[flang-commits] [flang] [flang][cuda] Restrict managed data transfers to whole-array assignments (PR #212593)

[Zhen Wang via flang-commits]

================
@@ -634,6 +634,8 @@ end subroutine
 ! CHECK-LABEL: func.func @_QPsub33
 ! CHECK-NOT: cuf.data_transfer 
 
+! Only a whole managed array is initialized on the device; a section is assigned on
+! the host (CUDA Fortran Programming Guide 3.4.1).
----------------
wangzpgi wrote:

You're right, it covers the whole array, the comment was misleading, fixed.

https://github.com/llvm/llvm-project/pull/212593