Branchy vs. branchless code on unpredictable data

arr[i]'s low bit is effectively random (no exploitable pattern) for every element. Version A uses a data-dependent branch; version B computes the same result with unconditional bitwise arithmetic every iteration (strictly more ALU ops, but no branch). For large, random input, which runs faster, and why?

main.cpp
// Version A: data-dependent branch
for (int i = 0; i < n; ++i) {
    if (arr[i] & 1) sum += arr[i];
    else            sum -= arr[i];
}

// Version B: branchless
for (int i = 0; i < n; ++i) {
    int mask = -(arr[i] & 1);              // all-1s if odd, all-0s if even
    sum += (arr[i] & mask) - (arr[i] & ~mask);
}

25 of 47 in Computer Architecture