Yen-Chen Chen authored 1 day ago and Mikael Stellio committed 1 day ago

## What is the new feature
Add a contribution guide. Yen-Chen's email is also updated after moving from KIT to TUM.

Approved-by: Mikael Stellio <mikael.stellio@c2sm.ethz.ch>
Merged-by: Mikael Stellio <mikael.stellio@c2sm.ethz.ch>
Changelog: other

Yen-Chen Chen authored 1 week ago and Mikael Stellio committed 1 week ago

## What is the new feature
Make changelog API to search for changelogs until the target commit. This way we can mange the versions more easily.
## How is it implemented
In the GitLab [Changelog API](https://docs.gitlab.com/api/repositories/#generate-changelog-data

), the default `from` attribute is the last tag, which is exactly what we wanted. The default `to` attribute is the latest commit at the default branch. We change it to the target commit where we create the tag.

Approved-by: Mikael Stellio <mikael.stellio@c2sm.ethz.ch>
Merged-by: Mikael Stellio <mikael.stellio@c2sm.ethz.ch>
Changelog: other

## What is the new feature
Hand maintenance over to Mikel Stellio

Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: other

Dylan Kierans authored 1 month ago and Yen-Chen Chen committed 1 month ago

## What is the new feature
New copy variants for 2d,4d,5d fields from double-precision to single-precision.

Co-authored-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Approved-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Changelog: feature

Yen-Chen Chen authored 1 month ago and Pradipta Samanta committed 1 month ago

## What is the bug
The `insert_dimension` fails in ICON without the deleted if clause.
## How do you fix it
Recover the deleted code.

Merged-by: Pradipta Samanta <samanta@dkrz.de>
Changelog: bugfix

Yen-Chen Chen authored 1 month ago and Pradipta Samanta committed 1 month ago

## What is the bug
The C++ test code for `mo_util_stride` uses the experimental standard library.  
The filesystem standard library is automatically linked in C++17
## How do you fix it
Replace with the standard library


Merged-by: Pradipta Samanta <samanta@dkrz.de>
Changelog: bugfix

Yen-Chen Chen authored 1 month ago and Pradipta Samanta committed 1 month ago

## What is the bug
`OpenMP_C` is not used and cannot be found with Mac clang, which causes compile trouble. This MR solves #25

.
## How do you fix it
Only find OpenMP and the link for Fortran.
Along the way, one bug for `insert_dimension` is also fixed with this MR.


Approved-by: Pradipta Samanta <samanta@dkrz.de>
Merged-by: Pradipta Samanta <samanta@dkrz.de>
Changelog: bugfix

Sergey Kosukhin authored 1 month ago and Yen-Chen Chen committed 1 month ago

## What is the bug

The endianness detection in `src/util_hash.c` produces a false result on macOS with GCC 13 or later. For whatever reason, [`#include <sys/param.h>`](https://gitlab.dkrz.de/icon-libraries/libfortran-support/-/blob/316d3dc6edb6ef0f4f1b9adaf9f3cb27a7bb831a/src/util_hash.c#L49) does not define the `__DARWIN_BYTE_ORDER` and `__DARWIN_LITTLE_ENDIAN` macros when [`#define _POSIX_C_SOURCE 200112L`](https://gitlab.dkrz.de/icon-libraries/libfortran-support/-/blob/316d3dc6edb6ef0f4f1b9adaf9f3cb27a7bb831a/src/util_hash.c#L45). And that leads to the wrong values of the `HASH_LITTLE_ENDIAN` and `HASH_BIG_ENDIAN` macros. See also https://gitlab.dkrz.de/icon/icon-mpim/-/merge_requests/684#note_300951

.

## How do you fix it

The endianness detection is delegated to CMake.

Approved-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Changelog: bugfix

Dylan Kierans authored 1 month ago and Yen-Chen Chen committed 1 month ago

## What is the new feature
Support for single-precision as working-precision in libfortran-support. Related to icon-mpim MR (https://gitlab.dkrz.de/icon/icon-mpim/-/merge_requests/558), and libiconmath MR (libiconmath!13

).

## How is it implemented
- New CMake argument `FS_ENABLE_SINGLE_PRECISION`, disabled by default
- Precision handling: Introduction of `src/mo_iconlib_kind.f90` shared by fortran-support and iconmath:
  - Precision (dp+sp): Using iso_fortran_env variables `dp=>real64` and `sp=>real32`
  - Working precision (wp): `FS_ENABLE_SINGLE_PRECISION==TRUE` -> `wp=real32`, else `wp=real64`
  - Variable precision (vp): `FS_ENABLE_MIXED_PRECISION==TRUE` -> `vp=real32`, else `vp=wp`
- Extensions of some functions to include sp and dp interfaces
- Kind for `t_ptr` types is `wp` by default, with wp or dp or sp versions specified by suffix.

Co-authored-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Co-authored-by: Dylan Kierans <dylankierans53@gmail.com>
Co-authored-by: Pradipta Samanta <samanta@dkrz.de>
Approved-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Changelog: feature

Marek Jacob authored 1 month ago and Yen-Chen Chen committed 1 month ago

## What is the new feature
`assert_lacc_equals_i_am_accel_node` is not used in ICON any more and should be removed completely.
## How is it implemented
`grep -r assert_lacc_equals_i_am_accel_node` in icon only found the definition and test in libfortran-support.

Co-authored-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Co-authored-by: Marek Jacob <MeraX@users.noreply.github.com>
Approved-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@tum.de>
Changelog: feature

Yen-Chen Chen authored 2 months ago and Pradipta Samanta committed 2 months ago

## What is the new feature
Bump license year to 2025

Approved-by: Pradipta Samanta <samanta@dkrz.de>
Merged-by: Pradipta Samanta <samanta@dkrz.de>
Changelog: other

Yen-Chen Chen authored 2 months ago and Pradipta Samanta committed 2 months ago

## What is the bug
The `t_char_workaround` was missing from the public interface, which causes some BuildBots to fail.
## How do you fix it
Add `t_char_workaround` public interface to `fortran_support`.

Approved by: Pradipta Samanta <samanta@dkrz.de>
Merged-by: Pradipta Samanta <samanta@dkrz.de>
Changelog: default

Tuomas Lunttila authored 2 months ago and Yen-Chen Chen committed 2 months ago

## What is the new feature
Improve `minval_2d` performance.
## How is it implemented
Add COLLAPSE(2) to a nested loop in minval_2d to improve performance.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: feature

## What is the bug
The variable `nnml` is falsely deleted in !94

. Which causes error in dace that uses `nnml` from `mo_namelist`.
## How do you fix it
Add back the variable.

Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

## What is the bug
CMake install prefix using `${CMAKE_INSTALL_PREFIX}` does not take environmental install path, but rather hardcoded to the explicit path. See [Sergey's comment](!95 (comment 280056)

)
## How do you fix it
Use `$<INSTALL_PREFIX>` instead.

Approved-by: Sergey Kosukhin <sergey.kosukhin@mpimet.mpg.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

## What is the bug
Gather licenses of technical files and docs into REUSE.toml. This prevents licenses from getting into the commit with the MR templates.
## How do you fix it
Add `REUSE.toml` and remove license texts

Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

Sergey Kosukhin authored 4 months ago and Yen-Chen Chen committed 4 months ago

This removes an anti-pattern from `CMakeLists.txt` (I'm very sorry for introducing it). The most annoying part about it is that it makes it difficult for a parent project (e.g. ICON) to support multi-configuration backends properly. Basically, the only right place to set `CMAKE_BUILD_TYPE` is on the command line when calling `cmake` (and generating for a single-configuration backend). Also, see [here](https://stackoverflow.com/questions/24460486/cmake-build-type-is-not-being-used-in-cmakelists-txt/24470998#24470998

).

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

The present check in the `set_acc_host_or_device` is heavily used in ICON and cannot be changed easily.
This problem comes from the discussion at https://gitlab.dkrz.de/icon/icon-mpim/-/merge_requests/563#note_276277


Revert the changes.

Approved-by: Marek Jacob <1129-b380572@users.noreply.gitlab.dkrz.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

Claudia Frauen authored 4 months ago and Yen-Chen Chen committed 4 months ago

Workaround for the CCE 17 requiring explicit copy for reduction variables (!99)

CCE 17 needed for AMD GPUs on LUMI now requires an explicit copy of reduction variables.
Add copy clauses for the reduction variables.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

Daniel Reinert authored 5 months ago and Yen-Chen Chen committed 4 months ago

The subroutines `init_zero_4d_[dp,sp,i4]` do not vectorize properly on NEC SX AURORA.

Loop collapsing is enforced by a compiler directive for the subroutines `init_zero_4d_[dp,sp,i4]`, in order to ensure proper vectorization on NEC SX AURORA.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: feature

Pradipta Samanta authored 7 months ago and Yen-Chen Chen committed 7 months ago

Fixed installation; other projects can now find module files from fortran-support (!95)

## What is the bug
Any projects dependent on fortran-support are not able to find the associated module files of when the fortran-support is installed in the system
## How do you fix it
`CMAKE_INSTALL_PREFIX` was not used to define `INSTALL_INTERFACE`. I corrected it now.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

## What is the new feature
Move random number generator to `libmath-support`. (Adds in https://gitlab.dkrz.de/icon-libraries/libmath-support/-/merge_requests/33

)
## How is it implemented
The random number generator module is more suitable to be put into `libmath-support`. Thus we are moving it from `libfortran-support`.

**The next versions `libfortran-support` 2.0.0 and `libmath-support` 1.1.0 should be merged at the same time.**

Approved-by: Daniel Hupp <daniel.hupp@meteoswiss.ch>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: feature

Sergey Kosukhin authored 8 months ago and Yen-Chen Chen committed 8 months ago

This updates the `add_icon_c_test` macro in `cmake/gtest_helper.cmake`:
1. The macro is turned into a function to avoid namespace pollution.
2. The function is renamed to `fs_add_c_test`.
3. The function now accepts named options `SOURCES` (followed by a list of source files of the test) and `ARGS` (followed by a list of command-line arguments for the test).
4. The old signature of the macro is still supported, i.e. it is possible to call something like `add_c_test(UtilHashTest ctest_util_hash.cpp)`, but it made sense to me to switch to the new one.
5. The `UtilFileTest` test is now provided with an extra command-line argument and it runs on `${CMAKE_CURRENT_BINARY_DIR}` instead of `.` as it was before !28.

Co-authored-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Approved-by: Terry Cojean <terry.cojean@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: other

## What is the new feature
Add a public interface for `libfortran-support`. This is related to #9 
## How is it implemented
Add a `fortran_support.F90` collecting all subroutines and functions in the `libfortran-support` library.  
Now one can do `USE fortran_support` to access all public subroutines/functions in the library.

Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: feature

## What is the bug
A package-NOTFOUND is undetected and does not raise a problem. See #20
This MR closes #20
## How do you fix it
The OpenMP find package is made `REQUIRED`.  
The find package for OpenACC has to stay `QUIET` or else it will try to look for both `OpenACC_C` and `OpenACC_Fortran`. (LUMI does not have `OpenACC_C` at the moment.)

Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Approved-by: Sergey Kosukhin <sergey.kosukhin@mpimet.mpg.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: other

## What is the new feature
Make `lacc` mandatory.
## How is it implemented
With the changes in https://gitlab.dkrz.de/icon/icon-mpim/-/merge_requests/342, https://gitlab.dkrz.de/art/art/-/merge_requests/96, and https://gitlab.dkrz.de/jsbach/jsbach/-/merge_requests/165

, the `lacc` argument in all `fortran_tools` functions should be added explicitly.  
This MR makes sure that in the future, an error will pop up if `lacc` is not passed.

Approved-by: Marek Jacob <1129-b380572@users.noreply.gitlab.dkrz.de>
Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: other

## What is the bug
Running pipelines on images cannot do tests with OpenACC.
## How do you fix it
Change the CI configuration to using the `levante` shared runner. This runner submits jobs on Levante directly with the CI script.  
`Check typos` CI is also fixed by specifying the conda channel to search for the package.

Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: bugfix

## What is the new feature
Add a formatter for CMake files. And CI task for CMake formatting is also added.

## How is it implemented
We use the opensource tool https://github.com/cheshirekow/cmake_format

 for formatting.

Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: feature

Roland Wirth authored 10 months ago and Yen-Chen Chen committed 10 months ago

String lists grow dynamically, removing the need for preallocated arrays. This simplifies
handling of large variable groups in the icon initialization phase.

The output lists of `add_to_list` are now `ALLOCATABLE`. They may be passed an unallocated
argument, which is treated like an empty list and allocated using the new `new_list`
subroutine. That subroutine initially allocates space for 8 strings. Every time the list
needs to grow, the allocated size doubles. This ensures constant amortized runtime for
adding to the list. No provisions are made for shrinking the list size, because they are
usually short-lived.

The tests for `add_to_list` are adapted for the new interface, checking list expansion.
A new test for `new_list` is added.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: feature

## What is the bug
`mo_fortran_tools` was updated during the release preparation MR.

Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: bugfix

## What is the new feature
Improve README.

Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: feature

## What is the bug
FortUTF did not return correct error code for failed test.
## How do you fix it
[This PR(merged)](https://github.com/artemis-beta/FortUTF/pull/22

) fixed the problem.

Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: bugfix

Daniel Hupp authored 11 months ago and Yen-Chen Chen committed 11 months ago

## What is the new feature
This PR adds a new module for random number generations. It contains a parallelization uniform random number generator and a normal random number generator.
## How is it implemented
The uniform random number generator uses the Lehmer random number generator.
The normal random number generator uses a Box-Mueller transform using the uniform random number generator.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: feature

<!--
ICON
## What is the bug
LUMI does not have `OpenACC_C` and the BB fails because we are checking `OpenACC` for both `C` and `Fortran`. This MR fixes `lumi_gpu` BB.
## How do you fix it
Only check if `OpenACC_Fortran` is found.

Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: bugfix

## What is the new feature
Add a typo check pipeline.
## How is it implemented
Use https://github.com/crate-ci/typos



Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: feature

## What is the new feature
Make `lacc` arguments mandatory. This MR closes #17



Approved-by: Marek Jacob <1129-b380572@users.noreply.gitlab.dkrz.de>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: bugfix

## What is the bug
Fix a test for the if_associated subroutine.
## How do you fix it
The false test did not pop up because of a bug in FortUTF. See #18



Approved-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: bugfix

## What is the bug
The accuracy of the expression evaluation code was falsely downscaled in the code.
## How do you fix it
The false test did not pop up because of a bug in FortUTF. See #18



Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: bugfix

## What is the new feature
Add tests
## How is it implemented
Add tests in `mo_expression` and the test coverage badge has officially turned yellow. 



Merged-by: Jonas Jucker <jonas.jucker@env.ethz.ch>
Changelog: feature

Jonas Jucker authored 1 year ago and Yen-Chen Chen committed 1 year ago

New instructions for the merge request workflow.

Approved-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Merged-by: Yen-Chen Chen <yen-chen.chen@kit.edu>
Changelog: other