[Bug tree-optimization/82604] [8 Regression] SPEC CPU2006 410.bwaves ~50% performance regression with trunk@253679 when ftree-parallelize-loops is used

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82604

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Sth like the following which only works up to the point of dependence analysis
trying to disambiguate this ref against others ...  I suppose some
dr_may_alias_p tweaks to consider niter information and step/size to do
a "simple" offset based overlap test is missing.

Index: gcc/tree-data-ref.c
===================================================================
--- gcc/tree-data-ref.c (revision 254858)
+++ gcc/tree-data-ref.c (working copy)
@@ -4827,7 +4827,7 @@ get_references_in_stmt (gimple *stmt, ve
            clobbers_memory = true;
            break;
          }
-      else
+      else if (! gimple_call_builtin_p (stmt, BUILT_IN_MEMSET))
        clobbers_memory = true;
     }
   else if (stmt_code == GIMPLE_ASM
@@ -4888,6 +4888,12 @@ get_references_in_stmt (gimple *stmt, ve
          default:
            break;
          }
+      else if (gimple_call_builtin_p (stmt, BUILT_IN_MEMSET))
+       {
+         ref.ref = fold_build2 (MEM_REF, build_array_type (char_type_node,
build_index_type (gimple_call_arg (stmt, 2))), gimple_call_arg (stmt, 0),
build_zero_cst (ptr_type_node));
+         references->safe_push (ref);
+         return false;
+       }

       op0 = gimple_call_lhs (stmt);
       n = gimple_call_num_args (stmt);