diff --git a/src/tc_bi_ortho/normal_ordered_contractions.irp.f b/src/tc_bi_ortho/normal_ordered_contractions.irp.f index 855cfd17..6f70516d 100644 --- a/src/tc_bi_ortho/normal_ordered_contractions.irp.f +++ b/src/tc_bi_ortho/normal_ordered_contractions.irp.f @@ -1,7 +1,7 @@ ! --- -BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_num)] +BEGIN_PROVIDER [ double precision, no_aba_contraction_v0, (mo_num,mo_num,mo_num,mo_num)] use bitmasks ! you need to include the bitmasks_module.f90 features @@ -16,7 +16,7 @@ BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_ double precision, allocatable :: tmpval_1(:), tmpval_2(:), tmpvec_1(:,:), tmpvec_2(:,:) double precision, allocatable :: tmp_2d(:,:) - print*,' Providing no_aba_contraction ...' + print*,' Providing no_aba_contraction_v0 ...' call wall_time(wall0) PROVIDE N_int @@ -102,7 +102,7 @@ BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_ do p1 = 1, mo_num do h2 = 1, mo_num do p2 = 1, mo_num - no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_3d(p2,h2,p1) + no_aba_contraction_v0(p2,h2,p1,h1) = no_aba_contraction_v0(p2,h2,p1,h1) + tmp_3d(p2,h2,p1) enddo enddo enddo @@ -153,7 +153,7 @@ BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_ !$OMP PARALLEL DO PRIVATE(h2,p2) do h2 = 1, mo_num do p2 = 1, mo_num - no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_2d(p2,h2) + no_aba_contraction_v0(p2,h2,p1,h1) = no_aba_contraction_v0(p2,h2,p1,h1) + tmp_2d(p2,h2) enddo enddo !$OMP END PARALLEL DO @@ -220,7 +220,7 @@ BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_ do p1 = 1, mo_num do h2 = 1, mo_num do p2 = 1, mo_num - no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_3d(p2,h2,p1) + no_aba_contraction_v0(p2,h2,p1,h1) = no_aba_contraction_v0(p2,h2,p1,h1) + tmp_3d(p2,h2,p1) enddo enddo enddo @@ -270,7 +270,7 @@ BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_ !$OMP PARALLEL DO PRIVATE(h2,p2) do h2 = 1, mo_num do p2 = 1, mo_num - no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_2d(p2,h2) + no_aba_contraction_v0(p2,h2,p1,h1) = no_aba_contraction_v0(p2,h2,p1,h1) + tmp_2d(p2,h2) enddo enddo !$OMP END PARALLEL DO @@ -285,11 +285,11 @@ BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_ deallocate(tmpval_1, tmpval_2) deallocate(tmpvec_1, tmpvec_2) - no_aba_contraction = -0.5d0 * no_aba_contraction - call sum_A_At(no_aba_contraction(1,1,1,1), mo_num*mo_num) + no_aba_contraction_v0 = -0.5d0 * no_aba_contraction_v0 + call sum_A_At(no_aba_contraction_v0(1,1,1,1), mo_num*mo_num) call wall_time(wall1) - print*,' Wall time for no_aba_contraction', wall1-wall0 + print*,' Wall time for no_aba_contraction_v0', wall1-wall0 END_PROVIDER @@ -1060,3 +1060,274 @@ BEGIN_PROVIDER [ double precision, no_aaa_contraction, (mo_num,mo_num,mo_num,mo_ END_PROVIDER ! --- + +BEGIN_PROVIDER [ double precision, no_aba_contraction, (mo_num,mo_num,mo_num,mo_num)] + + use bitmasks ! you need to include the bitmasks_module.f90 features + + implicit none + integer :: i, ii, h1, p1, h2, p2, ipoint + integer :: Ne(2) + double precision :: wall0, wall1 + integer, allocatable :: occ(:,:) + integer(bit_kind), allocatable :: key_i_core(:,:) + double precision, allocatable :: tmp_3d(:,:,:) + double precision, allocatable :: tmp1(:,:,:), tmp2(:,:) + double precision, allocatable :: tmpval_1(:), tmpval_2(:), tmpvec_1(:,:), tmpvec_2(:,:) + double precision, allocatable :: tmp_2d(:,:) + + print*,' Providing no_aba_contraction ...' + call wall_time(wall0) + + PROVIDE N_int + + allocate(occ(N_int*bit_kind_size,2)) + allocate(key_i_core(N_int,2)) + + if(core_tc_op) then + do i = 1, N_int + key_i_core(i,1) = xor(ref_bitmask(i,1), core_bitmask(i,1)) + key_i_core(i,2) = xor(ref_bitmask(i,2), core_bitmask(i,2)) + enddo + call bitstring_to_list_ab(key_i_core, occ, Ne, N_int) + else + call bitstring_to_list_ab(ref_bitmask, occ, Ne, N_int) + endif + + !$OMP PARALLEL & + !$OMP DEFAULT (NONE) & + !$OMP PRIVATE (ipoint, h1, p1, h2, p2, i, ii, & + !$OMP tmp_3d, tmp_2d, tmp1, tmp2, & + !$OMP tmpval_1, tmpval_2, tmpvec_1, tmpvec_2) & + !$OMP SHARED (n_points_final_grid, Ne, occ, mo_num, & + !$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, & + !$OMP int2_grad1_u12_bimo_t, final_weight_at_r_vector, & + !$OMP no_aba_contraction) + + + allocate(tmp_3d(mo_num,mo_num,mo_num), tmp_2d(mo_num,mo_num)) + allocate(tmp1(n_points_final_grid,3,mo_num), tmp2(n_points_final_grid,mo_num)) + allocate(tmpval_1(n_points_final_grid), tmpval_2(n_points_final_grid)) + allocate(tmpvec_1(n_points_final_grid,3), tmpvec_2(n_points_final_grid,3)) + + !$OMP DO + + do ii = 1, Ne(2) + i = occ(ii,2) + + do h1 = 1, mo_num + + do ipoint = 1, n_points_final_grid + tmpval_1(ipoint) = final_weight_at_r_vector(ipoint) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint, i) + tmpval_2(ipoint) = final_weight_at_r_vector(ipoint) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_1(ipoint,1) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,i, i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_1(ipoint,2) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,i, i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_1(ipoint,3) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,i, i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_2(ipoint,1) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,i,h1) * mos_r_in_r_array_transp(ipoint, i) + tmpvec_2(ipoint,2) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,i,h1) * mos_r_in_r_array_transp(ipoint, i) + tmpvec_2(ipoint,3) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,i,h1) * mos_r_in_r_array_transp(ipoint, i) + enddo + + do p1 = 1, mo_num + do ipoint = 1, n_points_final_grid + tmp1(ipoint,1,p1) = mos_l_in_r_array_transp(ipoint,p1) * (tmpvec_1(ipoint,1) - tmpvec_2(ipoint,1)) & + + tmpval_1(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,p1,h1) - tmpval_2(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,p1,i) + tmp1(ipoint,2,p1) = mos_l_in_r_array_transp(ipoint,p1) * (tmpvec_1(ipoint,2) - tmpvec_2(ipoint,2)) & + + tmpval_1(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,p1,h1) - tmpval_2(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,p1,i) + tmp1(ipoint,3,p1) = mos_l_in_r_array_transp(ipoint,p1) * (tmpvec_1(ipoint,3) - tmpvec_2(ipoint,3)) & + + tmpval_1(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,p1,h1) - tmpval_2(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,p1,i) + enddo + enddo + + call dgemm( 'T', 'N', mo_num*mo_num, mo_num, 3*n_points_final_grid, 1.d0 & + , int2_grad1_u12_bimo_t(1,1,1,1), 3*n_points_final_grid & + , tmp1(1,1,1), 3*n_points_final_grid & + , 0.d0, tmp_3d(1,1,1), mo_num*mo_num) + + do p1 = 1, mo_num + do h2 = 1, mo_num + do p2 = 1, mo_num + no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_3d(p2,h2,p1) + enddo + enddo + enddo + + do p1 = 1, mo_num + + do ipoint = 1, n_points_final_grid + tmpval_1(ipoint) = final_weight_at_r_vector(ipoint) * & + ( int2_grad1_u12_bimo_t(ipoint,1, i,i) * int2_grad1_u12_bimo_t(ipoint,1,p1,h1) & + + int2_grad1_u12_bimo_t(ipoint,2, i,i) * int2_grad1_u12_bimo_t(ipoint,2,p1,h1) & + + int2_grad1_u12_bimo_t(ipoint,3, i,i) * int2_grad1_u12_bimo_t(ipoint,3,p1,h1) & + - int2_grad1_u12_bimo_t(ipoint,1,p1,i) * int2_grad1_u12_bimo_t(ipoint,1, i,h1) & + - int2_grad1_u12_bimo_t(ipoint,2,p1,i) * int2_grad1_u12_bimo_t(ipoint,2, i,h1) & + - int2_grad1_u12_bimo_t(ipoint,3,p1,i) * int2_grad1_u12_bimo_t(ipoint,3, i,h1) ) + enddo + + do h2 = 1, mo_num + do ipoint = 1, n_points_final_grid + tmp2(ipoint,h2) = mos_r_in_r_array_transp(ipoint,h2) * tmpval_1(ipoint) + enddo + enddo + + call dgemm( 'T', 'N', mo_num, mo_num, n_points_final_grid, 1.d0 & + , mos_l_in_r_array_transp(1,1), n_points_final_grid & + , tmp2(1,1), n_points_final_grid & + , 0.d0, tmp_2d(1,1), mo_num) + + do h2 = 1, mo_num + do p2 = 1, mo_num + no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_2d(p2,h2) + enddo + enddo + + enddo ! p1 + enddo ! h1 + enddo ! i + + !$OMP END DO + + deallocate(tmp_3d, tmp_2d) + deallocate(tmp1, tmp2) + deallocate(tmpval_1, tmpval_2) + deallocate(tmpvec_1, tmpvec_2) + + !$OMP END PARALLEL + + + allocate(tmp_3d(mo_num,mo_num,mo_num), tmp_2d(mo_num,mo_num)) + allocate(tmp1(n_points_final_grid,3,mo_num), tmp2(n_points_final_grid,mo_num)) + allocate(tmpval_1(n_points_final_grid), tmpval_2(n_points_final_grid)) + allocate(tmpvec_1(n_points_final_grid,3), tmpvec_2(n_points_final_grid,3)) + + + ! purely open-shell part + if(Ne(2) < Ne(1)) then + + do ii = Ne(2) + 1, Ne(1) + i = occ(ii,1) + + do h1 = 1, mo_num + + !$OMP PARALLEL & + !$OMP DEFAULT (NONE) & + !$OMP PRIVATE (ipoint) & + !$OMP SHARED (n_points_final_grid, i, h1, & + !$OMP mos_l_in_r_array_transp, mos_r_in_r_array_transp, & + !$OMP int2_grad1_u12_bimo_t, final_weight_at_r_vector, & + !$OMP tmpval_1, tmpval_2, tmpvec_1, tmpvec_2) + !$OMP DO + do ipoint = 1, n_points_final_grid + tmpval_1(ipoint) = final_weight_at_r_vector(ipoint) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint, i) + tmpval_2(ipoint) = final_weight_at_r_vector(ipoint) * mos_l_in_r_array_transp(ipoint,i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_1(ipoint,1) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,i, i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_1(ipoint,2) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,i, i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_1(ipoint,3) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,i, i) * mos_r_in_r_array_transp(ipoint,h1) + tmpvec_2(ipoint,1) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,i,h1) * mos_r_in_r_array_transp(ipoint, i) + tmpvec_2(ipoint,2) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,i,h1) * mos_r_in_r_array_transp(ipoint, i) + tmpvec_2(ipoint,3) = final_weight_at_r_vector(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,i,h1) * mos_r_in_r_array_transp(ipoint, i) + enddo + !$OMP END DO + !$OMP END PARALLEL + + !$OMP PARALLEL & + !$OMP DEFAULT (NONE) & + !$OMP PRIVATE (p1, ipoint) & + !$OMP SHARED (mo_num, n_points_final_grid, h1, i, & + !$OMP mos_l_in_r_array_transp, int2_grad1_u12_bimo_t, & + !$OMP tmpval_1, tmpval_2, tmpvec_1, tmpvec_2, tmp1) + !$OMP DO + do p1 = 1, mo_num + do ipoint = 1, n_points_final_grid + tmp1(ipoint,1,p1) = mos_l_in_r_array_transp(ipoint,p1) * (tmpvec_1(ipoint,1) - tmpvec_2(ipoint,1)) & + + tmpval_1(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,p1,h1) - tmpval_2(ipoint) * int2_grad1_u12_bimo_t(ipoint,1,p1,i) + tmp1(ipoint,2,p1) = mos_l_in_r_array_transp(ipoint,p1) * (tmpvec_1(ipoint,2) - tmpvec_2(ipoint,2)) & + + tmpval_1(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,p1,h1) - tmpval_2(ipoint) * int2_grad1_u12_bimo_t(ipoint,2,p1,i) + tmp1(ipoint,3,p1) = mos_l_in_r_array_transp(ipoint,p1) * (tmpvec_1(ipoint,3) - tmpvec_2(ipoint,3)) & + + tmpval_1(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,p1,h1) - tmpval_2(ipoint) * int2_grad1_u12_bimo_t(ipoint,3,p1,i) + enddo + enddo + !$OMP END DO + !$OMP END PARALLEL + + call dgemm( 'T', 'N', mo_num*mo_num, mo_num, 3*n_points_final_grid, 0.5d0 & + , int2_grad1_u12_bimo_t(1,1,1,1), 3*n_points_final_grid & + , tmp1(1,1,1), 3*n_points_final_grid & + , 0.d0, tmp_3d(1,1,1), mo_num*mo_num) + + !$OMP PARALLEL DO PRIVATE(p1,h2,p2) + do p1 = 1, mo_num + do h2 = 1, mo_num + do p2 = 1, mo_num + no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_3d(p2,h2,p1) + enddo + enddo + enddo + !$OMP END PARALLEL DO + + do p1 = 1, mo_num + + ! to minimize the number of operations + !$OMP PARALLEL & + !$OMP DEFAULT (NONE) & + !$OMP PRIVATE (ipoint) & + !$OMP SHARED (n_points_final_grid, i, h1, p1, & + !$OMP int2_grad1_u12_bimo_t, final_weight_at_r_vector, & + !$OMP tmpval_1) + !$OMP DO + do ipoint = 1, n_points_final_grid + tmpval_1(ipoint) = final_weight_at_r_vector(ipoint) * ( int2_grad1_u12_bimo_t(ipoint,1, i,i) * int2_grad1_u12_bimo_t(ipoint,1,p1,h1) & + + int2_grad1_u12_bimo_t(ipoint,2, i,i) * int2_grad1_u12_bimo_t(ipoint,2,p1,h1) & + + int2_grad1_u12_bimo_t(ipoint,3, i,i) * int2_grad1_u12_bimo_t(ipoint,3,p1,h1) & + - int2_grad1_u12_bimo_t(ipoint,1,p1,i) * int2_grad1_u12_bimo_t(ipoint,1, i,h1) & + - int2_grad1_u12_bimo_t(ipoint,2,p1,i) * int2_grad1_u12_bimo_t(ipoint,2, i,h1) & + - int2_grad1_u12_bimo_t(ipoint,3,p1,i) * int2_grad1_u12_bimo_t(ipoint,3, i,h1) ) + enddo + !$OMP END DO + !$OMP END PARALLEL + + !$OMP PARALLEL & + !$OMP DEFAULT (NONE) & + !$OMP PRIVATE (h2, ipoint) & + !$OMP SHARED (mo_num, n_points_final_grid, & + !$OMP mos_r_in_r_array_transp, & + !$OMP tmpval_1, tmp2) + !$OMP DO + do h2 = 1, mo_num + do ipoint = 1, n_points_final_grid + tmp2(ipoint,h2) = mos_r_in_r_array_transp(ipoint,h2) * tmpval_1(ipoint) + enddo + enddo + !$OMP END DO + !$OMP END PARALLEL + + call dgemm( 'T', 'N', mo_num, mo_num, n_points_final_grid, 0.5d0 & + , mos_l_in_r_array_transp(1,1), n_points_final_grid & + , tmp2(1,1), n_points_final_grid & + , 0.d0, tmp_2d(1,1), mo_num) + + !$OMP PARALLEL DO PRIVATE(h2,p2) + do h2 = 1, mo_num + do p2 = 1, mo_num + no_aba_contraction(p2,h2,p1,h1) = no_aba_contraction(p2,h2,p1,h1) + tmp_2d(p2,h2) + enddo + enddo + !$OMP END PARALLEL DO + + enddo ! p1 + enddo ! h1 + enddo !i + endif + + deallocate(tmp_2d, tmp_3d) + deallocate(tmp1, tmp2) + deallocate(tmpval_1, tmpval_2) + deallocate(tmpvec_1, tmpvec_2) + + no_aba_contraction = -0.5d0 * no_aba_contraction + call sum_A_At(no_aba_contraction(1,1,1,1), mo_num*mo_num) + + call wall_time(wall1) + print*,' Wall time for no_aba_contraction', wall1-wall0 + +END_PROVIDER