10
0
mirror of https://github.com/LCPQ/quantum_package synced 2024-11-13 17:43:55 +01:00
quantum_package/scripts/generate_h_apply.py

269 lines
9.2 KiB
Python
Executable File

#!/usr/bin/env python
import os
file = open(os.environ["QPACKAGE_ROOT"]+'/src/Dets/H_apply_template.f','r')
template = file.read()
file.close()
keywords = """
subroutine
parameters
params_main
initialization
declarations
decls_main
keys_work
copy_buffer
finalization
generate_psi_guess
init_thread
printout_now
printout_always
deinit_thread
skip
init_main
filter_integrals
filter2h2p
""".split()
class H_apply(object):
def __init__(self,sub,SingleRef=False,do_mono_exc=True, do_double_exc=True):
s = {}
for k in keywords:
s[k] = ""
s["subroutine"] = "H_apply_%s"%(sub)
s["params_post"] = ""
self.selection_pt2 = None
self.perturbation = None
#s["omp_parallel"] = """!$OMP PARALLEL DEFAULT(NONE) &
s["omp_parallel"] = """!$OMP PARALLEL DEFAULT(SHARED) &
!$OMP PRIVATE(i,j,k,l,keys_out,hole,particle, &
!$OMP occ_particle,occ_hole,j_a,k_a,other_spin, &
!$OMP hole_save,ispin,jj,l_a,ib_jb_pairs,array_pairs, &
!$OMP accu,i_a,hole_tmp,particle_tmp,occ_particle_tmp, &
!$OMP occ_hole_tmp,key_idx,i_b,j_b,key,N_elec_in_key_part_1,&
!$OMP N_elec_in_key_hole_1,N_elec_in_key_part_2, &
!$OMP N_elec_in_key_hole_2,ia_ja_pairs) &
!$OMP SHARED(key_in,N_int,elec_num_tab,mo_tot_num, &
!$OMP hole_1, particl_1, hole_2, particl_2, &
!$OMP elec_alpha_num,i_generator)"""
s["omp_end_parallel"] = "!$OMP END PARALLEL"
s["omp_master"] = "!$OMP MASTER"
s["omp_end_master"] = "!$OMP END MASTER"
s["omp_barrier"] = "!$OMP BARRIER"
s["omp_do"] = "!$OMP DO SCHEDULE (static)"
s["omp_enddo"] = "!$OMP ENDDO NOWAIT"
d = { True : '.True.', False : '.False.'}
s["do_mono_excitations"] = d[do_mono_exc]
s["do_double_excitations"] = d[do_double_exc]
s["keys_work"] += "call fill_H_apply_buffer_no_selection(key_idx,keys_out,N_int,iproc)"
s["filter_integrals"] = "array_pairs = .True."
if SingleRef:
s["filter_integrals"] = """
call get_mo_bielec_integrals_existing_ik(i_a,j_a,mo_tot_num,array_pairs,mo_integrals_map)
"""
s["generate_psi_guess"] = """
! Sort H_jj to find the N_states lowest states
integer :: i
integer, allocatable :: iorder(:)
double precision, allocatable :: H_jj(:)
double precision, external :: diag_h_mat_elem
allocate(H_jj(N_det),iorder(N_det))
!$OMP PARALLEL DEFAULT(NONE) &
!$OMP SHARED(psi_det,N_int,H_jj,iorder,N_det) &
!$OMP PRIVATE(i)
!$OMP DO
do i = 1, N_det
H_jj(i) = diag_h_mat_elem(psi_det(1,1,i),N_int)
iorder(i) = i
enddo
!$OMP END DO
!$OMP END PARALLEL
call dsort(H_jj,iorder,N_det)
do k=1,N_states
psi_coef(iorder(k),k) = 1.d0
enddo
deallocate(H_jj,iorder)
"""
s["size_max"] = str(1024*128)
s["copy_buffer"] = """call copy_H_apply_buffer_to_wf
if (s2_eig) then
call make_s2_eigenfunction
endif
SOFT_TOUCH psi_det psi_coef N_det
"""
s["printout_now"] = """write(output_Dets,*) &
100.*float(i_generator)/float(N_det_generators), '% in ', wall_1-wall_0, 's'"""
self.data = s
def __setitem__(self,key,value):
self.data[key] = value
def __getitem__(self,key):
return self.data[key]
def __repr__(self):
buffer = template
for key,value in self.data.items():
buffer = buffer.replace('$'+key, value)
return buffer
def set_filter_2h_2p(self):
self["filter2h2p"] = """
! ! DIR$ FORCEINLINE
if(is_a_two_holes_two_particles(key))cycle
"""
def set_perturbation(self,pert):
if self.perturbation is not None:
raise
self.perturbation = pert
if pert is not None:
self.data["parameters"] = ",sum_e_2_pert_in,sum_norm_pert_in,sum_H_pert_diag_in,N_st,Nint"
self.data["declarations"] = """
integer, intent(in) :: N_st,Nint
double precision, intent(inout) :: sum_e_2_pert_in(N_st)
double precision, intent(inout) :: sum_norm_pert_in(N_st)
double precision, intent(inout) :: sum_H_pert_diag_in(N_st)
double precision :: sum_e_2_pert(N_st)
double precision :: sum_norm_pert(N_st)
double precision :: sum_H_pert_diag(N_st)
double precision, allocatable :: e_2_pert_buffer(:,:)
double precision, allocatable :: coef_pert_buffer(:,:)
ASSERT (Nint == N_int)
"""
self.data["init_thread"] = """
allocate (e_2_pert_buffer(N_st,size_max), coef_pert_buffer(N_st,size_max))
do k=1,N_st
sum_e_2_pert(k) = 0.d0
sum_norm_pert(k) = 0.d0
sum_H_pert_diag(k) = 0.d0
enddo
"""
self.data["deinit_thread"] = """
!$ call omp_set_lock(lck)
do k=1,N_st
sum_e_2_pert_in(k) = sum_e_2_pert_in(k) + sum_e_2_pert(k)
sum_norm_pert_in(k) = sum_norm_pert_in(k) + sum_norm_pert(k)
sum_H_pert_diag_in(k) = sum_H_pert_diag_in(k) + sum_H_pert_diag(k)
enddo
!$ call omp_unset_lock(lck)
deallocate (e_2_pert_buffer, coef_pert_buffer)
"""
self.data["size_max"] = "256"
self.data["initialization"] = """
PROVIDE CI_electronic_energy psi_selectors_coef psi_selectors E_corr_per_selectors psi_det_sorted_bit
"""
self.data["keys_work"] = """
call perturb_buffer_%s(i_generator,keys_out,key_idx,e_2_pert_buffer,coef_pert_buffer,sum_e_2_pert, &
sum_norm_pert,sum_H_pert_diag,N_st,N_int)
"""%(pert,)
self.data["finalization"] = """
"""
self.data["copy_buffer"] = ""
self.data["generate_psi_guess"] = ""
self.data["params_main"] = "pt2, norm_pert, H_pert_diag, N_st"
self.data["params_post"] = ","+self.data["params_main"] +", N_int"
self.data["decls_main"] = """ integer, intent(in) :: N_st
double precision, intent(inout):: pt2(N_st)
double precision, intent(inout):: norm_pert(N_st)
double precision, intent(inout):: H_pert_diag(N_st)
double precision :: delta_pt2(N_st), norm_psi(N_st), pt2_old(N_st)
PROVIDE CI_electronic_energy N_det_generators
do k=1,N_st
pt2(k) = 0.d0
norm_pert(k) = 0.d0
H_pert_diag(k) = 0.d0
norm_psi(k) = 0.d0
delta_pt2(k) = 0.d0
pt2_old(k) = 0.d0
enddo
write(output_Dets,'(A12, X, A8, 3(2X, A9), 2X, A8, 2X, A8, 2X, A8)') &
'N_generators', 'Norm', 'Delta PT2', 'PT2', 'Est. PT2', 'secs'
write(output_Dets,'(A12, X, A8, 3(2X, A9), 2X, A8, 2X, A8, 2X, A8)') &
'============', '========', '=========', '=========', '=========', &
'========='
"""
self.data["printout_always"] = """
do k=1,N_st
norm_psi(k) = norm_psi(k) + psi_coef(i_generator,k)*psi_coef(i_generator,k)
delta_pt2(k) = pt2(k) - pt2_old(k)
enddo
"""
self.data["printout_now"] = """
do k=1,N_st
write(output_Dets,'(I10, 4(2X, F9.6), 2X, F8.1)') &
i_generator, norm_psi(k), delta_pt2(k), pt2(k), &
pt2(k)/(norm_psi(k)*norm_psi(k)), &
wall_1-wall_0
pt2_old(k) = pt2(k)
enddo
progress_value = norm_psi(1)
"""
self.data["omp_parallel"] += """&
!$OMP SHARED(N_st) PRIVATE(e_2_pert_buffer,coef_pert_buffer) &
!$OMP PRIVATE(sum_e_2_pert, sum_norm_pert, sum_H_pert_diag)"""
def set_selection_pt2(self,pert):
if self.selection_pt2 is not None:
raise
self.set_perturbation(pert)
self.selection_pt2 = pert
if pert is not None:
self.data["parameters"] += ",select_max_out"
self.data["declarations"] += """
double precision, intent(inout) :: select_max_out"""
self.data["params_post"] += ", select_max(min(i_generator,size(select_max,1)))"
self.data["size_max"] = str(1024*128)
self.data["copy_buffer"] = """
call copy_H_apply_buffer_to_wf
if (s2_eig) then
call make_s2_eigenfunction
endif
! SOFT_TOUCH psi_det psi_coef N_det
selection_criterion_min = min(selection_criterion_min, maxval(select_max))*0.1d0
selection_criterion = selection_criterion_min
call write_double(output_Dets,selection_criterion,'Selection criterion')
"""
self.data["keys_work"] = """
e_2_pert_buffer = 0.d0
coef_pert_buffer = 0.d0
""" + self.data["keys_work"]
self.data["keys_work"] += """
call fill_H_apply_buffer_selection(key_idx,keys_out,e_2_pert_buffer, &
coef_pert_buffer,N_st,N_int,iproc,select_max_out)
"""
self.data["omp_parallel"] += """&
!$OMP REDUCTION (max:select_max_out)"""
self.data["skip"] = """
if (i_generator < size_select_max) then
if (select_max(i_generator) < selection_criterion_min*selection_criterion_factor) then
!$ call omp_set_lock(lck)
do k=1,N_st
norm_psi(k) = norm_psi(k) + psi_coef(i_generator,k)*psi_coef(i_generator,k)
! delta_pt2(k) = 0.d0
! pt2_old(k) = 0.d0
! pt2(k) = select_max(i_generator)
enddo
!$ call omp_unset_lock(lck)
cycle
endif
select_max(i_generator) = 0.d0
endif
"""