Some cleanup of the "time_evp*" entries

We decided not to provide this information via the key/value store, but
rather to use the ftimings module for that.

@loh will take care to implement ftimings in the new API
@amarek wants to port the legacy API so that the old global module
  variables still contain valid values

elpa/elpa_constants.h.in

@@ -84,9 +84,10 @@ enum ELPA_COMPLEX_KERNELS {
         X(ELPA_OK, 0) \
         X(ELPA_ERROR, -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)
+        X(ELPA_ERROR_ENTRY_INVALID_VALUE, -3) \
+        X(ELPA_ERROR_ENTRY_ALREADY_SET, -4) \
+        X(ELPA_ERROR_ENTRY_NO_STRING_REPRESENTATION, -5) \
+        X(ELPA_ERROR_ENTRY_READONLY, -6)
 
 enum ELPA_ERRORS {
         ELPA_FOR_ALL_ERRORS(ELPA_ENUM_ENTRY)

src/elpa1/elpa1_template.X90

@@ -59,8 +59,7 @@ function elpa_solve_evp_&
 	 &_1stage_&
 	 &PRECISION&
 	 &_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)
+	        mpi_comm_cols, mpi_comm_all, useGPU) result(success)
    use precision
    use cuda_functions
    use mod_check_for_gpu
@@ -108,15 +107,11 @@ function elpa_solve_evp_&
    logical, intent(in)                             :: useGPU
    logical                                         :: success
 
-   real(kind=c_double)                             :: time_evp_fwd,     &
-                                                      time_evp_solve, time_evp_back
-   logical, intent(in)                             :: summary_timings
    logical                                         :: do_useGPU
    integer(kind=ik)                                :: numberOfGPUDevices
 
    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                                         :: wantDebug
    integer(kind=c_int)                             :: istat
    character(200)                                  :: errorMessage
@@ -210,19 +205,12 @@ function elpa_solve_evp_&
      stop 1
    endif
 
-   ttt0 = MPI_Wtime()
    call tridiag_&
    &MATH_DATATYPE&
    &_&
    &PRECISION&
    & (na, a, lda, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols, ev, e, tau, do_useGPU)
 
-   ttt1 = MPI_Wtime()
-   if(my_prow==0 .and. my_pcol==0 .and. summary_timings) write(error_unit,*) 'Time tridiag_real :',ttt1-ttt0
-   time_evp_fwd = ttt1-ttt0
-
-   ttt0 = MPI_Wtime()
-
    call solve_tridi_&
    &PRECISION&
    & (na, nev, ev, e,  &
@@ -235,11 +223,6 @@ function elpa_solve_evp_&
       nblk, matrixCols, mpi_comm_rows, mpi_comm_cols, wantDebug, success)
    if (.not.(success)) return
 
-   ttt1 = MPI_Wtime()
-   if(my_prow==0 .and. my_pcol==0 .and. summary_timings) write(error_unit,*) 'Time solve_tridi  :',ttt1-ttt0
-   time_evp_solve = ttt1-ttt0
-
-   ttt0 = MPI_Wtime()
 #if COMPLEXCASE == 1
    q(1:l_rows,1:l_cols_nev) = q_real(1:l_rows,1:l_cols_nev)
 #endif
@@ -249,9 +232,6 @@ function elpa_solve_evp_&
    &_&
    &PRECISION&
    & (na, nev, a, lda, tau, q, ldq, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols, do_useGPU)
-   ttt1 = MPI_Wtime()
-   if(my_prow==0 .and. my_pcol==0 .and. summary_timings) write(error_unit,*) 'Time trans_ev_real:',ttt1-ttt0
-   time_evp_back = ttt1-ttt0
 
 #if COMPLEXCASE == 1
    deallocate(q_real, stat=istat, errmsg=errorMessage)

src/elpa1/legacy_interface/elpa1_template.X90

@@ -163,13 +163,11 @@ function elpa_solve_evp_&
      endif
    endif
 
-   if (elpa_print_times) then
-     call e%set("summary_timings", 1,successInternal)
-     if (successInternal .ne. ELPA_OK) then
-       print *, "Cannot set summary_timings"
-       success = .false.
-       return
-     endif
+   call e%set("summary_timings", 1,successInternal)
+   if (successInternal .ne. ELPA_OK) then
+     print *, "Cannot set summary_timings"
+     success = .false.
+     return
    endif
 
    call e%solve(a(1:lda,1:matrixCols), ev, q(1:ldq,1:matrixCols), successInternal)
@@ -180,11 +178,10 @@ function elpa_solve_evp_&
      return
    endif
 
-   if (elpa_print_times) then
-     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
+   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")
+
    call elpa_deallocate(e)
 
    call elpa_uninit()

src/elpa2/compute_hh_trafo.X90

@@ -65,12 +65,7 @@
 #else
        last_stripe_width, &
 #endif
-#if REALCASE == 1
-       THIS_REAL_ELPA_KERNEL)
-#endif
-#if COMPLEXCASE == 1
-       THIS_COMPLEX_ELPA_KERNEL)
-#endif
+       kernel)
 
          use precision
          use elpa_api
@@ -154,12 +149,7 @@
 
 #endif /* WITH_OPENMP */
 
-#if REALCASE == 1
-         integer(kind=ik), intent(in)       :: THIS_REAL_ELPA_KERNEL
-#endif
-#if COMPLEXCASE ==1
-         integer(kind=ik), intent(in)       :: THIS_COMPLEX_ELPA_KERNEL
-#endif
+         integer(kind=ik), intent(in)       :: kernel
 
          integer(kind=c_intptr_t)           :: a_dev
 	 integer(kind=c_intptr_t)           :: bcast_buffer_dev
@@ -184,13 +174,13 @@
          real(kind=c_double)                :: ttt ! MPI_WTIME always needs double
 
 #if REALCASE == 1
-         if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GPU) then
+         if (kernel .eq. ELPA_2STAGE_REAL_GPU) then
            ! ncols - indicates the number of HH reflectors to apply; at least 1 must be available
            if (ncols < 1) return
          endif
 #endif
 #if COMPLEXCASE == 1
-         if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GPU) then
+         if (kernel .eq. ELPA_2STAGE_COMPLEX_GPU) then
            ! ncols - indicates the number of HH reflectors to apply; at least 1 must be available
            if (ncols < 1) return
          endif
@@ -221,7 +211,7 @@
 #else /* WITH_OPENMP */
 
 #if REALCASE == 1
-         if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GPU) then
+         if (kernel .eq. ELPA_2STAGE_REAL_GPU) then
            print *,"compute_hh_trafo_&
 	   &MATH_DATATYPE&
 	   &_GPU OPENMP: not yet implemented"
@@ -229,7 +219,7 @@
          endif
 #endif
 #if COMPLEXCASE == 1
-         if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GPU) then
+         if (kernel .eq. ELPA_2STAGE_COMPLEX_GPU) then
            print *,"compute_hh_trafo_&
 	   &MATH_DATATYPE&
 	   &_GPU OPENMP: not yet implemented"
@@ -259,7 +249,7 @@
 
 #if REALCASE == 1
 ! GPU kernel real
-         if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GPU) then
+         if (kernel .eq. ELPA_2STAGE_REAL_GPU) then
            dev_offset = (0 + (a_off * stripe_width) + ( (istripe - 1) * stripe_width *a_dim2 )) *size_of_&
 	                &PRECISION&
 			&_&
@@ -272,7 +262,7 @@
 #endif /* REALCASE */
 #if COMPLEXCASE == 1
 ! GPU kernel complex
-         if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GPU) then
+         if (kernel .eq. ELPA_2STAGE_COMPLEX_GPU) then
 
            dev_offset = (0 + ( (  a_off + off-1 )* stripe_width) + ( (istripe - 1)*stripe_width*a_dim2 )) * size_of_&
 	                &PRECISION&
@@ -303,15 +293,15 @@
 
 #if REALCASE == 1
 #ifndef WITH_FIXED_REAL_KERNEL
-         if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX_BLOCK2 .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX2_BLOCK2 .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX512_BLOCK2 .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_SSE_BLOCK2 .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GENERIC    .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GENERIC_SIMPLE .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_SSE_ASSEMBLY .or. &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_BGP .or.        &
-             THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_BGQ) then
+         if (kernel .eq. ELPA_2STAGE_REAL_AVX_BLOCK2 .or. &
+             kernel .eq. ELPA_2STAGE_REAL_AVX2_BLOCK2 .or. &
+             kernel .eq. ELPA_2STAGE_REAL_AVX512_BLOCK2 .or. &
+             kernel .eq. ELPA_2STAGE_REAL_SSE_BLOCK2 .or. &
+             kernel .eq. ELPA_2STAGE_REAL_GENERIC    .or. &
+             kernel .eq. ELPA_2STAGE_REAL_GENERIC_SIMPLE .or. &
+             kernel .eq. ELPA_2STAGE_REAL_SSE_ASSEMBLY .or. &
+             kernel .eq. ELPA_2STAGE_REAL_BGP .or.        &
+             kernel .eq. ELPA_2STAGE_REAL_BGQ) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #endif /* REALCASE */
@@ -323,7 +313,7 @@
 ! generic kernel real case
 #if defined(WITH_REAL_GENERIC_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-             if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GENERIC) then
+             if (kernel .eq. ELPA_2STAGE_REAL_GENERIC) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
                do j = ncols, 2, -2
@@ -379,9 +369,9 @@
 ! generic kernel complex case
 #if defined(WITH_COMPLEX_GENERIC_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GENERIC .or. &
-               THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_BGP .or. &
-               THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_BGQ ) then
+           if (kernel .eq. ELPA_2STAGE_COMPLEX_GENERIC .or. &
+               kernel .eq. ELPA_2STAGE_COMPLEX_BGP .or. &
+               kernel .eq. ELPA_2STAGE_COMPLEX_BGQ ) then
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
              ttt = mpi_wtime()
              do j = ncols, 1, -1
@@ -420,7 +410,7 @@
 
             enddo
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GENERIC .or. THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_BGP .or. THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_BGQ )
+          endif ! (kernel .eq. ELPA_2STAGE_COMPLEX_GENERIC .or. kernel .eq. ELPA_2STAGE_COMPLEX_BGP .or. kernel .eq. ELPA_2STAGE_COMPLEX_BGQ )
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_GENERIC_KERNEL */
 
@@ -430,7 +420,7 @@
 ! generic simple real kernel
 #if defined(WITH_REAL_GENERIC_SIMPLE_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-             if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_GENERIC_SIMPLE) then
+             if (kernel .eq. ELPA_2STAGE_REAL_GENERIC_SIMPLE) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
                do j = ncols, 2, -2
                  w(:,1) = bcast_buffer(1:nbw,j+off)
@@ -484,7 +474,7 @@
 
 #if defined(WITH_COMPLEX_GENERIC_SIMPLE_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-            if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE) then
+            if (kernel .eq. ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE) then
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
              ttt = mpi_wtime()
              do j = ncols, 1, -1
@@ -522,7 +512,7 @@
 #endif /* WITH_OPENMP */
              enddo
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE)
+           endif ! (kernel .eq. ELPA_2STAGE_COMPLEX_GENERIC_SIMPLE)
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_GENERIC_SIMPLE_KERNEL */
 #endif /* COMPLEXCASE */
@@ -531,7 +521,7 @@
 ! sse assembly kernel real case
 #if defined(WITH_REAL_SSE_ASSEMBLY_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-             if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_SSE_ASSEMBLY) then
+             if (kernel .eq. ELPA_2STAGE_REAL_SSE_ASSEMBLY) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
                do j = ncols, 2, -2
                  w(:,1) = bcast_buffer(1:nbw,j+off)
@@ -563,7 +553,7 @@
 ! sse assembly kernel complex case
 #if defined(WITH_COMPLEX_SSE_ASSEMBLY_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_SSE_ASSEMBLY) then
+           if (kernel .eq. ELPA_2STAGE_COMPLEX_SSE_ASSEMBLY) then
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
              ttt = mpi_wtime()
              do j = ncols, 1, -1
@@ -584,7 +574,7 @@
 #endif
             enddo
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_SSE)
+          endif ! (kernel .eq. ELPA_2STAGE_COMPLEX_SSE)
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_SSE_ASSEMBLY_KERNEL */
 #endif /* COMPLEXCASE */
@@ -597,7 +587,7 @@
 ! sse block1 complex kernel
 #if defined(WITH_COMPLEX_SSE_BLOCK1_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK1) then
+          if (kernel .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK1) then
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 
 #if (!defined(WITH_FIXED_COMPLEX_KERNEL)) || (defined(WITH_FIXED_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_SSE_BLOCK2_KERNEL))
@@ -620,7 +610,7 @@
 #endif /* (!defined(WITH_FIXED_COMPLEX_KERNEL)) || (defined(WITH_FIXED_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_SSE_BLOCK2_KERNEL)) */
 
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK1)
+          endif ! (kernel .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK1)
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_SSE_BLOCK1_KERNEL */
 
@@ -634,8 +624,8 @@
 ! avx block1 complex kernel
 #if defined(WITH_COMPLEX_AVX_BLOCK1_KERNEL) || defined(WITH_COMPLEX_AVX2_BLOCK1_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          if ((THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK1) .or. &
-              (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK1)) then
+          if ((kernel .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK1) .or. &
+              (kernel .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK1)) then
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 
 #if (!defined(WITH_FIXED_COMPLEX_KERNEL)) || (defined(WITH_FIXED_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX_BLOCK2_KERNEL) && !defined(WITH_COMPLEX_AVX2_BLOCK2_KERNEL))
@@ -658,7 +648,7 @@
 #endif /* (!defined(WITH_FIXED_COMPLEX_KERNEL)) || (defined(WITH_FIXED_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX_BLOCK2_KERNEL) && !defined(WITH_COMPLEX_AVX2_BLOCK2_KERNEL)) */
 
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          endif ! ((THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK1) .or. (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK1))
+          endif ! ((kernel .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK1) .or. (kernel .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK1))
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_AVX_BLOCK1_KERNEL || WITH_COMPLEX_AVX2_BLOCK1_KERNEL */
 
@@ -672,7 +662,7 @@
 ! avx512 block1 complex kernel
 #if defined(WITH_COMPLEX_AVX512_BLOCK1_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          if ((THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK1)) then
+          if ((kernel .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK1)) then
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 
 #if (!defined(WITH_FIXED_COMPLEX_KERNEL)) || (defined(WITH_FIXED_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL) )
@@ -695,7 +685,7 @@
 #endif /* (!defined(WITH_FIXED_COMPLEX_KERNEL)) || (defined(WITH_FIXED_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL) ) */
 
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-          endif ! ((THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK1))
+          endif ! ((kernel .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK1))
 #endif /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_AVX512_BLOCK1_KERNEL  */
 #endif /* COMPLEXCASE */
@@ -705,7 +695,7 @@
 #if defined(WITH_REAL_SSE_BLOCK2_KERNEL)
 
 #ifndef WITH_FIXED_REAL_KERNEL
-           if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_SSE_BLOCK2) then
+           if (kernel .eq. ELPA_2STAGE_REAL_SSE_BLOCK2) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #if (!defined(WITH_FIXED_REAL_KERNEL)) || (defined(WITH_FIXED_REAL_KERNEL) && !defined(WITH_REAL_SSE_BLOCK6_KERNEL) && !defined(WITH_REAL_SSE_BLOCK4_KERNEL))
@@ -740,7 +730,7 @@
 
 #if defined(WITH_COMPLEX_SSE_BLOCK2_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           if (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK2) then
+           if (kernel .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK2) then
 #endif  /* not WITH_FIXED_COMPLEX_KERNEL */
 
              ttt = mpi_wtime()
@@ -776,7 +766,7 @@
 #endif
 
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK2)
+           endif ! (kernel .eq. ELPA_2STAGE_COMPLEX_SSE_BLOCK2)
 #endif  /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_SSE_BLOCK2_KERNEL */
 #endif /* COMPLEXCASE == 1 */
@@ -787,8 +777,8 @@
 #if defined(WITH_REAL_AVX_BLOCK2_KERNEL) || defined(WITH_REAL_AVX2_BLOCK2_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
 
-           if ((THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX_BLOCK2) .or. &
-               (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX2_BLOCK2))  then
+           if ((kernel .eq. ELPA_2STAGE_REAL_AVX_BLOCK2) .or. &
+               (kernel .eq. ELPA_2STAGE_REAL_AVX2_BLOCK2))  then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #if (!defined(WITH_FIXED_REAL_KERNEL)) || (defined(WITH_FIXED_REAL_KERNEL) && !defined(WITH_REAL_AVX_BLOCK6_KERNEL) && !defined(WITH_REAL_AVX_BLOCK4_KERNEL) && !defined(WITH_REAL_AVX2_BLOCK6_KERNEL) && !defined(WITH_REAL_AVX2_BLOCK4_KERNEL))
@@ -823,8 +813,8 @@
 ! implementation of avx block 2 complex case
 #if defined(WITH_COMPLEX_AVX_BLOCK2_KERNEL) || defined(WITH_COMPLEX_AVX2_BLOCK2_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           if ( (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK2) .or. &
-                (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK2) ) then
+           if ( (kernel .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK2) .or. &
+                (kernel .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK2) ) then
 #endif  /* not WITH_FIXED_COMPLEX_KERNEL */
 
               ttt = mpi_wtime()
@@ -860,7 +850,7 @@
 #endif
 
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           endif ! ( (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK2) .or. (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK2) )
+           endif ! ( (kernel .eq. ELPA_2STAGE_COMPLEX_AVX_BLOCK2) .or. (kernel .eq. ELPA_2STAGE_COMPLEX_AVX2_BLOCK2) )
 #endif  /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_AVX_BLOCK2_KERNEL || WITH_COMPLEX_AVX2_BLOCK2_KERNEL */
 
@@ -872,7 +862,7 @@
 #if defined(WITH_REAL_AVX512_BLOCK2_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
 
-           if ((THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX512_BLOCK2)) then
+           if ((kernel .eq. ELPA_2STAGE_REAL_AVX512_BLOCK2)) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #if (!defined(WITH_FIXED_REAL_KERNEL)) || (defined(WITH_FIXED_REAL_KERNEL) && !defined(WITH_REAL_AVX512_BLOCK6_KERNEL) && !defined(WITH_REAL_AVX512_BLOCK4_KERNEL))
@@ -907,7 +897,7 @@
 ! implementation of avx512 block 2 complex case
 #if defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL)
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           if ( (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK2)) then
+           if ( (kernel .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK2)) then
 #endif  /* not WITH_FIXED_COMPLEX_KERNEL */
 
               ttt = mpi_wtime()
@@ -943,7 +933,7 @@
 #endif
 
 #ifndef WITH_FIXED_COMPLEX_KERNEL
-           endif ! ( (THIS_COMPLEX_ELPA_KERNEL .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK2))
+           endif ! ( (kernel .eq. ELPA_2STAGE_COMPLEX_AVX512_BLOCK2))
 #endif  /* not WITH_FIXED_COMPLEX_KERNEL */
 #endif /* WITH_COMPLEX_AVX512_BLOCK2_KERNEL */
 #endif /* COMPLEXCASE */
@@ -953,7 +943,7 @@
 
 #if defined(WITH_REAL_BGP_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-             if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_BGP) then
+             if (kernel .eq. ELPA_2STAGE_REAL_BGP) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
                do j = ncols, 2, -2
                  w(:,1) = bcast_buffer(1:nbw,j+off)
@@ -975,7 +965,7 @@
 
 #if defined(WITH_REAL_BGQ_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-             if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_BGQ) then
+             if (kernel .eq. ELPA_2STAGE_REAL_BGQ) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
                do j = ncols, 2, -2
                  w(:,1) = bcast_buffer(1:nbw,j+off)
@@ -1032,7 +1022,7 @@
 
 #if defined(WITH_REAL_SSE_BLOCK4_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-           if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_SSE_BLOCK4) then
+           if (kernel .eq. ELPA_2STAGE_REAL_SSE_BLOCK4) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #if (!defined(WITH_FIXED_REAL_KERNEL)) || (defined(WITH_FIXED_REAL_KERNEL) && !defined(WITH_REAL_SSE_BLOCK6_KERNEL))
@@ -1105,8 +1095,8 @@
 ! avx block4 real kernel
 #if defined(WITH_REAL_AVX_BLOCK4_KERNEL) || defined(WITH_REAL_AVX2_BLOCK4_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-           if ((THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX_BLOCK4) .or.  &
-               (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX2_BLOCK4)) then
+           if ((kernel .eq. ELPA_2STAGE_REAL_AVX_BLOCK4) .or.  &
+               (kernel .eq. ELPA_2STAGE_REAL_AVX2_BLOCK4)) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #if (!defined(WITH_FIXED_REAL_KERNEL)) || (defined(WITH_FIXED_REAL_KERNEL) && !defined(WITH_REAL_AVX_BLOCK6_KERNEL) && !defined(WITH_REAL_AVX2_BLOCK6_KERNEL))
@@ -1180,7 +1170,7 @@
 
 #if defined(WITH_REAL_AVX512_BLOCK4_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-           if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX512_BLOCK4) then
+           if (kernel .eq. ELPA_2STAGE_REAL_AVX512_BLOCK4) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
 
 #if (!defined(WITH_FIXED_REAL_KERNEL)) || (defined(WITH_FIXED_REAL_KERNEL) && !defined(WITH_REAL_AVX512_BLOCK6_KERNEL))
@@ -1254,7 +1244,7 @@
 !sse block6 real kernel
 #if defined(WITH_REAL_SSE_BLOCK6_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-           if (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_SSE_BLOCK6) then
+           if (kernel .eq. ELPA_2STAGE_REAL_SSE_BLOCK6) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
              ! X86 INTRINSIC CODE, USING 6 HOUSEHOLDER VECTORS
              do j = ncols, 6, -6
@@ -1345,8 +1335,8 @@
 
 #if defined(WITH_REAL_AVX_BLOCK6_KERNEL) || defined(WITH_REAL_AVX2_BLOCK6_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-           if ((THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX_BLOCK6) .or. &
-               (THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX2_BLOCK6)) then
+           if ((kernel .eq. ELPA_2STAGE_REAL_AVX_BLOCK6) .or. &
+               (kernel .eq. ELPA_2STAGE_REAL_AVX2_BLOCK6)) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
              ! X86 INTRINSIC CODE, USING 6 HOUSEHOLDER VECTORS
              do j = ncols, 6, -6
@@ -1436,7 +1426,7 @@
 ! avx512 block6 kernel
 #if defined(WITH_REAL_AVX512_BLOCK6_KERNEL)
 #ifndef WITH_FIXED_REAL_KERNEL
-           if ((THIS_REAL_ELPA_KERNEL .eq. ELPA_2STAGE_REAL_AVX512_BLOCK6)) then
+           if ((kernel .eq. ELPA_2STAGE_REAL_AVX512_BLOCK6)) then
 #endif /* not WITH_FIXED_REAL_KERNEL */
              ! X86 INTRINSIC CODE, USING 6 HOUSEHOLDER VECTORS
              do j = ncols, 6, -6

src/elpa2/elpa2.F90

@@ -109,7 +109,7 @@ module elpa2_impl
 !>  \param mpi_comm_cols                        MPI communicator for columns
 !>  \param mpi_comm_all                         MPI communicator for the total processor set
 !>
-!>  \param THIS_REAL_ELPA_KERNEL_API (optional) specify used ELPA2 kernel via API
+!>  \param kernel                               specify ELPA2 kernel to use
 !>
 !>  \param useQR (optional)                     use QR decomposition
 !>  \param useGPU (optional)                    decide whether to use GPUs or not
@@ -157,7 +157,7 @@ module elpa2_impl
 !>  \param mpi_comm_cols                        MPI communicator for columns
 !>  \param mpi_comm_all                         MPI communicator for the total processor set
 !>
-!>  \param THIS_REAL_ELPA_KERNEL_API (optional) specify used ELPA2 kernel via API
+!>  \param kernel                               specify ELPA2 kernel to use
 !>
 !>  \param useQR (optional)                     use QR decomposition
 !>  \param useGPU (optional)                    decide whether GPUs should be used or not
@@ -205,7 +205,7 @@ module elpa2_impl
 !>  \param mpi_comm_cols                        MPI communicator for columns
 !>  \param mpi_comm_all                         MPI communicator for the total processor set
 !>
-!>  \param THIS_REAL_ELPA_KERNEL_API (optional) specify used ELPA2 kernel via API
+!>  \param kernel                               specify ELPA2 kernel to use
 !>  \param useGPU (optional)                    decide whether GPUs should be used or not
 !>
 !>  \result success                             logical, false if error occured
@@ -252,8 +252,8 @@ module elpa2_impl
 !>  \param mpi_comm_cols                        MPI communicator for columns
 !>  \param mpi_comm_all                         MPI communicator for the total processor set
 !>
-!>  \param THIS_COMPLEX_ELPA_KERNEL_API (optional) specify used ELPA2 kernel via API
-!>  \param useGPU (optional)                   decide whether GPUs should be used or not
+!>  \param kernel                               specify ELPA2 kernel to use
+!>  \param useGPU (optional)                    decide whether GPUs should be used or not
 !>
 !>  \result success                             logical, false if error occured
 !-------------------------------------------------------------------------------

src/elpa2/elpa2_template.X90

@@ -55,7 +55,7 @@
   &2stage_&
   &PRECISION&
   &_impl (obj, na, nev, a, lda, ev, q, ldq, nblk, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm_all,   &
-         time_evp_fwd, time_evp_solve, time_evp_back, summary_timings, useGPU, kernel &
+         useGPU, kernel &
 #if REALCASE == 1
    , useQR   &
 #endif
@@ -106,10 +106,7 @@
 #if COMPLEXCASE == 1
    real(kind=C_DATATYPE_KIND), allocatable   :: q_real(:,:)
 #endif
-   real(kind=c_double)                       :: time_evp_fwd, time_evp_solve, time_evp_back
-   logical, intent(in)                       :: summary_timings
    integer(kind=c_intptr_t)                  :: tmat_dev, q_dev, a_dev
-   real(kind=c_double)                        :: ttt0, ttt1, ttts  ! MPI_WTIME always needs double
 
    integer(kind=c_int)                       :: i
    logical                                   :: success, successCUDA
@@ -225,8 +222,6 @@
         success = .false.
         return
       endif
-
-      ttts = MPI_Wtime()
     else
 
       ! Choose bandwidth, must be a multiple of nblk, set to a value >= 32
@@ -257,9 +252,6 @@
       endif
 
       ! Reduction full -> band
-
-      ttt0 = MPI_Wtime()
-      ttts = ttt0
       call bandred_&
       &MATH_DATATYPE&
       &_&
@@ -272,13 +264,6 @@
 #endif
        )
       if (.not.(success)) return
-      ttt1 = MPI_Wtime()
-      if (my_prow==0 .and. my_pcol==0 .and. summary_timings) &
-        write(error_unit,*) "Time " // "bandred_&
-	&MATH_DATATYPE&
-	&_&
-	&PRECISION " // "               :",ttt1-ttt0
-
     end if  ! matrix not already banded on input
 
      ! Reduction band -> tridiagonal
@@ -292,29 +277,18 @@
        stop 1
      endif
 
-     ttt0 = MPI_Wtime()
-
      call tridiag_band_&
      &MATH_DATATYPE&
      &_&
      &PRECISION&
      (na, nbw, nblk, a, a_dev, lda, ev, e, matrixCols, hh_trans, mpi_comm_rows, mpi_comm_cols, mpi_comm_all, do_useGPU)
 
-     ttt1 = MPI_Wtime()
-     if (my_prow==0 .and. my_pcol==0 .and. summary_timings) &
-       write(error_unit,*) "Time " // "tridiag_band_&
-       &MATH_DATATYPE&
-       &_&
-       &PRECISION " // "          :",ttt1-ttt0
-
 #ifdef WITH_MPI
      call timer%start("mpi_communication")
      call mpi_bcast(ev, na, MPI_REAL_PRECISION, 0, mpi_comm_all, mpierr)
      call mpi_bcast(e, na, MPI_REAL_PRECISION, 0, mpi_comm_all, mpierr)
      call timer%stop("mpi_communication")
 #endif /* WITH_MPI */
-     ttt1 = MPI_Wtime()
-     time_evp_fwd = ttt1-ttts
 
 #if COMPLEXCASE == 1
      l_rows = local_index(na, my_prow, np_rows, nblk, -1) ! Local rows of a and q
@@ -331,8 +305,6 @@
 #endif
 
      ! Solve tridiagonal system
-
-     ttt0 = MPI_Wtime()
      call solve_tridi_&
      &PRECISION &
      (na, nev, ev, e, &
@@ -346,12 +318,6 @@
 
      if (.not.(success)) return
 
-     ttt1 = MPI_Wtime()
-     if (my_prow==0 .and. my_pcol==0 .and. summary_timings) &
-     write(error_unit,*) 'Time solve_tridi                :',ttt1-ttt0
-     time_evp_solve = ttt1-ttt0
-     ttts = ttt1
-
      deallocate(e, stat=istat, errmsg=errorMessage)
      if (istat .ne. 0) then
        print *,"solve_evp_&
@@ -372,9 +338,6 @@
      endif
 #endif