Also, make sure New inlines so it does not cause an allocation. With this, we have a zero allocation poly1305 flow and a zero allocation generic chacha20poly1305 composition! \o/ While at it, remove some redundant code, and prepare to drop some complexity once the last assembly implementation of sum is dropped. Benchstat with "-tags purego" on amd64 name old time/op new time/op delta Open-64-8 461ns ± 2% 415ns ± 1% -9.93% (p=0.000 n=10+8) Seal-64-8 450ns ± 3% 412ns ± 3% -8.41% (p=0.000 n=10+10) Open-64-X-8 603ns ± 2% 544ns ± 2% -9.84% (p=0.000 n=10+10) Seal-64-X-8 580ns ± 3% 553ns ± 1% -4.56% (p=0.000 n=9+8) Open-1350-8 3.98µs ± 2% 3.65µs ± 2% -8.28% (p=0.000 n=9+10) Seal-1350-8 3.95µs ± 2% 3.64µs ± 1% -7.93% (p=0.000 n=9+10) Open-1350-X-8 4.06µs ± 1% 3.68µs ± 3% -9.31% (p=0.000 n=9+10) Seal-1350-X-8 4.08µs ± 4% 3.64µs ± 1% -10.71% (p=0.000 n=9+10) Open-8192-8 21.7µs ± 3% 18.9µs ± 2% -13.10% (p=0.000 n=9+10) Seal-8192-8 21.5µs ± 3% 18.8µs ± 1% -12.51% (p=0.000 n=9+9) Open-8192-X-8 21.4µs ± 1% 19.1µs ± 2% -10.88% (p=0.000 n=10+10) Seal-8192-X-8 21.3µs ± 2% 19.0µs ± 3% -10.92% (p=0.000 n=10+10) name old speed new speed delta Open-64-8 139MB/s ± 2% 154MB/s ± 2% +11.05% (p=0.000 n=10+8) Seal-64-8 142MB/s ± 3% 155MB/s ± 3% +9.11% (p=0.000 n=10+10) Open-64-X-8 106MB/s ± 2% 118MB/s ± 2% +10.93% (p=0.000 n=10+10) Seal-64-X-8 110MB/s ± 3% 116MB/s ± 1% +4.75% (p=0.000 n=9+8) Open-1350-8 339MB/s ± 2% 370MB/s ± 2% +9.04% (p=0.000 n=9+10) Seal-1350-8 342MB/s ± 2% 371MB/s ± 1% +8.60% (p=0.000 n=9+10) Open-1350-X-8 333MB/s ± 1% 367MB/s ± 3% +10.30% (p=0.000 n=9+10) Seal-1350-X-8 331MB/s ± 4% 371MB/s ± 2% +11.96% (p=0.000 n=9+10) Open-8192-8 377MB/s ± 3% 434MB/s ± 2% +15.07% (p=0.000 n=9+10) Seal-8192-8 381MB/s ± 3% 436MB/s ± 1% +14.29% (p=0.000 n=9+9) Open-8192-X-8 383MB/s ± 1% 429MB/s ± 2% +12.21% (p=0.000 n=10+10) Seal-8192-X-8 385MB/s ± 2% 432MB/s ± 3% +12.26% (p=0.000 n=10+10) name old alloc/op new alloc/op delta Open-64-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Seal-64-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Open-64-X-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Seal-64-X-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Open-1350-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-1350-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Open-1350-X-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-1350-X-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Open-8192-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-8192-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Open-8192-X-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-8192-X-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Open-64-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-64-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-64-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-64-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-1350-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-1350-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-1350-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-1350-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-8192-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-8192-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-8192-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-8192-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Change-Id: I2c30ddc960a889b49c8ee8ff8073ffc4e75f43af Reviewed-on: https://go-review.googlesource.com/c/crypto/+/206977 Run-TryBot: Filippo Valsorda <filippo@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Katie Hockman <katie@golang.org>
245 lines
6.3 KiB
Go
245 lines
6.3 KiB
Go
// Copyright 2012 The Go Authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style
|
|
// license that can be found in the LICENSE file.
|
|
|
|
package poly1305
|
|
|
|
import (
|
|
"crypto/rand"
|
|
"encoding/hex"
|
|
"flag"
|
|
"testing"
|
|
"unsafe"
|
|
)
|
|
|
|
var stressFlag = flag.Bool("stress", false, "run slow stress tests")
|
|
|
|
type test struct {
|
|
in string
|
|
key string
|
|
tag string
|
|
}
|
|
|
|
func (t *test) Input() []byte {
|
|
in, err := hex.DecodeString(t.in)
|
|
if err != nil {
|
|
panic(err)
|
|
}
|
|
return in
|
|
}
|
|
|
|
func (t *test) Key() [32]byte {
|
|
buf, err := hex.DecodeString(t.key)
|
|
if err != nil {
|
|
panic(err)
|
|
}
|
|
var key [32]byte
|
|
copy(key[:], buf[:32])
|
|
return key
|
|
}
|
|
|
|
func (t *test) Tag() [16]byte {
|
|
buf, err := hex.DecodeString(t.tag)
|
|
if err != nil {
|
|
panic(err)
|
|
}
|
|
var tag [16]byte
|
|
copy(tag[:], buf[:16])
|
|
return tag
|
|
}
|
|
|
|
func testSum(t *testing.T, unaligned bool, sumImpl func(tag *[TagSize]byte, msg []byte, key *[32]byte)) {
|
|
var tag [16]byte
|
|
for i, v := range testData {
|
|
in := v.Input()
|
|
if unaligned {
|
|
in = unalignBytes(in)
|
|
}
|
|
key := v.Key()
|
|
sumImpl(&tag, in, &key)
|
|
if tag != v.Tag() {
|
|
t.Errorf("%d: expected %x, got %x", i, v.Tag(), tag[:])
|
|
}
|
|
if !Verify(&tag, in, &key) {
|
|
t.Errorf("%d: tag didn't verify", i)
|
|
}
|
|
// If the key is zero, the tag will always be zero, independent of the input.
|
|
if len(in) > 0 && key != [32]byte{} {
|
|
in[0] ^= 0xff
|
|
if Verify(&tag, in, &key) {
|
|
t.Errorf("%d: tag verified after altering the input", i)
|
|
}
|
|
in[0] ^= 0xff
|
|
}
|
|
// If the input is empty, the tag only depends on the second half of the key.
|
|
if len(in) > 0 {
|
|
key[0] ^= 0xff
|
|
if Verify(&tag, in, &key) {
|
|
t.Errorf("%d: tag verified after altering the key", i)
|
|
}
|
|
key[0] ^= 0xff
|
|
}
|
|
tag[0] ^= 0xff
|
|
if Verify(&tag, in, &key) {
|
|
t.Errorf("%d: tag verified after altering the tag", i)
|
|
}
|
|
tag[0] ^= 0xff
|
|
}
|
|
}
|
|
|
|
func TestBurnin(t *testing.T) {
|
|
// This test can be used to sanity-check significant changes. It can
|
|
// take about many minutes to run, even on fast machines. It's disabled
|
|
// by default.
|
|
if !*stressFlag {
|
|
t.Skip("skipping without -stress")
|
|
}
|
|
|
|
var key [32]byte
|
|
var input [25]byte
|
|
var output [16]byte
|
|
|
|
for i := range key {
|
|
key[i] = 1
|
|
}
|
|
for i := range input {
|
|
input[i] = 2
|
|
}
|
|
|
|
for i := uint64(0); i < 1e10; i++ {
|
|
Sum(&output, input[:], &key)
|
|
copy(key[0:], output[:])
|
|
copy(key[16:], output[:])
|
|
copy(input[:], output[:])
|
|
copy(input[16:], output[:])
|
|
}
|
|
|
|
const expected = "5e3b866aea0b636d240c83c428f84bfa"
|
|
if got := hex.EncodeToString(output[:]); got != expected {
|
|
t.Errorf("expected %s, got %s", expected, got)
|
|
}
|
|
}
|
|
|
|
func TestSum(t *testing.T) { testSum(t, false, Sum) }
|
|
func TestSumUnaligned(t *testing.T) { testSum(t, true, Sum) }
|
|
func TestSumGeneric(t *testing.T) { testSum(t, false, sumGeneric) }
|
|
func TestSumGenericUnaligned(t *testing.T) { testSum(t, true, sumGeneric) }
|
|
|
|
func TestWriteGeneric(t *testing.T) { testWriteGeneric(t, false) }
|
|
func TestWriteGenericUnaligned(t *testing.T) { testWriteGeneric(t, true) }
|
|
func TestWrite(t *testing.T) { testWrite(t, false) }
|
|
func TestWriteUnaligned(t *testing.T) { testWrite(t, true) }
|
|
|
|
func testWriteGeneric(t *testing.T, unaligned bool) {
|
|
for i, v := range testData {
|
|
key := v.Key()
|
|
input := v.Input()
|
|
var out [16]byte
|
|
|
|
if unaligned {
|
|
input = unalignBytes(input)
|
|
}
|
|
h := newMACGeneric(&key)
|
|
n, err := h.Write(input[:len(input)/3])
|
|
if err != nil || n != len(input[:len(input)/3]) {
|
|
t.Errorf("#%d: unexpected Write results: n = %d, err = %v", i, n, err)
|
|
}
|
|
n, err = h.Write(input[len(input)/3:])
|
|
if err != nil || n != len(input[len(input)/3:]) {
|
|
t.Errorf("#%d: unexpected Write results: n = %d, err = %v", i, n, err)
|
|
}
|
|
h.Sum(&out)
|
|
if tag := v.Tag(); out != tag {
|
|
t.Errorf("%d: expected %x, got %x", i, tag[:], out[:])
|
|
}
|
|
}
|
|
}
|
|
|
|
func testWrite(t *testing.T, unaligned bool) {
|
|
for i, v := range testData {
|
|
key := v.Key()
|
|
input := v.Input()
|
|
var out [16]byte
|
|
|
|
if unaligned {
|
|
input = unalignBytes(input)
|
|
}
|
|
h := New(&key)
|
|
n, err := h.Write(input[:len(input)/3])
|
|
if err != nil || n != len(input[:len(input)/3]) {
|
|
t.Errorf("#%d: unexpected Write results: n = %d, err = %v", i, n, err)
|
|
}
|
|
n, err = h.Write(input[len(input)/3:])
|
|
if err != nil || n != len(input[len(input)/3:]) {
|
|
t.Errorf("#%d: unexpected Write results: n = %d, err = %v", i, n, err)
|
|
}
|
|
h.Sum(out[:0])
|
|
tag := v.Tag()
|
|
if out != tag {
|
|
t.Errorf("%d: expected %x, got %x", i, tag[:], out[:])
|
|
}
|
|
if !h.Verify(tag[:]) {
|
|
t.Errorf("%d: Verify failed", i)
|
|
}
|
|
tag[0] ^= 0xff
|
|
if h.Verify(tag[:]) {
|
|
t.Errorf("%d: Verify succeeded after modifying the tag", i)
|
|
}
|
|
}
|
|
}
|
|
|
|
func benchmarkSum(b *testing.B, size int, unaligned bool) {
|
|
var out [16]byte
|
|
var key [32]byte
|
|
in := make([]byte, size)
|
|
if unaligned {
|
|
in = unalignBytes(in)
|
|
}
|
|
rand.Read(in)
|
|
b.SetBytes(int64(len(in)))
|
|
b.ResetTimer()
|
|
for i := 0; i < b.N; i++ {
|
|
Sum(&out, in, &key)
|
|
}
|
|
}
|
|
|
|
func benchmarkWrite(b *testing.B, size int, unaligned bool) {
|
|
var key [32]byte
|
|
h := New(&key)
|
|
in := make([]byte, size)
|
|
if unaligned {
|
|
in = unalignBytes(in)
|
|
}
|
|
rand.Read(in)
|
|
b.SetBytes(int64(len(in)))
|
|
b.ResetTimer()
|
|
for i := 0; i < b.N; i++ {
|
|
h.Write(in)
|
|
}
|
|
}
|
|
|
|
func Benchmark64(b *testing.B) { benchmarkSum(b, 64, false) }
|
|
func Benchmark1K(b *testing.B) { benchmarkSum(b, 1024, false) }
|
|
func Benchmark2M(b *testing.B) { benchmarkSum(b, 2*1024*1024, false) }
|
|
func Benchmark64Unaligned(b *testing.B) { benchmarkSum(b, 64, true) }
|
|
func Benchmark1KUnaligned(b *testing.B) { benchmarkSum(b, 1024, true) }
|
|
func Benchmark2MUnaligned(b *testing.B) { benchmarkSum(b, 2*1024*1024, true) }
|
|
|
|
func BenchmarkWrite64(b *testing.B) { benchmarkWrite(b, 64, false) }
|
|
func BenchmarkWrite1K(b *testing.B) { benchmarkWrite(b, 1024, false) }
|
|
func BenchmarkWrite2M(b *testing.B) { benchmarkWrite(b, 2*1024*1024, false) }
|
|
func BenchmarkWrite64Unaligned(b *testing.B) { benchmarkWrite(b, 64, true) }
|
|
func BenchmarkWrite1KUnaligned(b *testing.B) { benchmarkWrite(b, 1024, true) }
|
|
func BenchmarkWrite2MUnaligned(b *testing.B) { benchmarkWrite(b, 2*1024*1024, true) }
|
|
|
|
func unalignBytes(in []byte) []byte {
|
|
out := make([]byte, len(in)+1)
|
|
if uintptr(unsafe.Pointer(&out[0]))&(unsafe.Alignof(uint32(0))-1) == 0 {
|
|
out = out[1:]
|
|
} else {
|
|
out = out[:len(in)]
|
|
}
|
|
copy(out, in)
|
|
return out
|
|
}
|