Better scaling of Davidson

plugins/dress_zmq/dress_stoch_routines.irp.f

@@ -17,14 +17,9 @@ END_PROVIDER
   e = elec_num - n_core_orb * 2
   pt2_n_tasks_max = 1 + min((e*(e-1))/2, int(dsqrt(dble(N_det_generators)))/10)
   do i=1,N_det_generators
-    if (maxval(dabs(psi_coef_sorted_gen(i,1:N_states))) > 0.01d0) then
-      pt2_F(i) = pt2_n_tasks_max
-    else
-      pt2_F(i) = 1
-    endif
+    pt2_F(i) = 1 + dble(pt2_n_tasks_max)*maxval(dsqrt(dabs(psi_coef_sorted_gen(i,1:N_states))))
   enddo
 
-  
   if(N_det_generators < 128) then
     pt2_minDetInFirstTeeth = 1
     pt2_N_teeth = 1

src/Davidson/davidson_parallel.irp.f

@@ -129,8 +129,11 @@ subroutine davidson_slave_work(zmq_to_qp_run_socket, zmq_socket_push, N_st, sze,
     endif
     if(task_id == 0) exit
     read (msg,*) imin, imax, ishift, istep
-    v_t = 0.d0
-    s_t = 0.d0
+    integer :: k
+    do k=imin,imax
+      v_t(:,k) = 0.d0
+      s_t(:,k) = 0.d0
+    enddo
     call H_S2_u_0_nstates_openmp_work(v_t,s_t,u_t,N_st,N_det,imin,imax,ishift,istep)
     if (task_done_to_taskserver(zmq_to_qp_run_socket,worker_id,task_id) == -1) then
         print *,  irp_here, 'Unable to send task_done'
@@ -362,7 +365,7 @@ subroutine H_S2_u_0_nstates_zmq(v_0,s_0,u_0,N_st,sze)
   ishift=0
   imin=1
   N_det_inv = 1.d0/dble(N_det)
-  max_workload = 50000.d0
+  max_workload = 10000.d0
   do imax=1,N_det
     w = w + 1.d0
     if (w > max_workload) then

src/Davidson/u0Hu0.irp.f

@@ -175,7 +175,7 @@ subroutine H_S2_u_0_nstates_openmp_work_$N_int(v_t,s_t,u_t,N_st,sze,istart,iend,
   ASSERT (istart > 0)
   ASSERT (istep  > 0)
 
-  !$OMP DO SCHEDULE(dynamic,64)
+  !$OMP DO SCHEDULE(guided,64)
   do k_a=istart+ishift,iend,istep
 
     krow = psi_bilinear_matrix_rows(k_a)
@@ -246,7 +246,7 @@ subroutine H_S2_u_0_nstates_openmp_work_$N_int(v_t,s_t,u_t,N_st,sze,istart,iend,
   enddo
   !$OMP END DO 
 
-  !$OMP DO SCHEDULE(dynamic,64)
+  !$OMP DO SCHEDULE(guided,64)
   do k_a=istart+ishift,iend,istep