Migen Amaranth

Why is LiteX Migen based and hasn't switched Amaranth?

To give a bit of context/history, Migen DSL has been around since 2012 and we have been involved in its development since then while also working on LiteX and creating the ecosystem of cores.

Since 2012, LiteX has been used to create very various FPGA based systems for open-source and commercial projects and most of the codebase is currently tightly linked to Migen since written with its DSL. Most of these projects are still maintained and can't for now justify a complete rewrite.

Amaranth has been developed since 2018, several years after the initial LiteX development, and is revisiting/improving some of the concepts introduced by Migen. Switching the DSL used by LiteX needs to be prepared and will require hard work, time (and/or money :))... and we are still evaluating if it's worth switching to Amaranth or have our own DSL.

Why isn't LiteX using Amaranth's Migen compat mode?

Amaranth has a Migen compatibility layer that works great for small/medium size projects, but:

• This is not a long term solution.
• This still requires changes to the codebase.
• This does not mix well with LiteX relying on some Migen internals for simulation, build, logic analyzer, etc...

At some point, using Amaranth compat mode was envisaged but for the foregoing reasons this does not seem practical.

Is it already possible to use Amaranth modules/cores with LiteX?

LiteX is a very permissive SoC builder and the fact that the main modules/cores are still written with Migen DSL does not prevent integrating Amaranth modules in a LiteX SoC by elaborating the Amaranth module to verilog and integrate it in the design. A good example for this is the Minerva CPU integration in LiteX, the CPU is elaborated at build-time from Amaranth sources with appropriate parameters, exported to verilog and integrated in the LiteX design as any other CPU.

Even new DSLs allow developers to avoid VHDL/Verilog with FOSS toolchains, VHDL/Verilog will probably remain interchange formats for quite some time and developers will probably create standalone verilog core generator to facilitate reuse from other flows, similarly to what we are already doing with LiteDRAM, LitePCIe, LiteEth, etc...

Reusing Amaranth modules with LiteX can already be done in two different ways:

• Reusing a pre-generated verilog provided by developers.
• Elaborating the verilog from the Amaranth during the LiteX build (as done with Minerva CPU).

The latter is more flexible/powerful since allows elaborating the Amaranth module with appropriate parameters.

Will I be limited by this?

Not really, Amaranth some advantages over Migen and LiteX has features and a rich ecosystem of cores not already available in others projects, so while waiting for the planets to be aligned, the best is probably to take the best of both worlds:

• Develop in the language you find the most appropriate for you (VHDL/Verilog/Amaranth/Migen, etc....)
• Reuse other open-source cores when you don't have the time or don't want to develop it yourself. (Amaranth core exported to verilog and integrated in a LiteX design, LiteX core exported to verilog and integrated in a Amaranth design, Amaranth/LiteX core exported to verilog and integrated in a traditional VHDL/Verilog flow, etc....)
• Use the right tool/language for the right job.
• ... and most important: Enjoy what you do, understand that things are generally not perfect and that your help would be welcome to go in that direction!

Note: VexRiscv, probably the most popular CPU used with LiteX is written in Spinal HDL and Linux on LiteX VexRiscv project is a good example of the possible interoperability between various high level HDLs.

What are the long term plans regarding LiteX?

Things are still not fully decided, if:

• current LiteX codebase should be converted to Amaranth/our own DSL.
• current LiteX codebase should stay in Migen and new modules/cores developed with Amaranth or our own DSL.

...For now, there are still lots of things to do and in the short term Migen is not really limiting the project and a full DSL switch not really yet justified, so the project is still going to evolves slowly but surely and if developers want to help us improving things on the different fronts (including the Amaranth one) we'll be happy to collaborate!