ableos/hblang/src/codegen.rs

2377 lines
77 KiB
Rust
Raw Normal View History

use std::{
cell::{Cell, RefCell},
ops::Range,
};
2024-05-12 13:10:50 -05:00
use hbvm::Vm;
2024-05-17 12:53:59 -05:00
use crate::{
ident::{self, Ident},
2024-05-20 07:11:58 -05:00
parser::{idfl, ExprRef},
HashMap,
2024-05-17 12:53:59 -05:00
};
2024-05-12 04:52:58 -05:00
2024-05-10 14:33:42 -05:00
use {
2024-05-10 15:54:12 -05:00
crate::{
2024-05-11 15:22:08 -05:00
instrs, lexer, log,
2024-05-12 04:52:58 -05:00
parser::{self},
2024-05-10 15:54:12 -05:00
},
2024-05-10 14:33:42 -05:00
std::rc::Rc,
};
2024-05-09 11:16:01 -05:00
2024-05-12 04:52:58 -05:00
use {lexer::TokenKind as T, parser::Expr as E};
2024-05-20 07:11:58 -05:00
type FuncId = u32;
type Reg = u8;
2024-05-11 15:22:08 -05:00
type Type = u32;
2024-05-19 11:20:42 -05:00
type GlobalId = u32;
2024-05-11 15:22:08 -05:00
2024-05-20 07:11:58 -05:00
const VM_STACK_SIZE: usize = 1024 * 1024 * 2;
2024-05-14 05:17:39 -05:00
fn align_up(value: u64, align: u64) -> u64 {
(value + align - 1) & !(align - 1)
}
2024-05-19 11:20:42 -05:00
struct ItemId {
file: parser::FileId,
expr: parser::ExprRef,
2024-05-20 07:11:58 -05:00
id: u32,
2024-05-19 11:20:42 -05:00
}
2024-05-13 06:36:29 -05:00
#[derive(Debug, PartialEq, Eq)]
2024-05-19 11:20:42 -05:00
struct LinReg(Reg, Rc<RefCell<RegAlloc>>);
2024-05-12 04:52:58 -05:00
#[cfg(debug_assertions)]
impl Drop for LinReg {
fn drop(&mut self) {
2024-05-19 11:20:42 -05:00
self.1.borrow_mut().free(self.0)
2024-05-12 04:52:58 -05:00
}
}
struct Stack {
2024-05-14 05:17:39 -05:00
offset: u64,
size: u64,
alloc: Cell<Option<Rc<RefCell<StackAlloc>>>>,
}
impl std::fmt::Debug for Stack {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("Stack")
.field("offset", &self.offset)
.field("size", &self.size)
.finish()
}
}
impl PartialEq for Stack {
fn eq(&self, other: &Self) -> bool {
self.offset == other.offset && self.size == other.size
}
}
impl Eq for Stack {}
impl Stack {
fn leak(&self) {
self.alloc.set(None);
}
}
impl Drop for Stack {
2024-05-14 05:17:39 -05:00
fn drop(&mut self) {
2024-05-19 11:20:42 -05:00
if let Some(f) = self.alloc.get_mut().as_mut() {
f.borrow_mut().free(self.offset, self.size)
}
2024-05-14 05:17:39 -05:00
}
}
#[derive(Default, PartialEq, Eq, Debug)]
2024-05-14 05:17:39 -05:00
struct StackAlloc {
ranges: Vec<Range<u64>>,
height: u64,
}
impl StackAlloc {
fn alloc(&mut self, size: u64) -> (u64, u64) {
2024-05-14 05:17:39 -05:00
if let Some((index, range)) = self
.ranges
.iter_mut()
.enumerate()
.filter(|(_, range)| range.end - range.start >= size)
.min_by_key(|(_, range)| range.end - range.start)
{
let offset = range.start;
range.start += size;
if range.start == range.end {
self.ranges.swap_remove(index);
}
return (offset, size);
}
2024-05-14 05:17:39 -05:00
let offset = self.height;
self.height += size;
(offset, size)
2024-05-14 05:17:39 -05:00
}
fn free(&mut self, offset: u64, size: u64) {
2024-05-14 05:17:39 -05:00
let range = offset..offset + size;
// FIXME: we do more wor then we need to, rather we keep the sequence sorted and only scan
// element before and after the modified range
self.ranges.push(range);
self.ranges.sort_by_key(|range| range.start);
self.ranges.dedup_by(|b, a| {
if a.end == b.start {
a.end = b.end;
true
} else {
false
}
});
}
fn clear(&mut self) {
assert!(self.ranges.len() <= 1, "{:?}", self.ranges);
self.ranges.clear();
self.height = 0;
}
}
#[derive(Default, Debug)]
enum Ctx {
#[default]
None,
Inferred(Type),
Dest(Value),
2024-05-19 11:20:42 -05:00
DestUntyped(Loc),
}
2024-05-14 05:17:39 -05:00
impl Ctx {
fn ty(&self) -> Option<Type> {
Some(match self {
Self::Inferred(ty) => *ty,
Self::Dest(Value { ty, .. }) => *ty,
_ => return None,
})
}
2024-05-14 16:07:32 -05:00
fn loc(self) -> Option<Loc> {
Some(match self {
Self::Dest(Value { loc, .. }) => loc,
Self::DestUntyped(loc, ..) => loc,
_ => return None,
})
}
}
2024-05-20 07:11:58 -05:00
mod traps {
macro_rules! traps {
($($name:ident;)*) => {$(
pub const $name: u64 = ${index(0)};
)*};
}
traps! {
MAKE_STRUCT;
}
}
2024-05-12 04:52:58 -05:00
pub mod bt {
2024-05-11 15:22:08 -05:00
use super::*;
2024-05-20 07:11:58 -05:00
const fn array_to_lower_case<const N: usize>(array: [u8; N]) -> [u8; N] {
let mut result = [0; N];
let mut i = 0;
while i < N {
result[i] = array[i].to_ascii_lowercase();
i += 1;
}
result
}
// const string to lower case
2024-05-11 15:22:08 -05:00
macro_rules! builtin_type {
2024-05-20 07:11:58 -05:00
($($name:ident;)*) => {
$(pub const $name: Type = ${index(0)};)*
mod __lc_names {
use super::*;
$(pub const $name: &[u8] = &array_to_lower_case(unsafe {
*(stringify!($name).as_ptr() as *const [u8; stringify!($name).len()]) });)*
}
pub fn from_str(name: &str) -> Option<Type> {
match name.as_bytes() {
$(__lc_names::$name => Some($name),)*
_ => None,
}
}
pub fn to_str(ty: Type) -> &'static str {
match ty {
$(${index(0)} => unsafe { std::str::from_utf8_unchecked(__lc_names::$name) },)*
v => unreachable!("invalid type: {}", v),
}
}
};
2024-05-11 15:22:08 -05:00
}
builtin_type! {
2024-05-19 11:20:42 -05:00
UNDECLARED;
NEVER;
VOID;
2024-05-20 07:11:58 -05:00
TYPE;
BOOL;
2024-05-12 15:40:28 -05:00
U8;
2024-05-13 06:36:29 -05:00
U16;
U32;
UINT;
I8;
I16;
I32;
INT;
2024-05-11 15:22:08 -05:00
}
2024-05-13 06:36:29 -05:00
pub fn is_signed(ty: Type) -> bool {
2024-05-19 11:20:42 -05:00
(I8..=INT).contains(&ty)
2024-05-13 06:36:29 -05:00
}
pub fn is_unsigned(ty: Type) -> bool {
2024-05-19 11:20:42 -05:00
(U8..=UINT).contains(&ty)
2024-05-13 06:36:29 -05:00
}
pub fn strip_pointer(ty: Type) -> Type {
match TypeKind::from_ty(ty) {
TypeKind::Pointer(_) => INT,
2024-05-13 06:36:29 -05:00
_ => ty,
}
}
2024-05-13 07:23:19 -05:00
pub fn is_pointer(ty: Type) -> bool {
matches!(TypeKind::from_ty(ty), TypeKind::Pointer(_))
}
pub fn try_upcast(oa: Type, ob: Type) -> Option<Type> {
let (oa, ob) = (oa.min(ob), oa.max(ob));
let (a, b) = (strip_pointer(oa), strip_pointer(ob));
Some(match () {
_ if oa == ob => oa,
_ if is_signed(a) && is_signed(b) || is_unsigned(a) && is_unsigned(b) => ob,
_ if is_unsigned(a) && is_signed(b) && a - U8 < b - I8 => ob,
2024-05-13 06:36:29 -05:00
_ => return None,
})
}
2024-05-11 15:22:08 -05:00
}
2024-05-20 07:11:58 -05:00
macro_rules! type_kind {
($name:ident {$( $variant:ident, )*}) => {
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum $name {
$($variant(Type),)*
}
2024-05-11 15:22:08 -05:00
2024-05-20 07:11:58 -05:00
impl $name {
const FLAG_BITS: u32 = (${count($variant)} as u32).next_power_of_two().ilog2();
const FLAG_OFFSET: u32 = std::mem::size_of::<Type>() as u32 * 8 - Self::FLAG_BITS;
const INDEX_MASK: u32 = (1 << (32 - Self::FLAG_BITS)) - 1;
fn from_ty(ty: Type) -> Self {
let (flag, index) = (ty >> Self::FLAG_OFFSET, ty & Self::INDEX_MASK);
match flag {
$(${index(0)} => Self::$variant(index),)*
_ => unreachable!(),
}
}
const fn encode(self) -> Type {
let (index, flag) = match self {
$(Self::$variant(index) => (index, ${index(0)}),)*
};
(flag << Self::FLAG_OFFSET) | index
}
2024-05-11 15:22:08 -05:00
}
2024-05-20 07:11:58 -05:00
};
}
type_kind! {
TypeKind {
Builtin,
Struct,
Pointer,
Func,
Global,
2024-05-11 15:22:08 -05:00
}
2024-05-20 07:11:58 -05:00
}
2024-05-12 04:52:58 -05:00
2024-05-20 07:11:58 -05:00
impl Default for TypeKind {
fn default() -> Self {
Self::Builtin(bt::UNDECLARED)
2024-05-12 04:52:58 -05:00
}
2024-05-11 15:22:08 -05:00
}
const STACK_PTR: Reg = 254;
const ZERO: Reg = 0;
const RET_ADDR: Reg = 31;
2024-05-10 14:33:42 -05:00
struct Reloc {
2024-05-20 07:11:58 -05:00
id: Type,
2024-05-10 14:33:42 -05:00
offset: u32,
2024-05-11 09:04:13 -05:00
instr_offset: u16,
size: u16,
2024-05-10 14:33:42 -05:00
}
#[derive(Default)]
2024-05-20 07:11:58 -05:00
pub struct Block {
2024-05-10 14:33:42 -05:00
code: Vec<u8>,
relocs: Vec<Reloc>,
}
2024-05-20 07:11:58 -05:00
impl Block {
2024-05-10 14:33:42 -05:00
pub fn extend(&mut self, bytes: &[u8]) {
self.code.extend_from_slice(bytes);
}
2024-05-20 07:11:58 -05:00
pub fn offset(&mut self, id: FuncId, instr_offset: u16, size: u16) {
2024-05-10 14:33:42 -05:00
self.relocs.push(Reloc {
2024-05-20 07:11:58 -05:00
id: TypeKind::Func(id).encode(),
2024-05-11 09:04:13 -05:00
offset: self.code.len() as u32,
instr_offset,
2024-05-10 14:33:42 -05:00
size,
});
}
fn encode(&mut self, (len, instr): (usize, [u8; instrs::MAX_SIZE])) {
let name = instrs::NAMES[instr[0] as usize];
2024-05-11 15:22:08 -05:00
log::dbg!(
"{:08x}: {}: {}",
self.code.len(),
name,
instr
.iter()
.take(len)
.skip(1)
.map(|b| format!("{:02x}", b))
.collect::<String>()
);
self.code.extend_from_slice(&instr[..len]);
}
2024-05-13 06:36:29 -05:00
fn short_cut_bin_op(&mut self, dest: Reg, src: Reg, imm: u64) -> bool {
2024-05-19 11:20:42 -05:00
if imm == 0 && dest != src {
self.encode(instrs::cp(dest, src));
2024-05-13 06:36:29 -05:00
}
imm != 0
}
fn addi64(&mut self, dest: Reg, src: Reg, imm: u64) {
if self.short_cut_bin_op(dest, src, imm) {
self.encode(instrs::addi64(dest, src, imm));
}
2024-05-10 14:33:42 -05:00
}
2024-05-20 07:11:58 -05:00
fn call(&mut self, func: FuncId) {
self.offset(func, 3, 4);
self.encode(instrs::jal(RET_ADDR, ZERO, 0));
2024-05-10 14:33:42 -05:00
}
fn ret(&mut self) {
self.encode(instrs::jala(ZERO, RET_ADDR, 0));
2024-05-10 14:33:42 -05:00
}
2024-05-20 07:11:58 -05:00
fn prelude(&mut self) {
self.encode(instrs::jal(RET_ADDR, ZERO, 0));
self.encode(instrs::tx());
2024-05-10 15:54:12 -05:00
}
2024-05-20 07:11:58 -05:00
fn relocate(&mut self, labels: &[Func], globals: &[Global], shift: i64, skip: usize) {
for reloc in self.relocs.iter().skip(skip) {
let offset = match TypeKind::from_ty(reloc.id) {
TypeKind::Func(id) => labels[id as usize].offset,
TypeKind::Global(id) => globals[id as usize].offset,
v => unreachable!("invalid reloc: {:?}", v),
2024-05-19 11:20:42 -05:00
};
2024-05-20 07:11:58 -05:00
let offset = if reloc.size == 8 {
2024-05-10 14:33:42 -05:00
reloc.offset as i64
} else {
2024-05-19 11:20:42 -05:00
offset as i64 - reloc.offset as i64
2024-05-10 14:33:42 -05:00
} + shift;
2024-05-20 07:11:58 -05:00
write_reloc(
&mut self.code,
reloc.offset as usize + reloc.instr_offset as usize,
offset,
reloc.size,
);
2024-05-10 14:33:42 -05:00
}
}
2024-05-19 11:20:42 -05:00
2024-05-20 07:11:58 -05:00
fn append(&mut self, data: &mut Block, code_offset: usize, reloc_offset: usize) {
2024-05-19 11:20:42 -05:00
for reloc in &mut data.relocs[reloc_offset..] {
reloc.offset += self.code.len() as u32;
reloc.offset -= code_offset as u32;
}
self.relocs.extend(data.relocs.drain(reloc_offset..));
self.code.extend(data.code.drain(code_offset..));
}
2024-05-10 14:33:42 -05:00
}
2024-05-20 07:11:58 -05:00
fn write_reloc(doce: &mut [u8], offset: usize, value: i64, size: u16) {
debug_assert!(size <= 8);
debug_assert!(size.is_power_of_two());
debug_assert!(
doce[offset..offset + size as usize].iter().all(|&b| b == 0),
"{:?}",
&doce[offset..offset + size as usize]
);
let value = value.to_ne_bytes();
doce[offset..offset + size as usize].copy_from_slice(&value[..size as usize]);
}
2024-05-16 05:42:11 -05:00
#[derive(Default, PartialEq, Eq)]
pub struct RegAlloc {
2024-05-12 15:40:28 -05:00
free: Vec<Reg>,
max_used: Reg,
}
2024-05-16 05:42:11 -05:00
impl std::fmt::Debug for RegAlloc {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("RegAlloc").finish()
}
}
impl RegAlloc {
2024-05-11 09:04:13 -05:00
fn init_callee(&mut self) {
2024-05-10 14:33:42 -05:00
self.free.clear();
2024-05-12 15:40:28 -05:00
self.free.extend((32..=253).rev());
self.max_used = RET_ADDR;
}
fn allocate(&mut self) -> Reg {
let reg = self.free.pop().expect("TODO: we need to spill");
2024-05-12 15:40:28 -05:00
self.max_used = self.max_used.max(reg);
reg
}
fn free(&mut self, reg: Reg) {
self.free.push(reg);
}
2024-05-12 13:10:50 -05:00
2024-05-12 15:40:28 -05:00
fn pushed_size(&self) -> usize {
2024-05-20 07:11:58 -05:00
((self.max_used as usize).saturating_sub(RET_ADDR as usize) + 1) * 8
2024-05-12 13:10:50 -05:00
}
}
2024-05-09 11:16:01 -05:00
2024-05-12 16:45:28 -05:00
#[derive(Clone)]
2024-05-20 07:11:58 -05:00
struct Func {
2024-05-10 14:33:42 -05:00
offset: u32,
2024-05-20 07:11:58 -05:00
relocs: u32,
2024-05-12 16:45:28 -05:00
args: Rc<[Type]>,
ret: Type,
2024-05-10 14:33:42 -05:00
}
2024-05-12 04:52:58 -05:00
struct Variable {
id: Ident,
value: Value,
}
2024-05-11 09:04:13 -05:00
struct RetReloc {
offset: u32,
instr_offset: u16,
size: u16,
}
2024-05-11 11:16:27 -05:00
struct Loop {
2024-05-14 07:01:40 -05:00
var_count: usize,
offset: u32,
relocs: Vec<RetReloc>,
2024-05-11 11:16:27 -05:00
}
2024-05-11 15:22:08 -05:00
struct Struct {
2024-05-12 06:13:36 -05:00
fields: Rc<[(Rc<str>, Type)]>,
2024-05-11 15:22:08 -05:00
}
2024-05-12 15:40:28 -05:00
struct TypeDisplay<'a> {
2024-05-19 11:20:42 -05:00
codegen: &'a Codegen,
2024-05-12 15:40:28 -05:00
ty: Type,
}
impl<'a> TypeDisplay<'a> {
2024-05-19 11:20:42 -05:00
fn new(codegen: &'a Codegen, ty: Type) -> Self {
2024-05-12 15:40:28 -05:00
Self { codegen, ty }
}
fn rety(&self, ty: Type) -> Self {
Self::new(self.codegen, ty)
}
}
impl<'a> std::fmt::Display for TypeDisplay<'a> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
use TypeKind as TK;
2024-05-20 07:11:58 -05:00
match TK::from_ty(self.ty) {
TK::Builtin(ty) => write!(f, "{}", bt::to_str(ty)),
2024-05-12 15:40:28 -05:00
TK::Pointer(ty) => {
2024-05-20 07:11:58 -05:00
write!(f, "^{}", self.rety(self.codegen.pointers[ty as usize]))
2024-05-12 15:40:28 -05:00
}
2024-05-20 07:11:58 -05:00
_ if let Some((key, _)) =
self.codegen.symbols.iter().find(|(_, &ty)| ty == self.ty)
&& let Some(name) = self.codegen.files[key.file as usize]
.exprs()
.iter()
.find_map(|expr| match expr {
E::BinOp {
left: &E::Ident { name, id, .. },
op: T::Decl,
..
} if id == key.id => Some(name),
_ => None,
}) =>
{
write!(f, "{name}")
}
TK::Struct(idx) => {
let record = &self.codegen.structs[idx as usize];
write!(f, "{{")?;
for (i, &(ref name, ty)) in record.fields.iter().enumerate() {
if i != 0 {
write!(f, ", ")?;
}
write!(f, "{}: {}", name, self.rety(ty))?;
}
write!(f, "}}")
}
TK::Func(idx) => write!(f, "fn{}", idx),
TK::Global(idx) => write!(f, "global{}", idx),
}
2024-05-12 15:40:28 -05:00
}
}
struct Global {
2024-05-19 11:20:42 -05:00
code: u32,
offset: u32,
dep: GlobalId,
ty: Type,
}
2024-05-20 07:11:58 -05:00
#[derive(Default)]
struct Linked {
globals: usize,
relocs: usize,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
struct SymKey {
id: Ident,
file: parser::FileId,
}
2024-05-12 17:02:32 -05:00
#[derive(Default)]
2024-05-19 11:20:42 -05:00
pub struct Codegen {
cf: parser::Ast,
cf_id: parser::FileId,
2024-05-20 07:11:58 -05:00
ret: Type,
ret_reg: Option<Reg>,
cur_item: TypeKind,
gpa: Rc<RefCell<RegAlloc>>,
sa: Rc<RefCell<StackAlloc>>,
2024-05-11 09:04:13 -05:00
ret_relocs: Vec<RetReloc>,
2024-05-12 13:10:50 -05:00
loops: Vec<Loop>,
2024-05-20 07:11:58 -05:00
vars: Vec<Variable>,
to_generate: Vec<ItemId>,
2024-05-20 07:11:58 -05:00
code: Block,
data: Block,
2024-05-17 12:53:59 -05:00
2024-05-20 07:11:58 -05:00
symbols: HashMap<SymKey, Type>,
funcs: Vec<Func>,
2024-05-19 11:20:42 -05:00
globals: Vec<Global>,
2024-05-20 07:11:58 -05:00
structs: Vec<Struct>,
2024-05-19 11:20:42 -05:00
pointers: Vec<Type>,
2024-05-09 11:16:01 -05:00
2024-05-19 11:20:42 -05:00
pub files: Vec<parser::Ast>,
2024-05-12 16:19:45 -05:00
2024-05-20 07:11:58 -05:00
vm: Vm<LoggedMem, 0>,
stack: Vec<u8>,
linked: Linked,
2024-05-19 11:20:42 -05:00
}
2024-05-19 11:20:42 -05:00
impl Codegen {
2024-05-20 07:11:58 -05:00
fn with_cached_progress(&mut self, f: impl FnOnce(&mut Self)) {
let ret = std::mem::take(&mut self.ret);
let ret_reg = std::mem::take(&mut self.ret_reg);
let cur_item = std::mem::take(&mut self.cur_item);
let gpa = std::mem::take(&mut *self.gpa.borrow_mut());
let sa = std::mem::take(&mut *self.sa.borrow_mut());
let ret_relocs = self.ret_relocs.len();
let loops = self.loops.len();
let vars = self.vars.len();
f(self);
self.ret = ret;
self.ret_reg = ret_reg;
self.cur_item = cur_item;
*self.gpa.borrow_mut() = gpa;
*self.sa.borrow_mut() = sa;
self.ret_relocs.truncate(ret_relocs);
self.loops.truncate(loops);
self.vars.truncate(vars);
}
fn lazy_init(&mut self) {
if self.stack.capacity() == 0 {
self.stack.reserve(VM_STACK_SIZE);
self.vm.write_reg(
STACK_PTR,
unsafe { self.stack.as_ptr().add(self.stack.capacity()) } as u64,
);
}
}
2024-05-19 11:20:42 -05:00
pub fn generate(&mut self) {
2024-05-20 07:11:58 -05:00
self.lazy_init();
2024-05-19 11:20:42 -05:00
self.find_and_declare(0, Err("main"));
2024-05-20 07:11:58 -05:00
self.code.prelude();
self.complete_call_graph();
}
fn complete_call_graph(&mut self) {
2024-05-19 11:20:42 -05:00
while let Some(item) = self.to_generate.pop() {
2024-05-20 07:11:58 -05:00
self.with_cached_progress(|s| s.generate_item(item));
2024-05-12 16:45:28 -05:00
}
2024-05-19 11:20:42 -05:00
}
2024-05-12 16:45:28 -05:00
2024-05-19 11:20:42 -05:00
fn generate_item(&mut self, item: ItemId) {
let ast = self.files[item.file as usize].clone();
let expr = item.expr.get(&ast).unwrap();
self.cf = ast.clone();
self.cf_id = item.file;
match expr {
E::BinOp {
2024-05-20 07:11:58 -05:00
left: E::Ident { name, .. },
2024-05-12 16:19:45 -05:00
op: T::Decl,
2024-05-19 11:20:42 -05:00
right: E::Closure { body, args, .. },
} => {
log::dbg!("fn: {}", name);
2024-05-20 07:11:58 -05:00
self.cur_item = TypeKind::Func(item.id);
self.funcs[item.id as usize].offset = self.code.code.len() as _;
self.funcs[item.id as usize].relocs = self.code.relocs.len() as _;
let func = self.funcs[item.id as usize].clone();
2024-05-19 11:20:42 -05:00
self.gpa.borrow_mut().init_callee();
2024-05-12 16:19:45 -05:00
2024-05-20 07:11:58 -05:00
self.gen_prelude();
2024-05-19 11:20:42 -05:00
log::dbg!("fn-args");
2024-05-20 07:11:58 -05:00
let mut parama = self.param_alloc(func.ret);
for (arg, &ty) in args.iter().zip(func.args.iter()) {
2024-05-19 11:20:42 -05:00
let sym = parser::find_symbol(&self.cf.symbols, arg.id);
let loc = self.load_arg(sym.flags, ty, &mut parama);
self.vars.push(Variable {
id: arg.id,
value: Value { ty, loc },
});
2024-05-12 16:45:28 -05:00
}
2024-05-20 07:11:58 -05:00
if self.size_of(func.ret) > 16 {
2024-05-19 11:20:42 -05:00
let reg = self.gpa.borrow_mut().allocate();
self.code.encode(instrs::cp(reg, 1));
self.ret_reg = Some(reg);
} else {
self.ret_reg = None;
}
2024-05-12 16:45:28 -05:00
2024-05-20 07:11:58 -05:00
self.ret = func.ret;
2024-05-12 16:45:28 -05:00
2024-05-19 11:20:42 -05:00
log::dbg!("fn-body");
if self.expr(body).is_some() {
self.report(body.pos(), "expected all paths in the fucntion to return");
}
2024-05-12 16:45:28 -05:00
2024-05-19 11:20:42 -05:00
log::dbg!("fn-prelude, stack: {:x}", self.sa.borrow().height);
2024-05-12 16:45:28 -05:00
2024-05-19 11:20:42 -05:00
log::dbg!("fn-relocs");
2024-05-20 07:11:58 -05:00
self.reloc_prelude(item.id);
2024-05-12 16:45:28 -05:00
2024-05-19 11:20:42 -05:00
log::dbg!("fn-ret");
self.reloc_rets();
self.ret();
self.sa.borrow_mut().clear();
2024-05-12 16:45:28 -05:00
}
2024-05-20 07:11:58 -05:00
value => todo!("{value:?}"),
2024-05-09 11:16:01 -05:00
}
}
2024-05-12 15:40:28 -05:00
fn align_of(&self, ty: Type) -> u64 {
use TypeKind as TK;
match TypeKind::from_ty(ty) {
2024-05-20 07:11:58 -05:00
TK::Struct(t) => self.structs[t as usize]
2024-05-12 15:40:28 -05:00
.fields
.iter()
.map(|&(_, ty)| self.align_of(ty))
.max()
.unwrap(),
_ => self.size_of(ty).max(1),
}
}
2024-05-12 04:52:58 -05:00
fn size_of(&self, ty: Type) -> u64 {
2024-05-12 15:40:28 -05:00
use TypeKind as TK;
match TK::from_ty(ty) {
TK::Pointer(_) => 8,
TK::Builtin(bt::VOID) => 0,
TK::Builtin(bt::NEVER) => unreachable!(),
2024-05-13 06:36:29 -05:00
TK::Builtin(bt::INT | bt::UINT) => 8,
2024-05-20 07:11:58 -05:00
TK::Builtin(bt::I32 | bt::U32 | bt::TYPE) => 4,
2024-05-12 15:40:28 -05:00
TK::Builtin(bt::I16 | bt::U16) => 2,
TK::Builtin(bt::I8 | bt::U8 | bt::BOOL) => 1,
TK::Struct(ty) => {
let mut offset = 0;
2024-05-20 07:11:58 -05:00
let record = &self.structs[ty as usize];
2024-05-12 15:40:28 -05:00
for &(_, ty) in record.fields.iter() {
let align = self.align_of(ty);
2024-05-14 05:17:39 -05:00
offset = align_up(offset, align);
2024-05-12 15:40:28 -05:00
offset += self.size_of(ty);
}
offset
}
2024-05-20 07:11:58 -05:00
_ => unimplemented!("size_of: {}", self.display_ty(ty)),
2024-05-12 04:52:58 -05:00
}
}
2024-05-12 15:40:28 -05:00
fn display_ty(&self, ty: Type) -> TypeDisplay {
TypeDisplay::new(self, ty)
}
2024-05-14 05:17:39 -05:00
fn unwrap_struct(&self, ty: Type, pos: parser::Pos, context: impl std::fmt::Display) -> Type {
match TypeKind::from_ty(ty) {
TypeKind::Struct(idx) => idx,
_ => self.report(
2024-05-12 16:19:45 -05:00
pos,
2024-05-14 05:17:39 -05:00
format_args!("expected struct, got {} ({context})", self.display_ty(ty)),
),
}
}
fn offset_of(&self, pos: parser::Pos, ty: Type, field: Result<&str, usize>) -> (u64, Type) {
let idx = self.unwrap_struct(ty, pos, "field access");
2024-05-20 07:11:58 -05:00
let record = &self.structs[idx as usize];
2024-05-12 15:40:28 -05:00
let mut offset = 0;
2024-05-20 07:11:58 -05:00
for (i, &(ref name, ty)) in record.fields.iter().enumerate() {
2024-05-14 05:17:39 -05:00
if Ok(name.as_ref()) == field || Err(i) == field {
2024-05-20 07:11:58 -05:00
return (offset, ty);
2024-05-12 15:40:28 -05:00
}
2024-05-20 07:11:58 -05:00
let align = self.align_of(ty);
2024-05-14 05:17:39 -05:00
offset = align_up(offset, align);
2024-05-20 07:11:58 -05:00
offset += self.size_of(ty);
2024-05-12 15:40:28 -05:00
}
2024-05-14 05:17:39 -05:00
match field {
Ok(i) => self.report(pos, format_args!("field not found: {i}")),
Err(field) => self.report(pos, format_args!("field not found: {field}")),
}
2024-05-12 15:40:28 -05:00
}
fn alloc_reg(&mut self) -> LinReg {
2024-05-19 11:20:42 -05:00
LinReg(self.gpa.borrow_mut().allocate(), self.gpa.clone())
}
fn alloc_stack(&mut self, size: u64) -> Rc<Stack> {
let (offset, size) = self.sa.borrow_mut().alloc(size);
Stack {
offset,
size,
alloc: Cell::new(Some(self.sa.clone())),
}
.into()
}
2024-05-13 06:36:29 -05:00
fn loc_to_reg(&mut self, loc: Loc, size: u64) -> LinReg {
match loc {
2024-05-12 04:52:58 -05:00
Loc::RegRef(rr) => {
let reg = self.alloc_reg();
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cp(reg.0, rr));
reg
}
Loc::Reg(reg) => reg,
Loc::Deref(dreg, .., offset) => {
let reg = self.alloc_reg();
2024-05-13 06:36:29 -05:00
self.code
.encode(instrs::ld(reg.0, dreg.0, offset, size as _));
2024-05-12 04:52:58 -05:00
reg
}
Loc::DerefRef(dreg, .., offset) => {
let reg = self.alloc_reg();
2024-05-13 06:36:29 -05:00
self.code.encode(instrs::ld(reg.0, dreg, offset, size as _));
2024-05-12 13:10:50 -05:00
reg
}
Loc::Imm(imm) => {
let reg = self.alloc_reg();
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::li64(reg.0, imm));
reg
}
2024-05-14 05:17:39 -05:00
Loc::Stack(stack, off) => {
let reg = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.load_stack(reg.0, stack.offset + off, size as _);
reg
}
}
}
fn store_stack(&mut self, reg: Reg, offset: u64, size: u16) {
self.code.encode(instrs::st(reg, STACK_PTR, offset, size));
}
fn load_stack(&mut self, reg: Reg, offset: u64, size: u16) {
self.code.encode(instrs::ld(reg, STACK_PTR, offset, size));
}
2024-05-11 09:04:13 -05:00
fn reloc_rets(&mut self) {
2024-05-20 07:11:58 -05:00
let len = self.code.code.len() as i64;
for reloc in self.ret_relocs.iter() {
write_reloc(
&mut self.code.code,
reloc.offset as usize + reloc.instr_offset as usize,
len - reloc.offset as i64,
reloc.size,
);
2024-05-11 09:04:13 -05:00
}
}
2024-05-19 11:20:42 -05:00
fn ty(&mut self, expr: &parser::Expr) -> Type {
2024-05-20 07:11:58 -05:00
let offset = self.code.code.len();
let reloc_offset = self.code.relocs.len();
let value = self.expr(expr).unwrap();
_ = self.assert_ty(expr.pos(), value.ty, bt::TYPE);
if let Loc::Imm(ty) = value.loc {
return ty as _;
}
self.code.encode(instrs::tx());
let mut curr_temp = Block::default();
curr_temp.append(&mut self.code, offset, reloc_offset);
let mut curr_fn = Block::default();
match self.cur_item {
TypeKind::Func(id) => {
let func = &self.funcs[id as usize];
curr_fn.append(&mut self.code, func.offset as _, func.relocs as _);
log::dbg!("{:?}", curr_fn.code);
2024-05-12 15:40:28 -05:00
}
2024-05-20 07:11:58 -05:00
foo => todo!("{foo:?}"),
}
let offset = self.code.code.len();
self.code.append(&mut curr_temp, 0, 0);
self.complete_call_graph();
self.link();
self.vm.pc = hbvm::mem::Address::new(&self.code.code[offset] as *const u8 as _);
loop {
match self.vm.run().unwrap() {
hbvm::VmRunOk::End => break,
hbvm::VmRunOk::Ecall => self.handle_ecall(),
_ => unreachable!(),
2024-05-12 06:13:36 -05:00
}
2024-05-20 07:11:58 -05:00
}
match self.cur_item {
TypeKind::Func(id) => {
self.funcs[id as usize].offset = self.code.code.len() as _;
self.funcs[id as usize].relocs = self.code.relocs.len() as _;
self.code.append(&mut curr_fn, 0, 0);
}
foo => todo!("{foo:?}"),
}
match value.loc {
Loc::RegRef(reg) | Loc::Reg(LinReg(reg, ..)) => self.vm.read_reg(reg).0 as _,
_ => unreachable!(),
}
}
fn handle_ecall(&mut self) {
// the ecalls have exception, we cant pass the return value in two registers otherwise its
// hard to tell where the trap code is
match self.vm.read_reg(2).0 {
traps::MAKE_STRUCT => unsafe {
let file_id = self.vm.read_reg(3).0 as u32;
let expr = std::mem::transmute::<_, parser::ExprRef>(self.vm.read_reg(4));
let mut captures_addr = (self.vm.read_reg(STACK_PTR).0 as *const u8)
.add(self.vm.read_reg(5).0 as usize);
let ast = self.files[file_id as usize].clone();
let &E::Struct {
pos,
fields,
captured,
} = expr.get(&ast).unwrap()
else {
unreachable!()
};
let prev_len = self.vars.len();
for &id in captured {
let ty: Type = std::ptr::read_unaligned(captures_addr.cast());
captures_addr = captures_addr.add(4);
let mut imm = [0u8; 8];
assert!(self.size_of(ty) as usize <= imm.len());
std::ptr::copy_nonoverlapping(
captures_addr,
imm.as_mut_ptr(),
self.size_of(ty) as usize,
);
self.vars.push(Variable {
id,
value: Value {
ty,
loc: Loc::Imm(u64::from_ne_bytes(imm)),
},
});
}
let Value {
loc: Loc::Imm(ty), ..
} = self
.expr(&E::Struct {
pos,
fields,
captured: &[],
})
.unwrap()
else {
unreachable!()
};
self.vars.truncate(prev_len);
self.vm.write_reg(1, ty);
},
trap => todo!("unknown trap: {trap}"),
2024-05-12 04:52:58 -05:00
}
}
2024-05-19 11:20:42 -05:00
fn expr(&mut self, expr: &parser::Expr) -> Option<Value> {
self.expr_ctx(expr, Ctx::default())
}
2024-05-20 07:11:58 -05:00
fn handle_global(&mut self, id: GlobalId) -> Option<Value> {
let ptr = self.alloc_reg();
let global = &mut self.globals[id as usize];
match self.cur_item {
TypeKind::Global(gl) => global.dep = global.dep.max(gl),
_ => {}
}
self.code.relocs.push(Reloc {
id: TypeKind::Global(id as _).encode(),
offset: self.code.code.len() as u32,
instr_offset: 3,
size: 4,
});
self.code.encode(instrs::lra(ptr.0, 0, 0));
Some(Value {
ty: global.ty,
loc: Loc::Deref(ptr, None, 0),
})
}
2024-05-19 11:20:42 -05:00
fn expr_ctx(&mut self, expr: &parser::Expr, mut ctx: Ctx) -> Option<Value> {
use instrs as i;
2024-05-20 07:11:58 -05:00
let value = match *expr {
2024-05-20 07:11:58 -05:00
E::Struct {
fields, captured, ..
} => {
if captured.is_empty() {
let fields = fields
.iter()
.map(|&(name, ty)| (name.into(), self.ty(&ty)))
.collect();
self.structs.push(Struct { fields });
Some(Value::ty(
TypeKind::Struct(self.structs.len() as u32 - 1).encode(),
))
} else {
let values = captured
.iter()
.map(|&id| E::Ident {
id,
name: "booodab",
index: u32::MAX,
})
.map(|expr| self.expr(&expr))
.collect::<Option<Vec<_>>>()?;
let values_size = values
.iter()
.map(|value| 4 + self.size_of(value.ty))
.sum::<u64>();
let stack = self.alloc_stack(values_size);
let ptr = Loc::DerefRef(STACK_PTR, None, stack.offset);
let mut offset = 0;
for value in values {
self.assign(bt::TYPE, ptr.offset_ref(offset), Loc::Imm(value.ty as _));
offset += 4;
self.assign(value.ty, ptr.offset_ref(offset), value.loc);
offset += self.size_of(value.ty);
}
// eca MAKE_STRUCT(FileId, ExprRef, *Captures) -> Type;
let mut parama = self.param_alloc(bt::TYPE);
self.pass_arg(&Value::imm(traps::MAKE_STRUCT), &mut parama);
self.pass_arg(&Value::imm(self.cf_id as _), &mut parama);
self.pass_arg(
&Value::imm(unsafe { std::mem::transmute(parser::ExprRef::new(expr)) }),
&mut parama,
);
self.pass_arg(&Value::imm(stack.offset), &mut parama);
self.code.encode(i::eca());
Some(Value {
ty: bt::TYPE,
loc: Loc::RegRef(1),
})
}
}
E::UnOp {
op: T::Xor, val, ..
} => {
let val = self.ty(val);
Some(Value::ty(self.alloc_pointer(val)))
}
E::Directive {
name: "TypeOf",
args: [expr],
..
} => {
let offset = self.code.code.len() as u32;
let reloc_offset = self.code.relocs.len();
let ty = self
.expr_ctx(expr, Ctx::DestUntyped(Loc::DerefRef(0, None, 0)))
.unwrap()
.ty;
self.code.code.truncate(offset as usize);
self.code.relocs.truncate(reloc_offset);
Some(Value {
ty: bt::TYPE,
loc: Loc::Imm(ty as _),
})
}
2024-05-14 16:07:32 -05:00
E::Directive {
name: "eca",
args: [ret_ty, args @ ..],
..
} => {
let ty = self.ty(ret_ty);
let mut parama = self.param_alloc(ty);
2024-05-14 16:07:32 -05:00
let mut values = Vec::with_capacity(args.len());
for arg in args {
let arg = self.expr(arg)?;
self.pass_arg(&arg, &mut parama);
values.push(arg.loc);
}
drop(values);
let loc = self.alloc_ret_loc(ty, ctx);
self.code.encode(i::eca());
2024-05-14 16:07:32 -05:00
self.post_process_ret_loc(ty, &loc);
return Some(Value { ty, loc });
}
E::Directive {
name: "sizeof",
args: [ty],
..
} => {
let ty = self.ty(ty);
let loc = Loc::Imm(self.size_of(ty));
return Some(Value { ty: bt::UINT, loc });
}
E::Directive {
name: "alignof",
args: [ty],
..
} => {
let ty = self.ty(ty);
let loc = Loc::Imm(self.align_of(ty));
return Some(Value { ty: bt::UINT, loc });
}
E::Directive {
name: "intcast",
args: [val],
..
} => {
let Some(ty) = ctx.ty() else {
self.report(
expr.pos(),
"type to cast to is unknown, use `@as(<type>, <expr>)`",
);
};
let mut val = self.expr(val)?;
let from_size = self.size_of(val.ty);
let to_size = self.size_of(ty);
if from_size < to_size && bt::is_signed(val.ty) {
let reg = self.loc_to_reg(val.loc, from_size);
let op = [i::sxt8, i::sxt16, i::sxt32][from_size.ilog2() as usize];
2024-05-14 16:07:32 -05:00
self.code.encode(op(reg.0, reg.0));
val.loc = Loc::Reg(reg);
}
Some(Value { ty, loc: val.loc })
}
E::Directive {
name: "bitcast",
args: [val],
..
} => {
let Some(ty) = ctx.ty() else {
self.report(
expr.pos(),
"type to cast to is unknown, use `@as(<type>, <expr>)`",
);
};
let size = self.size_of(ty);
ctx = match ctx {
Ctx::Dest(Value { loc, .. }) | Ctx::DestUntyped(loc, ..) => {
2024-05-19 11:20:42 -05:00
Ctx::DestUntyped(loc)
2024-05-14 16:07:32 -05:00
}
_ => Ctx::None,
};
let val = self.expr_ctx(val, ctx)?;
if self.size_of(val.ty) != size {
self.report(
expr.pos(),
format_args!(
"cannot bitcast {} to {} (different sizes: {} != {size})",
self.display_ty(val.ty),
self.display_ty(ty),
self.size_of(val.ty),
),
);
}
// TODO: maybe check align
return Some(Value { ty, loc: val.loc });
}
E::Directive {
name: "as",
args: [ty, val],
..
} => {
let ty = self.ty(ty);
let ctx = match ctx {
Ctx::Dest(dest) => Ctx::Dest(dest),
2024-05-19 11:20:42 -05:00
Ctx::DestUntyped(loc) => Ctx::Dest(Value { ty, loc }),
2024-05-14 16:07:32 -05:00
_ => Ctx::Inferred(ty),
};
return self.expr_ctx(val, ctx);
}
2024-05-12 13:10:50 -05:00
E::Bool { value, .. } => Some(Value {
ty: bt::BOOL,
loc: Loc::Imm(value as u64),
}),
2024-05-14 05:17:39 -05:00
E::Ctor {
pos, ty, fields, ..
} => {
2024-05-19 11:20:42 -05:00
let Some(ty) = ty.map(|ty| self.ty(ty)).or(ctx.ty()) else {
2024-05-14 05:17:39 -05:00
self.report(pos, "expected type, (it cannot be inferred)");
};
2024-05-12 13:10:50 -05:00
let size = self.size_of(ty);
2024-05-14 16:07:32 -05:00
let loc = match ctx.loc() {
Some(loc) => loc,
_ => Loc::Stack(self.alloc_stack(size), 0),
};
2024-05-14 05:17:39 -05:00
let stuct = self.unwrap_struct(ty, pos, "struct literal");
2024-05-20 07:11:58 -05:00
let field_count = self.structs[stuct as usize].fields.len();
2024-05-14 05:17:39 -05:00
if field_count != fields.len() {
self.report(
pos,
format_args!("expected {} fields, got {}", field_count, fields.len()),
);
}
for (i, (name, field)) in fields.iter().enumerate() {
let (offset, ty) = self.offset_of(field.pos(), ty, name.ok_or(i));
let loc = loc.offset_ref(offset);
self.expr_ctx(field, Ctx::Dest(Value { ty, loc }))?;
2024-05-12 06:13:36 -05:00
}
return Some(Value { ty, loc });
2024-05-12 06:13:36 -05:00
}
E::Field { target, field } => {
let mut tal = self.expr(target)?;
2024-05-12 16:19:45 -05:00
if let TypeKind::Pointer(ty) = TypeKind::from_ty(tal.ty) {
tal.ty = self.pointers[ty as usize];
tal.loc = match tal.loc {
Loc::Reg(r) => Loc::Deref(r, None, 0),
Loc::RegRef(r) => Loc::DerefRef(r, None, 0),
l => {
let ptr = self.loc_to_reg(l, 8);
Loc::Deref(ptr, None, 0)
}
2024-05-12 15:40:28 -05:00
};
}
2024-05-14 05:17:39 -05:00
let (offset, ty) = self.offset_of(target.pos(), tal.ty, Ok(field));
let loc = tal.loc.offset(offset);
2024-05-12 15:40:28 -05:00
Some(Value { ty, loc })
2024-05-12 06:13:36 -05:00
}
2024-05-12 04:52:58 -05:00
E::UnOp {
2024-05-15 03:37:39 -05:00
op: T::Band,
2024-05-12 16:19:45 -05:00
val,
pos,
2024-05-12 04:52:58 -05:00
} => {
let val = self.expr(val)?;
let loc = match val.loc {
Loc::Deref(r, stack, off) => {
2024-05-19 11:20:42 -05:00
if let Some(stack) = stack {
stack.leak()
}
2024-05-13 06:36:29 -05:00
self.code.addi64(r.0, r.0, off);
Loc::Reg(r)
}
Loc::DerefRef(r, stack, off) => {
2024-05-19 11:20:42 -05:00
if let Some(stack) = stack {
stack.leak()
}
let reg = self.alloc_reg();
2024-05-13 06:36:29 -05:00
self.code.addi64(reg.0, r, off);
Loc::Reg(reg)
2024-05-12 04:52:58 -05:00
}
2024-05-14 05:17:39 -05:00
Loc::Stack(stack, off) => {
stack.leak();
let reg = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.code
.encode(i::addi64(reg.0, STACK_PTR, stack.offset + off));
2024-05-14 05:17:39 -05:00
Loc::Reg(reg)
}
2024-05-12 16:19:45 -05:00
l => self.report(
pos,
format_args!("cant take pointer of {} ({:?})", self.display_ty(val.ty), l),
),
};
Some(Value {
ty: self.alloc_pointer(val.ty),
loc,
})
2024-05-12 04:52:58 -05:00
}
E::UnOp {
2024-05-15 03:37:39 -05:00
op: T::Mul,
2024-05-12 16:19:45 -05:00
val,
pos,
2024-05-12 04:52:58 -05:00
} => {
let val = self.expr(val)?;
2024-05-12 04:52:58 -05:00
match TypeKind::from_ty(val.ty) {
TypeKind::Pointer(ty) => Some(Value {
ty: self.pointers[ty as usize],
loc: Loc::Deref(self.loc_to_reg(val.loc, self.size_of(val.ty)), None, 0),
2024-05-12 04:52:58 -05:00
}),
2024-05-12 16:19:45 -05:00
_ => self.report(
pos,
format_args!("expected pointer, got {}", self.display_ty(val.ty)),
),
2024-05-12 04:52:58 -05:00
}
}
2024-05-11 15:22:08 -05:00
E::BinOp {
2024-05-17 12:53:59 -05:00
left: E::Ident { id, .. },
2024-05-11 15:22:08 -05:00
op: T::Decl,
right,
} => {
let val = self.expr(right)?;
2024-05-12 06:13:36 -05:00
let loc = self.make_loc_owned(val.loc, val.ty);
2024-05-19 11:20:42 -05:00
let sym = parser::find_symbol(&self.cf.symbols, *id);
2024-05-14 05:17:39 -05:00
let loc = match loc {
2024-05-20 07:11:58 -05:00
Loc::Reg(r) if sym.flags & idfl::REFERENCED != 0 => {
2024-05-14 05:17:39 -05:00
let size = self.size_of(val.ty);
let stack = self.alloc_stack(size);
2024-05-14 05:17:39 -05:00
self.store_stack(r.0, stack.offset, size as _);
Loc::Stack(stack, 0)
}
l => l,
};
2024-05-12 04:52:58 -05:00
self.vars.push(Variable {
id: *id,
2024-05-12 06:13:36 -05:00
value: Value { ty: val.ty, loc },
2024-05-12 04:52:58 -05:00
});
Some(Value::VOID)
2024-05-11 15:22:08 -05:00
}
2024-05-20 07:11:58 -05:00
E::Call { func, args } => {
let func = self.ty(func);
let TypeKind::Func(func) = TypeKind::from_ty(func) else {
todo!()
2024-05-19 11:20:42 -05:00
};
2024-05-20 07:11:58 -05:00
let fn_label = self.funcs[func as usize].clone();
2024-05-14 05:17:39 -05:00
let mut parama = self.param_alloc(fn_label.ret);
2024-05-14 05:17:39 -05:00
let mut values = Vec::with_capacity(args.len());
2024-05-12 16:45:28 -05:00
for (earg, &ty) in args.iter().zip(fn_label.args.iter()) {
let arg = self.expr_ctx(earg, Ctx::Inferred(ty))?;
2024-05-13 06:36:29 -05:00
_ = self.assert_ty(earg.pos(), ty, arg.ty);
2024-05-14 05:17:39 -05:00
self.pass_arg(&arg, &mut parama);
values.push(arg.loc);
}
drop(values);
2024-05-13 02:38:33 -05:00
2024-05-14 16:07:32 -05:00
let loc = self.alloc_ret_loc(fn_label.ret, ctx);
2024-05-13 02:38:33 -05:00
2024-05-11 09:04:13 -05:00
self.code.call(func);
2024-05-13 02:38:33 -05:00
2024-05-14 16:07:32 -05:00
self.post_process_ret_loc(fn_label.ret, &loc);
2024-05-13 02:38:33 -05:00
return Some(Value {
2024-05-13 02:38:33 -05:00
ty: fn_label.ret,
loc,
});
2024-05-11 09:04:13 -05:00
}
2024-05-20 07:11:58 -05:00
E::Ident { id, .. } if ident::is_null(id) => Some(Value::ty(id)),
2024-05-19 11:20:42 -05:00
E::Ident { id, index, .. }
if let Some((var_index, var)) =
self.vars.iter_mut().enumerate().find(|(_, v)| v.id == id) =>
{
let sym = parser::find_symbol(&self.cf.symbols, id);
2024-05-17 12:53:59 -05:00
2024-05-20 07:11:58 -05:00
let loc = match idfl::index(sym.flags) == index
&& !self.loops.last().is_some_and(|l| l.var_count > var_index)
2024-05-14 07:01:40 -05:00
{
2024-05-14 05:17:39 -05:00
true => std::mem::replace(&mut var.value.loc, Loc::Imm(0)),
false => var.value.loc.take_ref(),
};
Some(Value {
2024-05-14 05:17:39 -05:00
ty: var.value.ty,
loc,
})
}
2024-05-20 07:11:58 -05:00
E::Ident { id, .. } => match self
.symbols
.get(&SymKey { id, file: 0 })
.copied()
.map(TypeKind::from_ty)
.unwrap_or_else(|| self.find_and_declare(0, Ok(id)))
{
TypeKind::Global(id) => self.handle_global(id),
tk => Some(Value::ty(tk.encode())),
},
2024-05-14 05:17:39 -05:00
E::Return { val, .. } => {
2024-05-10 14:33:42 -05:00
if let Some(val) = val {
2024-05-14 05:17:39 -05:00
let size = self.size_of(self.ret);
let loc = match size {
0 => Loc::Imm(0),
..=16 => Loc::RegRef(1),
_ => Loc::DerefRef(1, None, 0),
2024-05-14 05:17:39 -05:00
};
self.expr_ctx(val, Ctx::Dest(Value { loc, ty: self.ret }))?;
2024-05-10 14:33:42 -05:00
}
2024-05-11 09:04:13 -05:00
self.ret_relocs.push(RetReloc {
offset: self.code.code.len() as u32,
instr_offset: 1,
size: 4,
});
self.code.encode(i::jmp(0));
2024-05-10 14:33:42 -05:00
None
2024-05-09 11:16:01 -05:00
}
2024-05-11 15:22:08 -05:00
E::Block { stmts, .. } => {
2024-05-09 16:41:59 -05:00
for stmt in stmts {
self.expr(stmt)?;
2024-05-09 11:16:01 -05:00
}
Some(Value::VOID)
2024-05-09 11:16:01 -05:00
}
2024-05-11 15:22:08 -05:00
E::Number { value, .. } => Some(Value {
ty: ctx.ty().map(bt::strip_pointer).unwrap_or(bt::INT),
2024-05-10 14:33:42 -05:00
loc: Loc::Imm(value),
}),
2024-05-11 15:22:08 -05:00
E::If {
cond, then, else_, ..
} => 'b: {
2024-05-12 04:52:58 -05:00
log::dbg!("if-cond");
let cond = self.expr_ctx(cond, Ctx::Inferred(bt::BOOL))?;
2024-05-13 06:36:29 -05:00
let reg = self.loc_to_reg(cond.loc, 1);
let jump_offset = self.code.code.len() as u32;
self.code.encode(i::jeq(reg.0, 0, 0));
2024-05-12 04:52:58 -05:00
log::dbg!("if-then");
let then_unreachable = self.expr(then).is_none();
let mut else_unreachable = false;
2024-05-11 11:16:27 -05:00
2024-05-20 07:11:58 -05:00
let mut jump = self.code.code.len() as i64 - jump_offset as i64;
2024-05-11 11:16:27 -05:00
if let Some(else_) = else_ {
2024-05-12 04:52:58 -05:00
log::dbg!("if-else");
2024-05-11 11:16:27 -05:00
let else_jump_offset = self.code.code.len() as u32;
if !then_unreachable {
self.code.encode(i::jmp(0));
2024-05-20 07:11:58 -05:00
jump = self.code.code.len() as i64 - jump_offset as i64;
}
2024-05-11 11:16:27 -05:00
else_unreachable = self.expr(else_).is_none();
2024-05-11 11:16:27 -05:00
if !then_unreachable {
2024-05-20 07:11:58 -05:00
let jump = self.code.code.len() as i64 - else_jump_offset as i64;
log::dbg!("if-else-jump: {}", jump);
2024-05-20 07:11:58 -05:00
write_reloc(&mut self.code.code, else_jump_offset as usize + 1, jump, 4);
}
2024-05-11 11:16:27 -05:00
}
2024-05-12 04:52:58 -05:00
log::dbg!("if-then-jump: {}", jump);
2024-05-20 07:11:58 -05:00
write_reloc(&mut self.code.code, jump_offset as usize + 3, jump, 2);
if then_unreachable && else_unreachable {
break 'b None;
}
Some(Value::VOID)
}
E::Loop { body, .. } => 'a: {
2024-05-12 04:52:58 -05:00
log::dbg!("loop");
2024-05-11 11:16:27 -05:00
let loop_start = self.code.code.len() as u32;
self.loops.push(Loop {
2024-05-14 07:01:40 -05:00
var_count: self.vars.len() as _,
offset: loop_start,
relocs: Default::default(),
2024-05-11 11:16:27 -05:00
});
let body_unreachable = self.expr(body).is_none();
2024-05-11 11:16:27 -05:00
2024-05-12 04:52:58 -05:00
log::dbg!("loop-end");
if !body_unreachable {
let loop_end = self.code.code.len();
self.code
.encode(i::jmp(loop_start as i32 - loop_end as i32));
}
2024-05-11 11:16:27 -05:00
let loop_end = self.code.code.len() as u32;
let loop_ = self.loops.pop().unwrap();
let is_unreachable = loop_.relocs.is_empty();
2024-05-11 11:16:27 -05:00
for reloc in loop_.relocs {
let dest = &mut self.code.code
[reloc.offset as usize + reloc.instr_offset as usize..]
[..reloc.size as usize];
let offset = loop_end as i32 - reloc.offset as i32;
dest.copy_from_slice(&offset.to_ne_bytes());
}
2024-05-19 11:20:42 -05:00
self.vars.drain(loop_.var_count..);
2024-05-14 07:01:40 -05:00
if is_unreachable {
log::dbg!("infinite loop");
break 'a None;
}
Some(Value::VOID)
2024-05-11 11:16:27 -05:00
}
2024-05-11 15:22:08 -05:00
E::Break { .. } => {
2024-05-11 11:16:27 -05:00
let loop_ = self.loops.last_mut().unwrap();
let offset = self.code.code.len() as u32;
self.code.encode(i::jmp(0));
2024-05-11 11:16:27 -05:00
loop_.relocs.push(RetReloc {
offset,
instr_offset: 1,
size: 4,
});
None
}
2024-05-11 15:22:08 -05:00
E::Continue { .. } => {
2024-05-11 11:16:27 -05:00
let loop_ = self.loops.last().unwrap();
let offset = self.code.code.len() as u32;
self.code
.encode(i::jmp(loop_.offset as i32 - offset as i32));
2024-05-11 11:16:27 -05:00
None
}
E::BinOp {
left,
op: op @ (T::And | T::Or),
right,
} => {
let lhs = self.expr_ctx(left, Ctx::Inferred(bt::BOOL))?;
let lhs = self.loc_to_reg(lhs.loc, 1);
2024-05-20 07:11:58 -05:00
let jump_offset = self.code.code.len() + 3;
let op = if op == T::And { i::jeq } else { i::jne };
self.code.encode(op(lhs.0, 0, 0));
if let Some(rhs) = self.expr_ctx(right, Ctx::Inferred(bt::BOOL)) {
let rhs = self.loc_to_reg(rhs.loc, 1);
self.code.encode(i::cp(lhs.0, rhs.0));
}
2024-05-13 06:36:29 -05:00
2024-05-20 07:11:58 -05:00
let jump = self.code.code.len() as i64 - jump_offset as i64;
write_reloc(&mut self.code.code, jump_offset, jump, 2);
Some(Value {
ty: bt::BOOL,
loc: Loc::Reg(lhs),
})
}
E::BinOp { left, op, right } => 'ops: {
let left = self.expr(left)?;
2024-05-12 04:52:58 -05:00
if op == T::Assign {
self.expr_ctx(right, Ctx::Dest(left)).unwrap();
2024-05-13 06:36:29 -05:00
return Some(Value::VOID);
2024-05-12 04:52:58 -05:00
}
2024-05-10 15:54:12 -05:00
2024-05-14 05:17:39 -05:00
if let TypeKind::Struct(_) = TypeKind::from_ty(left.ty) {
let right = self.expr_ctx(right, Ctx::Inferred(left.ty))?;
_ = self.assert_ty(expr.pos(), left.ty, right.ty);
return self.struct_op(op, left.ty, ctx, left.loc, right.loc);
}
2024-05-13 06:36:29 -05:00
let lsize = self.size_of(left.ty);
2024-05-16 05:56:33 -05:00
let ty = ctx.ty().unwrap_or(left.ty);
2024-05-16 06:29:16 -05:00
2024-05-19 11:20:42 -05:00
let (lhs, loc) = match std::mem::take(&mut ctx).loc() {
Some(Loc::RegRef(reg)) if Loc::RegRef(reg) == left.loc && reg != 1 => {
(reg, Loc::RegRef(reg))
}
2024-05-16 05:56:33 -05:00
Some(loc) => {
2024-05-19 11:20:42 -05:00
debug_assert!(!matches!(loc, Loc::Reg(LinReg(RET_ADDR, ..))));
2024-05-16 05:56:33 -05:00
ctx = Ctx::Dest(Value { ty, loc });
2024-05-19 11:20:42 -05:00
let reg = self.loc_to_reg(left.loc, lsize);
(reg.0, Loc::Reg(reg))
}
None => {
let reg = self.loc_to_reg(left.loc, lsize);
(reg.0, Loc::Reg(reg))
2024-05-16 05:56:33 -05:00
}
};
let right = self.expr_ctx(right, Ctx::Inferred(left.ty))?;
2024-05-13 06:36:29 -05:00
let rsize = self.size_of(right.ty);
let ty = self.assert_ty(expr.pos(), left.ty, right.ty);
let size = self.size_of(ty);
let signed = bt::is_signed(ty);
2024-05-13 07:23:19 -05:00
2024-05-16 06:29:16 -05:00
if let Loc::Imm(mut imm) = right.loc
2024-05-16 06:32:04 -05:00
&& let Some(oper) = Self::imm_math_op(op, signed, size)
2024-05-16 06:29:16 -05:00
{
2024-05-16 06:32:04 -05:00
if matches!(op, T::Add | T::Sub)
&& let TypeKind::Pointer(ty) = TypeKind::from_ty(ty)
{
let size = self.size_of(self.pointers[ty as usize]);
imm *= size;
2024-05-16 06:29:16 -05:00
}
2024-05-19 11:20:42 -05:00
self.code.encode(oper(lhs, lhs, imm));
break 'ops Some(Value { ty, loc });
2024-05-16 06:29:16 -05:00
}
let rhs = self.loc_to_reg(right.loc, rsize);
2024-05-15 03:37:39 -05:00
if matches!(op, T::Add | T::Sub) {
let min_size = lsize.min(rsize);
if bt::is_signed(ty) && min_size < size {
2024-05-19 11:20:42 -05:00
let operand = if lsize < rsize { lhs } else { rhs.0 };
let op = [i::sxt8, i::sxt16, i::sxt32][min_size.ilog2() as usize];
self.code.encode(op(operand, operand));
}
if bt::is_pointer(left.ty) ^ bt::is_pointer(right.ty) {
let (offset, ty) = if bt::is_pointer(left.ty) {
(rhs.0, left.ty)
} else {
2024-05-19 11:20:42 -05:00
(lhs, right.ty)
};
let TypeKind::Pointer(ty) = TypeKind::from_ty(ty) else {
unreachable!()
};
let size = self.size_of(self.pointers[ty as usize]);
self.code.encode(i::muli64(offset, offset, size as _));
}
2024-05-13 07:23:19 -05:00
}
2024-05-14 05:17:39 -05:00
if let Some(op) = Self::math_op(op, signed, size) {
2024-05-19 11:20:42 -05:00
self.code.encode(op(lhs, lhs, rhs.0));
break 'ops Some(Value { ty, loc });
2024-05-13 07:23:19 -05:00
}
2024-05-13 06:36:29 -05:00
'cmp: {
let against = match op {
2024-05-19 11:20:42 -05:00
T::Le | T::Gt => 1,
T::Ne | T::Eq => 0,
2024-05-19 11:20:42 -05:00
T::Ge | T::Lt => (-1i64) as _,
_ => break 'cmp,
};
let op_fn = if signed { i::cmps } else { i::cmpu };
2024-05-19 11:20:42 -05:00
self.code.encode(op_fn(lhs, lhs, rhs.0));
self.code.encode(i::cmpui(lhs, lhs, against));
if matches!(op, T::Eq | T::Lt | T::Gt) {
2024-05-19 11:20:42 -05:00
self.code.encode(i::not(lhs, lhs));
2024-05-11 11:16:27 -05:00
}
2024-05-10 15:54:12 -05:00
2024-05-19 11:20:42 -05:00
break 'ops Some(Value { ty: bt::BOOL, loc });
}
2024-05-10 15:54:12 -05:00
unimplemented!("{:#?}", op)
2024-05-10 15:54:12 -05:00
}
ast => unimplemented!("{:#?}", ast),
}?;
2024-05-14 16:07:32 -05:00
match ctx {
Ctx::Dest(dest) => {
_ = self.assert_ty(expr.pos(), dest.ty, value.ty);
self.assign(dest.ty, dest.loc, value.loc)?;
2024-05-19 11:20:42 -05:00
Some(Value {
ty: dest.ty,
loc: Loc::Imm(0),
})
2024-05-14 16:07:32 -05:00
}
2024-05-19 11:20:42 -05:00
Ctx::DestUntyped(loc) => {
2024-05-14 16:07:32 -05:00
// Wo dont check since bitcast does
2024-05-19 11:20:42 -05:00
self.assign(value.ty, loc, value.loc);
Some(Value {
ty: value.ty,
loc: Loc::Imm(0),
})
2024-05-14 16:07:32 -05:00
}
_ => Some(value),
2024-05-09 11:16:01 -05:00
}
}
2024-05-14 05:17:39 -05:00
fn math_op(
op: T,
signed: bool,
size: u64,
) -> Option<fn(u8, u8, u8) -> (usize, [u8; instrs::MAX_SIZE])> {
use instrs as i;
macro_rules! div { ($($op:ident),*) => {[$(|a, b, c| i::$op(a, ZERO, b, c)),*]}; }
macro_rules! rem { ($($op:ident),*) => {[$(|a, b, c| i::$op(ZERO, a, b, c)),*]}; }
let ops = match op {
T::Add => [i::add8, i::add16, i::add32, i::add64],
T::Sub => [i::sub8, i::sub16, i::sub32, i::sub64],
T::Mul => [i::mul8, i::mul16, i::mul32, i::mul64],
T::Div if signed => div!(dirs8, dirs16, dirs32, dirs64),
T::Div => div!(diru8, diru16, diru32, diru64),
T::Mod if signed => rem!(dirs8, dirs16, dirs32, dirs64),
T::Mod => rem!(diru8, diru16, diru32, diru64),
T::Band => return Some(i::and),
T::Bor => return Some(i::or),
T::Xor => return Some(i::xor),
T::Shl => [i::slu8, i::slu16, i::slu32, i::slu64],
T::Shr if signed => [i::srs8, i::srs16, i::srs32, i::srs64],
T::Shr => [i::sru8, i::sru16, i::sru32, i::sru64],
_ => return None,
};
Some(ops[size.ilog2() as usize])
2024-05-14 05:17:39 -05:00
}
2024-05-16 06:29:16 -05:00
fn imm_math_op(
op: T,
signed: bool,
size: u64,
) -> Option<fn(u8, u8, u64) -> (usize, [u8; instrs::MAX_SIZE])> {
use instrs as i;
macro_rules! def_op {
($name:ident |$a:ident, $b:ident, $c:ident| $($tt:tt)*) => {
macro_rules! $name {
($$($$op:ident),*) => {
[$$(
|$a, $b, $c: u64| i::$$op($($tt)*),
)*]
}
}
};
}
def_op!(basic_op | a, b, c | a, b, c as _);
def_op!(sub_op | a, b, c | b, a, c.wrapping_neg() as _);
let ops = match op {
T::Add => basic_op!(addi8, addi16, addi32, addi64),
2024-05-16 06:32:04 -05:00
T::Sub => sub_op!(addi8, addi16, addi32, addi64),
2024-05-16 06:29:16 -05:00
T::Mul => basic_op!(muli8, muli16, muli32, muli64),
T::Band => return Some(i::andi),
T::Bor => return Some(i::ori),
T::Xor => return Some(i::xori),
2024-05-16 06:32:04 -05:00
T::Shr if signed => basic_op!(srui8, srui16, srui32, srui64),
T::Shr => basic_op!(srui8, srui16, srui32, srui64),
T::Shl => basic_op!(slui8, slui16, slui32, slui64),
2024-05-16 06:29:16 -05:00
_ => return None,
};
Some(ops[size.ilog2() as usize])
}
2024-05-14 05:17:39 -05:00
fn struct_op(&mut self, op: T, ty: Type, ctx: Ctx, left: Loc, right: Loc) -> Option<Value> {
if let TypeKind::Struct(stuct) = TypeKind::from_ty(ty) {
let dst = match ctx {
Ctx::Dest(dest) => dest.loc,
_ => Loc::Stack(self.alloc_stack(self.size_of(ty)), 0),
2024-05-14 05:17:39 -05:00
};
let mut offset = 0;
2024-05-20 07:11:58 -05:00
for &(_, ty) in self.structs[stuct as usize].fields.clone().iter() {
2024-05-14 05:17:39 -05:00
let align = self.align_of(ty);
offset = align_up(offset, align);
let size = self.size_of(ty);
let ctx = Ctx::Dest(Value::new(ty, dst.offset_ref(offset)));
let left = left.offset_ref(offset);
let right = right.offset_ref(offset);
self.struct_op(op, ty, ctx, left, right)?;
2024-05-14 05:17:39 -05:00
offset += size;
}
return Some(Value { ty, loc: dst });
}
let size = self.size_of(ty);
2024-05-16 06:29:16 -05:00
let signed = bt::is_signed(ty);
2024-05-14 05:17:39 -05:00
let (lhs, owned) = self.loc_to_reg_ref(&left, size);
2024-05-16 06:29:16 -05:00
if let Loc::Imm(imm) = right
2024-05-16 06:32:04 -05:00
&& let Some(op) = Self::imm_math_op(op, signed, size)
2024-05-16 06:29:16 -05:00
{
self.code.encode(op(lhs, lhs, imm));
return if let Ctx::Dest(dest) = ctx {
self.assign(dest.ty, dest.loc, owned.map_or(Loc::RegRef(lhs), Loc::Reg));
Some(Value::VOID)
} else {
Some(Value::new(ty, owned.map_or(Loc::RegRef(lhs), Loc::Reg)))
};
}
2024-05-14 05:17:39 -05:00
let rhs = self.loc_to_reg(right, size);
if let Some(op) = Self::math_op(op, signed, size) {
self.code.encode(op(lhs, lhs, rhs.0));
return if let Ctx::Dest(dest) = ctx {
self.assign(dest.ty, dest.loc, owned.map_or(Loc::RegRef(lhs), Loc::Reg));
Some(Value::VOID)
} else {
Some(Value::new(ty, owned.map_or(Loc::RegRef(lhs), Loc::Reg)))
};
}
unimplemented!("{:#?}", op)
}
fn loc_to_reg_ref(&mut self, loc: &Loc, size: u64) -> (u8, Option<LinReg>) {
match *loc {
Loc::RegRef(reg) => (reg, None),
Loc::Reg(LinReg(reg, ..)) => (reg, None),
Loc::Deref(LinReg(reg, ..), .., off) | Loc::DerefRef(reg, .., off) => {
let new = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.code.encode(instrs::ld(new.0, reg, off, size as _));
(new.0, Some(new))
}
Loc::Stack(ref stack, off) => {
let new = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.load_stack(new.0, stack.offset + off, size as _);
(new.0, Some(new))
}
Loc::Imm(imm) => {
let new = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.code.encode(instrs::li64(new.0, imm));
(new.0, Some(new))
}
}
}
2024-05-14 16:07:32 -05:00
fn assign_opaque(&mut self, size: u64, right: Loc, left: Loc) -> Option<Value> {
2024-05-13 06:36:29 -05:00
if left == right {
return Some(Value::VOID);
}
2024-05-12 13:10:50 -05:00
match size {
2024-05-13 06:36:29 -05:00
0 => {}
..=8 => {
let lhs = self.loc_to_reg(left, size);
match right {
2024-05-13 06:36:29 -05:00
Loc::RegRef(reg) => self.code.encode(instrs::cp(reg, lhs.0)),
Loc::Deref(reg, .., off) => {
2024-05-13 06:36:29 -05:00
self.code.encode(instrs::st(lhs.0, reg.0, off, size as _));
2024-05-12 13:10:50 -05:00
}
Loc::DerefRef(reg, .., off) => {
2024-05-13 06:36:29 -05:00
self.code.encode(instrs::st(lhs.0, reg, off, size as _));
}
2024-05-14 05:17:39 -05:00
Loc::Stack(stack, off) => {
self.store_stack(lhs.0, stack.offset + off, size as _);
2024-05-12 13:10:50 -05:00
}
l => unimplemented!("{:?}", l),
}
}
2024-05-14 05:17:39 -05:00
..=16 if matches!(right, Loc::RegRef(1)) => {
let (lhs, loff) = left.ref_to_ptr();
self.code.encode(instrs::st(1, lhs, loff, 16));
}
2024-05-13 06:36:29 -05:00
..=u64::MAX => {
2024-05-14 05:17:39 -05:00
let rhs = self.to_ptr(right);
let lhs = self.to_ptr(left);
2024-05-13 02:55:09 -05:00
self.code
.encode(instrs::bmc(lhs.0, rhs.0, size.try_into().unwrap()));
2024-05-12 04:52:58 -05:00
}
}
2024-05-12 13:10:50 -05:00
Some(Value::VOID)
}
2024-05-14 16:07:32 -05:00
fn assign(&mut self, ty: Type, right: Loc, left: Loc) -> Option<Value> {
self.assign_opaque(self.size_of(ty), right, left)
}
2024-05-14 05:17:39 -05:00
fn to_ptr(&mut self, loc: Loc) -> LinReg {
2024-05-12 13:10:50 -05:00
match loc {
Loc::Deref(reg, .., off) => {
2024-05-14 05:17:39 -05:00
self.code.addi64(reg.0, reg.0, off);
reg
}
Loc::DerefRef(reg, .., off) => {
let new = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.code.addi64(new.0, reg, off);
new
2024-05-12 13:10:50 -05:00
}
2024-05-14 05:17:39 -05:00
Loc::Stack(stack, off) => {
let reg = self.alloc_reg();
2024-05-14 05:17:39 -05:00
self.code.addi64(reg.0, STACK_PTR, stack.offset + off);
reg
}
l => unreachable!("{:?}", l),
}
}
2024-05-20 07:11:58 -05:00
fn find_and_declare(&mut self, file: parser::FileId, name: Result<Ident, &str>) -> TypeKind {
2024-05-19 11:20:42 -05:00
let f = self.files[file as usize].clone();
2024-05-20 07:11:58 -05:00
let (expr, id) = f.find_decl(name).expect("TODO: error");
let sym = match expr {
2024-05-19 11:20:42 -05:00
E::BinOp {
2024-05-20 07:11:58 -05:00
left: &E::Ident { .. },
2024-05-19 11:20:42 -05:00
op: T::Decl,
right: E::Closure { args, ret, .. },
} => {
let args = args.iter().map(|arg| self.ty(&arg.ty)).collect::<Vec<_>>();
let ret = self.ty(ret);
2024-05-20 07:11:58 -05:00
let id = self.declare_fn_label(args.into(), ret);
2024-05-19 11:20:42 -05:00
self.to_generate.push(ItemId {
file,
expr: ExprRef::new(expr),
2024-05-20 07:11:58 -05:00
id,
2024-05-19 11:20:42 -05:00
});
2024-05-20 07:11:58 -05:00
TypeKind::Func(id)
2024-05-19 11:20:42 -05:00
}
E::BinOp {
2024-05-20 07:11:58 -05:00
left: &E::Ident { .. },
2024-05-19 11:20:42 -05:00
op: T::Decl,
right: E::Struct { fields, .. },
} => {
let fields = fields
.iter()
.map(|&(name, ty)| (name.into(), self.ty(&ty)))
.collect();
2024-05-20 07:11:58 -05:00
self.structs.push(Struct { fields });
TypeKind::Struct(self.structs.len() as u32 - 1)
2024-05-19 11:20:42 -05:00
}
E::BinOp {
2024-05-20 07:11:58 -05:00
left: &E::Ident { .. },
2024-05-19 11:20:42 -05:00
op: T::Decl,
right,
} => {
let gid = self.globals.len() as GlobalId;
2024-05-20 07:11:58 -05:00
let prev_in_global = std::mem::replace(&mut self.cur_item, TypeKind::Global(gid));
2024-05-19 11:20:42 -05:00
let prev_gpa = std::mem::replace(&mut *self.gpa.borrow_mut(), Default::default());
let prev_sa = std::mem::replace(&mut *self.sa.borrow_mut(), Default::default());
let offset = self.code.code.len();
let reloc_count = self.code.relocs.len();
self.globals.push(Global {
2024-05-20 07:11:58 -05:00
ty: bt::UNDECLARED,
code: 0,
dep: 0,
2024-05-19 11:20:42 -05:00
offset: u32::MAX,
});
self.gpa.borrow_mut().init_callee();
let ret = self.gpa.borrow_mut().allocate();
// TODO: detect is constant does not call anything
self.code.encode(instrs::cp(ret, 1));
let ret = self
.expr_ctx(right, Ctx::DestUntyped(Loc::DerefRef(ret, None, 0)))
.expect("TODO: unreachable constant/global");
self.code.encode(instrs::tx());
self.globals[gid as usize].ty = ret.ty;
self.globals[gid as usize].code = self.data.code.len() as u32;
self.data.append(&mut self.code, offset, reloc_count);
*self.sa.borrow_mut() = prev_sa;
*self.gpa.borrow_mut() = prev_gpa;
2024-05-20 07:11:58 -05:00
self.cur_item = prev_in_global;
2024-05-19 11:20:42 -05:00
2024-05-20 07:11:58 -05:00
TypeKind::Global(gid)
2024-05-19 11:20:42 -05:00
}
e => unimplemented!("{e:#?}"),
2024-05-20 07:11:58 -05:00
};
self.symbols.insert(SymKey { id, file }, sym.encode());
sym
2024-05-19 11:20:42 -05:00
}
2024-05-20 07:11:58 -05:00
fn declare_fn_label(&mut self, args: Rc<[Type]>, ret: Type) -> FuncId {
self.funcs.push(Func {
2024-05-12 16:45:28 -05:00
offset: 0,
2024-05-20 07:11:58 -05:00
relocs: 0,
2024-05-12 16:45:28 -05:00
args,
ret,
});
2024-05-20 07:11:58 -05:00
self.funcs.len() as u32 - 1
2024-05-09 11:16:01 -05:00
}
2024-05-20 07:11:58 -05:00
fn gen_prelude(&mut self) {
self.code.encode(instrs::addi64(STACK_PTR, STACK_PTR, 0));
self.code.encode(instrs::st(RET_ADDR, STACK_PTR, 0, 0));
2024-05-09 11:16:01 -05:00
}
2024-05-20 07:11:58 -05:00
fn reloc_prelude(&mut self, id: FuncId) {
let mut cursor = self.funcs[id as usize].offset as usize;
let mut allocate = |size| (cursor += size, cursor).1;
2024-05-20 07:11:58 -05:00
let pushed = self.gpa.borrow().pushed_size() as i64;
let stack = self.sa.borrow().height as i64;
2024-05-20 07:11:58 -05:00
write_reloc(&mut self.code.code, allocate(3), -(pushed + stack), 8);
write_reloc(&mut self.code.code, allocate(8 + 3), stack, 8);
write_reloc(&mut self.code.code, allocate(8), pushed, 2);
2024-05-10 14:33:42 -05:00
}
2024-05-10 14:33:42 -05:00
fn ret(&mut self) {
2024-05-20 07:11:58 -05:00
let pushed = self.gpa.borrow().pushed_size() as u64;
let stack = self.sa.borrow().height as u64;
self.code
.encode(instrs::ld(RET_ADDR, STACK_PTR, stack, pushed as _));
self.code
.encode(instrs::addi64(STACK_PTR, STACK_PTR, stack + pushed));
2024-05-10 14:33:42 -05:00
self.code.ret();
}
2024-05-20 07:11:58 -05:00
fn link(&mut self) {
2024-05-19 11:20:42 -05:00
let mut globals = std::mem::take(&mut self.globals);
2024-05-20 07:11:58 -05:00
for global in globals.iter_mut().skip(self.linked.globals as usize) {
2024-05-19 11:20:42 -05:00
let size = self.size_of(global.ty);
global.offset = self.code.code.len() as u32;
self.code.code.extend(std::iter::repeat(0).take(size as _));
}
self.globals = globals;
let prev_len = self.code.code.len();
self.code.append(&mut self.data, 0, 0);
2024-05-20 07:11:58 -05:00
self.code
.relocate(&self.funcs, &self.globals, 0, self.linked.relocs);
2024-05-19 11:20:42 -05:00
2024-05-20 07:11:58 -05:00
let mut var_order = self
.globals
.iter()
.map(|g| g.dep)
.zip(0u32..)
.skip(self.linked.globals as usize)
.collect::<Vec<_>>();
var_order.sort_unstable();
for (_, glob_id) in var_order.into_iter().rev() {
let global = &self.globals[glob_id as usize];
self.vm.pc = hbvm::mem::Address::new(
&mut self.code.code[global.code as usize + prev_len] as *mut _ as u64,
);
self.vm.write_reg(
1,
&mut self.code.code[global.offset as usize] as *mut _ as u64,
);
self.vm.run().unwrap();
2024-05-19 11:20:42 -05:00
}
self.code.code.truncate(prev_len);
2024-05-10 14:33:42 -05:00
2024-05-20 07:11:58 -05:00
self.linked.globals = self.globals.len();
self.linked.relocs = self.code.relocs.len();
}
pub fn dump(mut self, out: &mut impl std::io::Write) -> std::io::Result<()> {
self.code.relocs.push(Reloc {
offset: 0,
size: 4,
instr_offset: 3,
id: TypeKind::Func(0).encode() as _,
});
self.link();
2024-05-10 14:33:42 -05:00
out.write_all(&self.code.code)
}
2024-05-12 04:52:58 -05:00
fn alloc_pointer(&mut self, ty: Type) -> Type {
let ty = self
.pointers
.iter()
.position(|&p| p == ty)
.unwrap_or_else(|| {
self.pointers.push(ty);
self.pointers.len() - 1
});
TypeKind::Pointer(ty as Type).encode()
}
2024-05-12 06:13:36 -05:00
fn make_loc_owned(&mut self, loc: Loc, ty: Type) -> Loc {
match loc {
Loc::RegRef(rreg) => {
let reg = self.alloc_reg();
2024-05-12 06:13:36 -05:00
self.code.encode(instrs::cp(reg.0, rreg));
Loc::Reg(reg)
}
Loc::Imm(imm) => {
let reg = self.alloc_reg();
2024-05-12 06:13:36 -05:00
self.code.encode(instrs::li64(reg.0, imm));
Loc::Reg(reg)
}
2024-05-14 05:17:39 -05:00
l @ (Loc::DerefRef(..) | Loc::Deref(..)) => {
2024-05-12 06:13:36 -05:00
let size = self.size_of(ty);
let stack = self.alloc_stack(size);
self.assign(ty, Loc::DerefRef(STACK_PTR, None, stack.offset), l);
2024-05-14 05:17:39 -05:00
Loc::Stack(stack, 0)
2024-05-12 06:13:36 -05:00
}
l => l,
}
}
2024-05-14 05:17:39 -05:00
fn pass_arg(&mut self, value: &Value, parama: &mut Range<u8>) {
2024-05-12 13:10:50 -05:00
let size = self.size_of(value.ty);
let p = parama.next().unwrap() as Reg;
if size > 16 {
2024-05-14 05:17:39 -05:00
let (ptr, off) = value.loc.ref_to_ptr();
self.code.addi64(p, ptr, off);
2024-05-13 02:38:33 -05:00
return;
2024-05-12 13:10:50 -05:00
}
match value.loc {
2024-05-19 11:20:42 -05:00
Loc::Reg(LinReg(reg, ..)) | Loc::RegRef(reg) => self.code.encode(instrs::cp(p, reg)),
Loc::Deref(LinReg(reg, ..), .., off) | Loc::DerefRef(reg, .., off) => {
2024-05-12 15:40:28 -05:00
self.code.encode(instrs::ld(p, reg, off, size as _));
2024-05-12 13:10:50 -05:00
}
2024-05-19 11:20:42 -05:00
Loc::Imm(imm) => self.code.encode(instrs::li64(p, imm)),
2024-05-14 05:17:39 -05:00
Loc::Stack(ref stack, off) => {
self.load_stack(p, stack.offset + off, size as _);
2024-05-19 11:20:42 -05:00
if size > 8 {
parama.next().unwrap();
}
2024-05-12 13:10:50 -05:00
}
}
}
fn load_arg(&mut self, flags: parser::IdentFlags, ty: Type, parama: &mut Range<u8>) -> Loc {
2024-05-12 13:10:50 -05:00
let size = self.size_of(ty);
match size {
2024-05-13 06:36:29 -05:00
0 => Loc::Imm(0),
2024-05-20 07:11:58 -05:00
..=8 if flags & idfl::REFERENCED == 0 => {
2024-05-16 05:42:11 -05:00
let reg = self.alloc_reg();
self.code.encode(instrs::cp(reg.0, parama.next().unwrap()));
Loc::Reg(reg)
}
2024-05-13 06:36:29 -05:00
..=8 => {
let stack = self.alloc_stack(size as _);
2024-05-14 05:17:39 -05:00
self.store_stack(parama.next().unwrap(), stack.offset, size as _);
Loc::Stack(stack, 0)
2024-05-12 13:10:50 -05:00
}
2024-05-13 06:36:29 -05:00
..=16 => {
let stack = self.alloc_stack(size);
2024-05-14 05:17:39 -05:00
self.store_stack(parama.next().unwrap(), stack.offset, size as _);
2024-05-13 06:36:29 -05:00
parama.next().unwrap();
2024-05-14 05:17:39 -05:00
Loc::Stack(stack, 0)
2024-05-12 13:10:50 -05:00
}
2024-05-20 07:11:58 -05:00
_ if flags & (idfl::MUTABLE | idfl::REFERENCED) == 0 => {
let ptr = parama.next().unwrap();
let reg = self.alloc_reg();
self.code.encode(instrs::cp(reg.0, ptr));
Loc::Deref(reg, None, 0)
}
_ => {
2024-05-12 13:10:50 -05:00
let ptr = parama.next().unwrap();
let stack = self.alloc_stack(size);
2024-05-14 05:17:39 -05:00
self.assign(
ty,
Loc::DerefRef(STACK_PTR, None, stack.offset),
Loc::DerefRef(ptr, None, 0),
2024-05-14 05:17:39 -05:00
);
Loc::Stack(stack, 0)
2024-05-12 06:13:36 -05:00
}
}
}
2024-05-12 16:19:45 -05:00
2024-05-13 06:36:29 -05:00
#[must_use]
fn assert_ty(&self, pos: parser::Pos, ty: Type, expected: Type) -> Type {
if let Some(res) = bt::try_upcast(ty, expected) {
res
} else {
2024-05-12 16:45:28 -05:00
let ty = self.display_ty(ty);
let expected = self.display_ty(expected);
self.report(pos, format_args!("expected {ty}, got {expected}"));
}
}
2024-05-12 16:19:45 -05:00
fn report(&self, pos: parser::Pos, msg: impl std::fmt::Display) -> ! {
2024-05-19 11:20:42 -05:00
let (line, col) = self.cf.nlines.line_col(pos);
println!("{}:{}:{}: {}", self.cf.path, line, col, msg);
2024-05-12 16:19:45 -05:00
unreachable!();
}
2024-05-14 16:07:32 -05:00
fn alloc_ret_loc(&mut self, ret: Type, ctx: Ctx) -> Loc {
let size = self.size_of(ret);
match size {
0 => Loc::Imm(0),
..=8 => Loc::RegRef(1),
..=16 => match ctx {
Ctx::Dest(dest) => dest.loc,
_ => Loc::Stack(self.alloc_stack(size), 0),
},
..=u64::MAX => {
let val = match ctx {
Ctx::Dest(dest) => dest.loc,
_ => Loc::Stack(self.alloc_stack(size), 0),
};
let (ptr, off) = val.ref_to_ptr();
self.code.encode(instrs::cp(1, ptr));
self.code.addi64(1, ptr, off);
val
}
}
}
fn post_process_ret_loc(&mut self, ty: Type, loc: &Loc) {
let size = self.size_of(ty);
match size {
0 => {}
..=8 => {}
..=16 => {
if let Loc::Stack(ref stack, off) = loc {
self.store_stack(1, stack.offset + off, size as _);
} else {
unreachable!()
}
}
..=u64::MAX => {}
}
}
fn param_alloc(&self, ret: Type) -> Range<u8> {
2 + (9..=16).contains(&self.size_of(ret)) as u8..12
}
}
#[derive(Debug)]
2024-05-12 04:52:58 -05:00
pub struct Value {
ty: Type,
2024-05-10 14:33:42 -05:00
loc: Loc,
}
impl Value {
const VOID: Self = Self {
ty: bt::VOID,
loc: Loc::Imm(0),
};
2024-05-14 05:17:39 -05:00
fn new(ty: Type, loc: Loc) -> Self {
Self { ty, loc }
}
2024-05-20 07:11:58 -05:00
fn ty(ty: Type) -> Self {
Self {
ty: bt::TYPE,
loc: Loc::Imm(ty as _),
}
}
fn imm(imm: u64) -> Value {
Self {
ty: bt::UINT,
loc: Loc::Imm(imm),
}
}
}
2024-05-13 06:36:29 -05:00
#[derive(Debug, PartialEq, Eq)]
2024-05-12 04:52:58 -05:00
enum Loc {
Reg(LinReg),
RegRef(Reg),
Deref(LinReg, Option<Rc<Stack>>, u64),
DerefRef(Reg, Option<Rc<Stack>>, u64),
2024-05-10 14:33:42 -05:00
Imm(u64),
Stack(Rc<Stack>, u64),
}
2024-05-12 04:52:58 -05:00
impl Loc {
fn take_ref(&self) -> Loc {
2024-05-14 05:17:39 -05:00
match *self {
Self::Reg(LinReg(reg, ..), ..) | Self::RegRef(reg) => Self::RegRef(reg),
Self::Deref(LinReg(reg, ..), ref st, off) | Self::DerefRef(reg, ref st, off) => {
Self::DerefRef(reg, st.clone(), off)
}
Self::Stack(ref stack, off) => {
Self::DerefRef(STACK_PTR, Some(stack.clone()), stack.offset + off)
}
2024-05-14 05:17:39 -05:00
ref un => unreachable!("{:?}", un),
}
}
2024-05-14 05:17:39 -05:00
fn ref_to_ptr(&self) -> (Reg, u64) {
match *self {
Loc::Deref(LinReg(reg, ..), _, off) => (reg, off),
Loc::DerefRef(reg, _, off) => (reg, off),
2024-05-14 05:17:39 -05:00
Loc::Stack(ref stack, off) => (STACK_PTR, stack.offset + off),
ref l => panic!("expected stack location, got {:?}", l),
}
}
fn offset_ref(&self, offset: u64) -> Loc {
2024-05-14 05:17:39 -05:00
self.take_ref().offset(offset)
}
fn offset(self, offset: u64) -> Loc {
match self {
Loc::Deref(r, stack, off) => Loc::Deref(r, stack, off + offset),
Loc::DerefRef(r, stack, off) => Loc::DerefRef(r, stack, off + offset),
2024-05-14 05:17:39 -05:00
Loc::Stack(s, off) => Loc::Stack(s, off + offset),
l => unreachable!("{:?}", l),
}
}
2024-05-12 04:52:58 -05:00
}
2024-05-19 11:20:42 -05:00
#[derive(Default)]
pub struct LoggedMem {
pub mem: hbvm::mem::HostMemory,
}
2024-05-11 09:04:13 -05:00
2024-05-19 11:20:42 -05:00
impl hbvm::mem::Memory for LoggedMem {
unsafe fn load(
&mut self,
addr: hbvm::mem::Address,
target: *mut u8,
count: usize,
) -> Result<(), hbvm::mem::LoadError> {
log::dbg!(
"load: {:x} {:?}",
addr.get(),
core::slice::from_raw_parts(addr.get() as *const u8, count)
);
self.mem.load(addr, target, count)
}
2024-05-17 12:53:59 -05:00
2024-05-19 11:20:42 -05:00
unsafe fn store(
&mut self,
addr: hbvm::mem::Address,
source: *const u8,
count: usize,
) -> Result<(), hbvm::mem::StoreError> {
log::dbg!(
"store: {:x} {:?}",
addr.get(),
core::slice::from_raw_parts(source, count)
);
self.mem.store(addr, source, count)
}
2024-05-19 11:20:42 -05:00
unsafe fn prog_read<T: Copy>(&mut self, addr: hbvm::mem::Address) -> T {
log::dbg!(
"read-typed: {:x} {} {:?}",
addr.get(),
std::any::type_name::<T>(),
if core::mem::size_of::<T>() == 1 {
instrs::NAMES[std::ptr::read(addr.get() as *const u8) as usize].to_string()
} else {
core::slice::from_raw_parts(addr.get() as *const u8, core::mem::size_of::<T>())
2024-05-12 13:10:50 -05:00
.iter()
.map(|&b| format!("{:02x}", b))
.collect::<String>()
2024-05-19 11:20:42 -05:00
}
);
self.mem.prog_read(addr)
}
2024-05-19 11:20:42 -05:00
}
#[cfg(test)]
mod tests {
2024-05-20 07:11:58 -05:00
use crate::codegen::LoggedMem;
2024-05-19 11:20:42 -05:00
use super::parser;
2024-05-09 16:41:59 -05:00
fn generate(input: &'static str, output: &mut String) {
2024-05-12 17:02:32 -05:00
let path = "test";
let mut codegen = super::Codegen::default();
2024-05-19 11:20:42 -05:00
codegen.files = vec![parser::Ast::new(path, input, &parser::no_loader)];
codegen.generate();
2024-05-10 14:33:42 -05:00
let mut out = Vec::new();
codegen.dump(&mut out).unwrap();
2024-05-10 14:33:42 -05:00
use std::fmt::Write;
2024-05-12 04:52:58 -05:00
let mut stack = [0_u64; 128];
2024-05-10 14:33:42 -05:00
let mut vm = unsafe {
2024-05-19 11:20:42 -05:00
hbvm::Vm::<_, 0>::new(
2024-05-20 07:11:58 -05:00
LoggedMem::default(),
2024-05-19 11:20:42 -05:00
hbvm::mem::Address::new(out.as_ptr() as u64),
)
2024-05-10 14:33:42 -05:00
};
vm.write_reg(
super::STACK_PTR,
unsafe { stack.as_mut_ptr().add(stack.len()) } as u64,
);
2024-05-10 14:33:42 -05:00
let stat = loop {
match vm.run() {
Ok(hbvm::VmRunOk::End) => break Ok(()),
Ok(ev) => writeln!(output, "ev: {:?}", ev).unwrap(),
Err(e) => break Err(e),
}
};
2024-05-12 15:40:28 -05:00
writeln!(output, "code size: {}", out.len()).unwrap();
2024-05-11 09:04:13 -05:00
writeln!(output, "ret: {:?}", vm.read_reg(1).0).unwrap();
2024-05-10 14:33:42 -05:00
writeln!(output, "status: {:?}", stat).unwrap();
2024-05-09 11:16:01 -05:00
}
2024-05-09 16:41:59 -05:00
crate::run_tests! { generate:
example => include_str!("../examples/main_fn.hb");
2024-05-10 15:54:12 -05:00
arithmetic => include_str!("../examples/arithmetic.hb");
variables => include_str!("../examples/variables.hb");
2024-05-11 09:04:13 -05:00
functions => include_str!("../examples/functions.hb");
if_statements => include_str!("../examples/if_statement.hb");
2024-05-11 11:16:27 -05:00
loops => include_str!("../examples/loops.hb");
2024-05-12 04:52:58 -05:00
fb_driver => include_str!("../examples/fb_driver.hb");
pointers => include_str!("../examples/pointers.hb");
2024-05-12 05:16:40 -05:00
structs => include_str!("../examples/structs.hb");
2024-05-13 06:36:29 -05:00
different_types => include_str!("../examples/different_types.hb");
2024-05-14 05:17:39 -05:00
struct_operators => include_str!("../examples/struct_operators.hb");
2024-05-14 16:07:32 -05:00
directives => include_str!("../examples/directives.hb");
global_variables => include_str!("../examples/global_variables.hb");
2024-05-20 07:11:58 -05:00
geneic_types => include_str!("../examples/generic_types.hb");
2024-05-09 11:16:01 -05:00
}
}