Go support on Epiphany

[stordopoulos]

[simoncook]

[stordopoulos]

I'm afraid I'm not the person for the job. Go would be perfect for parallella and Adapteva should do their very best to support it out of the box. Since Go is implemented in GCC, Epiphany should support it as well. At least that's my humble opinion. The concurrent stuff in Go and their Epiphany architecture seems like a great "marriage". If anyone is interest in doing that please drop a line here.

Again, in my humble opinion, the people from Adapteva should be the ones to curry the weight since Go is officially under the GCC umbrella.

[9600]

[mortdeus]

I'll be working on this. Is there an emulator I can use to test on?

[LamsonNguyen]

Yes, it is included in the SDK, though it only emulates one core, not inter-core communications.

[stordopoulos]

[Folknology]

@mortdeus Hopefully you know more about this than I, but on a cursory glance it appears that Go has a quite large runtime requirement, due in part to its type system, does this larger runtime have to be replicated into each epiphany core memory in order for Goroutines to run there, or is there some sort of cross runtime magic that means it can sit in the common ARM memory map shared between cores and Arm. I ask because of the 32K limitations of the epiphany core memory which could be a show stopper if sharing isn't possible, I don't have the key knowledge however with Go and its implementations to know if this is the case or not, some insight here would be informative from someone more knowledgeable.

It would certainly be great to see a good CSP implementation running on epiphany.

regards
Al

[stordopoulos]

@Folknology I can't imagine that being the case. Some ARM chips have the same limitation (0-32K) but Go runs fine on them. Specifically the ARMv7 architecture (Cortex- A7,A8,A15).

[mortdeus]

http://golang.org/doc/faq#goroutines

Im still reading the spec, but as far as im aware there is nothing that will be a showstopper for Go. If anything Epiphany will truly shine using Go because of the built in concurrency semantics. Another ambition of mine with this project is to port plan9 to run on epiphany. Especially considering the plan 9 derived research project going on in http://code.google.com/p/nix-os/. It would be interesting to tailor the ideas used in this project to epiphany.

I have a question for anybody from the parallella team working on Linux. Ive been reading a few white paper specifically about tegra and the nix os and it seems like benchmarks suggest that 4 cores are the "sweet spot" in terms of SMP kernel schedulers. Are there any sort of limitations in epiphany like this that are strictly because of the OS's lack of ability to handle that much concurrency? I mean assuming that were truly aiming for 18 million cores. :p

http://lsub.org/ls/export/nixposterabs.pdf
http://www.nvidia.com/object/white-papers.html