------- Comment #20 from dberlin at gcc dot gnu dot org 2007-07-16 22:29 ------- Subject: Re: [4.2/4.3 Regression] -fstrict-aliasing causes skipped code
Oh, for 4.2 you need to add make_constraint_to_escaped_var On 16 Jul 2007 15:51:44 -0000, rguenth at gcc dot gnu dot org <[EMAIL PROTECTED]> wrote: > > > ------- Comment #19 from rguenth at gcc dot gnu dot org 2007-07-16 15:51 > ------- > The backported patch causes > > FAIL: gcc.c-torture/execute/pr28778.c execution, -O2 > FAIL: gcc.c-torture/execute/pr28778.c execution, -Os > > DCE removes > > - > -<L1>:; > - list[3] = 42; > > we have one added constraint: > > blist.0_7 = blist_1 > +*blist.0_7 = &ANYTHING > ESCAPED_VARS = blist.0_7 > > IL differences after alias1 are > > @@ -251,12 +247,11 @@ > list[3] = 42; > blist_6 = &list; > > - # list_8 = PHI <list_4(3), list_5(4)>; > # blist_1 = PHI <blist_3(3), blist_6(4)>; > <L2>:; > blist.0_7 = (const GLint *) blist_1; > - # list_10 = V_MAY_DEF <list_8>; > - # NONLOCAL.44_11 = V_MAY_DEF <NONLOCAL.44_9>; > + # NONLOCAL.44_10 = V_MAY_DEF <NONLOCAL.44_8>; > + # SMT.45_11 = V_MAY_DEF <SMT.45_9>; > aglChoosePixelFormat (blist.0_7); > return; This is something that was fixed on 4.3 by proper SMT pruning. What is happening is it decides to prune the list access even though it's illegal to do so. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32328
