GNU Compiler Collection adds support for China's LoongArch CPU family

Version 12.1 of the GNU Compiler Collection (GCC) was released this month, and among its many changes is support for China's LoongArch processor architecture. The announcement of the release is here; the LoongArch port was accepted as recently as March. China's Academy of Sciences developed a family of MIPS-compatible …

[martinusher]

Change doesn't happen overnight

These processors might be effectively obsolete because of RISC-V but there's still a lot of them out there and there will be plenty in the production pipeline. Its the same with actual MIPS processors -- the switch will occur over a period of years because of product life cycles.

Integrating this architecture with the GUN toolchain would be a first step to phasing it out since it would free a lot of developers who would have been otherwise engaged in developing and maintain a proprietary toolchain to move to new work. (...and given the depth and sophistication of the GNU toolchain it doesn't make sense to go it alone anyway, you're far better off contributing work to the open source project)

[herman]

Deja RISC

MIPS and RISCV are reincarnations of an old favourite of mine. It is nice to play with these things in a FPGA and instead of simply using a RPi, or a PIC, make your own processor.

[Clausewitz4.0]

Elbrus and Godson

Other nations should also focus on developing their own brand of CPU, with its own instruction set.

Dependence on x86/x64 makes it too easy for Intel/AMD market dominance, and national security messing around.

[rcxb1]

Re: Elbrus and Godson

> Dependence on x86/x64 makes it too easy for Intel/AMD market dominance, and national security messing around.

Why not make your own x64 CPU? First released in 1999, patents should be expired most everywhere. Then you can run most existing software without modifications, you can keep Intel/AMD in check, and will be able to sell it to other countries as well if it is any good.

[sreynolds]

Re: Elbrus and Godson

What's a patent got to do with an instruction set? And what about the extensions to x64 like MMX SSE1..4 AVX1..2 etc.

Isn't that like an interface from the JVM that you can copy. But interestingly if someone culled a lot of the useless x64 instructions and re-implemented them from scratch, IMHO they could proably come up to par with AMD and even the bottom 10th percentile would be able to implement something as bad as Intel's implementation.

[rcxb1]

Re: Elbrus and Godson

> What's a patent got to do with an instruction set?

See:

https://www.theregister.com/2000/10/30/intel_patents_ia64_instruction_set/

https://newsroom.intel.com/editorials/x86-approaching-40-still-going-strong/

> what about the extensions to x64 like MMX SSE1..4 AVX1..2 etc.

I didn't say you'd get 100% compatibility. That said:

*Code is already typically compiled for baseline x86-64

*Instructions unknown to the CPU can be trapped and handled in software

*Alternatives can be implemented (like AMD did with 3dnow) and if it works well, others will implement it.

*Compiler makers will quickly implement any functional differences that are clearly documented to enable wider compatibility.

*Developers doing low-level optimizations will be slower to adapt, but they will get there if the processor becomes popular at all.

[YARR]

Those Chinese commies are having their LoongMarch through the Linux institutions.