tilelang.contrib.hip_resource_info
==================================

.. py:module:: tilelang.contrib.hip_resource_info

.. autoapi-nested-parse::

   Parse AMD GPU per-kernel resource usage out of clang's
   ``-Rpass-analysis=kernel-resource-usage`` remarks and expose them on
   JITKernel.

   clang emits a block like::
       remark: src.cc:9:0: Function Name: main_kernel [-Rpass-analysis=kernel-resource-usage]
       remark: src.cc:9:0:     TotalSGPRs: 16 [-Rpass-analysis=kernel-resource-usage]
       remark: src.cc:9:0:     VGPRs: 5 [-Rpass-analysis=kernel-resource-usage]
       remark: src.cc:9:0:     ScratchSize [bytes/lane]: 0 [-Rpass-analysis=kernel-resource-usage]
       remark: src.cc:9:0:     SGPRs Spill: 0 [-Rpass-analysis=kernel-resource-usage]
       remark: src.cc:9:0:     VGPRs Spill: 0 [-Rpass-analysis=kernel-resource-usage]
       ...

   right alongside any real warnings/errors. We *parse and strip* those
   lines before printing or raising, so autotune logs don't drown in
   hundreds of remark blocks while real diagnostics still surface.



Classes
-------

.. autoapisummary::

   tilelang.contrib.hip_resource_info.KernelResourceUsage


Functions
---------

.. autoapisummary::

   tilelang.contrib.hip_resource_info.hipcc_remark_flag
   tilelang.contrib.hip_resource_info.reset_recorder
   tilelang.contrib.hip_resource_info.pop_recorded
   tilelang.contrib.hip_resource_info.filter_and_record
   tilelang.contrib.hip_resource_info.dump_to_file
   tilelang.contrib.hip_resource_info.load_from_file


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

.. py:class:: KernelResourceUsage

   Resource counts as reported by clang's kernel-resource-usage pass.

   Field names mirror the remark labels (lower-cased, normalized) so we
   can extend without breaking callers.


   .. py:attribute:: n_regs
      :type:  int
      :value: 0



   .. py:attribute:: n_spills
      :type:  int
      :value: 0



   .. py:attribute:: scratch_bytes
      :type:  int
      :value: 0



   .. py:attribute:: n_max_threads
      :type:  int | None
      :value: None



   .. py:attribute:: extra
      :type:  dict[str, str]


.. py:function:: hipcc_remark_flag()

   The clang flag callers should pass to hipcc to enable the remark
   output we parse here.


.. py:function:: reset_recorder()

   Begin a fresh recording window on this thread.


.. py:function:: pop_recorded()

   Return everything recorded since the last ``reset_recorder`` and
   clear the buffer.


.. py:function:: filter_and_record(output)

   Strip kernel-resource-usage remarks from ``output``, parse them,
   and append the parsed entries to the active recorder (if any).
   Returns the filtered output with the remark lines removed.


.. py:function:: dump_to_file(usage, path)

   Persist parsed resource usage so it survives kernel-cache hits.


.. py:function:: load_from_file(path)

   Inverse of ``dump_to_file``. Tolerant of missing / unknown fields
   so older cache entries keep working when the dataclass evolves.