Split utf8::read_codepoint between single byte and multibyte code

Make read_codepoint_multibyte noinline so that the common case single
byte case gets inlined.

1 files changed, 17 insertions, 11 deletions

diff --git a/src/utf8.hh b/src/utf8.hh
index 0ba8bdef..e5643214 100644
--- a/src/utf8.hh
+++ b/src/utf8.hh
@@ -39,22 +39,13 @@ struct Pass
 
 }
 
-// returns the codepoint of the character whose first byte
-// is pointed by it
 template<typename InvalidPolicy = utf8::InvalidPolicy::Pass,
          typename Iterator, typename Sentinel>
-Codepoint read_codepoint(Iterator& it, const Sentinel& end)
+[[gnu::noinline]]
+Codepoint read_codepoint_multibyte(Iterator& it, const Sentinel& end, char byte)
     noexcept(noexcept(InvalidPolicy{}(0)))
 {
     if (it == end)
-        return InvalidPolicy{}(-1);
-    // According to rfc3629, UTF-8 allows only up to 4 bytes.
-    // (21 bits codepoint)
-    unsigned char byte = read(it);
-    if ((byte & 0x80) == 0) // 0xxxxxxx
-        return byte;
-
-    if (it == end)
         return InvalidPolicy{}(byte);
 
     if ((byte & 0xE0) == 0xC0) // 110xxxxx
@@ -81,6 +72,21 @@ Codepoint read_codepoint(Iterator& it, const Sentinel& end)
     return InvalidPolicy{}(byte);
 }
 
+// returns the codepoint of the character whose first byte
+// is pointed by it
+template<typename InvalidPolicy = utf8::InvalidPolicy::Pass,
+         typename Iterator, typename Sentinel>
+Codepoint read_codepoint(Iterator& it, const Sentinel& end)
+    noexcept(noexcept(InvalidPolicy{}(0)))
+{
+    if (it == end)
+        return InvalidPolicy{}(-1);
+    unsigned char byte = read(it);
+    if ((byte & 0x80) == 0) [[likely]] // 0xxxxxxx
+        return byte;
+    return read_codepoint_multibyte(it, end, byte);
+}
+
 template<typename InvalidPolicy = utf8::InvalidPolicy::Pass,
          typename Iterator, typename Sentinel>
 Codepoint codepoint(Iterator it, const Sentinel& end)