tilelang.contrib.cutedsl.reduce
===============================

.. py:module:: tilelang.contrib.cutedsl.reduce

.. autoapi-nested-parse::

   Reduce operations for CuTeDSL backend.
   Based on tl_templates/cuda/reduce.h



Classes
-------

.. autoapisummary::

   tilelang.contrib.cutedsl.reduce.SumOp
   tilelang.contrib.cutedsl.reduce.MaxOp
   tilelang.contrib.cutedsl.reduce.MinOp
   tilelang.contrib.cutedsl.reduce.BitAndOp
   tilelang.contrib.cutedsl.reduce.BitOrOp
   tilelang.contrib.cutedsl.reduce.BitXorOp


Functions
---------

.. autoapisummary::

   tilelang.contrib.cutedsl.reduce.min
   tilelang.contrib.cutedsl.reduce.max
   tilelang.contrib.cutedsl.reduce.bar_sync
   tilelang.contrib.cutedsl.reduce.bar_sync_ptx
   tilelang.contrib.cutedsl.reduce.AllReduce


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

.. py:function:: min(a, b, c = None, *, loc=None, ip=None)

.. py:function:: max(a, b, c = None, *, loc=None, ip=None)

.. py:class:: SumOp

   Sum reduction operator


   .. py:method:: __call__(x, y)
      :staticmethod:



.. py:class:: MaxOp

   Max reduction operator


   .. py:method:: __call__(x, y)
      :staticmethod:



.. py:class:: MinOp

   Min reduction operator


   .. py:method:: __call__(x, y)
      :staticmethod:



.. py:class:: BitAndOp

   Bitwise AND reduction operator


   .. py:method:: __call__(x, y)
      :staticmethod:



.. py:class:: BitOrOp

   Bitwise OR reduction operator


   .. py:method:: __call__(x, y)
      :staticmethod:



.. py:class:: BitXorOp

   Bitwise XOR reduction operator


   .. py:method:: __call__(x, y)
      :staticmethod:



.. py:function:: bar_sync(barrier_id, number_of_threads)

.. py:function:: bar_sync_ptx(barrier_id, number_of_threads)

.. py:function:: AllReduce(reducer, threads, scale, thread_offset, all_threads=None)

   AllReduce operation implementing warp/block-level reduction.
   Based on tl::AllReduce from reduce.h

   :param reducer: Reducer operator class (SumOp, MaxOp, etc.)
   :param threads: Number of threads participating in reduction
   :param scale: Reduction scale factor
   :param thread_offset: Thread ID offset
   :param all_threads: Total number of threads in block

   :returns: A callable object with run() and run_hopper() methods