LCPQ/quantum_package
10
0
Fork 0
mirror of https://github.com/LCPQ/quantum_package synced 2024-12-25 13:53:49 +01:00
Code Issues Releases Wiki Activity

Tuning

Browse Source
This commit is contained in:
Anthony Scemama 2018-06-06 21:21:42 +02:00
parent 535a3402d8
commit c46635c5ed
8 changed files with 95 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
ocaml/Makefile
View File

@ -12,7 +12,8 @@ endif
LIBS= LIBS=
PKGS= PKGS=
OCAMLCFLAGS="-g" OCAMLCFLAGS="-g"
OCAMLBUILD=ocamlbuild -j 0 -cflags $(OCAMLCFLAGS) -lflags $(OCAMLCFLAGS) OCAMLOPTFLAGS="opt -O3 -nodynlink -remove-unused-arguments -rounds 16 -inline 100 -inline-max-unroll 100"
OCAMLBUILD=ocamlbuild -j 0 -cflags $(OCAMLCFLAGS) -lflags $(OCAMLCFLAGS) -ocamlopt $(OCAMLOPTFLAGS)
MLLFILES=$(wildcard *.mll) MLLFILES=$(wildcard *.mll)
MLFILES=$(wildcard *.ml) ezfio.ml Qptypes.ml Input_auto_generated.ml qp_edit.ml MLFILES=$(wildcard *.ml) ezfio.ml Qptypes.ml Input_auto_generated.ml qp_edit.ml
MLIFILES=$(wildcard *.mli) git MLIFILES=$(wildcard *.mli) git

8
plugins/Full_CI_ZMQ/pt2_stoch_routines.irp.f
View File

@ -396,7 +396,11 @@ end function
BEGIN_PROVIDER [ integer, comb_teeth ] BEGIN_PROVIDER [ integer, comb_teeth ]
implicit none implicit none
comb_teeth = 100 BEGIN_DOC
! Number of teeth in the comb
END_DOC
comb_teeth = min(1+N_det/10,100)
END_PROVIDER END_PROVIDER
@ -555,7 +559,7 @@ END_PROVIDER
comb_step = 1d0/dfloat(comb_teeth) comb_step = 1d0/dfloat(comb_teeth)
first_det_of_comb = 1 first_det_of_comb = 1
do i=1,N_det_generators do i=1,N_det_generators
if(pt2_weight(i)/norm_left < .5d0*comb_step) then if(pt2_weight(i)/norm_left < .25d0*comb_step) then
first_det_of_comb = i first_det_of_comb = i
exit exit
end if end if

4
plugins/Full_CI_ZMQ/run_pt2_slave.irp.f
View File

@ -82,8 +82,8 @@ subroutine run_pt2_slave(thread,iproc,energy)
endif endif
call push_pt2_results(zmq_socket_push, i_generator, pt2, task_id, n_tasks) call push_pt2_results(zmq_socket_push, i_generator, pt2, task_id, n_tasks)
! Try to adjust n_tasks around 10 seconds per job ! Try to adjust n_tasks around 1 second per job
n_tasks = int(10.d0 * dble(n_tasks) / (time1 - time0 + 1.d-9))+1 n_tasks = min(n_tasks,int( dble(n_tasks) / (time1 - time0 + 1.d-9)))+1
end do end do
integer, external :: disconnect_from_taskserver integer, external :: disconnect_from_taskserver

4
plugins/Full_CI_ZMQ/run_selection_slave.irp.f
View File

@ -97,8 +97,8 @@ subroutine run_selection_slave_new(thread,iproc,energy)
pt2(:,:) = 0d0 pt2(:,:) = 0d0
buf%cur = 0 buf%cur = 0
! Try to adjust n_tasks around 10 seconds per job ! Try to adjust n_tasks around 1 second per job
n_tasks = int(10.d0 * dble(n_tasks) / (time1 - time0 + 1.d-9))+1 n_tasks = min(n_tasks,int( dble(n_tasks) / (time1 - time0 + 1.d-9)))+1
end do end do
integer, external :: disconnect_from_taskserver integer, external :: disconnect_from_taskserver

67
plugins/Full_CI_ZMQ/selection.irp.f
View File

@ -5,7 +5,7 @@ BEGIN_PROVIDER [ integer, fragment_count ]
BEGIN_DOC BEGIN_DOC
! Number of fragments for the deterministic part ! Number of fragments for the deterministic part
END_DOC END_DOC
fragment_count = (elec_alpha_num-n_core_orb)*mo_tot_num fragment_count = (elec_alpha_num-n_core_orb)**2
END_PROVIDER END_PROVIDER
@ -690,8 +690,8 @@ subroutine splash_pq(mask, sp, det, i_gen, N_sel, bannedOrb, banned, mat, intere
if (interesting(i) == i_gen) then if (interesting(i) == i_gen) then
if(sp == 3) then if(sp == 3) then
do j=1,mo_tot_num
do k=1,mo_tot_num do k=1,mo_tot_num
do j=1,mo_tot_num
banned(j,k,2) = banned(k,j,1) banned(j,k,2) = banned(k,j,1)
enddo enddo
enddo enddo
@ -909,14 +909,22 @@ subroutine get_d1(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, coefs)
if(.not. bannedOrb(puti, mi)) then if(.not. bannedOrb(puti, mi)) then
tmp_row = 0d0 tmp_row = 0d0
do putj=1, hfix-1 do putj=1, hfix-1
if(lbanned(putj, ma) .or. banned(putj, puti,bant)) cycle if(lbanned(putj, ma)) cycle
hij = (mo_bielec_integral(putj, hfix, p1, p2)-mo_bielec_integral(putj,hfix,p2,p1)) * get_phase_bi(phasemask, ma, ma, putj, p1, hfix, p2) if(banned(putj, puti,bant)) cycle
hij = mo_bielec_integral(putj, hfix, p1, p2)-mo_bielec_integral(putj,hfix,p2,p1)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, ma, ma, putj, p1, hfix, p2)
tmp_row(1:N_states,putj) += hij * coefs(1:N_states) tmp_row(1:N_states,putj) += hij * coefs(1:N_states)
endif
end do end do
do putj=hfix+1, mo_tot_num do putj=hfix+1, mo_tot_num
if(lbanned(putj, ma) .or. banned(putj, puti,bant)) cycle if(lbanned(putj, ma)) cycle
hij = (mo_bielec_integral(hfix,putj,p1, p2)-mo_bielec_integral(hfix,putj,p2,p1)) * get_phase_bi(phasemask, ma, ma, hfix, p1, putj, p2) if(banned(putj, puti,bant)) cycle
hij = mo_bielec_integral(putj,hfix,p2, p1)-mo_bielec_integral(putj,hfix,p1,p2)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, ma, ma, hfix, p1, putj, p2)
tmp_row(1:N_states,putj) += hij * coefs(1:N_states) tmp_row(1:N_states,putj) += hij * coefs(1:N_states)
endif
end do end do
if(ma == 1) then if(ma == 1) then
@ -935,15 +943,21 @@ subroutine get_d1(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, coefs)
!p1 fixed !p1 fixed
putj = p1 putj = p1
if(.not. banned(putj,puti,bant)) then if(.not. banned(putj,puti,bant)) then
hij = mo_bielec_integral(puti,hfix,pfix,p2) * get_phase_bi(phasemask, ma, mi, hfix, p2, puti, pfix) hij = mo_bielec_integral(puti,hfix,pfix,p2)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, ma, mi, hfix, p2, puti, pfix)
tmp_row(:,puti) += hij * coefs(:) tmp_row(:,puti) += hij * coefs(:)
endif endif
end if
putj = p2 putj = p2
if(.not. banned(putj,puti,bant)) then if(.not. banned(putj,puti,bant)) then
hij = mo_bielec_integral(puti,hfix,pfix,p1) * get_phase_bi(phasemask, ma, mi, hfix, p1, puti, pfix) hij = mo_bielec_integral(puti,hfix,pfix,p1)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, ma, mi, hfix, p1, puti, pfix)
tmp_row2(:,puti) += hij * coefs(:) tmp_row2(:,puti) += hij * coefs(:)
endif endif
end if
end do end do
if(mi == 1) then if(mi == 1) then
@ -953,7 +967,9 @@ subroutine get_d1(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, coefs)
mat(:,p1,:) += tmp_row(:,:) mat(:,p1,:) += tmp_row(:,:)
mat(:,p2,:) += tmp_row2(:,:) mat(:,p2,:) += tmp_row2(:,:)
end if end if
else
else ! sp /= 3
if(p(0,ma) == 3) then if(p(0,ma) == 3) then
do i=1,3 do i=1,3
hfix = h(1,ma) hfix = h(1,ma)
@ -962,14 +978,24 @@ subroutine get_d1(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, coefs)
p2 = p(turn3(2,i), ma) p2 = p(turn3(2,i), ma)
tmp_row = 0d0 tmp_row = 0d0
do putj=1,hfix-1 do putj=1,hfix-1
if(lbanned(putj,ma) .or. banned(puti,putj,1)) cycle if(lbanned(putj,ma)) cycle
hij = (mo_bielec_integral(putj, hfix, p1, p2)-mo_bielec_integral(putj,hfix,p2,p1)) * get_phase_bi(phasemask, ma, ma, putj, p1, hfix, p2) ! if(banned(puti,putj,1)) cycle
if(banned(putj,puti,1)) cycle
hij = mo_bielec_integral(putj, hfix, p1, p2)-mo_bielec_integral(putj,hfix,p2,p1)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, ma, ma, putj, p1, hfix, p2)
tmp_row(:,putj) += hij * coefs(:) tmp_row(:,putj) += hij * coefs(:)
endif
end do end do
do putj=hfix+1,mo_tot_num do putj=hfix+1,mo_tot_num
if(lbanned(putj,ma) .or. banned(puti,putj,1)) cycle if(lbanned(putj,ma)) cycle
hij = (mo_bielec_integral(putj, hfix, p2, p1)-mo_bielec_integral(putj,hfix,p1,p2)) * get_phase_bi(phasemask, ma, ma, hfix, p1, putj, p2) ! if(banned(puti,putj,1)) cycle
if(banned(putj,puti,1)) cycle
hij = mo_bielec_integral(putj, hfix, p2, p1)-mo_bielec_integral(putj,hfix,p1,p2)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, ma, ma, hfix, p1, putj, p2)
tmp_row(:,putj) += hij * coefs(:) tmp_row(:,putj) += hij * coefs(:)
endif
end do end do
mat(:, :puti-1, puti) += tmp_row(:,:puti-1) mat(:, :puti-1, puti) += tmp_row(:,:puti-1)
@ -986,15 +1012,21 @@ subroutine get_d1(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, coefs)
if(lbanned(puti,ma)) cycle if(lbanned(puti,ma)) cycle
putj = p2 putj = p2
if(.not. banned(puti,putj,1)) then if(.not. banned(puti,putj,1)) then
hij = mo_bielec_integral(puti,hfix, p1,pfix) * get_phase_bi(phasemask, mi, ma, hfix, pfix, puti, p1) hij = mo_bielec_integral(puti,hfix, p1,pfix)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, mi, ma, hfix, pfix, puti, p1)
tmp_row(:,puti) += hij * coefs(:) tmp_row(:,puti) += hij * coefs(:)
endif endif
end if
putj = p1 putj = p1
if(.not. banned(puti,putj,1)) then if(.not. banned(puti,putj,1)) then
hij = mo_bielec_integral(puti, hfix, p2, pfix) * get_phase_bi(phasemask, mi, ma, hfix, pfix, puti, p2) hij = mo_bielec_integral(puti, hfix, p2, pfix)
if (hij /= 0.d0) then
hij = hij * get_phase_bi(phasemask, mi, ma, hfix, pfix, puti, p2)
tmp_row2(:,puti) += hij * coefs(:) tmp_row2(:,puti) += hij * coefs(:)
endif endif
end if
end do end do
mat(:,:p2-1,p2) += tmp_row(:,:p2-1) mat(:,:p2-1,p2) += tmp_row(:,:p2-1)
mat(:,p2,p2:) += tmp_row(:,p2:) mat(:,p2,p2:) += tmp_row(:,p2:)
@ -1016,10 +1048,11 @@ subroutine get_d1(gen, phasemask, bannedOrb, banned, mat, mask, h, p, sp, coefs)
do i1=1,p(0,s1) do i1=1,p(0,s1)
ib = 1 ib = 1
if(s1 == s2) ib = i1+1 if(s1 == s2) ib = i1+1
do i2=ib,p(0,s2)
p1 = p(i1,s1) p1 = p(i1,s1)
if(bannedOrb(p1, s1)) cycle
do i2=ib,p(0,s2)
p2 = p(i2,s2) p2 = p(i2,s2)
if(bannedOrb(p1, s1) .or. bannedOrb(p2, s2) .or. banned(p1, p2, 1)) cycle if(bannedOrb(p2, s2) .or. banned(p1, p2, 1)) cycle
call apply_particles(mask, s1, p1, s2, p2, det, ok, N_int) call apply_particles(mask, s1, p1, s2, p2, det, ok, N_int)
call i_h_j(gen, det, N_int, hij) call i_h_j(gen, det, N_int, hij)
mat(:, p1, p2) += coefs(:) * hij mat(:, p1, p2) += coefs(:) * hij

2
plugins/Full_CI_ZMQ/zmq_selection.irp.f
View File

@ -60,7 +60,7 @@ subroutine ZMQ_selection(N_in, pt2)
task = ' ' task = ' '
do i= 1, N_det_generators do i= 1, N_det_generators
if (i>ishft(N_det_generators,-7)) then if (i>ishft(N_det_generators,-2)) then
write(task(ipos:ipos+30),'(I9,1X,I9,1X,I9,''|'')') 0, i, N write(task(ipos:ipos+30),'(I9,1X,I9,1X,I9,''|'')') 0, i, N
ipos += 30 ipos += 30
if (ipos > 63970) then if (ipos > 63970) then

26
src/Integrals_Bielec/qp_ao_ints.irp.f
View File

@ -1,10 +1,16 @@
program qp_ao_ints program qp_ao_ints
use omp_lib use omp_lib
implicit none implicit none
IRP_IF MPI
include 'mpif.h'
IRP_ENDIF
integer :: ierr
BEGIN_DOC BEGIN_DOC
! Increments a running calculation to compute AO integrals ! Increments a running calculation to compute AO integrals
END_DOC END_DOC
integer :: i integer :: i
PROVIDE zmq_context mpi_master zmq_state zmq_context
call switch_qp_run_to_master call switch_qp_run_to_master
@ -17,15 +23,27 @@ program qp_ao_ints
double precision :: integral, ao_bielec_integral double precision :: integral, ao_bielec_integral
integral = ao_bielec_integral(1,1,1,1) integral = ao_bielec_integral(1,1,1,1)
character*(64) :: state do
call wait_for_state(zmq_state,state) call wait_for_state('ao_integrals',zmq_state)
do while (state /= 'Stopped') if (zmq_state(1:7) == 'Stopped') then
exit
endif
!$OMP PARALLEL DEFAULT(PRIVATE) PRIVATE(i) !$OMP PARALLEL DEFAULT(PRIVATE) PRIVATE(i)
i = omp_get_thread_num() i = omp_get_thread_num()
call ao_bielec_integrals_in_map_slave_tcp(i) call ao_bielec_integrals_in_map_slave_tcp(i)
!$OMP END PARALLEL !$OMP END PARALLEL
call wait_for_state(zmq_state,state) IRP_IF MPI
call MPI_BARRIER(MPI_COMM_WORLD, ierr)
if (ierr /= MPI_SUCCESS) then
print *, irp_here, 'error in barrier'
endif
IRP_ENDIF
enddo enddo
IRP_IF MPI
call MPI_finalize(i)
IRP_ENDIF
print *, 'Done' print *, 'Done'
end end

2
src/ZMQ/utils.irp.f
View File

@ -337,7 +337,7 @@ IRP_ENDIF
stop 'Unable to set ZMQ_LINGER on push socket' stop 'Unable to set ZMQ_LINGER on push socket'
endif endif
rc = f77_zmq_setsockopt(new_zmq_push_socket,ZMQ_SNDHWM,5,4) rc = f77_zmq_setsockopt(new_zmq_push_socket,ZMQ_SNDHWM,1,4)
if (rc /= 0) then if (rc /= 0) then
stop 'Unable to set ZMQ_SNDHWM on push socket' stop 'Unable to set ZMQ_SNDHWM on push socket'
endif endif