[Issue]: flang-new: OpenMP declare target arrays crash

[VeeEM]

Problem Description

I'm seeing unexpected crashes with OpenMP offload code that uses declare target arrays after commit 13b4a9d.

Operating System

SUSE Linux Enterprise Server 15 SP5 (Cray OS on LUMI)

CPU

AMD EPYC 7742 64-Core

GPU

AMD Instinct MI250X

ROCm Version

ROCm 6.2.2

ROCm Component

flang

Steps to Reproduce

I think the program below might reproduce the issue. The program should print "acc 500" but after 13b4a9d it prints:

Queue error - HSA_STATUS_ERROR_MEMORY_FAULT
Display only launched kernel:
Kernel 'omp target in _QQmain @ 11 (__omp_offloading_54bbb604_a0017379__QQmain_l11)'
OFFLOAD ERROR: Memory access fault by GPU 4 (agent 0x35ff90) at virtual address (nil). Reasons: Unknown (0)
Device pointer (nil) does not point into any (current or prior) host-issued allocation (might be a global, stack, or shared location).

Compiler flags: flang-new -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa --offload-arch=gfx90a

program repr1
  integer, dimension(:), allocatable :: xs
  integer :: i, acc
  !$omp declare target (xs)

  allocate(xs(500))
  !$omp target enter data map(alloc:xs)
  xs = 1
  acc = 0
  !$omp target update to(xs)

  !$omp target map(acc)
  do i=1, 500
    acc = acc + xs(i)
  end do
  !$omp end target 
  
  print *, "acc", acc
end program

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

[ppanchad-amd]

Hi @VeeEM. Internal ticket has been created to investigate your issue. Thanks!

[tcgu-amd]

Hi @VeeEM, thanks for reaching out! Just to clarify, did the program work before 13b4a9d?

[VeeEM]

Yes, the program works before 13b4a9d.