Deleted manual CUDA acceleration code that is no longer useful.

Author: Damian Rouson

src/tests/integration/gpu/co_dot.f90

@@ -5,23 +5,13 @@ module accelerated_module
   private
   public :: co_dot_accelerated
   public :: co_dot_unaccelerated
-  public :: co_dot_manually_accelerated
   public :: co_dot_mapped_manually_accelerated
   public :: CUDA,OpenACC,OpenMP
   public :: walltime
 
   ! Explicit interfaces for procedures that wrap accelerated kernels
   interface  
 
-     ! This is the wrapper a programmer would have to write today to manually accelerate calculations
-     subroutine manual_cudaDot(a,b,partial_dot,n,img) bind(C, name="manual_cudaDot")
-       use iso_c_binding, only : c_float,c_int
-       real(c_float) :: a(*),b(*)
-       real(c_float) :: partial_dot
-       integer(c_int),value :: n
-       integer(c_int),value :: img
-     end subroutine
-
      subroutine manual_mapped_cudaDot(a,b,partial_dot,n,img) bind(C, name="manual_mapped_cudaDot")
        use iso_c_binding, only : c_float,c_int
        real(c_float) :: a(*),b(*)
@@ -59,14 +49,6 @@ subroutine co_dot_unaccelerated(x,y,x_dot_y)
      call co_sum(x_dot_y) ! Call Fortarn 2015 collective sum
   end subroutine 
 
-  ! This parallel collective dot product uses manual acceleration
-  subroutine co_dot_manually_accelerated(x,y,x_dot_y)
-     real(c_float), intent(in) :: x(:),y(:)
-     real(c_float), intent(out) :: x_dot_y
-     call manual_cudaDot(x,y,x_dot_y,size(x),this_image()-1) 
-     call co_sum(x_dot_y) ! Call Fortarn 2015 collective sum
-  end subroutine 
-
   subroutine co_dot_mapped_manually_accelerated(x,y,x_dot_y)
      real(c_float), intent(in) :: x(:),y(:)
      real(c_float), intent(out) :: x_dot_y
@@ -106,7 +88,7 @@ program cu_dot_test
   real(c_float) :: dot
   real(c_double) :: t_start, t_end
 
-  ! Compiler/library-accelerated variables
+  ! Library-accelerated variables
   real(c_float), allocatable :: a_acc(:)[:], b_acc(:)[:]
   real(c_float) :: dot_acc[*]
 
@@ -127,7 +109,7 @@ program cu_dot_test
   sync all
 
   block 
-!    use accelerated_module, only : co_dot_accelerated,co_dot_unaccelerated,co_dot_manually_accelerated,CUDA,walltime,co_dot_mapped_manually_accelerated
+!    use accelerated_module, only : co_dot_accelerated,co_dot_unaccelerated,CUDA,walltime,co_dot_mapped_manually_accelerated
     use accelerated_module
 
     !Parallel execution
@@ -138,13 +120,6 @@ program cu_dot_test
 
     sync all
 
-    t_start = walltime()
-    call co_dot_manually_accelerated(a_man,b_man,dot_man)
-    t_end = walltime()
-    if(me==1) print *, 'Manually accelerated dot_prod',dot_man,'time:',t_end-t_start
-
-    sync all
-
     !Serial execution
     t_start = walltime()
     call co_dot_unaccelerated(a_man,b_man,dot)