If this is in a VM and @simd results in different behavior, this may be the VM and LLVM not agreeing on which instruction sets your system is capable of handling.
On Thursday, April 14, 2016 at 6:29:58 AM UTC-7, Isaiah wrote: > > Some library mismatch of the generic version causing the crashes sounds >> very likely, doesn't it? > > > No. > > On Thu, Apr 14, 2016 at 8:51 AM, K leo <[email protected] <javascript:>> > wrote: > >> I did more tests and runs. Now I put most @inbounds and @simd back and >> even added more in other places. A few runs over 3 hours did not get any >> problem. Someone mentioned about heat. I am actually running these on the >> deck with outside temperature over 34C. The MacBook Pro is hot but has no >> problem. >> >> So I am more inclined to attributing the problem to the Linux generic >> version of Julia (since the current reliable runs have been using the >> Ubuntu PPA version). Some library mismatch of the generic version causing >> the crashes sounds very likely, doesn't it? >> >> >> On Sunday, April 10, 2016 at 4:25:52 PM UTC+5:30, K leo wrote: >>> >>> These couple weeks I ran julia (0.4.5) on a Xubuntu guest of VirtualBox >>> hosted by a Macbook Pro. During the hour-long runs, the system crashed a >>> few times: user interface froze. This has happened a few times in the past >>> when I ran Julia natively on a Xubuntu computer. So hardware problem can >>> be ruled out. >>> >>> Trying to guess what could be the problem for the crashes, I put my >>> attention on the use of @inbounds and @simd in the code. The documentation >>> says @inbounds can cause crash when the index gets out of bound. @simd was >>> not said of possibly causing crashes, but it is said of being >>> experimental. So I took those out and re-run the code. >>> >>> After a few hours, the code finishes without crashing. Though this does >>> not lead to the conclusion that the two modifiers were the culprit, as the >>> code did not crash everytime in the past, this rather makes me wonder if >>> @inbounds can possibly be the cause. Look, the code finishes without >>> having an index out of bound problem. Is this enough to conclude that >>> @inbounds was not the problem? >>> >>> What can make @simd crash the system? >>> >>> Another possible cause might be the version of Julia I used. For the >>> past weeks, I used the Linux generic version of Julia. This successful run >>> was on the version I got from Ubuntu's PPA. In the past, I also juggled >>> between the two sources for Julia. I can't be certain in saying that the >>> generic version crashes on ubuntu, but my question is what are the real >>> differences between the two version of Julia? >>> >>> >>> >
