tilelang.contrib.cutedsl.utils
==============================

.. py:module:: tilelang.contrib.cutedsl.utils

.. autoapi-nested-parse::

   Utility functions for CuTeDSL backend.

   Provides common helpers used across the CuTeDSL codegen:
   bitcast, tensor construction, warp election, barrier sync, and FP16 packing.



Attributes
----------

.. autoapisummary::

   tilelang.contrib.cutedsl.utils.BYTES_PER_TENSORMAP
   tilelang.contrib.cutedsl.utils.BYTES_PER_POINTER
   tilelang.contrib.cutedsl.utils.type_map


Functions
---------

.. autoapisummary::

   tilelang.contrib.cutedsl.utils.bitcast
   tilelang.contrib.cutedsl.utils.make_filled_tensor
   tilelang.contrib.cutedsl.utils.make_tensor_at_offset
   tilelang.contrib.cutedsl.utils.shuffle_elect
   tilelang.contrib.cutedsl.utils.sync_thread_partial
   tilelang.contrib.cutedsl.utils.pack_half2


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

.. py:data:: BYTES_PER_TENSORMAP
   :value: 128


.. py:data:: BYTES_PER_POINTER
   :value: 8


.. py:data:: type_map

.. py:function:: bitcast(value, target_dtype)

   Reinterpret the bits of a value as a different type.
   Equivalent to C's (*(target_type *)(&value)).

   :param value: Source value (Numeric type from CuTeDSL)
   :param target_dtype: Target type (CuTeDSL type like Int8, Float16, etc.)

   :returns: Value reinterpreted as target type


.. py:function:: make_filled_tensor(shape, value)

.. py:function:: make_tensor_at_offset(ptr, offset, shape, div_by=1)

.. py:function:: shuffle_elect(thread_extent)

.. py:function:: sync_thread_partial(barrier_id=None, thread_count=None)

.. py:function:: pack_half2(x, y)

   Pack two half-precision (fp16) values into a single 32-bit value.
   Corresponds to CUDA's __pack_half2 intrinsic.

   This packs two fp16 values into a single int32 by treating the fp16 bits
   as raw data and concatenating them.