/srv/irclogs.linaro.org/2018/06/12/#linaro-meeting.txt

stooartSorry folks - completely forgot this morning.08:42
stooart#startmeeting08:42
linarobotMeeting started Tue Jun 12 08:42:26 2018 UTC.  The chair is stooart. Information about MeetBot at http://wiki.debian.org/MeetBot.08:42
linarobotUseful Commands: #action #agreed #help #info #idea #link #topic.08:42
stooart#topic OpenJDK (East)08:42
stooartnjian - would you like to go first?08:42
njianI am still working on SVE enablement work on OpenJDK c2 compiler. Currently try to understand c2 register definitions to map SVE registers.08:43
njianWith the help of vector api intrinsics, it is able to generate initial SVE instructions in a quick and dirty manner.08:43
stooartHow is it looking?08:43
njianI just created a Agnostic shape, and that will direct map to IR and backend codegen.08:45
njianWe need a clear and better register definition for SVE. I haven't had an idea yet.08:46
stooartRight.08:46
stooartAre you concerned about register spilling ?08:46
njianNo08:46
njianNot all08:47
njianspilling is one issue08:47
stooartIndeed. Quite frankly, it is best avoided.08:47
njianYes, spill will be a problem, but there's still a lot of others08:48
stooartDon't be afraid to discuss it with the Andrews - Dinn and Haley.08:48
njianWe break many assumptions - need to fix.08:48
njianOK, I have to understand current code first. :-)08:49
stooartOf course.08:49
njianI also took a quick look at SPECjbb crash issue with AOT enabled today08:50
njianI believe that was caused by some v8.1 LSE issue on Graal compiler.08:50
stooartThat'll be Daniel's work then.08:50
njianI am not sure whether that was caused by Daniel's patch, but I think Zhongwei will help to check that first.08:50
stooart(or at least tangentally affected by it)08:50
njianYes08:51
njianI've asked zhoyao01_ for help08:51
stooartok.08:52
njianThat's me.08:52
stooartOk, thanks njian.08:53
stooartzhoyao01_, would you like to go next?08:53
zhoyao01_Sure08:53
zhoyao01_I'm still working on CAS assertion error. And as upstream suggestted, I'm trying to fix the error in aarch64.ad.08:53
stooartYes, is it going ok?08:54
zhoyao01_I'm not totally sure yet.08:54
zhoyao01_For now, I see for "storeN/P" case, the modularization of Barrier Set patch doesn't have a impact on it.08:55
zhoyao01_But as my previous patch shows, it has a impact on "CAS on OOP" case.08:56
stooartYes. I haven't looked at it closely enough, but my original assumption was that it was caused by my ZGC code. Obivously not, of course...08:57
zhoyao01_And I've worked out a patch to match that change in aarch64.ad, but it turns out that the old pattern is still valid after the modularization of Barrier Set patch.08:58
stooartGiven the continued concern over memory barriers, I do wonder how this will change the generated code.08:58
stooartAre we certain it is the modularization?08:58
njianTo me, the barrier/membar code in the backend is complicated.08:58
zhoyao01_Because when it is "CAS on None-OOP" case, there would be no GC write barrier. So the old pattern is still valid in such case. I'm working on a patch to match both pattern.08:59
stooartnjian: I think that is a fair assessment - anything requiring that much documentation in comments is complicated.08:59
njianIf the membar graph changes, I don't know how other arches (e.g. ppc) can handle that.08:59
njianYes, a lot of comments. :-)09:00
zhoyao01_Yes, current CAS error is caused by the "the modularization of Barrier Set patch".09:00
stooartzhoyao01_: Ok.09:01
zhoyao01_And another thing I have not checked yet is: there is some checking for graph with CAS and GC write barrier combined in aarch64.ad, and I'm not sure that checking will be effected by the modularization.09:01
njianOk. I think if your new patch to match both patterns can pass tests, you can send out the patch to ask andrew(s) for review/suggestions.09:02
zhoyao01_Sure, for now, I feel fixing it in backend takes more effort than fixing it in middle-end.09:03
zhoyao01_That's me.09:03
stooartThanks zhoyao01_ .09:03
stooartI'll go next - I don't see YangZhang.09:03
zhoyao01_Yang is on leaving this afternoon.09:03
njianShe's not here09:03
stooartI've been working on a couple of bugs - nothing complex, JDK-8204628 and JDK-8204606.09:04
stooartThe latter was caused by my misusing registers - some of the advice I got wasn't quite right, and my testing didn't cover it properly. Just waiting on reviews and merges.09:05
stooartQuite why they haven't been picked up yet, I don't know, as you currently can't build debug builds, and running without C1 disabled can quite reliably crash.09:06
njianadinn has reviewed your patches just now. :-)09:06
stooartI've been trying to catch up with the ZGC code - I was a bit blocked by the broken code. Serves me right for assuming upstream works.09:07
stooartI'm currently working on the 64-bit literal oops testing with my fixed applied and no ZGC code - I'm testing with just G1GC as if 64-bit literal oops works with that, I'm not chasing ZGC simultaneously. There is an issue with C2 code - C1 code is ok.09:08
stooartnjian: Ah good. That must be really recent.09:08
stooartI was also checking the ZBackingFile support - it was suggested we only need file-backed heaps for multimapping, with with explict hugepages, we do need the files in some configurations.09:14
stooartThat's all from me.09:14
stooartSorry for starting so late. Is there any other business?09:14
njianDo you expect aarch64 ZGC to work on jdk11 - there's a lot of work to do...09:15
stooartaph was asking what was needed to be done, and what help I needed.09:15
stooartHe would like it in jdk11, but it is a bit late at this point. I hadn't expected Oracle would push it though so quickly, but they've had a couple of years to get to grips with it.09:16
stooartI'll speak with him today.09:17
njianOK09:17
njianPlease let us know, if you need help.09:17
stooartThanks njian.09:17
stooartOk, well, have a good evening, I'll talk to you later.09:18
stooart#endmeeting09:18
linarobotMeeting ended Tue Jun 12 09:18:54 2018 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)09:18
linarobotMinutes:        https://irclogs.linaro.org/meeting-logs/linaro-meeting/2018/linaro-meeting.2018-06-12-08.42.html09:18
linarobotMinutes (text): https://irclogs.linaro.org/meeting-logs/linaro-meeting/2018/linaro-meeting.2018-06-12-08.42.txt09:18
linarobotLog:            https://irclogs.linaro.org/meeting-logs/linaro-meeting/2018/linaro-meeting.2018-06-12-08.42.log.html09:18

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!