https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99079
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Attachment #50175|0 |1
is obsolete| |
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 50176
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50176&action=edit
gcc11-pr99079.patch
Updated patch. Actually, I think widening conversion from signed TREE_TYPE
(@1)
to signed type is fine too.
In the problematic case where @1 is negative (i.e. powerof2 << var which shifts
it into the sign bit, certainly valid at least in C++20), if we say have int to
long long conversion and we know the dividend is non-negative, it would be
@0 % (long long) INT_MIN, i.e. @0 % -0x80000000LL and that can still be
expanded
as @0 & 0x7fffffffLL.