10
0
mirror of https://github.com/LCPQ/quantum_package synced 2024-12-25 13:53:49 +01:00
quantum_package/plugins/Full_CI_ZMQ/run_pt2_slave.irp.f
Anthony Scemama 873035e016 Squashed commit of the following:
commit 4b9c435dce0f3b3078d573e66fd32b40fca26497
Merge: 74e559c8 093e3fd0
Author: Anthony Scemama <scemama@irsamc.ups-tlse.fr>
Date:   Tue Sep 4 16:58:51 2018 +0200

    Merge branch 'thesis' of git://github.com/garniron/quantum_package into garniron-thesis

commit 093e3fd021
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Sep 4 16:13:00 2018 +0200

    removed ungodly hack

commit 8529a0f3f6
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Sep 4 14:57:19 2018 +0200

    reduced prints in pt2_stoch

commit 03b8f353bd
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Sep 4 14:41:46 2018 +0200

    teeth building check for pt2_stoch

commit 0d91b9310a
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Sep 4 14:35:04 2018 +0200

    timestamp of first pull

commit 34d9fa0165
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Sep 4 14:27:10 2018 +0200

    potential numerical precision bug

commit 9a0f900d8c
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Sep 4 14:09:51 2018 +0200

    tests if teeth can be built

commit dda0dc34df
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Mon Sep 3 17:48:04 2018 +0200

    corrected pt2_find_sample

commit a521f0cb82
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Mon Sep 3 16:08:02 2018 +0200

    tasks get by batches of Nproc

commit 997a5a1265
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Mon Sep 3 14:18:04 2018 +0200

    buffered task_id send

commit 99ea7948e0
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Mon Sep 3 12:29:12 2018 +0200

    unbalanced fragmentation

commit abb3b7e08b
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Sun Sep 2 17:18:44 2018 +0200

    overflow of pt2_J

commit 8df49f394b
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Sun Sep 2 15:58:48 2018 +0200

    removed useless computation of intermediate checkpoints

commit 4ba5b79eb3
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Sun Sep 2 15:50:14 2018 +0200

    dressing only sent for chosen checkpoint

commit a4a6a69459
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Sat Sep 1 17:01:56 2018 +0200

    cumulative dot_F

commit 6a7f04cb79
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Sat Sep 1 16:58:07 2018 +0200

    simpler purge

commit 168ca2f2e2
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Fri Aug 31 21:07:01 2018 +0200

    task list optimized

commit de4a0d0caf
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Fri Aug 31 18:57:03 2018 +0200

    removed print

commit fee31d4e3e
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Fri Aug 31 18:56:23 2018 +0200

    dress fragmentation

commit 02893a419d
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Fri Aug 31 15:52:16 2018 +0200

    bug in blocked search - replaced with thesis version

commit bb6e073cf1
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Thu Aug 30 21:24:45 2018 +0200

    ungodly hack to prevent double providing

commit 0609e8c627
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Thu Aug 30 20:52:05 2018 +0200

    debugging

commit a254fdd7cf
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Thu Aug 30 15:24:07 2018 +0200

    parallel bug

commit 2a6c1941d4
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Thu Aug 30 11:43:11 2018 +0200

    corrected when relative_error=0d0

commit bac039bdf1
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Thu Aug 30 10:58:17 2018 +0200

    relative error 1d-5

commit aae9d203ec
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Thu Aug 30 10:07:02 2018 +0200

    potential fragmentation bug

commit ad69f39f99
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Wed Aug 29 20:54:58 2018 +0200

    dress_zmq re-implemented

commit d78f64732a
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Wed Aug 29 11:30:19 2018 +0200

    pt2_stoch re-implemented

commit 4b9b54e19a
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Tue Aug 28 10:24:38 2018 +0200

    removed test for phase_mask_bit

commit 3abccca5e3
Author: Yann Garniron <yann.garniron@yahoo.fr>
Date:   Fri Aug 3 23:44:05 2018 +0200

    phasemask_bit
2018-09-04 17:31:45 +02:00

205 lines
5.2 KiB
Fortran

subroutine run_pt2_slave(thread,iproc,energy)
use f77_zmq
use selection_types
implicit none
double precision, intent(in) :: energy(N_states_diag)
integer, intent(in) :: thread, iproc
integer :: rc, i
integer :: worker_id, ctask, ltask
character*(512), allocatable :: task(:)
integer, allocatable :: task_id(:)
integer(ZMQ_PTR),external :: new_zmq_to_qp_run_socket
integer(ZMQ_PTR) :: zmq_to_qp_run_socket
integer(ZMQ_PTR), external :: new_zmq_push_socket
integer(ZMQ_PTR) :: zmq_socket_push
type(selection_buffer) :: buf
logical :: done
double precision,allocatable :: pt2(:,:)
integer :: n_tasks, k
integer, allocatable :: i_generator(:), subset(:)
allocate(task_id(pt2_n_tasks_max), task(pt2_n_tasks_max))
allocate(pt2(N_states,pt2_n_tasks_max), i_generator(pt2_n_tasks_max), subset(pt2_n_tasks_max))
zmq_to_qp_run_socket = new_zmq_to_qp_run_socket()
integer, external :: connect_to_taskserver
if (connect_to_taskserver(zmq_to_qp_run_socket,worker_id,thread) == -1) then
call end_zmq_to_qp_run_socket(zmq_to_qp_run_socket)
return
endif
zmq_socket_push = new_zmq_push_socket(thread)
buf%N = 0
n_tasks = 1
call create_selection_buffer(0, 0, buf)
done = .False.
do while (.not.done)
n_tasks = max(1,n_tasks)
n_tasks = min(n_tasks,pt2_n_tasks_max)
integer, external :: get_tasks_from_taskserver
if (get_tasks_from_taskserver(zmq_to_qp_run_socket,worker_id, task_id, task, n_tasks) == -1) then
exit
endif
done = task_id(n_tasks) == 0
if (done) n_tasks = n_tasks-1
if (n_tasks == 0) exit
do k=1,n_tasks
read (task(k),*) subset(k), i_generator(k)
enddo
double precision :: time0, time1
call wall_time(time0)
do k=1,n_tasks
pt2(:,k) = 0.d0
buf%cur = 0
call select_connected(i_generator(k),energy,pt2(1,k),buf,subset(k),pt2_F(i_generator(k)))
enddo
call wall_time(time1)
integer, external :: tasks_done_to_taskserver
if (tasks_done_to_taskserver(zmq_to_qp_run_socket,worker_id,task_id,n_tasks) == -1) then
done = .true.
endif
call push_pt2_results(zmq_socket_push, i_generator, pt2, task_id, n_tasks)
! Try to adjust n_tasks around 1 second per job
n_tasks = min(n_tasks,int( 1.d0*dble(n_tasks) / (time1 - time0 + 1.d-9)))+1
! n_tasks = n_tasks+1
end do
integer, external :: disconnect_from_taskserver
do i=1,300
if (disconnect_from_taskserver(zmq_to_qp_run_socket,worker_id) /= -2) exit
call sleep(1)
print *, 'Retry disconnect...'
end do
call end_zmq_push_socket(zmq_socket_push,thread)
call end_zmq_to_qp_run_socket(zmq_to_qp_run_socket)
call delete_selection_buffer(buf)
end subroutine
subroutine push_pt2_results(zmq_socket_push, index, pt2, task_id, n_tasks)
use f77_zmq
use selection_types
implicit none
integer(ZMQ_PTR), intent(in) :: zmq_socket_push
double precision, intent(in) :: pt2(N_states,n_tasks)
integer, intent(in) :: n_tasks, index(n_tasks), task_id(n_tasks)
integer :: rc
rc = f77_zmq_send( zmq_socket_push, n_tasks, 4, ZMQ_SNDMORE)
if (rc == -1) then
return
endif
if(rc /= 4) stop 'push'
rc = f77_zmq_send( zmq_socket_push, index, 4*n_tasks, ZMQ_SNDMORE)
if (rc == -1) then
return
endif
if(rc /= 4*n_tasks) stop 'push'
rc = f77_zmq_send( zmq_socket_push, pt2, 8*N_states*n_tasks, ZMQ_SNDMORE)
if (rc == -1) then
return
endif
if(rc /= 8*N_states*n_tasks) stop 'push'
rc = f77_zmq_send( zmq_socket_push, task_id, n_tasks*4, 0)
if (rc == -1) then
return
endif
if(rc /= 4*n_tasks) stop 'push'
! Activate is zmq_socket_push is a REQ
IRP_IF ZMQ_PUSH
IRP_ELSE
character*(2) :: ok
rc = f77_zmq_recv( zmq_socket_push, ok, 2, 0)
if (rc == -1) then
return
endif
if ((rc /= 2).and.(ok(1:2) /= 'ok')) then
print *, irp_here//': error in receiving ok'
stop -1
endif
IRP_ENDIF
end subroutine
subroutine pull_pt2_results(zmq_socket_pull, index, pt2, task_id, n_tasks)
use f77_zmq
use selection_types
implicit none
integer(ZMQ_PTR), intent(in) :: zmq_socket_pull
double precision, intent(inout) :: pt2(N_states,*)
integer, intent(out) :: index(*)
integer, intent(out) :: n_tasks, task_id(*)
integer :: rc, rn, i
rc = f77_zmq_recv( zmq_socket_pull, n_tasks, 4, 0)
if (rc == -1) then
n_tasks = 1
task_id(1) = 0
endif
if(rc /= 4) stop 'pull'
rc = f77_zmq_recv( zmq_socket_pull, index, 4*n_tasks, 0)
if (rc == -1) then
n_tasks = 1
task_id(1) = 0
endif
if(rc /= 4*n_tasks) stop 'pull'
rc = f77_zmq_recv( zmq_socket_pull, pt2, N_states*8*n_tasks, 0)
if (rc == -1) then
n_tasks = 1
task_id(1) = 0
endif
if(rc /= 8*N_states*n_tasks) stop 'pull'
rc = f77_zmq_recv( zmq_socket_pull, task_id, n_tasks*4, 0)
if (rc == -1) then
n_tasks = 1
task_id(1) = 0
endif
if(rc /= 4*n_tasks) stop 'pull'
! Activate is zmq_socket_pull is a REP
IRP_IF ZMQ_PUSH
IRP_ELSE
rc = f77_zmq_send( zmq_socket_pull, 'ok', 2, 0)
if (rc == -1) then
n_tasks = 1
task_id(1) = 0
endif
if (rc /= 2) then
print *, irp_here//': error in sending ok'
stop -1
endif
IRP_ENDIF
end subroutine