Loongson Extending Etnaviv Driver For PCI Device Support

[WorBlux]

Yeah, based on the old, legacy MIPS architecture, as opposed to RISC-V which is a modern, clean-slate, legacy-free design. Also, RISC-V is made to be free, open, royalty-free and free of external patents, so it would be a great choice.

It's precisely because there is no patent pool, which makes if vulnerable to external patents, and fragmentation from vendor patenting certain features. (it's vulnerable to Embrace, extend extinguish). Also while the ISA is free, the on-chip details of implementation may not be.

[uid313]

I'm sure that's what they said when RISC-V was announced: “How stupid, they should just ditch RISC-V and adopt ARM or x64 instead.”.

Who knows, maybe Loongson will be as good as RISC-V in most areas.

No, they couldn't adopt x64 because it is a proprietary architecture owned by Intel and AMD.
They could adopt ARM because it possible to license the architecture, but then they would still have to pay royalties and pay for a license, and it is owned by a British company, which means in case of sanctions they could pressure Arm not to renew the licenses or to revoke the licenses.

So RISC-V is attractive because it is free, open, royalty-free, modern, legacy-free.

[uid313]

Again: that was the thing with ARM, too. Started out locally and is now backed by many of the largest companies. And RISC-V too, btw: started out locally in British RISC PC's and is only now backed by many of the largest companies. So the same could happen to Loongson.

RISC-V was backed by large universities since day one such as University of California, and since day one had one had prominent and famous designers, scientists and engineers such as David Patterson.

Based heavily on MIPS64... and not too difficult to port. Also I think it's not just about the ISA, but having a design that doesn't rely or touch external patents that they can use as a target for their own fab processes.

Yeah, based on the old, legacy MIPS architecture, as opposed to RISC-V which is a modern, clean-slate, legacy-free design. Also, RISC-V is made to be free, open, royalty-free and free of external patents, so it would be a great choice.

[WorBlux]

Yeah, Loongson probably isn't bad, and probably will be good, but its their own thing while RISC-V is internationally backed by many of the largest companies so it will have much better adoption, maturity, support and ecosystem with kernels, applications, tooling, compilers, linkers and debuggers.

Based heavily on MIPS64... and not too difficult to port. Also I think it's not just about the ISA, but having a design that doesn't rely or touch external patents that they can use as a target for their own fab processes.

[Vistaus]

Yeah, Loongson probably isn't bad, and probably will be good, but its their own thing while RISC-V is internationally backed by many of the largest companies so it will have much better adoption, maturity, support and ecosystem with kernels, applications, tooling, compilers, linkers and debuggers.

Again: that was the thing with ARM, too. Started out locally and is now backed by many of the largest companies. And RISC-V too, btw: started out locally in British RISC PC's and is only now backed by many of the largest companies. So the same could happen to Loongson.

[uid313]

I'm sure that's what they said when RISC-V was announced: “How stupid, they should just ditch RISC-V and adopt ARM or x64 instead.”.

Who knows, maybe Loongson will be as good as RISC-V in most areas.

Yeah, Loongson probably isn't bad, and probably will be good, but its their own thing while RISC-V is internationally backed by many of the largest companies so it will have much better adoption, maturity, support and ecosystem with kernels, applications, tooling, compilers, linkers and debuggers.

[SteamPunker]

How do you mean "existing" ? I'm not aware that such an interesting card exists yet. It could indeed solve some cases.

Escuse me, I meant to type "interesting" there. I totally understand the confusion. To my knowledge, such a card does not exist yet.

[phoron]

Interesting. Does this mean that someone could produce a discrete PCIe graphics card with a Vivante GC1000 graphics core? It might not win any awards in terms of performance, but it could be a very cheap and low-power GPU with increasingly solid open-source driver support, which might make it an existing option in quite a few use cases.

How do you mean "existing" ? I'm not aware that such an interesting card exists yet. It could indeed solve some cases.

[Dawn]

I'm keen to see the same treatment for RISC-V, so I can actually use my c910 board without binary blobs!

But really, the Etnaviv driver is in a bit of a sad state without anyone really supporting it. Is it abandoned?

Vivante really is the only other GPU that is supported on RISC-V SoCs, as, well, y'know, Mali is Arm's.

There have been Mali GPUs in non-ARM types (MIPS [Alchemy], x86 [SofIA], and ST40/SH4 [STMicro STiH.]) Unless something has changed, ARM does not care about the ISA of the apps core you plug the Mali into.

Imgtec's PowerVR also exists, and is in use in some RISC-V devices. I'd guess PowerVR core volume is many times higher than Vivante.

[pierce]

I'm keen to see the same treatment for RISC-V, so I can actually use my c910 board without binary blobs!

But really, the Etnaviv driver is in a bit of a sad state without anyone really supporting it. Is it abandoned?

Vivante really is the only other GPU that is supported on RISC-V SoCs, as, well, y'know, Mali is Arm's.