[petsc-users] VecView output to HDF5 in 3.12.0 broken ?

[Sajid Ali]

Hi Stefano/PETSc Developers,

The chunksize is indeed limited to 4GB as per this page :
https://portal.hdfgroup.org/pages/viewpage.action?pageId=48808714.

With a (complex) DMDA vector of size (16384,16384,2) I see that PETSc saves
it as a hdf5 file with chunks of size (1024,1024,2). But with a non DMDA
vector I don't see any chunking happening. I tried examining the chunk size
after running the same example as above and increasing the size of the
vector until it fails to write.

The output of the following case (first size to fail) is attached : mpirun
-np 16 ./ex10 -m 134217728 &> log. There's a slightly different error here
which states :
```
    minor: Some MPI function failed
  #018: H5Dchunk.c line 4706 in H5D__chunk_collective_fill(): Invalid
argument, error stack:
PMPI_Type_create_hvector(125): MPI_Type_create_hvector(count=0,
blocklength=-2147483648, stride=0, MPI_BYTE, newtype=0x7ffc177d5bf8) failed
PMPI_Type_create_hvector(80).: Invalid value for blocklen, must be
non-negative but is -2147483648
    major: Internal error (too specific to document in detail)
```

Strangely the same case works with 3.11.1 and the dataset has 4 chunks. I'm
not sure how, but it looks like the chunking logic somehow got broken in
3.12.


Thank You,
Sajid Ali
Applied Physics
Northwestern University
s-sajid-ali.github.io
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20191011/8967e306/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: log
Type: application/octet-stream
Size: 20448 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20191011/8967e306/attachment-0001.obj>