[Bug middle-end/115924] Failure to optimize back and forth shifts of several operands by the same amount with operations in the middle to ands

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115924

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note better example:
```
#include <stdint.h>

uint32_t f(uint32_t i2, uint32_t aa, uint32_t aa1)
{
    return ((i2 >> 17) + (aa >> 17) + (aa1 >> 17)) << 17;
}

uint32_t fa(uint32_t i2, uint32_t aa, uint32_t aa1)
{
    return ((i2 >> 17) + (aa1 >> 17) + (aa >> 17)) << 17;
}

uint32_t f2(uint32_t i2, uint32_t aa, uint32_t aa1)
{
    return ((((i2 & -(1 << 17)) + aa) & -(1 << 17)) + aa1) & -(1 << 17);
}
uint32_t f2a(uint32_t i2, uint32_t aa, uint32_t aa1)
{
    return ((((i2 & -(1 << 17)) + aa1) & -(1 << 17)) + aa) & -(1 << 17);
}

bool f3a(uint32_t i2, uint32_t aa, uint32_t aa1)
{
    return f(i2, aa, aa1) == fa(i2, aa, aa1);
}
bool f3(uint32_t i2, uint32_t aa, uint32_t aa1)
{
    return f(i2, aa, aa1) == f2a(i2, aa, aa1);
}
```

GCC is able to optimize f3a but clang/LLVM does not.