[Bug c++/115350] New: Missing optimzation: fold `n = std::min(f ? 0 : 3, -a)` to `n = -a`

[Bug c++/115350] New: Missing optimzation: fold `n = std::min(f ? 0 : 3, -a)` to `n = -a`

[zhiwuyazhe154 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115350

            Bug ID: 115350
           Summary: Missing optimzation: fold `n = std::min(f ? 0 : 3,
                    -a)` to `n = -a`
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zhiwuyazhe154 at gmail dot com
  Target Milestone: ---

Godbolt Example: https://godbolt.org/z/j9rj5q6W5

Code Example:
unsigned short n;
#include <algorithm>
void fn1( unsigned short f, unsigned char a) {
    n = std::min(f ? 0 : 3, -a); // equals to "n = -a"
}

During the operation, -a will be converted to int type, and its value range is
[-255, 0], which must be less than 0 or 3, so “n = std::min(f ? 0 : 3, -a)” is
equivalent to "n = -a".

GCC -O3:
fn1(unsigned short, unsigned char):
        test    di, di
        je      .L5
        xor     eax, eax
        test    sil, sil
        jne     .L5
        mov     WORD PTR n[rip], ax
        ret
.L5:
        movzx   eax, sil
        neg     eax
        mov     WORD PTR n[rip], ax
        ret
n:
        .zero   2

Expected code (CLANG -O3):
fn1(unsigned short, unsigned char):                           
        neg     esi
        mov     word ptr [rip + n], si
        ret
n:
        .short  0

[Bug c++/115350] Missing optimzation: fold `n = std::min(f ? 0 : 3, -a)` to `n = -a`

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115350

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-06-05
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
What is more interesting is the order of evulation of the function arguments
make a huge difference.
Testcase:
```
unsigned short n;
#include <algorithm>
void fn1( unsigned short f, unsigned char a) {
    auto t1 = -a;
    auto t = f ? 0 : 3;
    n = std::min(t, t1); // equals to "n = -a"
}
void fn2( unsigned short f, unsigned char a) {
    auto t = f ? 0 : 3;
    auto t1 = -a;
    n = std::min(t, t1); // equals to "n = -a"
}
```

fn2 works while f1 does not.

fn2 used to be only optimized at the rtl level too ..

[Bug tree-optimization/115350] Missing optimzation: fold `n = std::min(f ? 0 : 3, -a)` to `n = -a`

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115350

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
          Component|c++                         |tree-optimization

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note also disabling evrp (-fdisable-tree-evrp) GCC is able to optimize both ...