Multi Core PULP first time setup

[Firasramadan]

Hello all, 

Diring a research project we are planning on using the multi core pulp. We started digging and we have multiple questions and inquiries that we would appreciate your help for: 
1. Where can we find the most updated data for PULP ? The github repository was last updated 2 years ago (?) is this accurate ? Can we rely on that ? 
2. What are the simulation tools that were used and which version work for sure ? I saw alot of questions regarding this point but couldn’t find a decisive answer  
3. In the future, we plane to go for tape-out. Are there any scripts that were used for Synthesis and Place-and-Route ? 
4. Benchmarking: other than the regression tests and simple tests that are available in the repository, is there any other platform or benchmarking guidelines for the multi core pulp ? Or should we build our benchmarks from scratch ? 
5. The SDKs in the repository seems to updated last time in 2019, is that okay ? Does work properly? 
6. Do you have a support for PLLs inside the core or the implementation of IO rings that we could get help from ? 


Your help is much much appreciated, 
Looking forward to hear from you. 

Best, 
Firas

[kgf]

Hello,

Please note that PULP is not a single architecture, it is a series of building blocks we use to design novel computing architectures. So there is not one PULP, there are independent projects underneath it. We have developed different RISC-V cores:
- Ariane (now CVA6 by OpenHW)
- RI5CY (now CV32E40P by OpenHW)
- Zeroriscy (now Ibex by LowRISC)
- Snitch
And combined them to get single microcontrollers (PULPinio PULPissimo, Cheshire), many-core accelerators (Snitch clusters, PULP clusters), and largers systems (Mempool, Occamy)

Since I get this question a lot, we do not really have a traditional multi-core system (with the exception of Culsans https://github.com/pulp-platform/culsans), our many-core architectures are geared towards data centric applications and used shared scratchpad memories and not traditional caches.

We use the public git repositories for our work, so all of them are actually up to date, but some are no longer an active research area for us (i.e. we achieved what we set out to do), and some yu need to check the development branches.

In most cases, we use Siemens/Mentor Questasim, any recent version should do the trick. Notice that the Intel/Altera edition is not really 100% the same as the regular Questasim, so it may not simulate properly. We have been moving towards using Verilator for most of our projects, but there is a lot of code, and not all of them have been yet converted.

Commercial EDA vendors (so far) do not really allow us to publish the synthesis and PR scripts. For the open source versions (https://github.com/pulp-platform/cheshire-ihp130-o) everything is available

Benchmarks are usually project specific, i.e. you need to have your application that you consider running on your system.

SDK is again a complicated issue, different platforms have variations of the system. Check for example Deeploy https://github.com/pulp-platform/Deeploy for a system we are currently using.

Technology specific aspects like PLLs and I/O cells will need to be managed by you, we are not allowed to make these public.

I hope this helps a bit,

Happy New Year
KGF