[Bug target/97734] GCC using branches when a conditional move would be better

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97734

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-12-15
     Ever confirmed|0                           |1

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note ICX produces a branch too just like LLVM.
MSVC produces:
$LL2@branchfree:
; Line 12
        mov     eax, DWORD PTR [r8+r10*4]
        xor     r9d, r9d
        cmp     rcx, rax
        seta    r9b
        inc     r9
        lea     r10, QWORD PTR [r9+r10*2]
        cmp     r10, rbx
        jb      SHORT $LL2@branchfree


The reason why GCC does not produce the cmov is because gimple produces:
  if (x_20(D) <= _27)
    goto <bb 4>; [50.00%]
  else
    goto <bb 5>; [50.00%]

  <bb 4> [local count: 477815112]:
  _23 = i_31 * 2;
  iftmp.0_17 = _23 + 1;
  goto <bb 6>; [100.00%]

  <bb 5> [local count: 477815112]:
  _25 = i_31 + 1;
  iftmp.0_24 = _25 * 2;

  <bb 6> [local count: 955630224]:
  # i_11 = PHI <iftmp.0_24(5), iftmp.0_17(4)>

If we write the code like:

        uint64_t t22 = 2*i + 1;
        i = (x <= a[i]) ? t22 : t22+1;

Then GCC will produce what MSVC produces (MSVC will still produce it too).
If we add unused variable like:

        uint64_t t22 = 2*i + 1;

GCC will produce the conditional move.

I can't get LLVM to produce either the setb/seta nor conditional move either.