Port and integrate the irregexp engine from V8

runtime/vm/intrinsifier_ia32.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 //
 // The intrinsic code below is executed before a method has built its frame.
 // The return address is on the stack and the arguments below it.
 // Registers EDX (arguments descriptor) and ECX (function) must be preserved.
 // Each intrinsification method returns true if the corresponding
 // Dart method was intrinsified.
 
@@ skipping 1682 matching lines @@
 void Intrinsifier::OneByteString_equality(Assembler* assembler) {
   StringEquality(assembler, kOneByteStringCid);
 }
 
 
 void Intrinsifier::TwoByteString_equality(Assembler* assembler) {
   StringEquality(assembler, kTwoByteStringCid);
 }
 
 
+static void GenerateRegexpSpecializationFor(
+    Assembler* assembler,
+    intptr_t cid,
+    Label* entry,
+    Label* exit,
+    bool jump_kind = Assembler::kFarJump) {
+  ASSERT(cid == kOneByteStringCid || cid == kExternalOneByteStringCid ||
+         cid == kTwoByteStringCid || cid == kExternalTwoByteStringCid);
+
+  __ Bind(entry);
+
+  // The passed string is of class cid.
+
+  __ movl(EAX, FieldAddress(EBX, JSRegExp::function_offset(cid)));
+
+  // Set the subject if it is currently unset.
+
+  __ LoadObject(ECX, Object::null_object());
+  __ cmpl(ECX, FieldAddress(EBX, JSRegExp::subject_offset(cid)));
+  __ j(NOT_EQUAL, exit, jump_kind);
+
+  __ StoreIntoObject(EBX,
+                     FieldAddress(EBX, JSRegExp::subject_offset(cid)),
+                     EDI,
+                     false);
+
+  __ jmp(exit, jump_kind);
+}
+
+
+void Intrinsifier::JSRegExp_ExecuteMatch(Assembler* assembler) {
+  Label is_one_byte_string;
+  Label is_two_byte_string;
+  Label is_external_one_byte_string;
+  Label is_external_two_byte_string;
+  Label fallthrough;
+
+  static const intptr_t kRegExpParamOffset = + 3 * kWordSize;
+  static const intptr_t kStringParamOffset = + 2 * kWordSize;
+  // const Smi& start_index is located at   (+ 1 * kWordSize).
+
+  // Register assignments are as follows:
+  // EAX: The appropriate specialized matcher function.
+  // EBX: The regexp object.
+  // ECX: Temp.
+  // EDI: Temp, Pointer to the function's code which we then tail-call.
+
+  __ movl(EBX, Address(ESP, kRegExpParamOffset));
+  __ movl(EDI, Address(ESP, kStringParamOffset));
+
+  // Check which string class we have been passed.
+
+  __ CompareClassId(EDI, kOneByteStringCid, ECX);
+  __ j(EQUAL, &is_one_byte_string);
+
+  __ CompareClassId(EDI, kTwoByteStringCid, ECX);
+  __ j(EQUAL, &is_two_byte_string);
+
+  __ CompareClassId(EDI, kExternalOneByteStringCid, ECX);
+  __ j(EQUAL, &is_external_one_byte_string);
+
+  __ CompareClassId(EDI, kExternalTwoByteStringCid, ECX);
+  __ j(EQUAL, &is_external_two_byte_string);
+
+  __ Stop("Unexpected class id");
+
+  GenerateRegexpSpecializationFor(assembler, kExternalTwoByteStringCid,
+                                  &is_external_two_byte_string, &fallthrough);
+  GenerateRegexpSpecializationFor(assembler, kExternalOneByteStringCid,
+                                  &is_external_one_byte_string, &fallthrough);
+  GenerateRegexpSpecializationFor(assembler, kTwoByteStringCid,
+                                  &is_two_byte_string, &fallthrough,
+                                  Assembler::kNearJump);
+  GenerateRegexpSpecializationFor(assembler, kOneByteStringCid,
+                                  &is_one_byte_string, &fallthrough,
+                                  Assembler::kNearJump);
+
+  // EAX contains the appropriate specialized function.
+
+  __ Bind(&fallthrough);
+
+  // Registers have been set up for the lazy compile stub at this point.
+  // It expects the function in EAX, the argument descriptor in EDX, and
+  // IC-Data in ECX. Irregexp generated functions do not use either the

[Florian Schneider, 2014/10/01 17:04:13]

ECX is destroyed by GenerateRegexpSpecializationFor - it should not matter, as
long as there is a GC-safe value in there. 

Object::null_object() will work, but please update the comment. The IC data are
only required when calling the InlineCacheStub.

+  // arguments descriptor or IC-Data, and hence these are not set here.
+
+  // Tail-call the function.
+  __ movl(EDI, FieldAddress(EAX, Function::instructions_offset()));
+  __ addl(EDI, Immediate(Instructions::HeaderSize() - kHeapObjectTag));
+  __ jmp(EDI);
+}
+
+
 // On stack: user tag (+1), return-address (+0).
 void Intrinsifier::UserTag_makeCurrent(Assembler* assembler) {
   Isolate* isolate = Isolate::Current();
   const Address current_tag_addr =
       Address::Absolute(reinterpret_cast<uword>(isolate) +
                         Isolate::current_tag_offset());
   const Address user_tag_addr =
       Address::Absolute(reinterpret_cast<uword>(isolate) +
                         Isolate::user_tag_offset());
   // EAX: Current user tag.
@@ skipping 28 matching lines @@
                         Isolate::current_tag_offset());
   // Set return value to Isolate::current_tag_.
   __ movl(EAX, current_tag_addr);
   __ ret();
 }
 
 #undef __
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_IA32