#asahi-dev · 2022-04-20 03:14
YichaoYu[m]Sorry I guess that the hotel internet…. I’ll close my irc connection….#asahi-dev · 2022-04-18 11:00
YichaoYu[m]thx#asahi-dev · 2022-04-18 10:58
YichaoYu[m]= = ...... k did't know that...#asahi-dev · 2022-04-18 10:57
YichaoYu[m]I sent an email to linux-arm-kernel and cc'd you and everyone I found i maintainers that related to pmu#asahi-dev · 2022-04-18 10:56
YichaoYu[m]but as I said in the email, compared to other arm cores, there are indeed things that are still missing in the kernel code#asahi-dev · 2022-04-18 10:53
YichaoYu[m]But you told me to not do that since there's nothing to be done to the kernel#asahi-dev · 2022-04-17 21:16
YichaoYu[m]Still haven’t heard anything regarding how to upstream any information about the perf counters but at least the result from the test gives me a fairly reliable rr to use on m1 now….#asahi-dev · 2022-04-02 14:55
YichaoYu[m](and I assume for a single thread with no race it should still be deterministic......)#asahi-dev · 2022-04-02 14:54
YichaoYu[m]that I agree, but X2 and X1 should still be loaded from what X0 ends up being#asahi-dev · 2022-04-02 14:52
YichaoYu[m]yeah, which I interprete as memory_order_consume https://en.cppreference.com/w/cpp/atomic/memory_order and that seems to be what E2.3.2 is talking about#asahi-dev · 2022-04-02 14:40
YichaoYu[m](essentially that if the cpu predicted what X0 is correctly it could use a value it has cached for that from a minute ago, which it otherwise can't)#asahi-dev · 2022-04-02 14:39
YichaoYu[m]my interpretation was that on a single thread the address it access should be consistent, but the thread may decide to access [X0], with the correctly loaded address, before X0 is loaded from [X3].#asahi-dev · 2022-04-02 14:38
YichaoYu[m]in this case, I feel like in order to get the wrong address it'd have to also either run the add with a wrong input or ignore it, which is even more insane#asahi-dev · 2022-04-02 14:37
YichaoYu[m]cuz how does ldnp decide what the old value of [X3] is, especially, say, when you also have a `add X0, X0, #8` between the two#asahi-dev · 2022-04-02 14:37
YichaoYu[m]that'll still be very strange#asahi-dev · 2022-04-02 14:25
YichaoYu[m]I don't think the page I linked make sense = = ......#asahi-dev · 2022-04-02 14:21
YichaoYu[m](doubled page number.... ok)#asahi-dev · 2022-04-02 14:16
YichaoYu[m](ok for me it was C6.2.98 LDNP) and I do interpret `X[t] = data1; X[t2] = data2;` as writing the load result to the registers.#asahi-dev · 2022-04-02 14:14
YichaoYu[m](I wonder if there's a version difference... for B.b revision that's LDTRSW section)#asahi-dev · 2022-04-02 14:10
YichaoYu[m]yeah, since it says "by the other observers within ...."#asahi-dev · 2022-04-02 14:09
YichaoYu[m]I'm currently interpreting that as it doesn't satisfy the c++ memory_order_consume but I'm reading it again...#asahi-dev · 2022-04-02 14:09
YichaoYu[m]I assume by ARM ARM you mean C3.2.4?#asahi-dev · 2022-04-02 14:06
YichaoYu[m]what do you mean by it isn't an actual load? I thought it will load and write to the registers?#asahi-dev · 2022-04-02 13:33
YichaoYu[m]but is it suggesting that the re-ordering is visible for a single thread?#asahi-dev · 2022-04-02 13:32
YichaoYu[m]I assume it's my misread of https://developer.arm.com/documentation/den0024/a/The-A64-instruction-set/Memory-access-instructions/Non-temporal-load-and-store-pair#asahi-dev · 2022-04-02 13:31
YichaoYu[m]is anyone familiar with ldnp instruction?#asahi-dev · 2022-03-31 15:36
YichaoYu[m]<maz> "it would definitely be useful to..." <- here's the initial result of the test: https://github.com/yuyichao/m1_pmu_test/wiki #asahi-dev · 2022-03-30 15:24
YichaoYu[m]I’m not particularly interested in submitting any code for review (kernel or otherwise) but just playing with it to collect info. I do want to make type_hardware work though so I guess I can send an email to the kernel list….#asahi-dev · 2022-03-30 15:22
YichaoYu[m](And I was only doing that on the once in that list but I guess I should do all 256 of them )#asahi-dev · 2022-03-30 15:22
YichaoYu[m]I’m just running some loops and summarizing the results to take educated guesses.#asahi-dev · 2022-03-30 15:21
YichaoYu[m]Yes that’s for sure#asahi-dev · 2022-03-30 15:21
YichaoYu[m]marcan: I assume you know better (than I do)#asahi-dev · 2022-03-30 15:20
YichaoYu[m]Sure that’s fine. I assume doing complete blind testing should be fine and so I was asking if it worth doing it.#asahi-dev · 2022-03-30 15:05
YichaoYu[m]for that last question I'm asking if the json, or whatever other format, you use to make perf understand all the counters, can that go directly into perf/kernel source#asahi-dev · 2022-03-30 15:04
YichaoYu[m]no that's a generic question about copyright#asahi-dev · 2022-03-30 14:38
YichaoYu[m](I can try mailing list but so far I have very bad experience reporting any issues there)#asahi-dev · 2022-03-30 14:37
YichaoYu[m]I assume at least the last question (which was what I asked initially) belongs to here. Is any of these info upstremable?#asahi-dev · 2022-03-30 14:18
YichaoYu[m](also I'm at work now so I may not reply promptly....)#asahi-dev · 2022-03-30 14:18
YichaoYu[m]are you saying that even if I only care about clearly core-independent events like branch count or instruction count, I would have to do processor detection and figure out what kind of events I need?#asahi-dev · 2022-03-30 14:17
YichaoYu[m]I'm also talking from the perspective of directly calling `perf_event_open`#asahi-dev · 2022-03-30 14:16
YichaoYu[m]Well, I asked several questions and your answer to all of them has been that, even though some of them aren't only about kernel....... (full message at https://matrix.org/_matrix/media/r0/download/matrix.org/SceEeaNNujVEkPZciOwZHrBL)#asahi-dev · 2022-03-30 13:56
YichaoYu[m]Are you taking about getting perf to work with current kernel or finding information so it it can be put into the kernel?#asahi-dev · 2022-03-30 13:54
YichaoYu[m]Why do I need luck to get the instructions count working?#asahi-dev · 2022-03-30 13:37
YichaoYu[m]I'm not saying all of them should be mapped, only the ones that do have a cannonical counterpart#asahi-dev · 2022-03-30 13:35
YichaoYu[m]there's no way to do that without wasting counters#asahi-dev · 2022-03-30 13:35
YichaoYu[m]and I don't want to pin#asahi-dev · 2022-03-30 13:35
YichaoYu[m]if I want to cound instructions#asahi-dev · 2022-03-30 13:35
YichaoYu[m]but that also means#asahi-dev · 2022-03-30 13:35
YichaoYu[m]I've never heard of that argument before#asahi-dev · 2022-03-30 13:34
YichaoYu[m]so you are saying PERF_TYPE_HARDWARE is useless?