OpenACC unsupported operation: ARDF

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi,

The following sample program fails to compile with message: unsupported instruction: ARDF when using the function version of it(sm). The subroutine version (sm1) works fine though

program test
   implicit none
   real, dimension(30,30,30) :: xx
   real, dimension(30,30) :: val 
!$acc declare create(xx,val)
   real:: v
   integer:: i,j 

!$acc kernels
   xx(:,:,:) = 1.0 
   val(:,:) = 0.0 
!$acc end kernels

!$acc parallel loop collapse(2) independent
   do i = 1,30
     do j = 1,30
       val(i,j) = sm(xx(i,1:20,j))
     enddo
   enddo
!$acc end parallel

!!$acc parallel loop collapse(2) independent
!   do i = 1,30
!     do j = 1,30
!       call sm1(xx(i,1:20,j),val(i,j))
!     enddo
!   enddo
!!$acc end parallel

!$acc update host(val)
print*, val 
contains

  function sm(xx)
!$acc routine vector
     implicit none
     real,intent(in):: xx(20)
     real,intent(out)::sm

     real:: sum 
     integer k

     sum = 0.0 
     do k = 1, 20
       sum = sum + xx(k)
     enddo
     sm = sum 
   end function sm

  subroutine sm1(xx,v)
!$acc routine vector
     implicit none
     real,intent(in):: xx(20)
     real,intent(out)::v

     real:: sum
     integer k

     sum = 0.0
     do k = 1, 20
       sum = sum + xx(k)
     enddo
     v = sum
   end subroutine
end program test

Error compiling

$ pgf90 -acc -Minfo=accel -ta=tesla,cc60,cc70,cuda9.0 oo.F90 
PGF90-S-0155-Accelerator region ignored; see -Minfo messages  (oo.F90: 14)
test:
      5, Generating create(xx(:,:,:),val(:,:))
     10, Loop is parallelizable
         Accelerator kernel generated
         Generating Tesla code
         10, !$acc loop gang, vector(4) ! blockidx%z threadidx%y
             !$acc loop gang, vector(32) ! blockidx%x threadidx%x
             !$acc loop gang ! blockidx%y
     11, Loop is parallelizable
         Accelerator kernel generated
         Generating Tesla code
         11, !$acc loop gang, vector(4) ! blockidx%y threadidx%y
             !$acc loop gang, vector(32) ! blockidx%x threadidx%x
     14, Accelerator region ignored
     16, Accelerator restriction: an unsupported operation was found
     17, Accelerator restriction: unsupported operation: ARDF
  0 inform,   0 warnings,   1 severes, 0 fatal for test
sm:
     33, Generating Tesla code
         43, !$acc loop vector ! threadidx%x
     43, Loop is parallelizable
sm1:
     49, Generating Tesla code
         59, !$acc loop vector ! threadidx%x
     59, Loop is parallelizable

Daniel

2

Hi Daniel,

This is a known issue when using an OpenACC routine as a contained subroutine. The work around would to not have “sm” be contained.

Note that we have improved this support and I show that your program will compile and run correctly using the PGI 18.4 or later.

% pgf90 test.f90 -ta=tesla:cc70 -Minfo=accel -fast -V18.1
PGF90-S-0155-Accelerator region ignored; see -Minfo messages  (test.f90: 14)
test:
      5, Generating create(xx(:,:,:),val(:,:))
     10, Loop is parallelizable
         Accelerator kernel generated
         Generating Tesla code
         10, !$acc loop gang, vector(4) ! blockidx%z threadidx%y
             !$acc loop gang, vector(32) ! blockidx%x threadidx%x
             !$acc loop gang ! blockidx%y
     11, Loop is parallelizable
         Accelerator kernel generated
         Generating Tesla code
         11, !$acc loop gang, vector(4) ! blockidx%y threadidx%y
             !$acc loop gang, vector(32) ! blockidx%x threadidx%x
     14, Accelerator region ignored
     16, Accelerator restriction: an unsupported operation was found
     17, Accelerator restriction: unsupported operation: ARDF
     30, Generating update self(val(:,:))
  0 inform,   0 warnings,   1 severes, 0 fatal for test
sm:
     34, Generating Tesla code
         44, !$acc loop vector ! threadidx%x
     44, Loop is parallelizable
sm1:
     50, Generating Tesla code
         60, !$acc loop vector ! threadidx%x
     60, Loop is parallelizable
% pgf90 test.f90 -ta=tesla:cc70 -Minfo=accel -fast -V18.4
test:
      5, Generating create(val(:,:),xx(:,:,:))
     10, Loop is parallelizable
         Accelerator kernel generated
         Generating Tesla code
         10, !$acc loop gang, vector(4) ! blockidx%z threadidx%y
             !$acc loop gang, vector(32) ! blockidx%x threadidx%x
             !$acc loop gang ! blockidx%y
     11, Loop is parallelizable
         Accelerator kernel generated
         Generating Tesla code
         11, !$acc loop gang, vector(4) ! blockidx%y threadidx%y
             !$acc loop gang, vector(32) ! blockidx%x threadidx%x
     14, Accelerator kernel generated
         Generating Tesla code
         15, !$acc loop gang collapse(2) ! blockidx%x
         16,   ! blockidx%x collapsed
         17, !$acc loop vector(32) ! threadidx%x
     14, Generating implicit copy(tmp$r(:))
     17, Loop is parallelizable
     30, Generating update self(val(:,:))
sm:
     34, Generating Tesla code
         44, !$acc loop vector ! threadidx%x
             Vector barrier inserted for vector loop reduction
     44, Loop is parallelizable
sm1:
     50, Generating Tesla code
         60, !$acc loop vector ! threadidx%x
             Vector barrier inserted for vector loop reduction
     60, Loop is parallelizable
% a.out
    20.00000        20.00000        20.00000        20.00000
    20.00000        20.00000        20.00000        20.00000
    20.00000        20.00000        20.00000        20.00000
    20.00000        20.00000        20.00000        20.00000
    20.00000        20.00000        20.00000        20.00000
....

Hope this helps,
Mat