tilelang.layout.swizzle
=======================

.. py:module:: tilelang.layout.swizzle

.. autoapi-nested-parse::

   Wrapping Layouts.



Functions
---------

.. autoapisummary::

   tilelang.layout.swizzle.make_swizzled_layout
   tilelang.layout.swizzle.make_volta_swizzled_layout
   tilelang.layout.swizzle.make_wgmma_swizzled_layout
   tilelang.layout.swizzle.make_tcgen05mma_swizzled_layout
   tilelang.layout.swizzle.make_full_bank_swizzled_layout
   tilelang.layout.swizzle.make_half_bank_swizzled_layout
   tilelang.layout.swizzle.make_quarter_bank_swizzled_layout
   tilelang.layout.swizzle.make_linear_layout
   tilelang.layout.swizzle.make_gemm_fragment_8x8
   tilelang.layout.swizzle.make_gemm_fragment_8x8_transposed
   tilelang.layout.swizzle.make_fully_replicated_layout_fragment


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

.. py:function:: make_swizzled_layout(buffer, k_major = True, allow_pad = True)

.. py:function:: make_volta_swizzled_layout(buffer, is_a = True, k_inner = True)

.. py:function:: make_wgmma_swizzled_layout(buffer, continuity = None, k_major = True)

.. py:function:: make_tcgen05mma_swizzled_layout(buffer, continuity = None, k_major = True)

.. py:function:: make_full_bank_swizzled_layout(buffer)

   :param buffer: BufferLikeType

   .. rubric:: 範例

   make_full_bank_swizzled_layout(buffer)


.. py:function:: make_half_bank_swizzled_layout(buffer)

   :param buffer: BufferLikeType

   .. rubric:: 範例

   make_half_bank_swizzled_layout(buffer)


.. py:function:: make_quarter_bank_swizzled_layout(buffer)

   :param buffer: BufferLikeType

   .. rubric:: 範例

   make_quarter_bank_swizzled_layout(buffer)


.. py:function:: make_linear_layout(buffer_or_load_or_region)

   Create a row-major linear layout for any dimension.

   :param buffer_or_load_or_region: BufferLikeType

   :returns: A row-major linear layout
   :rtype: Layout


.. py:function:: make_gemm_fragment_8x8()

   Create a standard 8x8 GEMM fragment layout for ldmatrix/stmatrix.

   This layout matches the warp-level matrix multiplication pattern used in tensor cores.

   :returns: An 8x8 fragment layout
   :rtype: Fragment


.. py:function:: make_gemm_fragment_8x8_transposed()

   Create a transposed 8x8 GEMM fragment layout for ldmatrix/stmatrix.

   This layout is the transposed version of make_gemm_fragment_8x8, useful for
   different access patterns in matrix operations.

   :returns: A transposed 8x8 fragment layout
   :rtype: Fragment


.. py:function:: make_fully_replicated_layout_fragment(buffer, threads)

   Create a fully replicated layout for a fragment buffer.

   A fully replicated fragment means all threads hold identical copies of the
   entire buffer. This is useful for index buffers or masks that need to be
   accessed uniformly across all threads.

   :param buffer: BufferLikeType to get shape information
   :param threads: Number of threads (replicate extent)

   :returns: A fully replicated layout where each thread has a complete copy
   :rtype: Fragment

   .. rubric:: 範例

   >>> C_local = T.alloc_fragment((2,), T.float32)
   >>> layout = make_fully_replicated_layout_fragment(C_local, 256)
   >>> T.annotate_layout({C_local: layout})