tilelang.ascend.language.frame
==============================

.. py:module:: tilelang.ascend.language.frame

.. autoapi-nested-parse::

   Execution-region frames for Ascend NPU.



Classes
-------

.. autoapisummary::

   tilelang.ascend.language.frame.CubeFrame
   tilelang.ascend.language.frame.VectorFrame
   tilelang.ascend.language.frame.SimtVFFrame
   tilelang.ascend.language.frame.SimdVFFrame


Functions
---------

.. autoapisummary::

   tilelang.ascend.language.frame.Cube
   tilelang.ascend.language.frame.Vector
   tilelang.ascend.language.frame.SimtVF
   tilelang.ascend.language.frame.inside_simdvf
   tilelang.ascend.language.frame.SimdVF


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

.. py:class:: CubeFrame

   Bases: :py:obj:`tvm.tirx.script.builder.frame.TIRFrame`


   Frame for Cube (AIC) execution region.


.. py:function:: Cube()

   Construct a Cube (AIC) execution region frame.


.. py:class:: VectorFrame

   Bases: :py:obj:`tvm.tirx.script.builder.frame.TIRFrame`


   Frame for Vector (AIV) execution region.

   In Mix mode, provides a subblock ID variable via ``as`` binding::

       with T.Vector(vector=2) as sid:
           # sid = asc_get_sub_block_id()
           ...


   .. py:method:: __enter__()


.. py:function:: Vector(vector = 2)

   Construct a Vector (AIV) execution region frame.

   :param vector: Number of AIV cores (1 or 2). Default 2.
   :type vector: int

   :returns: **res** -- The VectorFrame used to denote an AIV execution region.
   :rtype: VectorFrame


.. py:class:: SimtVFFrame

   Bases: :py:obj:`tvm.tirx.script.builder.frame.TIRFrame`


   Frame for SimtVF (SIMT Vector Fixed) region.

   The C++ frame generates the ``SIMT_VF`` block and its thread-extent
   bindings. The Python frame additionally maintains the active SIMT context.


   .. py:method:: __enter__()


   .. py:method:: __exit__(ptype, value, trace)


.. py:function:: SimtVF(threads = 128, latency = 0)

   Construct a dedicated SIMT vector-fragment region frame.

   :param threads: Thread count for the SimtVF frame. A sequence may contain up to three
                   dimensions; omitted dimensions default to one.
   :type threads: int | list[int] | tuple
   :param latency: Measured latency in cycles. Zero uses the scheduler's latency model.
   :type latency: int

   :returns: **res** -- The SimtVFFrame used to denote a SIMT vector-fragment region.
   :rtype: SimtVFFrame


.. py:class:: SimdVFFrame

   Bases: :py:obj:`tvm.tirx.script.builder.frame.TIRFrame`


   Frame for SimdVF (SIMD Vector Fragment) region.

   This is a register-level SIMD region without thread dimensions. The C++
   frame generates the ``SIMD_VF`` block and its scope marker.


   .. py:method:: __enter__()


   .. py:method:: __exit__(ptype, value, trace)


.. py:function:: inside_simdvf()

   Return whether the current builder is inside a SimdVF region.


.. py:function:: SimdVF(latency = 0)

   Construct a SimdVF region for Ascend NPU MicroAPI operations.

   :param latency: Measured latency in cycles. Zero uses the scheduler's latency model.
   :type latency: int

   :returns: **res** -- The SimdVFFrame used to denote a register-level SIMD region.
   :rtype: SimdVFFrame