Dependencies

We run CI on Ubuntu 14.04, and these are the libraries requirements:

  • Python (>= 3.3)
  • numpy and development package (>= 1.10?)
  • Google glog development packages
  • verilator or ncverilog (TODO: how about other tools which also support VPI)

Build Nicotb

First you should build the C++ part of Nicotb.

python3 setup.py install --user

The examples are under the sim/ directory. To run the examples, we have prepared a script.

cd sim/
./run_all_verilator.sh
./run_all_ncverilog.sh

You can find more details in examples.

Build Nicotb on a very old Linux, without root

I use Arch as my primary OS, but many EDA tools are installed on a old Linux PC, and it's very hard to install the dependencies from official repos. In a very old system (e.g. gcc 4.1), it is necessary to compile everything from source code. The most difficult and time-consuming part is to compile gcc. Fortunately, there is a good article about installing gcc. In following parts I will give a brief introduction about installing the gcc dependencies, gcc, Python, glog, and numpy in order. First, you need these files, and any newer version should be fine:

If the links are invalidated, then try to find in these pages:

. Note that Github does not allow older protocol of OpenSSL/wget/git, so sometimes you might have to download them from other PCs. Also, you should notice that some very old PCs cannot unzip these files.

mkdir ${HOME}/install
mkdir /tmp/gcc
# binutils (Note: must configure under a clean directory)
mkdir build_binutils
cd build_binutils
/absolute_path/configure --prefix=${HOME}/install
make install
# GMP
./configure \
	--disable-shared --enable-static \
	--prefix=/tmp/gcc
make install
# MPFR
./configure \
	--disable-shared --enable-static \
	--prefix=/tmp/gcc --with-gmp=/tmp/gcc
make install
# MPC
./configure \
	--disable-shared --enable-static \
	--prefix=/tmp/gcc --with-gmp=/tmp/gcc --with-mpfr=/tmp/gcc
make install
# GCC (Note: must configure under a clean directory)
mkdir build_gcc
cd build_gcc
/absolute_path/configure \
	--prefix=$HOME/install/ --enable-languages=c,c++ \
	--enable-shared --enable-threads=posix -enable-__cxa_atexit \
	--with-gmp=/tmp/gcc --with-mpc=/tmp/gcc --with-mpfr=/tmp/gcc \
	--program-suffix=-8
# set path (setenv for csh)
# It's better to add this to your bashrc.
# FIXME: The 32 and 64 path might mix-up?
export PATH=$PATH:$HOME/install/bin
export CC=gcc-8
export LD_LIBRARY_PATH=$HOME/install/lib:$HOME/install/lib64

Python and Numpy

# Python
./configure --prefix=${HOME}/install --enable-optimizations --enable-shared
make
make install
# Numpy（If the following build fails.)
# export BLAS=None
# export LAPACK=None
# export ATLAS=None
# Numpy
~/install/bin/python3 setup.py build
~/install/bin/python3 setup.py build_ext
~/install/bin/python3 setup.py build_py
~/install/bin/python3 setup.py build_clib
~/install/bin/python3 setup.py build install

Glog

This is simpler to compile.

./autogen.sh --force
./configure --prefix=${HOME}/install
make install

Finally, you must change g++ to g++-8 (or whatever).

CC=g++-8 python3 setup.py install --user

Preload libraries

** [Updated] This is no longer necessary since we put this in the newer Makefile for ncverilog (sim/Makefile.ius). ** LD_PRELOAD must be used to prioritize new libraries, since EDA tools ship with their own python library, which commonly causes conflicts.

export LD_PRELOAD=${HOME}/install/lib64/libstdc++.so:${HOME}/install/lib/libpython3.6m.so:${HOME}/install/lib/libglog.so