[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #1 from rguenth at gcc dot gnu dot org 2009-01-25 13:21 --- As soon as we do not have to properly represent this in the virtual use-def chains (while still being semi-precise) we can deal with this properly. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38964
[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #2 from rguenth at gcc dot gnu dot org 2009-01-25 15:56 --- One thing to note is that the adjustment we do during inlining means that we miss that same adjustment iff inlining is disabled and the simple (placement) new wrappers are marked const (and thus do not appear as memory barriers in the virtual SSA form). -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Status|UNCONFIRMED |NEW Ever Confirmed|0 |1 Last reconfirmed|-00-00 00:00:00 |2009-01-25 15:56:59 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38964
[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #3 from rguenth at gcc dot gnu dot org 2009-01-25 16:02 --- Plan for attacking the problem: 1) Write a verifier that discovers illegal code motion. 1a) Each store and load is assigned a generation count. 1b) After code motion optimizations verify that out-of-order stores/loads were validly interchanged. If not, ICE. If so, re-compute the generation counters. 2) Fix the fallout. Tree loop store motion is known to randomly interchange stores. 3) Profit. ... n) Deal with RTL (properly export alias information to RTL). -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38964
[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #4 from rguenth at gcc dot gnu dot org 2009-02-02 10:24 ---
Nice testcase from PR38503:
inline void *operator new (__SIZE_TYPE__, void *__p) throw() { return __p; }
struct Y {
Y() {}
int i;
};
struct X {
X() {}
void construct(const Y& y)
{
new (&m_data[0]) Y(y);
}
template
Y& get() { return reinterpret_cast(m_data); }
bool initialized;
char m_data[sizeof (Y)];
};
X x;
void bar(const X&);
void foo(Y& y)
{
x.get() = y;
x.initialized = true;
bar(x);
}
we have to assume that any globally accessible decl has a different dynamic
type than its declared type. Thus we may not apply TBAA to any ESCAPED
symbol.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38964
[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #5 from rguenth at gcc dot gnu dot org 2009-02-03 09:08 --- The testcase lays the way to proof that whatever PTA ends up putting in the ESCAPED solution (which includes all globals) TBAA disambiguation on these is invalid. This basically means that TBAA is nearly useless, as if PTA has more precise information we can as well use that. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38964
[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #6 from rguenth at gcc dot gnu dot org 2009-05-22 20:41 --- Subject: Bug 38964 Author: rguenth Date: Fri May 22 20:41:31 2009 New Revision: 147805 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=147805 Log: 2009-05-22 Richard Guenther PR middle-end/38964 * alias.c (write_dependence_p): Do not use TBAA for answering anti-dependence or output-dependence. * tree-ssa-structalias.c (set_uids_in_ptset): Remove TBAA pruning code. (emit_pointer_definition): Remove. (emit_alias_warning): Likewise. (find_what_var_points_to): Remove TBAA pruning code. (find_what_p_points_to): Likewise. Do not warn about strict-aliasing violations. (compute_points_to_sets): Remove code computing the set of dereferenced pointers. * tree-data-ref.c (dr_may_alias_p): Properly use the split oracle for querying anti and output dependencies. * tree-ssa-alias.c (refs_may_alias_p_1): Add argument specifying if TBAA may be applied. (refs_anti_dependent_p): New function. (refs_output_dependent_p): Likewise. * tree-ssa-alias.h (refs_anti_dependent_p): Declare. (refs_output_dependent_p): Likewise. * doc/tree-ssa.texi (Memory model): New section. testsuite/ * g++.dg/warn/Wstrict-aliasing-float-ref-int-obj.C: XFAIL. * gcc.dg/Wstrict-aliasing-converted-assigned.c: Likewise. * gcc.dg/Wstrict-aliasing-float-ptr-int-obj.c: Likewise. * doc/c-tree.texi (CHANGE_DYNAMIC_TYPE_EXPR): Remove. * doc/gimple.texi (GIMPLE_CHANGE_DYNAMIC_TYPE): Remove. * cfgexpand.c (expand_gimple_basic_block): Do not handle GIMPLE_CHANGE_DYNAMIC_TYPE or CHANGE_DYNAMIC_TYPE_EXPR. * expr.c (expand_expr_real_1): Likewise. * gimple-low.c (lower_stmt): Likewise. * gimple-pretty-print.c (dump_gimple_stmt): Likewise. (dump_gimple_cdt): Remove. * gimple.c (gss_for_code): Do not handle GIMPLE_CHANGE_DYNAMIC_TYPE. (gimple_size): Likewise. (walk_gimple_op): Likewise. (is_gimple_stmt): Likewise. (walk_stmt_load_store_addr_ops): Likewise. (gimple_build_cdt): Remove. * gimple.def (GIMPLE_CHANGE_DYNAMIC_TYPE): Remove. * gimple.h (gimple_cdt_new_type): Remove. (gimple_cdt_new_type_ptr): Likewise. (gimple_cdt_set_new_type): Likewise. (gimple_cdt_location): Likewise. (gimple_cdt_location_ptr): Likewise. (gimple_cdt_set_location): Likewise. * gimplify.c (gimplify_expr): Do not handle CHANGE_DYNAMIC_TYPE_EXPR. * tree-cfg.c (remove_useless_stmts_1): Do not handle GIMPLE_CHANGE_DYNAMIC_TYPE. (verify_types_in_gimple_stmt): Likewise. * tree-inline.c (estimate_num_insns): Likewise. (expand_call_inline): Do not copy DECL_NO_TBAA_P. (copy_decl_to_var): Likewise. (copy_result_decl_to_var): Likewise. * tree-pretty-print.c (dump_generic_node): Do not handle CHANGE_DYNAMIC_TYPE_EXPR. * tree-ssa-dce.c (mark_stmt_if_obviously_necessary): Likewise. * tree-ssa-operands.c (get_expr_operands): Likewise. * tree-ssa-structalias.c (struct variable_info): Remove no_tbaa_pruning member. (new_var_info): Do not set it based on DECL_NO_TBAA_P. (unify_nodes): Do not copy it. (find_func_aliases): Do not handle GIMPLE_CHANGE_DYNAMIC_TYPE. (dump_solution_for_var): Do not dump no_tbaa_pruning state. (set_uids_in_ptset): Do not check it. (find_what_var_points_to): Likewise. (compute_tbaa_pruning): Remove. (compute_points_to_sets): Do not call it. * tree.c (walk_tree_1): Do not handle CHANGE_DYNAMIC_TYPE_EXPR. * tree.def (CHANGE_DYNAMIC_TYPE_EXPR): Remove. * tree.h (CHANGE_DYNAMIC_TYPE_NEW_TYPE): Remove. (CHANGE_DYNAMIC_TYPE_LOCATION): Likewise. (DECL_NO_TBAA_P): Likewise. (struct tree_decl_common): Move no_tbaa_flag to unused flags section. * omp-low.c (copy_var_decl): Do not copy DECL_NO_TBAA_P. (expand_omp_atomic_pipeline): Do not set it. * print-tree.c (print_node): Do not dump it. * tree-ssa-copyrename.c (copy_rename_partition_coalesce): Remove redundant check. cp/ * init.c (avoid_placement_new_aliasing): Remove. (build_new_1): Do not call it. Modified: trunk/gcc/ChangeLog trunk/gcc/alias.c trunk/gcc/cfgexpand.c trunk/gcc/cp/ChangeLog trunk/gcc/cp/init.c trunk/gcc/doc/c-tree.texi trunk/gcc/doc/gimple.texi trunk/gcc/doc/tree-ssa.texi trunk/gcc/expr.c trunk/gcc/gimple-low.c trunk/gcc/gimple-pretty-print.c trunk/gcc/gimple.c trunk/gcc/gimple.def trunk/gcc/gimple.h trunk/gcc/gimplify.c trunk/gcc/omp-low.c trunk/gcc/print-tree.c trunk/gcc/testsuite/ChangeLog trunk/gcc/tes
[Bug tree-optimization/38964] TBAA side-effects of C++ new still missing
--- Comment #7 from rguenth at gcc dot gnu dot org 2009-05-22 20:42 --- Fixed for 4.5. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Status|NEW |RESOLVED Resolution||FIXED Target Milestone|--- |4.5.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38964
