https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104601
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rguenth at gcc dot gnu.org
--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I believe the bug is in visit_reference_op_call:
5241 /* If we value numbered an indirect functions function to
5242 one not clobbering memory value number its VDEF to its
5243 VUSE. */
5244 tree fn = gimple_call_fn (stmt);
5245 if (fn && TREE_CODE (fn) == SSA_NAME)
5246 {
5247 fn = SSA_VAL (fn);
5248 if (TREE_CODE (fn) == ADDR_EXPR
5249 && TREE_CODE (TREE_OPERAND (fn, 0)) == FUNCTION_DECL
5250 && (flags_from_decl_or_type (TREE_OPERAND (fn, 0))
5251 & (ECF_CONST | ECF_PURE)))
5252 vdef_val = vuse_ssa_val (gimple_vuse (stmt));
5253 }
5254 changed |= set_ssa_val_to (vdef, vdef_val);
stmt in this case is:
# .MEM_140 = VDEF <.MEM_96>
*__pred$__d_43 = _50 (_49);
and _50 value numbers to &f where f is a const function.
I think we can value number the vdef to vuse_ssa_val (gimple_vuse (stmt))
only if the lhs of the call is not present or is an SSA_NAME, when the call
(const or pure) acts as a store, we shouldn't do that.
A few lines before this there is also:
if (vnresult->result_vdef && vdef)
changed |= set_ssa_val_to (vdef, vnresult->result_vdef);
else if (vdef)
/* If the call was discovered to be pure or const reflect
that as far as possible. */
changed |= set_ssa_val_to (vdef, vuse_ssa_val (gimple_vuse (stmt)));
I wonder if that doesn't suffer from the same problem.
