Cleanup get_oper.F90

[ponweist]

When re-running the testcase from #2 in order to compare the performance after parallelization of kpath (see #5), initialization phase is very inefficient again:

Testcase parameters

32sm running at 64 processes:

kpath = T
kpath_task = curv
kpath_num_points = 500
kpath_bands_colour = spin

kslice = F

berry = T
berry_task = ahc
berry_kmesh = 48 48 48

Analysis

The reason for the inefficiency is that for the specific testcase parameters instead of the optimized routine get_morb_R (see #3), get_ahc_R is called, containing again the inefficient loop (get_oper.F90, lines 402ff.):

          ! Wannier-gauge overlap matrix S in the projected subspace
          !
          call get_win_min(ik,winmin_q)
          call get_win_min(nnlist(ik,nn),winmin_qb)
          S=cmplx_0
          do m=1,num_wann
             do n=1,num_wann
                do i=1,num_states(ik)
                   ii=winmin_q+i-1
                   do j=1,num_states(nnlist(ik,nn))
                      jj=winmin_qb+j-1
                      S(n,m)=S(n,m)&
                           +conjg(v_matrix(i,n,ik))*S_o(ii,jj)&
                           *v_matrix(j,m,nnlist(ik,nn))
                   end do
                end do
             end do
          end do

TODO

• Cleanup get_oper.F90 and minimize duplicated code.
  • Possible approach: Join different get_* routines to a single routine, providing logical flags as parameters for indicating which matrices need to be initialized.
• Consistently use get_gauge_overlap_matrix instead of nested loops similar to the above code snippet.
  • Optoinal: Think about better names for get_gauge_overlap_matrix and its parameters.

[ponweist]

Trace after e6d520d:

Time for initialization is down from 143s to 8s.