tilelang.profiler
=================

.. py:module:: tilelang.profiler

.. autoapi-nested-parse::

   The profiler and convert to torch utils



Submodules
----------

.. toctree::
   :maxdepth: 1

   /autoapi/tilelang/profiler/bench/index


Classes
-------

.. autoapisummary::

   tilelang.profiler.Profiler


Package Contents
----------------

.. py:class:: Profiler

   A profiler class for benchmarking and validating kernel implementations.

   .. attribute:: params

      List of kernel parameters defining the input/output specifications

   .. attribute:: result_idx

      Indices indicating which parameters are output tensors

   .. attribute:: supply_type

      Type of tensor supply to use (e.g., random, zeros, etc.)

   .. attribute:: adapter

      Optional kernel adapter for interfacing with different backends


   .. py:attribute:: params
      :type:  List[tilelang.engine.param.KernelParam]


   .. py:attribute:: result_idx
      :type:  List[int]


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


   .. py:attribute:: adapter
      :type:  Optional[tilelang.jit.adapter.BaseKernelAdapter]
      :value: None



   .. py:method:: __post_init__()

      Initialize tensor supply after dataclass initialization



   .. py:method:: with_default_adapter(adapter)


   .. py:method:: assert_allclose(reference_program, input_tensors = None, atol = 0.01, rtol = 0.01, max_mismatched_ratio=0.01)

      Validates kernel output against a reference implementation.

      :param reference_program: Reference implementation to compare against
      :param input_tensors: Optional pre-generated input tensors
      :param atol: Absolute tolerance for comparison
      :param rtol: Relative tolerance for comparison
      :param max_mismatched_ratio: Maximum allowed ratio of mismatched elements



   .. py:method:: manual_assert_close(reference_program, input_tensors = None, manual_check_prog = None)

      Validates kernel output against a reference implementation.

      :param reference_program: Reference implementation to compare against
      :param input_tensors: Optional pre-generated input tensors
      :param atol: Absolute tolerance for comparison
      :param rtol: Relative tolerance for comparison
      :param max_mismatched_ratio: Maximum allowed ratio of mismatched elements



   .. py:method:: assert_consistent(repeat=10)

      Checks for kernel consistency across multiple runs.

      :param repeat: Number of times to repeat the consistency check



   .. py:method:: run_once(func = None)


   .. py:method:: determine_profiler(func = None)

      Determines which profiler backend to use based on function type.

      :param func: Function to be profiled
      :param profiler: Explicitly specified profiler type or "auto" for automatic detection

      :returns: The determined profiler type ("torch" or "tvm")
      :rtype: str



   .. py:method:: do_bench(func = None, warmup = 25, rep = 100, n_warmup = 1, n_repeat = 1, input_tensors = None)

      Benchmarks the execution time of a given function.

      :param func: Function to benchmark (uses adapter if None)
      :param warmup: Warmup time in milliseconds
      :param rep: Number of repetitions for timing
      :param n_warmup: Number of warmup iterations
      :param n_repeat: Number of timing iterations
      :param profiler: Which profiling backend to use
      :param input_tensors: Optional pre-generated input tensors

      :returns: Average execution time in milliseconds
      :rtype: float



   .. py:property:: func


   .. py:method:: __call__(*args, **kwds)