Documentation,drivers/aie: align machine learning docs with current code

Documentation/applications/math/gemmlowp/index.rst

@@ -0,0 +1,13 @@
+=======================
+``gemmlowp`` gemmlowp
+=======================
+
+``apps/math/gemmlowp`` vendors Google `gemmlowp
+<https://github.com/google/gemmlowp>`_, a small low-precision matrix
+library used by TensorFlow Lite Micro.
+
+Enable ``CONFIG_MATH_GEMMLOWP``. The build downloads a pinned snapshot
+(``719139ce755a0f31cbf1c37f7f98adcc7fc9f425``). Headers are added from
+``apps/math/gemmlowp/gemmlowp``.
+
+:doc:`../../mlearning/tflite-micro/index` depends on this package.

Documentation/applications/math/kissfft/index.rst

@@ -0,0 +1,12 @@
+===================
+``kissfft`` KissFFT
+===================
+
+``apps/math/kissfft`` vendors `KissFFT
+<https://github.com/mborgerding/kissfft>`_ (tag ``v130``) and applies
+``kissfft.patch``. It compiles ``kiss_fft.c`` and ``tools/kiss_fftr.c``.
+
+Enable ``CONFIG_MATH_KISSFFT``. Include path:
+``apps/math/kissfft/kissfft``.
+
+:doc:`../../mlearning/tflite-micro/index` depends on this package.

Documentation/applications/math/ruy/index.rst

@@ -0,0 +1,12 @@
+=============
+``ruy`` Ruy
+=============
+
+``apps/math/ruy`` vendors Google `Ruy <https://github.com/google/ruy>`_,
+a matrix-multiplication library used by TensorFlow Lite Micro.
+
+Enable ``CONFIG_MATH_RUY``. The build downloads a pinned snapshot
+(``d37128311b445e758136b8602d1bbd2a755e115d``). Headers are added from
+``apps/math/ruy/ruy``.
+
+:doc:`../../mlearning/tflite-micro/index` depends on this package.

Documentation/applications/mlearning/cmsis-nn/index.rst

@@ -9,8 +9,9 @@ TensorFlow Lite Micro uses when ``CONFIG_MLEARNING_CMSIS_NN`` is enabled.
 It is distinct from :doc:`../cmsis/index`, which vendors the older CMSIS 5
 tree (DSP plus NN) for NNABLA.
 
-Enable with ``CONFIG_MLEARNING_CMSIS_NN``. The Makefile downloads a pinned
-CMSIS-NN commit and compiles:
+Enable with ``CONFIG_MLEARNING_CMSIS_NN``. The Makefile and CMake build
+download a pinned CMSIS-NN commit
+(``72e1ebf623ab1660a3e14e4e36fdcddce46f1991``) and compile:
 
 - Activation functions
 - Basic math

Documentation/applications/mlearning/cmsis/index.rst

@@ -5,10 +5,14 @@
 ``apps/mlearning/cmsis`` fetches `CMSIS 5
 <https://github.com/ARM-software/CMSIS_5>`_ and builds CMSIS-DSP and
 CMSIS-NN from that tree. Default version is ``5.8.0``
-(``CONFIG_CMSIS_VER``).
+(``CONFIG_CMSIS_VER``). ARM archived the CMSIS 5 repository on 18
+December 2025; NuttX still vendors this tree because NNABLA uses the
+CHW helpers in ``cmsis-nn-support_nnabla.patch``. There is no CMake
+build for this package.
 
 This package is aimed at NNABLA and generic DSP use. For TensorFlow Lite
 Micro kernel acceleration, use :doc:`../cmsis-nn/index` instead.
+``CONFIG_CMSIS_NN`` here is not ``CONFIG_MLEARNING_CMSIS_NN``.
 
 Enable ``CONFIG_CMSIS`` from :menuselection:`Application Configuration -->
 Machine Learning Support --> CMSIS Libraries`.

Documentation/applications/mlearning/darknet/index.rst

@@ -7,9 +7,12 @@
 Only Look Once) object detection.
 
 Enable ``CONFIG_DARKNET_YOLO``. ``CONFIG_DARKNET_YOLO_VER`` is the Git
-branch fetched at build time (default ``master``). The Makefile compiles
-the core Darknet sources (network, parser, convolutional and connected
-layers, YOLO/region/detection layers, RNN/GRU/LSTM, and image helpers).
+branch fetched at build time (default ``master``), not a commit pin.
+Upstream ``pjreddie/darknet`` is unmaintained, so the fetched tree can
+change between builds. There is no NuttX example application and no
+CMake build; the Makefile compiles the core Darknet sources (network,
+parser, convolutional and connected layers, YOLO/region/detection
+layers, RNN/GRU/LSTM, and image helpers).
 
 Include path: ``apps/mlearning/darknet/darknet/include``.
 

Documentation/applications/mlearning/index.rst

@@ -15,14 +15,20 @@ Enable the libraries from::
 
 The following packages are available:
 
-- :doc:`cmsis/index` — ARM CMSIS 5 DSP and NN libraries
+- :doc:`cmsis/index` — ARM CMSIS 5 DSP and NN libraries (Makefile only)
 - :doc:`cmsis-nn/index` — standalone ARM CMSIS-NN kernels for TensorFlow Lite Micro
-- :doc:`darknet/index` — Darknet YOLO object detection
-- :doc:`libnnablart/index` — Sony NNABLA C inference runtime
+- :doc:`darknet/index` — Darknet YOLO object detection (Makefile only)
+- :doc:`libnnablart/index` — Sony NNABLA C inference runtime (Makefile only)
 - :doc:`tflite-micro/index` — TensorFlow Lite for Microcontrollers
 
+``CONFIG_CMSIS_NN`` (nested under CMSIS Libraries) is not the same
+option as ``CONFIG_MLEARNING_CMSIS_NN`` (standalone CMSIS-NN used by
+TFLM). CMake currently builds ``tflite-micro`` and ``cmsis-nn`` only.
+
 A ready-to-build simulator configuration is ``sim:tflm``. See
-:doc:`/platforms/sim/sim/boards/sim/index`.
+:doc:`/platforms/sim/sim/boards/sim/index`. The kernel also has an
+optional AI-engine character driver (``CONFIG_AI_ENGINE``); see
+:doc:`/components/drivers/character/aie`.
 
 .. toctree::
    :glob:

Documentation/applications/mlearning/libnnablart/index.rst

@@ -7,7 +7,10 @@
 networks produced by Neural Network Libraries.
 
 Enable ``CONFIG_NNABLA_RT``. ``CONFIG_NNABLA_RT_VER`` selects the upstream
-tag (default ``1.24.0``). The Makefile downloads that release and compiles
+tag (default ``1.24.0``). Sony's later ``nnabla-c-runtime`` tags go
+through ``v1.38.0``; NuttX has not moved the default. Sony has placed
+NNABLA in maintenance. There is no NuttX example application and no
+CMake build. The Makefile downloads the selected release and compiles
 the runtime plus a subset of operators, including:
 
 - Activations (ReLU, sigmoid, tanh, softmax, ELU, leaky ReLU, Swish, ...)

Documentation/applications/mlearning/tflite-micro/index.rst

@@ -6,7 +6,8 @@ TensorFlow Lite for Microcontrollers (TFLM) is a C++ interpreter for
 running ``.tflite`` models on memory-constrained targets. NuttX integrates
 it from ``apps/mlearning/tflite-micro``.
 
-The build downloads a pinned TFLM snapshot from
+The build downloads a pinned TFLM snapshot
+(``cfa4c91d1b36c37c7c104b9c664615e59f1abfe3``, 24 February 2024) from
 https://github.com/tensorflow/tflite-micro and applies NuttX patches that
 add INT8-only operator registrations and an ``extern "C"`` entry point for
 the hello-world example.
@@ -16,10 +17,14 @@ Dependencies
 
 ``CONFIG_TFLITEMICRO`` depends on all of:
 
-- ``CONFIG_SYSTEM_FLATBUFFERS`` — FlatBuffers headers (``apps/system/flatbuffers``)
-- ``CONFIG_MATH_GEMMLOWP`` — gemmlowp (``apps/math/gemmlowp``)
-- ``CONFIG_MATH_KISSFFT`` — KissFFT (``apps/math/kissfft``)
-- ``CONFIG_MATH_RUY`` — Ruy (``apps/math/ruy``)
+- ``CONFIG_SYSTEM_FLATBUFFERS`` — FlatBuffers headers
+  (:doc:`../../system/flatbuffers/index`)
+- ``CONFIG_MATH_GEMMLOWP`` — gemmlowp
+  (:doc:`../../math/gemmlowp/index`)
+- ``CONFIG_MATH_KISSFFT`` — KissFFT
+  (:doc:`../../math/kissfft/index`)
+- ``CONFIG_MATH_RUY`` — Ruy
+  (:doc:`../../math/ruy/index`)
 
 TFLM is C++, so the configuration also needs C++ support (``CONFIG_HAVE_CXX``
 and a C++ standard library such as ``CONFIG_LIBCXX``).
@@ -95,10 +100,12 @@ unpacked tree under ``apps/mlearning/tflite-micro/tflite-micro``.
 Using the ``tflm`` tool
 =======================
 
-``tflm`` loads a ``.tflite`` file from the filesystem, constructs a
-``tflite::MicroInterpreter``, calls ``AllocateTensors()``, and can invoke
-the model once for profiling or emit compiled C++ (when TFLM was built
-with ``TFLITE_MODEL_COMPILER``).
+``tflm`` is a host-oriented NSH helper for ``sim:tflm``. It loads a
+``.tflite`` file from the filesystem with ``ifstream``, allocates the
+tensor arena with ``new``, constructs a ``tflite::MicroInterpreter``,
+and calls ``AllocateTensors()``. That heap-and-filesystem path is
+intentional on the simulator. On-target applications should embed the
+model as a C array; see *Embedding a model in an application* below.
 
 .. code-block:: console
 
@@ -116,10 +123,16 @@ with ``TFLITE_MODEL_COMPILER``).
 ``-i`` is required. ``-o`` is required only with ``-C``. Defaults are
 prefix ``NXAI`` and arena size 8192 bytes.
 
-The built-in operator resolver registers eight generic (float and
+``-C`` appears in the help text but is not functional in NuttX builds.
+Model compilation requires ``TFLITE_MODEL_COMPILER``, which neither the
+Makefile nor the CMake integration defines. ``tflm -C`` prints
+``Not supported compiling``.
+
+The built-in operator resolver registers nine generic (float and
 quantized) ops:
 
 - ``CONV_2D``
+- ``DEPTHWISE_CONV_2D``
 - ``MAX_POOL_2D``
 - ``QUANTIZE``
 - ``DEQUANTIZE``
@@ -184,17 +197,19 @@ and the NuttX apps tree next to ``nuttx`` (``../apps`` or
      nsh> tflm -E -i /path/to/model.tflite -a 8192
 
    The tool fails with ``AllocateTensors failed`` if the arena is too
-   small or the model uses operators outside the eight registered ops.
+   small or the model uses operators outside the nine registered ops.
 
 CMake is equivalent: ``cmake -B build -DBOARD_CONFIG=sim:tflm -GNinja``
 then ``cmake --build build`` and ``./build/nuttx``.
 
 Embedding a model in an application
 ===================================
 
-TFLM is designed for targets without a filesystem and without dynamic
-allocation for the model itself. Typical NuttX applications compile the
-``.tflite`` file into a C array and pass it to ``tflite::GetModel()``.
+The TFLM library is designed for targets without a filesystem and
+without dynamic allocation for the model itself. The ``tflm`` NSH tool
+is an exception used on ``sim:tflm``. Typical on-target applications
+compile the ``.tflite`` file into a C array and pass it to
+``tflite::GetModel()``.
 
 The CMake helper ``tflite_generate_data()`` in
 ``apps/mlearning/tflite-micro/CMakeLists.txt`` wraps ``xxd -i`` for that

Documentation/applications/system/flatbuffers/index.rst

@@ -0,0 +1,13 @@
+=============================
+``flatbuffers`` FlatBuffers
+=============================
+
+``apps/system/flatbuffers`` vendors Google `FlatBuffers
+<https://github.com/google/flatbuffers>`_ ``v23.5.26`` and applies
+``flatbuffers.patch``. TensorFlow Lite Micro uses the headers to read
+``.tflite`` model schemas.
+
+Enable ``CONFIG_SYSTEM_FLATBUFFERS``. Include path:
+``apps/system/flatbuffers/flatbuffers/include``.
+
+:doc:`../../mlearning/tflite-micro/index` depends on this package.

Documentation/components/drivers/character/aie.rst

@@ -0,0 +1,22 @@
+========================
+AI Engine (AIE) Drivers
+========================
+
+``drivers/aie`` is an upper-half character driver for hardware neural
+processing units. Enable ``CONFIG_AI_ENGINE``. There is no in-tree
+lower-half or board configuration yet; a platform must implement
+``struct aie_ops_s`` and call ``aie_register()``.
+
+The public header is ``include/nuttx/aie/ai_engine.h``. Ioctl commands:
+
+- ``AIE_CMD_LOAD`` — load a model. The argument is a model pointer
+  passed to ``ops->init()``. A second load on the same file returns
+  ``-EINVAL``.
+- ``AIE_CMD_FEED_INPUT`` — feed one input tensor via ``ops->feed_input()``.
+- ``AIE_CMD_GET_OUTPUT`` — read one output tensor via ``ops->get_output()``.
+
+Other ioctl numbers are forwarded to ``ops->control()`` when that
+callback is provided, otherwise ``-ENOSYS``.
+
+This kernel driver is independent of the application-level TinyML
+packages under :doc:`/applications/mlearning/index`.

Documentation/components/drivers/character/index.rst

@@ -55,6 +55,7 @@ Character device drivers have these properties:
   :maxdepth: 2
 
   1wire.rst
+  aie.rst
   analog/index.rst
   bch.rst
   can.rst

Documentation/components/drivers/index.rst

@@ -41,6 +41,11 @@ Subdirectories of ``nuttx/drivers``
 
   1wire device drivers.
 
+* ``aie/`` :doc:`character/aie`
+
+  Upper-half character driver for hardware AI / NPU engines.
+  See ``include/nuttx/aie/ai_engine.h``.
+
 * ``analog/`` :doc:`character/analog/index`
 
   This directory holds implementations of analog device drivers.

drivers/aie/CMakeLists.txt

@@ -0,0 +1,25 @@
+# ##############################################################################
+# drivers/aie/CMakeLists.txt
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more contributor
+# license agreements.  See the NOTICE file distributed with this work for
+# additional information regarding copyright ownership.  The ASF licenses this
+# file to you under the Apache License, Version 2.0 (the "License"); you may not
+# use this file except in compliance with the License.  You may obtain a copy of
+# the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.  See the
+# License for the specific language governing permissions and limitations under
+# the License.
+#
+# ##############################################################################
+
+if(CONFIG_AI_ENGINE)
+  target_sources(drivers PRIVATE ai_engine.c)
+endif()

drivers/aie/Kconfig

@@ -7,4 +7,7 @@ config AI_ENGINE
 	bool "AI Engine Acceleration Support"
 	default n
 	---help---
-		Drivers for various AI engine devices.
+		Upper-half character driver for hardware AI / NPU engines.
+		A board or architecture lower-half must implement struct
+		aie_ops_s and call aie_register(). There is no in-tree
+		lower-half yet. See include/nuttx/aie/ai_engine.h.

include/nuttx/aie/ai_engine.h

@@ -1,6 +1,8 @@
 /****************************************************************************
  * include/nuttx/aie/ai_engine.h
  *
+ * SPDX-License-Identifier: Apache-2.0
+ *
  * Licensed to the Apache Software Foundation (ASF) under one or more
  * contributor license agreements.  See the NOTICE file distributed with
  * this work for additional information regarding copyright ownership.  The
@@ -85,10 +87,10 @@ struct aie_lowerhalf_s
  ****************************************************************************/
 
 /****************************************************************************
- * Name: ai_engine_register
+ * Name: aie_register
  *
  * Description:
- *   Register all ai engine related drivers.
+ *   Register an AI engine character driver at path.
  *
  ****************************************************************************/