- Removed all Fortran code from the org/qmckl_sherman_morrison_woodbury.org file to prevent the appearance of *_ symbol names in libqmckl.so due to the use of Fortran compiler.
- Removed all compiler warings due to imlpicit call to fabs() by including math.h
- Removed all the cppcheck warnings of unsused variables by casting return-value function calls to void when the return value is not used. #25
The previous value of epsilon, which was pulled form the qmckl_context
and reused fot the qmckl probes, was arout 4.10^-16. It's a good value
to use for the tests with IEEE arithmetic, however, MCA introduces some
noise which is at this order of magnitude, and caused one probe in
particular to fail, while its accuracy was actually satisfying.
This value is now multiplied by 16 if VFC_CI is defined (to ask for 4
less significant digits in base 2), which makes more sense for MCA.
Line breaks in the lists of the documentation appeared in the file
preview. They have been removed so that all of its text belongs
correctly to eac bullet point.
Some checks in the AO test were not correctly ignored by the
preprocessor when vfc_ci support was enabled. This caused the test to
fail unexpectedly and made it impossible to get data from it in vfc_ci.
Some asserts where not correctly ignored with vfc_ci and caused tests to
exit prematurely when compiling the library with Verificarlo and using
stochastic backends.
The target has been set to 1d.-14 (or epsilon), because this was the value already
used in the tests. However this might not be the best choice with the
non-deterministic backend, so it might be necessary to adjust it in the
future.
Also, the expectedValue argument has been set to 0 since it is not used.
(the tests checks have already been written anyway)
Since we don't need to place probs on all matrices, a simple solution
seems to monitor one element of the matrix. A solution could be to
compute a matrix norm / residual but this would make the tests more
complex.
* comment
* Update distance test code
The distance test has been updated to the latest version, with a first attempt at using vfc_probes inside it
* Functional implementation of vfc_probes in the distance tests
This commit has the first functional vfc_ci tests. Verificarlo tests
should be written over the existing tests, and they can be enabled with
the following configure command:
QMCKL_DEVEL=1 ./configure --prefix=$PWD/_install --enable-maintainer-mode --enable-vfc_ci CC="verificarlo-f -Mpreprocess -D VFC_CI" FC="verificarlo-f -Mpreprocess -D VFC_CI" --host=x86_64
The --enable-vfc_ci flag will trigger the linking of the vfc_ci
library. Moreover, as of now, the "-Mpreprocess" and "-D VFC_CI" flags
have to be specified directly here. There is probably an appropriate
macro to place those flags into but I couldn't find it yet, and could
only manage to build the tests this way.
When the VFC_CI preprocessor is defined, somme additional code to
register and export the test probes can be executed (see
qmckl_distance.org).
As of now, the tests are built as normal, even though they are expected
to fail :
make all
make check
From there, the test_qmckl_distance (and potentially the others)
executable can be called at will. This will typically be done
automatically by vfc_ci, but one could manually execute the executable
by defining the following env variables :
VFC_PROBES_OUTPUT="test.csv" VFC_BACKENDS="libinterflop_ieee.so"
depending on the export file and the Verificarlo backend to be used.
The next steps will be to define more tests such as this one, and to
integrate them into a Verificarlo CI workflow (by writing a
vfc_tests_config.json file and using the automatic CI setup
command).
* Error in FOrtran interface fixed
* Added missing Fortran interfaces
* Modify distance test and install process integration
All probes are now ignored using only the preprocessor (instead
of checking for a facultative argument) in the distance test.
Moreover,preprocessing can now be enabled correctly using FCFLAGS
(the issue seemed to come from the order of the arguments passed
to gfortran/verificarlo-f with the preprocessor arg having to come
first).
* Add vfc_probes to AO tests
vfc_probes have been added to qmckl_ao.org in the same way as
qmckl_distance.org, which means that it can be enabled or disabled at
compile time using the --enable-vfc_ci option.
qmckl_distance.org has been slightly modified with a better indentation,
and configure.ac now adds the "-D VFC_CI" flag to CFLAGS when vfc_ci is
enabled.
* Start work on the vfc tests config file and on a probes wrapper
The goal in the next few commits is to make the integration of
vfc_probes even easier by using a wrapper to vfc_probe dedicated to
qmckl. This will make it easier to create a call to vfc_probe that can be
ignored if VFC_CI is not defined in the preprocessor. Once this is done,
the integration will be completed by trying to create an actual workflow
to automatically build the library and execute CI tests.
* Moved qmckl_probes out of src
As of now, qmckl_probes have been moved to tools, and can be built via a
bash script. This approach seems to make more sense, as this should not
be a part of the library itself, but an additional tool to test it.
* Functional Makefile setup to enable qmckl_probes
The current setup builds qmck_probes by adding it to the main QMckl
libray (by adding it to the libtool sources). The Fortran interface's
module still need to be compiled separately.
TODO : Clean the build setup, improve integration in qmckl_tests and
update tests in qmckl_ao with the new syntax.
* New probes syntax in AO tests
* Clean the probes/Makefile setup
The Fortran module is now built a the same time than the main library.
The commit also adds a few fixes in the tests and probes wrapper.
Co-authored-by: Anthony Scemama <scemama@irsamc.ups-tlse.fr>
