[Bug tree-optimization/84859] [8 Regression] bogus -Warray-bounds on a memcpy in a loop

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84859

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2018-03-14
      Known to work|                            |7.3.1
   Target Milestone|---                         |8.0
            Summary|bogus -Warray-bounds on a   |[8 Regression] bogus
                   |memcpy in a loop            |-Warray-bounds on a memcpy
                   |                            |in a loop
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The reason is a missed phiopt to MIN_EXPR for h and VRP not propagating through
memory.

we have

  <bb 4> [local count: 955630223]:
  if (n_6 > 255)
    goto <bb 5>; [50.00%]
  else
    goto <bb 6>; [50.00%]

  <bb 5> [local count: 477815112]:
  a[0] = 255;
  goto <bb 7>; [100.00%]

  <bb 6> [local count: 477815112]:
  _1 = (unsigned char) n_6;
  a[0] = _1;

  <bb 7> [local count: 955630223]:
  _2 = a[0];

where we lack a pass replacing that with

   <bb 7>
   # tem = PHI <255(5), _1(6)>
   a[0] = tem;
   _2 = a[0];

and then a MIN_EXPR.

Not sure why thw warn_restrict pass ends up with this kind of value-range
here though.  In the IL we have threaded the n > 255 check though and
ended up with an explicit memcpy (..., 255);  but that's not [9, 255].