[SIESTA-L] Compile problem : redeclared MPI data types
Hello, I¹m trying to compile Siesta 2.0.1 for parallel processing using MPI. My system is an AMD Opteron cluster with InfiniPath interconnects. The compiler is the PathScale compiler and I¹m using the ACML 3.0.0 libraries. The BLACS and scalapack libraries are installed and tested. When I try to compile Siesta I¹m given errors about redeclared attributes. (See below) I have also attached my arch.make file. This problem is identical to that posted June2007 by Eb Na ( http://www.mail-archive.com/siesta-l@listserv.uam.es/msg00844.html). Unfortunately there was never a response or follow-up to post the solution. I¹ve tried making changes to the mpi.F and other source files in an attempt to understand the origin of this problem, but have been unsuccessful. Any help or suggestions would be very welcome. Thank you for your time with this problem. Sincerely, Scott Beckman Assistant Professor Department of Material Science and Engineering Iowa State University sbeck...@iastate.edu + + OUTPUT FROM COMPILE + + hpc4~/Software/SIESTA/siesta-2.0.1/Src> hpc4~/Software/SIESTA/siesta-2.0.1/Src> make Compilation architecture to be used: pathf90-2.0-mpi If this is not what you want, create the right arch.make file using the models in Sys Hit ^C to abort... ==> Incorporating information about present compilation (compiler and flags) make[1]: Entering directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src' f90 -c -DMPI -I/usr/include -L/usr/lib compinfo.F90 make[1]: Leaving directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src' #...@rm -f compinfo.F90 make[1]: Entering directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src/MPI' f90 -c -I/usr/include -DMPI -I/usr/include -L/usr/lib mpi.F MODULE MPI_SIESTA ^ pathf95-855 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 1, Column = 14 The compiler has detected errors in module "MPI_SIESTA". No module information file will be created for this module. & DAT_single => MPI_real, & ^ pathf95-197 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 36, Column = 7 Unexpected syntax: "ONLY or use-name" was expected but found " ". integer, parameter :: MPI_real = DAT_single ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 71, Column = 31 "MPI_REAL" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 71, Column = 55 "DAT_SINGLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_2real = DAT_2single ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 72, Column = 31 "MPI_2REAL" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 72, Column = 55 "DAT_2SINGLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_double_precision = DAT_double ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 73, Column = 31 "MPI_DOUBLE_PRECISION" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 73, Column = 55 "DAT_DOUBLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_2double_precision = DAT_2double ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 74, Column = 31 "MPI_2DOUBLE_PRECISION" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 74, Column = 55 "DAT_2DOUBLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_complex = DAT_complex ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 75, Column = 31 "MPI_COMPLEX" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 75, Column = 55 "DAT_COMPLEX" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_double_complex= DAT_dcomplex ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 76, Column = 31 "MPI_DOUBLE_COMPLEX"
Re: [SIESTA-L] Compile problem : redeclared MPI data types
There is some problem with how PathScale handles the modules in Src/MPI/mpi.F If I replace: USE MPI__INCLUDE , #ifdef OLD_CRAY & DAT_single => MPI_real, & & DAT_2single => MPI_2real,& & DAT_double => MPI_real, & & DAT_2double => MPI_2real,& & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_complex & #else & DAT_single => MPI_real, & & DAT_2single => MPI_2real,& & DAT_double => MPI_double_precision, & & DAT_2double => MPI_2double_precision,& & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_double_complex & #endif With: USE MPI__INCLUDE, ONLY : DAT_single => MPI_real, & DAT_2single => MPI_2real, & DAT_double => MPI_double_precision, & DAT_2double => MPI_2double_precision, & DAT_complex => MPI_complex, & DAT_dcomplex => MPI_double_complex USE MPI__INCLUDE The modules and objects all compile correctly. Of course then the linking doesn¹t work correctly due to: ³Unexpected references of formal parameter² I¹m not at all familiar with the PathScale compiler (this is my first use of it) There must be some reason that the line-wrap used above does not work, but my F77 style does... However, I¹m confused by this because lower in mpi.F the & & does appear to work as a line-wrap. Is there a flag that I¹m missing? On 7/1/09 9:14 AM, "Scott Beckman" wrote: > Hello, > > I¹m trying to compile Siesta 2.0.1 for parallel processing using MPI. My > system is an AMD Opteron cluster with InfiniPath interconnects. The compiler > is the PathScale compiler and I¹m using the ACML 3.0.0 libraries. The BLACS > and scalapack libraries are installed and tested. > > When I try to compile Siesta I¹m given errors about redeclared attributes. > (See below) I have also attached my arch.make file. > > This problem is identical to that posted June2007 by Eb Na ( > http://www.mail-archive.com/siesta-l@listserv.uam.es/msg00844.html). > Unfortunately there was never a response or follow-up to post the solution. > > I¹ve tried making changes to the mpi.F and other source files in an attempt to > understand the origin of this problem, but have been unsuccessful. Any help > or suggestions would be very welcome. > > Thank you for your time with this problem. > Sincerely, > > > Scott Beckman > > Assistant Professor > Department of Material Science and Engineering > Iowa State University > sbeck...@iastate.edu > > + > + > OUTPUT FROM COMPILE > + > + > > hpc4~/Software/SIESTA/siesta-2.0.1/Src> > hpc4~/Software/SIESTA/siesta-2.0.1/Src> make > > Compilation architecture to be used: pathf90-2.0-mpi > If this is not what you want, create the right > arch.make file using the models in Sys > > Hit ^C to abort... > > ==> Incorporating information about present compilation (compiler and flags) > make[1]: Entering directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src' > f90 -c -DMPI -I/usr/include -L/usr/lib compinfo.F90 > make[1]: Leaving directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src' > #...@rm -f compinfo.F90 > > make[1]: Entering directory > `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src/MPI' > f90 -c -I/usr/include -DMPI -I/usr/include -L/usr/lib mpi.F > > MODULE MPI_SIESTA > ^ > pathf95-855 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 1, Column = 14 > The compiler has detected errors in module "MPI_SIESTA". No module > information file will be created for this module. > > & DAT_single => MPI_real, & > ^ > pathf95-197 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 36, Column = 7 > Unexpected syntax: "ONLY or use-name" was expected but found " ". > > integer, parameter :: MPI_real = DAT_single > ^ > pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 71, Column = 31 > "MPI_REAL" has been use associated, therefore it must not be redeclared with > the INTEGER(KIND=4) attribute. > ^ > pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 71, Column = 55 > "DAT_SINGLE" is used in a constant expression, therefore it must be a > constant. > > integer, parameter :: MPI_2real = DAT_2single > ^ > pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 72, Column = 31 > "MPI_2REAL" has been u
Re: [SIESTA-L] Compile problem : redeclared MPI data types
Dear Scott: Edit the mpi.F file in the Src/MPI subdirectory, and include the "USE MPI_INCLUDE, " inside the two blocks of the preprocessor. The final lines should look like: #ifdef OLD_CRAY USE MPI__INCLUDE , & DAT_single => MPI_real, & & DAT_2single => MPI_2real, & & DAT_double => MPI_real, & & DAT_2double => MPI_2real, & & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_complex & #else USE MPI__INCLUDE , & DAT_single => MPI_real, & & DAT_2single => MPI_2real, & & DAT_double => MPI_double_precision, & & DAT_2double => MPI_2double_precision, & & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_double_complex & #endif It works for me, and I am using essentially the same architecture, compiler, and libraries as you. Hope this helps, Javier Beckman escribió: Hello, I’m trying to compile Siesta 2.0.1 for parallel processing using MPI. My system is an AMD Opteron cluster with InfiniPath interconnects. The compiler is the PathScale compiler and I’m using the ACML 3.0.0 libraries. The BLACS and scalapack libraries are installed and tested. When I try to compile Siesta I’m given errors about redeclared attributes. (See below) I have also attached my arch.make file. This problem is identical to that posted June2007 by Eb Na ( http://www.mail-archive.com/siesta-l@listserv.uam.es/msg00844.html). Unfortunately there was never a response or follow-up to post the solution. I’ve tried making changes to the mpi.F and other source files in an attempt to understand the origin of this problem, but have been unsuccessful. Any help or suggestions would be very welcome. Thank you for your time with this problem. Sincerely, Scott Beckman Assistant Professor Department of Material Science and Engineering Iowa State University sbeck...@iastate.edu + + OUTPUT FROM COMPILE + + hpc4~/Software/SIESTA/siesta-2.0.1/Src> hpc4~/Software/SIESTA/siesta-2.0.1/Src> make Compilation architecture to be used: pathf90-2.0-mpi If this is not what you want, create the right arch.make file using the models in Sys Hit ^C to abort... ==> Incorporating information about present compilation (compiler and flags) make[1]: Entering directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src' f90 -c -DMPI -I/usr/include -L/usr/lib compinfo.F90 make[1]: Leaving directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src' #...@rm -f compinfo.F90 make[1]: Entering directory `/home/sbeckman/Software/SIESTA/siesta-2.0.1/Src/MPI' f90 -c -I/usr/include -DMPI -I/usr/include -L/usr/lib mpi.F MODULE MPI_SIESTA ^ pathf95-855 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 1, Column = 14 The compiler has detected errors in module "MPI_SIESTA". No module information file will be created for this module. & DAT_single => MPI_real, & ^ pathf95-197 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 36, Column = 7 Unexpected syntax: "ONLY or use-name" was expected but found " ". integer, parameter :: MPI_real = DAT_single ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 71, Column = 31 "MPI_REAL" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 71, Column = 55 "DAT_SINGLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_2real = DAT_2single ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 72, Column = 31 "MPI_2REAL" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 72, Column = 55 "DAT_2SINGLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_double_precision = DAT_double ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 73, Column = 31 "MPI_DOUBLE_PRECISION" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 73, Column = 55 "DAT_DOUBLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_2double_precision = DAT_2double ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 74, Column = 31 "MPI_2DOUBLE_PRECISION" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 74, Column = 55 "DAT_2DOUBLE" is used in a constant expression, therefore it must be a constant. integer, parameter :: MPI_complex = DAT_complex ^ pathf95-922 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 75, Column = 31 "MPI_COMPLEX" has been use associated, therefore it must not be redeclared with the INTEGER(KIND=4) attribute. ^ pathf95-868 f90: ERROR MPI_SIESTA, File = mpi.F, Line = 75, Column = 55 "DAT_COMPLEX" is used in a constant expression, therefore it must be a const
Re: [SIESTA-L] Compile problem : redeclared MPI data types
Hello Javier, Thanks for your suggestion. Yes, this does solve the compile problem. But when the build process gets to the final step, the linker still complains about unexpected references (see below). This problem must not be limited to the mpi.F module but could be anywhere. Is there a flag to make the linker more verbose? I don't know where to even begin looking for the cause of this problem. Would you be willing to post your arch.make file so I could compare? I've made a few changes to mine (see below) I don't see any obvious mistakes, but that doesn't mean that there aren't any. Thanks for your help. Best, Scott + + OUTPUT FROM COMPILE + + ### Compiler Error during Inline processing phase: ### Unexpected references of formal parameter IPA processing aborted f90 INTERNAL ERROR: /opt/pathscale/lib/2.4/ipa_link returned non-zero status 1 + + MY arch.make FILE + + # SIESTA_ARCH=pathf90-2.0-mpi # FC=f90 FC_ASIS=$(FC) # FFLAGS= -m64 -ipa -Ofast -Wl,-R/opt/pathscale/lib/2.4/ FFLAGS_DEBUG= -g LDFLAGS= -m64 -ipa -Ofast # RANLIB=ranlib COMP_LIBS= # NETCDF_LIBS= # /usr/local/netcdf-3.5/lib/pgi/libnetcdf.a NETCDF_INTERFACE=# libnetcdf_f90.a DEFS_CDF=# -DCDF # KINDS=4 8 MPI_INTERFACE=libmpi_f90.a MPI_INCLUDE=/usr/include MPI_LIBS=/usr/lib DEFS_MPI=-DMPI # #LIBS=/work/TG3/LIBS/scalapack-1.8.0/libscalapack.a /work/TG3/LIBS/BLACS/LIB/blacsF77init_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacs_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacsCinit_MPI-LINUX-0.a /opt/acml3.0.0/pathscale64/lib/libacml.a LIBS=/work/TG3/LIBS/scalapack-1.8.0/libscalapack.a /work/TG3/LIBS/BLACS/LIB/blacsF77init_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacs_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacsCinit_MPI-LINUX-0.a SYS=nag DEFS= $(DEFS_CDF) $(DEFS_MPI) # -DGRID_DP # .F.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $(DEFS) $< .f.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $< .F90.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $(DEFS) $< .f90.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $< -Original Message- From: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta on behalf of Javier Junquera Sent: Thu 7/2/2009 3:55 AM To: SIESTA-L@listserv.uam.es Subject: Re: [SIESTA-L] Compile problem : redeclared MPI data types Dear Scott: Edit the mpi.F file in the Src/MPI subdirectory, and include the "USE MPI_INCLUDE, " inside the two blocks of the preprocessor. The final lines should look like: #ifdef OLD_CRAY USE MPI__INCLUDE , & DAT_single => MPI_real, & & DAT_2single => MPI_2real, & & DAT_double => MPI_real, & & DAT_2double => MPI_2real, & & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_complex & #else USE MPI__INCLUDE , & DAT_single => MPI_real, & & DAT_2single => MPI_2real, & & DAT_double => MPI_double_precision, & & DAT_2double => MPI_2double_precision, & & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_double_complex & #endif It works for me, and I am using essentially the same architecture, compiler, and libraries as you. Hope this helps, Javier Beckman escribió: > Hello, > > I'm trying to compile Siesta 2.0.1 for parallel processing using MPI. > My system is an AMD Opteron cluster with InfiniPath interconnects. The > compiler is the PathScale compiler and I'm using the ACML 3.0.0 > libraries. The BLACS and scalapack libraries are installed and tested. > > When I try to compile Siesta I'm given errors about redeclared > attributes. (See below) I have also attached my arch.make file. > > This problem is identical to that posted June2007 by Eb Na ( > http://www.mail-archive.com/siesta-l@listserv.uam.es/msg00844.html). > Unfortunately there was never a response or follow-up to post the > solution. > > I've tried making changes to the mpi.F and other source files in an > attempt to understand the origin of this problem, but have been > unsuccessful. Any help or suggestions would be very welcome. > > Thank you for your time with this problem. > Sincerely, > > > Scott Beckman > > Assistant Professor > Department of Material Science and Engineering > Iowa State University > sbeck...@iastate.edu > > + > + > OUTPUT FROM COMPILE > + > + > > hpc4~/Software/SIESTA/siesta-2.0.1/Src> > hpc4~/Software/SIESTA/siesta-2.0.1/Src> make > > Compilation architecture to be used: pathf90-2.0-mpi > If
Re: [SIESTA-L] Compile problem : redeclared MPI data types
Dear Scott: Please, find enclosed the arch.make you were interested in. I am including both, for the parallel and the serial version. Hope this helps, Javier -Mensaje original- De: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta en nombre de Beckman, Scott P [M S E] Enviado el: jue 02/07/2009 22:09 Para: SIESTA-L@listserv.uam.es Asunto: Re: [SIESTA-L] Compile problem : redeclared MPI data types Hello Javier, Thanks for your suggestion. Yes, this does solve the compile problem. But when the build process gets to the final step, the linker still complains about unexpected references (see below). This problem must not be limited to the mpi.F module but could be anywhere. Is there a flag to make the linker more verbose? I don't know where to even begin looking for the cause of this problem. Would you be willing to post your arch.make file so I could compare? I've made a few changes to mine (see below) I don't see any obvious mistakes, but that doesn't mean that there aren't any. Thanks for your help. Best, Scott + + OUTPUT FROM COMPILE + + ### Compiler Error during Inline processing phase: ### Unexpected references of formal parameter IPA processing aborted f90 INTERNAL ERROR: /opt/pathscale/lib/2.4/ipa_link returned non-zero status 1 + + MY arch.make FILE + + # SIESTA_ARCH=pathf90-2.0-mpi # FC=f90 FC_ASIS=$(FC) # FFLAGS= -m64 -ipa -Ofast -Wl,-R/opt/pathscale/lib/2.4/ FFLAGS_DEBUG= -g LDFLAGS= -m64 -ipa -Ofast # RANLIB=ranlib COMP_LIBS= # NETCDF_LIBS= # /usr/local/netcdf-3.5/lib/pgi/libnetcdf.a NETCDF_INTERFACE=# libnetcdf_f90.a DEFS_CDF=# -DCDF # KINDS=4 8 MPI_INTERFACE=libmpi_f90.a MPI_INCLUDE=/usr/include MPI_LIBS=/usr/lib DEFS_MPI=-DMPI # #LIBS=/work/TG3/LIBS/scalapack-1.8.0/libscalapack.a /work/TG3/LIBS/BLACS/LIB/blacsF77init_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacs_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacsCinit_MPI-LINUX-0.a /opt/acml3.0.0/pathscale64/lib/libacml.a LIBS=/work/TG3/LIBS/scalapack-1.8.0/libscalapack.a /work/TG3/LIBS/BLACS/LIB/blacsF77init_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacs_MPI-LINUX-0.a /work/TG3/LIBS/BLACS/LIB/blacsCinit_MPI-LINUX-0.a SYS=nag DEFS= $(DEFS_CDF) $(DEFS_MPI) # -DGRID_DP # .F.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $(DEFS) $< .f.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $< .F90.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $(DEFS) $< .f90.o: $(FC) -c $(FFLAGS) $(INCFLAGS) $< -Original Message- From: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta on behalf of Javier Junquera Sent: Thu 7/2/2009 3:55 AM To: SIESTA-L@listserv.uam.es Subject: Re: [SIESTA-L] Compile problem : redeclared MPI data types Dear Scott: Edit the mpi.F file in the Src/MPI subdirectory, and include the "USE MPI_INCLUDE, " inside the two blocks of the preprocessor. The final lines should look like: #ifdef OLD_CRAY USE MPI__INCLUDE , & DAT_single => MPI_real, & & DAT_2single => MPI_2real, & & DAT_double => MPI_real, & & DAT_2double => MPI_2real, & & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_complex & #else USE MPI__INCLUDE , & DAT_single => MPI_real, & & DAT_2single => MPI_2real, & & DAT_double => MPI_double_precision, & & DAT_2double => MPI_2double_precision, & & DAT_complex => MPI_complex, & & DAT_dcomplex => MPI_double_complex & #endif It works for me, and I am using essentially the same architecture, compiler, and libraries as you. Hope this helps, Javier Beckman escribió: > Hello, > > I'm trying to compile Siesta 2.0.1 for parallel processing using MPI. > My system is an AMD Opteron cluster with InfiniPath interconnects. The > compiler is the PathScale compiler and I'm using the ACML 3.0.0 > libraries. The BLACS and scalapack libraries are installed and tested. > > When I try to compile Siesta I'm given errors about redeclared > attributes. (See below) I have also attached my arch.make file. > > This problem is identical to that posted June2007 by Eb Na ( > http://www.mail-archive.com/siesta-l@listserv.uam.es/msg00844.html). > Unfortunately there was never a response or follow-up to post the > solution. > > I've tried making changes to the mpi.F and other source files in an > attempt to understand the origin of this problem, but have been > unsuccessful. Any help or suggestions would be very welcome. > > Thank you for your time with this problem. > Sincerely, > > > Scott Beckman > > Assistant Professor > Departme
Re: [SIESTA-L] Compile problem : redeclared MPI data types
Hi, Thank you Javier. The error regarding the "Unexpected references of formal parameter" was solved by passing the DEF -DGRID_DP. Now I have a compiled version of siesta, but I've not yet tested it. (Cross your fingers!) I have several questions about the flags that are being passed, but I don't see an explanation in the internal documentation. For instance, the lines: KINDS=4 8 DEFS= -DWXML_INIT_FIX -DGRID_DP SYS=cpu_time FC=f90 -ff2c-abi /opt/pathscale/etc/f2c-abi/acml-2.0 Is there a difference in accuracy if I modify the KINDS? Do I need both 4 and 8? Is this related to the -DGRID_DP pre-compile flag? What are the options for SYS? I've seen "bsd", "nag", and "cpu_time." Which SYS is needed on my machine and which will give optimal performance? Do I really need to pass a flag to use f2c conversion? Is there a document where I can read about the flags that are available and what they control? Once I've optimized my arch.make, I'll post is here. Best, Scott Beckman Assistant Professor Department of Material Science and Engineering Iowa State University sbeck...@iastate.edu On 7/3/09 8:09 AM, "Junquera Quintana, Francisco Javier" wrote: > > > Dear Scott: > > Please, find enclosed the arch.make you were interested in. > I am including both, for the parallel and the serial version. > > Hope this helps, > > Javier > > > > > -Mensaje original- > De: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta en > nombre de Beckman, Scott P [M S E] > Enviado el: jue 02/07/2009 22:09 > Para: SIESTA-L@listserv.uam.es > Asunto: Re: [SIESTA-L] Compile problem : redeclared MPI data types > > Hello Javier, > > Thanks for your suggestion. Yes, this does solve the compile problem. But > when the build process gets to the final step, the linker still complains > about unexpected references (see below). This problem must not be limited to > the mpi.F module but could be anywhere. Is there a flag to make the linker > more verbose? I don't know where to even begin looking for the cause of this > problem. > > Would you be willing to post your arch.make file so I could compare? I've > made a few changes to mine (see below) I don't see any obvious mistakes, but > that doesn't mean that there aren't any. > > Thanks for your help. > Best, > Scott > > > + > + > OUTPUT FROM COMPILE > + > + > > ### Compiler Error during Inline processing phase: > ### Unexpected references of formal parameter > IPA processing aborted > f90 INTERNAL ERROR: /opt/pathscale/lib/2.4/ipa_link returned non-zero status 1 > > + > + > MY arch.make FILE > + > + > > # > SIESTA_ARCH=pathf90-2.0-mpi > # > FC=f90 > FC_ASIS=$(FC) > # > FFLAGS= -m64 -ipa -Ofast -Wl,-R/opt/pathscale/lib/2.4/ > FFLAGS_DEBUG= -g > LDFLAGS= -m64 -ipa -Ofast > # > RANLIB=ranlib > COMP_LIBS= > # > NETCDF_LIBS= # /usr/local/netcdf-3.5/lib/pgi/libnetcdf.a > NETCDF_INTERFACE=# libnetcdf_f90.a > DEFS_CDF=# -DCDF > # > KINDS=4 8 > > MPI_INTERFACE=libmpi_f90.a > > MPI_INCLUDE=/usr/include > MPI_LIBS=/usr/lib > DEFS_MPI=-DMPI > # > #LIBS=/work/TG3/LIBS/scalapack-1.8.0/libscalapack.a > /work/TG3/LIBS/BLACS/LIB/blacsF77init_MPI-LINUX-0.a > /work/TG3/LIBS/BLACS/LIB/blacs_MPI-LINUX-0.a > /work/TG3/LIBS/BLACS/LIB/blacsCinit_MPI-LINUX-0.a > /opt/acml3.0.0/pathscale64/lib/libacml.a > LIBS=/work/TG3/LIBS/scalapack-1.8.0/libscalapack.a > /work/TG3/LIBS/BLACS/LIB/blacsF77init_MPI-LINUX-0.a > /work/TG3/LIBS/BLACS/LIB/blacs_MPI-LINUX-0.a > /work/TG3/LIBS/BLACS/LIB/blacsCinit_MPI-LINUX-0.a > > SYS=nag > DEFS= $(DEFS_CDF) $(DEFS_MPI) # -DGRID_DP > # > .F.o: > $(FC) -c $(FFLAGS) $(INCFLAGS) $(DEFS) $< > .f.o: > $(FC) -c $(FFLAGS) $(INCFLAGS) $< > .F90.o: > $(FC) -c $(FFLAGS) $(INCFLAGS) $(DEFS) $< > > .f90.o: > $(FC) -c $(FFLAGS) $(INCFLAGS) $< > > > > > -Original Message- > From: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta on > behalf of Javier Junquera > Sent: Thu 7/2/2009 3:55 AM > To: SIESTA-L@listserv.uam.es > Subject: Re: [SIESTA-L] Compile problem : redeclared MPI data types > > Dear Scott: > > Edit the mpi.F file in the Src/MPI subdirectory, > and include the "USE MPI_INCLUDE, " > inside the two blocks of the preprocessor
