clang is suboptimal for -a * (-a * a)

[k-arrows]

Consider the following two functions.
https://godbolt.org/z/Taza3zT3G

int foo(int a) {
    return -a * (-a * a);
}

int bar(int a) {
    return (-a * -a) * a;
}

I expect the following output for both functions.

        mov     eax, edi
        imul    eax, edi
        imul    eax, edi
        ret

But interestingly, the output of Clang is better than that of GCC for the following function.
https://godbolt.org/z/17qvs9xcd

int foo1(int a, int b, int c) {
    return -a * (-b * c);
}

[Kmeakin]

Alive proof: https://alive2.llvm.org/ce/z/VQZh2L

[pinskia]

Note GCC only optimizes this at the RTL level and not the gimple level, filed https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113189 to optimize it at the gimple level too.