Merge branch 'master' into ELPA_GPU

.gitlab-ci.yml

@@ -1618,6 +1618,8 @@ test_project:
     - ./test_real
     - make distclean
     - popd
+    - pushd build
     - make distclean
     - rm -rf installdest
+    - popd
 

configure.ac

@@ -943,11 +943,10 @@ AC_SUBST([FC_MODOUT])
 AC_SUBST([OPENMP_CFLAGS])
 AC_SUBST([OPENMP_FCFLAGS])
 AC_SUBST([OPENMP_LDFLAGS])
-#AC_SUBST(OPT_FCFLAGS)
 AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
 
-rm -rf modules/ .fortran_dependencies/
-mkdir modules
+rm -rf modules/ private_modules/ .fortran_dependencies/
+mkdir modules private_modules
 
 #gl_VISIBILITY
 #AH_BOTTOM([#if HAVE_VISIBILITY

generated_headers.am

@@ -4,10 +4,10 @@ define extract_interface
 endef
 
 elpa test:
-	mkdir $@
+	@mkdir $@
 
 test/shared_sources: | test
-	mkdir $@
+	@mkdir $@
 
 config-f90.h: config.h
 	@echo "Generating $@...";

src/elpa1.F90

@@ -84,13 +84,6 @@
 module ELPA1
   use precision
   use elpa_utilities
-  use elpa1_compute
-
-#ifdef HAVE_DETAILED_TIMINGS
-  use timings
-#endif
-  use iso_c_binding
-  use elpa_mpi
 
   implicit none
 
@@ -254,6 +247,7 @@ contains
 
 function get_elpa_communicators(mpi_comm_global, my_prow, my_pcol, mpi_comm_rows, mpi_comm_cols) result(mpierr)
    use precision
+   use elpa_mpi
    implicit none
 
    integer(kind=ik), intent(in)  :: mpi_comm_global, my_prow, my_pcol
@@ -318,6 +312,8 @@ function solve_evp_real_1stage_double(na, nev, a, lda, ev, q, ldq, nblk, &
    use timings
 #endif
    use iso_c_binding
+   use elpa_mpi
+   use elpa1_compute
    implicit none
 
    integer(kind=ik), intent(in)  :: na, nev, lda, ldq, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols
@@ -569,6 +565,8 @@ function solve_evp_complex_1stage_double(na, nev, a, lda, ev, q, ldq, nblk, matr
 #endif
    use precision
    use iso_c_binding
+   use elpa_mpi
+   use elpa1_compute
    implicit none
 
    integer(kind=ik), intent(in)     :: na, nev, lda, ldq, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols

src/elpa2.F90

@@ -67,20 +67,8 @@ module ELPA2
 ! Version 1.1.2, 2011-02-21
 
   use elpa_utilities
-  use elpa1_compute
   use elpa1, only : elpa_print_times, time_evp_back, time_evp_fwd, time_evp_solve
   use elpa2_utilities
-  use elpa2_compute
-  use elpa_pdgeqrf
-
-  use iso_c_binding
-
-#ifdef WITH_GPU_VERSION
-!  use cuda_routines
-!  use cuda_c_kernel
-!  use iso_c_binding
-#endif
-  use elpa_mpi
 
   implicit none
 
@@ -171,6 +159,9 @@ contains
     use timings
 #endif
 
+   use elpa1_compute
+   use elpa2_compute
+   use elpa_mpi
    use precision
    use cuda_functions
    use mod_check_for_gpu
@@ -860,6 +851,9 @@ function solve_evp_complex_2stage_single(na, nev, a, lda, ev, q, ldq, nblk, &
 #ifdef HAVE_DETAILED_TIMINGS
    use timings
 #endif
+   use elpa1_compute
+   use elpa2_compute
+   use elpa_mpi
    use precision
    use cuda_functions
    use mod_check_for_gpu

test_project/src/test_real.F90

@@ -154,6 +154,12 @@ program test_real_example
    na_rows = numroc(na, nblk, my_prow, 0, np_rows)
    na_cols = numroc(na, nblk, my_pcol, 0, np_cols)
 
+   ! All ELPA routines need MPI communicators for communicating within
+   ! rows or columns of processes, these are set in get_elpa_communicators.
+
+   mpierr = get_elpa_communicators(mpi_comm_world, my_prow, my_pcol, &
+                                   mpi_comm_rows, mpi_comm_cols)
+
    ! set up the scalapack descriptor for the checks below
    ! For ELPA the following restrictions hold:
    ! - block sizes in both directions must be identical (args 4 a. 5)