diff options
Diffstat (limited to 'src/crypto/sha256')
| -rw-r--r-- | src/crypto/sha256/example_test.go | 41 | ||||
| -rw-r--r-- | src/crypto/sha256/fallback_test.go | 35 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256.go | 270 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256_test.go | 329 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block.go | 128 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_386.s | 283 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_amd64.go | 9 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_amd64.s | 1031 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_arm64.go | 21 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_arm64.s | 119 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_decl.go | 11 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_generic.go | 11 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_ppc64le.s | 419 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_s390x.go | 9 | ||||
| -rw-r--r-- | src/crypto/sha256/sha256block_s390x.s | 20 |
15 files changed, 2736 insertions, 0 deletions
diff --git a/src/crypto/sha256/example_test.go b/src/crypto/sha256/example_test.go new file mode 100644 index 0000000..7d73120 --- /dev/null +++ b/src/crypto/sha256/example_test.go @@ -0,0 +1,41 @@ +// 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 sha256_test + +import ( + "crypto/sha256" + "fmt" + "io" + "log" + "os" +) + +func ExampleSum256() { + sum := sha256.Sum256([]byte("hello world\n")) + fmt.Printf("%x", sum) + // Output: a948904f2f0f479b8f8197694b30184b0d2ed1c1cd2a1ec0fb85d299a192a447 +} + +func ExampleNew() { + h := sha256.New() + h.Write([]byte("hello world\n")) + fmt.Printf("%x", h.Sum(nil)) + // Output: a948904f2f0f479b8f8197694b30184b0d2ed1c1cd2a1ec0fb85d299a192a447 +} + +func ExampleNew_file() { + f, err := os.Open("file.txt") + if err != nil { + log.Fatal(err) + } + defer f.Close() + + h := sha256.New() + if _, err := io.Copy(h, f); err != nil { + log.Fatal(err) + } + + fmt.Printf("%x", h.Sum(nil)) +} diff --git a/src/crypto/sha256/fallback_test.go b/src/crypto/sha256/fallback_test.go new file mode 100644 index 0000000..3f561aa --- /dev/null +++ b/src/crypto/sha256/fallback_test.go @@ -0,0 +1,35 @@ +// 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 s390x + +package sha256 + +import ( + "fmt" + "io" + "testing" +) + +// Tests the fallback code path in case the optimized asm +// implementation cannot be used. +// See also TestBlockGeneric. +func TestGenericPath(t *testing.T) { + if useAsm == false { + t.Skipf("assembly implementation unavailable") + } + useAsm = false + defer func() { useAsm = true }() + c := New() + in := "ΑΒΓΔΕϜΖΗΘΙΚΛΜΝΞΟΠϺϘΡΣΤΥΦΧΨΩ" + gold := "e93d84ec2b22383123be9f713697fb25" + + "338c86e2f7d8d1ddc2d89d332dd9d76c" + if _, err := io.WriteString(c, in); err != nil { + t.Fatalf("could not write to c: %v", err) + } + out := fmt.Sprintf("%x", c.Sum(nil)) + if out != gold { + t.Fatalf("mismatch: got %s, wanted %s", out, gold) + } +} diff --git a/src/crypto/sha256/sha256.go b/src/crypto/sha256/sha256.go new file mode 100644 index 0000000..659531d --- /dev/null +++ b/src/crypto/sha256/sha256.go @@ -0,0 +1,270 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package sha256 implements the SHA224 and SHA256 hash algorithms as defined +// in FIPS 180-4. +package sha256 + +import ( + "crypto" + "encoding/binary" + "errors" + "hash" +) + +func init() { + crypto.RegisterHash(crypto.SHA224, New224) + crypto.RegisterHash(crypto.SHA256, New) +} + +// The size of a SHA256 checksum in bytes. +const Size = 32 + +// The size of a SHA224 checksum in bytes. +const Size224 = 28 + +// The blocksize of SHA256 and SHA224 in bytes. +const BlockSize = 64 + +const ( + chunk = 64 + init0 = 0x6A09E667 + init1 = 0xBB67AE85 + init2 = 0x3C6EF372 + init3 = 0xA54FF53A + init4 = 0x510E527F + init5 = 0x9B05688C + init6 = 0x1F83D9AB + init7 = 0x5BE0CD19 + init0_224 = 0xC1059ED8 + init1_224 = 0x367CD507 + init2_224 = 0x3070DD17 + init3_224 = 0xF70E5939 + init4_224 = 0xFFC00B31 + init5_224 = 0x68581511 + init6_224 = 0x64F98FA7 + init7_224 = 0xBEFA4FA4 +) + +// digest represents the partial evaluation of a checksum. +type digest struct { + h [8]uint32 + x [chunk]byte + nx int + len uint64 + is224 bool // mark if this digest is SHA-224 +} + +const ( + magic224 = "sha\x02" + magic256 = "sha\x03" + marshaledSize = len(magic256) + 8*4 + chunk + 8 +) + +func (d *digest) MarshalBinary() ([]byte, error) { + b := make([]byte, 0, marshaledSize) + if d.is224 { + b = append(b, magic224...) + } else { + b = append(b, magic256...) + } + b = appendUint32(b, d.h[0]) + b = appendUint32(b, d.h[1]) + b = appendUint32(b, d.h[2]) + b = appendUint32(b, d.h[3]) + b = appendUint32(b, d.h[4]) + b = appendUint32(b, d.h[5]) + b = appendUint32(b, d.h[6]) + b = appendUint32(b, d.h[7]) + b = append(b, d.x[:d.nx]...) + b = b[:len(b)+len(d.x)-int(d.nx)] // already zero + b = appendUint64(b, d.len) + return b, nil +} + +func (d *digest) UnmarshalBinary(b []byte) error { + if len(b) < len(magic224) || (d.is224 && string(b[:len(magic224)]) != magic224) || (!d.is224 && string(b[:len(magic256)]) != magic256) { + return errors.New("crypto/sha256: invalid hash state identifier") + } + if len(b) != marshaledSize { + return errors.New("crypto/sha256: invalid hash state size") + } + b = b[len(magic224):] + b, d.h[0] = consumeUint32(b) + b, d.h[1] = consumeUint32(b) + b, d.h[2] = consumeUint32(b) + b, d.h[3] = consumeUint32(b) + b, d.h[4] = consumeUint32(b) + b, d.h[5] = consumeUint32(b) + b, d.h[6] = consumeUint32(b) + b, d.h[7] = consumeUint32(b) + b = b[copy(d.x[:], b):] + b, d.len = consumeUint64(b) + d.nx = int(d.len % chunk) + return nil +} + +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) { + _ = b[7] + x := uint64(b[7]) | uint64(b[6])<<8 | uint64(b[5])<<16 | uint64(b[4])<<24 | + uint64(b[3])<<32 | uint64(b[2])<<40 | uint64(b[1])<<48 | uint64(b[0])<<56 + return b[8:], x +} + +func consumeUint32(b []byte) ([]byte, uint32) { + _ = b[3] + x := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24 + return b[4:], x +} + +func (d *digest) Reset() { + if !d.is224 { + d.h[0] = init0 + d.h[1] = init1 + d.h[2] = init2 + d.h[3] = init3 + d.h[4] = init4 + d.h[5] = init5 + d.h[6] = init6 + d.h[7] = init7 + } else { + d.h[0] = init0_224 + d.h[1] = init1_224 + d.h[2] = init2_224 + d.h[3] = init3_224 + d.h[4] = init4_224 + d.h[5] = init5_224 + d.h[6] = init6_224 + d.h[7] = init7_224 + } + d.nx = 0 + d.len = 0 +} + +// New returns a new hash.Hash computing the SHA256 checksum. The Hash +// also implements encoding.BinaryMarshaler and +// encoding.BinaryUnmarshaler to marshal and unmarshal the internal +// state of the hash. +func New() hash.Hash { + d := new(digest) + d.Reset() + return d +} + +// New224 returns a new hash.Hash computing the SHA224 checksum. +func New224() hash.Hash { + d := new(digest) + d.is224 = true + d.Reset() + return d +} + +func (d *digest) Size() int { + if !d.is224 { + return Size + } + return Size224 +} + +func (d *digest) BlockSize() int { return BlockSize } + +func (d *digest) Write(p []byte) (nn int, err error) { + nn = len(p) + d.len += uint64(nn) + if d.nx > 0 { + n := copy(d.x[d.nx:], p) + d.nx += n + if d.nx == chunk { + block(d, d.x[:]) + d.nx = 0 + } + p = p[n:] + } + if len(p) >= chunk { + n := len(p) &^ (chunk - 1) + block(d, p[:n]) + p = p[n:] + } + if len(p) > 0 { + d.nx = copy(d.x[:], p) + } + return +} + +func (d *digest) Sum(in []byte) []byte { + // Make a copy of d so that caller can keep writing and summing. + d0 := *d + hash := d0.checkSum() + if d0.is224 { + return append(in, hash[:Size224]...) + } + return append(in, hash[:]...) +} + +func (d *digest) checkSum() [Size]byte { + len := d.len + // Padding. Add a 1 bit and 0 bits until 56 bytes mod 64. + var tmp [64]byte + tmp[0] = 0x80 + if len%64 < 56 { + d.Write(tmp[0 : 56-len%64]) + } else { + d.Write(tmp[0 : 64+56-len%64]) + } + + // Length in bits. + len <<= 3 + binary.BigEndian.PutUint64(tmp[:], len) + d.Write(tmp[0:8]) + + if d.nx != 0 { + panic("d.nx != 0") + } + + var digest [Size]byte + + binary.BigEndian.PutUint32(digest[0:], d.h[0]) + binary.BigEndian.PutUint32(digest[4:], d.h[1]) + binary.BigEndian.PutUint32(digest[8:], d.h[2]) + binary.BigEndian.PutUint32(digest[12:], d.h[3]) + binary.BigEndian.PutUint32(digest[16:], d.h[4]) + binary.BigEndian.PutUint32(digest[20:], d.h[5]) + binary.BigEndian.PutUint32(digest[24:], d.h[6]) + if !d.is224 { + binary.BigEndian.PutUint32(digest[28:], d.h[7]) + } + + return digest +} + +// Sum256 returns the SHA256 checksum of the data. +func Sum256(data []byte) [Size]byte { + var d digest + d.Reset() + d.Write(data) + return d.checkSum() +} + +// Sum224 returns the SHA224 checksum of the data. +func Sum224(data []byte) [Size224]byte { + var d digest + d.is224 = true + d.Reset() + d.Write(data) + sum := d.checkSum() + ap := (*[Size224]byte)(sum[:]) + return *ap +} diff --git a/src/crypto/sha256/sha256_test.go b/src/crypto/sha256/sha256_test.go new file mode 100644 index 0000000..702aa0b --- /dev/null +++ b/src/crypto/sha256/sha256_test.go @@ -0,0 +1,329 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// SHA256 hash algorithm. See FIPS 180-2. + +package sha256 + +import ( + "bytes" + "crypto/rand" + "encoding" + "fmt" + "hash" + "io" + "testing" +) + +type sha256Test struct { + out string + in string + halfState string // marshaled hash state after first half of in written, used by TestGoldenMarshal +} + +var golden = []sha256Test{ + {"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"}, + {"ca978112ca1bbdcafac231b39a23dc4da786eff8147c4e72b9807785afee48bb", "a", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"}, + {"fb8e20fc2e4c3f248c60c39bd652f3c1347298bb977b8b4d5903b85055620603", "ab", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"}, + {"ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad", "abc", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"}, + {"88d4266fd4e6338d13b845fcf289579d209c897823b9217da3e161936f031589", "abcd", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19ab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"}, + {"36bbe50ed96841d10443bcb670d6554f0a34b761be67ec9c4a8ad2c0c44ca42c", "abcde", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19ab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"}, + {"bef57ec7f53a6d40beb640a780a639c83bc29ac8a9816f1fc6c5c6dcd93c4721", "abcdef", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19abc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"}, + {"7d1a54127b222502f5b79b5fb0803061152a44f92b37e23c6527baf665d4da9a", "abcdefg", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19abc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"}, + {"9c56cc51b374c3ba189210d5b6d4bf57790d351c96c47c02190ecf1e430635ab", "abcdefgh", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19abcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04"}, + {"19cc02f26df43cc571bc9ed7b0c4d29224a3ec229529221725ef76d021c8326f", "abcdefghi", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19abcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04"}, + {"72399361da6a7754fec986dca5b7cbaf1c810a28ded4abaf56b2106d06cb78b0", "abcdefghij", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19abcde\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x05"}, + {"a144061c271f152da4d151034508fed1c138b8c976339de229c3bb6d4bbb4fce", "Discard medicine more than two years old.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19Discard medicine mor\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x14"}, + {"6dae5caa713a10ad04b46028bf6dad68837c581616a1589a265a11288d4bb5c4", "He who has a shady past knows that nice guys finish last.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19He who has a shady past know\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, + {"ae7a702a9509039ddbf29f0765e70d0001177914b86459284dab8b348c2dce3f", "I wouldn't marry him with a ten foot pole.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19I wouldn't marry him \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x15"}, + {"6748450b01c568586715291dfa3ee018da07d36bb7ea6f180c1af6270215c64f", "Free! Free!/A trip/to Mars/for 900/empty jars/Burma Shave", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19Free! Free!/A trip/to Mars/f\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, + {"14b82014ad2b11f661b5ae6a99b75105c2ffac278cd071cd6c05832793635774", "The days of the digital watch are numbered. -Tom Stoppard", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19The days of the digital watch\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1d"}, + {"7102cfd76e2e324889eece5d6c41921b1e142a4ac5a2692be78803097f6a48d8", "Nepal premier won't resign.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19Nepal premier\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\r"}, + {"23b1018cd81db1d67983c5f7417c44da9deb582459e378d7a068552ea649dc9f", "For every action there is an equal and opposite government program.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19For every action there is an equa\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00!"}, + {"8001f190dfb527261c4cfcab70c98e8097a7a1922129bc4096950e57c7999a5a", "His money is twice tainted: 'taint yours and 'taint mine.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19His money is twice tainted: \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, + {"8c87deb65505c3993eb24b7a150c4155e82eee6960cf0c3a8114ff736d69cad5", "There is no reason for any individual to have a computer in their home. -Ken Olsen, 1977", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19There is no reason for any individual to hav\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00,"}, + {"bfb0a67a19cdec3646498b2e0f751bddc41bba4b7f30081b0b932aad214d16d7", "It's a tiny change to the code and not completely disgusting. - Bob Manchek", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19It's a tiny change to the code and no\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00%"}, + {"7f9a0b9bf56332e19f5a0ec1ad9c1425a153da1c624868fda44561d6b74daf36", "size: a.out: bad magic", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19size: a.out\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\f"}, + {"b13f81b8aad9e3666879af19886140904f7f429ef083286195982a7588858cfc", "The major problem is with sendmail. -Mark Horton", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19The major problem is wit\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x18"}, + {"b26c38d61519e894480c70c8374ea35aa0ad05b2ae3d6674eec5f52a69305ed4", "Give me a rock, paper and scissors and I will move the world. CCFestoon", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19Give me a rock, paper and scissors a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00$"}, + {"049d5e26d4f10222cd841a119e38bd8d2e0d1129728688449575d4ff42b842c1", "If the enemy is within range, then so are you.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19If the enemy is within \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x17"}, + {"0e116838e3cc1c1a14cd045397e29b4d087aa11b0853fc69ec82e90330d60949", "It's well we cannot hear the screams/That we create in others' dreams.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19It's well we cannot hear the scream\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00#"}, + {"4f7d8eb5bcf11de2a56b971021a444aa4eafd6ecd0f307b5109e4e776cd0fe46", "You remind me of a TV show, but that's all right: I watch it anyway.", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19You remind me of a TV show, but th\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\""}, + {"61c0cc4c4bd8406d5120b3fb4ebc31ce87667c162f29468b3c779675a85aebce", "C is as portable as Stonehedge!!", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19C is as portable\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x10"}, + {"1fb2eb3688093c4a3f80cd87a5547e2ce940a4f923243a79a2a1e242220693ac", "Even if I could be Shakespeare, I think I should still choose to be Faraday. - A. Huxley", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19Even if I could be Shakespeare, I think I sh\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00,"}, + {"395585ce30617b62c80b93e8208ce866d4edc811a177fdb4b82d3911d8696423", "The fugacity of a constituent in a mixture of gases at a given temperature is proportional to its mole fraction. Lewis-Randall Rule", "sha\x03\x93\x14\xc8z\x87\x0e\vo\xf1E\x0f\xa4V\xb2a\x00\x87\xb5ǔ\xfc\xeaV\u009eg\xbc\x17\xb1\x85њem\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00B"}, + {"4f9b189a13d030838269dce846b16a1ce9ce81fe63e65de2f636863336a98fe6", "How can you write a big system without C++? -Paul Glick", "sha\x03j\t\xe6g\xbbg\xae\x85<n\xf3r\xa5O\xf5:Q\x0eR\u007f\x9b\x05h\x8c\x1f\x83٫[\xe0\xcd\x19How can you write a big syst\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, +} + +var golden224 = []sha256Test{ + {"d14a028c2a3a2bc9476102bb288234c415a2b01f828ea62ac5b3e42f", "", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"}, + {"abd37534c7d9a2efb9465de931cd7055ffdb8879563ae98078d6d6d5", "a", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"}, + {"db3cda86d4429a1d39c148989566b38f7bda0156296bd364ba2f878b", "ab", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"}, + {"23097d223405d8228642a477bda255b32aadbce4bda0b3f7e36c9da7", "abc", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01"}, + {"a76654d8e3550e9a2d67a0eeb6c67b220e5885eddd3fde135806e601", "abcd", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4ab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"}, + {"bdd03d560993e675516ba5a50638b6531ac2ac3d5847c61916cfced6", "abcde", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4ab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02"}, + {"7043631cb415556a275a4ebecb802c74ee9f6153908e1792a90b6a98", "abcdef", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4abc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"}, + {"d1884e711701ad81abe0c77a3b0ea12e19ba9af64077286c72fc602d", "abcdefg", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4abc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03"}, + {"17eb7d40f0356f8598e89eafad5f6c759b1f822975d9c9b737c8a517", "abcdefgh", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4abcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04"}, + {"aeb35915346c584db820d2de7af3929ffafef9222a9bcb26516c7334", "abcdefghi", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4abcd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04"}, + {"d35e1e5af29ddb0d7e154357df4ad9842afee527c689ee547f753188", "abcdefghij", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4abcde\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x05"}, + {"19297f1cef7ddc8a7e947f5c5a341e10f7245045e425db67043988d7", "Discard medicine more than two years old.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4Discard medicine mor\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x14"}, + {"0f10c2eb436251f777fbbd125e260d36aecf180411726c7c885f599a", "He who has a shady past knows that nice guys finish last.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4He who has a shady past know\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, + {"4d1842104919f314cad8a3cd20b3cba7e8ed3e7abed62b57441358f6", "I wouldn't marry him with a ten foot pole.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4I wouldn't marry him \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x15"}, + {"a8ba85c6fe0c48fbffc72bbb2f03fcdbc87ae2dc7a56804d1590fb3b", "Free! Free!/A trip/to Mars/for 900/empty jars/Burma Shave", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4Free! Free!/A trip/to Mars/f\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, + {"5543fbab26e67e8885b1a852d567d1cb8b9bfe42e0899584c50449a9", "The days of the digital watch are numbered. -Tom Stoppard", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4The days of the digital watch\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1d"}, + {"65ca107390f5da9efa05d28e57b221657edc7e43a9a18fb15b053ddb", "Nepal premier won't resign.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4Nepal premier\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\r"}, + {"84953962be366305a9cc9b5cd16ed019edc37ac96c0deb3e12cca116", "For every action there is an equal and opposite government program.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4For every action there is an equa\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00!"}, + {"35a189ce987151dfd00b3577583cc6a74b9869eecf894459cb52038d", "His money is twice tainted: 'taint yours and 'taint mine.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4His money is twice tainted: \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, + {"2fc333713983edfd4ef2c0da6fb6d6415afb94987c91e4069eb063e6", "There is no reason for any individual to have a computer in their home. -Ken Olsen, 1977", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4There is no reason for any individual to hav\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00,"}, + {"cbe32d38d577a1b355960a4bc3c659c2dc4670859a19777a875842c4", "It's a tiny change to the code and not completely disgusting. - Bob Manchek", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4It's a tiny change to the code and no\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00%"}, + {"a2dc118ce959e027576413a7b440c875cdc8d40df9141d6ef78a57e1", "size: a.out: bad magic", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4size: a.out\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\f"}, + {"d10787e24052bcff26dc484787a54ed819e4e4511c54890ee977bf81", "The major problem is with sendmail. -Mark Horton", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4The major problem is wit\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x18"}, + {"62efcf16ab8a893acdf2f348aaf06b63039ff1bf55508c830532c9fb", "Give me a rock, paper and scissors and I will move the world. CCFestoon", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4Give me a rock, paper and scissors a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00$"}, + {"3e9b7e4613c59f58665104c5fa86c272db5d3a2ff30df5bb194a5c99", "If the enemy is within range, then so are you.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4If the enemy is within \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x17"}, + {"5999c208b8bdf6d471bb7c359ac5b829e73a8211dff686143a4e7f18", "It's well we cannot hear the screams/That we create in others' dreams.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4It's well we cannot hear the scream\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00#"}, + {"3b2d67ff54eabc4ef737b14edf87c64280ef582bcdf2a6d56908b405", "You remind me of a TV show, but that's all right: I watch it anyway.", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4You remind me of a TV show, but th\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\""}, + {"d0733595d20e4d3d6b5c565a445814d1bbb2fd08b9a3b8ffb97930c6", "C is as portable as Stonehedge!!", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4C is as portable\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x10"}, + {"43fb8aeed8a833175c9295c1165415f98c866ef08a4922959d673507", "Even if I could be Shakespeare, I think I should still choose to be Faraday. - A. Huxley", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4Even if I could be Shakespeare, I think I sh\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00,"}, + {"ec18e66e93afc4fb1604bc2baedbfd20b44c43d76e65c0996d7851c6", "The fugacity of a constituent in a mixture of gases at a given temperature is proportional to its mole fraction. Lewis-Randall Rule", "sha\x02\xea\xc9\xc2e\xddH\x0f\\.\xeb\xc4G\xda\xea\xd5TX\x17\xca3l\xfaV\x9d\x9d\x056\x85&1\rDem\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00B"}, + {"86ed2eaa9c75ba98396e5c9fb2f679ecf0ea2ed1e0ee9ceecb4a9332", "How can you write a big system without C++? -Paul Glick", "sha\x02\xc1\x05\x9e\xd86|\xd5\a0p\xdd\x17\xf7\x0eY9\xff\xc0\v1hX\x15\x11d\xf9\x8f\xa7\xbe\xfaO\xa4How can you write a big syst\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x1c"}, +} + +func TestGolden(t *testing.T) { + for i := 0; i < len(golden); i++ { + g := golden[i] + s := fmt.Sprintf("%x", Sum256([]byte(g.in))) + if s != g.out { + t.Fatalf("Sum256 function: sha256(%s) = %s want %s", g.in, s, g.out) + } + c := New() + for j := 0; j < 3; j++ { + if j < 2 { + io.WriteString(c, g.in) + } else { + io.WriteString(c, g.in[0:len(g.in)/2]) + c.Sum(nil) + io.WriteString(c, g.in[len(g.in)/2:]) + } + s := fmt.Sprintf("%x", c.Sum(nil)) + if s != g.out { + t.Fatalf("sha256[%d](%s) = %s want %s", j, g.in, s, g.out) + } + c.Reset() + } + } + for i := 0; i < len(golden224); i++ { + g := golden224[i] + s := fmt.Sprintf("%x", Sum224([]byte(g.in))) + if s != g.out { + t.Fatalf("Sum224 function: sha224(%s) = %s want %s", g.in, s, g.out) + } + c := New224() + for j := 0; j < 3; j++ { + if j < 2 { + io.WriteString(c, g.in) + } else { + io.WriteString(c, g.in[0:len(g.in)/2]) + c.Sum(nil) + io.WriteString(c, g.in[len(g.in)/2:]) + } + s := fmt.Sprintf("%x", c.Sum(nil)) + if s != g.out { + t.Fatalf("sha224[%d](%s) = %s want %s", j, g.in, s, g.out) + } + c.Reset() + } + } +} + +func TestGoldenMarshal(t *testing.T) { + tests := []struct { + name string + newHash func() hash.Hash + gold []sha256Test + }{ + {"256", New, golden}, + {"224", New224, golden224}, + } + + for _, tt := range tests { + t.Run(tt.name, func(t *testing.T) { + for _, g := range tt.gold { + h := tt.newHash() + h2 := tt.newHash() + + io.WriteString(h, g.in[:len(g.in)/2]) + + state, err := h.(encoding.BinaryMarshaler).MarshalBinary() + if err != nil { + t.Errorf("could not marshal: %v", err) + continue + } + + if string(state) != g.halfState { + t.Errorf("sha%s(%q) state = %q, want %q", tt.name, g.in, state, g.halfState) + continue + } + + if err := h2.(encoding.BinaryUnmarshaler).UnmarshalBinary(state); err != nil { + t.Errorf("could not unmarshal: %v", err) + continue + } + + io.WriteString(h, g.in[len(g.in)/2:]) + io.WriteString(h2, g.in[len(g.in)/2:]) + + if actual, actual2 := h.Sum(nil), h2.Sum(nil); !bytes.Equal(actual, actual2) { + t.Errorf("sha%s(%q) = 0x%x != marshaled 0x%x", tt.name, g.in, actual, actual2) + } + } + }) + } +} + +func TestMarshalTypeMismatch(t *testing.T) { + h1 := New() + h2 := New224() + + state1, err := h1.(encoding.BinaryMarshaler).MarshalBinary() + if err != nil { + t.Errorf("could not marshal: %v", err) + } + + if err := h2.(encoding.BinaryUnmarshaler).UnmarshalBinary(state1); err == nil { + t.Errorf("no error when one was expected") + } +} + +func TestSize(t *testing.T) { + c := New() + if got := c.Size(); got != Size { + t.Errorf("Size = %d; want %d", got, Size) + } + c = New224() + if got := c.Size(); got != Size224 { + t.Errorf("New224.Size = %d; want %d", got, Size224) + } +} + +func TestBlockSize(t *testing.T) { + c := New() + if got := c.BlockSize(); got != BlockSize { + t.Errorf("BlockSize = %d want %d", got, BlockSize) + } +} + +// Tests that blockGeneric (pure Go) and block (in assembly for some architectures) match. +func TestBlockGeneric(t *testing.T) { + gen, asm := New().(*digest), New().(*digest) + buf := make([]byte, BlockSize*20) // arbitrary factor + rand.Read(buf) + blockGeneric(gen, buf) + block(asm, buf) + if *gen != *asm { + t.Error("block and blockGeneric resulted in different states") + } +} + +// Tests for unmarshaling hashes that have hashed a large amount of data +// The initial hash generation is omitted from the test, because it takes a long time. +// The test contains some already-generated states, and their expected sums +// Tests a problem that is outlined in GitHub issue #29517 +// The problem is triggered when an amount of data has been hashed for which +// the data length has a 1 in the 32nd bit. When casted to int, this changes +// the sign of the value, and causes the modulus operation to return a +// different result. +type unmarshalTest struct { + state string + sum string +} + +var largeUnmarshalTests = []unmarshalTest{ + // Data length: 7_115_087_207 + { + state: "sha\x03yX\xaf\xb7\x04*\x8f\xaa\x9bx\xc5#\x1f\xeb\x94\xfdz1\xaf\xfbk֗\n\xc93\xcf\x02\v.\xa5\xe4\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xa8\x17\x9dg", + sum: "f5e06371f0c115e9968455c8e48a318aba548b9f15676fa41de123f7d1c99c55", + }, + + // Data length: 7_070_038_086 + { + state: "sha\x03$\x933u\nV\v\xe2\xf7:0!ʳ\xa4\x13\xd3 6\xdcBB\xb5\x19\xcd=\xc1h\xee=\xb4\x9c@ABCDE\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xa5h8F", + sum: "a280b08df5eba060fcd0eb3d29320bbc038afb95781661f91bbfd0a6fc9fdd6e", + }, + + // Data length: 6_464_878_887 + { + state: "sha\x03\x9f\x12\x87G\xf2\xdf<\x82\xa0\x11/*W\x02&IKWlh\x03\x95\xb1\xab\f\n\xf6Ze\xf9\x1d\x1b\x00\x01\x02\x03\x04\x05\x06\a\b\t\n\v\f\r\x0e\x0f\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f !\"#$%&\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x81V9'", + sum: "d2fffb762f105ab71e2d70069346c44c38c4fe183aad8cfcf5a76397c0457806", + }, +} + +func safeSum(h hash.Hash) (sum []byte, err error) { + defer func() { + if r := recover(); r != nil { + err = fmt.Errorf("sum panic: %v", r) + } + }() + + return h.Sum(nil), nil +} +func TestLargeHashes(t *testing.T) { + for i, test := range largeUnmarshalTests { + + h := New() + if err := h.(encoding.BinaryUnmarshaler).UnmarshalBinary([]byte(test.state)); err != nil { + t.Errorf("test %d could not unmarshal: %v", i, err) + continue + } + + sum, err := safeSum(h) + if err != nil { + t.Errorf("test %d could not sum: %v", i, err) + continue + } + + if fmt.Sprintf("%x", sum) != test.sum { + t.Errorf("test %d sum mismatch: expect %s got %x", i, test.sum, sum) + } + } +} + +func TestAllocations(t *testing.T) { + in := []byte("hello, world!") + out := make([]byte, 0, Size) + h := New() + n := int(testing.AllocsPerRun(10, func() { + h.Reset() + h.Write(in) + out = h.Sum(out[:0]) + })) + if n > 0 { + t.Errorf("allocs = %d, want 0", n) + } +} + +var bench = New() +var buf = make([]byte, 8192) + +func benchmarkSize(b *testing.B, size int) { + b.SetBytes(int64(size)) + sum := make([]byte, bench.Size()) + for i := 0; i < b.N; i++ { + bench.Reset() + bench.Write(buf[:size]) + bench.Sum(sum[:0]) + } +} + +func BenchmarkHash8Bytes(b *testing.B) { + benchmarkSize(b, 8) +} + +func BenchmarkHash1K(b *testing.B) { + benchmarkSize(b, 1024) +} + +func BenchmarkHash8K(b *testing.B) { + benchmarkSize(b, 8192) +} diff --git a/src/crypto/sha256/sha256block.go b/src/crypto/sha256/sha256block.go new file mode 100644 index 0000000..bd2f9da --- /dev/null +++ b/src/crypto/sha256/sha256block.go @@ -0,0 +1,128 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// SHA256 block step. +// In its own file so that a faster assembly or C version +// can be substituted easily. + +package sha256 + +import "math/bits" + +var _K = []uint32{ + 0x428a2f98, + 0x71374491, + 0xb5c0fbcf, + 0xe9b5dba5, + 0x3956c25b, + 0x59f111f1, + 0x923f82a4, + 0xab1c5ed5, + 0xd807aa98, + 0x12835b01, + 0x243185be, + 0x550c7dc3, + 0x72be5d74, + 0x80deb1fe, + 0x9bdc06a7, + 0xc19bf174, + 0xe49b69c1, + 0xefbe4786, + 0x0fc19dc6, + 0x240ca1cc, + 0x2de92c6f, + 0x4a7484aa, + 0x5cb0a9dc, + 0x76f988da, + 0x983e5152, + 0xa831c66d, + 0xb00327c8, + 0xbf597fc7, + 0xc6e00bf3, + 0xd5a79147, + 0x06ca6351, + 0x14292967, + 0x27b70a85, + 0x2e1b2138, + 0x4d2c6dfc, + 0x53380d13, + 0x650a7354, + 0x766a0abb, + 0x81c2c92e, + 0x92722c85, + 0xa2bfe8a1, + 0xa81a664b, + 0xc24b8b70, + 0xc76c51a3, + 0xd192e819, + 0xd6990624, + 0xf40e3585, + 0x106aa070, + 0x19a4c116, + 0x1e376c08, + 0x2748774c, + 0x34b0bcb5, + 0x391c0cb3, + 0x4ed8aa4a, + 0x5b9cca4f, + 0x682e6ff3, + 0x748f82ee, + 0x78a5636f, + 0x84c87814, + 0x8cc70208, + 0x90befffa, + 0xa4506ceb, + 0xbef9a3f7, + 0xc67178f2, +} + +func blockGeneric(dig *digest, p []byte) { + var w [64]uint32 + h0, h1, h2, h3, h4, h5, h6, h7 := dig.h[0], dig.h[1], dig.h[2], dig.h[3], dig.h[4], dig.h[5], dig.h[6], dig.h[7] + for len(p) >= chunk { + // Can interlace the computation of w with the + // rounds below if needed for speed. + for i := 0; i < 16; i++ { + j := i * 4 + w[i] = uint32(p[j])<<24 | uint32(p[j+1])<<16 | uint32(p[j+2])<<8 | uint32(p[j+3]) + } + for i := 16; i < 64; i++ { + v1 := w[i-2] + t1 := (bits.RotateLeft32(v1, -17)) ^ (bits.RotateLeft32(v1, -19)) ^ (v1 >> 10) + v2 := w[i-15] + t2 := (bits.RotateLeft32(v2, -7)) ^ (bits.RotateLeft32(v2, -18)) ^ (v2 >> 3) + w[i] = t1 + w[i-7] + t2 + w[i-16] + } + + a, b, c, d, e, f, g, h := h0, h1, h2, h3, h4, h5, h6, h7 + + for i := 0; i < 64; i++ { + t1 := h + ((bits.RotateLeft32(e, -6)) ^ (bits.RotateLeft32(e, -11)) ^ (bits.RotateLeft32(e, -25))) + ((e & f) ^ (^e & g)) + _K[i] + w[i] + + t2 := ((bits.RotateLeft32(a, -2)) ^ (bits.RotateLeft32(a, -13)) ^ (bits.RotateLeft32(a, -22))) + ((a & b) ^ (a & c) ^ (b & c)) + + h = g + g = f + f = e + e = d + t1 + d = c + c = b + b = a + a = t1 + t2 + } + + h0 += a + h1 += b + h2 += c + h3 += d + h4 += e + h5 += f + h6 += g + h7 += h + + p = p[chunk:] + } + + dig.h[0], dig.h[1], dig.h[2], dig.h[3], dig.h[4], dig.h[5], dig.h[6], dig.h[7] = h0, h1, h2, h3, h4, h5, h6, h7 +} diff --git a/src/crypto/sha256/sha256block_386.s b/src/crypto/sha256/sha256block_386.s new file mode 100644 index 0000000..086a0ab --- /dev/null +++ b/src/crypto/sha256/sha256block_386.s @@ -0,0 +1,283 @@ +// 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. + +// SHA256 block routine. See sha256block.go for Go equivalent. +// +// The algorithm is detailed in FIPS 180-4: +// +// https://csrc.nist.gov/publications/fips/fips180-4/fips-180-4.pdf +// +// Wt = Mt; for 0 <= t <= 15 +// Wt = SIGMA1(Wt-2) + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 63 +// +// a = H0 +// b = H1 +// c = H2 +// d = H3 +// e = H4 +// f = H5 +// g = H6 +// h = H7 +// +// for t = 0 to 63 { +// T1 = h + BIGSIGMA1(e) + Ch(e,f,g) + Kt + Wt +// T2 = BIGSIGMA0(a) + Maj(a,b,c) +// h = g +// g = f +// f = e +// e = d + T1 +// d = c +// c = b +// b = a +// a = T1 + T2 +// } +// +// H0 = a + H0 +// H1 = b + H1 +// H2 = c + H2 +// H3 = d + H3 +// H4 = e + H4 +// H5 = f + H5 +// H6 = g + H6 +// H7 = h + H7 + +// Wt = Mt; for 0 <= t <= 15 +#define MSGSCHEDULE0(index) \ + MOVL (index*4)(SI), AX; \ + BSWAPL AX; \ + MOVL AX, (index*4)(BP) + +// Wt = SIGMA1(Wt-2) + Wt-7 + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 63 +// SIGMA0(x) = ROTR(7,x) XOR ROTR(18,x) XOR SHR(3,x) +// SIGMA1(x) = ROTR(17,x) XOR ROTR(19,x) XOR SHR(10,x) +#define MSGSCHEDULE1(index) \ + MOVL ((index-2)*4)(BP), AX; \ + MOVL AX, CX; \ + RORL $17, AX; \ + MOVL CX, DX; \ + RORL $19, CX; \ + SHRL $10, DX; \ + MOVL ((index-15)*4)(BP), BX; \ + XORL CX, AX; \ + MOVL BX, CX; \ + XORL DX, AX; \ + RORL $7, BX; \ + MOVL CX, DX; \ + SHRL $3, DX; \ + RORL $18, CX; \ + ADDL ((index-7)*4)(BP), AX; \ + XORL CX, BX; \ + XORL DX, BX; \ + ADDL ((index-16)*4)(BP), BX; \ + ADDL BX, AX; \ + MOVL AX, ((index)*4)(BP) + +// Calculate T1 in AX - uses AX, BX, CX and DX registers. +// Wt is passed in AX. +// T1 = h + BIGSIGMA1(e) + Ch(e, f, g) + Kt + Wt +// BIGSIGMA1(x) = ROTR(6,x) XOR ROTR(11,x) XOR ROTR(25,x) +// Ch(x, y, z) = (x AND y) XOR (NOT x AND z) +#define SHA256T1(const, e, f, g, h) \ + MOVL (h*4)(DI), BX; \ + ADDL AX, BX; \ + MOVL (e*4)(DI), AX; \ + ADDL $const, BX; \ + MOVL (e*4)(DI), CX; \ + RORL $6, AX; \ + MOVL (e*4)(DI), DX; \ + RORL $11, CX; \ + XORL CX, AX; \ + MOVL (e*4)(DI), CX; \ + RORL $25, DX; \ + ANDL (f*4)(DI), CX; \ + XORL AX, DX; \ + MOVL (e*4)(DI), AX; \ + NOTL AX; \ + ADDL DX, BX; \ + ANDL (g*4)(DI), AX; \ + XORL CX, AX; \ + ADDL BX, AX + +// Calculate T2 in BX - uses AX, BX, CX and DX registers. +// T2 = BIGSIGMA0(a) + Maj(a, b, c) +// BIGSIGMA0(x) = ROTR(2,x) XOR ROTR(13,x) XOR ROTR(22,x) +// Maj(x, y, z) = (x AND y) XOR (x AND z) XOR (y AND z) +#define SHA256T2(a, b, c) \ + MOVL (a*4)(DI), AX; \ + MOVL (c*4)(DI), BX; \ + RORL $2, AX; \ + MOVL (a*4)(DI), DX; \ + ANDL (b*4)(DI), BX; \ + RORL $13, DX; \ + MOVL (a*4)(DI), CX; \ + ANDL (c*4)(DI), CX; \ + XORL DX, AX; \ + XORL CX, BX; \ + MOVL (a*4)(DI), DX; \ + MOVL (b*4)(DI), CX; \ + RORL $22, DX; \ + ANDL (a*4)(DI), CX; \ + XORL CX, BX; \ + XORL DX, AX; \ + ADDL AX, BX + +// Calculate T1 and T2, then e = d + T1 and a = T1 + T2. +// The values for e and a are stored in d and h, ready for rotation. +#define SHA256ROUND(index, const, a, b, c, d, e, f, g, h) \ + SHA256T1(const, e, f, g, h); \ + MOVL AX, 292(SP); \ + SHA256T2(a, b, c); \ + MOVL 292(SP), AX; \ + ADDL AX, BX; \ + ADDL AX, (d*4)(DI); \ + MOVL BX, (h*4)(DI) + +#define SHA256ROUND0(index, const, a, b, c, d, e, f, g, h) \ + MSGSCHEDULE0(index); \ + SHA256ROUND(index, const, a, b, c, d, e, f, g, h) + +#define SHA256ROUND1(index, const, a, b, c, d, e, f, g, h) \ + MSGSCHEDULE1(index); \ + SHA256ROUND(index, const, a, b, c, d, e, f, g, h) + +TEXT ·block(SB),0,$296-16 + MOVL p_base+4(FP), SI + MOVL p_len+8(FP), DX + SHRL $6, DX + SHLL $6, DX + + LEAL (SI)(DX*1), DI + MOVL DI, 288(SP) + CMPL SI, DI + JEQ end + + LEAL 256(SP), DI // variables + + MOVL dig+0(FP), BP + MOVL (0*4)(BP), AX // a = H0 + MOVL AX, (0*4)(DI) + MOVL (1*4)(BP), BX // b = H1 + MOVL BX, (1*4)(DI) + MOVL (2*4)(BP), CX // c = H2 + MOVL CX, (2*4)(DI) + MOVL (3*4)(BP), DX // d = H3 + MOVL DX, (3*4)(DI) + MOVL (4*4)(BP), AX // e = H4 + MOVL AX, (4*4)(DI) + MOVL (5*4)(BP), BX // f = H5 + MOVL BX, (5*4)(DI) + MOVL (6*4)(BP), CX // g = H6 + MOVL CX, (6*4)(DI) + MOVL (7*4)(BP), DX // h = H7 + MOVL DX, (7*4)(DI) + +loop: + MOVL SP, BP // message schedule + + SHA256ROUND0(0, 0x428a2f98, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND0(1, 0x71374491, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND0(2, 0xb5c0fbcf, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND0(3, 0xe9b5dba5, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND0(4, 0x3956c25b, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND0(5, 0x59f111f1, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND0(6, 0x923f82a4, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND0(7, 0xab1c5ed5, 1, 2, 3, 4, 5, 6, 7, 0) + SHA256ROUND0(8, 0xd807aa98, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND0(9, 0x12835b01, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND0(10, 0x243185be, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND0(11, 0x550c7dc3, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND0(12, 0x72be5d74, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND0(13, 0x80deb1fe, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND0(14, 0x9bdc06a7, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND0(15, 0xc19bf174, 1, 2, 3, 4, 5, 6, 7, 0) + + SHA256ROUND1(16, 0xe49b69c1, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND1(17, 0xefbe4786, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND1(18, 0x0fc19dc6, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND1(19, 0x240ca1cc, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND1(20, 0x2de92c6f, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND1(21, 0x4a7484aa, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND1(22, 0x5cb0a9dc, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND1(23, 0x76f988da, 1, 2, 3, 4, 5, 6, 7, 0) + SHA256ROUND1(24, 0x983e5152, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND1(25, 0xa831c66d, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND1(26, 0xb00327c8, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND1(27, 0xbf597fc7, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND1(28, 0xc6e00bf3, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND1(29, 0xd5a79147, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND1(30, 0x06ca6351, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND1(31, 0x14292967, 1, 2, 3, 4, 5, 6, 7, 0) + SHA256ROUND1(32, 0x27b70a85, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND1(33, 0x2e1b2138, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND1(34, 0x4d2c6dfc, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND1(35, 0x53380d13, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND1(36, 0x650a7354, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND1(37, 0x766a0abb, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND1(38, 0x81c2c92e, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND1(39, 0x92722c85, 1, 2, 3, 4, 5, 6, 7, 0) + SHA256ROUND1(40, 0xa2bfe8a1, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND1(41, 0xa81a664b, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND1(42, 0xc24b8b70, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND1(43, 0xc76c51a3, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND1(44, 0xd192e819, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND1(45, 0xd6990624, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND1(46, 0xf40e3585, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND1(47, 0x106aa070, 1, 2, 3, 4, 5, 6, 7, 0) + SHA256ROUND1(48, 0x19a4c116, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND1(49, 0x1e376c08, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND1(50, 0x2748774c, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND1(51, 0x34b0bcb5, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND1(52, 0x391c0cb3, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND1(53, 0x4ed8aa4a, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND1(54, 0x5b9cca4f, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND1(55, 0x682e6ff3, 1, 2, 3, 4, 5, 6, 7, 0) + SHA256ROUND1(56, 0x748f82ee, 0, 1, 2, 3, 4, 5, 6, 7) + SHA256ROUND1(57, 0x78a5636f, 7, 0, 1, 2, 3, 4, 5, 6) + SHA256ROUND1(58, 0x84c87814, 6, 7, 0, 1, 2, 3, 4, 5) + SHA256ROUND1(59, 0x8cc70208, 5, 6, 7, 0, 1, 2, 3, 4) + SHA256ROUND1(60, 0x90befffa, 4, 5, 6, 7, 0, 1, 2, 3) + SHA256ROUND1(61, 0xa4506ceb, 3, 4, 5, 6, 7, 0, 1, 2) + SHA256ROUND1(62, 0xbef9a3f7, 2, 3, 4, 5, 6, 7, 0, 1) + SHA256ROUND1(63, 0xc67178f2, 1, 2, 3, 4, 5, 6, 7, 0) + + MOVL dig+0(FP), BP + MOVL (0*4)(BP), AX // H0 = a + H0 + ADDL (0*4)(DI), AX + MOVL AX, (0*4)(DI) + MOVL AX, (0*4)(BP) + MOVL (1*4)(BP), BX // H1 = b + H1 + ADDL (1*4)(DI), BX + MOVL BX, (1*4)(DI) + MOVL BX, (1*4)(BP) + MOVL (2*4)(BP), CX // H2 = c + H2 + ADDL (2*4)(DI), CX + MOVL CX, (2*4)(DI) + MOVL CX, (2*4)(BP) + MOVL (3*4)(BP), DX // H3 = d + H3 + ADDL (3*4)(DI), DX + MOVL DX, (3*4)(DI) + MOVL DX, (3*4)(BP) + MOVL (4*4)(BP), AX // H4 = e + H4 + ADDL (4*4)(DI), AX + MOVL AX, (4*4)(DI) + MOVL AX, (4*4)(BP) + MOVL (5*4)(BP), BX // H5 = f + H5 + ADDL (5*4)(DI), BX + MOVL BX, (5*4)(DI) + MOVL BX, (5*4)(BP) + MOVL (6*4)(BP), CX // H6 = g + H6 + ADDL (6*4)(DI), CX + MOVL CX, (6*4)(DI) + MOVL CX, (6*4)(BP) + MOVL (7*4)(BP), DX // H7 = h + H7 + ADDL (7*4)(DI), DX + MOVL DX, (7*4)(DI) + MOVL DX, (7*4)(BP) + + ADDL $64, SI + CMPL SI, 288(SP) + JB loop + +end: + RET diff --git a/src/crypto/sha256/sha256block_amd64.go b/src/crypto/sha256/sha256block_amd64.go new file mode 100644 index 0000000..27464e2 --- /dev/null +++ b/src/crypto/sha256/sha256block_amd64.go @@ -0,0 +1,9 @@ +// 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 sha256 + +import "internal/cpu" + +var useAVX2 = cpu.X86.HasAVX2 && cpu.X86.HasBMI2 diff --git a/src/crypto/sha256/sha256block_amd64.s b/src/crypto/sha256/sha256block_amd64.s new file mode 100644 index 0000000..f6af47c --- /dev/null +++ b/src/crypto/sha256/sha256block_amd64.s @@ -0,0 +1,1031 @@ +// 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. + +#include "textflag.h" + +// SHA256 block routine. See sha256block.go for Go equivalent. +// +// The algorithm is detailed in FIPS 180-4: +// +// https://csrc.nist.gov/publications/fips/fips180-4/fips-180-4.pdf + +// The avx2-version is described in an Intel White-Paper: +// "Fast SHA-256 Implementations on Intel Architecture Processors" +// To find it, surf to http://www.intel.com/p/en_US/embedded +// and search for that title. +// AVX2 version by Intel, same algorithm as code in Linux kernel: +// https://github.com/torvalds/linux/blob/master/arch/x86/crypto/sha256-avx2-asm.S +// by +// James Guilford <james.guilford@intel.com> +// Kirk Yap <kirk.s.yap@intel.com> +// Tim Chen <tim.c.chen@linux.intel.com> + +// Wt = Mt; for 0 <= t <= 15 +// Wt = SIGMA1(Wt-2) + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 63 +// +// a = H0 +// b = H1 +// c = H2 +// d = H3 +// e = H4 +// f = H5 +// g = H6 +// h = H7 +// +// for t = 0 to 63 { +// T1 = h + BIGSIGMA1(e) + Ch(e,f,g) + Kt + Wt +// T2 = BIGSIGMA0(a) + Maj(a,b,c) +// h = g +// g = f +// f = e +// e = d + T1 +// d = c +// c = b +// b = a +// a = T1 + T2 +// } +// +// H0 = a + H0 +// H1 = b + H1 +// H2 = c + H2 +// H3 = d + H3 +// H4 = e + H4 +// H5 = f + H5 +// H6 = g + H6 +// H7 = h + H7 + +// Wt = Mt; for 0 <= t <= 15 +#define MSGSCHEDULE0(index) \ + MOVL (index*4)(SI), AX; \ + BSWAPL AX; \ + MOVL AX, (index*4)(BP) + +// Wt = SIGMA1(Wt-2) + Wt-7 + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 63 +// SIGMA0(x) = ROTR(7,x) XOR ROTR(18,x) XOR SHR(3,x) +// SIGMA1(x) = ROTR(17,x) XOR ROTR(19,x) XOR SHR(10,x) +#define MSGSCHEDULE1(index) \ + MOVL ((index-2)*4)(BP), AX; \ + MOVL AX, CX; \ + RORL $17, AX; \ + MOVL CX, DX; \ + RORL $19, CX; \ + SHRL $10, DX; \ + MOVL ((index-15)*4)(BP), BX; \ + XORL CX, AX; \ + MOVL BX, CX; \ + XORL DX, AX; \ + RORL $7, BX; \ + MOVL CX, DX; \ + SHRL $3, DX; \ + RORL $18, CX; \ + ADDL ((index-7)*4)(BP), AX; \ + XORL CX, BX; \ + XORL DX, BX; \ + ADDL ((index-16)*4)(BP), BX; \ + ADDL BX, AX; \ + MOVL AX, ((index)*4)(BP) + +// Calculate T1 in AX - uses AX, CX and DX registers. +// h is also used as an accumulator. Wt is passed in AX. +// T1 = h + BIGSIGMA1(e) + Ch(e, f, g) + Kt + Wt +// BIGSIGMA1(x) = ROTR(6,x) XOR ROTR(11,x) XOR ROTR(25,x) +// Ch(x, y, z) = (x AND y) XOR (NOT x AND z) +#define SHA256T1(const, e, f, g, h) \ + ADDL AX, h; \ + MOVL e, AX; \ + ADDL $const, h; \ + MOVL e, CX; \ + RORL $6, AX; \ + MOVL e, DX; \ + RORL $11, CX; \ + XORL CX, AX; \ + MOVL e, CX; \ + RORL $25, DX; \ + ANDL f, CX; \ + XORL AX, DX; \ + MOVL e, AX; \ + NOTL AX; \ + ADDL DX, h; \ + ANDL g, AX; \ + XORL CX, AX; \ + ADDL h, AX + +// Calculate T2 in BX - uses BX, CX, DX and DI registers. +// T2 = BIGSIGMA0(a) + Maj(a, b, c) +// BIGSIGMA0(x) = ROTR(2,x) XOR ROTR(13,x) XOR ROTR(22,x) +// Maj(x, y, z) = (x AND y) XOR (x AND z) XOR (y AND z) +#define SHA256T2(a, b, c) \ + MOVL a, DI; \ + MOVL c, BX; \ + RORL $2, DI; \ + MOVL a, DX; \ + ANDL b, BX; \ + RORL $13, DX; \ + MOVL a, CX; \ + ANDL c, CX; \ + XORL DX, DI; \ + XORL CX, BX; \ + MOVL a, DX; \ + MOVL b, CX; \ + RORL $22, DX; \ + ANDL a, CX; \ + XORL CX, BX; \ + XORL DX, DI; \ + ADDL DI, BX + +// Calculate T1 and T2, then e = d + T1 and a = T1 + T2. +// The values for e and a are stored in d and h, ready for rotation. +#define SHA256ROUND(index, const, a, b, c, d, e, f, g, h) \ + SHA256T1(const, e, f, g, h); \ + SHA256T2(a, b, c); \ + MOVL BX, h; \ + ADDL AX, d; \ + ADDL AX, h + +#define SHA256ROUND0(index, const, a, b, c, d, e, f, g, h) \ + MSGSCHEDULE0(index); \ + SHA256ROUND(index, const, a, b, c, d, e, f, g, h) + +#define SHA256ROUND1(index, const, a, b, c, d, e, f, g, h) \ + MSGSCHEDULE1(index); \ + SHA256ROUND(index, const, a, b, c, d, e, f, g, h) + + +// Definitions for AVX2 version + +// addm (mem), reg +// Add reg to mem using reg-mem add and store +#define addm(P1, P2) \ + ADDL P2, P1; \ + MOVL P1, P2 + +#define XDWORD0 Y4 +#define XDWORD1 Y5 +#define XDWORD2 Y6 +#define XDWORD3 Y7 + +#define XWORD0 X4 +#define XWORD1 X5 +#define XWORD2 X6 +#define XWORD3 X7 + +#define XTMP0 Y0 +#define XTMP1 Y1 +#define XTMP2 Y2 +#define XTMP3 Y3 +#define XTMP4 Y8 +#define XTMP5 Y11 + +#define XFER Y9 + +#define BYTE_FLIP_MASK Y13 // mask to convert LE -> BE +#define X_BYTE_FLIP_MASK X13 + +#define NUM_BYTES DX +#define INP DI + +#define CTX SI // Beginning of digest in memory (a, b, c, ... , h) + +#define a AX +#define b BX +#define c CX +#define d R8 +#define e DX +#define f R9 +#define g R10 +#define h R11 + +#define old_h R11 + +#define TBL BP + +#define SRND SI // SRND is same register as CTX + +#define T1 R12 + +#define y0 R13 +#define y1 R14 +#define y2 R15 +#define y3 DI + +// Offsets +#define XFER_SIZE 2*64*4 +#define INP_END_SIZE 8 +#define INP_SIZE 8 + +#define _XFER 0 +#define _INP_END _XFER + XFER_SIZE +#define _INP _INP_END + INP_END_SIZE +#define STACK_SIZE _INP + INP_SIZE + +#define ROUND_AND_SCHED_N_0(disp, a, b, c, d, e, f, g, h, XDWORD0, XDWORD1, XDWORD2, XDWORD3) \ + ; \ // ############################# RND N + 0 ############################// + MOVL a, y3; \ // y3 = a // MAJA + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + ; \ + ADDL (disp + 0*4)(SP)(SRND*1), h; \ // h = k + w + h // disp = k + w + ORL c, y3; \ // y3 = a|c // MAJA + VPALIGNR $4, XDWORD2, XDWORD3, XTMP0; \ // XTMP0 = W[-7] + MOVL f, y2; \ // y2 = f // CH + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + XORL g, y2; \ // y2 = f^g // CH + VPADDD XDWORD0, XTMP0, XTMP0; \ // XTMP0 = W[-7] + W[-16] // y1 = (e >> 6) // S1 + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + ; \ + ANDL e, y2; \ // y2 = (f^g)&e // CH + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + ADDL h, d; \ // d = k + w + h + d // -- + ; \ + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + VPALIGNR $4, XDWORD0, XDWORD1, XTMP1; \ // XTMP1 = W[-15] + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + ; \ + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + VPSRLD $7, XTMP1, XTMP2; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL c, T1; \ // T1 = a&c // MAJB + ; \ + ADDL y0, y2; \ // y2 = S1 + CH // -- + VPSLLD $(32-7), XTMP1, XTMP3; \ + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ; \ + ADDL y2, d; \ // d = k + w + h + d + S1 + CH = d + t1 // -- + VPOR XTMP2, XTMP3, XTMP3; \ // XTMP3 = W[-15] ror 7 + ; \ + VPSRLD $18, XTMP1, XTMP2; \ + ADDL y2, h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + ADDL y3, h // h = t1 + S0 + MAJ // -- + +#define ROUND_AND_SCHED_N_1(disp, a, b, c, d, e, f, g, h, XDWORD0, XDWORD1, XDWORD2, XDWORD3) \ + ; \ // ################################### RND N + 1 ############################ + ; \ + MOVL a, y3; \ // y3 = a // MAJA + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + ADDL (disp + 1*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ORL c, y3; \ // y3 = a|c // MAJA + ; \ + VPSRLD $3, XTMP1, XTMP4; \ // XTMP4 = W[-15] >> 3 + MOVL f, y2; \ // y2 = f // CH + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + XORL g, y2; \ // y2 = f^g // CH + ; \ + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + ANDL e, y2; \ // y2 = (f^g)&e // CH + ADDL h, d; \ // d = k + w + h + d // -- + ; \ + VPSLLD $(32-18), XTMP1, XTMP1; \ + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + ; \ + VPXOR XTMP1, XTMP3, XTMP3; \ + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + ; \ + VPXOR XTMP2, XTMP3, XTMP3; \ // XTMP3 = W[-15] ror 7 ^ W[-15] ror 18 + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL c, T1; \ // T1 = a&c // MAJB + ADDL y0, y2; \ // y2 = S1 + CH // -- + ; \ + VPXOR XTMP4, XTMP3, XTMP1; \ // XTMP1 = s0 + VPSHUFD $0xFA, XDWORD3, XTMP2; \ // XTMP2 = W[-2] {BBAA} + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ; \ + VPADDD XTMP1, XTMP0, XTMP0; \ // XTMP0 = W[-16] + W[-7] + s0 + ADDL y2, d; \ // d = k + w + h + d + S1 + CH = d + t1 // -- + ADDL y2, h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + ADDL y3, h; \ // h = t1 + S0 + MAJ // -- + ; \ + VPSRLD $10, XTMP2, XTMP4 // XTMP4 = W[-2] >> 10 {BBAA} + +#define ROUND_AND_SCHED_N_2(disp, a, b, c, d, e, f, g, h, XDWORD0, XDWORD1, XDWORD2, XDWORD3) \ + ; \ // ################################### RND N + 2 ############################ + ; \ + MOVL a, y3; \ // y3 = a // MAJA + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + ADDL (disp + 2*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ; \ + VPSRLQ $19, XTMP2, XTMP3; \ // XTMP3 = W[-2] ror 19 {xBxA} + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + ORL c, y3; \ // y3 = a|c // MAJA + MOVL f, y2; \ // y2 = f // CH + XORL g, y2; \ // y2 = f^g // CH + ; \ + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + VPSRLQ $17, XTMP2, XTMP2; \ // XTMP2 = W[-2] ror 17 {xBxA} + ANDL e, y2; \ // y2 = (f^g)&e // CH + ; \ + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + VPXOR XTMP3, XTMP2, XTMP2; \ + ADDL h, d; \ // d = k + w + h + d // -- + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + VPXOR XTMP2, XTMP4, XTMP4; \ // XTMP4 = s1 {xBxA} + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + ; \ + VPSHUFB shuff_00BA<>(SB), XTMP4, XTMP4;\ // XTMP4 = s1 {00BA} + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + VPADDD XTMP4, XTMP0, XTMP0; \ // XTMP0 = {..., ..., W[1], W[0]} + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL c, T1; \ // T1 = a&c // MAJB + ADDL y0, y2; \ // y2 = S1 + CH // -- + VPSHUFD $80, XTMP0, XTMP2; \ // XTMP2 = W[-2] {DDCC} + ; \ + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ADDL y2, d; \ // d = k + w + h + d + S1 + CH = d + t1 // -- + ADDL y2, h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + ; \ + ADDL y3, h // h = t1 + S0 + MAJ // -- + +#define ROUND_AND_SCHED_N_3(disp, a, b, c, d, e, f, g, h, XDWORD0, XDWORD1, XDWORD2, XDWORD3) \ + ; \ // ################################### RND N + 3 ############################ + ; \ + MOVL a, y3; \ // y3 = a // MAJA + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + ADDL (disp + 3*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ORL c, y3; \ // y3 = a|c // MAJA + ; \ + VPSRLD $10, XTMP2, XTMP5; \ // XTMP5 = W[-2] >> 10 {DDCC} + MOVL f, y2; \ // y2 = f // CH + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + XORL g, y2; \ // y2 = f^g // CH + ; \ + VPSRLQ $19, XTMP2, XTMP3; \ // XTMP3 = W[-2] ror 19 {xDxC} + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + ANDL e, y2; \ // y2 = (f^g)&e // CH + ADDL h, d; \ // d = k + w + h + d // -- + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + ; \ + VPSRLQ $17, XTMP2, XTMP2; \ // XTMP2 = W[-2] ror 17 {xDxC} + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + ; \ + VPXOR XTMP3, XTMP2, XTMP2; \ + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + ADDL y0, y2; \ // y2 = S1 + CH // -- + ; \ + VPXOR XTMP2, XTMP5, XTMP5; \ // XTMP5 = s1 {xDxC} + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + ADDL y2, d; \ // d = k + w + h + d + S1 + CH = d + t1 // -- + ; \ + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + ; \ + VPSHUFB shuff_DC00<>(SB), XTMP5, XTMP5;\ // XTMP5 = s1 {DC00} + ; \ + VPADDD XTMP0, XTMP5, XDWORD0; \ // XDWORD0 = {W[3], W[2], W[1], W[0]} + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL c, T1; \ // T1 = a&c // MAJB + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ; \ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ADDL y2, h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + ADDL y3, h // h = t1 + S0 + MAJ // -- + +#define DO_ROUND_N_0(disp, a, b, c, d, e, f, g, h, old_h) \ + ; \ // ################################### RND N + 0 ########################### + MOVL f, y2; \ // y2 = f // CH + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + XORL g, y2; \ // y2 = f^g // CH + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + ANDL e, y2; \ // y2 = (f^g)&e // CH + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + MOVL a, y3; \ // y3 = a // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + ADDL (disp + 0*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ORL c, y3; \ // y3 = a|c // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + ANDL c, T1; \ // T1 = a&c // MAJB + ADDL y0, y2; \ // y2 = S1 + CH // -- + ; \ + ADDL h, d; \ // d = k + w + h + d // -- + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ADDL y2, d // d = k + w + h + d + S1 + CH = d + t1 // -- + +#define DO_ROUND_N_1(disp, a, b, c, d, e, f, g, h, old_h) \ + ; \ // ################################### RND N + 1 ########################### + ADDL y2, old_h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0 // -- + MOVL f, y2; \ // y2 = f // CH + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + XORL g, y2; \ // y2 = f^g // CH + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + ANDL e, y2; \ // y2 = (f^g)&e // CH + ADDL y3, old_h; \ // h = t1 + S0 + MAJ // -- + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + MOVL a, y3; \ // y3 = a // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + ADDL (disp + 1*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ORL c, y3; \ // y3 = a|c // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + ANDL c, T1; \ // T1 = a&c // MAJB + ADDL y0, y2; \ // y2 = S1 + CH // -- + ; \ + ADDL h, d; \ // d = k + w + h + d // -- + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ; \ + ADDL y2, d // d = k + w + h + d + S1 + CH = d + t1 // -- + +#define DO_ROUND_N_2(disp, a, b, c, d, e, f, g, h, old_h) \ + ; \ // ################################### RND N + 2 ############################## + ADDL y2, old_h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + MOVL f, y2; \ // y2 = f // CH + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + XORL g, y2; \ // y2 = f^g // CH + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + ANDL e, y2; \ // y2 = (f^g)&e // CH + ADDL y3, old_h; \ // h = t1 + S0 + MAJ // -- + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + MOVL a, y3; \ // y3 = a // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + ADDL (disp + 2*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ORL c, y3; \ // y3 = a|c // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + ANDL c, T1; \ // T1 = a&c // MAJB + ADDL y0, y2; \ // y2 = S1 + CH // -- + ; \ + ADDL h, d; \ // d = k + w + h + d // -- + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ; \ + ADDL y2, d // d = k + w + h + d + S1 + CH = d + t1 // -- + +#define DO_ROUND_N_3(disp, a, b, c, d, e, f, g, h, old_h) \ + ; \ // ################################### RND N + 3 ########################### + ADDL y2, old_h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + MOVL f, y2; \ // y2 = f // CH + RORXL $25, e, y0; \ // y0 = e >> 25 // S1A + RORXL $11, e, y1; \ // y1 = e >> 11 // S1B + XORL g, y2; \ // y2 = f^g // CH + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) // S1 + RORXL $6, e, y1; \ // y1 = (e >> 6) // S1 + ANDL e, y2; \ // y2 = (f^g)&e // CH + ADDL y3, old_h; \ // h = t1 + S0 + MAJ // -- + ; \ + XORL y1, y0; \ // y0 = (e>>25) ^ (e>>11) ^ (e>>6) // S1 + RORXL $13, a, T1; \ // T1 = a >> 13 // S0B + XORL g, y2; \ // y2 = CH = ((f^g)&e)^g // CH + RORXL $22, a, y1; \ // y1 = a >> 22 // S0A + MOVL a, y3; \ // y3 = a // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) // S0 + RORXL $2, a, T1; \ // T1 = (a >> 2) // S0 + ADDL (disp + 3*4)(SP)(SRND*1), h; \ // h = k + w + h // -- + ORL c, y3; \ // y3 = a|c // MAJA + ; \ + XORL T1, y1; \ // y1 = (a>>22) ^ (a>>13) ^ (a>>2) // S0 + MOVL a, T1; \ // T1 = a // MAJB + ANDL b, y3; \ // y3 = (a|c)&b // MAJA + ANDL c, T1; \ // T1 = a&c // MAJB + ADDL y0, y2; \ // y2 = S1 + CH // -- + ; \ + ADDL h, d; \ // d = k + w + h + d // -- + ORL T1, y3; \ // y3 = MAJ = (a|c)&b)|(a&c) // MAJ + ADDL y1, h; \ // h = k + w + h + S0 // -- + ; \ + ADDL y2, d; \ // d = k + w + h + d + S1 + CH = d + t1 // -- + ; \ + ADDL y2, h; \ // h = k + w + h + S0 + S1 + CH = t1 + S0// -- + ; \ + ADDL y3, h // h = t1 + S0 + MAJ // -- + +TEXT ·block(SB), 0, $536-32 + CMPB ·useAVX2(SB), $1 + JE avx2 + + MOVQ p_base+8(FP), SI + MOVQ p_len+16(FP), DX + SHRQ $6, DX + SHLQ $6, DX + + LEAQ (SI)(DX*1), DI + MOVQ DI, 256(SP) + CMPQ SI, DI + JEQ end + + MOVQ dig+0(FP), BP + MOVL (0*4)(BP), R8 // a = H0 + MOVL (1*4)(BP), R9 // b = H1 + MOVL (2*4)(BP), R10 // c = H2 + MOVL (3*4)(BP), R11 // d = H3 + MOVL (4*4)(BP), R12 // e = H4 + MOVL (5*4)(BP), R13 // f = H5 + MOVL (6*4)(BP), R14 // g = H6 + MOVL (7*4)(BP), R15 // h = H7 + +loop: + MOVQ SP, BP + + SHA256ROUND0(0, 0x428a2f98, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND0(1, 0x71374491, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND0(2, 0xb5c0fbcf, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND0(3, 0xe9b5dba5, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND0(4, 0x3956c25b, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND0(5, 0x59f111f1, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND0(6, 0x923f82a4, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND0(7, 0xab1c5ed5, R9, R10, R11, R12, R13, R14, R15, R8) + SHA256ROUND0(8, 0xd807aa98, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND0(9, 0x12835b01, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND0(10, 0x243185be, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND0(11, 0x550c7dc3, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND0(12, 0x72be5d74, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND0(13, 0x80deb1fe, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND0(14, 0x9bdc06a7, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND0(15, 0xc19bf174, R9, R10, R11, R12, R13, R14, R15, R8) + + SHA256ROUND1(16, 0xe49b69c1, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND1(17, 0xefbe4786, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND1(18, 0x0fc19dc6, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND1(19, 0x240ca1cc, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND1(20, 0x2de92c6f, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND1(21, 0x4a7484aa, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND1(22, 0x5cb0a9dc, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND1(23, 0x76f988da, R9, R10, R11, R12, R13, R14, R15, R8) + SHA256ROUND1(24, 0x983e5152, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND1(25, 0xa831c66d, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND1(26, 0xb00327c8, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND1(27, 0xbf597fc7, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND1(28, 0xc6e00bf3, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND1(29, 0xd5a79147, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND1(30, 0x06ca6351, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND1(31, 0x14292967, R9, R10, R11, R12, R13, R14, R15, R8) + SHA256ROUND1(32, 0x27b70a85, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND1(33, 0x2e1b2138, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND1(34, 0x4d2c6dfc, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND1(35, 0x53380d13, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND1(36, 0x650a7354, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND1(37, 0x766a0abb, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND1(38, 0x81c2c92e, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND1(39, 0x92722c85, R9, R10, R11, R12, R13, R14, R15, R8) + SHA256ROUND1(40, 0xa2bfe8a1, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND1(41, 0xa81a664b, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND1(42, 0xc24b8b70, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND1(43, 0xc76c51a3, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND1(44, 0xd192e819, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND1(45, 0xd6990624, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND1(46, 0xf40e3585, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND1(47, 0x106aa070, R9, R10, R11, R12, R13, R14, R15, R8) + SHA256ROUND1(48, 0x19a4c116, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND1(49, 0x1e376c08, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND1(50, 0x2748774c, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND1(51, 0x34b0bcb5, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND1(52, 0x391c0cb3, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND1(53, 0x4ed8aa4a, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND1(54, 0x5b9cca4f, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND1(55, 0x682e6ff3, R9, R10, R11, R12, R13, R14, R15, R8) + SHA256ROUND1(56, 0x748f82ee, R8, R9, R10, R11, R12, R13, R14, R15) + SHA256ROUND1(57, 0x78a5636f, R15, R8, R9, R10, R11, R12, R13, R14) + SHA256ROUND1(58, 0x84c87814, R14, R15, R8, R9, R10, R11, R12, R13) + SHA256ROUND1(59, 0x8cc70208, R13, R14, R15, R8, R9, R10, R11, R12) + SHA256ROUND1(60, 0x90befffa, R12, R13, R14, R15, R8, R9, R10, R11) + SHA256ROUND1(61, 0xa4506ceb, R11, R12, R13, R14, R15, R8, R9, R10) + SHA256ROUND1(62, 0xbef9a3f7, R10, R11, R12, R13, R14, R15, R8, R9) + SHA256ROUND1(63, 0xc67178f2, R9, R10, R11, R12, R13, R14, R15, R8) + + MOVQ dig+0(FP), BP + ADDL (0*4)(BP), R8 // H0 = a + H0 + MOVL R8, (0*4)(BP) + ADDL (1*4)(BP), R9 // H1 = b + H1 + MOVL R9, (1*4)(BP) + ADDL (2*4)(BP), R10 // H2 = c + H2 + MOVL R10, (2*4)(BP) + ADDL (3*4)(BP), R11 // H3 = d + H3 + MOVL R11, (3*4)(BP) + ADDL (4*4)(BP), R12 // H4 = e + H4 + MOVL R12, (4*4)(BP) + ADDL (5*4)(BP), R13 // H5 = f + H5 + MOVL R13, (5*4)(BP) + ADDL (6*4)(BP), R14 // H6 = g + H6 + MOVL R14, (6*4)(BP) + ADDL (7*4)(BP), R15 // H7 = h + H7 + MOVL R15, (7*4)(BP) + + ADDQ $64, SI + CMPQ SI, 256(SP) + JB loop + +end: + RET + +avx2: + MOVQ dig+0(FP), CTX // d.h[8] + MOVQ p_base+8(FP), INP + MOVQ p_len+16(FP), NUM_BYTES + + LEAQ -64(INP)(NUM_BYTES*1), NUM_BYTES // Pointer to the last block + MOVQ NUM_BYTES, _INP_END(SP) + + CMPQ NUM_BYTES, INP + JE avx2_only_one_block + + // Load initial digest + MOVL 0(CTX), a // a = H0 + MOVL 4(CTX), b // b = H1 + MOVL 8(CTX), c // c = H2 + MOVL 12(CTX), d // d = H3 + MOVL 16(CTX), e // e = H4 + MOVL 20(CTX), f // f = H5 + MOVL 24(CTX), g // g = H6 + MOVL 28(CTX), h // h = H7 + +avx2_loop0: // at each iteration works with one block (512 bit) + + VMOVDQU (0*32)(INP), XTMP0 + VMOVDQU (1*32)(INP), XTMP1 + VMOVDQU (2*32)(INP), XTMP2 + VMOVDQU (3*32)(INP), XTMP3 + + VMOVDQU flip_mask<>(SB), BYTE_FLIP_MASK + + // Apply Byte Flip Mask: LE -> BE + VPSHUFB BYTE_FLIP_MASK, XTMP0, XTMP0 + VPSHUFB BYTE_FLIP_MASK, XTMP1, XTMP1 + VPSHUFB BYTE_FLIP_MASK, XTMP2, XTMP2 + VPSHUFB BYTE_FLIP_MASK, XTMP3, XTMP3 + + // Transpose data into high/low parts + VPERM2I128 $0x20, XTMP2, XTMP0, XDWORD0 // w3, w2, w1, w0 + VPERM2I128 $0x31, XTMP2, XTMP0, XDWORD1 // w7, w6, w5, w4 + VPERM2I128 $0x20, XTMP3, XTMP1, XDWORD2 // w11, w10, w9, w8 + VPERM2I128 $0x31, XTMP3, XTMP1, XDWORD3 // w15, w14, w13, w12 + + MOVQ $K256<>(SB), TBL // Loading address of table with round-specific constants + +avx2_last_block_enter: + ADDQ $64, INP + MOVQ INP, _INP(SP) + XORQ SRND, SRND + +avx2_loop1: // for w0 - w47 + // Do 4 rounds and scheduling + VPADDD 0*32(TBL)(SRND*1), XDWORD0, XFER + VMOVDQU XFER, (_XFER + 0*32)(SP)(SRND*1) + ROUND_AND_SCHED_N_0(_XFER + 0*32, a, b, c, d, e, f, g, h, XDWORD0, XDWORD1, XDWORD2, XDWORD3) + ROUND_AND_SCHED_N_1(_XFER + 0*32, h, a, b, c, d, e, f, g, XDWORD0, XDWORD1, XDWORD2, XDWORD3) + ROUND_AND_SCHED_N_2(_XFER + 0*32, g, h, a, b, c, d, e, f, XDWORD0, XDWORD1, XDWORD2, XDWORD3) + ROUND_AND_SCHED_N_3(_XFER + 0*32, f, g, h, a, b, c, d, e, XDWORD0, XDWORD1, XDWORD2, XDWORD3) + + // Do 4 rounds and scheduling + VPADDD 1*32(TBL)(SRND*1), XDWORD1, XFER + VMOVDQU XFER, (_XFER + 1*32)(SP)(SRND*1) + ROUND_AND_SCHED_N_0(_XFER + 1*32, e, f, g, h, a, b, c, d, XDWORD1, XDWORD2, XDWORD3, XDWORD0) + ROUND_AND_SCHED_N_1(_XFER + 1*32, d, e, f, g, h, a, b, c, XDWORD1, XDWORD2, XDWORD3, XDWORD0) + ROUND_AND_SCHED_N_2(_XFER + 1*32, c, d, e, f, g, h, a, b, XDWORD1, XDWORD2, XDWORD3, XDWORD0) + ROUND_AND_SCHED_N_3(_XFER + 1*32, b, c, d, e, f, g, h, a, XDWORD1, XDWORD2, XDWORD3, XDWORD0) + + // Do 4 rounds and scheduling + VPADDD 2*32(TBL)(SRND*1), XDWORD2, XFER + VMOVDQU XFER, (_XFER + 2*32)(SP)(SRND*1) + ROUND_AND_SCHED_N_0(_XFER + 2*32, a, b, c, d, e, f, g, h, XDWORD2, XDWORD3, XDWORD0, XDWORD1) + ROUND_AND_SCHED_N_1(_XFER + 2*32, h, a, b, c, d, e, f, g, XDWORD2, XDWORD3, XDWORD0, XDWORD1) + ROUND_AND_SCHED_N_2(_XFER + 2*32, g, h, a, b, c, d, e, f, XDWORD2, XDWORD3, XDWORD0, XDWORD1) + ROUND_AND_SCHED_N_3(_XFER + 2*32, f, g, h, a, b, c, d, e, XDWORD2, XDWORD3, XDWORD0, XDWORD1) + + // Do 4 rounds and scheduling + VPADDD 3*32(TBL)(SRND*1), XDWORD3, XFER + VMOVDQU XFER, (_XFER + 3*32)(SP)(SRND*1) + ROUND_AND_SCHED_N_0(_XFER + 3*32, e, f, g, h, a, b, c, d, XDWORD3, XDWORD0, XDWORD1, XDWORD2) + ROUND_AND_SCHED_N_1(_XFER + 3*32, d, e, f, g, h, a, b, c, XDWORD3, XDWORD0, XDWORD1, XDWORD2) + ROUND_AND_SCHED_N_2(_XFER + 3*32, c, d, e, f, g, h, a, b, XDWORD3, XDWORD0, XDWORD1, XDWORD2) + ROUND_AND_SCHED_N_3(_XFER + 3*32, b, c, d, e, f, g, h, a, XDWORD3, XDWORD0, XDWORD1, XDWORD2) + + ADDQ $4*32, SRND + CMPQ SRND, $3*4*32 + JB avx2_loop1 + +avx2_loop2: + // w48 - w63 processed with no scheduling (last 16 rounds) + VPADDD 0*32(TBL)(SRND*1), XDWORD0, XFER + VMOVDQU XFER, (_XFER + 0*32)(SP)(SRND*1) + DO_ROUND_N_0(_XFER + 0*32, a, b, c, d, e, f, g, h, h) + DO_ROUND_N_1(_XFER + 0*32, h, a, b, c, d, e, f, g, h) + DO_ROUND_N_2(_XFER + 0*32, g, h, a, b, c, d, e, f, g) + DO_ROUND_N_3(_XFER + 0*32, f, g, h, a, b, c, d, e, f) + + VPADDD 1*32(TBL)(SRND*1), XDWORD1, XFER + VMOVDQU XFER, (_XFER + 1*32)(SP)(SRND*1) + DO_ROUND_N_0(_XFER + 1*32, e, f, g, h, a, b, c, d, e) + DO_ROUND_N_1(_XFER + 1*32, d, e, f, g, h, a, b, c, d) + DO_ROUND_N_2(_XFER + 1*32, c, d, e, f, g, h, a, b, c) + DO_ROUND_N_3(_XFER + 1*32, b, c, d, e, f, g, h, a, b) + + ADDQ $2*32, SRND + + VMOVDQU XDWORD2, XDWORD0 + VMOVDQU XDWORD3, XDWORD1 + + CMPQ SRND, $4*4*32 + JB avx2_loop2 + + MOVQ dig+0(FP), CTX // d.h[8] + MOVQ _INP(SP), INP + + addm( 0(CTX), a) + addm( 4(CTX), b) + addm( 8(CTX), c) + addm( 12(CTX), d) + addm( 16(CTX), e) + addm( 20(CTX), f) + addm( 24(CTX), g) + addm( 28(CTX), h) + + CMPQ _INP_END(SP), INP + JB done_hash + + XORQ SRND, SRND + +avx2_loop3: // Do second block using previously scheduled results + DO_ROUND_N_0(_XFER + 0*32 + 16, a, b, c, d, e, f, g, h, a) + DO_ROUND_N_1(_XFER + 0*32 + 16, h, a, b, c, d, e, f, g, h) + DO_ROUND_N_2(_XFER + 0*32 + 16, g, h, a, b, c, d, e, f, g) + DO_ROUND_N_3(_XFER + 0*32 + 16, f, g, h, a, b, c, d, e, f) + + DO_ROUND_N_0(_XFER + 1*32 + 16, e, f, g, h, a, b, c, d, e) + DO_ROUND_N_1(_XFER + 1*32 + 16, d, e, f, g, h, a, b, c, d) + DO_ROUND_N_2(_XFER + 1*32 + 16, c, d, e, f, g, h, a, b, c) + DO_ROUND_N_3(_XFER + 1*32 + 16, b, c, d, e, f, g, h, a, b) + + ADDQ $2*32, SRND + CMPQ SRND, $4*4*32 + JB avx2_loop3 + + MOVQ dig+0(FP), CTX // d.h[8] + MOVQ _INP(SP), INP + ADDQ $64, INP + + addm( 0(CTX), a) + addm( 4(CTX), b) + addm( 8(CTX), c) + addm( 12(CTX), d) + addm( 16(CTX), e) + addm( 20(CTX), f) + addm( 24(CTX), g) + addm( 28(CTX), h) + + CMPQ _INP_END(SP), INP + JA avx2_loop0 + JB done_hash + +avx2_do_last_block: + + VMOVDQU 0(INP), XWORD0 + VMOVDQU 16(INP), XWORD1 + VMOVDQU 32(INP), XWORD2 + VMOVDQU 48(INP), XWORD3 + + VMOVDQU flip_mask<>(SB), BYTE_FLIP_MASK + + VPSHUFB X_BYTE_FLIP_MASK, XWORD0, XWORD0 + VPSHUFB X_BYTE_FLIP_MASK, XWORD1, XWORD1 + VPSHUFB X_BYTE_FLIP_MASK, XWORD2, XWORD2 + VPSHUFB X_BYTE_FLIP_MASK, XWORD3, XWORD3 + + MOVQ $K256<>(SB), TBL + + JMP avx2_last_block_enter + +avx2_only_one_block: + // Load initial digest + MOVL 0(CTX), a // a = H0 + MOVL 4(CTX), b // b = H1 + MOVL 8(CTX), c // c = H2 + MOVL 12(CTX), d // d = H3 + MOVL 16(CTX), e // e = H4 + MOVL 20(CTX), f // f = H5 + MOVL 24(CTX), g // g = H6 + MOVL 28(CTX), h // h = H7 + + JMP avx2_do_last_block + +done_hash: + VZEROUPPER + RET + +// shuffle byte order from LE to BE +DATA flip_mask<>+0x00(SB)/8, $0x0405060700010203 +DATA flip_mask<>+0x08(SB)/8, $0x0c0d0e0f08090a0b +DATA flip_mask<>+0x10(SB)/8, $0x0405060700010203 +DATA flip_mask<>+0x18(SB)/8, $0x0c0d0e0f08090a0b +GLOBL flip_mask<>(SB), 8, $32 + +// shuffle xBxA -> 00BA +DATA shuff_00BA<>+0x00(SB)/8, $0x0b0a090803020100 +DATA shuff_00BA<>+0x08(SB)/8, $0xFFFFFFFFFFFFFFFF +DATA shuff_00BA<>+0x10(SB)/8, $0x0b0a090803020100 +DATA shuff_00BA<>+0x18(SB)/8, $0xFFFFFFFFFFFFFFFF +GLOBL shuff_00BA<>(SB), 8, $32 + +// shuffle xDxC -> DC00 +DATA shuff_DC00<>+0x00(SB)/8, $0xFFFFFFFFFFFFFFFF +DATA shuff_DC00<>+0x08(SB)/8, $0x0b0a090803020100 +DATA shuff_DC00<>+0x10(SB)/8, $0xFFFFFFFFFFFFFFFF +DATA shuff_DC00<>+0x18(SB)/8, $0x0b0a090803020100 +GLOBL shuff_DC00<>(SB), 8, $32 + +// Round specific constants +DATA K256<>+0x00(SB)/4, $0x428a2f98 // k1 +DATA K256<>+0x04(SB)/4, $0x71374491 // k2 +DATA K256<>+0x08(SB)/4, $0xb5c0fbcf // k3 +DATA K256<>+0x0c(SB)/4, $0xe9b5dba5 // k4 +DATA K256<>+0x10(SB)/4, $0x428a2f98 // k1 +DATA K256<>+0x14(SB)/4, $0x71374491 // k2 +DATA K256<>+0x18(SB)/4, $0xb5c0fbcf // k3 +DATA K256<>+0x1c(SB)/4, $0xe9b5dba5 // k4 + +DATA K256<>+0x20(SB)/4, $0x3956c25b // k5 - k8 +DATA K256<>+0x24(SB)/4, $0x59f111f1 +DATA K256<>+0x28(SB)/4, $0x923f82a4 +DATA K256<>+0x2c(SB)/4, $0xab1c5ed5 +DATA K256<>+0x30(SB)/4, $0x3956c25b +DATA K256<>+0x34(SB)/4, $0x59f111f1 +DATA K256<>+0x38(SB)/4, $0x923f82a4 +DATA K256<>+0x3c(SB)/4, $0xab1c5ed5 + +DATA K256<>+0x40(SB)/4, $0xd807aa98 // k9 - k12 +DATA K256<>+0x44(SB)/4, $0x12835b01 +DATA K256<>+0x48(SB)/4, $0x243185be +DATA K256<>+0x4c(SB)/4, $0x550c7dc3 +DATA K256<>+0x50(SB)/4, $0xd807aa98 +DATA K256<>+0x54(SB)/4, $0x12835b01 +DATA K256<>+0x58(SB)/4, $0x243185be +DATA K256<>+0x5c(SB)/4, $0x550c7dc3 + +DATA K256<>+0x60(SB)/4, $0x72be5d74 // k13 - k16 +DATA K256<>+0x64(SB)/4, $0x80deb1fe +DATA K256<>+0x68(SB)/4, $0x9bdc06a7 +DATA K256<>+0x6c(SB)/4, $0xc19bf174 +DATA K256<>+0x70(SB)/4, $0x72be5d74 +DATA K256<>+0x74(SB)/4, $0x80deb1fe +DATA K256<>+0x78(SB)/4, $0x9bdc06a7 +DATA K256<>+0x7c(SB)/4, $0xc19bf174 + +DATA K256<>+0x80(SB)/4, $0xe49b69c1 // k17 - k20 +DATA K256<>+0x84(SB)/4, $0xefbe4786 +DATA K256<>+0x88(SB)/4, $0x0fc19dc6 +DATA K256<>+0x8c(SB)/4, $0x240ca1cc +DATA K256<>+0x90(SB)/4, $0xe49b69c1 +DATA K256<>+0x94(SB)/4, $0xefbe4786 +DATA K256<>+0x98(SB)/4, $0x0fc19dc6 +DATA K256<>+0x9c(SB)/4, $0x240ca1cc + +DATA K256<>+0xa0(SB)/4, $0x2de92c6f // k21 - k24 +DATA K256<>+0xa4(SB)/4, $0x4a7484aa +DATA K256<>+0xa8(SB)/4, $0x5cb0a9dc +DATA K256<>+0xac(SB)/4, $0x76f988da +DATA K256<>+0xb0(SB)/4, $0x2de92c6f +DATA K256<>+0xb4(SB)/4, $0x4a7484aa +DATA K256<>+0xb8(SB)/4, $0x5cb0a9dc +DATA K256<>+0xbc(SB)/4, $0x76f988da + +DATA K256<>+0xc0(SB)/4, $0x983e5152 // k25 - k28 +DATA K256<>+0xc4(SB)/4, $0xa831c66d +DATA K256<>+0xc8(SB)/4, $0xb00327c8 +DATA K256<>+0xcc(SB)/4, $0xbf597fc7 +DATA K256<>+0xd0(SB)/4, $0x983e5152 +DATA K256<>+0xd4(SB)/4, $0xa831c66d +DATA K256<>+0xd8(SB)/4, $0xb00327c8 +DATA K256<>+0xdc(SB)/4, $0xbf597fc7 + +DATA K256<>+0xe0(SB)/4, $0xc6e00bf3 // k29 - k32 +DATA K256<>+0xe4(SB)/4, $0xd5a79147 +DATA K256<>+0xe8(SB)/4, $0x06ca6351 +DATA K256<>+0xec(SB)/4, $0x14292967 +DATA K256<>+0xf0(SB)/4, $0xc6e00bf3 +DATA K256<>+0xf4(SB)/4, $0xd5a79147 +DATA K256<>+0xf8(SB)/4, $0x06ca6351 +DATA K256<>+0xfc(SB)/4, $0x14292967 + +DATA K256<>+0x100(SB)/4, $0x27b70a85 +DATA K256<>+0x104(SB)/4, $0x2e1b2138 +DATA K256<>+0x108(SB)/4, $0x4d2c6dfc +DATA K256<>+0x10c(SB)/4, $0x53380d13 +DATA K256<>+0x110(SB)/4, $0x27b70a85 +DATA K256<>+0x114(SB)/4, $0x2e1b2138 +DATA K256<>+0x118(SB)/4, $0x4d2c6dfc +DATA K256<>+0x11c(SB)/4, $0x53380d13 + +DATA K256<>+0x120(SB)/4, $0x650a7354 +DATA K256<>+0x124(SB)/4, $0x766a0abb +DATA K256<>+0x128(SB)/4, $0x81c2c92e +DATA K256<>+0x12c(SB)/4, $0x92722c85 +DATA K256<>+0x130(SB)/4, $0x650a7354 +DATA K256<>+0x134(SB)/4, $0x766a0abb +DATA K256<>+0x138(SB)/4, $0x81c2c92e +DATA K256<>+0x13c(SB)/4, $0x92722c85 + +DATA K256<>+0x140(SB)/4, $0xa2bfe8a1 +DATA K256<>+0x144(SB)/4, $0xa81a664b +DATA K256<>+0x148(SB)/4, $0xc24b8b70 +DATA K256<>+0x14c(SB)/4, $0xc76c51a3 +DATA K256<>+0x150(SB)/4, $0xa2bfe8a1 +DATA K256<>+0x154(SB)/4, $0xa81a664b +DATA K256<>+0x158(SB)/4, $0xc24b8b70 +DATA K256<>+0x15c(SB)/4, $0xc76c51a3 + +DATA K256<>+0x160(SB)/4, $0xd192e819 +DATA K256<>+0x164(SB)/4, $0xd6990624 +DATA K256<>+0x168(SB)/4, $0xf40e3585 +DATA K256<>+0x16c(SB)/4, $0x106aa070 +DATA K256<>+0x170(SB)/4, $0xd192e819 +DATA K256<>+0x174(SB)/4, $0xd6990624 +DATA K256<>+0x178(SB)/4, $0xf40e3585 +DATA K256<>+0x17c(SB)/4, $0x106aa070 + +DATA K256<>+0x180(SB)/4, $0x19a4c116 +DATA K256<>+0x184(SB)/4, $0x1e376c08 +DATA K256<>+0x188(SB)/4, $0x2748774c +DATA K256<>+0x18c(SB)/4, $0x34b0bcb5 +DATA K256<>+0x190(SB)/4, $0x19a4c116 +DATA K256<>+0x194(SB)/4, $0x1e376c08 +DATA K256<>+0x198(SB)/4, $0x2748774c +DATA K256<>+0x19c(SB)/4, $0x34b0bcb5 + +DATA K256<>+0x1a0(SB)/4, $0x391c0cb3 +DATA K256<>+0x1a4(SB)/4, $0x4ed8aa4a +DATA K256<>+0x1a8(SB)/4, $0x5b9cca4f +DATA K256<>+0x1ac(SB)/4, $0x682e6ff3 +DATA K256<>+0x1b0(SB)/4, $0x391c0cb3 +DATA K256<>+0x1b4(SB)/4, $0x4ed8aa4a +DATA K256<>+0x1b8(SB)/4, $0x5b9cca4f +DATA K256<>+0x1bc(SB)/4, $0x682e6ff3 + +DATA K256<>+0x1c0(SB)/4, $0x748f82ee +DATA K256<>+0x1c4(SB)/4, $0x78a5636f +DATA K256<>+0x1c8(SB)/4, $0x84c87814 +DATA K256<>+0x1cc(SB)/4, $0x8cc70208 +DATA K256<>+0x1d0(SB)/4, $0x748f82ee +DATA K256<>+0x1d4(SB)/4, $0x78a5636f +DATA K256<>+0x1d8(SB)/4, $0x84c87814 +DATA K256<>+0x1dc(SB)/4, $0x8cc70208 + +DATA K256<>+0x1e0(SB)/4, $0x90befffa +DATA K256<>+0x1e4(SB)/4, $0xa4506ceb +DATA K256<>+0x1e8(SB)/4, $0xbef9a3f7 +DATA K256<>+0x1ec(SB)/4, $0xc67178f2 +DATA K256<>+0x1f0(SB)/4, $0x90befffa +DATA K256<>+0x1f4(SB)/4, $0xa4506ceb +DATA K256<>+0x1f8(SB)/4, $0xbef9a3f7 +DATA K256<>+0x1fc(SB)/4, $0xc67178f2 + +GLOBL K256<>(SB), (NOPTR + RODATA), $512 diff --git a/src/crypto/sha256/sha256block_arm64.go b/src/crypto/sha256/sha256block_arm64.go new file mode 100644 index 0000000..e5da566 --- /dev/null +++ b/src/crypto/sha256/sha256block_arm64.go @@ -0,0 +1,21 @@ +// 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 sha256 + +import "internal/cpu" + +var k = _K + +//go:noescape +func sha256block(h []uint32, p []byte, k []uint32) + +func block(dig *digest, p []byte) { + if !cpu.ARM64.HasSHA2 { + blockGeneric(dig, p) + } else { + h := dig.h[:] + sha256block(h, p, k) + } +} diff --git a/src/crypto/sha256/sha256block_arm64.s b/src/crypto/sha256/sha256block_arm64.s new file mode 100644 index 0000000..d5c1eb0 --- /dev/null +++ b/src/crypto/sha256/sha256block_arm64.s @@ -0,0 +1,119 @@ +// 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. + +#include "textflag.h" + +#define HASHUPDATE \ + SHA256H V9.S4, V3, V2 \ + SHA256H2 V9.S4, V8, V3 \ + VMOV V2.B16, V8.B16 + +// func sha256block(h []uint32, p []byte, k []uint32) +TEXT ·sha256block(SB),NOSPLIT,$0 + MOVD h_base+0(FP), R0 // Hash value first address + MOVD p_base+24(FP), R1 // message first address + MOVD k_base+48(FP), R2 // k constants first address + MOVD p_len+32(FP), R3 // message length + VLD1 (R0), [V0.S4, V1.S4] // load h(a,b,c,d,e,f,g,h) + VLD1.P 64(R2), [V16.S4, V17.S4, V18.S4, V19.S4] + VLD1.P 64(R2), [V20.S4, V21.S4, V22.S4, V23.S4] + VLD1.P 64(R2), [V24.S4, V25.S4, V26.S4, V27.S4] + VLD1 (R2), [V28.S4, V29.S4, V30.S4, V31.S4] //load 64*4bytes K constant(K0-K63) + +blockloop: + + VLD1.P 16(R1), [V4.B16] // load 16bytes message + VLD1.P 16(R1), [V5.B16] // load 16bytes message + VLD1.P 16(R1), [V6.B16] // load 16bytes message + VLD1.P 16(R1), [V7.B16] // load 16bytes message + VMOV V0.B16, V2.B16 // backup: VO h(dcba) + VMOV V1.B16, V3.B16 // backup: V1 h(hgfe) + VMOV V2.B16, V8.B16 + VREV32 V4.B16, V4.B16 // prepare for using message in Byte format + VREV32 V5.B16, V5.B16 + VREV32 V6.B16, V6.B16 + VREV32 V7.B16, V7.B16 + + VADD V16.S4, V4.S4, V9.S4 // V18(W0+K0...W3+K3) + SHA256SU0 V5.S4, V4.S4 // V4: (su0(W1)+W0,...,su0(W4)+W3) + HASHUPDATE // H4 + + VADD V17.S4, V5.S4, V9.S4 // V18(W4+K4...W7+K7) + SHA256SU0 V6.S4, V5.S4 // V5: (su0(W5)+W4,...,su0(W8)+W7) + SHA256SU1 V7.S4, V6.S4, V4.S4 // V4: W16-W19 + HASHUPDATE // H8 + + VADD V18.S4, V6.S4, V9.S4 // V18(W8+K8...W11+K11) + SHA256SU0 V7.S4, V6.S4 // V6: (su0(W9)+W8,...,su0(W12)+W11) + SHA256SU1 V4.S4, V7.S4, V5.S4 // V5: W20-W23 + HASHUPDATE // H12 + + VADD V19.S4, V7.S4, V9.S4 // V18(W12+K12...W15+K15) + SHA256SU0 V4.S4, V7.S4 // V7: (su0(W13)+W12,...,su0(W16)+W15) + SHA256SU1 V5.S4, V4.S4, V6.S4 // V6: W24-W27 + HASHUPDATE // H16 + + VADD V20.S4, V4.S4, V9.S4 // V18(W16+K16...W19+K19) + SHA256SU0 V5.S4, V4.S4 // V4: (su0(W17)+W16,...,su0(W20)+W19) + SHA256SU1 V6.S4, V5.S4, V7.S4 // V7: W28-W31 + HASHUPDATE // H20 + + VADD V21.S4, V5.S4, V9.S4 // V18(W20+K20...W23+K23) + SHA256SU0 V6.S4, V5.S4 // V5: (su0(W21)+W20,...,su0(W24)+W23) + SHA256SU1 V7.S4, V6.S4, V4.S4 // V4: W32-W35 + HASHUPDATE // H24 + + VADD V22.S4, V6.S4, V9.S4 // V18(W24+K24...W27+K27) + SHA256SU0 V7.S4, V6.S4 // V6: (su0(W25)+W24,...,su0(W28)+W27) + SHA256SU1 V4.S4, V7.S4, V5.S4 // V5: W36-W39 + HASHUPDATE // H28 + + VADD V23.S4, V7.S4, V9.S4 // V18(W28+K28...W31+K31) + SHA256SU0 V4.S4, V7.S4 // V7: (su0(W29)+W28,...,su0(W32)+W31) + SHA256SU1 V5.S4, V4.S4, V6.S4 // V6: W40-W43 + HASHUPDATE // H32 + + VADD V24.S4, V4.S4, V9.S4 // V18(W32+K32...W35+K35) + SHA256SU0 V5.S4, V4.S4 // V4: (su0(W33)+W32,...,su0(W36)+W35) + SHA256SU1 V6.S4, V5.S4, V7.S4 // V7: W44-W47 + HASHUPDATE // H36 + + VADD V25.S4, V5.S4, V9.S4 // V18(W36+K36...W39+K39) + SHA256SU0 V6.S4, V5.S4 // V5: (su0(W37)+W36,...,su0(W40)+W39) + SHA256SU1 V7.S4, V6.S4, V4.S4 // V4: W48-W51 + HASHUPDATE // H40 + + VADD V26.S4, V6.S4, V9.S4 // V18(W40+K40...W43+K43) + SHA256SU0 V7.S4, V6.S4 // V6: (su0(W41)+W40,...,su0(W44)+W43) + SHA256SU1 V4.S4, V7.S4, V5.S4 // V5: W52-W55 + HASHUPDATE // H44 + + VADD V27.S4, V7.S4, V9.S4 // V18(W44+K44...W47+K47) + SHA256SU0 V4.S4, V7.S4 // V7: (su0(W45)+W44,...,su0(W48)+W47) + SHA256SU1 V5.S4, V4.S4, V6.S4 // V6: W56-W59 + HASHUPDATE // H48 + + VADD V28.S4, V4.S4, V9.S4 // V18(W48+K48,...,W51+K51) + HASHUPDATE // H52 + SHA256SU1 V6.S4, V5.S4, V7.S4 // V7: W60-W63 + + VADD V29.S4, V5.S4, V9.S4 // V18(W52+K52,...,W55+K55) + HASHUPDATE // H56 + + VADD V30.S4, V6.S4, V9.S4 // V18(W59+K59,...,W59+K59) + HASHUPDATE // H60 + + VADD V31.S4, V7.S4, V9.S4 // V18(W60+K60,...,W63+K63) + HASHUPDATE // H64 + + SUB $64, R3, R3 // message length - 64bytes, then compare with 64bytes + VADD V2.S4, V0.S4, V0.S4 + VADD V3.S4, V1.S4, V1.S4 + CBNZ R3, blockloop + +sha256ret: + + VST1 [V0.S4, V1.S4], (R0) // store hash value H + RET + diff --git a/src/crypto/sha256/sha256block_decl.go b/src/crypto/sha256/sha256block_decl.go new file mode 100644 index 0000000..c9c1194 --- /dev/null +++ b/src/crypto/sha256/sha256block_decl.go @@ -0,0 +1,11 @@ +// Copyright 2013 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +//go:build 386 || amd64 || s390x || ppc64le + +package sha256 + +//go:noescape + +func block(dig *digest, p []byte) diff --git a/src/crypto/sha256/sha256block_generic.go b/src/crypto/sha256/sha256block_generic.go new file mode 100644 index 0000000..a8878c2 --- /dev/null +++ b/src/crypto/sha256/sha256block_generic.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 && !386 && !s390x && !ppc64le && !arm64 + +package sha256 + +func block(dig *digest, p []byte) { + blockGeneric(dig, p) +} diff --git a/src/crypto/sha256/sha256block_ppc64le.s b/src/crypto/sha256/sha256block_ppc64le.s new file mode 100644 index 0000000..77e63c0 --- /dev/null +++ b/src/crypto/sha256/sha256block_ppc64le.s @@ -0,0 +1,419 @@ +// 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. + +// 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/. +// # ==================================================================== + +#include "textflag.h" + +// SHA256 block routine. See sha256block.go for Go equivalent. +// +// The algorithm is detailed in FIPS 180-4: +// +// https://csrc.nist.gov/publications/fips/fips180-4/fips-180-4.pdf +// +// Wt = Mt; for 0 <= t <= 15 +// Wt = SIGMA1(Wt-2) + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 63 +// +// a = H0 +// b = H1 +// c = H2 +// d = H3 +// e = H4 +// f = H5 +// g = H6 +// h = H7 +// +// for t = 0 to 63 { +// T1 = h + BIGSIGMA1(e) + Ch(e,f,g) + Kt + Wt +// T2 = BIGSIGMA0(a) + Maj(a,b,c) +// h = g +// g = f +// f = e +// e = d + T1 +// d = c +// c = b +// b = a +// a = T1 + T2 +// } +// +// H0 = a + H0 +// H1 = b + H1 +// H2 = c + H2 +// H3 = d + H3 +// H4 = e + H4 +// H5 = f + H5 +// H6 = g + H6 +// H7 = h + H7 + +#define CTX R3 +#define INP R4 +#define END R5 +#define TBL R6 +#define IDX R7 +#define CNT R8 +#define LEN R9 +#define OFFLOAD R11 +#define TEMP R12 + +#define HEX00 R0 +#define HEX10 R10 +#define HEX20 R25 +#define HEX30 R26 +#define HEX40 R27 +#define HEX50 R28 +#define HEX60 R29 +#define HEX70 R31 + +// V0-V7 are A-H +// V8-V23 are used for the message schedule +#define KI V24 +#define FUNC V25 +#define S0 V26 +#define S1 V27 +#define s0 V28 +#define s1 V29 +#define LEMASK V31 // Permutation control register for little endian + +// 4 copies of each Kt, to fill all 4 words of a vector register +DATA ·kcon+0x000(SB)/8, $0x428a2f98428a2f98 +DATA ·kcon+0x008(SB)/8, $0x428a2f98428a2f98 +DATA ·kcon+0x010(SB)/8, $0x7137449171374491 +DATA ·kcon+0x018(SB)/8, $0x7137449171374491 +DATA ·kcon+0x020(SB)/8, $0xb5c0fbcfb5c0fbcf +DATA ·kcon+0x028(SB)/8, $0xb5c0fbcfb5c0fbcf +DATA ·kcon+0x030(SB)/8, $0xe9b5dba5e9b5dba5 +DATA ·kcon+0x038(SB)/8, $0xe9b5dba5e9b5dba5 +DATA ·kcon+0x040(SB)/8, $0x3956c25b3956c25b +DATA ·kcon+0x048(SB)/8, $0x3956c25b3956c25b +DATA ·kcon+0x050(SB)/8, $0x59f111f159f111f1 +DATA ·kcon+0x058(SB)/8, $0x59f111f159f111f1 +DATA ·kcon+0x060(SB)/8, $0x923f82a4923f82a4 +DATA ·kcon+0x068(SB)/8, $0x923f82a4923f82a4 +DATA ·kcon+0x070(SB)/8, $0xab1c5ed5ab1c5ed5 +DATA ·kcon+0x078(SB)/8, $0xab1c5ed5ab1c5ed5 +DATA ·kcon+0x080(SB)/8, $0xd807aa98d807aa98 +DATA ·kcon+0x088(SB)/8, $0xd807aa98d807aa98 +DATA ·kcon+0x090(SB)/8, $0x12835b0112835b01 +DATA ·kcon+0x098(SB)/8, $0x12835b0112835b01 +DATA ·kcon+0x0A0(SB)/8, $0x243185be243185be +DATA ·kcon+0x0A8(SB)/8, $0x243185be243185be +DATA ·kcon+0x0B0(SB)/8, $0x550c7dc3550c7dc3 +DATA ·kcon+0x0B8(SB)/8, $0x550c7dc3550c7dc3 +DATA ·kcon+0x0C0(SB)/8, $0x72be5d7472be5d74 +DATA ·kcon+0x0C8(SB)/8, $0x72be5d7472be5d74 +DATA ·kcon+0x0D0(SB)/8, $0x80deb1fe80deb1fe +DATA ·kcon+0x0D8(SB)/8, $0x80deb1fe80deb1fe +DATA ·kcon+0x0E0(SB)/8, $0x9bdc06a79bdc06a7 +DATA ·kcon+0x0E8(SB)/8, $0x9bdc06a79bdc06a7 +DATA ·kcon+0x0F0(SB)/8, $0xc19bf174c19bf174 +DATA ·kcon+0x0F8(SB)/8, $0xc19bf174c19bf174 +DATA ·kcon+0x100(SB)/8, $0xe49b69c1e49b69c1 +DATA ·kcon+0x108(SB)/8, $0xe49b69c1e49b69c1 +DATA ·kcon+0x110(SB)/8, $0xefbe4786efbe4786 +DATA ·kcon+0x118(SB)/8, $0xefbe4786efbe4786 +DATA ·kcon+0x120(SB)/8, $0x0fc19dc60fc19dc6 +DATA ·kcon+0x128(SB)/8, $0x0fc19dc60fc19dc6 +DATA ·kcon+0x130(SB)/8, $0x240ca1cc240ca1cc +DATA ·kcon+0x138(SB)/8, $0x240ca1cc240ca1cc +DATA ·kcon+0x140(SB)/8, $0x2de92c6f2de92c6f +DATA ·kcon+0x148(SB)/8, $0x2de92c6f2de92c6f +DATA ·kcon+0x150(SB)/8, $0x4a7484aa4a7484aa +DATA ·kcon+0x158(SB)/8, $0x4a7484aa4a7484aa +DATA ·kcon+0x160(SB)/8, $0x5cb0a9dc5cb0a9dc +DATA ·kcon+0x168(SB)/8, $0x5cb0a9dc5cb0a9dc +DATA ·kcon+0x170(SB)/8, $0x76f988da76f988da +DATA ·kcon+0x178(SB)/8, $0x76f988da76f988da +DATA ·kcon+0x180(SB)/8, $0x983e5152983e5152 +DATA ·kcon+0x188(SB)/8, $0x983e5152983e5152 +DATA ·kcon+0x190(SB)/8, $0xa831c66da831c66d +DATA ·kcon+0x198(SB)/8, $0xa831c66da831c66d +DATA ·kcon+0x1A0(SB)/8, $0xb00327c8b00327c8 +DATA ·kcon+0x1A8(SB)/8, $0xb00327c8b00327c8 +DATA ·kcon+0x1B0(SB)/8, $0xbf597fc7bf597fc7 +DATA ·kcon+0x1B8(SB)/8, $0xbf597fc7bf597fc7 +DATA ·kcon+0x1C0(SB)/8, $0xc6e00bf3c6e00bf3 +DATA ·kcon+0x1C8(SB)/8, $0xc6e00bf3c6e00bf3 +DATA ·kcon+0x1D0(SB)/8, $0xd5a79147d5a79147 +DATA ·kcon+0x1D8(SB)/8, $0xd5a79147d5a79147 +DATA ·kcon+0x1E0(SB)/8, $0x06ca635106ca6351 +DATA ·kcon+0x1E8(SB)/8, $0x06ca635106ca6351 +DATA ·kcon+0x1F0(SB)/8, $0x1429296714292967 +DATA ·kcon+0x1F8(SB)/8, $0x1429296714292967 +DATA ·kcon+0x200(SB)/8, $0x27b70a8527b70a85 +DATA ·kcon+0x208(SB)/8, $0x27b70a8527b70a85 +DATA ·kcon+0x210(SB)/8, $0x2e1b21382e1b2138 +DATA ·kcon+0x218(SB)/8, $0x2e1b21382e1b2138 +DATA ·kcon+0x220(SB)/8, $0x4d2c6dfc4d2c6dfc +DATA ·kcon+0x228(SB)/8, $0x4d2c6dfc4d2c6dfc +DATA ·kcon+0x230(SB)/8, $0x53380d1353380d13 +DATA ·kcon+0x238(SB)/8, $0x53380d1353380d13 +DATA ·kcon+0x240(SB)/8, $0x650a7354650a7354 +DATA ·kcon+0x248(SB)/8, $0x650a7354650a7354 +DATA ·kcon+0x250(SB)/8, $0x766a0abb766a0abb +DATA ·kcon+0x258(SB)/8, $0x766a0abb766a0abb +DATA ·kcon+0x260(SB)/8, $0x81c2c92e81c2c92e +DATA ·kcon+0x268(SB)/8, $0x81c2c92e81c2c92e +DATA ·kcon+0x270(SB)/8, $0x92722c8592722c85 +DATA ·kcon+0x278(SB)/8, $0x92722c8592722c85 +DATA ·kcon+0x280(SB)/8, $0xa2bfe8a1a2bfe8a1 +DATA ·kcon+0x288(SB)/8, $0xa2bfe8a1a2bfe8a1 +DATA ·kcon+0x290(SB)/8, $0xa81a664ba81a664b +DATA ·kcon+0x298(SB)/8, $0xa81a664ba81a664b +DATA ·kcon+0x2A0(SB)/8, $0xc24b8b70c24b8b70 +DATA ·kcon+0x2A8(SB)/8, $0xc24b8b70c24b8b70 +DATA ·kcon+0x2B0(SB)/8, $0xc76c51a3c76c51a3 +DATA ·kcon+0x2B8(SB)/8, $0xc76c51a3c76c51a3 +DATA ·kcon+0x2C0(SB)/8, $0xd192e819d192e819 +DATA ·kcon+0x2C8(SB)/8, $0xd192e819d192e819 +DATA ·kcon+0x2D0(SB)/8, $0xd6990624d6990624 +DATA ·kcon+0x2D8(SB)/8, $0xd6990624d6990624 +DATA ·kcon+0x2E0(SB)/8, $0xf40e3585f40e3585 +DATA ·kcon+0x2E8(SB)/8, $0xf40e3585f40e3585 +DATA ·kcon+0x2F0(SB)/8, $0x106aa070106aa070 +DATA ·kcon+0x2F8(SB)/8, $0x106aa070106aa070 +DATA ·kcon+0x300(SB)/8, $0x19a4c11619a4c116 +DATA ·kcon+0x308(SB)/8, $0x19a4c11619a4c116 +DATA ·kcon+0x310(SB)/8, $0x1e376c081e376c08 +DATA ·kcon+0x318(SB)/8, $0x1e376c081e376c08 +DATA ·kcon+0x320(SB)/8, $0x2748774c2748774c +DATA ·kcon+0x328(SB)/8, $0x2748774c2748774c +DATA ·kcon+0x330(SB)/8, $0x34b0bcb534b0bcb5 +DATA ·kcon+0x338(SB)/8, $0x34b0bcb534b0bcb5 +DATA ·kcon+0x340(SB)/8, $0x391c0cb3391c0cb3 +DATA ·kcon+0x348(SB)/8, $0x391c0cb3391c0cb3 +DATA ·kcon+0x350(SB)/8, $0x4ed8aa4a4ed8aa4a +DATA ·kcon+0x358(SB)/8, $0x4ed8aa4a4ed8aa4a +DATA ·kcon+0x360(SB)/8, $0x5b9cca4f5b9cca4f +DATA ·kcon+0x368(SB)/8, $0x5b9cca4f5b9cca4f +DATA ·kcon+0x370(SB)/8, $0x682e6ff3682e6ff3 +DATA ·kcon+0x378(SB)/8, $0x682e6ff3682e6ff3 +DATA ·kcon+0x380(SB)/8, $0x748f82ee748f82ee +DATA ·kcon+0x388(SB)/8, $0x748f82ee748f82ee +DATA ·kcon+0x390(SB)/8, $0x78a5636f78a5636f +DATA ·kcon+0x398(SB)/8, $0x78a5636f78a5636f +DATA ·kcon+0x3A0(SB)/8, $0x84c8781484c87814 +DATA ·kcon+0x3A8(SB)/8, $0x84c8781484c87814 +DATA ·kcon+0x3B0(SB)/8, $0x8cc702088cc70208 +DATA ·kcon+0x3B8(SB)/8, $0x8cc702088cc70208 +DATA ·kcon+0x3C0(SB)/8, $0x90befffa90befffa +DATA ·kcon+0x3C8(SB)/8, $0x90befffa90befffa +DATA ·kcon+0x3D0(SB)/8, $0xa4506ceba4506ceb +DATA ·kcon+0x3D8(SB)/8, $0xa4506ceba4506ceb +DATA ·kcon+0x3E0(SB)/8, $0xbef9a3f7bef9a3f7 +DATA ·kcon+0x3E8(SB)/8, $0xbef9a3f7bef9a3f7 +DATA ·kcon+0x3F0(SB)/8, $0xc67178f2c67178f2 +DATA ·kcon+0x3F8(SB)/8, $0xc67178f2c67178f2 +DATA ·kcon+0x400(SB)/8, $0x0000000000000000 +DATA ·kcon+0x408(SB)/8, $0x0000000000000000 +DATA ·kcon+0x410(SB)/8, $0x1011121310111213 // permutation control vectors +DATA ·kcon+0x418(SB)/8, $0x1011121300010203 +DATA ·kcon+0x420(SB)/8, $0x1011121310111213 +DATA ·kcon+0x428(SB)/8, $0x0405060700010203 +DATA ·kcon+0x430(SB)/8, $0x1011121308090a0b +DATA ·kcon+0x438(SB)/8, $0x0405060700010203 +GLOBL ·kcon(SB), RODATA, $1088 + +#define SHA256ROUND0(a, b, c, d, e, f, g, h, xi) \ + VSEL g, f, e, FUNC; \ + VSHASIGMAW $15, e, $1, S1; \ + VADDUWM xi, h, h; \ + VSHASIGMAW $0, a, $1, S0; \ + VADDUWM FUNC, h, h; \ + VXOR b, a, FUNC; \ + VADDUWM S1, h, h; \ + VSEL b, c, FUNC, FUNC; \ + VADDUWM KI, g, g; \ + VADDUWM h, d, d; \ + VADDUWM FUNC, S0, S0; \ + LVX (TBL)(IDX), KI; \ + ADD $16, IDX; \ + VADDUWM S0, h, h + +#define SHA256ROUND1(a, b, c, d, e, f, g, h, xi, xj, xj_1, xj_9, xj_14) \ + VSHASIGMAW $0, xj_1, $0, s0; \ + VSEL g, f, e, FUNC; \ + VSHASIGMAW $15, e, $1, S1; \ + VADDUWM xi, h, h; \ + VSHASIGMAW $0, a, $1, S0; \ + VSHASIGMAW $15, xj_14, $0, s1; \ + VADDUWM FUNC, h, h; \ + VXOR b, a, FUNC; \ + VADDUWM xj_9, xj, xj; \ + VADDUWM S1, h, h; \ + VSEL b, c, FUNC, FUNC; \ + VADDUWM KI, g, g; \ + VADDUWM h, d, d; \ + VADDUWM FUNC, S0, S0; \ + VADDUWM s0, xj, xj; \ + LVX (TBL)(IDX), KI; \ + ADD $16, IDX; \ + VADDUWM S0, h, h; \ + VADDUWM s1, xj, xj + +// func block(dig *digest, p []byte) +TEXT ·block(SB),0,$128-32 + MOVD dig+0(FP), CTX + MOVD p_base+8(FP), INP + MOVD p_len+16(FP), LEN + + SRD $6, LEN + SLD $6, LEN + + ADD INP, LEN, END + + CMP INP, END + BEQ end + + MOVD $·kcon(SB), TBL + MOVD R1, OFFLOAD + + MOVD R0, CNT + MOVWZ $0x10, HEX10 + MOVWZ $0x20, HEX20 + MOVWZ $0x30, HEX30 + MOVWZ $0x40, HEX40 + MOVWZ $0x50, HEX50 + MOVWZ $0x60, HEX60 + MOVWZ $0x70, HEX70 + + MOVWZ $8, IDX + LVSL (IDX)(R0), LEMASK + VSPLTISB $0x0F, KI + VXOR KI, LEMASK, LEMASK + + LXVW4X (CTX)(HEX00), VS32 // v0 = vs32 + LXVW4X (CTX)(HEX10), VS36 // v4 = vs36 + + // unpack the input values into vector registers + VSLDOI $4, V0, V0, V1 + VSLDOI $8, V0, V0, V2 + VSLDOI $12, V0, V0, V3 + VSLDOI $4, V4, V4, V5 + VSLDOI $8, V4, V4, V6 + VSLDOI $12, V4, V4, V7 + +loop: + LVX (TBL)(HEX00), KI + MOVWZ $16, IDX + + LXVD2X (INP)(R0), VS40 // load v8 (=vs40) in advance + ADD $16, INP + + STVX V0, (OFFLOAD+HEX00) + STVX V1, (OFFLOAD+HEX10) + STVX V2, (OFFLOAD+HEX20) + STVX V3, (OFFLOAD+HEX30) + STVX V4, (OFFLOAD+HEX40) + STVX V5, (OFFLOAD+HEX50) + STVX V6, (OFFLOAD+HEX60) + STVX V7, (OFFLOAD+HEX70) + + VADDUWM KI, V7, V7 // h+K[i] + LVX (TBL)(IDX), KI + ADD $16, IDX + + VPERM V8, V8, LEMASK, V8 + SHA256ROUND0(V0, V1, V2, V3, V4, V5, V6, V7, V8) + VSLDOI $4, V8, V8, V9 + SHA256ROUND0(V7, V0, V1, V2, V3, V4, V5, V6, V9) + VSLDOI $4, V9, V9, V10 + SHA256ROUND0(V6, V7, V0, V1, V2, V3, V4, V5, V10) + LXVD2X (INP)(R0), VS44 // load v12 (=vs44) in advance + ADD $16, INP, INP + VSLDOI $4, V10, V10, V11 + SHA256ROUND0(V5, V6, V7, V0, V1, V2, V3, V4, V11) + VPERM V12, V12, LEMASK, V12 + SHA256ROUND0(V4, V5, V6, V7, V0, V1, V2, V3, V12) + VSLDOI $4, V12, V12, V13 + SHA256ROUND0(V3, V4, V5, V6, V7, V0, V1, V2, V13) + VSLDOI $4, V13, V13, V14 + SHA256ROUND0(V2, V3, V4, V5, V6, V7, V0, V1, V14) + LXVD2X (INP)(R0), VS48 // load v16 (=vs48) in advance + ADD $16, INP, INP + VSLDOI $4, V14, V14, V15 + SHA256ROUND0(V1, V2, V3, V4, V5, V6, V7, V0, V15) + VPERM V16, V16, LEMASK, V16 + SHA256ROUND0(V0, V1, V2, V3, V4, V5, V6, V7, V16) + VSLDOI $4, V16, V16, V17 + SHA256ROUND0(V7, V0, V1, V2, V3, V4, V5, V6, V17) + VSLDOI $4, V17, V17, V18 + SHA256ROUND0(V6, V7, V0, V1, V2, V3, V4, V5, V18) + VSLDOI $4, V18, V18, V19 + LXVD2X (INP)(R0), VS52 // load v20 (=vs52) in advance + ADD $16, INP, INP + SHA256ROUND0(V5, V6, V7, V0, V1, V2, V3, V4, V19) + VPERM V20, V20, LEMASK, V20 + SHA256ROUND0(V4, V5, V6, V7, V0, V1, V2, V3, V20) + VSLDOI $4, V20, V20, V21 + SHA256ROUND0(V3, V4, V5, V6, V7, V0, V1, V2, V21) + VSLDOI $4, V21, V21, V22 + SHA256ROUND0(V2, V3, V4, V5, V6, V7, V0, V1, V22) + VSLDOI $4, V22, V22, V23 + SHA256ROUND1(V1, V2, V3, V4, V5, V6, V7, V0, V23, V8, V9, V17, V22) + + MOVWZ $3, TEMP + MOVWZ TEMP, CTR + +L16_xx: + SHA256ROUND1(V0, V1, V2, V3, V4, V5, V6, V7, V8, V9, V10, V18, V23) + SHA256ROUND1(V7, V0, V1, V2, V3, V4, V5, V6, V9, V10, V11, V19, V8) + SHA256ROUND1(V6, V7, V0, V1, V2, V3, V4, V5, V10, V11, V12, V20, V9) + SHA256ROUND1(V5, V6, V7, V0, V1, V2, V3, V4, V11, V12, V13, V21, V10) + SHA256ROUND1(V4, V5, V6, V7, V0, V1, V2, V3, V12, V13, V14, V22, V11) + SHA256ROUND1(V3, V4, V5, V6, V7, V0, V1, V2, V13, V14, V15, V23, V12) + SHA256ROUND1(V2, V3, V4, V5, V6, V7, V0, V1, V14, V15, V16, V8, V13) + SHA256ROUND1(V1, V2, V3, V4, V5, V6, V7, V0, V15, V16, V17, V9, V14) + SHA256ROUND1(V0, V1, V2, V3, V4, V5, V6, V7, V16, V17, V18, V10, V15) + SHA256ROUND1(V7, V0, V1, V2, V3, V4, V5, V6, V17, V18, V19, V11, V16) + SHA256ROUND1(V6, V7, V0, V1, V2, V3, V4, V5, V18, V19, V20, V12, V17) + SHA256ROUND1(V5, V6, V7, V0, V1, V2, V3, V4, V19, V20, V21, V13, V18) + SHA256ROUND1(V4, V5, V6, V7, V0, V1, V2, V3, V20, V21, V22, V14, V19) + SHA256ROUND1(V3, V4, V5, V6, V7, V0, V1, V2, V21, V22, V23, V15, V20) + SHA256ROUND1(V2, V3, V4, V5, V6, V7, V0, V1, V22, V23, V8, V16, V21) + SHA256ROUND1(V1, V2, V3, V4, V5, V6, V7, V0, V23, V8, V9, V17, V22) + + BC 0x10, 0, L16_xx // bdnz + + LVX (OFFLOAD)(HEX00), V10 + + LVX (OFFLOAD)(HEX10), V11 + VADDUWM V10, V0, V0 + LVX (OFFLOAD)(HEX20), V12 + VADDUWM V11, V1, V1 + LVX (OFFLOAD)(HEX30), V13 + VADDUWM V12, V2, V2 + LVX (OFFLOAD)(HEX40), V14 + VADDUWM V13, V3, V3 + LVX (OFFLOAD)(HEX50), V15 + VADDUWM V14, V4, V4 + LVX (OFFLOAD)(HEX60), V16 + VADDUWM V15, V5, V5 + LVX (OFFLOAD)(HEX70), V17 + VADDUWM V16, V6, V6 + VADDUWM V17, V7, V7 + + CMPU INP, END + BLT loop + + LVX (TBL)(IDX), V8 + ADD $16, IDX + VPERM V0, V1, KI, V0 + LVX (TBL)(IDX), V9 + VPERM V4, V5, KI, V4 + VPERM V0, V2, V8, V0 + VPERM V4, V6, V8, V4 + VPERM V0, V3, V9, V0 + VPERM V4, V7, V9, V4 + STXVD2X VS32, (CTX+HEX00) // v0 = vs32 + STXVD2X VS36, (CTX+HEX10) // v4 = vs36 + +end: + RET + diff --git a/src/crypto/sha256/sha256block_s390x.go b/src/crypto/sha256/sha256block_s390x.go new file mode 100644 index 0000000..1a376c5 --- /dev/null +++ b/src/crypto/sha256/sha256block_s390x.go @@ -0,0 +1,9 @@ +// 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 sha256 + +import "internal/cpu" + +var useAsm = cpu.S390X.HasSHA256 diff --git a/src/crypto/sha256/sha256block_s390x.s b/src/crypto/sha256/sha256block_s390x.s new file mode 100644 index 0000000..81b1b38 --- /dev/null +++ b/src/crypto/sha256/sha256block_s390x.s @@ -0,0 +1,20 @@ +// 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. + +#include "textflag.h" + +// func block(dig *digest, p []byte) +TEXT ·block(SB), NOSPLIT|NOFRAME, $0-32 + MOVBZ ·useAsm(SB), R4 + LMG dig+0(FP), R1, R3 // R2 = &p[0], R3 = len(p) + MOVBZ $2, R0 // SHA-256 function code + CMPBEQ R4, $0, generic + +loop: + WORD $0xB93E0002 // KIMD R2 + BVS loop // continue if interrupted + RET + +generic: + BR ·blockGeneric(SB) |