GCC Compiler Accepts China's MIPS-Derived LoongArch CPU Port

[phoronix]

Phoronix: GCC Compiler Accepts China's MIPS-Derived LoongArch CPU Port

The GCC steering committee has signed off on the LoongArch compiler port and could still land for the GCC 12 stable compiler release in a few weeks...

https://www.phoronix.com/scan.php?page=news_item&px=LoongArch-For-GCC-12

[tildearrow]

This is just MIPS under a different name....

[coder]

This is just MIPS under a different name....

In a previous thread, one of the developers outlined a number of differences that sounded fairly significant, in aggregate.

I guess the main issue is how much code is duplicated from the MIPS path, to the point that it would just be easier to maintain it as a newer version of MIPS? I wonder what approach GCC took, although it's worth noting that the ISA could differ substantially while the system-level architecture could be nearly identical to MIPS. So, the right answer for GCC might differ from the right answer for the kernel.

[xen0n]

The LoongArch ISA is semantically closer to MIPS R6 than RISC-V, but the ABI is decidedly modeled after that of RISC-V. Its privileged architecture is mostly MIPS-style though. The implementation details are different enough to make code sharing with either MIPS or RISC-V unfeasible.

The LoongArch gcc port is an amalgamation of MIPS and RISC-V code, but honestly, every new gcc port invariably copies lots of code, maybe because the gcc codebase is simply too antiquated and complex for anyone to port from scratch.

[xen0n]

Is it possible to buy a LoongArch CPU+motherboard in Europe (or from Europe)?

They apparently haven't solved the problems of sales channels even domestically yet... but we'll see.

For now it's not recommended for end users to buy LoongArch products, because many essential things are still in flux -- at least a mandatory firmware upgrade / compatibility shim needed, and a full re-bootstrap (basically re-installation from scratch with binary distributions) already pending on the horizon, so you're going to jump through multiple hoops if all you want is to "regularly use" it.

P.S.: I wrote an FAQ around LoongArch things back in February; hope that is informative as much information about LoongArch is not translated. (Note that I plan to revise it to include more opinions criticizing such development; in its current form the article is probably not neutral from a (stereo-)typical Western viewpoint. There are significant criticism even inside China, and the article was written to not enrage any of the "pro-Loongson" (many of them nationalists), "pro-MIPS" (many of them liberals) or "pro-community" people, though.)

[scottishduck]

I think the Chinese & Russian alt-arch projects remain very compelling. Especially when they both seem to consider OSS the most secure path forward.

[coder]

P.S.: I wrote an FAQ around LoongArch things back in February; hope that is informative as much information about LoongArch is not translated.

Thank you for writing this and for answering questions & sharing information on these forums.

It's always good to have solid information. And perhaps it can better inform some views and opinions.

[tuxd3v]

P.S.: I wrote an FAQ around LoongArch things back in February;

thanks

Only a question..
Code compiled for 3A4000 can run on 3A5000, you speak about the instruction set being incompatible?
3A5000 - is fabed at 16nm process right?

thanks in advance

[xen0n]

thanks

Only a question..
Code compiled for 3A4000 can run on 3A5000, you speak about the instruction set being incompatible?

No, that's definitely not the case; the early materials from Loongson suggesting compatibility with MIPS binaries are not translated AFAIK (so the English-speaking world mostly wouldn't know about this), and this compatibility is never demonstrated in any publicly available distro, be it commercial or community. All you would get is the same "Exec format error" when you try to mix and match like this.

3A5000 - is fabed at 16nm process right?

thanks in advance

The 3A5000/3C5000 is likely fabbed at 12nm -- Loongson didn't release this particular info, unlike with earlier models, but this is easily revealed by simple searches, I checked both Google (English sources) and Baidu (Chinese sources). They always artificially cap their process node to the highest achievable/available in China mainland, so it's pretty predictable after all.

[tuxd3v]

The 3A5000/3C5000 is likely fabbed at 12nm
They always artificially cap their process node to the highest achievable/available in China mainland, so it's pretty predictable after all.

Thanks
I read a bit about China struggle to get more advanced process nodes, and I also read about China already having a 28nm process that could work at 16nm, with Chinese technology, for what I read they are trying to advance it..

But now you are talking about 12nm, would it be ASML technology, or is the developments of in-house technology that is progressing at high speed.
It would be interesting to know.

I know that there are already 3nm chips made in China with Chinese technology, but the problem is maintain same quality for mass scale production..
Anyway if China Mainland is already at 12 nm( for mass scale production ) means they are advancing at crazy high speed!