https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82083
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Isn't that the in the lambda_matrix* stuff?
I've recently discussed it with Richard:
<jakub> do you know something about lambda_matrix_right_hermite and functions
it calls? It calculates everything in int type, which on various testcases
results in various signed integer overflows (both + and *)
<jakub> shall it just use HWIs, or wide_ints, something different instead?
<richi> I "recently" changed similar code in tree-chrec.c to use widest_ints
(tree_fold_binomial)
<richi> I think the lambda code wants to use "infinite" precision integer math
<jakub> e.g. gcc.dg/torture/pr60183.c triggers many of these
<richi> yeah, I spotted the lambda stuff as well but thought w/o a miscompile
testcase I won't bother refactoring it (and making it slow)
<richi> bumping lambda_vector from int * to int64_t * would help a bit but not
really solve the underlying issue
<jakub> guess that would be some middle-ground, not have stuff really slow and
large, and avoid miscompiling the large majority of what is miscompiled right
now
<richi> possibly
No immediate plans to do anything about that though.
