[clang] [compiler-rt] Reland "[Clang][CodeGen] Report when an alias points to an incompatible target" (PR #195550)

[Martin Storsjö via llvm-commits]

mstorsjo wrote:

Just for reference, I noticed a small discrepancy to which cases GCC warns about:
```c
struct A {
    float a;
};

struct B {
    long long a, b, c, d;
};

void ptrFunc(struct A *ptr) { }
void ptrFuncAlias(struct B *ptr) __attribute__((alias("ptrFunc")));
```
Here, the functions differ in the struct that the pointer parameter points to. GCC doesn't warn about this at all, but Clang does.

Actual cases in real world code where I ran into this obviously won't have quite this different structs, but do have structs that map to the exact same layout.

I think this probably is reasonable to warn about though, given other cases that both GCC and Clang warn about:
```c
typedef struct C {
    int a;
} C;

typedef struct D {
    int a;
} D;

C structfunc(int a) {
    C ret;
    ret.a = a;
    return ret;
}

D structfuncalias(int a) __attribute__((alias("structfunc")));
```

---

I also did note that neither GCC nor Clang warns if the functions take different types that are typedeffed to the same base type:
```c
typedef long long intmax_t;

intmax_t imaxfunc(intmax_t a) { return a; }

long long llfunc(long long a) __attribute__((alias("imaxfunc")));
```

So this applies to types, but differently named structs with identical contents aren't considered equal. (If I change `typedef struct D { int a; } D;` above into `typdef C D;`, then both GCC and Clang stop warning about it.)

See https://gcc.godbolt.org/z/W79cdcYTc for my comparison example in action.

https://github.com/llvm/llvm-project/pull/195550