1 files changed, 69 insertions, 0 deletions
diff --git a/include/m_string.h b/include/m_string.h
index 3c4c6ea088a..a6a36f32ca4 100644
--- a/include/m_string.h
+++ b/include/m_string.h
@@ -260,4 +260,73 @@ typedef struct st_mysql_lex_string LEX_STRING;
 #define USTRING_WITH_LEN(X) ((uchar*) X), ((size_t) (sizeof(X) - 1))
 #define C_STRING_WITH_LEN(X) ((char *) (X)), ((size_t) (sizeof(X) - 1))
 
+/* SPACE_INT is a word that contains only spaces */
+#if SIZEOF_INT == 4
+#define SPACE_INT 0x20202020
+#elif SIZEOF_INT == 8
+#define SPACE_INT 0x2020202020202020
+#else
+#error define the appropriate constant for a word full of spaces
+#endif
+
+/**
+  Skip trailing space.
+
+  On most systems reading memory in larger chunks (ideally equal to the size of
+  the chinks that the machine physically reads from memory) causes fewer memory
+  access loops and hence increased performance.
+  This is why the 'int' type is used : it's closest to that (according to how
+  it's defined in C).
+  So when we determine the amount of whitespace at the end of a string we do
+  the following :
+    1. We divide the string into 3 zones :
+      a) from the start of the string (__start) to the first multiple
+        of sizeof(int)  (__start_words)
+      b) from the end of the string (__end) to the last multiple of sizeof(int)
+        (__end_words)
+      c) a zone that is aligned to sizeof(int) and can be safely accessed
+        through an int *
+    2. We start comparing backwards from (c) char-by-char. If all we find is
+       space then we continue
+    3. If there are elements in zone (b) we compare them as unsigned ints to a
+       int mask (SPACE_INT) consisting of all spaces
+    4. Finally we compare the remaining part (a) of the string char by char.
+       This covers for the last non-space unsigned int from 3. (if any)
+
+   This algorithm works well for relatively larger strings, but it will slow
+   the things down for smaller strings (because of the additional calculations
+   and checks compared to the naive method). Thus the barrier of length 20
+   is added.
+
+   @param     ptr   pointer to the input string
+   @param     len   the length of the string
+   @return          the last non-space character
+*/
+
+static inline const uchar *skip_trailing_space(const uchar *ptr,size_t len)
+{
+  const uchar *end= ptr + len;
+
+  if (len > 20)
+  {
+    const uchar *end_words= (const uchar *)(intptr)
+      (((ulonglong)(intptr)end) / SIZEOF_INT * SIZEOF_INT);
+    const uchar *start_words= (const uchar *)(intptr)
+       ((((ulonglong)(intptr)ptr) + SIZEOF_INT - 1) / SIZEOF_INT * SIZEOF_INT);
+
+    DBUG_ASSERT(((ulonglong)(intptr)ptr) >= SIZEOF_INT);
+    if (end_words > ptr)
+    {
+      while (end > end_words && end[-1] == 0x20)
+        end--;
+      if (end[-1] == 0x20 && start_words < end_words)
+        while (end > start_words && ((unsigned *)end)[-1] == SPACE_INT)
+          end -= SIZEOF_INT;
+    }
+  }
+  while (end > ptr && end[-1] == 0x20)
+    end--;
+  return (end);
+}
+
 #endif