2fb368cf8c
consistent space (as in 'end module' etc)
2021-12-06 07:59:40 +01:00
08a709c6d9
capitalize HDF5 prefix
2021-12-06 07:55:22 +01:00
c818756f68
wrong check
...
iand is used for checking, see https://support.hdfgroup.org/HDF5/doc/RM/H5Z/H5Zget_filter_info.htm
2021-12-02 20:51:33 +01:00
731e20261c
bugfix for HDF5 without deflate filter
...
before inquiring whether a filter supports compression/decompression, we
first need to check whether it is available at all to avoid raising an
error.
Extended the check to the shuffle filter
2021-12-02 13:00:59 +01:00
da9fdf53d2
consistent indentation and line-spacings in reporting
2021-11-15 12:35:44 -05:00
497bf2c645
avoid errors for empty datasets
...
empty dataset cannot be chunked, but filters can only be applied to
chunked datasets
2021-08-14 20:44:59 +02:00
30d9f57db0
also compress string datasets
...
results in different type (numpy string instead of numpy object), tests need to be adjusted
2021-08-14 16:28:51 +02:00
595cc4e59e
always chunk data
...
for small datasets:
- only one chunk
- no compression
2021-08-14 15:48:07 +02:00
68eb930bf0
Merge remote-tracking branch 'origin/development' into report-systems
2021-08-03 21:57:50 +02:00
044a048944
taking care of corner cases (e.g. restart)
...
adjusting tests to take care of new 'setup' group
2021-08-01 22:46:11 +02:00
a6f7e4f1a6
play it safe
2021-07-27 10:18:55 +02:00
820d590b6b
write string to result
...
requires to open file without MPI support
2021-07-27 07:25:44 +02:00
58c1f5fdc1
write string, only from one MPI process
...
needed to store simulation setup. I don't see a situation where strings
are distributed over multiple processes
2021-07-27 07:23:02 +02:00
a891fe4281
keep it simple
2021-07-26 22:59:53 +02:00
d2b5a4b339
pointer voodoo for ifort
2021-07-25 10:30:44 +02:00
c388ab97c9
trying to make sense of HDF5 pointer magic
2021-07-25 09:43:52 +02:00
d9ef1ef5e4
write variable length string (as other, best compatibility with h5py)
2021-07-25 09:36:56 +02:00
26e1e979f5
length is known
2021-07-24 23:08:05 +02:00
71e17ba917
WIP: report active slip systems to DADF5
...
needs further work, currently an array of fixed length strings is
written while all single strings are of variable length type
2021-07-24 22:51:57 +02:00
59d09d708e
string array support for attributes
2021-07-24 17:18:31 +02:00
db07ee2113
supported by recent compilers
2021-07-24 14:35:17 +02:00
136a4b1377
PETSc defines are rather complicated
...
now mpi_f08 can be used on newer PETSc installations if old MPI modules
are not exposed
2021-07-09 18:48:25 +02:00
3870a70a4e
HDF5 seems to use MPI (F90 style)
2021-07-09 12:37:39 +02:00
139f2c177a
use MPI_f08 if possible
...
most PETSc installations provide outdated MPI (f90 version)
MPI_COMM_WORLD is now of derived type (Fortran 08 style)
PETSC_COMM_WORLD is the plain integer (f90 style) alias.
Note that HDF5 is assumed to have f90 interfaces
2021-07-08 16:27:37 +02:00
4c3ff4bef4
consistent capitalization
2021-07-08 15:05:01 +02:00
58bc6e2ba6
avoid chained inclusions
2021-07-08 14:27:04 +02:00
5d0fc4fca3
more meaningful order
...
and intent(out) variables for read are at the front
2021-06-01 16:46:24 +02:00
b353f8ed92
workaround for HDF5: https://forum.hdfgroup.org/t/6186
...
writing chunked/compressed data can cause problems with MPI. Even though not 100%
clear from the reference, it seems that the issue only appears for HDF5
< 1.12.
Of special importance for Ubuntu since the Debian package is still at 1.10.6
2021-05-03 18:19:06 +02:00
d54e49e3bc
restore functionality to write non-parallel
...
not needed at the moment, but in general useful. If PETSc = parallel
should always hold, we can simplify much more
2021-02-22 13:37:21 +01:00
a51b32ccdf
only input
2021-01-26 20:18:04 +01:00
3ad741dbeb
only used in one function
2020-12-10 00:31:58 +01:00
97ee7e6ee5
not needed anymore
2020-12-07 17:26:50 +01:00
429b84004d
more filters
...
- shuffle: significanlty better compression
- Fletcher32: checksum to detect errors
computational overhead is very small
2020-12-06 19:57:00 +01:00
bc4361c2ae
use variable string length
...
for array, padding is needed to get same length
2020-12-06 13:32:20 +01:00
52e3fb50bc
compress.
...
Datasets are chunked along first timension. Chunk size (1MB for real) is
probably not optimal
2020-12-06 11:44:52 +01:00
018709bfcf
error stop better suited for internal errors
...
error stop gives full stack trace, which is very helpful for debugging.
2020-11-11 10:19:47 +01:00
03a68248b5
polishing
2020-11-06 13:25:00 +01:00
e649fb195f
update to use variable-length strings
2020-11-05 22:34:41 +01:00
2b6001222d
black magic
...
compiles also with Intel (complained about non-existing interface
before).
Direct pointer crashes ..
2020-11-05 17:59:05 +01:00
b8c97c5b7e
change string type
...
normal fortran strings are fixed-length, spacepad. h5py (at least v3.x)
creates variable-length, nullterm
2020-11-05 15:36:13 +01:00
e848590c5c
Merge remote-tracking branch 'origin/development' into Fortran-simplifications
2020-09-17 19:39:31 +02:00
5b83c8ad3c
use 'error stop'
...
- does not require IO
- prints stack trace
2020-09-13 12:09:32 +02:00
da0e16520c
handling of MPI in 'parallelization'
2020-09-13 10:26:33 +02:00
7754a1ea56
Restructuring for material.yaml
2020-08-15 16:08:46 +02:00
93adee2581
no need to have fixed string here
2020-08-09 06:27:29 +02:00
9565f15415
clearer description
2020-04-28 10:30:03 +02:00
ae49e6710d
documenting for doxygen
2020-03-15 12:54:35 +01:00
18ccc49d6f
avoid duplicated public/private statements
...
variables should be private
2020-02-29 15:15:41 +01:00
62865e779d
polishing
2020-01-26 18:31:56 +01:00
41650cdd11
use standard string length
2019-12-21 07:38:17 +01:00
Martin Diehl