https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117043
Bug ID: 117043
Summary: missed vectorization opportunity: data[i] = data[i -
a[i]] + 1; (a[i]=0)
Product: gcc
Version: 15.0
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 there seems to be a missing vectorization for the code
below.
reduced code: https://godbolt.org/z/zxKhKWxGq
int data[100];
void f(int a[100])
{
for(int i = 0; i < 100; i++){
a[i] = 0;
}
for(int i = 0; i < 100; i++){
data[i] = data[i - a[i]] + 1;
}
}
GCC -O3 -fno-vect-cost-model:
f(int*):
mov QWORD PTR [rdi], 0
mov rsi, rdi
lea rdi, [rdi+8]
xor eax, eax
mov QWORD PTR [rdi+384], 0
mov rcx, rsi
and rdi, -8
sub rcx, rdi
add ecx, 400
shr ecx, 3
rep stosq
.L2:
mov edx, eax
sub edx, DWORD PTR [rsi+rax*4]
movsx rdx, edx
mov edx, DWORD PTR data[0+rdx*4]
add edx, 1
mov DWORD PTR data[0+rax*4], edx
add rax, 1
cmp rax, 100
jne .L2
ret
<bb 3> [local count: 1063004408]:
# ivtmp.6_33 = PHI <ivtmp.6_37(3), 0(2)>
# DEBUG i => NULL
i_26 = (int) ivtmp.6_33;
# DEBUG i => i_26
# DEBUG BEGIN_STMT
_7 = MEM[(int *)a_16(D) + ivtmp.6_33 * 4];
_8 = i_26 - _7;
_9 = data[_8];
_10 = _9 + 1;
MEM[(int *)&data + ivtmp.6_33 * 4] = _10;
Expected code: vectorize the loop body.
For example:
<bb 3> [local count: 265751102]:
# ivtmp.14_19 = PHI <ivtmp.14_21(3), ivtmp.14_4(2)>
# DEBUG BEGIN_STMT
_5 = (void *) ivtmp.14_19;
vect__4.7_22 = MEM <vector(4) int> [(int *)_5];
vect__5.8_20 = vect__4.7_22 + { 1, 1, 1, 1 };
MEM <vector(4) int> [(int *)_5] = vect__5.8_20;
# DEBUG BEGIN_STMT
# DEBUG BEGIN_STMT
ivtmp.14_21 = ivtmp.14_19 + 16;
if (_11 != ivtmp.14_21)
goto <bb 3>; [95.96%]
else
goto <bb 4>; [4.04%]
Thank you very much for your time and effort! We look forward to hearing from
you.