https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123479
Bug ID: 123479
Summary: missed optimization for redundant conditional
computation
Product: gcc
Version: 16.0
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,
For the reduced test case below, GCC -O3 appears to generate a redundant
conditional computation.
Reduced code (from ABC project
https://github.com/berkeley-abc/abc/blob/c18b9a24de46d3b0209afa416511b7a1127344f9/src/sat/cadical/radix.hpp#L45):
https://godbolt.org/z/zEKoWx6Ta
void c(unsigned long* e, unsigned long* i) {
unsigned long n = i - e;
if (n <= 1)
return;
for (unsigned long g = 0; g < n; g++)
e[n] = g;
}
Optimized code of GCC -O3:
```
<bb 2> [local count: 178956971]:
_1 = i_6(D) - e_7(D);
_9 = (long unsigned int) _1;
if (_9 <= 8)
goto <bb 4>; [34.00%]
else
goto <bb 3>; [66.00%]
<bb 3> [local count: 118111600]:
_2 = _1 /[ex] 8;
_17 = (long unsigned int) _2;
_18 = _1 != 0; #redundant
_11 = (long unsigned int) _18;
g_13 = _17 - _11;
*i_6(D) = g_13;
<bb 4> [local count: 178956971]:
return;
```
GCC -O3:
"c(unsigned long*, unsigned long*)":
mov rax, rsi
sub rax, rdi
cmp rax, 8
jbe .L1
mov rdx, rax
sar rdx, 3
cmp rax, 1
mov rax, rdx
adc rax, -1
mov QWORD PTR [rsi], rax
.L1:
ret
Expected:
_Z1cPmS_:
.LFB0:
.cfi_startproc
movq %rsi, %rax
subq %rdi, %rax
cmpq $8, %rax
jbe .L1
sarq $3, %rax
subq $1, %rax
movq %rax, (%rsi)
.L1:
ret
