Skip to content

nod-ai/iree-amd-aie

Repository files navigation

AMD AIE Plugin for IREE

This repository contains an early-phase IREE compiler and runtime plugin for interfacing the AMD AIE accelerator to IREE.

Architectural Overview

image

Developer Setup

These instructions assume that you have an appropriate IREE checkout side by side with this repository have an IREE build setup in an iree-build directory that is also a sibling. This is not a requirement, but instructions will need to be changed for different paths.

Preparing repository:

git submodule update --init

Building (along with IREE)

Just show me the CMake

cmake -B $WHERE_YOU_WOULD_LIKE_TO_BUILD -S $IREE_REPO_SRC_DIR \
-DIREE_CMAKE_PLUGIN_PATHS=$IREE_AMD_AIE_REPO_SRC_DIR -DIREE_BUILD_PYTHON_BINDINGS=ON \
-DIREE_INPUT_STABLEHLO=OFF -DIREE_INPUT_TORCH=OFF -DIREE_INPUT_TOSA=OFF \
-DIREE_HAL_DRIVER_DEFAULTS=OFF -DIREE_TARGET_BACKEND_DEFAULTS=OFF -DIREE_TARGET_BACKEND_LLVM_CPU=ON \
-DIREE_BUILD_TESTS=ON -DIREE_EXTERNAL_HAL_DRIVERS=xrt \
-DCMAKE_INSTALL_PREFIX=$WHERE_YOU_WOULD_LIKE_TO_INSTALL

Instructions

To pin IREE and its submodules (LLVM, etc) to commits which are compatible with this plugin, run

python3 sync_deps.py

from within the iree-amd-aie root directory. Then the bare minimum CMake configure command is

cd ../iree-build
cmake -DIREE_BUILD_PYTHON_BINDINGS=ON -DIREE_CMAKE_PLUGIN_PATHS=$PWD/../iree-amd-aie .
ninja

to build IREE with amd-aie plugin. Very likely, you will want to use ccache and lld (or some other modern linker like mold)

-DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache
-DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=lld" -DCMAKE_SHARED_LINKER_FLAGS="-fuse-ld=lld"

Note, if you don't plan on using any of IREE's frontends or backends/targets (e.g., you're doing work on this code base itself), you can opt-out of everything (except the llvm-cpu backend) with

-DIREE_INPUT_STABLEHLO=OFF -DIREE_INPUT_TORCH=OFF -DIREE_INPUT_TOSA=OFF
-DIREE_HAL_DRIVER_DEFAULTS=OFF -DIREE_TARGET_BACKEND_DEFAULTS=OFF
-DIREE_TARGET_BACKEND_LLVM_CPU=ON

With the above you can also skip cloning the stablehlo and torch-mlir submodules/repos but in this case you will need to add

-DIREE_ERROR_ON_MISSING_SUBMODULES=OFF

Finally, if you're "bringing your own LLVM", i.e., you have a prebuilt/compiled distribution of LLVM you'd like to use, you can add

-DIREE_BUILD_BUNDLED_LLVM=OFF

Note, in this case you will need to supply -DLLVM_EXTERNAL_LIT=$SOMEWHERE (e.g., pip install lit; SOMEWHERE=$(which lit)).

Lit tests specific to AIE can be run with something like:

ctest -R amd-aie

Runtime driver setup

To enable the runtime driver, you need to also enable the XRT HAL:

cd ../iree-build
cmake . -DIREE_CMAKE_PLUGIN_PATHS=../iree-amd-aie \
  -DIREE_EXTERNAL_HAL_DRIVERS=xrt
ninja

Ubuntu Dependencies

XRT requires a number of packages. Here are the requirements for various operating systems:

apt install \
  libcurl4-openssl-dev \
  libdrm-dev \
  libelf-dev \
  libprotobuf-dev \
  libudev-dev \
  pkg-config \
  protobuf-compiler \
  python3-pybind11 \
  systemtap-sdt-dev

RH Based Deps

This is an incomplete list derived by adding what is needed to our development base manylinux (AlmaLinux 8) image.

yum install \
  libcurl-devel \
  libdrm-devel \
  libudev-devel \
  libuuid-devel \
  ncurses-devel \
  pkgconfig \
  protobuf-compiler \
  protobuf-devel \
  systemtap-sdt-devel