[Bug tree-optimization/111724] New: [Regression] Missed optimizations probably because of too early arithmetic optimization

[Bug tree-optimization/111724] New: [Regression] Missed optimizations probably because of too early arithmetic optimization

[652023330028 at smail dot nju.edu.cn]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111724

            Bug ID: 111724
           Summary: [Regression] Missed optimizations probably because of
                    too early arithmetic optimization
           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 found some optimizations (probably because of too early arithmetic
optimization) that GCC may have missed. We would greatly appreicate if you can
take a look and let us know what you think.

Given the following code:
https://godbolt.org/z/8EW8fx78K
int n;
void func(int w, int a, int b){
    for(int i=0;i<w;i++){
        n += (a+a)+(b+b);
        b += b;
    }
}

`a+a` is a loop invariant. But gcc-trunk -O3:
func(int, int, int):
        push    {r4, lr}
        subs    lr, r0, #0
        ble     .L1
        movw    r4, #:lower16:.LANCHOR0
        movt    r4, #:upper16:.LANCHOR0
        mov     ip, #0
        ldr     r0, [r4]
.L3:
        adds    r3, r1, r2
        add     ip, ip, #1
        lsls    r2, r2, #1
        cmp     lr, ip
        add     r0, r0, r3, lsl #1
        bne     .L3
        str     r0, [r4]

From original(tree):
   (void) (n = (a + b) * 2 + n) >>>>>;

This leads to:
  _1 = a_11(D) + b_18;
  _2 = _1 * 2;
  _4 = _2 + n_lsm.4_7;

So, it looks like the missed LICM is due to too early arithmetic optimization.

On gcc-6.4, it works as expected.

We found that this also affects other optimizations, such as common
subexpression elimination. We can provide examples if required.

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

[Bug tree-optimization/111724] [Regression] Missed optimizations probably because of too early arithmetic optimization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111724

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2023-10-09
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Yup, it's difficult.  reassoc doesn't handle signed arithmetic, that's usually
the pass that optimizes association for invariant motion.

[Bug tree-optimization/111724] [Regression] Missed optimizations probably because of too early arithmetic optimization

[652023330028 at smail dot nju.edu.cn]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111724

--- Comment #2 from Yi <652023330028 at smail dot nju.edu.cn> ---
(In reply to Richard Biener from comment #1)
> Yup, it's difficult.  reassoc doesn't handle signed arithmetic, that's
> usually the pass that optimizes association for invariant motion.

Yes. This issue has the same result for 'unsigned'.