vectorizable_load allocates n_copies+1 dr_chains, even though only the first n_copies are needed. This patch removes the extra one and IMO makes the flow a bit more obvious.
Tested on x86_64-linux-gnu and arm-linux-gnueabi. OK to install? Richard gcc/ * tree-vect-stmts.c (vectorizable_load): Allocate and free dr_chain within the per-copy loop. Index: gcc/tree-vect-stmts.c =================================================================== --- gcc/tree-vect-stmts.c 2011-04-12 11:55:07.000000000 +0100 +++ gcc/tree-vect-stmts.c 2011-04-12 11:55:08.000000000 +0100 @@ -3961,8 +3961,6 @@ vectorizable_load (gimple stmt, gimple_s } else vec_num = group_size; - - dr_chain = VEC_alloc (tree, heap, vec_num); } else { @@ -4116,6 +4114,9 @@ vectorizable_load (gimple stmt, gimple_s dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE); + if (strided_load || slp_perm) + dr_chain = VEC_alloc (tree, heap, vec_num); + for (i = 0; i < vec_num; i++) { if (i > 0) @@ -4325,8 +4326,6 @@ vectorizable_load (gimple stmt, gimple_s return false; *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); - VEC_free (tree, heap, dr_chain); - dr_chain = VEC_alloc (tree, heap, group_size); } else { @@ -4337,11 +4336,10 @@ vectorizable_load (gimple stmt, gimple_s prev_stmt_info = vinfo_for_stmt (new_stmt); } } + if (dr_chain) + VEC_free (tree, heap, dr_chain); } - if (dr_chain) - VEC_free (tree, heap, dr_chain); - return true; }