num_bigint/bigint/
shift.rsuse super::BigInt;
use super::Sign::NoSign;
use core::ops::{Shl, ShlAssign, Shr, ShrAssign};
use num_traits::{PrimInt, Signed, Zero};
macro_rules! impl_shift {
(@ref $Shx:ident :: $shx:ident, $ShxAssign:ident :: $shx_assign:ident, $rhs:ty) => {
impl<'b> $Shx<&'b $rhs> for BigInt {
type Output = BigInt;
#[inline]
fn $shx(self, rhs: &'b $rhs) -> BigInt {
$Shx::$shx(self, *rhs)
}
}
impl<'a, 'b> $Shx<&'b $rhs> for &'a BigInt {
type Output = BigInt;
#[inline]
fn $shx(self, rhs: &'b $rhs) -> BigInt {
$Shx::$shx(self, *rhs)
}
}
impl<'b> $ShxAssign<&'b $rhs> for BigInt {
#[inline]
fn $shx_assign(&mut self, rhs: &'b $rhs) {
$ShxAssign::$shx_assign(self, *rhs);
}
}
};
($($rhs:ty),+) => {$(
impl Shl<$rhs> for BigInt {
type Output = BigInt;
#[inline]
fn shl(self, rhs: $rhs) -> BigInt {
BigInt::from_biguint(self.sign, self.data << rhs)
}
}
impl<'a> Shl<$rhs> for &'a BigInt {
type Output = BigInt;
#[inline]
fn shl(self, rhs: $rhs) -> BigInt {
BigInt::from_biguint(self.sign, &self.data << rhs)
}
}
impl ShlAssign<$rhs> for BigInt {
#[inline]
fn shl_assign(&mut self, rhs: $rhs) {
self.data <<= rhs
}
}
impl_shift! { @ref Shl::shl, ShlAssign::shl_assign, $rhs }
impl Shr<$rhs> for BigInt {
type Output = BigInt;
#[inline]
fn shr(self, rhs: $rhs) -> BigInt {
let round_down = shr_round_down(&self, rhs);
let data = self.data >> rhs;
let data = if round_down { data + 1u8 } else { data };
BigInt::from_biguint(self.sign, data)
}
}
impl<'a> Shr<$rhs> for &'a BigInt {
type Output = BigInt;
#[inline]
fn shr(self, rhs: $rhs) -> BigInt {
let round_down = shr_round_down(self, rhs);
let data = &self.data >> rhs;
let data = if round_down { data + 1u8 } else { data };
BigInt::from_biguint(self.sign, data)
}
}
impl ShrAssign<$rhs> for BigInt {
#[inline]
fn shr_assign(&mut self, rhs: $rhs) {
let round_down = shr_round_down(self, rhs);
self.data >>= rhs;
if round_down {
self.data += 1u8;
} else if self.data.is_zero() {
self.sign = NoSign;
}
}
}
impl_shift! { @ref Shr::shr, ShrAssign::shr_assign, $rhs }
)*};
}
impl_shift! { u8, u16, u32, u64, u128, usize }
impl_shift! { i8, i16, i32, i64, i128, isize }
fn shr_round_down<T: PrimInt>(i: &BigInt, shift: T) -> bool {
if i.is_negative() {
let zeros = i.trailing_zeros().expect("negative values are non-zero");
shift > T::zero() && shift.to_u64().map(|shift| zeros < shift).unwrap_or(true)
} else {
false
}
}