nac3/nac3core/src/codegen/mod.rs

534 lines
18 KiB
Rust
Raw Normal View History

2021-08-11 14:37:26 +08:00
use crate::{
2021-11-20 19:50:25 +08:00
symbol_resolver::{StaticValue, SymbolResolver},
2021-09-30 17:07:48 +08:00
toplevel::{TopLevelContext, TopLevelDef},
2021-08-11 14:37:26 +08:00
typecheck::{
type_inferencer::{CodeLocation, PrimitiveStore},
typedef::{CallId, FuncArg, Type, TypeEnum, Unifier},
2021-08-11 14:37:26 +08:00
},
};
use crossbeam::channel::{unbounded, Receiver, Sender};
2021-08-11 14:37:26 +08:00
use inkwell::{
basic_block::BasicBlock,
builder::Builder,
context::Context,
module::Module,
passes::{PassManager, PassManagerBuilder},
2021-08-11 14:37:26 +08:00
types::{BasicType, BasicTypeEnum},
values::{BasicValueEnum, FunctionValue, PhiValue, PointerValue},
AddressSpace, OptimizationLevel,
2021-08-11 14:37:26 +08:00
};
use itertools::Itertools;
2021-11-03 17:11:00 +08:00
use nac3parser::ast::{Stmt, StrRef};
2021-11-20 19:50:25 +08:00
use parking_lot::{Condvar, Mutex};
use std::collections::{HashMap, HashSet};
2021-08-13 16:20:14 +08:00
use std::sync::{
atomic::{AtomicBool, Ordering},
Arc,
};
use std::thread;
2021-08-11 14:37:26 +08:00
pub mod concrete_type;
pub mod expr;
mod generator;
pub mod irrt;
2022-02-21 18:27:46 +08:00
pub mod stmt;
2021-08-11 14:37:26 +08:00
2021-08-12 13:55:15 +08:00
#[cfg(test)]
mod test;
use concrete_type::{ConcreteType, ConcreteTypeEnum, ConcreteTypeStore};
pub use generator::{CodeGenerator, DefaultCodeGenerator};
2021-11-20 19:50:25 +08:00
#[derive(Default)]
pub struct StaticValueStore {
pub lookup: HashMap<Vec<(usize, u64)>, usize>,
pub store: Vec<HashMap<usize, Arc<dyn StaticValue + Send + Sync>>>,
}
pub type VarValue<'ctx> = (PointerValue<'ctx>, Option<Arc<dyn StaticValue + Send + Sync>>, i64);
2021-08-12 13:55:15 +08:00
pub struct CodeGenContext<'ctx, 'a> {
2021-08-11 14:37:26 +08:00
pub ctx: &'ctx Context,
pub builder: Builder<'ctx>,
pub module: Module<'ctx>,
2021-08-12 13:55:15 +08:00
pub top_level: &'a TopLevelContext,
2021-08-11 14:37:26 +08:00
pub unifier: Unifier,
2021-10-16 18:08:13 +08:00
pub resolver: Arc<dyn SymbolResolver + Send + Sync>,
2021-11-20 19:50:25 +08:00
pub static_value_store: Arc<Mutex<StaticValueStore>>,
pub var_assignment: HashMap<StrRef, VarValue<'ctx>>,
2021-08-11 14:37:26 +08:00
pub type_cache: HashMap<Type, BasicTypeEnum<'ctx>>,
pub primitives: PrimitiveStore,
2021-09-22 16:04:25 +08:00
pub calls: Arc<HashMap<CodeLocation, CallId>>,
pub registry: &'a WorkerRegistry,
// const string cache
pub const_strings: HashMap<String, BasicValueEnum<'ctx>>,
2021-08-11 14:37:26 +08:00
// stores the alloca for variables
pub init_bb: BasicBlock<'ctx>,
// the first one is the test_bb, and the second one is bb after the loop
pub loop_target: Option<(BasicBlock<'ctx>, BasicBlock<'ctx>)>,
// unwind target bb
pub unwind_target: Option<BasicBlock<'ctx>>,
// return target bb, just emit ret if no such target
pub return_target: Option<BasicBlock<'ctx>>,
pub return_buffer: Option<PointerValue<'ctx>>,
// outer catch clauses
pub outer_catch_clauses:
Option<(Vec<Option<BasicValueEnum<'ctx>>>, BasicBlock<'ctx>, PhiValue<'ctx>)>,
}
impl<'ctx, 'a> CodeGenContext<'ctx, 'a> {
pub fn is_terminated(&self) -> bool {
self.builder.get_insert_block().unwrap().get_terminator().is_some()
}
2021-08-11 14:37:26 +08:00
}
type Fp = Box<dyn Fn(&Module) + Send + Sync>;
pub struct WithCall {
fp: Fp,
}
impl WithCall {
pub fn new(fp: Fp) -> WithCall {
WithCall { fp }
}
2021-08-13 16:20:14 +08:00
pub fn run<'ctx>(&self, m: &Module<'ctx>) {
(self.fp)(m)
}
}
pub struct WorkerRegistry {
sender: Arc<Sender<Option<CodeGenTask>>>,
receiver: Arc<Receiver<Option<CodeGenTask>>>,
2021-08-13 16:20:14 +08:00
panicked: AtomicBool,
task_count: Mutex<usize>,
thread_count: usize,
wait_condvar: Condvar,
2021-11-20 19:50:25 +08:00
top_level_ctx: Arc<TopLevelContext>,
static_value_store: Arc<Mutex<StaticValueStore>>,
}
impl WorkerRegistry {
pub fn create_workers<G: CodeGenerator + Send + 'static>(
generators: Vec<Box<G>>,
top_level_ctx: Arc<TopLevelContext>,
f: Arc<WithCall>,
2021-08-13 16:20:14 +08:00
) -> (Arc<WorkerRegistry>, Vec<thread::JoinHandle<()>>) {
let (sender, receiver) = unbounded();
let task_count = Mutex::new(0);
let wait_condvar = Condvar::new();
2021-11-20 19:50:25 +08:00
// init: 0 to be empty
let mut static_value_store: StaticValueStore = Default::default();
static_value_store.lookup.insert(Default::default(), 0);
static_value_store.store.push(Default::default());
let registry = Arc::new(WorkerRegistry {
sender: Arc::new(sender),
receiver: Arc::new(receiver),
thread_count: generators.len(),
2021-08-13 16:20:14 +08:00
panicked: AtomicBool::new(false),
2021-11-20 19:50:25 +08:00
static_value_store: Arc::new(Mutex::new(static_value_store)),
task_count,
wait_condvar,
2021-11-20 19:50:25 +08:00
top_level_ctx,
});
2021-08-13 16:20:14 +08:00
let mut handles = Vec::new();
for mut generator in generators.into_iter() {
let registry = registry.clone();
2021-08-13 16:20:14 +08:00
let registry2 = registry.clone();
let f = f.clone();
2021-08-13 16:20:14 +08:00
let handle = thread::spawn(move || {
2021-11-20 19:50:25 +08:00
registry.worker_thread(generator.as_mut(), f);
});
2021-08-13 16:20:14 +08:00
let handle = thread::spawn(move || {
if let Err(e) = handle.join() {
if let Some(e) = e.downcast_ref::<&'static str>() {
eprintln!("Got an error: {}", e);
} else {
eprintln!("Got an unknown error: {:?}", e);
}
registry2.panicked.store(true, Ordering::SeqCst);
registry2.wait_condvar.notify_all();
}
});
handles.push(handle);
}
2021-08-13 16:20:14 +08:00
(registry, handles)
}
2021-08-13 16:20:14 +08:00
pub fn wait_tasks_complete(&self, handles: Vec<thread::JoinHandle<()>>) {
{
let mut count = self.task_count.lock();
while *count != 0 {
2021-08-13 16:20:14 +08:00
if self.panicked.load(Ordering::SeqCst) {
break;
}
self.wait_condvar.wait(&mut count);
}
}
for _ in 0..self.thread_count {
self.sender.send(None).unwrap();
}
{
let mut count = self.task_count.lock();
while *count != self.thread_count {
2021-08-13 16:20:14 +08:00
if self.panicked.load(Ordering::SeqCst) {
break;
}
self.wait_condvar.wait(&mut count);
}
}
2021-08-13 16:20:14 +08:00
for handle in handles {
handle.join().unwrap();
}
if self.panicked.load(Ordering::SeqCst) {
panic!("tasks panicked");
}
}
pub fn add_task(&self, task: CodeGenTask) {
*self.task_count.lock() += 1;
self.sender.send(Some(task)).unwrap();
}
2021-11-20 19:50:25 +08:00
fn worker_thread<G: CodeGenerator>(&self, generator: &mut G, f: Arc<WithCall>) {
let context = Context::create();
let mut builder = context.create_builder();
let module = context.create_module(generator.get_name());
let pass_builder = PassManagerBuilder::create();
pass_builder.set_optimization_level(OptimizationLevel::Default);
let passes = PassManager::create(&module);
pass_builder.populate_function_pass_manager(&passes);
let mut errors = HashSet::new();
while let Some(task) = self.receiver.recv().unwrap() {
let tmp_module = context.create_module("tmp");
match gen_func(&context, generator, self, builder, tmp_module, task) {
Ok(result) => {
builder = result.0;
passes.run_on(&result.2);
module.link_in_module(result.1).unwrap();
}
Err((old_builder, e)) => {
builder = old_builder;
errors.insert(e);
}
}
*self.task_count.lock() -= 1;
self.wait_condvar.notify_all();
}
if !errors.is_empty() {
panic!("Codegen error: {}", errors.into_iter().sorted().join("\n----------\n"));
}
let result = module.verify();
if let Err(err) = result {
println!("{}", module.print_to_string().to_str().unwrap());
println!("{}", err);
panic!()
}
f.run(&module);
let mut lock = self.task_count.lock();
*lock += 1;
self.wait_condvar.notify_all();
}
}
2021-08-11 14:37:26 +08:00
pub struct CodeGenTask {
pub subst: Vec<(Type, ConcreteType)>,
pub store: ConcreteTypeStore,
2021-08-11 14:37:26 +08:00
pub symbol_name: String,
pub signature: ConcreteType,
2021-09-22 16:04:25 +08:00
pub body: Arc<Vec<Stmt<Option<Type>>>>,
pub calls: Arc<HashMap<CodeLocation, CallId>>,
pub unifier_index: usize,
2021-10-16 18:08:13 +08:00
pub resolver: Arc<dyn SymbolResolver + Send + Sync>,
2021-11-20 19:50:25 +08:00
pub id: usize,
2021-08-11 14:37:26 +08:00
}
fn get_llvm_type<'ctx>(
ctx: &'ctx Context,
generator: &mut dyn CodeGenerator,
2021-08-11 14:37:26 +08:00
unifier: &mut Unifier,
top_level: &TopLevelContext,
type_cache: &mut HashMap<Type, BasicTypeEnum<'ctx>>,
ty: Type,
) -> BasicTypeEnum<'ctx> {
use TypeEnum::*;
// we assume the type cache should already contain primitive types,
// and they should be passed by value instead of passing as pointer.
2021-08-12 16:36:23 +08:00
type_cache.get(&unifier.get_representative(ty)).cloned().unwrap_or_else(|| {
let ty_enum = unifier.get_ty(ty);
let result = match &*ty_enum {
2021-08-12 16:36:23 +08:00
TObj { obj_id, fields, .. } => {
// check to avoid treating primitives as classes
if obj_id.0 <= 7 {
unreachable!();
}
2021-08-12 16:36:23 +08:00
// a struct with fields in the order of declaration
let top_level_defs = top_level.definitions.read();
let definition = top_level_defs.get(obj_id.0).unwrap();
2022-02-21 18:27:46 +08:00
let ty = if let TopLevelDef::Class { name, fields: fields_list, .. } =
&*definition.read()
2021-08-12 16:36:23 +08:00
{
let struct_type = ctx.opaque_struct_type(&name.to_string());
2021-08-12 16:36:23 +08:00
let fields = fields_list
.iter()
2022-02-21 18:27:46 +08:00
.map(|f| {
get_llvm_type(
ctx,
generator,
unifier,
top_level,
type_cache,
fields[&f.0].0,
)
})
2021-08-12 16:36:23 +08:00
.collect_vec();
struct_type.set_body(&fields, false);
struct_type.ptr_type(AddressSpace::Generic).into()
2021-08-12 16:36:23 +08:00
} else {
unreachable!()
};
ty
}
TTuple { ty } => {
// a struct with fields in the order present in the tuple
let fields = ty
2021-08-11 14:37:26 +08:00
.iter()
.map(|ty| get_llvm_type(ctx, generator, unifier, top_level, type_cache, *ty))
2021-08-11 14:37:26 +08:00
.collect_vec();
2022-01-20 04:14:22 +08:00
ctx.struct_type(&fields, false).into()
2021-08-12 16:36:23 +08:00
}
TList { ty } => {
// a struct with an integer and a pointer to an array
2022-02-21 18:27:46 +08:00
let element_type =
get_llvm_type(ctx, generator, unifier, top_level, type_cache, *ty);
let fields = [
element_type.ptr_type(AddressSpace::Generic).into(),
generator.get_size_type(ctx).into(),
];
2021-08-12 16:36:23 +08:00
ctx.struct_type(&fields, false).ptr_type(AddressSpace::Generic).into()
}
TVirtual { .. } => unimplemented!(),
_ => unreachable!("{}", ty_enum.get_type_name()),
};
type_cache.insert(unifier.get_representative(ty), result);
result
2021-08-11 14:37:26 +08:00
})
}
pub fn gen_func<'ctx, G: CodeGenerator>(
2021-08-12 16:36:23 +08:00
context: &'ctx Context,
generator: &mut G,
registry: &WorkerRegistry,
2021-08-12 16:36:23 +08:00
builder: Builder<'ctx>,
module: Module<'ctx>,
task: CodeGenTask,
) -> Result<(Builder<'ctx>, Module<'ctx>, FunctionValue<'ctx>), (Builder<'ctx>, String)> {
2021-11-20 19:50:25 +08:00
let top_level_ctx = registry.top_level_ctx.clone();
let static_value_store = registry.static_value_store.clone();
2021-08-11 14:37:26 +08:00
let (mut unifier, primitives) = {
let (unifier, primitives) = &top_level_ctx.unifiers.read()[task.unifier_index];
(Unifier::from_shared_unifier(unifier), *primitives)
2021-08-11 14:37:26 +08:00
};
let mut cache = HashMap::new();
2021-08-11 14:37:26 +08:00
for (a, b) in task.subst.iter() {
// this should be unification between variables and concrete types
// and should not cause any problem...
let b = task.store.to_unifier_type(&mut unifier, &primitives, *b, &mut cache);
2022-02-21 18:27:46 +08:00
unifier
.unify(*a, b)
.or_else(|err| {
if matches!(&*unifier.get_ty(*a), TypeEnum::TRigidVar { .. }) {
unifier.replace_rigid_var(*a, b);
Ok(())
} else {
Err(err)
}
})
.unwrap()
2021-08-11 14:37:26 +08:00
}
// rebuild primitive store with unique representatives
let primitives = PrimitiveStore {
int32: unifier.get_representative(primitives.int32),
int64: unifier.get_representative(primitives.int64),
float: unifier.get_representative(primitives.float),
bool: unifier.get_representative(primitives.bool),
none: unifier.get_representative(primitives.none),
2021-10-23 23:53:36 +08:00
range: unifier.get_representative(primitives.range),
2021-11-02 23:22:37 +08:00
str: unifier.get_representative(primitives.str),
exception: unifier.get_representative(primitives.exception),
2021-08-11 14:37:26 +08:00
};
let mut type_cache: HashMap<_, _> = [
(primitives.int32, context.i32_type().into()),
(primitives.int64, context.i64_type().into()),
(primitives.float, context.f64_type().into()),
(primitives.bool, context.bool_type().into()),
(primitives.str, {
let str_type = context.opaque_struct_type("str");
let fields = [
context.i8_type().ptr_type(AddressSpace::Generic).into(),
generator.get_size_type(context).into(),
];
str_type.set_body(&fields, false);
str_type.into()
}),
2022-02-21 18:27:46 +08:00
(primitives.range, context.i32_type().array_type(3).ptr_type(AddressSpace::Generic).into()),
2021-08-11 14:37:26 +08:00
]
.iter()
.cloned()
.collect();
type_cache.insert(primitives.exception, {
let exception = context.opaque_struct_type("Exception");
let int32 = context.i32_type().into();
let int64 = context.i64_type().into();
let str_ty = *type_cache.get(&primitives.str).unwrap();
2022-02-21 18:27:46 +08:00
let fields = [int32, str_ty, int32, int32, str_ty, str_ty, int64, int64, int64];
exception.set_body(&fields, false);
exception.ptr_type(AddressSpace::Generic).into()
});
2021-08-11 14:37:26 +08:00
let (args, ret) = if let ConcreteTypeEnum::TFunc { args, ret, .. } =
task.store.get(task.signature)
{
(
args.iter()
.map(|arg| FuncArg {
name: arg.name,
ty: task.store.to_unifier_type(&mut unifier, &primitives, arg.ty, &mut cache),
default_value: arg.default_value.clone(),
})
.collect_vec(),
task.store.to_unifier_type(&mut unifier, &primitives, *ret, &mut cache),
)
} else {
unreachable!()
};
let params = args
2021-08-11 14:37:26 +08:00
.iter()
.map(|arg| {
2022-02-21 18:27:46 +08:00
get_llvm_type(
context,
generator,
&mut unifier,
top_level_ctx.as_ref(),
&mut type_cache,
arg.ty,
)
.into()
2021-08-11 14:37:26 +08:00
})
.collect_vec();
let fn_type = if unifier.unioned(ret, primitives.none) {
2021-08-11 14:37:26 +08:00
context.void_type().fn_type(&params, false)
} else {
2022-02-21 18:27:46 +08:00
get_llvm_type(
context,
generator,
&mut unifier,
top_level_ctx.as_ref(),
&mut type_cache,
ret,
)
.fn_type(&params, false)
2021-08-11 14:37:26 +08:00
};
let symbol = &task.symbol_name;
2021-09-08 19:45:36 +08:00
let fn_val =
module.get_function(symbol).unwrap_or_else(|| module.add_function(symbol, fn_type, None));
if let Some(personality) = &top_level_ctx.personality_symbol {
2021-10-16 18:08:13 +08:00
let personality = module.get_function(personality).unwrap_or_else(|| {
let ty = context.i32_type().fn_type(&[], true);
2021-10-16 18:08:13 +08:00
module.add_function(personality, ty, None)
});
fn_val.set_personality_function(personality);
}
2021-08-11 14:37:26 +08:00
let init_bb = context.append_basic_block(fn_val, "init");
builder.position_at_end(init_bb);
let body_bb = context.append_basic_block(fn_val, "body");
let mut var_assignment = HashMap::new();
for (n, arg) in args.iter().enumerate() {
2021-08-11 14:37:26 +08:00
let param = fn_val.get_nth_param(n as u32).unwrap();
let alloca = builder.build_alloca(
2022-02-21 18:27:46 +08:00
get_llvm_type(
context,
generator,
&mut unifier,
top_level_ctx.as_ref(),
&mut type_cache,
arg.ty,
),
2021-09-22 17:19:27 +08:00
&arg.name.to_string(),
2021-08-11 14:37:26 +08:00
);
builder.build_store(alloca, param);
2021-11-20 19:50:25 +08:00
var_assignment.insert(arg.name, (alloca, None, 0));
}
let return_buffer = fn_type.get_return_type().map(|v| builder.build_alloca(v, "$ret"));
2021-11-20 19:50:25 +08:00
let static_values = {
let store = registry.static_value_store.lock();
store.store[task.id].clone()
};
for (k, v) in static_values.into_iter() {
let (_, static_val, _) = var_assignment.get_mut(&args[k].name).unwrap();
*static_val = Some(v);
2021-08-11 14:37:26 +08:00
}
2021-11-20 19:50:25 +08:00
2021-08-11 14:37:26 +08:00
builder.build_unconditional_branch(body_bb);
builder.position_at_end(body_bb);
let mut code_gen_context = CodeGenContext {
2021-10-16 18:08:13 +08:00
ctx: context,
2021-08-11 14:37:26 +08:00
resolver: task.resolver,
top_level: top_level_ctx.as_ref(),
calls: task.calls,
loop_target: None,
return_target: None,
return_buffer,
unwind_target: None,
outer_catch_clauses: None,
const_strings: Default::default(),
registry,
2021-08-11 14:37:26 +08:00
var_assignment,
type_cache,
primitives,
init_bb,
builder,
module,
unifier,
2021-11-20 19:50:25 +08:00
static_value_store,
2021-08-11 14:37:26 +08:00
};
let mut err = None;
2021-08-12 13:55:15 +08:00
for stmt in task.body.iter() {
if let Err(e) = generator.gen_stmt(&mut code_gen_context, stmt) {
err = Some(e);
break;
}
if code_gen_context.is_terminated() {
2021-09-19 22:54:06 +08:00
break;
}
}
// after static analysis, only void functions can have no return at the end.
if !code_gen_context.is_terminated() {
2021-09-19 22:54:06 +08:00
code_gen_context.builder.build_return(None);
2021-08-11 14:37:26 +08:00
}
2021-08-12 13:55:15 +08:00
2021-08-13 13:33:59 +08:00
let CodeGenContext { builder, module, .. } = code_gen_context;
if let Some(e) = err {
return Err((builder, e));
}
2021-08-13 13:33:59 +08:00
Ok((builder, module, fn_val))
2021-08-11 14:37:26 +08:00
}