https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113301
Bug ID: 113301
Summary: [12/13/14 Regression] Missed optimization: (1/(x+1))/2
=> 0 since gcc-12
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 GCC seems to have missed optimizations as stated in the
title, and it works as expected on gcc-11.4.
After analysis, we found that they (gcc-trunk and gcc-11.4) are different in
evrp(tree), and further, the difference is in the calculation of the value
range.
reduced code:
https://godbolt.org/z/Y7v1jsTKf
int c;
void func(int x){
c=(1/(x+1))/2;
}
gcc(trunk) -O3 -fwrapv:
func(int):
lea edx, [rdi+1]
add edi, 2
xor eax, eax
cmp edi, 2
cmova edx, eax
mov eax, edx
shr eax, 31
add eax, edx
sar eax
mov DWORD PTR c[rip], eax
ret
evrp (tree):
Folding statement: _7 = _6 <= 2;
Not folded
Folding statement: _2 = _7 ? _1 : 0;
Possible COND_EXPR adjustment. Range op1 : [irange] int VARYING and Range op2:
[irange] int [0, 0] NONZERO 0x0
Not folded
Folding statement: _3 = _2 / 2;
Global Exported: _3 = [irange] int [-1073741824, 1073741823]
Not folded
Folding statement: c = _3;
Not folded
Folding statement: return;
Not folded
Expected code:
gcc(11.4) -O3 -fwrapv:
func(int):
mov DWORD PTR c[rip], 0
ret
evrp (tree):
=========== BB 2 ============
<bb 2> :
# DEBUG BEGIN_STMT
_1 = x_4(D) + 1;
_2 = 1 / _1;
c = 0;
return;
_2 : int [-1, 1]
Non-varying global ranges:
=========================:
_2 : int [-1, 1]
Also, the following code works as expected (gcc-trunk):
void func1(int x){
c=(1/x)/2;
}
Thank you very much for your time and effort! We look forward to hearing from
you.