Remove assumed size arrays from complex generic simple kernel

This commit is performance critical and has to be timed carefully.
Thus one can switch back to the old implementation. The new one,
however is more safe and better to debug

src/elpa2_kernels/elpa2_kernels_complex_simple.F90

@@ -71,9 +71,13 @@ contains
     implicit none
 
     integer(kind=ik), intent(in)    :: nb, nq, ldq
+#ifdef DESPERATELY_WANT_ASSUMED_SIZE
     complex(kind=ck), intent(inout) :: q(ldq,*)
     complex(kind=ck), intent(in)    :: hh(*)
-
+#else
+    complex(kind=ck), intent(inout) :: q(1:ldq,1:nb)
+    complex(kind=ck), intent(in)    :: hh(1:nb)
+#endif
     integer(kind=ik)                :: i
     complex(kind=ck)                :: h1, tau1, x(nq)
 
@@ -110,9 +114,13 @@ contains
     implicit none
 
     integer(kind=ik), intent(in)    :: nb, nq, ldq, ldh
+#ifdef DESPERATELY_WANT_ASSUMED_SIZE
     complex(kind=ck), intent(inout) :: q(ldq,*)
     complex(kind=ck), intent(in)    :: hh(ldh,*)
-
+#else
+    complex(kind=ck), intent(inout) :: q(1:ldq,1:nb+1)
+    complex(kind=ck), intent(in)    :: hh(1:ldh,1:2)
+#endif
     complex(kind=ck)                :: s, h1, h2, tau1, tau2, x(nq), y(nq)
     integer(kind=ik)                :: i
 

src/mod_compute_hh_trafo_complex.F90

@@ -128,12 +128,24 @@ module compute_hh_trafo_complex
              ttt = mpi_wtime()
              do j = ncols, 1, -1
 #ifdef WITH_OPENMP
+#ifdef DESPERATELY_WANT_ASSUMED_SIZE
                call single_hh_trafo_complex_generic_simple(a(1,j+off+a_off,istripe,my_thread), &
                                                           bcast_buffer(1,j+off),nbw,nl,stripe_width)
 #else
+               call single_hh_trafo_complex_generic_simple(a(1:stripe_width,j+off+a_off:j+off_a_off+nbw-1,istripe,my_thread), &
+                                                           bcast_buffer(1:nbw,j+off),nbw,nl,stripe_width)
+#endif
+
+#else /* WITH_OPENMP */
+#ifdef DESPERATELY_WANT_ASSUMED_SIZE
                call single_hh_trafo_complex_generic_simple(a(1,j+off+a_off,istripe), &
                                                           bcast_buffer(1,j+off),nbw,nl,stripe_width)
+#else
+               call single_hh_trafo_complex_generic_simple(a(1:stripe_width,j+off+a_off:j+off+a_off+nbw-1,istripe), &
+                                                          bcast_buffer(1:nbw,j+off),nbw,nl,stripe_width)
 #endif
+
+#endif /* WITH_OPENMP */
              enddo
 #if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
            endif