microbenchmark results

By: dmcq (dmcq.delete@this.fano.co.uk), September 24, 2021 12:05 pm
Room: Moderated Discussions
Andrey (andrey.semashev.delete@this.gmail.com) on September 24, 2021 8:29 am wrote:
> -.- (blarg.delete@this.mailinator.com) on September 23, 2021 8:01 pm wrote:
> > Andrey (andrey.semashev.delete@this.gmail.com) on September 23, 2021 10:11 am wrote:
> > > SVE is a spec, it simply allows a variety of implementations. It doesn't mean each and
> > > every one of them, down to pathological ones, will be implemented. But the code written
> > > for SVE will work, even in pathological cases. My point above was that there's no reason
> > > to worry about pathological cases because they will actually not exist in hardware.
> >
> > If they will never exist in hardware, why even allow it in the spec?
>
> There may be valid reasons. The spec has to strike a balance between being overly restrictive and
> overly permissive. And it has a limited scope. As I said before, vector size is not chosen in isolation,
> it should be aligned with other subsystems, which SVE doesn't define. Should SVE prohibit 136-bit
> vectors? No, because your cache line and page sizes may be a multiple of 17. In such an implementation,
> 136-bit vectors would be a reasonable choice. But in the power-of-2 world we live in, 136-bit vectors
> are nonsensical (possibly, aside from some very special purpose hardware).
>
> > > No, I'm not suggesting to hardcode a set of supported widths. However, the code could take into
> > > account the native vector length, including to achieve alignment - if this proves to be useful for
> > > performance on a given set of hardware implementations. As to how you would do this - there are
> > > many ways, starting from testing for a given set of CPU models that are known to benefit from this,
> > > and ending with all sorts of heuristics (e.g. check if the cache line size is a multiple of the
> > > native vector size).
> >
> > That sounds very much like hard-coding for specific cases - whether you base it purely from vector
> > width, or take other aspects into consideration. Which I think is fine, but it isn't and doesn't
> > really suggest a generic approach - i.e. to use your example, what to do if the cache line size is
> > not a multiple of the native vector size (especially when no such hardware currently exists)?
>
> By default - nothing. I mean, so long as you're targeting an unknown abstract implementation,
> you may as well forget about alignment, instruction choices and scheduling and other micro-optimizations
> and just write the unaligned loop that maintains correctness.
>
> When you don't know the hardware, but can reasonably guess its properties (e.g. because
> literally every CPU on the planet behaves the way you expect), you can apply the optimizations
> that are considered generally useful, like aligning your memory accesses.
>
> When the hardware is weird, and you want to optimize for it, then you have to get to know it. You
> would read its optimization manuals and benchmark to see what makes it tick, and choose your optimizations
> accordingly. This includes weird cases like cache line not being a multiple of the native vector
> size - in this case you would test how the hardware behaves wrt. data alignment.
>
> > > I mean, take x86 for example. There were times when unaligned vector memory accesses were so slow that
> > > programmers had to go an extra mile to just avoid movdqu.
> > > And they still do in performance critical routines.
> > > There were CPUs that were very slow with certain instructions (e.g. pshufb in some Atoms and older AMD
> > > processors), but still supported them. Programs using those instructions were still correct, and possibly
> > > even faster than their scalar equivalents, but still optimizing the code in account for such hardware
> > > quirks allowed to improve performance, and therefore considered a useful thing to do.
> >
> > x86 has the benefit of CPUID being able to disclose the CPU model.
> > ARM doesn't really have any nice way of determining
> > the model. And even if there was, heterogeneous core setups
> > are much more common on ARM, often with substantially
> > different core types, which complicates code which specifically tries to target one uarch.
> >
> > You can of course still try this type of optimisation with ARM, but, along with the diversity in the
> > ARM ecosystem, the general impression I get is that trying these sorts of optimisations (along with
> > trying to attain alignment) are much more difficult to pull off and often just isn't worth it.
>
> Yes, heterogeneous cores are a pain, and it looks like x86 will follow suit in the near future. But
> optimization is a two way road. As much as software wants to run fast on existing hardware, the future
> hardware wants to run the existing software as fast as possible, too. So, having a CPU with half of
> the cores having 128-bit vectors and the other half - 136-bit ones would be a terrible idea because
> it would perform worse on the existing code (more precisely, half of the cores would perform worse).
> There is no sense in making such a CPU as opposed to a more traditional design with power-of-2 sizes.
> That is unless there is a very strong reason to have specifically 136-bit vectors, and even then a sane
> CPU designer would make everything possible to make sure 128-bit vectors still work fast.
>
> So, while different cores may be different, I would not expect them to be incompatibly different. Yes, I'm
> aware of some ARM CPUs that have different cache line sizes, but (a) they are multiple of one another (i.e.
> not quite incompatible) and (b) that already proved to be a PITA, so probably was not such a good idea.
>
> Side note about hybrid designs where different cores are radically different, like x86+ARM that AMD did.
> That sort of combination is a somewhat different story. The cores in such a hybrid are inherently incompatible,
> but that's not a problem. A thread that is running on one kind of core could never run on the other kind
> of core because of the different instruction sets. But even then, both kinds of cores still interface the
> same system memory and IO, possibly via the same hardware blocks in the CPU, so they are not completely unrelated.
> So, if x86 cores follow the power-of-2 design, ARM cores will most likely have to as well.

SVE is in multiples of 128 bits so not so bad! I' guess the first hetrogenous system with a size greater than 128 bits will be an Apple one and I guess they'l go for having the same size in both, perhaps they'll share an SVE unit amongst the small cores like ARM. But they haven't even announced a system with SVE yet.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Armv8.8-A and Armv9.3-Aanonymou52021/09/16 02:25 PM
  Armv8.8-A and Armv9.3-ADoug S2021/09/16 09:57 PM
    Armv8.8-A and Armv9.3-ABrett2021/09/16 10:32 PM
      Armv8.8-A and Armv9.3-Aanon2021/09/16 10:55 PM
        Armv8.8-A and Armv9.3-Anone2021/09/16 11:51 PM
      Armv8.8-A and Armv9.3-AJörn Engel2021/09/17 04:42 AM
        Armv8.8-A and Armv9.3-AMichael S2021/09/17 06:48 AM
          Armv8.8-A and Armv9.3-AJörn Engel2021/09/18 01:01 PM
            Armv8.8-A and Armv9.3-AMichael S2021/09/18 02:58 PM
              microbenchmark resultsMichael S2021/09/19 03:46 PM
                microbenchmark source codeMichael S2021/09/19 03:58 PM
                  microbenchmark source code-.-2021/09/20 03:49 PM
                    microbenchmark source codeMichael S2021/09/21 09:17 AM
                      microbenchmark source code-.-2021/09/21 03:33 PM
                        microbenchmark source codeMichael S2021/09/21 05:05 PM
                microbenchmark resultsAnon2021/09/19 04:32 PM
                  microbenchmark resultsJörn Engel2021/09/19 07:46 PM
                    microbenchmark resultsdmcq2021/09/20 01:19 AM
                      microbenchmark resultsMichael S2021/09/20 04:12 AM
                      microbenchmark results-.-2021/09/20 03:44 PM
                        microbenchmark resultsMichael S2021/09/21 09:23 AM
                          microbenchmark results-.-2021/09/21 03:35 PM
                            microbenchmark resultsAndrey2021/09/21 04:25 PM
                              I agree (NT)Michael S2021/09/21 05:07 PM
                              microbenchmark results-.-2021/09/22 04:56 PM
                                microbenchmark resultsMichael S2021/09/23 05:11 AM
                                  microbenchmark resultsdmcq2021/09/23 06:53 AM
                                  microbenchmark resultsAndrey2021/09/23 09:20 AM
                                microbenchmark resultsAndrey2021/09/23 09:11 AM
                                  microbenchmark results-.-2021/09/23 07:01 PM
                                    microbenchmark resultsSimon Farnsworth2021/09/24 01:47 AM
                                      microbenchmark results-.-2021/09/24 05:00 PM
                                    microbenchmark resultsAndrey2021/09/24 07:29 AM
                                      microbenchmark resultsdmcq2021/09/24 12:05 PM
                                        microbenchmark resultsDoug S2021/09/24 01:12 PM
                                          microbenchmark results---2021/09/24 06:06 PM
                                            microbenchmark resultsDoug S2021/09/24 10:46 PM
                                              microbenchmark results---2021/09/25 08:56 AM
                                                microbenchmark resultsJukka Larja2021/09/26 01:01 AM
                                                microbenchmark resultsDoug S2021/09/26 08:41 AM
                                                  microbenchmark resultsdmcq2021/09/26 12:37 PM
                                                    microbenchmark resultsDoug S2021/09/27 09:32 AM
                                                      microbenchmark resultsdmcq2021/09/28 06:56 AM
                                              microbenchmark resultsDummond D. Slow2021/09/25 11:49 AM
                                                microbenchmark resultsBrett2021/09/25 02:31 PM
                                              microbenchmark resultsdmcq2021/09/25 11:51 AM
                                                microbenchmark resultsDoug S2021/09/26 08:45 AM
                                            microbenchmark resultsRichard S2021/09/25 12:51 AM
                                              microbenchmark resultsDummond D. Slow2021/09/25 11:52 AM
                                                microbenchmark results---2021/09/25 02:04 PM
                                      SVE alignment with non power-of-2 widths-.-2021/09/24 05:10 PM
                                        SVE alignment with non power-of-2 widthsAndrey2021/09/25 03:46 AM
                                          SVE alignment with non power-of-2 widths-.-2021/09/25 04:35 PM
                                          SVE alignment with non power-of-2 widthsKevin G2021/09/27 08:46 AM
                                            SVE alignment with non power-of-2 widths-.-2021/09/27 08:06 PM
                                              SVE alignment with non power-of-2 widthsJukka Larja2021/09/28 05:37 AM
                                                SVE alignment with non power-of-2 widthsAndrey2021/09/28 11:12 AM
                                                  SVE alignment with non power-of-2 widthsdmcq2021/09/28 01:29 PM
                                                SVE alignment with non power-of-2 widths-.-2021/09/28 05:37 PM
                                                  SVE alignment with non power-of-2 widthsJukka Larja2021/09/29 05:50 AM
                    microbenchmark results---2021/09/20 06:11 AM
                    microbenchmark resultsJörn Engel2021/09/23 04:10 AM
                      microbenchmark resultsMichael S2021/09/23 04:55 AM
                        microbenchmark resultsJörn Engel2021/09/23 08:24 AM
                          microbenchmark resultsRoyi2021/09/26 03:25 PM
                      microbenchmark resultsdmcq2021/09/23 09:42 AM
                        microbenchmark results---2021/09/23 10:53 AM
                      microbenchmark resultsanon22021/09/23 01:40 PM
                microbenchmark results: Zen 3Adrian2021/09/22 12:57 AM
                  microbenchmark results: Zen 3Adrian2021/09/22 01:08 AM
                    microbenchmark results: Zen 3Michael S2021/09/22 04:48 AM
                      microbenchmark results: Zen 3Adrian2021/09/22 05:05 AM
        Armv8.8-A and Armv9.3-AKonrad Schwarz2021/09/28 04:45 AM
    Armv8.8-A and Armv9.3-ALinus Torvalds2021/09/17 07:59 AM
      Armv8.8-A and Armv9.3-ADoug S2021/09/17 10:35 AM
        Armv8.8-A and Armv9.3-Anksingh2021/09/17 11:23 AM
          Armv8.8-A and Armv9.3-ADoug S2021/09/17 01:35 PM
            Armv8.8-A and Armv9.3-AKonrad Schwarz2021/10/15 05:23 AM
              Armv8.8-A and Armv9.3-Arwessel2021/10/15 05:49 AM
        Armv8.8-A and Armv9.3-AAdrian2021/09/17 10:07 PM
          Armv8.8-A and Armv9.3-ADoug S2021/09/18 06:34 AM
            Armv8.8-A and Armv9.3-AAdrian2021/09/18 06:38 AM
      Armv8.8-A and Armv9.3-Ablaine2021/09/18 09:37 AM
      Armv8.8-A and Armv9.3-ABrett2021/09/19 12:06 PM
        Armv8.8-A and Armv9.3-Admcq2021/09/19 12:36 PM
        Armv8.8-A and Armv9.3-ADoug S2021/09/19 05:07 PM
          Armv8.8-A and Armv9.3-A - movesdmcq2021/09/28 08:54 AM
            Armv8.8-A and Armv9.3-A - movesDoug S2021/09/28 12:57 PM
              Armv8.8-A and Armv9.3-A - movesdmcq2021/09/28 01:21 PM
                Armv8.8-A and Armv9.3-A - movesNoSpammer2021/09/29 02:53 AM
                  Armv8.8-A and Armv9.3-A - movesrwessel2021/09/29 05:55 AM
                    Armv8.8-A and Armv9.3-A - movesdmcq2021/09/29 06:53 AM
                      Armv8.8-A and Armv9.3-A - movesrwessel2021/09/29 10:35 AM
                        Armv8.8-A and Armv9.3-A - movesdmcq2021/09/29 12:44 PM
                          Armv8.8-A and Armv9.3-A - movesrwessel2021/09/29 12:58 PM
                            Armv8.8-A and Armv9.3-A - movesdmcq2021/09/29 02:52 PM
                              Armv8.8-A and Armv9.3-A - movesrwessel2021/09/29 05:36 PM
                              Armv8.8-A and Armv9.3-A - movesAndrey2021/09/29 06:58 PM
                    Armv8.8-A and Armv9.3-A - movesDoug S2021/09/29 09:10 AM
                      Armv8.8-A and Armv9.3-A - movesrwessel2021/09/29 10:30 AM
                        Armv8.8-A and Armv9.3-A - movesDoug S2021/09/29 09:02 PM
                          Armv8.8-A and Armv9.3-A - movesrwessel2021/09/29 10:22 PM
                            Armv8.8-A and Armv9.3-A - movesMark Roulo2021/09/30 06:37 AM
                              Armv8.8-A and Armv9.3-A - movesrwessel2021/09/30 07:02 AM
                                Did they publish a full description? (NT)Michael S2021/09/30 07:12 AM
                                  Did they publish a full description?rwessel2021/09/30 08:18 AM
                                    Did they publish a full description?Michael S2021/09/30 09:24 AM
                                      Did they publish a full description?rwessel2021/09/30 09:42 AM
                                    Did they publish a full description?Adrian2021/09/30 11:22 PM
                                  Do we even okiw it's three instructions per move?Carson2021/09/30 09:28 PM
                                    Do we even okiw it's three instructions per move?Adrian2021/09/30 11:27 PM
                                    Do we even okiw it's three instructions per move?rwessel2021/10/01 03:19 AM
                            Armv8.8-A and Armv9.3-A - movesDoug S2021/09/30 08:48 AM
                              Armv8.8-A and Armv9.3-A - movesrwessel2021/09/30 09:39 AM
                                Armv8.8-A and Armv9.3-A - movesDoug S2021/09/30 01:56 PM
                                  Armv8.8-A and Armv9.3-A - movesrwessel2021/09/30 04:20 PM
                                    Armv8.8-A and Armv9.3-A - movesdmcq2021/10/01 03:38 AM
                                      Armv8.8-A and Armv9.3-A - movesMichael S2021/10/01 04:04 AM
                                        Armv8.8-A and Armv9.3-A - movesLinus Torvalds2021/10/01 10:01 AM
                                          memcpy - instruction cracking vs DMArpg2021/10/02 01:51 AM
                                            memcpy - instruction cracking vs DMAAdrian2021/10/02 02:45 AM
                                              memcpy - instruction cracking vs DMADoug S2021/10/02 08:47 AM
                                                memcpy - instruction cracking vs DMAAdrian2021/10/02 09:15 AM
                                                memcpy - instruction cracking vs DMArwessel2021/10/02 10:37 AM
                                                  memcpy - instruction cracking vs DMADoug S2021/10/02 05:49 PM
                                            memcpy - instruction cracking vs DMALinus Torvalds2021/10/02 09:43 AM
                                              memcpy - instruction cracking vs DMAdmcq2021/10/02 10:32 AM
                                              memcpy - instruction cracking vs DMABrett2021/10/02 10:45 AM
                                              memcpy - instruction cracking vs DMA---2021/10/02 02:03 PM
                                                memcpy - instruction cracking vs DMA---2021/10/02 02:12 PM
                                                  Moving copy to DRAM doesn't help for small copiesMark Roulo2021/10/02 02:59 PM
                                                    Moving copy to DRAM doesn't help for small copies---2021/10/02 06:32 PM
                                                      Moving copy to DRAM doesn't help for small copiesMichael S2021/10/03 12:40 AM
                                                        Moving copy to DRAM doesn't help for small copiesDoug S2021/10/03 09:09 AM
                                                          Moving copy to DRAM doesn't help for small copiesrwessel2021/10/03 09:51 AM
                                                          Moving copy to DRAM doesn't help for small copiesLinus Torvalds2021/10/03 10:09 AM
                                                            How about environments such as Java?Mark Roulo2021/10/03 11:41 AM
                                                              How about environments such as Java?rwessel2021/10/03 11:49 AM
                                                                How about environments such as Java?Mark Roulo2021/10/03 12:22 PM
                                                              How about environments such as Java?anon22021/10/03 06:58 PM
                                                                How about environments such as Java?Etienne Lorrain2021/10/04 04:08 AM
                                                                  Apart from "It depends" there is no short answer. (NT)Michael S2021/10/04 04:30 AM
                                                                  How about environments such as Java?Andrey2021/10/04 05:04 AM
                                                                  How about environments such as Java?anon22021/10/04 05:32 AM
                                                                How about environments such as Java?Mark Roulo2021/10/04 06:31 AM
                                                                How about environments such as Java?---2021/10/04 08:41 AM
                                                                  How about environments such as Java?Doug S2021/10/04 09:23 AM
                                                                    How about environments such as Java?Andrey2021/10/04 11:14 AM
                                                                      How about environments such as Java?Doug S2021/10/04 12:20 PM
                                                                  How about environments such as Java?anon22021/10/04 01:23 PM
                                                                  How about environments such as Java?rwessel2021/10/04 03:54 PM
                                                            Moving copy to DRAM doesn't help for small copiesJörn Engel2021/10/04 04:52 AM
                                                            Early software zeroing !=== early hardware zeroingPaul A. Clayton2021/10/05 10:19 AM
                                                              Early software zeroing !=== early hardware zeroingDoug S2021/10/05 11:21 AM
                                                memcpy - instruction cracking vs DMABrendan2021/10/02 03:53 PM
                                                  memcpy - instruction cracking vs DMALinus Torvalds2021/10/03 09:48 AM
                                                    memcpy - instruction cracking vs DMAdmcq2021/10/03 12:54 PM
                                              memcpy - instruction cracking vs DMAYuhong Bao2021/10/03 12:30 AM
                                                memcpy - instruction cracking vs DMADavid Hess2021/10/05 04:19 PM
                                                  memcpy - instruction cracking vs DMAAdrian2021/10/05 10:28 PM
                                                    memcpy - instruction cracking vs DMAEtienne Lorrain2021/10/06 01:24 AM
                                                    memcpy - instruction cracking vs DMArwessel2021/10/06 02:38 AM
                                                      memcpy - instruction cracking vs DMAAdrian2021/10/06 03:04 AM
                                                        memcpy - instruction cracking vs DMArwessel2021/10/06 04:59 AM
                                                    memcpy - instruction cracking vs DMA---2021/10/06 08:07 AM
                                                      memcpy - instruction cracking vs DMAAndrey2021/10/06 01:59 PM
                                                    memcpy - instruction cracking vs DMAgallier22021/10/06 10:06 PM
                                                      memcpy - instruction cracking vs DMAAdrian2021/10/06 10:59 PM
                                              memcpy - instruction cracking vs DMAMichael S2021/10/03 12:51 AM
                                                memcpy - instruction cracking vs DMArwessel2021/10/03 04:06 AM
                                                  memcpy - instruction cracking vs DMAMichael S2021/10/03 04:24 AM
                                                    memcpy - instruction cracking vs DMAMatt Sayler2021/10/03 07:02 AM
                                                    memcpy - instruction cracking vs DMADoug S2021/10/03 09:14 AM
                                      Armv8.8-A and Armv9.3-A - movesrwessel2021/10/01 04:10 AM
                                        Armv8.8-A and Armv9.3-A - movesEtienne Lorrain2021/10/01 06:55 AM
                                          Armv8.8-A and Armv9.3-A - movesrwessel2021/10/01 07:14 AM
                                            Armv8.8-A and Armv9.3-A - movesDoug S2021/10/01 10:17 AM
                                              Armv8.8-A and Armv9.3-A - movesrwessel2021/10/02 03:57 AM
  Armv8.8-A and Armv9.3-Anone2021/10/13 05:06 AM
    Armv8.8-A and Armv9.3-AAdrian2021/10/13 05:22 AM
      Armv8.8-A and Armv9.3-ADoug S2021/10/13 08:01 AM
        Armv8.8-A and Armv9.3-Admcq2021/10/13 09:17 AM
          Armv8.8-A and Armv9.3-Anone2021/10/13 09:26 PM
            Armv8.8-A and Armv9.3-Admcq2021/10/14 07:22 AM
    Armv8.8-A and Armv9.3-Arwessel2021/10/14 08:01 AM
      Armv8.8-A and Armv9.3-AAnon2021/10/14 10:08 AM
        Armv8.8-A and Armv9.3-AMichael S2021/10/14 12:25 PM
      Armv8.8-A and Armv9.3-ADoug S2021/10/14 10:18 AM
        Armv8.8-A and Armv9.3-Arwessel2021/10/14 06:07 PM
          Armv8.8-A and Armv9.3-ADoug S2021/10/14 09:23 PM
            Armv8.8-A and Armv9.3-Admcq2021/10/15 12:41 AM
              Armv8.8-A and Armv9.3-AGabriele Svelto2021/10/15 04:07 AM
            Armv8.8-A and Armv9.3-Arwessel2021/10/15 03:49 AM
              Armv8.8-A and Armv9.3-ADoug S2021/10/15 09:44 AM
                Armv8.8-A and Armv9.3-Ame2021/10/15 05:34 PM
                  Armv8.8-A and Armv9.3-ADoug S2021/10/16 08:47 AM
                    Armv8.8-A and Armv9.3-Ame2021/10/17 04:19 AM
                      Armv8.8-A and Armv9.3-ADoug S2021/10/17 09:17 AM
                        Armv8.8-A and Armv9.3-Ame2021/10/17 11:31 AM
                          Armv8.8-A and Armv9.3-ADoug S2021/10/17 12:33 PM
                            Armv8.8-A and Armv9.3-AzArchJon2021/10/18 09:35 AM
                              Armv8.8-A and Armv9.3-ADoug S2021/10/18 01:35 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?