Pass a_dev to elpa2_tridiag

src/elpa2_template.X90

@@ -323,7 +323,7 @@
      &MATH_DATATYPE&
      &_&
      &PRECISION&
-     (na, nbw, nblk, a, lda, ev, e, matrixCols, hh_trans, mpi_comm_rows, mpi_comm_cols, mpi_comm_all)
+     (na, nbw, nblk, a, a_dev, lda, ev, e, matrixCols, hh_trans, mpi_comm_rows, mpi_comm_cols, mpi_comm_all)
 
      ttt1 = MPI_Wtime()
      if (my_prow==0 .and. my_pcol==0 .and. elpa_print_times) &

src/elpa2_trans_ev_band_to_full_template.X90

@@ -55,8 +55,7 @@
     &_&
     &PRECISION &
     (na, nqc, nblk, nbw, a, a_dev, lda, tmat, tmat_dev, q, &
-    q_dev,                  &
-     ldq, matrixCols, numBlocks, mpi_comm_rows, mpi_comm_cols, useGPU &
+     q_dev, ldq, matrixCols, numBlocks, mpi_comm_rows, mpi_comm_cols, useGPU &
 #if REALCASE == 1
      ,useQr)
 #endif

src/elpa2_tridiag_band_template.X90

@@ -56,7 +56,7 @@
     subroutine tridiag_band_complex_&
 #endif
     &PRECISION &
-    (na, nb, nblk, a, lda, d, e, matrixCols, &
+    (na, nb, nblk, aMatrix, a_dev, lda, d, e, matrixCols, &
 #if REALCASE == 1
     hh_trans_real, &
 #endif
@@ -74,7 +74,7 @@
     !
     !  nblk        blocksize of cyclic distribution, must be the same in both directions!
     !
-    !  a(lda,matrixCols)    Distributed system matrix reduced to banded form in the upper diagonal
+    !  aMatrix(lda,matrixCols)    Distributed system matrix reduced to banded form in the upper diagonal
     !
     !  lda         Leading dimension of a
     !  matrixCols  local columns of matrix a
@@ -98,23 +98,25 @@
 #endif
       use elpa2_workload
       use precision
+      use iso_c_binding
       implicit none
 
       integer(kind=ik), intent(in)                 ::  na, nb, nblk, lda, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm
 #if REALCASE == 1
 #ifdef USE_ASSUMED_SIZE
-      real(kind=REAL_DATATYPE), intent(in)         :: a(lda,*)
+      real(kind=REAL_DATATYPE), intent(in)         :: aMatrix(lda,*)
 #else
-      real(kind=REAL_DATATYPE), intent(in)         :: a(lda,matrixCols)
+      real(kind=REAL_DATATYPE), intent(in)         :: aMatrix(lda,matrixCols)
 #endif
 #endif /* REALCASE */
 #if COMPLEXCASE == 1
 #ifdef USE_ASSUMED_SIZE
-      complex(kind=COMPLEX_DATATYPE),intent(in)    :: a(lda,*)
+      complex(kind=COMPLEX_DATATYPE),intent(in)    :: aMatrix(lda,*)
 #else
-      complex(kind=COMPLEX_DATATYPE), intent(in)   :: a(lda,matrixCols)
+      complex(kind=COMPLEX_DATATYPE), intent(in)   :: aMatrix(lda,matrixCols)
 #endif
 #endif /* COMPLEXCASE */
+      integer(kind=c_intptr_t)                     :: a_dev
       real(kind=REAL_DATATYPE), intent(out)        :: d(na), e(na) ! set only on PE 0
 #if REALCASE == 1
       real(kind=REAL_DATATYPE), intent(out), &
@@ -267,7 +269,7 @@
       &MATH_DATATYPE&
       &_&
       &PRECISION&
-      &(a, lda, na, nblk, nb, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, ab)
+      &(aMatrix, a_dev, lda, na, nblk, nb, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, ab)
 
       ! Calculate the workload for each sweep in the back transformation
       ! and the space requirements to hold the HH vectors

src/redist_band.X90

@@ -47,26 +47,6 @@
 
 #include "config-f90.h"
 
-!#if REALCASE==1
-!
-!#ifdef DOUBLE_PRECISION_REAL
-!subroutine redist_band_real_double(lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, r_ab)
-!#else
-!subroutine redist_band_real_single(lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, r_ab)
-!#endif
-!
-!#endif /* REALCASE == 1 */
-!
-!#if COMPLEXCASE==1
-!
-!#ifdef DOUBLE_PRECISION_COMPLEX
-!subroutine redist_band_complex_double(lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, c_ab)
-!#else
-!subroutine redist_band_complex_single(lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, c_ab)
-!#endif
-!
-!#endif /* COMPLEXCASE == 1 */
-
 subroutine redist_band_&
 &MATH_DATATYPE&
 &_&
@@ -78,7 +58,7 @@ subroutine redist_band_&
 #if COMPLEXCASE == 1
             c_a, &
 #endif
-            lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, &
+            a_dev, lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm, &
 #if REALCASE == 1
             r_ab)
 #endif
@@ -93,22 +73,23 @@ subroutine redist_band_&
 #endif
    use elpa2_workload
    use precision
+   use iso_c_binding
    implicit none
 
-   integer(kind=ik), intent(in)     :: lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm
+   integer(kind=ik), intent(in)                :: lda, na, nblk, nbw, matrixCols, mpi_comm_rows, mpi_comm_cols, mpi_comm
 #if REALCASE == 1
    real(kind=REAL_DATATYPE), intent(in)        :: r_a(lda, matrixCols)
 #endif
 #if COMPLEXCASE == 1
-   complex(kind=COMPLEX_DATATYPE), intent(in)     :: c_a(lda, matrixCols)
+   complex(kind=COMPLEX_DATATYPE), intent(in)  :: c_a(lda, matrixCols)
 #endif
 
 #if REALCASE == 1
    real(kind=REAL_DATATYPE), intent(out)       :: r_ab(:,:)
 #endif
-
+   integer(kind=c_intptr_t)                    :: a_dev
 #if COMPLEXCASE == 1
-   complex(kind=COMPLEX_DATATYPE), intent(out)    :: c_ab(:,:)
+   complex(kind=COMPLEX_DATATYPE), intent(out) :: c_ab(:,:)
 #endif
 
    integer(kind=ik), allocatable    :: ncnt_s(:), nstart_s(:), ncnt_r(:), nstart_r(:), &