https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118129
Bug ID: 118129
Summary: Missed optimization for division with constant
comparison x / y CMP C0
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: antoshkka at gmail dot com
Target Milestone: ---
Consider the example:
bool test0(unsigned x, unsigned y) {
if (y == 0) __builtin_unreachable();
return x / y < 3;
}
The sample produces an assembly with an expensive `div` instruction:
test0(unsigned int, unsigned int):
mov eax, edi
xor edx, edx
div esi
cmp eax, 2
setbe al
ret
However, knowing that `y` is greater than 0 both sides of the equation could be
multiplied by `y` and divided by `3`, leading to an equivalent code:
bool test01(unsigned x, unsigned y) {
if (y == 0) __builtin_unreachable();
return x / 3 < y;
}
The assembly becomes better:
test01(unsigned int, unsigned int):
mov eax, 2863311531
mov edi, edi
imul rdi, rax
shr rdi, 33
cmp edi, esi
setb al
ret
Moreover, knowing that `y <= std::numeric_limits<unsigned>::max() / 3` the
function could be optimized even further to an equivalent of
bool test02(unsigned x, unsigned y) {
return x < y * 3;
}
That produces a very short and fast assembly:
test02(unsigned int, unsigned int):
lea eax, [rsi+rsi*2]
cmp edi, eax
setb al
ret
Godbolt playground: https://godbolt.org/z/d7qereKsb
