initial commit

8 files changed, 1750 insertions, 0 deletions

diff --git a/vendor/golang.org/x/crypto/blake2b/blake2b.go b/vendor/golang.org/x/crypto/blake2b/blake2b.go
new file mode 100644
index 0000000..d2e98d4
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2b.go
@@ -0,0 +1,291 @@
+// 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 blake2b implements the BLAKE2b hash algorithm defined by RFC 7693
+// and the extendable output function (XOF) BLAKE2Xb.
+//
+// BLAKE2b is optimized for 64-bit platforms—including NEON-enabled ARMs—and
+// produces digests of any size between 1 and 64 bytes.
+// For a detailed specification of BLAKE2b see https://blake2.net/blake2.pdf
+// and for BLAKE2Xb see https://blake2.net/blake2x.pdf
+//
+// If you aren't sure which function you need, use BLAKE2b (Sum512 or New512).
+// If you need a secret-key MAC (message authentication code), use the New512
+// function with a non-nil key.
+//
+// BLAKE2X is a construction to compute hash values larger than 64 bytes. It
+// can produce hash values between 0 and 4 GiB.
+package blake2b
+
+import (
+	"encoding/binary"
+	"errors"
+	"hash"
+)
+
+const (
+	// The blocksize of BLAKE2b in bytes.
+	BlockSize = 128
+	// The hash size of BLAKE2b-512 in bytes.
+	Size = 64
+	// The hash size of BLAKE2b-384 in bytes.
+	Size384 = 48
+	// The hash size of BLAKE2b-256 in bytes.
+	Size256 = 32
+)
+
+var (
+	useAVX2 bool
+	useAVX  bool
+	useSSE4 bool
+)
+
+var (
+	errKeySize  = errors.New("blake2b: invalid key size")
+	errHashSize = errors.New("blake2b: invalid hash size")
+)
+
+var iv = [8]uint64{
+	0x6a09e667f3bcc908, 0xbb67ae8584caa73b, 0x3c6ef372fe94f82b, 0xa54ff53a5f1d36f1,
+	0x510e527fade682d1, 0x9b05688c2b3e6c1f, 0x1f83d9abfb41bd6b, 0x5be0cd19137e2179,
+}
+
+// Sum512 returns the BLAKE2b-512 checksum of the data.
+func Sum512(data []byte) [Size]byte {
+	var sum [Size]byte
+	checkSum(&sum, Size, data)
+	return sum
+}
+
+// Sum384 returns the BLAKE2b-384 checksum of the data.
+func Sum384(data []byte) [Size384]byte {
+	var sum [Size]byte
+	var sum384 [Size384]byte
+	checkSum(&sum, Size384, data)
+	copy(sum384[:], sum[:Size384])
+	return sum384
+}
+
+// Sum256 returns the BLAKE2b-256 checksum of the data.
+func Sum256(data []byte) [Size256]byte {
+	var sum [Size]byte
+	var sum256 [Size256]byte
+	checkSum(&sum, Size256, data)
+	copy(sum256[:], sum[:Size256])
+	return sum256
+}
+
+// New512 returns a new hash.Hash computing the BLAKE2b-512 checksum. A non-nil
+// key turns the hash into a MAC. The key must be between zero and 64 bytes long.
+func New512(key []byte) (hash.Hash, error) { return newDigest(Size, key) }
+
+// New384 returns a new hash.Hash computing the BLAKE2b-384 checksum. A non-nil
+// key turns the hash into a MAC. The key must be between zero and 64 bytes long.
+func New384(key []byte) (hash.Hash, error) { return newDigest(Size384, key) }
+
+// New256 returns a new hash.Hash computing the BLAKE2b-256 checksum. A non-nil
+// key turns the hash into a MAC. The key must be between zero and 64 bytes long.
+func New256(key []byte) (hash.Hash, error) { return newDigest(Size256, key) }
+
+// New returns a new hash.Hash computing the BLAKE2b checksum with a custom length.
+// A non-nil key turns the hash into a MAC. The key must be between zero and 64 bytes long.
+// The hash size can be a value between 1 and 64 but it is highly recommended to use
+// values equal or greater than:
+// - 32 if BLAKE2b is used as a hash function (The key is zero bytes long).
+// - 16 if BLAKE2b is used as a MAC function (The key is at least 16 bytes long).
+// When the key is nil, the returned hash.Hash implements BinaryMarshaler
+// and BinaryUnmarshaler for state (de)serialization as documented by hash.Hash.
+func New(size int, key []byte) (hash.Hash, error) { return newDigest(size, key) }
+
+func newDigest(hashSize int, key []byte) (*digest, error) {
+	if hashSize < 1 || hashSize > Size {
+		return nil, errHashSize
+	}
+	if len(key) > Size {
+		return nil, errKeySize
+	}
+	d := &digest{
+		size:   hashSize,
+		keyLen: len(key),
+	}
+	copy(d.key[:], key)
+	d.Reset()
+	return d, nil
+}
+
+func checkSum(sum *[Size]byte, hashSize int, data []byte) {
+	h := iv
+	h[0] ^= uint64(hashSize) | (1 << 16) | (1 << 24)
+	var c [2]uint64
+
+	if length := len(data); length > BlockSize {
+		n := length &^ (BlockSize - 1)
+		if length == n {
+			n -= BlockSize
+		}
+		hashBlocks(&h, &c, 0, data[:n])
+		data = data[n:]
+	}
+
+	var block [BlockSize]byte
+	offset := copy(block[:], data)
+	remaining := uint64(BlockSize - offset)
+	if c[0] < remaining {
+		c[1]--
+	}
+	c[0] -= remaining
+
+	hashBlocks(&h, &c, 0xFFFFFFFFFFFFFFFF, block[:])
+
+	for i, v := range h[:(hashSize+7)/8] {
+		binary.LittleEndian.PutUint64(sum[8*i:], v)
+	}
+}
+
+type digest struct {
+	h      [8]uint64
+	c      [2]uint64
+	size   int
+	block  [BlockSize]byte
+	offset int
+
+	key    [BlockSize]byte
+	keyLen int
+}
+
+const (
+	magic         = "b2b"
+	marshaledSize = len(magic) + 8*8 + 2*8 + 1 + BlockSize + 1
+)
+
+func (d *digest) MarshalBinary() ([]byte, error) {
+	if d.keyLen != 0 {
+		return nil, errors.New("crypto/blake2b: cannot marshal MACs")
+	}
+	b := make([]byte, 0, marshaledSize)
+	b = append(b, magic...)
+	for i := 0; i < 8; i++ {
+		b = appendUint64(b, d.h[i])
+	}
+	b = appendUint64(b, d.c[0])
+	b = appendUint64(b, d.c[1])
+	// Maximum value for size is 64
+	b = append(b, byte(d.size))
+	b = append(b, d.block[:]...)
+	b = append(b, byte(d.offset))
+	return b, nil
+}
+
+func (d *digest) UnmarshalBinary(b []byte) error {
+	if len(b) < len(magic) || string(b[:len(magic)]) != magic {
+		return errors.New("crypto/blake2b: invalid hash state identifier")
+	}
+	if len(b) != marshaledSize {
+		return errors.New("crypto/blake2b: invalid hash state size")
+	}
+	b = b[len(magic):]
+	for i := 0; i < 8; i++ {
+		b, d.h[i] = consumeUint64(b)
+	}
+	b, d.c[0] = consumeUint64(b)
+	b, d.c[1] = consumeUint64(b)
+	d.size = int(b[0])
+	b = b[1:]
+	copy(d.block[:], b[:BlockSize])
+	b = b[BlockSize:]
+	d.offset = int(b[0])
+	return nil
+}
+
+func (d *digest) BlockSize() int { return BlockSize }
+
+func (d *digest) Size() int { return d.size }
+
+func (d *digest) Reset() {
+	d.h = iv
+	d.h[0] ^= uint64(d.size) | (uint64(d.keyLen) << 8) | (1 << 16) | (1 << 24)
+	d.offset, d.c[0], d.c[1] = 0, 0, 0
+	if d.keyLen > 0 {
+		d.block = d.key
+		d.offset = BlockSize
+	}
+}
+
+func (d *digest) Write(p []byte) (n int, err error) {
+	n = len(p)
+
+	if d.offset > 0 {
+		remaining := BlockSize - d.offset
+		if n <= remaining {
+			d.offset += copy(d.block[d.offset:], p)
+			return
+		}
+		copy(d.block[d.offset:], p[:remaining])
+		hashBlocks(&d.h, &d.c, 0, d.block[:])
+		d.offset = 0
+		p = p[remaining:]
+	}
+
+	if length := len(p); length > BlockSize {
+		nn := length &^ (BlockSize - 1)
+		if length == nn {
+			nn -= BlockSize
+		}
+		hashBlocks(&d.h, &d.c, 0, p[:nn])
+		p = p[nn:]
+	}
+
+	if len(p) > 0 {
+		d.offset += copy(d.block[:], p)
+	}
+
+	return
+}
+
+func (d *digest) Sum(sum []byte) []byte {
+	var hash [Size]byte
+	d.finalize(&hash)
+	return append(sum, hash[:d.size]...)
+}
+
+func (d *digest) finalize(hash *[Size]byte) {
+	var block [BlockSize]byte
+	copy(block[:], d.block[:d.offset])
+	remaining := uint64(BlockSize - d.offset)
+
+	c := d.c
+	if c[0] < remaining {
+		c[1]--
+	}
+	c[0] -= remaining
+
+	h := d.h
+	hashBlocks(&h, &c, 0xFFFFFFFFFFFFFFFF, block[:])
+
+	for i, v := range h {
+		binary.LittleEndian.PutUint64(hash[8*i:], v)
+	}
+}
+
+func appendUint64(b []byte, x uint64) []byte {
+	var a [8]byte
+	binary.BigEndian.PutUint64(a[:], x)
+	return append(b, a[:]...)
+}
+
+func appendUint32(b []byte, x uint32) []byte {
+	var a [4]byte
+	binary.BigEndian.PutUint32(a[:], x)
+	return append(b, a[:]...)
+}
+
+func consumeUint64(b []byte) ([]byte, uint64) {
+	x := binary.BigEndian.Uint64(b)
+	return b[8:], x
+}
+
+func consumeUint32(b []byte) ([]byte, uint32) {
+	x := binary.BigEndian.Uint32(b)
+	return b[4:], x
+}
diff --git a/vendor/golang.org/x/crypto/blake2b/blake2bAVX2_amd64.go b/vendor/golang.org/x/crypto/blake2b/blake2bAVX2_amd64.go
new file mode 100644
index 0000000..199c21d
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2bAVX2_amd64.go
@@ -0,0 +1,37 @@
+// 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.
+
+//go:build amd64 && gc && !purego
+
+package blake2b
+
+import "golang.org/x/sys/cpu"
+
+func init() {
+	useAVX2 = cpu.X86.HasAVX2
+	useAVX = cpu.X86.HasAVX
+	useSSE4 = cpu.X86.HasSSE41
+}
+
+//go:noescape
+func hashBlocksAVX2(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte)
+
+//go:noescape
+func hashBlocksAVX(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte)
+
+//go:noescape
+func hashBlocksSSE4(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte)
+
+func hashBlocks(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte) {
+	switch {
+	case useAVX2:
+		hashBlocksAVX2(h, c, flag, blocks)
+	case useAVX:
+		hashBlocksAVX(h, c, flag, blocks)
+	case useSSE4:
+		hashBlocksSSE4(h, c, flag, blocks)
+	default:
+		hashBlocksGeneric(h, c, flag, blocks)
+	}
+}
diff --git a/vendor/golang.org/x/crypto/blake2b/blake2bAVX2_amd64.s b/vendor/golang.org/x/crypto/blake2b/blake2bAVX2_amd64.s
new file mode 100644
index 0000000..9ae8206
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2bAVX2_amd64.s
@@ -0,0 +1,744 @@
+// 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.
+
+//go:build amd64 && gc && !purego
+
+#include "textflag.h"
+
+DATA ·AVX2_iv0<>+0x00(SB)/8, $0x6a09e667f3bcc908
+DATA ·AVX2_iv0<>+0x08(SB)/8, $0xbb67ae8584caa73b
+DATA ·AVX2_iv0<>+0x10(SB)/8, $0x3c6ef372fe94f82b
+DATA ·AVX2_iv0<>+0x18(SB)/8, $0xa54ff53a5f1d36f1
+GLOBL ·AVX2_iv0<>(SB), (NOPTR+RODATA), $32
+
+DATA ·AVX2_iv1<>+0x00(SB)/8, $0x510e527fade682d1
+DATA ·AVX2_iv1<>+0x08(SB)/8, $0x9b05688c2b3e6c1f
+DATA ·AVX2_iv1<>+0x10(SB)/8, $0x1f83d9abfb41bd6b
+DATA ·AVX2_iv1<>+0x18(SB)/8, $0x5be0cd19137e2179
+GLOBL ·AVX2_iv1<>(SB), (NOPTR+RODATA), $32
+
+DATA ·AVX2_c40<>+0x00(SB)/8, $0x0201000706050403
+DATA ·AVX2_c40<>+0x08(SB)/8, $0x0a09080f0e0d0c0b
+DATA ·AVX2_c40<>+0x10(SB)/8, $0x0201000706050403
+DATA ·AVX2_c40<>+0x18(SB)/8, $0x0a09080f0e0d0c0b
+GLOBL ·AVX2_c40<>(SB), (NOPTR+RODATA), $32
+
+DATA ·AVX2_c48<>+0x00(SB)/8, $0x0100070605040302
+DATA ·AVX2_c48<>+0x08(SB)/8, $0x09080f0e0d0c0b0a
+DATA ·AVX2_c48<>+0x10(SB)/8, $0x0100070605040302
+DATA ·AVX2_c48<>+0x18(SB)/8, $0x09080f0e0d0c0b0a
+GLOBL ·AVX2_c48<>(SB), (NOPTR+RODATA), $32
+
+DATA ·AVX_iv0<>+0x00(SB)/8, $0x6a09e667f3bcc908
+DATA ·AVX_iv0<>+0x08(SB)/8, $0xbb67ae8584caa73b
+GLOBL ·AVX_iv0<>(SB), (NOPTR+RODATA), $16
+
+DATA ·AVX_iv1<>+0x00(SB)/8, $0x3c6ef372fe94f82b
+DATA ·AVX_iv1<>+0x08(SB)/8, $0xa54ff53a5f1d36f1
+GLOBL ·AVX_iv1<>(SB), (NOPTR+RODATA), $16
+
+DATA ·AVX_iv2<>+0x00(SB)/8, $0x510e527fade682d1
+DATA ·AVX_iv2<>+0x08(SB)/8, $0x9b05688c2b3e6c1f
+GLOBL ·AVX_iv2<>(SB), (NOPTR+RODATA), $16
+
+DATA ·AVX_iv3<>+0x00(SB)/8, $0x1f83d9abfb41bd6b
+DATA ·AVX_iv3<>+0x08(SB)/8, $0x5be0cd19137e2179
+GLOBL ·AVX_iv3<>(SB), (NOPTR+RODATA), $16
+
+DATA ·AVX_c40<>+0x00(SB)/8, $0x0201000706050403
+DATA ·AVX_c40<>+0x08(SB)/8, $0x0a09080f0e0d0c0b
+GLOBL ·AVX_c40<>(SB), (NOPTR+RODATA), $16
+
+DATA ·AVX_c48<>+0x00(SB)/8, $0x0100070605040302
+DATA ·AVX_c48<>+0x08(SB)/8, $0x09080f0e0d0c0b0a
+GLOBL ·AVX_c48<>(SB), (NOPTR+RODATA), $16
+
+#define VPERMQ_0x39_Y1_Y1 BYTE $0xc4; BYTE $0xe3; BYTE $0xfd; BYTE $0x00; BYTE $0xc9; BYTE $0x39
+#define VPERMQ_0x93_Y1_Y1 BYTE $0xc4; BYTE $0xe3; BYTE $0xfd; BYTE $0x00; BYTE $0xc9; BYTE $0x93
+#define VPERMQ_0x4E_Y2_Y2 BYTE $0xc4; BYTE $0xe3; BYTE $0xfd; BYTE $0x00; BYTE $0xd2; BYTE $0x4e
+#define VPERMQ_0x93_Y3_Y3 BYTE $0xc4; BYTE $0xe3; BYTE $0xfd; BYTE $0x00; BYTE $0xdb; BYTE $0x93
+#define VPERMQ_0x39_Y3_Y3 BYTE $0xc4; BYTE $0xe3; BYTE $0xfd; BYTE $0x00; BYTE $0xdb; BYTE $0x39
+
+#define ROUND_AVX2(m0, m1, m2, m3, t, c40, c48) \
+	VPADDQ  m0, Y0, Y0;   \
+	VPADDQ  Y1, Y0, Y0;   \
+	VPXOR   Y0, Y3, Y3;   \
+	VPSHUFD $-79, Y3, Y3; \
+	VPADDQ  Y3, Y2, Y2;   \
+	VPXOR   Y2, Y1, Y1;   \
+	VPSHUFB c40, Y1, Y1;  \
+	VPADDQ  m1, Y0, Y0;   \
+	VPADDQ  Y1, Y0, Y0;   \
+	VPXOR   Y0, Y3, Y3;   \
+	VPSHUFB c48, Y3, Y3;  \
+	VPADDQ  Y3, Y2, Y2;   \
+	VPXOR   Y2, Y1, Y1;   \
+	VPADDQ  Y1, Y1, t;    \
+	VPSRLQ  $63, Y1, Y1;  \
+	VPXOR   t, Y1, Y1;    \
+	VPERMQ_0x39_Y1_Y1;    \
+	VPERMQ_0x4E_Y2_Y2;    \
+	VPERMQ_0x93_Y3_Y3;    \
+	VPADDQ  m2, Y0, Y0;   \
+	VPADDQ  Y1, Y0, Y0;   \
+	VPXOR   Y0, Y3, Y3;   \
+	VPSHUFD $-79, Y3, Y3; \
+	VPADDQ  Y3, Y2, Y2;   \
+	VPXOR   Y2, Y1, Y1;   \
+	VPSHUFB c40, Y1, Y1;  \
+	VPADDQ  m3, Y0, Y0;   \
+	VPADDQ  Y1, Y0, Y0;   \
+	VPXOR   Y0, Y3, Y3;   \
+	VPSHUFB c48, Y3, Y3;  \
+	VPADDQ  Y3, Y2, Y2;   \
+	VPXOR   Y2, Y1, Y1;   \
+	VPADDQ  Y1, Y1, t;    \
+	VPSRLQ  $63, Y1, Y1;  \
+	VPXOR   t, Y1, Y1;    \
+	VPERMQ_0x39_Y3_Y3;    \
+	VPERMQ_0x4E_Y2_Y2;    \
+	VPERMQ_0x93_Y1_Y1
+
+#define VMOVQ_SI_X11_0 BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x1E
+#define VMOVQ_SI_X12_0 BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x26
+#define VMOVQ_SI_X13_0 BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x2E
+#define VMOVQ_SI_X14_0 BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x36
+#define VMOVQ_SI_X15_0 BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x3E
+
+#define VMOVQ_SI_X11(n) BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x5E; BYTE $n
+#define VMOVQ_SI_X12(n) BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x66; BYTE $n
+#define VMOVQ_SI_X13(n) BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x6E; BYTE $n
+#define VMOVQ_SI_X14(n) BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x76; BYTE $n
+#define VMOVQ_SI_X15(n) BYTE $0xC5; BYTE $0x7A; BYTE $0x7E; BYTE $0x7E; BYTE $n
+
+#define VPINSRQ_1_SI_X11_0 BYTE $0xC4; BYTE $0x63; BYTE $0xA1; BYTE $0x22; BYTE $0x1E; BYTE $0x01
+#define VPINSRQ_1_SI_X12_0 BYTE $0xC4; BYTE $0x63; BYTE $0x99; BYTE $0x22; BYTE $0x26; BYTE $0x01
+#define VPINSRQ_1_SI_X13_0 BYTE $0xC4; BYTE $0x63; BYTE $0x91; BYTE $0x22; BYTE $0x2E; BYTE $0x01
+#define VPINSRQ_1_SI_X14_0 BYTE $0xC4; BYTE $0x63; BYTE $0x89; BYTE $0x22; BYTE $0x36; BYTE $0x01
+#define VPINSRQ_1_SI_X15_0 BYTE $0xC4; BYTE $0x63; BYTE $0x81; BYTE $0x22; BYTE $0x3E; BYTE $0x01
+
+#define VPINSRQ_1_SI_X11(n) BYTE $0xC4; BYTE $0x63; BYTE $0xA1; BYTE $0x22; BYTE $0x5E; BYTE $n; BYTE $0x01
+#define VPINSRQ_1_SI_X12(n) BYTE $0xC4; BYTE $0x63; BYTE $0x99; BYTE $0x22; BYTE $0x66; BYTE $n; BYTE $0x01
+#define VPINSRQ_1_SI_X13(n) BYTE $0xC4; BYTE $0x63; BYTE $0x91; BYTE $0x22; BYTE $0x6E; BYTE $n; BYTE $0x01
+#define VPINSRQ_1_SI_X14(n) BYTE $0xC4; BYTE $0x63; BYTE $0x89; BYTE $0x22; BYTE $0x76; BYTE $n; BYTE $0x01
+#define VPINSRQ_1_SI_X15(n) BYTE $0xC4; BYTE $0x63; BYTE $0x81; BYTE $0x22; BYTE $0x7E; BYTE $n; BYTE $0x01
+
+#define VMOVQ_R8_X15 BYTE $0xC4; BYTE $0x41; BYTE $0xF9; BYTE $0x6E; BYTE $0xF8
+#define VPINSRQ_1_R9_X15 BYTE $0xC4; BYTE $0x43; BYTE $0x81; BYTE $0x22; BYTE $0xF9; BYTE $0x01
+
+// load msg: Y12 = (i0, i1, i2, i3)
+// i0, i1, i2, i3 must not be 0
+#define LOAD_MSG_AVX2_Y12(i0, i1, i2, i3) \
+	VMOVQ_SI_X12(i0*8);           \
+	VMOVQ_SI_X11(i2*8);           \
+	VPINSRQ_1_SI_X12(i1*8);       \
+	VPINSRQ_1_SI_X11(i3*8);       \
+	VINSERTI128 $1, X11, Y12, Y12
+
+// load msg: Y13 = (i0, i1, i2, i3)
+// i0, i1, i2, i3 must not be 0
+#define LOAD_MSG_AVX2_Y13(i0, i1, i2, i3) \
+	VMOVQ_SI_X13(i0*8);           \
+	VMOVQ_SI_X11(i2*8);           \
+	VPINSRQ_1_SI_X13(i1*8);       \
+	VPINSRQ_1_SI_X11(i3*8);       \
+	VINSERTI128 $1, X11, Y13, Y13
+
+// load msg: Y14 = (i0, i1, i2, i3)
+// i0, i1, i2, i3 must not be 0
+#define LOAD_MSG_AVX2_Y14(i0, i1, i2, i3) \
+	VMOVQ_SI_X14(i0*8);           \
+	VMOVQ_SI_X11(i2*8);           \
+	VPINSRQ_1_SI_X14(i1*8);       \
+	VPINSRQ_1_SI_X11(i3*8);       \
+	VINSERTI128 $1, X11, Y14, Y14
+
+// load msg: Y15 = (i0, i1, i2, i3)
+// i0, i1, i2, i3 must not be 0
+#define LOAD_MSG_AVX2_Y15(i0, i1, i2, i3) \
+	VMOVQ_SI_X15(i0*8);           \
+	VMOVQ_SI_X11(i2*8);           \
+	VPINSRQ_1_SI_X15(i1*8);       \
+	VPINSRQ_1_SI_X11(i3*8);       \
+	VINSERTI128 $1, X11, Y15, Y15
+
+#define LOAD_MSG_AVX2_0_2_4_6_1_3_5_7_8_10_12_14_9_11_13_15() \
+	VMOVQ_SI_X12_0;                   \
+	VMOVQ_SI_X11(4*8);                \
+	VPINSRQ_1_SI_X12(2*8);            \
+	VPINSRQ_1_SI_X11(6*8);            \
+	VINSERTI128 $1, X11, Y12, Y12;    \
+	LOAD_MSG_AVX2_Y13(1, 3, 5, 7);    \
+	LOAD_MSG_AVX2_Y14(8, 10, 12, 14); \
+	LOAD_MSG_AVX2_Y15(9, 11, 13, 15)
+
+#define LOAD_MSG_AVX2_14_4_9_13_10_8_15_6_1_0_11_5_12_2_7_3() \
+	LOAD_MSG_AVX2_Y12(14, 4, 9, 13); \
+	LOAD_MSG_AVX2_Y13(10, 8, 15, 6); \
+	VMOVQ_SI_X11(11*8);              \
+	VPSHUFD     $0x4E, 0*8(SI), X14; \
+	VPINSRQ_1_SI_X11(5*8);           \
+	VINSERTI128 $1, X11, Y14, Y14;   \
+	LOAD_MSG_AVX2_Y15(12, 2, 7, 3)
+
+#define LOAD_MSG_AVX2_11_12_5_15_8_0_2_13_10_3_7_9_14_6_1_4() \
+	VMOVQ_SI_X11(5*8);              \
+	VMOVDQU     11*8(SI), X12;      \
+	VPINSRQ_1_SI_X11(15*8);         \
+	VINSERTI128 $1, X11, Y12, Y12;  \
+	VMOVQ_SI_X13(8*8);              \
+	VMOVQ_SI_X11(2*8);              \
+	VPINSRQ_1_SI_X13_0;             \
+	VPINSRQ_1_SI_X11(13*8);         \
+	VINSERTI128 $1, X11, Y13, Y13;  \
+	LOAD_MSG_AVX2_Y14(10, 3, 7, 9); \
+	LOAD_MSG_AVX2_Y15(14, 6, 1, 4)
+
+#define LOAD_MSG_AVX2_7_3_13_11_9_1_12_14_2_5_4_15_6_10_0_8() \
+	LOAD_MSG_AVX2_Y12(7, 3, 13, 11); \
+	LOAD_MSG_AVX2_Y13(9, 1, 12, 14); \
+	LOAD_MSG_AVX2_Y14(2, 5, 4, 15);  \
+	VMOVQ_SI_X15(6*8);               \
+	VMOVQ_SI_X11_0;                  \
+	VPINSRQ_1_SI_X15(10*8);          \
+	VPINSRQ_1_SI_X11(8*8);           \
+	VINSERTI128 $1, X11, Y15, Y15
+
+#define LOAD_MSG_AVX2_9_5_2_10_0_7_4_15_14_11_6_3_1_12_8_13() \
+	LOAD_MSG_AVX2_Y12(9, 5, 2, 10);  \
+	VMOVQ_SI_X13_0;                  \
+	VMOVQ_SI_X11(4*8);               \
+	VPINSRQ_1_SI_X13(7*8);           \
+	VPINSRQ_1_SI_X11(15*8);          \
+	VINSERTI128 $1, X11, Y13, Y13;   \
+	LOAD_MSG_AVX2_Y14(14, 11, 6, 3); \
+	LOAD_MSG_AVX2_Y15(1, 12, 8, 13)
+
+#define LOAD_MSG_AVX2_2_6_0_8_12_10_11_3_4_7_15_1_13_5_14_9() \
+	VMOVQ_SI_X12(2*8);                \
+	VMOVQ_SI_X11_0;                   \
+	VPINSRQ_1_SI_X12(6*8);            \
+	VPINSRQ_1_SI_X11(8*8);            \
+	VINSERTI128 $1, X11, Y12, Y12;    \
+	LOAD_MSG_AVX2_Y13(12, 10, 11, 3); \
+	LOAD_MSG_AVX2_Y14(4, 7, 15, 1);   \
+	LOAD_MSG_AVX2_Y15(13, 5, 14, 9)
+
+#define LOAD_MSG_AVX2_12_1_14_4_5_15_13_10_0_6_9_8_7_3_2_11() \
+	LOAD_MSG_AVX2_Y12(12, 1, 14, 4);  \
+	LOAD_MSG_AVX2_Y13(5, 15, 13, 10); \
+	VMOVQ_SI_X14_0;                   \
+	VPSHUFD     $0x4E, 8*8(SI), X11;  \
+	VPINSRQ_1_SI_X14(6*8);            \
+	VINSERTI128 $1, X11, Y14, Y14;    \
+	LOAD_MSG_AVX2_Y15(7, 3, 2, 11)
+
+#define LOAD_MSG_AVX2_13_7_12_3_11_14_1_9_5_15_8_2_0_4_6_10() \
+	LOAD_MSG_AVX2_Y12(13, 7, 12, 3); \
+	LOAD_MSG_AVX2_Y13(11, 14, 1, 9); \
+	LOAD_MSG_AVX2_Y14(5, 15, 8, 2);  \
+	VMOVQ_SI_X15_0;                  \
+	VMOVQ_SI_X11(6*8);               \
+	VPINSRQ_1_SI_X15(4*8);           \
+	VPINSRQ_1_SI_X11(10*8);          \
+	VINSERTI128 $1, X11, Y15, Y15
+
+#define LOAD_MSG_AVX2_6_14_11_0_15_9_3_8_12_13_1_10_2_7_4_5() \
+	VMOVQ_SI_X12(6*8);              \
+	VMOVQ_SI_X11(11*8);             \
+	VPINSRQ_1_SI_X12(14*8);         \
+	VPINSRQ_1_SI_X11_0;             \
+	VINSERTI128 $1, X11, Y12, Y12;  \
+	LOAD_MSG_AVX2_Y13(15, 9, 3, 8); \
+	VMOVQ_SI_X11(1*8);              \
+	VMOVDQU     12*8(SI), X14;      \
+	VPINSRQ_1_SI_X11(10*8);         \
+	VINSERTI128 $1, X11, Y14, Y14;  \
+	VMOVQ_SI_X15(2*8);              \
+	VMOVDQU     4*8(SI), X11;       \
+	VPINSRQ_1_SI_X15(7*8);          \
+	VINSERTI128 $1, X11, Y15, Y15
+
+#define LOAD_MSG_AVX2_10_8_7_1_2_4_6_5_15_9_3_13_11_14_12_0() \
+	LOAD_MSG_AVX2_Y12(10, 8, 7, 1);  \
+	VMOVQ_SI_X13(2*8);               \
+	VPSHUFD     $0x4E, 5*8(SI), X11; \
+	VPINSRQ_1_SI_X13(4*8);           \
+	VINSERTI128 $1, X11, Y13, Y13;   \
+	LOAD_MSG_AVX2_Y14(15, 9, 3, 13); \
+	VMOVQ_SI_X15(11*8);              \
+	VMOVQ_SI_X11(12*8);              \
+	VPINSRQ_1_SI_X15(14*8);          \
+	VPINSRQ_1_SI_X11_0;              \
+	VINSERTI128 $1, X11, Y15, Y15
+
+// func hashBlocksAVX2(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte)
+TEXT ·hashBlocksAVX2(SB), 4, $320-48 // frame size = 288 + 32 byte alignment
+	MOVQ h+0(FP), AX
+	MOVQ c+8(FP), BX
+	MOVQ flag+16(FP), CX
+	MOVQ blocks_base+24(FP), SI
+	MOVQ blocks_len+32(FP), DI
+
+	MOVQ SP, DX
+	ADDQ $31, DX
+	ANDQ $~31, DX
+
+	MOVQ CX, 16(DX)
+	XORQ CX, CX
+	MOVQ CX, 24(DX)
+
+	VMOVDQU ·AVX2_c40<>(SB), Y4
+	VMOVDQU ·AVX2_c48<>(SB), Y5
+
+	VMOVDQU 0(AX), Y8
+	VMOVDQU 32(AX), Y9
+	VMOVDQU ·AVX2_iv0<>(SB), Y6
+	VMOVDQU ·AVX2_iv1<>(SB), Y7
+
+	MOVQ 0(BX), R8
+	MOVQ 8(BX), R9
+	MOVQ R9, 8(DX)
+
+loop:
+	ADDQ $128, R8
+	MOVQ R8, 0(DX)
+	CMPQ R8, $128
+	JGE  noinc
+	INCQ R9
+	MOVQ R9, 8(DX)
+
+noinc:
+	VMOVDQA Y8, Y0
+	VMOVDQA Y9, Y1
+	VMOVDQA Y6, Y2
+	VPXOR   0(DX), Y7, Y3
+
+	LOAD_MSG_AVX2_0_2_4_6_1_3_5_7_8_10_12_14_9_11_13_15()
+	VMOVDQA Y12, 32(DX)
+	VMOVDQA Y13, 64(DX)
+	VMOVDQA Y14, 96(DX)
+	VMOVDQA Y15, 128(DX)
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_14_4_9_13_10_8_15_6_1_0_11_5_12_2_7_3()
+	VMOVDQA Y12, 160(DX)
+	VMOVDQA Y13, 192(DX)
+	VMOVDQA Y14, 224(DX)
+	VMOVDQA Y15, 256(DX)
+
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_11_12_5_15_8_0_2_13_10_3_7_9_14_6_1_4()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_7_3_13_11_9_1_12_14_2_5_4_15_6_10_0_8()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_9_5_2_10_0_7_4_15_14_11_6_3_1_12_8_13()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_2_6_0_8_12_10_11_3_4_7_15_1_13_5_14_9()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_12_1_14_4_5_15_13_10_0_6_9_8_7_3_2_11()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_13_7_12_3_11_14_1_9_5_15_8_2_0_4_6_10()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_6_14_11_0_15_9_3_8_12_13_1_10_2_7_4_5()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+	LOAD_MSG_AVX2_10_8_7_1_2_4_6_5_15_9_3_13_11_14_12_0()
+	ROUND_AVX2(Y12, Y13, Y14, Y15, Y10, Y4, Y5)
+
+	ROUND_AVX2(32(DX), 64(DX), 96(DX), 128(DX), Y10, Y4, Y5)
+	ROUND_AVX2(160(DX), 192(DX), 224(DX), 256(DX), Y10, Y4, Y5)
+
+	VPXOR Y0, Y8, Y8
+	VPXOR Y1, Y9, Y9
+	VPXOR Y2, Y8, Y8
+	VPXOR Y3, Y9, Y9
+
+	LEAQ 128(SI), SI
+	SUBQ $128, DI
+	JNE  loop
+
+	MOVQ R8, 0(BX)
+	MOVQ R9, 8(BX)
+
+	VMOVDQU Y8, 0(AX)
+	VMOVDQU Y9, 32(AX)
+	VZEROUPPER
+
+	RET
+
+#define VPUNPCKLQDQ_X2_X2_X15 BYTE $0xC5; BYTE $0x69; BYTE $0x6C; BYTE $0xFA
+#define VPUNPCKLQDQ_X3_X3_X15 BYTE $0xC5; BYTE $0x61; BYTE $0x6C; BYTE $0xFB
+#define VPUNPCKLQDQ_X7_X7_X15 BYTE $0xC5; BYTE $0x41; BYTE $0x6C; BYTE $0xFF
+#define VPUNPCKLQDQ_X13_X13_X15 BYTE $0xC4; BYTE $0x41; BYTE $0x11; BYTE $0x6C; BYTE $0xFD
+#define VPUNPCKLQDQ_X14_X14_X15 BYTE $0xC4; BYTE $0x41; BYTE $0x09; BYTE $0x6C; BYTE $0xFE
+
+#define VPUNPCKHQDQ_X15_X2_X2 BYTE $0xC4; BYTE $0xC1; BYTE $0x69; BYTE $0x6D; BYTE $0xD7
+#define VPUNPCKHQDQ_X15_X3_X3 BYTE $0xC4; BYTE $0xC1; BYTE $0x61; BYTE $0x6D; BYTE $0xDF
+#define VPUNPCKHQDQ_X15_X6_X6 BYTE $0xC4; BYTE $0xC1; BYTE $0x49; BYTE $0x6D; BYTE $0xF7
+#define VPUNPCKHQDQ_X15_X7_X7 BYTE $0xC4; BYTE $0xC1; BYTE $0x41; BYTE $0x6D; BYTE $0xFF
+#define VPUNPCKHQDQ_X15_X3_X2 BYTE $0xC4; BYTE $0xC1; BYTE $0x61; BYTE $0x6D; BYTE $0xD7
+#define VPUNPCKHQDQ_X15_X7_X6 BYTE $0xC4; BYTE $0xC1; BYTE $0x41; BYTE $0x6D; BYTE $0xF7
+#define VPUNPCKHQDQ_X15_X13_X3 BYTE $0xC4; BYTE $0xC1; BYTE $0x11; BYTE $0x6D; BYTE $0xDF
+#define VPUNPCKHQDQ_X15_X13_X7 BYTE $0xC4; BYTE $0xC1; BYTE $0x11; BYTE $0x6D; BYTE $0xFF
+
+#define SHUFFLE_AVX() \
+	VMOVDQA X6, X13;         \
+	VMOVDQA X2, X14;         \
+	VMOVDQA X4, X6;          \
+	VPUNPCKLQDQ_X13_X13_X15; \
+	VMOVDQA X5, X4;          \
+	VMOVDQA X6, X5;          \
+	VPUNPCKHQDQ_X15_X7_X6;   \
+	VPUNPCKLQDQ_X7_X7_X15;   \
+	VPUNPCKHQDQ_X15_X13_X7;  \
+	VPUNPCKLQDQ_X3_X3_X15;   \
+	VPUNPCKHQDQ_X15_X2_X2;   \
+	VPUNPCKLQDQ_X14_X14_X15; \
+	VPUNPCKHQDQ_X15_X3_X3;   \
+
+#define SHUFFLE_AVX_INV() \
+	VMOVDQA X2, X13;         \
+	VMOVDQA X4, X14;         \
+	VPUNPCKLQDQ_X2_X2_X15;   \
+	VMOVDQA X5, X4;          \
+	VPUNPCKHQDQ_X15_X3_X2;   \
+	VMOVDQA X14, X5;         \
+	VPUNPCKLQDQ_X3_X3_X15;   \
+	VMOVDQA X6, X14;         \
+	VPUNPCKHQDQ_X15_X13_X3;  \
+	VPUNPCKLQDQ_X7_X7_X15;   \
+	VPUNPCKHQDQ_X15_X6_X6;   \
+	VPUNPCKLQDQ_X14_X14_X15; \
+	VPUNPCKHQDQ_X15_X7_X7;   \
+
+#define HALF_ROUND_AVX(v0, v1, v2, v3, v4, v5, v6, v7, m0, m1, m2, m3, t0, c40, c48) \
+	VPADDQ  m0, v0, v0;   \
+	VPADDQ  v2, v0, v0;   \
+	VPADDQ  m1, v1, v1;   \
+	VPADDQ  v3, v1, v1;   \
+	VPXOR   v0, v6, v6;   \
+	VPXOR   v1, v7, v7;   \
+	VPSHUFD $-79, v6, v6; \
+	VPSHUFD $-79, v7, v7; \
+	VPADDQ  v6, v4, v4;   \
+	VPADDQ  v7, v5, v5;   \
+	VPXOR   v4, v2, v2;   \
+	VPXOR   v5, v3, v3;   \
+	VPSHUFB c40, v2, v2;  \
+	VPSHUFB c40, v3, v3;  \
+	VPADDQ  m2, v0, v0;   \
+	VPADDQ  v2, v0, v0;   \
+	VPADDQ  m3, v1, v1;   \
+	VPADDQ  v3, v1, v1;   \
+	VPXOR   v0, v6, v6;   \
+	VPXOR   v1, v7, v7;   \
+	VPSHUFB c48, v6, v6;  \
+	VPSHUFB c48, v7, v7;  \
+	VPADDQ  v6, v4, v4;   \
+	VPADDQ  v7, v5, v5;   \
+	VPXOR   v4, v2, v2;   \
+	VPXOR   v5, v3, v3;   \
+	VPADDQ  v2, v2, t0;   \
+	VPSRLQ  $63, v2, v2;  \
+	VPXOR   t0, v2, v2;   \
+	VPADDQ  v3, v3, t0;   \
+	VPSRLQ  $63, v3, v3;  \
+	VPXOR   t0, v3, v3
+
+// load msg: X12 = (i0, i1), X13 = (i2, i3), X14 = (i4, i5), X15 = (i6, i7)
+// i0, i1, i2, i3, i4, i5, i6, i7 must not be 0
+#define LOAD_MSG_AVX(i0, i1, i2, i3, i4, i5, i6, i7) \
+	VMOVQ_SI_X12(i0*8);     \
+	VMOVQ_SI_X13(i2*8);     \
+	VMOVQ_SI_X14(i4*8);     \
+	VMOVQ_SI_X15(i6*8);     \
+	VPINSRQ_1_SI_X12(i1*8); \
+	VPINSRQ_1_SI_X13(i3*8); \
+	VPINSRQ_1_SI_X14(i5*8); \
+	VPINSRQ_1_SI_X15(i7*8)
+
+// load msg: X12 = (0, 2), X13 = (4, 6), X14 = (1, 3), X15 = (5, 7)
+#define LOAD_MSG_AVX_0_2_4_6_1_3_5_7() \
+	VMOVQ_SI_X12_0;        \
+	VMOVQ_SI_X13(4*8);     \
+	VMOVQ_SI_X14(1*8);     \
+	VMOVQ_SI_X15(5*8);     \
+	VPINSRQ_1_SI_X12(2*8); \
+	VPINSRQ_1_SI_X13(6*8); \
+	VPINSRQ_1_SI_X14(3*8); \
+	VPINSRQ_1_SI_X15(7*8)
+
+// load msg: X12 = (1, 0), X13 = (11, 5), X14 = (12, 2), X15 = (7, 3)
+#define LOAD_MSG_AVX_1_0_11_5_12_2_7_3() \
+	VPSHUFD $0x4E, 0*8(SI), X12; \
+	VMOVQ_SI_X13(11*8);          \
+	VMOVQ_SI_X14(12*8);          \
+	VMOVQ_SI_X15(7*8);           \
+	VPINSRQ_1_SI_X13(5*8);       \
+	VPINSRQ_1_SI_X14(2*8);       \
+	VPINSRQ_1_SI_X15(3*8)
+
+// load msg: X12 = (11, 12), X13 = (5, 15), X14 = (8, 0), X15 = (2, 13)
+#define LOAD_MSG_AVX_11_12_5_15_8_0_2_13() \
+	VMOVDQU 11*8(SI), X12;  \
+	VMOVQ_SI_X13(5*8);      \
+	VMOVQ_SI_X14(8*8);      \
+	VMOVQ_SI_X15(2*8);      \
+	VPINSRQ_1_SI_X13(15*8); \
+	VPINSRQ_1_SI_X14_0;     \
+	VPINSRQ_1_SI_X15(13*8)
+
+// load msg: X12 = (2, 5), X13 = (4, 15), X14 = (6, 10), X15 = (0, 8)
+#define LOAD_MSG_AVX_2_5_4_15_6_10_0_8() \
+	VMOVQ_SI_X12(2*8);      \
+	VMOVQ_SI_X13(4*8);      \
+	VMOVQ_SI_X14(6*8);      \
+	VMOVQ_SI_X15_0;         \
+	VPINSRQ_1_SI_X12(5*8);  \
+	VPINSRQ_1_SI_X13(15*8); \
+	VPINSRQ_1_SI_X14(10*8); \
+	VPINSRQ_1_SI_X15(8*8)
+
+// load msg: X12 = (9, 5), X13 = (2, 10), X14 = (0, 7), X15 = (4, 15)
+#define LOAD_MSG_AVX_9_5_2_10_0_7_4_15() \
+	VMOVQ_SI_X12(9*8);      \
+	VMOVQ_SI_X13(2*8);      \
+	VMOVQ_SI_X14_0;         \
+	VMOVQ_SI_X15(4*8);      \
+	VPINSRQ_1_SI_X12(5*8);  \
+	VPINSRQ_1_SI_X13(10*8); \
+	VPINSRQ_1_SI_X14(7*8);  \
+	VPINSRQ_1_SI_X15(15*8)
+
+// load msg: X12 = (2, 6), X13 = (0, 8), X14 = (12, 10), X15 = (11, 3)
+#define LOAD_MSG_AVX_2_6_0_8_12_10_11_3() \
+	VMOVQ_SI_X12(2*8);      \
+	VMOVQ_SI_X13_0;         \
+	VMOVQ_SI_X14(12*8);     \
+	VMOVQ_SI_X15(11*8);     \
+	VPINSRQ_1_SI_X12(6*8);  \
+	VPINSRQ_1_SI_X13(8*8);  \
+	VPINSRQ_1_SI_X14(10*8); \
+	VPINSRQ_1_SI_X15(3*8)
+
+// load msg: X12 = (0, 6), X13 = (9, 8), X14 = (7, 3), X15 = (2, 11)
+#define LOAD_MSG_AVX_0_6_9_8_7_3_2_11() \
+	MOVQ    0*8(SI), X12;        \
+	VPSHUFD $0x4E, 8*8(SI), X13; \
+	MOVQ    7*8(SI), X14;        \
+	MOVQ    2*8(SI), X15;        \
+	VPINSRQ_1_SI_X12(6*8);       \
+	VPINSRQ_1_SI_X14(3*8);       \
+	VPINSRQ_1_SI_X15(11*8)
+
+// load msg: X12 = (6, 14), X13 = (11, 0), X14 = (15, 9), X15 = (3, 8)
+#define LOAD_MSG_AVX_6_14_11_0_15_9_3_8() \
+	MOVQ 6*8(SI), X12;      \
+	MOVQ 11*8(SI), X13;     \
+	MOVQ 15*8(SI), X14;     \
+	MOVQ 3*8(SI), X15;      \
+	VPINSRQ_1_SI_X12(14*8); \
+	VPINSRQ_1_SI_X13_0;     \
+	VPINSRQ_1_SI_X14(9*8);  \
+	VPINSRQ_1_SI_X15(8*8)
+
+// load msg: X12 = (5, 15), X13 = (8, 2), X14 = (0, 4), X15 = (6, 10)
+#define LOAD_MSG_AVX_5_15_8_2_0_4_6_10() \
+	MOVQ 5*8(SI), X12;      \
+	MOVQ 8*8(SI), X13;      \
+	MOVQ 0*8(SI), X14;      \
+	MOVQ 6*8(SI), X15;      \
+	VPINSRQ_1_SI_X12(15*8); \
+	VPINSRQ_1_SI_X13(2*8);  \
+	VPINSRQ_1_SI_X14(4*8);  \
+	VPINSRQ_1_SI_X15(10*8)
+
+// load msg: X12 = (12, 13), X13 = (1, 10), X14 = (2, 7), X15 = (4, 5)
+#define LOAD_MSG_AVX_12_13_1_10_2_7_4_5() \
+	VMOVDQU 12*8(SI), X12;  \
+	MOVQ    1*8(SI), X13;   \
+	MOVQ    2*8(SI), X14;   \
+	VPINSRQ_1_SI_X13(10*8); \
+	VPINSRQ_1_SI_X14(7*8);  \
+	VMOVDQU 4*8(SI), X15
+
+// load msg: X12 = (15, 9), X13 = (3, 13), X14 = (11, 14), X15 = (12, 0)
+#define LOAD_MSG_AVX_15_9_3_13_11_14_12_0() \
+	MOVQ 15*8(SI), X12;     \
+	MOVQ 3*8(SI), X13;      \
+	MOVQ 11*8(SI), X14;     \
+	MOVQ 12*8(SI), X15;     \
+	VPINSRQ_1_SI_X12(9*8);  \
+	VPINSRQ_1_SI_X13(13*8); \
+	VPINSRQ_1_SI_X14(14*8); \
+	VPINSRQ_1_SI_X15_0
+
+// func hashBlocksAVX(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte)
+TEXT ·hashBlocksAVX(SB), 4, $288-48 // frame size = 272 + 16 byte alignment
+	MOVQ h+0(FP), AX
+	MOVQ c+8(FP), BX
+	MOVQ flag+16(FP), CX
+	MOVQ blocks_base+24(FP), SI
+	MOVQ blocks_len+32(FP), DI
+
+	MOVQ SP, R10
+	ADDQ $15, R10
+	ANDQ $~15, R10
+
+	VMOVDQU ·AVX_c40<>(SB), X0
+	VMOVDQU ·AVX_c48<>(SB), X1
+	VMOVDQA X0, X8
+	VMOVDQA X1, X9
+
+	VMOVDQU ·AVX_iv3<>(SB), X0
+	VMOVDQA X0, 0(R10)
+	XORQ    CX, 0(R10)          // 0(R10) = ·AVX_iv3 ^ (CX || 0)
+
+	VMOVDQU 0(AX), X10
+	VMOVDQU 16(AX), X11
+	VMOVDQU 32(AX), X2
+	VMOVDQU 48(AX), X3
+
+	MOVQ 0(BX), R8
+	MOVQ 8(BX), R9
+
+loop:
+	ADDQ $128, R8
+	CMPQ R8, $128
+	JGE  noinc
+	INCQ R9
+
+noinc:
+	VMOVQ_R8_X15
+	VPINSRQ_1_R9_X15
+
+	VMOVDQA X10, X0
+	VMOVDQA X11, X1
+	VMOVDQU ·AVX_iv0<>(SB), X4
+	VMOVDQU ·AVX_iv1<>(SB), X5
+	VMOVDQU ·AVX_iv2<>(SB), X6
+
+	VPXOR   X15, X6, X6
+	VMOVDQA 0(R10), X7
+
+	LOAD_MSG_AVX_0_2_4_6_1_3_5_7()
+	VMOVDQA X12, 16(R10)
+	VMOVDQA X13, 32(R10)
+	VMOVDQA X14, 48(R10)
+	VMOVDQA X15, 64(R10)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX(8, 10, 12, 14, 9, 11, 13, 15)
+	VMOVDQA X12, 80(R10)
+	VMOVDQA X13, 96(R10)
+	VMOVDQA X14, 112(R10)
+	VMOVDQA X15, 128(R10)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX(14, 4, 9, 13, 10, 8, 15, 6)
+	VMOVDQA X12, 144(R10)
+	VMOVDQA X13, 160(R10)
+	VMOVDQA X14, 176(R10)
+	VMOVDQA X15, 192(R10)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX_1_0_11_5_12_2_7_3()
+	VMOVDQA X12, 208(R10)
+	VMOVDQA X13, 224(R10)
+	VMOVDQA X14, 240(R10)
+	VMOVDQA X15, 256(R10)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX_11_12_5_15_8_0_2_13()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX(10, 3, 7, 9, 14, 6, 1, 4)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX(7, 3, 13, 11, 9, 1, 12, 14)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX_2_5_4_15_6_10_0_8()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX_9_5_2_10_0_7_4_15()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX(14, 11, 6, 3, 1, 12, 8, 13)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX_2_6_0_8_12_10_11_3()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX(4, 7, 15, 1, 13, 5, 14, 9)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX(12, 1, 14, 4, 5, 15, 13, 10)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX_0_6_9_8_7_3_2_11()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX(13, 7, 12, 3, 11, 14, 1, 9)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX_5_15_8_2_0_4_6_10()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX_6_14_11_0_15_9_3_8()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX_12_13_1_10_2_7_4_5()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	LOAD_MSG_AVX(10, 8, 7, 1, 2, 4, 6, 5)
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX()
+	LOAD_MSG_AVX_15_9_3_13_11_14_12_0()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, X12, X13, X14, X15, X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, 16(R10), 32(R10), 48(R10), 64(R10), X15, X8, X9)
+	SHUFFLE_AVX()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, 80(R10), 96(R10), 112(R10), 128(R10), X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, 144(R10), 160(R10), 176(R10), 192(R10), X15, X8, X9)
+	SHUFFLE_AVX()
+	HALF_ROUND_AVX(X0, X1, X2, X3, X4, X5, X6, X7, 208(R10), 224(R10), 240(R10), 256(R10), X15, X8, X9)
+	SHUFFLE_AVX_INV()
+
+	VMOVDQU 32(AX), X14
+	VMOVDQU 48(AX), X15
+	VPXOR   X0, X10, X10
+	VPXOR   X1, X11, X11
+	VPXOR   X2, X14, X14
+	VPXOR   X3, X15, X15
+	VPXOR   X4, X10, X10
+	VPXOR   X5, X11, X11
+	VPXOR   X6, X14, X2
+	VPXOR   X7, X15, X3
+	VMOVDQU X2, 32(AX)
+	VMOVDQU X3, 48(AX)
+
+	LEAQ 128(SI), SI
+	SUBQ $128, DI
+	JNE  loop
+
+	VMOVDQU X10, 0(AX)
+	VMOVDQU X11, 16(AX)
+
+	MOVQ R8, 0(BX)
+	MOVQ R9, 8(BX)
+	VZEROUPPER
+
+	RET
diff --git a/vendor/golang.org/x/crypto/blake2b/blake2b_amd64.s b/vendor/golang.org/x/crypto/blake2b/blake2b_amd64.s
new file mode 100644
index 0000000..adfac00
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2b_amd64.s
@@ -0,0 +1,278 @@
+// 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.
+
+//go:build amd64 && gc && !purego
+
+#include "textflag.h"
+
+DATA ·iv0<>+0x00(SB)/8, $0x6a09e667f3bcc908
+DATA ·iv0<>+0x08(SB)/8, $0xbb67ae8584caa73b
+GLOBL ·iv0<>(SB), (NOPTR+RODATA), $16
+
+DATA ·iv1<>+0x00(SB)/8, $0x3c6ef372fe94f82b
+DATA ·iv1<>+0x08(SB)/8, $0xa54ff53a5f1d36f1
+GLOBL ·iv1<>(SB), (NOPTR+RODATA), $16
+
+DATA ·iv2<>+0x00(SB)/8, $0x510e527fade682d1
+DATA ·iv2<>+0x08(SB)/8, $0x9b05688c2b3e6c1f
+GLOBL ·iv2<>(SB), (NOPTR+RODATA), $16
+
+DATA ·iv3<>+0x00(SB)/8, $0x1f83d9abfb41bd6b
+DATA ·iv3<>+0x08(SB)/8, $0x5be0cd19137e2179
+GLOBL ·iv3<>(SB), (NOPTR+RODATA), $16
+
+DATA ·c40<>+0x00(SB)/8, $0x0201000706050403
+DATA ·c40<>+0x08(SB)/8, $0x0a09080f0e0d0c0b
+GLOBL ·c40<>(SB), (NOPTR+RODATA), $16
+
+DATA ·c48<>+0x00(SB)/8, $0x0100070605040302
+DATA ·c48<>+0x08(SB)/8, $0x09080f0e0d0c0b0a
+GLOBL ·c48<>(SB), (NOPTR+RODATA), $16
+
+#define SHUFFLE(v2, v3, v4, v5, v6, v7, t1, t2) \
+	MOVO       v4, t1; \
+	MOVO       v5, v4; \
+	MOVO       t1, v5; \
+	MOVO       v6, t1; \
+	PUNPCKLQDQ v6, t2; \
+	PUNPCKHQDQ v7, v6; \
+	PUNPCKHQDQ t2, v6; \
+	PUNPCKLQDQ v7, t2; \
+	MOVO       t1, v7; \
+	MOVO       v2, t1; \
+	PUNPCKHQDQ t2, v7; \
+	PUNPCKLQDQ v3, t2; \
+	PUNPCKHQDQ t2, v2; \
+	PUNPCKLQDQ t1, t2; \
+	PUNPCKHQDQ t2, v3
+
+#define SHUFFLE_INV(v2, v3, v4, v5, v6, v7, t1, t2) \
+	MOVO       v4, t1; \
+	MOVO       v5, v4; \
+	MOVO       t1, v5; \
+	MOVO       v2, t1; \
+	PUNPCKLQDQ v2, t2; \
+	PUNPCKHQDQ v3, v2; \
+	PUNPCKHQDQ t2, v2; \
+	PUNPCKLQDQ v3, t2; \
+	MOVO       t1, v3; \
+	MOVO       v6, t1; \
+	PUNPCKHQDQ t2, v3; \
+	PUNPCKLQDQ v7, t2; \
+	PUNPCKHQDQ t2, v6; \
+	PUNPCKLQDQ t1, t2; \
+	PUNPCKHQDQ t2, v7
+
+#define HALF_ROUND(v0, v1, v2, v3, v4, v5, v6, v7, m0, m1, m2, m3, t0, c40, c48) \
+	PADDQ  m0, v0;        \
+	PADDQ  m1, v1;        \
+	PADDQ  v2, v0;        \
+	PADDQ  v3, v1;        \
+	PXOR   v0, v6;        \
+	PXOR   v1, v7;        \
+	PSHUFD $0xB1, v6, v6; \
+	PSHUFD $0xB1, v7, v7; \
+	PADDQ  v6, v4;        \
+	PADDQ  v7, v5;        \
+	PXOR   v4, v2;        \
+	PXOR   v5, v3;        \
+	PSHUFB c40, v2;       \
+	PSHUFB c40, v3;       \
+	PADDQ  m2, v0;        \
+	PADDQ  m3, v1;        \
+	PADDQ  v2, v0;        \
+	PADDQ  v3, v1;        \
+	PXOR   v0, v6;        \
+	PXOR   v1, v7;        \
+	PSHUFB c48, v6;       \
+	PSHUFB c48, v7;       \
+	PADDQ  v6, v4;        \
+	PADDQ  v7, v5;        \
+	PXOR   v4, v2;        \
+	PXOR   v5, v3;        \
+	MOVOU  v2, t0;        \
+	PADDQ  v2, t0;        \
+	PSRLQ  $63, v2;       \
+	PXOR   t0, v2;        \
+	MOVOU  v3, t0;        \
+	PADDQ  v3, t0;        \
+	PSRLQ  $63, v3;       \
+	PXOR   t0, v3
+
+#define LOAD_MSG(m0, m1, m2, m3, src, i0, i1, i2, i3, i4, i5, i6, i7) \
+	MOVQ   i0*8(src), m0;     \
+	PINSRQ $1, i1*8(src), m0; \
+	MOVQ   i2*8(src), m1;     \
+	PINSRQ $1, i3*8(src), m1; \
+	MOVQ   i4*8(src), m2;     \
+	PINSRQ $1, i5*8(src), m2; \
+	MOVQ   i6*8(src), m3;     \
+	PINSRQ $1, i7*8(src), m3
+
+// func hashBlocksSSE4(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte)
+TEXT ·hashBlocksSSE4(SB), 4, $288-48 // frame size = 272 + 16 byte alignment
+	MOVQ h+0(FP), AX
+	MOVQ c+8(FP), BX
+	MOVQ flag+16(FP), CX
+	MOVQ blocks_base+24(FP), SI
+	MOVQ blocks_len+32(FP), DI
+
+	MOVQ SP, R10
+	ADDQ $15, R10
+	ANDQ $~15, R10
+
+	MOVOU ·iv3<>(SB), X0
+	MOVO  X0, 0(R10)
+	XORQ  CX, 0(R10)     // 0(R10) = ·iv3 ^ (CX || 0)
+
+	MOVOU ·c40<>(SB), X13
+	MOVOU ·c48<>(SB), X14
+
+	MOVOU 0(AX), X12
+	MOVOU 16(AX), X15
+
+	MOVQ 0(BX), R8
+	MOVQ 8(BX), R9
+
+loop:
+	ADDQ $128, R8
+	CMPQ R8, $128
+	JGE  noinc
+	INCQ R9
+
+noinc:
+	MOVQ R8, X8
+	PINSRQ $1, R9, X8
+
+	MOVO X12, X0
+	MOVO X15, X1
+	MOVOU 32(AX), X2
+	MOVOU 48(AX), X3
+	MOVOU ·iv0<>(SB), X4
+	MOVOU ·iv1<>(SB), X5
+	MOVOU ·iv2<>(SB), X6
+
+	PXOR X8, X6
+	MOVO 0(R10), X7
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 0, 2, 4, 6, 1, 3, 5, 7)
+	MOVO X8, 16(R10)
+	MOVO X9, 32(R10)
+	MOVO X10, 48(R10)
+	MOVO X11, 64(R10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 8, 10, 12, 14, 9, 11, 13, 15)
+	MOVO X8, 80(R10)
+	MOVO X9, 96(R10)
+	MOVO X10, 112(R10)
+	MOVO X11, 128(R10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 14, 4, 9, 13, 10, 8, 15, 6)
+	MOVO X8, 144(R10)
+	MOVO X9, 160(R10)
+	MOVO X10, 176(R10)
+	MOVO X11, 192(R10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 1, 0, 11, 5, 12, 2, 7, 3)
+	MOVO X8, 208(R10)
+	MOVO X9, 224(R10)
+	MOVO X10, 240(R10)
+	MOVO X11, 256(R10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 11, 12, 5, 15, 8, 0, 2, 13)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 10, 3, 7, 9, 14, 6, 1, 4)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 7, 3, 13, 11, 9, 1, 12, 14)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 2, 5, 4, 15, 6, 10, 0, 8)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 9, 5, 2, 10, 0, 7, 4, 15)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 14, 11, 6, 3, 1, 12, 8, 13)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 2, 6, 0, 8, 12, 10, 11, 3)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 4, 7, 15, 1, 13, 5, 14, 9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 12, 1, 14, 4, 5, 15, 13, 10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 0, 6, 9, 8, 7, 3, 2, 11)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 13, 7, 12, 3, 11, 14, 1, 9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 5, 15, 8, 2, 0, 4, 6, 10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 6, 14, 11, 0, 15, 9, 3, 8)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 12, 13, 1, 10, 2, 7, 4, 5)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 10, 8, 7, 1, 2, 4, 6, 5)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 15, 9, 3, 13, 11, 14, 12, 0)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, 16(R10), 32(R10), 48(R10), 64(R10), X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, 80(R10), 96(R10), 112(R10), 128(R10), X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, 144(R10), 160(R10), 176(R10), 192(R10), X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, 208(R10), 224(R10), 240(R10), 256(R10), X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	MOVOU 32(AX), X10
+	MOVOU 48(AX), X11
+	PXOR  X0, X12
+	PXOR  X1, X15
+	PXOR  X2, X10
+	PXOR  X3, X11
+	PXOR  X4, X12
+	PXOR  X5, X15
+	PXOR  X6, X10
+	PXOR  X7, X11
+	MOVOU X10, 32(AX)
+	MOVOU X11, 48(AX)
+
+	LEAQ 128(SI), SI
+	SUBQ $128, DI
+	JNE  loop
+
+	MOVOU X12, 0(AX)
+	MOVOU X15, 16(AX)
+
+	MOVQ R8, 0(BX)
+	MOVQ R9, 8(BX)
+
+	RET
diff --git a/vendor/golang.org/x/crypto/blake2b/blake2b_generic.go b/vendor/golang.org/x/crypto/blake2b/blake2b_generic.go
new file mode 100644
index 0000000..3168a8a
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2b_generic.go
@@ -0,0 +1,182 @@
+// 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 blake2b
+
+import (
+	"encoding/binary"
+	"math/bits"
+)
+
+// the precomputed values for BLAKE2b
+// there are 12 16-byte arrays - one for each round
+// the entries are calculated from the sigma constants.
+var precomputed = [12][16]byte{
+	{0, 2, 4, 6, 1, 3, 5, 7, 8, 10, 12, 14, 9, 11, 13, 15},
+	{14, 4, 9, 13, 10, 8, 15, 6, 1, 0, 11, 5, 12, 2, 7, 3},
+	{11, 12, 5, 15, 8, 0, 2, 13, 10, 3, 7, 9, 14, 6, 1, 4},
+	{7, 3, 13, 11, 9, 1, 12, 14, 2, 5, 4, 15, 6, 10, 0, 8},
+	{9, 5, 2, 10, 0, 7, 4, 15, 14, 11, 6, 3, 1, 12, 8, 13},
+	{2, 6, 0, 8, 12, 10, 11, 3, 4, 7, 15, 1, 13, 5, 14, 9},
+	{12, 1, 14, 4, 5, 15, 13, 10, 0, 6, 9, 8, 7, 3, 2, 11},
+	{13, 7, 12, 3, 11, 14, 1, 9, 5, 15, 8, 2, 0, 4, 6, 10},
+	{6, 14, 11, 0, 15, 9, 3, 8, 12, 13, 1, 10, 2, 7, 4, 5},
+	{10, 8, 7, 1, 2, 4, 6, 5, 15, 9, 3, 13, 11, 14, 12, 0},
+	{0, 2, 4, 6, 1, 3, 5, 7, 8, 10, 12, 14, 9, 11, 13, 15}, // equal to the first
+	{14, 4, 9, 13, 10, 8, 15, 6, 1, 0, 11, 5, 12, 2, 7, 3}, // equal to the second
+}
+
+func hashBlocksGeneric(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte) {
+	var m [16]uint64
+	c0, c1 := c[0], c[1]
+
+	for i := 0; i < len(blocks); {
+		c0 += BlockSize
+		if c0 < BlockSize {
+			c1++
+		}
+
+		v0, v1, v2, v3, v4, v5, v6, v7 := h[0], h[1], h[2], h[3], h[4], h[5], h[6], h[7]
+		v8, v9, v10, v11, v12, v13, v14, v15 := iv[0], iv[1], iv[2], iv[3], iv[4], iv[5], iv[6], iv[7]
+		v12 ^= c0
+		v13 ^= c1
+		v14 ^= flag
+
+		for j := range m {
+			m[j] = binary.LittleEndian.Uint64(blocks[i:])
+			i += 8
+		}
+
+		for j := range precomputed {
+			s := &(precomputed[j])
+
+			v0 += m[s[0]]
+			v0 += v4
+			v12 ^= v0
+			v12 = bits.RotateLeft64(v12, -32)
+			v8 += v12
+			v4 ^= v8
+			v4 = bits.RotateLeft64(v4, -24)
+			v1 += m[s[1]]
+			v1 += v5
+			v13 ^= v1
+			v13 = bits.RotateLeft64(v13, -32)
+			v9 += v13
+			v5 ^= v9
+			v5 = bits.RotateLeft64(v5, -24)
+			v2 += m[s[2]]
+			v2 += v6
+			v14 ^= v2
+			v14 = bits.RotateLeft64(v14, -32)
+			v10 += v14
+			v6 ^= v10
+			v6 = bits.RotateLeft64(v6, -24)
+			v3 += m[s[3]]
+			v3 += v7
+			v15 ^= v3
+			v15 = bits.RotateLeft64(v15, -32)
+			v11 += v15
+			v7 ^= v11
+			v7 = bits.RotateLeft64(v7, -24)
+
+			v0 += m[s[4]]
+			v0 += v4
+			v12 ^= v0
+			v12 = bits.RotateLeft64(v12, -16)
+			v8 += v12
+			v4 ^= v8
+			v4 = bits.RotateLeft64(v4, -63)
+			v1 += m[s[5]]
+			v1 += v5
+			v13 ^= v1
+			v13 = bits.RotateLeft64(v13, -16)
+			v9 += v13
+			v5 ^= v9
+			v5 = bits.RotateLeft64(v5, -63)
+			v2 += m[s[6]]
+			v2 += v6
+			v14 ^= v2
+			v14 = bits.RotateLeft64(v14, -16)
+			v10 += v14
+			v6 ^= v10
+			v6 = bits.RotateLeft64(v6, -63)
+			v3 += m[s[7]]
+			v3 += v7
+			v15 ^= v3
+			v15 = bits.RotateLeft64(v15, -16)
+			v11 += v15
+			v7 ^= v11
+			v7 = bits.RotateLeft64(v7, -63)
+
+			v0 += m[s[8]]
+			v0 += v5
+			v15 ^= v0
+			v15 = bits.RotateLeft64(v15, -32)
+			v10 += v15
+			v5 ^= v10
+			v5 = bits.RotateLeft64(v5, -24)
+			v1 += m[s[9]]
+			v1 += v6
+			v12 ^= v1
+			v12 = bits.RotateLeft64(v12, -32)
+			v11 += v12
+			v6 ^= v11
+			v6 = bits.RotateLeft64(v6, -24)
+			v2 += m[s[10]]
+			v2 += v7
+			v13 ^= v2
+			v13 = bits.RotateLeft64(v13, -32)
+			v8 += v13
+			v7 ^= v8
+			v7 = bits.RotateLeft64(v7, -24)
+			v3 += m[s[11]]
+			v3 += v4
+			v14 ^= v3
+			v14 = bits.RotateLeft64(v14, -32)
+			v9 += v14
+			v4 ^= v9
+			v4 = bits.RotateLeft64(v4, -24)
+
+			v0 += m[s[12]]
+			v0 += v5
+			v15 ^= v0
+			v15 = bits.RotateLeft64(v15, -16)
+			v10 += v15
+			v5 ^= v10
+			v5 = bits.RotateLeft64(v5, -63)
+			v1 += m[s[13]]
+			v1 += v6
+			v12 ^= v1
+			v12 = bits.RotateLeft64(v12, -16)
+			v11 += v12
+			v6 ^= v11
+			v6 = bits.RotateLeft64(v6, -63)
+			v2 += m[s[14]]
+			v2 += v7
+			v13 ^= v2
+			v13 = bits.RotateLeft64(v13, -16)
+			v8 += v13
+			v7 ^= v8
+			v7 = bits.RotateLeft64(v7, -63)
+			v3 += m[s[15]]
+			v3 += v4
+			v14 ^= v3
+			v14 = bits.RotateLeft64(v14, -16)
+			v9 += v14
+			v4 ^= v9
+			v4 = bits.RotateLeft64(v4, -63)
+
+		}
+
+		h[0] ^= v0 ^ v8
+		h[1] ^= v1 ^ v9
+		h[2] ^= v2 ^ v10
+		h[3] ^= v3 ^ v11
+		h[4] ^= v4 ^ v12
+		h[5] ^= v5 ^ v13
+		h[6] ^= v6 ^ v14
+		h[7] ^= v7 ^ v15
+	}
+	c[0], c[1] = c0, c1
+}
diff --git a/vendor/golang.org/x/crypto/blake2b/blake2b_ref.go b/vendor/golang.org/x/crypto/blake2b/blake2b_ref.go
new file mode 100644
index 0000000..6e28668
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2b_ref.go
@@ -0,0 +1,11 @@
+// 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.
+
+//go:build !amd64 || purego || !gc
+
+package blake2b
+
+func hashBlocks(h *[8]uint64, c *[2]uint64, flag uint64, blocks []byte) {
+	hashBlocksGeneric(h, c, flag, blocks)
+}
diff --git a/vendor/golang.org/x/crypto/blake2b/blake2x.go b/vendor/golang.org/x/crypto/blake2b/blake2x.go
new file mode 100644
index 0000000..52c414d
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/blake2x.go
@@ -0,0 +1,177 @@
+// Copyright 2017 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 blake2b
+
+import (
+	"encoding/binary"
+	"errors"
+	"io"
+)
+
+// XOF defines the interface to hash functions that
+// support arbitrary-length output.
+type XOF interface {
+	// Write absorbs more data into the hash's state. It panics if called
+	// after Read.
+	io.Writer
+
+	// Read reads more output from the hash. It returns io.EOF if the limit
+	// has been reached.
+	io.Reader
+
+	// Clone returns a copy of the XOF in its current state.
+	Clone() XOF
+
+	// Reset resets the XOF to its initial state.
+	Reset()
+}
+
+// OutputLengthUnknown can be used as the size argument to NewXOF to indicate
+// the length of the output is not known in advance.
+const OutputLengthUnknown = 0
+
+// magicUnknownOutputLength is a magic value for the output size that indicates
+// an unknown number of output bytes.
+const magicUnknownOutputLength = (1 << 32) - 1
+
+// maxOutputLength is the absolute maximum number of bytes to produce when the
+// number of output bytes is unknown.
+const maxOutputLength = (1 << 32) * 64
+
+// NewXOF creates a new variable-output-length hash. The hash either produce a
+// known number of bytes (1 <= size < 2**32-1), or an unknown number of bytes
+// (size == OutputLengthUnknown). In the latter case, an absolute limit of
+// 256GiB applies.
+//
+// A non-nil key turns the hash into a MAC. The key must between
+// zero and 32 bytes long.
+func NewXOF(size uint32, key []byte) (XOF, error) {
+	if len(key) > Size {
+		return nil, errKeySize
+	}
+	if size == magicUnknownOutputLength {
+		// 2^32-1 indicates an unknown number of bytes and thus isn't a
+		// valid length.
+		return nil, errors.New("blake2b: XOF length too large")
+	}
+	if size == OutputLengthUnknown {
+		size = magicUnknownOutputLength
+	}
+	x := &xof{
+		d: digest{
+			size:   Size,
+			keyLen: len(key),
+		},
+		length: size,
+	}
+	copy(x.d.key[:], key)
+	x.Reset()
+	return x, nil
+}
+
+type xof struct {
+	d                digest
+	length           uint32
+	remaining        uint64
+	cfg, root, block [Size]byte
+	offset           int
+	nodeOffset       uint32
+	readMode         bool
+}
+
+func (x *xof) Write(p []byte) (n int, err error) {
+	if x.readMode {
+		panic("blake2b: write to XOF after read")
+	}
+	return x.d.Write(p)
+}
+
+func (x *xof) Clone() XOF {
+	clone := *x
+	return &clone
+}
+
+func (x *xof) Reset() {
+	x.cfg[0] = byte(Size)
+	binary.LittleEndian.PutUint32(x.cfg[4:], uint32(Size)) // leaf length
+	binary.LittleEndian.PutUint32(x.cfg[12:], x.length)    // XOF length
+	x.cfg[17] = byte(Size)                                 // inner hash size
+
+	x.d.Reset()
+	x.d.h[1] ^= uint64(x.length) << 32
+
+	x.remaining = uint64(x.length)
+	if x.remaining == magicUnknownOutputLength {
+		x.remaining = maxOutputLength
+	}
+	x.offset, x.nodeOffset = 0, 0
+	x.readMode = false
+}
+
+func (x *xof) Read(p []byte) (n int, err error) {
+	if !x.readMode {
+		x.d.finalize(&x.root)
+		x.readMode = true
+	}
+
+	if x.remaining == 0 {
+		return 0, io.EOF
+	}
+
+	n = len(p)
+	if uint64(n) > x.remaining {
+		n = int(x.remaining)
+		p = p[:n]
+	}
+
+	if x.offset > 0 {
+		blockRemaining := Size - x.offset
+		if n < blockRemaining {
+			x.offset += copy(p, x.block[x.offset:])
+			x.remaining -= uint64(n)
+			return
+		}
+		copy(p, x.block[x.offset:])
+		p = p[blockRemaining:]
+		x.offset = 0
+		x.remaining -= uint64(blockRemaining)
+	}
+
+	for len(p) >= Size {
+		binary.LittleEndian.PutUint32(x.cfg[8:], x.nodeOffset)
+		x.nodeOffset++
+
+		x.d.initConfig(&x.cfg)
+		x.d.Write(x.root[:])
+		x.d.finalize(&x.block)
+
+		copy(p, x.block[:])
+		p = p[Size:]
+		x.remaining -= uint64(Size)
+	}
+
+	if todo := len(p); todo > 0 {
+		if x.remaining < uint64(Size) {
+			x.cfg[0] = byte(x.remaining)
+		}
+		binary.LittleEndian.PutUint32(x.cfg[8:], x.nodeOffset)
+		x.nodeOffset++
+
+		x.d.initConfig(&x.cfg)
+		x.d.Write(x.root[:])
+		x.d.finalize(&x.block)
+
+		x.offset = copy(p, x.block[:todo])
+		x.remaining -= uint64(todo)
+	}
+	return
+}
+
+func (d *digest) initConfig(cfg *[Size]byte) {
+	d.offset, d.c[0], d.c[1] = 0, 0, 0
+	for i := range d.h {
+		d.h[i] = iv[i] ^ binary.LittleEndian.Uint64(cfg[i*8:])
+	}
+}
diff --git a/vendor/golang.org/x/crypto/blake2b/register.go b/vendor/golang.org/x/crypto/blake2b/register.go
new file mode 100644
index 0000000..54e446e
--- /dev/null
+++ b/vendor/golang.org/x/crypto/blake2b/register.go
@@ -0,0 +1,30 @@
+// Copyright 2017 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 blake2b
+
+import (
+	"crypto"
+	"hash"
+)
+
+func init() {
+	newHash256 := func() hash.Hash {
+		h, _ := New256(nil)
+		return h
+	}
+	newHash384 := func() hash.Hash {
+		h, _ := New384(nil)
+		return h
+	}
+
+	newHash512 := func() hash.Hash {
+		h, _ := New512(nil)
+		return h
+	}
+
+	crypto.RegisterHash(crypto.BLAKE2b_256, newHash256)
+	crypto.RegisterHash(crypto.BLAKE2b_384, newHash384)
+	crypto.RegisterHash(crypto.BLAKE2b_512, newHash512)
+}