Error when using pinned memory

Accelerated Computing HPC Compilers Legacy PGI Compilers
#1

Dear NVIDIA developer，
I want to use the pinned memory in module. An example is as follows:

program test
use cudafor
use pintest

implicit none
integer::nx,ny,nz

nx=8; ny=8; nz=3
allocate(a(nx,ny,nz))
a=3

call addpin()

write(*,"(8(8i3,/))") a(:,:,1)

end program test

module pintest
    integer,allocatable,pinned::a(:,:,:)
    contains
    subroutine addpin()
        implicit none
        integer::b
        b=1
        a=a+b
    end subroutine addpin
end module pintest

Then I build it use:
nvfortran pintest.cuf test.cuf
then ./a.out
There is Segmentation fault (core dumped)

However, if I move the pinned in the module, it works well.

So could anyone let me know, how could I use pinned memory in module or in a subroutine?

Thanks a lot!

#2

Hi chen_yang,

Unfortunately I’m not sure what’s wrong since your program works fine for me with 20.11 on a x86_64 and ARM system. Is there any other information you can provide such as the compiler version, OS, CUDA driver version, and system that you’re running on?

Thanks,
Mat

% cat pintest.cuf
module pintest
    integer,allocatable,pinned::a(:,:,:)
    contains
    subroutine addpin()
        implicit none
        integer::b
        b=1
        a=a+b
    end subroutine addpin
end module pintest

% cat test.cuf
program test
use cudafor
use pintest

implicit none
integer::nx,ny,nz

nx=8; ny=8; nz=3
allocate(a(nx,ny,nz))
a=3

call addpin()

write(*,"(8(8i3,/))") a(:,:,1)

end program test

% nvfortran pintest.cuf test.cuf -V20.11 ; a.out
pintest.cuf:
test.cuf:
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
  4  4  4  4  4  4  4  4
#3

Thanks much, Mat!
Yes, it’s my fault. My computer lose the connection with GPU. Now the code can run successfully!
Thanks again!
Chen