Conlang/stdlib/num.cl

463 lines
10 KiB
Common Lisp

//! The primitive numeric types
#[intrinsic = "bool"]
pub type bool;
#[intrinsic = "char"]
pub type char;
#[intrinsic = "i8"]
pub type i8;
#[intrinsic = "i16"]
pub type i16;
#[intrinsic = "i32"]
pub type i32;
#[intrinsic = "i64"]
pub type i64;
#[intrinsic = "isize"]
pub type isize;
#[intrinsic = "u8"]
pub type u8;
#[intrinsic = "u16"]
pub type u16;
#[intrinsic = "u32"]
pub type u32;
#[intrinsic = "u64"]
pub type u64;
#[intrinsic = "usize"]
pub type usize;
// Contains implementations for (TODO) overloaded operators on num types
pub mod ops {
use super::*;
// Represents an ordering comparison verdict
pub enum Ordering {
Less,
Equal,
Greater,
}
impl bool {
pub const MIN: Self = false;
pub const MAX: Self = true;
pub const BIT_WIDTH: u32 = 1;
pub fn default() -> Self {
false
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl char {
pub fn default() -> Self {
'\u{1f988}'
}
}
impl u8 {
pub const MIN: Self = 0;
pub const MAX: Self = !0;
pub const BIT_WIDTH: u32 = 1;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl u16 {
pub const MIN: Self = 0;
pub const MAX: Self = !0;
pub const BIT_WIDTH: u32 = 2;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl u32 {
pub const MIN: Self = 0;
pub const MAX: Self = !0;
pub const BIT_WIDTH: u32 = 4;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl u64 {
pub const MIN: Self = 0;
pub const MAX: Self = !0;
pub const BIT_WIDTH: u32 = 8;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl usize {
pub const MIN: Self = __march_ptr_width_unsigned_min();
pub const MAX: Self = __march_ptr_width_unsigned_max();
pub const BIT_WIDTH: u32 = __march_ptr_width_bits();
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl i8 {
pub const MIN: Self = -128;
pub const MAX: Self = 127;
pub const BIT_WIDTH: u32 = 1;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl i16 {
pub const MIN: Self = -32768;
pub const MAX: Self = 32767;
pub const BIT_WIDTH: u32 = 2;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl i32 {
pub const MIN: Self = -2147483648;
pub const MAX: Self = 2147483647;
pub const BIT_WIDTH: u32 = 4;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl i64 {
pub const MIN: Self = -9223372036854775808;
pub const MAX: Self = 9223372036854775807;
pub const BIT_WIDTH: u32 = 8;
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
impl isize {
pub const MIN: Self = __march_ptr_width_signed_min();
pub const MAX: Self = __march_ptr_width_signed_max();
pub const BIT_WIDTH: u32 = __march_ptr_width_bits();
pub fn default() -> Self {
0
}
pub fn mul(a: Self, b: Self) -> Self {
a * b
}
pub fn div(a: Self, b: Self) -> Self {
a / b
}
pub fn rem(a: Self, b: Self) -> Self {
a % b
}
pub fn add(a: Self, b: Self) -> Self {
a + b
}
pub fn sub(a: Self, b: Self) -> Self {
a - b
}
pub fn shl(a: Self, b: u32) -> Self {
a << b
}
pub fn shr(a: Self, b: u32) -> Self {
a >> b
}
pub fn and(a: Self, b: Self) -> Self {
a & b
}
pub fn or(a: Self, b: Self) -> Self {
a | b
}
pub fn xor(a: Self, b: Self) -> Self {
a ^ b
}
}
}