https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108586
Bug ID: 108586
Summary: Using std::array instead of a plain array results in
different (worse?) code in array zero initialization
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: technicallyanonymous at proton dot me
Target Milestone: ---
This code, compiled with GCC 12.2 (-O2 -std=c++20):
#include <array>
int foo(char *arr);
int square(int num) {
std::array<char,96> arr = {};
return foo(arr.data());
}
results in this:
square(int):
sub rsp, 104
xor eax, eax
mov ecx, 12
mov rdi, rsp
rep stosq
mov rdi, rsp
call foo(char*)
add rsp, 104
ret
However, this very similar code:
int foo(char *arr);
int square(int num) {
char arr[96] = {};
return foo(arr);
}
is compiled as:
square(int):
sub rsp, 104
pxor xmm0, xmm0
mov rdi, rsp
movaps XMMWORD PTR [rsp], xmm0
movaps XMMWORD PTR [rsp+16], xmm0
movaps XMMWORD PTR [rsp+32], xmm0
movaps XMMWORD PTR [rsp+48], xmm0
movaps XMMWORD PTR [rsp+64], xmm0
movaps XMMWORD PTR [rsp+80], xmm0
call foo(char*)
add rsp, 104
ret
As you can see, SSE instructions are emitted only with a plain array. I also
tested clang and in both cases it emits movaps instructions.