use { super::{AssemblySpec, Backend, Nid, Node, Nodes}, crate::{ lexer::TokenKind, parser, reg, son::{debug_assert_matches, write_reloc, Kind, MEM}, ty::{self, Loc}, Offset, Reloc, Size, TypedReloc, Types, }, alloc::{boxed::Box, collections::BTreeMap, string::String, vec::Vec}, core::mem, hbbytecode::{self as instrs, *}, }; mod my_regalloc; mod their_regalloc; struct FuncDt { offset: Offset, // TODO: change to indices into common vec relocs: Vec, code: Vec, } impl Default for FuncDt { fn default() -> Self { Self { offset: u32::MAX, relocs: Default::default(), code: Default::default() } } } struct GlobalDt { offset: Offset, } impl Default for GlobalDt { fn default() -> Self { Self { offset: u32::MAX } } } #[derive(Default)] struct Assembler { frontier: Vec, globals: Vec, funcs: Vec, } #[derive(Default)] pub struct HbvmBackend { funcs: Vec, globals: Vec, asm: Assembler, ralloc: their_regalloc::Regalloc, ralloc_my: my_regalloc::Res, ret_relocs: Vec, relocs: Vec, jump_relocs: Vec<(Nid, Reloc)>, code: Vec, offsets: Vec, } impl HbvmBackend { fn emit(&mut self, instr: (usize, [u8; instrs::MAX_SIZE])) { emit(&mut self.code, instr); } } impl Backend for HbvmBackend { fn assemble_bin(&mut self, entry: ty::Func, types: &Types, to: &mut Vec) { to.extend([0u8; HEADER_SIZE]); binary_prelude(to); let AssemblySpec { code_length, data_length, entry } = self.assemble_reachable(entry, types, to); let exe = AbleOsExecutableHeader { magic_number: [0x15, 0x91, 0xD2], executable_version: 0, code_length, data_length, debug_length: 0, config_length: 0, metadata_length: 0, }; Reloc::new(HEADER_SIZE, 3, 4).apply_jump(to, entry, 0); unsafe { *to.as_mut_ptr().cast::() = exe } } fn assemble_reachable( &mut self, from: ty::Func, types: &Types, to: &mut Vec, ) -> AssemblySpec { debug_assert!(self.asm.frontier.is_empty()); debug_assert!(self.asm.funcs.is_empty()); debug_assert!(self.asm.globals.is_empty()); self.globals.resize_with(types.ins.globals.len(), Default::default); self.asm.frontier.push(ty::Kind::Func(from).compress()); while let Some(itm) = self.asm.frontier.pop() { match itm.expand() { ty::Kind::Func(func) => { let fuc = &mut self.funcs[func as usize]; debug_assert!(!fuc.code.is_empty()); if fuc.offset != u32::MAX { continue; } fuc.offset = 0; self.asm.funcs.push(func); self.asm.frontier.extend(fuc.relocs.iter().map(|r| r.target)); } ty::Kind::Global(glob) => { let glb = &mut self.globals[glob as usize]; if glb.offset != u32::MAX { continue; } glb.offset = 0; self.asm.globals.push(glob); } _ => unreachable!(), } } let init_len = to.len(); for &func in &self.asm.funcs { let fuc = &mut self.funcs[func as usize]; fuc.offset = to.len() as _; debug_assert!(!fuc.code.is_empty()); to.extend(&fuc.code); } let code_length = to.len() - init_len; for global in self.asm.globals.drain(..) { self.globals[global as usize].offset = to.len() as _; to.extend(&types.ins.globals[global as usize].data); } let data_length = to.len() - code_length - init_len; for func in self.asm.funcs.drain(..) { let fuc = &self.funcs[func as usize]; for rel in &fuc.relocs { let offset = match rel.target.expand() { ty::Kind::Func(fun) => self.funcs[fun as usize].offset, ty::Kind::Global(glo) => self.globals[glo as usize].offset, _ => unreachable!(), }; rel.reloc.apply_jump(to, offset, fuc.offset); } } AssemblySpec { code_length: code_length as _, data_length: data_length as _, entry: self.funcs[from as usize].offset, } } fn disasm<'a>( &'a self, mut sluce: &[u8], eca_handler: &mut dyn FnMut(&mut &[u8]), types: &'a Types, files: &'a [parser::Ast], output: &mut String, ) -> Result<(), hbbytecode::DisasmError<'a>> { use hbbytecode::DisasmItem; let functions = types .ins .funcs .iter() .zip(&self.funcs) .filter(|(_, f)| f.offset != u32::MAX) .map(|(f, fd)| { let name = if f.file != u32::MAX { let file = &files[f.file as usize]; file.ident_str(f.name) } else { "target_fn" }; (fd.offset, (name, fd.code.len() as u32, DisasmItem::Func)) }) .chain( types .ins .globals .iter() .zip(&self.globals) .filter(|(_, g)| g.offset != u32::MAX) .map(|(g, gd)| { let name = if g.file == u32::MAX { core::str::from_utf8(&g.data).unwrap_or("invalid utf-8") } else { let file = &files[g.file as usize]; file.ident_str(g.name) }; (gd.offset, (name, g.data.len() as Size, DisasmItem::Global)) }), ) .collect::>(); hbbytecode::disasm(&mut sluce, &functions, output, eca_handler) } fn emit_ct_body( &mut self, id: ty::Func, nodes: &mut Nodes, tys: &Types, files: &[parser::Ast], ) { self.emit_body(id, nodes, tys, files); let fd = &mut self.funcs[id as usize]; fd.code.truncate(fd.code.len() - instrs::jala(0, 0, 0).0); emit(&mut fd.code, instrs::tx()); } fn emit_body(&mut self, id: ty::Func, nodes: &mut Nodes, tys: &Types, files: &[parser::Ast]) { let sig = tys.ins.funcs[id as usize].sig.unwrap(); debug_assert!(self.code.is_empty()); self.offsets.clear(); self.offsets.resize(nodes.values.len(), Offset::MAX); let mut stack_size = 0; '_compute_stack: { let mems = mem::take(&mut nodes[MEM].outputs); for &stck in mems.iter() { if !matches!(nodes[stck].kind, Kind::Stck | Kind::Arg) { debug_assert_matches!( nodes[stck].kind, Kind::Phi | Kind::Return | Kind::Load | Kind::Call { .. } | Kind::Stre ); continue; } stack_size += tys.size_of(nodes[stck].ty); self.offsets[stck as usize] = stack_size; } for &stck in mems.iter() { if !matches!(nodes[stck].kind, Kind::Stck | Kind::Arg) { continue; } self.offsets[stck as usize] = stack_size - self.offsets[stck as usize]; } nodes[MEM].outputs = mems; } let (saved, tail) = self.emit_body_code(nodes, sig, tys, files); //let (saved, tail) = self.emit_body_code_my(nodes, sig, tys, files); if let Some(last_ret) = self.ret_relocs.last() && last_ret.offset as usize == self.code.len() - 5 && self .jump_relocs .last() .map_or(true, |&(r, _)| self.offsets[r as usize] as usize != self.code.len()) { self.code.truncate(self.code.len() - 5); self.ret_relocs.pop(); } // FIXME: maybe do this incrementally for (nd, rel) in self.jump_relocs.drain(..) { let offset = self.offsets[nd as usize]; //debug_assert!(offset < self.code.len() as u32 - 1); rel.apply_jump(&mut self.code, offset, 0); } let end = self.code.len(); for ret_rel in self.ret_relocs.drain(..) { ret_rel.apply_jump(&mut self.code, end as _, 0); } let mut stripped_prelude_size = 0; '_close_function: { let pushed = (saved as i64 + !tail as i64) * 8; let stack = stack_size as i64; let add_len = instrs::addi64(0, 0, 0).0; let st_len = instrs::st(0, 0, 0, 0).0; match (pushed, stack) { (0, 0) => { stripped_prelude_size = add_len + st_len; self.code.drain(0..stripped_prelude_size); break '_close_function; } (0, stack) => { write_reloc(&mut self.code, 3, -stack, 8); stripped_prelude_size = st_len; let end = add_len + st_len; self.code.drain(add_len..end); self.emit(instrs::addi64(reg::STACK_PTR, reg::STACK_PTR, stack as _)); break '_close_function; } _ => {} } write_reloc(&mut self.code, 3, -(pushed + stack), 8); write_reloc(&mut self.code, 3 + 8 + 3, stack, 8); write_reloc(&mut self.code, 3 + 8 + 3 + 8, pushed, 2); self.emit(instrs::ld( reg::RET_ADDR + tail as u8, reg::STACK_PTR, stack as _, pushed as _, )); self.emit(instrs::addi64(reg::STACK_PTR, reg::STACK_PTR, (pushed + stack) as _)); } self.relocs.iter_mut().for_each(|r| r.reloc.offset -= stripped_prelude_size as u32); if sig.ret != ty::Id::NEVER { self.emit(instrs::jala(reg::ZERO, reg::RET_ADDR, 0)); } if self.funcs.get(id as usize).is_none() { self.funcs.resize_with(id as usize + 1, Default::default); } self.funcs[id as usize].code = mem::take(&mut self.code); self.funcs[id as usize].relocs = mem::take(&mut self.relocs); debug_assert_eq!(self.ret_relocs.len(), 0); debug_assert_eq!(self.relocs.len(), 0); debug_assert_eq!(self.jump_relocs.len(), 0); debug_assert_eq!(self.code.len(), 0); } } impl Nodes { fn is_never_used(&self, nid: Nid, tys: &Types) -> bool { let node = &self[nid]; match node.kind { Kind::CInt { .. } => node.outputs.iter().all(|&o| { matches!(self[o].kind, Kind::BinOp { op } if op.imm_binop(self[o].ty).is_some() && self.is_const(self[o].inputs[2]) && op.cond_op(self[o].ty).is_none()) }), Kind::BinOp { op: TokenKind::Add | TokenKind::Sub } => { self[node.inputs[1]].lock_rc != 0 || (self.is_const(node.inputs[2]) && node.outputs.iter().all(|&n| self[n].uses_direct_offset_of(nid, tys))) } Kind::BinOp { op } => { op.cond_op(node.ty).is_some() && node.outputs.iter().all(|&n| self[n].kind == Kind::If) } Kind::Stck if tys.size_of(node.ty) == 0 => true, Kind::Stck | Kind::Arg => node.outputs.iter().all(|&n| { self[n].uses_direct_offset_of(nid, tys) || (matches!(self[n].kind, Kind::BinOp { op: TokenKind::Add }) && self.is_never_used(n, tys)) }), Kind::Load { .. } => node.ty.loc(tys) == Loc::Stack, _ => false, } } } impl Node { fn uses_direct_offset_of(&self, nid: Nid, tys: &Types) -> bool { ((self.kind == Kind::Stre && self.inputs[2] == nid) || (self.kind == Kind::Load && self.inputs[1] == nid)) && self.ty.loc(tys) == Loc::Reg } } impl TokenKind { fn cmp_against(self) -> Option { Some(match self { TokenKind::Le | TokenKind::Gt => 1, TokenKind::Ne | TokenKind::Eq => 0, TokenKind::Ge | TokenKind::Lt => (-1i64) as _, _ => return None, }) } fn float_cmp(self, ty: ty::Id) -> Option EncodedInstr> { if !ty.is_float() { return None; } let size = ty.simple_size().unwrap(); let ops = match self { TokenKind::Gt => [instrs::fcmpgt32, instrs::fcmpgt64], TokenKind::Lt => [instrs::fcmplt32, instrs::fcmplt64], _ => return None, }; Some(ops[size.ilog2() as usize - 2]) } #[expect(clippy::type_complexity)] fn cond_op(self, ty: ty::Id) -> Option<(fn(u8, u8, i16) -> EncodedInstr, bool)> { if ty.is_float() { return None; } let signed = ty.is_signed(); Some(( match self { Self::Le if signed => instrs::jgts, Self::Le => instrs::jgtu, Self::Lt if signed => instrs::jlts, Self::Lt => instrs::jltu, Self::Ge if signed => instrs::jlts, Self::Ge => instrs::jltu, Self::Gt if signed => instrs::jgts, Self::Gt => instrs::jgtu, Self::Eq => instrs::jne, Self::Ne => instrs::jeq, _ => return None, }, matches!(self, Self::Lt | TokenKind::Gt), )) } fn binop(self, ty: ty::Id) -> Option EncodedInstr> { let size = ty.simple_size().unwrap(); if ty.is_integer() || ty == ty::Id::BOOL || ty.is_pointer() { macro_rules! div { ($($op:ident),*) => {[$(|a, b, c| $op(a, 0, b, c)),*]}; } macro_rules! rem { ($($op:ident),*) => {[$(|a, b, c| $op(0, a, b, c)),*]}; } let signed = ty.is_signed(); let ops = match self { Self::Add => [add8, add16, add32, add64], Self::Sub => [sub8, sub16, sub32, sub64], Self::Mul => [mul8, mul16, mul32, mul64], Self::Div if signed => div!(dirs8, dirs16, dirs32, dirs64), Self::Div => div!(diru8, diru16, diru32, diru64), Self::Mod if signed => rem!(dirs8, dirs16, dirs32, dirs64), Self::Mod => rem!(diru8, diru16, diru32, diru64), Self::Band => return Some(and), Self::Bor => return Some(or), Self::Xor => return Some(xor), Self::Shl => [slu8, slu16, slu32, slu64], Self::Shr if signed => [srs8, srs16, srs32, srs64], Self::Shr => [sru8, sru16, sru32, sru64], _ => return None, }; Some(ops[size.ilog2() as usize]) } else { debug_assert!(ty.is_float(), "{self} {ty:?}"); let ops = match self { Self::Add => [fadd32, fadd64], Self::Sub => [fsub32, fsub64], Self::Mul => [fmul32, fmul64], Self::Div => [fdiv32, fdiv64], _ => return None, }; Some(ops[size.ilog2() as usize - 2]) } } fn imm_binop(self, ty: ty::Id) -> Option EncodedInstr> { macro_rules! def_op { ($name:ident |$a:ident, $b:ident, $c:ident| $($tt:tt)*) => { macro_rules! $name { ($$($$op:ident),*) => { [$$( |$a, $b, $c: u64| $$op($($tt)*), )*] } } }; } if ty.is_float() { return None; } def_op!(basic_op | a, b, c | a, b, c as _); def_op!(sub_op | a, b, c | a, b, c.wrapping_neg() as _); let signed = ty.is_signed(); let ops = match self { Self::Add => basic_op!(addi8, addi16, addi32, addi64), Self::Sub => sub_op!(addi8, addi16, addi32, addi64), Self::Mul => basic_op!(muli8, muli16, muli32, muli64), Self::Band => return Some(andi), Self::Bor => return Some(ori), Self::Xor => return Some(xori), Self::Shr if signed => basic_op!(srui8, srui16, srui32, srui64), Self::Shr => basic_op!(srui8, srui16, srui32, srui64), Self::Shl => basic_op!(slui8, slui16, slui32, slui64), _ => return None, }; let size = ty.simple_size().unwrap(); Some(ops[size.ilog2() as usize]) } fn unop(&self, dst: ty::Id, src: ty::Id) -> Option EncodedInstr> { let src_idx = src.simple_size().unwrap().ilog2() as usize - 2; Some(match self { Self::Sub => instrs::neg, Self::Float if dst.is_float() && src.is_integer() => { debug_assert_eq!(dst.simple_size(), src.simple_size()); [instrs::itf32, instrs::itf64][src_idx] } Self::Number if src.is_float() && dst.is_integer() => { [|a, b| instrs::fti32(a, b, 1), |a, b| instrs::fti64(a, b, 1)][src_idx] } Self::Float if dst.is_float() && src.is_float() => { [instrs::fc32t64, |a, b| instrs::fc64t32(a, b, 1)][src_idx] } _ => return None, }) } } type EncodedInstr = (usize, [u8; instrs::MAX_SIZE]); fn emit(out: &mut Vec, (len, instr): EncodedInstr) { out.extend_from_slice(&instr[..len]); } fn binary_prelude(to: &mut Vec) { emit(to, instrs::jal(reg::RET_ADDR, reg::ZERO, 0)); emit(to, instrs::tx()); } #[derive(Default)] pub struct LoggedMem { pub mem: hbvm::mem::HostMemory, op_buf: Vec, disp_buf: String, prev_instr: Option, } impl LoggedMem { unsafe fn display_instr(&mut self, instr: hbbytecode::Instr, addr: hbvm::mem::Address) { let novm: *const hbvm::Vm = core::ptr::null(); let offset = core::ptr::addr_of!((*novm).memory) as usize; let regs = unsafe { &*core::ptr::addr_of!( (*(((self as *mut _ as *mut u8).sub(offset)) as *const hbvm::Vm)) .registers ) }; let mut bytes = core::slice::from_raw_parts( (addr.get() - 1) as *const u8, core::mem::size_of::() + 1, ); use core::fmt::Write; hbbytecode::parse_args(&mut bytes, instr, &mut self.op_buf).unwrap(); debug_assert!(bytes.is_empty()); self.disp_buf.clear(); write!(self.disp_buf, "{:<10}", format!("{instr:?}")).unwrap(); for (i, op) in self.op_buf.drain(..).enumerate() { if i != 0 { write!(self.disp_buf, ", ").unwrap(); } write!(self.disp_buf, "{op:?}").unwrap(); if let hbbytecode::Oper::R(r) = op { write!(self.disp_buf, "({})", regs[r as usize].0).unwrap() } } log::trace!("read-typed: {:x}: {}", addr.get(), self.disp_buf); } } 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::trace!( "load: {:x} {}", addr.get(), AsHex(core::slice::from_raw_parts(addr.get() as *const u8, count)) ); self.mem.load(addr, target, count) } unsafe fn store( &mut self, addr: hbvm::mem::Address, source: *const u8, count: usize, ) -> Result<(), hbvm::mem::StoreError> { log::trace!( "store: {:x} {}", addr.get(), AsHex(core::slice::from_raw_parts(source, count)) ); self.mem.store(addr, source, count) } unsafe fn prog_read(&mut self, addr: hbvm::mem::Address) -> T { if log::log_enabled!(log::Level::Trace) { if core::any::TypeId::of::() == core::any::TypeId::of::() { if let Some(instr) = self.prev_instr { self.display_instr::<()>(instr, addr); } self.prev_instr = hbbytecode::Instr::try_from(*(addr.get() as *const u8)).ok(); } else { let instr = self.prev_instr.take().unwrap(); self.display_instr::(instr, addr); } } self.mem.prog_read(addr) } } struct AsHex<'a>(&'a [u8]); impl core::fmt::Display for AsHex<'_> { fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result { for &b in self.0 { write!(f, "{b:02x}")?; } Ok(()) } } const VM_STACK_SIZE: usize = 1024 * 64; pub struct Comptime { pub vm: hbvm::Vm, stack: Box<[u8; VM_STACK_SIZE]>, pub code: Vec, depth: usize, } impl Comptime { pub fn run(&mut self, ret_loc: &mut [u8], offset: u32) -> u64 { self.vm.write_reg(reg::RET, ret_loc.as_mut_ptr() as u64); let prev_pc = self.push_pc(offset); loop { match self.vm.run().expect("TODO") { hbvm::VmRunOk::End => break, hbvm::VmRunOk::Timer => todo!(), hbvm::VmRunOk::Ecall => todo!(), hbvm::VmRunOk::Breakpoint => todo!(), } } self.pop_pc(prev_pc); if let len @ 1..=8 = ret_loc.len() { ret_loc.copy_from_slice(&self.vm.read_reg(reg::RET).0.to_ne_bytes()[..len]) } self.vm.read_reg(reg::RET).0 } pub fn reset(&mut self) { let ptr = unsafe { self.stack.as_mut_ptr().cast::().add(VM_STACK_SIZE) as u64 }; self.vm.registers.fill(hbvm::value::Value(0)); self.vm.write_reg(reg::STACK_PTR, ptr); self.vm.pc = hbvm::mem::Address::new(self.code.as_ptr() as u64 + HEADER_SIZE as u64); } fn push_pc(&mut self, offset: Offset) -> hbvm::mem::Address { let entry = &mut self.code[offset as usize] as *mut _ as _; core::mem::replace(&mut self.vm.pc, hbvm::mem::Address::new(entry)) - self.code.as_ptr() as usize } fn pop_pc(&mut self, prev_pc: hbvm::mem::Address) { self.vm.pc = prev_pc + self.code.as_ptr() as usize; } pub fn clear(&mut self) { self.code.clear(); } #[must_use] pub fn active(&self) -> bool { self.depth != 0 } pub fn activate(&mut self) { self.depth += 1; } pub fn deactivate(&mut self) { self.depth -= 1; } } impl Default for Comptime { fn default() -> Self { let mut stack = Box::<[u8; VM_STACK_SIZE]>::new_uninit(); let mut vm = hbvm::Vm::default(); let ptr = unsafe { stack.as_mut_ptr().cast::().add(VM_STACK_SIZE) as u64 }; vm.write_reg(reg::STACK_PTR, ptr); Self { vm, stack: unsafe { stack.assume_init() }, code: Default::default(), depth: 0 } } } const HEADER_SIZE: usize = core::mem::size_of::(); #[repr(packed)] #[expect(dead_code)] pub struct AbleOsExecutableHeader { magic_number: [u8; 3], executable_version: u32, code_length: u64, data_length: u64, debug_length: u64, config_length: u64, metadata_length: u64, } #[cfg(test)] pub fn test_run_vm(out: &[u8], output: &mut String) { use core::fmt::Write; let mut stack = [0_u64; 1024 * 20]; let mut vm = unsafe { hbvm::Vm::<_, { 1024 * 100 }>::new( LoggedMem::default(), hbvm::mem::Address::new(out.as_ptr() as u64).wrapping_add(HEADER_SIZE), ) }; vm.write_reg(reg::STACK_PTR, unsafe { stack.as_mut_ptr().add(stack.len()) } as u64); let stat = loop { match vm.run() { Ok(hbvm::VmRunOk::End) => break Ok(()), Ok(hbvm::VmRunOk::Ecall) => match vm.read_reg(2).0 { 1 => writeln!(output, "ev: Ecall").unwrap(), // compatibility with a test 69 => { let [size, align] = [vm.read_reg(3).0 as usize, vm.read_reg(4).0 as usize]; let layout = core::alloc::Layout::from_size_align(size, align).unwrap(); let ptr = unsafe { alloc::alloc::alloc(layout) }; vm.write_reg(1, ptr as u64); } 96 => { let [ptr, size, align] = [ vm.read_reg(3).0 as usize, vm.read_reg(4).0 as usize, vm.read_reg(5).0 as usize, ]; let layout = core::alloc::Layout::from_size_align(size, align).unwrap(); unsafe { alloc::alloc::dealloc(ptr as *mut u8, layout) }; } 3 => vm.write_reg(1, 42), 8 => {} unknown => writeln!(output, "unknown ecall: {unknown:?}").unwrap(), }, Ok(hbvm::VmRunOk::Timer) => { writeln!(output, "timed out").unwrap(); break Ok(()); } Ok(ev) => writeln!(output, "ev: {:?}", ev).unwrap(), Err(e) => break Err(e), } }; writeln!(output, "code size: {}", out.len() - HEADER_SIZE).unwrap(); writeln!(output, "ret: {:?}", vm.read_reg(1).0).unwrap(); writeln!(output, "status: {:?}", stat).unwrap(); }