https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85888
--- Comment #3 from Martin Sebor <msebor at gcc dot gnu.org> --- The reason why the powerpc64 target expands the call at a different (later) point than other targets is because the first attempt to expand it ends up in the "cmpstrsi" pattern where the expansion fails (as does the expand_strn_compare() function called later). The later expansion is done with a modified call expression where the bound originally in SSA_NAME form is wrapped in a MIN_EXPR. On targets like x86 the first expansion attempt succeeds (resulting in a call top strncmp()). The subsequent call to maybe_warn_nonstring_arg() is made with the original call expression with the SSA_NAME bound.
