Bug fix: The program was compiled with the warnings indicating value overflow. The warnings pointed to an error in the code that caused the program to fail while reading data from the file. $ gcc -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 create_large_file.c create_large_file.c: In function 'main': create_large_file.c:48: warning: integer overflow i... $ a.out
Error in second seek because Invalid argument
Fixed the problem by declaring the correct constant.
Platforms tested: jam, amani and linew
Eliminated a compiler warning because printf does not have a prototype definition. Adding stdio.h took care of it. Also, edited the Copyrights notice.Tested: heiwa.
Added these tests: Verify that created file is deleted upon closing. Verify that MPI_MODE_EXCL works correctly with non-existing file. (see Bug 1468). Verify MPI_File_delete works.Tested:
Glory of Sandia.
Added missing copyright notices. Note that some arecreated after June 2006 and they would be THG solely
copyrighted unless they were a copied from pre-existing
files.
Purpose: bug fixDescription:
stdlib.h should be included for AIX 5.1.
malloc is defined inside stdlib.h, otherwise, unexpected results occur.
Solution:
include this header file.
Platforms tested:
Misc. update:
Purpose: Modify the code a little bit to make it more portableDescription:
Change the buf type from int to char.
Solution:
Platforms tested:
Misc. update:
Purpose: An MPI-IO program to test non-contiguous MPI derived datatype combined with collective IO.Description:
There is a bug at IBM machine with parallel HDF5. Hopefully this simple program can verify the problem is inside MPI-IO rather than inside HDF5.
Solution:
Platforms tested:
works fine at heping with mpich 1.2.6.
Fails at copper and bluesky, two IBM machines we have accesses.
Misc. update:
Purpose: bug fix Description:It seems this bug only happens when the number of process is 4.
It is very strange. Since we don't know why this happens, we still
keep this test as a bug test.
Solution:
Platforms tested:
Misc. update:
Purpose: Add a collective IO test when at least two processors contribute none for IO.Description:
Solution:
Platforms tested:
Should fail at mpich 1.2.6 and work at mpich 1.2.7.
Misc. update:
Purpose: While in the implementing collective chunk IO inside HDF5, we find various MPI-IO problems related to the collective IO support.Description:
Add a standard collective write program to verify the robustness of different MPI-IO packages.
Solution:
Platforms tested:
mpich 1.2.6 at Linux 2.4
Misc. update:
Purpose: Add a new C code to test the case when one or some processes don't contribute any IOs in collective code.Description:
Solution:
Platforms tested:
Linux 2.4(heping)
Misc. update:
Purpose: A bug fixDescription:
AIX 5.1 poe version 3.2.0.20,
The second value of the output array is not 0 but 2. This is not right
although the third value of the array is 2.
Solution:
Add another check to see whether the second value = the third value when
the third value is expected(2), if the second = the third, derived datatype
is still failed.
Platforms tested:
UCAR bluesky(AIX 5.1, poe 3.2.0.20)
NCSA c...
Purpose: bug fixDescription:
This test assumed the fill value was 0. This is not true for all platforms.
Solution:
Reworked the test a bit so that it no longer checks against the fill value but uses a known value.
Platforms tested:
copper and MCR. We no longer have a platform on which this test passes, so we could not
test the failure condition.
Misc. update:
Purpose: Bug fixDescription:
ret is not initialized to anything. On some systems, it defaults 0.
So, when the program returns 'ret' at the end, it may erroneously return 0,
which is equivalent to success.
Solution:
Initialize ret to -1.
Platforms tested:
MCR
Misc. update:
Albert Cheng
Binh-Minh Ribler
Elena Pourmal
Matthew Needham
Auto Bot
Muqun Yang
Raymond Lu
Leon Arber