with parallel/sequential support

Behavior of parallel and sequential:
Each function call (indirectly, can be inside a sequential block) within a parallel
block will update the end variable to the maximum now_mu in the block.
Each function call directly inside a parallel block will reset the timeline after
execution. A parallel block within a sequential block (or not within any block) will
set the timeline to the max now_mu within the block (and the outer max now_mu will also
be updated).

Implementation: We track the start and end separately.
- If there is a start variable, it indicates that we are directly inside a
parallel block and we have to reset the timeline after every function call.
- If there is a end variable, it indicates that we are (indirectly) inside a
parallel block, and we should update the max end value.

Note: requires testing, it is difficult to inspect the output IR
escape-analysis
pca006132 2021-10-31 17:16:21 +08:00 committed by Gitea
parent 137efebb33
commit 083eacc268
8 changed files with 347 additions and 13 deletions

View File

@ -5,7 +5,8 @@ from numpy import int32, int64
import nac3artiq
__all__ = ["KernelInvariant", "extern", "kernel", "portable", "ms", "us", "ns", "Core", "TTLOut"]
__all__ = ["KernelInvariant", "extern", "kernel", "portable", "ms", "us", "ns",
"Core", "TTLOut", "parallel", "sequential"]
import device_db
@ -15,7 +16,6 @@ nac3 = nac3artiq.NAC3(core_arguments["target"])
allow_module_registration = True
registered_ids = set()
def KernelInvariant(t):
return t
@ -83,6 +83,14 @@ def rtio_input_timestamp(timeout_mu: int64, channel: int32) -> int64:
def rtio_input_data(channel: int32) -> int32:
raise NotImplementedError("syscall not simulated")
def at_mu(_):
raise NotImplementedError("at_mu not simulated")
def now_mu() -> int32:
raise NotImplementedError("now_mu not simulated")
def delay_mu(_):
raise NotImplementedError("delay_mu not simulated")
@kernel
class Core:
@ -169,3 +177,17 @@ class TTLOut:
self.on()
self.core.delay(duration)
self.off()
@portable
class KernelContextManager:
@kernel
def __enter__(self):
pass
@kernel
def __exit__(self):
pass
parallel = KernelContextManager()
sequential = KernelContextManager()

196
nac3artiq/src/codegen.rs Normal file
View File

@ -0,0 +1,196 @@
use nac3core::{
codegen::{expr::gen_call, stmt::gen_with, CodeGenContext, CodeGenerator},
toplevel::DefinitionId,
typecheck::typedef::{FunSignature, Type},
};
use rustpython_parser::ast::{Expr, ExprKind, Located, Stmt, StmtKind, StrRef};
use inkwell::values::BasicValueEnum;
use crate::timeline::TimeFns;
pub struct ArtiqCodeGenerator<'a> {
name: String,
name_counter: u32,
start: Option<Expr<Option<Type>>>,
end: Option<Expr<Option<Type>>>,
timeline: &'a (dyn TimeFns + Sync),
}
impl<'a> ArtiqCodeGenerator<'a> {
pub fn new(name: String, timeline: &'a (dyn TimeFns + Sync)) -> ArtiqCodeGenerator<'a> {
ArtiqCodeGenerator {
name,
name_counter: 0,
start: None,
end: None,
timeline,
}
}
}
impl<'b> CodeGenerator for ArtiqCodeGenerator<'b> {
fn get_name(&self) -> &str {
&self.name
}
fn gen_call<'ctx, 'a>(
&mut self,
ctx: &mut CodeGenContext<'ctx, 'a>,
obj: Option<(Type, BasicValueEnum<'ctx>)>,
fun: (&FunSignature, DefinitionId),
params: Vec<(Option<StrRef>, BasicValueEnum<'ctx>)>,
) -> Option<BasicValueEnum<'ctx>> {
let result = gen_call(self, ctx, obj, fun, params);
if let Some(end) = self.end.clone() {
let old_end = self.gen_expr(ctx, &end).unwrap();
let now = self.timeline.emit_now_mu(ctx);
let smax = ctx.module.get_function("llvm.smax.i64").unwrap_or_else(|| {
let i64 = ctx.ctx.i64_type();
ctx.module.add_function(
"llvm.smax.i64",
i64.fn_type(&[i64.into(), i64.into()], false),
None,
)
});
let max = ctx
.builder
.build_call(smax, &[old_end, now], "smax")
.try_as_basic_value()
.left()
.unwrap();
let end_store = self.gen_store_target(ctx, &end);
ctx.builder.build_store(end_store, max);
}
if let Some(start) = self.start.clone() {
let start_val = self.gen_expr(ctx, &start).unwrap();
self.timeline.emit_at_mu(ctx, start_val);
}
result
}
fn gen_with<'ctx, 'a>(
&mut self,
ctx: &mut CodeGenContext<'ctx, 'a>,
stmt: &Stmt<Option<Type>>,
) -> bool {
if let StmtKind::With { items, body, .. } = &stmt.node {
if items.len() == 1 && items[0].optional_vars.is_none() {
let item = &items[0];
// Behavior of parallel and sequential:
// Each function call (indirectly, can be inside a sequential block) within a parallel
// block will update the end variable to the maximum now_mu in the block.
// Each function call directly inside a parallel block will reset the timeline after
// execution. A parallel block within a sequential block (or not within any block) will
// set the timeline to the max now_mu within the block (and the outer max now_mu will also
// be updated).
//
// Implementation: We track the start and end separately.
// - If there is a start variable, it indicates that we are directly inside a
// parallel block and we have to reset the timeline after every function call.
// - If there is a end variable, it indicates that we are (indirectly) inside a
// parallel block, and we should update the max end value.
if let ExprKind::Name { id, ctx: name_ctx } = &item.context_expr.node {
if id == &"parallel".into() {
let old_start = self.start.take();
let old_end = self.end.take();
let now = if let Some(old_start) = &old_start {
self.gen_expr(ctx, old_start).unwrap()
} else {
self.timeline.emit_now_mu(ctx)
};
// Emulate variable allocation, as we need to use the CodeGenContext
// HashMap to store our variable due to lifetime limitation
// Note: we should be able to store variables directly if generic
// associative type is used by limiting the lifetime of CodeGenerator to
// the LLVM Context.
// The name is guaranteed to be unique as users cannot use this as variable
// name.
self.start = old_start.clone().or_else(|| {
let start = format!("with-{}-start", self.name_counter).into();
let start_expr = Located {
// location does not matter at this point
location: stmt.location,
node: ExprKind::Name {
id: start,
ctx: name_ctx.clone(),
},
custom: Some(ctx.primitives.int64),
};
let start = self.gen_store_target(ctx, &start_expr);
ctx.builder.build_store(start, now);
Some(start_expr)
});
let end = format!("with-{}-end", self.name_counter).into();
let end_expr = Located {
// location does not matter at this point
location: stmt.location,
node: ExprKind::Name {
id: end,
ctx: name_ctx.clone(),
},
custom: Some(ctx.primitives.int64),
};
let end = self.gen_store_target(ctx, &end_expr);
ctx.builder.build_store(end, now);
self.end = Some(end_expr);
self.name_counter += 1;
let mut exited = false;
for stmt in body.iter() {
if self.gen_stmt(ctx, stmt) {
exited = true;
break;
}
}
// set duration
let end_expr = self.end.take().unwrap();
let end_val = self.gen_expr(ctx, &end_expr).unwrap();
// inside an sequential block
if old_start.is_none() {
self.timeline.emit_at_mu(ctx, end_val);
}
// inside a parallel block, should update the outer max now_mu
if let Some(old_end) = &old_end {
let outer_end_val = self.gen_expr(ctx, old_end).unwrap();
let smax = ctx.module.get_function("llvm.smax.i64").unwrap_or_else(|| {
let i64 = ctx.ctx.i64_type();
ctx.module.add_function(
"llvm.smax.i64",
i64.fn_type(&[i64.into(), i64.into()], false),
None,
)
});
let max = ctx
.builder
.build_call(smax, &[end_val, outer_end_val], "smax")
.try_as_basic_value()
.left()
.unwrap();
let outer_end = self.gen_store_target(ctx, old_end);
ctx.builder.build_store(outer_end, max);
}
self.start = old_start;
self.end = old_end;
return exited;
} else if id == &"sequential".into() {
let start = self.start.take();
for stmt in body.iter() {
if self.gen_stmt(ctx, stmt) {
self.start = start;
return true;
}
}
self.start = start;
return false;
}
}
}
// not parallel/sequential
gen_with(self, ctx, stmt)
} else {
unreachable!()
}
}
}

View File

@ -19,18 +19,16 @@ use rustpython_parser::{
use parking_lot::{Mutex, RwLock};
use nac3core::{
codegen::{
concrete_type::ConcreteTypeStore, CodeGenTask, DefaultCodeGenerator, WithCall,
WorkerRegistry,
},
codegen::{concrete_type::ConcreteTypeStore, CodeGenTask, WithCall, WorkerRegistry},
symbol_resolver::SymbolResolver,
toplevel::{composer::TopLevelComposer, DefinitionId, GenCall, TopLevelContext, TopLevelDef},
typecheck::typedef::{FunSignature, FuncArg},
typecheck::{type_inferencer::PrimitiveStore, typedef::Type},
};
use crate::symbol_resolver::Resolver;
use crate::{codegen::ArtiqCodeGenerator, symbol_resolver::Resolver};
mod codegen;
mod symbol_resolver;
mod timeline;
@ -440,10 +438,14 @@ impl Nac3 {
)
.expect("couldn't write module to file");
})));
let time_fns: &(dyn TimeFns + Sync) = match isa {
Isa::RiscV => &timeline::NOW_PINNING_TIME_FNS,
Isa::CortexA9 => &timeline::EXTERN_TIME_FNS,
};
let thread_names: Vec<String> = (0..4).map(|i| format!("module{}", i)).collect();
let threads: Vec<_> = thread_names
.iter()
.map(|s| Box::new(DefaultCodeGenerator::new(s.to_string())))
.map(|s| Box::new(ArtiqCodeGenerator::new(s.to_string(), time_fns)))
.collect();
py.allow_threads(|| {

View File

@ -125,6 +125,14 @@ pub trait CodeGenerator {
gen_if(self, ctx, stmt)
}
fn gen_with<'ctx, 'a>(
&mut self,
ctx: &mut CodeGenContext<'ctx, 'a>,
stmt: &Stmt<Option<Type>>,
) -> bool {
gen_with(self, ctx, stmt)
}
/// Generate code for a statement
/// Return true if the statement must early return
fn gen_stmt<'ctx, 'a>(

View File

@ -28,9 +28,9 @@ use std::sync::{
use std::thread;
pub mod concrete_type;
mod expr;
pub mod expr;
pub mod stmt;
mod generator;
mod stmt;
#[cfg(test)]
mod test;

View File

@ -291,6 +291,15 @@ pub fn gen_if<'ctx, 'a, G: CodeGenerator + ?Sized>(
}
}
pub fn gen_with<'ctx, 'a, G: CodeGenerator + ?Sized>(
_: &mut G,
_: &mut CodeGenContext<'ctx, 'a>,
_: &Stmt<Option<Type>>,
) -> bool {
// TODO: Implement with statement after finishing exceptions
unimplemented!()
}
pub fn gen_stmt<'ctx, 'a, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, 'a>,
@ -330,6 +339,7 @@ pub fn gen_stmt<'ctx, 'a, G: CodeGenerator + ?Sized>(
StmtKind::If { .. } => return generator.gen_if(ctx, stmt),
StmtKind::While { .. } => return generator.gen_while(ctx, stmt),
StmtKind::For { .. } => return generator.gen_for(ctx, stmt),
StmtKind::With { .. } => return generator.gen_with(ctx, stmt),
_ => unimplemented!(),
};
false

View File

@ -219,6 +219,17 @@ impl<'a> Inferencer<'a> {
self.check_block(orelse, &mut defined_identifiers)?;
Ok(false)
}
StmtKind::With { items, body, .. } => {
let mut new_defined_identifiers = defined_identifiers.clone();
for item in items.iter() {
self.check_expr(&item.context_expr, defined_identifiers)?;
if let Some(var) = item.optional_vars.as_ref() {
self.check_pattern(var, &mut new_defined_identifiers)?;
}
}
self.check_block(body, &mut new_defined_identifiers)?;
Ok(false)
}
StmtKind::Expr { value } => {
self.check_expr(value, defined_identifiers)?;
Ok(false)

View File

@ -64,6 +64,10 @@ impl fold::Fold<()> for NaiveFolder {
}
}
fn report_error<T>(msg: &str, location: Location) -> Result<T, String> {
Err(format!("{} at {}", msg, location))
}
impl<'a> fold::Fold<()> for Inferencer<'a> {
type TargetU = Option<Type>;
type Error = String;
@ -165,6 +169,14 @@ impl<'a> fold::Fold<()> for Inferencer<'a> {
}
fold::fold_stmt(self, node)?
}
ast::StmtKind::With { ref items, .. } => {
for item in items.iter() {
if let Some(var) = &item.optional_vars {
self.infer_pattern(var)?;
}
}
fold::fold_stmt(self, node)?
}
_ => fold::fold_stmt(self, node)?,
};
match &stmt.node {
@ -186,19 +198,92 @@ impl<'a> fold::Fold<()> for Inferencer<'a> {
}
ast::StmtKind::AnnAssign { .. } | ast::StmtKind::Expr { .. } => {}
ast::StmtKind::Break | ast::StmtKind::Continue | ast::StmtKind::Pass => {}
ast::StmtKind::With { items, .. } => {
for item in items.iter() {
let ty = item.context_expr.custom.unwrap();
// if we can simply unify without creating new types...
let mut fast_path = false;
if let TypeEnum::TObj { fields, .. } = &*self.unifier.get_ty(ty) {
let fields = fields.borrow();
fast_path = true;
if let Some(enter) = fields.get(&"__enter__".into()).cloned() {
if let TypeEnum::TFunc(signature) = &*self.unifier.get_ty(enter) {
let signature = signature.borrow();
if !signature.args.is_empty() {
return report_error(
"__enter__ method should take no argument other than self",
stmt.location
)
}
if let Some(var) = &item.optional_vars {
if signature.vars.is_empty() {
self.unify(signature.ret, var.custom.unwrap(), &stmt.location)?;
} else {
fast_path = false;
}
}
} else {
fast_path = false;
}
} else {
return report_error(
"__enter__ method is required for context manager",
stmt.location
);
}
if let Some(exit) = fields.get(&"__exit__".into()).cloned() {
if let TypeEnum::TFunc(signature) = &*self.unifier.get_ty(exit) {
let signature = signature.borrow();
if !signature.args.is_empty() {
return report_error(
"__exit__ method should take no argument other than self",
stmt.location
)
}
} else {
fast_path = false;
}
} else {
return report_error(
"__exit__ method is required for context manager",
stmt.location
);
}
}
if !fast_path {
let enter = TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![],
ret: item.optional_vars.as_ref().map_or_else(|| self.unifier.get_fresh_var().0, |var| var.custom.unwrap()),
vars: Default::default()
}));
let enter = self.unifier.add_ty(enter);
let exit = TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![],
ret: self.unifier.get_fresh_var().0,
vars: Default::default()
}));
let exit = self.unifier.add_ty(exit);
let mut fields = HashMap::new();
fields.insert("__enter__".into(), enter);
fields.insert("__exit__".into(), exit);
let record = self.unifier.add_record(fields);
self.unify(ty, record, &stmt.location)?;
}
}
}
ast::StmtKind::Return { value } => match (value, self.function_data.return_type) {
(Some(v), Some(v1)) => {
self.unify(v.custom.unwrap(), v1, &v.location)?;
}
(Some(_), None) => {
return Err("Unexpected return value".to_string());
return report_error("Unexpected return value", stmt.location);
}
(None, Some(_)) => {
return Err("Expected return value".to_string());
return report_error("Expected return value", stmt.location);
}
(None, None) => {}
},
_ => return Err("Unsupported statement type".to_string()),
_ => return report_error("Unsupported statement type", stmt.location),
};
Ok(stmt)
}