[Bug tree-optimization/124902] sum reduction of 2 masked_loads not vectorized

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124902

--- Comment #2 from Drea Pinski <pinskia at gcc dot gnu.org> ---
Even this should be vectorizable:
```
unsigned int foo(
    const unsigned * __restrict__ a,
    const unsigned * __restrict__ b,
    const unsigned * __restrict__ c,
    int n
) {
    unsigned sum = 0;
    for (int i = 0; i<n; i++)
    {
        if (a[i] != 0 || b[i] != 0) 
        {
            sum += a[i];
            sum += b[i];
        }
    }
    return sum;
}
```

Note both GCC and LLVM does not vectorize the above case.

LLVM is able to handle this while GCC does not:
```

unsigned int foo(
    const unsigned * __restrict__ a,
    const unsigned * __restrict__ b,
    const unsigned * __restrict__ c,
    int n
) {
    unsigned sum = 0;
    for (int i = 0; i<n; i++)
    {
        int t = a[i] + b[i];
        if (a[i] != 0 || b[i] != 0) 
        {
            sum += t;
        }
    }
    return sum;
}
```

I think is_cond_scalar_reduction is improved to handle this case then these all
would be vectorized.