holey-bytes/hblang/src/codegen.rs

1068 lines
33 KiB
Rust
Raw Normal View History

2024-05-12 04:52:58 -05:00
use crate::ident::Ident;
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};
2024-05-10 14:33:42 -05:00
type LabelId = u32;
type Reg = u8;
type MaskElem = u64;
2024-05-11 15:22:08 -05:00
type Type = u32;
2024-05-12 04:52:58 -05:00
#[derive(Debug)]
struct LinReg(Reg);
#[cfg(debug_assertions)]
impl Drop for LinReg {
fn drop(&mut self) {
if !std::thread::panicking() {
panic!("reg leaked");
}
}
}
pub mod bt {
2024-05-11 15:22:08 -05:00
use super::*;
const fn builtin_type(id: u32) -> Type {
Type::MAX - id
}
macro_rules! builtin_type {
($($name:ident;)*) => {$(
2024-05-12 04:52:58 -05:00
pub const $name: Type = TypeKind::Builtin(${index(0)}).encode();
2024-05-11 15:22:08 -05:00
)*};
}
builtin_type! {
2024-05-12 04:52:58 -05:00
VOID;
NEVER;
2024-05-11 15:22:08 -05:00
INT;
BOOL;
}
}
2024-05-12 04:52:58 -05:00
#[derive(Debug)]
2024-05-11 15:22:08 -05:00
enum TypeKind {
Builtin(Type),
Struct(Type),
2024-05-12 04:52:58 -05:00
Pointer(Type),
2024-05-11 15:22:08 -05:00
}
impl TypeKind {
2024-05-12 04:52:58 -05:00
const fn from_ty(ty: Type) -> Self {
let (flag, index) = (ty & 0b11, ty >> 2);
match flag {
0 => Self::Builtin(index),
1 => Self::Pointer(index),
2 => Self::Struct(index),
_ => unreachable!(),
2024-05-11 15:22:08 -05:00
}
}
2024-05-12 04:52:58 -05:00
const fn encode(self) -> Type {
let (index, flag) = match self {
Self::Builtin(index) => (index, 0),
Self::Pointer(index) => (index, 1),
Self::Struct(index) => (index, 2),
};
index << 2 | flag
}
2024-05-11 15:22:08 -05:00
}
const STACK_PTR: Reg = 254;
const ZERO: Reg = 0;
const RET_ADDR: Reg = 31;
const ELEM_WIDTH: usize = std::mem::size_of::<MaskElem>() * 8;
2024-05-10 14:33:42 -05:00
struct Frame {
label: LabelId,
prev_relocs: usize,
offset: u32,
}
struct Reloc {
2024-05-11 09:04:13 -05:00
id: LabelId,
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
}
struct StackReloc {
offset: u32,
size: u16,
}
2024-05-10 14:33:42 -05:00
#[derive(Default)]
pub struct Func {
code: Vec<u8>,
relocs: Vec<Reloc>,
}
impl Func {
pub fn extend(&mut self, bytes: &[u8]) {
self.code.extend_from_slice(bytes);
}
2024-05-11 09:04:13 -05:00
pub fn offset(&mut self, id: LabelId, instr_offset: u16, size: u16) {
2024-05-10 14:33:42 -05:00
self.relocs.push(Reloc {
id,
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]);
}
2024-05-10 14:33:42 -05:00
fn push(&mut self, value: Reg, size: usize) {
self.subi64(STACK_PTR, STACK_PTR, size as _);
self.encode(instrs::st(value, STACK_PTR, 0, size as _));
2024-05-10 14:33:42 -05:00
}
fn pop(&mut self, value: Reg, size: usize) {
self.encode(instrs::ld(value, STACK_PTR, 0, size as _));
self.encode(instrs::addi64(STACK_PTR, STACK_PTR, size as _));
}
fn subi64(&mut self, dest: Reg, src: Reg, imm: u64) {
self.encode(instrs::addi64(dest, src, imm.wrapping_neg()));
2024-05-10 14:33:42 -05:00
}
fn call(&mut self, func: LabelId) {
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.pop(RET_ADDR, 8);
self.encode(instrs::jala(ZERO, RET_ADDR, 0));
2024-05-10 14:33:42 -05:00
}
fn prelude(&mut self, entry: LabelId) {
self.call(entry);
self.encode(instrs::tx());
2024-05-10 15:54:12 -05:00
}
2024-05-11 09:04:13 -05:00
fn relocate(&mut self, labels: &[FnLabel], shift: i64) {
2024-05-10 14:33:42 -05:00
for reloc in self.relocs.drain(..) {
let label = &labels[reloc.id as usize];
let offset = if reloc.size == 8 {
reloc.offset as i64
} else {
label.offset as i64 - reloc.offset as i64
} + shift;
2024-05-11 15:22:08 -05:00
log::dbg!(
2024-05-12 04:52:58 -05:00
label.name,
2024-05-11 09:04:13 -05:00
offset,
reloc.size,
reloc.instr_offset,
reloc.offset,
shift,
label.offset
);
let dest = &mut self.code[reloc.offset as usize + reloc.instr_offset as usize..]
[..reloc.size as usize];
2024-05-10 14:33:42 -05:00
match reloc.size {
2 => dest.copy_from_slice(&(offset as i16).to_le_bytes()),
4 => dest.copy_from_slice(&(offset as i32).to_le_bytes()),
8 => dest.copy_from_slice(&(offset as i64).to_le_bytes()),
_ => unreachable!(),
};
}
}
}
#[derive(Default)]
pub struct RegAlloc {
free: Vec<Reg>,
// TODO:use 256 bit mask instead
2024-05-10 14:33:42 -05:00
used: Vec<Reg>,
}
impl RegAlloc {
2024-05-11 09:04:13 -05:00
fn init_callee(&mut self) {
2024-05-10 14:33:42 -05:00
self.clear();
2024-05-11 09:04:13 -05:00
self.free.extend(32..=253);
2024-05-10 14:33:42 -05:00
}
fn clear(&mut self) {
self.free.clear();
self.used.clear();
}
2024-05-12 04:52:58 -05:00
fn allocate(&mut self) -> LinReg {
let reg = self.free.pop().expect("TODO: we need to spill");
2024-05-10 14:33:42 -05:00
if self.used.binary_search_by_key(&!reg, |&r| !r).is_err() {
self.used.push(reg);
}
2024-05-12 04:52:58 -05:00
LinReg(reg)
}
2024-05-12 04:52:58 -05:00
fn free(&mut self, reg: LinReg) {
self.free.push(reg.0);
std::mem::forget(reg);
}
}
2024-05-09 11:16:01 -05:00
2024-05-11 09:04:13 -05:00
struct FnLabel {
2024-05-10 14:33:42 -05:00
offset: u32,
// TODO: use different stile of identifier that does not allocate, eg. index + length into a
// file
2024-05-12 04:52:58 -05:00
name: Ident,
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 {
offset: u32,
relocs: Vec<RetReloc>,
}
2024-05-11 15:22:08 -05:00
struct Struct {
2024-05-12 05:21:40 -05:00
id: Ident,
2024-05-12 06:13:36 -05:00
fields: Rc<[(Rc<str>, Type)]>,
2024-05-11 15:22:08 -05:00
}
2024-05-09 16:41:59 -05:00
pub struct Codegen<'a> {
2024-05-11 09:04:13 -05:00
path: &'a std::path::Path,
2024-05-12 04:52:58 -05:00
ret: Type,
2024-05-11 09:04:13 -05:00
gpa: RegAlloc,
code: Func,
temp: Func,
labels: Vec<FnLabel>,
stack_size: u64,
2024-05-12 04:52:58 -05:00
vars: Vec<Variable>,
stack_relocs: Vec<StackReloc>,
2024-05-11 09:04:13 -05:00
ret_relocs: Vec<RetReloc>,
2024-05-11 11:16:27 -05:00
loops: Vec<Loop>,
2024-05-11 15:22:08 -05:00
records: Vec<Struct>,
2024-05-12 04:52:58 -05:00
pointers: Vec<Type>,
main: Option<LabelId>,
2024-05-09 11:16:01 -05:00
}
2024-05-09 16:41:59 -05:00
impl<'a> Codegen<'a> {
pub fn new() -> Self {
2024-05-09 11:16:01 -05:00
Self {
2024-05-12 04:52:58 -05:00
path: std::path::Path::new(""),
ret: bt::VOID,
gpa: Default::default(),
code: Default::default(),
temp: Default::default(),
labels: Default::default(),
stack_size: 0,
2024-05-12 04:52:58 -05:00
vars: Default::default(),
stack_relocs: Default::default(),
2024-05-12 04:52:58 -05:00
ret_relocs: Default::default(),
loops: Default::default(),
records: Default::default(),
pointers: Default::default(),
main: None,
2024-05-09 11:16:01 -05:00
}
}
2024-05-10 14:33:42 -05:00
pub fn file(
&mut self,
path: &'a std::path::Path,
exprs: &'a [parser::Expr<'a>],
) -> std::fmt::Result {
self.path = path;
2024-05-09 16:41:59 -05:00
for expr in exprs {
2024-05-10 14:33:42 -05:00
self.expr(expr, None);
2024-05-09 11:16:01 -05:00
}
2024-05-09 16:41:59 -05:00
Ok(())
2024-05-09 11:16:01 -05:00
}
2024-05-12 04:52:58 -05:00
fn size_of(&self, ty: Type) -> u64 {
2024-05-12 06:13:36 -05:00
match ty {
bt::INT => 8,
bt::BOOL => 1,
_ => match TypeKind::from_ty(ty) {
TypeKind::Pointer(_) => 8,
TypeKind::Builtin(e) => unreachable!("{:?}", e),
TypeKind::Struct(ty) => self.records[ty as usize]
.fields
.iter()
.map(|(_, ty)| self.size_of(*ty))
.sum(),
},
2024-05-12 04:52:58 -05:00
}
}
fn loc_to_reg(&mut self, loc: Loc) -> LinReg {
match loc {
2024-05-12 04:52:58 -05:00
Loc::RegRef(rr) => {
let reg = self.gpa.allocate();
self.code.encode(instrs::cp(reg.0, rr));
reg
}
Loc::Reg(reg) => reg,
2024-05-12 04:52:58 -05:00
Loc::Deref(dreg) => {
let reg = self.gpa.allocate();
self.code.encode(instrs::ld(reg.0, dreg.0, 0, 8));
self.gpa.free(dreg);
reg
}
Loc::Imm(imm) => {
let reg = self.gpa.allocate();
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::li64(reg.0, imm));
reg
}
2024-05-12 06:13:36 -05:00
Loc::Stack(offset) | Loc::StackRef(offset) => {
let reg = self.gpa.allocate();
2024-05-12 04:52:58 -05:00
self.load_stack(reg.0, offset, 8);
reg
}
}
}
fn alloc_stack(&mut self, size: u32) -> u64 {
let offset = self.stack_size;
self.stack_size += size as u64;
offset
}
fn store_stack(&mut self, reg: Reg, offset: u64, size: u16) {
self.stack_relocs.push(StackReloc {
offset: self.code.code.len() as u32 + 3,
size,
});
self.code.encode(instrs::st(reg, STACK_PTR, offset, size));
}
fn load_stack(&mut self, reg: Reg, offset: u64, size: u16) {
self.stack_relocs.push(StackReloc {
offset: self.code.code.len() as u32 + 3,
size,
});
self.code.encode(instrs::ld(reg, STACK_PTR, offset, size));
}
2024-05-12 04:52:58 -05:00
fn reloc_stack(&mut self) {
for reloc in self.stack_relocs.drain(..) {
let dest = &mut self.code.code[reloc.offset as usize..][..reloc.size as usize];
let value = u64::from_ne_bytes(dest.try_into().unwrap());
2024-05-12 04:52:58 -05:00
let offset = self.stack_size - value;
dest.copy_from_slice(&offset.to_ne_bytes());
}
}
2024-05-11 09:04:13 -05:00
fn reloc_rets(&mut self) {
let len = self.code.code.len() as i32;
for reloc in self.ret_relocs.drain(..) {
let dest = &mut self.code.code[reloc.offset as usize + reloc.instr_offset as usize..]
[..reloc.size as usize];
debug_assert!(dest.iter().all(|&b| b == 0));
let offset = len - reloc.offset as i32;
dest.copy_from_slice(&offset.to_ne_bytes());
}
}
2024-05-12 04:52:58 -05:00
fn ty(&mut self, expr: &parser::Expr<'a>) -> Type {
match *expr {
2024-05-12 04:52:58 -05:00
E::Ident { name: "int", .. } => bt::INT,
E::Ident { name: "bool", .. } => bt::BOOL,
E::Ident { name: "void", .. } => bt::VOID,
E::Ident { name: "never", .. } => bt::NEVER,
2024-05-12 06:13:36 -05:00
E::Ident { id, .. } => {
let index = self
.records
.iter()
.position(|r| r.id == id)
.unwrap_or_else(|| {
panic!("type not found: {:?}", id);
});
TypeKind::Struct(index as Type).encode()
}
2024-05-12 04:52:58 -05:00
expr => unimplemented!("type: {:#?}", expr),
}
}
fn expr(&mut self, expr: &'a parser::Expr<'a>, expeted: Option<Type>) -> Option<Value> {
match *expr {
2024-05-12 06:13:36 -05:00
E::Ctor { ty, fields } => {
let ty = self.ty(&ty);
let TypeKind::Struct(idx) = TypeKind::from_ty(ty) else {
panic!("expected struct, got {:?}", ty);
};
let size = self.size_of(ty);
let stack = self.alloc_stack(size as u32);
let mut field_values = fields
.iter()
.map(|(name, field)| self.expr(field, None).map(|v| (*name, v)))
.collect::<Option<Vec<_>>>()?;
2024-05-12 06:13:36 -05:00
let decl_fields = self.records[idx as usize].fields.clone();
let mut offset = 0;
for (name, ty) in decl_fields.as_ref() {
let index = field_values
.iter()
.position(|(n, _)| *n == name.as_ref())
.unwrap();
let (_, value) = field_values.remove(index);
if value.ty != *ty {
panic!("expected {:?}, got {:?}", ty, value.ty);
}
let reg = self.loc_to_reg(value.loc);
self.store_stack(reg.0, stack + offset, 8);
self.gpa.free(reg);
offset += 8;
}
Some(Value {
ty,
loc: Loc::Stack(stack),
})
}
E::Field { target, field } => {
let target = self.expr(target, None)?;
2024-05-12 06:13:36 -05:00
let TypeKind::Struct(idx) = TypeKind::from_ty(target.ty) else {
panic!("expected struct, got {:?}", target.ty);
};
let decl_fields = self.records[idx as usize].fields.clone();
let index = decl_fields
.iter()
.position(|(name, _)| name.as_ref() == field)
.unwrap();
let size = self.size_of(decl_fields[index].1);
assert_eq!(size, 8, "TODO: implement other sizes");
let value = match target.loc {
Loc::Reg(_) => todo!(),
Loc::RegRef(_) => todo!(),
Loc::Deref(r) => {
self.code.encode(instrs::addi64(r.0, r.0, index as u64 * 8));
Loc::Deref(r)
}
Loc::Imm(_) => todo!(),
Loc::Stack(stack) => Loc::Stack(stack + index as u64 * 8),
Loc::StackRef(stack) => Loc::StackRef(stack + index as u64 * 8),
};
Some(Value {
ty: decl_fields[index].1,
loc: value,
})
}
2024-05-12 05:21:40 -05:00
E::BinOp {
left: E::Ident { id, .. },
op: T::Decl,
right: E::Struct { fields, .. },
} => {
let fields = fields
.iter()
.map(|&(name, ty)| (name.into(), self.ty(&ty)))
.collect();
self.records.push(Struct { id: *id, fields });
Some(Value::VOID)
2024-05-12 05:21:40 -05:00
}
2024-05-12 04:52:58 -05:00
E::UnOp {
op: T::Amp, val, ..
} => {
let val = self.expr(val, None)?;
2024-05-12 04:52:58 -05:00
match val.loc {
2024-05-12 06:13:36 -05:00
Loc::StackRef(off) => {
2024-05-12 04:52:58 -05:00
let reg = self.gpa.allocate();
self.stack_relocs.push(StackReloc {
offset: self.code.code.len() as u32 + 3,
size: 8,
});
self.code.encode(instrs::addi64(reg.0, STACK_PTR, off));
Some(Value {
ty: self.alloc_pointer(val.ty),
loc: Loc::Reg(reg),
})
}
l => panic!("cant take pointer of {:?}", l),
}
}
E::UnOp {
op: T::Star, val, ..
} => {
let val = self.expr(val, None)?;
2024-05-12 04:52:58 -05:00
let reg = self.loc_to_reg(val.loc);
match TypeKind::from_ty(val.ty) {
TypeKind::Pointer(ty) => Some(Value {
ty: self.pointers[ty as usize],
loc: Loc::Deref(reg),
}),
_ => panic!("cant deref {:?}", val.ty),
}
}
2024-05-11 15:22:08 -05:00
E::BinOp {
2024-05-12 04:52:58 -05:00
left: E::Ident { name, id, .. },
2024-05-11 15:22:08 -05:00
op: T::Decl,
right: E::Closure {
ret, body, args, ..
},
2024-05-09 16:41:59 -05:00
} => {
2024-05-12 04:52:58 -05:00
log::dbg!("fn: {}", name);
let frame = self.add_label(*id);
if *name == "main" {
self.main = Some(frame.label);
}
log::dbg!("fn-args");
for (i, arg) in args.iter().enumerate() {
2024-05-11 09:04:13 -05:00
let offset = self.alloc_stack(8);
2024-05-12 04:52:58 -05:00
let ty = self.ty(&arg.ty);
self.vars.push(Variable {
id: arg.id,
value: Value {
ty,
loc: Loc::Stack(offset),
},
});
2024-05-11 09:04:13 -05:00
self.store_stack(i as Reg + 2, offset, 8);
}
self.gpa.init_callee();
2024-05-12 04:52:58 -05:00
self.ret = self.ty(ret);
log::dbg!("fn-body");
if self.expr(body, None).is_some() {
panic!("expected all paths in the fucntion {name} to return");
}
2024-05-11 09:04:13 -05:00
self.vars.clear();
2024-05-12 04:52:58 -05:00
log::dbg!("fn-relocs");
self.reloc_stack();
log::dbg!("fn-prelude");
2024-05-10 14:33:42 -05:00
self.write_fn_prelude(frame);
2024-05-12 04:52:58 -05:00
log::dbg!("fn-ret");
2024-05-11 09:04:13 -05:00
self.reloc_rets();
self.ret();
2024-05-12 04:52:58 -05:00
self.stack_size = 0;
self.vars.clear();
Some(Value::VOID)
2024-05-10 14:33:42 -05:00
}
2024-05-11 15:22:08 -05:00
E::BinOp {
2024-05-12 04:52:58 -05:00
left: E::Ident { id, .. },
2024-05-11 15:22:08 -05:00
op: T::Decl,
right,
} => {
let val = self.expr(right, None)?;
2024-05-12 06:13:36 -05:00
let loc = self.make_loc_owned(val.loc, val.ty);
let loc = self.ensure_spilled(loc);
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
}
2024-05-11 09:04:13 -05:00
E::Call {
2024-05-12 04:52:58 -05:00
func: E::Ident { id, .. },
2024-05-11 09:04:13 -05:00
args,
} => {
for (i, arg) in args.iter().enumerate() {
let arg = self.expr(arg, None)?;
2024-05-11 09:04:13 -05:00
let reg = self.loc_to_reg(arg.loc);
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cp(i as Reg + 2, reg.0));
self.gpa.free(reg);
2024-05-11 09:04:13 -05:00
}
2024-05-12 04:52:58 -05:00
let func = self.get_or_reserve_label(*id);
2024-05-11 09:04:13 -05:00
self.code.call(func);
let reg = self.gpa.allocate();
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cp(reg.0, 1));
2024-05-11 09:04:13 -05:00
Some(Value {
ty: self.ret,
loc: Loc::Reg(reg),
})
}
2024-05-12 04:52:58 -05:00
E::Ident { name, id, .. } => {
let var = self
.vars
.iter()
.find(|v| v.id == id)
.unwrap_or_else(|| panic!("variable not found: {:?}", name));
Some(Value {
2024-05-12 04:52:58 -05:00
ty: var.value.ty,
loc: var.value.loc.take_ref(),
})
}
2024-05-11 15:22:08 -05:00
E::Return { val, .. } => {
2024-05-10 14:33:42 -05:00
if let Some(val) = val {
let val = self.expr(val, Some(self.ret))?;
2024-05-10 14:33:42 -05:00
if val.ty != self.ret {
2024-05-12 04:52:58 -05:00
panic!("expected {:?}, got {:?}", self.ret, val.ty);
2024-05-10 14:33:42 -05:00
}
self.assign(
Value {
ty: self.ret,
2024-05-12 04:52:58 -05:00
loc: Loc::RegRef(1),
},
val,
);
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(instrs::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 {
if let Loc::Reg(reg) = self.expr(stmt, None)?.loc {
2024-05-12 04:52:58 -05:00
self.gpa.free(reg);
}
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 {
2024-05-12 04:52:58 -05:00
ty: expeted.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(cond, Some(bt::BOOL))?;
let reg = self.loc_to_reg(cond.loc);
let jump_offset = self.code.code.len() as u32;
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::jeq(reg.0, 0, 0));
self.gpa.free(reg);
2024-05-12 04:52:58 -05:00
log::dbg!("if-then");
let then_unreachable = self.expr(then, None).is_none();
let mut else_unreachable = false;
2024-05-11 11:16:27 -05:00
let mut jump = self.code.code.len() as i16 - jump_offset as i16;
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(instrs::jmp(0));
jump = self.code.code.len() as i16 - jump_offset as i16;
}
2024-05-11 11:16:27 -05:00
else_unreachable = self.expr(else_, None).is_none();
2024-05-11 11:16:27 -05:00
if !then_unreachable {
let jump = self.code.code.len() as i32 - else_jump_offset as i32;
log::dbg!("if-else-jump: {}", jump);
self.code.code[else_jump_offset as usize + 1..][..4]
.copy_from_slice(&jump.to_ne_bytes());
}
2024-05-11 11:16:27 -05:00
}
2024-05-12 04:52:58 -05:00
log::dbg!("if-then-jump: {}", jump);
self.code.code[jump_offset as usize + 3..][..2]
.copy_from_slice(&jump.to_ne_bytes());
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 {
offset: loop_start,
relocs: Default::default(),
});
let body_unreachable = self.expr(body, None).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(instrs::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());
}
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(instrs::jmp(0));
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(instrs::jmp(loop_.offset as i32 - offset as i32));
None
}
2024-05-10 15:54:12 -05:00
E::BinOp { left, op, right } => {
let left = self.expr(left, expeted)?;
let right = self.expr(right, Some(left.ty))?;
2024-05-12 04:52:58 -05:00
if op == T::Assign {
return self.assign(left, right);
}
2024-05-10 15:54:12 -05:00
let lhs = self.loc_to_reg(left.loc);
let rhs = self.loc_to_reg(right.loc);
2024-05-10 15:54:12 -05:00
let op = match op {
T::Plus => instrs::add64,
T::Minus => instrs::sub64,
T::Star => instrs::mul64,
T::Le => {
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cmpu(lhs.0, lhs.0, rhs.0));
self.gpa.free(rhs);
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cmpui(lhs.0, lhs.0, 1));
return Some(Value {
2024-05-12 04:52:58 -05:00
ty: bt::BOOL,
loc: Loc::Reg(lhs),
});
}
2024-05-11 11:16:27 -05:00
T::Eq => {
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cmpu(lhs.0, lhs.0, rhs.0));
2024-05-11 11:16:27 -05:00
self.gpa.free(rhs);
2024-05-12 04:52:58 -05:00
self.code.encode(instrs::cmpui(lhs.0, lhs.0, 0));
self.code.encode(instrs::not(lhs.0, lhs.0));
2024-05-11 11:16:27 -05:00
return Some(Value {
2024-05-12 04:52:58 -05:00
ty: bt::BOOL,
2024-05-11 11:16:27 -05:00
loc: Loc::Reg(lhs),
});
}
T::FSlash => |reg0, reg1, reg2| instrs::diru64(reg0, ZERO, reg1, reg2),
2024-05-10 15:54:12 -05:00
_ => unimplemented!("{:#?}", op),
};
2024-05-12 04:52:58 -05:00
self.code.encode(op(lhs.0, lhs.0, rhs.0));
2024-05-10 15:54:12 -05:00
self.gpa.free(rhs);
Some(Value {
ty: left.ty,
loc: Loc::Reg(lhs),
})
}
ast => unimplemented!("{:#?}", ast),
2024-05-09 11:16:01 -05:00
}
}
2024-05-12 04:52:58 -05:00
fn assign(&mut self, left: Value, right: Value) -> Option<Value> {
let rhs = self.loc_to_reg(right.loc);
match left.loc {
2024-05-12 04:52:58 -05:00
Loc::Deref(reg) => {
self.code.encode(instrs::st(rhs.0, reg.0, 0, 8));
self.gpa.free(reg);
}
Loc::RegRef(reg) => self.code.encode(instrs::cp(reg, rhs.0)),
2024-05-12 06:13:36 -05:00
Loc::StackRef(offset) => self.store_stack(rhs.0, offset, 8),
_ => unimplemented!(),
}
self.gpa.free(rhs);
Some(Value::VOID)
}
2024-05-12 04:52:58 -05:00
fn get_or_reserve_label(&mut self, name: Ident) -> LabelId {
if let Some(label) = self.labels.iter().position(|l| l.name == name) {
2024-05-10 14:33:42 -05:00
label as u32
} else {
2024-05-12 04:52:58 -05:00
self.labels.push(FnLabel { offset: 0, name });
2024-05-10 14:33:42 -05:00
self.labels.len() as u32 - 1
}
2024-05-09 11:16:01 -05:00
}
2024-05-12 04:52:58 -05:00
fn add_label(&mut self, name: Ident) -> Frame {
2024-05-10 14:33:42 -05:00
let offset = self.code.code.len() as u32;
2024-05-12 04:52:58 -05:00
let label = if let Some(label) = self.labels.iter().position(|l| l.name == name) {
2024-05-10 14:33:42 -05:00
self.labels[label].offset = offset;
label as u32
} else {
2024-05-11 09:04:13 -05:00
self.labels.push(FnLabel {
2024-05-10 14:33:42 -05:00
offset,
name: name.into(),
});
self.labels.len() as u32 - 1
};
Frame {
label,
prev_relocs: self.code.relocs.len(),
offset,
}
2024-05-09 11:16:01 -05:00
}
2024-05-12 04:52:58 -05:00
fn get_label(&self, name: Ident) -> LabelId {
self.labels.iter().position(|l| l.name == name).unwrap() as _
2024-05-09 11:16:01 -05:00
}
2024-05-10 14:33:42 -05:00
fn write_fn_prelude(&mut self, frame: Frame) {
self.temp.push(RET_ADDR, 8);
2024-05-10 14:33:42 -05:00
for &reg in self.gpa.used.clone().iter() {
self.temp.push(reg, 8);
}
2024-05-12 04:52:58 -05:00
self.temp.subi64(STACK_PTR, STACK_PTR, self.stack_size);
2024-05-10 14:33:42 -05:00
for reloc in &mut self.code.relocs[frame.prev_relocs..] {
reloc.offset += self.temp.code.len() as u32;
}
2024-05-11 09:04:13 -05:00
for reloc in &mut self.ret_relocs {
reloc.offset += self.temp.code.len() as u32;
}
2024-05-10 14:33:42 -05:00
self.code.code.splice(
frame.offset as usize..frame.offset as usize,
self.temp.code.drain(..),
);
}
2024-05-10 14:33:42 -05:00
fn ret(&mut self) {
2024-05-11 09:04:13 -05:00
self.code
2024-05-12 04:52:58 -05:00
.encode(instrs::addi64(STACK_PTR, STACK_PTR, self.stack_size));
2024-05-10 14:33:42 -05:00
for reg in self.gpa.used.clone().iter().rev() {
self.code.pop(*reg, 8);
}
self.code.ret();
}
pub fn dump(mut self, out: &mut impl std::io::Write) -> std::io::Result<()> {
2024-05-12 04:52:58 -05:00
self.temp.prelude(self.main.unwrap());
2024-05-10 14:33:42 -05:00
self.temp
.relocate(&self.labels, self.temp.code.len() as i64);
self.code.relocate(&self.labels, 0);
out.write_all(&self.temp.code)?;
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.gpa.allocate();
self.code.encode(instrs::cp(reg.0, rreg));
Loc::Reg(reg)
}
Loc::Imm(imm) => {
let reg = self.gpa.allocate();
self.code.encode(instrs::li64(reg.0, imm));
Loc::Reg(reg)
}
Loc::StackRef(mut off) => {
let size = self.size_of(ty);
assert!(size % 8 == 0, "TODO: implement other sizes");
let stack = self.alloc_stack(size as u32);
let reg = self.gpa.allocate();
while size > 0 {
self.load_stack(reg.0, off, 8);
self.store_stack(reg.0, stack, 8);
off += 8;
}
self.gpa.free(reg);
Loc::Stack(stack)
}
l => l,
}
}
fn ensure_spilled(&mut self, loc: Loc) -> Loc {
match loc {
Loc::Reg(reg) => {
let stack = self.alloc_stack(8);
self.store_stack(reg.0, stack, 8);
self.gpa.free(reg);
Loc::Stack(stack)
}
l => l,
}
}
}
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),
};
}
2024-05-12 04:52:58 -05:00
#[derive(Debug)]
enum Loc {
Reg(LinReg),
RegRef(Reg),
Deref(LinReg),
2024-05-10 14:33:42 -05:00
Imm(u64),
Stack(u64),
2024-05-12 06:13:36 -05:00
StackRef(u64),
}
2024-05-12 04:52:58 -05:00
impl Loc {
fn take_ref(&self) -> Loc {
match self {
Self::Reg(reg) => Self::RegRef(reg.0),
2024-05-12 06:13:36 -05:00
Self::Stack(off) => Self::StackRef(*off),
un => unreachable!("{:?}", un),
2024-05-12 04:52:58 -05:00
}
}
}
2024-05-09 16:41:59 -05:00
#[cfg(test)]
mod tests {
2024-05-11 15:22:08 -05:00
use crate::{instrs, log};
2024-05-11 09:04:13 -05:00
struct TestMem;
impl hbvm::mem::Memory for TestMem {
#[inline]
unsafe fn load(
&mut self,
addr: hbvm::mem::Address,
target: *mut u8,
count: usize,
) -> Result<(), hbvm::mem::LoadError> {
2024-05-11 15:22:08 -05:00
log::dbg!(
2024-05-11 09:04:13 -05:00
"read: {:x} {} {:?}",
addr.get(),
count,
core::slice::from_raw_parts(target, count)
.iter()
.rev()
.skip_while(|&&b| b == 0)
.map(|&b| format!("{:02x}", b))
.collect::<String>()
);
unsafe { core::ptr::copy(addr.get() as *const u8, target, count) }
Ok(())
}
#[inline]
unsafe fn store(
&mut self,
addr: hbvm::mem::Address,
source: *const u8,
count: usize,
) -> Result<(), hbvm::mem::StoreError> {
2024-05-11 15:22:08 -05:00
log::dbg!("write: {:x} {}", addr.get(), count);
unsafe { core::ptr::copy(source, addr.get() as *mut u8, count) }
Ok(())
}
#[inline]
unsafe fn prog_read<T: Copy>(&mut self, addr: hbvm::mem::Address) -> T {
2024-05-11 15:22:08 -05:00
log::dbg!(
2024-05-11 09:04:13 -05:00
"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>())
.iter()
.map(|&b| format!("{:02x}", b))
.collect::<String>()
}
);
unsafe { core::ptr::read(addr.get() as *const T) }
}
}
2024-05-09 16:41:59 -05:00
fn generate(input: &'static str, output: &mut String) {
let path = std::path::Path::new("test");
let arena = crate::parser::Arena::default();
2024-05-12 04:52:58 -05:00
let mut parser = super::parser::Parser::new(input, path, &arena);
2024-05-09 16:41:59 -05:00
let exprs = parser.file();
let mut codegen = super::Codegen::new();
codegen.file(path, &exprs).unwrap();
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 {
hbvm::Vm::<TestMem, 0>::new(TestMem, hbvm::mem::Address::new(out.as_ptr() as u64))
};
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),
}
};
2024-05-11 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");
2024-05-09 11:16:01 -05:00
}
}