Speedup stringsearch for two byte strings

Uses the lower byte with memchr which is
significantly faster than a naive compare

Performance difference with bench (http://hastebin.com/xuxexataso.js):

old                             new

single character                single character
Κ found at 922                  Κ found at 922
3324                            616
㎡ found at 13217               ㎡ found at 13217
42366                           4931
က found at 4096                 က found at 4096
13369                           9836
＀ found at 65280                ＀ found at 65280
207472                          36149
ﾥ found at 65445                ﾥ found at 65445
209344                          36666
  found at 8197                   found at 8197
26731                           11757
倂 found at 20482               倂 found at 20482
66071                           17193

linear search                   linear search
ΚΛ found at 922                 ΚΛ found at 922
4112                            504
㎡㎢ found at 13217             ㎡㎢ found at 13217
55105                           5119
ﾥﾦ found at 65445               ﾥﾦ found at 65445
268016                          35496

linear + bmh search             linear + bmh search
ΚΛΜΝΞΟΠΡ found at 922           ΚΛΜΝΞΟΠΡ found at 922
2897                            522
ﾥﾦﾧﾨﾩﾪﾫﾬ found at 65445         ﾥﾦﾧﾨﾩﾪﾫﾬ found at 65445
167687                          158465

Review URL: https://codereview.chromium.org/1303033012

Cr-Commit-Position: refs/heads/master@{#30587}

diff --git a/AUTHORS b/AUTHORS
index 72c23bc..82a9c16 100644 (file)
--- a/AUTHORS
+++ b/AUTHORS
@@ -67,6 +67,7 @@ Johan Bergström <johan@bergstroem.nu>
 Jonathan Liu <net147@gmail.com>
 JunHo Seo <sejunho@gmail.com>
 Kang-Hao (Kenny) Lu <kennyluck@csail.mit.edu>
+Karl Skomski <karl@skomski.com>
 Luis Reis <luis.m.reis@gmail.com>
 Luke Zarko <lukezarko@gmail.com>
 Maciej Małecki <me@mmalecki.com>

diff --git a/src/string-search.h b/src/string-search.h
index 349d4fd..aeaa9b3 100644 (file)
--- a/src/string-search.h
+++ b/src/string-search.h
@@ -190,6 +190,38 @@ class StringSearch : private StringSearchBase {
 };
 
 
+template <typename PatternChar, typename SubjectChar>
+int FindFirstCharacter(Vector<const PatternChar> pattern,
+                       Vector<const SubjectChar> subject, int index) {
+  PatternChar pattern_first_char = pattern[0];
+
+  if (sizeof(SubjectChar) == 1 && sizeof(PatternChar) == 1) {
+    const SubjectChar* char_pos = reinterpret_cast<const SubjectChar*>(memchr(
+        subject.start() + index, pattern_first_char, subject.length() - index));
+    if (char_pos == NULL) return -1;
+    return static_cast<int>(char_pos - subject.start());
+  } else {
+    const uint8_t search_low_byte =
+        static_cast<uint8_t>(pattern_first_char & 0xFF);
+    const SubjectChar search_char =
+        static_cast<SubjectChar>(pattern_first_char);
+    int pos = index;
+    do {
+      const SubjectChar* char_pos = reinterpret_cast<const SubjectChar*>(
+          memchr(subject.start() + pos, search_low_byte,
+                 (subject.length() - pos) * sizeof(SubjectChar)));
+      if (char_pos == NULL) return -1;
+      pos = static_cast<int>(char_pos - subject.start());
+      if (IsAligned(reinterpret_cast<uintptr_t>(char_pos),
+                    sizeof(SubjectChar))) {
+        if (subject[pos] == search_char) return pos;
+      }
+    } while (++pos < subject.length());
+  }
+  return -1;
+}
+
+
 //---------------------------------------------------------------------
 // Single Character Pattern Search Strategy
 //---------------------------------------------------------------------
@@ -201,26 +233,15 @@ int StringSearch<PatternChar, SubjectChar>::SingleCharSearch(
     int index) {
   DCHECK_EQ(1, search->pattern_.length());
   PatternChar pattern_first_char = search->pattern_[0];
-  int i = index;
   if (sizeof(SubjectChar) == 1 && sizeof(PatternChar) == 1) {
-    const SubjectChar* pos = reinterpret_cast<const SubjectChar*>(
-        memchr(subject.start() + i,
-               pattern_first_char,
-               subject.length() - i));
-    if (pos == NULL) return -1;
-    return static_cast<int>(pos - subject.start());
+    return FindFirstCharacter(search->pattern_, subject, index);
   } else {
     if (sizeof(PatternChar) > sizeof(SubjectChar)) {
       if (exceedsOneByte(pattern_first_char)) {
         return -1;
       }
     }
-    SubjectChar search_char = static_cast<SubjectChar>(pattern_first_char);
-    int n = subject.length();
-    while (i < n) {
-      if (subject[i++] == search_char) return i - 1;
-    }
-    return -1;
+    return FindFirstCharacter(search->pattern_, subject, index);
   }
 }
 
@@ -254,20 +275,12 @@ int StringSearch<PatternChar, SubjectChar>::LinearSearch(
   Vector<const PatternChar> pattern = search->pattern_;
   DCHECK(pattern.length() > 1);
   int pattern_length = pattern.length();
-  PatternChar pattern_first_char = pattern[0];
   int i = index;
   int n = subject.length() - pattern_length;
   while (i <= n) {
-    if (sizeof(SubjectChar) == 1 && sizeof(PatternChar) == 1) {
-      const SubjectChar* pos = reinterpret_cast<const SubjectChar*>(
-          memchr(subject.start() + i,
-                 pattern_first_char,
-                 n - i + 1));
-      if (pos == NULL) return -1;
-      i = static_cast<int>(pos - subject.start()) + 1;
-    } else {
-      if (subject[i++] != pattern_first_char) continue;
-    }
+    i = FindFirstCharacter(pattern, subject, i);
+    if (i == -1) return -1;
+    i++;
     // Loop extracted to separate function to allow using return to do
     // a deeper break.
     if (CharCompare(pattern.start() + 1,
@@ -505,22 +518,11 @@ int StringSearch<PatternChar, SubjectChar>::InitialSearch(
 
   // We know our pattern is at least 2 characters, we cache the first so
   // the common case of the first character not matching is faster.
-  PatternChar pattern_first_char = pattern[0];
   for (int i = index, n = subject.length() - pattern_length; i <= n; i++) {
     badness++;
     if (badness <= 0) {
-      if (sizeof(SubjectChar) == 1 && sizeof(PatternChar) == 1) {
-        const SubjectChar* pos = reinterpret_cast<const SubjectChar*>(
-            memchr(subject.start() + i,
-                   pattern_first_char,
-                   n - i + 1));
-        if (pos == NULL) {
-          return -1;
-        }
-        i = static_cast<int>(pos - subject.start());
-      } else {
-        if (subject[i] != pattern_first_char) continue;
-      }
+      i = FindFirstCharacter(pattern, subject, i);
+      if (i == -1) return -1;
       int j = 1;
       do {
         if (pattern[j] != subject[i + j]) {

diff --git a/test/mjsunit/string-indexof-1.js b/test/mjsunit/string-indexof-1.js
index db3623f..366437a 100644 (file)
--- a/test/mjsunit/string-indexof-1.js
+++ b/test/mjsunit/string-indexof-1.js
@@ -77,6 +77,15 @@ assertEquals(-1, twoByteString.indexOf("\u0391\u03a3\u0395"),
 //single char pattern
 assertEquals(4, twoByteString.indexOf("\u0395"));
 
+// test string with alignment traps
+var alignmentString = "\u1122\u2211\u2222\uFF00\u00FF\u00FF";
+assertEquals(2, alignmentString.indexOf("\u2222"));
+assertEquals(4, alignmentString.indexOf("\u00FF\u00FF"));
+
+var longAlignmentString = "\uFF00" + "\u00FF".repeat(10);
+assertEquals(1,
+    longAlignmentString.indexOf("\u00FF".repeat(10)));
+
 // Test complex string indexOf algorithms. Only trigger for long strings.
 
 // Long string that isn't a simple repeat of a shorter string.