[Bug tree-optimization/91425] Ordered compares aren't optimised

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91425

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2019-08-12
          Component|middle-end                  |tree-optimization
            Version|unknown                     |10.0
            Summary|Ordered compares aren't     |Ordered compares aren't
                   |optimised properly          |optimised
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  No idea why improving GIMPLE is a requirement for making targets
behave better though.

Btw, on x86_64 ll and llx would have to be handled by phiopt since we have
for example

ll (double a, double b)
{
  _Bool _6;
  _Bool prephitmp_8;

  <bb 2> [local count: 1073741824]:
  if (a_3(D) u>= b_4(D))
    goto <bb 3>; [50.00%]
  else
    goto <bb 4>; [50.00%]

  <bb 3> [local count: 536870913]:
  _6 = a_3(D) < b_4(D);

  <bb 4> [local count: 1073741824]:
  # prephitmp_8 = PHI <_6(3), 1(2)>
  return prephitmp_8;


maybe_fold_or_comparisons could be used here eventually.  Btw. it's the
same missed optimization for non-ordered:

_Bool lt(double a, double b) { return a > b; }
_Bool lo(double a, double b) { return a < b; }
_Bool ll(double a, double b) { return lt(a,b) || lo(a,b); }
_Bool lx(double a, double b) { return lo(a,b) || lt(a,b); }

where we end up with

  <bb 2> [local count: 1073741824]:
  if (a_3(D) < b_4(D))
    goto <bb 4>; [50.00%]
  else
    goto <bb 3>; [50.00%]

  <bb 3> [local count: 536870913]:
  _6 = a_3(D) > b_4(D);

  <bb 4> [local count: 1073741824]:
  # prephitmp_8 = PHI <_6(3), 1(2)>

instead of != compares.