[Bug c++/49279] [4.5/4.6/4.7 Regression] Optimization incorrectly presuming constant variable inside loop in g++ 4.5 and 4.6 with -O2 and -O3 for x86_64 targets

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49279

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-06-06 
14:03:27 UTC ---
  # PT = nonlocal escaped
  D.7613_101 = MEM[(struct ei_matrix_storage *)this_30(D)];
  # PT = nonlocal escaped
  D.7616_104 = D.7613_101 + D.7582_90;
  # PT = nonlocal escaped { D.7934 } (restr)
  SR.167_105 = (const Scalar * restrict) D.7616_104;

  # PT = nonlocal escaped
  D.7671_127 = MEM[(struct ei_matrix_storage *)this_30(D)];
  D.7672_129 = D.7554_68 * 8;
  # PT = nonlocal escaped
  D.7674_130 = D.7671_127 + D.7672_129;
  # PT = nonlocal escaped { D.7936 } (restr)
  SR.169_131 = (const Scalar * restrict) D.7674_130;

so while the fix for PR48764 makes this_30 point to a single tag that
would make derived pointers derived it doesn't seem to work for this
case.  I have a variant that does.