https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126899
--- Comment #3 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
BTW, found another one that fails because the cost models differ:
gcc.dg/harden-cfr-pr111943.c (with --param=dom-jump-threading=0). This one
pins --param=max-jump-thread-duplication-stmts=0, and the backward threader
refuses to thread even paths that copy zero statements (its estimate is
compared with >=, so 0 * 2 >= 0 fails every path), while DOM's forward threader
counts the statements it would actually copy and allows the thread.
>From threadfull1:
Overall: 0 insns
FAIL: Did not thread around loop and would copy too many statements.
