Parallel and time macro segfault  7261 Closed pfitzseb opened this Issue on Jun 14, 2014 · 6 comments Projects None yet Labels parallel windows Milestone No milestone Assignees No one assigned 5 participants  pfitzseb  vtjnash  JeffBezanson  jiahao  ihnorton Notifications    pfitzseb pfitzseb commented on Jun 14, 2014  Hey everyone, I found a kinda strange problem. Suppose there's some code that  times a  parallel loop as follows:   time begin        a   zeros 100_000_000          parallel for i 1:100_000_000             a i    i        end end  FIrst problem: The time macro returns wrong values  I think , since it clearly takes a lot more time until the REPL responds than indicated.  Second problem: After hitting Enter, elapsed time: will be displayed as output, but it takes some time until the actual time can be seen. Now, if during that time the evaluation is interrupted ,the following is displayed:  elapsed time: exception on 0.1257680236:  seconds  ERROR: interrupt  in anonymous at no file:4  in anonymous at multi-jl:1261  in run_work_thunk at multi-jl:613  in run_work_thunk at multi-jl:622 800003120 bytes allocated   If the exact same loop is then started again, Julia segfaults:  elapsed time: Please submit a bug report with steps to reproduce this fault, and any error messages that follow  in their entirety . Thanks. Exception: EXCEPTION_ACCESS_VIOLATION at 0x6dba0333 ---jl_profile_is_running at ???:1840907059-jl_profile_is_running at ???:1840907059-jl_profile_is_running at ???:1840906724-jl_profile_is_running at ???:1840906906-jl_gc_collect at ???:1840909837 alloc_2w at ???:1840913957-jl_box_int64 at ???:1840821383 anonymous at no file:4 anonymous at multi-jl:1261 run_work_thunk at multi-jl:613 run_work_thunk at multi-jl:622-jlcall_run_work_thunk;1114 at ???:52354747 ??? at ???:2 ??? at ???:76992432 Please submit a bug report with steps to reproduce this fault, and any error messages that follow  in their entirety . Thanks. Exception: EXCEPTION_ACCESS_VIOLATION at 0x6dba0333 ---jl_profile_is_running at ???:1840907059 ??? at ???:76992432   the last five lines will be repeated several times . Without the  parallel macro the loop works flawlessly. I'm using the latest prebuilt Julia binary for Windows, versioninfo   below:  Julia Version 0.3.0-prerelease+3675 Commit 253cbac   2014-06-13 21:49 UTC  Platform Info:   System: Windows  x86_64-w64-mingw32    CPU: Intel R  Core TM  i7-3517U CPU   1.90GHz   WORD_SIZE: 64   BLAS: libopenblas  USE64BITINT DYNAMIC_ARCH NO_AFFINITY    LAPACK: libopenblas   LIBM: libopenlib  Edit: After reading a bit about  parallel in the docs, I realize that what I did above is not very... correct, but it still shouldn't segfault, right?! And, furthermore, above code seems to produce the correct result: Doing the same operation with  parallel and without leads to the exact same array  although memory consumption predictably doubles with a parallel loop .  Edit No 2: Oookay, this happens without  time, too, so it most likely is just because of my wrong usage of  parallel. Still, I'll leave the issue open because segfaults should be adressed imo.  vtjnash The Julia Language member vtjnash commented on Jun 14, 2014   JeffBezanson adding  sync makes the  time work as expected, but is this expected behavior?  jiahao jiahao added the parallel label on Jun 15, 2014  JeffBezanson The Julia Language member JeffBezanson commented on Jun 16, 2014   parallel with no reduction does not return a result so it is asynchronous; you have to wait for it explicitly.  JeffBezanson The Julia Language member JeffBezanson commented on Jun 16, 2014  see also  6752 for problems combining  Clanguage and parallelism.  JeffBezanson The Julia Language member JeffBezanson commented on Jun 16, 2014  I don't get the segfault on linux.  pfitzseb pfitzseb commented on Jun 17, 2014  Yeah, I can't reproduce it on linux either, although it's running in a VM and on a different machine. And about  time in conjuction with  parallel: Wouldn't it still be desirable to get accurate measuremeants of the longest running process?  ihnorton ihnorton added the windows label on Jan 11, 2015  pfitzseb pfitzseb commented on May 21  Just found this issue again, and can't reproduce the segfault anymore. Closing.  pfitzseb pfitzseb closed this on May 21