On 10/14/25 5:40 AM, Siddhesh Poyarekar wrote:
On 2025-10-12 00:30, Sam James wrote:
Linsen Zhou <[email protected]> writes:
After commit 51b85dfeb19652bf3e0aaec08828ba7cee1e641c, when the pointer
offset is a variable in the loop, the object size of point may also need
to be reexamined.
Which make gcc_assert in the check_for_plus_in_loops failed.
gcc/ChangeLog:
tree-object-size.cc (check_for_plus_in_loops): Skip check for
the variable offset
Signed-off-by: Linsen Zhou <[email protected]>
---
Could you add a testcase to the patch, and also add a reference to
PR122012 in the commit message and in the ChangeLog?
I can't review this though, cc'd Sid whose patch exposed this.
Related Alpine Linux issues:
https://gitlab.alpinelinux.org/alpine/aports/-/issues/17416
https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/87912
gcc/tree-object-size.cc | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/gcc/tree-object-size.cc b/gcc/tree-object-size.cc
index 8545eff61a3..413e67a8125 100644
--- a/gcc/tree-object-size.cc
+++ b/gcc/tree-object-size.cc
@@ -2147,10 +2147,9 @@ check_for_plus_in_loops (struct
object_size_info *osi, tree var)
tree basevar = gimple_assign_rhs1 (stmt);
tree cst = gimple_assign_rhs2 (stmt);
- gcc_assert (TREE_CODE (cst) == INTEGER_CST);
-
I think keeping the tree in 'cst' is confusing if we can on longer rely
on it being a constant expression, but please let others comment first.
Agreed, lets call this `offset`.
/* Skip non-positive offsets. */
- if (integer_zerop (cst) || compare_tree_int (cst,
offset_limit) > 0)
+ if (TREE_CODE (cst) != INTEGER_CST
+ || integer_zerop (cst) || compare_tree_int (cst, offset_limit)
> 0)
return;
osi->depths[SSA_NAME_VERSION (basevar)] = 1;
This looks good to me; I had made the incorrect assumption that only
constant expressions would make it here. I can't approve though, you'll
need a maintainer to do that.
Sid, given you wrote this code and ACK'd, I'll rubber-stamp.
Ideally this would include a testcase we can use in our testsuite.
Given it triggers an ICE, just adding a suitable file to gcc.dg/torture
should be sufficient.
Given this changes target independent code it needs to be bootstrapped
and regression tested on one of the primary platforms. Most folks use
x86-64 or aarch64 for the bootstrap & regression test.
Can you do these things Linsen?
jeff
