Assertion `ncp->head != ((void *)0)' failed.

[Rob Latham]

On Sat, Feb 16, 2013 at 04:45:33AM +0000, Liu, Jaln wrote:
> Hi,
> 
> >The assertion failure indicates there is no successful nonblocking calls before
> the ncmpi_wait_all().
> 
> This is interesting, I traced the program and finally found a bug in my codes. Nothing wrong with the function itself.
> The mistake I made is that I didn't store the requests in the right way, therefore, when it comes to the ncmpi_wait, the function couldn't find the data.

Sounds like we need to come up with a better error message !

==rob

> Thanks,
> 
> Jialin
> ________________________________________
> From: parallel-netcdf-bounces at lists.mcs.anl.gov [parallel-netcdf-bounces at lists.mcs.anl.gov] on behalf of Wei-keng Liao [wkliao at ece.northwestern.edu]
> Sent: Wednesday, February 13, 2013 2:44 PM
> To: parallel-netcdf at lists.mcs.anl.gov
> Subject: Re: Assertion `ncp->head != ((void *)0)' failed.
> 
> The assertion failure indicates there is no successful nonblocking calls before
> the ncmpi_wait_all().
> 
> Can you extract from your codes? I can check if it is PnetCDF internal.
> 
> Wei-keng
> 
> On Feb 13, 2013, at 2:39 PM, Liu, Jaln wrote:
> 
> > Hi Dr. Liao,
> >
> > I printed the error information, as shown in the attachment,  "`ncp->head != ((void *)0)' failed" is all I can get.
> >
> > Jialin
> > ________________________________________
> > From: parallel-netcdf-bounces at lists.mcs.anl.gov [parallel-netcdf-bounces at lists.mcs.anl.gov] on behalf of Wei-keng Liao [wkliao at ece.northwestern.edu]
> > Sent: Wednesday, February 13, 2013 1:52 PM
> > To: parallel-netcdf at lists.mcs.anl.gov
> > Subject: Re: Assertion `ncp->head != ((void *)0)' failed.
> >
> > I suggest you check the return value of your nonblocking call like below.
> >  err = ncmpi_iget_vara( ... );
> >  if (err != NC_NOERR)
> >      printf("Error: %s\n", ncmpi_strerror(err));
> >
> > You can use as many ncmpi_iget_vara() calls to the same/different variables as you like.
> >
> > I also suggest you to give a try to the latest SVN source. It has a new improvement
> > for the same-variable cases.
> > See instructions in http://trac.mcs.anl.gov/projects/parallel-netcdf/wiki/Download
> >
> > Wei-keng
> >
> > On Feb 13, 2013, at 1:43 PM, Liu, Jaln wrote:
> >
> >> Hi Dr. Liao,
> >>
> >> Thank you. Other ncmpi call works fine, like  ncmpi_get_vara_float and ncmpi_get_vara_float_all,
> >>
> >>> Also, I am seeing this line in your codes:
> >>>      nb_temp_in[rank]=calloc(varasize,sizeof(float));
> >>
> >> Yes, it should be `nb_temp'.
> >>
> >> I wonder can we use ncmpi_iget_vara() on the same variable but different starts&lengths. From the sample program, it is used for multiple variables.
> >>
> >> Jialin
> >> ________________________________________
> >> From: parallel-netcdf-bounces at lists.mcs.anl.gov [parallel-netcdf-bounces at lists.mcs.anl.gov] on behalf of Wei-keng Liao [wkliao at ece.northwestern.edu]
> >> Sent: Wednesday, February 13, 2013 1:28 PM
> >> To: parallel-netcdf at lists.mcs.anl.gov
> >> Subject: Re: Assertion `ncp->head != ((void *)0)' failed.
> >>
> >> Hi, Jialin,
> >>
> >> I suspect it is because the call to ncmpi_iget_vara() failed.
> >> Please check the return values of all your ncmpi_ calls.
> >> It may tell you if you got an error before you would like to continue.
> >>
> >> Also, I am seeing this line in your codes:
> >>>      nb_temp_in[rank]=calloc(varasize,sizeof(float));
> >>
> >> But you are using nb_temp[rank] in ncmpi_iget_vara()
> >>
> >>
> >> Wei-keng
> >>
> >> On Feb 13, 2013, at 12:13 PM, Liu, Jaln wrote:
> >>
> >>> Hi,
> >>>
> >>> When I run the codes below, I got an error " Assertion `ncp->head != ((void *)0)' failed."
> >>> (I looked into the source codes, I saw a operation of checking Id, seems each request should be checked and matched with the file handle? I don't know the reason)
> >>>
> >>>      float ** nb_temp;
> >>>      nb_temp=malloc(nprocs*sizeof(float *));
> >>>      int * request=calloc(nprocs, sizeof(int));
> >>>      int * status=calloc(nprocs,sizeof(int));
> >>>      mpi_start[1]=rank*5;//starts on other dimensions are specified to 0.
> >>>      int varasize=100*10*NLAT*NLON;
> >>>      nb_temp_in[rank]=calloc(varasize,sizeof(float));
> >>>
> >>>      ncmpi_iget_vara(ncid, temp_varid, mpi_start,mpi_count,nb_temp[rank],varasize,MPI_FLOAT,&(request[rank]);
> >>>
> >>>      ncmpi_wait_all(ncid, nprocs, request, status);
> >>>
> >>> A detailed return is in the attachment,
> >>>
> >>> Can you please tell me what's wrong with the codes?
> >>> I appreciate it.
> >>>
> >>> Jialin
> >>> <56034.e>

-- 
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA