Files
lang/src/arch/x86_64/encode.rs
T

251 lines
8.8 KiB
Rust

use super::*;
use crate::backend::Symbol;
type ERes = Result<(), CompilerMsg>;
/// machine code
#[derive(Default)]
pub struct Code {
pub(super) bytes: Vec<u8>,
pub(super) missing: Vec<(usize, Symbol)>,
}
impl Code {
pub fn mov(&mut self, dst: impl Into<RegMem>, src: impl Into<RegImmMem>) -> ERes {
let dst = dst.into();
let src = src.into();
match dst {
RegMem::Reg(mut dst) => match src {
RegImmMem::Reg(src) => {
if dst.width() != src.width() {
return Err("src and dst are not same width".into());
}
if dst.incompatible(&src) {
return Err("incompatible registers due to rex".into());
}
let width = dst.width();
self.prefix16(width);
if src.requires_rex() || dst.requires_rex() {
self.bytes.push(rex(width, src, 0, dst));
}
self.bytes.push(0x88 | width.not8());
self.bytes.push(modrm_regs(src, dst));
}
RegImmMem::Imm(src) => {
let src_width = src.width_unsigned()?;
if src_width > dst.width() {
return Err("immediate cannot fit in register".into());
}
self.prefix16(dst);
if dst.width() == Width::B64 && src_width <= Width::B32 && src.0 < 0 {
// use different op that sign extends for less bytes
self.bytes
.extend([rex(dst.width(), 0, 0, dst), 0xc7, 0xc0 | dst.base()]);
self.imm(src, Width::B32);
} else {
if src_width <= Width::B32 {
dst = dst.lower64();
}
if dst.requires_rex() {
self.bytes.push(rex(dst.width(), 0, 0, dst));
}
self.bytes.push(0xb0 | (dst.not8() << 3) | dst.base());
self.imm(src, dst.width());
}
}
RegImmMem::Mem(src) => {
if src.width != dst.width() {
return Err("register & memory sizes don't match".into());
}
if dst.high() && src.reg.gt8() {
return Err("registers incompatible (REX)".into());
}
self.prefix32(&src)?;
self.prefix16(dst);
if dst.requires_rex() | src.reg.gt8() {
self.bytes.push(rex(dst.width(), dst, 0, src.reg));
}
self.bytes.push(0x8a | dst.not8());
self.modrm_regdisp(dst, src);
}
},
RegMem::Mem(dst) => match src {
RegImmMem::Reg(src) => {
if src.width() != dst.width {
return Err("register & memory sizes don't match".into());
}
if src.high() && dst.reg.gt8() {
return Err("registers incompatible (REX)".into());
}
self.prefix32(&dst)?;
self.prefix16(src);
if src.requires_rex() | dst.reg.gt8() {
self.bytes.push(rex(src.width(), src, 0, dst.reg));
}
self.bytes.push(0x88 | src.not8());
self.modrm_regdisp(src, dst);
}
RegImmMem::Imm(src) => {
let encode_width = dst.width.min(Width::B32);
let src_width = if dst.width == Width::B64 {
src.width_signed()
} else {
src.width_unsigned()
}?;
if src_width == Width::B64 {
return Err("cannot move 64 bit immediate into memory".into());
}
if src_width > dst.width {
return Err("source cannot fit in destination".into());
}
self.prefix32(&dst)?;
self.prefix16(encode_width);
if dst.reg.requires_mem_rex() || dst.width == Width::B64 {
self.bytes.push(rex(dst.width, 0, 0, dst.reg));
}
self.bytes.push(0xc6 | encode_width.not8());
self.modrm_regdisp(None, dst);
self.imm(src, encode_width);
}
RegImmMem::Mem(_) => return Err("cannot move memory to memory".into()),
},
}
Ok(())
}
pub fn push(&mut self, reg: impl Into<RegImmMem>) -> ERes {
match reg.into() {
RegImmMem::Reg(reg) => match reg.width() {
Width::B64 => {
if reg.gt8() {
self.bytes.push(0x41);
}
self.bytes.push(0x50 | reg.base());
}
Width::B16 => {}
_ => return Err("register must be 64 or 16 bit".into()),
},
RegImmMem::Imm(imm) => match imm.width_unsigned()? {
Width::B8 => {
self.bytes.push(0x6a);
self.bytes.push(imm.0 as u8);
}
Width::B16 | Width::B32 => {
self.bytes.push(0x68);
self.bytes.extend((imm.0 as u32).to_le_bytes());
}
Width::B64 => return Err("immediate must be 32 bit or less".into()),
},
RegImmMem::Mem(mem) => todo!(),
}
Ok(())
}
pub fn pop(&mut self, reg: Reg) -> ERes {
match reg.width() {
Width::B64 | Width::B16 => (),
_ => return Err("register must be 64 or 16 bit".into()),
}
self.prefix16(reg);
if reg.gt8() {
self.bytes.push(0x41);
}
self.bytes.push(0x58 | reg.base());
Ok(())
}
pub fn lea(&mut self, dst: Reg, sym: Symbol) {
self.bytes
.extend([rex(1, dst, 0, 0), 0x8d, modrm_disp32(dst)]);
self.sym_offset4(sym);
}
pub fn int(&mut self, code: u8) {
self.bytes.extend([0xcd, code])
}
pub fn syscall(&mut self) {
self.bytes.extend([0x0f, 0x05])
}
pub fn call(&mut self, sym: Symbol) {
self.bytes.push(0xe8);
self.sym_offset4(sym);
}
pub fn call_mem(&mut self, sym: Symbol) {
self.bytes.extend([0xff, 0x15]);
self.sym_offset4(sym);
}
pub fn ret(&mut self) {
self.bytes.push(0xc3);
}
pub fn sub(&mut self) {
// sub esp 40 iirc
self.bytes.extend([0x48, 0x83, 0xec, 0x28]);
}
fn prefix16(&mut self, width: impl Into<Width>) {
if width.into() == Width::B16 {
self.bytes.push(0x66);
}
}
fn prefix32(&mut self, mem: &Mem) -> Result<(), CompilerMsg> {
match mem.reg.width() {
Width::B8 | Width::B16 => return Err("invalid register width".into()),
Width::B32 => self.bytes.push(0x67),
Width::B64 => (),
}
Ok(())
}
fn modrm_regdisp(&mut self, reg: impl Into<Option<Reg>>, mem: Mem) {
const I8_MIN: i32 = i8::MIN as i32;
const I8_MAX: i32 = i8::MAX as i32;
let mod_ = match mem.disp {
0 => 0b00,
I8_MIN..=I8_MAX => 0b01,
_ => 0b10,
};
let r = reg.into().map(|r| Reg::base(&r)).unwrap_or(0);
self.bytes.push(modrm(mod_, r, mem.reg.base()));
if mem.reg.base() == rsp.base() {
// SIB
self.bytes.push(0x24);
}
match mod_ {
0b00 => (),
0b01 => self.bytes.push(mem.disp as u8),
0b10 => self.bytes.extend(mem.disp.to_le_bytes()),
_ => unreachable!(),
}
}
/// inserts a 32 bit offset from a symbol
fn sym_offset4(&mut self, sym: Symbol) {
let pos = self.bytes.len();
self.bytes.extend([0; 4]);
self.missing.push((pos, sym));
}
pub fn extend(&mut self, other: &Code) {
let pos = self.bytes.len();
self.bytes.extend(&other.bytes);
self.missing
.extend(other.missing.iter().map(|&(p, s)| (pos + p, s)));
}
fn imm(&mut self, imm: Imm, width: Width) {
self.bytes.extend(&imm.0.to_le_bytes()[..width.bytes()]);
}
}
pub fn encode(f: impl FnOnce(&mut Code) -> Result<(), CompilerMsg>) -> Result<Code, CompilerMsg> {
let mut code = Code::default();
f(&mut code)?;
Ok(code)
}