diff --git a/gogf127/gogf127.go b/gf127/arith.go similarity index 59% rename from gogf127/gogf127.go rename to gf127/arith.go index 10af5c4..7977440 100644 --- a/gogf127/gogf127.go +++ b/gf127/arith.go @@ -3,22 +3,10 @@ // Package gf127 implements the GF(2^127) arithmetic // modulo reduction polynomial x^127 + x^63 + 1 . // Implementation is in pure Go. -package gogf127 +package gf127 import ( - "encoding/binary" - "encoding/hex" - "errors" "math/bits" - "math/rand" -) - -// GF127 represents element of GF(2^127) -type GF127 [2]uint64 - -const ( - msb64 = uint64(0x8000000000000000) - byteSize = 16 ) var ( @@ -32,50 +20,6 @@ func New(lo, hi uint64) *GF127 { return &GF127{lo, hi} } -// Random returns random element from GF(2^127). -// Is used mostly for testing. -func Random() *GF127 { - return &GF127{rand.Uint64(), rand.Uint64() >> 1} -} - -// String returns hex-encoded representation, starting with MSB. -func (c *GF127) String() string { - return hex.EncodeToString(c.ByteArray()) -} - -// Equals checks if two reduced (zero MSB) elements of GF(2^127) are equal -func (c *GF127) Equals(b *GF127) bool { - return c[0] == b[0] && c[1] == b[1] -} - -// ByteArray represents element of GF(2^127) as byte array of length 16. -func (c *GF127) ByteArray() (buf []byte) { - buf = make([]byte, 16) - binary.BigEndian.PutUint64(buf[:8], c[1]) - binary.BigEndian.PutUint64(buf[8:], c[0]) - return -} - -// MarshalBinary implements encoding.BinaryMarshaler. -func (c *GF127) MarshalBinary() (data []byte, err error) { - return c.ByteArray(), nil -} - -// UnmarshalBinary implements encoding.BinaryUnmarshaler. -func (c *GF127) UnmarshalBinary(data []byte) error { - if len(data) != byteSize { - return errors.New("data must be 16-bytes long") - } - - c[0] = binary.BigEndian.Uint64(data[8:]) - c[1] = binary.BigEndian.Uint64(data[:8]) - if c[1]&msb64 != 0 { - return errors.New("MSB must be zero") - } - - return nil -} - // Inv sets b to a^-1 // Algorithm is based on Extended Euclidean Algorithm // and is described by Hankerson, Hernandez, Menezes in @@ -129,18 +73,6 @@ func msb(a *GF127) (x int) { return 127 - x } -// Mul sets c to the product a*b and returns c. -func (c *GF127) Mul(a, b *GF127) *GF127 { - Mul(a, b, c) - return c -} - -// Add sets c to the sum a+b and returns c. -func (c *GF127) Add(a, b *GF127) *GF127 { - Add(a, b, c) - return c -} - // Mul1 copies a to b. func Mul1(a, b *GF127) { b[0] = a[0] diff --git a/gogf127/gf127_test.go b/gf127/arith_test.go similarity index 98% rename from gogf127/gf127_test.go rename to gf127/arith_test.go index a6addb6..4cac7f7 100644 --- a/gogf127/gf127_test.go +++ b/gf127/arith_test.go @@ -1,4 +1,4 @@ -package gogf127 +package gf127 import ( "testing" @@ -6,8 +6,6 @@ import ( "github.com/stretchr/testify/require" ) -const maxUint64 = ^uint64(0) - func TestAdd(t *testing.T) { var ( a = Random() diff --git a/gf127/avx/gf127.go b/gf127/avx/gf127.go new file mode 100644 index 0000000..4072981 --- /dev/null +++ b/gf127/avx/gf127.go @@ -0,0 +1,34 @@ +// Copyright 2018 (c) NSPCC +// +// Package gf127 implements the GF(2^127) arithmetic +// modulo reduction polynomial x^127 + x^63 + 1 . +// This is rather straight-forward re-implementation of C library +// available here https://github.com/srijs/hwsl2-core . +// Interfaces are highly influenced by math/big . +package avx + +import ( + "github.com/nspcc-dev/tzhash/gf127" +) + +// GF127 is an alias for a main type. +type GF127 = gf127.GF127 + +const msb64 = uint64(1) << 63 + +var ( + // x127x63 represents x^127 + x^63. Used in assembly file. + x127x63 = GF127{msb64, msb64} +) + +// Add sets c to a+b. +func Add(a, b, c *GF127) + +// Mul sets c to a*b. +func Mul(a, b, c *GF127) + +// Mul10 sets b to a*x. +func Mul10(a, b *GF127) + +// Mul11 sets b to a*(x+1). +func Mul11(a, b *GF127) diff --git a/gf127/gf127_amd64.s b/gf127/avx/gf127_amd64.s similarity index 75% rename from gf127/gf127_amd64.s rename to gf127/avx/gf127_amd64.s index ada6f24..d159814 100644 --- a/gf127/gf127_amd64.s +++ b/gf127/avx/gf127_amd64.s @@ -27,22 +27,6 @@ TEXT ·Mul10(SB),NOSPLIT,$0 MOVUPD X1, (AX) RET -// func Mul10x2(a, b) *[4]uint64 -TEXT ·Mul10x2(SB),NOSPLIT,$0 - MOVQ a+0(FP), AX - VMOVDQA (AX), Y0 - VPSLLQ $1, Y0, Y1 - VPALIGNR $8, Y1, Y0, Y2 - VPSRLQ $63, Y2, Y2 - VPXOR Y1, Y2, Y2 - VPSRLQ $63, Y1, Y3 - VPSLLQ $63, Y3, Y3 - VPUNPCKHQDQ Y3, Y3, Y3 - VPXOR Y2, Y3, Y3 - MOVQ b+8(FP), AX - VMOVDQA Y3, (AX) - RET - // func Mul11(a, b *[2]uint64) TEXT ·Mul11(SB),NOSPLIT,$0 MOVQ a+0(FP), AX @@ -60,23 +44,6 @@ TEXT ·Mul11(SB),NOSPLIT,$0 MOVUPD X1, (AX) RET -// func Mul11x2(a, b) *[4]uint64 -TEXT ·Mul11x2(SB),NOSPLIT,$0 - MOVQ a+0(FP), AX - VMOVDQA (AX), Y0 - VPSLLQ $1, Y0, Y1 - VPALIGNR $8, Y1, Y0, Y2 - VPSRLQ $63, Y2, Y2 - VPXOR Y1, Y2, Y2 - VPSRLQ $63, Y1, Y3 - VPSLLQ $63, Y3, Y3 - VPUNPCKHQDQ Y3, Y3, Y3 - VPXOR Y2, Y3, Y3 - VPXOR Y0, Y3, Y3 - MOVQ b+8(FP), AX - VMOVDQA Y3, (AX) - RET - // func Mul(a, b, c *[2]uint64) TEXT ·Mul(SB),NOSPLIT,$0 MOVQ a+0(FP), AX // X0 = a0 . a1 diff --git a/gf127/gf127_test.go b/gf127/avx/gf127_test.go similarity index 56% rename from gf127/gf127_test.go rename to gf127/avx/gf127_test.go index e24533d..9da629c 100644 --- a/gf127/gf127_test.go +++ b/gf127/avx/gf127_test.go @@ -1,8 +1,9 @@ -package gf127 +package avx import ( "testing" + "github.com/nspcc-dev/tzhash/gf127" "github.com/stretchr/testify/require" ) @@ -10,12 +11,12 @@ const maxUint64 = ^uint64(0) func TestAdd(t *testing.T) { var ( - a = Random() - b = Random() + a = gf127.Random() + b = gf127.Random() e = &GF127{a[0] ^ b[0], a[1] ^ b[1]} c = new(GF127) ) - c.Add(a, b) + Add(a, b, c) require.Equal(t, e, c) } @@ -65,46 +66,3 @@ func TestMul11(t *testing.T) { require.Equal(t, tc[1], c) } } - -var testCasesInv = [][2]*GF127{ - {&GF127{1, 0}, &GF127{1, 0}}, - {&GF127{3, 0}, &GF127{msb64, ^msb64}}, - {&GF127{54321, 12345}, &GF127{8230555108620784737, 3929873967650665114}}, -} - -func TestInv(t *testing.T) { - var a, b, c = new(GF127), new(GF127), new(GF127) - for _, tc := range testCasesInv { - Inv(tc[0], c) - require.Equal(t, tc[1], c) - } - - for i := 0; i < 3; i++ { - // 0 has no inverse - if a = Random(); a.Equals(&GF127{0, 0}) { - continue - } - Inv(a, b) - Mul(a, b, c) - require.Equal(t, &GF127{1, 0}, c) - } -} - -func TestGF127_MarshalBinary(t *testing.T) { - a := New(0xFF, 0xEE) - data, err := a.MarshalBinary() - require.NoError(t, err) - require.Equal(t, data, []byte{0, 0, 0, 0, 0, 0, 0, 0xEE, 0, 0, 0, 0, 0, 0, 0, 0xFF}) - - a = Random() - data, err = a.MarshalBinary() - require.NoError(t, err) - - b := new(GF127) - err = b.UnmarshalBinary(data) - require.NoError(t, err) - require.Equal(t, a, b) - - err = b.UnmarshalBinary([]byte{0, 1, 2, 3}) - require.Error(t, err) -} diff --git a/gf127/gf127x2.go b/gf127/avx2/gf127x2.go similarity index 91% rename from gf127/gf127x2.go rename to gf127/avx2/gf127x2.go index 22ab9a8..7dbca58 100644 --- a/gf127/gf127x2.go +++ b/gf127/avx2/gf127x2.go @@ -1,10 +1,15 @@ -package gf127 +package avx2 import ( "encoding/binary" "encoding/hex" + + "github.com/nspcc-dev/tzhash/gf127" ) +// GF127 is an alias for a main type. +type GF127 = gf127.GF127 + // GF127x2 represents a pair of elements of GF(2^127) stored together. type GF127x2 [2]GF127 diff --git a/gf127/avx2/gf127x2_amd64.s b/gf127/avx2/gf127x2_amd64.s new file mode 100644 index 0000000..05f3700 --- /dev/null +++ b/gf127/avx2/gf127x2_amd64.s @@ -0,0 +1,35 @@ +#include "textflag.h" + +// func Mul10x2(a, b) *[4]uint64 +TEXT ·Mul10x2(SB),NOSPLIT,$0 + MOVQ a+0(FP), AX + VMOVDQA (AX), Y0 + VPSLLQ $1, Y0, Y1 + VPALIGNR $8, Y1, Y0, Y2 + VPSRLQ $63, Y2, Y2 + VPXOR Y1, Y2, Y2 + VPSRLQ $63, Y1, Y3 + VPSLLQ $63, Y3, Y3 + VPUNPCKHQDQ Y3, Y3, Y3 + VPXOR Y2, Y3, Y3 + MOVQ b+8(FP), AX + VMOVDQA Y3, (AX) + RET + + +// func Mul11x2(a, b) *[4]uint64 +TEXT ·Mul11x2(SB),NOSPLIT,$0 + MOVQ a+0(FP), AX + VMOVDQA (AX), Y0 + VPSLLQ $1, Y0, Y1 + VPALIGNR $8, Y1, Y0, Y2 + VPSRLQ $63, Y2, Y2 + VPXOR Y1, Y2, Y2 + VPSRLQ $63, Y1, Y3 + VPSLLQ $63, Y3, Y3 + VPUNPCKHQDQ Y3, Y3, Y3 + VPXOR Y2, Y3, Y3 + VPXOR Y0, Y3, Y3 + MOVQ b+8(FP), AX + VMOVDQA Y3, (AX) + RET diff --git a/gf127/gf127x2_test.go b/gf127/avx2/gf127x2_test.go similarity index 97% rename from gf127/gf127x2_test.go rename to gf127/avx2/gf127x2_test.go index 73a4a68..5de5865 100644 --- a/gf127/gf127x2_test.go +++ b/gf127/avx2/gf127x2_test.go @@ -1,4 +1,4 @@ -package gf127 +package avx2 import ( "testing" @@ -6,6 +6,8 @@ import ( "github.com/stretchr/testify/require" ) +const maxUint64 = ^uint64(0) + var testCasesSplit = []struct { num *GF127x2 h1 *GF127 diff --git a/gf127/gf127.go b/gf127/gf127.go index 40e7965..de3c276 100644 --- a/gf127/gf127.go +++ b/gf127/gf127.go @@ -1,42 +1,21 @@ -// Copyright 2018 (c) NSPCC -// -// Package gf127 implements the GF(2^127) arithmetic -// modulo reduction polynomial x^127 + x^63 + 1 . -// This is rather straight-forward re-implementation of C library -// available here https://github.com/srijs/hwsl2-core . -// Interfaces are highly influenced by math/big . package gf127 import ( "encoding/binary" "encoding/hex" "errors" - "math/bits" "math/rand" ) +const ( + byteSize = 16 + maxUint64 = ^uint64(0) + msb64 = uint64(1) << 63 +) + // GF127 represents element of GF(2^127) type GF127 [2]uint64 -const ( - msb64 = uint64(0x8000000000000000) - byteSize = 16 -) - -var ( - // x127x63 represents x^127 + x^63. Used in assembly file. - x127x63 = GF127{msb64, msb64} - - // x126x631 is reduction polynomial x^127+x^63+1 - x127x631 = GF127{msb64 + 1, msb64} -) - -// New constructs new element of GF(2^127) as hi*x^64 + lo. -// It is assumed that hi has zero MSB. -func New(lo, hi uint64) *GF127 { - return &GF127{lo, hi} -} - // Random returns random element from GF(2^127). // Is used mostly for testing. func Random() *GF127 { @@ -80,92 +59,3 @@ func (c *GF127) UnmarshalBinary(data []byte) error { return nil } - -// Inv sets b to a^-1 -// Algorithm is based on Extended Euclidean Algorithm -// and is described by Hankerson, Hernandez, Menezes in -// https://link.springer.com/content/pdf/10.1007/3-540-44499-8_1.pdf -func Inv(a, b *GF127) { - var ( - v = x127x631 - u = *a - c, d = &GF127{1, 0}, &GF127{0, 0} - t = new(GF127) - x *GF127 - ) - - // degree of polynomial is a position of most significant bit - for du, dv := msb(&u), msb(&v); du != 0; du, dv = msb(&u), msb(&v) { - if du < dv { - v, u = u, v - dv, du = du, dv - d, c = c, d - } - - x = xN(du - dv) - - Mul(x, &v, t) - Add(&u, t, &u) - - // becasuse mul performs reduction on t, we need - // manually reduce u at first step - if msb(&u) == 127 { - Add(&u, &x127x631, &u) - } - - Mul(x, d, t) - Add(c, t, c) - } - *b = *c -} - -func xN(n int) *GF127 { - if n < 64 { - return &GF127{1 << uint(n), 0} - } - return &GF127{0, 1 << uint(n-64)} -} - -func msb(a *GF127) (x int) { - x = bits.LeadingZeros64(a[1]) - if x == 64 { - x = bits.LeadingZeros64(a[0]) + 64 - } - return 127 - x -} - -// Mul sets c to the product a*b and returns c. -func (c *GF127) Mul(a, b *GF127) *GF127 { - Mul(a, b, c) - return c -} - -// Add sets c to the sum a+b and returns c. -func (c *GF127) Add(a, b *GF127) *GF127 { - Add(a, b, c) - return c -} - -// Mul1 copies a to b. -func Mul1(a, b *GF127) { - b[0] = a[0] - b[1] = a[1] -} - -// And sets c to a & b (bitwise-and). -func And(a, b, c *GF127) { - c[0] = a[0] & b[0] - c[1] = a[1] & b[1] -} - -// Add sets c to a+b. -func Add(a, b, c *GF127) - -// Mul sets c to a*b. -func Mul(a, b, c *GF127) - -// Mul10 sets b to a*x. -func Mul10(a, b *GF127) - -// Mul11 sets b to a*(x+1). -func Mul11(a, b *GF127) diff --git a/tz/avx.go b/tz/avx.go index 4a61702..09793ef 100644 --- a/tz/avx.go +++ b/tz/avx.go @@ -6,20 +6,18 @@ package tz import ( "hash" "math" - - "github.com/nspcc-dev/tzhash/gf127" ) type digest struct { - x [4]gf127.GF127 + x [4]GF127 } // type assertion var _ hash.Hash = (*digest)(nil) var ( - minmax = [2]gf127.GF127{{0, 0}, {math.MaxUint64, math.MaxUint64}} - x127x63 = gf127.GF127{1 << 63, 1 << 63} + minmax = [2]GF127{{0, 0}, {math.MaxUint64, math.MaxUint64}} + x127x63 = GF127{1 << 63, 1 << 63} ) func newAVX() *digest { @@ -48,10 +46,10 @@ func (d *digest) byteArray() (b [hashSize]byte) { } func (d *digest) Reset() { - d.x[0] = gf127.GF127{1, 0} - d.x[1] = gf127.GF127{0, 0} - d.x[2] = gf127.GF127{0, 0} - d.x[3] = gf127.GF127{1, 0} + d.x[0] = GF127{1, 0} + d.x[1] = GF127{0, 0} + d.x[2] = GF127{0, 0} + d.x[3] = GF127{1, 0} } func (d *digest) Write(data []byte) (n int, err error) { @@ -77,4 +75,4 @@ func (d *digest) BlockSize() int { return hashBlockSize } -func mulBitRight(c00, c01, c10, c11, e *gf127.GF127) +func mulBitRight(c00, c01, c10, c11, e *GF127) diff --git a/tz/avx2.go b/tz/avx2.go index 00ec3b3..788baf2 100644 --- a/tz/avx2.go +++ b/tz/avx2.go @@ -6,11 +6,11 @@ package tz import ( "hash" - "github.com/nspcc-dev/tzhash/gf127" + "github.com/nspcc-dev/tzhash/gf127/avx2" ) type digest2 struct { - x [2]gf127.GF127x2 + x [2]avx2.GF127x2 } // type assertion @@ -44,8 +44,8 @@ func (d *digest2) Sum(in []byte) []byte { return append(in, h[:]...) } func (d *digest2) Reset() { - d.x[0] = gf127.GF127x2{gf127.GF127{1, 0}, gf127.GF127{0, 0}} - d.x[1] = gf127.GF127x2{gf127.GF127{0, 0}, gf127.GF127{1, 0}} + d.x[0] = avx2.GF127x2{GF127{1, 0}, GF127{0, 0}} + d.x[1] = avx2.GF127x2{GF127{0, 0}, GF127{1, 0}} } func (d *digest2) Size() int { return hashSize } func (d *digest2) BlockSize() int { return hashBlockSize } @@ -62,4 +62,4 @@ func (d *digest2) checkSum() (b [hashSize]byte) { return } -func mulBitRightx2(c00c10 *gf127.GF127x2, c01c11 *gf127.GF127x2, e *gf127.GF127) +func mulBitRightx2(c00c10 *avx2.GF127x2, c01c11 *avx2.GF127x2, e *GF127) diff --git a/tz/avx2_inline.go b/tz/avx2_inline.go index 286e1b7..09afbc1 100644 --- a/tz/avx2_inline.go +++ b/tz/avx2_inline.go @@ -7,11 +7,11 @@ package tz import ( "hash" - "github.com/nspcc-dev/tzhash/gf127" + "github.com/nspcc-dev/tzhash/gf127/avx2" ) type digest3 struct { - x [2]gf127.GF127x2 + x [2]avx2.GF127x2 } // type assertion @@ -38,8 +38,8 @@ func (d *digest3) Sum(in []byte) []byte { return append(in, h[:]...) } func (d *digest3) Reset() { - d.x[0] = gf127.GF127x2{gf127.GF127{1, 0}, gf127.GF127{0, 0}} - d.x[1] = gf127.GF127x2{gf127.GF127{0, 0}, gf127.GF127{1, 0}} + d.x[0] = avx2.GF127x2{GF127{1, 0}, GF127{0, 0}} + d.x[1] = avx2.GF127x2{GF127{0, 0}, GF127{1, 0}} } func (d *digest3) Size() int { return hashSize } func (d *digest3) BlockSize() int { return hashBlockSize } @@ -56,4 +56,4 @@ func (d *digest3) checkSum() (b [hashSize]byte) { return } -func mulByteRightx2(c00c10 *gf127.GF127x2, c01c11 *gf127.GF127x2, b byte) +func mulByteRightx2(c00c10 *avx2.GF127x2, c01c11 *avx2.GF127x2, b byte) diff --git a/tz/pure.go b/tz/pure.go index d1a5961..8c5dad9 100644 --- a/tz/pure.go +++ b/tz/pure.go @@ -1,11 +1,11 @@ package tz import ( - "github.com/nspcc-dev/tzhash/gogf127" + "github.com/nspcc-dev/tzhash/gf127" ) type digestp struct { - x [4]gogf127.GF127 + x [4]GF127 } // New returns a new hash.Hash computing the Tillich-Zémor checksum. @@ -35,15 +35,15 @@ func (d *digestp) byteArray() (b [hashSize]byte) { } func (d *digestp) Reset() { - d.x[0] = gogf127.GF127{1, 0} - d.x[1] = gogf127.GF127{0, 0} - d.x[2] = gogf127.GF127{0, 0} - d.x[3] = gogf127.GF127{1, 0} + d.x[0] = GF127{1, 0} + d.x[1] = GF127{0, 0} + d.x[2] = GF127{0, 0} + d.x[3] = GF127{1, 0} } func (d *digestp) Write(data []byte) (n int, err error) { n = len(data) - tmp := new(gogf127.GF127) + tmp := new(GF127) for _, b := range data { mulBitRightPure(&d.x[0], &d.x[1], &d.x[2], &d.x[3], b&0x80 != 0, tmp) mulBitRightPure(&d.x[0], &d.x[1], &d.x[2], &d.x[3], b&0x40 != 0, tmp) @@ -65,28 +65,28 @@ func (d *digestp) BlockSize() int { return hashBlockSize } -func mulBitRightPure(c00, c01, c10, c11 *gogf127.GF127, bit bool, tmp *gogf127.GF127) { +func mulBitRightPure(c00, c01, c10, c11 *GF127, bit bool, tmp *GF127) { if bit { *tmp = *c00 - gogf127.Mul10(c00, c00) - gogf127.Add(c00, c01, c00) - gogf127.Mul11(tmp, tmp) - gogf127.Add(c01, tmp, c01) + gf127.Mul10(c00, c00) + gf127.Add(c00, c01, c00) + gf127.Mul11(tmp, tmp) + gf127.Add(c01, tmp, c01) *tmp = *c10 - gogf127.Mul10(c10, c10) - gogf127.Add(c10, c11, c10) - gogf127.Mul11(tmp, tmp) - gogf127.Add(c11, tmp, c11) + gf127.Mul10(c10, c10) + gf127.Add(c10, c11, c10) + gf127.Mul11(tmp, tmp) + gf127.Add(c11, tmp, c11) } else { *tmp = *c00 - gogf127.Mul10(c00, c00) - gogf127.Add(c00, c01, c00) + gf127.Mul10(c00, c00) + gf127.Add(c00, c01, c00) *c01 = *tmp *tmp = *c10 - gogf127.Mul10(c10, c10) - gogf127.Add(c10, c11, c10) + gf127.Mul10(c10, c10) + gf127.Add(c10, c11, c10) *c11 = *tmp } } diff --git a/tz/sl2.go b/tz/sl2.go index 30c0264..e0a4d8a 100644 --- a/tz/sl2.go +++ b/tz/sl2.go @@ -4,17 +4,21 @@ import ( "errors" "github.com/nspcc-dev/tzhash/gf127" - "github.com/nspcc-dev/tzhash/gogf127" + "github.com/nspcc-dev/tzhash/gf127/avx" ) -type sl2 [2][2]gf127.GF127 +type ( + GF127 = gf127.GF127 + + sl2 [2][2]GF127 +) var id = sl2{ - {gf127.GF127{1, 0}, gf127.GF127{0, 0}}, - {gf127.GF127{0, 0}, gf127.GF127{1, 0}}, + {GF127{1, 0}, GF127{0, 0}}, + {GF127{0, 0}, GF127{1, 0}}, } -var mul func(a, b, c *sl2, x *[4]gf127.GF127) +var mul func(a, b, c *sl2, x *[4]GF127) func init() { if hasAVX { @@ -50,130 +54,130 @@ func (c *sl2) UnmarshalBinary(data []byte) (err error) { return } -func (c *sl2) mulStrassen(a, b *sl2, x *[8]gf127.GF127) *sl2 { +func (c *sl2) mulStrassen(a, b *sl2, x *[8]GF127) *sl2 { // strassen algorithm - gf127.Add(&a[0][0], &a[1][1], &x[0]) - gf127.Add(&b[0][0], &b[1][1], &x[1]) - gf127.Mul(&x[0], &x[1], &x[0]) + avx.Add(&a[0][0], &a[1][1], &x[0]) + avx.Add(&b[0][0], &b[1][1], &x[1]) + avx.Mul(&x[0], &x[1], &x[0]) - gf127.Add(&a[1][0], &a[1][1], &x[1]) - gf127.Mul(&x[1], &b[0][0], &x[1]) + avx.Add(&a[1][0], &a[1][1], &x[1]) + avx.Mul(&x[1], &b[0][0], &x[1]) - gf127.Add(&b[0][1], &b[1][1], &x[2]) - gf127.Mul(&x[2], &a[0][0], &x[2]) + avx.Add(&b[0][1], &b[1][1], &x[2]) + avx.Mul(&x[2], &a[0][0], &x[2]) - gf127.Add(&b[1][0], &b[0][0], &x[3]) - gf127.Mul(&x[3], &a[1][1], &x[3]) + avx.Add(&b[1][0], &b[0][0], &x[3]) + avx.Mul(&x[3], &a[1][1], &x[3]) - gf127.Add(&a[0][0], &a[0][1], &x[4]) - gf127.Mul(&x[4], &b[1][1], &x[4]) + avx.Add(&a[0][0], &a[0][1], &x[4]) + avx.Mul(&x[4], &b[1][1], &x[4]) - gf127.Add(&a[1][0], &a[0][0], &x[5]) - gf127.Add(&b[0][0], &b[0][1], &x[6]) - gf127.Mul(&x[5], &x[6], &x[5]) + avx.Add(&a[1][0], &a[0][0], &x[5]) + avx.Add(&b[0][0], &b[0][1], &x[6]) + avx.Mul(&x[5], &x[6], &x[5]) - gf127.Add(&a[0][1], &a[1][1], &x[6]) - gf127.Add(&b[1][0], &b[1][1], &x[7]) - gf127.Mul(&x[6], &x[7], &x[6]) + avx.Add(&a[0][1], &a[1][1], &x[6]) + avx.Add(&b[1][0], &b[1][1], &x[7]) + avx.Mul(&x[6], &x[7], &x[6]) - gf127.Add(&x[2], &x[4], &c[0][1]) - gf127.Add(&x[1], &x[3], &c[1][0]) + avx.Add(&x[2], &x[4], &c[0][1]) + avx.Add(&x[1], &x[3], &c[1][0]) - gf127.Add(&x[4], &x[6], &x[4]) - gf127.Add(&x[0], &x[3], &c[0][0]) - gf127.Add(&c[0][0], &x[4], &c[0][0]) + avx.Add(&x[4], &x[6], &x[4]) + avx.Add(&x[0], &x[3], &c[0][0]) + avx.Add(&c[0][0], &x[4], &c[0][0]) - gf127.Add(&x[0], &x[1], &x[0]) - gf127.Add(&x[2], &x[5], &c[1][1]) - gf127.Add(&c[1][1], &x[0], &c[1][1]) + avx.Add(&x[0], &x[1], &x[0]) + avx.Add(&x[2], &x[5], &c[1][1]) + avx.Add(&c[1][1], &x[0], &c[1][1]) return c } -func mulSL2AVX(a, b, c *sl2, x *[4]gf127.GF127) { - gf127.Mul(&a[0][0], &b[0][0], &x[0]) - gf127.Mul(&a[0][0], &b[0][1], &x[1]) - gf127.Mul(&a[1][0], &b[0][0], &x[2]) - gf127.Mul(&a[1][0], &b[0][1], &x[3]) +func mulSL2AVX(a, b, c *sl2, x *[4]GF127) { + avx.Mul(&a[0][0], &b[0][0], &x[0]) + avx.Mul(&a[0][0], &b[0][1], &x[1]) + avx.Mul(&a[1][0], &b[0][0], &x[2]) + avx.Mul(&a[1][0], &b[0][1], &x[3]) - gf127.Mul(&a[0][1], &b[1][0], &c[0][0]) - gf127.Add(&c[0][0], &x[0], &c[0][0]) - gf127.Mul(&a[0][1], &b[1][1], &c[0][1]) - gf127.Add(&c[0][1], &x[1], &c[0][1]) - gf127.Mul(&a[1][1], &b[1][0], &c[1][0]) - gf127.Add(&c[1][0], &x[2], &c[1][0]) - gf127.Mul(&a[1][1], &b[1][1], &c[1][1]) - gf127.Add(&c[1][1], &x[3], &c[1][1]) + avx.Mul(&a[0][1], &b[1][0], &c[0][0]) + avx.Add(&c[0][0], &x[0], &c[0][0]) + avx.Mul(&a[0][1], &b[1][1], &c[0][1]) + avx.Add(&c[0][1], &x[1], &c[0][1]) + avx.Mul(&a[1][1], &b[1][0], &c[1][0]) + avx.Add(&c[1][0], &x[2], &c[1][0]) + avx.Mul(&a[1][1], &b[1][1], &c[1][1]) + avx.Add(&c[1][1], &x[3], &c[1][1]) } -func mulSL2Pure(a, b, c *sl2, x *[4]gf127.GF127) { - gogf127.Mul((*gogf127.GF127)(&a[0][0]), (*gogf127.GF127)(&b[0][0]), (*gogf127.GF127)(&x[0])) - gogf127.Mul((*gogf127.GF127)(&a[0][0]), (*gogf127.GF127)(&b[0][1]), (*gogf127.GF127)(&x[1])) - gogf127.Mul((*gogf127.GF127)(&a[1][0]), (*gogf127.GF127)(&b[0][0]), (*gogf127.GF127)(&x[2])) - gogf127.Mul((*gogf127.GF127)(&a[1][0]), (*gogf127.GF127)(&b[0][1]), (*gogf127.GF127)(&x[3])) +func mulSL2Pure(a, b, c *sl2, x *[4]GF127) { + gf127.Mul((*GF127)(&a[0][0]), (*GF127)(&b[0][0]), (*GF127)(&x[0])) + gf127.Mul((*GF127)(&a[0][0]), (*GF127)(&b[0][1]), (*GF127)(&x[1])) + gf127.Mul((*GF127)(&a[1][0]), (*GF127)(&b[0][0]), (*GF127)(&x[2])) + gf127.Mul((*GF127)(&a[1][0]), (*GF127)(&b[0][1]), (*GF127)(&x[3])) - gogf127.Mul((*gogf127.GF127)(&a[0][1]), (*gogf127.GF127)(&b[1][0]), (*gogf127.GF127)(&c[0][0])) - gogf127.Add((*gogf127.GF127)(&c[0][0]), (*gogf127.GF127)(&x[0]), (*gogf127.GF127)(&c[0][0])) - gogf127.Mul((*gogf127.GF127)(&a[0][1]), (*gogf127.GF127)(&b[1][1]), (*gogf127.GF127)(&c[0][1])) - gogf127.Add((*gogf127.GF127)(&c[0][1]), (*gogf127.GF127)(&x[1]), (*gogf127.GF127)(&c[0][1])) - gogf127.Mul((*gogf127.GF127)(&a[1][1]), (*gogf127.GF127)(&b[1][0]), (*gogf127.GF127)(&c[1][0])) - gogf127.Add((*gogf127.GF127)(&c[1][0]), (*gogf127.GF127)(&x[2]), (*gogf127.GF127)(&c[1][0])) - gogf127.Mul((*gogf127.GF127)(&a[1][1]), (*gogf127.GF127)(&b[1][1]), (*gogf127.GF127)(&c[1][1])) - gogf127.Add((*gogf127.GF127)(&c[1][1]), (*gogf127.GF127)(&x[3]), (*gogf127.GF127)(&c[1][1])) + gf127.Mul((*GF127)(&a[0][1]), (*GF127)(&b[1][0]), (*GF127)(&c[0][0])) + gf127.Add((*GF127)(&c[0][0]), (*GF127)(&x[0]), (*GF127)(&c[0][0])) + gf127.Mul((*GF127)(&a[0][1]), (*GF127)(&b[1][1]), (*GF127)(&c[0][1])) + gf127.Add((*GF127)(&c[0][1]), (*GF127)(&x[1]), (*GF127)(&c[0][1])) + gf127.Mul((*GF127)(&a[1][1]), (*GF127)(&b[1][0]), (*GF127)(&c[1][0])) + gf127.Add((*GF127)(&c[1][0]), (*GF127)(&x[2]), (*GF127)(&c[1][0])) + gf127.Mul((*GF127)(&a[1][1]), (*GF127)(&b[1][1]), (*GF127)(&c[1][1])) + gf127.Add((*GF127)(&c[1][1]), (*GF127)(&x[3]), (*GF127)(&c[1][1])) } func (c *sl2) MulA() *sl2 { - var a gf127.GF127 + var a GF127 - gf127.Mul10(&c[0][0], &a) + avx.Mul10(&c[0][0], &a) gf127.Mul1(&c[0][0], &c[0][1]) - gf127.Add(&a, &c[0][1], &c[0][0]) + avx.Add(&a, &c[0][1], &c[0][0]) - gf127.Mul10(&c[1][0], &a) + avx.Mul10(&c[1][0], &a) gf127.Mul1(&c[1][0], &c[1][1]) - gf127.Add(&a, &c[1][1], &c[1][0]) + avx.Add(&a, &c[1][1], &c[1][0]) return c } func (c *sl2) MulB() *sl2 { - var a gf127.GF127 + var a GF127 gf127.Mul1(&c[0][0], &a) - gf127.Mul10(&c[0][0], &c[0][0]) - gf127.Add(&c[0][1], &c[0][0], &c[0][0]) - gf127.Add(&c[0][0], &a, &c[0][1]) + avx.Mul10(&c[0][0], &c[0][0]) + avx.Add(&c[0][1], &c[0][0], &c[0][0]) + avx.Add(&c[0][0], &a, &c[0][1]) gf127.Mul1(&c[1][0], &a) - gf127.Mul10(&c[1][0], &c[1][0]) - gf127.Add(&c[1][1], &c[1][0], &c[1][0]) - gf127.Add(&c[1][0], &a, &c[1][1]) + avx.Mul10(&c[1][0], &c[1][0]) + avx.Add(&c[1][1], &c[1][0], &c[1][0]) + avx.Add(&c[1][0], &a, &c[1][1]) return c } func (c *sl2) Mul(a, b *sl2) *sl2 { - mul(a, b, c, new([4]gf127.GF127)) + mul(a, b, c, new([4]GF127)) return c } // Inv returns inverse of a in GL_2(GF(2^127)) func Inv(a *sl2) (b *sl2) { b = new(sl2) - inv(a, b, new([2]gf127.GF127)) + inv(a, b, new([2]GF127)) return } -func inv(a, b *sl2, t *[2]gf127.GF127) { - gf127.Mul(&a[0][0], &a[1][1], &t[0]) - gf127.Mul(&a[0][1], &a[1][0], &t[1]) - gf127.Add(&t[0], &t[1], &t[0]) +func inv(a, b *sl2, t *[2]GF127) { + avx.Mul(&a[0][0], &a[1][1], &t[0]) + avx.Mul(&a[0][1], &a[1][0], &t[1]) + avx.Add(&t[0], &t[1], &t[0]) gf127.Inv(&t[0], &t[1]) - gf127.Mul(&t[1], &a[0][0], &b[1][1]) - gf127.Mul(&t[1], &a[0][1], &b[0][1]) - gf127.Mul(&t[1], &a[1][0], &b[1][0]) - gf127.Mul(&t[1], &a[1][1], &b[0][0]) + avx.Mul(&t[1], &a[0][0], &b[1][1]) + avx.Mul(&t[1], &a[0][1], &b[0][1]) + avx.Mul(&t[1], &a[1][0], &b[1][0]) + avx.Mul(&t[1], &a[1][1], &b[0][0]) } func (c *sl2) String() string { diff --git a/tz/sl2_test.go b/tz/sl2_test.go index 884d64c..d43fec8 100644 --- a/tz/sl2_test.go +++ b/tz/sl2_test.go @@ -6,6 +6,7 @@ import ( "time" "github.com/nspcc-dev/tzhash/gf127" + "github.com/nspcc-dev/tzhash/gf127/avx" "github.com/stretchr/testify/require" ) @@ -21,12 +22,12 @@ func random() (a *sl2) { // so that result is in SL2 // d = a^-1*(1+b*c) - gf127.Mul(&a[0][1], &a[1][0], &a[1][1]) - gf127.Add(&a[1][1], gf127.New(1, 0), &a[1][1]) + avx.Mul(&a[0][1], &a[1][0], &a[1][1]) + avx.Add(&a[1][1], gf127.New(1, 0), &a[1][1]) t := gf127.New(0, 0) gf127.Inv(&a[0][0], t) - gf127.Mul(t, &a[1][1], &a[1][1]) + avx.Mul(t, &a[1][1], &a[1][1]) return }