The split of the versioning condition assumes the definition is in the condition block which is ensured by the versioning code. But that only works when we actually have to insert any statements for the versioning condition. The following adjusts the guard accordingly and asserts this condition.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed. PR tree-optimization/108816 * tree-vect-loop-manip.cc (vect_loop_versioning): Adjust versioning condition split prerequesite, assert required invariant. * gcc.dg/torture/pr108816.c: New testcase. --- gcc/testsuite/gcc.dg/torture/pr108816.c | 18 ++++++++++++++++++ gcc/tree-vect-loop-manip.cc | 3 ++- 2 files changed, 20 insertions(+), 1 deletion(-) create mode 100644 gcc/testsuite/gcc.dg/torture/pr108816.c diff --git a/gcc/testsuite/gcc.dg/torture/pr108816.c b/gcc/testsuite/gcc.dg/torture/pr108816.c new file mode 100644 index 00000000000..4c24d5584c1 --- /dev/null +++ b/gcc/testsuite/gcc.dg/torture/pr108816.c @@ -0,0 +1,18 @@ +/* { dg-do compile } */ +/* { dg-additional-options "-fvect-cost-model=dynamic" } */ + +int m; + +void +foo (int p[][16], unsigned int x) +{ + while (x < 4) + { + int j; + + for (j = x * 4; j < (x + 1) * 4 - 2; j++) + p[0][j] = p[m][j]; + + ++x; + } +} diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc index c04fcf40c44..6aa3d2ed0bf 100644 --- a/gcc/tree-vect-loop-manip.cc +++ b/gcc/tree-vect-loop-manip.cc @@ -3477,7 +3477,7 @@ vect_loop_versioning (loop_vec_info loop_vinfo, tree cost_name = NULL_TREE; profile_probability prob2 = profile_probability::uninitialized (); if (cond_expr - && !integer_truep (cond_expr) + && EXPR_P (cond_expr) && (version_niter || version_align || version_alias @@ -3711,6 +3711,7 @@ vect_loop_versioning (loop_vec_info loop_vinfo, if (cost_name && TREE_CODE (cost_name) == SSA_NAME) { gimple *def = SSA_NAME_DEF_STMT (cost_name); + gcc_assert (gimple_bb (def) == condition_bb); /* All uses of the cost check are 'true' after the check we are going to insert. */ replace_uses_by (cost_name, boolean_true_node); -- 2.35.3