save a little in some cost models.
-//
-// Most multiplications by 2 and 19 are 32-bit precomputations;
-// cheaper than 64-bit postcomputations.
-//
-// There is one remaining multiplication by 19 in the carry chain;
-// one *19 precomputation can be merged into this,
-// but the resulting data flow is considerably less clean.
-//
-// There are 12 carries below.
-// 10 of them are 2-way parallelizable and vectorizable.
-// Can get away with 11 carries, but then data flow is much deeper.
-//
-// With tighter constraints on inputs can squeeze carries into int32.
-func feMul(h, f, g *fieldElement) {
-	f0 := f[0]
-	f1 := f[1]
-	f2 := f[2]
-	f3 := f[3]
-	f4 := f[4]
-	f5 := f[5]
-	f6 := f[6]
-	f7 := f[7]
-	f8 := f[8]
-	f9 := f[9]
-	g0 := g[0]
-	g1 := g[1]
-	g2 := g[2]
-	g3 := g[3]
-	g4 := g[4]
-	g5 := g[5]
-	g6 := g[6]
-	g7 := g[7]
-	g8 := g[8]
-	g9 := g[9]
-	g1_19 := 19 * g1 // 1.4*2^29
-	g2_19 := 19 * g2 // 1.4*2^30; still ok
-	g3_19 := 19 * g3
-	g4_19 := 19 * g4
-	g5_19 := 19 * g5
-	g6_19 := 19 * g6
-	g7_19 := 19 * g7
-	g8_19 := 19 * g8
-	g9_19 := 19 * g9
-	f1_2 := 2 * f1
-	f3_2 := 2 * f3
-	f5_2 := 2 * f5
-	f7_2 := 2 * f7
-	f9_2 := 2 * f9
-	f0g0 := int64(f0) * int64(g0)
-	f0g1 := int64(f0) * int64(g1)
-	f0g2 := int64(f0) * int64(g2)
-	f0g3 := int64(f0) * int64(g3)
-	f0g4 := int64(f0) * int64(g4)
-	f0g5 := int64(f0) * int64(g5)
-	f0g6 := int64(f0) * int64(g6)
-	f0g7 := int64(f0) * int64(g7)
-	f0g8 := int64(f0) * int64(g8)
-	f0g9 := int64(f0) * int64(g9)
-	f1g0 := int64(f1) * int64(g0)
-	f1g1_2 := int64(f1_2) * int64(g1)
-	f1g2 := int64(f1) * int64(g2)
-	f1g3_2 := int64(f1_2) * int64(g3)
-	f1g4 := int64(f1) * int64(g4)
-	f1g5_2 := int64(f1_2) * int64(g5)
-	f1g6 := int64(f1) * int64(g6)
-	f1g7_2 := int64(f1_2) * int64(g7)
-	f1g8 := int64(f1) * int64(g8)
-	f1g9_38 := int64(f1_2) * int64(g9_19)
-	f2g0 := int64(f2) * int64(g0)
-	f2g1 := int64(f2) * int64(g1)
-	f2g2 := int64(f2) * int64(g2)
-	f2g3 := int64(f2) * int64(g3)
-	f2g4 := int64(f2) * int64(g4)
-	f2g5 := int64(f2) * int64(g5)
-	f2g6 := int64(f2) * int64(g6)
-	f2g7 := int64(f2) * int64(g7)
-	f2g8_19 := int64(f2) * int64(g8_19)
-	f2g9_19 := int64(f2) * int64(g9_19)
-	f3g0 := int64(f3) * int64(g0)
-	f3g1_2 := int64(f3_2) * int64(g1)
-	f3g2 := int64(f3) * int64(g2)
-	f3g3_2 := int64(f3_2) * int64(g3)
-	f3g4 := int64(f3) * int64(g4)
-	f3g5_2 := int64(f3_2) * int64(g5)
-	f3g6 := int64(f3) * int64(g6)
-	f3g7_38 := int64(f3_2) * int64(g7_19)
-	f3g8_19 := int64(f3) * int64(g8_19)
-	f3g9_38 := int64(f3_2) * int64(g9_19)
-	f4g0 := int64(f4) * int64(g0)
-	f4g1 := int64(f4) * int64(g1)
-	f4g2 := int64(f4) * int64(g2)
-	f4g3 := int64(f4) * int64(g3)
-	f4g4 := int64(f4) * int64(g4)
-	f4g5 := int64(f4) * int64(g5)
-	f4g6_19 := int64(f4) * int64(g6_19)
-	f4g7_19 := int64(f4) * int64(g7_19)
-	f4g8_19 := int64(f4) * int64(g8_19)
-	f4g9_19 := int64(f4) * int64(g9_19)
-	f5g0 := int64(f5) * int64(g0)
-	f5g1_2 := int64(f5_2) * int64(g1)
-	f5g2 := int64(f5) * int64(g2)
-	f5g3_2 := int64(f5_2) * int64(g3)
-	f5g4 := int64(f5) * int64(g4)
-	f5g5_38 := int64(f5_2) * int64(g5_19)
-	f5g6_19 := int64(f5) * int64(g6_19)
-	f5g7_38 := int64(f5_2) * int64(g7_19)
-	f5g8_19 := int64(f5) * int64(g8_19)
-	f5g9_38 := int64(f5_2) * int64(g9_19)
-	f6g0 := int64(f6) * int64(g0)
-	f6g1 := int64(f6) * int64(g1)
-	f6g2 := int64(f6) * int64(g2)
-	f6g3 := int64(f6) * int64(g3)
-	f6g4_19 := int64(f6) * int64(g4_19)
-	f6g5_19 := int64(f6) * int64(g5_19)
-	f6g6_19 := int64(f6) * int64(g6_19)
-	f6g7_19 := int64(f6) * int64(g7_19)
-	f6g8_19 := int64(f6) * int64(g8_19)
-	f6g9_19 := int64(f6) * int64(g9_19)
-	f7g0 := int64(f7) * int64(g0)
-	f7g1_2 := int64(f7_2) * int64(g1)
-	f7g2 := int64(f7) * int64(g2)
-	f7g3_38 := int64(f7_2) * int64(g3_19)
-	f7g4_19 := int64(f7) * int64(g4_19)
-	f7g5_38 := int64(f7_2) * int64(g5_19)
-	f7g6_19 := int64(f7) * int64(g6_19)
-	f7g7_38 := int64(f7_2) * int64(g7_19)
-	f7g8_19 := int64(f7) * int64(g8_19)
-	f7g9_38 := int64(f7_2) * int64(g9_19)
-	f8g0 := int64(f8) * int64(g0)
-	f8g1 := int64(f8) * int64(g1)
-	f8g2_19 := int64(f8) * int64(g2_19)
-	f8g3_19 := int64(f8) * int64(g3_19)
-	f8g4_19 := int64(f8) * int64(g4_19)
-	f8g5_19 := int64(f8) * int64(g5_19)
-	f8g6_19 := int64(f8) * int64(g6_19)
-	f8g7_19 := int64(f8) * int64(g7_19)
-	f8g8_19 := int64(f8) * int64(g8_19)
-	f8g9_19 := int64(f8) * int64(g9_19)
-	f9g0 := int64(f9) * int64(g0)
-	f9g1_38 := int64(f9_2) * int64(g1_19)
-	f9g2_19 := int64(f9) * int64(g2_19)
-	f9g3_38 := int64(f9_2) * int64(g3_19)
-	f9g4_19 := int64(f9) * int64(g4_19)
-	f9g5_38 := int64(f9_2) * int64(g5_19)
-	f9g6_19 := int64(f9) * int64(g6_19)
-	f9g7_38 := int64(f9_2) * int64(g7_19)
-	f9g8_19 := int64(f9) * int64(g8_19)
-	f9g9_38 := int64(f9_2) * int64(g9_19)
-	h0 := f0g0 + f1g9_38 + f2g8_19 + f3g7_38 + f4g6_19 + f5g5_38 + f6g4_19 + f7g3_38 + f8g2_19 + f9g1_38
-	h1 := f0g1 + f1g0 + f2g9_19 + f3g8_19 + f4g7_19 + f5g6_19 + f6g5_19 + f7g4_19 + f8g3_19 + f9g2_19
-	h2 := f0g2 + f1g1_2 + f2g0 + f3g9_38 + f4g8_19 + f5g7_38 + f6g6_19 + f7g5_38 + f8g4_19 + f9g3_38
-	h3 := f0g3 + f1g2 + f2g1 + f3g0 + f4g9_19 + f5g8_19 + f6g7_19 + f7g6_19 + f8g5_19 + f9g4_19
-	h4 := f0g4 + f1g3_2 + f2g2 + f3g1_2 + f4g0 + f5g9_38 + f6g8_19 + f7g7_38 + f8g6_19 + f9g5_38
-	h5 := f0g5 + f1g4 + f2g3 + f3g2 + f4g1 + f5g0 + f6g9_19 + f7g8_19 + f8g7_19 + f9g6_19
-	h6 := f0g6 + f1g5_2 + f2g4 + f3g3_2 + f4g2 + f5g1_2 + f6g0 + f7g9_38 + f8g8_19 + f9g7_38
-	h7 := f0g7 + f1g6 + f2g5 + f3g4 + f4g3 + f5g2 + f6g1 + f7g0 + f8g9_19 + f9g8_19
-	h8 := f0g8 + f1g7_2 + f2g6 + f3g5_2 + f4g4 + f5g3_2 + f6g2 + f7g1_2 + f8g0 + f9g9_38
-	h9 := f0g9 + f1g8 + f2g7 + f3g6 + f4g5 + f5g4 + f6g3 + f7g2 + f8g1 + f9g0
-	var carry [10]int64
-
-	// |h0| <= (1.1*1.1*2^52*(1+19+19+19+19)+1.1*1.1*2^50*(38+38+38+38+38))
-	//   i.e. |h0| <= 1.2*2^59; narrower ranges for h2, h4, h6, h8
-	// |h1| <= (1.1*1.1*2^51*(1+1+19+19+19+19+19+19+19+19))
-	//   i.e. |h1| <= 1.5*2^58; narrower ranges for h3, h5, h7, h9
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	// |h0| <= 2^25
-	// |h4| <= 2^25
-	// |h1| <= 1.51*2^58
-	// |h5| <= 1.51*2^58
-
-	carry[1] = (h1 + (1 << 24)) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[5] = (h5 + (1 << 24)) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-	// |h1| <= 2^24; from now on fits into int32
-	// |h5| <= 2^24; from now on fits into int32
-	// |h2| <= 1.21*2^59
-	// |h6| <= 1.21*2^59
-
-	carry[2] = (h2 + (1 << 25)) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[6] = (h6 + (1 << 25)) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-	// |h2| <= 2^25; from now on fits into int32 unchanged
-	// |h6| <= 2^25; from now on fits into int32 unchanged
-	// |h3| <= 1.51*2^58
-	// |h7| <= 1.51*2^58
-
-	carry[3] = (h3 + (1 << 24)) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[7] = (h7 + (1 << 24)) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-	// |h3| <= 2^24; from now on fits into int32 unchanged
-	// |h7| <= 2^24; from now on fits into int32 unchanged
-	// |h4| <= 1.52*2^33
-	// |h8| <= 1.52*2^33
-
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[8] = (h8 + (1 << 25)) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-	// |h4| <= 2^25; from now on fits into int32 unchanged
-	// |h8| <= 2^25; from now on fits into int32 unchanged
-	// |h5| <= 1.01*2^24
-	// |h9| <= 1.51*2^58
-
-	carry[9] = (h9 + (1 << 24)) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-	// |h9| <= 2^24; from now on fits into int32 unchanged
-	// |h0| <= 1.8*2^37
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	// |h0| <= 2^25; from now on fits into int32 unchanged
-	// |h1| <= 1.01*2^24
-
-	h[0] = int32(h0)
-	h[1] = int32(h1)
-	h[2] = int32(h2)
-	h[3] = int32(h3)
-	h[4] = int32(h4)
-	h[5] = int32(h5)
-	h[6] = int32(h6)
-	h[7] = int32(h7)
-	h[8] = int32(h8)
-	h[9] = int32(h9)
+// It is recommended to use the X25519 function with Basepoint instead, as
+// copying into fixed size arrays can lead to unexpected bugs.
+func ScalarBaseMult(dst, scalar *[32]byte) {
+	ScalarMult(dst, scalar, &basePoint)
 }
 
-// feSquare calculates h = f*f. Can overlap h with f.
-//
-// Preconditions:
-//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//
-// Postconditions:
-//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-func feSquare(h, f *fieldElement) {
-	f0 := f[0]
-	f1 := f[1]
-	f2 := f[2]
-	f3 := f[3]
-	f4 := f[4]
-	f5 := f[5]
-	f6 := f[6]
-	f7 := f[7]
-	f8 := f[8]
-	f9 := f[9]
-	f0_2 := 2 * f0
-	f1_2 := 2 * f1
-	f2_2 := 2 * f2
-	f3_2 := 2 * f3
-	f4_2 := 2 * f4
-	f5_2 := 2 * f5
-	f6_2 := 2 * f6
-	f7_2 := 2 * f7
-	f5_38 := 38 * f5 // 1.31*2^30
-	f6_19 := 19 * f6 // 1.31*2^30
-	f7_38 := 38 * f7 // 1.31*2^30
-	f8_19 := 19 * f8 // 1.31*2^30
-	f9_38 := 38 * f9 // 1.31*2^30
-	f0f0 := int64(f0) * int64(f0)
-	f0f1_2 := int64(f0_2) * int64(f1)
-	f0f2_2 := int64(f0_2) * int64(f2)
-	f0f3_2 := int64(f0_2) * int64(f3)
-	f0f4_2 := int64(f0_2) * int64(f4)
-	f0f5_2 := int64(f0_2) * int64(f5)
-	f0f6_2 := int64(f0_2) * int64(f6)
-	f0f7_2 := int64(f0_2) * int64(f7)
-	f0f8_2 := int64(f0_2) * int64(f8)
-	f0f9_2 := int64(f0_2) * int64(f9)
-	f1f1_2 := int64(f1_2) * int64(f1)
-	f1f2_2 := int64(f1_2) * int64(f2)
-	f1f3_4 := int64(f1_2) * int64(f3_2)
-	f1f4_2 := int64(f1_2) * int64(f4)
-	f1f5_4 := int64(f1_2) * int64(f5_2)
-	f1f6_2 := int64(f1_2) * int64(f6)
-	f1f7_4 := int64(f1_2) * int64(f7_2)
-	f1f8_2 := int64(f1_2) * int64(f8)
-	f1f9_76 := int64(f1_2) * int64(f9_38)
-	f2f2 := int64(f2) * int64(f2)
-	f2f3_2 := int64(f2_2) * int64(f3)
-	f2f4_2 := int64(f2_2) * int64(f4)
-	f2f5_2 := int64(f2_2) * int64(f5)
-	f2f6_2 := int64(f2_2) * int64(f6)
-	f2f7_2 := int64(f2_2) * int64(f7)
-	f2f8_38 := int64(f2_2) * int64(f8_19)
-	f2f9_38 := int64(f2) * int64(f9_38)
-	f3f3_2 := int64(f3_2) * int64(f3)
-	f3f4_2 := int64(f3_2) * int64(f4)
-	f3f5_4 := int64(f3_2) * int64(f5_2)
-	f3f6_2 := int64(f3_2) * int64(f6)
-	f3f7_76 := int64(f3_2) * int64(f7_38)
-	f3f8_38 := int64(f3_2) * int64(f8_19)
-	f3f9_76 := int64(f3_2) * int64(f9_38)
-	f4f4 := int64(f4) * int64(f4)
-	f4f5_2 := int64(f4_2) * int64(f5)
-	f4f6_38 := int64(f4_2) * int64(f6_19)
-	f4f7_38 := int64(f4) * int64(f7_38)
-	f4f8_38 := int64(f4_2) * int64(f8_19)
-	f4f9_38 := int64(f4) * int64(f9_38)
-	f5f5_38 := int64(f5) * int64(f5_38)
-	f5f6_38 := int64(f5_2) * int64(f6_19)
-	f5f7_76 := int64(f5_2) * int64(f7_38)
-	f5f8_38 := int64(f5_2) * int64(f8_19)
-	f5f9_76 := int64(f5_2) * int64(f9_38)
-	f6f6_19 := int64(f6) * int64(f6_19)
-	f6f7_38 := int64(f6) * int64(f7_38)
-	f6f8_38 := int64(f6_2) * int64(f8_19)
-	f6f9_38 := int64(f6) * int64(f9_38)
-	f7f7_38 := int64(f7) * int64(f7_38)
-	f7f8_38 := int64(f7_2) * int64(f8_19)
-	f7f9_76 := int64(f7_2) * int64(f9_38)
-	f8f8_19 := int64(f8) * int64(f8_19)
-	f8f9_38 := int64(f8) * int64(f9_38)
-	f9f9_38 := int64(f9) * int64(f9_38)
-	h0 := f0f0 + f1f9_76 + f2f8_38 + f3f7_76 + f4f6_38 + f5f5_38
-	h1 := f0f1_2 + f2f9_38 + f3f8_38 + f4f7_38 + f5f6_38
-	h2 := f0f2_2 + f1f1_2 + f3f9_76 + f4f8_38 + f5f7_76 + f6f6_19
-	h3 := f0f3_2 + f1f2_2 + f4f9_38 + f5f8_38 + f6f7_38
-	h4 := f0f4_2 + f1f3_4 + f2f2 + f5f9_76 + f6f8_38 + f7f7_38
-	h5 := f0f5_2 + f1f4_2 + f2f3_2 + f6f9_38 + f7f8_38
-	h6 := f0f6_2 + f1f5_4 + f2f4_2 + f3f3_2 + f7f9_76 + f8f8_19
-	h7 := f0f7_2 + f1f6_2 + f2f5_2 + f3f4_2 + f8f9_38
-	h8 := f0f8_2 + f1f7_4 + f2f6_2 + f3f5_4 + f4f4 + f9f9_38
-	h9 := f0f9_2 + f1f8_2 + f2f7_2 + f3f6_2 + f4f5_2
-	var carry [10]int64
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-
-	carry[1] = (h1 + (1 << 24)) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[5] = (h5 + (1 << 24)) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-
-	carry[2] = (h2 + (1 << 25)) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[6] = (h6 + (1 << 25)) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-
-	carry[3] = (h3 + (1 << 24)) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[7] = (h7 + (1 << 24)) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
+const (
+	// ScalarSize is the size of the scalar input to X25519.
+	ScalarSize = 32
+	// PointSize is the size of the point input to X25519.
+	PointSize = 32
+)
 
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[8] = (h8 + (1 << 25)) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
+// Basepoint is the canonical Curve25519 generator.
+var Basepoint []byte
 
-	carry[9] = (h9 + (1 << 24)) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
+var basePoint = [32]byte{9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
 
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
+func init() { Basepoint = basePoint[:] }
 
-	h[0] = int32(h0)
-	h[1] = int32(h1)
-	h[2] = int32(h2)
-	h[3] = int32(h3)
-	h[4] = int32(h4)
-	h[5] = int32(h5)
-	h[6] = int32(h6)
-	h[7] = int32(h7)
-	h[8] = int32(h8)
-	h[9] = int32(h9)
+func checkBasepoint() {
+	if subtle.ConstantTimeCompare(Basepoint, []byte{
+		0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	}) != 1 {
+		panic("curve25519: global Basepoint value was modified")
+	}
 }
 
-// feMul121666 calculates h = f * 121666. Can overlap h with f.
+// X25519 returns the result of the scalar multiplication (scalar * point),
+// according to RFC 7748, Section 5. scalar, point and the return value are
+// slices of 32 bytes.
 //
-// Preconditions:
-//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+// scalar can be generated at random, for example with crypto/rand. point should
+// be either Basepoint or the output of another X25519 call.
 //
-// Postconditions:
-//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-func feMul121666(h, f *fieldElement) {
-	h0 := int64(f[0]) * 121666
-	h1 := int64(f[1]) * 121666
-	h2 := int64(f[2]) * 121666
-	h3 := int64(f[3]) * 121666
-	h4 := int64(f[4]) * 121666
-	h5 := int64(f[5]) * 121666
-	h6 := int64(f[6]) * 121666
-	h7 := int64(f[7]) * 121666
-	h8 := int64(f[8]) * 121666
-	h9 := int64(f[9]) * 121666
-	var carry [10]int64
-
-	carry[9] = (h9 + (1 << 24)) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-	carry[1] = (h1 + (1 << 24)) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[3] = (h3 + (1 << 24)) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[5] = (h5 + (1 << 24)) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-	carry[7] = (h7 + (1 << 24)) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[2] = (h2 + (1 << 25)) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[6] = (h6 + (1 << 25)) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-	carry[8] = (h8 + (1 << 25)) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-
-	h[0] = int32(h0)
-	h[1] = int32(h1)
-	h[2] = int32(h2)
-	h[3] = int32(h3)
-	h[4] = int32(h4)
-	h[5] = int32(h5)
-	h[6] = int32(h6)
-	h[7] = int32(h7)
-	h[8] = int32(h8)
-	h[9] = int32(h9)
+// If point is Basepoint (but not if it's a different slice with the same
+// contents) a precomputed implementation might be used for performance.
+func X25519(scalar, point []byte) ([]byte, error) {
+	// Outline the body of function, to let the allocation be inlined in the
+	// caller, and possibly avoid escaping to the heap.
+	var dst [32]byte
+	return x25519(&dst, scalar, point)
 }
 
-// feInvert sets out = z^-1.
-func feInvert(out, z *fieldElement) {
-	var t0, t1, t2, t3 fieldElement
-	var i int
-
-	feSquare(&t0, z)
-	for i = 1; i < 1; i++ {
-		feSquare(&t0, &t0)
-	}
-	feSquare(&t1, &t0)
-	for i = 1; i < 2; i++ {
-		feSquare(&t1, &t1)
-	}
-	feMul(&t1, z, &t1)
-	feMul(&t0, &t0, &t1)
-	feSquare(&t2, &t0)
-	for i = 1; i < 1; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t1, &t1, &t2)
-	feSquare(&t2, &t1)
-	for i = 1; i < 5; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t1, &t2, &t1)
-	feSquare(&t2, &t1)
-	for i = 1; i < 10; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t2, &t2, &t1)
-	feSquare(&t3, &t2)
-	for i = 1; i < 20; i++ {
-		feSquare(&t3, &t3)
-	}
-	feMul(&t2, &t3, &t2)
-	feSquare(&t2, &t2)
-	for i = 1; i < 10; i++ {
-		feSquare(&t2, &t2)
+func x25519(dst *[32]byte, scalar, point []byte) ([]byte, error) {
+	var in [32]byte
+	if l := len(scalar); l != 32 {
+		return nil, fmt.Errorf("bad scalar length: %d, expected %d", l, 32)
 	}
-	feMul(&t1, &t2, &t1)
-	feSquare(&t2, &t1)
-	for i = 1; i < 50; i++ {
-		feSquare(&t2, &t2)
+	if l := len(point); l != 32 {
+		return nil, fmt.Errorf("bad point length: %d, expected %d", l, 32)
 	}
-	feMul(&t2, &t2, &t1)
-	feSquare(&t3, &t2)
-	for i = 1; i < 100; i++ {
-		feSquare(&t3, &t3)
+	copy(in[:], scalar)
+	if &point[0] == &Basepoint[0] {
+		checkBasepoint()
+		ScalarBaseMult(dst, &in)
+	} else {
+		var base, zero [32]byte
+		copy(base[:], point)
+		ScalarMult(dst, &in, &base)
+		if subtle.ConstantTimeCompare(dst[:], zero[:]) == 1 {
+			return nil, fmt.Errorf("bad input point: low order point")
 		}
-	feMul(&t2, &t3, &t2)
-	feSquare(&t2, &t2)
-	for i = 1; i < 50; i++ {
-		feSquare(&t2, &t2)
 	}
-	feMul(&t1, &t2, &t1)
-	feSquare(&t1, &t1)
-	for i = 1; i < 5; i++ {
-		feSquare(&t1, &t1)
-	}
-	feMul(out, &t1, &t0)
-}
-
-func scalarMult(out, in, base *[32]byte) {
-	var e [32]byte
-
-	copy(e[:], in[:])
-	e[0] &= 248
-	e[31] &= 127
-	e[31] |= 64
-
-	var x1, x2, z2, x3, z3, tmp0, tmp1 fieldElement
-	feFromBytes(&x1, base)
-	feOne(&x2)
-	feCopy(&x3, &x1)
-	feOne(&z3)
-
-	swap := int32(0)
-	for pos := 254; pos >= 0; pos-- {
-		b := e[pos/8] >> uint(pos&7)
-		b &= 1
-		swap ^= int32(b)
-		feCSwap(&x2, &x3, swap)
-		feCSwap(&z2, &z3, swap)
-		swap = int32(b)
-
-		feSub(&tmp0, &x3, &z3)
-		feSub(&tmp1, &x2, &z2)
-		feAdd(&x2, &x2, &z2)
-		feAdd(&z2, &x3, &z3)
-		feMul(&z3, &tmp0, &x2)
-		feMul(&z2, &z2, &tmp1)
-		feSquare(&tmp0, &tmp1)
-		feSquare(&tmp1, &x2)
-		feAdd(&x3, &z3, &z2)
-		feSub(&z2, &z3, &z2)
-		feMul(&x2, &tmp1, &tmp0)
-		feSub(&tmp1, &tmp1, &tmp0)
-		feSquare(&z2, &z2)
-		feMul121666(&z3, &tmp1)
-		feSquare(&x3, &x3)
-		feAdd(&tmp0, &tmp0, &z3)
-		feMul(&z3, &x1, &z2)
-		feMul(&z2, &tmp1, &tmp0)
-	}
-
-	feCSwap(&x2, &x3, swap)
-	feCSwap(&z2, &z3, swap)
-
-	feInvert(&z2, &z2)
-	feMul(&x2, &x2, &z2)
-	feToBytes(out, &x2)
+	return dst[:], nil
 }

src/vendor/golang.org/x/crypto/curve25519/mont25519_amd64.go → src/vendor/golang.org/x/crypto/curve25519/curve25519_amd64.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build amd64,!gccgo,!appengine
+// +build amd64,!gccgo,!appengine,!purego
 
 package curve25519
 

src/vendor/golang.org/x/crypto/curve25519/ladderstep_amd64.s → src/vendor/golang.org/x/crypto/curve25519/curve25519_amd64.s

@@ -5,9 +5,84 @@
 // This code was translated into a form compatible with 6a from the public
 // domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
 
-// +build amd64,!gccgo,!appengine
+// +build amd64,!gccgo,!appengine,!purego
 
-#include "const_amd64.h"
+#define REDMASK51     0x0007FFFFFFFFFFFF
+
+// These constants cannot be encoded in non-MOVQ immediates.
+// We access them directly from memory instead.
+
+DATA ·_121666_213(SB)/8, $996687872
+GLOBL ·_121666_213(SB), 8, $8
+
+DATA ·_2P0(SB)/8, $0xFFFFFFFFFFFDA
+GLOBL ·_2P0(SB), 8, $8
+
+DATA ·_2P1234(SB)/8, $0xFFFFFFFFFFFFE
+GLOBL ·_2P1234(SB), 8, $8
+
+// func freeze(inout *[5]uint64)
+TEXT ·freeze(SB),7,$0-8
+	MOVQ inout+0(FP), DI
+
+	MOVQ 0(DI),SI
+	MOVQ 8(DI),DX
+	MOVQ 16(DI),CX
+	MOVQ 24(DI),R8
+	MOVQ 32(DI),R9
+	MOVQ $REDMASK51,AX
+	MOVQ AX,R10
+	SUBQ $18,R10
+	MOVQ $3,R11
+REDUCELOOP:
+	MOVQ SI,R12
+	SHRQ $51,R12
+	ANDQ AX,SI
+	ADDQ R12,DX
+	MOVQ DX,R12
+	SHRQ $51,R12
+	ANDQ AX,DX
+	ADDQ R12,CX
+	MOVQ CX,R12
+	SHRQ $51,R12
+	ANDQ AX,CX
+	ADDQ R12,R8
+	MOVQ R8,R12
+	SHRQ $51,R12
+	ANDQ AX,R8
+	ADDQ R12,R9
+	MOVQ R9,R12
+	SHRQ $51,R12
+	ANDQ AX,R9
+	IMUL3Q $19,R12,R12
+	ADDQ R12,SI
+	SUBQ $1,R11
+	JA REDUCELOOP
+	MOVQ $1,R12
+	CMPQ R10,SI
+	CMOVQLT R11,R12
+	CMPQ AX,DX
+	CMOVQNE R11,R12
+	CMPQ AX,CX
+	CMOVQNE R11,R12
+	CMPQ AX,R8
+	CMOVQNE R11,R12
+	CMPQ AX,R9
+	CMOVQNE R11,R12
+	NEGQ R12
+	ANDQ R12,AX
+	ANDQ R12,R10
+	SUBQ R10,SI
+	SUBQ AX,DX
+	SUBQ AX,CX
+	SUBQ AX,R8
+	SUBQ AX,R9
+	MOVQ SI,0(DI)
+	MOVQ DX,8(DI)
+	MOVQ CX,16(DI)
+	MOVQ R8,24(DI)
+	MOVQ R9,32(DI)
+	RET
 
 // func ladderstep(inout *[5][5]uint64)
 TEXT ·ladderstep(SB),0,$296-8
@@ -1375,3 +1450,344 @@ TEXT ·ladderstep(SB),0,$296-8
 	MOVQ AX,104(DI)
 	MOVQ R10,112(DI)
 	RET
+
+// func cswap(inout *[4][5]uint64, v uint64)
+TEXT ·cswap(SB),7,$0
+	MOVQ inout+0(FP),DI
+	MOVQ v+8(FP),SI
+
+	SUBQ $1, SI
+	NOTQ SI
+	MOVQ SI, X15
+	PSHUFD $0x44, X15, X15
+
+	MOVOU 0(DI), X0
+	MOVOU 16(DI), X2
+	MOVOU 32(DI), X4
+	MOVOU 48(DI), X6
+	MOVOU 64(DI), X8
+	MOVOU 80(DI), X1
+	MOVOU 96(DI), X3
+	MOVOU 112(DI), X5
+	MOVOU 128(DI), X7
+	MOVOU 144(DI), X9
+
+	MOVO X1, X10
+	MOVO X3, X11
+	MOVO X5, X12
+	MOVO X7, X13
+	MOVO X9, X14
+
+	PXOR X0, X10
+	PXOR X2, X11
+	PXOR X4, X12
+	PXOR X6, X13
+	PXOR X8, X14
+	PAND X15, X10
+	PAND X15, X11
+	PAND X15, X12
+	PAND X15, X13
+	PAND X15, X14
+	PXOR X10, X0
+	PXOR X10, X1
+	PXOR X11, X2
+	PXOR X11, X3
+	PXOR X12, X4
+	PXOR X12, X5
+	PXOR X13, X6
+	PXOR X13, X7
+	PXOR X14, X8
+	PXOR X14, X9
+
+	MOVOU X0, 0(DI)
+	MOVOU X2, 16(DI)
+	MOVOU X4, 32(DI)
+	MOVOU X6, 48(DI)
+	MOVOU X8, 64(DI)
+	MOVOU X1, 80(DI)
+	MOVOU X3, 96(DI)
+	MOVOU X5, 112(DI)
+	MOVOU X7, 128(DI)
+	MOVOU X9, 144(DI)
+	RET
+
+// func mul(dest, a, b *[5]uint64)
+TEXT ·mul(SB),0,$16-24
+	MOVQ dest+0(FP), DI
+	MOVQ a+8(FP), SI
+	MOVQ b+16(FP), DX
+
+	MOVQ DX,CX
+	MOVQ 24(SI),DX
+	IMUL3Q $19,DX,AX
+	MOVQ AX,0(SP)
+	MULQ 16(CX)
+	MOVQ AX,R8
+	MOVQ DX,R9
+	MOVQ 32(SI),DX
+	IMUL3Q $19,DX,AX
+	MOVQ AX,8(SP)
+	MULQ 8(CX)
+	ADDQ AX,R8
+	ADCQ DX,R9
+	MOVQ 0(SI),AX
+	MULQ 0(CX)
+	ADDQ AX,R8
+	ADCQ DX,R9
+	MOVQ 0(SI),AX
+	MULQ 8(CX)
+	MOVQ AX,R10
+	MOVQ DX,R11
+	MOVQ 0(SI),AX
+	MULQ 16(CX)
+	MOVQ AX,R12
+	MOVQ DX,R13
+	MOVQ 0(SI),AX
+	MULQ 24(CX)
+	MOVQ AX,R14
+	MOVQ DX,R15
+	MOVQ 0(SI),AX
+	MULQ 32(CX)
+	MOVQ AX,BX
+	MOVQ DX,BP
+	MOVQ 8(SI),AX
+	MULQ 0(CX)
+	ADDQ AX,R10
+	ADCQ DX,R11
+	MOVQ 8(SI),AX
+	MULQ 8(CX)
+	ADDQ AX,R12
+	ADCQ DX,R13
+	MOVQ 8(SI),AX
+	MULQ 16(CX)
+	ADDQ AX,R14
+	ADCQ DX,R15
+	MOVQ 8(SI),AX
+	MULQ 24(CX)
+	ADDQ AX,BX
+	ADCQ DX,BP
+	MOVQ 8(SI),DX
+	IMUL3Q $19,DX,AX
+	MULQ 32(CX)
+	ADDQ AX,R8
+	ADCQ DX,R9
+	MOVQ 16(SI),AX
+	MULQ 0(CX)
+	ADDQ AX,R12
+	ADCQ DX,R13
+	MOVQ 16(SI),AX
+	MULQ 8(CX)
+	ADDQ AX,R14
+	ADCQ DX,R15
+	MOVQ 16(SI),AX
+	MULQ 16(CX)
+	ADDQ AX,BX
+	ADCQ DX,BP
+	MOVQ 16(SI),DX
+	IMUL3Q $19,DX,AX
+	MULQ 24(CX)
+	ADDQ AX,R8
+	ADCQ DX,R9
+	MOVQ 16(SI),DX
+	IMUL3Q $19,DX,AX
+	MULQ 32(CX)
+	ADDQ AX,R10
+	ADCQ DX,R11
+	MOVQ 24(SI),AX
+	MULQ 0(CX)
+	ADDQ AX,R14
+	ADCQ DX,R15
+	MOVQ 24(SI),AX
+	MULQ 8(CX)
+	ADDQ AX,BX
+	ADCQ DX,BP
+	MOVQ 0(SP),AX
+	MULQ 24(CX)
+	ADDQ AX,R10
+	ADCQ DX,R11
+	MOVQ 0(SP),AX
+	MULQ 32(CX)
+	ADDQ AX,R12
+	ADCQ DX,R13
+	MOVQ 32(SI),AX
+	MULQ 0(CX)
+	ADDQ AX,BX
+	ADCQ DX,BP
+	MOVQ 8(SP),AX
+	MULQ 16(CX)
+	ADDQ AX,R10
+	ADCQ DX,R11
+	MOVQ 8(SP),AX
+	MULQ 24(CX)
+	ADDQ AX,R12
+	ADCQ DX,R13
+	MOVQ 8(SP),AX
+	MULQ 32(CX)
+	ADDQ AX,R14
+	ADCQ DX,R15
+	MOVQ $REDMASK51,SI
+	SHLQ $13,R8,R9
+	ANDQ SI,R8
+	SHLQ $13,R10,R11
+	ANDQ SI,R10
+	ADDQ R9,R10
+	SHLQ $13,R12,R13
+	ANDQ SI,R12
+	ADDQ R11,R12
+	SHLQ $13,R14,R15
+	ANDQ SI,R14
+	ADDQ R13,R14
+	SHLQ $13,BX,BP
+	ANDQ SI,BX
+	ADDQ R15,BX
+	IMUL3Q $19,BP,DX
+	ADDQ DX,R8
+	MOVQ R8,DX
+	SHRQ $51,DX
+	ADDQ R10,DX
+	MOVQ DX,CX
+	SHRQ $51,DX
+	ANDQ SI,R8
+	ADDQ R12,DX
+	MOVQ DX,R9
+	SHRQ $51,DX
+	ANDQ SI,CX
+	ADDQ R14,DX
+	MOVQ DX,AX
+	SHRQ $51,DX
+	ANDQ SI,R9
+	ADDQ BX,DX
+	MOVQ DX,R10
+	SHRQ $51,DX
+	ANDQ SI,AX
+	IMUL3Q $19,DX,DX
+	ADDQ DX,R8
+	ANDQ SI,R10
+	MOVQ R8,0(DI)
+	MOVQ CX,8(DI)
+	MOVQ R9,16(DI)
+	MOVQ AX,24(DI)
+	MOVQ R10,32(DI)
+	RET
+
+// func square(out, in *[5]uint64)
+TEXT ·square(SB),7,$0-16
+	MOVQ out+0(FP), DI
+	MOVQ in+8(FP), SI
+
+	MOVQ 0(SI),AX
+	MULQ 0(SI)
+	MOVQ AX,CX
+	MOVQ DX,R8
+	MOVQ 0(SI),AX
+	SHLQ $1,AX
+	MULQ 8(SI)
+	MOVQ AX,R9
+	MOVQ DX,R10
+	MOVQ 0(SI),AX
+	SHLQ $1,AX
+	MULQ 16(SI)
+	MOVQ AX,R11
+	MOVQ DX,R12
+	MOVQ 0(SI),AX
+	SHLQ $1,AX
+	MULQ 24(SI)
+	MOVQ AX,R13
+	MOVQ DX,R14
+	MOVQ 0(SI),AX
+	SHLQ $1,AX
+	MULQ 32(SI)
+	MOVQ AX,R15
+	MOVQ DX,BX
+	MOVQ 8(SI),AX
+	MULQ 8(SI)
+	ADDQ AX,R11
+	ADCQ DX,R12
+	MOVQ 8(SI),AX
+	SHLQ $1,AX
+	MULQ 16(SI)
+	ADDQ AX,R13
+	ADCQ DX,R14
+	MOVQ 8(SI),AX
+	SHLQ $1,AX
+	MULQ 24(SI)
+	ADDQ AX,R15
+	ADCQ DX,BX
+	MOVQ 8(SI),DX
+	IMUL3Q $38,DX,AX
+	MULQ 32(SI)
+	ADDQ AX,CX
+	ADCQ DX,R8
+	MOVQ 16(SI),AX
+	MULQ 16(SI)
+	ADDQ AX,R15
+	ADCQ DX,BX
+	MOVQ 16(SI),DX
+	IMUL3Q $38,DX,AX
+	MULQ 24(SI)
+	ADDQ AX,CX
+	ADCQ DX,R8
+	MOVQ 16(SI),DX
+	IMUL3Q $38,DX,AX
+	MULQ 32(SI)
+	ADDQ AX,R9
+	ADCQ DX,R10
+	MOVQ 24(SI),DX
+	IMUL3Q $19,DX,AX
+	MULQ 24(SI)
+	ADDQ AX,R9
+	ADCQ DX,R10
+	MOVQ 24(SI),DX
+	IMUL3Q $38,DX,AX
+	MULQ 32(SI)
+	ADDQ AX,R11
+	ADCQ DX,R12
+	MOVQ 32(SI),DX
+	IMUL3Q $19,DX,AX
+	MULQ 32(SI)
+	ADDQ AX,R13
+	ADCQ DX,R14
+	MOVQ $REDMASK51,SI
+	SHLQ $13,CX,R8
+	ANDQ SI,CX
+	SHLQ $13,R9,R10
+	ANDQ SI,R9
+	ADDQ R8,R9
+	SHLQ $13,R11,R12
+	ANDQ SI,R11
+	ADDQ R10,R11
+	SHLQ $13,R13,R14
+	ANDQ SI,R13
+	ADDQ R12,R13
+	SHLQ $13,R15,BX
+	ANDQ SI,R15
+	ADDQ R14,R15
+	IMUL3Q $19,BX,DX
+	ADDQ DX,CX
+	MOVQ CX,DX
+	SHRQ $51,DX
+	ADDQ R9,DX
+	ANDQ SI,CX
+	MOVQ DX,R8
+	SHRQ $51,DX
+	ADDQ R11,DX
+	ANDQ SI,R8
+	MOVQ DX,R9
+	SHRQ $51,DX
+	ADDQ R13,DX
+	ANDQ SI,R9
+	MOVQ DX,AX
+	SHRQ $51,DX
+	ADDQ R15,DX
+	ANDQ SI,AX
+	MOVQ DX,R10
+	SHRQ $51,DX
+	IMUL3Q $19,DX,DX
+	ADDQ DX,CX
+	ANDQ SI,R10
+	MOVQ CX,0(DI)
+	MOVQ R8,8(DI)
+	MOVQ R9,16(DI)
+	MOVQ AX,24(DI)
+	MOVQ R10,32(DI)
+	RET

src/vendor/golang.org/x/crypto/curve25519/curve25519_generic.go

+// 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 curve25519
+
+import "encoding/binary"
+
+// This code is a port of the public domain, "ref10" implementation of
+// curve25519 from SUPERCOP 20130419 by D. J. Bernstein.
+
+// fieldElement represents an element of the field GF(2^255 - 19). An element
+// t, entries t[0]...t[9], represents the integer t[0]+2^26 t[1]+2^51 t[2]+2^77
+// t[3]+2^102 t[4]+...+2^230 t[9]. Bounds on each t[i] vary depending on
+// context.
+type fieldElement [10]int32
+
+func feZero(fe *fieldElement) {
+	for i := range fe {
+		fe[i] = 0
+	}
+}
+
+func feOne(fe *fieldElement) {
+	feZero(fe)
+	fe[0] = 1
+}
+
+func feAdd(dst, a, b *fieldElement) {
+	for i := range dst {
+		dst[i] = a[i] + b[i]
+	}
+}
+
+func feSub(dst, a, b *fieldElement) {
+	for i := range dst {
+		dst[i] = a[i] - b[i]
+	}
+}
+
+func feCopy(dst, src *fieldElement) {
+	for i := range dst {
+		dst[i] = src[i]
+	}
+}
+
+// feCSwap replaces (f,g) with (g,f) if b == 1; replaces (f,g) with (f,g) if b == 0.
+//
+// Preconditions: b in {0,1}.
+func feCSwap(f, g *fieldElement, b int32) {
+	b = -b
+	for i := range f {
+		t := b & (f[i] ^ g[i])
+		f[i] ^= t
+		g[i] ^= t
+	}
+}
+
+// load3 reads a 24-bit, little-endian value from in.
+func load3(in []byte) int64 {
+	var r int64
+	r = int64(in[0])
+	r |= int64(in[1]) << 8
+	r |= int64(in[2]) << 16
+	return r
+}
+
+// load4 reads a 32-bit, little-endian value from in.
+func load4(in []byte) int64 {
+	return int64(binary.LittleEndian.Uint32(in))
+}
+
+func feFromBytes(dst *fieldElement, src *[32]byte) {
+	h0 := load4(src[:])
+	h1 := load3(src[4:]) << 6
+	h2 := load3(src[7:]) << 5
+	h3 := load3(src[10:]) << 3
+	h4 := load3(src[13:]) << 2
+	h5 := load4(src[16:])
+	h6 := load3(src[20:]) << 7
+	h7 := load3(src[23:]) << 5
+	h8 := load3(src[26:]) << 4
+	h9 := (load3(src[29:]) & 0x7fffff) << 2
+
+	var carry [10]int64
+	carry[9] = (h9 + 1<<24) >> 25
+	h0 += carry[9] * 19
+	h9 -= carry[9] << 25
+	carry[1] = (h1 + 1<<24) >> 25
+	h2 += carry[1]
+	h1 -= carry[1] << 25
+	carry[3] = (h3 + 1<<24) >> 25
+	h4 += carry[3]
+	h3 -= carry[3] << 25
+	carry[5] = (h5 + 1<<24) >> 25
+	h6 += carry[5]
+	h5 -= carry[5] << 25
+	carry[7] = (h7 + 1<<24) >> 25
+	h8 += carry[7]
+	h7 -= carry[7] << 25
+
+	carry[0] = (h0 + 1<<25) >> 26
+	h1 += carry[0]
+	h0 -= carry[0] << 26
+	carry[2] = (h2 + 1<<25) >> 26
+	h3 += carry[2]
+	h2 -= carry[2] << 26
+	carry[4] = (h4 + 1<<25) >> 26
+	h5 += carry[4]
+	h4 -= carry[4] << 26
+	carry[6] = (h6 + 1<<25) >> 26
+	h7 += carry[6]
+	h6 -= carry[6] << 26
+	carry[8] = (h8 + 1<<25) >> 26
+	h9 += carry[8]
+	h8 -= carry[8] << 26
+
+	dst[0] = int32(h0)
+	dst[1] = int32(h1)
+	dst[2] = int32(h2)
+	dst[3] = int32(h3)
+	dst[4] = int32(h4)
+	dst[5] = int32(h5)
+	dst[6] = int32(h6)
+	dst[7] = int32(h7)
+	dst[8] = int32(h8)
+	dst[9] = int32(h9)
+}
+
+// feToBytes marshals h to s.
+// Preconditions:
+//   |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+//
+// Write p=2^255-19; q=floor(h/p).
+// Basic claim: q = floor(2^(-255)(h + 19 2^(-25)h9 + 2^(-1))).
+//
+// Proof:
+//   Have |h|<=p so |q|<=1 so |19^2 2^(-255) q|<1/4.
+//   Also have |h-2^230 h9|<2^230 so |19 2^(-255)(h-2^230 h9)|<1/4.
+//
+//   Write y=2^(-1)-19^2 2^(-255)q-19 2^(-255)(h-2^230 h9).
+//   Then 0<y<1.
+//
+//   Write r=h-pq.
+//   Have 0<=r<=p-1=2^255-20.
+//   Thus 0<=r+19(2^-255)r<r+19(2^-255)2^255<=2^255-1.
+//
+//   Write x=r+19(2^-255)r+y.
+//   Then 0<x<2^255 so floor(2^(-255)x) = 0 so floor(q+2^(-255)x) = q.
+//
+//   Have q+2^(-255)x = 2^(-255)(h + 19 2^(-25) h9 + 2^(-1))
+//   so floor(2^(-255)(h + 19 2^(-25) h9 + 2^(-1))) = q.
+func feToBytes(s *[32]byte, h *fieldElement) {
+	var carry [10]int32
+
+	q := (19*h[9] + (1 << 24)) >> 25
+	q = (h[0] + q) >> 26
+	q = (h[1] + q) >> 25
+	q = (h[2] + q) >> 26
+	q = (h[3] + q) >> 25
+	q = (h[4] + q) >> 26
+	q = (h[5] + q) >> 25
+	q = (h[6] + q) >> 26
+	q = (h[7] + q) >> 25
+	q = (h[8] + q) >> 26
+	q = (h[9] + q) >> 25
+
+	// Goal: Output h-(2^255-19)q, which is between 0 and 2^255-20.
+	h[0] += 19 * q
+	// Goal: Output h-2^255 q, which is between 0 and 2^255-20.
+
+	carry[0] = h[0] >> 26
+	h[1] += carry[0]
+	h[0] -= carry[0] << 26
+	carry[1] = h[1] >> 25
+	h[2] += carry[1]
+	h[1] -= carry[1] << 25
+	carry[2] = h[2] >> 26
+	h[3] += carry[2]
+	h[2] -= carry[2] << 26
+	carry[3] = h[3] >> 25
+	h[4] += carry[3]
+	h[3] -= carry[3] << 25
+	carry[4] = h[4] >> 26
+	h[5] += carry[4]
+	h[4] -= carry[4] << 26
+	carry[5] = h[5] >> 25
+	h[6] += carry[5]
+	h[5] -= carry[5] << 25
+	carry[6] = h[6] >> 26
+	h[7] += carry[6]
+	h[6] -= carry[6] << 26
+	carry[7] = h[7] >> 25
+	h[8] += carry[7]
+	h[7] -= carry[7] << 25
+	carry[8] = h[8] >> 26
+	h[9] += carry[8]
+	h[8] -= carry[8] << 26
+	carry[9] = h[9] >> 25
+	h[9] -= carry[9] << 25
+	// h10 = carry9
+
+	// Goal: Output h[0]+...+2^255 h10-2^255 q, which is between 0 and 2^255-20.
+	// Have h[0]+...+2^230 h[9] between 0 and 2^255-1;
+	// evidently 2^255 h10-2^255 q = 0.
+	// Goal: Output h[0]+...+2^230 h[9].
+
+	s[0] = byte(h[0] >> 0)
+	s[1] = byte(h[0] >> 8)
+	s[2] = byte(h[0] >> 16)
+	s[3] = byte((h[0] >> 24) | (h[1] << 2))
+	s[4] = byte(h[1] >> 6)
+	s[5] = byte(h[1] >> 14)
+	s[6] = byte((h[1] >> 22) | (h[2] << 3))
+	s[7] = byte(h[2] >> 5)
+	s[8] = byte(h[2] >> 13)
+	s[9] = byte((h[2] >> 21) | (h[3] << 5))
+	s[10] = byte(h[3] >> 3)
+	s[11] = byte(h[3] >> 11)
+	s[12] = byte((h[3] >> 19) | (h[4] << 6))
+	s[13] = byte(h[4] >> 2)
+	s[14] = byte(h[4] >> 10)
+	s[15] = byte(h[4] >> 18)
+	s[16] = byte(h[5] >> 0)
+	s[17] = byte(h[5] >> 8)
+	s[18] = byte(h[5] >> 16)
+	s[19] = byte((h[5] >> 24) | (h[6] << 1))
+	s[20] = byte(h[6] >> 7)
+	s[21] = byte(h[6] >> 15)
+	s[22] = byte((h[6] >> 23) | (h[7] << 3))
+	s[23] = byte(h[7] >> 5)
+	s[24] = byte(h[7] >> 13)
+	s[25] = byte((h[7] >> 21) | (h[8] << 4))
+	s[26] = byte(h[8] >> 4)
+	s[27] = byte(h[8] >> 12)
+	s[28] = byte((h[8] >> 20) | (h[9] << 6))
+	s[29] = byte(h[9] >> 2)
+	s[30] = byte(h[9] >> 10)
+	s[31] = byte(h[9] >> 18)
+}
+
+// feMul calculates h = f * g
+// Can overlap h with f or g.
+//
+// Preconditions:
+//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+//    |g| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+//
+// Postconditions:
+//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+//
+// Notes on implementation strategy:
+//
+// Using schoolbook multiplication.
+// Karatsuba would save a little in some cost models.
+//
+// Most multiplications by 2 and 19 are 32-bit precomputations;
+// cheaper than 64-bit postcomputations.
+//
+// There is one remaining multiplication by 19 in the carry chain;
+// one *19 precomputation can be merged into this,
+// but the resulting data flow is considerably less clean.
+//
+// There are 12 carries below.
+// 10 of them are 2-way parallelizable and vectorizable.
+// Can get away with 11 carries, but then data flow is much deeper.
+//
+// With tighter constraints on inputs can squeeze carries into int32.
+func feMul(h, f, g *fieldElement) {
+	f0 := f[0]
+	f1 := f[1]
+	f2 := f[2]
+	f3 := f[3]
+	f4 := f[4]
+	f5 := f[5]
+	f6 := f[6]
+	f7 := f[7]
+	f8 := f[8]
+	f9 := f[9]
+	g0 := g[0]
+	g1 := g[1]
+	g2 := g[2]
+	g3 := g[3]
+	g4 := g[4]
+	g5 := g[5]
+	g6 := g[6]
+	g7 := g[7]
+	g8 := g[8]
+	g9 := g[9]
+	g1_19 := 19 * g1 // 1.4*2^29
+	g2_19 := 19 * g2 // 1.4*2^30; still ok
+	g3_19 := 19 * g3
+	g4_19 := 19 * g4
+	g5_19 := 19 * g5
+	g6_19 := 19 * g6
+	g7_19 := 19 * g7
+	g8_19 := 19 * g8
+	g9_19 := 19 * g9
+	f1_2 := 2 * f1
+	f3_2 := 2 * f3
+	f5_2 := 2 * f5
+	f7_2 := 2 * f7
+	f9_2 := 2 * f9
+	f0g0 := int64(f0) * int64(g0)
+	f0g1 := int64(f0) * int64(g1)
+	f0g2 := int64(f0) * int64(g2)
+	f0g3 := int64(f0) * int64(g3)
+	f0g4 := int64(f0) * int64(g4)
+	f0g5 := int64(f0) * int64(g5)
+	f0g6 := int64(f0) * int64(g6)
+	f0g7 := int64(f0) * int64(g7)
+	f0g8 := int64(f0) * int64(g8)
+	f0g9 := int64(f0) * int64(g9)
+	f1g0 := int64(f1) * int64(g0)
+	f1g1_2 := int64(f1_2) * int64(g1)
+	f1g2 := int64(f1) * int64(g2)
+	f1g3_2 := int64(f1_2) * int64(g3)
+	f1g4 := int64(f1) * int64(g4)
+	f1g5_2 := int64(f1_2) * int64(g5)
+	f1g6 := int64(f1) * int64(g6)
+	f1g7_2 := int64(f1_2) * int64(g7)
+	f1g8 := int64(f1) * int64(g8)
+	f1g9_38 := int64(f1_2) * int64(g9_19)
+	f2g0 := int64(f2) * int64(g0)
+	f2g1 := int64(f2) * int64(g1)
+	f2g2 := int64(f2) * int64(g2)
+	f2g3 := int64(f2) * int64(g3)
+	f2g4 := int64(f2) * int64(g4)
+	f2g5 := int64(f2) * int64(g5)
+	f2g6 := int64(f2) * int64(g6)
+	f2g7 := int64(f2) * int64(g7)
+	f2g8_19 := int64(f2) * int64(g8_19)
+	f2g9_19 := int64(f2) * int64(g9_19)
+	f3g0 := int64(f3) * int64(g0)
+	f3g1_2 := int64(f3_2) * int64(g1)
+	f3g2 := int64(f3) * int64(g2)
+	f3g3_2 := int64(f3_2) * int64(g3)
+	f3g4 := int64(f3) * int64(g4)
+	f3g5_2 := int64(f3_2) * int64(g5)
+	f3g6 := int64(f3) * int64(g6)
+	f3g7_38 := int64(f3_2) * int64(g7_19)
+	f3g8_19 := int64(f3) * int64(g8_19)
+	f3g9_38 := int64(f3_2) * int64(g9_19)
+	f4g0 := int64(f4) * int64(g0)
+	f4g1 := int64(f4) * int64(g1)
+	f4g2 := int64(f4) * int64(g2)
+	f4g3 := int64(f4) * int64(g3)
+	f4g4 := int64(f4) * int64(g4)
+	f4g5 := int64(f4) * int64(g5)
+	f4g6_19 := int64(f4) * int64(g6_19)
+	f4g7_19 := int64(f4) * int64(g7_19)
+	f4g8_19 := int64(f4) * int64(g8_19)
+	f4g9_19 := int64(f4) * int64(g9_19)
+	f5g0 := int64(f5) * int64(g0)
+	f5g1_2 := int64(f5_2) * int64(g1)
+	f5g2 := int64(f5) * int64(g2)
+	f5g3_2 := int64(f5_2) * int64(g3)
+	f5g4 := int64(f5) * int64(g4)
+	f5g5_38 := int64(f5_2) * int64(g5_19)
+	f5g6_19 := int64(f5) * int64(g6_19)
+	f5g7_38 := int64(f5_2) * int64(g7_19)
+	f5g8_19 := int64(f5) * int64(g8_19)
+	f5g9_38 := int64(f5_2) * int64(g9_19)
+	f6g0 := int64(f6) * int64(g0)
+	f6g1 := int64(f6) * int64(g1)
+	f6g2 := int64(f6) * int64(g2)
+	f6g3 := int64(f6) * int64(g3)
+	f6g4_19 := int64(f6) * int64(g4_19)
+	f6g5_19 := int64(f6) * int64(g5_19)
+	f6g6_19 := int64(f6) * int64(g6_19)
+	f6g7_19 := int64(f6) * int64(g7_19)
+	f6g8_19 := int64(f6) * int64(g8_19)
+	f6g9_19 := int64(f6) * int64(g9_19)
+	f7g0 := int64(f7) * int64(g0)
+	f7g1_2 := int64(f7_2) * int64(g1)
+	f7g2 := int64(f7) * int64(g2)
+	f7g3_38 := int64(f7_2) * int64(g3_19)
+	f7g4_19 := int64(f7) * int64(g4_19)
+	f7g5_38 := int64(f7_2) * int64(g5_19)
+	f7g6_19 := int64(f7) * int64(g6_19)
+	f7g7_38 := int64(f7_2) * int64(g7_19)
+	f7g8_19 := int64(f7) * int64(g8_19)
+	f7g9_38 := int64(f7_2) * int64(g9_19)
+	f8g0 := int64(f8) * int64(g0)
+	f8g1 := int64(f8) * int64(g1)
+	f8g2_19 := int64(f8) * int64(g2_19)
+	f8g3_19 := int64(f8) * int64(g3_19)
+	f8g4_19 := int64(f8) * int64(g4_19)
+	f8g5_19 := int64(f8) * int64(g5_19)
+	f8g6_19 := int64(f8) * int64(g6_19)
+	f8g7_19 := int64(f8) * int64(g7_19)
+	f8g8_19 := int64(f8) * int64(g8_19)
+	f8g9_19 := int64(f8) * int64(g9_19)
+	f9g0 := int64(f9) * int64(g0)
+	f9g1_38 := int64(f9_2) * int64(g1_19)
+	f9g2_19 := int64(f9) * int64(g2_19)
+	f9g3_38 := int64(f9_2) * int64(g3_19)
+	f9g4_19 := int64(f9) * int64(g4_19)
+	f9g5_38 := int64(f9_2) * int64(g5_19)
+	f9g6_19 := int64(f9) * int64(g6_19)
+	f9g7_38 := int64(f9_2) * int64(g7_19)
+	f9g8_19 := int64(f9) * int64(g8_19)
+	f9g9_38 := int64(f9_2) * int64(g9_19)
+	h0 := f0g0 + f1g9_38 + f2g8_19 + f3g7_38 + f4g6_19 + f5g5_38 + f6g4_19 + f7g3_38 + f8g2_19 + f9g1_38
+	h1 := f0g1 + f1g0 + f2g9_19 + f3g8_19 + f4g7_19 + f5g6_19 + f6g5_19 + f7g4_19 + f8g3_19 + f9g2_19
+	h2 := f0g2 + f1g1_2 + f2g0 + f3g9_38 + f4g8_19 + f5g7_38 + f6g6_19 + f7g5_38 + f8g4_19 + f9g3_38
+	h3 := f0g3 + f1g2 + f2g1 + f3g0 + f4g9_19 + f5g8_19 + f6g7_19 + f7g6_19 + f8g5_19 + f9g4_19
+	h4 := f0g4 + f1g3_2 + f2g2 + f3g1_2 + f4g0 + f5g9_38 + f6g8_19 + f7g7_38 + f8g6_19 + f9g5_38
+	h5 := f0g5 + f1g4 + f2g3 + f3g2 + f4g1 + f5g0 + f6g9_19 + f7g8_19 + f8g7_19 + f9g6_19
+	h6 := f0g6 + f1g5_2 + f2g4 + f3g3_2 + f4g2 + f5g1_2 + f6g0 + f7g9_38 + f8g8_19 + f9g7_38
+	h7 := f0g7 + f1g6 + f2g5 + f3g4 + f4g3 + f5g2 + f6g1 + f7g0 + f8g9_19 + f9g8_19
+	h8 := f0g8 + f1g7_2 + f2g6 + f3g5_2 + f4g4 + f5g3_2 + f6g2 + f7g1_2 + f8g0 + f9g9_38
+	h9 := f0g9 + f1g8 + f2g7 + f3g6 + f4g5 + f5g4 + f6g3 + f7g2 + f8g1 + f9g0
+	var carry [10]int64
+
+	// |h0| <= (1.1*1.1*2^52*(1+19+19+19+19)+1.1*1.1*2^50*(38+38+38+38+38))
+	//   i.e. |h0| <= 1.2*2^59; narrower ranges for h2, h4, h6, h8
+	// |h1| <= (1.1*1.1*2^51*(1+1+19+19+19+19+19+19+19+19))
+	//   i.e. |h1| <= 1.5*2^58; narrower ranges for h3, h5, h7, h9
+
+	carry[0] = (h0 + (1 << 25)) >> 26
+	h1 += carry[0]
+	h0 -= carry[0] << 26
+	carry[4] = (h4 + (1 << 25)) >> 26
+	h5 += carry[4]
+	h4 -= carry[4] << 26
+	// |h0| <= 2^25
+	// |h4| <= 2^25
+	// |h1| <= 1.51*2^58
+	// |h5| <= 1.51*2^58
+
+	carry[1] = (h1 + (1 << 24)) >> 25
+	h2 += carry[1]
+	h1 -= carry[1] << 25
+	carry[5] = (h5 + (1 << 24)) >> 25
+	h6 += carry[5]
+	h5 -= carry[5] << 25
+	// |h1| <= 2^24; from now on fits into int32
+	// |h5| <= 2^24; from now on fits into int32
+	// |h2| <= 1.21*2^59
+	// |h6| <= 1.21*2^59
+
+	carry[2] = (h2 + (1 << 25)) >> 26
+	h3 += carry[2]
+	h2 -= carry[2] << 26
+	carry[6] = (h6 + (1 << 25)) >> 26
+	h7 += carry[6]
+	h6 -= carry[6] << 26
+	// |h2| <= 2^25; from now on fits into int32 unchanged
+	// |h6| <= 2^25; from now on fits into int32 unchanged
+	// |h3| <= 1.51*2^58
+	// |h7| <= 1.51*2^58
+
+	carry[3] = (h3 + (1 << 24)) >> 25
+	h4 += carry[3]
+	h3 -= carry[3] << 25
+	carry[7] = (h7 + (1 << 24)) >> 25
+	h8 += carry[7]
+	h7 -= carry[7] << 25
+	// |h3| <= 2^24; from now on fits into int32 unchanged
+	// |h7| <= 2^24; from now on fits into int32 unchanged
+	// |h4| <= 1.52*2^33
+	// |h8| <= 1.52*2^33
+
+	carry[4] = (h4 + (1 << 25)) >> 26
+	h5 += carry[4]
+	h4 -= carry[4] << 26
+	carry[8] = (h8 + (1 << 25)) >> 26
+	h9 += carry[8]
+	h8 -= carry[8] << 26
+	// |h4| <= 2^25; from now on fits into int32 unchanged
+	// |h8| <= 2^25; from now on fits into int32 unchanged
+	// |h5| <= 1.01*2^24
+	// |h9| <= 1.51*2^58
+
+	carry[9] = (h9 + (1 << 24)) >> 25
+	h0 += carry[9] * 19
+	h9 -= carry[9] << 25
+	// |h9| <= 2^24; from now on fits into int32 unchanged
+	// |h0| <= 1.8*2^37
+
+	carry[0] = (h0 + (1 << 25)) >> 26
+	h1 += carry[0]
+	h0 -= carry[0] << 26
+	// |h0| <= 2^25; from now on fits into int32 unchanged
+	// |h1| <= 1.01*2^24
+
+	h[0] = int32(h0)
+	h[1] = int32(h1)
+	h[2] = int32(h2)
+	h[3] = int32(h3)
+	h[4] = int32(h4)
+	h[5] = int32(h5)
+	h[6] = int32(h6)
+	h[7] = int32(h7)
+	h[8] = int32(h8)
+	h[9] = int32(h9)
+}
+
+// feSquare calculates h = f*f. Can overlap h with f.
+//
+// Preconditions:
+//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+//
+// Postconditions:
+//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+func feSquare(h, f *fieldElement) {
+	f0 := f[0]
+	f1 := f[1]
+	f2 := f[2]
+	f3 := f[3]
+	f4 := f[4]
+	f5 := f[5]
+	f6 := f[6]
+	f7 := f[7]
+	f8 := f[8]
+	f9 := f[9]
+	f0_2 := 2 * f0
+	f1_2 := 2 * f1
+	f2_2 := 2 * f2
+	f3_2 := 2 * f3
+	f4_2 := 2 * f4
+	f5_2 := 2 * f5
+	f6_2 := 2 * f6
+	f7_2 := 2 * f7
+	f5_38 := 38 * f5 // 1.31*2^30
+	f6_19 := 19 * f6 // 1.31*2^30
+	f7_38 := 38 * f7 // 1.31*2^30
+	f8_19 := 19 * f8 // 1.31*2^30
+	f9_38 := 38 * f9 // 1.31*2^30
+	f0f0 := int64(f0) * int64(f0)
+	f0f1_2 := int64(f0_2) * int64(f1)
+	f0f2_2 := int64(f0_2) * int64(f2)
+	f0f3_2 := int64(f0_2) * int64(f3)
+	f0f4_2 := int64(f0_2) * int64(f4)
+	f0f5_2 := int64(f0_2) * int64(f5)
+	f0f6_2 := int64(f0_2) * int64(f6)
+	f0f7_2 := int64(f0_2) * int64(f7)
+	f0f8_2 := int64(f0_2) * int64(f8)
+	f0f9_2 := int64(f0_2) * int64(f9)
+	f1f1_2 := int64(f1_2) * int64(f1)
+	f1f2_2 := int64(f1_2) * int64(f2)
+	f1f3_4 := int64(f1_2) * int64(f3_2)
+	f1f4_2 := int64(f1_2) * int64(f4)
+	f1f5_4 := int64(f1_2) * int64(f5_2)
+	f1f6_2 := int64(f1_2) * int64(f6)
+	f1f7_4 := int64(f1_2) * int64(f7_2)
+	f1f8_2 := int64(f1_2) * int64(f8)
+	f1f9_76 := int64(f1_2) * int64(f9_38)
+	f2f2 := int64(f2) * int64(f2)
+	f2f3_2 := int64(f2_2) * int64(f3)
+	f2f4_2 := int64(f2_2) * int64(f4)
+	f2f5_2 := int64(f2_2) * int64(f5)
+	f2f6_2 := int64(f2_2) * int64(f6)
+	f2f7_2 := int64(f2_2) * int64(f7)
+	f2f8_38 := int64(f2_2) * int64(f8_19)
+	f2f9_38 := int64(f2) * int64(f9_38)
+	f3f3_2 := int64(f3_2) * int64(f3)
+	f3f4_2 := int64(f3_2) * int64(f4)
+	f3f5_4 := int64(f3_2) * int64(f5_2)
+	f3f6_2 := int64(f3_2) * int64(f6)
+	f3f7_76 := int64(f3_2) * int64(f7_38)
+	f3f8_38 := int64(f3_2) * int64(f8_19)
+	f3f9_76 := int64(f3_2) * int64(f9_38)
+	f4f4 := int64(f4) * int64(f4)
+	f4f5_2 := int64(f4_2) * int64(f5)
+	f4f6_38 := int64(f4_2) * int64(f6_19)
+	f4f7_38 := int64(f4) * int64(f7_38)
+	f4f8_38 := int64(f4_2) * int64(f8_19)
+	f4f9_38 := int64(f4) * int64(f9_38)
+	f5f5_38 := int64(f5) * int64(f5_38)
+	f5f6_38 := int64(f5_2) * int64(f6_19)
+	f5f7_76 := int64(f5_2) * int64(f7_38)
+	f5f8_38 := int64(f5_2) * int64(f8_19)
+	f5f9_76 := int64(f5_2) * int64(f9_38)
+	f6f6_19 := int64(f6) * int64(f6_19)
+	f6f7_38 := int64(f6) * int64(f7_38)
+	f6f8_38 := int64(f6_2) * int64(f8_19)
+	f6f9_38 := int64(f6) * int64(f9_38)
+	f7f7_38 := int64(f7) * int64(f7_38)
+	f7f8_38 := int64(f7_2) * int64(f8_19)
+	f7f9_76 := int64(f7_2) * int64(f9_38)
+	f8f8_19 := int64(f8) * int64(f8_19)
+	f8f9_38 := int64(f8) * int64(f9_38)
+	f9f9_38 := int64(f9) * int64(f9_38)
+	h0 := f0f0 + f1f9_76 + f2f8_38 + f3f7_76 + f4f6_38 + f5f5_38
+	h1 := f0f1_2 + f2f9_38 + f3f8_38 + f4f7_38 + f5f6_38
+	h2 := f0f2_2 + f1f1_2 + f3f9_76 + f4f8_38 + f5f7_76 + f6f6_19
+	h3 := f0f3_2 + f1f2_2 + f4f9_38 + f5f8_38 + f6f7_38
+	h4 := f0f4_2 + f1f3_4 + f2f2 + f5f9_76 + f6f8_38 + f7f7_38
+	h5 := f0f5_2 + f1f4_2 + f2f3_2 + f6f9_38 + f7f8_38
+	h6 := f0f6_2 + f1f5_4 + f2f4_2 + f3f3_2 + f7f9_76 + f8f8_19
+	h7 := f0f7_2 + f1f6_2 + f2f5_2 + f3f4_2 + f8f9_38
+	h8 := f0f8_2 + f1f7_4 + f2f6_2 + f3f5_4 + f4f4 + f9f9_38
+	h9 := f0f9_2 + f1f8_2 + f2f7_2 + f3f6_2 + f4f5_2
+	var carry [10]int64
+
+	carry[0] = (h0 + (1 << 25)) >> 26
+	h1 += carry[0]
+	h0 -= carry[0] << 26
+	carry[4] = (h4 + (1 << 25)) >> 26
+	h5 += carry[4]
+	h4 -= carry[4] << 26
+
+	carry[1] = (h1 + (1 << 24)) >> 25
+	h2 += carry[1]
+	h1 -= carry[1] << 25
+	carry[5] = (h5 + (1 << 24)) >> 25
+	h6 += carry[5]
+	h5 -= carry[5] << 25
+
+	carry[2] = (h2 + (1 << 25)) >> 26
+	h3 += carry[2]
+	h2 -= carry[2] << 26
+	carry[6] = (h6 + (1 << 25)) >> 26
+	h7 += carry[6]
+	h6 -= carry[6] << 26
+
+	carry[3] = (h3 + (1 << 24)) >> 25
+	h4 += carry[3]
+	h3 -= carry[3] << 25
+	carry[7] = (h7 + (1 << 24)) >> 25
+	h8 += carry[7]
+	h7 -= carry[7] << 25
+
+	carry[4] = (h4 + (1 << 25)) >> 26
+	h5 += carry[4]
+	h4 -= carry[4] << 26
+	carry[8] = (h8 + (1 << 25)) >> 26
+	h9 += carry[8]
+	h8 -= carry[8] << 26
+
+	carry[9] = (h9 + (1 << 24)) >> 25
+	h0 += carry[9] * 19
+	h9 -= carry[9] << 25
+
+	carry[0] = (h0 + (1 << 25)) >> 26
+	h1 += carry[0]
+	h0 -= carry[0] << 26
+
+	h[0] = int32(h0)
+	h[1] = int32(h1)
+	h[2] = int32(h2)
+	h[3] = int32(h3)
+	h[4] = int32(h4)
+	h[5] = int32(h5)
+	h[6] = int32(h6)
+	h[7] = int32(h7)
+	h[8] = int32(h8)
+	h[9] = int32(h9)
+}
+
+// feMul121666 calculates h = f * 121666. Can overlap h with f.
+//
+// Preconditions:
+//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+//
+// Postconditions:
+//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+func feMul121666(h, f *fieldElement) {
+	h0 := int64(f[0]) * 121666
+	h1 := int64(f[1]) * 121666
+	h2 := int64(f[2]) * 121666
+	h3 := int64(f[3]) * 121666
+	h4 := int64(f[4]) * 121666
+	h5 := int64(f[5]) * 121666
+	h6 := int64(f[6]) * 121666
+	h7 := int64(f[7]) * 121666
+	h8 := int64(f[8]) * 121666
+	h9 := int64(f[9]) * 121666
+	var carry [10]int64
+
+	carry[9] = (h9 + (1 << 24)) >> 25
+	h0 += carry[9] * 19
+	h9 -= carry[9] << 25
+	carry[1] = (h1 + (1 << 24)) >> 25
+	h2 += carry[1]
+	h1 -= carry[1] << 25
+	carry[3] = (h3 + (1 << 24)) >> 25
+	h4 += carry[3]
+	h3 -= carry[3] << 25
+	carry[5] = (h5 + (1 << 24)) >> 25
+	h6 += carry[5]
+	h5 -= carry[5] << 25
+	carry[7] = (h7 + (1 << 24)) >> 25
+	h8 += carry[7]
+	h7 -= carry[7] << 25
+
+	carry[0] = (h0 + (1 << 25)) >> 26
+	h1 += carry[0]
+	h0 -= carry[0] << 26
+	carry[2] = (h2 + (1 << 25)) >> 26
+	h3 += carry[2]
+	h2 -= carry[2] << 26
+	carry[4] = (h4 + (1 << 25)) >> 26
+	h5 += carry[4]
+	h4 -= carry[4] << 26
+	carry[6] = (h6 + (1 << 25)) >> 26
+	h7 += carry[6]
+	h6 -= carry[6] << 26
+	carry[8] = (h8 + (1 << 25)) >> 26
+	h9 += carry[8]
+	h8 -= carry[8] << 26
+
+	h[0] = int32(h0)
+	h[1] = int32(h1)
+	h[2] = int32(h2)
+	h[3] = int32(h3)
+	h[4] = int32(h4)
+	h[5] = int32(h5)
+	h[6] = int32(h6)
+	h[7] = int32(h7)
+	h[8] = int32(h8)
+	h[9] = int32(h9)
+}
+
+// feInvert sets out = z^-1.
+func feInvert(out, z *fieldElement) {
+	var t0, t1, t2, t3 fieldElement
+	var i int
+
+	feSquare(&t0, z)
+	for i = 1; i < 1; i++ {
+		feSquare(&t0, &t0)
+	}
+	feSquare(&t1, &t0)
+	for i = 1; i < 2; i++ {
+		feSquare(&t1, &t1)
+	}
+	feMul(&t1, z, &t1)
+	feMul(&t0, &t0, &t1)
+	feSquare(&t2, &t0)
+	for i = 1; i < 1; i++ {
+		feSquare(&t2, &t2)
+	}
+	feMul(&t1, &t1, &t2)
+	feSquare(&t2, &t1)
+	for i = 1; i < 5; i++ {
+		feSquare(&t2, &t2)
+	}
+	feMul(&t1, &t2, &t1)
+	feSquare(&t2, &t1)
+	for i = 1; i < 10; i++ {
+		feSquare(&t2, &t2)
+	}
+	feMul(&t2, &t2, &t1)
+	feSquare(&t3, &t2)
+	for i = 1; i < 20; i++ {
+		feSquare(&t3, &t3)
+	}
+	feMul(&t2, &t3, &t2)
+	feSquare(&t2, &t2)
+	for i = 1; i < 10; i++ {
+		feSquare(&t2, &t2)
+	}
+	feMul(&t1, &t2, &t1)
+	feSquare(&t2, &t1)
+	for i = 1; i < 50; i++ {
+		feSquare(&t2, &t2)
+	}
+	feMul(&t2, &t2, &t1)
+	feSquare(&t3, &t2)
+	for i = 1; i < 100; i++ {
+		feSquare(&t3, &t3)
+	}
+	feMul(&t2, &t3, &t2)
+	feSquare(&t2, &t2)
+	for i = 1; i < 50; i++ {
+		feSquare(&t2, &t2)
+	}
+	feMul(&t1, &t2, &t1)
+	feSquare(&t1, &t1)
+	for i = 1; i < 5; i++ {
+		feSquare(&t1, &t1)
+	}
+	feMul(out, &t1, &t0)
+}
+
+func scalarMultGeneric(out, in, base *[32]byte) {
+	var e [32]byte
+
+	copy(e[:], in[:])
+	e[0] &= 248
+	e[31] &= 127
+	e[31] |= 64
+
+	var x1, x2, z2, x3, z3, tmp0, tmp1 fieldElement
+	feFromBytes(&x1, base)
+	feOne(&x2)
+	feCopy(&x3, &x1)
+	feOne(&z3)
+
+	swap := int32(0)
+	for pos := 254; pos >= 0; pos-- {
+		b := e[pos/8] >> uint(pos&7)
+		b &= 1
+		swap ^= int32(b)
+		feCSwap(&x2, &x3, swap)
+		feCSwap(&z2, &z3, swap)
+		swap = int32(b)
+
+		feSub(&tmp0, &x3, &z3)
+		feSub(&tmp1, &x2, &z2)
+		feAdd(&x2, &x2, &z2)
+		feAdd(&z2, &x3, &z3)
+		feMul(&z3, &tmp0, &x2)
+		feMul(&z2, &z2, &tmp1)
+		feSquare(&tmp0, &tmp1)
+		feSquare(&tmp1, &x2)
+		feAdd(&x3, &z3, &z2)
+		feSub(&z2, &z3, &z2)
+		feMul(&x2, &tmp1, &tmp0)
+		feSub(&tmp1, &tmp1, &tmp0)
+		feSquare(&z2, &z2)
+		feMul121666(&z3, &tmp1)
+		feSquare(&x3, &x3)
+		feAdd(&tmp0, &tmp0, &z3)
+		feMul(&z3, &x1, &z2)
+		feMul(&z2, &tmp1, &tmp0)
+	}
+
+	feCSwap(&x2, &x3, swap)
+	feCSwap(&z2, &z3, swap)
+
+	feInvert(&z2, &z2)
+	feMul(&x2, &x2, &z2)
+	feToBytes(out, &x2)
+}

src/vendor/golang.org/x/crypto/curve25519/const_amd64.h → src/vendor/golang.org/x/crypto/curve25519/curve25519_noasm.go

-// Copyright 2012 The Go Authors. All rights reserved.
+// Copyright 2019 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.
 
-// This code was translated into a form compatible with 6a from the public
-// domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
+// +build !amd64 gccgo appengine purego
 
-#define REDMASK51     0x0007FFFFFFFFFFFF
+package curve25519
+
+func scalarMult(out, in, base *[32]byte) {
+	scalarMultGeneric(out, in, base)
+}

src/vendor/golang.org/x/crypto/curve25519/doc.go

-// 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 curve25519 provides an implementation of scalar multiplication on
-// the elliptic curve known as curve25519. See https://cr.yp.to/ecdh.html
-package curve25519 // import "golang.org/x/crypto/curve25519"
-
-// basePoint is the x coordinate of the generator of the curve.
-var basePoint = [32]byte{9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
-
-// ScalarMult sets dst to the product in*base where dst and base are the x
-// coordinates of group points and all values are in little-endian form.
-func ScalarMult(dst, in, base *[32]byte) {
-	scalarMult(dst, in, base)
-}
-
-// ScalarBaseMult sets dst to the product in*base where dst and base are the x
-// coordinates of group points, base is the standard generator and all values
-// are in little-endian form.
-func ScalarBaseMult(dst, in *[32]byte) {
-	ScalarMult(dst, in, &basePoint)
-}

src/vendor/golang.org/x/crypto/curve25519/freeze_amd64.s

-// 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.
-
-// This code was translated into a form compatible with 6a from the public
-// domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
-
-// +build amd64,!gccgo,!appengine
-
-#include "const_amd64.h"
-
-// func freeze(inout *[5]uint64)
-TEXT ·freeze(SB),7,$0-8
-	MOVQ inout+0(FP), DI
-
-	MOVQ 0(DI),SI
-	MOVQ 8(DI),DX
-	MOVQ 16(DI),CX
-	MOVQ 24(DI),R8
-	MOVQ 32(DI),R9
-	MOVQ $REDMASK51,AX
-	MOVQ AX,R10
-	SUBQ $18,R10
-	MOVQ $3,R11
-REDUCELOOP:
-	MOVQ SI,R12
-	SHRQ $51,R12
-	ANDQ AX,SI
-	ADDQ R12,DX
-	MOVQ DX,R12
-	SHRQ $51,R12
-	ANDQ AX,DX
-	ADDQ R12,CX
-	MOVQ CX,R12
-	SHRQ $51,R12
-	ANDQ AX,CX
-	ADDQ R12,R8
-	MOVQ R8,R12
-	SHRQ $51,R12
-	ANDQ AX,R8
-	ADDQ R12,R9
-	MOVQ R9,R12
-	SHRQ $51,R12
-	ANDQ AX,R9
-	IMUL3Q $19,R12,R12
-	ADDQ R12,SI
-	SUBQ $1,R11
-	JA REDUCELOOP
-	MOVQ $1,R12
-	CMPQ R10,SI
-	CMOVQLT R11,R12
-	CMPQ AX,DX
-	CMOVQNE R11,R12
-	CMPQ AX,CX
-	CMOVQNE R11,R12
-	CMPQ AX,R8
-	CMOVQNE R11,R12
-	CMPQ AX,R9
-	CMOVQNE R11,R12
-	NEGQ R12
-	ANDQ R12,AX
-	ANDQ R12,R10
-	SUBQ R10,SI
-	SUBQ AX,DX
-	SUBQ AX,CX
-	SUBQ AX,R8
-	SUBQ AX,R9
-	MOVQ SI,0(DI)
-	MOVQ DX,8(DI)
-	MOVQ CX,16(DI)
-	MOVQ R8,24(DI)
-	MOVQ R9,32(DI)
-	RET

src/vendor/golang.org/x/crypto/curve25519/mul_amd64.s

-// 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.
-
-// This code was translated into a form compatible with 6a from the public
-// domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
-
-// +build amd64,!gccgo,!appengine
-
-#include "const_amd64.h"
-
-// func mul(dest, a, b *[5]uint64)
-TEXT ·mul(SB),0,$16-24
-	MOVQ dest+0(FP), DI
-	MOVQ a+8(FP), SI
-	MOVQ b+16(FP), DX
-
-	MOVQ DX,CX
-	MOVQ 24(SI),DX
-	IMUL3Q $19,DX,AX
-	MOVQ AX,0(SP)
-	MULQ 16(CX)
-	MOVQ AX,R8
-	MOVQ DX,R9
-	MOVQ 32(SI),DX
-	IMUL3Q $19,DX,AX
-	MOVQ AX,8(SP)
-	MULQ 8(CX)
-	ADDQ AX,R8
-	ADCQ DX,R9
-	MOVQ 0(SI),AX
-	MULQ 0(CX)
-	ADDQ AX,R8
-	ADCQ DX,R9
-	MOVQ 0(SI),AX
-	MULQ 8(CX)
-	MOVQ AX,R10
-	MOVQ DX,R11
-	MOVQ 0(SI),AX
-	MULQ 16(CX)
-	MOVQ AX,R12
-	MOVQ DX,R13
-	MOVQ 0(SI),AX
-	MULQ 24(CX)
-	MOVQ AX,R14
-	MOVQ DX,R15
-	MOVQ 0(SI),AX
-	MULQ 32(CX)
-	MOVQ AX,BX
-	MOVQ DX,BP
-	MOVQ 8(SI),AX
-	MULQ 0(CX)
-	ADDQ AX,R10
-	ADCQ DX,R11
-	MOVQ 8(SI),AX
-	MULQ 8(CX)
-	ADDQ AX,R12
-	ADCQ DX,R13
-	MOVQ 8(SI),AX
-	MULQ 16(CX)
-	ADDQ AX,R14
-	ADCQ DX,R15
-	MOVQ 8(SI),AX
-	MULQ 24(CX)
-	ADDQ AX,BX
-	ADCQ DX,BP
-	MOVQ 8(SI),DX
-	IMUL3Q $19,DX,AX
-	MULQ 32(CX)
-	ADDQ AX,R8
-	ADCQ DX,R9
-	MOVQ 16(SI),AX
-	MULQ 0(CX)
-	ADDQ AX,R12
-	ADCQ DX,R13
-	MOVQ 16(SI),AX
-	MULQ 8(CX)
-	ADDQ AX,R14
-	ADCQ DX,R15
-	MOVQ 16(SI),AX
-	MULQ 16(CX)
-	ADDQ AX,BX
-	ADCQ DX,BP
-	MOVQ 16(SI),DX
-	IMUL3Q $19,DX,AX
-	MULQ 24(CX)
-	ADDQ AX,R8
-	ADCQ DX,R9
-	MOVQ 16(SI),DX
-	IMUL3Q $19,DX,AX
-	MULQ 32(CX)
-	ADDQ AX,R10
-	ADCQ DX,R11
-	MOVQ 24(SI),AX
-	MULQ 0(CX)
-	ADDQ AX,R14
-	ADCQ DX,R15
-	MOVQ 24(SI),AX
-	MULQ 8(CX)
-	ADDQ AX,BX
-	ADCQ DX,BP
-	MOVQ 0(SP),AX
-	MULQ 24(CX)
-	ADDQ AX,R10
-	ADCQ DX,R11
-	MOVQ 0(SP),AX
-	MULQ 32(CX)
-	ADDQ AX,R12
-	ADCQ DX,R13
-	MOVQ 32(SI),AX
-	MULQ 0(CX)
-	ADDQ AX,BX
-	ADCQ DX,BP
-	MOVQ 8(SP),AX
-	MULQ 16(CX)
-	ADDQ AX,R10
-	ADCQ DX,R11
-	MOVQ 8(SP),AX
-	MULQ 24(CX)
-	ADDQ AX,R12
-	ADCQ DX,R13
-	MOVQ 8(SP),AX
-	MULQ 32(CX)
-	ADDQ AX,R14
-	ADCQ DX,R15
-	MOVQ $REDMASK51,SI
-	SHLQ $13,R8,R9
-	ANDQ SI,R8
-	SHLQ $13,R10,R11
-	ANDQ SI,R10
-	ADDQ R9,R10
-	SHLQ $13,R12,R13
-	ANDQ SI,R12
-	ADDQ R11,R12
-	SHLQ $13,R14,R15
-	ANDQ SI,R14
-	ADDQ R13,R14
-	SHLQ $13,BX,BP
-	ANDQ SI,BX
-	ADDQ R15,BX
-	IMUL3Q $19,BP,DX
-	ADDQ DX,R8
-	MOVQ R8,DX
-	SHRQ $51,DX
-	ADDQ R10,DX
-	MOVQ DX,CX
-	SHRQ $51,DX
-	ANDQ SI,R8
-	ADDQ R12,DX
-	MOVQ DX,R9
-	SHRQ $51,DX
-	ANDQ SI,CX
-	ADDQ R14,DX
-	MOVQ DX,AX
-	SHRQ $51,DX
-	ANDQ SI,R9
-	ADDQ BX,DX
-	MOVQ DX,R10
-	SHRQ $51,DX
-	ANDQ SI,AX
-	IMUL3Q $19,DX,DX
-	ADDQ DX,R8
-	ANDQ SI,R10
-	MOVQ R8,0(DI)
-	MOVQ CX,8(DI)
-	MOVQ R9,16(DI)
-	MOVQ AX,24(DI)
-	MOVQ R10,32(DI)
-	RET

src/vendor/golang.org/x/crypto/curve25519/square_amd64.s

-// 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.
-
-// This code was translated into a form compatible with 6a from the public
-// domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
-
-// +build amd64,!gccgo,!appengine
-
-#include "const_amd64.h"
-
-// func square(out, in *[5]uint64)
-TEXT ·square(SB),7,$0-16
-	MOVQ out+0(FP), DI
-	MOVQ in+8(FP), SI
-
-	MOVQ 0(SI),AX
-	MULQ 0(SI)
-	MOVQ AX,CX
-	MOVQ DX,R8
-	MOVQ 0(SI),AX
-	SHLQ $1,AX
-	MULQ 8(SI)
-	MOVQ AX,R9
-	MOVQ DX,R10
-	MOVQ 0(SI),AX
-	SHLQ $1,AX
-	MULQ 16(SI)
-	MOVQ AX,R11
-	MOVQ DX,R12
-	MOVQ 0(SI),AX
-	SHLQ $1,AX
-	MULQ 24(SI)
-	MOVQ AX,R13
-	MOVQ DX,R14
-	MOVQ 0(SI),AX
-	SHLQ $1,AX
-	MULQ 32(SI)
-	MOVQ AX,R15
-	MOVQ DX,BX
-	MOVQ 8(SI),AX
-	MULQ 8(SI)
-	ADDQ AX,R11
-	ADCQ DX,R12
-	MOVQ 8(SI),AX
-	SHLQ $1,AX
-	MULQ 16(SI)
-	ADDQ AX,R13
-	ADCQ DX,R14
-	MOVQ 8(SI),AX
-	SHLQ $1,AX
-	MULQ 24(SI)
-	ADDQ AX,R15
-	ADCQ DX,BX
-	MOVQ 8(SI),DX
-	IMUL3Q $38,DX,AX
-	MULQ 32(SI)
-	ADDQ AX,CX
-	ADCQ DX,R8
-	MOVQ 16(SI),AX
-	MULQ 16(SI)
-	ADDQ AX,R15
-	ADCQ DX,BX
-	MOVQ 16(SI),DX
-	IMUL3Q $38,DX,AX
-	MULQ 24(SI)
-	ADDQ AX,CX
-	ADCQ DX,R8
-	MOVQ 16(SI),DX
-	IMUL3Q $38,DX,AX
-	MULQ 32(SI)
-	ADDQ AX,R9
-	ADCQ DX,R10
-	MOVQ 24(SI),DX
-	IMUL3Q $19,DX,AX
-	MULQ 24(SI)
-	ADDQ AX,R9
-	ADCQ DX,R10
-	MOVQ 24(SI),DX
-	IMUL3Q $38,DX,AX
-	MULQ 32(SI)
-	ADDQ AX,R11
-	ADCQ DX,R12
-	MOVQ 32(SI),DX
-	IMUL3Q $19,DX,AX
-	MULQ 32(SI)
-	ADDQ AX,R13
-	ADCQ DX,R14
-	MOVQ $REDMASK51,SI
-	SHLQ $13,CX,R8
-	ANDQ SI,CX
-	SHLQ $13,R9,R10
-	ANDQ SI,R9
-	ADDQ R8,R9
-	SHLQ $13,R11,R12
-	ANDQ SI,R11
-	ADDQ R10,R11
-	SHLQ $13,R13,R14
-	ANDQ SI,R13
-	ADDQ R12,R13
-	SHLQ $13,R15,BX
-	ANDQ SI,R15
-	ADDQ R14,R15
-	IMUL3Q $19,BX,DX
-	ADDQ DX,CX
-	MOVQ CX,DX
-	SHRQ $51,DX
-	ADDQ R9,DX
-	ANDQ SI,CX
-	MOVQ DX,R8
-	SHRQ $51,DX
-	ADDQ R11,DX
-	ANDQ SI,R8
-	MOVQ DX,R9
-	SHRQ $51,DX
-	ADDQ R13,DX
-	ANDQ SI,R9
-	MOVQ DX,AX
-	SHRQ $51,DX
-	ADDQ R15,DX
-	ANDQ SI,AX
-	MOVQ DX,R10
-	SHRQ $51,DX
-	IMUL3Q $19,DX,DX
-	ADDQ DX,CX
-	ANDQ SI,R10
-	MOVQ CX,0(DI)
-	MOVQ R8,8(DI)
-	MOVQ R9,16(DI)
-	MOVQ AX,24(DI)
-	MOVQ R10,32(DI)
-	RET

src/vendor/golang.org/x/crypto/internal/chacha20/asm_ppc64le.s

-// Copyright 2019 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.
-
-// Based on CRYPTOGAMS code with the following comment:
-// # ====================================================================
-// # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
-// # project. The module is, however, dual licensed under OpenSSL and
-// # CRYPTOGAMS licenses depending on where you obtain it. For further
-// # details see http://www.openssl.org/~appro/cryptogams/.
-// # ====================================================================
-
-// Original code can be found at the link below:
-// https://github.com/dot-asm/cryptogams/commit/a60f5b50ed908e91e5c39ca79126a4a876d5d8ff
-
-// There are some differences between CRYPTOGAMS code and this one. The round
-// loop for "_int" isn't the same as the original. Some adjustments were
-// necessary because there are less vector registers available.  For example, some
-// X variables (r12, r13, r14, and r15) share the same register used by the
-// counter. The original code uses ctr to name the counter. Here we use CNT
-// because golang uses CTR as the counter register name.
-
-// +build ppc64le,!gccgo,!appengine
-
-#include "textflag.h"
-
-#define OUT  R3
-#define INP  R4
-#define LEN  R5
-#define KEY  R6
-#define CNT  R7
-
-#define TEMP R8
-
-#define X0   R11
-#define X1   R12
-#define X2   R14
-#define X3   R15
-#define X4   R16
-#define X5   R17
-#define X6   R18
-#define X7   R19
-#define X8   R20
-#define X9   R21
-#define X10  R22
-#define X11  R23
-#define X12  R24
-#define X13  R25
-#define X14  R26
-#define X15  R27
-
-#define CON0 X0
-#define CON1 X1
-#define CON2 X2
-#define CON3 X3
-
-#define KEY0 X4
-#define KEY1 X5
-#define KEY2 X6
-#define KEY3 X7
-#define KEY4 X8
-#define KEY5 X9
-#define KEY6 X10
-#define KEY7 X11
-
-#define CNT0 X12
-#define CNT1 X13
-#define CNT2 X14
-#define CNT3 X15
-
-#define TMP0 R9
-#define TMP1 R10
-#define TMP2 R28
-#define TMP3 R29
-
-#define CONSTS  R8
-
-#define A0      V0
-#define B0      V1
-#define C0      V2
-#define D0      V3
-#define A1      V4
-#define B1      V5
-#define C1      V6
-#define D1      V7
-#define A2      V8
-#define B2      V9
-#define C2      V10
-#define D2      V11
-#define T0      V12
-#define T1      V13
-#define T2      V14
-
-#define K0      V15
-#define K1      V16
-#define K2      V17
-#define K3      V18
-#define K4      V19
-#define K5      V20
-
-#define FOUR    V21
-#define SIXTEEN V22
-#define TWENTY4 V23
-#define TWENTY  V24
-#define TWELVE  V25
-#define TWENTY5 V26
-#define SEVEN   V27
-
-#define INPPERM V28
-#define OUTPERM V29
-#define OUTMASK V30
-
-#define DD0     V31
-#define DD1     SEVEN
-#define DD2     T0
-#define DD3     T1
-#define DD4     T2
-
-DATA  ·consts+0x00(SB)/8, $0x3320646e61707865
-DATA  ·consts+0x08(SB)/8, $0x6b20657479622d32
-DATA  ·consts+0x10(SB)/8, $0x0000000000000001
-DATA  ·consts+0x18(SB)/8, $0x0000000000000000
-DATA  ·consts+0x20(SB)/8, $0x0000000000000004
-DATA  ·consts+0x28(SB)/8, $0x0000000000000000
-DATA  ·consts+0x30(SB)/8, $0x0a0b08090e0f0c0d
-DATA  ·consts+0x38(SB)/8, $0x0203000106070405
-DATA  ·consts+0x40(SB)/8, $0x090a0b080d0e0f0c
-DATA  ·consts+0x48(SB)/8, $0x0102030005060704
-GLOBL ·consts(SB), RODATA, $80
-
-//func chaCha20_ctr32_vmx(out, inp *byte, len int, key *[32]byte, counter *[16]byte)
-TEXT ·chaCha20_ctr32_vmx(SB),NOSPLIT|NOFRAME,$0
-	// Load the arguments inside the registers
-	MOVD out+0(FP), OUT
-	MOVD inp+8(FP), INP
-	MOVD len+16(FP), LEN
-	MOVD key+24(FP), KEY
-	MOVD counter+32(FP), CNT
-
-	MOVD $·consts(SB), CONSTS // point to consts addr
-
-	MOVD $16, X0
-	MOVD $32, X1
-	MOVD $48, X2
-	MOVD $64, X3
-	MOVD $31, X4
-	MOVD $15, X5
-
-	// Load key
-	LVX  (KEY)(R0), K1
-	LVSR (KEY)(R0), T0
-	LVX  (KEY)(X0), K2
-	LVX  (KEY)(X4), DD0
-
-	// Load counter
-	LVX  (CNT)(R0), K3
-	LVSR (CNT)(R0), T1
-	LVX  (CNT)(X5), DD1
-
-	// Load constants
-	LVX (CONSTS)(R0), K0
-	LVX (CONSTS)(X0), K5
-	LVX (CONSTS)(X1), FOUR
-	LVX (CONSTS)(X2), SIXTEEN
-	LVX (CONSTS)(X3), TWENTY4
-
-	// Align key and counter
-	VPERM K2,  K1, T0, K1
-	VPERM DD0, K2, T0, K2
-	VPERM DD1, K3, T1, K3
-
-	// Load counter to GPR
-	MOVWZ 0(CNT), CNT0
-	MOVWZ 4(CNT), CNT1
-	MOVWZ 8(CNT), CNT2
-	MOVWZ 12(CNT), CNT3
-
-	// Adjust vectors for the initial state
-	VADDUWM K3, K5, K3
-	VADDUWM K3, K5, K4
-	VADDUWM K4, K5, K5
-
-	// Synthesized constants
-	VSPLTISW $-12, TWENTY
-	VSPLTISW $12, TWELVE
-	VSPLTISW $-7, TWENTY5
-
-	VXOR T0, T0, T0
-	VSPLTISW $-1, OUTMASK
-	LVSR (INP)(R0), INPPERM
-	LVSL (OUT)(R0), OUTPERM
-	VPERM OUTMASK, T0, OUTPERM, OUTMASK
-
-loop_outer_vmx:
-	// Load constant
-	MOVD $0x61707865, CON0
-	MOVD $0x3320646e, CON1
-	MOVD $0x79622d32, CON2
-	MOVD $0x6b206574, CON3
-
-	VOR K0, K0, A0
-	VOR K0, K0, A1
-	VOR K0, K0, A2
-	VOR K1, K1, B0
-
-	MOVD $10, TEMP
-
-	// Load key to GPR
-	MOVWZ 0(KEY), X4
-	MOVWZ 4(KEY), X5
-	MOVWZ 8(KEY), X6
-	MOVWZ 12(KEY), X7
-	VOR K1, K1, B1
-	VOR K1, K1, B2
-	MOVWZ 16(KEY), X8
-	MOVWZ  0(CNT), X12
-	MOVWZ 20(KEY), X9
-	MOVWZ 4(CNT), X13
-	VOR K2, K2, C0
-	VOR K2, K2, C1
-	MOVWZ 24(KEY), X10
-	MOVWZ 8(CNT), X14
-	VOR K2, K2, C2
-	VOR K3, K3, D0
-	MOVWZ 28(KEY), X11
-	MOVWZ 12(CNT), X15
-	VOR K4, K4, D1
-	VOR K5, K5, D2
-
-	MOVD X4, TMP0
-	MOVD X5, TMP1
-	MOVD X6, TMP2
-	MOVD X7, TMP3
-	VSPLTISW $7, SEVEN
-
-	MOVD TEMP, CTR
-
-loop_vmx:
-	// CRYPTOGAMS uses a macro to create a loop using perl. This isn't possible
-	// using assembly macros.  Therefore, the macro expansion result was used
-	// in order to maintain the algorithm efficiency.
-	// This loop generates three keystream blocks using VMX instructions and,
-	// in parallel, one keystream block using scalar instructions.
-	ADD X4, X0, X0
-	ADD X5, X1, X1
-	VADDUWM A0, B0, A0
-	VADDUWM A1, B1, A1
-	ADD X6, X2, X2
-	ADD X7, X3, X3
-	VADDUWM A2, B2, A2
-	VXOR D0, A0, D0
-	XOR X0, X12, X12
-	XOR X1, X13, X13
-	VXOR D1, A1, D1
-	VXOR D2, A2, D2
-	XOR X2, X14, X14
-	XOR X3, X15, X15
-	VPERM D0, D0, SIXTEEN, D0
-	VPERM D1, D1, SIXTEEN, D1
-	ROTLW $16, X12, X12
-	ROTLW $16, X13, X13
-	VPERM D2, D2, SIXTEEN, D2
-	VADDUWM C0, D0, C0
-	ROTLW $16, X14, X14
-	ROTLW $16, X15, X15
-	VADDUWM C1, D1, C1
-	VADDUWM C2, D2, C2
-	ADD X12, X8, X8
-	ADD X13, X9, X9
-	VXOR B0, C0, T0
-	VXOR B1, C1, T1
-	ADD X14, X10, X10
-	ADD X15, X11, X11
-	VXOR B2, C2, T2
-	VRLW T0, TWELVE, B0
-	XOR X8, X4, X4
-	XOR X9, X5, X5
-	VRLW T1, TWELVE, B1
-	VRLW T2, TWELVE, B2
-	XOR X10, X6, X6
-	XOR X11, X7, X7
-	VADDUWM A0, B0, A0
-	VADDUWM A1, B1, A1
-	ROTLW $12, X4, X4
-	ROTLW $12, X5, X5
-	VADDUWM A2, B2, A2
-	VXOR D0, A0, D0
-	ROTLW $12, X6, X6
-	ROTLW $12, X7, X7
-	VXOR D1, A1, D1
-	VXOR D2, A2, D2
-	ADD X4, X0, X0
-	ADD X5, X1, X1
-	VPERM D0, D0, TWENTY4, D0
-	VPERM D1, D1, TWENTY4, D1
-	ADD X6, X2, X2
-	ADD X7, X3, X3
-	VPERM D2, D2, TWENTY4, D2
-	VADDUWM C0, D0, C0
-	XOR X0, X12, X12
-	XOR X1, X13, X13
-	VADDUWM C1, D1, C1
-	VADDUWM C2, D2, C2
-	XOR X2, X14, X14
-	XOR X3, X15, X15
-	VXOR B0, C0, T0
-	VXOR B1, C1, T1
-	ROTLW $8, X12, X12
-	ROTLW $8, X13, X13
-	VXOR B2, C2, T2
-	VRLW T0, SEVEN, B0
-	ROTLW $8, X14, X14
-	ROTLW $8, X15, X15
-	VRLW T1, SEVEN, B1
-	VRLW T2, SEVEN, B2
-	ADD X12, X8, X8
-	ADD X13, X9, X9
-	VSLDOI $8, C0, C0, C0
-	VSLDOI $8, C1, C1, C1
-	ADD X14, X10, X10
-	ADD X15, X11, X11
-	VSLDOI $8, C2, C2, C2
-	VSLDOI $12, B0, B0, B0
-	XOR X8, X4, X4
-	XOR X9, X5, X5
-	VSLDOI $12, B1, B1, B1
-	VSLDOI $12, B2, B2, B2
-	XOR X10, X6, X6
-	XOR X11, X7, X7
-	VSLDOI $4, D0, D0, D0
-	VSLDOI $4, D1, D1, D1
-	ROTLW $7, X4, X4
-	ROTLW $7, X5, X5
-	VSLDOI $4, D2, D2, D2
-	VADDUWM A0, B0, A0
-	ROTLW $7, X6, X6
-	ROTLW $7, X7, X7
-	VADDUWM A1, B1, A1
-	VADDUWM A2, B2, A2
-	ADD X5, X0, X0
-	ADD X6, X1, X1
-	VXOR D0, A0, D0
-	VXOR D1, A1, D1
-	ADD X7, X2, X2
-	ADD X4, X3, X3
-	VXOR D2, A2, D2
-	VPERM D0, D0, SIXTEEN, D0
-	XOR X0, X15, X15
-	XOR X1, X12, X12
-	VPERM D1, D1, SIXTEEN, D1
-	VPERM D2, D2, SIXTEEN, D2
-	XOR X2, X13, X13
-	XOR X3, X14, X14
-	VADDUWM C0, D0, C0
-	VADDUWM C1, D1, C1
-	ROTLW $16, X15, X15
-	ROTLW $16, X12, X12
-	VADDUWM C2, D2, C2
-	VXOR B0, C0, T0
-	ROTLW $16, X13, X13
-	ROTLW $16, X14, X14
-	VXOR B1, C1, T1
-	VXOR B2, C2, T2
-	ADD X15, X10, X10
-	ADD X12, X11, X11
-	VRLW T0, TWELVE, B0
-	VRLW T1, TWELVE, B1
-	ADD X13, X8, X8
-	ADD X14, X9, X9
-	VRLW T2, TWELVE, B2
-	VADDUWM A0, B0, A0
-	XOR X10, X5, X5
-	XOR X11, X6, X6
-	VADDUWM A1, B1, A1
-	VADDUWM A2, B2, A2
-	XOR X8, X7, X7
-	XOR X9, X4, X4
-	VXOR D0, A0, D0
-	VXOR D1, A1, D1
-	ROTLW $12, X5, X5
-	ROTLW $12, X6, X6
-	VXOR D2, A2, D2
-	VPERM D0, D0, TWENTY4, D0
-	ROTLW $12, X7, X7
-	ROTLW $12, X4, X4
-	VPERM D1, D1, TWENTY4, D1
-	VPERM D2, D2, TWENTY4, D2
-	ADD X5, X0, X0
-	ADD X6, X1, X1
-	VADDUWM C0, D0, C0
-	VADDUWM C1, D1, C1
-	ADD X7, X2, X2
-	ADD X4, X3, X3
-	VADDUWM C2, D2, C2
-	VXOR B0, C0, T0
-	XOR X0, X15, X15
-	XOR X1, X12, X12
-	VXOR B1, C1, T1
-	VXOR B2, C2, T2
-	XOR X2, X13, X13
-	XOR X3, X14, X14
-	VRLW T0, SEVEN, B0
-	VRLW T1, SEVEN, B1
-	ROTLW $8, X15, X15
-	ROTLW $8, X12, X12
-	VRLW T2, SEVEN, B2
-	VSLDOI $8, C0, C0, C0
-	ROTLW $8, X13, X13
-	ROTLW $8, X14, X14
-	VSLDOI $8, C1, C1, C1
-	VSLDOI $8, C2, C2, C2
-	ADD X15, X10, X10
-	ADD X12, X11, X11
-	VSLDOI $4, B0, B0, B0
-	VSLDOI $4, B1, B1, B1
-	ADD X13, X8, X8
-	ADD X14, X9, X9
-	VSLDOI $4, B2, B2, B2
-	VSLDOI $12, D0, D0, D0
-	XOR X10, X5, X5
-	XOR X11, X6, X6
-	VSLDOI $12, D1, D1, D1
-	VSLDOI $12, D2, D2, D2
-	XOR X8, X7, X7
-	XOR X9, X4, X4
-	ROTLW $7, X5, X5
-	ROTLW $7, X6, X6
-	ROTLW $7, X7, X7
-	ROTLW $7, X4, X4
-	BC 0x10, 0, loop_vmx
-
-	SUB $256, LEN, LEN
-
-	// Accumulate key block
-	ADD $0x61707865, X0, X0
-	ADD $0x3320646e, X1, X1
-	ADD $0x79622d32, X2, X2
-	ADD $0x6b206574, X3, X3
-	ADD TMP0, X4, X4
-	ADD TMP1, X5, X5
-	ADD TMP2, X6, X6
-	ADD TMP3, X7, X7
-	MOVWZ 16(KEY), TMP0
-	MOVWZ 20(KEY), TMP1
-	MOVWZ 24(KEY), TMP2
-	MOVWZ 28(KEY), TMP3
-	ADD TMP0, X8, X8
-	ADD TMP1, X9, X9
-	ADD TMP2, X10, X10
-	ADD TMP3, X11, X11
-
-	MOVWZ 12(CNT), TMP0
-	MOVWZ 8(CNT), TMP1
-	MOVWZ 4(CNT), TMP2
-	MOVWZ 0(CNT), TEMP
-	ADD TMP0, X15, X15
-	ADD TMP1, X14, X14
-	ADD TMP2, X13, X13
-	ADD TEMP, X12, X12
-
-	// Accumulate key block
-	VADDUWM A0, K0, A0
-	VADDUWM A1, K0, A1
-	VADDUWM A2, K0, A2
-	VADDUWM B0, K1, B0
-	VADDUWM B1, K1, B1
-	VADDUWM B2, K1, B2
-	VADDUWM C0, K2, C0
-	VADDUWM C1, K2, C1
-	VADDUWM C2, K2, C2
-	VADDUWM D0, K3, D0
-	VADDUWM D1, K4, D1
-	VADDUWM D2, K5, D2
-
-	// Increment counter
-	ADD $4, TEMP, TEMP
-	MOVW TEMP, 0(CNT)
-
-	VADDUWM K3, FOUR, K3
-	VADDUWM K4, FOUR, K4
-	VADDUWM K5, FOUR, K5
-
-	// XOR the input slice (INP) with the keystream, which is stored in GPRs (X0-X3).
-
-	// Load input (aligned or not)
-	MOVWZ 0(INP), TMP0
-	MOVWZ 4(INP), TMP1
-	MOVWZ 8(INP), TMP2
-	MOVWZ 12(INP), TMP3
-
-	// XOR with input
-	XOR TMP0, X0, X0
-	XOR TMP1, X1, X1
-	XOR TMP2, X2, X2
-	XOR TMP3, X3, X3
-	MOVWZ 16(INP), TMP0
-	MOVWZ 20(INP), TMP1
-	MOVWZ 24(INP), TMP2
-	MOVWZ 28(INP), TMP3
-	XOR TMP0, X4, X4
-	XOR TMP1, X5, X5
-	XOR TMP2, X6, X6
-	XOR TMP3, X7, X7
-	MOVWZ 32(INP), TMP0
-	MOVWZ 36(INP), TMP1
-	MOVWZ 40(INP), TMP2
-	MOVWZ 44(INP), TMP3
-	XOR TMP0, X8, X8
-	XOR TMP1, X9, X9
-	XOR TMP2, X10, X10
-	XOR TMP3, X11, X11
-	MOVWZ 48(INP), TMP0
-	MOVWZ 52(INP), TMP1
-	MOVWZ 56(INP), TMP2
-	MOVWZ 60(INP), TMP3
-	XOR TMP0, X12, X12
-	XOR TMP1, X13, X13
-	XOR TMP2, X14, X14
-	XOR TMP3, X15, X15
-
-	// Store output (aligned or not)
-	MOVW X0, 0(OUT)
-	MOVW X1, 4(OUT)
-	MOVW X2, 8(OUT)
-	MOVW X3, 12(OUT)
-
-	ADD $64, INP, INP // INP points to the end of the slice for the alignment code below
-
-	MOVW X4, 16(OUT)
-	MOVD $16, TMP0
-	MOVW X5, 20(OUT)
-	MOVD $32, TMP1
-	MOVW X6, 24(OUT)
-	MOVD $48, TMP2
-	MOVW X7, 28(OUT)
-	MOVD $64, TMP3
-	MOVW X8, 32(OUT)
-	MOVW X9, 36(OUT)
-	MOVW X10, 40(OUT)
-	MOVW X11, 44(OUT)
-	MOVW X12, 48(OUT)
-	MOVW X13, 52(OUT)
-	MOVW X14, 56(OUT)
-	MOVW X15, 60(OUT)
-	ADD $64, OUT, OUT
-
-	// Load input
-	LVX (INP)(R0), DD0
-	LVX (INP)(TMP0), DD1
-	LVX (INP)(TMP1), DD2
-	LVX (INP)(TMP2), DD3
-	LVX (INP)(TMP3), DD4
-	ADD $64, INP, INP
-
-	VPERM DD1, DD0, INPPERM, DD0 // Align input
-	VPERM DD2, DD1, INPPERM, DD1
-	VPERM DD3, DD2, INPPERM, DD2
-	VPERM DD4, DD3, INPPERM, DD3
-	VXOR A0, DD0, A0 // XOR with input
-	VXOR B0, DD1, B0
-	LVX (INP)(TMP0), DD1 // Keep loading input
-	VXOR C0, DD2, C0
-	LVX (INP)(TMP1), DD2
-	VXOR D0, DD3, D0
-	LVX (INP)(TMP2), DD3
-	LVX (INP)(TMP3), DD0
-	ADD $64, INP, INP
-	MOVD $63, TMP3 // 63 is not a typo
-	VPERM A0, A0, OUTPERM, A0
-	VPERM B0, B0, OUTPERM, B0
-	VPERM C0, C0, OUTPERM, C0
-	VPERM D0, D0, OUTPERM, D0
-
-	VPERM DD1, DD4, INPPERM, DD4 // Align input
-	VPERM DD2, DD1, INPPERM, DD1
-	VPERM DD3, DD2, INPPERM, DD2
-	VPERM DD0, DD3, INPPERM, DD3
-	VXOR A1, DD4, A1
-	VXOR B1, DD1, B1
-	LVX (INP)(TMP0), DD1 // Keep loading
-	VXOR C1, DD2, C1
-	LVX (INP)(TMP1), DD2
-	VXOR D1, DD3, D1
-	LVX (INP)(TMP2), DD3
-
-	// Note that the LVX address is always rounded down to the nearest 16-byte
-	// boundary, and that it always points to at most 15 bytes beyond the end of
-	// the slice, so we cannot cross a page boundary.
-	LVX (INP)(TMP3), DD4 // Redundant in aligned case.
-	ADD $64, INP, INP
-	VPERM A1, A1, OUTPERM, A1 // Pre-misalign output
-	VPERM B1, B1, OUTPERM, B1
-	VPERM C1, C1, OUTPERM, C1
-	VPERM D1, D1, OUTPERM, D1
-
-	VPERM DD1, DD0, INPPERM, DD0 // Align Input
-	VPERM DD2, DD1, INPPERM, DD1
-	VPERM DD3, DD2, INPPERM, DD2
-	VPERM DD4, DD3, INPPERM, DD3
-	VXOR A2, DD0, A2
-	VXOR B2, DD1, B2
-	VXOR C2, DD2, C2
-	VXOR D2, DD3, D2
-	VPERM A2, A2, OUTPERM, A2
-	VPERM B2, B2, OUTPERM, B2
-	VPERM C2, C2, OUTPERM, C2
-	VPERM D2, D2, OUTPERM, D2
-
-	ANDCC $15, OUT, X1 // Is out aligned?
-	MOVD OUT, X0
-
-	VSEL A0, B0, OUTMASK, DD0 // Collect pre-misaligned output
-	VSEL B0, C0, OUTMASK, DD1
-	VSEL C0, D0, OUTMASK, DD2
-	VSEL D0, A1, OUTMASK, DD3
-	VSEL A1, B1, OUTMASK, B0
-	VSEL B1, C1, OUTMASK, C0
-	VSEL C1, D1, OUTMASK, D0
-	VSEL D1, A2, OUTMASK, A1
-	VSEL A2, B2, OUTMASK, B1
-	VSEL B2, C2, OUTMASK, C1
-	VSEL C2, D2, OUTMASK, D1
-
-	STVX DD0, (OUT+TMP0)
-	STVX DD1, (OUT+TMP1)
-	STVX DD2, (OUT+TMP2)
-	ADD $64, OUT, OUT
-	STVX DD3, (OUT+R0)
-	STVX B0, (OUT+TMP0)
-	STVX C0, (OUT+TMP1)
-	STVX D0, (OUT+TMP2)
-	ADD $64, OUT, OUT
-	STVX A1, (OUT+R0)
-	STVX B1, (OUT+TMP0)
-	STVX C1, (OUT+TMP1)
-	STVX D1, (OUT+TMP2)
-	ADD $64, OUT, OUT
-
-	BEQ aligned_vmx
-
-	SUB X1, OUT, X2 // in misaligned case edges
-	MOVD $0, X3 // are written byte-by-byte
-
-unaligned_tail_vmx:
-	STVEBX D2, (X2+X3)
-	ADD $1, X3, X3
-	CMPW X3, X1
-	BNE unaligned_tail_vmx
-	SUB X1, X0, X2
-
-unaligned_head_vmx:
-	STVEBX A0, (X2+X1)
-	CMPW X1, $15
-	ADD $1, X1, X1
-	BNE unaligned_head_vmx
-
-	CMPU LEN, $255 // done with 256-byte block yet?
-	BGT loop_outer_vmx
-
-	JMP done_vmx
-
-aligned_vmx:
-	STVX A0, (X0+R0)
-	CMPU LEN, $255 // done with 256-byte block yet?
-	BGT loop_outer_vmx
-
-done_vmx:
-	RET

src/vendor/golang.org/x/crypto/internal/chacha20/chacha_generic.go

-// Copyright 2016 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 ChaCha20 implements the core ChaCha20 function as specified
-// in https://tools.ietf.org/html/rfc7539#section-2.3.
-package chacha20
-
-import (
-	"crypto/cipher"
-	"encoding/binary"
-
-	"golang.org/x/crypto/internal/subtle"
-)
-
-// assert that *Cipher implements cipher.Stream
-var _ cipher.Stream = (*Cipher)(nil)
-
-// Cipher is a stateful instance of ChaCha20 using a particular key
-// and nonce. A *Cipher implements the cipher.Stream interface.
-type Cipher struct {
-	key     [8]uint32
-	counter uint32 // incremented after each block
-	nonce   [3]uint32
-	buf     [bufSize]byte // buffer for unused keystream bytes
-	len     int           // number of unused keystream bytes at end of buf
-}
-
-// New creates a new ChaCha20 stream cipher with the given key and nonce.
-// The initial counter value is set to 0.
-func New(key [8]uint32, nonce [3]uint32) *Cipher {
-	return &Cipher{key: key, nonce: nonce}
-}
-
-// ChaCha20 constants spelling "expand 32-byte k"
-const (
-	j0 uint32 = 0x61707865
-	j1 uint32 = 0x3320646e
-	j2 uint32 = 0x79622d32
-	j3 uint32 = 0x6b206574
-)
-
-func quarterRound(a, b, c, d uint32) (uint32, uint32, uint32, uint32) {
-	a += b
-	d ^= a
-	d = (d << 16) | (d >> 16)
-	c += d
-	b ^= c
-	b = (b << 12) | (b >> 20)
-	a += b
-	d ^= a
-	d = (d << 8) | (d >> 24)
-	c += d
-	b ^= c
-	b = (b << 7) | (b >> 25)
-	return a, b, c, d
-}
-
-// XORKeyStream XORs each byte in the given slice with a byte from the
-// cipher's key stream. Dst and src must overlap entirely or not at all.
-//
-// If len(dst) < len(src), XORKeyStream will panic. It is acceptable
-// to pass a dst bigger than src, and in that case, XORKeyStream will
-// only update dst[:len(src)] and will not touch the rest of dst.
-//
-// Multiple calls to XORKeyStream behave as if the concatenation of
-// the src buffers was passed in a single run. That is, Cipher
-// maintains state and does not reset at each XORKeyStream call.
-func (s *Cipher) XORKeyStream(dst, src []byte) {
-	if len(dst) < len(src) {
-		panic("chacha20: output smaller than input")
-	}
-	if subtle.InexactOverlap(dst[:len(src)], src) {
-		panic("chacha20: invalid buffer overlap")
-	}
-
-	// xor src with buffered keystream first
-	if s.len != 0 {
-		buf := s.buf[len(s.buf)-s.len:]
-		if len(src) < len(buf) {
-			buf = buf[:len(src)]
-		}
-		td, ts := dst[:len(buf)], src[:len(buf)] // BCE hint
-		for i, b := range buf {
-			td[i] = ts[i] ^ b
-		}
-		s.len -= len(buf)
-		if s.len != 0 {
-			return
-		}
-		s.buf = [len(s.buf)]byte{} // zero the empty buffer
-		src = src[len(buf):]
-		dst = dst[len(buf):]
-	}
-
-	if len(src) == 0 {
-		return
-	}
-	if haveAsm {
-		if uint64(len(src))+uint64(s.counter)*64 > (1<<38)-64 {
-			panic("chacha20: counter overflow")
-		}
-		s.xorKeyStreamAsm(dst, src)
-		return
-	}
-
-	// set up a 64-byte buffer to pad out the final block if needed
-	// (hoisted out of the main loop to avoid spills)
-	rem := len(src) % 64  // length of final block
-	fin := len(src) - rem // index of final block
-	if rem > 0 {
-		copy(s.buf[len(s.buf)-64:], src[fin:])
-	}
-
-	// pre-calculate most of the first round
-	s1, s5, s9, s13 := quarterRound(j1, s.key[1], s.key[5], s.nonce[0])
-	s2, s6, s10, s14 := quarterRound(j2, s.key[2], s.key[6], s.nonce[1])
-	s3, s7, s11, s15 := quarterRound(j3, s.key[3], s.key[7], s.nonce[2])
-
-	n := len(src)
-	src, dst = src[:n:n], dst[:n:n] // BCE hint
-	for i := 0; i < n; i += 64 {
-		// calculate the remainder of the first round
-		s0, s4, s8, s12 := quarterRound(j0, s.key[0], s.key[4], s.counter)
-
-		// execute the second round
-		x0, x5, x10, x15 := quarterRound(s0, s5, s10, s15)
-		x1, x6, x11, x12 := quarterRound(s1, s6, s11, s12)
-		x2, x7, x8, x13 := quarterRound(s2, s7, s8, s13)
-		x3, x4, x9, x14 := quarterRound(s3, s4, s9, s14)
-
-		// execute the remaining 18 rounds
-		for i := 0; i < 9; i++ {
-			x0, x4, x8, x12 = quarterRound(x0, x4, x8, x12)
-			x1, x5, x9, x13 = quarterRound(x1, x5, x9, x13)
-			x2, x6, x10, x14 = quarterRound(x2, x6, x10, x14)
-			x3, x7, x11, x15 = quarterRound(x3, x7, x11, x15)
-
-			x0, x5, x10, x15 = quarterRound(x0, x5, x10, x15)
-			x1, x6, x11, x12 = quarterRound(x1, x6, x11, x12)
-			x2, x7, x8, x13 = quarterRound(x2, x7, x8, x13)
-			x3, x4, x9, x14 = quarterRound(x3, x4, x9, x14)
-		}
-
-		x0 += j0
-		x1 += j1
-		x2 += j2
-		x3 += j3
-
-		x4 += s.key[0]
-		x5 += s.key[1]
-		x6 += s.key[2]
-		x7 += s.key[3]
-		x8 += s.key[4]
-		x9 += s.key[5]
-		x10 += s.key[6]
-		x11 += s.key[7]
-
-		x12 += s.counter
-		x13 += s.nonce[0]
-		x14 += s.nonce[1]
-		x15 += s.nonce[2]
-
-		// increment the counter
-		s.counter += 1
-		if s.counter == 0 {
-			panic("chacha20: counter overflow")
-		}
-
-		// pad to 64 bytes if needed
-		in, out := src[i:], dst[i:]
-		if i == fin {
-			// src[fin:] has already been copied into s.buf before
-			// the main loop
-			in, out = s.buf[len(s.buf)-64:], s.buf[len(s.buf)-64:]
-		}
-		in, out = in[:64], out[:64] // BCE hint
-
-		// XOR the key stream with the source and write out the result
-		xor(out[0:], in[0:], x0)
-		xor(out[4:], in[4:], x1)
-		xor(out[8:], in[8:], x2)
-		xor(out[12:], in[12:], x3)
-		xor(out[16:], in[16:], x4)
-		xor(out[20:], in[20:], x5)
-		xor(out[24:], in[24:], x6)
-		xor(out[28:], in[28:], x7)
-		xor(out[32:], in[32:], x8)
-		xor(out[36:], in[36:], x9)
-		xor(out[40:], in[40:], x10)
-		xor(out[44:], in[44:], x11)
-		xor(out[48:], in[48:], x12)
-		xor(out[52:], in[52:], x13)
-		xor(out[56:], in[56:], x14)
-		xor(out[60:], in[60:], x15)
-	}
-	// copy any trailing bytes out of the buffer and into dst
-	if rem != 0 {
-		s.len = 64 - rem
-		copy(dst[fin:], s.buf[len(s.buf)-64:])
-	}
-}
-
-// Advance discards bytes in the key stream until the next 64 byte block
-// boundary is reached and updates the counter accordingly. If the key
-// stream is already at a block boundary no bytes will be discarded and
-// the counter will be unchanged.
-func (s *Cipher) Advance() {
-	s.len -= s.len % 64
-	if s.len == 0 {
-		s.buf = [len(s.buf)]byte{}
-	}
-}
-
-// XORKeyStream crypts bytes from in to out using the given key and counters.
-// In and out must overlap entirely or not at all. Counter contains the raw
-// ChaCha20 counter bytes (i.e. block counter followed by nonce).
-func XORKeyStream(out, in []byte, counter *[16]byte, key *[32]byte) {
-	s := Cipher{
-		key: [8]uint32{
-			binary.LittleEndian.Uint32(key[0:4]),
-			binary.LittleEndian.Uint32(key[4:8]),
-			binary.LittleEndian.Uint32(key[8:12]),
-			binary.LittleEndian.Uint32(key[12:16]),
-			binary.LittleEndian.Uint32(key[16:20]),
-			binary.LittleEndian.Uint32(key[20:24]),
-			binary.LittleEndian.Uint32(key[24:28]),
-			binary.LittleEndian.Uint32(key[28:32]),
-		},
-		nonce: [3]uint32{
-			binary.LittleEndian.Uint32(counter[4:8]),
-			binary.LittleEndian.Uint32(counter[8:12]),
-			binary.LittleEndian.Uint32(counter[12:16]),
-		},
-		counter: binary.LittleEndian.Uint32(counter[0:4]),
-	}
-	s.XORKeyStream(out, in)
-}
-
-// HChaCha20 uses the ChaCha20 core to generate a derived key from a key and a
-// nonce. It should only be used as part of the XChaCha20 construction.
-func HChaCha20(key *[8]uint32, nonce *[4]uint32) [8]uint32 {
-	x0, x1, x2, x3 := j0, j1, j2, j3
-	x4, x5, x6, x7 := key[0], key[1], key[2], key[3]
-	x8, x9, x10, x11 := key[4], key[5], key[6], key[7]
-	x12, x13, x14, x15 := nonce[0], nonce[1], nonce[2], nonce[3]
-
-	for i := 0; i < 10; i++ {
-		x0, x4, x8, x12 = quarterRound(x0, x4, x8, x12)
-		x1, x5, x9, x13 = quarterRound(x1, x5, x9, x13)
-		x2, x6, x10, x14 = quarterRound(x2, x6, x10, x14)
-		x3, x7, x11, x15 = quarterRound(x3, x7, x11, x15)
-
-		x0, x5, x10, x15 = quarterRound(x0, x5, x10, x15)
-		x1, x6, x11, x12 = quarterRound(x1, x6, x11, x12)
-		x2, x7, x8, x13 = quarterRound(x2, x7, x8, x13)
-		x3, x4, x9, x14 = quarterRound(x3, x4, x9, x14)
-	}
-
-	var out [8]uint32
-	out[0], out[1], out[2], out[3] = x0, x1, x2, x3
-	out[4], out[5], out[6], out[7] = x12, x13, x14, x15
-	return out
-}

src/vendor/golang.org/x/crypto/internal/chacha20/chacha_ppc64le.go

-// Copyright 2019 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.
-
-// +build ppc64le,!gccgo,!appengine
-
-package chacha20
-
-import "encoding/binary"
-
-const (
-	bufSize = 256
-	haveAsm = true
-)
-
-//go:noescape
-func chaCha20_ctr32_vmx(out, inp *byte, len int, key *[8]uint32, counter *uint32)
-
-func (c *Cipher) xorKeyStreamAsm(dst, src []byte) {
-	if len(src) >= bufSize {
-		chaCha20_ctr32_vmx(&dst[0], &src[0], len(src)-len(src)%bufSize, &c.key, &c.counter)
-	}
-	if len(src)%bufSize != 0 {
-		chaCha20_ctr32_vmx(&c.buf[0], &c.buf[0], bufSize, &c.key, &c.counter)
-		start := len(src) - len(src)%bufSize
-		ts, td, tb := src[start:], dst[start:], c.buf[:]
-		// Unroll loop to XOR 32 bytes per iteration.
-		for i := 0; i < len(ts)-32; i += 32 {
-			td, tb = td[:len(ts)], tb[:len(ts)] // bounds check elimination
-			s0 := binary.LittleEndian.Uint64(ts[0:8])
-			s1 := binary.LittleEndian.Uint64(ts[8:16])
-			s2 := binary.LittleEndian.Uint64(ts[16:24])
-			s3 := binary.LittleEndian.Uint64(ts[24:32])
-			b0 := binary.LittleEndian.Uint64(tb[0:8])
-			b1 := binary.LittleEndian.Uint64(tb[8:16])
-			b2 := binary.LittleEndian.Uint64(tb[16:24])
-			b3 := binary.LittleEndian.Uint64(tb[24:32])
-			binary.LittleEndian.PutUint64(td[0:8], s0^b0)
-			binary.LittleEndian.PutUint64(td[8:16], s1^b1)
-			binary.LittleEndian.PutUint64(td[16:24], s2^b2)
-			binary.LittleEndian.PutUint64(td[24:32], s3^b3)
-			ts, td, tb = ts[32:], td[32:], tb[32:]
-		}
-		td, tb = td[:len(ts)], tb[:len(ts)] // bounds check elimination
-		for i, v := range ts {
-			td[i] = tb[i] ^ v
-		}
-		c.len = bufSize - (len(src) % bufSize)
-
-	}
-
-}

src/vendor/golang.org/x/crypto/poly1305/bits_compat.go

+// Copyright 2019 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.
+
+// +build !go1.13
+
+package poly1305
+
+// Generic fallbacks for the math/bits intrinsics, copied from
+// src/math/bits/bits.go. They were added in Go 1.12, but Add64 and Sum64 had
+// variable time fallbacks until Go 1.13.
+
+func bitsAdd64(x, y, carry uint64) (sum, carryOut uint64) {
+	sum = x + y + carry
+	carryOut = ((x & y) | ((x | y) &^ sum)) >> 63
+	return
+}
+
+func bitsSub64(x, y, borrow uint64) (diff, borrowOut uint64) {
+	diff = x - y - borrow
+	borrowOut = ((^x & y) | (^(x ^ y) & diff)) >> 63
+	return
+}
+
+func bitsMul64(x, y uint64) (hi, lo uint64) {
+	const mask32 = 1<<32 - 1
+	x0 := x & mask32
+	x1 := x >> 32
+	y0 := y & mask32
+	y1 := y >> 32
+	w0 := x0 * y0
+	t := x1*y0 + w0>>32
+	w1 := t & mask32
+	w2 := t >> 32
+	w1 += x0 * y1
+	hi = x1*y1 + w2 + w1>>32
+	lo = x * y
+	return
+}

src/vendor/golang.org/x/crypto/poly1305/bits_go1.13.go

+// Copyright 2019 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.
+
+// +build go1.13
+
+package poly1305
+
+import "math/bits"
+
+func bitsAdd64(x, y, carry uint64) (sum, carryOut uint64) {
+	return bits.Add64(x, y, carry)
+}
+
+func bitsSub64(x, y, borrow uint64) (diff, borrowOut uint64) {
+	return bits.Sub64(x, y, borrow)
+}
+
+func bitsMul64(x, y uint64) (hi, lo uint64) {
+	return bits.Mul64(x, y)
+}

src/vendor/golang.org/x/crypto/poly1305/poly1305.go

@@ -22,8 +22,14 @@ import "crypto/subtle"
 // TagSize is the size, in bytes, of a poly1305 authenticator.
 const TagSize = 16
 
-// Verify returns true if mac is a valid authenticator for m with the given
-// key.
+// Sum generates an authenticator for msg 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 Sum(out *[16]byte, m []byte, key *[32]byte) {
+	sum(out, m, key)
+}
+
+// Verify returns true if mac is a valid authenticator for m with the given key.
 func Verify(mac *[16]byte, m []byte, key *[32]byte) bool {
 	var tmp [16]byte
 	Sum(&tmp, m, key)

src/vendor/golang.org/x/crypto/poly1305/sum_amd64.go

@@ -7,62 +7,52 @@
 package poly1305
 
 //go:noescape
-func initialize(state *[7]uint64, key *[32]byte)
+func update(state *macState, msg []byte)
 
-//go:noescape
-func update(state *[7]uint64, msg []byte)
-
-//go:noescape
-func finalize(tag *[TagSize]byte, state *[7]uint64)
-
-// Sum 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 Sum(out *[16]byte, m []byte, key *[32]byte) {
+func sum(out *[16]byte, m []byte, key *[32]byte) {
 	h := newMAC(key)
 	h.Write(m)
 	h.Sum(out)
 }
 
 func newMAC(key *[32]byte) (h mac) {
-	initialize(&h.state, key)
+	initialize(key, &h.r, &h.s)
 	return
 }
 
-type mac struct {
-	state [7]uint64 // := uint64{ h0, h1, h2, r0, r1, pad0, pad1 }
+// mac is a wrapper for macGeneric that redirects calls that would have gone to
+// updateGeneric to update.
+//
+// Its Write and Sum methods are otherwise identical to the macGeneric ones, but
+// using function pointers would carry a major performance cost.
+type mac struct{ macGeneric }
 
-	buffer [TagSize]byte
-	offset int
-}
-
-func (h *mac) Write(p []byte) (n int, err error) {
-	n = len(p)
+func (h *mac) Write(p []byte) (int, error) {
+	nn := len(p)
 	if h.offset > 0 {
-		remaining := TagSize - h.offset
-		if n < remaining {
-			h.offset += copy(h.buffer[h.offset:], p)
-			return n, nil
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
 		}
-		copy(h.buffer[h.offset:], p[:remaining])
-		p = p[remaining:]
+		p = p[n:]
 		h.offset = 0
-		update(&h.state, h.buffer[:])
+		update(&h.macState, h.buffer[:])
 	}
-	if nn := len(p) - (len(p) % TagSize); nn > 0 {
-		update(&h.state, p[:nn])
-		p = p[nn:]
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		update(&h.macState, p[:n])
+		p = p[n:]
 	}
 	if len(p) > 0 {
 		h.offset += copy(h.buffer[h.offset:], p)
 	}
-	return n, nil
+	return nn, nil
 }
 
 func (h *mac) Sum(out *[16]byte) {
-	state := h.state
+	state := h.macState
 	if h.offset > 0 {
 		update(&state, h.buffer[:h.offset])
 	}
-	finalize(out, &state)
+	finalize(out, &state.h, &state.s)
 }

src/vendor/golang.org/x/crypto/poly1305/sum_amd64.s

@@ -54,10 +54,6 @@
 	ADCQ  t3, h1;                  \
 	ADCQ  $0, h2
 
-DATA ·poly1305Mask<>+0x00(SB)/8, $0x0FFFFFFC0FFFFFFF
-DATA ·poly1305Mask<>+0x08(SB)/8, $0x0FFFFFFC0FFFFFFC
-GLOBL ·poly1305Mask<>(SB), RODATA, $16
-
 // func update(state *[7]uint64, msg []byte)
 TEXT ·update(SB), $0-32
 	MOVQ state+0(FP), DI
@@ -110,39 +106,3 @@ done:
 	MOVQ R9, 8(DI)
 	MOVQ R10, 16(DI)
 	RET
-
-// func initialize(state *[7]uint64, key *[32]byte)
-TEXT ·initialize(SB), $0-16
-	MOVQ state+0(FP), DI
-	MOVQ key+8(FP), SI
-
-	// state[0...7] is initialized with zero
-	MOVOU 0(SI), X0
-	MOVOU 16(SI), X1
-	MOVOU ·poly1305Mask<>(SB), X2
-	PAND  X2, X0
-	MOVOU X0, 24(DI)
-	MOVOU X1, 40(DI)
-	RET
-
-// func finalize(tag *[TagSize]byte, state *[7]uint64)
-TEXT ·finalize(SB), $0-16
-	MOVQ tag+0(FP), DI
-	MOVQ state+8(FP), SI
-
-	MOVQ    0(SI), AX
-	MOVQ    8(SI), BX
-	MOVQ    16(SI), CX
-	MOVQ    AX, R8
-	MOVQ    BX, R9
-	SUBQ    $0xFFFFFFFFFFFFFFFB, AX
-	SBBQ    $0xFFFFFFFFFFFFFFFF, BX
-	SBBQ    $3, CX
-	CMOVQCS R8, AX
-	CMOVQCS R9, BX
-	ADDQ    40(SI), AX
-	ADCQ    48(SI), BX
-
-	MOVQ AX, 0(DI)
-	MOVQ BX, 8(DI)
-	RET

src/vendor/golang.org/x/crypto/poly1305/sum_arm.go

@@ -6,14 +6,11 @@
 
 package poly1305
 
-// This function is implemented in sum_arm.s
+// poly1305_auth_armv6 is implemented in sum_arm.s
 //go:noescape
 func poly1305_auth_armv6(out *[16]byte, m *byte, mlen uint32, key *[32]byte)
 
-// Sum 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 Sum(out *[16]byte, m []byte, key *[32]byte) {
+func sum(out *[16]byte, m []byte, key *[32]byte) {
 	var mPtr *byte
 	if len(m) > 0 {
 		mPtr = &m[0]

src/vendor/golang.org/x/crypto/poly1305/sum_generic.go

@@ -2,18 +2,29 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+// This file provides the generic implementation of Sum and MAC. Other files
+// might provide optimized assembly implementations of some of this code.
+
 package poly1305
 
 import "encoding/binary"
 
-const (
-	msgBlock   = uint32(1 << 24)
-	finalBlock = uint32(0)
-)
+// Poly1305 [RFC 7539] is a relatively simple algorithm: the authentication tag
+// for a 64 bytes message is approximately
+//
+//     s + m[0:16] * r⁴ + m[16:32] * r³ + m[32:48] * r² + m[48:64] * r  mod  2¹³⁰ - 5
+//
+// for some secret r and s. It can be computed sequentially like
+//
+//     for len(msg) > 0:
+//         h += read(msg, 16)
+//         h *= r
+//         h %= 2¹³⁰ - 5
+//     return h + s
+//
+// All the complexity is about doing performant constant-time math on numbers
+// larger than any available numeric type.
 
-// sumGeneric generates an authenticator for msg using a one-time key and
-// puts the 16-byte result into out. This is the generic implementation of
-// Sum and should be called if no assembly implementation is available.
 func sumGeneric(out *[TagSize]byte, msg []byte, key *[32]byte) {
 	h := newMACGeneric(key)
 	h.Write(msg)
@@ -21,152 +32,276 @@ func sumGeneric(out *[TagSize]byte, msg []byte, key *[32]byte) {
 }
 
 func newMACGeneric(key *[32]byte) (h macGeneric) {
-	h.r[0] = binary.LittleEndian.Uint32(key[0:]) & 0x3ffffff
-	h.r[1] = (binary.LittleEndian.Uint32(key[3:]) >> 2) & 0x3ffff03
-	h.r[2] = (binary.LittleEndian.Uint32(key[6:]) >> 4) & 0x3ffc0ff
-	h.r[3] = (binary.LittleEndian.Uint32(key[9:]) >> 6) & 0x3f03fff
-	h.r[4] = (binary.LittleEndian.Uint32(key[12:]) >> 8) & 0x00fffff
-
-	h.s[0] = binary.LittleEndian.Uint32(key[16:])
-	h.s[1] = binary.LittleEndian.Uint32(key[20:])
-	h.s[2] = binary.LittleEndian.Uint32(key[24:])
-	h.s[3] = binary.LittleEndian.Uint32(key[28:])
+	initialize(key, &h.r, &h.s)
 	return
 }
 
+// macState holds numbers in saturated 64-bit little-endian limbs. That is,
+// the value of [x0, x1, x2] is x[0] + x[1] * 2⁶⁴ + x[2] * 2¹²⁸.
+type macState struct {
+	// h is the main accumulator. It is to be interpreted modulo 2¹³⁰ - 5, but
+	// can grow larger during and after rounds.
+	h [3]uint64
+	// r and s are the private key components.
+	r [2]uint64
+	s [2]uint64
+}
+
 type macGeneric struct {
-	h, r [5]uint32
-	s    [4]uint32
+	macState
 
 	buffer [TagSize]byte
 	offset int
 }
 
-func (h *macGeneric) Write(p []byte) (n int, err error) {
-	n = len(p)
+// Write splits the incoming message into TagSize chunks, and passes them to
+// update. It buffers incomplete chunks.
+func (h *macGeneric) Write(p []byte) (int, error) {
+	nn := len(p)
 	if h.offset > 0 {
-		remaining := TagSize - h.offset
-		if n < remaining {
-			h.offset += copy(h.buffer[h.offset:], p)
-			return n, nil
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
 		}
-		copy(h.buffer[h.offset:], p[:remaining])
-		p = p[remaining:]
+		p = p[n:]
 		h.offset = 0
-		updateGeneric(h.buffer[:], msgBlock, &(h.h), &(h.r))
+		updateGeneric(&h.macState, h.buffer[:])
 	}
-	if nn := len(p) - (len(p) % TagSize); nn > 0 {
-		updateGeneric(p, msgBlock, &(h.h), &(h.r))
-		p = p[nn:]
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		updateGeneric(&h.macState, p[:n])
+		p = p[n:]
 	}
 	if len(p) > 0 {
 		h.offset += copy(h.buffer[h.offset:], p)
 	}
-	return n, nil
+	return nn, nil
 }
 
-func (h *macGeneric) Sum(out *[16]byte) {
-	H, R := h.h, h.r
+// Sum flushes the last incomplete chunk from the buffer, if any, and generates
+// the MAC output. It does not modify its state, in order to allow for multiple
+// calls to Sum, even if no Write is allowed after Sum.
+func (h *macGeneric) Sum(out *[TagSize]byte) {
+	state := h.macState
 	if h.offset > 0 {
-		var buffer [TagSize]byte
-		copy(buffer[:], h.buffer[:h.offset])
-		buffer[h.offset] = 1 // invariant: h.offset < TagSize
-		updateGeneric(buffer[:], finalBlock, &H, &R)
+		updateGeneric(&state, h.buffer[:h.offset])
+	}
+	finalize(out, &state.h, &state.s)
+}
+
+// [rMask0, rMask1] is the specified Poly1305 clamping mask in little-endian. It
+// clears some bits of the secret coefficient to make it possible to implement
+// multiplication more efficiently.
+const (
+	rMask0 = 0x0FFFFFFC0FFFFFFF
+	rMask1 = 0x0FFFFFFC0FFFFFFC
+)
+
+func initialize(key *[32]byte, r, s *[2]uint64) {
+	r[0] = binary.LittleEndian.Uint64(key[0:8]) & rMask0
+	r[1] = binary.LittleEndian.Uint64(key[8:16]) & rMask1
+	s[0] = binary.LittleEndian.Uint64(key[16:24])
+	s[1] = binary.LittleEndian.Uint64(key[24:32])
+}
+
+// uint128 holds a 128-bit number as two 64-bit limbs, for use with the
+// bits.Mul64 and bits.Add64 intrinsics.
+type uint128 struct {
+	lo, hi uint64
+}
+
+func mul64(a, b uint64) uint128 {
+	hi, lo := bitsMul64(a, b)
+	return uint128{lo, hi}
+}
+
+func add128(a, b uint128) uint128 {
+	lo, c := bitsAdd64(a.lo, b.lo, 0)
+	hi, c := bitsAdd64(a.hi, b.hi, c)
+	if c != 0 {
+		panic("poly1305: unexpected overflow")
 	}
-	finalizeGeneric(out, &H, &(h.s))
+	return uint128{lo, hi}
+}
+
+func shiftRightBy2(a uint128) uint128 {
+	a.lo = a.lo>>2 | (a.hi&3)<<62
+	a.hi = a.hi >> 2
+	return a
 }
 
-func updateGeneric(msg []byte, flag uint32, h, r *[5]uint32) {
-	h0, h1, h2, h3, h4 := h[0], h[1], h[2], h[3], h[4]
-	r0, r1, r2, r3, r4 := uint64(r[0]), uint64(r[1]), uint64(r[2]), uint64(r[3]), uint64(r[4])
-	R1, R2, R3, R4 := r1*5, r2*5, r3*5, r4*5
-
-	for len(msg) >= TagSize {
-		// h += msg
-		h0 += binary.LittleEndian.Uint32(msg[0:]) & 0x3ffffff
-		h1 += (binary.LittleEndian.Uint32(msg[3:]) >> 2) & 0x3ffffff
-		h2 += (binary.LittleEndian.Uint32(msg[6:]) >> 4) & 0x3ffffff
-		h3 += (binary.LittleEndian.Uint32(msg[9:]) >> 6) & 0x3ffffff
-		h4 += (binary.LittleEndian.Uint32(msg[12:]) >> 8) | flag
-
-		// h *= r
-		d0 := (uint64(h0) * r0) + (uint64(h1) * R4) + (uint64(h2) * R3) + (uint64(h3) * R2) + (uint64(h4) * R1)
-		d1 := (d0 >> 26) + (uint64(h0) * r1) + (uint64(h1) * r0) + (uint64(h2) * R4) + (uint64(h3) * R3) + (uint64(h4) * R2)
-		d2 := (d1 >> 26) + (uint64(h0) * r2) + (uint64(h1) * r1) + (uint64(h2) * r0) + (uint64(h3) * R4) + (uint64(h4) * R3)
-		d3 := (d2 >> 26) + (uint64(h0) * r3) + (uint64(h1) * r2) + (uint64(h2) * r1) + (uint64(h3) * r0) + (uint64(h4) * R4)
-		d4 := (d3 >> 26) + (uint64(h0) * r4) + (uint64(h1) * r3) + (uint64(h2) * r2) + (uint64(h3) * r1) + (uint64(h4) * r0)
-
-		// h %= p
-		h0 = uint32(d0) & 0x3ffffff
-		h1 = uint32(d1) & 0x3ffffff
-		h2 = uint32(d2) & 0x3ffffff
-		h3 = uint32(d3) & 0x3ffffff
-		h4 = uint32(d4) & 0x3ffffff
-
-		h0 += uint32(d4>>26) * 5
-		h1 += h0 >> 26
-		h0 = h0 & 0x3ffffff
+// updateGeneric absorbs msg into the state.h accumulator. For each chunk m of
+// 128 bits of message, it computes
+//
+//     h₊ = (h + m) * r  mod  2¹³⁰ - 5
+//
+// If the msg length is not a multiple of TagSize, it assumes the last
+// incomplete chunk is the final one.
+func updateGeneric(state *macState, msg []byte) {
+	h0, h1, h2 := state.h[0], state.h[1], state.h[2]
+	r0, r1 := state.r[0], state.r[1]
+
+	for len(msg) > 0 {
+		var c uint64
+
+		// For the first step, h + m, we use a chain of bits.Add64 intrinsics.
+		// The resulting value of h might exceed 2¹³⁰ - 5, but will be partially
+		// reduced at the end of the multiplication below.
+		//
+		// The spec requires us to set a bit just above the message size, not to
+		// hide leading zeroes. For full chunks, that's 1 << 128, so we can just
+		// add 1 to the most significant (2¹²⁸) limb, h2.
+		if len(msg) >= TagSize {
+			h0, c = bitsAdd64(h0, binary.LittleEndian.Uint64(msg[0:8]), 0)
+			h1, c = bitsAdd64(h1, binary.LittleEndian.Uint64(msg[8:16]), c)
+			h2 += c + 1
 
 			msg = msg[TagSize:]
+		} else {
+			var buf [TagSize]byte
+			copy(buf[:], msg)
+			buf[len(msg)] = 1
+
+			h0, c = bitsAdd64(h0, binary.LittleEndian.Uint64(buf[0:8]), 0)
+			h1, c = bitsAdd64(h1, binary.LittleEndian.Uint64(buf[8:16]), c)
+			h2 += c
+
+			msg = nil
 		}
 
-	h[0], h[1], h[2], h[3], h[4] = h0, h1, h2, h3, h4
+		// Multiplication of big number limbs is similar to elementary school
+		// columnar multiplication. Instead of digits, there are 64-bit limbs.
+		//
+		// We are multiplying a 3 limbs number, h, by a 2 limbs number, r.
+		//
+		//                        h2    h1    h0  x
+		//                              r1    r0  =
+		//                       ----------------
+		//                      h2r0  h1r0  h0r0     <-- individual 128-bit products
+		//            +   h2r1  h1r1  h0r1
+		//               ------------------------
+		//                 m3    m2    m1    m0      <-- result in 128-bit overlapping limbs
+		//               ------------------------
+		//         m3.hi m2.hi m1.hi m0.hi           <-- carry propagation
+		//     +         m3.lo m2.lo m1.lo m0.lo
+		//        -------------------------------
+		//           t4    t3    t2    t1    t0      <-- final result in 64-bit limbs
+		//
+		// The main difference from pen-and-paper multiplication is that we do
+		// carry propagation in a separate step, as if we wrote two digit sums
+		// at first (the 128-bit limbs), and then carried the tens all at once.
+
+		h0r0 := mul64(h0, r0)
+		h1r0 := mul64(h1, r0)
+		h2r0 := mul64(h2, r0)
+		h0r1 := mul64(h0, r1)
+		h1r1 := mul64(h1, r1)
+		h2r1 := mul64(h2, r1)
+
+		// Since h2 is known to be at most 7 (5 + 1 + 1), and r0 and r1 have their
+		// top 4 bits cleared by rMask{0,1}, we know that their product is not going
+		// to overflow 64 bits, so we can ignore the high part of the products.
+		//
+		// This also means that the product doesn't have a fifth limb (t4).
+		if h2r0.hi != 0 {
+			panic("poly1305: unexpected overflow")
+		}
+		if h2r1.hi != 0 {
+			panic("poly1305: unexpected overflow")
+		}
+
+		m0 := h0r0
+		m1 := add128(h1r0, h0r1) // These two additions don't overflow thanks again
+		m2 := add128(h2r0, h1r1) // to the 4 masked bits at the top of r0 and r1.
+		m3 := h2r1
+
+		t0 := m0.lo
+		t1, c := bitsAdd64(m1.lo, m0.hi, 0)
+		t2, c := bitsAdd64(m2.lo, m1.hi, c)
+		t3, _ := bitsAdd64(m3.lo, m2.hi, c)
+
+		// Now we have the result as 4 64-bit limbs, and we need to reduce it
+		// modulo 2¹³⁰ - 5. The special shape of this Crandall prime lets us do
+		// a cheap partial reduction according to the reduction identity
+		//
+		//     c * 2¹³⁰ + n  =  c * 5 + n  mod  2¹³⁰ - 5
+		//
+		// because 2¹³⁰ = 5 mod 2¹³⁰ - 5. Partial reduction since the result is
+		// likely to be larger than 2¹³⁰ - 5, but still small enough to fit the
+		// assumptions we make about h in the rest of the code.
+		//
+		// See also https://speakerdeck.com/gtank/engineering-prime-numbers?slide=23
+
+		// We split the final result at the 2¹³⁰ mark into h and cc, the carry.
+		// Note that the carry bits are effectively shifted left by 2, in other
+		// words, cc = c * 4 for the c in the reduction identity.
+		h0, h1, h2 = t0, t1, t2&maskLow2Bits
+		cc := uint128{t2 & maskNotLow2Bits, t3}
+
+		// To add c * 5 to h, we first add cc = c * 4, and then add (cc >> 2) = c.
+
+		h0, c = bitsAdd64(h0, cc.lo, 0)
+		h1, c = bitsAdd64(h1, cc.hi, c)
+		h2 += c
+
+		cc = shiftRightBy2(cc)
+
+		h0, c = bitsAdd64(h0, cc.lo, 0)
+		h1, c = bitsAdd64(h1, cc.hi, c)
+		h2 += c
+
+		// h2 is at most 3 + 1 + 1 = 5, making the whole of h at most
+		//
+		//     5 * 2¹²⁸ + (2¹²⁸ - 1) = 6 * 2¹²⁸ - 1
+	}
+
+	state.h[0], state.h[1], state.h[2] = h0, h1, h2
 }
 
-func finalizeGeneric(out *[TagSize]byte, h *[5]uint32, s *[4]uint32) {
-	h0, h1, h2, h3, h4 := h[0], h[1], h[2], h[3], h[4]
-
-	// h %= p reduction
-	h2 += h1 >> 26
-	h1 &= 0x3ffffff
-	h3 += h2 >> 26
-	h2 &= 0x3ffffff
-	h4 += h3 >> 26
-	h3 &= 0x3ffffff
-	h0 += 5 * (h4 >> 26)
-	h4 &= 0x3ffffff
-	h1 += h0 >> 26
-	h0 &= 0x3ffffff
-
-	// h - p
-	t0 := h0 + 5
-	t1 := h1 + (t0 >> 26)
-	t2 := h2 + (t1 >> 26)
-	t3 := h3 + (t2 >> 26)
-	t4 := h4 + (t3 >> 26) - (1 << 26)
-	t0 &= 0x3ffffff
-	t1 &= 0x3ffffff
-	t2 &= 0x3ffffff
-	t3 &= 0x3ffffff
-
-	// select h if h < p else h - p
-	t_mask := (t4 >> 31) - 1
-	h_mask := ^t_mask
-	h0 = (h0 & h_mask) | (t0 & t_mask)
-	h1 = (h1 & h_mask) | (t1 & t_mask)
-	h2 = (h2 & h_mask) | (t2 & t_mask)
-	h3 = (h3 & h_mask) | (t3 & t_mask)
-	h4 = (h4 & h_mask) | (t4 & t_mask)
-
-	// h %= 2^128
-	h0 |= h1 << 26
-	h1 = ((h1 >> 6) | (h2 << 20))
-	h2 = ((h2 >> 12) | (h3 << 14))
-	h3 = ((h3 >> 18) | (h4 << 8))
-
-	// s: the s part of the key
-	// tag = (h + s) % (2^128)
-	t := uint64(h0) + uint64(s[0])
-	h0 = uint32(t)
-	t = uint64(h1) + uint64(s[1]) + (t >> 32)
-	h1 = uint32(t)
-	t = uint64(h2) + uint64(s[2]) + (t >> 32)
-	h2 = uint32(t)
-	t = uint64(h3) + uint64(s[3]) + (t >> 32)
-	h3 = uint32(t)
-
-	binary.LittleEndian.PutUint32(out[0:], h0)
-	binary.LittleEndian.PutUint32(out[4:], h1)
-	binary.LittleEndian.PutUint32(out[8:], h2)
-	binary.LittleEndian.PutUint32(out[12:], h3)
+const (
+	maskLow2Bits    uint64 = 0x0000000000000003
+	maskNotLow2Bits uint64 = ^maskLow2Bits
+)
+
+// select64 returns x if v == 1 and y if v == 0, in constant time.
+func select64(v, x, y uint64) uint64 { return ^(v-1)&x | (v-1)&y }
+
+// [p0, p1, p2] is 2¹³⁰ - 5 in little endian order.
+const (
+	p0 = 0xFFFFFFFFFFFFFFFB
+	p1 = 0xFFFFFFFFFFFFFFFF
+	p2 = 0x0000000000000003
+)
+
+// finalize completes the modular reduction of h and computes
+//
+//     out = h + s  mod  2¹²⁸
+//
+func finalize(out *[TagSize]byte, h *[3]uint64, s *[2]uint64) {
+	h0, h1, h2 := h[0], h[1], h[2]
+
+	// After the partial reduction in updateGeneric, h might be more than
+	// 2¹³⁰ - 5, but will be less than 2 * (2¹³⁰ - 5). To complete the reduction
+	// in constant time, we compute t = h - (2¹³⁰ - 5), and select h as the
+	// result if the subtraction underflows, and t otherwise.
+
+	hMinusP0, b := bitsSub64(h0, p0, 0)
+	hMinusP1, b := bitsSub64(h1, p1, b)
+	_, b = bitsSub64(h2, p2, b)
+
+	// h = h if h < p else h - p
+	h0 = select64(b, h0, hMinusP0)
+	h1 = select64(b, h1, hMinusP1)
+
+	// Finally, we compute the last Poly1305 step
+	//
+	//     tag = h + s  mod  2¹²⁸
+	//
+	// by just doing a wide addition with the 128 low bits of h and discarding
+	// the overflow.
+	h0, c := bitsAdd64(h0, s[0], 0)
+	h1, _ = bitsAdd64(h1, s[1], c)
+
+	binary.LittleEndian.PutUint64(out[0:8], h0)
+	binary.LittleEndian.PutUint64(out[8:16], h1)
 }

src/vendor/golang.org/x/crypto/poly1305/sum_noasm.go

@@ -6,10 +6,7 @@
 
 package poly1305
 
-// Sum generates an authenticator for msg 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 Sum(out *[TagSize]byte, msg []byte, key *[32]byte) {
+func sum(out *[TagSize]byte, msg []byte, key *[32]byte) {
 	h := newMAC(key)
 	h.Write(msg)
 	h.Sum(out)

src/vendor/golang.org/x/crypto/poly1305/sum_ppc64le.go

@@ -7,62 +7,52 @@
 package poly1305
 
 //go:noescape
-func initialize(state *[7]uint64, key *[32]byte)
+func update(state *macState, msg []byte)
 
-//go:noescape
-func update(state *[7]uint64, msg []byte)
-
-//go:noescape
-func finalize(tag *[TagSize]byte, state *[7]uint64)
-
-// Sum 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 Sum(out *[16]byte, m []byte, key *[32]byte) {
+func sum(out *[16]byte, m []byte, key *[32]byte) {
 	h := newMAC(key)
 	h.Write(m)
 	h.Sum(out)
 }
 
 func newMAC(key *[32]byte) (h mac) {
-	initialize(&h.state, key)
+	initialize(key, &h.r, &h.s)
 	return
 }
 
-type mac struct {
-	state [7]uint64 // := uint64{ h0, h1, h2, r0, r1, pad0, pad1 }
+// mac is a wrapper for macGeneric that redirects calls that would have gone to
+// updateGeneric to update.
+//
+// Its Write and Sum methods are otherwise identical to the macGeneric ones, but
+// using function pointers would carry a major performance cost.
+type mac struct{ macGeneric }
 
-	buffer [TagSize]byte
-	offset int
-}
-
-func (h *mac) Write(p []byte) (n int, err error) {
-	n = len(p)
+func (h *mac) Write(p []byte) (int, error) {
+	nn := len(p)
 	if h.offset > 0 {
-		remaining := TagSize - h.offset
-		if n < remaining {
-			h.offset += copy(h.buffer[h.offset:], p)
-			return n, nil
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
 		}
-		copy(h.buffer[h.offset:], p[:remaining])
-		p = p[remaining:]
+		p = p[n:]
 		h.offset = 0
-		update(&h.state, h.buffer[:])
+		update(&h.macState, h.buffer[:])
 	}
-	if nn := len(p) - (len(p) % TagSize); nn > 0 {
-		update(&h.state, p[:nn])
-		p = p[nn:]
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		update(&h.macState, p[:n])
+		p = p[n:]
 	}
 	if len(p) > 0 {
 		h.offset += copy(h.buffer[h.offset:], p)
 	}
-	return n, nil
+	return nn, nil
 }
 
 func (h *mac) Sum(out *[16]byte) {
-	state := h.state
+	state := h.macState
 	if h.offset > 0 {
 		update(&state, h.buffer[:h.offset])
 	}
-	finalize(out, &state)
+	finalize(out, &state.h, &state.s)
 }

src/vendor/golang.org/x/crypto/poly1305/sum_ppc64le.s

@@ -58,7 +58,6 @@ DATA ·poly1305Mask<>+0x08(SB)/8, $0x0FFFFFFC0FFFFFFC
 GLOBL ·poly1305Mask<>(SB), RODATA, $16
 
 // func update(state *[7]uint64, msg []byte)
-
 TEXT ·update(SB), $0-32
 	MOVD state+0(FP), R3
 	MOVD msg_base+8(FP), R4
@@ -180,68 +179,3 @@ done:
 	MOVD R9, 8(R3)
 	MOVD R10, 16(R3)
 	RET
-
-// func initialize(state *[7]uint64, key *[32]byte)
-TEXT ·initialize(SB), $0-16
-	MOVD state+0(FP), R3
-	MOVD key+8(FP), R4
-
-	// state[0...7] is initialized with zero
-	// Load key
-	MOVD 0(R4), R5
-	MOVD 8(R4), R6
-	MOVD 16(R4), R7
-	MOVD 24(R4), R8
-
-	// Address of key mask
-	MOVD $·poly1305Mask<>(SB), R9
-
-	// Save original key in state
-	MOVD R7, 40(R3)
-	MOVD R8, 48(R3)
-
-	// Get mask
-	MOVD (R9), R7
-	MOVD 8(R9), R8
-
-	// And with key
-	AND R5, R7, R5
-	AND R6, R8, R6
-
-	// Save masked key in state
-	MOVD R5, 24(R3)
-	MOVD R6, 32(R3)
-	RET
-
-// func finalize(tag *[TagSize]byte, state *[7]uint64)
-TEXT ·finalize(SB), $0-16
-	MOVD tag+0(FP), R3
-	MOVD state+8(FP), R4
-
-	// Get h0, h1, h2 from state
-	MOVD 0(R4), R5
-	MOVD 8(R4), R6
-	MOVD 16(R4), R7
-
-	// Save h0, h1
-	MOVD  R5, R8
-	MOVD  R6, R9
-	MOVD  $3, R20
-	MOVD  $-1, R21
-	SUBC  $-5, R5
-	SUBE  R21, R6
-	SUBE  R20, R7
-	MOVD  $0, R21
-	SUBZE R21
-
-	// Check for carry
-	CMP  $0, R21
-	ISEL $2, R5, R8, R5
-	ISEL $2, R6, R9, R6
-	MOVD 40(R4), R8
-	MOVD 48(R4), R9
-	ADDC R8, R5
-	ADDE R9, R6
-	MOVD R5, 0(R3)
-	MOVD R6, 8(R3)
-	RET

src/vendor/golang.org/x/crypto/poly1305/sum_s390x.go

@@ -22,10 +22,7 @@ func poly1305vx(out *[16]byte, m *byte, mlen uint64, key *[32]byte)
 //go:noescape
 func poly1305vmsl(out *[16]byte, m *byte, mlen uint64, key *[32]byte)
 
-// Sum 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 Sum(out *[16]byte, m []byte, key *[32]byte) {
+func sum(out *[16]byte, m []byte, key *[32]byte) {
 	if cpu.S390X.HasVX {
 		var mPtr *byte
 		if len(m) > 0 {

src/vendor/modules.txt

-# golang.org/x/crypto v0.0.0-20190611184440-5c40567a22f8
+# golang.org/x/crypto v0.0.0-20191111213947-16651526fdb4
 ## explicit
+golang.org/x/crypto/chacha20
 golang.org/x/crypto/chacha20poly1305
 golang.org/x/crypto/cryptobyte
 golang.org/x/crypto/cryptobyte/asn1
 golang.org/x/crypto/curve25519
 golang.org/x/crypto/hkdf
-golang.org/x/crypto/internal/chacha20
 golang.org/x/crypto/internal/subtle
 golang.org/x/crypto/poly1305
 # golang.org/x/net v0.0.0-20191105084925-a882066a44e0