Port and integrate the irregexp engine from V8

runtime/vm/intrinsifier_ia32.cc

Index: runtime/vm/intrinsifier_ia32.cc
diff --git a/runtime/vm/intrinsifier_ia32.cc b/runtime/vm/intrinsifier_ia32.cc
index 6e24ad845dd1c0ccf5158b1bb2f25db8a77d20ee..79e54d403581108bc22f78c23ea2684f77b33ff9 100644
--- a/runtime/vm/intrinsifier_ia32.cc
+++ b/runtime/vm/intrinsifier_ia32.cc
@@ -1700,6 +1700,95 @@ void Intrinsifier::TwoByteString_equality(Assembler* assembler) {
 }
 
 
+static void GenerateRegexpSpecializationFor(
+    Assembler* assembler,
+    intptr_t cid,
+    Label* entry,
+    Label* exit,
+    bool jump_kind = Assembler::kFarJump) {
+  ASSERT(cid == kOneByteStringCid || cid == kTwoByteStringCid ||
+         cid == kExternalOneByteStringCid || 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 (one- or two-byte 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 if the string is a one byte string.
+
+  __ 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);
+
+  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 (one- or two-byte string) function.
+
+  __ Bind(&fallthrough);
+
+  // Registers have been set up for the lazy compile stub at this point.
+  // It expects the function in EAX while; EDX and ECX are unused.
+

[srdjan, 2014/09/22 22:34:07]

Registers EDX (arguments descriptor) and ECX (function) must be normally
preserved. Please explain here why it is OK not do so.

[jgruber1, 2014/09/23 10:58:47]

Done.

+  // 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();