Merge branch 'master_pre_stage' of https://gitlab.mpcdf.mpg.de/elpa/elpa into master_pre_stage

src/GPU/check_for_gpu.F90

@@ -62,6 +62,19 @@ module mod_check_for_gpu
 
       gpuAvailable = .false.
 
+      if(cublasHandle .ne. -1) then
+	gpuAvailable = .true.
+	numberOfDevices = -1
+	if(myid == 0) then
+          print *, "Skipping GPU init, should have already been initialized "
+	endif
+	return
+      else
+	if(myid == 0) then
+	  print *, "Initializing the GPU devices"
+	endif
+      endif
+
       if (.not.(present(wantDebug))) then
         wantDebugMessage = .false.
       else

src/GPU/mod_cuda.F90

@@ -55,7 +55,8 @@ module cuda_functions
   integer(kind=ik) :: cudaHostRegisterMapped
   integer(kind=ik) :: cudaMemcpyDeviceToDevice
 
-  integer(kind=C_intptr_T) :: cublasHandle
+  ! TODO global variable, has to be changed
+  integer(kind=C_intptr_T) :: cublasHandle = -1
 
   integer(kind=c_intptr_t), parameter :: size_of_double_real    = 8_rk8
 #ifdef WANT_SINGLE_PRECISION_REAL

src/elpa1/elpa1_template.F90

@@ -220,10 +220,10 @@ function elpa_solve_evp_&
    wantDebug = debug == 1
    do_useGPU = .false.
 
-
+   
    if (useGPU) then
+     call obj%timer%start("check_for_gpu")
      if (check_for_gpu(my_pe,numberOfGPUDevices, wantDebug=wantDebug)) then
-
        do_useGPU = .true.
        ! set the neccessary parameters
        cudaMemcpyHostToDevice   = cuda_memcpyHostToDevice()
@@ -236,8 +236,10 @@ function elpa_solve_evp_&
        success = .false.
        return
      endif
+     call obj%timer%stop("check_for_gpu")
    endif
 
+
    do_useGPU_tridiag = do_useGPU
    do_useGPU_solve_tridi = do_useGPU
    do_useGPU_trans_ev = do_useGPU

src/elpa2/elpa2_template.F90

@@ -239,6 +239,7 @@
 
     do_useGPU = .false.
     if (useGPU) then
+      call obj%timer%start("check_for_gpu")
       if (check_for_gpu(my_pe,numberOfGPUDevices, wantDebug=wantDebug)) then
 
          do_useGPU = .true.
@@ -254,6 +255,7 @@
         success = .false.
         return
       endif
+      call obj%timer%stop("check_for_gpu")
     endif
 
     do_useGPU_bandred = do_useGPU