Announcement

Collapse
No announcement yet.

Loongson Volleys Latest Patches For LoongArch Linux Support

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Loongson Volleys Latest Patches For LoongArch Linux Support

    Phoronix: Loongson Volleys Latest Patches For LoongArch Linux Support

    Chinese vendor Loongson continues working on their Linux kernel patches enabling the LoongArch processor ISA as their fork from MIPS. While early on when copying existing MIPS open-source code they were quick to call their new ISA "not MIPS", in these later patch series they continue to refer to their ISA as "a bit like MIPS or RISC-V."..

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    A bit like MIPS? Erm.... a lot like MIPS. Same amount of registers (bar minor differences), same instructions.
    Just different syntax

    Comment


    • #3
      Originally posted by tildearrow View Post
      A bit like MIPS? Erm.... a lot like MIPS. Same amount of registers (bar minor differences), same instructions.
      Just different syntax
      I can even imagine the meeting between marketing and engineering about that:

      Marketing: Okay, we decided...
      Engineering: We who?
      Marketing: We Marketing®, the ones that bring the money to the company, my fellow simpleton.
      Engineering: sigh...
      Marketing: We decided to sell it as "LoongArch".
      Engineering: Sell what as "LoongArch"?
      Marketing: The thing you guys call architecture.
      Engineering: Why?
      Marketing: Because it strengthen the brand. Our research show people see that MIPS thing as a obsolete and undesirable product.
      Engineering: But it is just MIPS with a few additions. People making programs for it will see right trough your bullshit. They will still call it MIPS.
      Marketing: Oh my marketing impaired friend, people will call it what we decide to call it. That is the POWER® only Marketing® have.
      Engineering: sigh...

      Comment


      • #4
        Originally posted by M@GOid View Post

        I can even imagine the meeting between marketing and engineering about that:

        Marketing: Okay, we decided...
        Engineering: We who?
        Marketing: We Marketing®, the ones that bring the money to the company, my fellow simpleton.
        Engineering: sigh...
        Marketing: We decided to sell it as "LoongArch".
        Engineering: Sell what as "LoongArch"?
        Marketing: The thing you guys call architecture.
        Engineering: Why?
        Marketing: Because it strengthen the brand. Our research show people see that MIPS thing as a obsolete and undesirable product.
        Engineering: But it is just MIPS with a few additions. People making programs for it will see right trough your bullshit. They will still call it MIPS.
        Marketing: Oh my marketing impaired friend, people will call it what we decide to call it. That is the POWER® only Marketing® have.
        Engineering: sigh...
        They're also using the Intel method.
        Intel Post-Skylake Marketing Release Strategy
        DO NOT RELEASE TO THE GENERAL PUBLIC
        Take a previously existing Skylake product, give it a new name, add a + to 14nm, change the socket to break compatibility, repeat for 7 years.

        Comment


        • #5
          Originally posted by tildearrow View Post
          A bit like MIPS? Erm.... a lot like MIPS. Same amount of registers (bar minor differences), same instructions.
          Just different syntax
          Actually it's more of a Frankenstein with bits of MIPS and bits of RISC-V, and totally different encoding of course. For example it doesn't have any notion of delay slots; its long jump ("j") is PC-relative instead of PC-regional like the MIPS j; its "pcaddu12i" is actually the same as RISC-V auipc, "jirl" the same as RISC-V jalr, etc etc etc.

          IMHO the way this whole Loongson thing got its present reputation is at least partly due to language barrier. Most people I know that are familiar with Loongson systems/people can only barely express in English; they seldom jump in the conversation as a result, only doing infrequent code drops, announcing conclusions without reasoning behind everything. Not quite acceptable but that's about the best thing they can do.

          Comment


          • #6
            they claim a performance comparable to the first Ryzen generation, but probably that is with toolchains optimized for the arch, and that means a ecosystem prepared for it..
            Which is not available outside the company, for the general purpose distros..
            So probably we will never see a distro optimized for this arch..
            Last edited by tuxd3v; 13 October 2021, 09:56 AM. Reason: bugs

            Comment


            • #7
              Originally posted by xen0n View Post
              Actually it's more of a Frankenstein with bits of MIPS and bits of RISC-V, and totally different encoding of course. For example it doesn't have any notion of delay slots; its long jump ("j") is PC-relative instead of PC-regional like the MIPS j; its "pcaddu12i" is actually the same as RISC-V auipc, "jirl" the same as RISC-V jalr, etc etc etc.
              Well, ARMv7 has different instruction encodings, but they don't use a completely different top-level ISA, do they?

              I think that if something like >= 90% of the Loongarch path is identical to MIPS, then maybe a compromise can be struct by renaming the kernel's MIPS ISA to "MIPS/Loongarch"?

              Originally posted by xen0n View Post
              IMHO the way this whole Loongson thing got its present reputation is at least partly due to language barrier. Most people I know that are familiar with Loongson systems/people can only barely express in English;
              Interesting insight, but still a bit surprising when the Kernel's official language is English. I'm sure you need to have pretty good English comprehension to understand the kernel's docs, comments, and even variable & function names.

              I've worked with developers in China with very workable English proficiency. I know it's hard to learn & maintain good language proficiency without living in a country where it's commonly spoken, but they do exist. I wonder if they could maybe hire someone with a background as a technical writer/translater, to act as liaison, because the communication issues are clearly costing them.

              Comment


              • #8
                Originally posted by coder View Post
                Well, ARMv7 has different instruction encodings, but they don't use a completely different top-level ISA, do they?

                I think that if something like >= 90% of the Loongarch path is identical to MIPS, then maybe a compromise can be struct by renaming the kernel's MIPS ISA to "MIPS/Loongarch"?


                Interesting insight, but still a bit surprising when the Kernel's official language is English. I'm sure you need to have pretty good English comprehension to understand the kernel's docs, comments, and even variable & function names.

                I've worked with developers in China with very workable English proficiency. I know it's hard to learn & maintain good language proficiency without living in a country where it's commonly spoken, but they do exist. I wonder if they could maybe hire someone with a background as a technical writer/translater, to act as liaison, because the communication issues are clearly costing them.
                Understanding written texts and communicating are two totally different beasts. I can attest to that in my country: we have a high score on the woldwide English proficiency index, but a lot of people can barely communicate in proper English, if at all, esp. in real life communication (i.e. face-to-face).

                Comment


                • #9
                  Originally posted by coder View Post
                  Well, ARMv7 has different instruction encodings, but they don't use a completely different top-level ISA, do they?

                  I think that if something like >= 90% of the Loongarch path is identical to MIPS, then maybe a compromise can be struct by renaming the kernel's MIPS ISA to "MIPS/Loongarch"?
                  LoongArch is not 90%+ identical to MIPS. This is a very different ISA, and things like "no delay slots" are not mere encoding differences.

                  ARM and ARM64 are considered entirely distinct ports by the Linux kernel. Take a look at the arch/ directory.

                  Comment


                  • #10
                    Originally posted by tuxd3v View Post
                    they claim a performance comparable to the first Ryzen generation, but probably that is with toolchains optimized for the arch, and that means a ecosystem prepared for it..
                    Which is not available outside the company, for the general purpose distros..
                    So probably we will never see a distro optimized for this arch..
                    last time I've heard they were roughly as performant as pre-ryzen cpus

                    Comment

                    Working...
                    X