January 2011 - Mpich2-maint - mailman3.cels.anl.gov

Re: [mpich2-maint] #1418: sub-communicator operation sometime fails if node failure occurs
by 王睿 27 Jan '11

27 Jan '11

Yep, it was the problem. It works well after adding a barrier. Best Regards, Rui 2011/1/27 mpich2 <mpich2-maint(a)mcs.anl.gov> > #1418: sub-communicator operation sometime fails if node failure occurs > > ----------------------------------+----------------------------------------- > Reporter: wangraying@… | Owner: buntinas > Type: bug | Status: assigned > Priority: major | Milestone: mpich2-1.3.3 > Component: mpich2 | Keywords: > > ----------------------------------+----------------------------------------- > > Comment(by buntinas): > > I'm not seeing the same errors. One thing I noticed is that sometimes > proc 2 exits before all processes have completed the comm_split. This > will result in an invalid col_comm communicator at some nodes. Check the > return code for comm_split to check for that. > > I added a barrier between the comm_split and the exit. That ensured that > the comm_split would succeed. (Sometimes the barrier returned an error, > but that's OK). > > > Can you try this again with the latest trunk and let us know if you still > get this error? > > Thanks, > -d > > -- > Ticket URL: <http://trac.mcs.anl.gov/projects/mpich2/ticket/1418#comment:3 > > > mpich2 <http://www.mcs.anl.gov/research/projects/mpich2> > MPICH2 is a high-performance and widely portable implementation of MPI > -- Rui Wang Key Laboratory of Computing Systems and Architecture, Institute of Computing technology, Chinese Academy of Science, Beijing, P.R.China

1 0

windows console functions using mpich2
by S.Banen 27 Jan '11

27 Jan '11

Hi, I have a problem that has been bugging me for several days now, and I can't find a solution or a clue in any forum, so I hope that you can help me. I am using windows console functions (SetConsoleCursorPosition, etc.) in a c++ win32 console application to control the location of the cursor in the console window. One example for using this is that I print every five seconds to screen the updated remaining time for all nodes participating in the computation: Node 0 timeleft: 1 day(s), 06:25:43 Node 1 timeleft: 1 day(s), 06:24:21 Etc. I don't want to use a new line each time, so the remaining time is printed at the same window position every time. Executing the application without MPI works fine. Executing the application from the development IDE (Visual Studio 2005) using the "MPI Cluster Debugger" also works fine. But when executing the application from the command line using mpiexec, it seems as if the code for controlling the cursor is simply ignored; the output is not printed at the desired location, but at the current position of the cursor. Below code that points out the problem: In stdafx.h: #include "mpi.h" #include <stdio.h> #include <iostream> Using namespace std; #include "windows.h" In MPITest.cpp: #include "stdafx.h" int main(int argc,char *argv []) { int numtasks, rank, rc, wait; rc = MPI_Init(&argc,&argv); MPI_Comm_size(MPI_COMM_WORLD,&numtasks); MPI_Comm_rank(MPI_COMM_WORLD,&rank); printf ("Number of tasks= %d My rank= %d\n", numtasks,rank); HANDLE hStdout = GetStdHandle(STD_ERROR_HANDLE); COORD pos = {7, 7}; int val = SetConsoleCursorPosition( hStdout, pos); cerr<<"This text should appear at console window location of pos"<<endl; MPI_Finalize(); } My questions are: - Is it possible to control the cursor location using mpich2 together with the windows console functions? - If the answer is yes, what am I doing wrong (or, what do I have to do to get it right)? - If the answer is no, is there any way to work around this problem? Thank you for your time. Regards, Steven Banen

1 0

Mpich2 for Windows and PGI compilers
by Pierre-Olivier Dallaire 25 Jan '11

25 Jan '11

Good afternoon, I was wondering if somebody was able to use the PGI Fortran compiler under windows and link it the application with the mpich2 windows binary provided on the website. Regards, PO

1 0

Aborting: Unable to connect to localhost
by xuan liu 25 Jan '11

25 Jan '11

Hallo mpich2 team: i wil ask you, why cannt my test-programe to connect to my localhost? my system is windows xp sp3 amd64 *1 core no firewell no virsscaner pilifischy

1 0

PathScale EKOPath 4 - We're back!
by PathScale Inc. 21 Jan '11

21 Jan '11

EKOPath 4 and ENZO 2011 Beta are live! Brief reminder that you're receiving this email because you subscribed via www.pathscale.com. Our updated privacy policy requires you to subscribe again if you wish to continue receiving further news and updates. Short link http://eepurl.com/cloQH EKOPath 4 http://www.pathscale.com/ekopath-compiler-suite Email sales(a)pathscale.com to receive 35% off MSRP before January 31st Now you can give us your feedback! http://vote.pathscale.com/ Note : Email support(a)pathscale.zendesk.com if you need access the new support portal What's new in EKOPath 4 * Completely new C++ STL, runtime and unwind libraries * Reference counted basic_string using atomic locking * Thread-safety, including iostream and locale objects * Fortran 2003 procedure pointers * Optimized exception handling * Overall performance improvements * Improved Interprocedural Optimization * Please see documentation for complete list For a free 15 day trial please email sales(a)pathscale.com http://www.pathscale.com/ekopath-compiler-suite Core Highlights * High performance C, C++, and Fortran 77/90/95/2003 compilers * OpenMP 2.5 for C, C++ and Fortran * Comprehensive support for Intel64 and AMD64 processors * Industry leading optimizations and performance * Advanced PathDB serial debugger for C, C++ and Fortran * Advanced PathAS assembler with GPGPU support * Compatible with GNU/gcc toolchain and third party debuggers * Industry leading performance and robustness Supported OS * Linux (Redhat 5.x, 6.x, SLES 9, 10, 11 and Ubuntu 10) * Solaris 10 and OpenSolaris * FreeBSD 8+ * Please see documentation for complete list Beta testing for OSX and Windows 64 starts Q1 2011_ Contact - sales(a)pathscale.com if you're interested in early access http://www.pathscale.com/ekopath-compiler-suite ENZO 2011 - Smarter GPGPU programming ENZO is a complete kernel driver to compiler GPGPU solution designed from the ground up for HPC. PathScale has been working hard for over a year to combine the ease of HMPP programming model with the performance of a native CUDA compiler. * HMPP C, C++ and Fortran directive * Optimized for NVIDIA Tesla * PathScale CUDA compiler * Assembler/disassembler (accurate to instructions level) * Much more Visit http://www.pathscale.com/ENZO for more information Copyright © 2011 PathScale Inc., All rights reserved. This email was sent to mpich2-maint(a)mcs.anl.gov. If you are no longer interested you can unsubscribe instantly: http://marketing.pathscale.com/t/r/u/ydjiel/ojhhyvku/

1 0

Re: [mpich2-maint] #1139: mpiexec's "-genv" does not work on Windows for strings longer than 260 chars
by Petrov, Vladimir S 14 Jan '11

14 Jan '11

Hi, The name is short. It's the value's length that is causing the problem. Best regards, Vladimir Petrov -----Original Message----- From: mpich2 [mailto:[email protected]] Sent: Friday, January 14, 2011 10:30 PM Subject: Re: [mpich2-maint] #1139: mpiexec's "-genv" does not work on Windows for strings longer than 260 chars #1139: mpiexec's "-genv" does not work on Windows for strings longer than 260 chars -----------------------------------------+---------------------------------- Reporter: vladimir.s.petrov@… | Owner: jayesh Type: bug | Status: accepted Priority: major | Milestone: mpich2-1.3.2 Component: mpich2 | Keywords: -----------------------------------------+---------------------------------- Comment(by jayesh): Hi, Is the name of the environment variable longer than 260 chars (Currently MPICH2 has a limit of 256 chars for environment variable names & 8K chars for environment variable values)? I tried setting environment variables with values larger than 260 chars with MPICH2 1.3.1 and it worked for me. -Jayesh -- Ticket URL: <http://trac.mcs.anl.gov/projects/mpich2/ticket/1139#comment:3> mpich2 <http://www.mcs.anl.gov/research/projects/mpich2> MPICH2 is a high-performance and widely portable implementation of MPI -------------------------------------------------------------------- Closed Joint Stock Company Intel A/O Registered legal address: Krylatsky Hills Business Park, 17 Krylatskaya Str., Bldg 4, Moscow 121614, Russian Federation This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

1 0

Re: [mpich2-maint] #1158: Fortran-Binding for MPI_Comm_get_attr(MPI_APPNUM) broken on bigendian/64 bits
by Rajeev Thakur 14 Jan '11

14 Jan '11

Rick Bradshaw and Narayan. Rajeev On Jan 14, 2011, at 10:18 AM, William Gropp wrote: > I might have an account, but I can't log in (it wants a password and the last one that I set doesn't work). Who should I contact about niagara1? > > Bill > > On Jan 12, 2011, at 4:38 PM, Darius Buntinas wrote: > >> >> Bill, >> >> There's niagara1.mcs.anl.gov you can use. I think you should have an account there. >> >> -d >> >> On Jan 12, 2011, at 4:32 PM, mpich2 wrote: >> >>> #1158: Fortran-Binding for MPI_Comm_get_attr(MPI_APPNUM) broken on bigendian/64 >>> bits >>> ------------------------------------+--------------------------------------- >>> Reporter: nicolai.stange@… | Owner: gropp >>> Type: bug | Status: assigned >>> Priority: major | Milestone: mpich2-1.3.2 >>> Component: mpich2 | Keywords: >>> ------------------------------------+--------------------------------------- >>> >>> Comment(by gropp): >>> >>> I'll have a look at it. >>> >>> However, note that the Fortran interface for attribute values is >>> >>> The answer to the questions about the attribute values is pretty simple - >>> as the attributes in C and in Fortran are different (in one case they are >>> pointers to the values and in the others they are the values), the >>> attribute keys are and should be different. >>> >>> Is there a Solaris system around on which I could test? The code should >>> have been moving the bit around correctly, but there may be a missing >>> cast, either here or in the init code. >>> >>> Bill >>> >>> -- >>> Ticket URL: <http://trac.mcs.anl.gov/projects/mpich2/ticket/1158#comment:4> >>> mpich2 <http://www.mcs.anl.gov/research/projects/mpich2> >>> MPICH2 is a high-performance and widely portable implementation of MPI >> > > William Gropp > Deputy Director for Research > Institute for Advanced Computing Applications and Technologies > Paul and Cynthia Saylor Professor of Computer Science > University of Illinois Urbana-Champaign > > > > > _______________________________________________ > mpich2-maint mailing list > mpich2-maint(a)lists.mcs.anl.gov > https://lists.mcs.anl.gov/mailman/listinfo/mpich2-maint

1 0

Re: [mpich2-maint] #1158: Fortran-Binding for MPI_Comm_get_attr(MPI_APPNUM) broken on bigendian/64 bits
by William Gropp 14 Jan '11

14 Jan '11

I might have an account, but I can't log in (it wants a password and the last one that I set doesn't work). Who should I contact about niagara1? Bill On Jan 12, 2011, at 4:38 PM, Darius Buntinas wrote: > > Bill, > > There's niagara1.mcs.anl.gov you can use. I think you should have > an account there. > > -d > > On Jan 12, 2011, at 4:32 PM, mpich2 wrote: > >> #1158: Fortran-Binding for MPI_Comm_get_attr(MPI_APPNUM) broken on >> bigendian/64 >> bits >> ------------------------------------ >> +--------------------------------------- >> Reporter: nicolai.stange@… | Owner: gropp >> Type: bug | Status: assigned >> Priority: major | Milestone: mpich2-1.3.2 >> Component: mpich2 | Keywords: >> ------------------------------------ >> +--------------------------------------- >> >> Comment(by gropp): >> >> I'll have a look at it. >> >> However, note that the Fortran interface for attribute values is >> >> The answer to the questions about the attribute values is pretty >> simple - >> as the attributes in C and in Fortran are different (in one case >> they are >> pointers to the values and in the others they are the values), the >> attribute keys are and should be different. >> >> Is there a Solaris system around on which I could test? The code >> should >> have been moving the bit around correctly, but there may be a missing >> cast, either here or in the init code. >> >> Bill >> >> -- >> Ticket URL: <http://trac.mcs.anl.gov/projects/mpich2/ticket/1158#comment:4 >> > >> mpich2 <http://www.mcs.anl.gov/research/projects/mpich2> >> MPICH2 is a high-performance and widely portable implementation of >> MPI > William Gropp Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign

1 0

Re: [mpich2-maint] #1062: iprobe problem
by Professor A.D. Brown 14 Jan '11

14 Jan '11

Well, changing to nemesis made a difference; I get the probe back in 1, 2 or 3 attempts, but this is still independent of the explicit delay in the loop - thus with the loop delay set to 1000ms is can take 3 seconds for Iprobe to see the message? Still, it's much better. Thanks for your efforts. I don't think there is a ticket for this - or if there is I didn't create it - your website kept falling over so I mailed the problem in. mpich2 wrote: > #1062: iprobe problem > --------------------+------------------------------------------------------- > Reporter: gropp | Owner: jayesh > Type: bug | Status: accepted > Priority: minor | Milestone: mpich2-1.3.3 > Component: mpich2 | Keywords: > --------------------+------------------------------------------------------- > > Comment(by jayesh): > > Hi, > If I run your code with MPICH2 1.3.1 on my Windows 7 laptop (64-bit, core > i7) I get the data in two iprobes. > The older versions of MPICH2 had sock as the default channel and the > newer ones use Nemesis (newer channel, better performance). You might also > want to try your code with 1.2.1p1 by explicitly specifying Nemesis as the > channel for your job (mpiexec -n 2 -channel nemesis mpi_snark.exe). > > (PS: Go ahead and update the ticket directly - instead of sending an email > - with your findings) > Regards, > Jayesh > -- --------------------------- Professor A.D. Brown Electronic System Design Group Electronics and Computer Science University of Southampton SO17 1BJ, Hampshire, UK T: +44 (2380) 593374 F: +44 (2380) 592901

1 0

Limpe ja o seu nome SPC/SERASA
by Nome Limpo 14 Jan '11

14 Jan '11

Nada vale mais do que ter seu crédito de volta! Renove sua vida, Tenha seu CNPJ ou CPF sem restrições de uma maneira rapida, facil e segura. Informaçoes: nomelimpo.sac(a)live.com

1 0