Fix erroneous reference to MPI_COMM_WORLD in checks for GPU

src/GPU/check_for_gpu.F90

@@ -48,18 +48,21 @@ module mod_check_for_gpu
 
   contains
 
-    function check_for_gpu(myid, numberOfDevices, wantDebug) result(gpuAvailable)
+    function check_for_gpu(obj, myid, numberOfDevices, wantDebug) result(gpuAvailable)
       use cuda_functions
       use precision
       use elpa_mpi
+      use elpa_abstract_impl
       implicit none
 
+      class(elpa_abstract_impl_t), intent(inout)                         :: obj
       integer(kind=ik), intent(in)  :: myid
       logical, optional, intent(in) :: wantDebug
       logical                       :: success, wantDebugMessage
       integer(kind=ik), intent(out) :: numberOfDevices
       integer(kind=ik)              :: deviceNumber, mpierr, maxNumberOfDevices
       logical                       :: gpuAvailable
+      integer(kind=ik)              :: error, mpi_comm_all
       !character(len=1024)           :: envname
 
       if (.not.(present(wantDebug))) then
@@ -74,6 +77,12 @@ module mod_check_for_gpu
 
       gpuAvailable = .false.
 
+      call obj%get("mpi_comm_parent",mpi_comm_all,error)
+      if (error .ne. ELPA_OK) then
+        print *,"Problem getting option for mpi_comm_parent. Aborting..."
+        stop
+      endif
+
       if (cublasHandle .ne. -1) then
         gpuAvailable = .true.
         numberOfDevices = -1
@@ -98,7 +107,7 @@ module mod_check_for_gpu
       ! make sure that all nodes have the same number of GPU's, otherwise
       ! we run into loadbalancing trouble
 #ifdef WITH_MPI
-      call mpi_allreduce(numberOfDevices, maxNumberOfDevices, 1, MPI_INTEGER, MPI_MAX, MPI_COMM_WORLD, mpierr)
+      call mpi_allreduce(numberOfDevices, maxNumberOfDevices, 1, MPI_INTEGER, MPI_MAX, mpi_comm_all, mpierr)
 
       if (maxNumberOfDevices .ne. numberOfDevices) then
         print *,"Different number of GPU devices on MPI tasks!"

src/elpa1/elpa1_template.F90

@@ -329,7 +329,7 @@ function elpa_solve_evp_&
    if (useGPU) then
      call obj%timer%start("check_for_gpu")
 
-     if (check_for_gpu(my_pe,numberOfGPUDevices, wantDebug=wantDebug)) then
+     if (check_for_gpu(obj, my_pe, numberOfGPUDevices, wantDebug=wantDebug)) then
        do_useGPU = .true.
        ! set the neccessary parameters
        cudaMemcpyHostToDevice   = cuda_memcpyHostToDevice()

src/elpa1/elpa_multiply_a_b.F90

@@ -176,7 +176,7 @@
 
   if (useGPU) then
     call obj%timer%start("check_for_gpu")
-    if (check_for_gpu(myid,numGPU)) then
+    if (check_for_gpu(obj, myid, numGPU)) then
       ! set the neccessary parameters
       cudaMemcpyHostToDevice   = cuda_memcpyHostToDevice()
       cudaMemcpyDeviceToHost   = cuda_memcpyDeviceToHost()

src/elpa2/elpa2_template.F90

@@ -360,7 +360,7 @@
     do_useGPU = .false.
     if (useGPU) then
       call obj%timer%start("check_for_gpu")
-      if (check_for_gpu(my_pe,numberOfGPUDevices, wantDebug=wantDebug)) then
+      if (check_for_gpu(obj, my_pe, numberOfGPUDevices, wantDebug=wantDebug)) then
 
          do_useGPU = .true.