I am trying o use oneapi together with Quantum Espresso. I've installed QE that successfully detected oneapi environment, linked to mpi, fortran, c++, and MKL. After installation, all tests of QE's test-suit were passed. However, when trying to run QE with bigger structures, the code stuck in the initial diagonalization steps. I've changed the diagonalization method, the problem persisted. I thought the error was in the structure itself, but using old versions of compilers and MKL the code successfully performed the diagonalization.
Please find attached the QE's input file, the pseudopotentials, and the output produced.
Anyone already faced this kind of problem?
Thanks for reaching out to us.
>> After installation, all tests of QE's test-suit were passed. when trying to run QE with bigger structures
Could you please provide us the exact steps that you have followed to run the code so that we can work on it from our end?
>> old versions of compilers and MKL the code successfully performed the diagonalization
Could you please let us know the versions of the compilers with which the diagonalization was performed successfully?
It would be a great help if you also let us know your environment details
OS & version
I'm sorry for the latter response, I was in trouble login into my Intel account.
>>Could you please provide us the exact steps that you have followed to run the code so that we can work on it from our end?
To run the code I've used:
mpirun -np 18 pw.x -i inputfile.in > outputfile.out
the input file is the one I've sent with the original post.
>>Could you please let us know the versions of the compilers with which the diagonalization was performed successfully?
Diagonalization is part of the self-consistent calculation. It was performed successfully using compilers from Parallel Studio 2017 and using free compilers such as mpif90, gcc, gfortran, and algebra libraries downloaded and compiled from
Please find attached the make.inc file with the configuration options to which QE ran the simulation. QE ran succesfull using my personal computer Asus notebook with Intel I7 8Gb RAM running Ubuntu 18 LTS, and in my Dell workstation with Intel E52400 64Gb RAM running Ubuntu 20 LTS.
Could you please provide us with the input matrix that you have used to perform the diagonalization? So that it will help us to get more insights on why the code is getting stuck with the latest version of MKL.
Unfortunately, I can't. I'm only a user, not a developer. Even though Quantum Espresso is an open-source project, the codes are quite complex for a normal user to know exactly where the magic happens. The only thing I can say is that Quantum Espresso employs at least two iterative diagonalization schemes (from https://www.quantum-espresso.org/Doc/INPUT_PW.html#idm845
Davidson iterative diagonalization with overlap matrix (default). Fast, may in some rare cases fail.
Conjugate-gradient-like band-by-band diagonalization. MUCH slower than 'david' but uses less memory and is (a little bit) more robust.
Could you please share the results of MKL_VERBOSE for both current and previous versions of MKL?
To do this, use the below command
and then run your executable.
For more details regarding the above environment variable, please refer to the below link.
We are closing this thread for now assuming that your issue has been resolved. Please post a new question if you need any additional information from Intel as this thread will no longer be monitored.