Hi!
On Tue, 16 May 2017 20:55:46 +0800, Chung-Lin Tang <chunglin_t...@mentor.com>
wrote:
> finalize clause of the exit data directive, and the
> corresponding API routines.
> --- libgomp/oacc-parallel.c (revision 248095)
> +++ libgomp/oacc-parallel.c (revision 248096)
> @@ -355,7 +355,22 @@
> }
> }
> else
> - tgt->device_descr->openacc.register_async_cleanup_func (tgt, async);
> + {
> + bool async_unmap = false;
> + for (size_t i = 0; i < tgt->list_count; i++)
> + {
> + splay_tree_key k = tgt->list[i].key;
> + if (k && k->refcount == 1)
> + {
> + async_unmap = true;
> + break;
> + }
> + }
> + if (async_unmap)
> + tgt->device_descr->openacc.register_async_cleanup_func (tgt, async);
> + else
> + gomp_unmap_vars (tgt, false);
> + }
>
> acc_dev->openacc.async_set_async_func (acc_async_sync);
This additional gomp_unmap_vars call also needs be instrumented for the
OpenACC Profiling Interface.
> --- libgomp/openacc.h (revision 248095)
> +++ libgomp/openacc.h (revision 248096)
> +/* Finalize versions of copyout/delete functions, specified in OpenACC 2.5.
> */
> +void acc_copyout_finalize (void *, size_t) __GOACC_NOTHROW;
> +void acc_copyout_finalize_async (void *, size_t, int) __GOACC_NOTHROW;
> +void acc_delete_finalize (void *, size_t) __GOACC_NOTHROW;
> +void acc_delete_finalize_async (void *, size_t, int) __GOACC_NOTHROW;
And for these, the OpenACC Profiling Interface status needs to be
documented.
Committed to gomp-4_0-branch in r248150:
commit dc97f798ad7f7f44f45b2b8e0ece81f3926fa1c2
Author: tschwinge <tschwinge@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Wed May 17 12:07:30 2017 +0000
OpenACC 2.5 Profiling Interface changes for "finalize" handling
libgomp/
* libgomp.texi (OpenACC Profiling Interface): Update.
* oacc-parallel.c (GOACC_parallel_keyed): Update profiling event
generation.
git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gomp-4_0-branch@248150
138bc75d-0d04-0410-961f-82ee72b054a4
---
libgomp/ChangeLog.gomp | 4 ++++
libgomp/libgomp.texi | 4 ++--
libgomp/oacc-parallel.c | 9 +++++++--
3 files changed, 13 insertions(+), 4 deletions(-)
diff --git libgomp/ChangeLog.gomp libgomp/ChangeLog.gomp
index 2ea7215..996c1f9 100644
--- libgomp/ChangeLog.gomp
+++ libgomp/ChangeLog.gomp
@@ -1,5 +1,9 @@
2017-05-17 Thomas Schwinge <tho...@codesourcery.com>
+ * libgomp.texi (OpenACC Profiling Interface): Update.
+ * oacc-parallel.c (GOACC_parallel_keyed): Update profiling event
+ generation.
+
* oacc-parallel.c (GOACC_enter_exit_data): Locally compute
"finalize", and remove the formal parameter. Adjust all users.
(GOACC_declare): Don't replace GOMP_MAP_FROM with
diff --git libgomp/libgomp.texi libgomp/libgomp.texi
index 69fb3be..1dea1e2 100644
--- libgomp/libgomp.texi
+++ libgomp/libgomp.texi
@@ -3459,8 +3459,8 @@ offloading devices (it's not clear if they should be):
@item @code{acc_free}
@item @code{acc_copyin}, @code{acc_present_or_copyin}, @code{acc_copyin_async}
@item @code{acc_create}, @code{acc_present_or_create}, @code{acc_create_async}
-@item @code{acc_copyout}, @code{acc_copyout_async}
-@item @code{acc_delete}, @code{acc_delete_async}
+@item @code{acc_copyout}, @code{acc_copyout_async},
@code{acc_copyout_finalize}, @code{acc_copyout_finalize_async}
+@item @code{acc_delete}, @code{acc_delete_async}, @code{acc_delete_finalize},
@code{acc_delete_finalize_async}
@item @code{acc_update_device}, @code{acc_update_device_async}
@item @code{acc_update_self}, @code{acc_update_self_async}
@item @code{acc_map_data}, @code{acc_unmap_data}
diff --git libgomp/oacc-parallel.c libgomp/oacc-parallel.c
index ff6e96c..622c711 100644
--- libgomp/oacc-parallel.c
+++ libgomp/oacc-parallel.c
@@ -333,8 +333,10 @@ GOACC_parallel_keyed (int device, void (*fn) (void *),
async, dims, tgt);
/* If running synchronously, unmap immediately. */
+ bool copyfrom = true;
if (async < acc_async_noval)
{
+ unmap:
if (profiling_dispatch_p)
{
prof_info.event_type = acc_ev_exit_data_start;
@@ -344,7 +346,7 @@ GOACC_parallel_keyed (int device, void (*fn) (void *),
goacc_profiling_dispatch (&prof_info, &enter_exit_data_event_info,
&api_info);
}
- gomp_unmap_vars (tgt, true);
+ gomp_unmap_vars (tgt, copyfrom);
if (profiling_dispatch_p)
{
prof_info.event_type = acc_ev_exit_data_end;
@@ -369,7 +371,10 @@ GOACC_parallel_keyed (int device, void (*fn) (void *),
if (async_unmap)
tgt->device_descr->openacc.register_async_cleanup_func (tgt, async);
else
- gomp_unmap_vars (tgt, false);
+ {
+ copyfrom = false;
+ goto unmap;
+ }
}
acc_dev->openacc.async_set_async_func (acc_async_sync);
(That one can certainly do with some restructuring, to avoid the "goto".)
;-)
Grüße
Thomas
