- better definition of northo/ndiag

git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@6696 c92efa57-630b-4861-b058-cf58834340f0

CPV/ortho_base.f90

@@ -139,7 +139,7 @@ END SUBROUTINE diagonalize_parallel
    SUBROUTINE mesure_diag_perf( n )
       !
       USE mp_global,   ONLY: nproc_image, me_image, intra_image_comm, root_image
-      USE mp_global,   ONLY: np_ortho2, np_ortho, me_ortho, ortho_comm, ortho_comm_id
+      USE mp_global,   ONLY: nproc_ortho, np_ortho, me_ortho, ortho_comm, ortho_comm_id
       USE io_global,   ONLY: ionode, stdout
       USE mp,          ONLY: mp_sum, mp_bcast, mp_barrier
       USE mp,          ONLY: mp_max
@@ -159,7 +159,7 @@ END SUBROUTINE diagonalize_parallel
       !
       ! Check if number of PEs for orthogonalization/diagonalization is given from the input
       !
-      IF( np_ortho2 > 0 ) THEN
+      IF( nproc_ortho > 0 ) THEN
          use_parallel_diag = .TRUE. 
          RETURN
       END IF
@@ -284,7 +284,7 @@ END SUBROUTINE diagonalize_parallel
    SUBROUTINE mesure_mmul_perf( n )
       !
       USE mp_global,   ONLY: nproc_image, me_image, intra_image_comm, &
-                             root_image, ortho_comm, np_ortho2, np_ortho, &
+                             root_image, ortho_comm, nproc_ortho, np_ortho, &
                              me_ortho, init_ortho_group, ortho_comm_id
       USE io_global,   ONLY: ionode, stdout
       USE mp,          ONLY: mp_sum, mp_bcast, mp_barrier
@@ -303,27 +303,14 @@ END SUBROUTINE diagonalize_parallel
       REAL(DP) :: cclock
       EXTERNAL :: cclock
       !
-      IF( np_ortho2 > 0 ) THEN
-         !
-         !  Here the number of processors is suggested on input
-         !
-         np    = np_ortho2 
-         if( np > MIN( n, nproc_image ) ) np = MIN( n, nproc_image )
-         np    = MAX( INT( SQRT( DBLE( np ) + 0.1d0 ) ), 1 ) 
-         !
-      ELSE
-         !
-         !  Take the maximum number of processors
-         !
-         np    = MIN( n, nproc_image )
-         np    = MAX( INT( SQRT( DBLE( np ) + 0.1d0 ) ), 1 ) 
-         !
-      END IF
-
-      ! 
-      !  Now test the allowed processors mesh sizes
+      np    = MAX( INT( SQRT( DBLE( nproc_ortho ) + 0.1d0 ) ), 1 ) 
+      !
+      !  Make ortho group compatible with the number of electronic states
+      !
+      np    = MIN( np, n )
+      !
+      !  Now re-define the ortho group and test the performance
       !
-
       CALL init_ortho_group( np * np, intra_image_comm )
 
       CALL descla_init( desc, n, n, np_ortho, me_ortho, ortho_comm, ortho_comm_id )
@@ -368,8 +355,6 @@ END SUBROUTINE diagonalize_parallel
 
 #endif
 
-      CALL init_ortho_group( np*np, intra_image_comm )
-
 #if defined __PARA
       IF( ionode ) THEN
          WRITE( stdout, '(/,3X,"Constraints matrixes will be distributed block like on")' )

PP/projwfc.f90

@@ -109,7 +109,7 @@ PROGRAM projwfc
   USE io_files,   ONLY : nd_nmbr, prefix, tmp_dir, trimcheck
   USE noncollin_module, ONLY : noncolin
   USE mp,               ONLY : mp_bcast      
-  USE mp_global,        ONLY : mp_startup, np_ortho2
+  USE mp_global,        ONLY : mp_startup, nproc_ortho
   USE environment,      ONLY : environment_start
   !
   ! for GWW
@@ -216,7 +216,7 @@ PROGRAM projwfc
   IF (noncolin) THEN 
      CALL projwave_nc(filproj, lsym )
   ELSE
-     IF( np_ortho2 > 1 ) THEN
+     IF( nproc_ortho > 1 ) THEN
         CALL pprojwave (filproj, lsym)
      ELSE
         CALL projwave (filproj, lsym, lgww)

PW/setup.f90

@@ -752,13 +752,14 @@ SUBROUTINE check_para_diag( nelec )
   USE kinds,            ONLY : DP
   USE control_flags,    ONLY : use_para_diag, gamma_only
   USE io_global,        ONLY : stdout, ionode, ionode_id
-  USE mp_global,        ONLY : nproc_pool, init_ortho_group, np_ortho2, &
+  USE mp_global,        ONLY : nproc_pool, init_ortho_group, nproc_ortho, &
                                np_ortho, intra_pool_comm
 
   IMPLICIT NONE
 
   REAL(DP), INTENT(IN) :: nelec
   LOGICAL, SAVE :: first = .TRUE.
+  INTEGER :: np
 
   !  avoid synchronization problems when more images are active
 
@@ -768,28 +769,18 @@ SUBROUTINE check_para_diag( nelec )
 
   use_para_diag = .TRUE.
   !
-  !  here we initialize the sub group of processors that will take part
+  !  here we re-initialize the sub group of processors that will take part
   !  in the matrix diagonalization. 
   !  NOTE that the maximum number of processors may not be the optimal one,
   !  and -ndiag N argument can be used to force a given number N of processors
   !
-  IF( np_ortho2 < 1 ) THEN
-     !
-     !  use all the available processors
-     !
-     np_ortho2 = MIN( INT( nelec )/2, nproc_pool )
-     !
-     !  avoid problems with systems with a single electron
-     !
-     IF ( np_ortho2 < 1) np_ortho2 = 1
-     !
-  ELSE
-     !
-     np_ortho2 = MIN( INT( nelec )/2, np_ortho2 )
-     !
-  END IF
+  np = MAX( INT( SQRT( DBLE( nproc_ortho ) + 0.1d0 ) ), 1 )
+  !
+  !  Make ortho group compatible with the number of electronic states
+  !
+  np = MIN( INT( nelec )/2, np )
 
-  CALL init_ortho_group( np_ortho2, intra_pool_comm )
+  CALL init_ortho_group( np * np, intra_pool_comm )
 
   IF ( ionode ) THEN
      !