Merge branch 'develop' into bdevorem/geg_heuristics

Author: bdevorem

CHANGELOG.md

@@ -5,10 +5,25 @@ Full documentation for MIGraphX is available at
 
 ## Develop Branch
 
+
+### Added
+
+
+### Changed
+
+* Updated README `rbuild` installation instructions to use python venv to avoid warning (#4405)
+
 ### Fixed
 
 * Fixed `pointwise: Wrong number of arguments` error when quantizing certain models to `int8` [(#4398)](https://github.com/ROCm/AMDMIGraphX/pull/4398).
 
+### Optimized
+
+
+### Removed
+
+
+
 ## MIGraphX 2.14 for ROCm 7.1.0
 
 ### Added

README.md

@@ -24,7 +24,7 @@ version.
 
 You have three options for building from source:
 
-* [ROCm build tool](#use-the-rocm-build-tool-rbuild): Uses
+* [ROCm build tool](#compile-using-the-rocm-build-tool-rbuild): Uses
   [rbuild](https://github.com/RadeonOpenCompute/rbuild) to install prerequisites, then you can build
   the libraries with a single command.
 
@@ -50,26 +50,41 @@ The following is a list of prerequisites for building MIGraphX.
 * [MessagePack](https://msgpack.org/index.html) for model serialization to binary format
 * [SQLite3](https://www.sqlite.org/index.html) to create database of kernels' tuning information or run queries on existing database
 
-### Use the ROCm build tool [rbuild](https://github.com/RadeonOpenCompute/rbuild).
+### Installing the ROCm build tool [rbuild](https://github.com/RadeonOpenCompute/rbuild).
 
 1. Install `rocm-cmake`, `pip3`, `rocblas`, and `miopen-hip`:
 
     ```bash
     sudo apt install -y rocm-cmake python3-pip rocblas miopen-hip
     ```
 
-2. Install [rbuild](https://github.com/RadeonOpenCompute/rbuild) (sudo may be required):
+2. Set up and activate python venv for rbuild:
+    ```bash
+    python3 -m venv venv_rbuild
+    source venv_rbuild/bin/activate
+    ```
+
+3. Install [rbuild](https://github.com/RadeonOpenCompute/rbuild) in the python venv:
 
     ```bash
-    pip3 install https://github.com/RadeonOpenCompute/rbuild/archive/master.tar.gz
+    pip install https://github.com/RadeonOpenCompute/rbuild/archive/master.tar.gz
     ```
 
-3. Build MIGraphX source code:
+> [!NOTE]
+> You can deactivate the python virtual environment with `deactivate`.
+
+
+### Compile using the ROCm build tool [rbuild](https://github.com/RadeonOpenCompute/rbuild).
+
+1. Follow instructions in [installing rbuild](#installing-the-rocm-build-tool-rbuild)
+
+2. Build MIGraphX source code:
 
     ```bash
     rbuild build -d depend -B build -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')
     ```
 
+
 Once completed, all prerequisites are in the `depend` folder and MIGraphX is in the `build` directory.
 
 > [!NOTE]
@@ -79,7 +94,9 @@ Once completed, all prerequisites are in the `depend` folder and MIGraphX is in
 
 ### Use CMake to build MIGraphX
 
-1. Install the prerequisites:
+1. Follow instructions in [installing rbuild](#installing-the-rocm-build-tool)
+
+2. Install the prerequisites:
 
     ```bash
     rbuild prepare -d depend
@@ -95,22 +112,22 @@ Once completed, all prerequisites are in the `depend` folder and MIGraphX is in
     users. For the default location, `sudo` is required to run the script. You can also specify a different
     location using `./tools/install_prereqs.sh $custom_location`.
 
-2. Go to the project folder and create a `build` directory:
+3. Go to the project folder and create a `build` directory:
 
     ```bash
     mkdir build
     cd build
     ```
 
-3. Configure CMake. If the prerequisites are installed at the default location `/usr/local`, use:
+4. Configure CMake. If the prerequisites are installed at the default location `/usr/local`, use:
 
     ```bash
     CXX=/opt/rocm/llvm/bin/clang++ cmake .. -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')
     ```
 
     Otherwise, you need to set `-DCMAKE_PREFIX_PATH=$your_loc` to configure CMake.
 
-4. Build MIGraphX source code:
+5. Build MIGraphX source code:
 
     ```cpp
     make -j$(nproc)
@@ -122,7 +139,7 @@ Once completed, all prerequisites are in the `depend` folder and MIGraphX is in
     make -j$(nproc) check
     ```
 
-5. Install MIGraphX libraries:
+6. Install MIGraphX libraries:
 
     ```cpp
     make install
@@ -210,7 +227,7 @@ This will build a local searchable web site inside the docs/html folder.
 
 Documentation is built using [Doxygen](http://www.stack.nl/~dimitri/doxygen/download.html) and [rocm-docs-core](https://github.com/RadeonOpenCompute/rocm-docs-core)
 
-Run the steps below to build documentation locally.
+Run the steps below to build documentation locally. You can reuse the same venv from [installing rbuild](#installing-the-rocm-build-tool-rbuild).
 
 ```bash
 cd docs

docs/conceptual/deep-learning-compilation.rst

@@ -0,0 +1,49 @@
+.. meta::
+  :description: Deep learning compilation with MIGraphX
+  :keywords: MIGraphX, ROCm, library, API
+
+.. _deep-learning-compilation:
+
+********************************************************************
+Deep learning compilation with MIGraphX
+********************************************************************
+
+The MIGraphX deep learning (DL) compiler improves inference by analyzing 
+a model’s compute graph and applying program transformations. After 
+optimization, the compiler lowers graph operations to device kernels (from 
+libraries or via code generation) for efficient execution.
+
+A common transformation is kernel fusion; where compatible operations are 
+merged into a single kernel launch. Fusion reduces launch overhead and 
+avoids extra reads or writes between host and device, which typically 
+improves latency and throughput. By applying graph-level optimizations 
+and choosing or generating efficient device kernels, MIGraphX delivers 
+high-performance over uncompiled models and less optimized compiled solutions.
+
+An overview of the compilation process for MIGraphX is shown below in 
+:numref:`compilation-label`. One type of optimization that MIGraphX 
+performs are kernel fusions such as the Attention fusion seen in 
+:numref:`attention-label`.
+
+.. image:: ../data/migraphx_compilation_flow.png
+   :scale: 50%
+   :alt: Simplified overview of the MIGraphX compilation process
+   :name: compilation-label
+
+.. image:: ../data/attention_fusion.png
+   :scale: 50%
+   :alt: Attention operations fused into a single kernel
+   :name: attention-label
+
+What MIGraphX provides
+====================================================================
+
+* **End-to-end:** compilation and execution of DL models on AMD GPUs
+* **C++ implementation:** with Python and C++ APIs
+* **Model inputs:**
+    * ONNX and TensorFlow
+    * PyTorch through `https://github.com/ROCm/torch_migraphx <https://github.com/ROCm/torch_migraphx>`__
+    * ONNX Runtime execution provider
+* **Hardware targets:** AMD Navi (consumer) and MI (server) GPUs
+* **Supported data types:** FP16, BF16, OCP FP8, INT8, INT4
+

docs/data/attention_fusion.png

@@ -8,15 +8,8 @@
 MIGraphX documentation
 ===========================
 
-MIGraphX is a graph inference engine and graph compiler. MIGraphX accelerates machine-learning models by leveraging several graph-level transformations and optimizations. These optimizations include:
-
-* Operator fusion
-* Arithmetic simplifications
-* Dead-code elimination
-* Common subexpression elimination (CSE)
-* Constant propagation
-
-After optimization, MIGraphX generates code for AMD GPUs by calling various ROCm libraries to create the fastest combinations of HIP kernels.
+MIGraphX is a graph compiler and inference engine for high performance machine learning model inference.
+It compiles trained models from end-to-end to optimize for inference performance on AMD hardware.
 
 The MIGraphX public repository is located at `https://github.com/ROCm/AMDMIGraphX/ <https://github.com/ROCm/AMDMIGraphX/>`_
 
@@ -25,8 +18,11 @@ The MIGraphX public repository is located at `https://github.com/ROCm/AMDMIGraph
 
   .. grid-item-card:: Install
 
-    * :doc:`Installing MIGraphX with the package installer <./install/installing_with_package>`
-    * :doc:`Building and installing MIGraphX from source code <./install/building_migraphx>`
+    * :doc:`MIGraphX on ROCm installation <./install/install-migraphx>`
+
+  .. grid-item-card:: Conceptual
+
+    * :doc:`Deep learning compilation with MIGraphX <./conceptual/deep-learning-compilation>`
 
   .. grid-item-card:: Reference
 
@@ -51,3 +47,4 @@ To contribute to the documentation refer to
 
 Licensing information can be found on the
 `Licensing <https://rocm.docs.amd.com/en/latest/about/license.html>`_ page.
+

docs/data/attention_fusion.svg

@@ -0,0 +1,51 @@
+.. meta::
+  :description: Installing MIGraphX for ROCm
+  :keywords: install, build, MIGraphX, AMD, ROCm, package installer, development, contributing
+
+********************************************************************
+MIGraphX on ROCm installation
+********************************************************************
+
+ROCm must be installed before installing MIGraphX. See `ROCm installation 
+for Linux <https://rocm.docs.amd.com/projects/install-on-linux/en/latest/>`__ 
+for instructions.
+
+Installing MIGraphX using a package installer is sufficient for most users 
+who want to use the MIGraphX API. If you plan to develop for MIGraphX or 
+contribute to the source code, see `Developing for MIGraphX <../dev/contributing-to-migraphx.html>`_
+
+Install MIGraphX with a package installer
+====================================================================
+
+The package installer will install all the prerequisites you need for MIGraphX.
+
+Use the following command to install MIGraphX: 
+
+  .. code:: shell
+  
+    sudo apt update && sudo apt install -y migraphx
+
+Build MIGraphX from source
+====================================================================
+
+.. note::
+  
+  This method for building MIGraphX requires using ``sudo``.
+
+1. Install `rocm-cmake`, `pip3`, `rocblas`, and `miopen-hip`:
+
+    .. code:: shell
+
+        sudo apt install -y rocm-cmake python3-pip rocblas miopen-hip
+    
+2. Install `rbuild <https://github.com/RadeonOpenCompute/rbuild>`_:
+
+    .. code:: shell
+
+        pip3 install --prefix /usr/local https://github.com/RadeonOpenCompute/rbuild/archive/master.tar.gz
+    
+3. Build MIGraphX source code:
+
+    .. code:: shell
+
+        sudo rbuild build -d depend -B build -DGPU_TARGETS=$(/opt/rocm/bin/rocminfo | grep -o -m1 'gfx.*')