08:32:23 <bero> #startmeeting
08:32:23 <linarobot> Meeting started Tue Nov 24 08:32:23 2015 UTC.  The chair is bero. Information about MeetBot at http://wiki.debian.org/MeetBot.
08:32:23 <linarobot> Useful Commands: #action #agreed #help #info #idea #link #topic.
08:32:36 <bero> No big news today -- let's get into progress updates...
08:32:40 <bero> sunao: do you want to go first?
08:32:50 <sunao> What did you do?
08:32:52 <sunao> 1. work on LMG-933, font and glyphs cache.
08:32:53 <sunao> what are you going to do?
08:32:55 <sunao> 1.  Continue the investigate on memory allocate ? or work on LMG-933, fonts and glyphs cache.
08:32:56 <sunao> what problems do you have?
08:32:58 <sunao> 1.  Only hikey hdmi problem.
08:33:33 * luther is here
08:33:34 <sunao> bero, do you think we need to do more work on the memory allocator?
08:34:19 <bero> sunao: thanks, sounds good. The info on the memory allocator looks really interesting (guess we should make dlmalloc the default on low memory then). Do you think there's anything left there to do like try to tweak dlmalloc, or would you rather move on to font/glyph cache?
08:34:57 <sunao> I think we can fallback to dlmalloc
08:35:44 <bero> sunao: yes, that's what I'm thinking too - seems like a relatively obvious choice with the numbers you found
08:36:51 <bero> sunao: Do you think there's any tweaks we can make to dlmalloc to make it save even more?
08:37:14 <sunao> OK, I will try
08:38:03 <bero> ok, let's take another day or two to try tweaking dlmalloc before moving on to fonts and glyph cache.
08:38:12 <bero> Qian__: do you want to go next?
08:38:27 <Qian__> What did you do?
08:38:27 <Qian__> 1) Finish uploading SPEC fruit
08:38:27 <Qian__> 2) Reproduce the hash map error on x86(using perl-5.22.0)
08:38:29 <Qian__> 3) Debug it whether it can be replaced by script
08:38:31 <Qian__> a) $res->is_success can NOT be replaced by $res->{is_success}, maybe it’s NOT hash map referenced, Maybe there are othres.
08:38:33 <Qian__> b) Some interfaces may be changed (ex: return of get_url(), a little different dumps                     map)
08:38:35 <Qian__> c) $config->xxx(); works well.
08:38:37 <Qian__> What are you going to do?
08:38:39 <Qian__> Discussion about the further step about SPEC
08:38:41 <Qian__> Write script to have a try
08:38:43 <Qian__> What problems do you have?
08:38:45 <Qian__> Not an expert of regular expression, maybe spend a little long time to write the script.
08:38:47 <Qian__> Is there an expert to help?
08:39:08 <Qian__> I think we need a discussion
08:39:44 <bero> yes, bogden mentioned something about using SPEC without all the perl tools, do you think that's an option?
08:40:20 <Qian__> maybe, But it seems  a hard work ,too
08:41:10 <Qian__> I'd like writing a script first to have a try
08:41:30 <bero> Is most of the stuff that fails in spec itself or in the external perl modules it pulls in? If the latter, maybe updating them will fix it
08:42:40 <Qian__> yes, that's what bogden  wants to do
08:43:16 <Qian__> But, maybe cause other imcompatible problems
08:43:20 <bero> and it may be what makes most sense, given there may be issues beyond the $a->b vs. $a->{b} issue
08:43:24 <xruxa> bero, Qian__ - hi guys. Tom suggested that we stop work on SPEC now and ask LMG-SC to defer it - and have Qian__ work on something else
08:43:47 <xruxa> bero, Qian__ - okay with you?
08:44:01 <Qian__> ok, will we go on working on it?
08:44:07 <bero> xruxa: definitely ok with me -- this turned out to be a MUCH bigger effort than anyone expected
08:44:17 <Qian__> or it's NOT very important
08:44:34 <xruxa> bero, ack that, let me update the roadmap card and JIRA tasks, and you have Qian__ working on something else right away.
08:44:35 <bero> xruxa: And really the SPEC guys should be fixing this mess, nobody in the real world wants to still use a perl version from 2 decades ago
08:44:36 <Qian__> or it's NOT very important
08:45:00 <bero> Qian__:
08:45:23 <bero> the other task currently assigned to you is LMG-914, Check for static linkage and copied code where shared linkage should be used
08:45:26 <Qian__> i think should write a script first to have  a try
08:45:47 <bero> Qian__: Do you want to move ahead with that or would you prefer to do something else?
08:45:55 <bero> Qian__: how long do you think writing the script would take?
08:46:01 <Qian__> Ok, I will do the new work, and do the spec on spare time
08:46:35 <Qian__> Maybe two or three days/
08:47:07 <Qian__> I expect there are some issues can't be found now
08:47:15 <bero> I think we can give it 2 or 3 days... But if it just unmasks further problems, let's put spec on hold for now
08:48:04 <Qian__> Ok, I will try
08:48:12 <bero> Qian__: thanks
08:48:16 <bero> luther: do you want to go next?
08:48:59 <luther> ok
08:49:06 * luther What did you do?
08:49:07 <luther> 1. There are some company's urgent tasks needing to do
08:49:07 * luther What are you going to do?
08:49:07 <luther> 1. Go on company's urgent tasks
08:49:07 * luther What problems do you have?
08:49:07 <luther> 1. So many company's tasks are blocking there
08:49:21 <xruxa> hehehe
08:49:32 <xruxa> luther, thanks for letting us know.
08:49:52 <bero> luther: ok - how long do you think this will take?
08:50:49 <luther> xruxa, bero: i don't know how long, because i have a team in the company, in recent days there many new features needs to be develop.
08:51:44 <bero> luther: ok, let me know once there's any estimates so we know if we need to change anything in December planning...
08:51:44 <xruxa> luther, ok, keep us posted with updates.
08:51:55 <bero> xavierhsu1: do you want to go next?
08:52:05 <xavierhsu1> ok
08:52:05 <luther> xruxa, bero: ok :-)
08:53:07 <xavierhsu1> What did you do? 1. The original jpeg-turbo libs(modify by YongQin) also can not normally be used on Hikey(Using official image - Android L).   2. Because the official image (Android L) of Hikey isn't using jpeg-turbo, I need to build the Android source code from my local side.   3. The android source code (Android L) can be builded from my local side but it can not be normally booted.   4. After Xinliang help, now I can buil
08:53:31 <xavierhsu1> What are you going to do? 1. I need to check the my source code because the encoder (libjpeg) not be called.
08:53:45 <xavierhsu1> What problems do you have? 1. Debug now.
08:54:17 <bero> ok, thanks
08:54:33 <bero> liuyq: do you want to go next?
08:54:46 <xavierhsu1> bero, thanks.
08:54:55 <liuyq> What did you do?
08:54:56 <liuyq> Finished the benchmark tests on Nexus7 for both arm and thumb mode, Float related operations have more regression in thumb mode
08:54:56 <liuyq> Investigated problems for running hikey jobs on LAVA.
08:54:56 <liuyq> Fixed one LAVA bug, waiting for deployed.
08:54:57 <liuyq> What are you going to do?
08:55:00 <liuyq> Finish the HOWTOs for Hikey board
08:55:02 <liuyq> Send out draft version release notes
08:55:04 <liuyq> Re-submit lava jobs after lava-dispatcher fix deployed.
08:55:06 <liuyq> What problems do you have?
08:55:08 <liuyq> None
08:55:36 <bero> liuyq: thanks, looks good
08:55:41 <bero> zhizhoutian__: do you want to go next?
08:56:02 <zhizhoutian__> What did you do?
08:56:03 <zhizhoutian__> LMG-939	 Re-investigate zramfs to reduce memory usage
08:56:03 <zhizhoutian__> - Learn swap document, zram document, Spreadtrum memory analysis framework, Investigate costmem tools(a small memory cost tools written by mingling@Spreadturm)
08:56:03 <zhizhoutian__> - About save 14M memory when oom oocurred, this is log:
08:56:03 <zhizhoutian__> [ 1582.988708] c0 Enhanced Mem-info :ZRAM
08:56:03 <zhizhoutian__> [ 1582.988739] c0 Detail:
08:56:03 <zhizhoutian__> [ 1582.988739] c0 Total used:1808 pages, 7232 kB
08:56:04 <zhizhoutian__> zram can always save 2*(used memory), so it is 2*7232KB ~= 14M
08:56:04 <zhizhoutian__> What are you going to do?
08:56:05 <zhizhoutian__> What problems do you have?
08:56:08 <liuyq> bero, for the thumb mode, there are some test related browser have regressions, but I am not sure what caused that, probably float operations, but I have no proof
08:57:03 <bero> liuyq: I'd guess -ffast-math because that allows for some inaccuracy -- let's turn it off and just enable it in extreme memory saving mode...
08:57:26 <bero> zhizhoutian__: thanks, that looks interesting. Do you have any recommendations on how much memory we should allocate to zramfs already?
08:57:37 <bero> zhizhoutian__: And is costmem useful/available to the rest of us?
08:57:59 <zhizhoutian__> bero: costmem is a very small tools
08:58:18 <zhizhoutian__> bero: sure, you can see this tools' source code after short of time.
08:58:45 <zhizhoutian__> bero: about recommendations, i need to investigate.
08:58:55 <zhizhoutian__> bero: it is 400M in current syste,
08:59:30 <zhizhoutian__> bero: is the capacity of /dev/block/zram0
09:00:03 <bero> zhizhoutian__: ok, I think the next step would be figuring out the optimal size...
09:00:27 <zhizhoutian__> bero: yes, i think the more the better.
09:00:38 <zhizhoutian__> bero: but the speed of compress is a issue.
09:00:55 <zhizhoutian__> bero: so we are investigating the hardware compress
09:01:35 <zhizhoutian__> bero: if we can speed up this, then we can "exchange more space with time"
09:02:20 <bero> zhizhoutian__: yes, as long as compression is slow we need to keep some more memory uncompressed - if we can speed it up, we don't need that much uncompressed memory...
09:02:28 <bero> hardware compression there would be REALLY interesting
09:02:31 <zhizhoutian__> bero: btw, we are pushing legal department to share our source code to linaro
09:03:30 <bero> yes, that would be really good, right now we're pretty much expecting to just do most related development on Nexus 7 because it's similar enough and available, but having the real target hw with the needed code is obviously better
09:03:54 <zhizhoutian__> bero: you interest on this? so good:) very happy we can work on this together
09:04:35 <bero> me too
09:04:43 <bero> eric____: do you want to go next?
09:05:03 <eric____> fine
09:05:50 <eric____> bero: I've told you about the problem what I met
09:06:34 <eric____> I tried to get some help form the forum, but nobody answered yesterday
09:07:09 <bero> eric____: yes, let me know if you're blocked on anything there
09:07:31 <eric____> bero: I'm trying to do scan build in a small scale, and try to find a way
09:08:16 <eric____> I also want to trace it in a log.txt file and find what happend then
09:08:31 <eric____> bero: yes, I will
09:08:40 <bero> eric____: thanks
09:09:00 <bero> lijixin: do you want to go next?
09:10:44 <bero> Probably not there...
09:10:49 <bero> wuhai: do you want to go next?
09:11:03 <wuhai> What did you do?
09:11:03 <wuhai> 1.Analyzes PowerManagerService architecture code, transforming a part of the code to C/C++ by JNI, but the compiler error, then fix it, to be tested
09:11:03 <wuhai> 2.Learning the android memory tuning from https://01.org/android-ia/user-guides/android-memory-tuning-android-5.0-and-5.1
09:11:03 <wuhai> What are you going to do?
09:11:03 <wuhai> 1.To be test PowerManagerService after modified, to see whether there is bug?
09:11:04 <wuhai> 2.Analysis of architectural codes other android system services.
09:11:04 <wuhai> What problems do you have?
09:11:05 <wuhai> None
09:13:06 <bero> thanks
09:13:12 <bero> wangjian: do you want to go next?
09:13:20 <wangjian> ok
09:13:34 <wangjian> What did you do?
09:13:34 <wangjian> 1.Have got the hackbox account and downloading MLCR code.
09:13:34 <wangjian> 2.Discuss with Tom about how to get help from linaro to start work on reduce memory usage.
09:13:34 <wangjian> 3.Try to analysis launcher's feature,such as WallpaperPicker,WidgetsManage,to find some way to reduce memory usage.I found that these funtion don't malloc memory only if they to be called. But most of the time they will not be called.
09:13:34 <wangjian> What are you going to do?
09:13:35 <wangjian> Find and optimied some features which are often used in launcher.
09:13:35 <wangjian> What problems do you have?
09:13:36 <wangjian> None.
09:15:40 <bero> thanks
09:16:01 <bero> yspan mentioned earlier he has a conflicting meeting
09:16:10 <bero> so I think that was everyone who is around
09:16:14 <xruxa> ok
09:16:19 <bero> no, actually forgot myself
09:16:23 <bero> so it was just everyone who matters ;)
09:16:42 <bero> anyway:
09:16:45 <bero> What did you do?
09:16:45 <bero> Toolchain update
09:16:45 <bero> Started taking another look at LTO
09:16:45 <bero> Started trying to strip out unneeded symbols with -fvisibility=hidden -fvisibility-inlines-hidden
09:16:45 <bero> Debugged Opteron A1100 issues
09:16:45 <bero> What are you going to do?
09:16:45 <bero> More LTO debugging
09:16:46 <bero> Try to build AOSP with out-of-tree clang (+clang LTO)
09:16:46 <bero> Hopefully get the A1100 box up and running
09:16:47 <bero> Fix up issues with global -fvisibility=hidden -fvisibility-inlines-hidden
09:16:47 <bero> What problems do you have?
09:16:48 <bero> none
09:17:33 <bero> guess we're done -- have a nice rest of the day everyone, and keep us posted if there's any news
09:17:35 <bero> #endmeeting