For performance reasons make usage of assumed-size arrays default

This can be switched off via --disable-assumed-size

configure.ac

@@ -983,17 +983,17 @@ DX_MAN_FEATURE(ON)
 DX_HTML_FEATURE(ON)
 DX_INIT_DOXYGEN([ELPA], [Doxyfile], [docs])
 
-USE_ASSUMED_SIZE=no
-AC_MSG_CHECKING(whether --enable-assumed-size is specified)
+USE_ASSUMED_SIZE=yes
+AC_MSG_CHECKING(whether --disable-assumed-size is specified)
 AC_ARG_ENABLE([assumed-size],
-                AS_HELP_STRING([--enable-assumed-size],
-                               [use assumed-size Fortran arrays]),
-              [],
-              [USE_ASSUMED_SIZE=no])
+                AS_HELP_STRING([--disable-assumed-size],
+			       [do NOT use assumed-size Fortran arrays]),
+              [USE_ASSUMED_SIZE=no],
+              [USE_ASSUMED_SIZE=yes])
 AC_MSG_RESULT([${USE_ASSUMED_SIZE}])
 AM_CONDITIONAL([WITH_USE_ASSUMED_SIZE],[test x"$USE_ASSUMED_SIZE" = x"yes"])
 if test x"${USE_ASSUMED_SIZE}" = x"yes" ; then
-  AC_DEFINE([USE_ASSUMED_SIZE],[1],[use assumed size Fortran arrays, even if not debuggable])
+  AC_DEFINE([USE_ASSUMED_SIZE],[1],[for performance reasons use assumed size Fortran arrays, even if not debuggable])
 fi
 
 if test x"${want_single_precision}" = x"yes" ; then

src/elpa2_compute_complex_template.X90

@@ -217,9 +217,12 @@
       l_cols_tile = tile_size/np_cols ! local cols of a tile
 
       if (useGPU) then
+#if !defined(USE_ASSUMED_SIZE)
         if (size(a,dim=1) .ne. lda .or. size(a,dim=2) .ne. na_cols) then
           print *,"bandred_complex: sizes of a wrong ? ",lda,size(a,dim=1),na_cols,size(a,dim=2)
         endif
+#endif
+
 #ifdef DOUBLE_PRECISION_COMPLEX
         successCUDA = cuda_memcpy(a_dev, loc(a(1,1)),(lda)*(na_cols)*size_of_double_complex_datatype,cudaMemcpyHostToDevice)
 #else
@@ -1046,10 +1049,12 @@
        enddo ! istep
 
        if (useGPU) then
-
+#if !(defined(USE_ASSUMED_SIZE))
          if (size(a,dim=1)*size(a,dim=2) .ne. lda*na_cols) then
            print *,"bandred_complex: size a ",size(a,dim=1)*size(a,dim=2) , lda*na_cols
          endif
+#endif
+
 #ifdef DOUBLE_PRECISION_COMPLEX
          successCUDA = cuda_memcpy ( loc(a(1,1)), a_dev, lda*na_cols*size_of_double_complex_datatype,cudaMemcpyDeviceToHost)
 #else

src/elpa2_kernels/elpa2_kernels_real_simple.F90

@@ -58,6 +58,7 @@
 
 #include "config-f90.h"
 
+#ifndef USE_ASSUMED_SIZE
 module real_generic_simple_kernel
 
   private
@@ -68,6 +69,7 @@ module real_generic_simple_kernel
 #endif
 
   contains
+#endif
 
 #define REALCASE 1
 #define DOUBLE_PRECISION_REAL 1
@@ -87,5 +89,7 @@ module real_generic_simple_kernel
 #undef REALCASE
 #endif
 
+#ifndef USE_ASSUMED_SIZE
 end module real_generic_simple_kernel
+#endif
 ! --------------------------------------------------------------------------------------------------

src/elpa2_kernels/elpa2_kernels_real_template.X90

@@ -213,7 +213,7 @@
 
     integer(kind=ik), intent(in)    :: nb, ldq, ldh
 #ifdef USE_ASSUMED_SIZE
-    complex(kind=COMPLEX_DATATPYE), intent(inout) :: q(ldq/2,*)
+    complex(kind=COMPLEX_DATATYPE), intent(inout) :: q(ldq/2,*)
     real(kind=REAL_DATATYPE), intent(in)       :: hh(ldh,*)
 #else
     real(kind=REAL_DATATYPE), intent(inout)    :: q(:,:)
@@ -222,7 +222,7 @@
     real(kind=REAL_DATATYPE), intent(in)       :: s
 
 #ifdef PACK_REAL_TO_COMPLEX
-    complex(kind=COMPLEX_DATATPYE)                :: x1, x2, x3, x4, x5, x6, y1, y2, y3, y4, y5, y6
+    complex(kind=COMPLEX_DATATYPE)                :: x1, x2, x3, x4, x5, x6, y1, y2, y3, y4, y5, y6
 #else
     real(kind=REAL_DATATYPE)                   :: x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, &
                                        y1, y2, y3, y4, y5, y6, y7, y8, y9, y10, y11, y12
@@ -463,7 +463,7 @@
 
     integer(kind=ik), intent(in)     :: nb, ldq, ldh
 #ifdef USE_ASSUMED_SIZE
-    complex(kind=COMPLEX_DATATPYE), intent(inout)  :: q(ldq/2,*)
+    complex(kind=COMPLEX_DATATYPE), intent(inout)  :: q(ldq/2,*)
     real(kind=REAL_DATATYPE), intent(in)        :: hh(ldh,*)
 #else
     real(kind=REAL_DATATYPE),   intent(inout)   :: q(:,:)
@@ -471,7 +471,7 @@
 #endif
     real(kind=REAL_DATATYPE), intent(in)        :: s
 #ifdef PACK_REAL_TO_COMPLEX
-    complex(kind=COMPLEX_DATATPYE)                 :: x1, x2, x3, x4, y1, y2, y3, y4
+    complex(kind=COMPLEX_DATATYPE)                 :: x1, x2, x3, x4, y1, y2, y3, y4
 #else
     real(kind=REAL_DATATYPE)                    :: x1, x2, x3, x4, x5, x6, x7, x8, &
                                         y1, y2, y3, y4, y5, y6, y7, y8
@@ -654,7 +654,7 @@
 
     integer(kind=ik), intent(in)    :: nb, ldq, ldh
 #ifdef USE_ASSUMED_SIZE
-    complex(kind=COMPLEX_DATATPYE), intent(inout) :: q(ldq/2,*)
+    complex(kind=COMPLEX_DATATYPE), intent(inout) :: q(ldq/2,*)
     real(kind=REAL_DATATYPE), intent(in)       :: hh(ldh,*)
 #else
     real(kind=REAL_DATATYPE), intent(inout)    :: q(:,:) !q(1:ldq/2,1:nb+1)
@@ -663,7 +663,7 @@
     real(kind=REAL_DATATYPE), intent(in)       :: s
 
 #ifdef PACK_REAL_TO_COMPLEX
-    complex(kind=COMPLEX_DATATPYE)                :: x1, x2, y1, y2
+    complex(kind=COMPLEX_DATATYPE)                :: x1, x2, y1, y2
 #else
     real(kind=REAL_DATATYPE)                   :: x1, x2, x3, x4, y1, y2, y3, y4
 #endif

src/elpa2_kernels/elpa2_kernels_simple_template.X90

@@ -202,7 +202,7 @@
     call timer%start("kernel real generic simple: double_hh_trafo_real_generic_simple_single")
 #endif
 
-#endif 
+#endif
 
 #endif /* REALCASE==1 */
 

src/mod_compute_hh_trafo_real.F90

@@ -84,7 +84,7 @@ module compute_hh_trafo_real
          use cuda_c_kernel
          use cuda_functions
 
-#if defined(WITH_REAL_GENERIC_SIMPLE_KERNEL)
+#if defined(WITH_REAL_GENERIC_SIMPLE_KERNEL) && !(defined(USE_ASSUMED_SIZE))
          use real_generic_simple_kernel !, only : double_hh_trafo_generic_simple
 #endif
 
@@ -668,7 +668,7 @@ module compute_hh_trafo_real
          use cuda_c_kernel
          use cuda_functions
 
-#if defined(WITH_REAL_GENERIC_SIMPLE_KERNEL)
+#if defined(WITH_REAL_GENERIC_SIMPLE_KERNEL) && !(defined(USE_ASSUMED_SIZE))
          use real_generic_simple_kernel !, only : double_hh_trafo_generic_simple
 #endif