OpenACC與CUDA Fortran交互(2)

在帖子《OpenACC與CUDA Fortran交互(1)》中，舉了一個在openacc中嵌套cuda fortran的例子。現在舉一個cuda fortran中嵌套openacc的例子。上代碼：

1. 
   
2. ! cuf_main.cuf
   
3. program main
   
4.   integer, parameter :: N = 2**20
   
5.   ! Allocate X and Y only on the device
   
6.   real, device, dimension(N) :: X, Y
   
7.   integer :: i
   
8.   real :: tmp
   
9. 
   
10.   ! CUDA Fortran will automatically convert these to run on the device
    
11.   X(:) = 1.0
    
12.   Y(:) = 0.0
    
13. 
    
14.   !$acc kernels deviceptr(x,y)
    
15.   y(:) = y(:) + 2.0*x(:)
    
16.   !$acc end kernels
    
17. 
    
18.   ! Copy the first element back from Y for correctness checking
    
19.   tmp = y(1)
    
20.   print *, tmp
    
21. end program
    
22. 
    

編譯：

1. pgf90  -o cuda_fortran_openacc cuf_main.cuf -acc -Mcuda=cc3.5 -Minfo