Skip to content

Flatten Python eval loop with trampoline for reduced call overhead #8431

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
youknowone merged 10 commits into from
Aug 5, 2026
+783 −79
Merged

Flatten Python eval loop with trampoline for reduced call overhead #8431

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
2fc29ce
Factor with_iframe into enter_iframe/exit_iframe helpers
youknowone Aug 1, 2026
b581c90
Allocate InterpreterFrame and LocalsPlus together on the datastack
youknowone Aug 1, 2026
b089987
Implement trampoline loop for Python-to-Python calls
youknowone Aug 1, 2026
cdc0d32
Optimize trampoline: skip recursion check, avoid temporary_refs mutex…
youknowone Aug 1, 2026
d00c250
Use UnsafeCell for pending_tailcall_refs
youknowone Aug 1, 2026
e0fea23
Remove .claude/settings.json from tracking
youknowone Aug 1, 2026
80028a8
Replace SendPtr with Option<SendNonNull> for type-safe pending tailcall
youknowone Aug 2, 2026
0bb5668
Restrict PendingFrame to private, expose only set/take methods
youknowone Aug 2, 2026
a561f48
Restore C-stack overflow check in trampoline enter_iframe_unchecked
youknowone Aug 2, 2026
a2e9a49
Address code review: entry frame double-free, panic safety, cleanup
youknowone Aug 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 0 additions & 15 deletions .claude/settings.json
View file
Open in desktop

This file was deleted.

41 changes: 16 additions & 25 deletions crates/vm/src/builtins/function.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,9 +64,9 @@ fn format_missing_args(
#[pyclass(module = false, name = "function", traverse = "manual")]
#[derive(Debug)]
pub struct PyFunction {
code: PyAtomicRef<PyCode>,
globals: PyDictRef,
builtins: PyObjectRef,
pub(crate) code: PyAtomicRef<PyCode>,
pub(crate) globals: PyDictRef,
pub(crate) builtins: PyObjectRef,
pub(crate) closure: Option<PyRef<PyTuple<PyCellRef>>>,
defaults_and_kwdefaults: PyMutex<(Option<PyTupleRef>, Option<PyDictRef>)>,
name: PyMutex<PyStrRef>,
Expand Down Expand Up @@ -617,11 +617,6 @@ impl Py<PyFunction> {

// Fast path: stack-allocated InterpreterFrame, no FrameObject.
// No refcount inc for code — it's alive via self.code for the call duration.
let nlocalsplus = code.localspluskinds.len();
let max_stackdepth = code.max_stackdepth as usize;
let localsplus =
crate::frame::LocalsPlus::new_on_datastack(nlocalsplus, max_stackdepth, vm);

let locals = if code.flags.contains(bytecode::CodeFlags::NEWLOCALS) {
crate::frame::FrameLocals::lazy()
} else if let Some(locals) = locals {
Expand All @@ -634,22 +629,21 @@ impl Py<PyFunction> {

// Use self.as_object() as raw pointer — no refcount inc/dec.
// The function is alive on the caller's stack for the call duration.
let mut iframe = crate::frame::InterpreterFrame::new(
let iframe = crate::frame::InterpreterFrame::new_on_datastack(
&self.code,
&self.globals,
&self.builtins,
Some(self.as_object()),
localsplus,
locals,
self.closure.as_ref().map_or(&[], |c| c.as_slice()),
crate::frame::FrameOwner::Thread,
vm,
);
let result = self
.fill_locals_from_args_iframe(&mut iframe, func_args, vm)
.and_then(|()| vm.run_frame_fast(&mut iframe));
.fill_locals_from_args_iframe(iframe, func_args, vm)
.and_then(|()| vm.run_frame_fast(iframe));
// Release data stack memory — must happen on both success and error.
unsafe {
if let Some(base) = iframe.localsplus.release_datastack() {
if let Some(base) = iframe.release_datastack_frame() {
vm.datastack_pop(base);
}
}
Expand Down Expand Up @@ -797,10 +791,6 @@ impl Py<PyFunction> {
vm: &VirtualMachine,
) -> PyResult {
let code = &*self.code;
let nlocalsplus = code.localspluskinds.len();
let max_stackdepth = code.max_stackdepth as usize;
let localsplus =
crate::frame::LocalsPlus::new_on_datastack(nlocalsplus, max_stackdepth, vm);

let locals = if code.flags.contains(bytecode::CodeFlags::NEWLOCALS) {
crate::frame::FrameLocals::lazy()
Expand All @@ -810,15 +800,14 @@ impl Py<PyFunction> {
))
};

let mut iframe = crate::frame::InterpreterFrame::new(
let iframe = crate::frame::InterpreterFrame::new_on_datastack(
code,
&self.globals,
&self.builtins,
Some(self.as_object()),
localsplus,
locals,
self.closure.as_ref().map_or(&[], |c| c.as_slice()),
crate::frame::FrameOwner::Thread,
vm,
);

// Fill arguments directly into fastlocals
Expand All @@ -829,9 +818,9 @@ impl Py<PyFunction> {
}
}

let result = vm.run_frame_fast(&mut iframe);
let result = vm.run_frame_fast(iframe);
unsafe {
if let Some(base) = iframe.localsplus.release_datastack() {
if let Some(base) = iframe.release_datastack_frame() {
vm.datastack_pop(base);
}
}
Expand Down Expand Up @@ -887,8 +876,10 @@ pub(crate) fn datastack_frame_size_bytes_for_code(code: &Py<PyCode>) -> Option<u
return None;
}
let nlocalsplus = code.localspluskinds.len();
let capacity = nlocalsplus.checked_add(code.max_stackdepth as usize)?;
capacity.checked_mul(core::mem::size_of::<usize>())
Some(crate::frame::datastack_iframe_total_bytes(
nlocalsplus,
code.max_stackdepth as usize,
))
}

impl PyPayload for PyFunction {
Expand Down
2 changes: 2 additions & 0 deletions crates/vm/src/coroutine.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@ impl ExecutionResult {
};
PyIterReturn::StopIteration(arg)
}
Self::TailCall => unreachable!("TailCall in generator/coroutine"),
}
}
}
Expand Down Expand Up @@ -104,6 +105,7 @@ impl Coro {
self.clear_frame_locals_on_close();
}
Ok(ExecutionResult::Yield(_)) => {}
Ok(ExecutionResult::TailCall) => unreachable!("TailCall in generator/coroutine"),
}
}

Expand Down
Loading
Loading