[Bug tree-optimization/114722] New: Missed optimization: !e*d*e=>0, affected by useless instructions

[Bug tree-optimization/114722] New: Missed optimization: !e*d*e=>0, affected by useless instructions

[652023330028 at smail dot nju.edu.cn]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114722

            Bug ID: 114722
           Summary: Missed optimization: !e*d*e=>0, affected by useless
                    instructions
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: 652023330028 at smail dot nju.edu.cn
  Target Milestone: ---

Hello, we noticed that the code below can be optimized as stated in the title
(!e*d*e=>0), but gcc -O3 -fwrapv missed it. It looks like it's affected by some
useless instructions.

https://godbolt.org/z/5vPqcbGWr

extern int x;
extern int y;
void func1(int a, int b, int c, int d, int e) {
    a = c;
    a = b;
    a = d;
    x = c;
    y = !e * d * e;
}

GCC -O3 -fwrapv:
func1(int, int, int, int, int):
        xor     eax, eax
        test    r8d, r8d
        mov     DWORD PTR x[rip], edx
        cmovne  ecx, eax
        imul    ecx, r8d
        mov     DWORD PTR y[rip], ecx
        ret


Expected code (when we remove 'a=c; a=b; a=d;'):
func1(int, int, int, int, int):
        mov     DWORD PTR x[rip], edx
        mov     DWORD PTR y[rip], 0
        ret

By the way, when the exact same fun1 and func2 appear together, func2 gets the
expected result:

extern int x;
extern int y;
void func1(int a, int b, int c, int d, int e) {
    a = c;
    a = b;
    a = d;
    x = c;
    y = !e * d * e;
}

void func2(int a, int b, int c, int d, int e) {
    a = c;
    a = b;
    a = d;
    x = c;
    y = !e * d * e;
}

GCC -O3 -fwrapv:
func1(int, int, int, int, int):
        xor     eax, eax
        test    r8d, r8d
        mov     DWORD PTR x[rip], edx
        cmovne  ecx, eax
        imul    ecx, r8d
        mov     DWORD PTR y[rip], ecx
        ret
func2(int, int, int, int, int):
        mov     DWORD PTR x[rip], edx
        mov     DWORD PTR y[rip], 0
        ret

Thank you very much for your time and effort! We look forward to hearing from
you.

[Bug tree-optimization/114722] Missed optimization: !e*d*e=>0, affected by useless instructions

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114722

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-04-15
           Severity|normal                      |enhancement
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |missed-optimization, TREE

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---

Confirmed.

Working case:
  x = c_6(D);
  _1 = e_8(D) == 0;
  _11 = e_8(D) * d_9(D);
  _4 = _1 ? _11 : 0;

vs not working case:

  x = c_5(D);
  _1 = e_10(D) == 0;
  _3 = _1 ? d_7(D) : 0;
  _4 = _3 * e_10(D);


In the first case, the rtl optimizers can figure it out back to 0. While in the
second case, it can't. 

Not a regression and has only been improved lately.