https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81065
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |wrong-code
Status|UNCONFIRMED |NEW
Last reconfirmed| |2017-06-12
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Thanks for reporting these bugs, they are all latent wrong-code even w/o UBSAN.
.original w/o ubsan:
;; Function foo (null)
;; enabled by -tree-original
{
return (int) x * 14165900 + -77663775;
}
and as usual, it's extract_muldiv ...
(gdb) p debug_generic_expr (op0)
(int) x * 1652 + -9057
$1 = void
(gdb) p debug_generic_expr (op1)
8575
turning that into
(int) x * 14165900 + -77663775
it really means that this kind of distribution is never safe unless we rewrite
the inner multiplication into unsigned arithmetic (given the cast of x we
do have an idea about the value range of the other operand so we could
handle some cases -- but I'd rather not do that in extract_muldiv but
in a match.pd pattern).
I'd love to say bye-bye to extract_muldiv in it's current state...
