|
From: Philippe W. <phi...@sk...> - 2011-11-07 21:17:57
|
gcc compile farm has received a bbbbbigggggg power7 computer (64 processors, 64Gb memory) from IBM, installed with fedora16. Unless someone is already busy setting up a nightly test on a power7, I can setup a nightly ppc64/fedora16 test on this computer. Note: I just have one problem : I did one build + regtest on this computer, and had drd running on annotate_hbefore that took about one hour of cpu (then I killed it). gcc compile farm highly recommends to avoid such looping tests (so, I will put a ulimit in cpu when launching the nightly test). But still it would be nice to fix the annotate_hbefore, otherwise we will have random full failures of the nightly run on this setup. I did run on this ppc64 annotate_hbefore with helgrind and drd a few times, and the time the test takes has varied between 5 seconds to 2 min 34 seconds. drd has shown the biggest difference but with helgrind, there is also a huge variation. I also encountered a looping annotate_hbefore on amd64, and Florian got it on s390x. A change described in https://bugs.kde.org/show_bug.cgi?id=268623 is improving, but seems it can still hang (or at least run during a long time or have huge variations in time). So, it looks like there is a common cause (not fixed yet) to these hangs/big time variations (unfair scheduling ?). Bart, do you expect to continue the work on the fair scheduling in https://bugs.kde.org/show_bug.cgi?id=270006 ? Philippe |