[Clang] optimization: clang does not know memcpy returns its first arg

[rockeet]

Both compile with -O3

void foo(char* buf, const char* src, size_t n) {
    memcpy(buf, src, n);
    memcpy(buf + n, &n, sizeof(n));
}

gcc generate code:

foo(char*, char const*, unsigned long):
        push    rbx
        mov     rbx, rdx
        call    memcpy
        mov     QWORD PTR [rax+rbx], rbx
        pop     rbx
        ret

gcc knows memcpy returns its first arg.

clang generate code:

foo(char*, char const*, unsigned long):
        push    r14
        push    rbx
        push    rax
        mov     rbx, rdx
        mov     r14, rdi
        call    memcpy@PLT
        mov     qword ptr [r14 + rbx], rbx
        add     rsp, 8
        pop     rbx
        pop     r14
        ret

clang does not know memcpy returns its first arg. It is worse that even when we tell clang to use return value from memcpy:

void foo2(char* buf, const char* src, size_t n) {
    char* ptr = (char*)memcpy(buf, src, n);
    memcpy(ptr + n, &n, sizeof(n));
}

clang still generate same code as original foo. gcc also generate same code for foo and foo2.

[frederick-vs-ja]

This looks like duplicate of #46610.