gstaichi 2.1.1rc3__cp313-cp313-macosx_11_0_arm64.whl

gstaichi/lang/kernel_impl.py

@@ -0,0 +1,1341 @@
+import ast
+import dataclasses
+import functools
+import inspect
+import json
+import operator
+import os
+import pathlib
+import re
+import sys
+import textwrap
+import time
+import types
+import typing
+import warnings
+from typing import Any, Callable, Type
+
+import numpy as np
+
+import gstaichi.lang
+import gstaichi.lang._ndarray
+import gstaichi.lang._texture
+import gstaichi.types.annotations
+from gstaichi import _logging
+from gstaichi._lib import core as _ti_core
+from gstaichi._lib.core.gstaichi_python import (
+    ASTBuilder,
+    CompiledKernelData,
+    FunctionKey,
+    KernelCxx,
+    KernelLaunchContext,
+)
+from gstaichi.lang import _kernel_impl_dataclass, impl, ops, runtime_ops
+from gstaichi.lang._fast_caching import src_hasher
+from gstaichi.lang._template_mapper import TemplateMapper
+from gstaichi.lang._wrap_inspect import FunctionSourceInfo, get_source_info_and_src
+from gstaichi.lang.any_array import AnyArray
+from gstaichi.lang.ast import (
+    ASTTransformerContext,
+    KernelSimplicityASTChecker,
+    transform_tree,
+)
+from gstaichi.lang.ast.ast_transformer_utils import ReturnStatus
+from gstaichi.lang.exception import (
+    GsTaichiCompilationError,
+    GsTaichiRuntimeError,
+    GsTaichiRuntimeTypeError,
+    GsTaichiSyntaxError,
+    GsTaichiTypeError,
+    handle_exception_from_cpp,
+)
+from gstaichi.lang.expr import Expr
+from gstaichi.lang.kernel_arguments import ArgMetadata
+from gstaichi.lang.matrix import MatrixType
+from gstaichi.lang.shell import _shell_pop_print
+from gstaichi.lang.struct import StructType
+from gstaichi.lang.util import cook_dtype, has_pytorch
+from gstaichi.types import (
+    ndarray_type,
+    primitive_types,
+    sparse_matrix_builder,
+    template,
+    texture_type,
+)
+from gstaichi.types.compound_types import CompoundType
+from gstaichi.types.enums import AutodiffMode, Layout
+from gstaichi.types.utils import is_signed
+
+CompiledKernelKeyType = tuple[Callable, int, AutodiffMode]
+
+
+class GsTaichiCallable:
+    """
+    BoundGsTaichiCallable is used to enable wrapping a bindable function with a class.
+
+    Design requirements for GsTaichiCallable:
+    - wrap/contain a reference to a class Func instance, and allow (the GsTaichiCallable) being passed around
+      like normal function pointer
+    - expose attributes of the wrapped class Func, such as `_if_real_function`, `_primal`, etc
+    - allow for (now limited) strong typing, and enable type checkers, such as pyright/mypy
+        - currently GsTaichiCallable is a shared type used for all functions marked with @ti.func, @ti.kernel,
+          python functions (?)
+        - note: current type-checking implementation does not distinguish between different type flavors of
+          GsTaichiCallable, with different values of `_if_real_function`, `_primal`, etc
+    - handle not only class-less functions, but also class-instance methods (where determining the `self`
+      reference is a challenge)
+
+    Let's take the following example:
+
+    def test_ptr_class_func():
+    @ti.data_oriented
+    class MyClass:
+        def __init__(self):
+            self.a = ti.field(dtype=ti.f32, shape=(3))
+
+        def add2numbers_py(self, x, y):
+            return x + y
+
+        @ti.func
+        def add2numbers_func(self, x, y):
+            return x + y
+
+        @ti.kernel
+        def func(self):
+            a, add_py, add_func = ti.static(self.a, self.add2numbers_py, self.add2numbers_func)
+            a[0] = add_py(2, 3)
+            a[1] = add_func(3, 7)
+
+    (taken from test_ptr_assign.py).
+
+    When the @ti.func decorator is parsed, the function `add2numbers_func` exists, but there is not yet any `self`
+    - it is not possible for the method to be bound, to a `self` instance
+    - however, the @ti.func annotation, runs the kernel_imp.py::func function --- it is at this point
+      that GsTaichi's original code creates a class Func instance (that wraps the add2numbers_func)
+      and immediately we create a GsTaichiCallable instance that wraps the Func instance.
+    - effectively, we have two layers of wrapping GsTaichiCallable->Func->function pointer
+      (actual function definition)
+    - later on, when we call self.add2numbers_py, here:
+
+            a, add_py, add_func = ti.static(self.a, self.add2numbers_py, self.add2numbers_func)
+
+      ... we want to call the bound method, `self.add2numbers_py`.
+    - an actual python function reference, created by doing somevar = MyClass.add2numbers, can automatically
+      binds to self, when called from self in this way (however, add2numbers_py is actually a class
+      Func instance, wrapping python function reference -- now also all wrapped by a GsTaichiCallable
+      instance -- returned by the kernel_impl.py::func function, run by @ti.func)
+    - however, in order to be able to add strongly typed attributes to the wrapped python function, we need
+      to wrap the wrapped python function in a class
+    - the wrapped python function, wrapped in a GsTaichiCallable class (which is callable, and will
+      execute the underlying double-wrapped python function), will NOT automatically bind
+    - when we invoke GsTaichiCallable, the wrapped function is invoked. The wrapped function is unbound, and
+      so `self` is not automatically passed in, as an argument, and things break
+
+    To address this we need to use the `__get__` method, in our function wrapper, ie GsTaichiCallable,
+    and have the `__get__` method return the `BoundGsTaichiCallable` object. The `__get__` method handles
+    running the binding for us, and effectively binds `BoundFunc` object to `self` object, by passing
+    in the instance, as an argument into `BoundGsTaichiCallable.__init__`.
+
+    `BoundFunc` can then be used as a normal bound func - even though it's just an object instance -
+    using its `__call__` method. Effectively, at the time of actually invoking the underlying python
+    function, we have 3 layers of wrapper instances:
+        BoundGsTaichiCallabe -> GsTaichiCallable -> Func -> python function reference/definition
+    """
+
+    def __init__(self, fn: Callable, wrapper: Callable) -> None:
+        self.fn: Callable = fn
+        self.wrapper: Callable = wrapper
+        self._is_real_function: bool = False
+        self._is_gstaichi_function: bool = False
+        self._is_wrapped_kernel: bool = False
+        self._is_classkernel: bool = False
+        self._primal: Kernel | None = None
+        self._adjoint: Kernel | None = None
+        self.grad: Kernel | None = None
+        self._is_staticmethod: bool = False
+        self.is_pure = False
+        functools.update_wrapper(self, fn)
+
+    def __call__(self, *args, **kwargs):
+        return self.wrapper.__call__(*args, **kwargs)
+
+    def __get__(self, instance, owner):
+        if instance is None:
+            return self
+        return BoundGsTaichiCallable(instance, self)
+
+
+class BoundGsTaichiCallable:
+    def __init__(self, instance: Any, gstaichi_callable: "GsTaichiCallable"):
+        self.wrapper = gstaichi_callable.wrapper
+        self.instance = instance
+        self.gstaichi_callable = gstaichi_callable
+
+    def __call__(self, *args, **kwargs):
+        return self.wrapper(self.instance, *args, **kwargs)
+
+    def __getattr__(self, k: str) -> Any:
+        res = getattr(self.gstaichi_callable, k)
+        return res
+
+    def __setattr__(self, k: str, v: Any) -> None:
+        # Note: these have to match the name of any attributes on this class.
+        if k in ("wrapper", "instance", "gstaichi_callable"):
+            object.__setattr__(self, k, v)
+        else:
+            setattr(self.gstaichi_callable, k, v)
+
+
+def func(fn: Callable, is_real_function: bool = False) -> GsTaichiCallable:
+    """Marks a function as callable in GsTaichi-scope.
+
+    This decorator transforms a Python function into a GsTaichi one. GsTaichi
+    will JIT compile it into native instructions.
+
+    Args:
+        fn (Callable): The Python function to be decorated
+        is_real_function (bool): Whether the function is a real function
+
+    Returns:
+        Callable: The decorated function
+
+    Example::
+
+        >>> @ti.func
+        >>> def foo(x):
+        >>>     return x + 2
+        >>>
+        >>> @ti.kernel
+        >>> def run():
+        >>>     print(foo(40))  # 42
+    """
+    is_classfunc = _inside_class(level_of_class_stackframe=3 + is_real_function)
+
+    fun = Func(fn, _classfunc=is_classfunc, is_real_function=is_real_function)
+    gstaichi_callable = GsTaichiCallable(fn, fun)
+    gstaichi_callable._is_gstaichi_function = True
+    gstaichi_callable._is_real_function = is_real_function
+    return gstaichi_callable
+
+
+def real_func(fn: Callable) -> GsTaichiCallable:
+    return func(fn, is_real_function=True)
+
+
+def pyfunc(fn: Callable) -> GsTaichiCallable:
+    """Marks a function as callable in both GsTaichi and Python scopes.
+
+    When called inside the GsTaichi scope, GsTaichi will JIT compile it into
+    native instructions. Otherwise it will be invoked directly as a
+    Python function.
+
+    See also :func:`~gstaichi.lang.kernel_impl.func`.
+
+    Args:
+        fn (Callable): The Python function to be decorated
+
+    Returns:
+        Callable: The decorated function
+    """
+    is_classfunc = _inside_class(level_of_class_stackframe=3)
+    fun = Func(fn, _classfunc=is_classfunc, _pyfunc=True)
+    gstaichi_callable = GsTaichiCallable(fn, fun)
+    gstaichi_callable._is_gstaichi_function = True
+    gstaichi_callable._is_real_function = False
+    return gstaichi_callable
+
+
+def _populate_global_vars_for_templates(
+    template_slot_locations: list[int],
+    argument_metas: list[ArgMetadata],
+    global_vars: dict[str, Any],
+    fn: Callable,
+    py_args: tuple[Any, ...],
+):
+    """
+    Inject template parameters into globals
+
+    Globals are being abused to store the python objects associated
+    with templates. We continue this approach, and in addition this function
+    handles injecting expanded python variables from dataclasses.
+    """
+    for i in template_slot_locations:
+        template_var_name = argument_metas[i].name
+        global_vars[template_var_name] = py_args[i]
+    parameters = inspect.signature(fn).parameters
+    for i, (parameter_name, parameter) in enumerate(parameters.items()):
+        if dataclasses.is_dataclass(parameter.annotation):
+            _kernel_impl_dataclass.populate_global_vars_from_dataclass(
+                parameter_name,
+                parameter.annotation,
+                py_args[i],
+                global_vars=global_vars,
+            )
+
+
+def _get_tree_and_ctx(
+    self: "Func | Kernel",
+    args: tuple[Any, ...],
+    excluded_parameters=(),
+    is_kernel: bool = True,
+    arg_features=None,
+    ast_builder: "ASTBuilder | None" = None,
+    is_real_function: bool = False,
+    current_kernel: "Kernel | None" = None,
+) -> tuple[ast.Module, ASTTransformerContext]:
+    function_source_info, src = get_source_info_and_src(self.func)
+    src = [textwrap.fill(line, tabsize=4, width=9999) for line in src]
+    tree = ast.parse(textwrap.dedent("\n".join(src)))
+
+    func_body = tree.body[0]
+    func_body.decorator_list = []  # type: ignore , kick that can down the road...
+
+    global_vars = _get_global_vars(self.func)
+
+    if is_kernel or is_real_function:
+        _populate_global_vars_for_templates(
+            template_slot_locations=self.template_slot_locations,
+            argument_metas=self.arg_metas,
+            global_vars=global_vars,
+            fn=self.func,
+            py_args=args,
+        )
+
+    if current_kernel is not None:  # Kernel
+        current_kernel.kernel_function_info = function_source_info
+    if current_kernel is None:
+        current_kernel = impl.get_runtime()._current_kernel
+    assert current_kernel is not None
+    current_kernel.visited_functions.add(function_source_info)
+
+    return tree, ASTTransformerContext(
+        excluded_parameters=excluded_parameters,
+        is_kernel=is_kernel,
+        func=self,
+        arg_features=arg_features,
+        global_vars=global_vars,
+        argument_data=args,
+        src=src,
+        start_lineno=function_source_info.start_lineno,
+        end_lineno=function_source_info.end_lineno,
+        file=function_source_info.filepath,
+        ast_builder=ast_builder,
+        is_real_function=is_real_function,
+    )
+
+
+def _process_args(self: "Func | Kernel", is_func: bool, args: tuple[Any, ...], kwargs) -> tuple[Any, ...]:
+    if is_func:
+        self.arg_metas = _kernel_impl_dataclass.expand_func_arguments(self.arg_metas)
+
+    fused_args: list[Any] = [arg_meta.default for arg_meta in self.arg_metas]
+    len_args = len(args)
+
+    if len_args > len(fused_args):
+        arg_str = ", ".join(map(str, args))
+        expected_str = ", ".join(f"{arg.name} : {arg.annotation}" for arg in self.arg_metas)
+        msg = f"Too many arguments. Expected ({expected_str}), got ({arg_str})."
+        raise GsTaichiSyntaxError(msg)
+
+    for i, arg in enumerate(args):
+        fused_args[i] = arg
+
+    for key, value in kwargs.items():
+        for i, arg in enumerate(self.arg_metas):
+            if key == arg.name:
+                if i < len_args:
+                    raise GsTaichiSyntaxError(f"Multiple values for argument '{key}'.")
+                fused_args[i] = value
+                break
+        else:
+            raise GsTaichiSyntaxError(f"Unexpected argument '{key}'.")
+
+    for i, arg in enumerate(fused_args):
+        if arg is inspect.Parameter.empty:
+            if self.arg_metas[i].annotation is inspect._empty:
+                raise GsTaichiSyntaxError(f"Parameter `{self.arg_metas[i].name}` missing.")
+            else:
+                raise GsTaichiSyntaxError(
+                    f"Parameter `{self.arg_metas[i].name} : {self.arg_metas[i].annotation}` missing."
+                )
+
+    return tuple(fused_args)
+
+
+class Func:
+    function_counter = 0
+
+    def __init__(self, _func: Callable, _classfunc=False, _pyfunc=False, is_real_function=False) -> None:
+        self.func = _func
+        self.func_id = Func.function_counter
+        Func.function_counter += 1
+        self.compiled = {}
+        self.classfunc = _classfunc
+        self.pyfunc = _pyfunc
+        self.is_real_function = is_real_function
+        self.arg_metas: list[ArgMetadata] = []
+        self.orig_arguments: list[ArgMetadata] = []
+        self.return_type: tuple[Type, ...] | None = None
+        self.extract_arguments()
+        self.template_slot_locations: list[int] = []
+        for i, arg in enumerate(self.arg_metas):
+            if arg.annotation == template or isinstance(arg.annotation, template):
+                self.template_slot_locations.append(i)
+        self.mapper = TemplateMapper(self.arg_metas, self.template_slot_locations)
+        self.gstaichi_functions = {}  # The |Function| class in C++
+        self.has_print = False
+
+    def __call__(self: "Func", *args, **kwargs) -> Any:
+        args = _process_args(self, is_func=True, args=args, kwargs=kwargs)
+
+        if not impl.inside_kernel():
+            if not self.pyfunc:
+                raise GsTaichiSyntaxError("GsTaichi functions cannot be called from Python-scope.")
+            return self.func(*args)
+
+        current_kernel = impl.get_runtime().current_kernel
+        if self.is_real_function:
+            if current_kernel.autodiff_mode != AutodiffMode.NONE:
+                raise GsTaichiSyntaxError("Real function in gradient kernels unsupported.")
+            instance_id, arg_features = self.mapper.lookup(args)
+            key = _ti_core.FunctionKey(self.func.__name__, self.func_id, instance_id)
+            if key.instance_id not in self.compiled:
+                self.do_compile(key=key, args=args, arg_features=arg_features)
+            return self.func_call_rvalue(key=key, args=args)
+        tree, ctx = _get_tree_and_ctx(
+            self,
+            is_kernel=False,
+            args=args,
+            ast_builder=current_kernel.ast_builder(),
+            is_real_function=self.is_real_function,
+        )
+
+        struct_locals = _kernel_impl_dataclass.extract_struct_locals_from_context(ctx)
+
+        tree = _kernel_impl_dataclass.unpack_ast_struct_expressions(tree, struct_locals=struct_locals)
+        ret = transform_tree(tree, ctx)
+        if not self.is_real_function:
+            if self.return_type and ctx.returned != ReturnStatus.ReturnedValue:
+                raise GsTaichiSyntaxError("Function has a return type but does not have a return statement")
+        return ret
+
+    def func_call_rvalue(self, key: FunctionKey, args: tuple[Any, ...]) -> Any:
+        # Skip the template args, e.g., |self|
+        assert self.is_real_function
+        non_template_args = []
+        dbg_info = _ti_core.DebugInfo(impl.get_runtime().get_current_src_info())
+        for i, kernel_arg in enumerate(self.arg_metas):
+            anno = kernel_arg.annotation
+            if not isinstance(anno, template):
+                if id(anno) in primitive_types.type_ids:
+                    non_template_args.append(ops.cast(args[i], anno))
+                elif isinstance(anno, primitive_types.RefType):
+                    non_template_args.append(_ti_core.make_reference(args[i].ptr, dbg_info))
+                elif isinstance(anno, ndarray_type.NdarrayType):
+                    if not isinstance(args[i], AnyArray):
+                        raise GsTaichiTypeError(
+                            f"Expected ndarray in the kernel argument for argument {kernel_arg.name}, got {args[i]}"
+                        )
+                    non_template_args += _ti_core.get_external_tensor_real_func_args(args[i].ptr, dbg_info)
+                else:
+                    non_template_args.append(args[i])
+        non_template_args = impl.make_expr_group(non_template_args)
+        compiling_callable = impl.get_runtime().compiling_callable
+        assert compiling_callable is not None
+        func_call = compiling_callable.ast_builder().insert_func_call(
+            self.gstaichi_functions[key.instance_id], non_template_args, dbg_info
+        )
+        if self.return_type is None:
+            return None
+        func_call = Expr(func_call)
+        ret = []
+
+        for i, return_type in enumerate(self.return_type):
+            if id(return_type) in primitive_types.type_ids:
+                ret.append(
+                    Expr(
+                        _ti_core.make_get_element_expr(
+                            func_call.ptr, (i,), _ti_core.DebugInfo(impl.get_runtime().get_current_src_info())
+                        )
+                    )
+                )
+            elif isinstance(return_type, (StructType, MatrixType)):
+                ret.append(return_type.from_gstaichi_object(func_call, (i,)))
+            else:
+                raise GsTaichiTypeError(f"Unsupported return type for return value {i}: {return_type}")
+        if len(ret) == 1:
+            return ret[0]
+        return tuple(ret)
+
+    def do_compile(self, key: FunctionKey, args: tuple[Any, ...], arg_features: tuple[Any, ...]) -> None:
+        tree, ctx = _get_tree_and_ctx(
+            self, is_kernel=False, args=args, arg_features=arg_features, is_real_function=self.is_real_function
+        )
+        fn = impl.get_runtime().prog.create_function(key)
+
+        def func_body():
+            old_callable = impl.get_runtime().compiling_callable
+            impl.get_runtime()._compiling_callable = fn
+            ctx.ast_builder = fn.ast_builder()
+            transform_tree(tree, ctx)
+            impl.get_runtime()._compiling_callable = old_callable
+
+        self.gstaichi_functions[key.instance_id] = fn
+        self.compiled[key.instance_id] = func_body
+        self.gstaichi_functions[key.instance_id].set_function_body(func_body)
+
+    def extract_arguments(self) -> None:
+        sig = inspect.signature(self.func)
+        if sig.return_annotation not in (inspect.Signature.empty, None):
+            self.return_type = sig.return_annotation
+            if (
+                isinstance(self.return_type, (types.GenericAlias, typing._GenericAlias))  # type: ignore
+                and self.return_type.__origin__ is tuple  # type: ignore
+            ):
+                self.return_type = self.return_type.__args__  # type: ignore
+            if self.return_type is None:
+                return
+            if not isinstance(self.return_type, (list, tuple)):
+                self.return_type = (self.return_type,)
+            for i, return_type in enumerate(self.return_type):
+                if return_type is Ellipsis:
+                    raise GsTaichiSyntaxError("Ellipsis is not supported in return type annotations")
+        params = sig.parameters
+        arg_names = params.keys()
+        for i, arg_name in enumerate(arg_names):
+            param = params[arg_name]
+            if param.kind == inspect.Parameter.VAR_KEYWORD:
+                raise GsTaichiSyntaxError(
+                    "GsTaichi functions do not support variable keyword parameters (i.e., **kwargs)"
+                )
+            if param.kind == inspect.Parameter.VAR_POSITIONAL:
+                raise GsTaichiSyntaxError(
+                    "GsTaichi functions do not support variable positional parameters (i.e., *args)"
+                )
+            if param.kind == inspect.Parameter.KEYWORD_ONLY:
+                raise GsTaichiSyntaxError("GsTaichi functions do not support keyword parameters")
+            if param.kind != inspect.Parameter.POSITIONAL_OR_KEYWORD:
+                raise GsTaichiSyntaxError('GsTaichi functions only support "positional or keyword" parameters')
+            annotation = param.annotation
+            if annotation is inspect.Parameter.empty:
+                if i == 0 and self.classfunc:
+                    annotation = template()
+                # TODO: pyfunc also need type annotation check when real function is enabled,
+                #       but that has to happen at runtime when we know which scope it's called from.
+                elif not self.pyfunc and self.is_real_function:
+                    raise GsTaichiSyntaxError(
+                        f"GsTaichi function `{self.func.__name__}` parameter `{arg_name}` must be type annotated"
+                    )
+            else:
+                if isinstance(annotation, ndarray_type.NdarrayType):
+                    pass
+                elif isinstance(annotation, MatrixType):
+                    pass
+                elif isinstance(annotation, StructType):
+                    pass
+                elif id(annotation) in primitive_types.type_ids:
+                    pass
+                elif type(annotation) == gstaichi.types.annotations.Template:
+                    pass
+                elif isinstance(annotation, template) or annotation == gstaichi.types.annotations.Template:
+                    pass
+                elif isinstance(annotation, primitive_types.RefType):
+                    pass
+                elif isinstance(annotation, type) and dataclasses.is_dataclass(annotation):
+                    pass
+                else:
+                    raise GsTaichiSyntaxError(
+                        f"Invalid type annotation (argument {i}) of GsTaichi function: {annotation}"
+                    )
+            self.arg_metas.append(ArgMetadata(annotation, param.name, param.default))
+            self.orig_arguments.append(ArgMetadata(annotation, param.name, param.default))
+
+
+def _get_global_vars(_func: Callable) -> dict[str, Any]:
+    # Discussions: https://github.com/taichi-dev/gstaichi/issues/282
+    global_vars = _func.__globals__.copy()
+
+    freevar_names = _func.__code__.co_freevars
+    closure = _func.__closure__
+    if closure:
+        freevar_values = list(map(lambda x: x.cell_contents, closure))
+        for name, value in zip(freevar_names, freevar_values):
+            global_vars[name] = value
+
+    return global_vars
+
+
+@dataclasses.dataclass
+class SrcLlCacheObservations:
+    cache_key_generated: bool = False
+    cache_validated: bool = False
+    cache_loaded: bool = False
+    cache_stored: bool = False
+
+
+class Kernel:
+    counter = 0
+
+    def __init__(self, _func: Callable, autodiff_mode: AutodiffMode, _classkernel=False) -> None:
+        self.func = _func
+        self.kernel_counter = Kernel.counter
+        Kernel.counter += 1
+        assert autodiff_mode in (
+            AutodiffMode.NONE,
+            AutodiffMode.VALIDATION,
+            AutodiffMode.FORWARD,
+            AutodiffMode.REVERSE,
+        )
+        self.autodiff_mode = autodiff_mode
+        self.grad: "Kernel | None" = None
+        self.arg_metas: list[ArgMetadata] = []
+        self.return_type = None
+        self.classkernel = _classkernel
+        self.extract_arguments()
+        self.template_slot_locations = []
+        for i, arg in enumerate(self.arg_metas):
+            if arg.annotation == template or isinstance(arg.annotation, template):
+                self.template_slot_locations.append(i)
+        self.mapper = TemplateMapper(self.arg_metas, self.template_slot_locations)
+        impl.get_runtime().kernels.append(self)
+        self.reset()
+        self.kernel_cpp = None
+        # A materialized kernel is a KernelCxx object which may or may not have
+        # been compiled. It generally has been converted at least as far as AST
+        # and front-end IR, but not necessarily any further.
+        self.materialized_kernels: dict[CompiledKernelKeyType, KernelCxx] = {}
+        self.has_print = False
+        self.gstaichi_callable: GsTaichiCallable | None = None
+        self.visited_functions: set[FunctionSourceInfo] = set()
+        self.kernel_function_info: FunctionSourceInfo | None = None
+        self.compiled_kernel_data_by_key: dict[CompiledKernelKeyType, CompiledKernelData] = {}
+
+        self.src_ll_cache_observations: SrcLlCacheObservations = SrcLlCacheObservations()
+
+    def ast_builder(self) -> ASTBuilder:
+        assert self.kernel_cpp is not None
+        return self.kernel_cpp.ast_builder()
+
+    def reset(self) -> None:
+        self.runtime = impl.get_runtime()
+        self.materialized_kernels = {}
+
+    def extract_arguments(self) -> None:
+        sig = inspect.signature(self.func)
+        if sig.return_annotation not in (inspect._empty, None):
+            self.return_type = sig.return_annotation
+            if (
+                isinstance(self.return_type, (types.GenericAlias, typing._GenericAlias))  # type: ignore
+                and self.return_type.__origin__ is tuple
+            ):
+                self.return_type = self.return_type.__args__
+            if not isinstance(self.return_type, (list, tuple)):
+                self.return_type = (self.return_type,)
+            for return_type in self.return_type:
+                if return_type is Ellipsis:
+                    raise GsTaichiSyntaxError("Ellipsis is not supported in return type annotations")
+        params = dict(sig.parameters)
+        arg_names = params.keys()
+        for i, arg_name in enumerate(arg_names):
+            param = params[arg_name]
+            if param.kind == inspect.Parameter.VAR_KEYWORD:
+                raise GsTaichiSyntaxError(
+                    "GsTaichi kernels do not support variable keyword parameters (i.e., **kwargs)"
+                )
+            if param.kind == inspect.Parameter.VAR_POSITIONAL:
+                raise GsTaichiSyntaxError(
+                    "GsTaichi kernels do not support variable positional parameters (i.e., *args)"
+                )
+            if param.default is not inspect.Parameter.empty:
+                raise GsTaichiSyntaxError("GsTaichi kernels do not support default values for arguments")
+            if param.kind == inspect.Parameter.KEYWORD_ONLY:
+                raise GsTaichiSyntaxError("GsTaichi kernels do not support keyword parameters")
+            if param.kind != inspect.Parameter.POSITIONAL_OR_KEYWORD:
+                raise GsTaichiSyntaxError('GsTaichi kernels only support "positional or keyword" parameters')
+            annotation = param.annotation
+            if param.annotation is inspect.Parameter.empty:
+                if i == 0 and self.classkernel:  # The |self| parameter
+                    annotation = template()
+                else:
+                    raise GsTaichiSyntaxError("GsTaichi kernels parameters must be type annotated")
+            else:
+                if isinstance(
+                    annotation,
+                    (
+                        template,
+                        ndarray_type.NdarrayType,
+                        texture_type.TextureType,
+                        texture_type.RWTextureType,
+                    ),
+                ):
+                    pass
+                elif annotation is ndarray_type.NdarrayType:
+                    # convert from ti.types.NDArray into ti.types.NDArray()
+                    annotation = annotation()
+                elif id(annotation) in primitive_types.type_ids:
+                    pass
+                elif isinstance(annotation, sparse_matrix_builder):
+                    pass
+                elif isinstance(annotation, MatrixType):
+                    pass
+                elif isinstance(annotation, StructType):
+                    pass
+                elif annotation == template:
+                    pass
+                elif isinstance(annotation, type) and dataclasses.is_dataclass(annotation):
+                    pass
+                else:
+                    raise GsTaichiSyntaxError(f"Invalid type annotation (argument {i}) of Taichi kernel: {annotation}")
+            self.arg_metas.append(ArgMetadata(annotation, param.name, param.default))
+
+    def materialize(self, key: CompiledKernelKeyType | None, args: tuple[Any, ...], arg_features=None):
+        if key is None:
+            key = (self.func, 0, self.autodiff_mode)
+        self.runtime.materialize()
+        self.fast_checksum = None
+
+        if key in self.materialized_kernels:
+            return
+
+        if self.runtime.src_ll_cache and self.gstaichi_callable and self.gstaichi_callable.is_pure:
+            kernel_source_info, _src = get_source_info_and_src(self.func)
+            self.fast_checksum = src_hasher.create_cache_key(kernel_source_info, args)
+            if self.fast_checksum:
+                self.src_ll_cache_observations.cache_key_generated = True
+            if self.fast_checksum and src_hasher.validate_cache_key(self.fast_checksum):
+                self.src_ll_cache_observations.cache_validated = True
+                prog = impl.get_runtime().prog
+                self.compiled_kernel_data_by_key[key] = prog.load_fast_cache(
+                    self.fast_checksum,
+                    self.func.__name__,
+                    prog.config(),
+                    prog.get_device_caps(),
+                )
+                if self.compiled_kernel_data_by_key[key]:
+                    self.src_ll_cache_observations.cache_loaded = True
+
+        kernel_name = f"{self.func.__name__}_c{self.kernel_counter}_{key[1]}"
+        _logging.trace(f"Materializing kernel {kernel_name} in {self.autodiff_mode}...")
+
+        tree, ctx = _get_tree_and_ctx(
+            self,
+            args=args,
+            excluded_parameters=self.template_slot_locations,
+            arg_features=arg_features,
+            current_kernel=self,
+        )
+
+        if self.autodiff_mode != AutodiffMode.NONE:
+            KernelSimplicityASTChecker(self.func).visit(tree)
+
+        # Do not change the name of 'gstaichi_ast_generator'
+        # The warning system needs this identifier to remove unnecessary messages
+        def gstaichi_ast_generator(kernel_cxx: KernelCxx):
+            nonlocal tree
+            if self.runtime.inside_kernel:
+                raise GsTaichiSyntaxError(
+                    "Kernels cannot call other kernels. I.e., nested kernels are not allowed. "
+                    "Please check if you have direct/indirect invocation of kernels within kernels. "
+                    "Note that some methods provided by the GsTaichi standard library may invoke kernels, "
+                    "and please move their invocations to Python-scope."
+                )
+            self.kernel_cpp = kernel_cxx
+            self.runtime.inside_kernel = True
+            self.runtime._current_kernel = self
+            assert self.runtime._compiling_callable is None
+            self.runtime._compiling_callable = kernel_cxx
+            try:
+                ctx.ast_builder = kernel_cxx.ast_builder()
+
+                def ast_to_dict(node: ast.AST | list | primitive_types._python_primitive_types):
+                    if isinstance(node, ast.AST):
+                        fields = {k: ast_to_dict(v) for k, v in ast.iter_fields(node)}
+                        return {
+                            "type": node.__class__.__name__,
+                            "fields": fields,
+                            "lineno": getattr(node, "lineno", None),
+                            "col_offset": getattr(node, "col_offset", None),
+                        }
+                    if isinstance(node, list):
+                        return [ast_to_dict(x) for x in node]
+                    return node  # Basic types (str, int, None, etc.)
+
+                if os.environ.get("TI_DUMP_AST", "") == "1":
+                    target_dir = pathlib.Path("/tmp/ast")
+                    target_dir.mkdir(parents=True, exist_ok=True)
+
+                    start = time.time()
+                    ast_str = ast.dump(tree, indent=2)
+                    output_file = target_dir / f"{kernel_name}_ast.txt"
+                    output_file.write_text(ast_str)
+                    elapsed_txt = time.time() - start
+
+                    start = time.time()
+                    json_str = json.dumps(ast_to_dict(tree), indent=2)
+                    output_file = target_dir / f"{kernel_name}_ast.json"
+                    output_file.write_text(json_str)
+                    elapsed_json = time.time() - start
+
+                    output_file = target_dir / f"{kernel_name}_gen_time.json"
+                    output_file.write_text(
+                        json.dumps({"elapsed_txt": elapsed_txt, "elapsed_json": elapsed_json}, indent=2)
+                    )
+                struct_locals = _kernel_impl_dataclass.extract_struct_locals_from_context(ctx)
+                tree = _kernel_impl_dataclass.unpack_ast_struct_expressions(tree, struct_locals=struct_locals)
+                ctx.only_parse_function_def = self.compiled_kernel_data_by_key.get(key) is not None
+                transform_tree(tree, ctx)
+                if not ctx.is_real_function:
+                    if self.return_type and ctx.returned != ReturnStatus.ReturnedValue:
+                        raise GsTaichiSyntaxError("Kernel has a return type but does not have a return statement")
+            finally:
+                self.runtime.inside_kernel = False
+                self.runtime._current_kernel = None
+                self.runtime._compiling_callable = None
+
+        gstaichi_kernel = impl.get_runtime().prog.create_kernel(gstaichi_ast_generator, kernel_name, self.autodiff_mode)
+        assert key not in self.materialized_kernels
+        self.materialized_kernels[key] = gstaichi_kernel
+
+    def launch_kernel(self, t_kernel: KernelCxx, compiled_kernel_data: CompiledKernelData | None, *args) -> Any:
+        assert len(args) == len(self.arg_metas), f"{len(self.arg_metas)} arguments needed but {len(args)} provided"
+
+        tmps = []
+        callbacks = []
+
+        actual_argument_slot = 0
+        launch_ctx = t_kernel.make_launch_context()
+        max_arg_num = 512
+        exceed_max_arg_num = False
+
+        def set_arg_ndarray(indices: tuple[int, ...], v: gstaichi.lang._ndarray.Ndarray) -> None:
+            v_primal = v.arr
+            v_grad = v.grad.arr if v.grad else None
+            if v_grad is None:
+                launch_ctx.set_arg_ndarray(indices, v_primal)  # type: ignore , solvable probably, just not today
+            else:
+                launch_ctx.set_arg_ndarray_with_grad(indices, v_primal, v_grad)  # type: ignore
+
+        def set_arg_texture(indices: tuple[int, ...], v: gstaichi.lang._texture.Texture) -> None:
+            launch_ctx.set_arg_texture(indices, v.tex)
+
+        def set_arg_rw_texture(indices: tuple[int, ...], v: gstaichi.lang._texture.Texture) -> None:
+            launch_ctx.set_arg_rw_texture(indices, v.tex)
+
+        def set_arg_ext_array(indices: tuple[int, ...], v: Any, needed: ndarray_type.NdarrayType) -> None:
+            # v is things like torch Tensor and numpy array
+            # Not adding type for this, since adds additional dependencies
+            #
+            # Element shapes are already specialized in GsTaichi codegen.
+            # The shape information for element dims are no longer needed.
+            # Therefore we strip the element shapes from the shape vector,
+            # so that it only holds "real" array shapes.
+            is_soa = needed.layout == Layout.SOA
+            array_shape = v.shape
+            if functools.reduce(operator.mul, array_shape, 1) > np.iinfo(np.int32).max:
+                warnings.warn("Ndarray index might be out of int32 boundary but int64 indexing is not supported yet.")
+            if needed.dtype is None or id(needed.dtype) in primitive_types.type_ids:
+                element_dim = 0
+            else:
+                element_dim = needed.dtype.ndim
+                array_shape = v.shape[element_dim:] if is_soa else v.shape[:-element_dim]
+            if isinstance(v, np.ndarray):
+                # numpy
+                if v.flags.c_contiguous:
+                    launch_ctx.set_arg_external_array_with_shape(indices, int(v.ctypes.data), v.nbytes, array_shape, 0)
+                elif v.flags.f_contiguous:
+                    # TODO: A better way that avoids copying is saving strides info.
+                    tmp = np.ascontiguousarray(v)
+                    # Purpose: DO NOT GC |tmp|!
+                    tmps.append(tmp)
+
+                    def callback(original, updated):
+                        np.copyto(original, np.asfortranarray(updated))
+
+                    callbacks.append(functools.partial(callback, v, tmp))
+                    launch_ctx.set_arg_external_array_with_shape(
+                        indices, int(tmp.ctypes.data), tmp.nbytes, array_shape, 0
+                    )
+                else:
+                    raise ValueError(
+                        "Non contiguous numpy arrays are not supported, please call np.ascontiguousarray(arr) "
+                        "before passing it into gstaichi kernel."
+                    )
+            elif has_pytorch():
+                import torch  # pylint: disable=C0415
+
+                if isinstance(v, torch.Tensor):
+                    if not v.is_contiguous():
+                        raise ValueError(
+                            "Non contiguous tensors are not supported, please call tensor.contiguous() before "
+                            "passing it into gstaichi kernel."
+                        )
+                    gstaichi_arch = self.runtime.prog.config().arch
+
+                    def get_call_back(u, v):
+                        def call_back():
+                            u.copy_(v)
+
+                        return call_back
+
+                    # FIXME: only allocate when launching grad kernel
+                    if v.requires_grad and v.grad is None:
+                        v.grad = torch.zeros_like(v)
+
+                    if v.requires_grad:
+                        if not isinstance(v.grad, torch.Tensor):
+                            raise ValueError(
+                                f"Expecting torch.Tensor for gradient tensor, but getting {v.grad.__class__.__name__} instead"
+                            )
+                        if not v.grad.is_contiguous():
+                            raise ValueError(
+                                "Non contiguous gradient tensors are not supported, please call tensor.grad.contiguous() before passing it into gstaichi kernel."
+                            )
+
+                    tmp = v
+                    if (str(v.device) != "cpu") and not (
+                        str(v.device).startswith("cuda") and gstaichi_arch == _ti_core.Arch.cuda
+                    ):
+                        # Getting a torch CUDA tensor on GsTaichi non-cuda arch:
+                        # We just replace it with a CPU tensor and by the end of kernel execution we'll use the
+                        # callback to copy the values back to the original CUDA tensor.
+                        host_v = v.to(device="cpu", copy=True)
+                        tmp = host_v
+                        callbacks.append(get_call_back(v, host_v))
+
+                    launch_ctx.set_arg_external_array_with_shape(
+                        indices,
+                        int(tmp.data_ptr()),
+                        tmp.element_size() * tmp.nelement(),
+                        array_shape,
+                        int(v.grad.data_ptr()) if v.grad is not None else 0,
+                    )
+                else:
+                    raise GsTaichiRuntimeTypeError(
+                        f"Argument of type {type(v)} cannot be converted into required type {needed}"
+                    )
+            else:
+                raise GsTaichiRuntimeTypeError(f"Argument {needed} cannot be converted into required type {v}")
+
+        def set_arg_matrix(indices: tuple[int, ...], v, needed) -> None:
+            def cast_float(x: float | np.floating | np.integer | int) -> float:
+                if not isinstance(x, (int, float, np.integer, np.floating)):
+                    raise GsTaichiRuntimeTypeError(
+                        f"Argument {needed.dtype} cannot be converted into required type {type(x)}"
+                    )
+                return float(x)
+
+            def cast_int(x: int | np.integer) -> int:
+                if not isinstance(x, (int, np.integer)):
+                    raise GsTaichiRuntimeTypeError(
+                        f"Argument {needed.dtype} cannot be converted into required type {type(x)}"
+                    )
+                return int(x)
+
+            cast_func = None
+            if needed.dtype in primitive_types.real_types:
+                cast_func = cast_float
+            elif needed.dtype in primitive_types.integer_types:
+                cast_func = cast_int
+            else:
+                raise ValueError(f"Matrix dtype {needed.dtype} is not integer type or real type.")
+
+            if needed.ndim == 2:
+                v = [cast_func(v[i, j]) for i in range(needed.n) for j in range(needed.m)]
+            else:
+                v = [cast_func(v[i]) for i in range(needed.n)]
+            v = needed(*v)
+            needed.set_kernel_struct_args(v, launch_ctx, indices)
+
+        def set_arg_sparse_matrix_builder(indices: tuple[int, ...], v) -> None:
+            # Pass only the base pointer of the ti.types.sparse_matrix_builder() argument
+            launch_ctx.set_arg_uint(indices, v._get_ndarray_addr())
+
+        set_later_list = []
+
+        def recursive_set_args(needed_arg_type: Type, provided_arg_type: Type, v: Any, indices: tuple[int, ...]) -> int:
+            """
+            Returns the number of kernel args set
+            e.g. templates don't set kernel args, so returns 0
+            a single ndarray is 1 kernel arg, so returns 1
+            a struct of 3 ndarrays would set 3 kernel args, so return 3
+            note: len(indices) > 1 only happens with argpack (which we are removing support for)
+            """
+            nonlocal actual_argument_slot, exceed_max_arg_num, set_later_list
+            if actual_argument_slot >= max_arg_num:
+                exceed_max_arg_num = True
+                return 0
+            actual_argument_slot += 1
+            # Note: do not use sth like "needed == f32". That would be slow.
+            if id(needed_arg_type) in primitive_types.real_type_ids:
+                if not isinstance(v, (float, int, np.floating, np.integer)):
+                    raise GsTaichiRuntimeTypeError.get(indices, needed_arg_type.to_string(), provided_arg_type)
+                launch_ctx.set_arg_float(indices, float(v))
+                return 1
+            if id(needed_arg_type) in primitive_types.integer_type_ids:
+                if not isinstance(v, (int, np.integer)):
+                    raise GsTaichiRuntimeTypeError.get(indices, needed_arg_type.to_string(), provided_arg_type)
+                if is_signed(cook_dtype(needed_arg_type)):
+                    launch_ctx.set_arg_int(indices, int(v))
+                else:
+                    launch_ctx.set_arg_uint(indices, int(v))
+                return 1
+            if isinstance(needed_arg_type, sparse_matrix_builder):
+                set_arg_sparse_matrix_builder(indices, v)
+                return 1
+            if dataclasses.is_dataclass(needed_arg_type):
+                assert provided_arg_type == needed_arg_type
+                idx = 0
+                for j, field in enumerate(dataclasses.fields(needed_arg_type)):
+                    assert not isinstance(field.type, str)
+                    field_value = getattr(v, field.name)
+                    idx += recursive_set_args(field.type, field.type, field_value, (indices[0] + idx,))
+                return idx
+            if isinstance(needed_arg_type, ndarray_type.NdarrayType) and isinstance(v, gstaichi.lang._ndarray.Ndarray):
+                set_arg_ndarray(indices, v)
+                return 1
+            if isinstance(needed_arg_type, texture_type.TextureType) and isinstance(v, gstaichi.lang._texture.Texture):
+                set_arg_texture(indices, v)
+                return 1
+            if isinstance(needed_arg_type, texture_type.RWTextureType) and isinstance(
+                v, gstaichi.lang._texture.Texture
+            ):
+                set_arg_rw_texture(indices, v)
+                return 1
+            if isinstance(needed_arg_type, ndarray_type.NdarrayType):
+                set_arg_ext_array(indices, v, needed_arg_type)
+                return 1
+            if isinstance(needed_arg_type, MatrixType):
+                set_arg_matrix(indices, v, needed_arg_type)
+                return 1
+            if isinstance(needed_arg_type, StructType):
+                # Unclear how to make the following pass typing checks
+                # StructType implements __instancecheck__, which should be a classmethod, but
+                # is currently an instance method
+                # TODO: look into this more deeply at some point
+                if not isinstance(v, needed_arg_type):  # type: ignore
+                    raise GsTaichiRuntimeTypeError(
+                        f"Argument {provided_arg_type} cannot be converted into required type {needed_arg_type}"
+                    )
+                needed_arg_type.set_kernel_struct_args(v, launch_ctx, indices)
+                return 1
+            if needed_arg_type == template or isinstance(needed_arg_type, template):
+                return 0
+            raise ValueError(f"Argument type mismatch. Expecting {needed_arg_type}, got {type(v)}.")
+
+        template_num = 0
+        i_out = 0
+        for i_in, val in enumerate(args):
+            needed_ = self.arg_metas[i_in].annotation
+            if needed_ == template or isinstance(needed_, template):
+                template_num += 1
+                i_out += 1
+                continue
+            i_out += recursive_set_args(needed_, type(val), val, (i_out - template_num,))
+
+        for i, (set_arg_func, params) in enumerate(set_later_list):
+            set_arg_func((len(args) - template_num + i,), *params)
+
+        if exceed_max_arg_num:
+            raise GsTaichiRuntimeError(
+                f"The number of elements in kernel arguments is too big! Do not exceed {max_arg_num} on {_ti_core.arch_name(impl.current_cfg().arch)} backend."
+            )
+
+        try:
+            prog = impl.get_runtime().prog
+            if not compiled_kernel_data:
+                compiled_kernel_data = prog.compile_kernel(prog.config(), prog.get_device_caps(), t_kernel)
+                if self.fast_checksum:
+                    src_hasher.store(self.fast_checksum, self.visited_functions)
+                    prog.store_fast_cache(
+                        self.fast_checksum,
+                        self.kernel_cpp,
+                        prog.config(),
+                        prog.get_device_caps(),
+                        compiled_kernel_data,
+                    )
+                    self.src_ll_cache_observations.cache_stored = True
+            prog.launch_kernel(compiled_kernel_data, launch_ctx)
+        except Exception as e:
+            e = handle_exception_from_cpp(e)
+            if impl.get_runtime().print_full_traceback:
+                raise e
+            raise e from None
+
+        ret = None
+        ret_dt = self.return_type
+        has_ret = ret_dt is not None
+
+        if has_ret or self.has_print:
+            runtime_ops.sync()
+
+        if has_ret:
+            ret = []
+            for i, ret_type in enumerate(ret_dt):
+                ret.append(self.construct_kernel_ret(launch_ctx, ret_type, (i,)))
+            if len(ret_dt) == 1:
+                ret = ret[0]
+        if callbacks:
+            for c in callbacks:
+                c()
+
+        return ret
+
+    def construct_kernel_ret(self, launch_ctx: KernelLaunchContext, ret_type: Any, index: tuple[int, ...] = ()):
+        if isinstance(ret_type, CompoundType):
+            return ret_type.from_kernel_struct_ret(launch_ctx, index)
+        if ret_type in primitive_types.integer_types:
+            if is_signed(cook_dtype(ret_type)):
+                return launch_ctx.get_struct_ret_int(index)
+            return launch_ctx.get_struct_ret_uint(index)
+        if ret_type in primitive_types.real_types:
+            return launch_ctx.get_struct_ret_float(index)
+        raise GsTaichiRuntimeTypeError(f"Invalid return type on index={index}")
+
+    def ensure_compiled(self, *args: tuple[Any, ...]) -> tuple[Callable, int, AutodiffMode]:
+        instance_id, arg_features = self.mapper.lookup(args)
+        key = (self.func, instance_id, self.autodiff_mode)
+        self.materialize(key=key, args=args, arg_features=arg_features)
+        return key
+
+    # For small kernels (< 3us), the performance can be pretty sensitive to overhead in __call__
+    # Thus this part needs to be fast. (i.e. < 3us on a 4 GHz x64 CPU)
+    @_shell_pop_print
+    def __call__(self, *args, **kwargs) -> Any:
+        args = _process_args(self, is_func=False, args=args, kwargs=kwargs)
+
+        # Transform the primal kernel to forward mode grad kernel
+        # then recover to primal when exiting the forward mode manager
+        if self.runtime.fwd_mode_manager and not self.runtime.grad_replaced:
+            # TODO: if we would like to compute 2nd-order derivatives by forward-on-reverse in a nested context manager fashion,
+            # i.e., a `Tape` nested in the `FwdMode`, we can transform the kernels with `mode_original == AutodiffMode.REVERSE` only,
+            # to avoid duplicate computation for 1st-order derivatives
+            self.runtime.fwd_mode_manager.insert(self)
+
+        # Both the class kernels and the plain-function kernels are unified now.
+        # In both cases, |self.grad| is another Kernel instance that computes the
+        # gradient. For class kernels, args[0] is always the kernel owner.
+
+        # No need to capture grad kernels because they are already bound with their primal kernels
+        if (
+            self.autodiff_mode in (AutodiffMode.NONE, AutodiffMode.VALIDATION)
+            and self.runtime.target_tape
+            and not self.runtime.grad_replaced
+        ):
+            self.runtime.target_tape.insert(self, args)
+
+        if self.autodiff_mode != AutodiffMode.NONE and impl.current_cfg().opt_level == 0:
+            _logging.warn("""opt_level = 1 is enforced to enable gradient computation.""")
+            impl.current_cfg().opt_level = 1
+        key = self.ensure_compiled(*args)
+        kernel_cpp = self.materialized_kernels[key]
+        compiled_kernel_data = self.compiled_kernel_data_by_key.get(key, None)
+        return self.launch_kernel(kernel_cpp, compiled_kernel_data, *args)
+
+
+# For a GsTaichi class definition like below:
+#
+# @ti.data_oriented
+# class X:
+#   @ti.kernel
+#   def foo(self):
+#     ...
+#
+# When ti.kernel runs, the stackframe's |code_context| of Python 3.8(+) is
+# different from that of Python 3.7 and below. In 3.8+, it is 'class X:',
+# whereas in <=3.7, it is '@ti.data_oriented'. More interestingly, if the class
+# inherits, i.e. class X(object):, then in both versions, |code_context| is
+# 'class X(object):'...
+_KERNEL_CLASS_STACKFRAME_STMT_RES = [
+    re.compile(r"@(\w+\.)?data_oriented"),
+    re.compile(r"class "),
+]
+
+
+def _inside_class(level_of_class_stackframe: int) -> bool:
+    try:
+        maybe_class_frame = sys._getframe(level_of_class_stackframe)
+        statement_list = inspect.getframeinfo(maybe_class_frame)[3]
+        if statement_list is None:
+            return False
+        first_statment = statement_list[0].strip()
+        for pat in _KERNEL_CLASS_STACKFRAME_STMT_RES:
+            if pat.match(first_statment):
+                return True
+    except:
+        pass
+    return False
+
+
+def _kernel_impl(_func: Callable, level_of_class_stackframe: int, verbose: bool = False) -> GsTaichiCallable:
+    # Can decorators determine if a function is being defined inside a class?
+    # https://stackoverflow.com/a/8793684/12003165
+    is_classkernel = _inside_class(level_of_class_stackframe + 1)
+
+    if verbose:
+        print(f"kernel={_func.__name__} is_classkernel={is_classkernel}")
+    primal = Kernel(_func, autodiff_mode=AutodiffMode.NONE, _classkernel=is_classkernel)
+    adjoint = Kernel(_func, autodiff_mode=AutodiffMode.REVERSE, _classkernel=is_classkernel)
+    # Having |primal| contains |grad| makes the tape work.
+    primal.grad = adjoint
+
+    wrapped: GsTaichiCallable
+    if is_classkernel:
+        # For class kernels, their primal/adjoint callables are constructed
+        # when the kernel is accessed via the instance inside
+        # _BoundedDifferentiableMethod.
+        # This is because we need to bind the kernel or |grad| to the instance
+        # owning the kernel, which is not known until the kernel is accessed.
+        #
+        # See also: _BoundedDifferentiableMethod, data_oriented.
+        @functools.wraps(_func)
+        def wrapped_classkernel(*args, **kwargs):
+            # If we reach here (we should never), it means the class is not decorated
+            # with @ti.data_oriented, otherwise getattr would have intercepted the call.
+            clsobj = type(args[0])
+            assert not hasattr(clsobj, "_data_oriented")
+            raise GsTaichiSyntaxError(f"Please decorate class {clsobj.__name__} with @ti.data_oriented")
+
+        wrapped = GsTaichiCallable(_func, wrapped_classkernel)
+    else:
+
+        @functools.wraps(_func)
+        def wrapped_func(*args, **kwargs):
+            try:
+                return primal(*args, **kwargs)
+            except (GsTaichiCompilationError, GsTaichiRuntimeError) as e:
+                if impl.get_runtime().print_full_traceback:
+                    raise e
+                raise type(e)("\n" + str(e)) from None
+
+        wrapped = GsTaichiCallable(_func, wrapped_func)
+        wrapped.grad = adjoint
+
+    wrapped._is_wrapped_kernel = True
+    wrapped._is_classkernel = is_classkernel
+    wrapped._primal = primal
+    wrapped._adjoint = adjoint
+    primal.gstaichi_callable = wrapped
+    return wrapped
+
+
+def kernel(fn: Callable):
+    """Marks a function as a GsTaichi kernel.
+
+    A GsTaichi kernel is a function written in Python, and gets JIT compiled by
+    GsTaichi into native CPU/GPU instructions (e.g. a series of CUDA kernels).
+    The top-level ``for`` loops are automatically parallelized, and distributed
+    to either a CPU thread pool or massively parallel GPUs.
+
+    Kernel's gradient kernel would be generated automatically by the AutoDiff system.
+
+    See also https://docs.taichi-lang.org/docs/syntax#kernel.
+
+    Args:
+        fn (Callable): the Python function to be decorated
+
+    Returns:
+        Callable: The decorated function
+
+    Example::
+
+        >>> x = ti.field(ti.i32, shape=(4, 8))
+        >>>
+        >>> @ti.kernel
+        >>> def run():
+        >>>     # Assigns all the elements of `x` in parallel.
+        >>>     for i in x:
+        >>>         x[i] = i
+    """
+    return _kernel_impl(fn, level_of_class_stackframe=3)
+
+
+class _BoundedDifferentiableMethod:
+    def __init__(self, kernel_owner: Any, wrapped_kernel_func: GsTaichiCallable | BoundGsTaichiCallable):
+        clsobj = type(kernel_owner)
+        if not getattr(clsobj, "_data_oriented", False):
+            raise GsTaichiSyntaxError(f"Please decorate class {clsobj.__name__} with @ti.data_oriented")
+        self._kernel_owner = kernel_owner
+        self._primal = wrapped_kernel_func._primal
+        self._adjoint = wrapped_kernel_func._adjoint
+        self._is_staticmethod = wrapped_kernel_func._is_staticmethod
+        self.__name__: str | None = None
+
+    def __call__(self, *args, **kwargs):
+        try:
+            assert self._primal is not None
+            if self._is_staticmethod:
+                return self._primal(*args, **kwargs)
+            return self._primal(self._kernel_owner, *args, **kwargs)
+
+        except (GsTaichiCompilationError, GsTaichiRuntimeError) as e:
+            if impl.get_runtime().print_full_traceback:
+                raise e
+            raise type(e)("\n" + str(e)) from None
+
+    def grad(self, *args, **kwargs) -> Kernel:
+        assert self._adjoint is not None
+        return self._adjoint(self._kernel_owner, *args, **kwargs)
+
+
+def data_oriented(cls):
+    """Marks a class as GsTaichi compatible.
+
+    To allow for modularized code, GsTaichi provides this decorator so that
+    GsTaichi kernels can be defined inside a class.
+
+    See also https://docs.taichi-lang.org/docs/odop
+
+    Example::
+
+        >>> @ti.data_oriented
+        >>> class TiArray:
+        >>>     def __init__(self, n):
+        >>>         self.x = ti.field(ti.f32, shape=n)
+        >>>
+        >>>     @ti.kernel
+        >>>     def inc(self):
+        >>>         for i in self.x:
+        >>>             self.x[i] += 1.0
+        >>>
+        >>> a = TiArray(32)
+        >>> a.inc()
+
+    Args:
+        cls (Class): the class to be decorated
+
+    Returns:
+        The decorated class.
+    """
+
+    def _getattr(self, item):
+        method = cls.__dict__.get(item, None)
+        is_property = method.__class__ == property
+        is_staticmethod = method.__class__ == staticmethod
+        if is_property:
+            x = method.fget
+        else:
+            x = super(cls, self).__getattribute__(item)
+        if hasattr(x, "_is_wrapped_kernel"):
+            if inspect.ismethod(x):
+                wrapped = x.__func__
+            else:
+                wrapped = x
+            assert isinstance(wrapped, (BoundGsTaichiCallable, GsTaichiCallable))
+            wrapped._is_staticmethod = is_staticmethod
+            if wrapped._is_classkernel:
+                ret = _BoundedDifferentiableMethod(self, wrapped)
+                ret.__name__ = wrapped.__name__  # type: ignore
+                if is_property:
+                    return ret()
+                return ret
+        if is_property:
+            return x(self)
+        return x
+
+    cls.__getattribute__ = _getattr
+    cls._data_oriented = True
+
+    return cls
+
+
+__all__ = ["data_oriented", "func", "kernel", "pyfunc", "real_func"]