mirror of
https://github.com/TwiN/gatus.git
synced 2024-11-22 07:53:38 +01:00
441 lines
9.9 KiB
Go
441 lines
9.9 KiB
Go
package uint128 // import "lukechampine.com/uint128"
|
|
|
|
import (
|
|
"encoding/binary"
|
|
"errors"
|
|
"fmt"
|
|
"math"
|
|
"math/big"
|
|
"math/bits"
|
|
)
|
|
|
|
// Zero is a zero-valued uint128.
|
|
var Zero Uint128
|
|
|
|
// Max is the largest possible uint128 value.
|
|
var Max = New(math.MaxUint64, math.MaxUint64)
|
|
|
|
// A Uint128 is an unsigned 128-bit number.
|
|
type Uint128 struct {
|
|
Lo, Hi uint64
|
|
}
|
|
|
|
// IsZero returns true if u == 0.
|
|
func (u Uint128) IsZero() bool {
|
|
// NOTE: we do not compare against Zero, because that is a global variable
|
|
// that could be modified.
|
|
return u == Uint128{}
|
|
}
|
|
|
|
// Equals returns true if u == v.
|
|
//
|
|
// Uint128 values can be compared directly with ==, but use of the Equals method
|
|
// is preferred for consistency.
|
|
func (u Uint128) Equals(v Uint128) bool {
|
|
return u == v
|
|
}
|
|
|
|
// Equals64 returns true if u == v.
|
|
func (u Uint128) Equals64(v uint64) bool {
|
|
return u.Lo == v && u.Hi == 0
|
|
}
|
|
|
|
// Cmp compares u and v and returns:
|
|
//
|
|
// -1 if u < v
|
|
// 0 if u == v
|
|
// +1 if u > v
|
|
//
|
|
func (u Uint128) Cmp(v Uint128) int {
|
|
if u == v {
|
|
return 0
|
|
} else if u.Hi < v.Hi || (u.Hi == v.Hi && u.Lo < v.Lo) {
|
|
return -1
|
|
} else {
|
|
return 1
|
|
}
|
|
}
|
|
|
|
// Cmp64 compares u and v and returns:
|
|
//
|
|
// -1 if u < v
|
|
// 0 if u == v
|
|
// +1 if u > v
|
|
//
|
|
func (u Uint128) Cmp64(v uint64) int {
|
|
if u.Hi == 0 && u.Lo == v {
|
|
return 0
|
|
} else if u.Hi == 0 && u.Lo < v {
|
|
return -1
|
|
} else {
|
|
return 1
|
|
}
|
|
}
|
|
|
|
// And returns u&v.
|
|
func (u Uint128) And(v Uint128) Uint128 {
|
|
return Uint128{u.Lo & v.Lo, u.Hi & v.Hi}
|
|
}
|
|
|
|
// And64 returns u&v.
|
|
func (u Uint128) And64(v uint64) Uint128 {
|
|
return Uint128{u.Lo & v, u.Hi & 0}
|
|
}
|
|
|
|
// Or returns u|v.
|
|
func (u Uint128) Or(v Uint128) Uint128 {
|
|
return Uint128{u.Lo | v.Lo, u.Hi | v.Hi}
|
|
}
|
|
|
|
// Or64 returns u|v.
|
|
func (u Uint128) Or64(v uint64) Uint128 {
|
|
return Uint128{u.Lo | v, u.Hi | 0}
|
|
}
|
|
|
|
// Xor returns u^v.
|
|
func (u Uint128) Xor(v Uint128) Uint128 {
|
|
return Uint128{u.Lo ^ v.Lo, u.Hi ^ v.Hi}
|
|
}
|
|
|
|
// Xor64 returns u^v.
|
|
func (u Uint128) Xor64(v uint64) Uint128 {
|
|
return Uint128{u.Lo ^ v, u.Hi ^ 0}
|
|
}
|
|
|
|
// Add returns u+v.
|
|
func (u Uint128) Add(v Uint128) Uint128 {
|
|
lo, carry := bits.Add64(u.Lo, v.Lo, 0)
|
|
hi, carry := bits.Add64(u.Hi, v.Hi, carry)
|
|
if carry != 0 {
|
|
panic("overflow")
|
|
}
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// AddWrap returns u+v with wraparound semantics; for example,
|
|
// Max.AddWrap(From64(1)) == Zero.
|
|
func (u Uint128) AddWrap(v Uint128) Uint128 {
|
|
lo, carry := bits.Add64(u.Lo, v.Lo, 0)
|
|
hi, _ := bits.Add64(u.Hi, v.Hi, carry)
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// Add64 returns u+v.
|
|
func (u Uint128) Add64(v uint64) Uint128 {
|
|
lo, carry := bits.Add64(u.Lo, v, 0)
|
|
hi, carry := bits.Add64(u.Hi, 0, carry)
|
|
if carry != 0 {
|
|
panic("overflow")
|
|
}
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// AddWrap64 returns u+v with wraparound semantics; for example,
|
|
// Max.AddWrap64(1) == Zero.
|
|
func (u Uint128) AddWrap64(v uint64) Uint128 {
|
|
lo, carry := bits.Add64(u.Lo, v, 0)
|
|
hi := u.Hi + carry
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// Sub returns u-v.
|
|
func (u Uint128) Sub(v Uint128) Uint128 {
|
|
lo, borrow := bits.Sub64(u.Lo, v.Lo, 0)
|
|
hi, borrow := bits.Sub64(u.Hi, v.Hi, borrow)
|
|
if borrow != 0 {
|
|
panic("underflow")
|
|
}
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// SubWrap returns u-v with wraparound semantics; for example,
|
|
// Zero.SubWrap(From64(1)) == Max.
|
|
func (u Uint128) SubWrap(v Uint128) Uint128 {
|
|
lo, borrow := bits.Sub64(u.Lo, v.Lo, 0)
|
|
hi, _ := bits.Sub64(u.Hi, v.Hi, borrow)
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// Sub64 returns u-v.
|
|
func (u Uint128) Sub64(v uint64) Uint128 {
|
|
lo, borrow := bits.Sub64(u.Lo, v, 0)
|
|
hi, borrow := bits.Sub64(u.Hi, 0, borrow)
|
|
if borrow != 0 {
|
|
panic("underflow")
|
|
}
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// SubWrap64 returns u-v with wraparound semantics; for example,
|
|
// Zero.SubWrap64(1) == Max.
|
|
func (u Uint128) SubWrap64(v uint64) Uint128 {
|
|
lo, borrow := bits.Sub64(u.Lo, v, 0)
|
|
hi := u.Hi - borrow
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// Mul returns u*v, panicking on overflow.
|
|
func (u Uint128) Mul(v Uint128) Uint128 {
|
|
hi, lo := bits.Mul64(u.Lo, v.Lo)
|
|
p0, p1 := bits.Mul64(u.Hi, v.Lo)
|
|
p2, p3 := bits.Mul64(u.Lo, v.Hi)
|
|
hi, c0 := bits.Add64(hi, p1, 0)
|
|
hi, c1 := bits.Add64(hi, p3, c0)
|
|
if (u.Hi != 0 && v.Hi != 0) || p0 != 0 || p2 != 0 || c1 != 0 {
|
|
panic("overflow")
|
|
}
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// MulWrap returns u*v with wraparound semantics; for example,
|
|
// Max.MulWrap(Max) == 1.
|
|
func (u Uint128) MulWrap(v Uint128) Uint128 {
|
|
hi, lo := bits.Mul64(u.Lo, v.Lo)
|
|
hi += u.Hi*v.Lo + u.Lo*v.Hi
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// Mul64 returns u*v, panicking on overflow.
|
|
func (u Uint128) Mul64(v uint64) Uint128 {
|
|
hi, lo := bits.Mul64(u.Lo, v)
|
|
p0, p1 := bits.Mul64(u.Hi, v)
|
|
hi, c0 := bits.Add64(hi, p1, 0)
|
|
if p0 != 0 || c0 != 0 {
|
|
panic("overflow")
|
|
}
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// MulWrap64 returns u*v with wraparound semantics; for example,
|
|
// Max.MulWrap64(2) == Max.Sub64(1).
|
|
func (u Uint128) MulWrap64(v uint64) Uint128 {
|
|
hi, lo := bits.Mul64(u.Lo, v)
|
|
hi += u.Hi * v
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// Div returns u/v.
|
|
func (u Uint128) Div(v Uint128) Uint128 {
|
|
q, _ := u.QuoRem(v)
|
|
return q
|
|
}
|
|
|
|
// Div64 returns u/v.
|
|
func (u Uint128) Div64(v uint64) Uint128 {
|
|
q, _ := u.QuoRem64(v)
|
|
return q
|
|
}
|
|
|
|
// QuoRem returns q = u/v and r = u%v.
|
|
func (u Uint128) QuoRem(v Uint128) (q, r Uint128) {
|
|
if v.Hi == 0 {
|
|
var r64 uint64
|
|
q, r64 = u.QuoRem64(v.Lo)
|
|
r = From64(r64)
|
|
} else {
|
|
// generate a "trial quotient," guaranteed to be within 1 of the actual
|
|
// quotient, then adjust.
|
|
n := uint(bits.LeadingZeros64(v.Hi))
|
|
v1 := v.Lsh(n)
|
|
u1 := u.Rsh(1)
|
|
tq, _ := bits.Div64(u1.Hi, u1.Lo, v1.Hi)
|
|
tq >>= 63 - n
|
|
if tq != 0 {
|
|
tq--
|
|
}
|
|
q = From64(tq)
|
|
// calculate remainder using trial quotient, then adjust if remainder is
|
|
// greater than divisor
|
|
r = u.Sub(v.Mul64(tq))
|
|
if r.Cmp(v) >= 0 {
|
|
q = q.Add64(1)
|
|
r = r.Sub(v)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// QuoRem64 returns q = u/v and r = u%v.
|
|
func (u Uint128) QuoRem64(v uint64) (q Uint128, r uint64) {
|
|
if u.Hi < v {
|
|
q.Lo, r = bits.Div64(u.Hi, u.Lo, v)
|
|
} else {
|
|
q.Hi, r = bits.Div64(0, u.Hi, v)
|
|
q.Lo, r = bits.Div64(r, u.Lo, v)
|
|
}
|
|
return
|
|
}
|
|
|
|
// Mod returns r = u%v.
|
|
func (u Uint128) Mod(v Uint128) (r Uint128) {
|
|
_, r = u.QuoRem(v)
|
|
return
|
|
}
|
|
|
|
// Mod64 returns r = u%v.
|
|
func (u Uint128) Mod64(v uint64) (r uint64) {
|
|
_, r = u.QuoRem64(v)
|
|
return
|
|
}
|
|
|
|
// Lsh returns u<<n.
|
|
func (u Uint128) Lsh(n uint) (s Uint128) {
|
|
if n > 64 {
|
|
s.Lo = 0
|
|
s.Hi = u.Lo << (n - 64)
|
|
} else {
|
|
s.Lo = u.Lo << n
|
|
s.Hi = u.Hi<<n | u.Lo>>(64-n)
|
|
}
|
|
return
|
|
}
|
|
|
|
// Rsh returns u>>n.
|
|
func (u Uint128) Rsh(n uint) (s Uint128) {
|
|
if n > 64 {
|
|
s.Lo = u.Hi >> (n - 64)
|
|
s.Hi = 0
|
|
} else {
|
|
s.Lo = u.Lo>>n | u.Hi<<(64-n)
|
|
s.Hi = u.Hi >> n
|
|
}
|
|
return
|
|
}
|
|
|
|
// LeadingZeros returns the number of leading zero bits in u; the result is 128
|
|
// for u == 0.
|
|
func (u Uint128) LeadingZeros() int {
|
|
if u.Hi > 0 {
|
|
return bits.LeadingZeros64(u.Hi)
|
|
}
|
|
return 64 + bits.LeadingZeros64(u.Lo)
|
|
}
|
|
|
|
// TrailingZeros returns the number of trailing zero bits in u; the result is
|
|
// 128 for u == 0.
|
|
func (u Uint128) TrailingZeros() int {
|
|
if u.Lo > 0 {
|
|
return bits.TrailingZeros64(u.Lo)
|
|
}
|
|
return 64 + bits.TrailingZeros64(u.Hi)
|
|
}
|
|
|
|
// OnesCount returns the number of one bits ("population count") in u.
|
|
func (u Uint128) OnesCount() int {
|
|
return bits.OnesCount64(u.Hi) + bits.OnesCount64(u.Lo)
|
|
}
|
|
|
|
// RotateLeft returns the value of u rotated left by (k mod 128) bits.
|
|
func (u Uint128) RotateLeft(k int) Uint128 {
|
|
const n = 128
|
|
s := uint(k) & (n - 1)
|
|
return u.Lsh(s).Or(u.Rsh(n - s))
|
|
}
|
|
|
|
// RotateRight returns the value of u rotated left by (k mod 128) bits.
|
|
func (u Uint128) RotateRight(k int) Uint128 {
|
|
return u.RotateLeft(-k)
|
|
}
|
|
|
|
// Reverse returns the value of u with its bits in reversed order.
|
|
func (u Uint128) Reverse() Uint128 {
|
|
return Uint128{bits.Reverse64(u.Hi), bits.Reverse64(u.Lo)}
|
|
}
|
|
|
|
// ReverseBytes returns the value of u with its bytes in reversed order.
|
|
func (u Uint128) ReverseBytes() Uint128 {
|
|
return Uint128{bits.ReverseBytes64(u.Hi), bits.ReverseBytes64(u.Lo)}
|
|
}
|
|
|
|
// Len returns the minimum number of bits required to represent u; the result is
|
|
// 0 for u == 0.
|
|
func (u Uint128) Len() int {
|
|
return 128 - u.LeadingZeros()
|
|
}
|
|
|
|
// String returns the base-10 representation of u as a string.
|
|
func (u Uint128) String() string {
|
|
if u.IsZero() {
|
|
return "0"
|
|
}
|
|
buf := []byte("0000000000000000000000000000000000000000") // log10(2^128) < 40
|
|
for i := len(buf); ; i -= 19 {
|
|
q, r := u.QuoRem64(1e19) // largest power of 10 that fits in a uint64
|
|
var n int
|
|
for ; r != 0; r /= 10 {
|
|
n++
|
|
buf[i-n] += byte(r % 10)
|
|
}
|
|
if q.IsZero() {
|
|
return string(buf[i-n:])
|
|
}
|
|
u = q
|
|
}
|
|
}
|
|
|
|
// PutBytes stores u in b in little-endian order. It panics if len(b) < 16.
|
|
func (u Uint128) PutBytes(b []byte) {
|
|
binary.LittleEndian.PutUint64(b[:8], u.Lo)
|
|
binary.LittleEndian.PutUint64(b[8:], u.Hi)
|
|
}
|
|
|
|
// Big returns u as a *big.Int.
|
|
func (u Uint128) Big() *big.Int {
|
|
i := new(big.Int).SetUint64(u.Hi)
|
|
i = i.Lsh(i, 64)
|
|
i = i.Xor(i, new(big.Int).SetUint64(u.Lo))
|
|
return i
|
|
}
|
|
|
|
// Scan implements fmt.Scanner.
|
|
func (u *Uint128) Scan(s fmt.ScanState, ch rune) error {
|
|
i := new(big.Int)
|
|
if err := i.Scan(s, ch); err != nil {
|
|
return err
|
|
} else if i.Sign() < 0 {
|
|
return errors.New("value cannot be negative")
|
|
} else if i.BitLen() > 128 {
|
|
return errors.New("value overflows Uint128")
|
|
}
|
|
u.Lo = i.Uint64()
|
|
u.Hi = i.Rsh(i, 64).Uint64()
|
|
return nil
|
|
}
|
|
|
|
// New returns the Uint128 value (lo,hi).
|
|
func New(lo, hi uint64) Uint128 {
|
|
return Uint128{lo, hi}
|
|
}
|
|
|
|
// From64 converts v to a Uint128 value.
|
|
func From64(v uint64) Uint128 {
|
|
return New(v, 0)
|
|
}
|
|
|
|
// FromBytes converts b to a Uint128 value.
|
|
func FromBytes(b []byte) Uint128 {
|
|
return New(
|
|
binary.LittleEndian.Uint64(b[:8]),
|
|
binary.LittleEndian.Uint64(b[8:]),
|
|
)
|
|
}
|
|
|
|
// FromBig converts i to a Uint128 value. It panics if i is negative or
|
|
// overflows 128 bits.
|
|
func FromBig(i *big.Int) (u Uint128) {
|
|
if i.Sign() < 0 {
|
|
panic("value cannot be negative")
|
|
} else if i.BitLen() > 128 {
|
|
panic("value overflows Uint128")
|
|
}
|
|
u.Lo = i.Uint64()
|
|
u.Hi = i.Rsh(i, 64).Uint64()
|
|
return u
|
|
}
|
|
|
|
// FromString parses s as a Uint128 value.
|
|
func FromString(s string) (u Uint128, err error) {
|
|
_, err = fmt.Sscan(s, &u)
|
|
return
|
|
}
|