[libcxx-commits] [libcxx] [libc++] Speed up vector<bool> copy/move-ctors [1/3] (PR #120132)

[Louis Dionne via libcxx-commits]

================
@@ -674,25 +676,30 @@ vector<bool, _Allocator>::vector(initializer_list<value_type> __il, const alloca
 
 #endif // _LIBCPP_CXX03_LANG
 
+// This function copies each storage word as a whole, which is substantially more efficient than copying
+// individual bits within each word
+template <class _Allocator>
+_LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_HIDE_FROM_ABI void vector<bool, _Allocator>::__alloc_and_copy(const vector& __v) {
+  if (__v.__size_) {
+    __vallocate(__v.__size_);
+    std::copy(__v.__begin_, __v.__begin_ + __external_cap_to_internal(__v.__size_), __begin_);
+  }
+  __size_ = __v.__size_;
+}
+
 template <class _Allocator>
 _LIBCPP_CONSTEXPR_SINCE_CXX20 vector<bool, _Allocator>::vector(const vector& __v)
     : __begin_(nullptr),
       __size_(0),
       __cap_(0),
       __alloc_(__storage_traits::select_on_container_copy_construction(__v.__alloc_)) {
-  if (__v.size() > 0) {
-    __vallocate(__v.size());
-    __construct_at_end(__v.begin(), __v.end(), __v.size());
-  }
+  __alloc_and_copy(__v);
----------------
ldionne wrote:

I have a bit of trouble justifying it, but I would find the code easier to understand if you inlined `__alloc_and_copy` into its call sites. It's a simple function and some call sites are slightly different (e.g. wouldn't require setting the size outside of the `if` branch). IMO that would simplify the code.

https://github.com/llvm/llvm-project/pull/120132