`<algorithm>`: std::uninitialized_copy used memmove instead of memcpy

[TymianekPL]

Describe the bug

std::uninitialized_copy uses memmove instead of memcpy. This is a codegen issue. memcpy is faster as it is an intrinsic.

Command-line test case

C:\Temp>type repro.cpp
#include <iostream>
struct thirtytwo {
    char bytes[32];
};

void f(thirtytwo& target, const thirtytwo& source) {
    std::uninitialized_copy(std::begin(source.bytes), std::end(source.bytes), std::begin(target.bytes));
}

Expected behavior

memcpy

Additional context

Discussed in MSVC STL discord server in #stl

[frederick-vs-ja]

UCRT implements memcpy in the same way as memmove. I'm not sure whether switching to memcpy will improve performance for our implementation.

[AlexGuteniev]

MSVC implements memcpy but not memmove as intrinsic. See /Oi and #pragma intrinsic.
In practice this means significant optimization of memcpy for any small size known at compile time.

When intrinsic memcpy optimization does not happen, they are indeed the same function.

[YexuanXiao]

It can be observed that MSVC generates SSE/AVX/ARM64 SIMD instructions when using memcpy. However, further testing is needed to evaluate the performance impact under different constant sizes. https://godbolt.org/z/5hehq8svh