[QE-users] gfortran vs intel lead to different results of electron-phonon coupling by ph.x

[Arena Konta]

Hi everyone,
 
Using the same input files (from one of the tutorials, step-by-step), I got different output files of dynamical matrices when using different intel compilators (Intel 13, Intel 15 and Intel 16). Only the results using GNU gfortran are correct. In all calculations, there is no significant error of freqs. In turn, all scf calculations (intel and gnu) also work fine.
 
Intel 15, QE 6.3
 
 q = (    0.000000000   0.000000000   0.000000000 )
 
 **************************************************************************
     freq (    1) =       0.485694 [THz] =      16.201000 [cm-1]
 (  0.833379  0.000000 -0.410303  0.000000  0.370314  0.000000 )
     freq (    2) =       0.485694 [THz] =      16.201000 [cm-1]
 ( -0.037325  0.000000  0.626695  0.000000  0.778370  0.000000 )
     freq (    3) =       0.485694 [THz] =      16.201000 [cm-1]
 ( -0.551441  0.000000 -0.662499  0.000000  0.506960  0.000000 )
 
  
 Intel 16, QE 6.3
 q = (    0.000000000   0.000000000   0.000000000 )
 
 **************************************************************************
     freq (    1) =       0.485682 [THz] =      16.200616 [cm-1]
 ( -0.635098  0.000000  0.108968  0.000000  0.764707  0.000000 )
     freq (    2) =       0.485682 [THz] =      16.200616 [cm-1]
 ( -0.688313  0.000000 -0.529109  0.000000 -0.496255  0.000000 )
     freq (    3) =       0.485682 [THz] =      16.200616 [cm-1]
 (  0.350537  0.000000 -0.841528  0.000000  0.411040  0.000000 )
 
 
 
 
 Intel 15, QE 6.2.1
  q = (    0.000000000   0.000000000   0.000000000 )
 
 **************************************************************************
     freq (    1) =       0.485689 [THz] =      16.200826 [cm-1]
 (  0.164642  0.000000  0.968532  0.000000 -0.186653  0.000000 )
     freq (    2) =       0.485689 [THz] =      16.200826 [cm-1]
 (  0.014161  0.000000  0.186894  0.000000  0.982278  0.000000 )
     freq (    3) =       0.485689 [THz] =      16.200826 [cm-1]
 ( -0.986252  0.000000  0.164367  0.000000 -0.017055  0.000000 )
 
 
The gfortran results are correct and the same like ones in the tutorial:
 
     q = (    0.000000000   0.000000000   0.000000000 )
 
 **************************************************************************
     freq (    1) =       0.486780 [THz] =      16.237227 [cm-1]
 (  0.550619  0.000000  0.765144  0.000000 -0.333726  0.000000 )
     freq (    2) =       0.486780 [THz] =      16.237227 [cm-1]
 (  0.063986  0.000000  0.359925  0.000000  0.930784  0.000000 )
     freq (    3) =       0.486780 [THz] =      16.237227 [cm-1]
 ( -0.832301  0.000000  0.533862 -0.000000 -0.149223  0.000000 )
 
 
Default compilations (We use Intel Xeon E5-2670):
DFLAGS         =  -D__FFTW3 -D__MPI -D__SCALAPACK
MPIF90         = mpif90
F90           = ifort
CC             = icc
F77            = ifort
CPP            = cpp
CPPFLAGS       = -P -traditional $(DFLAGS) $(IFLAGS)
CFLAGS         = -O3 $(DFLAGS) $(IFLAGS)
F90FLAGS       = $(FFLAGS) -nomodule -fpp $(FDFLAGS) $(CUDA_F90FLAGS) $(IFLAGS) $(MODFLAGS)
FFLAGS         = -O2 -assume byterecl -g -traceback
FFLAGS_NOOPT   = -O0 -assume byterecl -g -traceback
LD             = mpif90
LDFLAGS        = -static-intel
LD_LIBS        =
BLAS_LIBS      =   -lmkl_intel_lp64  -lmkl_sequential -lmkl_core
BLAS_LIBS_SWITCH = external
LAPACK_LIBS    =
LAPACK_LIBS_SWITCH = external
SCALAPACK_LIBS = -lmkl_scalapack_lp64 -lmkl_blacs_openmpi_lp64
FFT_LIBS       =  -lfftw3
 
 
Everything run on 1 processor. Any suggestions?

Tnak you in advance.


Arena Konta
The Institute of Thermophysics in Novosibirsk Scientific Center