https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115483
Bug ID: 115483
Summary: Missed optimization for loop invariant code
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: zhiwuyazhe154 at gmail dot com
Target Milestone: ---
Godbolt example: https://godbolt.org/z/s75d4GfTT
code example:
unsigned m,n;
unsigned func(unsigned a, unsigned b,
unsigned c, unsigned d)
{
for(unsigned i=0; i<100; i++)
{
m += 1024;
b += c/n + (c + d) + m;
m += m;
}
return b;
}
In this case, c/n + (c + d) is loop invariant code.
It seems that reassociate breaks the optimization of licm.
There are a few other issues that mention the shortcomings of reassociate, but
there doesn't seem to be an issue that shows its negative impact.
GCC -O3:
func(unsigned int, unsigned int, unsigned int, unsigned int):
mov eax, edx
mov r9d, edx
xor edx, edx
mov r8d, esi
div DWORD PTR n[rip]
mov esi, DWORD PTR m[rip]
mov edi, 100
.L2:
add esi, 1024
lea edx, [rsi+rcx]
add esi, esi
add edx, r9d
add edx, eax
add r8d, edx
sub edi, 1
jne .L2
mov DWORD PTR m[rip], esi
mov eax, r8d
ret
