onsdag 2 november 2016

IP-XACT: The good, the bad and the outright madness

I've found myself in a strange position recently defending IP-XACT on a number of occasions. Most recently there were some heated discussions with lots of hand gestures at orconf. The heat and hand gestures however might have been due to the fact that we were in Italy. Still, there were discussions.

So let me get this straight once and for all by quoting one of the great thinkers of the 21st century.
IP-XACT is in most parts horrible, but it's the best chance we have for a vendor-neutral standard.

O. Kindgren, 2016

So what is this mostly horrible but still kind of useful standard? It's an IEEE standard developed by Accellera aimed to create an interchange format for IP cores. You can find the standard here, but the Wikipedia article does a better job at capturing the purpose and extents of the standard in a few sentences.

IP-XACT is an XML format that defines and describes electronic components and their designs. IP-XACT was created by the SPIRIT Consortium as a standard to enable automated configuration and integration through tools.

The goals of the standard are

  • to ensure delivery of compatible component descriptions from multiple component vendors,
  • to enable exchanging complex component libraries between electronic design automation (EDA) tools for SoC design (design environments),
  • to describe configurable components using metadata, and
  • to enable the provision of EDA vendor-neutral scripts for component creation and configuration (generators, configurators).

These are all noble goals, so why the criticism? First of all, it's an XML-based standard. This alone will turn away many people, but frankly, don't tell me that json is any better. For some reason though, json has managed to attract a lot less negativity while still being affected by some severe shortcomings and compatibility problems. My personal pet peeve is that there isn't any way to add comments to a JSON file, which is very annoying when you want to temporarily remove a part of the file, or explain something. For those of you who don't know json that well, it looks a bit like LISP code with wrinkles.

With that out of the way, another problem is tool compatibility. As with most standards in the FPGA world, the two giants, Xilinx and Intel (still haven't gotten used to saying Intel instead of Altera) does a superb job of never backing the same standard, to ensure an absolute minimum of interoperability. Prior examples being Xilinx endorsing the FMC standard for daughter boards with Altera going for HSMC. Xilinx used to have their UCF constraint format which didn't look like any other constraint format, but have now moved to XDC, which is almost but not quite compatible with the more common SDC format. In the case of IP-XACT, Xilinx supports it, but not Altera

Actually, supports is a bit too strong a word in this case, as Vivado only works with up to the 2009 standard, not IP-XACT 2014. ISE of course doesn't handle IP-XACT at all, but I guess that's ok since it has been obsoleted since some years. Actually, when I say that Vivado supports up to the 2009 standard I'm still being a bit too kind, and this brings me to the first issue with IP-XACT

Vendor extension

Vendor extensions means that tools are free to define their own non-standard XML tags in the IP-XACT files. This allows them to implement additions where they feel the standard is lacking. This can be a good thing, but the drawback is that it opens up for a lot of non-standard additions, where every vendor implements the same functionality in different ways. I've seen examples where most of the file is just encapsulated in vendorExtensions. Incidentally, this is a bit similar to what happened to the other vendor-neutral EDA standard, EDIF, where no two files from different vendors were really compatible with each other. Wikipedia has a fascinating article about EDIF, which should serve as a cautious reminder to the future of IP-XACT

Tool support

What about IP-XACT support in other tools then? Unfortunately most commercial IP-XACT tools are out of reach for someone not deeply entrenched in ASIC development, so my other examples will be open source tools. The likely most used open source tool for working with IP-XACT is Kactus2. Kactus2 is a graphical application where you can fill in the various fields of an IP-XACT component file in a more or less user-friendly manner. It can parse Verilog/VHDL code to help you import existing HDL code into an IP-XACT model. Once you have built up a library of files, you can hook them up together in a block diagram view and export the top-levels as verilog or VHDL together with documentation, C header files for registers and a few other things. Kactus2 up to version 2.8 only handled the older IP-XACT 1.5 version, while Kactus2 3.0 (yes, it's Kactus2 3.0, not Kactus3. I asked the developers) and onwards only handle IP-XACT 2014. The attentive reader will notice that this means that Kactus2 and Vivado are no longer able to work with the same files.

For my own humble uses of IP-XACT, I decided to write a small python library called ipyxact to easier interface the parts of IP-XACT that I care about. This means it only support a limited subset of the standard. On the upside, it reads and writes 1.5, 1.6, 2009 and 2014 files, and there are some ideas how to make it cover the full standard. So, as you can see, there are still tools to be written for IP-XACT. For the IP-XACT files themselves Accellera provides XSL scripts which can be used to automatically upgrade files to a newer version. (Hey JSON, you got something like XSLT? No? Didn't think so either)

What about the good parts then? There must be something good to say about IP-XACT too. Fortunately, there are some good parts, and these are enough for me to endorse it, even with its numerous other faults.

VLNV tags

Each named object, such as a component or bus in IP-XACT have an identifier called VLNV. VLNV stands for Vendor, Library, Name, Version and is used to uniquely describe an object. These are often written as a colon-separated string, such as librecores.org:peripherals:uart16550:1.5 Even though it sometimes can be a bit awkward to come up with a good naming scheme for Vendor and Library, it will work out fine for most use cases, and the four fields are enough to avoid any namespace conflicts. Well done IP-XACT!

Block diagrams

One part of the IP-XACT standard allows you to define the interfaces of your IP cores. This can be single wires, but more common is to group these together into a bus, such as Wishbone or SPI, which can be connected between components. There are several graphical editors to help with this too. Kactus2 is the main example, and is also Open Source. Vivado comes with a built-in block diagram tool which is based on IP-XACT too. I love this part of the standard. I always draw my designs on paper before getting to work, so why not be able to use this for both documentation and implementation? I know that many of the new HDL such as Chisel and Migen was born partly out of frustration of connecting components in Verilog and VHDL. These languages makes it a bit easier, but it's still programming code, and in my opinion this is a graphical problem best solved with a graphical tool. Do your business logic in your language of choice, but leave the interconnections to a drag-and-drop application. Kactus2 still has some ways to go to make this really comfortable, but it works fine for many cases and I have done a few Proof of Concept designs to get better acquainted with the process. The drawback here is that it's not that flexible. Even though interfaces can be parametized with regards to widths and types, I haven't found a way to create components with a variable number of buses, which is handy for things like bus interconnects where you want to quickly hook up or remove master and slaves. My solution for that is to use my trusty ipyxact library to generate application-specific wrappers for these components. It works, but requires an extra step. There is unfortunately another problem with the way bus interfaces work in IP-XACT. Each bus type is identified by a VLNV tag. This means that a one_vendor:buslibrary:busname:version is not compatible to other_vendor:buslibrary:busname:version. And you know what? IP vendors are happy to put their own names on things, but not their competitors, so this will happen a lot.

In all fairness, this problem comes down to management of the shared resources and Accellera are providing bus definitions for a few common buses, like SPI and the various MII variants to connect ethernet Phys and MACs, but it would be great if other owners of buses would do the same as well. For LibreCores we are planning to host open standards such as Wishbone and we're happy to help if other want to host their standard bus defintions here as well

 File sets

File sets lists the files of an IP. Each file contain information such as file type, if it's an include file, which library it belongs to and other metadata. The filesets themselves have additional properties such as describing its purpose. This can for example be used to expose different filesets for simulation or synthesis. The filesets are actually the only part of IP-XACT that is currently supported in FuseSoC. Instead of listing the files directly in the FuseSoC .core files, a .core file can point to an IP-XACT component file and FuseSoC will automatically pick up the files from there. And even if you prefer to use only .core files, the fileset sections in the .core files are modeled after the structure in IP-XACT. Sounds good? Mostly, yes, but there are some drawbacks. First of all, the files can contain a lot more metadata than what is listed above, but I will return to that later. Instead, I will focus on the file types. The IP-XACT standard decided to enumerate all valid file types in the standard. Here's the complete list for the latest (2014) revision.

unknown, cSource, cppSource, asmSource, vhdlSource, vhdlSource-87, vhdlSource-93, verilogSource, verilogSource-95, verilogSource-2001, swObject, swObjectLibrary, vhdlBinaryLibrary, verilogBinaryLibrary, unelaboratedHdl, executableHdl, systemVerilogSource, systemVerilogSource-3.0, systemVerilogSource-3.1, systemCSource, systemCSource-2.0, systemCSource-2.0.1, systemCSource-2.1, systemCSource-2.2, veraSource, eSource, perlSource, tclSource, OVASource, SVASource, pslSource, systemVerilogSource-3.1a, SDC, vhdlAmsSource, verilogAmsSource, systemCAmsSource, libertySource, user

Quite some list, right? Both SystemVerilog 3.0 and SystemVerilog 3.1 is there. There are also something called executableHdl which I have no idea what that is. Looks like everyting is covered. Well, what about verilog 2005? Not a major language revision, but they added $clog2 in that version which is quite handy. Or VHDL 2008? That one is really big. Turns out that they forgot about those. There are also a gazillion new HDL languages such as Chisel, MyHDL, parC as well as all the vendor-specific IP formats, which have no entries in this list. For FuseSoC, I have added a few new file types (currently QIP, verilogSource-2005, vhdlSource-2008, xci and xdc) which have been required, which already makes it a superset of IP-XACT. Enumerating types like this is just a bad idea, and we will see even better examples of this later on in this article

 Register maps

Register maps is one of those areas where you want to make sure that the documentation matches the implementation. For this reason it makes sense to have a single source format from which you can create nice looking tables, C header files, slave interfaces for various CPU buses and perhaps test vectors. Most projects I have been working on have had different strategies for this. Many just do an implementation for a single bus interface and use a separate document where the designer has to do the tedious and error-prone work of writing down the same information and remember to tell the software developers when and how to update their C header files. This is of course very often not kept in sync. One magnificent example is a company, which shall be left unnamed, which stored all their registers in a spreadsheet file which was put under version control as a binary file and parsed this file with some magic scripts to extract the information. Nothing has ever gone wrong when parsing spreadsheets, right? And comparing two versions of a spreadsheet is always a joy.

IP-XACT, while being a XML-based format which some smarty-pants will always claim is actually a binary format, makes it a lot easier to keep the different target formats in sync. The register maps in IP-XACT capture enough information to describe most common access types, such as read-only, clear-on-write, single-shot and different bits slices of a register can be fully independent from each other. As usual with IP-XACT it's also possible to set some less common properties, such as marking a register as readWriteOnce. Might be useful if you're implementing a CD-ROM perhaps, I don't know.

Let's discuss the drawbacks then. Well, it turns out that the IP-XACT standard committee themselves didn't think register maps were good enough, so they came up with yet another format called SystemRDL, which they recommend using to *drum roll* generate IP-XACT register maps! So despite having added all these really complex features to express all details about registers they recommend moving to another, even more expressive format, to create the register maps. Maybe they should consider, perhaps, dropping register maps all together from the standard if that wasn't good enough? It's not like IP-XACT would be too small without the register parts. Instead, it seems like they are adding more features to registers for every standard revision. And this thing about adding features to the standard brings me to the really scary parts. Let me present:

IP-XACT as a build system

Every file listed in a fileset has some metadata, as I mentioned above. One of these metadatums (not at all sure that's the correct word for a single piece of metadata) is the buildCommand. It is a whole XML tree structure used to describe how to build said file. It comes with a ton of complex options used to describe which flags to use for building and then linking the file. Flags can be taken from a default build command, overridden by the build command specified for the file or taken from its sibling files, all according to some vaguely defined rules. Having a single buildcommand is of course impossible to combine with tool-independence since there is no standard command to deal with HDL files. Each tool has its own syntax and structures. Embedding this in the IP-XACT files is just wrong! Ok, let's assume that this isn't for HDL files then. Since building and linking is explictly mentioned in the tag names, we can guess that this is aimed at C code. Problem is, this can work totally different for other languages. Even for C files, this is a problem, since it's not really possible to switch between gcc, llvm or other compilers. There are also no way to do conditional compilations based on which flags are being set at a higher level. Please, please, please IP-XACT. Leave this task to experienced software developers. There is already an endless list of build systems for software (autotools, cmake, meson, scons, waf...) with decades of work to get them to the state where they are. Many of them still have some really bad limitations and the last thing the world needs is another half-assed build system designed by RTL designers who's only exposure to software structure was taking a class in Java back when Spice Girls topped the charts. This is unfortunately all too symptomatic of the EDA standards, and no amount of craziness in the IP-XACT standard manages to come close to the proposal to add inline Perl in SystemVerilog. Sometimes I think we're just doomed.

Speaking of software, this leads me to another favorite part of IP-XACT

IP-XACT to describe software API

Digging deep into the dark corners of IP-XACT, you will find that each fileset can be matched to a "function". The documentation of this particular tag says "Generator information if this file set describes a function. For example, this file set may describe diagnostics for which the DE can generate a diagnostics driver". Ok, not all that clear what this means, but let's move on and see what's beneath this function tag. There is a tag called entryPoint, which apparently is the Optional name for the function. There's also a fileRef, which is used to describe A reference to the file that contains the entry point function. Likely, those two are related somehow. There are a couple other tags as well, but the real gem is returnType, with the helpful documentation saying Function return type. Possible values are void and int. Yes, as we all know, int and void are the only two types a function will ever return. Let me illustrate this by a short conversation

- Hi Mr ipxact_square_root_function!
- Hi!
- Do you think you could give me the square root of three
- Certainly! It's my pleasure. The answer is.... let me think... one!
- One?
- Yes, one.
- Are you.. I mean, shouldn't that be more like... is that really...?
- Look, you should be happy that I gave you something. Or would you prefer I give you void instead?
- No no no! One is fine, one is fine. My bad
- I thought so
- (grumble grumble, I'm not asking that one again)
- What did you say?
- Oh! No, nothing. One is fine. Moving along

So my point here is that large parts of this standard manages to be extremely complex, while at the same time having strange limitations that makes them mostly useless. Couldn't we just fix these parts of the standard then? Well, this leads me to

IP-XACT, the free standard

To take part of the decision making progress, all we need to do is becoming members of Accellera, which can be done by registering on their home page. Oh, and you need to pay $25000 too. As much as I'd love to take part in steering this standard away from it's very uncertain path, I have a hard time imagining myself going to my boss and say "Hey, I need $25000! Why? Someone is WRONG on the internet. I need to fix this". Instead we have participants from the usual suspects in the EDA industry, and ideas from academia and the general software industry are nowhere to be found

The verdict

Having worked as a contractor on FPGA/ASIC designs for many years as well as digging into dozens of different Open Source Silicon projects, often with the purpose of packaging them for FuseSoC, I have come across many many, ways to do the exactly same thing. I'm dead tired of having to learn yet another register map generator or IP core description format only to find that I need to write another set of tools to automate the process. It's incredibly time-consuming, and this is why I cheer for IP-XACT, despite its numerous faults. It's a standard way of dsecribing IP cores. Given some time and effort, I think that there will grow a healthy ecosystem around the standard, and I think in fact that it's absolutely required to have everything from full-blown environments to small utilities for dealing with IP-XACT. Otherwise, no one will see the point of supporting this multiheaded beast of XML.

I also hope that the EDA industry will come to its senses and see that much innovation and knowledge never reaches the standards as long as the organizations charge these sums of money to participate. Also, if the standard continues to branch out with more of these half-baked ideas, it will die under its own weight. If that happens I hope the good ideas will be forked off to a new standard, hopefully driven by a more inclusive organization. But I hope that won't happen at all, because it would likely result in ten similar standards fighting for attention, and as I mentioned in the outset, this is the best chance we have for a vendor-neutral standard. Let's go with the flow and see where it takes us. We can always jump ship later.

måndag 8 augusti 2016

FuseSoC 1.5

Finally! FuseSoC 1.5 is now released. I was just about to release it a few weeks ago when I discovered two quite serious bugs. Then I was just about to release it when I had written a lengthy blog post describing some of the new backends. At that point I realized that it would be better to write those bits as part of the documentation. Yes, you read it right. Documentation. This release has 45% more documentation than previous releases. A massive increase from 245 lines to 368 lines. Oh well. One day it will all be properly documented...I hope. At least the new backends are documented and I will try to give the old ones the same treatment. Promise!

Now for the good stuff...

 New backends

This version adds support for one new simulator, one new FPGA flow and three EDA vendor IP package formats.

The new FPGA flow is for the awesome project IceStorm, which provides a complete open source flow for building FPGA images for Lattice iCE40 devices. The tools for this consists of yosys to handle synthesis, arachne-pnr for place & route and icepack for creating the binary file. Both yosys and arachne-pnr can handle other vendor's devices as well, and I hope to use these tools in other flows as well in the future.

There is support for an open source VHDL simulator called GHDL, which is most likely the open source EDA tool with the best VHDL support.

The new IP package formats being handled are Altera QIP files, Xilinx Coregen files (for ISE) and Xilinx xci files (for Vivado). The last two deserves a few extra words since they use the provider mechanism as a plugin system in a clever way. Providers, in the FuseSoC world, are modules that are normally responsible for downloading a core and put it into the cache. There are providers already for getting cores from git and svn repos as well as from simple URLs. The providers are each responsible for their own options but typically look something like this

name = github
user = olofk
repo = wb_bfm
version = v1.0

One FuseSoC user and contributor however realized that the provider doesn't really have to download anything, and can instead be used to do transformations on source code instead. He implemented this to generate HDL code from vendor IP core descriptions and added the new coregen and logicore providers. They each look like this

name = coregen
script_file = <coregen ip description>.xco
project_file = <coregen project file>.cgp

name = logicore
script_file = <generator script>.tcl
project_file = <vivado ip description>.xci

When a core with one of these providers is requested, the Xilinx IP generation tools for ISE or Vivado are invoked to generate the HDL code, which is then placed in the cache. For the coregen-based cores, the generated HDL files are listed in the .core file as would have been done for any core.

For the Vivado-based logicore IP however there is an extra trick. As Vivado generates an IP-XACT component description for its generated IP cores, we can simply point to that one, and automatically let FuseSoC get all the HDL files for us automatically.

I have been planning for several years to do some kind of plugin mechanism to run custom preprocessing commands on source files, and using the provider mechanism for this solves the problem in most cases. Examples of future providers could be to turn MyHDL, Migen or Chisel code into verilog, or build a verilog top-level from an IP-XACT design file. As this was not originally my idea, I was naturally extremely sceptic of the whole idea. It does have some limitations, but for most cases it gets the job done and is a clever solution for the problem.

Continious integration

Another great addition brought in by a new FuseSoC user is the continious integration testing with Travis and AppVeyor. Every time something is commited to the FuseSoC repository, Travis and Appveyor are triggered to run some basic tests. As an interesting note, it took less than a week after adding the CI support until this found a bug caused by a minor difference between python 2 and python 3. That bug would probably had gone unnoticed for a long time if not for this, so it's worth its money already. Especially so, since both services are free to use.

Windows compatibility

The same user who brought in the CI testing also done some changes to the build system to increase the windows compatibility. It still doesn't run properly on Windows, but it can now at least install, list the cores properly and run a few simulations

Improved parameter handling

There had been several improvements to the handling of command-line parameters for compile-time or run-time configurations. It's now possible to specify a default value for parameters. This can be useful to override default settings in the upstream code, for example to specify a different path to a memory initizalization file than what the upstream project uses. There is also some visual improvements to the output of fusesoc {sim,build} <core> --help to make it clearer what kind of parameter is being changed. Several bugs has been fixed in this area as well, and it should now be possible to use top-level parameters with all tools.

The old [plusarg] sections that were deprecated in favor of the [parameter *] sections are no longer parsed as I have decided that users have had enough time to migrate to the parameter sections. The warning still remains to remind everyone that it's time to move on and the article about FuseSoC 1.4 has more info on what the [parameter] sections should look like.

Improved ModelSim backend

The Modelsim backend has been rewritten, both to support loading user-specified TCL files and to make it easier to rebuild a design from the exported sources. And believe it or not, there is even documentation for the modelsim backend now!

Build system changes

Finally, I also decided to remove the old autotools scripts for installation. For some time it has been possible to install FuseSoC with either python's setuptools (e.g. pip install fusesoc) or autotools (e.g. ./configure && make && make install). With the improvements in the setuptools based system, it makes little sense to keep autotools. I don't want to keep two tools updated, so from now on, FuseSoC will only use setuptools for installations. Farewell autotools. It's not you, it's me


...as always, tons of bugs were killed during the making of this release. A few notable ones are that VHDL and SystemVerilog are now supported properly in ISIM, IP-XACT 2009 and 2014 files are supported, RTL library affiliation is read from IP-XACT logicalName tags and FuseSoC doesn't crash anymore if it encounters an unknown simulator in the simulators list

I'm really excited to start working on FuseSoC 1.6 now. There are already a bunch of new things ready to be commited, that I have been holding off until FuseSoC 1.5 was released. Make sure to check the git repo for daily progress if you want to get all the new things before the next version is released.

Thanks for listening!

torsdag 2 juni 2016

FuseSoC and your custom workflow

Last week a colleague made a horrible mistake. He casually asked me to tell him a bit about FuseSoC. This made me very happy and after an hour he probably regretted asking (sorry Johan).

Anyway, it turned out that he wanted to know if FuseSoC was the right tool for him. As most other RTL developers, he had his own set of custom scripts to launch simulations - in this case it was Makefiles. After discussing a bit back and forth we both realized that switching to a complete FuseSoC-based workflow wouldn't really be all that great for him, as he would lose some of the tight tool-integration with his existing scripts. Talking a bit more however revealed that he had no great solution for bringing in all the source files that his makefiles should digest, and as I have written about before, FuseSoC is not only meant as a end-to-end solution for running simulations or making FPGA images. It's also meant to be used as a library so that it can be hooked up to other tools. And what my colleague needed in this case was mainly a list of all RTL source files, so I said that I could make such a tool based on FuseSoC in fifteen minutes. It turned that I was wrong. Starting out with the script I did for VUnit integration, it took less than ten minutes to get it working. Without further ado, I therefore present the FuseSoC file dumper script as an example on how to use just the parts of FuseSoC that brings in dependencies, sorts them in order and lists all files, as long as your cores use the FuseSoC .core format to describe them. Use it, adapt it and share it.

# FuseSoC source file dumper. fusesoc_file_dump_demo
# Copyright (C) 2016  Olof Kindgren <olof.kindgren@gmail.com>
# Permission to use, copy, modify, and/or distribute this software for any
# purpose with or without fee is hereby granted, provided that the above
# copyright notice and this permission notice appear in all copies.

from collections import OrderedDict
import os.path
import sys

from fusesoc.config import Config
from fusesoc.coremanager import CoreManager, DependencyError

if len(sys.argv) != 2:
    print("Usage: {} <top level core>".format(sys.argv[0]))
top_core = sys.argv[1]

#Create singleton instances for core manager and configuration handler
#Configuration manager is not needed in this example
cm = CoreManager()
config = Config()

#Add core libraries that were picked up from fusesoc.conf by the config handler
except (RuntimeError, IOError) as e:
    pr_warn("Failed to register cores root '{}'".format(str(e)))

#Get the sorted list of dependencies starting from the top-level core
    cores = cm.get_depends(top_core)
except DependencyError as e:
    print("'{}' or any of its dependencies requires '{}', but this core was not found".format(top_core, e.value))

#Iterate over cores, filesets and files and add all relevant sources files to a list
incdirs = set()
libraries = []
src_files = []

#'usage' is a list of tags to look for in the filesets. Only look at filesets where any of these tags are present
usage = ['sim']
for core_name in cores:
    core = cm.get_core(core_name)
    basepath = core.files_root
    for fs in core.file_sets:
        if (set(fs.usage) & set(usage)) and ((core_name == top_core) or not fs.private):
            for file in fs.file:
                if file.is_include_file:
                    incdirs.add(os.path.join(basepath, os.path.dirname(file.name)))
                    src_files.append("{},{},{}".format(os.path.join(basepath, file.name),

print("==Include directories==")
print("==Source files==")

Granted that you have a fusesoc.conf file in your current directory or ~/.config/fusesoc and the FuseSoC standard core library installed, both of which you get if you install FuseSoC, you can now run python fusesoc_file_dump_demo.py de0_nano to get a list of all source files, include directories and libraries needed to build the de0_nano system.

onsdag 3 februari 2016

FuseSoC 1.4

When I consider the magnitude of the subject which I am to bring before my readers-a subject, in which the interests, not of this country, nor of Europe alone, but of the whole world, and of posterity, are involved: and when I think, at the same time, on the weakness of the advocate who has undertaken this great cause-when these reflections press upon my mind, it is impossible for me not to feel both terrified and concerned at my own inadequacy to such a task. But when I reflect, however, on the encouragement which I have had, through the whole course of a long and laborious examination of this question, and how much candour I have experienced, and how conviction has increased within my own mind, in proportion as I have advanced in my labours;-when I reflect, especially, that however averse any gentleman may now be, yet we shall all be of one opinion in the end;-when I turn myself to these thoughts, I take courage-I determine to forget all my other fears, and I march forward with a firmer step in the full assurance that my cause will bear me out, and that I shall be able to justify my decision to release FuseSoC 1.4.

William Wilberforce, a man who apparently had a thing for extremely long sentences, might have said something a bit similar about the end of slavery, but that's not the topic for today. Today's topic is the new FuseSoC release. We now live in a world of git snapshots (and less slavery than 1789), and I don't expect an increase in the number after the decimal point in the configure script matter to most people. Still, I think it's nice to do a stable release once in a while. It helps packagers (where are you by the way? I'm still waiting for someone to make a FuseSoC rpm or deb), and more importantly, I can take this as an opportunity to break things. This will be the first version where a part of CAPI1 gets deprecated. There are some other noteworthy changes since FuseSoC 1.3, and although I have summarized the changes in the NEWS file, and I recently wrote a post about the new IP-Xact support there are other changes that could be worth explaining in further detail. I do realize that this would not be needed if someone had written proper documentation (where are you by the way? I'm still waiting for someone to write the documentation for me), but no one has, so this will have to do for now.

Prominent features


 File sets

Up until now, source code has been mainly verilog, with some support for C(++) for Verilator testbenches and VPI modules, together with a very limited support for VHDL in the Quartus backend flow. This part has been completely overhauled in this release, where filesets are introduced. Filesets is modeled on the IP-Xact filesets, and these will have a stronger connection in the future, as IP-Xact support is also introduced in this version. The verilog section will be deprecated at some point, so I suggest moving to filesets already. Here's a quick example on how to make the switch. The verilog section of the de0 nano system...

src_files =

tb_private_src_files =
include_files =

tb_include_files =


[fileset rtl_files]
files =
file_type = verilogSource
usage = sim synth

[fileset tb_files]
files =
file_type = verilogSource
usage = sim

[fileset include_files]
files =
file_type = verilogSource
is_include_file = true
usage = sim synth

[fileset tb_include_files]
files =
file_type = verilogSource
is_include_file = true
usage = sim

It's a few more lines, but the added flexibility outweighs the few extra bytes of ascii characters. It's now possible to set other file types, such as vhdlSource, or request specific language versions such as verilogSource-2001. The available file types are enumerated in the IP-Xact standard with some addtional ones allowed by FuseSoC (verilogSource-2005 and vhdlSource-2008 for now). The name of a fileset is not important, but the filesets are parsed in the order they appear in the .core file, so make sure that they are correctly ordered when dealing with languages where this is important.

The usage tag can be set to one or more items, and determines which tools that will use the fileset. These can be either a category (sim, synth) or a specific tool (verilator, quartus, modelsim...)

For languages that has a concept of libraries (VHDL, not verilog), the logical_name tag can be used to indicate library affiliation.

Another new feature is per-file attributes, which can be used to override the default value for the fileset. These attributes are placed inside square brackets at the end of the file. They are comma-separated and are either of the form attribute=value, or just attribute, to set boolean attributes. With per-file attributes, the above example can be changed to:

[fileset rtl_files]
files =
file_type = verilogSource
usage = sim synth

[fileset tb_files]
files =
file_type = verilogSource
usage = sim

Note that the file_type attribute for spi_image.vh was only added to show how multiple attributes can be set.

If there is an IP-Xact component file for the core, FuseSoC can parse that for the filesets instead. In that case, the above example will look like this

component =

Yep, that's right. We can leave it all to IP-Xact. Except for when we need file types that aren't in the IP-Xact standard (again Verilog 2005 and VHDL 2008 are the primary cases here), and the usage will default to both sim and synth. Future versions of FuseSoC will allow overriding options in the IP-Xact files to handle these cases. The IP-Xact support is described in greater detail in an earlier post

Mixed-language support

With the new file set features in place, most of the simulators and synthesis tools now accept both verilog, systemVerilog and VHDL files simultaneously for mixed-language projects.


Compile-time parameter enhancements

Both build backends now support setting top-level verilog parameters at synthesis-time. This can be used for example to initialize a boot ROM with different contents, or setting a version tag at compile-time. Using the de0 nano example once again, the following can be added to the .core file to register a parameter

[parameter bootrom_file]
datatype    = file
description = Boot ROM contents in Verilog hex format
paramtype   = vlogparam
scope       = public

To select a different boot ROM, run fusesoc build de0_nano --bootrom_file=/path/to/bootrom.vh

Slightly related are some enhancements to the simulator backends. Both verilator and XSIM now accepts parameters which are turned into plusargs, and all simulator backends now analyse the CLI arguments before building the simulator model to save some time.

Distutils and pypi

Apparently all the cool kids put their Python code on pypi, and it was suggested that I should do the same with FuseSoC. The easy way to do that is to replace autotools with a Python-based build system called distutils...or setuptools... which at some point gets used by pip... to create an egg or a wheel... which can be installed with pip.... or easy_install. I'm really not sure. Python packaging is a complete mess with around ten competing build systems. Try to Google it if you don't believe me. It's insane. I think this tweet pretty much sums it up.

So what are the benefits of this new Python-based build system. A small reduction in line count of the configuration files and the possibility to upload to pypi. It doesn't handle dependencies unfortunately, and there is no way to uninstall a package. How I love progress! The autotools-based build system is still there, until I figure out what to do.

That's all for now. I have already plenty of things lined up for the FuseSoC 1.5 release

torsdag 14 januari 2016

FuseSoC and IP-Xact

When I started the work on what would become FuseSoC, I had the ambition to somehow take advantage of the IP-Xact standard. Unfortunately my time to work on FuseSoC is limited, and there has been many other features that got higher priority. In the end, it might actually have been a good thing that it wasn't added until now, since it gave me plenty of time to get a feel for IP-Xact and figure out just how to best integrate it with FuseSoC.

Let's begin with a small introduction to IP-Xact.

According to Accellera, who oversees the standard, IP-Xact is "a well-defined XML Schema for meta-data that documents the characteristics of Intellectual Property (IP) required for the automation of the configuration and integration of IP blocks; and to define an Application Programming Interface (API) to make this meta-data directly accessible to automation tools"

So IP-Xact is one or several XML files that you can add to your IP core to describe certain properties of the core. There are tons of features in IP-Xact to handle different aspects of the design. These include things like register maps, external ports, source files, parameters, build commands and much more. IP-Xact files are becoming quite common as a method of encapsulating IP so that they can be integrated more easily in different EDA flows. Many of the EDA vendors are using it, even though they sometimes tend to use so many vendor-specific extensions that the IP in practice are of limited use outside of their own tools.

The observant reader will notice that some of IP-Xact's features are directly overlapping with the FuseSoC .core files. With the premise that double documentation is a bad thing, let's take a look at different options for only having one source for metadata.

  1. Just use FuseSoC .core files. This is what we have today, but the whole idea of the IP-Xact integration  is to expand the FuseSoC universe, take advantage of an existing standard and trying to not reinvent things. Also, this article would have been terribly short if I had decided on this option
  2. Just use IP-Xact files. This is a solid proposition, but my opinion on IP-Xact is that it is a flawed standard.  It's currently the best (only?) chance of a vendor-neutral standard at all though, so I can forgive some of its drawbacks, but not all of them. Some of the more annyoing problems are the file_type parameter that can be attached to each file. The latest IP-Xact standard currently defines 38 language types. Verilog is represented by verilogSource, verilogSource-95 and verilogSource-2001. VHDL has three similar options. C only has cSource and there are a few more generic such as swObject, SDC, unknown and user. The problem here is that this list doesn't stand a chance of keeping up with all the new and existing file types that people wants to use for digital design. It leaves out both VHDL 2008 and Verilog 2005, which makes it harder to decide on which compile flags to use for the EDA tools. It's also not aware of the new school of HDL like Chisel, MyHDL, Migen or Cx to just mention a few (a better solution was suggested here).

    I'm also not completely convinced by the attempt of becoming a build manager à la CMake or Apache ANT. The build management part of IP-Xact already have quite a lot of options, but not nearly enough to be flexible enough to support complex builds. I would rather see that they left this part out to avoid that the standard becomes to unwieldy.

    There are also plenty of options in the .core files that are not in IP-Xact. "A-ha!", says the experienced IP-Xact user here. "But IP-Xact has a built-in mechanism for specifying vendor-specific extension. All FuseSoC nonsense could just be marked as extensions". True, but first of all, that clashes a bit with the idea that FuseSoC shouldn't make unnecessary demands on the IP cores. One thing in particular that would be a bit complicated is how to specify where to find dependencies for a core. All this might be solvable, but I also find IP-Xact a bit too heavy and at the same time not specific enough compared to the existing .core files for some parts.

  3. My preferred option is instead to keep the .core files, but allow specifying an accompanying IP-Xact file from where FuseSoC can get additional metadata. This allows us to work with cores both with and without IP-Xact files and we don't have to make any changes to the upstream cores. We can avoid double documentation and add things to the .core files that are hard to do with IP-Xact. Win, win, win, win!

    To further motivate this decision, I'm drawing on the experience of software package managers here. To make a comparasion with the software world, IP-Xact is our AppStream file, that contains metadata for applications, that can be used in different package managers. The program itself is still put in a .deb or a .rpm file or specified by a .ebuild or pkgbuild file (the .core file). These are different layers that have some overlapping information.

For FuseSoC I'm currently using the file sets and core description, since these are options that are already available in FuseSoC and can be easily fetched from the IP-Xact file instead. There are other things in IP-Xact that could be used in the future as well, but we all know that the future is scary, so let's not talk about that.

To make FuseSoC aware of a core's IP-Xact component file, this file has to be added to the main section with the new option 'component=<file.xml>'. When the core is loaded, the file sets in the IP-Xact file are parsed and added to the list of filesets found in the .core file. There are some extra features on the roadmap here to make it possible to merge or replace filesets with similar names in both the .core and the IP-Xact file. Merging filesets can be useful as some file types (such as verilogSource-2005 and vhdlSource-2008) and other options are only available when the filesets are specified in the .core file. The group tag from the IP-Xact should probably be merged with the usage tag in the .core file as well at some point.

Note also that the  component option doesn't have to be a single file, but can also be a space-separated list of files. In this case, file sets are appended, but the description is taken from the first file that sets it.

The first IP-Xact features will be available in FuseSoC 1.4, but they are already available in git, and as a bonus, I have also put together a Proof of Concept system (a PoC SoC :)) , together with some instructions here, that uses the new features. It's a stripped down version of the de0 nano system that is available in orpsoc-cores and can be both simulated and built to an FPGA bitstream.

For those of you who haven't used FuseSoC before, you can find installation instructions in the FuseSoC repo

After installation, add the path to de0_nano_ipxact to your core library path and run
fusesoc build de0_nano_ipxact to build an FPGA bitstream or fusesoc sim de0_nano_ipxact to run a simulation in Icarus Verilog

Have fun and let me know what's good and what can be improved

lördag 19 december 2015

FuseSoC and VUnit

I recently improved the VHDL support in FuseSoC and since I've been using vunit a bit lately I thought it could be a fun experiment to see if I could combine the strengths of these projects.

For those not aware, FuseSoC is a package manager and build system for FPGA/ASIC systems. It handles dependencies between IP cores and has functionality to build FPGA images or run testbenches a single core or a complete SoC. Each core has a .core file to describe which source files it contains, which dependencies it has on other cores and lots of other things. The aim of FuseSoC is to make it easier to reuse components and create SoCs that can target different FPGA vendors and technologies. FuseSoC is Open Souce, written in Python and can be found at https://github.com/olofk/fusesoc

VUnit is an open source unit testing framework for VHDL released under the terms of Mozilla Public License, v. 2.0. It features the functionality needed to realize continuous and automated testing of your VHDL code. VUnit doesn't replace but rather complements traditional testing methodologies by supporting a "test early and often" approach through automation.

VUnit has a lot of great functionality for writing unit tests in VHDL, but requires the users to set up the source tree with all dependecies and their libraries themselves

FuseSoC on the other hand has knowledge of each cores files and dependencies, but very little convenience functions for writing unit tests. The ones that are existing are mainly targeting verilog users.

Given these preconditions, my idea was to let FuseSoC collect all source code and give it to VUnit to run unit tests on them.

Let's get started

VUnit requires the user to write a small Python script that sets up simulation settings, collects all source files, put them into libraries and starts an external simulator. This script is then launched from the command line with options to decide which unit tests to run, which simulator to use and where the output should go among other things. Here's the example script from VUnit's user guide
# file: run.py
from vunit import VUnit

# Create VUnit instance by parsing command line arguments
vu = VUnit.from_argv()

# Create library 'lib'
lib = vu.add_library("lib")

# Add all files ending in .vhd in current working directory

# Run vunit function

FuseSoC is also launched from the command line and expects to be instructed which core or system to use, if it should do synthesis+P&R or run a simulator and other options. FuseSoC however was always meant to be used as a library as well as a command-line tool, so to make these work together, we create a VUnit run script that imports the necessary functions from FuseSoC. Thankfully both tools are written in Python, or I would have given up at this point.

The first inconvenient difference between FuseSoC and VUnit is that the vunit run script need all source files that should be compiled for any testbench to run. FuseSoC on the other hand don't know which source files to use until we tell it which core to use as it's top-level core. To work around this I decided to look at the VUnit's -o option, which is used to tell VUnit which output directory to use. I simply peek at the output directory and use that as the FuseSoC top-level core name. We now got the first lines of the new script.

from vunit import VUnit
vu = VUnit.from_argv()
top_core = os.path.basename(vu._output_path)

Now we need to do some basic FuseSoC initialization. First we create a core manager, which is a singleton instance that is handling the database of cores and their dependencies.

from fusesoc.coremanager import CoreManager, DependencyError
cm = CoreManager()

Next step is to register a core library in the core manager. Normally FuseSoC picks up locations of core libraries from the fusesoc.conf file, which can be in the current directory or ~/.config/fusesoc, or from the --cores-root=/path/to/library command-line options.

We don't have any command-line options for this, but we can get fusesoc.conf by using the Config() singleton class.

from fusesoc.config import Config
config = Config()

We can also add any known directories directly with


The core manager will scan the added directories recursively and pick up any FuseSoC .core files it finds. (Note: If a .core file is found in a directory, its subdirectories will not be searched for other .core files).

It's now time to sort out the dependency chain of the top-level core we requested earlier

    cores = cm.get_depends(top_core)
except DependencyError as e:
    print("'{}' or any of its dependencies requires '{}', but this core was not found".format(top_core, e.value))

If a dependency is missing, we tell the user and exit. If all was well, we now have a sorted list of cores in the 'cores' variable. Each element is a FuseSoC Core class that contains all necessary information about the core.

With all the cores found, we can now start iterating over them in order to get all the source files and other information we need and hand it over to VUnit. Some notes to the code below:

1. 'usage' is a list of tags to look for in the filesets. Only look at filesets where any of these tags are present. FuseSoC itself looks for the names 'sim' and 'synth' to indicate if the files should be used for simulation and synthesis. We can also choose to only use a fileset with a certain tool by for example setting the tag 'modelsim' or 'icarus' instead of 'sim'
2. File sets in FuseSoC can be public or private. Default is public and indicates that other cores might find the files in there useful. This is for example the files for synthesis and testbench helper functions such as BFMs or bus monitors. Private filesets are used for things like core-specifc testbenches and target-specific top-level files.
3. Even though the most common way to use libraries is to have one library for each core, there's nothing stopping us from splitting a library into several FuseSoC cores, or let one core contain multiple libraries.

usage = ['sim']
libs = OrderedDict()
for core_name in cores:
    core = cm.get_core(core_name)
    basepath = core.files_root
    for fs in core.file_sets:
        if (set(fs.usage) & set(usage)) and ((core_name == top_core) or not fs.private):
            for file in fs.file:
                if file.is_include_file:
                    #TODO: incdirs not used right now
                    incdirs.add(os.path.join(basepath, os.path.dirname(file.name)))
                    except KeyError:
                    vu.add_source_file(os.path.join(basepath, file.name), file.logical_name)

With that we are done, and can safely leave it to VUnit to do the rest


It all works fine, and to make it more interactive, I set up a demo project at https://github.com/olofk/fusesoc_vunit_demo. This contains a packet generator for a made-up frame format together with a testbench. The packet generator has dependencies on two utility libraries (libstorage and libaxis) that I wrote a while ago to handle some common tasks that I got fed up with rewriting everytime I started a new project. The packet generator test bench uses some VUnit functions to make the example a bit more educational.

I hope this will be useful for all people using, or thinking of using, VUnit. For all you others, you can still use FuseSoC without VUnit for simulating and building systems and cores. The FuseSoC standard core library that can be downloaded as part of the FuseSoC installation contains about 60 cores, and there are several more core libraries on the internet that can be combined with this.

Happy hacking!

söndag 11 maj 2014

OpenRISC health report April 2014

About once a year, I'm at some conference to present the OpenRISC project for a new audience. One part of the presentation is often a quick rundown of all the things that has happened since the year before. Problem is that there's just too much stuff going on as the project continues to grow. To remedy this, I'm planning to write a small summary every month, or every quarter (time will tell), both to remind myself of what we are doing, and to make it more visible to the casual observer that there is pretty cool stuff going on with this nifty little CPU and the things around it. As this is the first installment, I would really have to present everything that has happened over the last 15 years to give a complete picture. I will not do that. You will have to settle with a back log from last year's OpenRISC conference.

While writing this article, I had to constantly ask for help explaining some of the concepts. This once again showed me how large this project has become and how much in-depth expertise that is shared by the participants. It also shows how fast things are moving, since most of the above items are things that has happened only during the last six months! I probably need to write a bit more often to avoid having to write a novel each time, but enough introduction. I will now leave the stage to OpenRISC and it's friends...

Debian for OpenRISC 

Perhaps the news item that has had the most coverage is the work done mainly by Christian Svensson to create a Debian port for OpenRISC. While OpenRISC, or rather the or1k architecture to be more precise, has been a supported target in the official Linux kernel since version 3.1, users have had very little support for creating a complete OpenRISC-based system. With the Debian port, things should be much simpler, and once the basic system is installed, your favorite package is just an apt-get install away. Naturally all packages aren't supported yet, and the package creation process has uncovered several bugs and missing features in the OpenRISC toolchain. The upside of this is that most of these bugs were fixed extremely fast, mostly by Stefan Kristiansson and Christian Svensson. As of now roughly 5000 packages have been built and packaged so far, thanks to Manuel A. Fernandez Montecelo. Many of the packages have been tested both in qemu and on a Digilent Atlys board with mor1kx. Apart from a multitude of library dependencies, the package list includes X.org, qt, scummvm, irssi and Fluxbox which means that you can now play games and chat somewhat comfortably on an OpenRISC system.

More information can be found here

Atomic operations

When running multiple cores, or multiple threads on the same CPU, each thread or core needs to know if someone else is currently trying to access an exclusive resource, such as an UART or a part of a shared memory. Normally this is done in software with mutex operation, such as semaphores or spinlocks, and the requirement for implementing a mutex is that an mutex operation is never allowed to be interrupted. Previously on OpenRISC this was done by making a syscall that disabled all interrupts as it's first instructions. That way it could make sure it wasn't interrupted by something else while it was busy doing it's work. Unfortunately this method is quite slow, and just doing ls generates  about 1700 system calls of which roughly 90% are mutex operations. Also, the syscall method won't work at all for multiple cores.

To improve the situation, Stefan Kristiansson has added two completely new instructions to the OpenRISC instruction called l.lwa (Load Word Atomic) and l.swa (Store Word Atomic). These two instructions form a load-link/store-conditional atomic operation, which can be read about in further detail on Wikipedia. Apart from the updates to the architecture specification, Stefan has implemented the operations in cgen CPU description in binutils so that the toolchain knows about them and can use them, the instruction set simulator or1ksim and in the RTL implementation mor1kx together with test cases.

An interesting side note is that while no OpenRISC implementation has ever had atomic operations before, Stefan found mentions of a pair of operations in a very early version of the architecture specification, and decided to reuse these names.


OpenRISC has traditionally been used as a single core CPU in a SoC, but there is an increasing demand for spreading workloads over several cores in a system. Of course, it has always been possible to instantiate as many OpenRISC cores as you can fit in an FPGA or ASIC. The problems arise when they need to cooperate to share resources like main memory or peripherals (exclusive accesses), spread a workload evenly between them (load balancing) and making sure that their local caches are kept in sync with each others (cache coherency).

For several years, Stefan Wallentowitz has been working on the necessary hardware and software additions required for multicore OpenRISC as part of his research project OpTiMSoC. A few weeks ago he made an announcement that there now was a multicore OpenRISC demo SoC based on mor1kx running under FuseSoC. The required changes have been added to mor1kx, and apart from a special version of the newlib toolchain it uses unmodified versions of the software and hardware IP cores.


During the last two or three years I have been working on a successor to the OpenRISC Reference Platform System on Chip (ORPSoC) that is used to run RTL simulations and provide a base for porting an OpenRISC SoC to a new FPGA board. The project was called ORPSoCv3 (to replace the old ORPSoCv2) and I did three releases before swiftly renaming it to FuseSoC. Why? you ask. Well, it turned out that there wasn't a single line of OpenRISC-specific code in FuseSoC, and I want it to be used as a general-purpose tool for building and simulating cores and SoCs for FPGA. This has become reality as it is now used for a SoC based on the eco32 CPU in addition to being the OpenRISC standard tool and other Open Source projects have shown interest as well.

Apart from the renaming and the never ending bug fixing, FuseSoC has gained many new features over the last six months. Some of the highlights include support for using SystemC in Verilator test benches, VHDL and using ISE to create images for Xilinx FPGAs. More detail can be found in the FuseSoC repository on github. The number of contributors continue to increase for every release which is a good indication that it is used by many people... or at least that many people find bugs.

ORPSoC Cores

The other part of the FuseSoC infrastructure is the collection of cores and systems that can be utilized by FuseSoC. The main repository is called orpsoc-cores and contains ready-made OpenRISC SoCs for different FPGA boards as well as the cores that are used on the boards.

Since October last year there has been added support for the very popular Digilent Atlys board, with ports for the LX9 Microboard, DE2 70, SoCKIT, NEEK and ordb2a in the works.

Many new cores have been added as well, and there is currently 39 supported cores in total. For more details see the ORPSoC Cores repository on github.


The de-facto standard open source toolchain for creating programs consist of two parts. The compiler (GCC, with LLVM getting more common as well) and binutils. Binutils can be summarized as everything but the compiler. This means the linker, assembler, archiver and tools for modifying ELF files. OpenRISC support in binutils was added 14 years ago, but since then, no improvements have been sent upstream.

There has been a huge amount of work over the last years to improve the OpenRISC binutils support, and earlier this year it was decided that these changes should be sent upstream. Having them in the official binutils distribution means that anyone who wants to make binutils for OpenRISC can now use the official binutils package.

In the OpenRISC case, sending the changes upstream wasn't completely trivial even though the code itself was in very good shape, as FSF (Free Software Foundation) who owns binutils need to have the written agreement of all contributors that they are OK with giving away to copyright of the code to FSF. That means that the first thing that had to be done was to go through all changes and hunt down all the people who have contributed.

I volunteered to take on this mind-numbing job, which wasn't made any easier as we switched VCS from SVN on OpenCores to GIT on github a few years ago, and renamed the architecture from or32 to or1k. Luckily, most people had been very good with updating changelogs, which made things a lot easier. In the meantime, Christian Svensson had split up the OpenRISC-specific changes to a set of patches that was sent in for review. When that was done, all contributors had to get a copyright assignment document mailed from FSF to sign and send back. As the patches were already reviewed, the rest of the process went quickly by when all the mails had been sent back.

Stefan Kristiansson and Christian Svensson are now the proud caretakers of the OpenRISC port which should be included in the next binutils release.


While there are many things that most people agree on, there seem to be no such consensus when it comes to standard C libraries. Everyone seems to have their own specific use case that requires a brand new C library. OpenRISC already has support for glibc, uClibc and newlib, but the new kid on the block when it comes to embedded software is musl. musl is designed to be small and portable which suits the OpenRISC use case just fine. Stefan Kristiansson has prepared a musl port which is mostly done apart from some details about syscall handling that needs to be resolved first. Now we only need to port bionic and we're ready to take on Android.


The up-and-coming mor1kx OpenRISC RTL implementation is rapidly gaining new features, perhaps as a testament to it's more modular design compared to the original or1200 implementation. In addition to the atomic operations mentioned above and multiple bug fixes it now supports caches with more than two ways,  and the ability to reading the GPRs through the debug port.

ORconf 2014

Last year's OpenRISC conference in Cambridge, England was a great success and attracted participants from both academia and the industry. As for this year, Stefan Wallentowitz has kindly offered to host the conference at Munich University. Initial planning has started and nothing is set in stone yet, but it will probably be carried out the weekend October 11-12. More information will become available at http://orconf.org over the coming months.