[Bug tree-optimization/96244] Redudant mask load generated

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96244

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2020-07-20
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |missed-optimization

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
So one thing to note here is that a masked load of a by-value parameter decl
doesn't make much sense since a load from it cannot trap (unless there's
out-of-bound accesses - which I'm not sure we may simply disregard here).

if-conversion sees

VIEW_CONVERT_EXPR<int[8]>(d)[i_16]

and indeed considers it

iftmp.0_11 = VIEW_CONVERT_EXPR<int[8]>(c)[i_17];
tree could trap...

  <bb 3> [local count: 954449105]:
  # RANGE [0, 8] NONZERO 15
  # i_17 = PHI <0(2), i_13(8)>
  # ivtmp_6 = PHI <8(2), ivtmp_4(8)>
  _1 = VIEW_CONVERT_EXPR<int[8]>(a)[i_17];
  _2 = VIEW_CONVERT_EXPR<int[8]>(b)[i_17];

so range-info is one index too pessimistic here.  So IMHO it's not about
"redundant" masked loads, it's about the fact that we end up with loads
at all here.  If c and d would not be register arguments we would have to
perform loads and if they might trap we could not elide the masked load.