[CP2K-user] [CP2K:22050] Re: Problems when running stress tensor calculations with MKL compiled CP2K

[Yu Zhu]

Problem solved. Thanks to Chatgpt.

This problem arises when the mkl version is too old (from intel2020u4) to 
be compatible with the intalled gcc-13.2.0.

However, switching to oneapi-2022.3 and use the classic Intel compiler, I 
found that the automatically generated .arch files wrongly linked " 
libmkl_gf_lp64.so " rather than " libmkl_intel_lp64.so ", both for versions 
2024.1 and 2025.1. To amend this, one only needs to execute " sed -i 
's/gf/intel/' ". And I have to say, CP2K compiled with Intel compiler and 
MKL runs much faster (about 20%) than that  compiled with GNU compiler and 
openblas.

在2025年3月13日星期四 UTC+8 18:36:50<Yu Zhu> 写道：

> Dear CP2K developers and users, 
>
> I am testing a stress tensor calculation example with my newly compiled 
> CP2K-2025.1 on Platinum 9282 CPUs and I found something wrong:
>
> With the binary compiled with "--with-gcc=install --with-intel=no 
> --with-mpich=install* --with-mkl=system*" tags , the MPI threw an error 
> and the calculation cannot be performed if multiple k-points (say, 2*1*1) 
> were to be sampled, but normally terminated if only gamma point was 
> considered. Since the version compiled with "--with-gcc= 
> install  --with-intel=no --with-mpich=install *--with-openblas=install*" 
> can perform the calculation with multiple k-points without error  (though 
> much solwer than the apptainer image), I suspect that there is something 
> wrong in the version compiled with MKL and wonder if there is any 
> workaround owning to the high performance of MKL.
>
> The input file looks like this:
> '''
> &GLOBAL
>   PROJECT StressTensor
>   PRINT_LEVEL LOW
>   RUN_TYPE GEO_OPT
> &END GLOBAL
>
> &FORCE_EVAL
>   METHOD Quickstep
>   STRESS_TENSOR ANALYTICAL
>   &SUBSYS
>     &CELL
>       A    
>       B     
>       C     
>       PERIODIC XYZ #Direction(s) of applied PBC (geometry aspect)
>     &END CELL
>     &COORD
>
>     &END COORD
>     &KIND Cu
>       ELEMENT Cu
>       BASIS_SET DZVP-MOLOPT-SR-GTH-q11
>       POTENTIAL GTH-PBE
>     &END KIND
>     &KIND H
>       ELEMENT H
>       BASIS_SET TZVP-MOLOPT-GTH-q1
>       POTENTIAL GTH-PBE
>     &END KIND
>     &KIND O
>       ELEMENT O
>       BASIS_SET TZVP-MOLOPT-GTH-q6
>       POTENTIAL GTH-PBE
>     &END KIND
>   &END SUBSYS
>
>   &DFT
>     BASIS_SET_FILE_NAME  BASIS_MOLOPT
>     POTENTIAL_FILE_NAME  POTENTIAL
>     CHARGE    0 #Net charge
>     MULTIPLICITY    1 #Spin multiplicity
>     &KPOINTS
>       SCHEME MONKHORST-PACK  2  1  1
>     &END KPOINTS
>     &QS
>       EPS_DEFAULT 1.0E-12 #Set all EPS_xxx to values such that the energy 
> will be correct up to this value
>     &END QS
>     &POISSON
>       PERIODIC XYZ #Direction(s) of PBC for calculating electrostatics
>       PSOLVER PERIODIC #The way to solve Poisson equation
>     &END POISSON
>     &XC
>       &XC_FUNCTIONAL PBE
>       &END XC_FUNCTIONAL
>       &VDW_POTENTIAL
>         POTENTIAL_TYPE PAIR_POTENTIAL
>         &PAIR_POTENTIAL
>           PARAMETER_FILE_NAME dftd3.dat
>           TYPE DFTD3(BJ)
>           REFERENCE_FUNCTIONAL PBE
>           CALCULATE_C9_TERM T #Calculate C9-related three-body term, more 
> accurate for large system
>         &END PAIR_POTENTIAL
>       &END VDW_POTENTIAL
>     &END XC
>     &MGRID
>       CUTOFF  480
>       REL_CUTOFF  60
>       NGRIDS 5 #The number of multigrids to use. 5 is optimal for 
> MOLOPT-GTH basis sets
>     &END MGRID
>     &SCF
>       MAX_SCF 256
>       EPS_SCF 1.0E-06 #Convergence threshold of density matrix of inner SCF
>       &DIAGONALIZATION
>         ALGORITHM STANDARD #Algorithm for diagonalization
>       &END DIAGONALIZATION
>       &MIXING #How to mix old and new density matrices
>         METHOD BROYDEN_MIXING #PULAY_MIXING is also a good alternative
>         ALPHA 0.4 #Default. Mixing 40% of new density matrix with the old 
> one
>         NBROYDEN 8 #Default is 4. Number of previous steps stored for the 
> actual mixing scheme
>       &END MIXING
>       &SMEAR
>         METHOD FERMI_DIRAC
>         ELECTRONIC_TEMPERATURE 30 #Electronic temperature of Fermi-Dirac 
> smearing in K
>       &END SMEAR
>       ADDED_MOS   504 #Number of virtual MOs to solve
>       &PRINT
>         &RESTART #Note: Use "&RESTART OFF" can prevent generating .wfn file
>           BACKUP_COPIES 0 #Maximum number of backup copies of wfn file. 0 
> means never
>         &END RESTART
>       &END PRINT
>     &END SCF
>   &END DFT
>
>   &PRINT
>     &STRESS_TENSOR
>     &END STRESS_TENSOR
>   &END PRINT
> &END FORCE_EVAL
>
> &MOTION
>   &GEO_OPT
>     TYPE MINIMIZATION #Search for minimum
>     KEEP_SPACE_GROUP F #If T, then space group will be detected and 
> preserved
>     OPTIMIZER CG #Can also be BFGS or LBFGS
>     &CG
>       &LINE_SEARCH
>         TYPE 2PNT #Two-point extrapolation, cheap while acceptable. Can 
> also be FIT, GOLD
>       &END LINE_SEARCH
>     &END CG
>     MAX_ITER 0 #Maximum number of geometry optimization
>     MAX_DR 3E-3 #Maximum geometry change
>     RMS_DR 1.5E-3 #RMS geometry change
>     MAX_FORCE 4.5E-4 #Maximum force
>     RMS_FORCE 3E-4 #RMS force
>   &END GEO_OPT
>   &CONSTRAINT
>     &FIXED_ATOMS #Set atoms to be fixed
>       COMPONENTS_TO_FIX XYZ #Which fractional components will be fixed, 
> can be X, Y, Z, XY, XZ, YZ, XYZ
>       LIST       5 6 11 12 17 18 23 24 29 30 35 36 41 42 47 48 53 54 59 60 
> 65 66 71 72 77 78 83 84 89 90 95 96 101 102 107 108 113 114 119 120 125 126 
> 131 132 137 138 143 144 149 150 155 156 161 162 167 168 173 174 179 180 185 
> 186 191 192 197 198 203 204 209 210 215 216 221 222 227 228 233 234 239 240 
> 245 246 251 252 257 258 263 264 269 270 275 276 281 282 287 288 293 294 299 
> 300 305 306 311 312 317 318 323 324 329 330 335 336 341 342 347 348 353 354 
> 359 360 365 366 371 372 377 378 383 384 389 390 395 396 401 402 407 408 413 
> 414 419 420 425 426 431 432
>     &END FIXED_ATOMS
>     &FIXED_ATOMS #Set atoms to be fixed
>       COMPONENTS_TO_FIX Z #Which fractional components will be fixed, can 
> be X, Y, Z, XY, XZ, YZ, XYZ
>       LIST       721..1008
>     &END FIXED_ATOMS
>   &END CONSTRAINT
>   &PRINT
>     &TRAJECTORY
>       FORMAT xyz
>     &END TRAJECTORY
>     &RESTART
>       BACKUP_COPIES 0 #Maximum number of backing up restart file, 0 means 
> never
>     &END RESTART
>   &END PRINT
> &END MOTION
> '''
>
> BTW, it's found that the binary compiled with openmpi-5.0.6 shows severe 
> performance problems compared to openmpi-4.1.5 or mpich-4.0.3 in our 
> cluster, and so did the apptainer image, but this is not so urgent :)
>
> All your suggestions are greatly appreciated.
>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/a4939d41-2c45-4de2-a924-1784f49283d8n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260120/93705629/attachment.htm>