[Bug target/105670] New: [x86] suboptimal code for branch over two bools

[Bug target/105670] New: [x86] suboptimal code for branch over two bools

[fent at in dot tum.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105670

            Bug ID: 105670
           Summary: [x86] suboptimal code for branch over two bools
           Product: gcc
           Version: 12.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: fent at in dot tum.de
  Target Milestone: ---

void test(bool a, bool b, int* t) {
  if (a & b)
    *t = 42;
}

With -O1, -O2, -O3, and -Os this produces:

test(bool, bool, int*):
        test    dil, dil
        je      .L1
        test    sil, sil
        je      .L1
        mov     DWORD PTR [rdx], 42
.L1:
        ret

The following would be 5B shorter and also faster, since it minimizes branch
misses:

test(bool, bool, int*):
        test    sil, dil
        je      .L1
        mov     DWORD PTR [rdx], 42
.L1:
        ret

Note that this is (somewhat) ABI dependent, but works on x86-64 System V,
since:
> bit 0 contains the truth value and bits 1 to 7 shall be zero

[Bug target/105670] [x86] suboptimal code for branch over two bools

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105670

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-05-20
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Gimple level is good:
  _7 = a_2(D) & b_3(D);
  if (_7 != 0)
    goto <bb 3>; [50.00%]
  else
    goto <bb 4>; [50.00%]

But at expand it changes the above to:
(insn 11 10 12 (set (reg:CCZ 17 flags)
        (compare:CCZ (reg/v:QI 83 [ a ])
            (const_int 0 [0]))) "/app/example.cpp":2:3 -1
     (nil))

(jump_insn 12 11 13 (set (pc)
        (if_then_else (eq (reg:CCZ 17 flags)
                (const_int 0 [0]))
            (label_ref 0)
            (pc))) "/app/example.cpp":2:3 -1
     (int_list:REG_BR_PROB 268435460 (nil)))

(insn 13 12 14 (set (reg:CCZ 17 flags)
        (compare:CCZ (reg/v:QI 85 [ b ])
            (const_int 0 [0]))) "/app/example.cpp":2:3 -1
     (nil))

(jump_insn 14 13 0 (set (pc)
        (if_then_else (eq (reg:CCZ 17 flags)
                (const_int 0 [0]))
            (label_ref 0)
            (pc))) "/app/example.cpp":2:3 -1
     (int_list:REG_BR_PROB 357913948 (nil)))


I noticed LLVM does the same as GCC here ...
While ICC does the one test.

[Bug target/105670] [x86] suboptimal code for branch over two bools

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105670

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement

[Bug target/105670] [x86] suboptimal code for branch over two bools

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105670

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Dup of bug 79045.

*** This bug has been marked as a duplicate of bug 79045 ***