molpro/library-manager

library-manager's Issues

LibraryManager_Append should capture missing target name

In GitLab by @mxs301 on Aug 25, 2020, 17:29

Correct entries in generated pkgconfig file

In GitLab by @mxs301 on Sep 16, 2020, 14:27

We have an option to dump all C and Fortran flags to a pkgconfig file. Currently we just make list of different options by recursively going through target dependencies and appending. There is no clean up which means linker flags are wrong. It's time to process those options to generate a pkgconfig file that actually works.

TODO:

• Remove duplicate CFlags.
• Remove duplicate linker flags. Link flags are currently appended in order of dependence, so only leaving the right most entry should lead to correct order.
• Use "-L /path/to/lib -l lib.a" idiom to make the output intelligible.

Add Find HDF5 function

In GitLab by @mxs301 on Jun 28, 2020, 10:01

See issue https://gitlab.com/molpro/gci/-/issues/21.

• Implement LibraryManager_FindHDF5 which returns an alias target and makes sure that only on library is ever used

Fixed sphinx docs - [merged]

In GitLab by @mxs301 on May 29, 2020, 09:50

Merges LM-funcitons-sphinx-docs -> master

correct detection of fortran i8 option - [merged]

In GitLab by @KnowlesPJ on Jan 17, 2021, 12:50

Merges fortran-i8 -> master

Support blas/lapacke on macos with Apple or OpenBLAS vendor types - [merged]

In GitLab by @KnowlesPJ on Feb 16, 2022, 17:36

Merges macos-lapacke -> master

LibraryManager_LAPACK and LibraryManager_BLAS should have well-defined outcomes, and should be controllable by options

In GitLab by @KnowlesPJ on Aug 30, 2020, 12:35

cmake 3.16 and 3.18 behave quite differently. Sometimes you get an imported target BLAS::BLAS, which needs the GLOBAL property setting; sometimes (I think 3.16) you don't get this target at all. We need to look at this and either fix it up, or mandate a sufficiently new cmake. Whatever happens, the outcome of LibraryManager_BLAS and LibraryManager_LAPACK should be uniform.

Perhaps it would be best if all the logic in https://gitlab.com/Knowles-group/Molpro/-/blob/14511fdcb5ba3964a05246563da3fc81908581a9/external/src/CMakeLists.txt#L10-26 would be best done inside the function - in other words express, optionally, preference for integer size and threading as parameters to LibraryManager_LAPACK

• Ensure that the outcome is always independent of cmake version, if necessary insisting on a higher minimum version than at present
• Provide options to express preference for integer length and threading
• Decide, implement, document and use what is expected to be set by these functions. Target name BLAS::BLAS? Local or global? Information on integer length?

Related to #3.

Mechanism to override flags of some targets

In GitLab by @mxs301 on Sep 18, 2020, 09:37

It might be necessary to build one target with debugging and no optimisation, while the rest of the targets have full optimisation. This type of scenario is very common in things like game development where speed of debugging is crucial.
We should provide a mechanism for this.

To override the flags at CMake level, we need to know the target name and the flags. LibraryManager can accept a variable "OVERRIDE_FLAGS_${TARGET_NAME}" with a new list of flags. This can be passed during configure, and in external/src/CMakeLists.txt we can add a function call "LibraryManager_OverrideFlags" that will loop over all targets defined in "${BUILDSYSTEM_TARGETS}" and apply override flags if the corresponding variable is defined.

Install fortran modules to include

In GitLab by @mxs301 on Aug 29, 2020, 13:53

Currently we install them to include/fortran but there is no need for that and it is simpler to install them to include/.

correct detection of fortran i8 option - [merged]

In GitLab by @KnowlesPJ on Jan 17, 2021, 14:05

Merges fortran-i8 -> master

Find lapacke lib - [merged]

In GitLab by @KnowlesPJ on Mar 11, 2021, 16:44

Merges find-lapacke-lib -> master

Add Options to pass BLAS and LAPACK flags by user

In GitLab by @mxs301 on Jun 28, 2020, 10:03

There should be an option for the user to pass BLAS and LAPACK flags, so that find_package() is not called.

Introduce testing of Eigen, and fixes for MacOS - [merged]

In GitLab by @KnowlesPJ on Feb 20, 2022, 07:07

Merges eigen -> master

Introduce testing of Eigen.
For BLA_VENDOR=Apple, force LAPACK to come from OpenBLAS

Improved LAPACKE - [merged]

In GitLab by @KnowlesPJ on Feb 25, 2022, 17:46

Merges no-lapacke -> master

Repair fortran i8 detection - [merged]

In GitLab by @KnowlesPJ on Jan 17, 2021, 12:14

Merges repair-fortran-i8-detection -> master

Bring FindBLAS and FindLAPACK modules into LibraryManager

In GitLab by @mxs301 on Sep 18, 2020, 09:49

There is quite a bit of variation in those modules depending on CMake version. Most importantly in 3.18 they create targets, and this is exactly the mechanism we introduced ourselves. Those modules are pretty much stand-alone and we should attempt to provide the most recent versions, irrespective of installed cmake. This speaks in favour of copying those modules into LibraryManager.

BLAS/LAPACK discovery does not signal MKL unless MKL was explicitly requested

In GitLab by @KnowlesPJ on Sep 13, 2020, 20:45

https://gitlab.com/molpro/librarymanager/-/blob/d34819f57238b9efab14acb51ae3abfeccd64016/cmake/LibraryManager.cmake#L456-470 is not effective in the case that no arguments are given to LibraryManager_FindLAPACK and BLA_VENDOR is empty. At that point, BLA_VENDOR_FOUND contains All. Somehow one should discover that MKL has been found even though not asked for explicitly. Without this there is trouble on MacOS because of the lack of https://gitlab.com/molpro/librarymanager/-/blob/d34819f57238b9efab14acb51ae3abfeccd64016/cmake/LibraryManager.cmake#L458

hack away generator expressions bleeding in to pkg-config configuration - [merged]

In GitLab by @KnowlesPJ on Feb 16, 2022, 21:18

Merges generator-pkg-config -> master

Support for oneapi in the context of FindLAPACK not understanding its structure yet - [merged]

In GitLab by @KnowlesPJ on May 12, 2021, 17:21

Merges oneapi-discovery -> master

scoping of FORTRAN_INTEGER8

In GitLab by @KnowlesPJ on Jun 21, 2020, 12:57

https://gitlab.com/molpro/librarymanager/-/blob/b44b32d41842e03e655e89c887d123cab0729959/cmake/LibraryManager.cmake#L79-84 looks imperfect to me, because the scope of the outcome (CMAKE_Fortran_FLAGS) is local, but a global property is set to prevent it being done anywhere in the future. It is fine if, as intended, this is called at the top level, but not otherwise. At the very least, a trap would be good, as some hours of debugging can result otherwise for the uninitiated!

Better lapack cblas discovery - [merged]

In GitLab by @KnowlesPJ on Feb 24, 2022, 06:47

Merges better-lapack-cblas-discovery -> master