[regexp] do not assume short external strings have a minimum size.

src/x64/code-stubs-x64.cc

Index: src/x64/code-stubs-x64.cc
diff --git a/src/x64/code-stubs-x64.cc b/src/x64/code-stubs-x64.cc
index bb008295f01c01c92ecc1c703279c54b086f1580..f489a90c37d9e46a541f62083558e4d6b7efe478 100644
--- a/src/x64/code-stubs-x64.cc
+++ b/src/x64/code-stubs-x64.cc
@@ -644,35 +644,34 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
   __ movp(rdi, args.GetArgumentOperand(SUBJECT_STRING_ARGUMENT_INDEX));
   __ JumpIfSmi(rdi, &runtime);
   __ movp(r15, rdi);  // Make a copy of the original subject string.
-  __ movp(rbx, FieldOperand(rdi, HeapObject::kMapOffset));
-  __ movzxbl(rbx, FieldOperand(rbx, Map::kInstanceTypeOffset));
   // rax: RegExp data (FixedArray)
   // rdi: subject string
   // r15: subject string
   // Handle subject string according to its encoding and representation:
   // (1) Sequential two byte?  If yes, go to (9).
-  // (2) Sequential one byte?  If yes, go to (6).
-  // (3) Anything but sequential or cons?  If yes, go to (7).
-  // (4) Cons string.  If the string is flat, replace subject with first string.
-  //     Otherwise bailout.
-  // (5a) Is subject sequential two byte?  If yes, go to (9).
-  // (5b) Is subject external?  If yes, go to (8).
-  // (6) One byte sequential.  Load regexp code for one byte.
+  // (2) Sequential one byte?  If yes, go to (5).
+  // (3) Sequential or cons?  If not, go to (6).
+  // (4) Cons string.  If the string is flat, replace subject with first string
+  //     and go to (1). Otherwise bail out to runtime.
+  // (5) One byte sequential.  Load regexp code for one byte.
   // (E) Carry on.
   /// [...]
 
   // Deferred code at the end of the stub:
-  // (7) Not a long external string?  If yes, go to (10).
-  // (8) External string.  Make it, offset-wise, look like a sequential string.
-  // (8a) Is the external string one byte?  If yes, go to (6).
-  // (9) Two byte sequential.  Load regexp code for one byte. Go to (E).
+  // (6) Long external string?  If not, go to (10).
+  // (7) External string.  Make it, offset-wise, look like a sequential string.
+  // (8) Is the external string one byte?  If yes, go to (5).
+  // (9) Two byte sequential.  Load regexp code for two byte. Go to (E).
   // (10) Short external string or not a string?  If yes, bail out to runtime.
-  // (11) Sliced string.  Replace subject with parent. Go to (5a).
+  // (11) Sliced string.  Replace subject with parent. Go to (1).
+
+  Label seq_one_byte_string /* 5 */, seq_two_byte_string /* 9 */,
+      external_string /* 7 */, check_underlying /* 1 */,
+      not_seq_nor_cons /* 6 */, check_code /* E */, not_long_external /* 10 */;
 
-  Label seq_one_byte_string /* 6 */, seq_two_byte_string /* 9 */,
-        external_string /* 8 */, check_underlying /* 5a */,
-        not_seq_nor_cons /* 7 */, check_code /* E */,
-        not_long_external /* 10 */;
+  __ bind(&check_underlying);
+  __ movp(rbx, FieldOperand(rdi, HeapObject::kMapOffset));
+  __ movzxbl(rbx, FieldOperand(rbx, Map::kInstanceTypeOffset));
 
   // (1) Sequential two byte?  If yes, go to (9).
   __ andb(rbx, Immediate(kIsNotStringMask |
@@ -682,14 +681,14 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
   STATIC_ASSERT((kStringTag | kSeqStringTag | kTwoByteStringTag) == 0);
   __ j(zero, &seq_two_byte_string);  // Go to (9).
 
-  // (2) Sequential one byte?  If yes, go to (6).
+  // (2) Sequential one byte?  If yes, go to (5).
   // Any other sequential string must be one byte.
   __ andb(rbx, Immediate(kIsNotStringMask |
                          kStringRepresentationMask |
                          kShortExternalStringMask));
-  __ j(zero, &seq_one_byte_string, Label::kNear);  // Go to (6).
+  __ j(zero, &seq_one_byte_string, Label::kNear);  // Go to (5).
 
-  // (3) Anything but sequential or cons?  If yes, go to (7).
+  // (3) Sequential or cons?  If not, go to (6).
   // We check whether the subject string is a cons, since sequential strings
   // have already been covered.
   STATIC_ASSERT(kConsStringTag < kExternalStringTag);
@@ -697,7 +696,7 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
   STATIC_ASSERT(kIsNotStringMask > kExternalStringTag);
   STATIC_ASSERT(kShortExternalStringTag > kExternalStringTag);
   __ cmpp(rbx, Immediate(kExternalStringTag));
-  __ j(greater_equal, &not_seq_nor_cons);  // Go to (7).
+  __ j(greater_equal, &not_seq_nor_cons);  // Go to (6).
 
   // (4) Cons string.  Check that it's flat.
   // Replace subject with first string and reload instance type.
@@ -705,22 +704,9 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
                  Heap::kempty_stringRootIndex);
   __ j(not_equal, &runtime);
   __ movp(rdi, FieldOperand(rdi, ConsString::kFirstOffset));
-  __ bind(&check_underlying);
-  __ movp(rbx, FieldOperand(rdi, HeapObject::kMapOffset));
-  __ movp(rbx, FieldOperand(rbx, Map::kInstanceTypeOffset));
+  __ jmp(&check_underlying);
 
-  // (5a) Is subject sequential two byte?  If yes, go to (9).
-  __ testb(rbx, Immediate(kStringRepresentationMask | kStringEncodingMask));
-  STATIC_ASSERT((kSeqStringTag | kTwoByteStringTag) == 0);
-  __ j(zero, &seq_two_byte_string);  // Go to (9).
-  // (5b) Is subject external?  If yes, go to (8).
-  __ testb(rbx, Immediate(kStringRepresentationMask));
-  // The underlying external string is never a short external string.
-  STATIC_ASSERT(ExternalString::kMaxShortLength < ConsString::kMinLength);
-  STATIC_ASSERT(ExternalString::kMaxShortLength < SlicedString::kMinLength);
-  __ j(not_zero, &external_string);  // Go to (8)
-
-  // (6) One byte sequential.  Load regexp code for one byte.
+  // (5) One byte sequential.  Load regexp code for one byte.
   __ bind(&seq_one_byte_string);
   // rax: RegExp data (FixedArray)
   __ movp(r11, FieldOperand(rax, JSRegExp::kDataOneByteCodeOffset));
@@ -967,12 +953,12 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
   __ TailCallRuntime(Runtime::kRegExpExec);
 
   // Deferred code for string handling.
-  // (7) Not a long external string?  If yes, go to (10).
+  // (6) Long external string?  If not, go to (10).
   __ bind(&not_seq_nor_cons);
   // Compare flags are still set from (3).
   __ j(greater, &not_long_external, Label::kNear);  // Go to (10).
 
-  // (8) External string.  Short external strings have been ruled out.
+  // (7) External string.  Short external strings have been ruled out.
   __ bind(&external_string);
   __ movp(rbx, FieldOperand(rdi, HeapObject::kMapOffset));
   __ movzxbl(rbx, FieldOperand(rbx, Map::kInstanceTypeOffset));
@@ -987,13 +973,13 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
   STATIC_ASSERT(SeqTwoByteString::kHeaderSize == SeqOneByteString::kHeaderSize);
   __ subp(rdi, Immediate(SeqTwoByteString::kHeaderSize - kHeapObjectTag));
   STATIC_ASSERT(kTwoByteStringTag == 0);
-  // (8a) Is the external string one byte?  If yes, go to (6).
+  // (8) Is the external string one byte?  If yes, go to (5).
   __ testb(rbx, Immediate(kStringEncodingMask));
-  __ j(not_zero, &seq_one_byte_string);  // Goto (6).
+  __ j(not_zero, &seq_one_byte_string);  // Go to (5).
 
   // rdi: subject string (flat two-byte)
   // rax: RegExp data (FixedArray)
-  // (9) Two byte sequential.  Load regexp code for one byte.  Go to (E).
+  // (9) Two byte sequential.  Load regexp code for two byte.  Go to (E).
   __ bind(&seq_two_byte_string);
   __ movp(r11, FieldOperand(rax, JSRegExp::kDataUC16CodeOffset));
   __ Set(rcx, 0);  // Type is two byte.
@@ -1006,7 +992,7 @@ void RegExpExecStub::Generate(MacroAssembler* masm) {
   __ testb(rbx, Immediate(kIsNotStringMask | kShortExternalStringMask));
   __ j(not_zero, &runtime);
 
-  // (11) Sliced string.  Replace subject with parent. Go to (5a).
+  // (11) Sliced string.  Replace subject with parent. Go to (1).
   // Load offset into r14 and replace subject string with parent.
   __ SmiToInteger32(r14, FieldOperand(rdi, SlicedString::kOffsetOffset));
   __ movp(rdi, FieldOperand(rdi, SlicedString::kParentOffset));