diff mbox

[1/2] Fix miscompile of 416.gamess.

Message ID 1280434877-24563-2-git-send-email-sebpop@gmail.com
State New
Headers show

Commit Message

Sebastian Pop July 29, 2010, 8:21 p.m. UTC 
2010-07-29  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-sese-to-poly.c (rewrite_close_phi_out_of_ssa): Add back
	the case removed in the previous patch, when the only phi argument
	is defined in the same loop as the phi node itself.  Handle it
	separately from the invariant case by both propagating it outside
	the region and replacing the phi node with an assign.
---
 gcc/ChangeLog.graphite      |    8 ++++++++
 gcc/graphite-sese-to-poly.c |   10 ++++++++++
 2 files changed, 18 insertions(+), 0 deletions(-)
diff mbox

Patch

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 77aa85c..e040d6e 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,11 @@ 
+2010-07-29  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (rewrite_close_phi_out_of_ssa): Add back
+	the case removed in the previous patch, when the only phi argument
+	is defined in the same loop as the phi node itself.  Handle it
+	separately from the invariant case by both propagating it outside
+	the region and replacing the phi node with an assign.
+
 2010-07-28  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* graphite-sese-to-poly.c (rewrite_close_phi_out_of_ssa): The only
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 05a0e8f..1503e2c 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2248,6 +2248,16 @@  rewrite_close_phi_out_of_ssa (gimple_stmt_iterator *psi, sese region)
       return;
     }
 
+  else if (gimple_bb (SSA_NAME_DEF_STMT (arg))->loop_father == bb->loop_father)
+    {
+      propagate_expr_outside_region (res, arg, region);
+      stmt = gimple_build_assign (res, arg);
+      remove_phi_node (psi, false);
+      gsi_insert_before (&gsi, stmt, GSI_NEW_STMT);
+      SSA_NAME_DEF_STMT (res) = stmt;
+      return;
+    }
+
   /* If res is scev analyzable and is not a scalar value, it is safe
      to ignore the close phi node: it will be code generated in the
      out of Graphite pass.  */