https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112123
Bug ID: 112123
Summary: Missed optimization of loop deletion because loop
invariants are not identified
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 in the following loop, a+b is a loop invariant, so the
value of m can be computed directly. The values of a and b can also be computed
without loop iteration.
https://godbolt.org/z/6xTE7zrPj
int m;
int a, b;
void test() {
for (int c = 0; c < 4087153; c += 1) {
m = a + b;
a += b;
b += -b;
}
}
GCC:
test():
mov edx, DWORD PTR a[rip]
mov ecx, DWORD PTR b[rip]
mov eax, 4087153
.L2:
add edx, ecx
xor ecx, ecx
sub eax, 1
jne .L2
mov DWORD PTR m[rip], edx
mov DWORD PTR a[rip], edx
mov DWORD PTR b[rip], 0
ret
Expected code (Clang):
test(): # @test()
mov eax, dword ptr [rip + b]
add eax, dword ptr [rip + a]
mov dword ptr [rip + a], eax
mov dword ptr [rip + b], 0
mov dword ptr [rip + m], eax
ret
Thank you very much for your time and effort! We look forward to hearing from
you.
