[Bug tree-optimization/82885] memcpy does not propagate aliasing knowledge

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82885

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |tree-optimization
           Severity|normal                      |enhancement

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The original testcase is fixed in GCC 8+:
  <bb 2> [local count: 1073741825]:
  __builtin_memcpy (ch1_3(D), ch2_4(D), 1048576);
  _1 = MEM[(const char *)ch2_4(D) + 2B];
  *ch1_3(D) = _1;
  MEM[(char *)ch1_3(D) + 1B] = _1;

fre1 is able to figure this out but this is not exactly propagating aliasing
info, just that the value of ch2[2] cannot change if ch1[0] is stored with the
same value of what was in ch2[2]; memcpy is not needed to show that :).

A better testcase which shows there is still a missing aliasing case:
void foo3(char* ch1, char* ch2) {
    __builtin_memcpy(ch1, ch2, 1024*1024);
    ch1[0] = ch2[2] + 1;
    ch1[1] = ch2[2];
}