topic Re: crash in instrinsic function on win2k in Intel® Fortran Compiler

crash in instrinsic function on win2k

drb — Thu, 15 Sep 2005 18:09:33 GMT

Hi,

I have a multithreaded code with which I am runnin into some problems. I have calls to the intrinsic function DATE_AND_TIME and GETDAT. When I start a second thread, the function DATE_AND_TIME causes a crash, so I replaced it with CPU_TIME which seems to work. However, later on there is call to GETDAT which causes the crash in a similiar way. However, both of these subroutines were called in the original master thread without problem.

More confusingly, this does not crash on windows xp, and does not crash on windows2000 with the development env (compilers etc) installed, only when I move the exe over to another machine running win2k (it works on winxp without compilers) without the compilers installed do I get this problem.

This leads me to think that it is a runtime library problem, however, I have taken over the required dll's I think, so it is difficult to understand the problem. It isn't consistent however, i have two options, which are similiar, that start different subroutines in threads. It is only with one threaded subroutine that causes the problem.

Any tips greatly appreciated, I cannot debug since the problem doesn't happen when the compilers are installed

Thanks

Ben

Re: crash in instrinsic function on win2k

Steven_L_Intel1 — Thu, 15 Sep 2005 19:34:49 GMT

Please elaborate on "crash". What is the exact and complete error message? Are you building with the multithreaded libraries?

Re: crash in instrinsic function on win2k

drb — Thu, 15 Sep 2005 20:05:33 GMT

Sorry, I should have been clearer. I get the error:

"The instruction at "0x78003ef8" referenced memory at "0x00000046". The memory could not be "read".

I have tried to increase the stacksize for the new thread, but this has not made any effect.

Thanks,

Ben

edit: Yes, I have the multithreaded libraries switched on. I have tried the exe on two win2k machines without compilers, one with the dotnetfx.exe applied one without, I am not sure if this should make a difference. The project is a mixed C++/IVF project, the IVF being the main code, the C++ linked in. Using Intel 9.0 and C++.NET 2003

Message Edited by drb on 09-15-2005 06:07 AM

Re: crash in instrinsic function on win2k

Steven_L_Intel1 — Thu, 15 Sep 2005 20:39:28 GMT

I have no idea what dotnetfix.exe is.

Here's an experiment. On the system where this program doesn't work, boot into Safe Mode (typically you press F8 during the boot process to specify this) and then try running the program. What happens?

Re: crash in instrinsic function on win2k

Paul_Curtis — Fri, 16 Sep 2005 03:40:04 GMT

I run highly multithreaded F90 codes on all varieties of WinXP and Win2K without any crashes. These codes make extensive use of timing functions, using the Win32 API GetLocalTime (st) where st is a TYPE T_SYSTEMTIME. The systemtime data type provides millisecond resolution andeasily supports any sort of elapsed-time calculations.

Re: crash in instrinsic function on win2k

drb — Fri, 16 Sep 2005 17:52:35 GMT

Thank you for the replies. I think I have discovered a solution, but not the problem (which I suspect lies in some memory problems somewhere else in the code).
In an attempt to resolve the crash, I tried add the instrinsic call in various places to the subroutine. If I place the call to DATE_AND_TIME at the first executable line in the subroutine,then the program works fine.

Ben

Re: crash in instrinsic function on win2k

jim_dempsey — Tue, 20 Sep 2005 03:09:11 GMT

For the modules that contain the call the the intrinsic subroutine DATE_AND_TIME is the compiler option -Qauto used?.

i.e. is the 8 entry integer array for values on the stack?

Without the -Qauto a common name mangled block is created for the integer arrayfor exclusive use by the module (not the thread calling the module)and then used by all threads calling the particular routine?

Jim Dempsey