yaole/xgcl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

init: v1.0.0

Browse Source
This commit is contained in:
yaole
2026-05-27 23:03:00 +08:00
commit 8d97f750eb
466 changed files with 80067 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
internal/xor/export.go
+3
View File
@@ -0,0 +1,3 @@
package xor
var XorBytes = xorBytes
internal/xor/xor_amd64.go
+27
View File
@@ -0,0 +1,27 @@
// Copyright 2018 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 xor
// xorBytes xors the bytes in a and b. The destination should have enough
// space, otherwise xorBytes will panic. Returns the number of bytes xor'd.
func xorBytes(dst, a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
if n == 0 {
return 0
}
_ = dst[n-1]
xorBytesSSE2(&dst[0], &a[0], &b[0], n) // amd64 must have SSE2
return n
}
func xorWords(dst, a, b []byte) {
xorBytes(dst, a, b)
}
//go:noescape
func xorBytesSSE2(dst, a, b *byte, n int)
internal/xor/xor_amd64.s
+54
View File
@@ -0,0 +1,54 @@
// Copyright 2018 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 xorBytesSSE2(dst, a, b *byte, n int)
TEXT ·xorBytesSSE2(SB), NOSPLIT, $0
MOVQ dst+0(FP), BX
MOVQ a+8(FP), SI
MOVQ b+16(FP), CX
MOVQ n+24(FP), DX
TESTQ $15, DX // AND 15 & len, if not zero jump to not_aligned.
JNZ not_aligned
aligned:
MOVQ $0, AX // position in slices
loop16b:
MOVOU (SI)(AX*1), X0 // XOR 16byte forwards.
MOVOU (CX)(AX*1), X1
PXOR X1, X0
MOVOU X0, (BX)(AX*1)
ADDQ $16, AX
CMPQ DX, AX
JNE loop16b
RET
loop_1b:
SUBQ $1, DX // XOR 1byte backwards.
MOVB (SI)(DX*1), DI
MOVB (CX)(DX*1), AX
XORB AX, DI
MOVB DI, (BX)(DX*1)
TESTQ $7, DX // AND 7 & len, if not zero jump to loop_1b.
JNZ loop_1b
CMPQ DX, $0 // if len is 0, ret.
JE ret
TESTQ $15, DX // AND 15 & len, if zero jump to aligned.
JZ aligned
not_aligned:
TESTQ $7, DX // AND $7 & len, if not zero jump to loop_1b.
JNE loop_1b
SUBQ $8, DX // XOR 8bytes backwards.
MOVQ (SI)(DX*1), DI
MOVQ (CX)(DX*1), AX
XORQ AX, DI
MOVQ DI, (BX)(DX*1)
CMPQ DX, $16 // if len is greater or equal 16 here, it must be aligned.
JGE aligned
ret:
RET
internal/xor/xor_arm64.go
+29
View File
@@ -0,0 +1,29 @@
// Copyright 2020 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 xor
// xorBytes xors the bytes in a and b. The destination should have enough
// space, otherwise xorBytes will panic. Returns the number of bytes xor'd.
func xorBytes(dst, a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
if n == 0 {
return 0
}
// make sure dst has enough space
_ = dst[n-1]
xorBytesARM64(&dst[0], &a[0], &b[0], n)
return n
}
func xorWords(dst, a, b []byte) {
xorBytes(dst, a, b)
}
//go:noescape
func xorBytesARM64(dst, a, b *byte, n int)
internal/xor/xor_arm64.s
+67
View File
@@ -0,0 +1,67 @@
// Copyright 2020 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 xorBytesARM64(dst, a, b *byte, n int)
TEXT ·xorBytesARM64(SB), NOSPLIT|NOFRAME, $0
MOVD dst+0(FP), R0
MOVD a+8(FP), R1
MOVD b+16(FP), R2
MOVD n+24(FP), R3
CMP $64, R3
BLT tail
loop_64:
VLD1.P 64(R1), [V0.B16, V1.B16, V2.B16, V3.B16]
VLD1.P 64(R2), [V4.B16, V5.B16, V6.B16, V7.B16]
VEOR V0.B16, V4.B16, V4.B16
VEOR V1.B16, V5.B16, V5.B16
VEOR V2.B16, V6.B16, V6.B16
VEOR V3.B16, V7.B16, V7.B16
VST1.P [V4.B16, V5.B16, V6.B16, V7.B16], 64(R0)
SUBS $64, R3
CMP $64, R3
BGE loop_64
tail:
// quick end
CBZ R3, end
TBZ $5, R3, less_than32
VLD1.P 32(R1), [V0.B16, V1.B16]
VLD1.P 32(R2), [V2.B16, V3.B16]
VEOR V0.B16, V2.B16, V2.B16
VEOR V1.B16, V3.B16, V3.B16
VST1.P [V2.B16, V3.B16], 32(R0)
less_than32:
TBZ $4, R3, less_than16
LDP.P 16(R1), (R11, R12)
LDP.P 16(R2), (R13, R14)
EOR R11, R13, R13
EOR R12, R14, R14
STP.P (R13, R14), 16(R0)
less_than16:
TBZ $3, R3, less_than8
MOVD.P 8(R1), R11
MOVD.P 8(R2), R12
EOR R11, R12, R12
MOVD.P R12, 8(R0)
less_than8:
TBZ $2, R3, less_than4
MOVWU.P 4(R1), R13
MOVWU.P 4(R2), R14
EORW R13, R14, R14
MOVWU.P R14, 4(R0)
less_than4:
TBZ $1, R3, less_than2
MOVHU.P 2(R1), R15
MOVHU.P 2(R2), R16
EORW R15, R16, R16
MOVHU.P R16, 2(R0)
less_than2:
TBZ $0, R3, end
MOVBU (R1), R17
MOVBU (R2), R19
EORW R17, R19, R19
MOVBU R19, (R0)
end:
RET
internal/xor/xor_generic.go
+92
View File
@@ -0,0 +1,92 @@
// 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 !amd64 && !ppc64 && !ppc64le && !arm64
// +build !amd64,!ppc64,!ppc64le,!arm64
package xor
import (
"runtime"
"unsafe"
)
// xorBytes xors the bytes in a and b. The destination should have enough
// space, otherwise xorBytes will panic. Returns the number of bytes xor'd.
func xorBytes(dst, a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
if n == 0 {
return 0
}
switch {
case supportsUnaligned:
fastXORBytes(dst, a, b, n)
default:
// TODO(hanwen): if (dst, a, b) have common alignment
// we could still try fastXORBytes. It is not clear
// how often this happens, and it's only worth it if
// the block encryption itself is hardware
// accelerated.
safeXORBytes(dst, a, b, n)
}
return n
}
const wordSize = int(unsafe.Sizeof(uintptr(0)))
const supportsUnaligned = runtime.GOARCH == "386" || runtime.GOARCH == "ppc64" || runtime.GOARCH == "ppc64le" || runtime.GOARCH == "s390x"
// fastXORBytes xors in bulk. It only works on architectures that
// support unaligned read/writes.
// n needs to be smaller or equal than the length of a and b.
func fastXORBytes(dst, a, b []byte, n int) {
// Assert dst has enough space
_ = dst[n-1]
w := n / wordSize
if w > 0 {
dw := *(*[]uintptr)(unsafe.Pointer(&dst))
aw := *(*[]uintptr)(unsafe.Pointer(&a))
bw := *(*[]uintptr)(unsafe.Pointer(&b))
for i := 0; i < w; i++ {
dw[i] = aw[i] ^ bw[i]
}
}
for i := (n - n%wordSize); i < n; i++ {
dst[i] = a[i] ^ b[i]
}
}
// n needs to be smaller or equal than the length of a and b.
func safeXORBytes(dst, a, b []byte, n int) {
for i := 0; i < n; i++ {
dst[i] = a[i] ^ b[i]
}
}
// fastXORWords XORs multiples of 4 or 8 bytes (depending on architecture.)
// The arguments are assumed to be of equal length.
func fastXORWords(dst, a, b []byte) {
dw := *(*[]uintptr)(unsafe.Pointer(&dst))
aw := *(*[]uintptr)(unsafe.Pointer(&a))
bw := *(*[]uintptr)(unsafe.Pointer(&b))
n := len(b) / wordSize
for i := 0; i < n; i++ {
dw[i] = aw[i] ^ bw[i]
}
}
// fastXORWords XORs multiples of 4 or 8 bytes (depending on architecture.)
// The slice arguments a and b are assumed to be of equal length.
func xorWords(dst, a, b []byte) {
if supportsUnaligned {
fastXORWords(dst, a, b)
} else {
safeXORBytes(dst, a, b, len(b))
}
}
internal/xor/xor_ppc64x.go
+30
View File
@@ -0,0 +1,30 @@
// Copyright 2018 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 ppc64 || ppc64le
// +build ppc64 ppc64le
package xor
// xorBytes xors the bytes in a and b. The destination should have enough
// space, otherwise xorBytes will panic. Returns the number of bytes xor'd.
func xorBytes(dst, a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
if n == 0 {
return 0
}
_ = dst[n-1]
xorBytesVSX(&dst[0], &a[0], &b[0], n)
return n
}
func xorWords(dst, a, b []byte) {
xorBytes(dst, a, b)
}
//go:noescape
func xorBytesVSX(dst, a, b *byte, n int)
internal/xor/xor_ppc64x.s
+87
View File
@@ -0,0 +1,87 @@
// Copyright 2018 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.
// +build ppc64 ppc64le
#include "textflag.h"
// func xorBytesVSX(dst, a, b *byte, n int)
TEXT ·xorBytesVSX(SB), NOSPLIT, $0
MOVD dst+0(FP), R3 // R3 = dst
MOVD a+8(FP), R4 // R4 = a
MOVD b+16(FP), R5 // R5 = b
MOVD n+24(FP), R6 // R6 = n
CMPU R6, $32, CR7 // Check if n ≥ 32 bytes
MOVD R0, R8 // R8 = index
CMPU R6, $8, CR6 // Check if 8 ≤ n < 32 bytes
BLT CR6, small // Smaller than 8
BLT CR7, xor16 // Case for 16 ≤ n < 32 bytes
// Case for n ≥ 32 bytes
preloop32:
SRD $5, R6, R7 // Setup loop counter
MOVD R7, CTR
MOVD $16, R10
ANDCC $31, R6, R9 // Check for tailing bytes for later
loop32:
LXVD2X (R4)(R8), VS32 // VS32 = a[i,...,i+15]
LXVD2X (R4)(R10), VS34
LXVD2X (R5)(R8), VS33 // VS33 = b[i,...,i+15]
LXVD2X (R5)(R10), VS35
XXLXOR VS32, VS33, VS32 // VS34 = a[] ^ b[]
XXLXOR VS34, VS35, VS34
STXVD2X VS32, (R3)(R8) // Store to dst
STXVD2X VS34, (R3)(R10)
ADD $32, R8 // Update index
ADD $32, R10
BC 16, 0, loop32 // bdnz loop16
BEQ CR0, done
MOVD R9, R6
CMP R6, $8
BLT small
xor16:
CMP R6, $16
BLT xor8
LXVD2X (R4)(R8), VS32
LXVD2X (R5)(R8), VS33
XXLXOR VS32, VS33, VS32
STXVD2X VS32, (R3)(R8)
ADD $16, R8
ADD $-16, R6
CMP R6, $8
BLT small
xor8:
// Case for 8 ≤ n < 16 bytes
MOVD (R4)(R8), R14 // R14 = a[i,...,i+7]
MOVD (R5)(R8), R15 // R15 = b[i,...,i+7]
XOR R14, R15, R16 // R16 = a[] ^ b[]
SUB $8, R6 // n = n - 8
MOVD R16, (R3)(R8) // Store to dst
ADD $8, R8
// Check if we're finished
CMP R6, R0
BGT small
RET
// Case for n < 8 bytes and tailing bytes from the
// previous cases.
small:
CMP R6, R0
BEQ done
MOVD R6, CTR // Setup loop counter
loop:
MOVBZ (R4)(R8), R14 // R14 = a[i]
MOVBZ (R5)(R8), R15 // R15 = b[i]
XOR R14, R15, R16 // R16 = a[i] ^ b[i]
MOVB R16, (R3)(R8) // Store to dst
ADD $1, R8
BC 16, 0, loop // bdnz loop
done:
RET
internal/xor/xor_test.go
+76
View File
@@ -0,0 +1,76 @@
// 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.
package xor_test
import (
"bytes"
"crypto/rand"
"fmt"
"io"
"testing"
"xdx.jelly/xgcl/internal/xor"
)
func TestXOR(t *testing.T) {
for j := 1; j <= 1024; j++ {
if testing.Short() && j > 16 {
break
}
for alignP := 0; alignP < 2; alignP++ {
for alignQ := 0; alignQ < 2; alignQ++ {
for alignD := 0; alignD < 2; alignD++ {
p := make([]byte, j)[alignP:]
q := make([]byte, j)[alignQ:]
d1 := make([]byte, j+alignD)[alignD:]
d2 := make([]byte, j+alignD)[alignD:]
if _, err := io.ReadFull(rand.Reader, p); err != nil {
t.Fatal(err)
}
if _, err := io.ReadFull(rand.Reader, q); err != nil {
t.Fatal(err)
}
xor.XorBytes(d1, p, q)
n := min(p, q)
for i := 0; i < n; i++ {
d2[i] = p[i] ^ q[i]
}
if !bytes.Equal(d1, d2) {
t.Logf("p: %#v", p)
t.Logf("q: %#v", q)
t.Logf("expect: %#v", d2)
t.Logf("result: %#v", d1)
t.Fatal("not equal")
}
}
}
}
}
}
func min(a, b []byte) int {
n := len(a)
if len(b) < n {
n = len(b)
}
return n
}
func BenchmarkXORBytes(b *testing.B) {
dst := make([]byte, 1<<15)
data0 := make([]byte, 1<<15)
data1 := make([]byte, 1<<15)
sizes := []int64{1 << 3, 1 << 7, 1 << 11, 1 << 15}
for _, size := range sizes {
b.Run(fmt.Sprintf("%dBytes", size), func(b *testing.B) {
s0 := data0[:size]
s1 := data1[:size]
b.SetBytes(int64(size))
for i := 0; i < b.N; i++ {
xor.XorBytes(dst, s0, s1)
}
})
}
}