[libgomp,patch] Add omp_{integer,logical}_kind to omp_lib.h

[libgomp,patch] Add omp_{integer,logical}_kind to omp_lib.h

[FX]

[-- Attachment #1: Type: text/plain, Size: 406 bytes --]

PR 42041 is about definitions for omp_integer_kind and omp_logical_kind missing from omp_lib.h (even though they're in omp_lib.f90). This is strictly conforming to OpenMP 3.0 standard, but it was a bug in the standard, due to be fixed by 3.1.

In the process of cleaning some low-hanging bugs in the bugzilla, I propose this patch.
Built and regtested on x86_64-linux, OK to commit to trunk?

FX



[-- Attachment #2: libgomp.diff --]
[-- Type: application/octet-stream, Size: 1724 bytes --]

Index: omp_lib.h.in
===================================================================
--- omp_lib.h.in	(revision 160470)
+++ omp_lib.h.in	(working copy)
@@ -22,9 +22,12 @@
 !  see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
 !  <http://www.gnu.org/licenses/>.
 
+      integer omp_integer_kind, omp_logical_kind
       integer omp_lock_kind, omp_nest_lock_kind, openmp_version
       integer omp_sched_kind, omp_sched_static, omp_sched_dynamic
       integer omp_sched_guided, omp_sched_auto
+      parameter (omp_integer_kind = 4)
+      parameter (omp_logical_kind = 4)
       parameter (omp_lock_kind = @OMP_LOCK_KIND@)
       parameter (omp_nest_lock_kind = @OMP_NEST_LOCK_KIND@)
       parameter (omp_sched_kind = 4)
@@ -42,16 +45,16 @@
       external omp_set_num_threads
 
       external omp_get_dynamic, omp_get_nested
-      logical(4) omp_get_dynamic, omp_get_nested
+      logical(omp_logical_kind) omp_get_dynamic, omp_get_nested
       external omp_test_lock, omp_in_parallel
-      logical(4) omp_test_lock, omp_in_parallel
+      logical(omp_logical_kind) omp_test_lock, omp_in_parallel
 
       external omp_get_max_threads, omp_get_num_procs
-      integer(4) omp_get_max_threads, omp_get_num_procs
+      integer(omp_integer_kind) omp_get_max_threads, omp_get_num_procs
       external omp_get_num_threads, omp_get_thread_num
-      integer(4) omp_get_num_threads, omp_get_thread_num
+      integer(omp_integer_kind) omp_get_num_threads, omp_get_thread_num
       external omp_test_nest_lock
-      integer(4) omp_test_nest_lock
+      integer(omp_integer_kind) omp_test_nest_lock
 
       external omp_get_wtick, omp_get_wtime
       double precision omp_get_wtick, omp_get_wtime

[-- Attachment #3: libgomp.ChangeLog --]
[-- Type: application/octet-stream, Size: 205 bytes --]

2010-06-10  Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>

	PR fortran/42041
	* omp_lib.h.in: Add definitions of omp_integer_kind and
	omp_logical_kind, and use them in external function declarations.

Re: [libgomp,patch] Add omp_{integer,logical}_kind to omp_lib.h

[Jakub Jelinek]

On Thu, Jun 10, 2010 at 10:36:06PM +0200, FX wrote:
> PR 42041 is about definitions for omp_integer_kind and omp_logical_kind missing from omp_lib.h (even though they're in omp_lib.f90). This is strictly conforming to OpenMP 3.0 standard, but it was a bug in the standard, due to be fixed by 3.1.
> 
> In the process of cleaning some low-hanging bugs in the bugzilla, I propose this patch.
> Built and regtested on x86_64-linux, OK to commit to trunk?

Can't we wait here a week to see what OpenMP 3.1 draft actually says on
this?

	Jakub

Re: [libgomp,patch] Add omp_{integer,logical}_kind to omp_lib.h

[FX]

> To add: The OpenMP 3.1 draft is said to be released at the IWOMP 2010
> conference, which is next week.

Anyone has access to the document? I cannot find anything on the IWOMP website, nor on openmp.org.

FX

Re: [libgomp,patch] Add omp_{integer,logical}_kind to omp_lib.h

[Tobias Burnus]

On 07/02/2010 12:27 PM, FX wrote:
> Anyone has access to the document? I cannot find anything on the IWOMP website, nor on openmp.org.
>   

The publication of the spec got delayed as the spec as not ready. There
was one meeting after the IWOMP conference to discuss further details,
but seemingly some more items need to be solved. Bronis de Supinski
wrote on 17 June that he hopes that the OpenMP 3.1 draft will be
available in few weeks. Cf. also Michael Wong's report at
http://www-949.ibm.com/software/rational/cafe/blogs/ccpp-parallel-multicore/2010/06/21/the-view-from-iwomp-2010-trip-report

If you spend 50 bucks/year, you can become cOMPunity member [1] and get
early access to the draft... (Which neither I nor Jakub have.)

For the features planed for 3.1, see the link above - and the relevant
proceedings in LNCS (linked from the conference site).
The OpenMP 3.1 goals were to refine and extend the existing
specification, not break existing code, and to pose a minimal
implementation burden beyond 3.0. Planed are:
- User-defined reduction operators [according to the proceeding paper,
it looks quite powerful and thus requires quite some work to implement -
thus one wonders about the definition of "minimal"]
- Atomics: support capture/write
- Task scheduling: taskyield construct, final clause

Tobias

[1] http://www.compunity.org/contacts/membership/index.php