is_present doesn't seem to work in 15.10 (was fine in 15.7)

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi again. May have found another one. The following code will error out on 15.10 but works on 15.7. line 100 is the one "!$acc kernels " one in inline_kernels (I left away some comments).

./example_gpu
hostptr=0x12adaa0,stride=1,size=655360,eltsize=4,name=a,flags=0x200=present,async=-1,threadid=1
FATAL ERROR: data in PRESENT clause was not found on device 1: name=a host:0x12adaa0
 file:/home/michel/hybrid/examples/openACC_hybrid_hostonly/build/gpu/source/example.f90 inline_kernels line:100




pgf90 -acc -Minfo=accel,inline,ipa -Mneginfo -Mcuda=cc2x,cc3x -ta=nvidia,cc2x,cc3x -fast -DGPU -c example.f90 -o example.o




module example
contains
 subroutine host_only_subroutine(a, b, c, d)
  use openacc
  use cudafor
  real, dimension(256, 256, 10), intent(in) :: a, b
  real, dimension(256, 256, 10), intent(out) :: c, d
  call inline_kernels(a, b, c, d)
end subroutine
 subroutine wrapper_inline_kernels(a, b, c, d)
  use openacc
  use cudafor
  real, intent(in) :: a(256, 256, 10), b(256, 256, 10)
  real, intent(out) :: c(256, 256, 10), d(256, 256, 10)
! ****** additional symbols inserted by framework to emulate device support of language features
  integer(4) :: hf_symbols_are_device_present
!$acc enter data copyin(a), copyin(c), copyin(b), copyin(d)
  hf_symbols_are_device_present = acc_is_present(a)
! ****** end additional symbols

  call inline_kernels (a, b, c, d)
!$acc exit data delete(a), copyout(c), delete(b), copyout(d)
end subroutine
 subroutine inline_kernels(a, b, c, d)
  use openacc
  use cudafor
  real, intent(in) :: a(256, 256, 10), b(256, 256, 10)
  real, intent(out) :: c(256, 256, 10), d(256, 256, 10)
  integer(4) :: y, x
  integer(4) :: hf_symbols_are_device_present
  hf_symbols_are_device_present = acc_is_present(a)

!$acc kernels if(hf_symbols_are_device_present) present(a) present(c) present(b) present(d)
!$acc loop independent vector(16)
  do y=1,256
!$acc loop independent vector(16)
   do x=1,256
!$acc loop seq
    do z=1,10
     c(x, y, z)= a(x, y, z)+ b(x, y, z)
    end do
   end do
  end do
!$acc end kernels
!$acc kernels if(hf_symbols_are_device_present) present(a) present(c) present(b) present(d)
!$acc loop independent vector(16)
  do y=1,256
!$acc loop independent vector(16)
   do x=1,256
!$acc loop seq
    do z=1,10
     d(x, y, z)= a(x, y, z)* b(x, y, z)
    end do
   end do
  end do
!$acc end kernels
end subroutine
end module example
program main
use example
real, dimension(256, 256, 10) :: a, b, c, d, e, f
integer :: x, y, z
integer :: fail_x, fail_y, fail_z
logical test
a(:,:,:) = 1.0d0
b(:,:,:) = 2.0d0
c(:,:,:) = 0.0d0
d(:,:,:) = 0.0d0
e(:,:,:) = 0.0d0
f(:,:,:) = 0.0d0
test = .TRUE.
call host_only_subroutine(a, b, c, d)
call wrapper_inline_kernels(c, d, e, f)
write(6,*) "calculation complete"
do y=1,256
 do x=1,256
  do z=1,10
   if (test .EQ. .TRUE. .AND. e(x, y, z) .NE. 5.0d0) then
   test = .FALSE.
   fail_x = x
   fail_y = y
   fail_z = z
   end if
   if (test .EQ. .TRUE. .AND. f(x, y, z) .NE. 6.0d0) then
   test = .FALSE.
   fail_x = x
   fail_y = y
   fail_z = z
   end if
  end do
 end do
end do
if (test .EQ. .TRUE.) then
write(6,*) "test ok"
else
write(6,*) "test failed"
write(6,*) "fails at", fail_x, fail_y, fail_z, "E:", c(fail_x, fail_y, fail_z), "F:", d(fail_x, fail_y, fail_z)
stop 2
end if
stop
end program main
2

Hi Michel,

You are correct that there was a change of behavior after we added support for Multi-core CPU, but I believe 15.10 is correct and 15.7 was in error.

At the point when you call “acc_is_present”, the device type is “host” and “a” is present on the host.

Do you mind adding a call to “acc_init” at the start of your program and explicitly set the device type? The program works as expected once I add this.

  • Mat