Announcement

Collapse
No announcement yet.

Apple M2 Support Added To Upstream LLVM Along With The A15, A16

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

  • #11
    What I find shocking is that The A16 is still only ARMv8-A.

    I guess there's no reason Apple has to move to ARMv9-A until it's good & ready, but they're probably going to start missing out on some SVE2 optimizations, if they drag their feet too much longer. As much as Apple controls its own software ecosystem, app developers surely use a fair amount of open source libs that'll begin to gain SVE2 codepaths, as ARMv9 becomes the default for new Android devices and Windows-ARM laptops.

    In the server market, Graviton 3 is probably the last big server CPU not to use ARMv9, but even it has SVE1.

    Comment


    • #12
      Originally posted by coder View Post
      What I find shocking is that The A16 is still only ARMv8-A.

      I guess there's no reason Apple has to move to ARMv9-A until it's good & ready, but they're probably going to start missing out on some SVE2 optimizations, if they drag their feet too much longer. As much as Apple controls its own software ecosystem, app developers surely use a fair amount of open source libs that'll begin to gain SVE2 codepaths, as ARMv9 becomes the default for new Android devices and Windows-ARM laptops.

      In the server market, Graviton 3 is probably the last big server CPU not to use ARMv9, but even it has SVE1.
      Apple doesn't come with things first. They wait for others and then perfect it (and present as they invented it ). Currently no other available ARM processor for desktop or mobile has more than 128bit. Even in x86 Intel is struggling at offering 512bit.

      Comment


      • #13
        Originally posted by Ladis View Post
        Currently no other available ARM processor for desktop or mobile has more than 128bit. Even in x86 Intel is struggling at offering 512bit.
        At some point, we'll see code optimized with SVE2 instructions and not Neon. At that point, it won't be a question of vector width, but one of scalar vs. vector!

        Also, ARM has been keen to point out that SVE2 offers intrinsic benefits over Neon, even when implemented at the same width. Look at the left side of this graph. Both N1 and N2 are 2x 128-bit.



        "Arm here discloses that the performance advantages on auto-vectorizable code can be quite significant. In a 2x128b comparison between the N1 and the N2, we can see around 40th-percentile gains of at least 20% of performance, with some code reaching even much higher gains of up to +90%.​"

        Source: https://www.anandtech.com/show/16640...-cmn700-mesh/5

        Comment


        • #14
          Originally posted by coder View Post
          At some point, we'll see code optimized with SVE2 instructions and not Neon. At that point, it won't be a question of vector width, but one of scalar vs. vector!
          At that time, Apple will use RISC-V and x86 will be dead

          Comment


          • #15
            Originally posted by Ladis View Post
            At that time, Apple will use RISC-V and x86 will be dead
            You jest, but it's been no small undertaking for Apple to move its MacOS ecosystem over to ARM (and they're not even done, yet -- there's still the x86-based Power Mac to replace!), and that's building on the momentum of 1.5 decades of using ARM in phones, tablets, and iPods. So, there's basically no way Apple is moving off ARM. They'd sooner buy ARM than switch ISAs, within at least the next decade.

            Someday, I believe even AArch64 will become obsolete. However, the switch would be for technological reasons, and therefore definitely not to a roughly comparable ISA like RISC-V.

            Comment


            • #16
              Originally posted by coder View Post
              You jest, but it's been no small undertaking for Apple to move its MacOS ecosystem over to ARM (and they're not even done, yet -- there's still the x86-based Power Mac to replace!), and that's building on the momentum of 1.5 decades of using ARM in phones, tablets, and iPods. So, there's basically no way Apple is moving off ARM. They'd sooner buy ARM than switch ISAs, within at least the next decade.

              Someday, I believe even AArch64 will become obsolete. However, the switch would be for technological reasons, and therefore definitely not to a roughly comparable ISA like RISC-V.
              Nobody uses Mac Pros. Those are extremely overpriced and the hardware inside is very soon obsolete (Mac Pros come always the last in the line). I know companies, who buy the cheapest Mac Pros and don't even unbox them. Instead, they build Hackintosh (as they depend on macOS software). Fun fact: M1 Mac Pro was prepared, but at the end Apple wasn't sure people will accept switching to ARM so extremely well. And releasing it too late, when M2 was around the corner, didn't make much sense.

              PS: Apple makes own CPU (and GPU) cores. They don't depend on a specific ISA. Also their userbase is the most accepting of changes.

              PPS: The switch can be also for money reasons

              Comment


              • #17
                Originally posted by Ladis View Post
                Nobody uses Mac Pros. Those are extremely overpriced and the hardware inside is very soon obsolete (Mac Pros come always the last in the line).
                My point was just that their transition is incomplete until those are replaced, as well. And they will be, because it's the only Mac which can accommodate large amounts of RAM, a significant amount of storage, or PCIe cards. And they have deep-pocketed professional users who require those things.

                Originally posted by Ladis View Post
                I know companies, who buy the cheapest Mac Pros and don't even unbox them. Instead, they build Hackintosh (as they depend on macOS software).
                That will probably get a lot harder, when those companies want to start running ARM versions of the software.

                Originally posted by Ladis View Post
                Fun fact: M1 Mac Pro was prepared, but at the end Apple wasn't sure people will accept switching to ARM so extremely well. And releasing it too late, when M2 was around the corner, didn't make much sense.
                That doesn't any make sense. I won't bother asking you for a source, because the claim doesn't withstand even superficial analysis.

                Originally posted by Ladis View Post
                PS: Apple makes own CPU (and GPU) cores. They don't depend on a specific ISA.
                Okay, thanks for letting me know who I'm dealing with. I won't waste any further time on you.

                Comment


                • #18
                  Originally posted by coder View Post
                  That will probably get a lot harder, when those companies want to start running ARM versions of the software.
                  What software will have only ARM version?

                  Originally posted by coder View Post
                  Okay, thanks for letting me know who I'm dealing with. I won't waste any further time on you.
                  You probably didn't understand. Apple makes own cores and the ISA is only the instruction decoder on the hardware side and compiler + development environment + emulator (Rosetta) on the software side. Apple jumps ISAs like no one other.

                  Comment


                  • #19
                    Originally posted by coder View Post
                    What I find shocking is that The A16 is still only ARMv8-A.

                    I guess there's no reason Apple has to move to ARMv9-A until it's good & ready
                    And likely some good reasons not to move. While the details of the license Apple has with ARM is under NDA, an architecture license (which is what Apple has been presumed to have), does not allow use of a different variant without payments of additional fees. There are no (currently named) companies with an architecture license for ARMv9 (not that we would necessarily know if there are any).

                    Comment


                    • #20
                      Originally posted by rene View Post

                      does not exactly require a genius adding some cpu names and their publicly known supported ISA set to an processor model table, ...
                      While just adding a name is easy, there are also scheduling/tuning/feature values, and those are (primarily) known only to the chip designer. In this case they had to use some specific settings to enable features (as the commit comment explicitly mentions).

                      Comment

                      Working...
                      X