The problem isn't because of the compiler version, it's because of floating-point arithmetic.

Floating-point arithmetics(especially, float) are usually very likely to be inaccurate, because, unlike integer types, floating-point types can’t represent exact values. Also, they can give different results on different compilers or machines. For example, some hardware architectures may support advanced floating-point operations, and some compilers may use different floating-point instructions depending on the system.

If you run the codes below by the “Custom Invocation” feature, you can see that GNU G++17 7.3.0 and GNU G++20 11.2.0 (64 bit, winlibs) give different results.

// Input: 900000000 909090909
// 4545455.0000000000 (GNU G++17 7.3.0)
// 4545469.0000000000 (GNU G++20 11.2.0 (64 bit, winlibs))
#include <bits/stdc++.h>
using namespace std;

int main() {
    int l, r;
    scanf("%d %d", &l, &r);
    printf("%.10f", (r - l + 1) - ((r / 2) - ceil((float)l / 2) + 1));
    return 0;
}

Don’t use floating-point types if you can go with integer types. If you absolutely need them, use double instead.