tilelang.jit.adapter.cython.adapter module

The profiler and convert to torch utils

class tilelang.jit.adapter.cython.adapter.CythonKernelAdapter(params: List[KernelParam], result_idx: List[int], target: Union[str, Target], 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

buffer_device_map: Optional[Dict[Var, Tuple[int, torch.device]]] = None

buffer_dtype_map: Optional[Dict[Var, Tuple[int, torch.dtype]]] = None

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.

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

property prim_func: PrimFunc

Returns the primary TIR function from the IR module.

ptr_map: Optional[Dict[int, str]] = None

property srcpath

Returns the source path of the compiled library.

static_shape_map: Optional[Dict[Var, Tuple[int, List[Tuple[int, int]]]]] = None

target: Union[str, Target] = 'cuda'

wrapped_source: Optional[str] = None

tilelang.jit.adapter.cython.adapter.get_cache_dir() → Path

Get the cache directory for the current Python version.

tilelang.jit.adapter.cython.adapter.get_cached_lib(source_code: str) → Tuple[Optional[CDLL], Path]

Try to load cached library or return None if not found.

tilelang.jit.adapter.cython.adapter.get_cython_compiler() → Optional[str]

Return the path to the Cython compiler.

Returns:

out – The path to the Cython compiler, or None if none was found.

Return type:

Optional[str]