ECP integrals not working

[Peterson, Kirk A]

Hi all,

 

Has anyone used the ECP code lately? Even the test jobs in test/ecp fail for me with:

[c1:586555:0:586555] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x7ffca8205330)

==== backtrace (tid: 586555) ====

 0  /opt/ohpc/pub/mpi/ucx-ohpc/1.11.2/lib/libucs.so.0(ucs_handle_error+0x254) [0x7f7a8aaa0b94]

 1  /opt/ohpc/pub/mpi/ucx-ohpc/1.11.2/lib/libucs.so.0(+0x27d4c) [0x7f7a8aaa0d4c]

 2  /opt/ohpc/pub/mpi/ucx-ohpc/1.11.2/lib/libucs.so.0(+0x27ff8) [0x7f7a8aaa0ff8]

 3  /lib64/libpthread.so.0(+0x12ce0) [0x7f7aa3aa0ce0]

 4  /scratch/kipeters/DIRAC_HF_HI_sorep_586548/dirac.x(recp_1e_+0x2c) [0x98aa7c]

 5  /scratch/kipeters/DIRAC_HF_HI_sorep_586548/dirac.x(recp_+0x68) [0x98b2b8]

 

 

I’m using a pretty standard (I think) gfortran (v12.2.0) build.

 

Best regards,

 

-Kirk

[Visscher, L. (Luuk)]

Dear Kirk,

Hackathon today, so looking into this.

For me the two ECP tests pass with gcc 14.2 on MacOSX 15.1.1, which should be the latest versions of both. Will test later also the MPI one.

Luuk

-- 
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dirac-users/IA0PR01MB8308FA67BF2FF650E07F9B2ED61F2%40IA0PR01MB8308.prod.exchangelabs.com.

[peterso...@gmail.com]

Dear Luuk,

 

Thanks very much for looking into this. I’ll see if I can update gcc on my cluster.  Have you also taken a look at the H format for the basis sets?

 

Best regards,  -Kirk

 

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/634CFCB4-95AE-4CCD-BBCE-7C02CE6C36EE%40vu.nl.

[Visscher, L. (Luuk)]

Dear Kirk,

Hope this works out for the ECP code. Did not get round testing the MPI version yet, but cannot see why this would behave differently.

For the basis sets including a H in the first column. This was indeed a bug that we found and fixed. 

Will work in the upcoming DIRAC25, but also the current development version at gitlab already has this fix (https://gitlab.com/dirac/dirac/-/merge_requests/217).

best regards,

Luuk

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/F9FAFF31-3817-BD4A-90A7-8EFD4AFF699D%40hxcore.ol.

[peterso...@gmail.com]

Dear Luuk,

 

Great news about the H format. I will proceed to put all our ECP-based correlation consistent basis sets into the library. I haven’t been able to get back to the ECP integrals yet, but will tackle that this week.

 

Best regards,

 

-Kirk

 

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/517B67EE-BBF3-4D85-8CE8-EB29C8514FC8%40vu.nl.