https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123979
Bug ID: 123979
Summary: complex lowering should update vop manually
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: internal-improvement
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Target Milestone: ---
As mentioned in
https://gcc.gnu.org/pipermail/gcc-patches/2026-February/707575.html.
Complex lowering should update the vop for the new load/stores instead of
relying on the renamer.
The first thing is extract_component needs to be split into 2 pieces really.
!gimple_p and gimple_p case.
The !gimple_p case should only be handling COMPLEX_CST/SSA_NAME .
With that in mind, this code:
```
else if (is_gimple_assign (stmt)
&& (gimple_assign_rhs_code (stmt) == REALPART_EXPR
|| gimple_assign_rhs_code (stmt) == IMAGPART_EXPR)
&& TREE_CODE (lhs) == SSA_NAME)
{
rhs = gimple_assign_rhs1 (stmt);
rhs = extract_component (gsi, TREE_OPERAND (rhs, 0),
gimple_assign_rhs_code (stmt)
== IMAGPART_EXPR,
false);
gimple_assign_set_rhs_from_tree (gsi, rhs);
stmt = gsi_stmt (*gsi);
update_stmt (stmt);
}
```
Should be restricted to `TREE_CODE (TREE_OPERAND (rhs, 0)) == SSA_NAME` as it
is just creating a copy of the rhs for no reason.
Note I am not even sure `TREE_CODE (lhs) == SSA_NAME` will never be false here
either.
