Skip to content

Apply vectorcall for more places #7336

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 6 commits into from
Mar 4, 2026
Merged

Apply vectorcall for more places #7336

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
f584c47
Add PyType vectorcall and use vectorcall in all specialized call fall…
youknowone Mar 3, 2026
de9cbc4
Add vectorcall slot for PyMethodDescriptor and PyWrapper
youknowone Mar 3, 2026
02a6998
Add FuncArgs::from_vectorcall_owned and simplify vectorcall fallback …
youknowone Mar 3, 2026
07f129b
Use lazy locals allocation in vectorcall fast path, add debug_assert …
youknowone Mar 3, 2026
c47d1f2
Use from_vectorcall_owned in descriptor/type vectorcall, add stack de…
youknowone Mar 3, 2026
2885f80
Collapse duplicated branch tails and use checked_sub in CALL_KW
youknowone Mar 3, 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
Next Next commit
Add PyType vectorcall and use vectorcall in all specialized call fall… 
…backs
  • Loading branch information
@youknowone
youknowone committed Mar 4, 2026
commit f584c47f5da7d94b5e8b5a098fb012aab51bd763
29 changes: 29 additions & 0 deletions crates/vm/src/builtins/type.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2339,8 +2339,37 @@ fn subtype_set_dict(obj: PyObjectRef, value: PyObjectRef, vm: &VirtualMachine) -
* The magical type type
*/

/// Vectorcall for PyType (PEP 590).
/// Fast path: type(x) returns x.__class__ without constructing FuncArgs.
fn vectorcall_type(
zelf_obj: &PyObject,
args: Vec<PyObjectRef>,
nargs: usize,
kwnames: Option<&[PyObjectRef]>,
vm: &VirtualMachine,
) -> PyResult {
let zelf: &Py<PyType> = zelf_obj.downcast_ref().unwrap();

// type(x) fast path: single positional arg, no kwargs
if zelf.is(vm.ctx.types.type_type) {
let no_kwargs = kwnames.is_none_or(|kw| kw.is_empty());
if nargs == 1 && no_kwargs {
return Ok(args[0].obj_type());
}
}

// Fallback: construct FuncArgs and use standard call
let func_args = FuncArgs::from_vectorcall(&args, nargs, kwnames);
PyType::call(zelf, func_args, vm)
}

pub(crate) fn init(ctx: &'static Context) {
PyType::extend_class(ctx, ctx.types.type_type);
ctx.types
.type_type
.slots
.vectorcall
.store(Some(vectorcall_type));
}

pub(crate) fn call_slot_new(
Expand Down
47 changes: 12 additions & 35 deletions crates/vm/src/frame.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3847,8 +3847,7 @@ impl ExecutingFrame<'_> {
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallMethodDescriptorO => {
let instr_idx = self.lasti() as usize - 1;
Expand Down Expand Up @@ -3885,8 +3884,7 @@ impl ExecutingFrame<'_> {
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallMethodDescriptorFast => {
let instr_idx = self.lasti() as usize - 1;
Expand Down Expand Up @@ -3924,8 +3922,7 @@ impl ExecutingFrame<'_> {
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallBuiltinClass => {
let instr_idx = self.lasti() as usize - 1;
Expand All @@ -3935,25 +3932,12 @@ impl ExecutingFrame<'_> {
let callable = self.nth_value(nargs + 1);
let callable_tag = callable as *const PyObject as u32;
if cached_tag == callable_tag && callable.downcast_ref::<PyType>().is_some() {
let args = self.collect_positional_args(nargs);
let self_or_null = self.pop_value_opt();
let callable = self.pop_value();
let final_args = if let Some(self_val) = self_or_null {
let mut args = args;
args.prepend_arg(self_val);
args
} else {
args
};
let result = callable.call(final_args, vm)?;
self.push_value(result);
return Ok(None);
return self.execute_call_vectorcall(nargs, vm);
}
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallAllocAndEnterInit => {
let instr_idx = self.lasti() as usize - 1;
Expand Down Expand Up @@ -4013,8 +3997,7 @@ impl ExecutingFrame<'_> {
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallMethodDescriptorFastWithKeywords => {
// Native function interface is uniform regardless of keyword support
Expand Down Expand Up @@ -4053,8 +4036,7 @@ impl ExecutingFrame<'_> {
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallBuiltinFastWithKeywords => {
// Native function interface is uniform regardless of keyword support
Expand Down Expand Up @@ -4087,8 +4069,7 @@ impl ExecutingFrame<'_> {
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallNonPyGeneral => {
let instr_idx = self.lasti() as usize - 1;
Expand All @@ -4098,14 +4079,12 @@ impl ExecutingFrame<'_> {
let callable = self.nth_value(nargs + 1);
let callable_tag = callable as *const PyObject as u32;
if cached_tag == callable_tag {
let args = self.collect_positional_args(nargs);
return self.execute_call(args, vm);
return self.execute_call_vectorcall(nargs, vm);
}
self.deoptimize(Instruction::Call {
argc: Arg::marker(),
});
let args = self.collect_positional_args(nargs);
self.execute_call(args, vm)
self.execute_call_vectorcall(nargs, vm)
}
Instruction::CallKwPy => {
let instr_idx = self.lasti() as usize - 1;
Expand Down Expand Up @@ -4197,14 +4176,12 @@ impl ExecutingFrame<'_> {
let callable = self.nth_value(nargs + 2);
let callable_tag = callable as *const PyObject as u32;
if cached_tag == callable_tag {
let args = self.collect_keyword_args(nargs);
return self.execute_call(args, vm);
return self.execute_call_kw_vectorcall(nargs, vm);
}
self.deoptimize(Instruction::CallKw {
argc: Arg::marker(),
});
let args = self.collect_keyword_args(nargs);
self.execute_call(args, vm)
self.execute_call_kw_vectorcall(nargs, vm)
}
Instruction::LoadSuperAttrAttr => {
let oparg = u32::from(arg);
Expand Down