Speed up string-lessp for multibyte strings

Improve comparison speed when both arguments are multibyte strings,
at least one of them containing a non-ASCII character.  (All-ASCII
multibyte strings are already fast.)
The speed-up is about 2× for strings of 10 chars, 10× for strings of
100 chars.

* src/fns.c (Fstring_lessp): Quickly skip the common prefix by
comparing words.

1 files changed, 39 insertions, 12 deletions

diff --git a/src/fns.c b/src/fns.c
index 22e66d3653d..bc4915eb25b 100644
--- a/src/fns.c
+++ b/src/fns.c
@@ -454,23 +454,50 @@ Symbols are also allowed; their print names are used instead.  */)
       && (!STRING_MULTIBYTE (string2) || SCHARS (string2) == SBYTES (string2)))
     {
       /* Each argument is either unibyte or all-ASCII multibyte:
-         we can compare bytewise.
-         (Arbitrary multibyte strings cannot be compared bytewise because
-         that would give a different order for raw bytes 80..FF.)  */
+         we can compare bytewise.  */
       int d = memcmp (SSDATA (string1), SSDATA (string2), n);
       return d < 0 || (d == 0 && n < SCHARS (string2)) ? Qt : Qnil;
     }
   else if (STRING_MULTIBYTE (string1) && STRING_MULTIBYTE (string2))
     {
-      ptrdiff_t i1 = 0, i1_byte = 0, i2 = 0, i2_byte = 0;
-      while (i1 < n)
-        {
-          int c1 = fetch_string_char_advance_no_check (string1, &i1, &i1_byte);
-          int c2 = fetch_string_char_advance_no_check (string2, &i2, &i2_byte);
-          if (c1 != c2)
-            return c1 < c2 ? Qt : Qnil;
-        }
-      return i1 < SCHARS (string2) ? Qt : Qnil;
+      /* Two arbitrary multibyte strings: we cannot use memcmp because
+         the encoding for raw bytes would sort those between U+007F and U+0080
+         which isn't where we want them.
+         Instead, we skip the longest common prefix and look at
+         what follows.  */
+      ptrdiff_t nb1 = SBYTES (string1);
+      ptrdiff_t nb2 = SBYTES (string2);
+      ptrdiff_t nb = min (nb1, nb2);
+
+      /* First compare entire machine words.  (String data is allocated
+         with word alignment.)  */
+      typedef size_t word_t;
+      int ws = sizeof (word_t);
+      const word_t *w1 = (const word_t *) SDATA (string1);
+      const word_t *w2 = (const word_t *) SDATA (string2);
+      ptrdiff_t b = 0;
+      while (b < nb - ws + 1 && w1[b / ws] == w2[b / ws])
+        b += ws;
+
+      /* Scan forward to the differing byte (at most ws-1 bytes).  */
+      while (b < nb && SREF (string1, b) == SREF (string2, b))
+        b++;
+
+      if (b >= nb)
+        /* One string is a prefix of the other.  */
+        return b < nb2 ? Qt : Qnil;
+
+      /* Now back up to the start of the differing characters:
+         it's the last byte not having the bit pattern 10xxxxxx.  */
+      while ((SREF (string1, b) & 0xc0) == 0x80)
+        b--;
+
+      /* Compare the differing characters.  */
+      ptrdiff_t i1 = 0, i2 = 0;
+      ptrdiff_t i1_byte = b, i2_byte = b;
+      int c1 = fetch_string_char_advance_no_check (string1, &i1, &i1_byte);
+      int c2 = fetch_string_char_advance_no_check (string2, &i2, &i2_byte);
+      return c1 < c2 ? Qt : Qnil;
     }
   else if (STRING_MULTIBYTE (string1))
     {