public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [Committed, PR69039] Only allow single exit phi for reduction in try_create_reduction_list
@ 2016-01-10 12:18 Tom de Vries
  0 siblings, 0 replies; only message in thread
From: Tom de Vries @ 2016-01-10 12:18 UTC (permalink / raw)
  To: gcc-patches; +Cc: Richard Biener

[-- Attachment #1: Type: text/plain, Size: 1281 bytes --]

Hi,

Consider this test-case (reduced from ira-color.c):
...
unsigned int b;

unsigned int
fn1 (unsigned int d)
{
   int i;

   for (i = 0; i < 1000; i++)
     b |= d;

   return b;
}
...

When compiling with -O2 -ftree-parallelize-loops=2, we run into a segfault.


Before parloops, we have reduction stmt _6 with loop phi prephitmp_11:
...
   <bb 2>:
   pretmp_1 = b;

   <bb 3>:
   # i_13 = PHI <i_8(4), 0(2)>
   # prephitmp_11 = PHI <_6(4), pretmp_1(2)>
   # ivtmp_4 = PHI <ivtmp_12(4), 1000(2)>
   _6 = d_5(D) | prephitmp_11;
   i_8 = i_13 + 1;
   ivtmp_12 = ivtmp_4 - 1;
   if (ivtmp_12 != 0)
     goto <bb 4>;
   else
     goto <bb 5>;

   <bb 4>:
   goto <bb 3>;

   <bb 5>:
   # _9 = PHI <_6(3)>
   # b_lsm.4_17 = PHI <_6(3)>
   b = b_lsm.4_17;
   return _9;
...

There are however two corresponding loop exit phis: _9 and b_lsm.4_17.

We only track one exit phi in the keep_res field of struct 
reduction_info, and the segfault is a result of that discrepancy.

The seqfault only happens when using transform_to_exit_first_loop_alt, 
so only trunk needs this fix.


The patch fixes the segfault conservatively by bailing out of 
parallelizing the loop if a reduction has more than one exit phi.

Bootstrapped and reg-tested on x86_64.

Committed to trunk.

Thanks,
- Tom

[-- Attachment #2: 0001-Only-allow-single-exit-phi-for-reduction-in-try_create_reduction_list.patch --]
[-- Type: text/x-patch, Size: 1477 bytes --]

Only allow single exit phi for reduction in try_create_reduction_list

2016-01-07  Tom de Vries  <tom@codesourcery.com>

	PR tree-optimization/69039
	* tree-parloops.c (try_create_reduction_list): Only allow single exit
	phi for reduction.

	* gcc.dg/autopar/pr69039.c: New test.

---
 gcc/testsuite/gcc.dg/autopar/pr69039.c | 15 +++++++++++++++
 gcc/tree-parloops.c                    |  8 ++++++++
 2 files changed, 23 insertions(+)

diff --git a/gcc/testsuite/gcc.dg/autopar/pr69039.c b/gcc/testsuite/gcc.dg/autopar/pr69039.c
new file mode 100644
index 0000000..556f700
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/autopar/pr69039.c
@@ -0,0 +1,15 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-parallelize-loops=2" } */
+
+unsigned int b;
+
+unsigned int
+fn1 (unsigned int d)
+{
+  int i;
+
+  for (i = 0; i < 1000; i++)
+    b |= d;
+
+  return b;
+}
diff --git a/gcc/tree-parloops.c b/gcc/tree-parloops.c
index 84f18bd..394aba8 100644
--- a/gcc/tree-parloops.c
+++ b/gcc/tree-parloops.c
@@ -2595,6 +2595,14 @@ try_create_reduction_list (loop_p loop,
 			 "  FAILED: it is not a part of reduction.\n");
 	      return false;
 	    }
+	  if (red->keep_res != NULL)
+	    {
+	      if (dump_file && (dump_flags & TDF_DETAILS))
+		fprintf (dump_file,
+			 "  FAILED: reduction has multiple exit phis.\n");
+	      return false;
+	    }
+	  red->keep_res = phi;
 	  if (dump_file && (dump_flags & TDF_DETAILS))
 	    {
 	      fprintf (dump_file, "reduction phi is  ");

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

only message in thread, other threads:[~2016-01-10 12:18 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-01-10 12:18 [Committed, PR69039] Only allow single exit phi for reduction in try_create_reduction_list Tom de Vries

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).