10
0
mirror of https://github.com/LCPQ/quantum_package synced 2024-12-22 20:35:19 +01:00

Parallelization of Davidson

This commit is contained in:
Anthony Scemama 2017-04-19 19:45:18 +02:00
parent 27d07d7676
commit cbf8c54d70
3 changed files with 37 additions and 10 deletions

View File

@ -271,6 +271,11 @@ subroutine H_S2_u_0_nstates_zmq(v_0,s_0,u_0,N_st,sze)
double precision, allocatable :: u_t(:,:)
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: u_t
PROVIDE psi_det_beta_unique psi_bilinear_matrix_order_transp_reverse psi_det_alpha_unique
PROVIDE psi_bilinear_matrix_transp_values psi_bilinear_matrix_values psi_bilinear_matrix_columns_loc
PROVIDE ref_bitmask_energy nproc
allocate(u_t(N_st,N_det))
do k=1,N_st
call dset_order(u_0(1,k),psi_bilinear_matrix_order,N_det)
@ -290,7 +295,6 @@ subroutine H_S2_u_0_nstates_zmq(v_0,s_0,u_0,N_st,sze)
ASSERT (Nint > 0)
ASSERT (Nint == N_int)
ASSERT (n>0)
PROVIDE ref_bitmask_energy nproc
call new_parallel_job(zmq_to_qp_run_socket,'davidson')
@ -335,15 +339,39 @@ subroutine H_S2_u_0_nstates_zmq(v_0,s_0,u_0,N_st,sze)
deallocate(u_t)
! Create tasks
! ============
integer :: istep, imin, imax, ishift
istep=2
do imin=1,N_det, 1048576
double precision :: w, max_workload, N_det_inv, di
max_workload = N_det_beta_unique+N_det_alpha_unique
w = 0.d0
istep=4
ishift=0
imin=1
N_det_inv = 1.d0/dble(N_det)
di = dble(N_det)
do imax=1,N_det
di = di-1.d0
w = w + (di*N_det_inv)**2
if (w > max_workload) then
do ishift=0,istep-1
write(task,'(4(I9,1X),1A)') imin, imax, ishift, istep, '|'
call add_task_to_taskserver(zmq_to_qp_run_socket,trim(task))
enddo
istep = max(istep-1,1)
imin = imax+1
w = 0.d0
endif
enddo
if (w > 0.d0) then
imax = N_det
do ishift=0,istep-1
imax = min(N_det, imin+1048576-1)
write(task,'(4(I9,1X),1A)') imin, imax, ishift, istep, '|'
call add_task_to_taskserver(zmq_to_qp_run_socket,trim(task))
enddo
enddo
endif
v_0 = 0.d0
s_0 = 0.d0

View File

@ -86,7 +86,7 @@ subroutine H_S2_u_0_nstates_openmp_work(v_0,s_0,u_t,N_st,sze,istart,iend,ishift,
double precision, intent(out) :: v_0(sze,N_st), s_0(sze,N_st)
PROVIDE ref_bitmask_energy
PROVIDE ref_bitmask_energy N_int
select case (N_int)
case (1)
@ -136,7 +136,6 @@ subroutine H_S2_u_0_nstates_openmp_work_$N_int(v_0,s_0,u_t,N_st,sze,istart,iend,
integer*8 :: k8
double precision, allocatable :: v_t(:,:), s_t(:,:)
!DIR$ ATTRIBUTES ALIGN : $IRP_ALIGN :: v_t, s_t
PROVIDE N_int
maxab = max(N_det_alpha_unique, N_det_beta_unique)+1
allocate(idx0(maxab))
@ -148,6 +147,7 @@ subroutine H_S2_u_0_nstates_openmp_work_$N_int(v_0,s_0,u_t,N_st,sze,istart,iend,
! Prepare the array of all alpha single excitations
! -------------------------------------------------
PROVIDE N_int
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(psi_bilinear_matrix_rows, N_det, &
!$OMP psi_bilinear_matrix_columns, &
@ -157,9 +157,8 @@ subroutine H_S2_u_0_nstates_openmp_work_$N_int(v_0,s_0,u_t,N_st,sze,istart,iend,
!$OMP psi_bilinear_matrix_transp_columns, &
!$OMP psi_bilinear_matrix_transp_order, N_st, &
!$OMP psi_bilinear_matrix_order_transp_reverse, &
!$OMP singles_alpha_csc, singles_alpha_csc_idx, &
!$OMP psi_bilinear_matrix_columns_loc, &
!$OMP singles_alpha_size, istart, iend, istep, &
!$OMP istart, iend, istep, &
!$OMP ishift, idx0, u_t, maxab, v_0, s_0) &
!$OMP PRIVATE(krow, kcol, tmp_det, spindet, k_a, k_b, i, &
!$OMP lcol, lrow, l_a, l_b, nmax, &

View File

@ -541,7 +541,7 @@ BEGIN_PROVIDER [ integer, psi_bilinear_matrix_transp_rows_loc, (N_det_alpha_uniq
psi_bilinear_matrix_transp_rows_loc(l) = k
endif
enddo
psi_bilinear_matrix_transp_rows_loc(N_det_beta_unique+1) = N_det+1
psi_bilinear_matrix_transp_rows_loc(N_det_alpha_unique+1) = N_det+1
END_PROVIDER
BEGIN_PROVIDER [ integer, psi_bilinear_matrix_order_transp_reverse , (N_det) ]