[Bug target/113255] [11/12/13/14 Regression] wrong code with -O2 -mtune=k8

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113255

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
diff --git a/gcc/alias.cc b/gcc/alias.cc
index b2ec4806d22..0150dd699db 100644
--- a/gcc/alias.cc
+++ b/gcc/alias.cc
@@ -2272,6 +2272,8 @@ static bool
 base_alias_check (rtx x, rtx x_base, rtx y, rtx y_base,
                  machine_mode x_mode, machine_mode y_mode)
 {
+  return 1;
+
   /* If the address itself has no known base see if a known equivalent
      value has one.  If either address still has no known base, nothing
      is known about aliasing.  */

(an experiment I did many years ago already) gives clean testresults
besides

+FAIL: gcc.dg/guality/pr41447-1.c   -O2  -DPREVENT_OPTIMIZATION  execution test
+FAIL: gcc.dg/guality/pr41447-1.c   -O3 -g  -DPREVENT_OPTIMIZATION  execution
test
+FAIL: gcc.dg/guality/pr41447-1.c   -Os  -DPREVENT_OPTIMIZATION  execution test
+FAIL: gcc.dg/guality/pr41447-1.c   -O2 -flto -fno-use-linker-plugin
-flto-partition=none  -DPREVENT_OPTIMIZATION execution test

and only minor change in code generation:

   text    data     bss     dec     hex filename
48374841          64824 1939512 50379177        300b9a9 ../obj2/gcc/cc1plus
48375065          64824 1939512 50379401        300ba89 gcc/cc1plus

where the larger binary is the patched one.

Assembly-wise there are scheduling changes, missed scheduling over spills.

Recovering this and similar cases should be as easy as marking spill
MEMs with a flag (in MEM_EXPR) for example, distinguishing (classes of)
stack memory from the rest.  We should have this already (spill_slot_decl,
set_mem_attrs_for_spill), not sure why it doesn't look effective.

Improving test coverage for desired transforms would be nice as well.