vendor: github.com/cespare/xxhash/v2 v2.2.0

full diff: https://github.com/cespare/xxhash/compare/v2.1.2...v2.2.0

Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
This commit is contained in:
Sebastiaan van Stijn 2023-07-20 00:21:37 +02:00
parent c1d0657029
commit 5b138189b9
No known key found for this signature in database
GPG Key ID: 76698F39D527CE8C
12 changed files with 401 additions and 213 deletions

View File

@ -50,7 +50,7 @@ require (
github.com/Azure/go-ansiterm v0.0.0-20210617225240-d185dfc1b5a1 // indirect github.com/Azure/go-ansiterm v0.0.0-20210617225240-d185dfc1b5a1 // indirect
github.com/Microsoft/go-winio v0.5.2 // indirect github.com/Microsoft/go-winio v0.5.2 // indirect
github.com/beorn7/perks v1.0.1 // indirect github.com/beorn7/perks v1.0.1 // indirect
github.com/cespare/xxhash/v2 v2.1.2 // indirect github.com/cespare/xxhash/v2 v2.2.0 // indirect
github.com/docker/go v1.5.1-1.0.20160303222718-d30aec9fd63c // indirect github.com/docker/go v1.5.1-1.0.20160303222718-d30aec9fd63c // indirect
github.com/docker/go-events v0.0.0-20190806004212-e31b211e4f1c // indirect github.com/docker/go-events v0.0.0-20190806004212-e31b211e4f1c // indirect
github.com/docker/go-metrics v0.0.1 // indirect github.com/docker/go-metrics v0.0.1 // indirect

View File

@ -67,8 +67,9 @@ github.com/census-instrumentation/opencensus-proto v0.2.1/go.mod h1:f6KPmirojxKA
github.com/certifi/gocertifi v0.0.0-20191021191039-0944d244cd40/go.mod h1:sGbDF6GwGcLpkNXPUTkMRoywsNa/ol15pxFe6ERfguA= github.com/certifi/gocertifi v0.0.0-20191021191039-0944d244cd40/go.mod h1:sGbDF6GwGcLpkNXPUTkMRoywsNa/ol15pxFe6ERfguA=
github.com/certifi/gocertifi v0.0.0-20200922220541-2c3bb06c6054/go.mod h1:sGbDF6GwGcLpkNXPUTkMRoywsNa/ol15pxFe6ERfguA= github.com/certifi/gocertifi v0.0.0-20200922220541-2c3bb06c6054/go.mod h1:sGbDF6GwGcLpkNXPUTkMRoywsNa/ol15pxFe6ERfguA=
github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs= github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
github.com/cespare/xxhash/v2 v2.1.2 h1:YRXhKfTDauu4ajMg1TPgFO5jnlC2HCbmLXMcTG5cbYE=
github.com/cespare/xxhash/v2 v2.1.2/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs= github.com/cespare/xxhash/v2 v2.1.2/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
github.com/cespare/xxhash/v2 v2.2.0 h1:DC2CZ1Ep5Y4k3ZQ899DldepgrayRUGE6BBZ/cd9Cj44=
github.com/cespare/xxhash/v2 v2.2.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
github.com/chzyer/logex v1.1.10/go.mod h1:+Ywpsq7O8HXn0nuIou7OrIPyXbp3wmkHB+jjWRnGsAI= github.com/chzyer/logex v1.1.10/go.mod h1:+Ywpsq7O8HXn0nuIou7OrIPyXbp3wmkHB+jjWRnGsAI=
github.com/chzyer/readline v0.0.0-20180603132655-2972be24d48e/go.mod h1:nSuG5e5PlCu98SY8svDHJxuZscDgtXS6KTTbou5AhLI= github.com/chzyer/readline v0.0.0-20180603132655-2972be24d48e/go.mod h1:nSuG5e5PlCu98SY8svDHJxuZscDgtXS6KTTbou5AhLI=
github.com/chzyer/test v0.0.0-20180213035817-a1ea475d72b1/go.mod h1:Q3SI9o4m/ZMnBNeIyt5eFwwo7qiLfzFZmjNmxjkiQlU= github.com/chzyer/test v0.0.0-20180213035817-a1ea475d72b1/go.mod h1:Q3SI9o4m/ZMnBNeIyt5eFwwo7qiLfzFZmjNmxjkiQlU=

View File

@ -3,8 +3,7 @@
[![Go Reference](https://pkg.go.dev/badge/github.com/cespare/xxhash/v2.svg)](https://pkg.go.dev/github.com/cespare/xxhash/v2) [![Go Reference](https://pkg.go.dev/badge/github.com/cespare/xxhash/v2.svg)](https://pkg.go.dev/github.com/cespare/xxhash/v2)
[![Test](https://github.com/cespare/xxhash/actions/workflows/test.yml/badge.svg)](https://github.com/cespare/xxhash/actions/workflows/test.yml) [![Test](https://github.com/cespare/xxhash/actions/workflows/test.yml/badge.svg)](https://github.com/cespare/xxhash/actions/workflows/test.yml)
xxhash is a Go implementation of the 64-bit xxhash is a Go implementation of the 64-bit [xxHash] algorithm, XXH64. This is a
[xxHash](http://cyan4973.github.io/xxHash/) algorithm, XXH64. This is a
high-quality hashing algorithm that is much faster than anything in the Go high-quality hashing algorithm that is much faster than anything in the Go
standard library. standard library.
@ -25,8 +24,11 @@ func (*Digest) WriteString(string) (int, error)
func (*Digest) Sum64() uint64 func (*Digest) Sum64() uint64
``` ```
This implementation provides a fast pure-Go implementation and an even faster The package is written with optimized pure Go and also contains even faster
assembly implementation for amd64. assembly implementations for amd64 and arm64. If desired, the `purego` build tag
opts into using the Go code even on those architectures.
[xxHash]: http://cyan4973.github.io/xxHash/
## Compatibility ## Compatibility
@ -45,19 +47,20 @@ I recommend using the latest release of Go.
Here are some quick benchmarks comparing the pure-Go and assembly Here are some quick benchmarks comparing the pure-Go and assembly
implementations of Sum64. implementations of Sum64.
| input size | purego | asm | | input size | purego | asm |
| --- | --- | --- | | ---------- | --------- | --------- |
| 5 B | 979.66 MB/s | 1291.17 MB/s | | 4 B | 1.3 GB/s | 1.2 GB/s |
| 100 B | 7475.26 MB/s | 7973.40 MB/s | | 16 B | 2.9 GB/s | 3.5 GB/s |
| 4 KB | 17573.46 MB/s | 17602.65 MB/s | | 100 B | 6.9 GB/s | 8.1 GB/s |
| 10 MB | 17131.46 MB/s | 17142.16 MB/s | | 4 KB | 11.7 GB/s | 16.7 GB/s |
| 10 MB | 12.0 GB/s | 17.3 GB/s |
These numbers were generated on Ubuntu 18.04 with an Intel i7-8700K CPU using These numbers were generated on Ubuntu 20.04 with an Intel Xeon Platinum 8252C
the following commands under Go 1.11.2: CPU using the following commands under Go 1.19.2:
``` ```
$ go test -tags purego -benchtime 10s -bench '/xxhash,direct,bytes' benchstat <(go test -tags purego -benchtime 500ms -count 15 -bench 'Sum64$')
$ go test -benchtime 10s -bench '/xxhash,direct,bytes' benchstat <(go test -benchtime 500ms -count 15 -bench 'Sum64$')
``` ```
## Projects using this package ## Projects using this package

10
vendor/github.com/cespare/xxhash/v2/testall.sh generated vendored Normal file
View File

@ -0,0 +1,10 @@
#!/bin/bash
set -eu -o pipefail
# Small convenience script for running the tests with various combinations of
# arch/tags. This assumes we're running on amd64 and have qemu available.
go test ./...
go test -tags purego ./...
GOARCH=arm64 go test
GOARCH=arm64 go test -tags purego

View File

@ -16,19 +16,11 @@ const (
prime5 uint64 = 2870177450012600261 prime5 uint64 = 2870177450012600261
) )
// NOTE(caleb): I'm using both consts and vars of the primes. Using consts where // Store the primes in an array as well.
// possible in the Go code is worth a small (but measurable) performance boost //
// by avoiding some MOVQs. Vars are needed for the asm and also are useful for // The consts are used when possible in Go code to avoid MOVs but we need a
// convenience in the Go code in a few places where we need to intentionally // contiguous array of the assembly code.
// avoid constant arithmetic (e.g., v1 := prime1 + prime2 fails because the var primes = [...]uint64{prime1, prime2, prime3, prime4, prime5}
// result overflows a uint64).
var (
prime1v = prime1
prime2v = prime2
prime3v = prime3
prime4v = prime4
prime5v = prime5
)
// Digest implements hash.Hash64. // Digest implements hash.Hash64.
type Digest struct { type Digest struct {
@ -50,10 +42,10 @@ func New() *Digest {
// Reset clears the Digest's state so that it can be reused. // Reset clears the Digest's state so that it can be reused.
func (d *Digest) Reset() { func (d *Digest) Reset() {
d.v1 = prime1v + prime2 d.v1 = primes[0] + prime2
d.v2 = prime2 d.v2 = prime2
d.v3 = 0 d.v3 = 0
d.v4 = -prime1v d.v4 = -primes[0]
d.total = 0 d.total = 0
d.n = 0 d.n = 0
} }
@ -69,21 +61,23 @@ func (d *Digest) Write(b []byte) (n int, err error) {
n = len(b) n = len(b)
d.total += uint64(n) d.total += uint64(n)
memleft := d.mem[d.n&(len(d.mem)-1):]
if d.n+n < 32 { if d.n+n < 32 {
// This new data doesn't even fill the current block. // This new data doesn't even fill the current block.
copy(d.mem[d.n:], b) copy(memleft, b)
d.n += n d.n += n
return return
} }
if d.n > 0 { if d.n > 0 {
// Finish off the partial block. // Finish off the partial block.
copy(d.mem[d.n:], b) c := copy(memleft, b)
d.v1 = round(d.v1, u64(d.mem[0:8])) d.v1 = round(d.v1, u64(d.mem[0:8]))
d.v2 = round(d.v2, u64(d.mem[8:16])) d.v2 = round(d.v2, u64(d.mem[8:16]))
d.v3 = round(d.v3, u64(d.mem[16:24])) d.v3 = round(d.v3, u64(d.mem[16:24]))
d.v4 = round(d.v4, u64(d.mem[24:32])) d.v4 = round(d.v4, u64(d.mem[24:32]))
b = b[32-d.n:] b = b[c:]
d.n = 0 d.n = 0
} }
@ -133,21 +127,20 @@ func (d *Digest) Sum64() uint64 {
h += d.total h += d.total
i, end := 0, d.n b := d.mem[:d.n&(len(d.mem)-1)]
for ; i+8 <= end; i += 8 { for ; len(b) >= 8; b = b[8:] {
k1 := round(0, u64(d.mem[i:i+8])) k1 := round(0, u64(b[:8]))
h ^= k1 h ^= k1
h = rol27(h)*prime1 + prime4 h = rol27(h)*prime1 + prime4
} }
if i+4 <= end { if len(b) >= 4 {
h ^= uint64(u32(d.mem[i:i+4])) * prime1 h ^= uint64(u32(b[:4])) * prime1
h = rol23(h)*prime2 + prime3 h = rol23(h)*prime2 + prime3
i += 4 b = b[4:]
} }
for i < end { for ; len(b) > 0; b = b[1:] {
h ^= uint64(d.mem[i]) * prime5 h ^= uint64(b[0]) * prime5
h = rol11(h) * prime1 h = rol11(h) * prime1
i++
} }
h ^= h >> 33 h ^= h >> 33

View File

@ -1,215 +1,209 @@
//go:build !appengine && gc && !purego
// +build !appengine // +build !appengine
// +build gc // +build gc
// +build !purego // +build !purego
#include "textflag.h" #include "textflag.h"
// Register allocation: // Registers:
// AX h #define h AX
// SI pointer to advance through b #define d AX
// DX n #define p SI // pointer to advance through b
// BX loop end #define n DX
// R8 v1, k1 #define end BX // loop end
// R9 v2 #define v1 R8
// R10 v3 #define v2 R9
// R11 v4 #define v3 R10
// R12 tmp #define v4 R11
// R13 prime1v #define x R12
// R14 prime2v #define prime1 R13
// DI prime4v #define prime2 R14
#define prime4 DI
// round reads from and advances the buffer pointer in SI. #define round(acc, x) \
// It assumes that R13 has prime1v and R14 has prime2v. IMULQ prime2, x \
#define round(r) \ ADDQ x, acc \
MOVQ (SI), R12 \ ROLQ $31, acc \
ADDQ $8, SI \ IMULQ prime1, acc
IMULQ R14, R12 \
ADDQ R12, r \
ROLQ $31, r \
IMULQ R13, r
// mergeRound applies a merge round on the two registers acc and val. // round0 performs the operation x = round(0, x).
// It assumes that R13 has prime1v, R14 has prime2v, and DI has prime4v. #define round0(x) \
#define mergeRound(acc, val) \ IMULQ prime2, x \
IMULQ R14, val \ ROLQ $31, x \
ROLQ $31, val \ IMULQ prime1, x
IMULQ R13, val \
XORQ val, acc \ // mergeRound applies a merge round on the two registers acc and x.
IMULQ R13, acc \ // It assumes that prime1, prime2, and prime4 have been loaded.
ADDQ DI, acc #define mergeRound(acc, x) \
round0(x) \
XORQ x, acc \
IMULQ prime1, acc \
ADDQ prime4, acc
// blockLoop processes as many 32-byte blocks as possible,
// updating v1, v2, v3, and v4. It assumes that there is at least one block
// to process.
#define blockLoop() \
loop: \
MOVQ +0(p), x \
round(v1, x) \
MOVQ +8(p), x \
round(v2, x) \
MOVQ +16(p), x \
round(v3, x) \
MOVQ +24(p), x \
round(v4, x) \
ADDQ $32, p \
CMPQ p, end \
JLE loop
// func Sum64(b []byte) uint64 // func Sum64(b []byte) uint64
TEXT ·Sum64(SB), NOSPLIT, $0-32 TEXT ·Sum64(SB), NOSPLIT|NOFRAME, $0-32
// Load fixed primes. // Load fixed primes.
MOVQ ·prime1v(SB), R13 MOVQ ·primes+0(SB), prime1
MOVQ ·prime2v(SB), R14 MOVQ ·primes+8(SB), prime2
MOVQ ·prime4v(SB), DI MOVQ ·primes+24(SB), prime4
// Load slice. // Load slice.
MOVQ b_base+0(FP), SI MOVQ b_base+0(FP), p
MOVQ b_len+8(FP), DX MOVQ b_len+8(FP), n
LEAQ (SI)(DX*1), BX LEAQ (p)(n*1), end
// The first loop limit will be len(b)-32. // The first loop limit will be len(b)-32.
SUBQ $32, BX SUBQ $32, end
// Check whether we have at least one block. // Check whether we have at least one block.
CMPQ DX, $32 CMPQ n, $32
JLT noBlocks JLT noBlocks
// Set up initial state (v1, v2, v3, v4). // Set up initial state (v1, v2, v3, v4).
MOVQ R13, R8 MOVQ prime1, v1
ADDQ R14, R8 ADDQ prime2, v1
MOVQ R14, R9 MOVQ prime2, v2
XORQ R10, R10 XORQ v3, v3
XORQ R11, R11 XORQ v4, v4
SUBQ R13, R11 SUBQ prime1, v4
// Loop until SI > BX. blockLoop()
blockLoop:
round(R8)
round(R9)
round(R10)
round(R11)
CMPQ SI, BX MOVQ v1, h
JLE blockLoop ROLQ $1, h
MOVQ v2, x
ROLQ $7, x
ADDQ x, h
MOVQ v3, x
ROLQ $12, x
ADDQ x, h
MOVQ v4, x
ROLQ $18, x
ADDQ x, h
MOVQ R8, AX mergeRound(h, v1)
ROLQ $1, AX mergeRound(h, v2)
MOVQ R9, R12 mergeRound(h, v3)
ROLQ $7, R12 mergeRound(h, v4)
ADDQ R12, AX
MOVQ R10, R12
ROLQ $12, R12
ADDQ R12, AX
MOVQ R11, R12
ROLQ $18, R12
ADDQ R12, AX
mergeRound(AX, R8)
mergeRound(AX, R9)
mergeRound(AX, R10)
mergeRound(AX, R11)
JMP afterBlocks JMP afterBlocks
noBlocks: noBlocks:
MOVQ ·prime5v(SB), AX MOVQ ·primes+32(SB), h
afterBlocks: afterBlocks:
ADDQ DX, AX ADDQ n, h
// Right now BX has len(b)-32, and we want to loop until SI > len(b)-8. ADDQ $24, end
ADDQ $24, BX CMPQ p, end
JG try4
CMPQ SI, BX loop8:
JG fourByte MOVQ (p), x
ADDQ $8, p
round0(x)
XORQ x, h
ROLQ $27, h
IMULQ prime1, h
ADDQ prime4, h
wordLoop: CMPQ p, end
// Calculate k1. JLE loop8
MOVQ (SI), R8
ADDQ $8, SI
IMULQ R14, R8
ROLQ $31, R8
IMULQ R13, R8
XORQ R8, AX try4:
ROLQ $27, AX ADDQ $4, end
IMULQ R13, AX CMPQ p, end
ADDQ DI, AX JG try1
CMPQ SI, BX MOVL (p), x
JLE wordLoop ADDQ $4, p
IMULQ prime1, x
XORQ x, h
fourByte: ROLQ $23, h
ADDQ $4, BX IMULQ prime2, h
CMPQ SI, BX ADDQ ·primes+16(SB), h
JG singles
MOVL (SI), R8 try1:
ADDQ $4, SI ADDQ $4, end
IMULQ R13, R8 CMPQ p, end
XORQ R8, AX
ROLQ $23, AX
IMULQ R14, AX
ADDQ ·prime3v(SB), AX
singles:
ADDQ $4, BX
CMPQ SI, BX
JGE finalize JGE finalize
singlesLoop: loop1:
MOVBQZX (SI), R12 MOVBQZX (p), x
ADDQ $1, SI ADDQ $1, p
IMULQ ·prime5v(SB), R12 IMULQ ·primes+32(SB), x
XORQ R12, AX XORQ x, h
ROLQ $11, h
IMULQ prime1, h
ROLQ $11, AX CMPQ p, end
IMULQ R13, AX JL loop1
CMPQ SI, BX
JL singlesLoop
finalize: finalize:
MOVQ AX, R12 MOVQ h, x
SHRQ $33, R12 SHRQ $33, x
XORQ R12, AX XORQ x, h
IMULQ R14, AX IMULQ prime2, h
MOVQ AX, R12 MOVQ h, x
SHRQ $29, R12 SHRQ $29, x
XORQ R12, AX XORQ x, h
IMULQ ·prime3v(SB), AX IMULQ ·primes+16(SB), h
MOVQ AX, R12 MOVQ h, x
SHRQ $32, R12 SHRQ $32, x
XORQ R12, AX XORQ x, h
MOVQ AX, ret+24(FP) MOVQ h, ret+24(FP)
RET RET
// writeBlocks uses the same registers as above except that it uses AX to store
// the d pointer.
// func writeBlocks(d *Digest, b []byte) int // func writeBlocks(d *Digest, b []byte) int
TEXT ·writeBlocks(SB), NOSPLIT, $0-40 TEXT ·writeBlocks(SB), NOSPLIT|NOFRAME, $0-40
// Load fixed primes needed for round. // Load fixed primes needed for round.
MOVQ ·prime1v(SB), R13 MOVQ ·primes+0(SB), prime1
MOVQ ·prime2v(SB), R14 MOVQ ·primes+8(SB), prime2
// Load slice. // Load slice.
MOVQ b_base+8(FP), SI MOVQ b_base+8(FP), p
MOVQ b_len+16(FP), DX MOVQ b_len+16(FP), n
LEAQ (SI)(DX*1), BX LEAQ (p)(n*1), end
SUBQ $32, BX SUBQ $32, end
// Load vN from d. // Load vN from d.
MOVQ d+0(FP), AX MOVQ s+0(FP), d
MOVQ 0(AX), R8 // v1 MOVQ 0(d), v1
MOVQ 8(AX), R9 // v2 MOVQ 8(d), v2
MOVQ 16(AX), R10 // v3 MOVQ 16(d), v3
MOVQ 24(AX), R11 // v4 MOVQ 24(d), v4
// We don't need to check the loop condition here; this function is // We don't need to check the loop condition here; this function is
// always called with at least one block of data to process. // always called with at least one block of data to process.
blockLoop: blockLoop()
round(R8)
round(R9)
round(R10)
round(R11)
CMPQ SI, BX
JLE blockLoop
// Copy vN back to d. // Copy vN back to d.
MOVQ R8, 0(AX) MOVQ v1, 0(d)
MOVQ R9, 8(AX) MOVQ v2, 8(d)
MOVQ R10, 16(AX) MOVQ v3, 16(d)
MOVQ R11, 24(AX) MOVQ v4, 24(d)
// The number of bytes written is SI minus the old base pointer. // The number of bytes written is p minus the old base pointer.
SUBQ b_base+8(FP), SI SUBQ b_base+8(FP), p
MOVQ SI, ret+32(FP) MOVQ p, ret+32(FP)
RET RET

183
vendor/github.com/cespare/xxhash/v2/xxhash_arm64.s generated vendored Normal file
View File

@ -0,0 +1,183 @@
//go:build !appengine && gc && !purego
// +build !appengine
// +build gc
// +build !purego
#include "textflag.h"
// Registers:
#define digest R1
#define h R2 // return value
#define p R3 // input pointer
#define n R4 // input length
#define nblocks R5 // n / 32
#define prime1 R7
#define prime2 R8
#define prime3 R9
#define prime4 R10
#define prime5 R11
#define v1 R12
#define v2 R13
#define v3 R14
#define v4 R15
#define x1 R20
#define x2 R21
#define x3 R22
#define x4 R23
#define round(acc, x) \
MADD prime2, acc, x, acc \
ROR $64-31, acc \
MUL prime1, acc
// round0 performs the operation x = round(0, x).
#define round0(x) \
MUL prime2, x \
ROR $64-31, x \
MUL prime1, x
#define mergeRound(acc, x) \
round0(x) \
EOR x, acc \
MADD acc, prime4, prime1, acc
// blockLoop processes as many 32-byte blocks as possible,
// updating v1, v2, v3, and v4. It assumes that n >= 32.
#define blockLoop() \
LSR $5, n, nblocks \
PCALIGN $16 \
loop: \
LDP.P 16(p), (x1, x2) \
LDP.P 16(p), (x3, x4) \
round(v1, x1) \
round(v2, x2) \
round(v3, x3) \
round(v4, x4) \
SUB $1, nblocks \
CBNZ nblocks, loop
// func Sum64(b []byte) uint64
TEXT ·Sum64(SB), NOSPLIT|NOFRAME, $0-32
LDP b_base+0(FP), (p, n)
LDP ·primes+0(SB), (prime1, prime2)
LDP ·primes+16(SB), (prime3, prime4)
MOVD ·primes+32(SB), prime5
CMP $32, n
CSEL LT, prime5, ZR, h // if n < 32 { h = prime5 } else { h = 0 }
BLT afterLoop
ADD prime1, prime2, v1
MOVD prime2, v2
MOVD $0, v3
NEG prime1, v4
blockLoop()
ROR $64-1, v1, x1
ROR $64-7, v2, x2
ADD x1, x2
ROR $64-12, v3, x3
ROR $64-18, v4, x4
ADD x3, x4
ADD x2, x4, h
mergeRound(h, v1)
mergeRound(h, v2)
mergeRound(h, v3)
mergeRound(h, v4)
afterLoop:
ADD n, h
TBZ $4, n, try8
LDP.P 16(p), (x1, x2)
round0(x1)
// NOTE: here and below, sequencing the EOR after the ROR (using a
// rotated register) is worth a small but measurable speedup for small
// inputs.
ROR $64-27, h
EOR x1 @> 64-27, h, h
MADD h, prime4, prime1, h
round0(x2)
ROR $64-27, h
EOR x2 @> 64-27, h, h
MADD h, prime4, prime1, h
try8:
TBZ $3, n, try4
MOVD.P 8(p), x1
round0(x1)
ROR $64-27, h
EOR x1 @> 64-27, h, h
MADD h, prime4, prime1, h
try4:
TBZ $2, n, try2
MOVWU.P 4(p), x2
MUL prime1, x2
ROR $64-23, h
EOR x2 @> 64-23, h, h
MADD h, prime3, prime2, h
try2:
TBZ $1, n, try1
MOVHU.P 2(p), x3
AND $255, x3, x1
LSR $8, x3, x2
MUL prime5, x1
ROR $64-11, h
EOR x1 @> 64-11, h, h
MUL prime1, h
MUL prime5, x2
ROR $64-11, h
EOR x2 @> 64-11, h, h
MUL prime1, h
try1:
TBZ $0, n, finalize
MOVBU (p), x4
MUL prime5, x4
ROR $64-11, h
EOR x4 @> 64-11, h, h
MUL prime1, h
finalize:
EOR h >> 33, h
MUL prime2, h
EOR h >> 29, h
MUL prime3, h
EOR h >> 32, h
MOVD h, ret+24(FP)
RET
// func writeBlocks(d *Digest, b []byte) int
TEXT ·writeBlocks(SB), NOSPLIT|NOFRAME, $0-40
LDP ·primes+0(SB), (prime1, prime2)
// Load state. Assume v[1-4] are stored contiguously.
MOVD d+0(FP), digest
LDP 0(digest), (v1, v2)
LDP 16(digest), (v3, v4)
LDP b_base+8(FP), (p, n)
blockLoop()
// Store updated state.
STP (v1, v2), 0(digest)
STP (v3, v4), 16(digest)
BIC $31, n
MOVD n, ret+32(FP)
RET

View File

@ -1,3 +1,5 @@
//go:build (amd64 || arm64) && !appengine && gc && !purego
// +build amd64 arm64
// +build !appengine // +build !appengine
// +build gc // +build gc
// +build !purego // +build !purego

View File

@ -1,4 +1,5 @@
// +build !amd64 appengine !gc purego //go:build (!amd64 && !arm64) || appengine || !gc || purego
// +build !amd64,!arm64 appengine !gc purego
package xxhash package xxhash
@ -14,10 +15,10 @@ func Sum64(b []byte) uint64 {
var h uint64 var h uint64
if n >= 32 { if n >= 32 {
v1 := prime1v + prime2 v1 := primes[0] + prime2
v2 := prime2 v2 := prime2
v3 := uint64(0) v3 := uint64(0)
v4 := -prime1v v4 := -primes[0]
for len(b) >= 32 { for len(b) >= 32 {
v1 = round(v1, u64(b[0:8:len(b)])) v1 = round(v1, u64(b[0:8:len(b)]))
v2 = round(v2, u64(b[8:16:len(b)])) v2 = round(v2, u64(b[8:16:len(b)]))
@ -36,19 +37,18 @@ func Sum64(b []byte) uint64 {
h += uint64(n) h += uint64(n)
i, end := 0, len(b) for ; len(b) >= 8; b = b[8:] {
for ; i+8 <= end; i += 8 { k1 := round(0, u64(b[:8]))
k1 := round(0, u64(b[i:i+8:len(b)]))
h ^= k1 h ^= k1
h = rol27(h)*prime1 + prime4 h = rol27(h)*prime1 + prime4
} }
if i+4 <= end { if len(b) >= 4 {
h ^= uint64(u32(b[i:i+4:len(b)])) * prime1 h ^= uint64(u32(b[:4])) * prime1
h = rol23(h)*prime2 + prime3 h = rol23(h)*prime2 + prime3
i += 4 b = b[4:]
} }
for ; i < end; i++ { for ; len(b) > 0; b = b[1:] {
h ^= uint64(b[i]) * prime5 h ^= uint64(b[0]) * prime5
h = rol11(h) * prime1 h = rol11(h) * prime1
} }

View File

@ -1,3 +1,4 @@
//go:build appengine
// +build appengine // +build appengine
// This file contains the safe implementations of otherwise unsafe-using code. // This file contains the safe implementations of otherwise unsafe-using code.

View File

@ -1,3 +1,4 @@
//go:build !appengine
// +build !appengine // +build !appengine
// This file encapsulates usage of unsafe. // This file encapsulates usage of unsafe.
@ -11,7 +12,7 @@ import (
// In the future it's possible that compiler optimizations will make these // In the future it's possible that compiler optimizations will make these
// XxxString functions unnecessary by realizing that calls such as // XxxString functions unnecessary by realizing that calls such as
// Sum64([]byte(s)) don't need to copy s. See https://golang.org/issue/2205. // Sum64([]byte(s)) don't need to copy s. See https://go.dev/issue/2205.
// If that happens, even if we keep these functions they can be replaced with // If that happens, even if we keep these functions they can be replaced with
// the trivial safe code. // the trivial safe code.

2
vendor/modules.txt vendored
View File

@ -12,7 +12,7 @@ github.com/Microsoft/go-winio/pkg/guid
# github.com/beorn7/perks v1.0.1 # github.com/beorn7/perks v1.0.1
## explicit; go 1.11 ## explicit; go 1.11
github.com/beorn7/perks/quantile github.com/beorn7/perks/quantile
# github.com/cespare/xxhash/v2 v2.1.2 # github.com/cespare/xxhash/v2 v2.2.0
## explicit; go 1.11 ## explicit; go 1.11
github.com/cespare/xxhash/v2 github.com/cespare/xxhash/v2
# github.com/container-orchestrated-devices/container-device-interface v0.6.0 # github.com/container-orchestrated-devices/container-device-interface v0.6.0