[patch, vectorizer] Fix a bug in creation of reduction epilogue

[patch, vectorizer] Fix a bug in creation of reduction epilogue

[Ira Rosen]

Hi,

In case of multiple loop exit phis in vectorization of reduction we reduce
them to one vector. The result of this reduction is later ignored in case
we do the final value extraction with scalar code. This causes wrong code
generation for gfortran.dg/forall_7.f90 with -O3 -funroll-loops on Cell
SPU. This patch fixes this.

Bootstrapped on powerpc64-suse-linux and tested on powerpc64-suse-linux and
on spu-redhat-linux.
Committed.

Ira

ChangeLog:

	* tree-vect-loop.c (vect_create_epilog_for_reduction): Use the
	result of multiple results reduction when extracting the final
	value using scalar code.

Index: tree-vect-loop.c
===================================================================
--- tree-vect-loop.c    (revision 177266)
+++ tree-vect-loop.c    (working copy)
@@ -3683,13 +3683,13 @@ vect_create_epilog_for_reduction (VEC (tree, heap)
     {
       tree first_vect = PHI_RESULT (VEC_index (gimple, new_phis, 0));
       tree tmp;
+      gimple new_vec_stmt = NULL;

       vec_dest = vect_create_destination_var (scalar_dest, vectype);
       for (k = 1; k < VEC_length (gimple, new_phis); k++)
         {
           gimple next_phi = VEC_index (gimple, new_phis, k);
           tree second_vect = PHI_RESULT (next_phi);
-          gimple new_vec_stmt;

           tmp = build2 (code, vectype,  first_vect, second_vect);
           new_vec_stmt = gimple_build_assign (vec_dest, tmp);
@@ -3699,6 +3699,11 @@ vect_create_epilog_for_reduction (VEC (tree, heap)
         }

       new_phi_result = first_vect;
+      if (new_vec_stmt)
+        {
+          VEC_truncate (gimple, new_phis, 0);
+          VEC_safe_push (gimple, heap, new_phis, new_vec_stmt);
+        }
     }
   else
     new_phi_result = PHI_RESULT (VEC_index (gimple, new_phis, 0));
@@ -3809,7 +3814,10 @@ vect_create_epilog_for_reduction (VEC (tree, heap)
           vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1);
           FOR_EACH_VEC_ELT (gimple, new_phis, i, new_phi)
             {
-              vec_temp = PHI_RESULT (new_phi);
+              if (gimple_code (new_phi) == GIMPLE_PHI)
+                vec_temp = PHI_RESULT (new_phi);
+              else
+                vec_temp = gimple_assign_lhs (new_phi);
               rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize,
                             bitsize_zero_node);
               epilog_stmt = gimple_build_assign (new_scalar_dest, rhs);