Feature: enable scalapcack solver for LR-TDDFT

[maki49]

No description provided.

[maki49]

The bug occured in the integration test seems strange:

I tried to debug it on my machine and find that:

• It happens at the second call of complex eigensolver pzheev
• Compiling with intel oneapi has no such problem. It only happends with GNU compiler.
• Running with 1 processor encounters the same problem.
• Both jobtype = 'N' and V have such problem.
• pzheevd and pzheevx also have such problem.
• Enlarging work and rwork to 100 times of the queried value of lwork and rwork does not help.

gdb info:

Thread 1 "abacus" received signal SIGSEGV, Segmentation fault.
0x00001555447c8cbb in mkl_lapack_zladiv () from /opt/intel/oneapi/mkl/2024.2/lib/libmkl_core.so.2
(gdb) bt
#0  0x00001555447c8cbb in mkl_lapack_zladiv ()
   from /opt/intel/oneapi/mkl/2024.2/lib/libmkl_core.so.2
#1  0x0000555555f5bb2a in pzlarfg_.constprop ()
#2  0x0000555555f6736b in pzlatrd_.constprop ()
#3  0x0000555555f5aa8c in pzhetrd_.constprop ()
#4  0x0000555555f0637a in pzheev_ ()
#5  0x0000555555e3e5f0 in LR_Util::diag_scalapack (
    n=@0x7fffffff89a0: 80, mat=mat@entry=0x555559120830, 
    eigval=0x555558a4a560, eigvec=eigvec@entry=0x555559139840, 
    desc=...)
    at /home/fortneu49/abacus-fix/abacus-develop/source/module_lr/utils/lr_util.cpp:222

@caic99 @dyzheng do you have any idea?

[caic99]

Hi @maki49 ,
Would you check all the input params (and its shape for arrays) with extra care? Like, does the pointer and ld matches?

[mohanchen]

It seems that all tests have passed, if the issue has been solved and the PR can be accepted, let me know.

[maki49]

It havn't been solved. I'm still debugging. Here are somes test results:

compiler	solver	1 processor	multi-processor
intel	pzheevx	OK	segfault
gnu	pzheevx	OK	segfault
intel	pzhegvx	OK	segfault
gnu	pzhegvx	OK	segfault
intel	pzheev	OK	OK
gnu	pzheev	segfault	segfault
intel	pzheevd	OK	OK
gnu	pzheevd	OK	usually OK but occasional segfault