tilelang.jit.adapter.ctypes.adapter module

The profiler and convert to torch utils

class tilelang.jit.adapter.ctypes.adapter.CtypesKernelAdapter(params: List[TensorType], result_idx: List[int], target: str, func_or_mod: Union[PrimFunc, IRModule], host_mod: Optional[IRModule] = None, device_mod: Optional[IRModule] = None, kernel_global_source: Optional[str] = None, verbose: bool = False, pass_configs: Optional[Dict[str, Any]] = None)

Bases: BaseKernelAdapter

Adapter class that converts TVM/TIR functions to callable CUDA kernels using ctypes.

This adapter handles: 1. Converting TIR functions to compiled CUDA libraries 2. Managing dynamic shapes in tensor operations 3. Wrapping C++ kernels for Python/PyTorch usage

dynamic_symbolic_map: Optional[Dict[Var, Tuple[int, int]]] = None

classmethod from_database(params: List[TensorType], result_idx: List[int], target: str, func_or_mod: Union[PrimFunc, IRModule], kernel_global_source: str, kernel_lib_path: str, verbose: bool = False, pass_configs: Optional[Dict[str, Any]] = None)

get_kernel_source(kernel_only: bool = False)

Returns the source code of the compiled kernel.

ir_module: Optional[IRModule] = None

property is_dynamic

Indicates whether the kernel handles dynamic shapes.

kernel_global_source: Optional[str] = None

lib: Optional[CDLL] = None

property lib_code

Returns the code of the compiled library.

property libpath

Returns the path to the compiled library.

param_dtypes: Optional[List[torch.dtype]] = None

param_shapes: Optional[List[List]] = None

pass_configs: Optional[Dict[str, Any]] = None

property prim_func: PrimFunc

Returns the primary TIR function from the IR module.

property srcpath

Returns the source path of the compiled library.

target = 'cuda'

wrapped_source: Optional[str] = None