Working version of ELPA with new API

Still missing is the compatibility layer, currently it only compiles
when configure is called with

  --disable-legacy

Also, a more general solution to parameter passing via environment
variables would be nice.

configure.ac

@@ -4,7 +4,7 @@ AC_PREREQ([2.69])
 AC_INIT([elpa],m4_esyscmd_s([awk '/^ *Version:/ {print $2;}' elpa.spec]), [elpa-library@mpcdf.mpg.de])
 AC_SUBST([PACKAGE_VERSION])
 
-AC_CONFIG_SRCDIR([src/elpa_t.F90])
+AC_CONFIG_SRCDIR([src/elpa.F90])
 
 AM_INIT_AUTOMAKE([foreign -Wall subdir-objects])
 
@@ -12,6 +12,8 @@ AC_CONFIG_MACRO_DIR([m4])
 AC_CONFIG_HEADERS([config.h])
 AM_SILENT_RULES([yes])
 
+# ABI version
+#
 # Set the libtool library version, see LIBRARY_INTERFACE
 #
 # See http://www.gnu.org/software/libtool/manual/html_node/Updating-version-info.html
@@ -26,7 +28,10 @@ AM_SILENT_RULES([yes])
 #    were added by the new interface)
 #
 AC_SUBST([ELPA_SO_VERSION], [8:0:0])
-#
+
+# API Version
+AC_DEFINE([EARLIEST_API_VERSION], [20170403], [Earliest supported ELPA API version])
+AC_DEFINE([CURRENT_API_VERSION], [20170403], [Current ELPA API version])
 
 
 AX_CHECK_GNU_MAKE()
@@ -34,10 +39,12 @@ if test x$_cv_gnu_make_command = x ; then
         AC_MSG_ERROR([Need GNU Make])
 fi
 
-#AC_CHECK_PROG(CPP_FOUND,cpp,yes,no)
-#if test x"${CPP_FOUND}" = xno; then
-#  AC_MSG_ERROR([no cpp found])
-#fi
+AC_ARG_ENABLE([legacy],
+              AS_HELP_STRING([--disable-legacy],
+                             [do not build legacy API, default yes]),
+              [],
+              [enable_legacy=yes])
+AM_CONDITIONAL([ENABLE_LEGACY],[test x"$enable_legacy" = x"yes"])
 
 # gnu-make fortran module dependencies
 m4_include([fdep/fortran_dependencies.m4])
@@ -352,7 +359,8 @@ use_band_to_full_blocking=yes
 AC_MSG_CHECKING(whether BAND_TO_FLULL_BLOCKING is requested)
 AC_ARG_ENABLE(band-to-full-blocking,[AS_HELP_STRING([--enable-band-to-full-blocking],
                                                     [build ELPA2 with blocking in band_to_full (default: enabled)])],
-              want_band_to_full_blocking="yes", want_to_full_blocking="no")
+              [want_band_to_full_blocking="yes"],
+	      [want_to_full_blocking="no"])
 AC_MSG_RESULT([${want_band_to_full_blocking}])
 
 if test x"${enableval}" = x"no" ; then
@@ -633,7 +641,7 @@ m4_foreach_w([elpa_m4_kind],[real complex],[
         fi
         # find the number of this kernel
         ELPA_2STAGE_[]m4_toupper(elpa_m4_kind)[]_DEFAULT=`grep -i '^ *X(ELPA_2STAGE_'$default_[]elpa_m4_kind[]_kernel'\>' $srcdir/elpa/elpa_constants.h.in | \
-                                                                awk '// { gsub(/.*, */, ""); gsub(/).*/, ""); print;}'`
+                                                                awk '// { gsub(/^[[^,]]*, */, ""); gsub(/,.*/, ""); print;}'`
         AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kind)[_DEFAULT])
 ])
 
@@ -835,14 +843,26 @@ m4_foreach_w([elpa_m4_kernel],elpa_m4_all_kernels,[
   AM_CONDITIONAL([WITH_]m4_toupper(elpa_m4_kernel)[_KERNEL],[test x"$use_[]elpa_m4_kernel[]" = x"yes"])
   if test x"$use_[]elpa_m4_kernel[]" = x"yes" ; then
     AC_DEFINE([WITH_]m4_toupper(elpa_m4_kernel)[_KERNEL],[1],[Build elpa_m4_kernel kernel])
+    kernel_defined=1
+  else
+    kernel_defined=0
   fi
+  [ELPA_2STAGE_]m4_toupper(elpa_m4_kernel)[_COMPILED]=$kernel_defined
+  AC_SUBST([ELPA_2STAGE_]m4_toupper(elpa_m4_kernel)[_COMPILED])
 ])
 
 
 AM_CONDITIONAL([WITH_GPU_VERSION],[test x"$use_gpu" = x"yes"])
 if test x"${use_gpu}" = x"yes" ; then
   AC_DEFINE([WITH_GPU_VERSION],[1],[enable GPU support])
+  ELPA_2STAGE_COMPLEX_GPU_COMPILED=1
+  ELPA_2STAGE_REAL_GPU_COMPILED=1
+else
+  ELPA_2STAGE_COMPLEX_GPU_COMPILED=0
+  ELPA_2STAGE_REAL_GPU_COMPILED=0
 fi
+AC_SUBST([ELPA_2STAGE_COMPLEX_GPU_COMPILED])
+AC_SUBST([ELPA_2STAGE_REAL_GPU_COMPILED])
 
 LT_INIT
 
@@ -879,7 +899,6 @@ AM_CONDITIONAL([WANT_SINGLE_PRECISION_COMPLEX],[test x"$want_single_precision" =
 AC_SUBST([MPI_BINARY])
 AC_SUBST([WITH_MKL])
 AC_SUBST([WITH_BLACS])
-AC_SUBST([with_amd_bulldozer_kernel])
 AC_SUBST([FC_MODINC])
 AC_SUBST([FC_MODOUT])
 AC_SUBST([OPENMP_CFLAGS])
@@ -887,8 +906,9 @@ AC_SUBST([OPENMP_FCFLAGS])
 AC_SUBST([OPENMP_LDFLAGS])
 AC_SUBST([DOXYGEN_OUTPUT_DIR], [docs])
 
-rm -rf modules/ private_modules/ .fortran_dependencies/
-mkdir modules private_modules
+#rm -rf modules/ private_modules/ .fortran_dependencies/
+mkdir -p modules private_modules
+
 
 #gl_VISIBILITY
 #AH_BOTTOM([#if HAVE_VISIBILITY
@@ -990,3 +1010,6 @@ echo "*  the next stable release.                                           *"
 echo "***********************************************************************"
 echo " "
 make -f $srcdir/generated_headers.am generated-headers top_srcdir="$srcdir"
+
+
+echo "TODO: Check crazy BAND_TO_FULL_BLOCKING flags"

elpa/elpa.h

@@ -6,4 +6,6 @@
 #include <elpa/elpa_constants.h>
 #include <elpa/elpa_generated.h>
 
+const char *elpa_strerr(int elpa_error);
+
 #endif

elpa/elpa_constants.h.in

@@ -5,83 +5,95 @@
  */
 
 /* Private helper macros */
-#define ELPA_ENUM_ENTRY(name, value) \
+#define ELPA_ENUM_ENTRY(name, value, ...) \
         name = value,
-#define ELPA_ENUM_SUM(name, value) +1
+#define ELPA_ENUM_SUM(name, value, ...) +1
 
 
 /* Solver constants */
-#define ELPA_SOLVER_MAP(X) \
+#define ELPA_FOR_ALL_SOLVERS(X) \
         X(ELPA_SOLVER_1STAGE, 1) \
         X(ELPA_SOLVER_2STAGE, 2)
 
 enum ELPA_SOLVERS {
-        ELPA_SOLVER_MAP(ELPA_ENUM_ENTRY)
+        ELPA_FOR_ALL_SOLVERS(ELPA_ENUM_ENTRY)
 };
 
-#define ELPA_NUMBER_OF_SOLVERS (0 ELPA_SOLVER_MAP(ELPA_ENUM_SUM))
+#define ELPA_NUMBER_OF_SOLVERS (0 ELPA_FOR_ALL_SOLVERS(ELPA_ENUM_SUM))
 
 /* Kernel constants */
-#define ELPA_REAL_KERNEL_MAP(X) \
-        X(ELPA_2STAGE_REAL_GENERIC, 1) \
-        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE, 2) \
-        X(ELPA_2STAGE_REAL_BGP, 3) \
-        X(ELPA_2STAGE_REAL_BGQ, 4) \
-        X(ELPA_2STAGE_REAL_SSE, 5) \
-        X(ELPA_2STAGE_REAL_SSE_BLOCK2, 6) \
-        X(ELPA_2STAGE_REAL_SSE_BLOCK4, 7) \
-        X(ELPA_2STAGE_REAL_SSE_BLOCK6, 8) \
-        X(ELPA_2STAGE_REAL_AVX_BLOCK2, 9) \
-        X(ELPA_2STAGE_REAL_AVX_BLOCK4, 10) \
-        X(ELPA_2STAGE_REAL_AVX_BLOCK6, 11) \
-        X(ELPA_2STAGE_REAL_AVX2_BLOCK2, 12) \
-        X(ELPA_2STAGE_REAL_AVX2_BLOCK4, 13) \
-        X(ELPA_2STAGE_REAL_AVX2_BLOCK6, 14) \
-        X(ELPA_2STAGE_REAL_AVX512_BLOCK2, 15) \
-        X(ELPA_2STAGE_REAL_AVX512_BLOCK4, 16) \
-        X(ELPA_2STAGE_REAL_AVX512_BLOCK6, 17) \
-        X(ELPA_2STAGE_REAL_GPU, 18) \
-        X(ELPA_2STAGE_REAL_DEFAULT, @ELPA_2STAGE_REAL_DEFAULT@)
+#define ELPA_FOR_ALL_2STAGE_REAL_KERNELS(X) \
+        X(ELPA_2STAGE_REAL_GENERIC, 1, @ELPA_2STAGE_REAL_GENERIC_COMPILED@) \
+        X(ELPA_2STAGE_REAL_GENERIC_SIMPLE, 2, @ELPA_2STAGE_REAL_GENERIC_SIMPLE_COMPILED@) \
+        X(ELPA_2STAGE_REAL_BGP, 3, @ELPA_2STAGE_REAL_BGP_COMPILED@) \
+        X(ELPA_2STAGE_REAL_BGQ, 4, @ELPA_2STAGE_REAL_BGQ_COMPILED@) \
+        X(ELPA_2STAGE_REAL_SSE_ASSEMBLY, 5, @ELPA_2STAGE_REAL_SSE_ASSEMBLY_COMPILED@) \
+        X(ELPA_2STAGE_REAL_SSE_BLOCK2, 6, @ELPA_2STAGE_REAL_SSE_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_REAL_SSE_BLOCK4, 7, @ELPA_2STAGE_REAL_SSE_BLOCK4_COMPILED@) \
+        X(ELPA_2STAGE_REAL_SSE_BLOCK6, 8, @ELPA_2STAGE_REAL_SSE_BLOCK6_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX_BLOCK2, 9, @ELPA_2STAGE_REAL_AVX_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX_BLOCK4, 10, @ELPA_2STAGE_REAL_AVX_BLOCK4_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX_BLOCK6, 11, @ELPA_2STAGE_REAL_AVX_BLOCK6_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX2_BLOCK2, 12, @ELPA_2STAGE_REAL_AVX2_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX2_BLOCK4, 13, @ELPA_2STAGE_REAL_AVX2_BLOCK4_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX2_BLOCK6, 14, @ELPA_2STAGE_REAL_AVX2_BLOCK6_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX512_BLOCK2, 15, @ELPA_2STAGE_REAL_AVX512_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX512_BLOCK4, 16, @ELPA_2STAGE_REAL_AVX512_BLOCK4_COMPILED@) \
+        X(ELPA_2STAGE_REAL_AVX512_BLOCK6, 17, @ELPA_2STAGE_REAL_AVX512_BLOCK6_COMPILED@) \
+        X(ELPA_2STAGE_REAL_GPU, 18, @ELPA_2STAGE_REAL_GPU_COMPILED@)
+
+#define ELPA_FOR_ALL_2STAGE_REAL_KERNELS_AND_DEFAULT(X) \
+        ELPA_FOR_ALL_2STAGE_REAL_KERNELS(X) \
+        X(ELPA_2STAGE_REAL_INVALID, -1, choke me) \
+        X(ELPA_2STAGE_REAL_DEFAULT, @ELPA_2STAGE_REAL_DEFAULT@, choke me)
 
 enum ELPA_REAL_KERNELS {
-        ELPA_REAL_KERNEL_MAP(ELPA_ENUM_ENTRY)
+        ELPA_FOR_ALL_2STAGE_REAL_KERNELS_AND_DEFAULT(ELPA_ENUM_ENTRY)
 };
 
 
-#define ELPA_COMPLEX_KERNEL_MAP(X) \
-        X(ELPA_2STAGE_COMPLEX_GENERIC, 1) \
-        X(ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE, 2) \
-        X(ELPA_2STAGE_COMPLEX_BGP, 3) \
-        X(ELPA_2STAGE_COMPLEX_BGQ, 4) \
-        X(ELPA_2STAGE_COMPLEX_SSE, 5) \
-        X(ELPA_2STAGE_COMPLEX_SSE_BLOCK1, 6) \
-        X(ELPA_2STAGE_COMPLEX_SSE_BLOCK2, 7) \
-        X(ELPA_2STAGE_COMPLEX_AVX_BLOCK1, 8) \
-        X(ELPA_2STAGE_COMPLEX_AVX_BLOCK2, 9) \
-        X(ELPA_2STAGE_COMPLEX_AVX2_BLOCK1, 10) \
-        X(ELPA_2STAGE_COMPLEX_AVX2_BLOCK2, 11) \
-        X(ELPA_2STAGE_COMPLEX_AVX512_BLOCK1, 12) \
-        X(ELPA_2STAGE_COMPLEX_AVX512_BLOCK2, 13) \
-        X(ELPA_2STAGE_COMPLEX_GPU, 14) \
-        X(ELPA_2STAGE_COMPLEX_DEFAULT, @ELPA_2STAGE_COMPLEX_DEFAULT@)
+#define ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS(X) \
+        X(ELPA_2STAGE_COMPLEX_GENERIC, 1, @ELPA_2STAGE_COMPLEX_GENERIC_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE, 2, @ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_BGP, 3, @ELPA_2STAGE_COMPLEX_BGP_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_BGQ, 4, @ELPA_2STAGE_COMPLEX_BGQ_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_SSE_ASSEMBLY, 5, @ELPA_2STAGE_COMPLEX_SSE_ASSEMBLY_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_SSE_BLOCK1, 6, @ELPA_2STAGE_COMPLEX_SSE_BLOCK1_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_SSE_BLOCK2, 7, @ELPA_2STAGE_COMPLEX_SSE_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_AVX_BLOCK1, 8, @ELPA_2STAGE_COMPLEX_AVX_BLOCK1_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_AVX_BLOCK2, 9, @ELPA_2STAGE_COMPLEX_AVX_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_AVX2_BLOCK1, 10, @ELPA_2STAGE_COMPLEX_AVX2_BLOCK1_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_AVX2_BLOCK2, 11, @ELPA_2STAGE_COMPLEX_AVX2_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_AVX512_BLOCK1, 12, @ELPA_2STAGE_COMPLEX_AVX512_BLOCK1_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_AVX512_BLOCK2, 13, @ELPA_2STAGE_COMPLEX_AVX512_BLOCK2_COMPILED@) \
+        X(ELPA_2STAGE_COMPLEX_GPU, 14, @ELPA_2STAGE_COMPLEX_GPU_COMPILED@)
+
+#define ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS_AND_DEFAULT(X) \
+        ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS(X) \
+        X(ELPA_2STAGE_COMPLEX_INVALID, -1, choke me) \
+        X(ELPA_2STAGE_COMPLEX_DEFAULT, @ELPA_2STAGE_COMPLEX_DEFAULT@, choke me)
 
 enum ELPA_COMPLEX_KERNELS {
-        ELPA_COMPLEX_KERNEL_MAP(ELPA_ENUM_ENTRY)
+        ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS_AND_DEFAULT(ELPA_ENUM_ENTRY)
 };
 
 
 
 /* General constants */
-#define ELPA_CONSTANTS_MAP(X) \
+#define ELPA_FOR_ALL_ERRORS(X) \
+        X(ELPA_OK, 0) \
         X(ELPA_ERROR, -1) \
-        X(ELPA_NO, 0) \
-        X(ELPA_OK, 1)
+        X(ELPA_ERROR_ENTRY_NOT_FOUND, -2) \
+        X(ELPA_ERROR_INVALID_VALUE, -3) \
+        X(ELPA_ERROR_VALUE_ALREADY_SET, -4) \
+        X(ELPA_ERROR_NO_STRING_REPRESENTATION, -5)
 
-enum ELPA_CONSTANTS {
-        ELPA_CONSTANTS_MAP(ELPA_ENUM_ENTRY)
+enum ELPA_ERRORS {
+        ELPA_FOR_ALL_ERRORS(ELPA_ENUM_ENTRY)
+};
 
-        /* -1 to take out the default kernel again */
-        ELPA_2STAGE_NUMBER_OF_COMPLEX_KERNELS = (0 ELPA_COMPLEX_KERNEL_MAP(ELPA_ENUM_SUM) -1),
-        ELPA_2STAGE_NUMBER_OF_REAL_KERNELS = (0 ELPA_REAL_KERNEL_MAP(ELPA_ENUM_SUM) -1),
+enum ELPA_CONSTANTS {
+        ELPA_2STAGE_NUMBER_OF_COMPLEX_KERNELS = (0 ELPA_FOR_ALL_2STAGE_COMPLEX_KERNELS(ELPA_ENUM_SUM)),
+        ELPA_2STAGE_NUMBER_OF_REAL_KERNELS = (0 ELPA_FOR_ALL_2STAGE_REAL_KERNELS(ELPA_ENUM_SUM)),
 };
 

fdep/fortran_dependencies.mk

@@ -105,7 +105,7 @@ endef
 
 # $1 program
 define program_dependencies
-	$(_f90_only_verbose)rm -f .$p.dep.args
+	$(_f90_only_verbose)echo -n > .$p.dep.args
 	$(foreach argument,$(_$p_use_mods) $(_$p_def_mods) $(foreach l,$(call recursive_lib_deps,$p),$(_$l_use_mods) $(_$l_def_mods)),$(call append_to,$(argument),.$p.dep.args))
 	$(_f90_only_verbose)$(top_srcdir)/fdep/fortran_dependencies.pl $p < .$p.dep.args >> $@ || { rm $@; exit 1; }
 	$(_f90_only_verbose)rm -f .$p.dep.args

src/elpa.F90

+! The ELPA public API
+
+module elpa
+  use elpa_api
+  use elpa_constants
+
+  implicit none
+  public
+
+  contains
+
+    function elpa_allocate() result(obj)
+      use elpa_impl
+      class(elpa_t), pointer :: obj
+      obj => elpa_impl_allocate()
+    end function
+
+
+    subroutine elpa_deallocate(obj)
+      class(elpa_t), pointer :: obj
+      call obj%destroy()
+      deallocate(obj)
+    end subroutine
+
+end module

src/elpa1/elpa1_template.X90

@@ -58,7 +58,7 @@ function elpa_solve_evp_&
          &MATH_DATATYPE&
 	 &_1stage_&
 	 &PRECISION&
-	 &_impl ( na, nev, a, lda, ev, q, ldq, nblk, matrixCols, mpi_comm_rows, &
+	 &_impl (obj, na, nev, a, lda, ev, q, ldq, nblk, matrixCols, mpi_comm_rows, &
 	        mpi_comm_cols, mpi_comm_all, useGPU, time_evp_fwd,     &
                 time_evp_solve, time_evp_back, summary_timings) result(success)
    use precision
@@ -70,11 +70,13 @@ function elpa_solve_evp_&
    use timings_dummy
 #endif
    use iso_c_binding
+   use elpa_api
    use elpa_mpi
    use elpa1_compute
    use elpa1_utilities, only : gpu_usage_via_environment_variable
    implicit none
 
+   class(elpa_t), intent(in)                       :: obj
    integer(kind=c_int), intent(in)                 :: na, nev, lda, ldq, nblk, matrixCols, mpi_comm_rows, &
                                                       mpi_comm_cols, mpi_comm_all
    real(kind=REAL_DATATYPE), intent(out)           :: ev(na)
@@ -115,7 +117,6 @@ function elpa_solve_evp_&
    integer(kind=c_int)                             :: my_pe, n_pes, my_prow, my_pcol, mpierr
    real(kind=C_DATATYPE_KIND), allocatable         :: e(:)
    real(kind=c_double)                             :: ttt0, ttt1 ! MPI_WTIME always needs double
-   logical, save                                   :: firstCall = .true.
    logical                                         :: wantDebug
    integer(kind=c_int)                             :: istat
    character(200)                                  :: errorMessage
@@ -142,14 +143,8 @@ function elpa_solve_evp_&
    call timer%stop("mpi_communication")
    success = .true.
 
-   wantDebug = .false.
-   if (firstCall) then
-     ! are debug messages desired?
-     wantDebug = debug_messages_via_environment_variable()
-     firstCall = .false.
-   endif
-
-   do_useGPU      = .false.
+   wantDebug = obj%get("debug") == 1
+   do_useGPU = .false.
 
 
    if (useGPU) then

src/elpa1/legacy_interface/elpa1_template_legacy.X90

@@ -114,7 +114,7 @@ function elpa_solve_evp_&
    call mpi_comm_rank(mpi_comm_cols,my_pcol,mpierr)
 
    success = .true.
-   if (elpa_init(20170403) /= ELPA_OK) then
+   if (elpa_init(CURRENT_API_VERSION) /= ELPA_OK) then
      print *,  "ELPA API version not supported"
      success = .false.
      return
@@ -123,6 +123,7 @@ function elpa_solve_evp_&
    e = elpa_allocate()
 
    call e%set("na", na)
+   call e%set("nev", nev)
    call e%set("local_nrows", lda)
    call e%set("local_ncols", matrixCols)
    call e%set("nblk", nblk)
@@ -131,9 +132,7 @@ function elpa_solve_evp_&
    call e%set("process_row", my_prow)
    call e%set("process_col", my_pcol)
 
-   call e%setup(successInternal)
-
-   if (successInternal .ne. ELPA_OK) then
+   if (e%setup() .ne. ELPA_OK) then
      print *, "Cannot setup ELPA instance"
      success = .false.
      return
@@ -174,6 +173,7 @@ function elpa_solve_evp_&
    endif
 
    call e%solve(a(1:lda,1:matrixCols), ev, q(1:ldq,1:matrixCols), successInternal)
+
    if (successInternal .ne. ELPA_OK) then
      print *, "Cannot solve with ELPA 1stage"
      success = .false.
@@ -181,9 +181,9 @@ function elpa_solve_evp_&
    endif
 
    if (elpa_print_times) then
-     time_evp_fwd =   e%get("time_evp_fwd")
-     time_evp_solve = e%get("time_evp_solve")
-     time_evp_back =  e%get("time_evp_back")
+     time_evp_fwd =   e%get_double("time_evp_fwd")
+     time_evp_solve = e%get_double("time_evp_solve")
+     time_evp_back =  e%get_double("time_evp_back")
    endif
    call e%destroy()
 
@@ -197,4 +197,4 @@ function elpa_solve_evp_&
 
 end function
 
-
+! vim: syntax=fortran

src/elpa1/legacy_interface/elpa_cholesky_template_legacy.X90

@@ -99,16 +99,16 @@
 
       success = .true.
 
-      if (elpa_init(20170403) /= ELPA_OK) then
+      if (elpa_init(CURRENT_API_VERSION) /= ELPA_OK) then
         print *, "ELPA API version not supported"
         success = .false.
-        stop
         return
       endif
 
       e = elpa_allocate()
 
       call e%set("na", na)
+      call e%set("nev", nev)
       call e%set("local_nrows", lda)
       call e%set("local_ncols", matrixCols)
       call e%set("nblk", nblk)
@@ -116,12 +116,9 @@
       call e%set("mpi_comm_rows", mpi_comm_rows)
       call e%set("mpi_comm_cols", mpi_comm_cols)
 
-      call e%setup(successInternal)
-
-      if (successInternal .ne. ELPA_OK) then
+      if (e%setup() .ne. ELPA_OK) then
         print *, "Cannot setup ELPA instance"
         success = .false.
-        stop
         return
       endif
 
@@ -129,7 +126,6 @@
 
       if (successInternal .ne. ELPA_OK) then
         print *, "Cannot run cholesky"
-        stop
         success = .false.
         return
       else
@@ -149,3 +145,5 @@
 #undef COMPLEXCASE
 #undef DOUBLE_PRECISION
 #undef SINGLE_PRECISION
+
+! vim: syntax=fortran

src/elpa1/legacy_interface/elpa_invert_trm_legacy.X90

@@ -95,7 +95,6 @@
 !      integer(kind=ik)             :: istat
 !      character(200)               :: errorMessage
 
-       integer(kind=ik)            :: nev
        integer(kind=iK)            :: successInternal
        type(elpa_t)                :: e
 
@@ -107,10 +106,9 @@
 
        success = .true.
 
-       if (elpa_init(20170403) /= ELPA_OK) then
+       if (elpa_init(CURRENT_API_VERSION) /= ELPA_OK) then
          print *, "ELPA API version not supported"
          success = .false.
-         stop
          return
        endif
 
@@ -124,12 +122,9 @@
        call e%set("mpi_comm_rows", mpi_comm_rows)
        call e%set("mpi_comm_cols", mpi_comm_cols)
 
-       call e%setup(successInternal)
-
-       if (successInternal .ne. ELPA_OK) then
+       if (e%setup() .ne. ELPA_OK) then
          print *, "Cannot setup ELPA instance"
          success = .false.
-         stop
          return
        endif
 
@@ -137,7 +132,6 @@
 
        if (successInternal .ne. ELPA_OK) then
          print *, "Cannot run invert_trm"
-         stop
          success = .false.
          return
        else

src/elpa1/legacy_interface/elpa_multiply_a_b_legacy.X90

@@ -84,7 +84,6 @@
 #endif
 #endif
   !    integer(kind=ik)                         :: my_prow, my_pcol, np_rows, np_cols, mpierr
-       integer(kind=ik)                         :: nev
   !    integer(kind=ik)                         :: l_cols, l_rows, l_rows_np
 !      integer(kind=ik)                         :: np, n, nb, nblk_mult, lrs, lre, lcs, lce
 !      integer(kind=ik)                         :: gcol_min, gcol, goff
@@ -120,10 +119,9 @@
       !l_rows = local_index(na,  my_prow, np_rows, nblk, -1) ! Local rows of a and b
       !l_cols = local_index(ncb, my_pcol, np_cols, nblk, -1) ! Local cols of b
 
-      if (elpa_init(20170403) /= ELPA_OK) then
+      if (elpa_init(CURRENT_API_VERSION) /= ELPA_OK) then
         print *, "ELPA API version not supported"
         success = .false.
-        stop
         return
       endif
 
@@ -131,18 +129,15 @@
 
       call e%set("na", na)
       call e%set("local_nrows", lda)
-      call e%set("local_ncols", matrixCols)
+      call e%set("local_ncols", ldaCols)
       call e%set("nblk", nblk)
 
       call e%set("mpi_comm_rows", mpi_comm_rows)
       call e%set("mpi_comm_cols", mpi_comm_cols)
 
-      call e%setup(successInternal)
-      if (successInternal .ne. ELPA_OK) then
+      if (e%setup() .ne. ELPA_OK) then
         print *, "Cannot setup ELPA instance"
-        stop
         success = .false.
-        return
       endif
 
       call e%hermitian_multiply(uplo_a, uplo_c, na, ncb, a(1:lda,1:ldaCols), lda, ldaCols, &
@@ -151,7 +146,6 @@
 
       if (successInternal .ne. ELPA_OK) then
         print *, "Cannot run multiply_a_b"
-        stop
         success = .false.
         return
       endif

src/elpa1/legacy_interface/elpa_solve_tridi_legacy.X90

@@ -56,14 +56,10 @@
 
 #include "../../general/sanity.X90"
 
-      !use elpa1_compute, solve_tridi_&
-      !                   &PRECISION&
-!			 &_private => solve_tridi_&
-!			 &PRECISION
       use precision
-      use elpa_type
+      use elpa
 #ifdef HAVE_DETAILED_TIMINGS
-       use timings
+      use timings
 #else
        use timings_dummy
 #endif
@@ -79,7 +75,7 @@
       logical, intent(in)      :: wantDebug
       logical :: success
       integer(kind=iK)            :: successInternal
-      type(elpa_t)                :: e
+      type(elpa_t)                :: elpa
 
       call timer%start("elpa_solve_tridi_&
       &PRECISION&
@@ -87,32 +83,29 @@
 
       success = .false.
 
-      if (elpa_init(20170403) /= ELPA_OK) then
+      if (elpa_init(CURRENT_API_VERSION) /= ELPA_OK) then
         print *, "ELPA API version not supported"
         success = .false.
-        stop
-        return
       endif
 
-      e = elpa_allocate()
+      elpa = elpa_allocate()
 
-      call e%set("na", na)
-      call e%set("local_nrows", lda)
-      call e%set("local_ncols", matrixCols)
-      call e%set("nblk", nblk)
+      call elpa%set("na", na)
+      call elpa%set("nev", nev)
+      call elpa%set("local_nrows", ldq)
+      call elpa%set("local_ncols", matrixCols)
+      call elpa%set("nblk", nblk)
 
-      call e%set("mpi_comm_rows", mpi_comm_rows)
-      call e%set("mpi_comm_cols", mpi_comm_cols)
+      call elpa%set("mpi_comm_rows", mpi_comm_rows)
+      call elpa%set("mpi_comm_cols", mpi_comm_cols)
 
-      call e%setup(successInternal)
-      if (successInternal .ne. ELPA_OK) then
+      if (elpa%setup() .ne. ELPA_OK) then
         print *, "Cannot setup ELPA instance"
         success = .false.
-        stop
         return
       endif
 
-      call e%solve_tridi(d(1:na), e(1:na), q(1:ldq,1:matrixCols), successInternal)
+      call elpa%solve_tridi(d(1:na), e(1:na), q(1:ldq,1:matrixCols), successInternal)
 
       if (successInternal .ne. ELPA_OK) then
         print *, "Cannot run solve_tridi"
@@ -122,7 +115,7 @@
       else
         success =.true.
       endif
-      call e%destroy()
+      call elpa%destroy()
 
       call elpa_uninit()
 
@@ -135,3 +128,4 @@
 #undef DOUBLE_PRECISION
 #undef SINGLE_PRECISION
 
+! vim: syntax=fortran

src/elpa2/elpa2.F90

@@ -55,18 +55,11 @@
 #include "config-f90.h"
 !> \brief Fortran module which provides the routines to use the 2-stage ELPA solver
 module elpa2_impl
-
-! Version 1.1.2, 2011-02-21
-
-  use elpa_utilities
- !  use elpa1_impl, only : elpa_print_times, time_evp_back, time_evp_fwd, time_evp_solve
-  use elpa2_utilities