LCPQ/qp2
9
1
Fork 0
mirror of https://github.com/QuantumPackage/qp2.git synced 2025-04-25 17:54:44 +02:00
Code Releases Activity

Introduced GPU arrays

Browse Source
This commit is contained in:
Anthony Scemama 2025-02-05 13:59:37 +01:00
parent 3d46cde2e4
commit 243ee0ed14
2 changed files with 55 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/ccsd/ccsd_space_orb_sub.irp.f
View File

@ -177,9 +177,11 @@ subroutine run_ccsd_space_orb
call guess_t2(nO,nV,cc_space_f_o,cc_space_f_v,cc_space_v_oovv,h_t2)
call gpu_upload(h_t2, t2)
deallocate(h_t1, h_t2)
call update_tau_space(nO,nV,h_t1,t1,t2,tau)
call update_tau_space(nO,nV,t1%f,t1,t2,tau)
call update_tau_x_space(nO,nV,tau,tau_x)
call det_energy(psi_det(1,1,cc_ref),uncorr_energy)
print*,'Det energy', uncorr_energy
@ -310,7 +312,6 @@ subroutine run_ccsd_space_orb
call save_energy(uncorr_energy + energy, e_t)
deallocate(h_t1, h_t2)
if (do_mo_cholesky) then
call gpu_deallocate(d_cc_space_v_oo_chol)
call gpu_deallocate(d_cc_space_v_ov_chol)

89
src/ccsd/ccsd_spin_orb_sub.irp.f
View File

@ -1,6 +1,5 @@
! Code
subroutine run_ccsd_spin_orb
use gpu
implicit none
@ -8,8 +7,6 @@ subroutine run_ccsd_spin_orb
! CCSD in spin orbitals
END_DOC
double precision, allocatable :: t1(:,:), t2(:,:,:,:), tau(:,:,:,:), tau_t(:,:,:,:)
double precision, allocatable :: r1(:,:), r2(:,:,:,:)
double precision, allocatable :: cF_oo(:,:), cF_ov(:,:), cF_vv(:,:)
double precision, allocatable :: cW_oooo(:,:,:,:), cW_ovvo(:,:,:,:) !, cW_vvvv(:,:,:,:)
@ -23,6 +20,9 @@ subroutine run_ccsd_spin_orb
integer :: nb_iter, i,j,a,b
double precision :: uncorr_energy, energy, max_r, max_r1, max_r2, cc, ta, tb,ti,tf,tbi,tfi
type(gpu_double4) :: t2, r2, tau, tau_t
type(gpu_double2) :: t1, r1
if (do_mo_cholesky) then
PROVIDE cholesky_mo_transp
FREE cholesky_ao
@ -34,13 +34,18 @@ subroutine run_ccsd_spin_orb
call print_det(psi_det(1,1,cc_ref),N_int)
! Allocation
allocate(t1(cc_nOab,cc_nVab), t2(cc_nOab,cc_nOab,cc_nVab,cc_nVab), tau(cc_nOab,cc_nOab,cc_nVab,cc_nVab), tau_t(cc_nOab,cc_nOab,cc_nVab,cc_nVab))
allocate(r1(cc_nOab,cc_nVab), r2(cc_nOab,cc_nOab,cc_nVab,cc_nVab))
allocate(cF_oo(cc_nOab,cc_nOab), cF_ov(cc_nOab,cc_nVab), cF_vv(cc_nVab,cc_nVab))
allocate(cW_oooo(cc_nOab,cc_nOab,cc_nOab,cc_nOab), cW_ovvo(cc_nOab,cc_nVab,cc_nVab,cc_nOab))!, cW_vvvv(cc_nVab,cc_nVab,cc_nVab,cc_nVab))
allocate(f_o(cc_nOab), f_v(cc_nVab))
call gpu_allocate(t1, cc_nOab,cc_nVab)
call gpu_allocate(r1, cc_nOab,cc_nVab)
call gpu_allocate(t2, cc_nOab,cc_nOab,cc_nVab,cc_nVab)
call gpu_allocate(r2, cc_nOab,cc_nOab,cc_nVab,cc_nVab)
call gpu_allocate(tau, cc_nOab,cc_nOab,cc_nVab,cc_nVab)
call gpu_allocate(tau_t, cc_nOab,cc_nOab,cc_nVab,cc_nVab)
! Allocation for the diis
if (cc_update_method == 'diis') then
allocate(all_err(cc_nOab*cc_nVab+cc_nOab*cc_nOab*cc_nVab*cc_nVab,cc_diis_depth), all_t(cc_nOab*cc_nVab+cc_nOab*cc_nOab*cc_nVab*cc_nVab,cc_diis_depth))
@ -58,23 +63,29 @@ subroutine run_ccsd_spin_orb
! Init of T
t1 = 0d0
call guess_t1(cc_nOab,cc_nVab,f_o,f_v,cc_spin_f_ov,t1)
call guess_t2(cc_nOab,cc_nVab,f_o,f_v,cc_spin_v_oovv,t2)
call compute_tau_spin(cc_nOab,cc_nVab,t1,t2,tau)
call compute_tau_t_spin(cc_nOab,cc_nVab,t1,t2,tau_t)
double precision, allocatable :: h_t1(:,:), h_t2(:,:,:,:)
allocate(h_t1(cc_nOab,cc_nVab), h_t2(cc_nOab,cc_nOab,cc_nVab,cc_nVab))
h_t1 = 0d0
call guess_t1(cc_nOab,cc_nVab,f_o,f_v,cc_spin_f_ov,h_t1)
call gpu_upload(h_t1, t1)
call guess_t2(cc_nOab,cc_nVab,f_o,f_v,cc_spin_v_oovv,h_t2)
call gpu_upload(h_t2, t2)
deallocate(h_t1,h_t2)
call compute_tau_spin(cc_nOab,cc_nVab,t1%f,t2%f,tau%f)
call compute_tau_t_spin(cc_nOab,cc_nVab,t1%f,t2%f,tau_t%f)
call det_energy(psi_det(1,1,cc_ref),uncorr_energy)
print*,'Det energy', uncorr_energy
call ccsd_energy_spin(cc_nOab,cc_nVab,t1,t2,cc_spin_F_ov,cc_spin_v_oovv,energy)
call ccsd_energy_spin(cc_nOab,cc_nVab,t1%f,t2%f,cc_spin_F_ov,cc_spin_v_oovv,energy)
print*,'guess energy', uncorr_energy+energy, energy
! Loop init
nb_iter = 0
not_converged = .True.
r1 = 0d0
r2 = 0d0
max_r1 = 0d0
max_r2 = 0d0
@ -88,46 +99,46 @@ subroutine run_ccsd_spin_orb
do while (not_converged)
! Intermediates
call compute_cF_oo(cc_nOab,cc_nVab,t1,tau_t,cc_spin_F_oo,cc_spin_F_ov,cc_spin_v_ooov,cc_spin_v_oovv,cF_oo)
call compute_cF_ov(cc_nOab,cc_nVab,t1,cc_spin_F_ov,cc_spin_v_oovv,cF_ov)
call compute_cF_vv(cc_nOab,cc_nVab,t1,tau_t,cc_spin_F_ov,cc_spin_F_vv,cc_spin_v_oovv,cF_vv)
call compute_cF_oo(cc_nOab,cc_nVab,t1%f,tau_t%f,cc_spin_F_oo,cc_spin_F_ov,cc_spin_v_ooov,cc_spin_v_oovv,cF_oo)
call compute_cF_ov(cc_nOab,cc_nVab,t1%f,cc_spin_F_ov,cc_spin_v_oovv,cF_ov)
call compute_cF_vv(cc_nOab,cc_nVab,t1%f,tau_t%f,cc_spin_F_ov,cc_spin_F_vv,cc_spin_v_oovv,cF_vv)
call compute_cW_oooo(cc_nOab,cc_nVab,t1,t2,tau,cc_spin_v_oooo,cc_spin_v_ooov,cc_spin_v_oovv,cW_oooo)
call compute_cW_ovvo(cc_nOab,cc_nVab,t1,t2,tau,cc_spin_v_ovvo,cc_spin_v_oovo,cc_spin_v_oovv,cW_ovvo)
call compute_cW_oooo(cc_nOab,cc_nVab,t1%f,t2%f,tau%f,cc_spin_v_oooo,cc_spin_v_ooov,cc_spin_v_oovv,cW_oooo)
call compute_cW_ovvo(cc_nOab,cc_nVab,t1%f,t2%f,tau%f,cc_spin_v_ovvo,cc_spin_v_oovo,cc_spin_v_oovv,cW_ovvo)
! Residuals
call compute_r1_spin(cc_nOab,cc_nVab,t1,t2,f_o,f_v,cc_spin_F_ov,cF_oo,cF_ov,cF_vv,cc_spin_v_oovo,cc_spin_v_ovov,r1)
call compute_r2_spin(cc_nOab,cc_nVab,t1,t2,tau,f_o,f_v,cF_oo,cF_ov,cF_vv,cW_oooo,cW_ovvo,cc_spin_v_ovoo,cc_spin_v_oovv,cc_spin_v_ovvo,r2)
call compute_r1_spin(cc_nOab,cc_nVab,t1%f,t2%f,f_o,f_v,cc_spin_F_ov,cF_oo,cF_ov,cF_vv,cc_spin_v_oovo,cc_spin_v_ovov,r1%f)
call compute_r2_spin(cc_nOab,cc_nVab,t1%f,t2%f,tau%f,f_o,f_v,cF_oo,cF_ov,cF_vv,cW_oooo,cW_ovvo,cc_spin_v_ovoo,cc_spin_v_oovv,cc_spin_v_ovvo,r2%f)
! Max elements in the residuals
max_r1 = maxval(abs(r1(:,:)))
max_r2 = maxval(abs(r2(:,:,:,:)))
max_r1 = maxval(abs(r1%f(:,:)))
max_r2 = maxval(abs(r2%f(:,:,:,:)))
max_r = max(max_r1,max_r2)
call wall_time(ti)
! Update
if (cc_update_method == 'diis') then
!call update_t_ccsd(cc_nOab,cc_nVab,nb_iter,f_o,f_v,r1,r2,t1,t2,all_err1,all_err2,all_t1,all_t2)
!call update_t_ccsd_diis(cc_nOab,cc_nVab,nb_iter,f_o,f_v,r1,r2,t1,t2,all_err1,all_err2,all_t1,all_t2)
call update_t_ccsd_diis_v3(cc_nOab,cc_nVab,nb_iter,f_o,f_v,r1,r2,t1,t2,all_err,all_t)
!call update_t_ccsd(cc_nOab,cc_nVab,nb_iter,f_o,f_v,r1%f,r2%f,t1%f,t2%f,all_err1,all_err2,all_t1%f,all_t2)
!call update_t_ccsd_diis(cc_nOab,cc_nVab,nb_iter,f_o,f_v,r1%f,r2%f,t1%f,t2%f,all_err1,all_err2,all_t1%f,all_t2)
call update_t_ccsd_diis_v3(cc_nOab,cc_nVab,nb_iter,f_o,f_v,r1%f,r2%f,t1%f,t2%f,all_err,all_t)
! Standard update as T = T - Delta
elseif (cc_update_method == 'none') then
call update_t1(cc_nOab,cc_nVab,f_o,f_v,r1,t1)
call update_t2(cc_nOab,cc_nVab,f_o,f_v,r2,t2)
call update_t1(cc_nOab,cc_nVab,f_o,f_v,r1%f,t1%f)
call update_t2(cc_nOab,cc_nVab,f_o,f_v,r2%f,t2%f)
else
print*,'Unkonw cc_method_method: '//cc_update_method
endif
call compute_tau_spin(cc_nOab,cc_nVab,t1,t2,tau)
call compute_tau_t_spin(cc_nOab,cc_nVab,t1,t2,tau_t)
call compute_tau_spin(cc_nOab,cc_nVab,t1%f,t2%f,tau%f)
call compute_tau_t_spin(cc_nOab,cc_nVab,t1%f,t2%f,tau_t%f)
call wall_time(tf)
if (cc_dev) then
print*,'Update:',tf-ti,'s'
endif
! Print
call ccsd_energy_spin(cc_nOab,cc_nVab,t1,t2,cc_spin_F_ov,cc_spin_v_oovv,energy)
call ccsd_energy_spin(cc_nOab,cc_nVab,t1%f,t2%f,cc_spin_F_ov,cc_spin_v_oovv,energy)
call wall_time(tfi)
write(*,'(A3,I6,A3,F18.12,A3,F16.12,A3,ES10.2,A3,ES10.2,A2)') ' | ',nb_iter,' | ', &
@ -159,8 +170,8 @@ subroutine run_ccsd_spin_orb
print*,''
if (write_amplitudes) then
call write_t1(cc_nOab,cc_nVab,t1)
call write_t2(cc_nOab,cc_nVab,t2)
call write_t1(cc_nOab,cc_nVab,t1%f)
call write_t2(cc_nOab,cc_nVab,t2%f)
call ezfio_set_utils_cc_io_amplitudes('Read')
endif
@ -168,8 +179,6 @@ subroutine run_ccsd_spin_orb
if (cc_update_method == 'diis') then
deallocate(all_err,all_t)
endif
deallocate(tau,tau_t)
deallocate(r1,r2)
deallocate(cF_oo,cF_ov,cF_vv)
deallocate(cW_oooo,cW_ovvo)!,cW_vvvv)
@ -178,7 +187,7 @@ subroutine run_ccsd_spin_orb
if (cc_par_t .and. elec_alpha_num +elec_beta_num > 2) then
print*,'CCSD(T) calculation...'
call wall_time(ta)
call ccsd_par_t_spin_v2(cc_nOab,cc_nVab,t1,t2,f_o,f_v,cc_spin_f_ov,cc_spin_v_ooov,cc_spin_v_vvoo,t_corr)
call ccsd_par_t_spin_v2(cc_nOab,cc_nVab,t1%f,t2%f,f_o,f_v,cc_spin_f_ov,cc_spin_v_ooov,cc_spin_v_vvoo,t_corr)
call wall_time(tb)
print*,'Done'
print*,'Time: ',tb-ta, ' s'
@ -192,7 +201,13 @@ subroutine run_ccsd_spin_orb
call save_energy(uncorr_energy + energy, t_corr)
deallocate(f_o,f_v)
deallocate(t1,t2)
call gpu_deallocate(t1)
call gpu_deallocate(r1)
call gpu_deallocate(t2)
call gpu_deallocate(r2)
call gpu_deallocate(tau)
call gpu_deallocate(tau_t)
end