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

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

Index: src/arm/regexp-macro-assembler-arm.cc
diff --git a/src/arm/regexp-macro-assembler-arm.cc b/src/arm/regexp-macro-assembler-arm.cc
index 24b6a9c81a8b2dd6e09040a2cfe65075c32437a9..5b17dc18ad5a8d8018dcff5c96274f1c15e9139d 100644
--- a/src/arm/regexp-macro-assembler-arm.cc
+++ b/src/arm/regexp-macro-assembler-arm.cc
@@ -459,8 +459,6 @@ 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
@@ -469,11 +467,6 @@ 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);
@@ -487,11 +480,6 @@ 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(' '));
@@ -504,33 +492,18 @@ 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));
@@ -546,13 +519,71 @@ 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;
   }