Hi Rizwana, The regressions run fine on a rather large selection of hosts and compilers (OS X, RHE5, RHE6, Ubuntu 10.04, Ubuntu 12.04, with gcc 4.6 all the way to 4.9, and clang 3.1). I cannot imagine Ubuntu 14.04 should cause problems. The thing that makes me most suspicious is that you get different results when re-running the same tests. There is no non-determinism in the simulator, and the results should be the same every time. We have done quite a lot of “linting” and an ARM regression run with UBSan switched on shows no run-time errors at all.
Is anyone else having these issues? Andreas From: Rizwana Begum <[email protected]<mailto:[email protected]>> Date: Wednesday, 4 February 2015 05:23 To: Andreas Hansson <[email protected]<mailto:[email protected]>> Cc: gem5 users mailing list <[email protected]<mailto:[email protected]>> Subject: Re: [gem5-users] DRAMCtrl auto precharge logic simplification Hello Andreas, This is really strange. I cloned the repo and downloaded the system files again. The errors still don't go away. I tried for 'fast' binary as well and the errors persist. I am trying it all on Ubuntu 14.04. What host OS do you use? I also noticed that number of these errors vary from run to run. I am planning to run long ones tonight, will let you know if those run fine or result in errors. Thank you, -Rizwana On Tue, Feb 3, 2015 at 4:56 PM, Andreas Hansson <[email protected]<mailto:[email protected]>> wrote: Hi Rizwana, That is strange. I just cloned the repo and ran with the files from http://www.gem5.org/dist/current/arm/aarch-system-2014-10.tar.xz without any problems. Could you do a clean build and re-download just to be sure? Andreas From: Rizwana Begum <[email protected]<mailto:[email protected]>> Date: Tuesday, 3 February 2015 21:49 To: Andreas Hansson <[email protected]<mailto:[email protected]>> Cc: gem5 users mailing list <[email protected]<mailto:[email protected]>> Subject: Re: [gem5-users] DRAMCtrl auto precharge logic simplification Hi Andreas, Here are the ones that changed: ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview-simple-atomic CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview-simple-atomic-dual CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview-simple-timing CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview-simple-timing-dual CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview-switcheroo-atomic CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview64-simple-atomic CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview64-simple-atomic-dual CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview64-simple-timing CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview64-simple-timing-dual CHANGED! ***** build/ARM/tests/opt/quick/fs/10.linux-boot/arm/linux/realview64-switcheroo-atomic CHANGED! Thank you, -Rizwana On Tue, Feb 3, 2015 at 4:04 PM, Andreas Hansson <[email protected]<mailto:[email protected]>> wrote: Hi Rizwana, That doesn’t sound good. Let me double check the system files. Which regressions are “changed"? Andreas From: Rizwana Begum <[email protected]<mailto:[email protected]>> Date: Tuesday, 3 February 2015 21:01 To: Andreas Hansson <[email protected]<mailto:[email protected]>> Cc: gem5 users mailing list <[email protected]<mailto:[email protected]>> Subject: Re: [gem5-users] DRAMCtrl auto precharge logic simplification Hello Andreas, I ran the tests again and they result in similar errors (magnitude error %) in stats. When I don't specify the paths to system files, tests fail complaining about system files path. I ran tests on my local Ubuntu (14.04) as well as on a CentOS 5 server, both result in similar errors. Though none of the tests fail, I don't have clean regression tests (I only tried quick ones). I have following gem5 commit: changeset: 10667:e17949745150 tag: tip user: Malek Musleh <[email protected]<mailto:[email protected]>> date: Fri Jan 30 15:49:34 2015 -0600 summary: config: arm: fix os_flags All I did was cloned gem5-dev repository, compiled ARM opt binary (scons build/ARM/gem5.opt), downloaded system files, placed the binaries and disks folder in system folder, pointed to the system folder in SysPaths.py, and then ran quick regressions (scons build/ARM/tests/opt/quick). Any help to resolve this issue would be great. Thank you, -Rizwana On Tue, Feb 3, 2015 at 12:02 PM, Rizwana Begum <[email protected]<mailto:[email protected]>> wrote: Hello Andreas, I ran tests on clean repo (without any source code modifications), however, I had modified SysPaths.py to point to binaries, disks downloaded from gem5 download page (ARM Full-System Files<http://www.gem5.org/dist/current/arm/aarch-system-2014-10.tar.xz>). I am not sure, if that caused the tests to use different kernel/disk image than the default ones. I compiled opt binary and ran quick tests on opt (scons build/ARM/tests/opt/quick). I am running tests now without modifications to SysPaths.py. I will let you know if my results would change. Thank you, -Rizwana On Tue, Feb 3, 2015 at 11:04 AM, Andreas Hansson <[email protected]<mailto:[email protected]>> wrote: Hi Rizwana, There should be no stats changes at all on the trunk. The regression runs from yesterday confirms this as well. Are you sure you ran with a clean repo? Andreas From: Rizwana Begum <[email protected]<mailto:[email protected]>> Date: Tuesday, 3 February 2015 15:44 To: Andreas Hansson <[email protected]<mailto:[email protected]>> Cc: gem5 users mailing list <[email protected]<mailto:[email protected]>> Subject: Re: [gem5-users] DRAMCtrl auto precharge logic simplification Hello Andreas, Yes, the condition will still consider bank conflicts for open-adaptive policy. I made the fix and ran quick regression tests for ARM. All of the tests passed, however there are some errors(in % difference) in results. Here is the output before and after the fix (grep for error): gem5-dev tip (before fix): Maximum error magnitude: +68.750000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +89.743590% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +68.750000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% With fix: Maximum error magnitude: +6.666667% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +38.862333% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +68.750000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +89.743590% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +9999.000000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +68.750000% [... showing top 20 errors only, additional errors omitted ...] Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Maximum error magnitude: +0.000000% Is this a concern? or can I ignore these errors? PS: Attached full output of quick regression tests. Thank you, -Rizwana On Tue, Feb 3, 2015 at 3:57 AM, Andreas Hansson <[email protected]<mailto:[email protected]>> wrote: Hi Rizwana, I look forward to see the patch. Note that open-adaptive and closed-adaptive behave differently. In the open case we only close if there is no hit, and there is a bank conflict. For the closed one we close if there are no hits (ignoring if there are conflicts or not). If this is till captured then by all means go ahead and simplify the code. Perhaps add some more comments to make this more clear. Andreas From: Rizwana Begum via gem5-users <[email protected]<mailto:[email protected]>> Reply-To: Rizwana Begum <[email protected]<mailto:[email protected]>>, gem5 users mailing list <[email protected]<mailto:[email protected]>> Date: Tuesday, 3 February 2015 05:01 To: gem5 users mailing list <[email protected]<mailto:[email protected]>> Subject: [gem5-users] DRAMCtrl auto precharge logic simplification Hello All, I think the if condition that's checking to find the right condition for auto-precharge in doDRAMAccess() can be simplified Original : while (!(got_more_hits && (got_bank_conflict || pageMgmt == Enums::close_adaptive)) && p != queue.end()) { Simplified: while (!got_more_hits && p != queue.end()) { The above simplification comes as both close-adaptive and open-adaptive policies keep row open if a hit is found. Otherwise the search for a hit continues until the end of the queue and during the search got_bank_conflict gets updated anyways. I am planning to put this simplification on reviewboard (along with another bug fix that I have). I would appreciate it if anyone using DRAM controller has any comments regarding this simplification. Thank you, -Rizwana -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2557590 ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2548782 -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2557590 ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2548782 -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2557590 ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2548782 -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2557590 ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2548782 -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2557590 ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, Registered in England & Wales, Company No: 2548782
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
