YI Zeping

Try compiling with mpifort or gfortran? GCC compiler should work with ELPA, the developers did tested it. The toolchain contains nothing about Fortran compiler, maybe you should specify it manually,...

UPDATE: AOCC fails at all tests above and have one additional test failed. The addition test is: 189 - HSolver_bpcg (Failed) Error message is: ``` /home/yizeyi18/soft/abacus-develop-3.4.3/source/module_hsolver/test/diago_bpcg_test.cpp:144: Failure The difference between...

UPDATE: Clang-17.0.4 fails at all tests failed same as AOCC-4.1, with similar error message. One only differance form aocc is that clang would link to libstdc++.so without absolute path infomation,...

UPDATE: Clang-18.1.1 fails at more tests, including all tests with HF calculation. test log: [LastTest.log](https://github.com/deepmodeling/abacus-develop/files/14730932/LastTest.log) Failed Tests ```bash 1 - integrated_test (Failed) 6 - base_atom_in (Failed) 8 - base_timer (Failed)...

@caic99 `-DNDEBUG` is generated by `cmake` for all release build. Should I reconfigure the package with `-DCMAKE_BUILD_TYPE=Debug` (instead of `-DCMAKE_BUILD_TYPE=Release` or `-DCMAKE_BUILD_TYPE=RelWithDebInfo`)?

@caic99 I tried both `-DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS="-O3 -g"` and `-DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS="-g"`, both yields error in unit test. The no `-O3` one still hangs at test 154 `HSolver_bpcg`(for over 20 min), log...

@caic99 Till now, all HSolver tests timeout(in 1500 seconds), fail, or segfault, with a big memory comsumption(40-70GiB). I suppose it's from low compiler optimization level, I did not observe these...

I'm also facing this issue, and have more information could provide: This output comes from `ompi/mca/osc/ucx/osc_ucx_component.c:369`, which inferrs `component_set_priority()` implemented in the same file returns a non-success value: https://github.com/open-mpi/ompi/blob/45258b6cb7129df7c5387445a75f3791cfdc4e74/ompi/mca/osc/ucx/osc_ucx_component.c#L366-L371 And...

Would #9446 solve this issue?

As compiler management structure of spack varied a lot, this PR could get closed.