[libcxx-commits] [libcxx] [libc++] Optimize ranges::equal for vector<bool>::iterator (PR #121084)

[Louis Dionne via libcxx-commits]

================
@@ -33,6 +41,132 @@ _LIBCPP_PUSH_MACROS
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
+template <class _Cp, bool _IC1, bool _IC2>
+[[__nodiscard__]] _LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_HIDE_FROM_ABI bool __equal_unaligned(
+    __bit_iterator<_Cp, _IC1> __first1, __bit_iterator<_Cp, _IC1> __last1, __bit_iterator<_Cp, _IC2> __first2) {
+  using _It             = __bit_iterator<_Cp, _IC1>;
+  using difference_type = typename _It::difference_type;
+  using __storage_type  = typename _It::__storage_type;
+
+  const int __bits_per_word = _It::__bits_per_word;
+  difference_type __n       = __last1 - __first1;
+  if (__n > 0) {
+    // do first word
+    if (__first1.__ctz_ != 0) {
+      unsigned __clz_f     = __bits_per_word - __first1.__ctz_;
+      difference_type __dn = std::min(static_cast<difference_type>(__clz_f), __n);
+      __n -= __dn;
+      __storage_type __m   = (~__storage_type(0) << __first1.__ctz_) & (~__storage_type(0) >> (__clz_f - __dn));
+      __storage_type __b   = *__first1.__seg_ & __m;
+      unsigned __clz_r     = __bits_per_word - __first2.__ctz_;
+      __storage_type __ddn = std::min<__storage_type>(__dn, __clz_r);
+      __m                  = (~__storage_type(0) << __first2.__ctz_) & (~__storage_type(0) >> (__clz_r - __ddn));
+      if (__first2.__ctz_ > __first1.__ctz_) {
+        if ((*__first2.__seg_ & __m) != (__b << (__first2.__ctz_ - __first1.__ctz_)))
+          return false;
+      } else {
+        if ((*__first2.__seg_ & __m) != (__b >> (__first1.__ctz_ - __first2.__ctz_)))
+          return false;
+      }
+      __first2.__seg_ += (__ddn + __first2.__ctz_) / __bits_per_word;
+      __first2.__ctz_ = static_cast<unsigned>((__ddn + __first2.__ctz_) % __bits_per_word);
+      __dn -= __ddn;
+      if (__dn > 0) {
+        __m = ~__storage_type(0) >> (__bits_per_word - __dn);
+        if ((*__first2.__seg_ & __m) != (__b >> (__first1.__ctz_ + __ddn)))
+          return false;
+        __first2.__ctz_ = static_cast<unsigned>(__dn);
+      }
+      ++__first1.__seg_;
+      // __first1.__ctz_ = 0;
+    }
+    // __first1.__ctz_ == 0;
+    // do middle words
+    unsigned __clz_r   = __bits_per_word - __first2.__ctz_;
+    __storage_type __m = ~__storage_type(0) << __first2.__ctz_;
+    for (; __n >= __bits_per_word; __n -= __bits_per_word, ++__first1.__seg_) {
+      __storage_type __b = *__first1.__seg_;
+      if ((*__first2.__seg_ & __m) != (__b << __first2.__ctz_))
+        return false;
+      ++__first2.__seg_;
+      if ((*__first2.__seg_ & ~__m) != (__b >> __clz_r))
+        return false;
+    }
+    // do last word
+    if (__n > 0) {
+      __m                 = ~__storage_type(0) >> (__bits_per_word - __n);
+      __storage_type __b  = *__first1.__seg_ & __m;
+      __storage_type __dn = std::min(__n, static_cast<difference_type>(__clz_r));
+      __m                 = (~__storage_type(0) << __first2.__ctz_) & (~__storage_type(0) >> (__clz_r - __dn));
+      if ((*__first2.__seg_ & __m) != (__b << __first2.__ctz_))
+        return false;
+      __first2.__seg_ += (__dn + __first2.__ctz_) / __bits_per_word;
+      __first2.__ctz_ = static_cast<unsigned>((__dn + __first2.__ctz_) % __bits_per_word);
+      __n -= __dn;
+      if (__n > 0) {
+        __m = ~__storage_type(0) >> (__bits_per_word - __n);
+        if ((*__first2.__seg_ & __m) != (__b >> __dn))
+          return false;
+      }
+    }
+  }
+  return true;
+}
+
+template <class _Cp, bool _IC1, bool _IC2>
+[[__nodiscard__]] _LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_HIDE_FROM_ABI bool __equal_aligned(
+    __bit_iterator<_Cp, _IC1> __first1, __bit_iterator<_Cp, _IC1> __last1, __bit_iterator<_Cp, _IC2> __first2) {
+  using _It             = __bit_iterator<_Cp, _IC1>;
+  using difference_type = typename _It::difference_type;
+  using __storage_type  = typename _It::__storage_type;
+
+  const int __bits_per_word = _It::__bits_per_word;
+  difference_type __n       = __last1 - __first1;
+  if (__n > 0) {
+    // do first word
+    if (__first1.__ctz_ != 0) {
+      unsigned __clz       = __bits_per_word - __first1.__ctz_;
+      difference_type __dn = std::min(static_cast<difference_type>(__clz), __n);
+      __n -= __dn;
+      __storage_type __m = (~__storage_type(0) << __first1.__ctz_) & (~__storage_type(0) >> (__clz - __dn));
+      if ((*__first2.__seg_ & __m) != (*__first1.__seg_ & __m))
+        return false;
+      ++__first2.__seg_;
+      ++__first1.__seg_;
+      // __first1.__ctz_ = 0;
+      // __first2.__ctz_ = 0;
+    }
+    // __first1.__ctz_ == 0;
+    // __first2.__ctz_ == 0;
+    // do middle words
+    for (; __n >= __bits_per_word; __n -= __bits_per_word, ++__first1.__seg_, ++__first2.__seg_)
+      if (*__first2.__seg_ != *__first1.__seg_)
+        return false;
+    // do last word
+    if (__n > 0) {
+      __storage_type __m = ~__storage_type(0) >> (__bits_per_word - __n);
+      if ((*__first2.__seg_ & __m) != (*__first1.__seg_ & __m))
+        return false;
+    }
+  }
+  return true;
+}
+
+template <class _Cp,
+          bool _IC1,
+          bool _IC2,
+          class _BinaryPredicate,
+          __enable_if_t<std::is_same<_BinaryPredicate, __equal_to>::value, int> = 0>
+[[__nodiscard__]] inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 bool __equal_iter_impl(
----------------
ldionne wrote:

Let's inline this function into `__equal_impl(__bit_iterator, ...)` below, since the additional indirection doesn't serve much of a purpose.

https://github.com/llvm/llvm-project/pull/121084