cva6 VS ara

Hello. I'm looking into implementing RISC-V on an FPGA for a school project. The two repos I'm looking into using are the Ariane and RocketChip repos. Both look actively maintained, but RocketChip has more recent releases, and it's used by this other repo that creates a block design in Vivado with the RISC-V RTL. However, we would also like to be able to make changes to the core, and I'm afraid that scala/Chisel might be difficult to learn. Ariane looks like SystemVerilog while RocketChip is mostly Chisel. Does any have recommendations on which RISC-V repo would be good to use for a project?

There are 108 RISC-V cores that have been created so far (according to this list), but only a couple are 64 bit, open source and powerful enough that you would want to use them (like Shakti, CVA6 and NutShell)

The source code of Ara as well as Ariane, also known as CVA6 is available on GitHub.

I made a post about this here not too long ago, but I think it would be really useful to come up with a Vivado benchmark, in the form of a standardized large and representative design. I was curious about Alder Lake performance too, and compared my new 12700K workstation against my laptop with this open source RISC-V CPU: https://github.com/openhwgroup/cva6

When the OpenHW Group was created in 2019, I had some hope that Alibaba or NXP (who are in the OpenHW Group) would release an open hardware RISC-V processor, but it looks like they are not making any public commits to the CVA6 core, so I doubt that we are ever going to see the source code of Alibaba's XT910 or NXP's Chassis RISC-V processor.

Ariane is now cva6 (it moved to a industry supported non-profit).

At this stage, it could make sense for e.g. universities to start developing peripherals & controllers targeted at ASIC rather than creating yet-another-core (https://riscv.org/exchange/cores-socs/ has 107 lines already for cores), leveraging an OSHW ASIC-proven core from e.g. the OpenHW group (https://github.com/openhwgroup/cva6). Manufacturing in not-so-old processes is affordable for teaching institutions (e.g. https://europractice-ic.com/ in Europe), and taping out working cores is no longer a 'new' thing (e.g. http://asic.ethz.ch/all/years.html ).

Looking at the github of the openhw group looks like the license is granting patents to the project. So it looks ok.

yeah, ara also currently doesn't work, but that it exist is already really cool, and will likely get fixed and completed in the future

The ISA is variable length/scalable, but this implementation uses a 4096 wide register file.

They are a bit disingenuous in claiming they support rvv 1.0 while others only a subset, as they haven't implemented vrgather or vcompress yet, but there are open pull request for them [0].

Sadly there also seem to be a few bugs when simulating with verilator [1], so I couldn't measure all instructions, but here is `vadd.vv` and `vwaddu.vv` for the VLEN=4096, four lane configuration:

vadd.vv:

e32m1: 16 cycles

e32m2 32 cycles

e32m4 63 cycles

e32m8 126 cycles

vwaddu.vv:

e32m1: 34 cycles

e32m2: 69 cycles

e32m4: 140 cycles

[0] https://github.com/pulp-platform/ara/pull/180

[1] https://github.com/pulp-platform/ara/issues/250

The PULP Ara is a 64-bit Vector Unit

It's not made very clear until the Conclusion section that they have provided an open source implementation of a RVV 1.0 vector unit, available at https://github.com/pulp-platform/ara : "The PULP Ara is a 64-bit Vector Unit, compatible with the RISC-V Vector Extension Version 1.0, working as a coprocessor to CORE-V's CVA6 core."

The source code of Ara as well as Ariane, also known as CVA6 is available on GitHub.