October 2014 - devel - mailman3.cels.anl.gov

Re: [mpich-devel] mpich -mcmodel=medium
by Junchao Zhang 21 Oct '14

21 Oct '14

On Tue, Oct 21, 2014 at 9:04 AM, WeiGuang He <c212heweiguang(a)gmail.com> wrote: > 1 In the 'bin' directory containing mpif90, the following command shows > ifort is used. > $./mpif90 -v > > mpifort for MPICH version 3.1.2 > > ifort version 14.0.3 > > 2 ifort -mcmodel=medium won't work. > If ifort did not work, you may try to build MPICH with gfortran > Since large array could not work with mpich in my computer, > I downloaded openmpi and now it is working fine. The parallel code being > modified from the serial > version > is now working. > > Thank you for your attention >

1 0

Re: [mpich-devel] mpich -mcmodel=medium
by WeiGuang He 21 Oct '14

21 Oct '14

1 In the 'bin' directory containing mpif90, the following command shows ifort is used. $./mpif90 -v mpifort for MPICH version 3.1.2 ifort version 14.0.3 2 ifort -mcmodel=medium won't work. Since large array could not work with mpich in my computer, I downloaded openmpi and now it is working fine. The parallel code being modified from the serial version is now working. Thank you for your attention On Mon, Oct 20, 2014 at 8:35 PM, <devel-request(a)mpich.org> wrote: > Send devel mailing list submissions to > devel(a)mpich.org > > To subscribe or unsubscribe via the World Wide Web, visit > https://lists.mpich.org/mailman/listinfo/devel > or, via email, send a message with subject or body 'help' to > devel-request(a)mpich.org > > You can reach the person managing the list at > devel-owner(a)mpich.org > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of devel digest..." > > > Today's Topics: > > 1. Re: > mpich -mcmodel=medium (Junchao Zhang) > 2. new tutorial books on MPI (Rajeev Thakur) > 3. proper way to detect if program launched by mpiexec? > (Jeff Hammond) > 4. Re: proper way to detect if program launched by mpiexec? > (Balaji, Pavan) > 5. Re: proper way to detect if program launched by mpiexec? > (Jeff Hammond) > 6. Re: proper way to detect if program launched by mpiexec? > (Rob Latham) > 7. Re: proper way to detect if program launched by mpiexec? > (Balaji, Pavan) > 8. Re: proper way to detect if program launched by mpiexec? > (Jeff Hammond) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Wed, 8 Oct 2014 08:57:34 -0500 > From: Junchao Zhang <jczhang(a)mcs.anl.gov> > To: "<devel(a)mpich.org>" <devel(a)mpich.org> > Subject: Re: [mpich-devel] mpich -mcmodel=medium > Message-ID: > <CA+MQGp-=+_ > aK83ef4zdteyckS1nrX8nPQu2XB1yEmeHHWOrEQA(a)mail.gmail.com> > Content-Type: text/plain; charset="utf-8" > > Did you build mpich with Intel compilers? > Have you tried to use -mcmodel=large for a non-mpi program and compiled it > with ifort? > > --Junchao Zhang > > On Wed, Oct 8, 2014 at 8:02 AM, WeiGuang He <c212heweiguang(a)gmail.com> > wrote: > > > Hello, madam/sir > > It seems that "-mcmodel=medium" is not supported by mpif90 (I installed > > the latest mpich). > > How could I use large array? > > > > mpif90 mtestf90.f90 winfencef90.f90 -o abc -assume byterecl -shared-intel > > -mcmodel=large -fPIC > > > > %% The output is > > > > ./abc: symbol lookup error: ./abc: undefined symbol: > > __intel_new_feature_proc_init > > > > ./abc: symbol lookup error: ./abc: undefined symbol: > > __intel_new_feature_proc_init > > > > > > > > > =================================================================================== > > > > = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES > > > > = PID 27182 RUNNING AT lgmiris > > > > = EXIT CODE: 127 > > > > = CLEANING UP REMAINING PROCESSES > > > > = YOU CAN IGNORE THE BELOW CLEANUP MESSAGES > > > > > > > =================================================================================== > > > > Best Regards > > Weiguang HE > > > > _______________________________________________ > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/devel > > >

1 0

Re: [mpich-devel] proper way to detect if program launched by mpiexec?
by Jeff Hammond 20 Oct '14

20 Oct '14

On Mon, Oct 20, 2014 at 11:15 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > >> On Oct 20, 2014, at 12:43 PM, Jeff Hammond <jeff.science(a)gmail.com> wrote: >> Is this because MPICH can't spawn processes except through >> PMI/Hydra/whatever and you - justifiably - do not want to implement a >> special case workaround using execvp() directly? > > The new process needs a way to query for business cards and such, for which mpiexec is needed. So the current design of singleton init is to spawn mpiexec and connect to it. That’s a mess and is not clearly specified by the PMI spec. My suggestion was to execvp the executable “mpiexec” with the right arguments to relaunch the application again, but there are some problems with that: Thanks for the info. I have no problem accepting the terms of MPI-3, section 8.7, paragraph 1, and forcing myself to use mpiexec. Jeff -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Re: [mpich-devel] proper way to detect if program launched by mpiexec?
by Balaji, Pavan 20 Oct '14

20 Oct '14

> On Oct 20, 2014, at 12:43 PM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > Is this because MPICH can't spawn processes except through > PMI/Hydra/whatever and you - justifiably - do not want to implement a > special case workaround using execvp() directly? The new process needs a way to query for business cards and such, for which mpiexec is needed. So the current design of singleton init is to spawn mpiexec and connect to it. That’s a mess and is not clearly specified by the PMI spec. My suggestion was to execvp the executable “mpiexec” with the right arguments to relaunch the application again, but there are some problems with that: http://trac.mpich.org/projects/mpich/ticket/1074 — Pavan -- Pavan Balaji ✉️ http://www.mcs.anl.gov/~balaji

1 0

Re: [mpich-devel] proper way to detect if program launched by mpiexec?
by Rob Latham 20 Oct '14

20 Oct '14

On 10/20/2014 11:34 AM, Jeff Hammond wrote: > I wrote a trivial program that uses MPI_Comm_spawn. It works when I > launch with mpiexec but not if I run directly. > > 1) Is this a bug in MPICH? http://trac.mpich.org/projects/mpich/ticket/2175 ==rob > 2) What is the proper way for the program to detect how the program > was launched? I would like to provide a helpful warning message... > > Thanks! > > Jeff > > jrhammon-mac01:comm_spawn jrhammon$ ./parent > I am 0 of 1 for the parent program. > [mpiexec(a)jrhammon-mac01.local] match_arg > (../../../../src/pm/hydra/utils/args/args.c:159): unrecognized > argument pmi_args > [mpiexec(a)jrhammon-mac01.local] HYDU_parse_array > (../../../../src/pm/hydra/utils/args/args.c:174): argument matching > returned error > [mpiexec(a)jrhammon-mac01.local] parse_args > (../../../../src/pm/hydra/ui/mpich/utils.c:1596): error parsing input > array > [mpiexec(a)jrhammon-mac01.local] HYD_uii_mpx_get_parameters > (../../../../src/pm/hydra/ui/mpich/utils.c:1648): unable to parse user > arguments > [mpiexec(a)jrhammon-mac01.local] main > (../../../../src/pm/hydra/ui/mpich/mpiexec.c:153): error parsing > parameters > ^C > > jrhammon-mac01:comm_spawn jrhammon$ mpiexec -n 1 ./parent > I am 0 of 1 for the parent program. > I am 0 of 1 for the child program. > > ==> parent.c <== > #include <stdio.h> > #include <mpi.h> > > int main(int argc, char * argv[]) > { > MPI_Init(&argc,&argv); > > int size, rank; > MPI_Comm_size(MPI_COMM_WORLD, &size); > MPI_Comm_rank(MPI_COMM_WORLD, &rank); > > printf("I am %d of %d for the parent program.\n", rank, size); > > MPI_Comm intercomm; > int errors[size]; > MPI_Comm_spawn( (char*)"./child", MPI_ARGV_NULL, 1, MPI_INFO_NULL, > 0 /* root */, MPI_COMM_WORLD, &intercomm, errors); > > MPI_Finalize(); > return 0; > } > > ==> child.c <== > #include <stdio.h> > #include <mpi.h> > > int main(int argc, char * argv[]) > { > MPI_Init(&argc,&argv); > > int size, rank; > MPI_Comm_size(MPI_COMM_WORLD, &size); > MPI_Comm_rank(MPI_COMM_WORLD, &rank); > > printf("I am %d of %d for the child program.\n", rank, size); > > MPI_Finalize(); > return 0; > } > > ==> Makefile <== > CC = mpicc > CFLAGS = -g -O2 -Wall -std=c99 > > LD = $(CC) > LDFLAGS = $(COPT) > > LIBS = > > TESTS= parent child > > all: $(TESTS) > > %: %.o > $(CC) $(CFLAGS) $< $(LIBS) -o $@ > > clean: > -rm -f *.o > -rm -f $(TESTS) > -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA

1 0

Re: [mpich-devel] proper way to detect if program launched by mpiexec?
by Jeff Hammond 20 Oct '14

20 Oct '14

Is this because MPICH can't spawn processes except through PMI/Hydra/whatever and you - justifiably - do not want to implement a special case workaround using execvp() directly? Jeff On Mon, Oct 20, 2014 at 9:44 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > This is a known issue with MPICH. There’s no portable way to detect this, but for MPICH you can look for PMI variables in the environment. > > — Pavan > >> On Oct 20, 2014, at 11:34 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: >> >> I wrote a trivial program that uses MPI_Comm_spawn. It works when I >> launch with mpiexec but not if I run directly. >> >> 1) Is this a bug in MPICH? >> 2) What is the proper way for the program to detect how the program >> was launched? I would like to provide a helpful warning message... >> >> Thanks! >> >> Jeff >> >> jrhammon-mac01:comm_spawn jrhammon$ ./parent >> I am 0 of 1 for the parent program. >> [mpiexec(a)jrhammon-mac01.local] match_arg >> (../../../../src/pm/hydra/utils/args/args.c:159): unrecognized >> argument pmi_args >> [mpiexec(a)jrhammon-mac01.local] HYDU_parse_array >> (../../../../src/pm/hydra/utils/args/args.c:174): argument matching >> returned error >> [mpiexec(a)jrhammon-mac01.local] parse_args >> (../../../../src/pm/hydra/ui/mpich/utils.c:1596): error parsing input >> array >> [mpiexec(a)jrhammon-mac01.local] HYD_uii_mpx_get_parameters >> (../../../../src/pm/hydra/ui/mpich/utils.c:1648): unable to parse user >> arguments >> [mpiexec(a)jrhammon-mac01.local] main >> (../../../../src/pm/hydra/ui/mpich/mpiexec.c:153): error parsing >> parameters >> ^C >> >> jrhammon-mac01:comm_spawn jrhammon$ mpiexec -n 1 ./parent >> I am 0 of 1 for the parent program. >> I am 0 of 1 for the child program. >> >> ==> parent.c <== >> #include <stdio.h> >> #include <mpi.h> >> >> int main(int argc, char * argv[]) >> { >> MPI_Init(&argc,&argv); >> >> int size, rank; >> MPI_Comm_size(MPI_COMM_WORLD, &size); >> MPI_Comm_rank(MPI_COMM_WORLD, &rank); >> >> printf("I am %d of %d for the parent program.\n", rank, size); >> >> MPI_Comm intercomm; >> int errors[size]; >> MPI_Comm_spawn( (char*)"./child", MPI_ARGV_NULL, 1, MPI_INFO_NULL, >> 0 /* root */, MPI_COMM_WORLD, &intercomm, errors); >> >> MPI_Finalize(); >> return 0; >> } >> >> ==> child.c <== >> #include <stdio.h> >> #include <mpi.h> >> >> int main(int argc, char * argv[]) >> { >> MPI_Init(&argc,&argv); >> >> int size, rank; >> MPI_Comm_size(MPI_COMM_WORLD, &size); >> MPI_Comm_rank(MPI_COMM_WORLD, &rank); >> >> printf("I am %d of %d for the child program.\n", rank, size); >> >> MPI_Finalize(); >> return 0; >> } >> >> ==> Makefile <== >> CC = mpicc >> CFLAGS = -g -O2 -Wall -std=c99 >> >> LD = $(CC) >> LDFLAGS = $(COPT) >> >> LIBS = >> >> TESTS= parent child >> >> all: $(TESTS) >> >> %: %.o >> $(CC) $(CFLAGS) $< $(LIBS) -o $@ >> >> clean: >> -rm -f *.o >> -rm -f $(TESTS) >> >> -- >> Jeff Hammond >> jeff.science(a)gmail.com >> http://jeffhammond.github.io/ >> _______________________________________________ >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/devel > > -- > Pavan Balaji ✉️ > http://www.mcs.anl.gov/~balaji > > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Re: [mpich-devel] proper way to detect if program launched by mpiexec?
by Balaji, Pavan 20 Oct '14

20 Oct '14

This is a known issue with MPICH. There’s no portable way to detect this, but for MPICH you can look for PMI variables in the environment. — Pavan > On Oct 20, 2014, at 11:34 AM, Jeff Hammond <jeff.science(a)gmail.com> wrote: > > I wrote a trivial program that uses MPI_Comm_spawn. It works when I > launch with mpiexec but not if I run directly. > > 1) Is this a bug in MPICH? > 2) What is the proper way for the program to detect how the program > was launched? I would like to provide a helpful warning message... > > Thanks! > > Jeff > > jrhammon-mac01:comm_spawn jrhammon$ ./parent > I am 0 of 1 for the parent program. > [mpiexec(a)jrhammon-mac01.local] match_arg > (../../../../src/pm/hydra/utils/args/args.c:159): unrecognized > argument pmi_args > [mpiexec(a)jrhammon-mac01.local] HYDU_parse_array > (../../../../src/pm/hydra/utils/args/args.c:174): argument matching > returned error > [mpiexec(a)jrhammon-mac01.local] parse_args > (../../../../src/pm/hydra/ui/mpich/utils.c:1596): error parsing input > array > [mpiexec(a)jrhammon-mac01.local] HYD_uii_mpx_get_parameters > (../../../../src/pm/hydra/ui/mpich/utils.c:1648): unable to parse user > arguments > [mpiexec(a)jrhammon-mac01.local] main > (../../../../src/pm/hydra/ui/mpich/mpiexec.c:153): error parsing > parameters > ^C > > jrhammon-mac01:comm_spawn jrhammon$ mpiexec -n 1 ./parent > I am 0 of 1 for the parent program. > I am 0 of 1 for the child program. > > ==> parent.c <== > #include <stdio.h> > #include <mpi.h> > > int main(int argc, char * argv[]) > { > MPI_Init(&argc,&argv); > > int size, rank; > MPI_Comm_size(MPI_COMM_WORLD, &size); > MPI_Comm_rank(MPI_COMM_WORLD, &rank); > > printf("I am %d of %d for the parent program.\n", rank, size); > > MPI_Comm intercomm; > int errors[size]; > MPI_Comm_spawn( (char*)"./child", MPI_ARGV_NULL, 1, MPI_INFO_NULL, > 0 /* root */, MPI_COMM_WORLD, &intercomm, errors); > > MPI_Finalize(); > return 0; > } > > ==> child.c <== > #include <stdio.h> > #include <mpi.h> > > int main(int argc, char * argv[]) > { > MPI_Init(&argc,&argv); > > int size, rank; > MPI_Comm_size(MPI_COMM_WORLD, &size); > MPI_Comm_rank(MPI_COMM_WORLD, &rank); > > printf("I am %d of %d for the child program.\n", rank, size); > > MPI_Finalize(); > return 0; > } > > ==> Makefile <== > CC = mpicc > CFLAGS = -g -O2 -Wall -std=c99 > > LD = $(CC) > LDFLAGS = $(COPT) > > LIBS = > > TESTS= parent child > > all: $(TESTS) > > %: %.o > $(CC) $(CFLAGS) $< $(LIBS) -o $@ > > clean: > -rm -f *.o > -rm -f $(TESTS) > > -- > Jeff Hammond > jeff.science(a)gmail.com > http://jeffhammond.github.io/ > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel -- Pavan Balaji ✉️ http://www.mcs.anl.gov/~balaji

1 0

proper way to detect if program launched by mpiexec?
by Jeff Hammond 20 Oct '14

20 Oct '14

I wrote a trivial program that uses MPI_Comm_spawn. It works when I launch with mpiexec but not if I run directly. 1) Is this a bug in MPICH? 2) What is the proper way for the program to detect how the program was launched? I would like to provide a helpful warning message... Thanks! Jeff jrhammon-mac01:comm_spawn jrhammon$ ./parent I am 0 of 1 for the parent program. [mpiexec(a)jrhammon-mac01.local] match_arg (../../../../src/pm/hydra/utils/args/args.c:159): unrecognized argument pmi_args [mpiexec(a)jrhammon-mac01.local] HYDU_parse_array (../../../../src/pm/hydra/utils/args/args.c:174): argument matching returned error [mpiexec(a)jrhammon-mac01.local] parse_args (../../../../src/pm/hydra/ui/mpich/utils.c:1596): error parsing input array [mpiexec(a)jrhammon-mac01.local] HYD_uii_mpx_get_parameters (../../../../src/pm/hydra/ui/mpich/utils.c:1648): unable to parse user arguments [mpiexec(a)jrhammon-mac01.local] main (../../../../src/pm/hydra/ui/mpich/mpiexec.c:153): error parsing parameters ^C jrhammon-mac01:comm_spawn jrhammon$ mpiexec -n 1 ./parent I am 0 of 1 for the parent program. I am 0 of 1 for the child program. ==> parent.c <== #include <stdio.h> #include <mpi.h> int main(int argc, char * argv[]) { MPI_Init(&argc,&argv); int size, rank; MPI_Comm_size(MPI_COMM_WORLD, &size); MPI_Comm_rank(MPI_COMM_WORLD, &rank); printf("I am %d of %d for the parent program.\n", rank, size); MPI_Comm intercomm; int errors[size]; MPI_Comm_spawn( (char*)"./child", MPI_ARGV_NULL, 1, MPI_INFO_NULL, 0 /* root */, MPI_COMM_WORLD, &intercomm, errors); MPI_Finalize(); return 0; } ==> child.c <== #include <stdio.h> #include <mpi.h> int main(int argc, char * argv[]) { MPI_Init(&argc,&argv); int size, rank; MPI_Comm_size(MPI_COMM_WORLD, &size); MPI_Comm_rank(MPI_COMM_WORLD, &rank); printf("I am %d of %d for the child program.\n", rank, size); MPI_Finalize(); return 0; } ==> Makefile <== CC = mpicc CFLAGS = -g -O2 -Wall -std=c99 LD = $(CC) LDFLAGS = $(COPT) LIBS = TESTS= parent child all: $(TESTS) %: %.o $(CC) $(CFLAGS) $< $(LIBS) -o $@ clean: -rm -f *.o -rm -f $(TESTS) -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

new tutorial books on MPI
by Rajeev Thakur 13 Oct '14

13 Oct '14

Since many of you are interested in MPI, I wanted to bring to your attention two new tutorial books on MPI: 1. Using MPI: Portable Parallel Programming with the Message-Passing Interface, 3rd edition by William Gropp, Ewing Lusk, and Anthony Skjellum This is an updated version of their earlier book, and covers basic MPI. http://www.amazon.com/Using-MPI-Programming-Message-Passing-Engineering/dp/… 2. Using Advanced MPI: Modern Features of the Message-Passing Interface by William Gropp, Torsten Hoefler, Rajeev Thakur, and Ewing Lusk This is a new book on advanced features of MPI, including the new features in MPI-3. http://www.amazon.com/Using-Advanced-MPI-Message-Passing-Engineering/dp/026… Rajeev

1 0

Re: [mpich-devel] mpich -mcmodel=medium
by Junchao Zhang 08 Oct '14

08 Oct '14

Did you build mpich with Intel compilers? Have you tried to use -mcmodel=large for a non-mpi program and compiled it with ifort? --Junchao Zhang On Wed, Oct 8, 2014 at 8:02 AM, WeiGuang He <c212heweiguang(a)gmail.com> wrote: > Hello, madam/sir > It seems that "-mcmodel=medium" is not supported by mpif90 (I installed > the latest mpich). > How could I use large array? > > mpif90 mtestf90.f90 winfencef90.f90 -o abc -assume byterecl -shared-intel > -mcmodel=large -fPIC > > %% The output is > > ./abc: symbol lookup error: ./abc: undefined symbol: > __intel_new_feature_proc_init > > ./abc: symbol lookup error: ./abc: undefined symbol: > __intel_new_feature_proc_init > > > > =================================================================================== > > = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES > > = PID 27182 RUNNING AT lgmiris > > = EXIT CODE: 127 > > = CLEANING UP REMAINING PROCESSES > > = YOU CAN IGNORE THE BELOW CLEANUP MESSAGES > > > =================================================================================== > > Best Regards > Weiguang HE > > _______________________________________________ > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/devel >

1 0