| Issue |
86949
|
| Summary |
Missed optimization (inline breaks knowledge about [[pure]])
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
kelbon
|
This code (intuitively) seams easy to optimize: function 'sum_foo' is pure, it does not change its arguments, so its result will be always same. Expected calculate one time + multiply by 100'000
But looks like compiler inlines 'sum_foo' and forgets, that it is [[pure]]. Then it tries to do all smart things about loops, but do not understands, that loop does not change its body
So, i added [[gnu::noinline]] and both gcc and clang create 100'000 * sum_foo, as expected.
```asm
foo(int*, int): # @foo(int*, int)
push rax
call sum_foo(int const*, int)
imul rax, rax, 100000
pop rcx
ret
```
I think it is very confusing behavior and may be fixed
```cpp
[[gnu::pure]] [[gnu::noinline]] static long long sum_foo(const int* data, int arraySize) noexcept {
long long sum = 0;
for (int c = 0; c < arraySize; ++c)
sum += data[c];
return sum;
}
long long foo(int* data, int arraySize) noexcept {
long long sum = 0;
for (int i = 0; i < 100000; ++i)
sum += sum_foo(data, arraySize);
return sum;
}
```
https://godbolt.org/z/53Y8P34dx
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
