Loongson Begins Posting Linux Patches For 3A6000 Series CPUs

[phoronix]

Phoronix: Loongson Begins Posting Linux Patches For 3A6000 Series CPUs

While the Loongson 3A6000 processors have yet to be officially launched, rumors since last year put it on target for launching in the first half of this year and some claims that there is such performance uplift that these Chinese CPUs could rival AMD Zen 3 or Intel Tiger Lake levels of performance. Ahead of the 3A6000 series launch, Linux patches have begun appearing for these next-gen LoongArch processors...

https://www.phoronix.com/news/Loongson-3A6000-Linux-Patches

[kylew77]

It is worth noting that OpenBSD has support for Loongson but I don't know if it is the older Loongson or the modern reimplementation?

[saladin]

Hold on, does the 3A5000 not have page walkers?

Update: after a quick look at chipsandcheese and the kernel docs, there's no mention of either software managed TLBs or a page walker. It's a mystery.

[NovenTheHero]

I really want to get one of these with a compatible motherboard just because they are niche and not really available.

[brad0]

It is worth noting that OpenBSD has support for Loongson but I don't know if it is the older Loongson or the modern reimplementation?

There is no modern reimplementation. There are two different architectures. Older Loongson CPUs are MIPS. Newer CPUs are LoongArch.

[kylew77]

There is no modern reimplementation. There are two different architectures. Older Loongson CPUs are MIPS. Newer CPUs are LoongArch.

Thank you for the clarification!

[stormcrow]

There is no modern reimplementation. There are two different architectures. Older Loongson CPUs are MIPS. Newer CPUs are LoongArch.

That's the Chinese claim. Once the code was introduced to support LoongArch, however, kernel devs have pointed out that at least some of LoongArch ISA support code is a direct copy of MIPS and just a fork of MIPS64 R6. I'm more inclined to believe the kernel devs than I am marketing material from Loongson who's actually owned by their Institute of Computer Technology under the auspices of the Chinese Academies of Science itself a branch of the PRC government.

Kinda have to keep in mind that the PRC is desperate to wean themselves off US and other Western technology companies like Intel, AMD, and ARM but they don't have a lot of creative types. Every time a Chinese company makes a product announcement they compare it favorably to the point of outperforming the products they're trying replace to make sure their companies and especially the CCP save face in front of their internal populace. Western developers aren't beholden to follow the CCP script however, so sometimes big lies, like LoongArch and the LoongISA are entirely internally created, are exposed. No one in the West cares about China saving face when they try to spin a Big One, however, except those corporations with their hand in the PRC's cookie jar.

[Spacefish]

Newer CPUs are LoongArch.

Which is "heavily inspired" by MIPS

[Sonadow]

That's the Chinese claim. Once the code was introduced to support LoongArch, however, kernel devs have pointed out that at least some of LoongArch ISA support code is a direct copy of MIPS and just a fork of MIPS64 R6. I'm more inclined to believe the kernel devs than I am marketing material from Loongson who's actually owned by their Institute of Computer Technology under the auspices of the Chinese Academies of Science itself a branch of the PRC government.

Kinda have to keep in mind that the PRC is desperate to wean themselves off US and other Western technology companies like Intel, AMD, and ARM but they don't have a lot of creative types. Every time a Chinese company makes a product announcement they compare it favorably to the point of outperforming the products they're trying replace to make sure their companies and especially the CCP save face in front of their internal populace. Western developers aren't beholden to follow the CCP script however, so sometimes big lies, like LoongArch and the LoongISA are entirely internally created, are exposed. No one in the West cares about China saving face when they try to spin a Big One, however, except those corporations with their hand in the PRC's cookie jar.

Another dumbfuck who quotes one line from a kernel dev making claims that it's a copy of MIPS while demanding that Chinese developers make Loongarch behave more like MIPS by nitpicking on how Loongarch does things differently from MIPS.

[xen0n]

Hold on, does the 3A5000 not have page walkers?

Update: after a quick look at chipsandcheese and the kernel docs, there's no mention of either software managed TLBs or a page walker. It's a mystery.

Before the Loongson 3A6000 the TLB management approach is virtually identical to MIPS where most logic is managed by software. Compare arch/loongarch/mm/tlbex.S where things bear a certain resemblance to the Loongson-specific parts in arch/mips/mm/tlbex.c (the MIPS assembly has to be dynamically generated, because any given Linux/MIPS build has to adapt to a range of HW with slight differences here and there).