lsh256 lsh512 alignment

RyuaNerin · Jun 11, 2024 · 579d155 · 579d155
1 parent b20cd67
commit 579d155
Show file tree

Hide file tree

Showing 13 changed files with 202 additions and 54 deletions.
diff --git a/internal/aligned/aligned.go b/internal/aligned/aligned.go
@@ -0,0 +1,24 @@
+package aligned
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestIsAligned(t *testing.T, align uintptr, raw interface{}, fieldName string) {
+	v := reflect.ValueOf(raw)
+	for v.Kind() == reflect.Ptr {
+		v = v.Elem()
+	}
+	vt := v.Type()
+
+	vtf, ok := vt.FieldByName(fieldName)
+	if !ok {
+		t.Errorf("%s.%s not found", vt.Name(), fieldName)
+		return
+	}
+
+	if vtf.Offset&(align-1) != 0 {
+		t.Errorf("%s.%s is not Aligned. offset=%d", vt.Name(), fieldName, vtf.Offset)
+	}
+}
diff --git a/internal/binary.go b/internal/binary.go
@@ -2,10 +2,14 @@ package internal
 
 import "encoding/binary"
 
-func ConsumeUint64(b []byte) ([]byte, uint64) {
-	return b[8:], binary.BigEndian.Uint64(b)
+func ConsumeUint16(b []byte) ([]byte, uint16) {
+	return b[2:], binary.BigEndian.Uint16(b)
 }
 
 func ConsumeUint32(b []byte) ([]byte, uint32) {
 	return b[4:], binary.BigEndian.Uint32(b)
 }
+
+func ConsumeUint64(b []byte) ([]byte, uint64) {
+	return b[8:], binary.BigEndian.Uint64(b)
+}
diff --git a/internal/endian.go b/internal/endian.go
@@ -1,5 +1,11 @@
 package internal
 
+func AppendBigUint8(b []byte, v byte) []byte {
+	return append(b,
+		v,
+	)
+}
+
 func AppendBigUint16(b []byte, v uint16) []byte {
 	return append(b,
 		byte(v>>8),

diff --git a/lsh256/asm.go b/lsh256/asm.go
@@ -18,43 +18,48 @@ type simdSet struct {
 }
 
 func (simd *simdSet) NewContext(size int) hash.Hash {
-	ctx := new(lsh256ContextAsm)
-	ctx.simd = simd
-	ctx.size = size
+	ctx := &lsh256ContextAsm{
+		simd:    simd,
+		algType: size,
+	}
 	ctx.Reset()
 	return ctx
 }
 
 func (simd *simdSet) Sum(size int, data []byte) [Size]byte {
-	var ctx lsh256ContextAsm
-	ctx.simd = simd
-	ctx.size = size
+	ctx := lsh256ContextAsm{
+		simd:    simd,
+		algType: size,
+	}
 	ctx.Reset()
 	ctx.Write(data)
 
 	return ctx.checkSum()
 }
 
-const contextDataSize = 16 + 16 + 4*8 + 4*8 + 128
-
 type lsh256ContextAsm struct {
-	//nolint:unused
-	data [contextDataSize]byte // 최상단으로 배치하여 aligned 문제 수정...
+	// 16 bytes aligned
+	algType           int
+	_                 uintptr
+	remainDataByteLen int
+	_                 uintptr
+	cvL               [4]uint64
+	cvR               [4]uint64
+	lastBlock         [BlockSize]byte
 
 	simd *simdSet
-	size int
 }
 
 func (ctx *lsh256ContextAsm) Size() int {
-	return ctx.size
+	return ctx.algType
 }
 
 func (ctx *lsh256ContextAsm) BlockSize() int {
 	return BlockSize
 }
 
 func (ctx *lsh256ContextAsm) Reset() {
-	ctx.simd.init(ctx, uint64(ctx.size))
+	ctx.simd.init(ctx, uint64(ctx.algType))
 }
 
 func (ctx *lsh256ContextAsm) Write(data []byte) (n int, err error) {
@@ -69,7 +74,7 @@ func (ctx *lsh256ContextAsm) Write(data []byte) (n int, err error) {
 func (ctx *lsh256ContextAsm) Sum(p []byte) []byte {
 	ctx0 := *ctx
 	hash := ctx0.checkSum()
-	return append(p, hash[:ctx.size]...)
+	return append(p, hash[:ctx.algType]...)
 }
 
 func (ctx *lsh256ContextAsm) checkSum() (hash [Size]byte) {

diff --git a/lsh256/asm_amd64_test.go b/lsh256/asm_amd64_test.go
@@ -8,6 +8,8 @@ package lsh256
 import (
 	"testing"
 
+	"github.com/RyuaNerin/go-krypto/internal/aligned"
+
 	. "github.com/RyuaNerin/testingutil"
 )
 
@@ -17,6 +19,15 @@ var (
 	newAVX2  = simdSetAVX2.NewContext
 )
 
+func Test_Aligned(t *testing.T) {
+	var ctx lsh256ContextAsm
+	aligned.TestIsAligned(t, 16, &ctx, "algType")
+	aligned.TestIsAligned(t, 16, &ctx, "remainDataByteLen")
+	aligned.TestIsAligned(t, 16, &ctx, "cvL")
+	aligned.TestIsAligned(t, 16, &ctx, "cvR")
+	aligned.TestIsAligned(t, 16, &ctx, "lastBlock")
+}
+
 func Test_ShortWrite_SSE2(t *testing.T)  { HTSWA(t, as, newSSE2, !hasSSE2) }
 func Test_ShortWrite_SSSE3(t *testing.T) { HTSWA(t, as, newSSSE3, !hasSSSE3) }
 func Test_ShortWrite_AVX2(t *testing.T)  { HTSWA(t, as, newAVX2, !hasAVX2) }

diff --git a/lsh256/asm_arm64_test.go b/lsh256/asm_arm64_test.go
@@ -8,11 +8,22 @@ package lsh256
 import (
 	"testing"
 
+	"github.com/RyuaNerin/go-krypto/internal/aligned"
+
 	. "github.com/RyuaNerin/testingutil"
 )
 
 var newNEON = simdSetNEON.NewContext
 
+func Test_Aligned(t *testing.T) {
+	var ctx lsh256ContextAsm
+	aligned.TestIsAligned(t, 16, &ctx, "algType")
+	aligned.TestIsAligned(t, 16, &ctx, "remainDataByteLen")
+	aligned.TestIsAligned(t, 16, &ctx, "cvL")
+	aligned.TestIsAligned(t, 16, &ctx, "cvR")
+	aligned.TestIsAligned(t, 16, &ctx, "lastBlock")
+}
+
 func Test_ShortWrite_NEON(t *testing.T) { HTSWA(t, as, newNEON, !hasNEON) }
 
 func Test_LSH224_NEON(t *testing.T) { HT(t, newNEON(Size224), testCases224, !hasNEON) }

diff --git a/lsh256/encoding_asm.go b/lsh256/encoding_asm.go
@@ -7,7 +7,6 @@ package lsh256
 
 import (
 	"encoding"
-	"encoding/binary"
 	"errors"
 
 	"github.com/RyuaNerin/go-krypto/internal"
@@ -16,15 +15,32 @@ import (
 const (
 	magic         = "lsh\x00"
 	marshaledSize = len(magic) +
-		/*data*/ contextDataSize +
-		/*size*/ 2
+		1 + /* algType */
+		1 + /* remainDataByteLen */
+		4*8 + /* cvL */
+		4*8 + /* cvR */
+		BlockSize /* lastBlock */
 )
 
 func (ctx *lsh256ContextAsm) MarshalBinary() ([]byte, error) {
 	b := make([]byte, 0, marshaledSize)
 	b = append(b, magic...)
-	b = append(b, ctx.data[:]...)
-	b = internal.AppendBigUint16(b, uint16(ctx.size))
+
+	b = internal.AppendBigUint8(b, uint8(ctx.algType))
+
+	b = internal.AppendBigUint8(b, uint8(ctx.remainDataByteLen))
+
+	for _, v := range ctx.cvL {
+		b = internal.AppendBigUint64(b, v)
+	}
+
+	for _, v := range ctx.cvR {
+		b = internal.AppendBigUint64(b, v)
+	}
+
+	b = append(b, ctx.lastBlock[:ctx.remainDataByteLen]...)
+	b = b[:len(b)+len(ctx.lastBlock)-ctx.remainDataByteLen] // already zero
+
 	return b, nil
 }
 
@@ -37,8 +53,21 @@ func (ctx *lsh256ContextAsm) UnmarshalBinary(b []byte) error {
 	}
 
 	b = b[len(magic):]
-	b = b[copy(ctx.data[:], b[:contextDataSize]):]
-	ctx.size = int(binary.BigEndian.Uint16(b))
+
+	ctx.algType = int(b[0])
+	ctx.remainDataByteLen = int(b[1])
+	b = b[2:]
+
+	for i := range ctx.cvL {
+		b, ctx.cvL[i] = internal.ConsumeUint64(b)
+	}
+
+	for i := range ctx.cvR {
+		b, ctx.cvR[i] = internal.ConsumeUint64(b)
+	}
+
+	copy(ctx.lastBlock[:], b)
+
 	return nil
 }
 

diff --git a/lsh256/generic.go b/lsh256/generic.go
@@ -12,24 +12,24 @@ import (
 )
 
 func newContextGo(size int) hash.Hash {
-	ctx := new(lsh256ContextGo)
-	ctx.outlenbytes = size
+	ctx := &lsh256ContextGo{
+		outlenbytes: size,
+	}
 	ctx.Reset()
 
 	return ctx
 }
 
 func sumGo(size int, data []byte) [Size]byte {
-	var ctx lsh256ContextGo
-	ctx.outlenbytes = size
+	ctx := lsh256ContextGo{
+		outlenbytes: size,
+	}
 	ctx.Reset()
 	ctx.Write(data)
 
 	return ctx.checkSum()
 }
 
-// java to go
-
 const (
 	numStep = 26
 

diff --git a/lsh512/asm.go b/lsh512/asm.go
@@ -18,43 +18,48 @@ type simdSet struct {
 }
 
 func (simd *simdSet) NewContext(size int) hash.Hash {
-	ctx := new(lsh512ContextAsm)
-	ctx.simd = simd
-	ctx.size = size
+	ctx := &lsh512ContextAsm{
+		simd:    simd,
+		algType: size,
+	}
 	ctx.Reset()
 	return ctx
 }
 
 func (simd *simdSet) Sum(size int, data []byte) [Size]byte {
-	var ctx lsh512ContextAsm
-	ctx.simd = simd
-	ctx.size = size
+	ctx := lsh512ContextAsm{
+		simd:    simd,
+		algType: size,
+	}
 	ctx.Reset()
 	ctx.Write(data)
 
 	return ctx.checkSum()
 }
 
-const contextDataSize = 16 + 16 + 8*8 + 8*8 + 256
-
 type lsh512ContextAsm struct {
-	//nolint:unused
-	data [contextDataSize]byte // 최상단으로 배치하여 aligned 문제 수정...
+	// 16 bytes aligned
+	algType           int
+	_                 uintptr
+	remainDataByteLen int
+	_                 [1]uintptr
+	cvL               [8]uint64
+	cvR               [8]uint64
+	lastBlock         [BlockSize]byte
 
 	simd *simdSet
-	size int
 }
 
 func (ctx *lsh512ContextAsm) Size() int {
-	return ctx.size
+	return ctx.algType
 }
 
 func (ctx *lsh512ContextAsm) BlockSize() int {
 	return BlockSize
 }
 
 func (ctx *lsh512ContextAsm) Reset() {
-	ctx.simd.init(ctx, uint64(ctx.size))
+	ctx.simd.init(ctx, uint64(ctx.algType))
 }
 
 func (ctx *lsh512ContextAsm) Write(data []byte) (n int, err error) {
@@ -69,7 +74,7 @@ func (ctx *lsh512ContextAsm) Write(data []byte) (n int, err error) {
 func (ctx *lsh512ContextAsm) Sum(p []byte) []byte {
 	ctx0 := *ctx
 	hash := ctx0.checkSum()
-	return append(p, hash[:ctx.size]...)
+	return append(p, hash[:ctx.algType]...)
 }
 
 func (ctx *lsh512ContextAsm) checkSum() (hash [Size]byte) {

diff --git a/lsh512/asm_amd64_test.go b/lsh512/asm_amd64_test.go
@@ -8,6 +8,8 @@ package lsh512
 import (
 	"testing"
 
+	"github.com/RyuaNerin/go-krypto/internal/aligned"
+
 	. "github.com/RyuaNerin/testingutil"
 )
 
@@ -17,6 +19,15 @@ var (
 	newAVX2  = simdSetAVX2.NewContext
 )
 
+func Test_Aligned(t *testing.T) {
+	var ctx lsh512ContextAsm
+	aligned.TestIsAligned(t, 16, &ctx, "algType")
+	aligned.TestIsAligned(t, 16, &ctx, "remainDataByteLen")
+	aligned.TestIsAligned(t, 16, &ctx, "cvL")
+	aligned.TestIsAligned(t, 16, &ctx, "cvR")
+	aligned.TestIsAligned(t, 16, &ctx, "lastBlock")
+}
+
 func Test_ShortWrite_SSE2(t *testing.T)  { HTSWA(t, as, newSSE2, !hasSSE2) }
 func Test_ShortWrite_SSSE3(t *testing.T) { HTSWA(t, as, newSSSE3, !hasSSSE3) }
 func Test_ShortWrite_AVX2(t *testing.T)  { HTSWA(t, as, newAVX2, !hasAVX2) }

diff --git a/lsh512/asm_arm64_test.go b/lsh512/asm_arm64_test.go
@@ -8,11 +8,22 @@ package lsh512
 import (
 	"testing"
 
+	"github.com/RyuaNerin/go-krypto/internal/aligned"
+
 	. "github.com/RyuaNerin/testingutil"
 )
 
 var newNEON = simdSetNEON.NewContext
 
+func Test_Aligned(t *testing.T) {
+	var ctx lsh512ContextAsm
+	aligned.TestIsAligned(t, 16, &ctx, "algType")
+	aligned.TestIsAligned(t, 16, &ctx, "remainDataByteLen")
+	aligned.TestIsAligned(t, 16, &ctx, "cvL")
+	aligned.TestIsAligned(t, 16, &ctx, "cvR")
+	aligned.TestIsAligned(t, 16, &ctx, "lastBlock")
+}
+
 func Test_ShortWrite_NEON(t *testing.T) { HTSWA(t, as, newNEON, !hasNEON) }
 
 func Test_LSH512_224_NEON(t *testing.T) { HT(t, newNEON(Size224), testCases224, !hasNEON) }