scnc
/
quantum-espresso
mirror of https://gitlab.com/QEF/q-e.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Two bugs fixed by Taylor in new band parallelization of EXX at Gamma: 

- incorrect logic for some specific cases
- tiny error coming from conversion of two real numbers into a complex one:
  CMPLX has the bad habit of converting to single precision complex unless
  explicitly instructed to convert to double! This fixes mysterious crashes
  with negative dexx close to convergence.
Credits and references added


git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@13313 c92efa57-630b-4861-b058-cf58834340f0
This commit is contained in:
giannozz 2017-02-17 12:24:27 +00:00
parent 4503e67f24
commit 08b7224019
1 changed files with 32 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
PW/src/exx.f90
View File

@ -1,4 +1,4 @@
!! Copyright (C) 2005-2015 Quantum ESPRESSO group
! Copyright (C) 2005-2017 Quantum ESPRESSO group
! This file is distributed under the terms of the
! GNU General Public License. See the file `License'
! in the root directory of the present distribution,
@ -8,6 +8,12 @@
MODULE exx
!--------------------------------------
!
! Variables and subroutines for calculation of exact-exchange contribution
! Implements ACE: Lin Lin, J. Chem. Theory Comput. 2016, 12, 2242
! Contains code for band parallelization over pairs of bands: see T. Barnes,
! T. Kurth, P. Carrier, N. Wichmann, D. Prendergast, P.R.C. Kent, J. Deslippe
! Computer Physics Communications, 2017
!
USE kinds, ONLY : DP
USE coulomb_vcut_module, ONLY : vcut_init, vcut_type, vcut_info, &
vcut_get, vcut_spheric_get
@ -769,7 +775,7 @@ MODULE exx
INTEGER :: npw, current_ik
INTEGER, EXTERNAL :: global_kpoint_index
INTEGER :: ibnd_start_new, ibnd_end_new
INTEGER :: ibnd_exx
INTEGER :: ibnd_exx, evc_offset
!
CALL start_clock ('exxinit')
!
@ -851,10 +857,10 @@ MODULE exx
IF ( gamma_only ) THEN
ibnd_buff_start = ibnd_start_new/2
IF(mod(ibnd_start,2)==1) ibnd_buff_start = ibnd_buff_start +1
IF(mod(iexx_start,2)==1) ibnd_buff_start = ibnd_buff_start +1
!
ibnd_buff_end = ibnd_end_new/2
IF(mod(ibnd_end,2)==1) ibnd_buff_end = ibnd_buff_end +1
IF(mod(iexx_end,2)==1) ibnd_buff_end = ibnd_buff_end +1
ELSE
ibnd_buff_start = ibnd_start_new
ibnd_buff_end = ibnd_end_new
@ -903,6 +909,7 @@ MODULE exx
ibnd_loop_start=iexx_start
ENDIF
evc_offset = 0
DO ibnd = ibnd_loop_start, iexx_end, 2
ibnd_exx = ibnd
h_ibnd = h_ibnd + 1
@ -912,21 +919,22 @@ MODULE exx
IF ( ibnd < iexx_end ) THEN
IF ( ibnd == ibnd_loop_start .and. MOD(iexx_start,2) == 0 ) THEN
DO ig=1,exx_fft%npwt
psic_exx(exx_fft%nlt(ig)) = ( 0._dp, 1._dp )*evc_exx(ig,ibnd-ibnd_loop_start+2)
psic_exx(exx_fft%nltm(ig)) = ( 0._dp, 1._dp )*conjg(evc_exx(ig,ibnd-ibnd_loop_start+2))
psic_exx(exx_fft%nlt(ig)) = ( 0._dp, 1._dp )*evc_exx(ig,1)
psic_exx(exx_fft%nltm(ig)) = ( 0._dp, 1._dp )*conjg(evc_exx(ig,1))
ENDDO
evc_offset = -1
ELSE
DO ig=1,exx_fft%npwt
psic_exx(exx_fft%nlt(ig)) = evc_exx(ig,ibnd-ibnd_loop_start+1) &
+ ( 0._dp, 1._dp ) * evc_exx(ig,ibnd-ibnd_loop_start+2)
psic_exx(exx_fft%nltm(ig)) = conjg( evc_exx(ig,ibnd-ibnd_loop_start+1) ) &
+ ( 0._dp, 1._dp ) * conjg( evc_exx(ig,ibnd-ibnd_loop_start+2) )
psic_exx(exx_fft%nlt(ig)) = evc_exx(ig,ibnd-ibnd_loop_start+evc_offset+1) &
+ ( 0._dp, 1._dp ) * evc_exx(ig,ibnd-ibnd_loop_start+evc_offset+2)
psic_exx(exx_fft%nltm(ig)) = conjg( evc_exx(ig,ibnd-ibnd_loop_start+evc_offset+1) ) &
+ ( 0._dp, 1._dp ) * conjg( evc_exx(ig,ibnd-ibnd_loop_start+evc_offset+2) )
ENDDO
END IF
ELSE
DO ig=1,exx_fft%npwt
psic_exx(exx_fft%nlt (ig)) = evc_exx(ig,ibnd-ibnd_loop_start+1)
psic_exx(exx_fft%nltm(ig)) = conjg( evc_exx(ig,ibnd-ibnd_loop_start+1) )
psic_exx(exx_fft%nlt (ig)) = evc_exx(ig,ibnd-ibnd_loop_start+evc_offset+1)
psic_exx(exx_fft%nltm(ig)) = conjg( evc_exx(ig,ibnd-ibnd_loop_start+evc_offset+1) )
ENDDO
ENDIF
@ -1318,6 +1326,7 @@ MODULE exx
COMPLEX(DP), ALLOCATABLE :: exxtemp(:,:), psiwork(:)
INTEGER :: ijt, njt, jblock_start, jblock_end
INTEGER :: index_start, index_end, exxtemp_index
INTEGER :: ending_im
!
ialloc = nibands(my_egrp_id+1)
!
@ -1385,8 +1394,8 @@ MODULE exx
l_fft_doubleband = .false.
l_fft_singleband = .false.
!
IF ( mod(ii,2)==1 .and. (ii+1)<=nibands(my_egrp_id+1) ) l_fft_doubleband = .true.
IF ( mod(ii,2)==1 .and. ii==nibands(my_egrp_id+1) ) l_fft_singleband = .true.
IF ( mod(ii,2)==1 .and. (ii+1)<=min(m,nibands(my_egrp_id+1)) ) l_fft_doubleband = .true.
IF ( mod(ii,2)==1 .and. ii==min(m,nibands(my_egrp_id+1)) ) l_fft_singleband = .true.
!
IF( l_fft_doubleband ) THEN
!$omp parallel do default(shared), private(ig)
@ -1594,7 +1603,12 @@ MODULE exx
!
CALL result_sum(n*npol, m, big_result)
IF (iexx_istart(my_egrp_id+1).gt.0) THEN
DO im=1, iexx_iend(my_egrp_id+1) - iexx_istart(my_egrp_id+1) + 1
IF (negrp == 1) then
ending_im = m
ELSE
ending_im = iexx_iend(my_egrp_id+1) - iexx_istart(my_egrp_id+1) + 1
END IF
DO im=1, ending_im
!$omp parallel do default(shared), private(ig) firstprivate(im,n)
DO ig = 1, n
hpsi(ig,im)=hpsi(ig,im) + big_result(ig,im+iexx_istart(my_egrp_id+1)-1)
@ -5134,13 +5148,14 @@ END SUBROUTINE compute_becpsi
! two real bands can be coupled into a single complex one
DO ir=1, lda*npol
exxtemp(ir,1+(jbnd-jstart+1)/2) = CMPLX( work(ir,jbnd-jstart+1),&
work(ir,jbnd-jstart+2) )
work(ir,jbnd-jstart+2),&
KIND=dp )
END DO
ELSE
! case of lone last band
DO ir=1, lda*npol
exxtemp(ir,1+(jbnd-jstart+1)/2) = CMPLX( work(ir,jbnd-jstart+1),&
0.0_dp )
0.0_dp, KIND=dp )
END DO
ENDIF
END DO