gotosocial/vendor/github.com/zeebo/xxh3/hash64.go
kim 07207e71e9
[performance] cache library performance enhancements (updates go-structr => v0.2.0) (#2575)
* update go-structr => v0.2.0

* update readme

* whoops, fix the link
2024-01-26 12:14:10 +00:00

126 lines
3.7 KiB
Go

package xxh3
import "math/bits"
// Hash returns the hash of the byte slice.
func Hash(b []byte) uint64 {
return hashAny(*(*str)(ptr(&b)))
}
// Hash returns the hash of the string slice.
func HashString(s string) uint64 {
return hashAny(*(*str)(ptr(&s)))
}
func hashAny(s str) (acc u64) {
p, l := s.p, s.l
switch {
case l <= 16:
switch {
case l > 8: // 9-16
inputlo := readU64(p, 0) ^ (key64_024 ^ key64_032)
inputhi := readU64(p, ui(l)-8) ^ (key64_040 ^ key64_048)
folded := mulFold64(inputlo, inputhi)
return xxh3Avalanche(u64(l) + bits.ReverseBytes64(inputlo) + inputhi + folded)
case l > 3: // 4-8
input1 := readU32(p, 0)
input2 := readU32(p, ui(l)-4)
input64 := u64(input2) + u64(input1)<<32
keyed := input64 ^ (key64_008 ^ key64_016)
return rrmxmx(keyed, u64(l))
case l == 3: // 3
c12 := u64(readU16(p, 0))
c3 := u64(readU8(p, 2))
acc = c12<<16 + c3 + 3<<8
case l > 1: // 2
c12 := u64(readU16(p, 0))
acc = c12*(1<<24+1)>>8 + 2<<8
case l == 1: // 1
c1 := u64(readU8(p, 0))
acc = c1*(1<<24+1<<16+1) + 1<<8
default: // 0
return 0x2d06800538d394c2 // xxh_avalanche(key64_056 ^ key64_064)
}
acc ^= u64(key32_000 ^ key32_004)
return xxhAvalancheSmall(acc)
case l <= 128:
acc = u64(l) * prime64_1
if l > 32 {
if l > 64 {
if l > 96 {
acc += mulFold64(readU64(p, 6*8)^key64_096, readU64(p, 7*8)^key64_104)
acc += mulFold64(readU64(p, ui(l)-8*8)^key64_112, readU64(p, ui(l)-7*8)^key64_120)
} // 96
acc += mulFold64(readU64(p, 4*8)^key64_064, readU64(p, 5*8)^key64_072)
acc += mulFold64(readU64(p, ui(l)-6*8)^key64_080, readU64(p, ui(l)-5*8)^key64_088)
} // 64
acc += mulFold64(readU64(p, 2*8)^key64_032, readU64(p, 3*8)^key64_040)
acc += mulFold64(readU64(p, ui(l)-4*8)^key64_048, readU64(p, ui(l)-3*8)^key64_056)
} // 32
acc += mulFold64(readU64(p, 0*8)^key64_000, readU64(p, 1*8)^key64_008)
acc += mulFold64(readU64(p, ui(l)-2*8)^key64_016, readU64(p, ui(l)-1*8)^key64_024)
return xxh3Avalanche(acc)
case l <= 240:
acc = u64(l) * prime64_1
acc += mulFold64(readU64(p, 0*16+0)^key64_000, readU64(p, 0*16+8)^key64_008)
acc += mulFold64(readU64(p, 1*16+0)^key64_016, readU64(p, 1*16+8)^key64_024)
acc += mulFold64(readU64(p, 2*16+0)^key64_032, readU64(p, 2*16+8)^key64_040)
acc += mulFold64(readU64(p, 3*16+0)^key64_048, readU64(p, 3*16+8)^key64_056)
acc += mulFold64(readU64(p, 4*16+0)^key64_064, readU64(p, 4*16+8)^key64_072)
acc += mulFold64(readU64(p, 5*16+0)^key64_080, readU64(p, 5*16+8)^key64_088)
acc += mulFold64(readU64(p, 6*16+0)^key64_096, readU64(p, 6*16+8)^key64_104)
acc += mulFold64(readU64(p, 7*16+0)^key64_112, readU64(p, 7*16+8)^key64_120)
// avalanche
acc = xxh3Avalanche(acc)
// trailing groups after 128
top := ui(l) &^ 15
for i := ui(8 * 16); i < top; i += 16 {
acc += mulFold64(readU64(p, i+0)^readU64(key, i-125), readU64(p, i+8)^readU64(key, i-117))
}
// last 16 bytes
acc += mulFold64(readU64(p, ui(l)-16)^key64_119, readU64(p, ui(l)-8)^key64_127)
return xxh3Avalanche(acc)
default:
acc = u64(l) * prime64_1
accs := [8]u64{
prime32_3, prime64_1, prime64_2, prime64_3,
prime64_4, prime32_2, prime64_5, prime32_1,
}
if hasAVX512 && l >= avx512Switch {
accumAVX512(&accs, p, key, u64(l))
} else if hasAVX2 {
accumAVX2(&accs, p, key, u64(l))
} else if hasSSE2 {
accumSSE(&accs, p, key, u64(l))
} else {
accumScalar(&accs, p, key, u64(l))
}
// merge accs
acc += mulFold64(accs[0]^key64_011, accs[1]^key64_019)
acc += mulFold64(accs[2]^key64_027, accs[3]^key64_035)
acc += mulFold64(accs[4]^key64_043, accs[5]^key64_051)
acc += mulFold64(accs[6]^key64_059, accs[7]^key64_067)
return xxh3Avalanche(acc)
}
}