From 721064dc293d8915fbb33d83bd983a40dcca180f Mon Sep 17 00:00:00 2001
From: Roland Reichwein <mail@reichwein.it>
Date: Thu, 23 Dec 2021 13:27:34 +0100
Subject: Speed optimization

---
 include/unicode.h    | 73 ++++++++++++++++++++++++++++++++++++++++++++++++++--
 src/test-unicode.cpp |  3 +++
 2 files changed, 74 insertions(+), 2 deletions(-)

diff --git a/include/unicode.h b/include/unicode.h
index 43dc44e..5774db7 100644
--- a/include/unicode.h
+++ b/include/unicode.h
@@ -8,6 +8,7 @@
 #pragma once
 
 #include <algorithm>
+#include <cstdint>
 #include <iterator>
 #include <list>
 #include <memory>
@@ -203,6 +204,17 @@ namespace unicode::detail {
    return calculate_value();
   }
 
+  utf_iterator& operator+=(size_t distance)
+  {
+   std::advance(iterator, distance);
+   return *this;
+  }
+
+  size_t operator-(const utf_iterator& other) const
+  {
+   return iterator - other.iterator;
+  }
+
  private:
   typename string_type::const_iterator iterator;
   typename string_type::const_iterator end_iterator;
@@ -394,7 +406,7 @@ namespace unicode {
   }
 
   // return reference?
-  value_type operator*()
+  value_type operator*() const
   {
    input_type value{*m_it};
 
@@ -407,6 +419,17 @@ namespace unicode {
    return static_cast<value_type>(static_cast<uint8_t>(value));
   }
 
+  iso_iterator& operator+=(size_t distance)
+  {
+   std::advance(m_it, distance);
+   return *this;
+  }
+
+  difference_type operator-(const iso_iterator& other) const
+  {
+   return m_it - other.m_it;
+  }
+
  private:
   iterator m_it;
  };
@@ -518,13 +541,59 @@ namespace unicode {
  typedef UTF<utf_iterator<char16_t>, utf_back_insert_iterator<char16_t>> UTF_16;
  typedef UTF<utf_iterator<char32_t>, utf_back_insert_iterator<char32_t>> UTF_32;
 
+ // std::distance doesn't work here: it is based on "output" distance of iterators
+ template<class Iterator>
+ size_t input_distance(const Iterator& it1, const Iterator& it2)
+ {
+  return it2 - it1;
+ }
+
  // From and To are facets
  template<typename From, typename To, std::enable_if_t<std::is_empty<From>::value, bool> = true>
  typename To::string_type convert(const typename From::string_type& s)
  {
   typename To::string_type result;
 
-  std::copy(From::begin(s), From::end(s), To::back_inserter(result));
+  if constexpr(sizeof(typename From::string_type::value_type) == 1 &&
+               sizeof(typename To::value_type) == 1 &&
+               sizeof(size_t) >= 8) {
+   auto begin{From::begin(s)};
+   auto end{From::end(s)};
+   auto back_inserter{To::back_inserter(result)};
+   auto addr{reinterpret_cast<const uint64_t*>(&s.data()[s.size() - input_distance(begin, end)])};
+   while (input_distance(begin, end) >= 8) {
+    if (((uintptr_t)(void*)addr & 7) == 0) {
+     while (input_distance(begin, end) >= 8) {
+      uint64_t data{*addr};
+      if ((data & 0x8080808080808080ULL) == 0ULL) {
+       result.append(reinterpret_cast<const typename To::value_type*>(addr), 8);
+       begin += 8;
+       ++addr;
+      } else {
+       // just advance one code unit for now
+       back_inserter = *begin;
+       ++begin;
+       break;
+      }
+     }
+    }
+
+    // keep up after unaligned Non-ASCII code points
+    while (begin!= end && (uintptr_t)(void*)(addr = reinterpret_cast<const uint64_t*>(&s.data()[s.size() - input_distance(begin, end)])) & 7) {
+     back_inserter = *begin;
+     ++begin;
+    }
+   }
+
+   // remainder < 8 bytes   
+   while (begin != end) {
+    back_inserter = *begin;
+    ++begin;
+   }
+
+  } else {
+   std::copy(From::begin(s), From::end(s), To::back_inserter(result));
+  }
 
   return result;
  }
diff --git a/src/test-unicode.cpp b/src/test-unicode.cpp
index 59d55b9..d638cbb 100644
--- a/src/test-unicode.cpp
+++ b/src/test-unicode.cpp
@@ -566,6 +566,9 @@ BOOST_AUTO_TEST_CASE(convert)
  // deque
  BOOST_CHECK((unicode::convert<std::deque<char>, std::deque<wchar_t>>(std::deque<char>{})) == std::deque<wchar_t>{});
  BOOST_CHECK((unicode::convert<std::deque<char>, std::deque<wchar_t>>(std::deque<char>{'\xc3', '\xa4', '\xc3', '\xb6', '\xc3', '\xbc'})) == (std::deque<wchar_t>{L'ä', L'ö', L'ü'}));
+ // yet unsupported:
+ //BOOST_CHECK((unicode::convert<utf8_t, char16_t>(std::deque<utf8_t>{u8'\xc3', u8'\xa4', u8'\xc3', u8'\xb6', u8'\xc3', u8'\xbc'})) == (std::deque<char16_t>{u'ä', u'ö', u'ü'}));
+ //BOOST_CHECK((unicode::convert<unicode::UTF_8, unicode::UTF_16>(std::deque<utf8_t>{u8'\xc3', u8'\xa4', u8'\xc3', u8'\xb6', u8'\xc3', u8'\xbc'})) == (std::deque<char16_t>{u'ä', u'ö', u'ü'}));
  
  // deque with uint8_t, uint16_t
  BOOST_CHECK((unicode::convert<std::deque<uint8_t>, std::deque<uint16_t>>(std::deque<uint8_t>{})) == std::deque<uint16_t>{});
-- 
cgit v1.2.3