Building the code

Prerequisites

lld
cmake 3.20.6
ninja 1.8.2
Xilinx Vitis 2023.2
python 3.8.x and pip
virtualenv
pip3 install psutil rich pybind11 numpy
clang/llvm 14+ from source https://github.com/llvm/llvm-project

Xilinx Vitis can be downloaded and installed from the Xilinx Downloads site.

In order to successfully install Vitis on a fresh bare-bones Ubuntu install, some additional prerequisites are required, documented here. For Ubuntu 20.04, the installation should succeed if you additionally install the following packages: libncurses5 libtinfo5 libncurses5-dev libncursesw5-dev ncurses-compat-libs libstdc++6:i386 libgtk2.0-0:i386 dpkg-dev:i386 python3-pip Further note that the above mentioned cmake prerequisite is not satisfied by the package provided by Ubuntu; you will need to obtain a more current version.

NOTE: Using the Vitis recommended settings64.sh script to set up your environement can cause tool conflicts. Setup your environment in the following order for aietools and Vitis:

export PATH=$PATH:<Vitis_install_path>/Vitis/2023.2/aietools/bin:<Vitis_install_path>/Vitis/2023.2/bin

The cmake and python packages prerequisites can be satisfied by sourcing the utils/setup_python_packages.sh script. See step 2 of the build instructions. This script requires virtualenv.

clang/llvm 14+ are recommended to be built with the provided scripts. See step 3. of the build instructions.

In addition, the following optional packages may be useful:

LibXAIE is a backend target used to execute designs in hardware: https://github.com/Xilinx/embeddedsw/tree/master/XilinxProcessorIPLib/drivers/aiengine

Note that if you build one of the supported platforms like vck190_bare_prod, the generated sysroot already contains the LibXAIE drivers so you do not need to download the embeddedsw repository or define the LibXAIE_DIR cmake parameter.

Currently, the only supported target is the Xilinx VCK190 board, running Ubuntu-based Linux, however the tools are largely board and device independent and can be adapted to other environments.

Building on X86 for mlir-aie development

Clone the mlir-aie repository with its sub-modules:
```
 git clone --recurse-submodules https://github.com/Xilinx/mlir-aie.git
 cd mlir-aie
```
All subsequent steps should be run from inside the top-level directory of the mlir-aie repository cloned above.
Source utils/setup_python_packages.sh to setup the prerequisite python packages. This script creates and installs the python packages listed in utils/requirements.txt in a virtual python environment called ‘sandbox’, then it enters the sandbox:
```
 source utils/setup_python_packages.sh
```
If you need to exit the sandbox later, type deactivate. If you have a recent Linux distribution, you might not need this, as you are able to have all the required packages from the distribution.
Clone and compile LLVM, with the ability to target AArch64 as a cross-compiler, and with MLIR enabled: in addition, we make some common build optimizations to use a linker (lld or gold) other than ld (which tends to be quite slow on large link jobs) and to link against libLLVM.so and libClang.so. You may find that other options are also useful. Note that due to changing MLIR APIs, only a particular revision is expected to work.

To clone llvm, run utils/clone-llvm.sh (see utils/clone-llvm.sh for the correct llvm commit hash):
```
 ./utils/clone-llvm.sh
```
If you have already an LLVM repository, you can instead of cloning just make a new worktree from it by using:
```
 ./utils/clone-llvm.sh --llvm-worktree <directory-of-existing-LLVM-repository>
```
To build (compile and install) LLVM, run utils/build-llvm-local.sh in the directory where llvm has been cloned. See utils/build-llvm-local.sh for additional shell script arguments. (Note that build-llvm-local.sh and build-llvm.sh are a variation of the LLVM build script used for CI on GitHub and looking at the continuous integration recipe https://github.com/Xilinx/mlir-aie/blob/main/.github/workflows/buildAndTest.yml and output https://github.com/Xilinx/mlir-aie/actions/ might help in the case of compilation problem.)
```
 ./utils/build-llvm-local.sh
```
This will build LLVM in llvm/build and install the LLVM binaries under llvm/install.
Build the MLIR-AIE tools by calling utils/build-mlir-aie.sh for Versal or Ryzen AI with the path to the llvm/build directory. The Vitis environment will have to be set up for this to succeed. Note that when targetting the VCK5000 platform, please refer to the further instructions below titled Building on X86 targeting the VCK5000 as there are a few extra steps needed.
```
 source <Vitis Install Path>/settings64.sh
 ./utils/build-mlir-aie.sh <llvm dir>/<build dir>
```
This will create a build and install folder in the directory that you cloned MLIR AIE into.

The MLIR AIE tools will be able to generate binaries targetting a combination of AIEngine and ARM/x86 processors.
In order to run all the tools, it is necessary to add some paths into your environment. This can be done by sourcing the utils/env_setup.sh script with the paths to the install folders for mlir-aie and llvm.
```
 source utils/env_setup.sh <mlir-aie>/install <llvm dir>/install
```

Building on X86 targetting the VCK5000

In order to build and run on PCIe cards, you first have to build and install the aie-rt library. We chose to install the library in /opt/xaiengine but it is not required for the tools to be installed there. Just ensure that when building mlir-aie and mlir-air, that you point to the directory in which the aie-rt library was installed.

git clone https://github.com/stephenneuendorffer/aie-rt
cd aie-rt
git checkout phoenix_v2023.2
cd driver/src
make -f Makefile.Linux CFLAGS="-D__AIEAMDAIR__"
sudo cp -r ../include /opt/aiengine/
sudo cp libxaiengine.so* /opt/xaiengine/lib/
export LD_LIBRARY_PATH=/opt/xaiengine/lib:${LD_LIBRARY_PATH}

When targetting the VCK5000 Versal device, you must build and install our experimental ROCm runtime (ROCr) which allows us to communicate with the AIEs. The ROCm-air-platforms repository contains documentation on how to install ROCr. Run the following script to clone the ROCm-air-platforms repository:

./utils/clone-rocm-air-platforms.sh

Then, set ${ROCM_ROOT} to the ROCm install from the previous path. Then, run the following command to build the mlir-aie toolchain targetting the VCK5000.

./utils/build-mlir-aie-pcie.sh llvm/build/ build install /opt/xaienginev2 ${ROCM_ROOT}/lib/cmake/hsa-runtime64/ ${ROCM_ROOT}/lib/cmake/hsakmt/

The PCIe AIR runtime requires the use of the AIR PCIe kernel driver. The driver directory in the ROCm-air-platforms repository contains documentation on how to compile and load the AIR PCIe kernel driver.

Sysroot

Since the AIE tools are cross-compiling, in order to actually compile code, we need a ‘sysroot’ directory, containing an ARM rootfs. This rootfs must match what will be available in the runtime environment. Note that copying the rootfs is often insufficient, since many root file systems include absolute links. Absolute symbolic links can be converted to relative symbolic links using symlinks.

cd /
sudo symlinks -rc .

Following the platform build steps will create such a sysroot based on PetaLinux. Note that those instructions require Vitis 2021.2 – building a sysroot with Vitis 2023.2 will not currently succeed.