nextpnr

nextpnr -- a portable FPGA place and route tool

nextpnr aims to be a vendor neutral, timing driven, FOSS FPGA place and route tool.

Currently nextpnr supports:

• Lattice iCE40 devices supported by Project IceStorm
• Lattice ECP5 devices supported by Project Trellis
• Lattice Nexus devices supported by Project Oxide
• Gowin LittleBee and Aurora V devices supported by Project Apicula
• NanoXplore NG-Ultra devices supported by Project Beyond
• Cologne Chip GateMate devices supported by Project Peppercorn
• (experimental) Cyclone V devices supported by Mistral
• (experimental) Lattice MachXO2 devices supported by Project Trellis
• (experimental) Xilinx 7-series devices supported by Project X-Ray
• (experimental) a "generic" back-end for user-defined architectures

A brief (academic) paper describing the Yosys+nextpnr flow can be found on arXiv.

Here is a screenshot of nextpnr for iCE40. Build instructions and getting started notes can be found below.

See also:

• F.A.Q.
• Architecture API

Prerequisites

The following packages need to be installed for building nextpnr, independent of the selected architecture:

• CMake 3.25 or later
• Modern C++17 compiler (clang-format required for development)
• Python 3.5 or later, including development libraries (python3-dev for Ubuntu)
  • Python 3.9 or later is required for nextpnr-himbaechel
  • on Windows make sure to install same version as supported by vcpkg
• Boost libraries (libboost-dev libboost-filesystem-dev libboost-thread-dev libboost-program-options-dev libboost-iostreams-dev libboost-dev or libboost-all-dev for Ubuntu)
• Eigen3 (libeigen3-dev for Ubuntu)
• Yosys is required to synthesise the demo design
• For building on Windows with MSVC, usage of vcpkg is advised for dependency installation.
  • For 32 bit builds: vcpkg install boost-filesystem boost-program-options boost-thread eigen3
  • For 64 bit builds: vcpkg install boost-filesystem:x64-windows boost-program-options:x64-windows boost-thread:x64-windows eigen3:x64-windows
  • For static builds, add -static to each of the package names. For example, change eigen3:x64-windows to eigen3:x64-windows-static
  • A copy of Python that matches the version in vcpkg (currently Python 3.6.4). You can download the Embeddable Zip File and extract it. You may need to extract python36.zip within the embeddable zip file to a new directory called "Lib".
• For building on macOS, brew utility is needed.
  • Install all needed packages brew install cmake python boost eigen

Getting started

First of all, run:

git submodule update --init --recursive

nextpnr-ice40

For iCE40 support, install Project IceStorm to /usr/local or another location, which should be passed as -DICESTORM_INSTALL_PREFIX=/usr to CMake. Then build and install nextpnr-ice40 using the following commands:

mkdir -p build && cd build
cmake .. -DARCH=ice40
make -j$(nproc)
sudo make install

On Windows, you may specify paths explicitly:

cmake . -B build -DARCH=ice40 -DICESTORM_INSTALL_PREFIX=C:/ProgramData/icestorm -DCMAKE_TOOLCHAIN_FILE=C:/vcpkg/scripts/buildsystems/vcpkg.cmake -DVCPKG_TARGET_TRIPLET=x64-windows -G "Visual Studio 15 2017 Win64" -DPython3_EXECUTABLE=C:/Python364/python.exe -DPython3_LIBRARY=C:/vcpkg/packages/python3_x64-windows/lib/python36.lib -DPython3_INCLUDE_DIR=C:/vcpkg/packages/python3_x64-windows/include/python3.6
cmake --build build --config Release

To build a static release, change the target triplet from x64-windows to x64-windows-static and add -DBUILD_STATIC=ON.

A simple example that runs on the iCEstick dev board can be found in ice40/examples/blinky/blinky.*. Usage example:

cd ice40/examples/blinky
yosys -p 'synth_ice40 -top blinky -json blinky.json' blinky.v               # synthesize into blinky.json
nextpnr-ice40 --hx1k --json blinky.json --pcf blinky.pcf --asc blinky.asc   # run place and route
icepack blinky.asc blinky.bin                                               # generate binary bitstream file
iceprog blinky.bin                                                          # upload design to iCEstick

Running nextpnr in GUI mode (see below for instructions on building nextpnr with GUI support):

nextpnr-ice40 --json blinky.json --pcf blinky.pcf --asc blinky.asc --gui

(Use the toolbar buttons or the Python command console to perform actions such as pack, place, route, and write output files.)

nextpnr-ecp5

For ECP5 support, install Project Trellis to /usr/local or another location, which should be passed as -DTRELLIS_INSTALL_PREFIX=/usr/local to CMake. Then build and install nextpnr-ecp5 using the following commands:

mkdir -p build && cd build
cmake .. -DARCH=ecp5 -DTRELLIS_INSTALL_PREFIX=/usr/local
make -j$(nproc)
sudo make install

• Examples of the ECP5 flow for a range of boards can be found in the Project Trellis Examples.

nextpnr-nexus

For Nexus support, install Project Oxide to $HOME/.cargo or another location, which should be passed as -DOXIDE_INSTALL_PREFIX=$HOME/.cargo to CMake. Then build and install nextpnr-nexus using the following commands:

mkdir -p build && cd build
cmake .. -DARCH=nexus -DOXIDE_INSTALL_PREFIX=$HOME/.cargo
make -j$(nproc)
sudo make install

• Examples of the Nexus flow for a range of boards can be found in the Project Oxide Examples.

nextpnr-mistral

For Cyclone V support, clone Mistral to $HOME/mistral or another location and pass this path as -DMISTRAL_ROOT=$HOME/mistral to CMake. Then build and install nextpnr-mistral using the following commands:

mkdir -p build && cd build
cmake .. -DARCH=mistral -DMISTRAL_ROOT=$HOME/mistral
make -j$(nproc)
sudo make install

Cyclone V support is currently experimental and has limited testing. The backend is undergoing active API refactoring, and its structure, build requirements, and integration points may change between versions.

nextpnr-machxo2

Install Project Trellis as for nextpnr-ecp5 above.

mkdir -p build && cd build
cmake .. -DARCH=machxo2 -DTRELLIS_INSTALL_PREFIX=/usr/local
make -j$(nproc)
sudo make install

nextpnr-generic

The generic target allows running placement and routing for arbitrary custom architectures.

mkdir -p build && cd build
cmake .. -DARCH=generic
make -j$(nproc)
sudo make install

An example of how to use the generic flow is in generic/examples. See also the Generic Architecture docs.

nextpnr-himbaechel

The himbaechel target allows running placement and routing for larger architectures that share a common structure.

gowin

For Gowin support, install Project Apicula

mkdir -p build && cd build
cmake .. -DARCH="himbaechel" -DHIMBAECHEL_UARCH="gowin"
make -j$(nproc)
sudo make install

• Examples of the Gowin flow for a range of boards can be found in the Project Apicula Examples.

ng-ultra

For NanoXplore NG-Ultra support, clone Project Beyond DB repo

mkdir -p build && cd build
cmake .. -DARCH="himbaechel" -DHIMBAECHEL_UARCH="ng-ultra" -DHIMBAECHEL_PRJBEYOND_DB=/path/to/prjbeyond-db -DHIMBAECHEL_NGULTRA_DEVICES=ng-ultra
make -j$(nproc)
sudo make install

Please note that binary bitstream creation requires Impulse tool from NanoXplore.

gatemate

For Cologne Chip GateMate support, clone Project Peppercorn

mkdir -p build && cd build
cmake .. -DARCH="himbaechel" -DHIMBAECHEL_UARCH="gatemate" -DHIMBAECHEL_PEPPERCORN_PATH=/path/to/prjpeppercorn
make -j$(nproc)
sudo make install

GUI

The nextpnr GUI is not built by default, to reduce the number of dependencies for a standard headless build. To enable it, add -DBUILD_GUI=ON to the CMake command line and ensure that Qt6 and OpenGL are available:

For Qt6:

• On Ubuntu 22.04 LTS or later, install qt6-base-dev
• For MSVC vcpkg, install qt-base (32-bit) or qt-base:x64-windows (64-bit)
• For Homebrew, install qt6 and add qt6 in path: echo 'export PATH="/usr/local/opt/qt/bin:$PATH"' >> ~/.bash_profile ` - this change is effective in next terminal session, so please re-open terminal window before building

Multiple architectures

To build nextpnr for multiple architectures at once, a semicolon-separated list can be used with -DARCH.

mkdir -p build && cd build
cmake .. -DARCH="ice40;ecp5"
make -j$(nproc)
sudo make install

To build every available stable architecture, use -DARCH=all. To include experimental arches (currently nexus), use -DARCH=all+alpha.

Per-microarchitecture Himbächel

To build a single nextpnr-himbachel executable for each of the supported microarchitectures, use -DHIMBAECHEL_SPLIT.

mkdir -p build && cd build
cmake .. -DARCH="himbaechel" -DHIMBAECHEL_UARCH="gowin;ng-ultra"
make -j$(nproc)
sudo make install

In such a build, instead of a single nextpnr-himbaechel binary, two binaries nextpnr-himbaechel-gowin and nextpnr-himbaechel-ng-ultra are built. Although they are installed together, each microarchitecture is completely independent of the other, and only needs its corresponding .../share/himbaechel/<microarchitecture>/ chip database directory to run. Split build reduces the size of individual distributed artifacts (although the total size increases), and allows co-installation of artifacts of different versions.

Cross-compilation

Apart from chip databases, nextpnr requires the bba tool to be compiled for the build system. This tool can be compiled as a separate project:

cd bba
cmake .
make

This will create a bba-export.cmake file. Provide the path to this file when cross-building nextpnr by using -DBBA_IMPORT=/path/to/bba-export.cmake.

Additional notes for building nextpnr

The following runs a debug build of the iCE40 architecture without GUI, without Python support, without the HeAP analytic placer and only HX1K support:

mkdir -p build && cd build
cmake .. -DARCH=ice40 -DCMAKE_BUILD_TYPE=Debug -DBUILD_PYTHON=OFF -DICE40_DEVICES=1k
make -j$(nproc)

To make static build release for iCE40 architecture use the following:

mkdir -p build && cd build
cmake .. -DARCH=ice40 -DBUILD_PYTHON=OFF -DSTATIC_BUILD=ON
make -j$(nproc)

The HeAP placer's solver can optionally use OpenMP for a speedup on very large designs. Enable this by passing -DUSE_OPENMP=yes to cmake (compiler support may vary).

You can change the location where nextpnr will be installed (this will usually default to /usr/local) by using -DCMAKE_INSTALL_PREFIX=/install/prefix.

Notes for developers

• All code is formatted using clang-format according to the style rules in .clang-format (LLVM based with increased indent widths and brac