Compile pnetcdf library at a new Linux cluster at NERSC

[Rob Ross]

Hi,

Not at this time, no.  You'll have to use another compiler/MPI library.

We will look into options for allowing PnetCDF to operate without those 
calls.  Of course, MPI-IO itself is part of MPI-2, so we need a 
significant fraction of MPI-2 :)...

Rob

Yu-Heng Tseng wrote:
> Yes. Only part of MPI-2 is implemented in the compiler.
> Both MPI_Type_dup and MPI_Request_get_status 
> are not supported in the Linux cluster at NERSC (jacquard).
> Is it possible to test the library without both functionalities?
> 
> ---------------------------------------------------
> Yu-Heng Tseng
> 
> Computational Research Division
> Lawrence Berkeley National Laboratory
> One Cyclotron Rd, MS: 50F-1650
> Berkeley, CA94720
> YHTseng at lbl.gov
> 510.495.2904
> 
> ----- Original Message -----
> From: William Gropp <gropp at mcs.anl.gov>
> Date: Friday, December 9, 2005 5:02 am
> Subject: Re: Compile pnetcdf library at a new Linux cluster at NERSC
> 
>> At 02:44 AM 12/9/2005, Yu-Heng Tseng wrote:
>>> Hi,
>>>
>>> I couldn't compile the library using "pathf90/pathcc" compiler. It
>>> gives me the following error during "gmake". It seems "MPI_Type_dup"
>>> is not defined. A bug in the compiler or the library?
>> Both MPI_Type_dup and MPI_Request_get_status are part of MPI-2.  
>> What 
>> version of MPI does your system support?
>> It sounds like we should add tests for these routines to the 
>> configure step.
>>
>> Bill
>>
>>
>>> mpicc -o ncmpigen -g -O2 -D_LARGEFILE64_SOURCE -
> D_FILE_OFFSET_BITS=64
>>> main.o load.o ncgentab.o escapes.o getfill.o init.o genlib.o -
>>> L../../../src/lib -lpnetcdf
>>> ../../../src/lib/libpnetcdf.a(mpinetcdf.o)(.text+0x1041d): In 
>> function>`ncmpi_request_get_status':
>>> /scratch/scratchdirs/yhtseng/cam_PIO/CAMMODELS/parallel-netcdf-
>>> 1.0.1/src/lib/mpinetcdf.c:8218: undefined reference to
>>> `MPI_Request_get_status'
>>> ../../../src/lib/libpnetcdf.a(mpinetcdf.o)(.text+0x10cf4): In 
>> function>`ncmpi_iget_var1':
>>> /scratch/scratchdirs/yhtseng/cam_PIO/CAMMODELS/parallel-netcdf-
>>> 1.0.1/src/lib/mpinetcdf.c:8528: undefined reference to 
> `MPI_Type_dup'
>>> ---------------------------------------------------
>>> Yu-Heng Tseng
>>>
>>> Computational Research Division
>>> Lawrence Berkeley National Laboratory
>>> One Cyclotron Rd, MS: 50F-1650
>>> Berkeley, CA94720
>>> YHTseng at lbl.gov
>>> 510.495.2904
>>>
>>> ----- Original Message -----
>>> From: Jim Edwards <jedwards at ucar.edu>
>>> Date: Tuesday, December 6, 2005 0:40 am
>>> Subject: Re: Compile pnetcdf library in IBM SP at NCAR
>>>
>>>> There is a pnetcdf version installed in the /contrib file system
>>>> on the NCAR
>>>> IBM machines.  I have already reported this problem with 
> configure
>>>> to the
>>>> parallel netcdf developers.   They have provided a patched 
>> release to
>>>> address the problem.
>>>>
>>>> Jim Edwards
>>>> IBM applications support
>>>> NCAR
>>>>
>>>> On 12/6/05, Yu-Heng Tseng <YHTseng at lbl.gov> wrote:
>>>>> Hi,
>>>>>
>>>>> I couldn't compile the library in NCAR's IBM SP.
>>>>> It works fine in NERSC's IBM SP machine.
>>>>> I am not sure if this is compiler's or the library's problem.
>>>>> When I tried to "./configure" the
>>>>> it stops at the following line
>>>>>
>>>>> checking for Fortran-equivalent to netCDF "byte"... byte
>>>>> checking for Fortran-equivalent to netCDF "short"... integer*2
>>>>> checking for Fortran "integer*"... no
>>>>> configure: error: F77 does not support "integer*"
>>>>>
>>>>> The "integer*" is actually "integer8" in a successful check.
>>>>>
>>>>> Anyone has clues to solve the problem? The c compiler at NCAR
>>>>> seems different from that at NERSC.
>>>>> NCAR uses  VisualAge C++ Professional / C for AIX Compiler,
>>>> Version 6
>>>>> Thanks a lot for your help!
>>>>>
>>>>> Yu-heng
>>>>> ---------------------------------------------------
>>>>> Yu-Heng Tseng
>>>>>
>>>>> Computational Research Division
>>>>> Lawrence Berkeley National Laboratory
>>>>> One Cyclotron Rd, MS: 50F-1650
>>>>> Berkeley, CA94720
>>>>> YHTseng at lbl.gov
>>>>> 510.495.2904
>>>>>
>>>>>
>> William Gropp
>> http://www.mcs.anl.gov/~gropp 
>>
>>
>