public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [committed] Fix lastprivate gimplification in taskloop (PR middle-end/89002)
@ 2019-01-28 23:05 Jakub Jelinek
  0 siblings, 0 replies; only message in thread
From: Jakub Jelinek @ 2019-01-28 23:05 UTC (permalink / raw)
  To: gcc-patches

Hi!

When gimplifying lastprivate or linear clause with non-NULL
OMP_CLAUSE_*_STMT, we wrap it into a BIND_EXPR and push temporaries there,
but I haven't done that when we are adding that only during gimplification
for the loop's IV.  For worksharing loop or distribute or simd it isn't
that big deal, because either there is no outlining of the corresponding
code, or it is through outer parallel, teams etc. so the temporaries end up
there.  But, in the case of taskloop, there is always outlining and so we
need to make sure the temporaries are in the right spot.

Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux,
committed to trunk (so far).

2019-01-28  Jakub Jelinek  <jakub@redhat.com>

	PR middle-end/89002
	* gimplify.c (gimplify_omp_for): When adding OMP_CLAUSE_*_GIMPLE_SEQ
	for lastprivate/linear IV, push gimplify context around gimplify_assign
	and, if it needed any temporaries, pop it into a gimple bind around the
	sequence.

	* testsuite/libgomp.c/pr89002.c: New test.

--- gcc/gimplify.c.jj	2019-01-22 23:27:56.000000000 +0100
+++ gcc/gimplify.c	2019-01-28 17:32:58.149683211 +0100
@@ -11167,8 +11167,17 @@ gimplify_omp_for (tree *expr_p, gimple_s
 		  seq = &OMP_CLAUSE_LASTPRIVATE_GIMPLE_SEQ (c);
 		else
 		  seq = &OMP_CLAUSE_LINEAR_GIMPLE_SEQ (c);
+		push_gimplify_context ();
 		gimplify_assign (decl, t, seq);
-	    }
+		gimple *bind = NULL;
+		if (gimplify_ctxp->temps)
+		  {
+		    bind = gimple_build_bind (NULL_TREE, *seq, NULL_TREE);
+		    *seq = NULL;
+		    gimplify_seq_add_stmt (seq, bind);
+		  }
+		pop_gimplify_context (bind);
+	      }
 	}
     }
 
--- libgomp/testsuite/libgomp.c/pr89002.c.jj	2019-01-28 17:49:47.791086171 +0100
+++ libgomp/testsuite/libgomp.c/pr89002.c	2019-01-28 17:49:17.836600273 +0100
@@ -0,0 +1,43 @@
+/* PR middle-end/89002 */
+
+extern void abort (void);
+
+int
+foo (int x)
+{
+  int a;
+  int *p = &a;
+
+#pragma omp taskloop lastprivate (a)
+  for (a = 0; a < x; ++a)
+    ;
+  return *p;
+}
+
+int
+bar (int x)
+{
+  int a;
+  int *p = &a;
+
+#pragma omp parallel
+#pragma omp single
+#pragma omp taskloop lastprivate (a)
+  for (a = 0; a < x; ++a)
+    ;
+  return *p;
+}
+
+int
+main ()
+{
+#pragma omp parallel
+#pragma omp single
+  {
+    if (foo (4) != 4)
+      abort ();
+  }
+  if (bar (6) != 6)
+    abort ();
+  return 0;
+}

	Jakub

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2019-01-28 22:39 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-01-28 23:05 [committed] Fix lastprivate gimplification in taskloop (PR middle-end/89002) Jakub Jelinek

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).