Installation of 6.4.1 with Wannier90 3.1 using OneAPI 2022 on AMD 7643

[kalyankn]

Hi

I was able to compile and complete the installation of VASP 6.4.1 with Wannier90 3.1 (serial mode using ifort) and HDF5 using OneAPI 2022 on AMD EPYC 7643 based cluster without error. But, the testsuite got failed with following errors.

The following tests failed, please check the output file manually:
bulk_BN_PBE_Libxc bulk_BN_SCAN_Libxc bulk_InP_SOC_G0W0_nosym bulk_InP_SOC_G0W0_sym bulk_InP_SOC_G0W0_sym_RPR bulk_SiO2_LOPTICS bulk_SiO2_LOPTICS_nosym bulk_SiO2_LOPTICS_RPR bulk_SiO2_LPEAD bulk_SiO2_LPEAD_nosym bulk_SiO2_LPEAD_RPR C_2x2x2_CORE_CON C_2x2x2_CORE_CON_RPR C_2x2x2_RPAFORCE Hg_x Hg_x_RPR Hg_y Hg_y_RPR Hg_z Hg_z_RPR mlwf_mos2_soc_locproj mlwf_mos2_soc_wannier90 ML_FeCrNi_ISTART3 ML_PbICs_ISTART3 ML_Phosphene_ISTART1 ML_RhCO_ISTART3 ML_ZrO_ISTART3 SiC8_ACFDT SiC8_ACFDTR SiC8_ACFDTRK SiC8_ACFDTR_SCK SiC8_G0W0R SiC8_G0W0RK SiC8_GW0R SiC8_GW0RK SiC8_QPGW0R SrVO3_CRPA_band_removal SrVO3_CRPA_band_removal_RPR SrVO3_CRPA_proj SrVO3_CRPA_proj_RPR SrVO3_CRPA_weighted SrVO3_CRPA_weighted_RPR SrVO3_NLRPA SrVO3_NLRPA_RPR Tl_x Tl_x_RPR Tl_y Tl_y_RPR Tl_z Tl_z_RPR

My users are always complaining about running VASP with Wannier 90 using AMD processors for all possible versions with gcc, aocc and oneAPI. I am still not able to figure out a working combination of versions of all these software.

I am attaching the makefile and testsuite.log also for reference. I ran the testsuite as root with proper settings for MPI. But, still the error persists.

Any inputs/suggestions related to this will be helpful.

Thanks in advance,

Best,
Kalyan

[michael_wolloch]

Dear Kalyan,

Note that there was a bug in the wannier90 interface in version 6.4.1, which was fixed in version 6.4.2. Please consult the known issues list on the wiki. It seems that other things are at play here, however, since you have tests failing that do not use wannier90.

I would nevertheless try and use the current version 6.4.3, if you have access to it, while I try and reproduce your errors with 6.4.1.

Please also provide the link-line and cpp-options of your executables by using the -l and -c command line arguments.

The make.inc of your wannier90 installation could also be valuable.

One thing I can tell you immediately: you should expect the *_Libxc tests to fail if you do not compile with the -DUSELIBXC

I will reply again if I have, or have not reproduced your errors.
Cheers, Michael

[michael_wolloch]

Dear Kalyan,

I was able to reproduce your problems with the default stack size of 8192 KiB on our AMD EPYC 7713 node. So the issue is not the toolchain, but your stack size.

Please check your stack size with

Code: Select all

ulimit -s

and set it to unlimited with

Code: Select all

ulimit -s unlimited

You might want to do this in the .bashrc of all your users.

Then run the failed tests (other than the libXC ones) again using:

Code: Select all

export VASP_TESTSUITE_TESTS="bulk_InP_SOC_G0W0_nosym bulk_InP_SOC_G0W0_sym bulk_InP_SOC_G0W0_sym_RPR bulk_SiO2_LOPTICS bulk_SiO2_LOPTICS_nosym bulk_SiO2_LOPTICS_RPR bulk_SiO2_LPEAD bulk_SiO2_LPEAD_nosym bulk_SiO2_LPEAD_RPR C_2x2x2_CORE_CON C_2x2x2_CORE_CON_RPR C_2x2x2_RPAFORCE Hg_x Hg_x_RPR Hg_y Hg_y_RPR Hg_z Hg_z_RPR mlwf_mos2_soc_locproj mlwf_mos2_soc_wannier90 ML_FeCrNi_ISTART3 ML_PbICs_ISTART3 ML_Phosphene_ISTART1 ML_RhCO_ISTART3 ML_ZrO_ISTART3 SiC8_ACFDT SiC8_ACFDTR SiC8_ACFDTRK SiC8_ACFDTR_SCK SiC8_G0W0R SiC8_G0W0RK SiC8_GW0R SiC8_GW0RK SiC8_QPGW0R SrVO3_CRPA_band_removal SrVO3_CRPA_band_removal_RPR SrVO3_CRPA_proj SrVO3_CRPA_proj_RPR SrVO3_CRPA_weighted SrVO3_CRPA_weighted_RPR SrVO3_NLRPA SrVO3_NLRPA_RPR Tl_x Tl_x_RPR Tl_y Tl_y_RPR Tl_z Tl_z_RPR"

Since you compiled with openMP support, it is a good idea to source the respective config file in the testsuite folder before you run the tests, if you have not already done that:

Code: Select all

source impi+omp.conf

For me, all those tests passed with unlimited stack size using our OneAPI 2022 toolchain (ifort version 2021.5.0, mkl 2022.0.1, impi 2021.5.0)

We will adapt our install documentation on the wiki to include the importance of unlimited stack size.

Note that you should still upgrade your version to fix the known bug in the Wannier90 interface that is present in 6.4.1!

Please let me know if this works for you,
Michael