re PR tree-optimization/78856 (wrong code at -O3 on x86_64-linux-gnu (in both 32-bit and 64-bit modes))

PR tree-optimizatin/78856
	* tree-ssa-threadupdate.c: Include tree-vectorizer.h.
	(mark_threaded_blocks): Remove code to truncate thread paths that
	cross multiple loop headers.  Instead invalidate the cached loop
	iteration information and handle case of a thread path walking
	into an irreducible region.

	PR tree-optimization/78856
	* gcc.c-torture/execute/pr78856.c: New test.

From-SVN: r244045

gcc/ChangeLog

@@ -1,3 +1,12 @@
+2017-01-03  Jeff Law  <law@redhat.com>
+
+	PR tree-optimizatin/78856
+	* tree-ssa-threadupdate.c: Include tree-vectorizer.h.
+	(mark_threaded_blocks): Remove code to truncate thread paths that
+	cross multiple loop headers.  Instead invalidate the cached loop
+	iteration information and handle case of a thread path walking
+	into an irreducible region.
+
 2016-12-30  Michael Meissner  <meissner@linux.vnet.ibm.com>
 
 	PR target/78900

gcc/testsuite/ChangeLog

@@ -1,3 +1,8 @@
+2017-01-03  Jeff Law  <law@redhat.com>
+
+	PR tree-optimization/78856
+	* gcc.c-torture/execute/pr78856.c: New test.
+
 2017-01-03  Michael Meissner  <meissner@linux.vnet.ibm.com>
 
 	PR target/78953

gcc/testsuite/gcc.c-torture/execute/pr78856.c

@@ -0,0 +1,25 @@
+extern void exit (int);
+
+int a, b, c, d, e, f[3]; 
+
+int main() 
+{
+  while (d)
+    while (1)
+      ;
+  int g = 0, h, i = 0;
+  for (; g < 21; g += 9) 
+    {
+      int j = 1;
+      for (h = 0; h < 3; h++)
+	f[h] = 1;
+      for (; j < 10; j++) {
+	d = i && (b ? 0 : c); 
+	i = 1;
+	if (g)
+	  a = e;
+      }
+  }
+  exit (0);
+}
+

gcc/tree-ssa-threadupdate.c

@@ -34,6 +34,7 @@ along with GCC; see the file COPYING3.  If not see
 #include "cfgloop.h"
 #include "dbgcnt.h"
 #include "tree-cfg.h"
+#include "tree-vectorizer.h"
 
 /* Given a block B, update the CFG and SSA graph to reflect redirecting
    one or more in-edges to B to instead reach the destination of an
@@ -2084,10 +2085,8 @@ mark_threaded_blocks (bitmap threaded_blocks)
   /* Look for jump threading paths which cross multiple loop headers.
 
      The code to thread through loop headers will change the CFG in ways
-     that break assumptions made by the loop optimization code.
+     that invalidate the cached loop iteration information.  So we must
-
+     detect that case and wipe the cached information.  */
-     We don't want to blindly cancel the requests.  We can instead do better
-     by trimming off the end of the jump thread path.  */
   EXECUTE_IF_SET_IN_BITMAP (tmp, 0, i, bi)
     {
       basic_block bb = BASIC_BLOCK_FOR_FN (cfun, i);
@@ -2102,26 +2101,16 @@ mark_threaded_blocks (bitmap threaded_blocks)
 		   i++)
 		{
 		  basic_block dest = (*path)[i]->e->dest;
+		  basic_block src = (*path)[i]->e->src;
 		  crossed_headers += (dest == dest->loop_father->header);
+		  /* If we step from a block outside an irreducible region
+		     to a block inside an irreducible region, then we have
+		     crossed into a loop.  */
+		  crossed_headers += ((src->flags & BB_IRREDUCIBLE_LOOP)
+				      != (dest->flags & BB_IRREDUCIBLE_LOOP));
 		  if (crossed_headers > 1)
 		    {
-		      /* Trim from entry I onwards.  */
+		      vect_free_loop_info_assumptions (dest->loop_father);
-		      for (unsigned int j = i; j < path->length (); j++)
-			delete (*path)[j];
-		      path->truncate (i);
-
-		      /* Now that we've truncated the path, make sure
-			 what's left is still valid.   We need at least
-			 two edges on the path and the last edge can not
-			 be a joiner.  This should never happen, but let's
-			 be safe.  */
-		      if (path->length () < 2
-			  || (path->last ()->type
-			      == EDGE_COPY_SRC_JOINER_BLOCK))
-			{
-			  delete_jump_thread_path (path);
-			  e->aux = NULL;
-			}
 		      break;
 		    }
 		}