Guide – How to: setup

Latest update: 2020-10-21

This guide illustrates prerequisites and environment setup for ESP. For a quicker and more portable setup, we also provide Docker images for ESP as described in the Docker section of this guide.

Table of Contents

Software packets
CAD tools
Environment variables
ESP repository
Software toolchain
Patching Ariane for Xcelium Simulator
Docker

Software packets

The ESP design flow has been developed and tested on the CentOS 7 Linux distribution, which is our preferred OS. Alternatively, you can use Ubuntu 18.04 or Red Hat Enterprise Linux 7.8.

Note: Cadence tools do not officially support Ubuntu 18.04. If you choose to work with Ubuntu, the accelerators unit test simulation within the *Cadence Stratus HLS* environment may not function properly.

In order to support embedded scripts and CAD tools for simulation and implementation, the following packages should be installed using the system packet manager.

Centos 7

  # Miscellaneous
  sudo yum install -y git octave octave-io jq

  # Python
  sudo yum install -y python python-pip python3 python3-pip python3-tkinter
  sudo pip3 install Pmw

  # Perl
  sudo yum install -y perl perl-Env perl-YAML perl-XML-Simple perl-ExtUtils-MakeMaker perl-Thread-Queue perl-Capture-Tiny

  # CAD tools and SW toolchains dependencies
  sudo yum install -y xterm
  sudo yum install -y csh ksh zsh tcl
  sudo yum install -y glibc-devel glibc-devel.i686
  sudo yum install -y glibc-static glibc-static.i686
  sudo yum install -y mesa-libGL.i686 mesa-libGLU.i686
  sudo yum install -y mesa-libGL mesa-libGLU
  sudo yum install -y mesa-dri-drivers mesa-dri-drivers.i686
  sudo yum install -y readline-devel readline-devel.i686
  sudo yum install -y libXp libXp.i686
  sudo yum install -y openmotif
  sudo yum install -y ncurses
  sudo yum install -y gdbm-devel gdbm-devel.i686
  sudo yum install -y libSM libSM.i686
  sudo yum install -y libXcursor libXcursor.i686
  sudo yum install -y libXft libXft.i686
  sudo yum install -y libXrandr libXrandr.i686
  sudo yum install -y libXScrnSaver libXScrnSaver.i686
  sudo yum install -y libmpc-devel libmpc-devel.i686
  sudo yum install -y nspr nspr.i686
  sudo yum install -y nspr-devel nspr-devel.i686
  sudo yum install -y tk-devel
  sudo yum install -y dtc bison flex bzip2 patch bc
  sudo ln -s /lib64/libtiff.so.5 /lib64/libtiff.so.3
  sudo ln -s /usr/lib64/libmpc.so.3 /usr/lib64/libmpc.so.2

  # For older GUIs (e.g. Stratus 18.2)
  sudo yum install -y libpng12 libpng12.i686

  # QT
  sudo yum install -y qtcreator
  sudo ln -s /usr/bin/qmake-qt5 /usr/bin/qmake

Ubuntu 18.04

  # Miscellaneous
  sudo apt install -y git octave octave-io jq

  # Python
  sudo apt install -y python python-pip python3 python3-pip python3-tk
  sudo pip3 install Pmw # sudo may not be needed

  # Perl
  sudo apt install -y perl libyaml-perl libxml-perl

  # CAD tools and SW toolchains dependencies
  sudo apt install -y xterm
  sudo apt install -y csh ksh zsh tcl
  sudo apt install -y build-essential
  sudo apt install -y libgl1-mesa-dev libglu1-mesa libgl1-mesa-dri
  sudo apt install -y libreadline-dev
  sudo apt install -y libxpm-dev
  sudo apt install -y libmotif-dev
  sudo apt install -y libncurses-dev
  sudo apt install -y libgdbm-dev
  sudo apt install -y libsm-dev
  sudo apt install -y libxcursor-dev
  sudo apt install -y libxft-dev
  sudo apt install -y libxrandr-dev
  sudo apt install -y libxss-dev
  sudo apt install -y libmpc-dev
  sudo apt install -y libnspr4
  sudo apt install -y libnspr4-dev
  sudo apt install -y tk-dev
  sudo apt install -y flex
  sudo apt install -y rename
  sudo apt install -y zlib1g:i386
  sudo apt install -y gcc-multilib
  sudo apt install -y device-tree-compiler
  sudo apt install -y bison

  # For older GUIs (e.g. Stratus 18.2)
  echo 'deb http://security.ubuntu.com/ubuntu xenial-security main' | sudo tee -a /etc/apt/sources.list
  sudo apt install -y libpng12-0

  # QT
  sudo apt install -y qtcreator

Red Hat Enterprise Linux 7.8

  # Miscellaneous
  sudo yum install -y git octave octave-io jq

  # Python
  sudo yum install -y python python-pip python3 python3-pip python34-tkinter
  sudo pip3 install Pmw

  # Perl
  sudo yum install -y perl perl-YAML perl-ExtUtils-MakeMaker
  sudo yum install -y http://mirror.centos.org/centos/7/os/x86_64/Packages/perl-XML-Simple-2.20-5.el7.noarch.rpm

  # CAD tools and SW toolchains dependencies
  sudo yum install -y xterm
  sudo yum install -y csh ksh zsh tcl
  sudo yum install -y glibc-devel glibc-devel.i686
  sudo yum install -y glibc-static glibc-static.i686
  sudo yum install -y mesa-libGL.i686 mesa-libGLU.i686
  sudo yum install -y mesa-libGL mesa-libGLU
  sudo yum install -y mesa-dri-drivers mesa-dri-drivers.i686
  sudo yum install -y readline-devel readline-devel.i686
  sudo yum install -y libXp libXp.i686
  sudo yum install -y openmotif
  sudo yum install -y ncurses
  sudo yum install -y gdbm-devel gdbm-devel.i686
  sudo yum install -y libSM libSM.i686
  sudo yum install -y libXcursor libXcursor.i686
  sudo yum install -y libXft libXft.i686
  sudo yum install -y libXrandr libXrandr.i686
  sudo yum install -y libXScrnSaver libXScrnSaver.i686
  sudo yum install -y libmpc-devel
  sudo yum install -y http://mirror.centos.org/centos/7/os/x86_64/Packages/libmpc-1.0.1-3.el7.i686.rpm
  sudo yum install -y nspr nspr.i686
  sudo yum install -y nspr-devel nspr-devel.i686
  sudo yum install -y tk-devel
  sudo yum install -y http://mirror.centos.org/centos/7/extras/x86_64/Packages/dtc-1.4.6-1.el7.x86_64.rpm
  sudo ln -s /lib64/libtiff.so.5 /lib64/libtiff.so.3
  sudo ln -s /usr/lib64/libmpc.so.3 /usr/lib64/libmpc.so.2
  sudo yum install -y bison

  # For older GUIs (e.g. Stratus 18.2)
  sudo yum install -y libpng12 libpng12.i686

  # QT
  sudo yum install -y qtcreator
  sudo ln -s /usr/bin/qmake-qt5 /usr/bin/qmake

CAD tools

ESP leverages a mix of open-source tools and scripts, as well as commercial tools. The following list specifies which commercial tools are currently supported and which tools are required to complete some steps of the ESP design methodology. Support for the tools marked as beta in the following list has not been fully tested or completed.

Required

RTL simulation: requires one of the following tools (64-bit version)
- Mentor Graphics ModelSim SE 2019.2: RTL system-level simulator
- Cadence Incisive 15.2: RTL system-level simulator (LEON3 only)
- Cadence Xcelium 18.03: RTL system-level simulator (apply patch below)
FPGA prototyping: requires the following tool
- Xilinx Vivado 2019.2: logic synthesis and implementation for FPGA

Note: As of May 18 2020 we updated the ESP synthesis and simulation scripts to work with Vivado 2019.2 and Modelsim 2019.2. This update improves simulation speed and enables us to add support for the newest Xilinx development boards. If you wish to continue working with the older versions of the tools (Vivado 2018.2 and Modelsim 10.6c or 10.7a), you may keep doing so and still get the latest updates from the ESP repository: after you pull the latest changes, simply revert the updates to the ESP scripts: git revert 44b1753.

Optional

Accelerator design: the following tools are optional
- Cadence Stratus HLS 18.20: high-level synthesis of accelerators from SystemC
- Xilinx Vivado HLS 2019.2: high-level synthesis of accelerators from C/C++ (beta)
- Mentor Catapult HLS 10.5b: high-level synthesis of accelerators from SystemC and C/C++
  - Xilinx Vivado 2018.3: requirement of Catapult HLS
Cache hierarchy: the default ESP configuration selects an RTL implementation of the cache hierarchy (no additional tool required). This supports both multi-core execution and coherence for accelerators. Optionally, ESP provides a SystemC implementation of the caches that can be synthesized with Cadence Stratus HLS. The SystemC version of the caches is particularly helpful to conduct architectural research on coherence, as the SystemC model is significantly easier to modify than the RTL implementation. To synthesize this version of the ESP caches, the following tool is required:
- Cadence Stratus HLS 18.15: high-level synthesis of caches from SystemC
Note: we will soon make available an instance of the RTL for the caches synthesized with Cadence Stratus HLS (pending approval for release).

Environment variables

Here are the environment variables required by the CAD tools:

# Cadence: Stratus HLS, Incisive, Xcelium
  # e.g. <stratus_path> = /opt/cadence/stratus182
  # e.g. <incisive_path> = /opt/cadence/incisive152
  # e.g. <xcelium_path> = /opt/cadence/xcelium18
  export LM_LICENSE_FILE=$LM_LICENSE_FILE:<cadence_license_path>
  export PATH=$PATH:<stratus_path>/bin:<incisive_path>/tools/cdsgcc/gcc/bin:<xcelium_path>/tools/cdsgcc/gcc/bin
  export CDS_AUTO_64BIT=all
  export HOST=$(hostname) # for Ubuntu only

# Xilinx: Vivado, Vivado HLS
  # e.g. <vivado_path> = /opt/xilinx/Vivado/2018.2
  export XILINXD_LICENSE_FILE=<xilinx_license_path>
  source <vivado_path>/settings64.sh

# Mentor: Catapult HLS, Modelsim
  # e.g. <modelsim_path> = /opt/mentor/modeltech
  # e.g. <catapult_path> = /opt/mentor/catapult
  export LM_LICENSE_FILE=$LM_LICENSE_FILE:<mentor_license_path>
  export PATH=$PATH:<modelsim_path>/bin
  export AMS_MODEL_TECH=<modelsim_path>
  export PATH=$PATH:<catapult_path>/Mgc_home/bin
  export SYSTEMC=<catapult_path>/Mgc_home/shared
  export SYSTEMC_HOME=<catapult_path>/Mgc_home/shared
  export MGC_HOME=<catapult_path>/Mgc_home
  export LIBDIR=-L<catapult_path>/Mgc_home/shared/lib $LIBDIR

# RISC-V
  # e.g. <riscv_path> = /opt/riscv
  export RISCV=<riscv_path>
  export PATH=$PATH:<riscv_path>/bin

# Leon3
  # e.g. <leon3_path> = /opt/leon
  export PATH=$PATH:<leon3_path>/bin
  export PATH=$PATH:<leon3_path>/mklinuximg
  export PATH=$PATH:<leon3_path>/sparc-elf/bin

ESP repository

ESP is available as an open-source GitHub repository. All original files in ESP are governed by the Apache 2.0 License, whereas third-party open-source software maintains its original license and includes the copyright notice from the authors.

Use the following command to obtain the ESP source code from GitHub. The option recursive will initialize and download the source code from submodules that are linked by the main ESP repository. Please be patient, as one of the submodules is Linux, which embeds a very large history.

# Clone ESP and all Git submodules at once
git clone --recursive https://github.com/sld-columbia/esp.git

# Alternatively
git clone https://github.com/sld-columbia/esp.git
git submodule update --init --recursive

Software toolchain

Every ESP SoC must include at least one processor core to execute the operating system and the target applications. ESP currently supports the Leon3 core from GRLIB, implementing the 32-bit SPARC V8 instruction-set architecture (ISA) and the Ariane core from ETH Zurich, implementing the 64-bit version of the RISC-V ISA.

Since the target ISA is not x86, a cross compiler is required to build software that can execute on the target processor. In addition, Linux requires a root filesystem that hosts all necessary initialization scripts, header files and dynamically-linked libraries. A partial overlay for the root filesystem is embedded into the ESP code base, however, binaries from busybox and libraries must be compiled and combined to the overlay. Failure to complete this step will prevent Linux from completing the boot process.

ESP provides scripts that allow users to build the toolchain automatically.

cd <esp>

# Leon3 toolchain
./utils/scripts/build_leon3_toolchain.sh

# RISC-V toolchain
./utils/scripts/build_riscv_toolchain.sh

The scripts go through several interactive steps to build the toolchain. If one step fails due to missing packages on the host system, users can restart the script and skip phases that completed successfully. For a default installation the user can select all the default answers of the interactive script.

Patching Ariane for Xcelium Simulator

The source code of Ariane includes UVM and SVA features that can only work with Modelsim. These features are disabled by the ESP Makefile by defining the variable VERILATOR during compilation. As a result, the instruction tracer is not instantiated, because it requires UVM support. The developers of Ariane replaced the instruction tracer with custom RTL code when simulating with Verilator. This code, unfortunately, triggers non suppressible errors in the Xcelium compiler.

While we do recommend to use Modelsim for simulation and keep all of the UVM and SVA features enabled, if you need to simulate using Xcelium, please apply this patch to the Ariane repository in third-party/ariane

The patch is simply commenting the code for instruction tracing with Verilator.

Finally, please note that Incisive cannot compile the source code of Ariane, but it will work with Leon3-based ESP instances.

Docker

The ESP Docker images are available on Dockerhub: hub.docker.com/repository/docker/davidegiri/esp.

The repository containing the Dockerfiles for generating the ESP Docker images is available on Github: https://github.com/sld-columbia/esp-docker.

The ESP Docker images are based on a CentOS 7 Docker image and they contain:

the installation of all the software packets required by ESP;
the installation of utilities like vim, emacs, tmux, socat and minicom, useful when working with ESP;
an environment variables setup script to be customized with the correct CAD tools paths and licenses;
the ESP repository and all its submodules;
the installation of the software toolchains for RISC-V and Leon3.

The are two types of images, which are identified by the full and small strings in the tag. The complete images are labeled with full and they are over 5GB in size in the compressed format. We also offer smaller images labeled with the small string that are slightly above 1GB in size, because they do not include the installation of the RISC-V and Leon3 software toolchains.

We have tested the ESP Docker images on Windows 10, MacOS 10.15, CentOS 7, Ubuntu 18.04 and RedHat 7.8. However, they should work on other OS distributions as well.

Install Docker

If you haven’t already, install Docker by following the instructions for your OS: https://docs.docker.com/engine/install/.

To start the Docker daemon on Linux systems you need to run:

sudo systemctl start docker

On Linux systems you will need to use sudo to run all docker commands (sudo docker) unless you add your user to the docker group. To add your user to the docker group run the following:

sudo usermod -aG docker ${USER}

Then you need to logout and log back in and after that you can verify you are part of the docker group by running groups.

Download the Docker image

Download the Docker image by running the following command (on Windows 10 you should run it in the PowerShell). Replace <tag> with centos7-full for the full image and with centos7-small for the small image.

docker pull davidegiri/esp:<tag>

You may need to add login credentials to be able to pull. In that case you can add your credentials by running the following. A prompt will ask for your password as well.

docker login -u <your-dockerhub-username>

Start the Docker container

Linux

On Linux you can start a Docker container in interactive mode as follows. The security-opt, network, env and volume arguments are needed to connect the container to the host machine display, so that it’s possible to open graphical interfaces from inside the container. Replace <tag> with centos7-full for the full image and with centos7-small for the small image.

docker run -it --security-opt label=type:container_runtime_t --network=host -e DISPLAY=$DISPLAY -v "$HOME/.Xauthority:/root/.Xauthority:rw" davidegiri/esp:<tag> /bin/bash

MacOS

On MacOS you need to make sure that the X server is running and is configured properly to be able to open graphical interfaces from inside the Docker container:

Install XQuartz
Open a terminal and launch the application with open -a xQuartz.
From the XQuartz drop-down menu, open Preferences.. and select the Security tab. Make sure that the check box “Allow connections from network clients” is enabled, as shown in the picture below.

From the same drop-down menu, quit XQuartz. The application must exit completely, then you should restart XQuartz.

From a new terminal window, enable X forwarding for your current IP address:

  ip=$(ifconfig en0 | grep inet | awk '$1=="inet" {print $2}')
  xhost + $ip

Finally, you can run the container as follows. Replace <tag> with centos7-full for the full image and with centos7-small for the small image.

docker run -it --network=host -e DISPLAY=$ip:0 -v /tmp/.X11-unix:/tmp/.X11-unix davidegiri/esp:<tag> /bin/bash

Windows 10

On Windows 10 you can start a Docker container in interactive mode from the Docker dashboard GUI by clicking first on the RUN button next to the Docker image name and then on the CLI button next to the running container.

To be able to launch graphical interfaces from inside the container on Windows 10 you need a couple of extra steps (adapted from these guides: guide1, guide2):

Install VcXsrv Windows X Server and after that launch XLaunch. In the XLaunch configuration steps use all the default configurations apart from selecting the “Disable access control” option, which is not selected by default.
Find the IP of you Windows machine by running ipconfig in the PowerShell. Look for the IPv4 Address entry.
Run export DISPLAY=<host-ip-address>:0.0 in the Docker container that you previously started with the CLI button.

Test X forwarding

From inside the container you can test the connection to the host display by running xeyes or xclock.

CAD tools with Docker

The CAD tools section has the complete list of CAD tools required by ESP. The Docker image doesn’t contain any of those CAD tools. Hence, next we describe a few options to work with CAD tools in the Docker container.

Install the CAD tools on your host machine from inside the container

Installing the CAD tools from inside the Docker container is convenient because all required packages are already installed. Moreover, this strategy is especially useful if your host machine doesn’t run CentOS 7, which is the OS running in the Docker container. However, instead of installing the CAD tools inside the container, it’s preferable to install them on the host machine, so that they can be used also by other containers or even natively on the host machine in some case. In addition, it’s better to avoid committing to the Docker image the large CAD tools installation folders.

Installing the CAD tools on the host machine from inside the container can be done by using volumes or bind mounts, which are the two main ways for persisting data generated by and/or used by Docker containers. Use bind mounts if you want to specify the absolute path on the host machine where the CAD tools should be. Use volumes if you want the data to live inside Docker’s storage directory, which is managed by Docker (this is preferable on Windows and MacOS).

You can declare multiple volumes and bind mounts when you launch the container with docker run by adding the following arguments.

Volume: -v <volume-name>:/cad-tools-path/inside/container. The volume called <volume-name> lives inside the Docker’s storage directory and it will be accessible from inside the container at the path /cad-tools-path/inside/container.
Bind mount: -v /cad-tools-path/on/host:/cad-tools-path/inside/container. The file or folder /cad-tools-path/on/host will be accessible from inside the container at the path /cad-tools-path/inside/container.

The idea is that from inside the container you can install the CAD tools at the path(s) of the volumes or bind mounts that you defined. Once the tools are installed, every container can access them if it receives the proper volume or bind mount arguments.

Use CAD tools already installed on your host machine

This option is useful if you already have some of the CAD tools installed on your host machine and if the host machine OS matches the one of the Docker container. In that case you can simply give access to the CAD tools with a bind mount when you start a container, by passing the -v argument as decribed in the previous section: -v /cad-tools-path/on/host:/cad-tools-path/inside/container.

Install the CAD tools inside the container

If you want a fully self-contained container, you can install the tools directly inside the container, without defining any volumes or bind mounts. The issue with this solution is that the container size will increase considerably, making it less portable.

Environment variables with Docker

The ESP Docker image provides two scripts for setting the required environment variables, as specified in the environment variables section.

Source the esp_env.sh script if you don’t need to use any CAD tools.

cd /home/espuser
source esp_env.sh

If you need some CAD tools you can use the esp_env_cad.sh instead. You should customize the script by inserting the paths of your tools and licences and by commenting out the environment variables for the tools that you don’t have. Then you should source the script.

cd /home/espuser
source esp_env_cad.sh

Useful Docker commands

Exit a container:

# run from inside the container
exit

List all local containers and their IDs:

docker ps -a

Stop a container:

docker stop <container-ID>

Start a container:

docker start <container-ID>

Attach to a running container:

docker attach <container-ID>

Copy data from host machine to a container:

docker cp <path-on-host> <container-ID>:/<path-inside-container>

Delete a container:

docker rm -f <container-ID>

List all local images:

docker images

Delete an image:

docker rmi <image-name>

Here is the complete Docker documentation.