Attempt to make \b\w+ faster. Slight performance increase on, e.g., string unpacking.

Review URL: http://codereview.chromium.org/507051


git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3563 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/arm/regexp-macro-assembler-arm.cc

@@ -465,8 +465,6 @@ void RegExpMacroAssemblerARM::CheckNotCharacterAfterMinusAnd(
 
 
 bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type,
-                                                         int cp_offset,
-                                                         bool check_offset,
                                                          Label* on_no_match) {
   // Range checks (c in min..max) are generally implemented by an unsigned
   // (c - min) <= (max - min) check
@@ -475,11 +473,6 @@ bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type,
     // Match space-characters
     if (mode_ == ASCII) {
       // ASCII space characters are '\t'..'\r' and ' '.
-      if (check_offset) {
-        LoadCurrentCharacter(cp_offset, on_no_match);
-      } else {
-        LoadCurrentCharacterUnchecked(cp_offset, 1);
-      }
       Label success;
       __ cmp(current_character(), Operand(' '));
       __ b(eq, &success);
@@ -493,11 +486,6 @@ bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type,
     return false;
   case 'S':
     // Match non-space characters.
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
     if (mode_ == ASCII) {
       // ASCII space characters are '\t'..'\r' and ' '.
       __ cmp(current_character(), Operand(' '));
@@ -510,33 +498,18 @@ bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type,
     return false;
   case 'd':
     // Match ASCII digits ('0'..'9')
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
     __ sub(r0, current_character(), Operand('0'));
     __ cmp(current_character(), Operand('9' - '0'));
     BranchOrBacktrack(hi, on_no_match);
     return true;
   case 'D':
     // Match non ASCII-digits
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
     __ sub(r0, current_character(), Operand('0'));
     __ cmp(r0, Operand('9' - '0'));
     BranchOrBacktrack(ls, on_no_match);
     return true;
   case '.': {
     // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
     __ eor(r0, current_character(), Operand(0x01));
     // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
     __ sub(r0, r0, Operand(0x0b));
@@ -552,13 +525,71 @@ bool RegExpMacroAssemblerARM::CheckSpecialCharacterClass(uc16 type,
     }
     return true;
   }
+  case 'n': {
+      // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
+      __ eor(r0, current_character(), Operand(0x01));
+      // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
+      __ sub(r0, r0, Operand(0x0b));
+      __ cmp(r0, Operand(0x0c - 0x0b));
+      if (mode_ == ASCII) {
+        BranchOrBacktrack(hi, on_no_match);
+      } else {
+        Label done;
+        __ b(ls, &done);
+        // Compare original value to 0x2028 and 0x2029, using the already
+        // computed (current_char ^ 0x01 - 0x0b). I.e., check for
+        // 0x201d (0x2028 - 0x0b) or 0x201e.
+        __ sub(r0, r0, Operand(0x2028 - 0x0b));
+        __ cmp(r0, Operand(1));
+        BranchOrBacktrack(hi, on_no_match);
+        __ bind(&done);
+      }
+      return true;
+    }
+  case 'w': {
+    // Match word character (0-9, A-Z, a-z and _).
+    Label digits, done;
+    __ cmp(current_character(), Operand('9'));
+    __ b(ls, &digits);
+    __ cmp(current_character(), Operand('_'));
+    __ b(eq, &done);
+    __ orr(r0, current_character(), Operand(0x20));
+    __ sub(r0, r0, Operand('a'));
+    __ cmp(r0, Operand('z' - 'a'));
+    BranchOrBacktrack(hi, on_no_match);
+    __ jmp(&done);
+
+    __ bind(&digits);
+    __ cmp(current_character(), Operand('0'));
+    BranchOrBacktrack(lo, on_no_match);
+    __ bind(&done);
+
+    return true;
+  }
+  case 'W': {
+    // Match non-word character (not 0-9, A-Z, a-z and _).
+    Label digits, done;
+    __ cmp(current_character(), Operand('9'));
+    __ b(ls, &digits);
+    __ cmp(current_character(), Operand('_'));
+    BranchOrBacktrack(eq, on_no_match);
+    __ orr(r0, current_character(), Operand(0x20));
+    __ sub(r0, r0, Operand('a'));
+    __ cmp(r0, Operand('z' - 'a'));
+    BranchOrBacktrack(ls, on_no_match);
+    __ jmp(&done);
+
+    __ bind(&digits);
+    __ cmp(current_character(), Operand('0'));
+    BranchOrBacktrack(hs, on_no_match);
+    __ bind(&done);
+
+    return true;
+  }
   case '*':
     // Match any character.
-    if (check_offset) {
-      CheckPosition(cp_offset, on_no_match);
-    }
     return true;
-  // No custom implementation (yet): w, W, s(UC16), S(UC16).
+  // No custom implementation (yet): s(UC16), S(UC16).
   default:
     return false;
   }

src/arm/regexp-macro-assembler-arm.h

@@ -80,8 +80,6 @@ class RegExpMacroAssemblerARM: public NativeRegExpMacroAssembler {
   // the end of the string.
   virtual void CheckPosition(int cp_offset, Label* on_outside_input);
   virtual bool CheckSpecialCharacterClass(uc16 type,
-                                          int cp_offset,
-                                          bool check_offset,
                                           Label* on_no_match);
   virtual void Fail();
   virtual Handle<Object> GetCode(Handle<String> source);

src/ast.cc

@@ -433,7 +433,7 @@ void* RegExpUnparser::VisitQuantifier(RegExpQuantifier* that, void* data) {
   } else {
     stream()->Add("%i ", that->max());
   }
-  stream()->Add(that->is_greedy() ? "g " : "n ");
+  stream()->Add(that->is_greedy() ? "g " : that->is_possessive() ? "p " : "n ");
   that->body()->Accept(this, data);
   stream()->Add(")");
   return NULL;

src/ast.h

@@ -1526,6 +1526,7 @@ class CharacterSet BASE_EMBEDDED {
     standard_set_type_ = special_set_type;
   }
   bool is_standard() { return standard_set_type_ != 0; }
+  void Canonicalize();
  private:
   ZoneList<CharacterRange>* ranges_;
   // If non-zero, the value represents a standard set (e.g., all whitespace
@@ -1619,12 +1620,13 @@ class RegExpText: public RegExpTree {
 
 class RegExpQuantifier: public RegExpTree {
  public:
-  RegExpQuantifier(int min, int max, bool is_greedy, RegExpTree* body)
-      : min_(min),
+  enum Type { GREEDY, NON_GREEDY, POSSESSIVE };
+  RegExpQuantifier(int min, int max, Type type, RegExpTree* body)
+      : body_(body),
+        min_(min),
         max_(max),
-        is_greedy_(is_greedy),
-        body_(body),
-        min_match_(min * body->min_match()) {
+        min_match_(min * body->min_match()),
+        type_(type) {
     if (max > 0 && body->max_match() > kInfinity / max) {
       max_match_ = kInfinity;
     } else {
@@ -1648,15 +1650,17 @@ class RegExpQuantifier: public RegExpTree {
   virtual int max_match() { return max_match_; }
   int min() { return min_; }
   int max() { return max_; }
-  bool is_greedy() { return is_greedy_; }
+  bool is_possessive() { return type_ == POSSESSIVE; }
+  bool is_non_greedy() { return type_ == NON_GREEDY; }
+  bool is_greedy() { return type_ == GREEDY; }
   RegExpTree* body() { return body_; }
  private:
+  RegExpTree* body_;
   int min_;
   int max_;
-  bool is_greedy_;
-  RegExpTree* body_;
   int min_match_;
   int max_match_;
+  Type type_;
 };
 
 

src/flag-definitions.h

@@ -329,6 +329,9 @@ DEFINE_bool(collect_heap_spill_statistics, false,
             "(requires heap_stats)")
 
 // Regexp
+DEFINE_bool(regexp_possessive_quantifier,
+            false,
+            "enable possessive quantifier syntax for testing")
 DEFINE_bool(trace_regexp_bytecodes, false, "trace regexp bytecode execution")
 DEFINE_bool(trace_regexp_assembler,
             false,

src/ia32/regexp-macro-assembler-ia32.cc

@@ -477,8 +477,6 @@ void RegExpMacroAssemblerIA32::CheckNotCharacterAfterMinusAnd(
 
 
 bool RegExpMacroAssemblerIA32::CheckSpecialCharacterClass(uc16 type,
-                                                          int cp_offset,
-                                                          bool check_offset,
                                                           Label* on_no_match) {
   // Range checks (c in min..max) are generally implemented by an unsigned
   // (c - min) <= (max - min) check
@@ -487,17 +485,12 @@ bool RegExpMacroAssemblerIA32::CheckSpecialCharacterClass(uc16 type,
     // Match space-characters
     if (mode_ == ASCII) {
       // ASCII space characters are '\t'..'\r' and ' '.
-      if (check_offset) {
-        LoadCurrentCharacter(cp_offset, on_no_match);
-      } else {
-        LoadCurrentCharacterUnchecked(cp_offset, 1);
-      }
       Label success;
       __ cmp(current_character(), ' ');
       __ j(equal, &success);
       // Check range 0x09..0x0d
-      __ sub(Operand(current_character()), Immediate('\t'));
-      __ cmp(current_character(), '\r' - '\t');
+      __ lea(eax, Operand(current_character(), -'\t'));
+      __ cmp(eax, '\r' - '\t');
       BranchOrBacktrack(above, on_no_match);
       __ bind(&success);
       return true;
@@ -505,72 +498,118 @@ bool RegExpMacroAssemblerIA32::CheckSpecialCharacterClass(uc16 type,
     return false;
   case 'S':
     // Match non-space characters.
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
     if (mode_ == ASCII) {
       // ASCII space characters are '\t'..'\r' and ' '.
       __ cmp(current_character(), ' ');
       BranchOrBacktrack(equal, on_no_match);
-      __ sub(Operand(current_character()), Immediate('\t'));
-      __ cmp(current_character(), '\r' - '\t');
+      __ lea(eax, Operand(current_character(), -'\t'));
+      __ cmp(eax, '\r' - '\t');
       BranchOrBacktrack(below_equal, on_no_match);
       return true;
     }
     return false;
   case 'd':
     // Match ASCII digits ('0'..'9')
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
-    __ sub(Operand(current_character()), Immediate('0'));
-    __ cmp(current_character(), '9' - '0');
+    __ lea(eax, Operand(current_character(), -'0'));
+    __ cmp(eax, '9' - '0');
     BranchOrBacktrack(above, on_no_match);
     return true;
   case 'D':
     // Match non ASCII-digits
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
-    __ sub(Operand(current_character()), Immediate('0'));
-    __ cmp(current_character(), '9' - '0');
+    __ lea(eax, Operand(current_character(), -'0'));
+    __ cmp(eax, '9' - '0');
     BranchOrBacktrack(below_equal, on_no_match);
     return true;
   case '.': {
     // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
-    __ xor_(Operand(current_character()), Immediate(0x01));
+    __ mov(Operand(eax), current_character());
+    __ xor_(Operand(eax), Immediate(0x01));
     // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
-    __ sub(Operand(current_character()), Immediate(0x0b));
-    __ cmp(current_character(), 0x0c - 0x0b);
+    __ sub(Operand(eax), Immediate(0x0b));
+    __ cmp(eax, 0x0c - 0x0b);
     BranchOrBacktrack(below_equal, on_no_match);
     if (mode_ == UC16) {
       // Compare original value to 0x2028 and 0x2029, using the already
       // computed (current_char ^ 0x01 - 0x0b). I.e., check for
       // 0x201d (0x2028 - 0x0b) or 0x201e.
-      __ sub(Operand(current_character()), Immediate(0x2028 - 0x0b));
-      __ cmp(current_character(), 1);
+      __ sub(Operand(eax), Immediate(0x2028 - 0x0b));
+      __ cmp(eax, 0x2029 - 0x2028);
       BranchOrBacktrack(below_equal, on_no_match);
     }
     return true;
   }
+  case 'w': {
+    Label done, check_digits;
+    __ cmp(Operand(current_character()), Immediate('9'));
+    __ j(less_equal, &check_digits);
+    __ cmp(Operand(current_character()), Immediate('_'));
+    __ j(equal, &done);
+    // Convert to lower case if letter.
+    __ mov(Operand(eax), current_character());
+    __ or_(eax, 0x20);
+    // check current character in range ['a'..'z'], nondestructively.
+    __ sub(Operand(eax), Immediate('a'));
+    __ cmp(Operand(eax), Immediate('z' - 'a'));
+    BranchOrBacktrack(above, on_no_match);
+    __ jmp(&done);
+    __ bind(&check_digits);
+    // Check current character in range ['0'..'9'].
+    __ cmp(Operand(current_character()), Immediate('0'));
+    BranchOrBacktrack(below, on_no_match);
+    __ bind(&done);
+
+    return true;
+  }
+  case 'W': {
+    Label done, check_digits;
+    __ cmp(Operand(current_character()), Immediate('9'));
+    __ j(less_equal, &check_digits);
+    __ cmp(Operand(current_character()), Immediate('_'));
+    BranchOrBacktrack(equal, on_no_match);
+    // Convert to lower case if letter.
+    __ mov(Operand(eax), current_character());
+    __ or_(eax, 0x20);
+    // check current character in range ['a'..'z'], nondestructively.
+    __ sub(Operand(eax), Immediate('a'));
+    __ cmp(Operand(eax), Immediate('z' - 'a'));
+    BranchOrBacktrack(below_equal, on_no_match);
+    __ jmp(&done);
+    __ bind(&check_digits);
+    // Check current character in range ['0'..'9'].
+    __ cmp(Operand(current_character()), Immediate('0'));
+    BranchOrBacktrack(above_equal, on_no_match);
+    __ bind(&done);
+    return true;
+  }
+  // Non-standard classes (with no syntactic shorthand) used internally.
   case '*':
     // Match any character.
-    if (check_offset) {
-      CheckPosition(cp_offset, on_no_match);
+    return true;
+  case 'n': {
+    // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 or 0x2029).
+    // The opposite of '.'.
+    __ mov(Operand(eax), current_character());
+    __ xor_(Operand(eax), Immediate(0x01));
+    // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
+    __ sub(Operand(eax), Immediate(0x0b));
+    __ cmp(eax, 0x0c - 0x0b);
+    if (mode_ == ASCII) {
+      BranchOrBacktrack(above, on_no_match);
+    } else {
+      Label done;
+      BranchOrBacktrack(below_equal, &done);
+      ASSERT_EQ(UC16, mode_);
+      // Compare original value to 0x2028 and 0x2029, using the already
+      // computed (current_char ^ 0x01 - 0x0b). I.e., check for
+      // 0x201d (0x2028 - 0x0b) or 0x201e.
+      __ sub(Operand(eax), Immediate(0x2028 - 0x0b));
+      __ cmp(eax, 1);
+      BranchOrBacktrack(above, on_no_match);
+      __ bind(&done);
     }
     return true;
-  // No custom implementation (yet): w, W, s(UC16), S(UC16).
+  }
+  // No custom implementation (yet): s(UC16), S(UC16).
   default:
     return false;
   }

src/ia32/regexp-macro-assembler-ia32.h

@@ -78,10 +78,7 @@ class RegExpMacroAssemblerIA32: public NativeRegExpMacroAssembler {
   // Checks whether the given offset from the current position is before
   // the end of the string.
   virtual void CheckPosition(int cp_offset, Label* on_outside_input);
-  virtual bool CheckSpecialCharacterClass(uc16 type,
-                                          int cp_offset,
-                                          bool check_offset,
-                                          Label* on_no_match);
+  virtual bool CheckSpecialCharacterClass(uc16 type, Label* on_no_match);
   virtual void Fail();
   virtual Handle<Object> GetCode(Handle<String> source);
   virtual void GoTo(Label* label);

src/jsregexp.h

@@ -177,6 +177,57 @@ class RegExpImpl {
 };
 
 
+// Represents the location of one element relative to the intersection of
+// two sets. Corresponds to the four areas of a Venn diagram.
+enum ElementInSetsRelation {
+  kInsideNone = 0,
+  kInsideFirst = 1,
+  kInsideSecond = 2,
+  kInsideBoth = 3
+};
+
+
+// Represents the relation of two sets.
+// Sets can be either disjoint, partially or fully overlapping, or equal.
+class SetRelation BASE_EMBEDDED {
+ public:
+  // Relation is represented by a bit saying whether there are elements in
+  // one set that is not in the other, and a bit saying that there are elements
+  // that are in both sets.
+
+  // Location of an element. Corresponds to the internal areas of
+  // a Venn diagram.
+  enum {
+    kInFirst = 1 << kInsideFirst,
+    kInSecond = 1 << kInsideSecond,
+    kInBoth = 1 << kInsideBoth
+  };
+  SetRelation() : bits_(0) {}
+  ~SetRelation() {}
+  // Add the existence of objects in a particular
+  void SetElementsInFirstSet() { bits_ |= kInFirst; }
+  void SetElementsInSecondSet() { bits_ |= kInSecond; }
+  void SetElementsInBothSets() { bits_ |= kInBoth; }
+  // Check the currently known relation of the sets (common functions only,
+  // for other combinations, use value() to get the bits and check them
+  // manually).
+  // Sets are completely disjoint.
+  bool Disjoint() { return (bits_ & kInBoth) == 0; }
+  // Sets are equal.
+  bool Equals() { return (bits_ & (kInFirst | kInSecond)) == 0; }
+  // First set contains second.
+  bool Contains() { return (bits_ & kInSecond) == 0; }
+  // Second set contains first.
+  bool ContainedIn() { return (bits_ & kInFirst) == 0; }
+  bool NonTrivialIntersection() {
+    return (bits_ == (kInFirst | kInSecond | kInBoth));
+  }
+  int value() { return bits_; }
+ private:
+  int bits_;
+};
+
+
 class CharacterRange {
  public:
   CharacterRange() : from_(0), to_(0) { }
@@ -208,7 +259,39 @@ class CharacterRange {
                     Vector<const uc16> overlay,
                     ZoneList<CharacterRange>** included,
                     ZoneList<CharacterRange>** excluded);
-
+  // Whether a range list is in canonical form: Ranges ordered by from value,
+  // and ranges non-overlapping and non-adjacent.
+  static bool IsCanonical(ZoneList<CharacterRange>* ranges);
+  // Convert range list to canonical form. The characters covered by the ranges
+  // will still be the same, but no character is in more than one range, and
+  // adjacent ranges are merged. The resulting list may be shorter than the
+  // original, but cannot be longer.
+  static void Canonicalize(ZoneList<CharacterRange>* ranges);
+  // Check how the set of characters defined by a CharacterRange list relates
+  // to the set of word characters. List must be in canonical form.
+  static SetRelation WordCharacterRelation(ZoneList<CharacterRange>* ranges);
+  // Takes two character range lists (representing character sets) in canonical
+  // form and merges them.
+  // The characters that are only covered by the first set are added to
+  // first_set_only_out. the characters that are only in the second set are
+  // added to second_set_only_out, and the characters that are in both are
+  // added to both_sets_out.
+  // The pointers to first_set_only_out, second_set_only_out and both_sets_out
+  // should be to empty lists, but they need not be distinct, and may be NULL.
+  // If NULL, the characters are dropped, and if two arguments are the same
+  // pointer, the result is the union of the two sets that would be created
+  // if the pointers had been distinct.
+  // This way, the Merge function can compute all the usual set operations:
+  // union (all three out-sets are equal), intersection (only both_sets_out is
+  // non-NULL), and set difference (only first_set is non-NULL).
+  static void Merge(ZoneList<CharacterRange>* first_set,
+                    ZoneList<CharacterRange>* second_set,
+                    ZoneList<CharacterRange>* first_set_only_out,
+                    ZoneList<CharacterRange>* second_set_only_out,
+                    ZoneList<CharacterRange>* both_sets_out);
+  // Negate the contents of a character range in canonical form.
+  static void Negate(ZoneList<CharacterRange>* src,
+                     ZoneList<CharacterRange>* dst);
   static const int kRangeCanonicalizeMax = 0x346;
   static const int kStartMarker = (1 << 24);
   static const int kPayloadMask = (1 << 24) - 1;
@@ -482,7 +565,7 @@ class QuickCheckDetails {
 
 class RegExpNode: public ZoneObject {
  public:
-  RegExpNode() : trace_count_(0) { }
+  RegExpNode() : first_character_set_(NULL), trace_count_(0) { }
   virtual ~RegExpNode();
   virtual void Accept(NodeVisitor* visitor) = 0;
   // Generates a goto to this node or actually generates the code at this point.
@@ -533,8 +616,29 @@ class RegExpNode: public ZoneObject {
   SiblingList* siblings() { return &siblings_; }
   void set_siblings(SiblingList* other) { siblings_ = *other; }
 
+  // Return the set of possible next characters recognized by the regexp
+  // (or a safe subset, potentially the set of all characters).
+  ZoneList<CharacterRange>* FirstCharacterSet();
+
+  // Compute (if possible within the budget of traversed nodes) the
+  // possible first characters of the input matched by this node and
+  // its continuation. Returns the remaining budget after the computation.
+  // If the budget is spent, the result is negative, and the cached
+  // first_character_set_ value isn't set.
+  virtual int ComputeFirstCharacterSet(int budget);
+
+  // Get and set the cached first character set value.
+  ZoneList<CharacterRange>* first_character_set() {
+    return first_character_set_;
+  }
+  void set_first_character_set(ZoneList<CharacterRange>* character_set) {
+    first_character_set_ = character_set;
+  }
+
  protected:
   enum LimitResult { DONE, CONTINUE };
+  static const int kComputeFirstCharacterSetFail = -1;
+
   LimitResult LimitVersions(RegExpCompiler* compiler, Trace* trace);
 
   // Returns a sibling of this node whose interests and assumptions
@@ -555,9 +659,11 @@ class RegExpNode: public ZoneObject {
   virtual RegExpNode* Clone() = 0;
 
  private:
+  static const int kFirstCharBudget = 10;
   Label label_;
   NodeInfo info_;
   SiblingList siblings_;
+  ZoneList<CharacterRange>* first_character_set_;
   // This variable keeps track of how many times code has been generated for
   // this node (in different traces).  We don't keep track of where the
   // generated code is located unless the code is generated at the start of
@@ -648,7 +754,7 @@ class ActionNode: public SeqRegExpNode {
   // TODO(erikcorry): We should allow some action nodes in greedy loops.
   virtual int GreedyLoopTextLength() { return kNodeIsTooComplexForGreedyLoops; }
   virtual ActionNode* Clone() { return new ActionNode(*this); }
-
+  virtual int ComputeFirstCharacterSet(int budget);
  private:
   union {
     struct {
@@ -714,7 +820,7 @@ class TextNode: public SeqRegExpNode {
     return result;
   }
   void CalculateOffsets();
-
+  virtual int ComputeFirstCharacterSet(int budget);
  private:
   enum TextEmitPassType {
     NON_ASCII_MATCH,             // Check for characters that can't match.
@@ -744,7 +850,12 @@ class AssertionNode: public SeqRegExpNode {
     AT_START,
     AT_BOUNDARY,
     AT_NON_BOUNDARY,
-    AFTER_NEWLINE
+    AFTER_NEWLINE,
+    // Types not directly expressible in regexp syntax.
+    // Used for modifying a boundary node if its following character is
+    // known to be word and/or non-word.
+    AFTER_NONWORD_CHARACTER,
+    AFTER_WORD_CHARACTER
   };
   static AssertionNode* AtEnd(RegExpNode* on_success) {
     return new AssertionNode(AT_END, on_success);
@@ -768,8 +879,10 @@ class AssertionNode: public SeqRegExpNode {
                                     RegExpCompiler* compiler,
                                     int filled_in,
                                     bool not_at_start);
+  virtual int ComputeFirstCharacterSet(int budget);
   virtual AssertionNode* Clone() { return new AssertionNode(*this); }
   AssertionNodeType type() { return type_; }
+  void set_type(AssertionNodeType type) { type_ = type; }
  private:
   AssertionNode(AssertionNodeType t, RegExpNode* on_success)
       : SeqRegExpNode(on_success), type_(t) { }
@@ -797,7 +910,7 @@ class BackReferenceNode: public SeqRegExpNode {
     return;
   }
   virtual BackReferenceNode* Clone() { return new BackReferenceNode(*this); }
-
+  virtual int ComputeFirstCharacterSet(int budget);
  private:
   int start_reg_;
   int end_reg_;
@@ -819,7 +932,6 @@ class EndNode: public RegExpNode {
     UNREACHABLE();
   }
   virtual EndNode* Clone() { return new EndNode(*this); }
-
  private:
   Action action_;
 };
@@ -953,6 +1065,7 @@ class NegativeLookaheadChoiceNode: public ChoiceNode {
   // characters, but on a negative lookahead the negative branch did not take
   // part in that calculation (EatsAtLeast) so the assumptions don't hold.
   virtual bool try_to_emit_quick_check_for_alternative(int i) { return i != 0; }
+  virtual int ComputeFirstCharacterSet(int budget);
 };
 
 
@@ -971,6 +1084,7 @@ class LoopChoiceNode: public ChoiceNode {
                                     RegExpCompiler* compiler,
                                     int characters_filled_in,
                                     bool not_at_start);
+  virtual int ComputeFirstCharacterSet(int budget);
   virtual LoopChoiceNode* Clone() { return new LoopChoiceNode(*this); }
   RegExpNode* loop_node() { return loop_node_; }
   RegExpNode* continue_node() { return continue_node_; }
@@ -1126,7 +1240,7 @@ class Trace {
   void set_backtrack(Label* backtrack) { backtrack_ = backtrack; }
   void set_stop_node(RegExpNode* node) { stop_node_ = node; }
   void set_loop_label(Label* label) { loop_label_ = label; }
-  void set_characters_preloaded(int cpre) { characters_preloaded_ = cpre; }
+  void set_characters_preloaded(int count) { characters_preloaded_ = count; }
   void set_bound_checked_up_to(int to) { bound_checked_up_to_ = to; }
   void set_flush_budget(int to) { flush_budget_ = to; }
   void set_quick_check_performed(QuickCheckDetails* d) {

src/parser.cc

@@ -371,7 +371,7 @@ class RegExpBuilder: public ZoneObject {
   void AddAtom(RegExpTree* tree);
   void AddAssertion(RegExpTree* tree);
   void NewAlternative();  // '|'
-  void AddQuantifierToAtom(int min, int max, bool is_greedy);
+  void AddQuantifierToAtom(int min, int max, RegExpQuantifier::Type type);
   RegExpTree* ToRegExp();
  private:
   void FlushCharacters();
@@ -503,7 +503,9 @@ RegExpTree* RegExpBuilder::ToRegExp() {
 }
 
 
-void RegExpBuilder::AddQuantifierToAtom(int min, int max, bool is_greedy) {
+void RegExpBuilder::AddQuantifierToAtom(int min,
+                                        int max,
+                                        RegExpQuantifier::Type type) {
   if (pending_empty_) {
     pending_empty_ = false;
     return;
@@ -543,7 +545,7 @@ void RegExpBuilder::AddQuantifierToAtom(int min, int max, bool is_greedy) {
     UNREACHABLE();
     return;
   }
-  terms_.Add(new RegExpQuantifier(min, max, is_greedy, atom));
+  terms_.Add(new RegExpQuantifier(min, max, type, atom));
   LAST(ADD_TERM);
 }
 
@@ -4278,12 +4280,16 @@ RegExpTree* RegExpParser::ParseDisjunction() {
     default:
       continue;
     }
-    bool is_greedy = true;
+    RegExpQuantifier::Type type = RegExpQuantifier::GREEDY;
     if (current() == '?') {
-      is_greedy = false;
+      type = RegExpQuantifier::NON_GREEDY;
+      Advance();
+    } else if (FLAG_regexp_possessive_quantifier && current() == '+') {
+      // FLAG_regexp_possessive_quantifier is a debug-only flag.
+      type = RegExpQuantifier::POSSESSIVE;
       Advance();
     }
-    builder->AddQuantifierToAtom(min, max, is_greedy);
+    builder->AddQuantifierToAtom(min, max, type);
   }
 }
 

src/regexp-macro-assembler-tracer.cc

@@ -307,18 +307,11 @@ void RegExpMacroAssemblerTracer::CheckCharacters(Vector<const uc16> str,
 
 bool RegExpMacroAssemblerTracer::CheckSpecialCharacterClass(
     uc16 type,
-    int cp_offset,
-    bool check_offset,
     Label* on_no_match) {
   bool supported = assembler_->CheckSpecialCharacterClass(type,
-                                                          cp_offset,
-                                                          check_offset,
                                                           on_no_match);
-  PrintF(" CheckSpecialCharacterClass(type='%c', offset=%d, "
-             "check_offset=%s, label[%08x]): %s;\n",
+  PrintF(" CheckSpecialCharacterClass(type='%c', label[%08x]): %s;\n",
          type,
-         cp_offset,
-         check_offset ? "true" : "false",
          on_no_match,
          supported ? "true" : "false");
   return supported;

src/regexp-macro-assembler-tracer.h

@@ -69,8 +69,6 @@ class RegExpMacroAssemblerTracer: public RegExpMacroAssembler {
                                               uc16 and_with,
                                               Label* on_not_equal);
   virtual bool CheckSpecialCharacterClass(uc16 type,
-                                          int cp_offset,
-                                          bool check_offset,
                                           Label* on_no_match);
   virtual void Fail();
   virtual Handle<Object> GetCode(Handle<String> source);

src/regexp-macro-assembler.h

@@ -123,8 +123,6 @@ class RegExpMacroAssembler {
   // not have custom support.
   // May clobber the current loaded character.
   virtual bool CheckSpecialCharacterClass(uc16 type,
-                                          int cp_offset,
-                                          bool check_offset,
                                           Label* on_no_match) {
     return false;
   }

src/x64/regexp-macro-assembler-x64.cc

@@ -60,7 +60,7 @@ namespace internal {
  * - r8  : code object pointer. Used to convert between absolute and
  *         code-object-relative addresses.
  *
- * The registers rax, rbx, rcx, r9 and r11 are free to use for computations.
+ * The registers rax, rbx, r9 and r11 are free to use for computations.
  * If changed to use r12+, they should be saved as callee-save registers.
  *
  * Each call to a C++ method should retain these registers.
@@ -496,27 +496,22 @@ void RegExpMacroAssemblerX64::CheckNotCharacterAfterMinusAnd(
 
 
 bool RegExpMacroAssemblerX64::CheckSpecialCharacterClass(uc16 type,
-                                                         int cp_offset,
-                                                         bool check_offset,
                                                          Label* on_no_match) {
   // Range checks (c in min..max) are generally implemented by an unsigned
-  // (c - min) <= (max - min) check
+  // (c - min) <= (max - min) check, using the sequence:
+  //   lea(rax, Operand(current_character(), -min)) or sub(rax, Immediate(min))
+  //   cmp(rax, Immediate(max - min))
   switch (type) {
   case 's':
     // Match space-characters
     if (mode_ == ASCII) {
       // ASCII space characters are '\t'..'\r' and ' '.
-      if (check_offset) {
-        LoadCurrentCharacter(cp_offset, on_no_match);
-      } else {
-        LoadCurrentCharacterUnchecked(cp_offset, 1);
-      }
       Label success;
       __ cmpl(current_character(), Immediate(' '));
       __ j(equal, &success);
       // Check range 0x09..0x0d
-      __ subl(current_character(), Immediate('\t'));
-      __ cmpl(current_character(), Immediate('\r' - '\t'));
+      __ lea(rax, Operand(current_character(), -'\t'));
+      __ cmpl(rax, Immediate('\r' - '\t'));
       BranchOrBacktrack(above, on_no_match);
       __ bind(&success);
       return true;
@@ -524,72 +519,116 @@ bool RegExpMacroAssemblerX64::CheckSpecialCharacterClass(uc16 type,
     return false;
   case 'S':
     // Match non-space characters.
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
     if (mode_ == ASCII) {
       // ASCII space characters are '\t'..'\r' and ' '.
       __ cmpl(current_character(), Immediate(' '));
       BranchOrBacktrack(equal, on_no_match);
-      __ subl(current_character(), Immediate('\t'));
-      __ cmpl(current_character(), Immediate('\r' - '\t'));
+      __ lea(rax, Operand(current_character(), -'\t'));
+      __ cmpl(rax, Immediate('\r' - '\t'));
       BranchOrBacktrack(below_equal, on_no_match);
       return true;
     }
     return false;
   case 'd':
     // Match ASCII digits ('0'..'9')
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
-    __ subl(current_character(), Immediate('0'));
-    __ cmpl(current_character(), Immediate('9' - '0'));
+    __ lea(rax, Operand(current_character(), -'0'));
+    __ cmpl(rax, Immediate('9' - '0'));
     BranchOrBacktrack(above, on_no_match);
     return true;
   case 'D':
     // Match non ASCII-digits
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
-    __ subl(current_character(), Immediate('0'));
-    __ cmpl(current_character(), Immediate('9' - '0'));
+    __ lea(rax, Operand(current_character(), -'0'));
+    __ cmpl(rax, Immediate('9' - '0'));
     BranchOrBacktrack(below_equal, on_no_match);
     return true;
   case '.': {
     // Match non-newlines (not 0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
-    if (check_offset) {
-      LoadCurrentCharacter(cp_offset, on_no_match, 1);
-    } else {
-      LoadCurrentCharacterUnchecked(cp_offset, 1);
-    }
-    __ xor_(current_character(), Immediate(0x01));
+    __ movl(rax, current_character());
+    __ xor_(rax, Immediate(0x01));
     // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
-    __ subl(current_character(), Immediate(0x0b));
-    __ cmpl(current_character(), Immediate(0x0c - 0x0b));
+    __ subl(rax, Immediate(0x0b));
+    __ cmpl(rax, Immediate(0x0c - 0x0b));
     BranchOrBacktrack(below_equal, on_no_match);
     if (mode_ == UC16) {
       // Compare original value to 0x2028 and 0x2029, using the already
       // computed (current_char ^ 0x01 - 0x0b). I.e., check for
       // 0x201d (0x2028 - 0x0b) or 0x201e.
-      __ subl(current_character(), Immediate(0x2028 - 0x0b));
-      __ cmpl(current_character(), Immediate(1));
+      __ subl(rax, Immediate(0x2028 - 0x0b));
+      __ cmpl(rax, Immediate(0x2029 - 0x2028));
       BranchOrBacktrack(below_equal, on_no_match);
     }
     return true;
   }
+  case 'n': {
+    // Match newlines (0x0a('\n'), 0x0d('\r'), 0x2028 and 0x2029)
+    __ movl(rax, current_character());
+    __ xor_(rax, Immediate(0x01));
+    // See if current character is '\n'^1 or '\r'^1, i.e., 0x0b or 0x0c
+    __ subl(rax, Immediate(0x0b));
+    __ cmpl(rax, Immediate(0x0c - 0x0b));
+    if (mode_ == ASCII) {
+      BranchOrBacktrack(above, on_no_match);
+    } else {
+      Label done;
+      BranchOrBacktrack(below_equal, &done);
+      // Compare original value to 0x2028 and 0x2029, using the already
+      // computed (current_char ^ 0x01 - 0x0b). I.e., check for
+      // 0x201d (0x2028 - 0x0b) or 0x201e.
+      __ subl(rax, Immediate(0x2028 - 0x0b));
+      __ cmpl(rax, Immediate(0x2029 - 0x2028));
+      BranchOrBacktrack(above, on_no_match);
+      __ bind(&done);
+    }
+    return true;
+  }
+  case 'w': {
+    Label done, check_digits;
+    __ cmpl(current_character(), Immediate('9'));
+    __ j(less_equal, &check_digits);
+    __ cmpl(current_character(), Immediate('_'));
+    __ j(equal, &done);
+    // Convert to lower case if letter.
+    __ movl(rax, current_character());
+    __ orl(rax, Immediate(0x20));
+    // check rax in range ['a'..'z'].
+    __ subl(rax, Immediate('a'));
+    __ cmpl(rax, Immediate('z' - 'a'));
+    BranchOrBacktrack(above, on_no_match);
+    __ jmp(&done);
+    __ bind(&check_digits);
+    // Check current character in range ['0'..'9'].
+    __ cmpl(current_character(), Immediate('0'));
+    BranchOrBacktrack(below, on_no_match);
+    __ bind(&done);
+
+    return true;
+  }
+  case 'W': {
+    Label done, check_digits;
+    __ cmpl(current_character(), Immediate('9'));
+    __ j(less_equal, &check_digits);
+    __ cmpl(current_character(), Immediate('_'));
+    BranchOrBacktrack(equal, on_no_match);
+    // Convert to lower case if letter.
+    __ movl(rax, current_character());
+    __ orl(rax, Immediate(0x20));
+    // check current character in range ['a'..'z'], nondestructively.
+    __ subl(rax, Immediate('a'));
+    __ cmpl(rax, Immediate('z' - 'a'));
+    BranchOrBacktrack(below_equal, on_no_match);
+    __ jmp(&done);
+    __ bind(&check_digits);
+    // Check current character in range ['0'..'9'].
+    __ cmpl(current_character(), Immediate('0'));
+    BranchOrBacktrack(above_equal, on_no_match);
+    __ bind(&done);
+
+    return true;
+  }
   case '*':
     // Match any character.
-    if (check_offset) {
-      CheckPosition(cp_offset, on_no_match);
-    }
     return true;
-  // No custom implementation (yet): w, W, s(UC16), S(UC16).
+  // No custom implementation (yet): s(UC16), S(UC16).
   default:
     return false;
   }

src/x64/regexp-macro-assembler-x64.h

@@ -73,8 +73,6 @@ class RegExpMacroAssemblerX64: public NativeRegExpMacroAssembler {
   // the end of the string.
   virtual void CheckPosition(int cp_offset, Label* on_outside_input);
   virtual bool CheckSpecialCharacterClass(uc16 type,
-                                          int cp_offset,
-                                          bool check_offset,
                                           Label* on_no_match);
   virtual void Fail();
   virtual Handle<Object> GetCode(Handle<String> source);

test/cctest/test-regexp.cc

@@ -58,6 +58,16 @@
 using namespace v8::internal;
 
 
+static bool CheckParse(const char* input) {
+  V8::Initialize(NULL);
+  v8::HandleScope scope;
+  ZoneScope zone_scope(DELETE_ON_EXIT);
+  FlatStringReader reader(CStrVector(input));
+  RegExpCompileData result;
+  return v8::internal::ParseRegExp(&reader, false, &result);
+}
+
+
 static SmartPointer<const char> Parse(const char* input) {
   V8::Initialize(NULL);
   v8::HandleScope scope;
@@ -106,7 +116,7 @@ static MinMaxPair CheckMinMaxMatch(const char* input) {
 }
 
 
-
+#define CHECK_PARSE_ERROR(input) CHECK(!CheckParse(input))
 #define CHECK_PARSE_EQ(input, expected) CHECK_EQ(expected, *Parse(input))
 #define CHECK_SIMPLE(input, simple) CHECK_EQ(simple, CheckSimple(input));
 #define CHECK_MIN_MAX(input, min, max)                                         \
@@ -600,6 +610,34 @@ TEST(DispatchTableConstruction) {
   }
 }
 
+// Test of debug-only syntax.
+#ifdef DEBUG
+
+TEST(ParsePossessiveRepetition) {
+  bool old_flag_value = FLAG_regexp_possessive_quantifier;
+
+  // Enable possessive quantifier syntax.
+  FLAG_regexp_possessive_quantifier = true;
+
+  CHECK_PARSE_EQ("a*+", "(# 0 - p 'a')");
+  CHECK_PARSE_EQ("a++", "(# 1 - p 'a')");
+  CHECK_PARSE_EQ("a?+", "(# 0 1 p 'a')");
+  CHECK_PARSE_EQ("a{10,20}+", "(# 10 20 p 'a')");
+  CHECK_PARSE_EQ("za{10,20}+b", "(: 'z' (# 10 20 p 'a') 'b')");
+
+  // Disable possessive quantifier syntax.
+  FLAG_regexp_possessive_quantifier = false;
+
+  CHECK_PARSE_ERROR("a*+");
+  CHECK_PARSE_ERROR("a++");
+  CHECK_PARSE_ERROR("a?+");
+  CHECK_PARSE_ERROR("a{10,20}+");
+  CHECK_PARSE_ERROR("a{10,20}+b");
+
+  FLAG_regexp_possessive_quantifier = old_flag_value;
+}
+
+#endif
 
 // Tests of interpreter.
 
@@ -1550,7 +1588,68 @@ TEST(CharClassDifference) {
 }
 
 
+TEST(CanonicalizeCharacterSets) {
+  ZoneScope scope(DELETE_ON_EXIT);
+  ZoneList<CharacterRange>* list = new ZoneList<CharacterRange>(4);
+  CharacterSet set(list);
+
+  list->Add(CharacterRange(10, 20));
+  list->Add(CharacterRange(30, 40));
+  list->Add(CharacterRange(50, 60));
+  set.Canonicalize();
+  ASSERT_EQ(3, list->length());
+  ASSERT_EQ(10, list->at(0).from());
+  ASSERT_EQ(20, list->at(0).to());
+  ASSERT_EQ(30, list->at(1).from());
+  ASSERT_EQ(40, list->at(1).to());
+  ASSERT_EQ(50, list->at(2).from());
+  ASSERT_EQ(60, list->at(2).to());
+
+  list->Rewind(0);
+  list->Add(CharacterRange(10, 20));
+  list->Add(CharacterRange(50, 60));
+  list->Add(CharacterRange(30, 40));
+  set.Canonicalize();
+  ASSERT_EQ(3, list->length());
+  ASSERT_EQ(10, list->at(0).from());
+  ASSERT_EQ(20, list->at(0).to());
+  ASSERT_EQ(30, list->at(1).from());
+  ASSERT_EQ(40, list->at(1).to());
+  ASSERT_EQ(50, list->at(2).from());
+  ASSERT_EQ(60, list->at(2).to());
+
+  list->Rewind(0);
+  list->Add(CharacterRange(30, 40));
+  list->Add(CharacterRange(10, 20));
+  list->Add(CharacterRange(25, 25));
+  list->Add(CharacterRange(100, 100));
+  list->Add(CharacterRange(1, 1));
+  set.Canonicalize();
+  ASSERT_EQ(5, list->length());
+  ASSERT_EQ(1, list->at(0).from());
+  ASSERT_EQ(1, list->at(0).to());
+  ASSERT_EQ(10, list->at(1).from());
+  ASSERT_EQ(20, list->at(1).to());
+  ASSERT_EQ(25, list->at(2).from());
+  ASSERT_EQ(25, list->at(2).to());
+  ASSERT_EQ(30, list->at(3).from());
+  ASSERT_EQ(40, list->at(3).to());
+  ASSERT_EQ(100, list->at(4).from());
+  ASSERT_EQ(100, list->at(4).to());
+
+  list->Rewind(0);
+  list->Add(CharacterRange(10, 19));
+  list->Add(CharacterRange(21, 30));
+  list->Add(CharacterRange(20, 20));
+  set.Canonicalize();
+  ASSERT_EQ(1, list->length());
+  ASSERT_EQ(10, list->at(0).from());
+  ASSERT_EQ(30, list->at(0).to());
+}
+
+
+
 TEST(Graph) {
   V8::Initialize(NULL);
-  Execute("(?:(?:x(.))?\1)+$", false, true, true);
+  Execute("\\b\\w+\\b", false, true, true);
 }