tilelang.contrib.rocm
=====================

.. py:module:: tilelang.contrib.rocm

.. autoapi-nested-parse::

   Utility for ROCm backend



Functions
---------

.. autoapisummary::

   tilelang.contrib.rocm.find_lld
   tilelang.contrib.rocm.rocm_link
   tilelang.contrib.rocm.callback_rocm_link
   tilelang.contrib.rocm.callback_rocm_bitcode_path
   tilelang.contrib.rocm.parse_compute_version
   tilelang.contrib.rocm.have_matrixcore
   tilelang.contrib.rocm.get_rocm_arch
   tilelang.contrib.rocm.find_rocm_path


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

.. py:function:: find_lld(required=True)

   Find ld.lld in system.

   :param required: Whether it is required,
                    runtime error will be raised if the compiler is required.
   :type required: bool

   :returns: **valid_list** -- List of possible paths.
   :rtype: list of str

   .. note::

      This function will first search ld.lld that
      matches the major llvm version that built with tvm


.. py:function:: rocm_link(in_file, out_file, lld=None)

   Link relocatable ELF object to shared ELF object using lld

   :param in_file: Input file name (relocatable ELF object file)
   :type in_file: str
   :param out_file: Output file name (shared ELF object file)
   :type out_file: str
   :param lld: The lld linker, if not specified,
               we will try to guess the matched clang version.
   :type lld: str, optional


.. py:function:: callback_rocm_link(obj_bin)

   Links object file generated from LLVM to HSA Code Object

   :param obj_bin: The object file
   :type obj_bin: bytearray

   :returns: **cobj_bin** -- The HSA Code Object
   :rtype: bytearray


.. py:function:: callback_rocm_bitcode_path(rocdl_dir=None)

   Utility function to find ROCm device library bitcodes

   :param rocdl_dir: The path to rocm library directory
                     The default value is the standard location
   :type rocdl_dir: str


.. py:function:: parse_compute_version(compute_version)

   Parse compute capability string to divide major and minor version

   :param compute_version: compute capability of a GPU (e.g. "6.0")
   :type compute_version: str

   :returns: * **major** (*int*) -- major version number
             * **minor** (*int*) -- minor version number


.. py:function:: have_matrixcore(compute_version=None)

   Either MatrixCore support is provided in the compute capability or not

   :param compute_version: compute capability of a GPU (e.g. "7.0").
   :type compute_version: str, optional

   :returns: **have_matrixcore** -- True if MatrixCore support is provided, False otherwise
   :rtype: bool


.. py:function:: get_rocm_arch(rocm_path='/opt/rocm')

   Utility function to get the AMD GPU architecture

   :param rocm_path: The path to rocm installation directory
   :type rocm_path: str

   :returns: **gpu_arch** -- The AMD GPU architecture
   :rtype: str


.. py:function:: find_rocm_path()

   Utility function to find ROCm path

   :returns: **path** -- Path to ROCm root.
   :rtype: str