[libc-commits] [libc] [libc] memmove optimizations (PR #70043)

[Guillaume Chatelet via libc-commits]

gchatelet wrote:

And how about splitting `inline_memmove` into several functions:
 - `bool inline_memmove_small_size()` returning whether the size was handled
 - `inline_memmove_follow_up` (or `inline_memmove_larger_sizes`)

We'd have the following code
```
LLVM_LIBC_FUNCTION(void *, memmove, (void *dst, const void *src, size_t count)) {
  if (inline_memmove_small_size(dst, src, count))
    return dst;
  if (is_disjoint(dst, src, count))
    inline_memcpy(dst, src, count);
  else 
    inline_memmove_follow_up(dst, src, count);
  return dst;
}
```

Each arch would be responsible for providing the implementations side by side so we can visually check they complement each other. For arch where we don't have a fast path `inline_memmove_small_size` would be
```
constexpr bool inline_memmove_small_size(void *dst, const void *src, size_t count) {
  return false;
}
```

Inlining would remove the branch altogether.

WDTY?

https://github.com/llvm/llvm-project/pull/70043