[mpich-discuss] Compiling mpich2 with icc

[Rajeev Thakur]

I personally haven't measured, but I guess it will be more a function of
whether the application's compute part performs better with icc than gcc
(rather than MPI). 


> -----Original Message-----
> From: owner-mpich-discuss at mcs.anl.gov 
> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Robert Kubrick
> Sent: Wednesday, September 24, 2008 12:49 PM
> To: mpich-discuss at mcs.anl.gov
> Subject: Re: [mpich-discuss] Compiling mpich2 with icc
> 
> ok, thanks. Did you notice any substantial performance 
> improvement compiling mpich2 with icc versus g++?
> 
> On Sep 24, 2008, at 1:36 PM, Rajeev Thakur wrote:
> 
> > I am able to reproduce the problem if I use the -fast 
> option to icc. 
> > But without -fast it works. Wonder if it is a problem with 
> icc. I used 
> > version 10.1.
> >
> > Rajeev
> >
> >> -----Original Message-----
> >> From: owner-mpich-discuss at mcs.anl.gov 
> >> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of 
> Robert Kubrick
> >> Sent: Wednesday, September 24, 2008 11:54 AM
> >> To: mpich-discuss at mcs.anl.gov
> >> Subject: Re: [mpich-discuss] Compiling mpich2 with icc
> >>
> >> Compiling with ipcp returns this error:
> >>
> >> make[4]: Entering directory `/home/pts/archives/mpich2-1.1.0a1/src/
> >> mpid/common/locks'
> >>    CC              mpidu_process_locks.c
> >> ../../../include/pmi.h(740): error: parameter type 
> involves pointer 
> >> to array of unknown bound
> >>    int PMI_Args_to_keyval(int *argcp, char *((*argvp)[]), 
> >> PMI_keyval_t **keyvalp, int *size);
> >>                                       ^
> >>
> >> compilation aborted for mpidu_process_locks.c (code 2)
> >> make[4]: *** [mpidu_process_locks.o] Error 2
> >> make[4]: Leaving directory `/home/pts/archives/mpich2-1.1.0a1/src/
> >> mpid/common/locks'
> >> make[3]: *** [all-redirect] Error 2
> >> make[3]: Leaving directory `/home/pts/archives/mpich2-1.1.0a1/src/
> >> mpid/common'
> >> make[2]: *** [all-redirect] Error 2
> >> make[2]: Leaving directory
> >> `/home/pts/archives/mpich2-1.1.0a1/src/mpid'
> >> make[1]: *** [all-redirect] Error 2
> >> make[1]: Leaving directory `/home/pts/archives/mpich2-1.1.0a1/src'
> >> make: *** [all-redirect] Error 2
> >>
> >>
> >> I tried nm -a libmpich.a and it didn't return the required 
> symbols. I 
> >> am not too familiar with nm, but I am not sure this symbol table is
> >> right:
> >>
> >> me at harper1:~/archives/mpich2-1.1.0a1> nm -a lib/libmpich.a
> >> |grep MPIR_Version_date me at harper1:~/archives/mpich2-1.1.0a1>
> >> nm -a lib/libmpich.a
> >>
> >> mpidu_process_locks.o:
> >> 00000000 a .strtab
> >> 00000000 a .symtab
> >> 00000000 n __ildata
> >>           w __ildata_included
> >> 00000000 a mpidu_process_locks.c
> >>
> >> mpidu_atomic_primitives.o:
> >> 00000000 a .strtab
> >> 00000000 a .symtab
> >> 00000000 n __ildata
> >>           w __ildata_included
> >> 00000000 a mpidu_atomic_primitives.c
> >>
> >> mpidu_queue.o:
> >> 00000000 a .strtab
> >> 00000000 a .symtab
> >> 00000000 n __ildata
> >>           w __ildata_included
> >> 00000000 a mpidu_queue.c
> >>
> >> dataloop.o:
> >> 00000000 a .strtab
> >> 00000000 a .symtab
> >> 00000000 n __ildata
> >>           w __ildata_included
> >> 00000000 a dataloop.c
> >> ...
> >> ...
> >> ...
> >>
> >> On Sep 24, 2008, at 12:00 PM, Rajeev Thakur wrote:
> >>
> >>> Can you do an "nm" on libmpich.a and see if those symbols are in 
> >>> there.
> >>> Also, use icpc as the C++ compiler.
> >>>
> >>> Rajeev
> >>>
> >>>> -----Original Message-----
> >>>> From: owner-mpich-discuss at mcs.anl.gov 
> >>>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of
> >> Robert Kubrick
> >>>> Sent: Wednesday, September 24, 2008 10:55 AM
> >>>> To: mpich-discuss at mcs.anl.gov
> >>>> Subject: Re: [mpich-discuss] Compiling mpich2 with icc
> >>>>
> >>>> initinfo.c compiles fine, but there are several warnings during 
> >>>> compilation. I am attaching here the complete make log.
> >>>>
> >>>>
> >>>>
> >>>
> >>
> >>
> >
> 
>