[flang-commits] [flang] [llvm] [Flang][OpenMP] Support nested derived types in DO CONCURRENT device conversion (PR #218963)

[via flang-commits]

================
@@ -0,0 +1,50 @@
+! Verifies that `do concurrent` correctly lowers map for basic nested derived
+! types, correctly mapping the required components of the derived type.
+! REQUIRES: flang, amdgpu
+
+! RUN: %libomptarget-compile-fortran-generic -fdo-concurrent-to-openmp=device
+! RUN: %libomptarget-run-generic 2>&1 | %fcheck-generic
+program main
+   implicit none
+
+   type :: alloc_buffer
+     integer :: i
+     real, allocatable :: data(:)
+   end type alloc_buffer
+
+   type :: buffer
+     integer :: i
+     real :: data(8)
+   end type buffer
+
+   type :: array_dt
+      type(buffer) :: buf
+   end type array_dt
+
+   type :: alloc_array_dt
+      type(alloc_buffer) :: buf
----------------
MattPD wrote:

I confirmed that both elements are now updated correctly on the device. Thanks!


https://github.com/llvm/llvm-project/pull/218963