Re: [Patch] Fortran: Update use_device_ptr for OpenMP 5.1 [PR105318]

[Tobias Burnus <tobias@codesourcery.com>]

[-- Attachment #1: Type: text/plain, Size: 1674 bytes --]

On 30.09.22 12:41, Tobias Burnus wrote:

Fortran: Update use_device_ptr for OpenMP 5.1 [PR105318]

The following two lines slipped in – which I have now removed in my version of the patch:

--- a/gcc/testsuite/gfortran.dg/gomp/is_device_ptr-1.f90
+++ b/gcc/testsuite/gfortran.dg/gomp/is_device_ptr-1.f90
@@ -1,4 +1,5 @@
 ! { dg-do compile }
+! First


(Leftover - I wanted to refer to PR105318 - but then decided against it.)



+++ b/libgomp/testsuite/libgomp.fortran/is_device_ptr-2.f90
@@ -0,0 +1,167 @@
+! { dg-do compile }


Likewise leftover - I had an intermittent problem with device testing; hence, I disabled it. It turned out to be unrelated, but it remained.

Side remark: An issue I found with device testing was that I missed the "xx => null()"; in principle obvious (as GCC tried to map the pointer target of the uninitialized pointer), but still easy to miss.

Side note: The OpenMP spec (post-5.2) also clarified the mapping of undefined pointers / requiring the two null init here.

(The second null() is to avoid issues with firstprivatizing/copy-in of the internal pointer representation vs. simply using it (host fallback, unified shared memory). This forces the pointer after the target region to be the same as before the region (now required by the spec!), permitting either implementation without unexpected side effects for the user.)

Tobias

-----------------
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955