From b03709020df45806e59d2dfc18303285d69696d4 Mon Sep 17 00:00:00 2001 From: Anthony Scemama Date: Sun, 4 Jun 2023 16:45:38 +0200 Subject: [PATCH] 5idx arrays need O(N3) temp arrays --- src/bi_ort_ints/bi_ort_ints.irp.f | 34 +-- src/bi_ort_ints/three_body_ijmkl.irp.f | 330 +++++++++++++------------ 2 files changed, 187 insertions(+), 177 deletions(-) diff --git a/src/bi_ort_ints/bi_ort_ints.irp.f b/src/bi_ort_ints/bi_ort_ints.irp.f index 42bbe315..5618a2cd 100644 --- a/src/bi_ort_ints/bi_ort_ints.irp.f +++ b/src/bi_ort_ints/bi_ort_ints.irp.f @@ -8,8 +8,8 @@ program bi_ort_ints my_n_pt_a_grid = 14 touch my_grid_becke my_n_pt_r_grid my_n_pt_a_grid ! call test_3e + call test_5idx2 call test_5idx -! call test_5idx2 end subroutine test_5idx2 @@ -60,6 +60,8 @@ subroutine test_5idx k = 1 n = 0 accu = 0.d0 + PROVIDE three_e_5_idx_direct_bi_ort_old + do i = 1, mo_num do k = 1, mo_num do j = 1, mo_num @@ -69,28 +71,28 @@ subroutine test_5idx ! stop ! endif -! new = three_e_5_idx_direct_bi_ort(m,l,j,k,i) -! ref = three_e_5_idx_direct_bi_ort_old(m,l,j,k,i) + new = three_e_5_idx_direct_bi_ort(m,l,j,k,i) + ref = three_e_5_idx_direct_bi_ort_old(m,l,j,k,i) + contrib = dabs(new - ref) + accu += contrib + if(contrib .gt. 1.d-10)then + print*,'direct' + print*,i,k,j,l,m + print*,ref,new,contrib + stop + endif +! +! new = three_e_5_idx_exch12_bi_ort(m,l,j,k,i) +! ref = three_e_5_idx_exch12_bi_ort_old(m,l,j,k,i) ! contrib = dabs(new - ref) ! accu += contrib ! if(contrib .gt. 1.d-10)then -! print*,'direct' +! print*,'exch12' ! print*,i,k,j,l,m ! print*,ref,new,contrib ! stop ! endif ! - new = three_e_5_idx_exch12_bi_ort(m,l,j,k,i) - ref = three_e_5_idx_exch12_bi_ort_old(m,l,j,k,i) - contrib = dabs(new - ref) - accu += contrib - if(contrib .gt. 1.d-10)then - print*,'exch12' - print*,i,k,j,l,m - print*,ref,new,contrib - stop - endif - ! ! new = three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) ! ref = three_e_5_idx_cycle_1_bi_ort_old(m,l,j,k,i) @@ -135,7 +137,7 @@ subroutine test_5idx ! print*,ref,new,contrib ! stop ! endif -! + enddo enddo enddo diff --git a/src/bi_ort_ints/three_body_ijmkl.irp.f b/src/bi_ort_ints/three_body_ijmkl.irp.f index 7b39235b..6e46637d 100644 --- a/src/bi_ort_ints/three_body_ijmkl.irp.f +++ b/src/bi_ort_ints/three_body_ijmkl.irp.f @@ -15,7 +15,7 @@ end ! ! matrix element of the -L three-body operator FOR THE DIRECT TERMS OF DOUBLE EXCITATIONS AND BI ORTHO MOs ! - ! three_e_5_idx_direct_bi_ort(m,l,j,k,i) = ::: notice that i is the RIGHT MO and k is the LEFT MO + ! three_e_5_idx_direct_bi_ort(m,l,j,k,i) = :: : notice that i is the RIGHT MO and k is the LEFT MO ! ! notice the -1 sign: in this way three_e_3_idx_direct_bi_ort can be directly used to compute Slater rules with a + sign ! @@ -23,210 +23,218 @@ end END_DOC implicit none - integer :: i, j, k, m, l - double precision :: wall1, wall0 - integer :: ipoint - double precision, allocatable :: grad_mli(:,:,:), orb_mat(:,:,:) - double precision, allocatable :: lk_grad_mi(:,:,:,:), rk_grad_im(:,:,:,:) - double precision, allocatable :: lm_grad_ik(:,:,:,:), rm_grad_ik(:,:,:,:) - double precision, allocatable :: tmp_mat(:,:,:,:) - allocate(tmp_mat(mo_num,mo_num,mo_num,mo_num)) + integer :: i, j, k, m, l + double precision :: wall1, wall0 + integer :: ipoint + double precision, allocatable :: grad_mli(:,:), orb_mat(:,:,:) + double precision, allocatable :: lk_grad_mi(:,:,:,:), rk_grad_im(:,:,:) + double precision, allocatable :: lm_grad_ik(:,:,:,:), rm_grad_ik(:,:,:) + double precision, allocatable :: tmp_mat(:,:,:) provide mos_r_in_r_array_transp mos_l_in_r_array_transp PROVIDE mo_l_coef mo_r_coef int2_grad1_u12_bimo_t + call print_memory_usage print *, ' Providing the three_e_5_idx_bi_ort ...' call wall_time(wall0) - do m = 1, mo_num + three_e_5_idx_direct_bi_ort (:,:,:,:,:) = 0.d0 + three_e_5_idx_cycle_1_bi_ort(:,:,:,:,:) = 0.d0 + three_e_5_idx_cycle_2_bi_ort(:,:,:,:,:) = 0.d0 + three_e_5_idx_exch23_bi_ort (:,:,:,:,:) = 0.d0 + three_e_5_idx_exch13_bi_ort (:,:,:,:,:) = 0.d0 - allocate(grad_mli(n_points_final_grid,mo_num,mo_num)) + call print_memory_usage + + allocate(tmp_mat(mo_num,mo_num,mo_num)) allocate(orb_mat(n_points_final_grid,mo_num,mo_num)) - !$OMP PARALLEL & - !$OMP DEFAULT (NONE) & - !$OMP PRIVATE (i,l,ipoint) & - !$OMP SHARED (m,mo_num,n_points_final_grid, & - !$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, & - !$OMP int2_grad1_u12_bimo_t, final_weight_at_r_vector, & - !$OMP grad_mli, orb_mat) - !$OMP DO COLLAPSE(2) + + !$OMP PARALLEL DO PRIVATE (i,l,ipoint) do i=1,mo_num do l=1,mo_num - do ipoint=1, n_points_final_grid + do ipoint=1, n_points_final_grid - grad_mli(ipoint,l,i) = final_weight_at_r_vector(ipoint) * ( & - int2_grad1_u12_bimo_t(ipoint,1,m,m) * int2_grad1_u12_bimo_t(ipoint,1,l,i) + & - int2_grad1_u12_bimo_t(ipoint,2,m,m) * int2_grad1_u12_bimo_t(ipoint,2,l,i) + & - int2_grad1_u12_bimo_t(ipoint,3,m,m) * int2_grad1_u12_bimo_t(ipoint,3,l,i) ) + orb_mat(ipoint,l,i) = final_weight_at_r_vector(ipoint) & + * mos_l_in_r_array_transp(ipoint,l) & + * mos_r_in_r_array_transp(ipoint,i) - orb_mat(ipoint,l,i) = mos_l_in_r_array_transp(ipoint,l) * mos_r_in_r_array_transp(ipoint,i) - - enddo - enddo - enddo - !$OMP END DO - !$OMP END PARALLEL - - call dgemm('T','N', mo_num*mo_num, mo_num*mo_num, n_points_final_grid, 1.d0, & - orb_mat, n_points_final_grid, & - grad_mli, n_points_final_grid, 0.d0, & - tmp_mat, mo_num*mo_num) - - !$OMP PARALLEL DO PRIVATE(i,j,k,l) - do i = 1, mo_num - do k = 1, mo_num - do j = 1, mo_num - do l = 1, mo_num - three_e_5_idx_direct_bi_ort(m,l,j,k,i) = - tmp_mat(l,j,k,i) - tmp_mat(k,i,l,j) - enddo enddo enddo enddo !$OMP END PARALLEL DO - deallocate(orb_mat,grad_mli) + tmp_mat = 0.d0 + call print_memory_usage +! + do m = 1, mo_num - allocate(lm_grad_ik(n_points_final_grid,3,mo_num,mo_num)) - allocate(rm_grad_ik(n_points_final_grid,3,mo_num,mo_num)) - allocate(rk_grad_im(n_points_final_grid,3,mo_num,mo_num)) + allocate(grad_mli(n_points_final_grid,mo_num)) - !$OMP PARALLEL & - !$OMP DEFAULT (NONE) & - !$OMP PRIVATE (i,l,ipoint) & - !$OMP SHARED (m,mo_num,n_points_final_grid, & - !$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, & - !$OMP int2_grad1_u12_bimo_t, final_weight_at_r_vector, & - !$OMP rm_grad_ik, lm_grad_ik, rk_grad_im, lk_grad_mi) - !$OMP DO COLLAPSE(2) - do i=1,mo_num - do l=1,mo_num - do ipoint=1, n_points_final_grid + do i=1,mo_num + !$OMP PARALLEL DO PRIVATE (l,ipoint) + do l=1,mo_num + do ipoint=1, n_points_final_grid - lm_grad_ik(ipoint,1,l,i) = mos_l_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,1,l,i) * final_weight_at_r_vector(ipoint) - lm_grad_ik(ipoint,2,l,i) = mos_l_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,2,l,i) * final_weight_at_r_vector(ipoint) - lm_grad_ik(ipoint,3,l,i) = mos_l_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,3,l,i) * final_weight_at_r_vector(ipoint) + grad_mli(ipoint,l) = & + int2_grad1_u12_bimo_t(ipoint,1,m,m) * int2_grad1_u12_bimo_t(ipoint,1,l,i) +& + int2_grad1_u12_bimo_t(ipoint,2,m,m) * int2_grad1_u12_bimo_t(ipoint,2,l,i) +& + int2_grad1_u12_bimo_t(ipoint,3,m,m) * int2_grad1_u12_bimo_t(ipoint,3,l,i) - rm_grad_ik(ipoint,1,l,i) = mos_r_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,1,l,i) - rm_grad_ik(ipoint,2,l,i) = mos_r_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,2,l,i) - rm_grad_ik(ipoint,3,l,i) = mos_r_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,3,l,i) + enddo + enddo + !$OMP END PARALLEL DO - rk_grad_im(ipoint,1,l,i) = mos_r_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,1,i,m) - rk_grad_im(ipoint,2,l,i) = mos_r_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,2,i,m) - rk_grad_im(ipoint,3,l,i) = mos_r_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,3,i,m) + call dgemm('T','N', mo_num*mo_num, mo_num, n_points_final_grid, 1.d0,& + orb_mat, n_points_final_grid, & + grad_mli, n_points_final_grid, 0.d0, & + tmp_mat, mo_num*mo_num) - enddo - enddo - enddo - !$OMP END DO - !$OMP END PARALLEL - call dgemm('T','N', mo_num*mo_num, mo_num*mo_num, 3*n_points_final_grid, 1.d0, & - lm_grad_ik, 3*n_points_final_grid, & - rm_grad_ik, 3*n_points_final_grid, 0.d0, & - tmp_mat, mo_num*mo_num) - - !$OMP PARALLEL DO PRIVATE(i,j,k,l) - do i = 1, mo_num - do k = 1, mo_num + !$OMP PARALLEL PRIVATE(j,k,l) + !$OMP DO + do k = 1, mo_num + do j = 1, mo_num + do l = 1, mo_num + three_e_5_idx_direct_bi_ort(m,l,j,k,i) = three_e_5_idx_direct_bi_ort(m,l,j,k,i) - tmp_mat(l,j,k) + enddo + enddo + enddo + !$OMP END DO + !$OMP DO do j = 1, mo_num do l = 1, mo_num - three_e_5_idx_direct_bi_ort(m,l,j,k,i) = three_e_5_idx_direct_bi_ort(m,l,j,k,i) - tmp_mat(l,j,k,i) + do k = 1, mo_num + three_e_5_idx_direct_bi_ort(m,k,i,l,j) = three_e_5_idx_direct_bi_ort(m,k,i,l,j) - tmp_mat(l,j,k) + enddo + enddo + enddo + !$OMP END DO + !$OMP END PARALLEL + enddo + + deallocate(grad_mli) + + allocate(lm_grad_ik(n_points_final_grid,3,mo_num,mo_num)) + allocate(lk_grad_mi(n_points_final_grid,3,mo_num,mo_num)) + + !$OMP PARALLEL DO PRIVATE (i,l,ipoint) + do i=1,mo_num + do l=1,mo_num + do ipoint=1, n_points_final_grid + + lm_grad_ik(ipoint,1,l,i) = mos_l_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,1,l,i) * final_weight_at_r_vector(ipoint) + lm_grad_ik(ipoint,2,l,i) = mos_l_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,2,l,i) * final_weight_at_r_vector(ipoint) + lm_grad_ik(ipoint,3,l,i) = mos_l_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,3,l,i) * final_weight_at_r_vector(ipoint) + + lk_grad_mi(ipoint,1,l,i) = mos_l_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,1,m,i) * final_weight_at_r_vector(ipoint) + lk_grad_mi(ipoint,2,l,i) = mos_l_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,2,m,i) * final_weight_at_r_vector(ipoint) + lk_grad_mi(ipoint,3,l,i) = mos_l_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,3,m,i) * final_weight_at_r_vector(ipoint) + enddo enddo enddo - enddo - !$OMP END PARALLEL DO + !$OMP END PARALLEL DO - call dgemm('T','N', mo_num*mo_num, mo_num*mo_num, 3*n_points_final_grid, 1.d0, & - lm_grad_ik, 3*n_points_final_grid, & - rk_grad_im, 3*n_points_final_grid, 0.d0, & - tmp_mat, mo_num*mo_num) - !$OMP PARALLEL DO PRIVATE(i,j,k,l) - do i = 1, mo_num - do k = 1, mo_num - do j = 1, mo_num - do l = 1, mo_num - three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) = - tmp_mat(l,i,j,k) - three_e_5_idx_cycle_2_bi_ort(m,l,j,k,i) = - tmp_mat(k,j,i,l) - three_e_5_idx_exch23_bi_ort (m,l,j,k,i) = - tmp_mat(k,i,j,l) - three_e_5_idx_exch13_bi_ort (m,l,j,k,i) = - tmp_mat(l,j,i,k) + allocate(rm_grad_ik(n_points_final_grid,3,mo_num)) + allocate(rk_grad_im(n_points_final_grid,3,mo_num)) + + do i=1,mo_num + !$OMP PARALLEL DO PRIVATE (l,ipoint) + do l=1,mo_num + do ipoint=1, n_points_final_grid + + rm_grad_ik(ipoint,1,l) = mos_r_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,1,l,i) + rm_grad_ik(ipoint,2,l) = mos_r_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,2,l,i) + rm_grad_ik(ipoint,3,l) = mos_r_in_r_array_transp(ipoint,m) * int2_grad1_u12_bimo_t(ipoint,3,l,i) + + rk_grad_im(ipoint,1,l) = mos_r_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,1,i,m) + rk_grad_im(ipoint,2,l) = mos_r_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,2,i,m) + rk_grad_im(ipoint,3,l) = mos_r_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,3,i,m) + enddo enddo - enddo - enddo - !$OMP END PARALLEL DO + !$OMP END PARALLEL DO - deallocate(lm_grad_ik) + call dgemm('T','N', mo_num*mo_num, mo_num, 3*n_points_final_grid, 1.d0,& + lm_grad_ik, 3*n_points_final_grid, & + rm_grad_ik, 3*n_points_final_grid, 0.d0, & + tmp_mat, mo_num*mo_num) - allocate(lk_grad_mi(n_points_final_grid,3,mo_num,mo_num)) - - !$OMP PARALLEL & - !$OMP DEFAULT (NONE) & - !$OMP PRIVATE (i,l,ipoint) & - !$OMP SHARED (m,mo_num,n_points_final_grid, & - !$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, & - !$OMP int2_grad1_u12_bimo_t, final_weight_at_r_vector, & - !$OMP lk_grad_mi) - !$OMP DO COLLAPSE(2) - do i=1,mo_num - do l=1,mo_num - do ipoint=1, n_points_final_grid - - lk_grad_mi(ipoint,1,l,i) = mos_l_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,1,m,i) * final_weight_at_r_vector(ipoint) - lk_grad_mi(ipoint,2,l,i) = mos_l_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,2,m,i) * final_weight_at_r_vector(ipoint) - lk_grad_mi(ipoint,3,l,i) = mos_l_in_r_array_transp(ipoint,l) * int2_grad1_u12_bimo_t(ipoint,3,m,i) * final_weight_at_r_vector(ipoint) - - enddo - enddo - enddo - !$OMP END DO - !$OMP END PARALLEL - - call dgemm('T','N', mo_num*mo_num, mo_num*mo_num, 3*n_points_final_grid, 1.d0, & - lk_grad_mi, 3*n_points_final_grid, & - rm_grad_ik, 3*n_points_final_grid, 0.d0, & - tmp_mat, mo_num*mo_num) - - !$OMP PARALLEL DO PRIVATE(i,j,k,l) - do i = 1, mo_num - do k = 1, mo_num - do j = 1, mo_num - do l = 1, mo_num - three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) = three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) - tmp_mat(k,j,l,i) - three_e_5_idx_cycle_2_bi_ort(m,l,j,k,i) = three_e_5_idx_cycle_2_bi_ort(m,l,j,k,i) - tmp_mat(l,i,k,j) - three_e_5_idx_exch23_bi_ort (m,l,j,k,i) = three_e_5_idx_exch23_bi_ort (m,l,j,k,i) - tmp_mat(l,j,k,i) - three_e_5_idx_exch13_bi_ort (m,l,j,k,i) = three_e_5_idx_exch13_bi_ort (m,l,j,k,i) - tmp_mat(k,i,l,j) + !$OMP PARALLEL DO PRIVATE(j,k,l) + do k = 1, mo_num + do j = 1, mo_num + do l = 1, mo_num + three_e_5_idx_direct_bi_ort(m,l,j,k,i) = three_e_5_idx_direct_bi_ort(m,l,j,k,i) - tmp_mat(l,j,k) + enddo enddo enddo - enddo - enddo - !$OMP END PARALLEL DO + !$OMP END PARALLEL DO - call dgemm('T','N', mo_num*mo_num, mo_num*mo_num, 3*n_points_final_grid, 1.d0, & - lk_grad_mi, 3*n_points_final_grid, & - rk_grad_im, 3*n_points_final_grid, 0.d0, & - tmp_mat, mo_num*mo_num) + call dgemm('T','N', mo_num*mo_num, mo_num, 3*n_points_final_grid, 1.d0,& + lm_grad_ik, 3*n_points_final_grid, & + rk_grad_im, 3*n_points_final_grid, 0.d0, & + tmp_mat, mo_num*mo_num) - !$OMP PARALLEL DO PRIVATE(i,j,k,l) - do i = 1, mo_num - do k = 1, mo_num - do j = 1, mo_num - do l = 1, mo_num - three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) = three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) - tmp_mat(l,j,i,k) - three_e_5_idx_cycle_2_bi_ort(m,l,j,k,i) = three_e_5_idx_cycle_2_bi_ort(m,l,j,k,i) - tmp_mat(k,i,j,l) - three_e_5_idx_exch23_bi_ort (m,l,j,k,i) = three_e_5_idx_exch23_bi_ort (m,l,j,k,i) - tmp_mat(k,j,i,l) - three_e_5_idx_exch13_bi_ort (m,l,j,k,i) = three_e_5_idx_exch13_bi_ort (m,l,j,k,i) - tmp_mat(l,i,j,k) + !$OMP PARALLEL DO PRIVATE(j,k,l) + do k = 1, mo_num + do j = 1, mo_num + do l = 1, mo_num + three_e_5_idx_cycle_1_bi_ort(m,l,j,i,k) = three_e_5_idx_cycle_1_bi_ort(m,l,j,i,k) - tmp_mat(l,k,j) + three_e_5_idx_cycle_2_bi_ort(m,i,j,k,l) = three_e_5_idx_cycle_2_bi_ort(m,i,j,k,l) - tmp_mat(k,j,l) + three_e_5_idx_exch23_bi_ort (m,i,j,k,l) = three_e_5_idx_exch23_bi_ort (m,i,j,k,l) - tmp_mat(k,l,j) + three_e_5_idx_exch13_bi_ort (m,l,j,i,k) = three_e_5_idx_exch13_bi_ort (m,l,j,i,k) - tmp_mat(l,j,k) + enddo enddo enddo - enddo - enddo - !$OMP END PARALLEL DO + !$OMP END PARALLEL DO + + + call dgemm('T','N', mo_num*mo_num, mo_num, 3*n_points_final_grid, 1.d0,& + lk_grad_mi, 3*n_points_final_grid, & + rm_grad_ik, 3*n_points_final_grid, 0.d0, & + tmp_mat, mo_num*mo_num) + + !$OMP PARALLEL DO PRIVATE(j,k,l) + do k = 1, mo_num + do j = 1, mo_num + do l = 1, mo_num + three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) = three_e_5_idx_cycle_1_bi_ort(m,l,j,k,i) - tmp_mat(k,j,l) + three_e_5_idx_cycle_2_bi_ort(m,l,i,k,j) = three_e_5_idx_cycle_2_bi_ort(m,l,i,k,j) - tmp_mat(l,j,k) + three_e_5_idx_exch23_bi_ort (m,l,j,k,i) = three_e_5_idx_exch23_bi_ort (m,l,j,k,i) - tmp_mat(l,j,k) + three_e_5_idx_exch13_bi_ort (m,l,i,k,j) = three_e_5_idx_exch13_bi_ort (m,l,i,k,j) - tmp_mat(k,j,l) + enddo + enddo + enddo + !$OMP END PARALLEL DO + + call dgemm('T','N', mo_num*mo_num, mo_num, 3*n_points_final_grid, 1.d0,& + lk_grad_mi, 3*n_points_final_grid, & + rk_grad_im, 3*n_points_final_grid, 0.d0, & + tmp_mat, mo_num*mo_num) + + !$OMP PARALLEL DO PRIVATE(j,k,l) + do k = 1, mo_num + do j = 1, mo_num + do l = 1, mo_num + three_e_5_idx_cycle_1_bi_ort(m,l,j,i,k) = three_e_5_idx_cycle_1_bi_ort(m,l,j,i,k) - tmp_mat(l,j,k) + three_e_5_idx_cycle_2_bi_ort(m,i,j,k,l) = three_e_5_idx_cycle_2_bi_ort(m,i,j,k,l) - tmp_mat(k,l,j) + three_e_5_idx_exch23_bi_ort (m,i,j,k,l) = three_e_5_idx_exch23_bi_ort (m,i,j,k,l) - tmp_mat(k,j,l) + three_e_5_idx_exch13_bi_ort (m,l,j,i,k) = three_e_5_idx_exch13_bi_ort (m,l,j,i,k) - tmp_mat(l,k,j) + enddo + enddo + enddo + !$OMP END PARALLEL DO + + enddo + deallocate(rm_grad_ik) + deallocate(rk_grad_im) + deallocate(lk_grad_mi) + deallocate(lm_grad_ik) - deallocate(lk_grad_mi) - deallocate(rm_grad_ik) - deallocate(rk_grad_im) enddo + deallocate(orb_mat) + call wall_time(wall1) print *, ' wall time for three_e_5_idx_bi_ort', wall1 - wall0