x86 and ARM

By: Wilco (wilco.dijkstra.delete@this.ntlworld.com), November 27, 2020 3:46 am
Room: Moderated Discussions
none (none.delete@this.none.com) on November 27, 2020 12:07 am wrote:
> Chester (lamchester.delete@this.gmail.com) on November 26, 2020 6:38 pm wrote:
> [...]
> > Has anything changed with aarch64 that invalidates Henry's statement?
>
> Predication of all instructions has disappeared. Multiple load/store are no more (except
> for some SIMD instructions and load/sotre pair). The shift options to ALU operations have
> been vastly reduced (both in number of options, and in number of instructions to which it
> can be applied).

Actually AArch64 supports more shift+ALU instructions than Thumb-2. There is shift+add, shift+logical, extend+shift+add, shift+mask. They typically execute in a single cycle on modern cores (including shift+add), so it is wrong to call them "difficult to implement".

Predication and load/store multiple are also fast on OoO cores - it's just not a good idea to waste precious ISA encoding space to predication when not many instructions are predicated in modern code. As branch prediction becomes better, there is less benefit from predication (on early Arm it was a no brainer, you could predicate 3 instructions and still be faster than a branch).

> AArch64 is much cleaner than the older ARM 32-bit instruction set. But that doesn't make it
> a very "RISC" ISA. It's just a modern instruction set which IMHO is miles ahead of x86-64.

Yes, the purist view of RISC never made any sense - you do need powerful instructions that speed up your code.

> I nonetheless agree with David and others: with sufficient efforts and resources most ISAs
> should have similar performance for high-end chips. The only caveat is that if you have to
> dedicate too many resources to get around ISA complexities, you won't dedicate that HW to
> some other feature that might improve performance.

It's not just the extra design effort, but also the extra transistors and power. And these are very significant, just consider how many high-end CPUs Arm designs a year or how little power M1 uses while beating the fastest x86 CPUs.

So the argument that it's all the same at the high-end is obviously wrong.

Wilco
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon2020/11/19 07:17 PM
  David Kanter RWT article: "Why Apple Won’t ARM the MacBook"tarlinian2020/11/19 07:21 PM
  The article was spot on. Not possible in 2012-2014. Could become possible around 2016-2021.Jouni Osmala2020/11/19 10:29 PM
    The article was spot on. Not possible in 2012-2014. Could become possible around 2016-2021.anon2020/11/20 02:25 PM
  You're intellectually bankruptDaniel B2020/11/20 02:51 AM
    You're intellectually bankruptMichael S2020/11/20 05:22 AM
  David Kanter RWT article: "Why Apple Won’t ARM the MacBook"NaNon2020/11/20 04:08 AM
  He was right. ARM MBs weren't viable until 2019-2020, 8 years later (NT)JRB2020/11/20 08:34 AM
  David Kanter RWT article: "Why Apple Won’t ARM the MacBook"Rayla2020/11/20 09:37 AM
    David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon2020/11/20 02:26 PM
  David Kanter RWT article: "Why Apple Won’t ARM the MacBook"Jon Masters2020/11/20 10:24 AM
    David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon2020/11/20 02:31 PM
      David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon22020/11/20 04:57 PM
        David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon2020/11/20 09:29 PM
          David Kanter RWT article: "Why Apple Won’t ARM the MacBook"Jose2020/11/21 08:48 AM
    David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon22020/11/20 04:54 PM
      David Kanter RWT article: "Why Apple Won’t ARM the MacBook"Jan Vlietinck2020/11/20 11:23 PM
        David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon22020/11/21 12:41 AM
          David Kanter RWT article: "Why Apple Won’t ARM the MacBook"Jan Vlietinck2020/11/21 01:52 AM
            David Kanter RWT article: "Why Apple Won’t ARM the MacBook"anon22020/11/21 03:12 AM
              David Kanter RWT article: "Why Apple Won’t ARM the MacBook"Jan Vlietinck2020/11/21 03:25 AM
        Happy with my historical record :)David Kanter2020/11/21 09:30 AM
          Happy with my historical record :)Per Hesselgren2020/11/21 01:09 PM
            Happy with my historical record :)Maynard Handley2020/11/21 02:02 PM
              Happy with my historical record :)anon2020/11/22 12:23 AM
                Happy with my historical record :)anon22020/11/22 01:13 AM
                  Happy with my historical record :)Michael S2020/11/22 03:12 AM
                    Happy with my historical record :)anon2020/11/22 10:53 AM
                    Swift is 3-wide + custom layout (NT)vvid2020/11/23 04:59 AM
                      Swift is 3-wide + custom layoutanon2020/11/23 02:14 PM
                        No.Heikki Kultala2020/11/23 03:40 PM
                Happy with my historical record :)rwessel2020/11/22 09:14 AM
            Happy with my historical record :)David Kanter2020/11/21 02:51 PM
              Happy with my historical record :)Per Hesselgren2020/11/22 12:28 AM
                Happy with my historical record :)Dummond D. Slow2020/11/22 12:39 PM
                  Is ARM32 native in ARM64?Per Hesselgren2020/11/22 10:34 PM
                    Is ARM32 native in ARM64?Per Hesselgren2020/11/22 10:47 PM
                    Is ARM32 native in ARM64?vvid2020/11/23 06:48 AM
                      Is ARM32 native in ARM64?Per Hesselgren2020/11/23 07:15 AM
                        Is ARM32 native in ARM64?vvid2020/11/23 12:24 PM
                      Is ARM32 native in ARM64?Wilco2020/11/23 11:10 AM
                        Is ARM32 native in ARM64?vvid2020/11/23 01:21 PM
                          Is ARM32 native in ARM64?Per Hesselgren2020/11/23 01:54 PM
          Happy with my historical record :)anon2020/11/21 05:09 PM
            Happy with my historical record :)David Kanter2020/11/21 08:12 PM
          Happy with my historical record :)Emil Briggs2020/11/22 05:11 AM
          Happy with my historical record :)juanrga2020/11/22 08:22 AM
            Happy with my historical record :)Wilco2020/11/22 10:34 AM
              Happy with my historical record :)Dummond D. Slow2020/11/22 12:26 PM
                Happy with my historical record :)Dummond D. Slow2020/11/22 12:42 PM
          Happy with my historical record :) - but should you be?anon2020/11/22 02:35 PM
            Happy with my historical record :) - but should you be?anon.12020/11/22 03:42 PM
              Happy with my historical record :) - but should you be?Foo_2020/11/23 01:14 AM
                Happy with my historical record :) - but should you be?anon.12020/11/23 02:19 PM
            Happy with my historical record :) - but should you be?anon22020/11/22 04:24 PM
            Happy with my historical record :) - but should you be?David Kanter2020/11/22 11:01 PM
              Process technologyDavid Kanter2020/11/23 07:51 AM
                Process technologyDummond D. Slow2020/11/23 08:56 AM
                  Process technologyWilco2020/11/23 11:59 AM
                Process technologyDoug S2020/11/23 12:08 PM
                  Process technologyDummond D. Slow2020/11/23 12:24 PM
                  Process technologyDavid Kanter2020/11/23 05:17 PM
                    Process technologyDoug S2020/11/24 10:09 AM
                      Process technologyDavid Kanter2020/11/24 06:54 PM
              x86 and ARMChester2020/11/26 10:29 AM
                x86 and ARMMaynard Handley2020/11/26 01:05 PM
                x86 and ARMdmcq2020/11/26 02:31 PM
                  x86 and ARMChester2020/11/26 06:38 PM
                    x86 and ARMnone2020/11/27 12:07 AM
                      x86 and ARMJames2020/11/27 01:51 AM
                        x86 and ARMChester2020/11/27 05:22 PM
                          x86 and ARMMark Roulo2020/11/27 07:18 PM
                            x86 and ARMJames2020/11/28 03:22 AM
                            x86 and ARMWilco2020/11/28 06:54 AM
                              x86 and ARMSimon Farnsworth2020/11/30 02:20 AM
                            x86 and ARMgallier22020/11/29 11:42 PM
                          x86 and ARMWilco2020/11/28 06:46 AM
                            x86 and ARMChester2020/11/28 10:28 PM
                              and you don't understand the difference between 1995 and 2001 it seems (NT)anon2020/11/28 11:24 PM
                                sure, early 2000s thenChester2020/11/29 06:31 AM
                                Yes he doesHeikki Kultala2020/11/29 04:24 PM
                                  so not just any random binary (NT)anon2020/11/29 10:08 PM
                            x86 compatibility with 16-bit modesRob Thorpe2020/11/30 06:29 PM
                      x86 and ARMWilco2020/11/27 03:46 AM
                        x86 and ARMChester2020/11/28 10:46 PM
                          x86 and ARMJan Vlietinck2020/11/29 12:21 AM
                            x86 and ARMjuanrga2020/11/29 04:44 AM
                              x86 and ARMChester2020/11/29 06:30 AM
                                I didn't write that ARM is 7 times more efficient than x86. Don't make things up (NT)juanrga2020/11/29 08:04 AM
                                x86 and ARMDoug S2020/11/29 09:26 AM
                                  x86 and ARMwumpus2020/11/29 10:06 AM
                                  x86 and ARMDummond D. Slow2020/11/29 11:59 AM
                                    x86 and ARMjuanrga2020/11/30 01:28 AM
                                      x86 and ARMAdrian2020/11/30 04:50 AM
                                        x86 and ARMJan Vlietinck2020/11/30 06:04 AM
                                          x86 and ARMjuanrga2020/12/01 04:09 AM
                                        x86 and ARMDoug S2020/11/30 11:50 AM
                                      x86 and ARManon.12020/11/30 05:27 AM
                                        x86 and ARManon2020/11/30 08:33 PM
                                          x86 and ARManon.12020/12/01 07:11 AM
                                            x86 and ARManon2020/12/01 02:28 PM
                                              x86 and ARManon.12020/12/01 06:44 PM
                                      x86 and ARMDummond D. Slow2020/11/30 08:04 AM
                                      x86 and ARMGionatan Danti2020/11/30 08:44 AM
                                        x86 and ARMFoo_2020/12/01 12:26 AM
                                          x86 and ARMGionatan Danti2020/12/01 10:01 AM
                                        x86 and ARMjuanrga2020/12/01 04:17 AM
                    x86 and ARMUngo2020/11/27 01:55 AM
                    x86 and ARMjuanrga2020/11/27 03:58 AM
              Happy with my historical record :) - but should you be?anon2020/11/30 02:30 PM
                Happy with my historical record :) - but should you be?David Kanter2020/11/30 04:15 PM
                Happy with my historical record :) - but should you be?anon22020/12/01 03:22 AM
                  Happy with my historical record :) - but should you be?anon2020/12/01 02:16 PM
                    Happy with my historical record :) - but should you be?Maynard Handley2020/12/01 02:38 PM
                      Happy with my historical record :) - but should you be?Gabriele Svelto2020/12/08 07:51 AM
          Happy with my historical record :)Jan Vlietinck2020/11/23 12:55 AM
            Happy with my historical record :)anonymou52020/11/23 03:44 AM
              Happy with my historical record :)Jan Vlietinck2020/11/23 06:16 AM
                Happy with my historical record :)David Kanter2020/11/23 07:18 AM
                  Happy with my historical record :)Jan Vlietinck2020/11/23 09:37 AM
                  Happy with my historical record :)Ungo2020/11/23 08:48 PM
                  Happy with my historical record :)Maynard Handley2020/11/23 09:01 PM
                    Happy with my historical record :)Dummond D. Slow2020/11/24 07:48 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?