I am testing the following patch to fix a latent bug in the vectorizer
dealing with redundant DRs.
Bootstrap and regtest pending on x86_64-unknown-linux-gnu.
Richard.
2015-01-12 Richard Biener <rguent...@suse.de>
PR tree-optimization/64404
* tree-vect-stmts.c (vectorizable_load): Use the proper
vectorized stmts for CSEing loads with the same DR.
* gcc.dg/vect/pr64404.c: New testcase.
Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c (revision 219446)
+++ gcc/tree-vect-stmts.c (working copy)
@@ -6155,7 +6155,7 @@ vectorizable_load (gimple stmt, gimple_s
is even wrong code. See PR56270. */
&& !slp)
{
- *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
+ *vec_stmt = STMT_VINFO_VEC_STMT (vinfo_for_stmt (first_stmt));
return true;
}
first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
Index: gcc/testsuite/gcc.dg/vect/pr64404.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/pr64404.c (revision 0)
+++ gcc/testsuite/gcc.dg/vect/pr64404.c (working copy)
@@ -0,0 +1,26 @@
+/* { dg-do compile } */
+/* { dg-additional-options "--param=sccvn-max-alias-queries-per-access=1" } */
+
+typedef struct
+{
+ float l, h;
+} tFPinterval;
+
+tFPinterval X[1024];
+tFPinterval Y[1024];
+tFPinterval Z[1024];
+
+void
+Compute (void)
+{
+ int d;
+ for (d = 0; d < 1024; d++)
+ {
+ Y[d].l = X[d].l + X[d].h;
+ Y[d].h = Y[d].l;
+ Z[d].l = X[d].l;
+ Z[d].h = X[d].h;
+ }
+}
+
+/* { dg-final { cleanup-tree-dump "vect" } } */
