perso/syncthing-arm
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

lib/connections: Use our own fork of kcp (fixes #4063)

Browse Source 
This updates kcp and uses our own fork which:

1. Keys sessions not just by remote address, but by remote address +
conversation id 2. Allows not to close connections that were passed directly
to the library. 3. Resets cache key if the session gets terminated.

GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4339
LGTM: calmh
This commit is contained in:
Audrius Butkevicius
2017-09-02 06:04:35 +00:00
committed by Jakob Borg
parent ab132ff6fe
commit cbcc3ea132
19 changed files with 1551 additions and 188 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
vendor/github.com/templexxx/xor/LICENSE generated vendored Normal file
View File

@@ -0,0 +1,21 @@
MIT License
Copyright (c) 2017 Temple3x
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

442
vendor/github.com/templexxx/xor/avx2_amd64.s generated vendored Normal file
View File

@@ -0,0 +1,442 @@
#include "textflag.h"
// addr of mem
#define DST BX
#define SRC SI
#define SRC0 TMP4
#define SRC1 TMP5
// loop args
// num of vect
#define VECT CX
#define LEN DX
// pos of matrix
#define POS R8
// tmp store
// num of vect or ...
#define TMP1 R9
// pos of matrix or ...
#define TMP2 R10
// store addr of data/parity or ...
#define TMP3 R11
#define TMP4 R12
#define TMP5 R13
#define TMP6 R14
// func bytesAVX2mini(dst, src0, src1 []byte, size int)
TEXT ·bytesAVX2mini(SB), NOSPLIT, $0
MOVQ len+72(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $31, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop32b:
VMOVDQU (SRC0)(POS*1), Y0
VPXOR (SRC1)(POS*1), Y0, Y0
VMOVDQU Y0, (DST)(POS*1)
ADDQ $32, POS
CMPQ LEN, POS
JNE loop32b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $31, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $31, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $32
JGE aligned
RET
ret:
RET
// func bytesAVX2small(dst, src0, src1 []byte, size int)
TEXT ·bytesAVX2small(SB), NOSPLIT, $0
MOVQ len+72(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $127, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop128b:
VMOVDQU (SRC0)(POS*1), Y0
VMOVDQU 32(SRC0)(POS*1), Y1
VMOVDQU 64(SRC0)(POS*1), Y2
VMOVDQU 96(SRC0)(POS*1), Y3
VPXOR (SRC1)(POS*1), Y0, Y0
VPXOR 32(SRC1)(POS*1), Y1, Y1
VPXOR 64(SRC1)(POS*1), Y2, Y2
VPXOR 96(SRC1)(POS*1), Y3, Y3
VMOVDQU Y0, (DST)(POS*1)
VMOVDQU Y1, 32(DST)(POS*1)
VMOVDQU Y2, 64(DST)(POS*1)
VMOVDQU Y3, 96(DST)(POS*1)
ADDQ $128, POS
CMPQ LEN, POS
JNE loop128b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $127, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $127, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $128
JGE aligned
RET
ret:
RET
// func bytesAVX2big(dst, src0, src1 []byte, size int)
TEXT ·bytesAVX2big(SB), NOSPLIT, $0
MOVQ len+72(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $127, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop128b:
VMOVDQU (SRC0)(POS*1), Y0
VMOVDQU 32(SRC0)(POS*1), Y1
VMOVDQU 64(SRC0)(POS*1), Y2
VMOVDQU 96(SRC0)(POS*1), Y3
VPXOR (SRC1)(POS*1), Y0, Y0
VPXOR 32(SRC1)(POS*1), Y1, Y1
VPXOR 64(SRC1)(POS*1), Y2, Y2
VPXOR 96(SRC1)(POS*1), Y3, Y3
LONG $0xe77da1c4; WORD $0x0304
LONG $0xe77da1c4; WORD $0x034c; BYTE $0x20
LONG $0xe77da1c4; WORD $0x0354; BYTE $0x40
LONG $0xe77da1c4; WORD $0x035c; BYTE $0x60
ADDQ $128, POS
CMPQ LEN, POS
JNE loop128b
SFENCE
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $127, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $127, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $128
JGE aligned
RET
ret:
RET
// func matrixAVX2small(dst []byte, src [][]byte)
TEXT ·matrixAVX2small(SB), NOSPLIT, $0
MOVQ dst+0(FP), DST
MOVQ src+24(FP), SRC
MOVQ vec+32(FP), VECT
MOVQ len+8(FP), LEN
TESTQ $127, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop128b:
MOVQ VECT, TMP1
SUBQ $2, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
VMOVDQU (TMP3)(POS*1), Y0
VMOVDQU 32(TMP4)(POS*1), Y1
VMOVDQU 64(TMP3)(POS*1), Y2
VMOVDQU 96(TMP4)(POS*1), Y3
next_vect:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
VMOVDQU (TMP3)(POS*1), Y4
VMOVDQU 32(TMP4)(POS*1), Y5
VMOVDQU 64(TMP3)(POS*1), Y6
VMOVDQU 96(TMP4)(POS*1), Y7
VPXOR Y4, Y0, Y0
VPXOR Y5, Y1, Y1
VPXOR Y6, Y2, Y2
VPXOR Y7, Y3, Y3
SUBQ $1, TMP1
JGE next_vect
VMOVDQU Y0, (DST)(POS*1)
VMOVDQU Y1, 32(DST)(POS*1)
VMOVDQU Y2, 64(DST)(POS*1)
VMOVDQU Y3, 96(DST)(POS*1)
ADDQ $128, POS
CMPQ LEN, POS
JNE loop128b
RET
loop_1b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVB -1(TMP3)(LEN*1), TMP5
next_vect_1b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVB -1(TMP3)(LEN*1), TMP6
XORB TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_1b
MOVB TMP5, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $127, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP4
ANDQ $127, TMP4
loop_8b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVQ -8(TMP3)(LEN*1), TMP5
next_vect_8b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ -8(TMP3)(LEN*1), TMP6
XORQ TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_8b
MOVQ TMP5, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP4
JG loop_8b
CMPQ LEN, $128
JGE aligned
RET
ret:
RET
// func matrixAVX2big(dst []byte, src [][]byte)
TEXT ·matrixAVX2big(SB), NOSPLIT, $0
MOVQ dst+0(FP), DST
MOVQ src+24(FP), SRC
MOVQ vec+32(FP), VECT
MOVQ len+8(FP), LEN
TESTQ $127, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop128b:
MOVQ VECT, TMP1
SUBQ $2, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
VMOVDQU (TMP3)(POS*1), Y0
VMOVDQU 32(TMP4)(POS*1), Y1
VMOVDQU 64(TMP3)(POS*1), Y2
VMOVDQU 96(TMP4)(POS*1), Y3
next_vect:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
VMOVDQU (TMP3)(POS*1), Y4
VMOVDQU 32(TMP4)(POS*1), Y5
VMOVDQU 64(TMP3)(POS*1), Y6
VMOVDQU 96(TMP4)(POS*1), Y7
VPXOR Y4, Y0, Y0
VPXOR Y5, Y1, Y1
VPXOR Y6, Y2, Y2
VPXOR Y7, Y3, Y3
SUBQ $1, TMP1
JGE next_vect
LONG $0xe77da1c4; WORD $0x0304 // VMOVNTDQ go1.8 has
LONG $0xe77da1c4; WORD $0x034c; BYTE $0x20
LONG $0xe77da1c4; WORD $0x0354; BYTE $0x40
LONG $0xe77da1c4; WORD $0x035c; BYTE $0x60
ADDQ $128, POS
CMPQ LEN, POS
JNE loop128b
RET
loop_1b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVB -1(TMP3)(LEN*1), TMP5
next_vect_1b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVB -1(TMP3)(LEN*1), TMP6
XORB TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_1b
MOVB TMP5, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $127, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP4
ANDQ $127, TMP4
loop_8b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVQ -8(TMP3)(LEN*1), TMP5
next_vect_8b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ -8(TMP3)(LEN*1), TMP6
XORQ TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_8b
MOVQ TMP5, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP4
JG loop_8b
CMPQ LEN, $128
JGE aligned
RET
ret:
RET
TEXT ·hasAVX2(SB), NOSPLIT, $0
XORQ AX, AX
XORQ CX, CX
ADDL $7, AX
CPUID
SHRQ $5, BX
ANDQ $1, BX
MOVB BX, ret+0(FP)
RET

116
vendor/github.com/templexxx/xor/nosimd.go generated vendored Normal file
View File

@@ -0,0 +1,116 @@
// 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
import (
"runtime"
"unsafe"
)
const wordSize = int(unsafe.Sizeof(uintptr(0)))
const supportsUnaligned = runtime.GOARCH == "386" || runtime.GOARCH == "amd64" || runtime.GOARCH == "ppc64" || runtime.GOARCH == "ppc64le" || runtime.GOARCH == "s390x"
// xor the bytes in a and b. The destination is assumed to have enough space.
func bytesNoSIMD(dst, a, b []byte, size int) {
if supportsUnaligned {
fastXORBytes(dst, a, b, size)
} else {
// 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, size)
}
}
// split slice for cache-friendly
const unitSize = 16 * 1024
func matrixNoSIMD(dst []byte, src [][]byte) {
size := len(src[0])
start := 0
do := unitSize
for start < size {
end := start + do
if end <= size {
partNoSIMD(start, end, dst, src)
start = start + do
} else {
partNoSIMD(start, size, dst, src)
start = size
}
}
}
// split vect will improve performance with big data by reducing cache pollution
func partNoSIMD(start, end int, dst []byte, src [][]byte) {
bytesNoSIMD(dst[start:end], src[0][start:end], src[1][start:end], end-start)
for i := 2; i < len(src); i++ {
bytesNoSIMD(dst[start:end], dst[start:end], src[i][start:end], end-start)
}
}
// fastXORBytes xor in bulk. It only works on architectures that
// support unaligned read/writes.
func fastXORBytes(dst, a, b []byte, n int) {
w := n / wordSize
if w > 0 {
wordBytes := w * wordSize
fastXORWords(dst[:wordBytes], a[:wordBytes], b[:wordBytes])
}
for i := n - n%wordSize; i < n; i++ {
dst[i] = a[i] ^ b[i]
}
}
func safeXORBytes(dst, a, b []byte, n int) {
ex := n % 8
for i := 0; i < ex; i++ {
dst[i] = a[i] ^ b[i]
}
for i := ex; i < n; i += 8 {
_dst := dst[i : i+8]
_a := a[i : i+8]
_b := b[i : i+8]
_dst[0] = _a[0] ^ _b[0]
_dst[1] = _a[1] ^ _b[1]
_dst[2] = _a[2] ^ _b[2]
_dst[3] = _a[3] ^ _b[3]
_dst[4] = _a[4] ^ _b[4]
_dst[5] = _a[5] ^ _b[5]
_dst[6] = _a[6] ^ _b[6]
_dst[7] = _a[7] ^ _b[7]
}
}
// 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
ex := n % 8
for i := 0; i < ex; i++ {
dw[i] = aw[i] ^ bw[i]
}
for i := ex; i < n; i += 8 {
_dw := dw[i : i+8]
_aw := aw[i : i+8]
_bw := bw[i : i+8]
_dw[0] = _aw[0] ^ _bw[0]
_dw[1] = _aw[1] ^ _bw[1]
_dw[2] = _aw[2] ^ _bw[2]
_dw[3] = _aw[3] ^ _bw[3]
_dw[4] = _aw[4] ^ _bw[4]
_dw[5] = _aw[5] ^ _bw[5]
_dw[6] = _aw[6] ^ _bw[6]
_dw[7] = _aw[7] ^ _bw[7]
}
}

574
vendor/github.com/templexxx/xor/sse2_amd64.s generated vendored Normal file
View File

@@ -0,0 +1,574 @@
#include "textflag.h"
// addr of mem
#define DST BX
#define SRC SI
#define SRC0 TMP4
#define SRC1 TMP5
// loop args
// num of vect
#define VECT CX
#define LEN DX
// pos of matrix
#define POS R8
// tmp store
// num of vect or ...
#define TMP1 R9
// pos of matrix or ...
#define TMP2 R10
// store addr of data/parity or ...
#define TMP3 R11
#define TMP4 R12
#define TMP5 R13
#define TMP6 R14
// func bytesSrc0(dst, src0, src1 []byte)
TEXT ·xorSrc0(SB), NOSPLIT, $0
MOVQ len+32(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $15, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop16b:
MOVOU (SRC0)(POS*1), X0
XORPD (SRC1)(POS*1), X0
MOVOU X0, (DST)(POS*1)
ADDQ $16, POS
CMPQ LEN, POS
JNE loop16b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $15, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $15, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $16
JGE aligned
RET
ret:
RET
// func bytesSrc1(dst, src0, src1 []byte)
TEXT ·xorSrc1(SB), NOSPLIT, $0
MOVQ len+56(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $15, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop16b:
MOVOU (SRC0)(POS*1), X0
XORPD (SRC1)(POS*1), X0
MOVOU X0, (DST)(POS*1)
ADDQ $16, POS
CMPQ LEN, POS
JNE loop16b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $15, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $15, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $16
JGE aligned
RET
ret:
RET
// func bytesSSE2mini(dst, src0, src1 []byte, size int)
TEXT ·bytesSSE2mini(SB), NOSPLIT, $0
MOVQ len+72(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $15, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop16b:
MOVOU (SRC0)(POS*1), X0
XORPD (SRC1)(POS*1), X0
// MOVOU (SRC1)(POS*1), X4
// PXOR X4, X0
MOVOU X0, (DST)(POS*1)
ADDQ $16, POS
CMPQ LEN, POS
JNE loop16b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $15, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $15, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $16
JGE aligned
RET
ret:
RET
// func bytesSSE2small(dst, src0, src1 []byte, size int)
TEXT ·bytesSSE2small(SB), NOSPLIT, $0
MOVQ len+72(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $63, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop64b:
MOVOU (SRC0)(POS*1), X0
MOVOU 16(SRC0)(POS*1), X1
MOVOU 32(SRC0)(POS*1), X2
MOVOU 48(SRC0)(POS*1), X3
MOVOU (SRC1)(POS*1), X4
MOVOU 16(SRC1)(POS*1), X5
MOVOU 32(SRC1)(POS*1), X6
MOVOU 48(SRC1)(POS*1), X7
PXOR X4, X0
PXOR X5, X1
PXOR X6, X2
PXOR X7, X3
MOVOU X0, (DST)(POS*1)
MOVOU X1, 16(DST)(POS*1)
MOVOU X2, 32(DST)(POS*1)
MOVOU X3, 48(DST)(POS*1)
ADDQ $64, POS
CMPQ LEN, POS
JNE loop64b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $63, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $63, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $64
JGE aligned
RET
ret:
RET
// func bytesSSE2big(dst, src0, src1 []byte, size int)
TEXT ·bytesSSE2big(SB), NOSPLIT, $0
MOVQ len+72(FP), LEN
CMPQ LEN, $0
JE ret
MOVQ dst+0(FP), DST
MOVQ src0+24(FP), SRC0
MOVQ src1+48(FP), SRC1
TESTQ $63, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop64b:
MOVOU (SRC0)(POS*1), X0
MOVOU 16(SRC0)(POS*1), X1
MOVOU 32(SRC0)(POS*1), X2
MOVOU 48(SRC0)(POS*1), X3
MOVOU (SRC1)(POS*1), X4
MOVOU 16(SRC1)(POS*1), X5
MOVOU 32(SRC1)(POS*1), X6
MOVOU 48(SRC1)(POS*1), X7
PXOR X4, X0
PXOR X5, X1
PXOR X6, X2
PXOR X7, X3
LONG $0xe70f4266; WORD $0x0304 // MOVNTDQ
LONG $0xe70f4266; WORD $0x034c; BYTE $0x10
LONG $0xe70f4266; WORD $0x0354; BYTE $0x20
LONG $0xe70f4266; WORD $0x035c; BYTE $0x30
ADDQ $64, POS
CMPQ LEN, POS
JNE loop64b
RET
loop_1b:
MOVB -1(SRC0)(LEN*1), TMP1
MOVB -1(SRC1)(LEN*1), TMP2
XORB TMP1, TMP2
MOVB TMP2, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $63, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP1
ANDQ $63, TMP1
loop_8b:
MOVQ -8(SRC0)(LEN*1), TMP2
MOVQ -8(SRC1)(LEN*1), TMP3
XORQ TMP2, TMP3
MOVQ TMP3, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP1
JG loop_8b
CMPQ LEN, $64
JGE aligned
RET
ret:
RET
// func matrixSSE2small(dst []byte, src [][]byte)
TEXT ·matrixSSE2small(SB), NOSPLIT, $0
MOVQ dst+0(FP), DST
MOVQ src+24(FP), SRC
MOVQ vec+32(FP), VECT
MOVQ len+8(FP), LEN
TESTQ $63, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop64b:
MOVQ VECT, TMP1
SUBQ $2, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
MOVOU (TMP3)(POS*1), X0
MOVOU 16(TMP4)(POS*1), X1
MOVOU 32(TMP3)(POS*1), X2
MOVOU 48(TMP4)(POS*1), X3
next_vect:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
MOVOU (TMP3)(POS*1), X4
MOVOU 16(TMP4)(POS*1), X5
MOVOU 32(TMP3)(POS*1), X6
MOVOU 48(TMP4)(POS*1), X7
PXOR X4, X0
PXOR X5, X1
PXOR X6, X2
PXOR X7, X3
SUBQ $1, TMP1
JGE next_vect
MOVOU X0, (DST)(POS*1)
MOVOU X1, 16(DST)(POS*1)
MOVOU X2, 32(DST)(POS*1)
MOVOU X3, 48(DST)(POS*1)
ADDQ $64, POS
CMPQ LEN, POS
JNE loop64b
RET
loop_1b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVB -1(TMP3)(LEN*1), TMP5
next_vect_1b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVB -1(TMP3)(LEN*1), TMP6
XORB TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_1b
MOVB TMP5, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $63, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP4
ANDQ $63, TMP4
loop_8b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVQ -8(TMP3)(LEN*1), TMP5
next_vect_8b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ -8(TMP3)(LEN*1), TMP6
XORQ TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_8b
MOVQ TMP5, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP4
JG loop_8b
CMPQ LEN, $64
JGE aligned
RET
ret:
RET
// func matrixSSE2big(dst []byte, src [][]byte)
TEXT ·matrixSSE2big(SB), NOSPLIT, $0
MOVQ dst+0(FP), DST
MOVQ src+24(FP), SRC
MOVQ vec+32(FP), VECT
MOVQ len+8(FP), LEN
TESTQ $63, LEN
JNZ not_aligned
aligned:
MOVQ $0, POS
loop64b:
MOVQ VECT, TMP1
SUBQ $2, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
MOVOU (TMP3)(POS*1), X0
MOVOU 16(TMP4)(POS*1), X1
MOVOU 32(TMP3)(POS*1), X2
MOVOU 48(TMP4)(POS*1), X3
next_vect:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ TMP3, TMP4
MOVOU (TMP3)(POS*1), X4
MOVOU 16(TMP4)(POS*1), X5
MOVOU 32(TMP3)(POS*1), X6
MOVOU 48(TMP4)(POS*1), X7
PXOR X4, X0
PXOR X5, X1
PXOR X6, X2
PXOR X7, X3
SUBQ $1, TMP1
JGE next_vect
LONG $0xe70f4266; WORD $0x0304
LONG $0xe70f4266; WORD $0x034c; BYTE $0x10
LONG $0xe70f4266; WORD $0x0354; BYTE $0x20
LONG $0xe70f4266; WORD $0x035c; BYTE $0x30
ADDQ $64, POS
CMPQ LEN, POS
JNE loop64b
RET
loop_1b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVB -1(TMP3)(LEN*1), TMP5
next_vect_1b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVB -1(TMP3)(LEN*1), TMP6
XORB TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_1b
MOVB TMP5, -1(DST)(LEN*1)
SUBQ $1, LEN
TESTQ $7, LEN
JNZ loop_1b
CMPQ LEN, $0
JE ret
TESTQ $63, LEN
JZ aligned
not_aligned:
TESTQ $7, LEN
JNE loop_1b
MOVQ LEN, TMP4
ANDQ $63, TMP4
loop_8b:
MOVQ VECT, TMP1
MOVQ $0, TMP2
MOVQ (SRC)(TMP2*1), TMP3
SUBQ $2, TMP1
MOVQ -8(TMP3)(LEN*1), TMP5
next_vect_8b:
ADDQ $24, TMP2
MOVQ (SRC)(TMP2*1), TMP3
MOVQ -8(TMP3)(LEN*1), TMP6
XORQ TMP6, TMP5
SUBQ $1, TMP1
JGE next_vect_8b
MOVQ TMP5, -8(DST)(LEN*1)
SUBQ $8, LEN
SUBQ $8, TMP4
JG loop_8b
CMPQ LEN, $64
JGE aligned
RET
ret:
RET
TEXT ·hasSSE2(SB), NOSPLIT, $0
XORQ AX, AX
INCL AX
CPUID
SHRQ $26, DX
ANDQ $1, DX
MOVB DX, ret+0(FP)
RET

49
vendor/github.com/templexxx/xor/xor.go generated vendored Normal file
View File

@@ -0,0 +1,49 @@
package xor
// SIMD Extensions
const (
none = iota
avx2
// first introduced by Intel with the initial version of the Pentium 4 in 2001
// so I think we can assume all amd64 has sse2
sse2
)
var extension = none
// Bytes : chose the shortest one as xor size
// it's better to use it for big data ( > 64bytes )
func Bytes(dst, src0, src1 []byte) {
size := len(dst)
if size > len(src0) {
size = len(src0)
}
if size > len(src1) {
size = len(src1)
}
xorBytes(dst, src0, src1, size)
}
// BytesSameLen : all slice's length must be equal
// cut size branch, save time for small data
func BytesSameLen(dst, src0, src1 []byte) {
xorSrc1(dst, src0, src1)
}
// BytesSrc0 : src1 >= src0, dst >= src0
// xor src0's len bytes
func BytesSrc0(dst, src0, src1 []byte) {
xorSrc0(dst, src0, src1)
}
// BytesSrc1 : src0 >= src1, dst >= src1
// xor src1's len bytes
func BytesSrc1(dst, src0, src1 []byte) {
xorSrc1(dst, src0, src1)
}
// Matrix : all slice's length must be equal && != 0
// len(src) must >= 2
func Matrix(dst []byte, src [][]byte) {
xorMatrix(dst, src)
}

118
vendor/github.com/templexxx/xor/xor_amd64.go generated vendored Normal file
View File

@@ -0,0 +1,118 @@
package xor
func init() {
getEXT()
}
func getEXT() {
if hasAVX2() {
extension = avx2
} else {
extension = sse2
}
return
}
func xorBytes(dst, src0, src1 []byte, size int) {
switch extension {
case avx2:
bytesAVX2(dst, src0, src1, size)
default:
bytesSSE2(dst, src0, src1, size)
}
}
// non-temporal hint store
const nontmp = 8 * 1024
const avx2loopsize = 128
func bytesAVX2(dst, src0, src1 []byte, size int) {
if size < avx2loopsize {
bytesAVX2mini(dst, src0, src1, size)
} else if size >= avx2loopsize && size <= nontmp {
bytesAVX2small(dst, src0, src1, size)
} else {
bytesAVX2big(dst, src0, src1, size)
}
}
const sse2loopsize = 64
func bytesSSE2(dst, src0, src1 []byte, size int) {
if size < sse2loopsize {
bytesSSE2mini(dst, src0, src1, size)
} else if size >= sse2loopsize && size <= nontmp {
bytesSSE2small(dst, src0, src1, size)
} else {
bytesSSE2big(dst, src0, src1, size)
}
}
func xorMatrix(dst []byte, src [][]byte) {
switch extension {
case avx2:
matrixAVX2(dst, src)
default:
matrixSSE2(dst, src)
}
}
func matrixAVX2(dst []byte, src [][]byte) {
size := len(dst)
if size > nontmp {
matrixAVX2big(dst, src)
} else {
matrixAVX2small(dst, src)
}
}
func matrixSSE2(dst []byte, src [][]byte) {
size := len(dst)
if size > nontmp {
matrixSSE2big(dst, src)
} else {
matrixSSE2small(dst, src)
}
}
//go:noescape
func xorSrc0(dst, src0, src1 []byte)
//go:noescape
func xorSrc1(dst, src0, src1 []byte)
//go:noescape
func bytesAVX2mini(dst, src0, src1 []byte, size int)
//go:noescape
func bytesAVX2big(dst, src0, src1 []byte, size int)
//go:noescape
func bytesAVX2small(dst, src0, src1 []byte, size int)
//go:noescape
func bytesSSE2mini(dst, src0, src1 []byte, size int)
//go:noescape
func bytesSSE2small(dst, src0, src1 []byte, size int)
//go:noescape
func bytesSSE2big(dst, src0, src1 []byte, size int)
//go:noescape
func matrixAVX2small(dst []byte, src [][]byte)
//go:noescape
func matrixAVX2big(dst []byte, src [][]byte)
//go:noescape
func matrixSSE2small(dst []byte, src [][]byte)
//go:noescape
func matrixSSE2big(dst []byte, src [][]byte)
//go:noescape
func hasAVX2() bool
//go:noescape
func hasSSE2() bool

19
vendor/github.com/templexxx/xor/xor_other.go generated vendored Normal file
View File

@@ -0,0 +1,19 @@
// +build !amd64 noasm
package xor
func xorBytes(dst, src0, src1 []byte, size int) {
bytesNoSIMD(dst, src0, src1, size)
}
func xorMatrix(dst []byte, src [][]byte) {
matrixNoSIMD(dst, src)
}
func xorSrc0(dst, src0, src1 []byte) {
bytesNoSIMD(dst, src0, src1, len(src0))
}
func xorSrc1(dst, src0, src1 []byte) {
bytesNoSIMD(dst, src0, src1, len(src1))
}