tilelang.autotuner.param
========================

.. py:module:: tilelang.autotuner.param

.. autoapi-nested-parse::

   The auto-tune parameters.



Attributes
----------

.. autoapisummary::

   tilelang.autotuner.param.BEST_CONFIG_PATH
   tilelang.autotuner.param.FUNCTION_PATH
   tilelang.autotuner.param.LATENCY_PATH
   tilelang.autotuner.param.KERNEL_PATH
   tilelang.autotuner.param.WRAPPED_KERNEL_PATH
   tilelang.autotuner.param.KERNEL_LIB_PATH
   tilelang.autotuner.param.PARAMS_PATH


Classes
-------

.. autoapisummary::

   tilelang.autotuner.param.CompileArgs
   tilelang.autotuner.param.ProfileArgs
   tilelang.autotuner.param.AutotuneResult


Module Contents
---------------

.. py:data:: BEST_CONFIG_PATH
   :value: 'best_config.json'


.. py:data:: FUNCTION_PATH
   :value: 'function.pkl'


.. py:data:: LATENCY_PATH
   :value: 'latency.json'


.. py:data:: KERNEL_PATH
   :value: 'kernel.cu'


.. py:data:: WRAPPED_KERNEL_PATH
   :value: 'wrapped_kernel.cu'


.. py:data:: KERNEL_LIB_PATH
   :value: 'kernel_lib.so'


.. py:data:: PARAMS_PATH
   :value: 'params.pkl'


.. py:class:: CompileArgs

   Compile arguments for the auto-tuner. Detailed description can be found in `tilelang.jit.compile`.
   .. attribute:: out_idx

      List of output tensor indices.

   .. attribute:: execution_backend

      Execution backend to use for kernel execution (default: "cython").

   .. attribute:: target

      Compilation target, either as a string or a TVM Target object (default: "auto").

   .. attribute:: target_host

      Target host for cross-compilation (default: None).

   .. attribute:: verbose

      Whether to enable verbose output (default: False).

   .. attribute:: pass_configs

      Additional keyword arguments to pass to the Compiler PassContext.

   .. attribute:: Available options

      "tir.disable_vectorize": bool, default: False
      "tl.disable_tma_lower": bool, default: False
      "tl.disable_warp_specialized": bool, default: False
      "tl.config_index_bitwidth": int, default: None
      "tl.disable_dynamic_tail_split": bool, default: False
      "tl.dynamic_vectorize_size_bits": int, default: 128
      "tl.disable_safe_memory_legalize": bool, default: False


   .. py:attribute:: out_idx
      :type:  Optional[Union[List[int], int]]
      :value: None



   .. py:attribute:: execution_backend
      :type:  Literal['dlpack', 'ctypes', 'cython']
      :value: 'cython'



   .. py:attribute:: target
      :type:  Literal['auto', 'cuda', 'hip']
      :value: 'auto'



   .. py:attribute:: target_host
      :type:  Union[str, tvm.target.Target]
      :value: None



   .. py:attribute:: verbose
      :type:  bool
      :value: False



   .. py:attribute:: pass_configs
      :type:  Optional[Dict[str, Any]]
      :value: None



   .. py:method:: compile_program(program)


   .. py:method:: __hash__()


.. py:class:: ProfileArgs

   Profile arguments for the auto-tuner.

   .. attribute:: warmup

      Number of warmup iterations.

   .. attribute:: rep

      Number of repetitions for timing.

   .. attribute:: timeout

      Maximum time per configuration.

   .. attribute:: supply_type

      Type of tensor supply mechanism.

   .. attribute:: ref_prog

      Reference program for correctness validation.

   .. attribute:: supply_prog

      Supply program for input tensors.

   .. attribute:: out_idx

      Union[List[int], int] = -1

   .. attribute:: supply_type

      tilelang.TensorSupplyType = tilelang.TensorSupplyType.Auto

   .. attribute:: ref_prog

      Callable = None

   .. attribute:: supply_prog

      Callable = None

   .. attribute:: rtol

      float = 1e-2

   .. attribute:: atol

      float = 1e-2

   .. attribute:: max_mismatched_ratio

      float = 0.01

   .. attribute:: skip_check

      bool = False

   .. attribute:: manual_check_prog

      Callable = None

   .. attribute:: cache_input_tensors

      bool = True


   .. py:attribute:: warmup
      :type:  int
      :value: 25



   .. py:attribute:: rep
      :type:  int
      :value: 100



   .. py:attribute:: timeout
      :type:  int
      :value: 30



   .. py:attribute:: supply_type
      :type:  tilelang.TensorSupplyType


   .. py:attribute:: ref_prog
      :type:  Callable
      :value: None



   .. py:attribute:: supply_prog
      :type:  Callable
      :value: None



   .. py:attribute:: rtol
      :type:  float
      :value: 0.01



   .. py:attribute:: atol
      :type:  float
      :value: 0.01



   .. py:attribute:: max_mismatched_ratio
      :type:  float
      :value: 0.01



   .. py:attribute:: skip_check
      :type:  bool
      :value: False



   .. py:attribute:: manual_check_prog
      :type:  Callable
      :value: None



   .. py:attribute:: cache_input_tensors
      :type:  bool
      :value: True



   .. py:method:: __hash__()


.. py:class:: AutotuneResult

   Results from auto-tuning process.

   .. attribute:: latency

      Best achieved execution latency.

   .. attribute:: config

      Configuration that produced the best result.

   .. attribute:: ref_latency

      Reference implementation latency.

   .. attribute:: libcode

      Generated library code.

   .. attribute:: func

      Optimized function.

   .. attribute:: kernel

      Compiled kernel function.


   .. py:attribute:: latency
      :type:  Optional[float]
      :value: None



   .. py:attribute:: config
      :type:  Optional[dict]
      :value: None



   .. py:attribute:: ref_latency
      :type:  Optional[float]
      :value: None



   .. py:attribute:: libcode
      :type:  Optional[str]
      :value: None



   .. py:attribute:: func
      :type:  Optional[Callable]
      :value: None



   .. py:attribute:: kernel
      :type:  Optional[Callable]
      :value: None



   .. py:method:: save_to_disk(path, verbose = False)


   .. py:method:: load_from_disk(path, compile_args)
      :classmethod: