from float + tests

This commit is contained in:
Quaternions 2024-10-01 15:00:30 -07:00
parent 9b3dde66bd
commit 0d05540a6b
2 changed files with 144 additions and 0 deletions

View File

@ -197,6 +197,97 @@ macro_rules! impl_into_float {
impl_into_float!(f32,u32,8,24); impl_into_float!(f32,u32,8,24);
impl_into_float!(f64,u64,11,53); impl_into_float!(f64,u64,11,53);
#[inline]
fn integer_decode_f32(f: f32) -> (u64, i16, bool) {
let bits: u32 = f.to_bits();
let sign: bool = bits & (1<<31) != 0;
let mut exponent: i16 = ((bits >> 23) & 0xff) as i16;
let mantissa = if exponent == 0 {
(bits & 0x7fffff) << 1
} else {
(bits & 0x7fffff) | 0x800000
};
// Exponent bias + mantissa shift
exponent -= 127 + 23;
(mantissa as u64, exponent, sign)
}
#[inline]
fn integer_decode_f64(f: f64) -> (u64, i16, bool) {
let bits: u64 = f.to_bits();
let sign: bool = bits & (1u64<<63) != 0;
let mut exponent: i16 = ((bits >> 52) & 0x7ff) as i16;
let mantissa = if exponent == 0 {
(bits & 0xfffffffffffff) << 1
} else {
(bits & 0xfffffffffffff) | 0x10000000000000
};
// Exponent bias + mantissa shift
exponent -= 1023 + 52;
(mantissa, exponent, sign)
}
#[derive(Debug,Eq,PartialEq)]
pub enum FixedFromFloatError{
Nan,
Infinite,
Overflow,
Underflow,
}
impl FixedFromFloatError{
pub fn underflow_to_zero<const N:usize,const F:usize>(self)->Result<Fixed<N,F>,Self>{
match self{
FixedFromFloatError::Underflow=>Ok(Fixed::ZERO),
_=>Err(self),
}
}
}
macro_rules! impl_from_float {
( $decode:ident, $input: ty, $mantissa_bits:expr ) => {
impl<const N:usize,const F:usize> TryFrom<$input> for Fixed<N,F>{
type Error=FixedFromFloatError;
#[inline]
fn try_from(value:$input)->Result<Self,Self::Error>{
const DIGIT_SHIFT:u32=6;
match value.classify(){
std::num::FpCategory::Nan=>Err(FixedFromFloatError::Nan),
std::num::FpCategory::Infinite=>Err(FixedFromFloatError::Infinite),
std::num::FpCategory::Zero=>Ok(Self::ZERO),
std::num::FpCategory::Subnormal
|std::num::FpCategory::Normal
=>{
let (m,e,s)=$decode(value);
let mut digits=[0u64;N];
let most_significant_bit=e as i32+$mantissa_bits as i32+F as i32;
if most_significant_bit<0{
return Err(FixedFromFloatError::Underflow);
}
let digit_index=most_significant_bit>>DIGIT_SHIFT;
let digit=digits.get_mut(digit_index as usize).ok_or(FixedFromFloatError::Overflow)?;
let take_bits=most_significant_bit-(digit_index<<DIGIT_SHIFT);
let rest_of_mantissa=-($mantissa_bits as i32-(take_bits as i32));
*digit=signed_shift(m,rest_of_mantissa);
if rest_of_mantissa<0&&digit_index!=0{
//we don't care if some float bits are partially truncated
if let Some(digit)=digits.get_mut((digit_index-1) as usize){
let take_bits=most_significant_bit-((digit_index-1)<<DIGIT_SHIFT);
let rest_of_mantissa=-($mantissa_bits as i32-(take_bits as i32));
*digit=signed_shift(m,rest_of_mantissa);
}
}
let bits=BInt::from_bits(bnum::BUint::from_digits(digits));
Ok(if s{
Self::from_bits(bits.overflowing_neg().0)
}else{
Self::from_bits(bits)
})
},
}
}
}
}
}
impl_from_float!(integer_decode_f32,f32,24);
impl_from_float!(integer_decode_f64,f64,53);
impl<const N:usize,const F:usize> core::fmt::Display for Fixed<N,F>{ impl<const N:usize,const F:usize> core::fmt::Display for Fixed<N,F>{
#[inline] #[inline]
fn fmt(&self,f:&mut core::fmt::Formatter)->Result<(),core::fmt::Error>{ fn fmt(&self,f:&mut core::fmt::Formatter)->Result<(),core::fmt::Error>{

View File

@ -37,6 +37,59 @@ fn to_f64(){
assert_eq!(f,237946589723468975f64*2.0f64.powi(16)); assert_eq!(f,237946589723468975f64*2.0f64.powi(16));
} }
#[test]
fn from_f32(){
let a=I256F256::from(1)>>2;
let b:Result<I256F256,_>=0.25f32.try_into();
assert_eq!(b,Ok(a));
let a=I256F256::from(-1)>>2;
let b:Result<I256F256,_>=(-0.25f32).try_into();
assert_eq!(b,Ok(a));
let a=I256F256::from(0);
let b:Result<I256F256,_>=0.try_into();
assert_eq!(b,Ok(a));
let a=I256F256::from(0b101011110101001010101010000000000000000000000000000i64)<<16;
let b:Result<I256F256,_>=(0b101011110101001010101010000000000000000000000000000u64 as f32*2.0f32.powi(16)).try_into();
assert_eq!(b,Ok(a));
//I32F32::MAX into f32 is truncated into this value
let a=I32F32::raw(0b111111111111111111111111000000000000000000000000000000000000000i64);
let b:Result<I32F32,_>=Into::<f32>::into(I32F32::MAX).try_into();
assert_eq!(b,Ok(a));
//I32F32::MIN hits a special case since it's not representable as a positive signed integer
//TODO: don't return an overflow because this is technically possible
let a=I32F32::MIN;
let b:Result<I32F32,_>=Into::<f32>::into(I32F32::MIN).try_into();
assert_eq!(b,Err(crate::fixed::FixedFromFloatError::Overflow));
//16 is within the 24 bits of float precision
let b:Result<I32F32,_>=Into::<f32>::into(-I32F32::MIN.fix_2()).try_into();
assert_eq!(b,Err(crate::fixed::FixedFromFloatError::Overflow));
let b:Result<I32F32,_>=f32::MIN_POSITIVE.try_into();
assert_eq!(b,Err(crate::fixed::FixedFromFloatError::Underflow));
//test many cases
for i in 0..64{
let a=crate::fixed::Fixed::<2,64>::raw_digit(0b111111111111111111111111000000000000000000000000000000000000000i64)<<i;
let f:f32=a.into();
let b:Result<crate::fixed::Fixed<2,64>,_>=f.try_into();
assert_eq!(b,Ok(a));
}
}
#[test]
fn from_f64(){
let a=I256F256::from(1)>>2;
let b:Result<I256F256,_>=0.25f64.try_into();
assert_eq!(b,Ok(a));
let a=I256F256::from(-1)>>2;
let b:Result<I256F256,_>=(-0.25f64).try_into();
assert_eq!(b,Ok(a));
let a=I256F256::from(0);
let b:Result<I256F256,_>=0.try_into();
assert_eq!(b,Ok(a));
let a=I256F256::from(0b101011110101001010101010000000000000000000000000000i64)<<16;
let b:Result<I256F256,_>=(0b101011110101001010101010000000000000000000000000000u64 as f64*2.0f64.powi(16)).try_into();
assert_eq!(b,Ok(a));
}
#[test] #[test]
fn you_can_shr_numbers(){ fn you_can_shr_numbers(){
let a=I32F32::from(4); let a=I32F32::from(4);