1 // Copyright 2017 the V8 project authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4
5 #include "src/builtins/builtins-string-gen.h"
6
7 #include "src/builtins/builtins-regexp-gen.h"
8 #include "src/builtins/builtins-utils-gen.h"
9 #include "src/builtins/builtins.h"
10 #include "src/codegen/code-factory.h"
11 #include "src/execution/protectors.h"
12 #include "src/heap/factory-inl.h"
13 #include "src/heap/heap-inl.h"
14 #include "src/logging/counters.h"
15 #include "src/objects/objects.h"
16 #include "src/objects/property-cell.h"
17
18 namespace v8 {
19 namespace internal {
20
21 using Node = compiler::Node;
22
DirectStringData(TNode<String> string,TNode<Word32T> string_instance_type)23 TNode<RawPtrT> StringBuiltinsAssembler::DirectStringData(
24 TNode<String> string, TNode<Word32T> string_instance_type) {
25 // Compute the effective offset of the first character.
26 TVARIABLE(RawPtrT, var_data);
27 Label if_sequential(this), if_external(this), if_join(this);
28 Branch(Word32Equal(Word32And(string_instance_type,
29 Int32Constant(kStringRepresentationMask)),
30 Int32Constant(kSeqStringTag)),
31 &if_sequential, &if_external);
32
33 BIND(&if_sequential);
34 {
35 var_data = RawPtrAdd(
36 ReinterpretCast<RawPtrT>(BitcastTaggedToWord(string)),
37 IntPtrConstant(SeqOneByteString::kHeaderSize - kHeapObjectTag));
38 Goto(&if_join);
39 }
40
41 BIND(&if_external);
42 {
43 // This is only valid for ExternalStrings where the resource data
44 // pointer is cached (i.e. no uncached external strings).
45 CSA_ASSERT(this, Word32NotEqual(
46 Word32And(string_instance_type,
47 Int32Constant(kUncachedExternalStringMask)),
48 Int32Constant(kUncachedExternalStringTag)));
49 var_data = LoadExternalStringResourceDataPtr(CAST(string));
50 Goto(&if_join);
51 }
52
53 BIND(&if_join);
54 return var_data.value();
55 }
56
DispatchOnStringEncodings(TNode<Word32T> const lhs_instance_type,TNode<Word32T> const rhs_instance_type,Label * if_one_one,Label * if_one_two,Label * if_two_one,Label * if_two_two)57 void StringBuiltinsAssembler::DispatchOnStringEncodings(
58 TNode<Word32T> const lhs_instance_type,
59 TNode<Word32T> const rhs_instance_type, Label* if_one_one,
60 Label* if_one_two, Label* if_two_one, Label* if_two_two) {
61 STATIC_ASSERT(kStringEncodingMask == 0x8);
62 STATIC_ASSERT(kTwoByteStringTag == 0x0);
63 STATIC_ASSERT(kOneByteStringTag == 0x8);
64
65 // First combine the encodings.
66
67 const TNode<Int32T> encoding_mask = Int32Constant(kStringEncodingMask);
68 const TNode<Word32T> lhs_encoding =
69 Word32And(lhs_instance_type, encoding_mask);
70 const TNode<Word32T> rhs_encoding =
71 Word32And(rhs_instance_type, encoding_mask);
72
73 const TNode<Word32T> combined_encodings =
74 Word32Or(lhs_encoding, Word32Shr(rhs_encoding, 1));
75
76 // Then dispatch on the combined encoding.
77
78 Label unreachable(this, Label::kDeferred);
79
80 int32_t values[] = {
81 kOneByteStringTag | (kOneByteStringTag >> 1),
82 kOneByteStringTag | (kTwoByteStringTag >> 1),
83 kTwoByteStringTag | (kOneByteStringTag >> 1),
84 kTwoByteStringTag | (kTwoByteStringTag >> 1),
85 };
86 Label* labels[] = {
87 if_one_one, if_one_two, if_two_one, if_two_two,
88 };
89
90 STATIC_ASSERT(arraysize(values) == arraysize(labels));
91 Switch(combined_encodings, &unreachable, values, labels, arraysize(values));
92
93 BIND(&unreachable);
94 Unreachable();
95 }
96
97 template <typename SubjectChar, typename PatternChar>
CallSearchStringRaw(const TNode<RawPtrT> subject_ptr,const TNode<IntPtrT> subject_length,const TNode<RawPtrT> search_ptr,const TNode<IntPtrT> search_length,const TNode<IntPtrT> start_position)98 TNode<IntPtrT> StringBuiltinsAssembler::CallSearchStringRaw(
99 const TNode<RawPtrT> subject_ptr, const TNode<IntPtrT> subject_length,
100 const TNode<RawPtrT> search_ptr, const TNode<IntPtrT> search_length,
101 const TNode<IntPtrT> start_position) {
102 const TNode<ExternalReference> function_addr = ExternalConstant(
103 ExternalReference::search_string_raw<SubjectChar, PatternChar>());
104 const TNode<ExternalReference> isolate_ptr =
105 ExternalConstant(ExternalReference::isolate_address(isolate()));
106
107 MachineType type_ptr = MachineType::Pointer();
108 MachineType type_intptr = MachineType::IntPtr();
109
110 const TNode<IntPtrT> result = UncheckedCast<IntPtrT>(CallCFunction(
111 function_addr, type_intptr, std::make_pair(type_ptr, isolate_ptr),
112 std::make_pair(type_ptr, subject_ptr),
113 std::make_pair(type_intptr, subject_length),
114 std::make_pair(type_ptr, search_ptr),
115 std::make_pair(type_intptr, search_length),
116 std::make_pair(type_intptr, start_position)));
117
118 return result;
119 }
120
PointerToStringDataAtIndex(TNode<RawPtrT> string_data,TNode<IntPtrT> index,String::Encoding encoding)121 TNode<RawPtrT> StringBuiltinsAssembler::PointerToStringDataAtIndex(
122 TNode<RawPtrT> string_data, TNode<IntPtrT> index,
123 String::Encoding encoding) {
124 const ElementsKind kind = (encoding == String::ONE_BYTE_ENCODING)
125 ? UINT8_ELEMENTS
126 : UINT16_ELEMENTS;
127 TNode<IntPtrT> offset_in_bytes = ElementOffsetFromIndex(index, kind);
128 return RawPtrAdd(string_data, offset_in_bytes);
129 }
130
GenerateStringEqual(TNode<String> left,TNode<String> right)131 void StringBuiltinsAssembler::GenerateStringEqual(TNode<String> left,
132 TNode<String> right) {
133 TVARIABLE(String, var_left, left);
134 TVARIABLE(String, var_right, right);
135 Label if_equal(this), if_notequal(this), if_indirect(this, Label::kDeferred),
136 restart(this, {&var_left, &var_right});
137
138 TNode<IntPtrT> lhs_length = LoadStringLengthAsWord(left);
139 TNode<IntPtrT> rhs_length = LoadStringLengthAsWord(right);
140
141 // Strings with different lengths cannot be equal.
142 GotoIf(WordNotEqual(lhs_length, rhs_length), &if_notequal);
143
144 Goto(&restart);
145 BIND(&restart);
146 TNode<String> lhs = var_left.value();
147 TNode<String> rhs = var_right.value();
148
149 TNode<Uint16T> lhs_instance_type = LoadInstanceType(lhs);
150 TNode<Uint16T> rhs_instance_type = LoadInstanceType(rhs);
151
152 StringEqual_Core(lhs, lhs_instance_type, rhs, rhs_instance_type, lhs_length,
153 &if_equal, &if_notequal, &if_indirect);
154
155 BIND(&if_indirect);
156 {
157 // Try to unwrap indirect strings, restart the above attempt on success.
158 MaybeDerefIndirectStrings(&var_left, lhs_instance_type, &var_right,
159 rhs_instance_type, &restart);
160
161 TailCallRuntime(Runtime::kStringEqual, NoContextConstant(), lhs, rhs);
162 }
163
164 BIND(&if_equal);
165 Return(TrueConstant());
166
167 BIND(&if_notequal);
168 Return(FalseConstant());
169 }
170
StringEqual_Core(TNode<String> lhs,TNode<Word32T> lhs_instance_type,TNode<String> rhs,TNode<Word32T> rhs_instance_type,TNode<IntPtrT> length,Label * if_equal,Label * if_not_equal,Label * if_indirect)171 void StringBuiltinsAssembler::StringEqual_Core(
172 TNode<String> lhs, TNode<Word32T> lhs_instance_type, TNode<String> rhs,
173 TNode<Word32T> rhs_instance_type, TNode<IntPtrT> length, Label* if_equal,
174 Label* if_not_equal, Label* if_indirect) {
175 CSA_ASSERT(this, WordEqual(LoadStringLengthAsWord(lhs), length));
176 CSA_ASSERT(this, WordEqual(LoadStringLengthAsWord(rhs), length));
177 // Fast check to see if {lhs} and {rhs} refer to the same String object.
178 GotoIf(TaggedEqual(lhs, rhs), if_equal);
179
180 // Combine the instance types into a single 16-bit value, so we can check
181 // both of them at once.
182 TNode<Word32T> both_instance_types = Word32Or(
183 lhs_instance_type, Word32Shl(rhs_instance_type, Int32Constant(8)));
184
185 // Check if both {lhs} and {rhs} are internalized. Since we already know
186 // that they're not the same object, they're not equal in that case.
187 int const kBothInternalizedMask =
188 kIsNotInternalizedMask | (kIsNotInternalizedMask << 8);
189 int const kBothInternalizedTag = kInternalizedTag | (kInternalizedTag << 8);
190 GotoIf(Word32Equal(Word32And(both_instance_types,
191 Int32Constant(kBothInternalizedMask)),
192 Int32Constant(kBothInternalizedTag)),
193 if_not_equal);
194
195 // Check if both {lhs} and {rhs} are direct strings, and that in case of
196 // ExternalStrings the data pointer is cached.
197 STATIC_ASSERT(kUncachedExternalStringTag != 0);
198 STATIC_ASSERT(kIsIndirectStringTag != 0);
199 int const kBothDirectStringMask =
200 kIsIndirectStringMask | kUncachedExternalStringMask |
201 ((kIsIndirectStringMask | kUncachedExternalStringMask) << 8);
202 GotoIfNot(Word32Equal(Word32And(both_instance_types,
203 Int32Constant(kBothDirectStringMask)),
204 Int32Constant(0)),
205 if_indirect);
206
207 // Dispatch based on the {lhs} and {rhs} string encoding.
208 int const kBothStringEncodingMask =
209 kStringEncodingMask | (kStringEncodingMask << 8);
210 int const kOneOneByteStringTag = kOneByteStringTag | (kOneByteStringTag << 8);
211 int const kTwoTwoByteStringTag = kTwoByteStringTag | (kTwoByteStringTag << 8);
212 int const kOneTwoByteStringTag = kOneByteStringTag | (kTwoByteStringTag << 8);
213 Label if_oneonebytestring(this), if_twotwobytestring(this),
214 if_onetwobytestring(this), if_twoonebytestring(this);
215 TNode<Word32T> masked_instance_types =
216 Word32And(both_instance_types, Int32Constant(kBothStringEncodingMask));
217 GotoIf(
218 Word32Equal(masked_instance_types, Int32Constant(kOneOneByteStringTag)),
219 &if_oneonebytestring);
220 GotoIf(
221 Word32Equal(masked_instance_types, Int32Constant(kTwoTwoByteStringTag)),
222 &if_twotwobytestring);
223 Branch(
224 Word32Equal(masked_instance_types, Int32Constant(kOneTwoByteStringTag)),
225 &if_onetwobytestring, &if_twoonebytestring);
226
227 BIND(&if_oneonebytestring);
228 StringEqual_Loop(lhs, lhs_instance_type, MachineType::Uint8(), rhs,
229 rhs_instance_type, MachineType::Uint8(), length, if_equal,
230 if_not_equal);
231
232 BIND(&if_twotwobytestring);
233 StringEqual_Loop(lhs, lhs_instance_type, MachineType::Uint16(), rhs,
234 rhs_instance_type, MachineType::Uint16(), length, if_equal,
235 if_not_equal);
236
237 BIND(&if_onetwobytestring);
238 StringEqual_Loop(lhs, lhs_instance_type, MachineType::Uint8(), rhs,
239 rhs_instance_type, MachineType::Uint16(), length, if_equal,
240 if_not_equal);
241
242 BIND(&if_twoonebytestring);
243 StringEqual_Loop(lhs, lhs_instance_type, MachineType::Uint16(), rhs,
244 rhs_instance_type, MachineType::Uint8(), length, if_equal,
245 if_not_equal);
246 }
247
StringEqual_Loop(TNode<String> lhs,TNode<Word32T> lhs_instance_type,MachineType lhs_type,TNode<String> rhs,TNode<Word32T> rhs_instance_type,MachineType rhs_type,TNode<IntPtrT> length,Label * if_equal,Label * if_not_equal)248 void StringBuiltinsAssembler::StringEqual_Loop(
249 TNode<String> lhs, TNode<Word32T> lhs_instance_type, MachineType lhs_type,
250 TNode<String> rhs, TNode<Word32T> rhs_instance_type, MachineType rhs_type,
251 TNode<IntPtrT> length, Label* if_equal, Label* if_not_equal) {
252 CSA_ASSERT(this, WordEqual(LoadStringLengthAsWord(lhs), length));
253 CSA_ASSERT(this, WordEqual(LoadStringLengthAsWord(rhs), length));
254
255 // Compute the effective offset of the first character.
256 TNode<RawPtrT> lhs_data = DirectStringData(lhs, lhs_instance_type);
257 TNode<RawPtrT> rhs_data = DirectStringData(rhs, rhs_instance_type);
258
259 // Loop over the {lhs} and {rhs} strings to see if they are equal.
260 TVARIABLE(IntPtrT, var_offset, IntPtrConstant(0));
261 Label loop(this, &var_offset);
262 Goto(&loop);
263 BIND(&loop);
264 {
265 // If {offset} equals {end}, no difference was found, so the
266 // strings are equal.
267 GotoIf(WordEqual(var_offset.value(), length), if_equal);
268
269 // Load the next characters from {lhs} and {rhs}.
270 TNode<Word32T> lhs_value = UncheckedCast<Word32T>(
271 Load(lhs_type, lhs_data,
272 WordShl(var_offset.value(),
273 ElementSizeLog2Of(lhs_type.representation()))));
274 TNode<Word32T> rhs_value = UncheckedCast<Word32T>(
275 Load(rhs_type, rhs_data,
276 WordShl(var_offset.value(),
277 ElementSizeLog2Of(rhs_type.representation()))));
278
279 // Check if the characters match.
280 GotoIf(Word32NotEqual(lhs_value, rhs_value), if_not_equal);
281
282 // Advance to next character.
283 var_offset = IntPtrAdd(var_offset.value(), IntPtrConstant(1));
284 Goto(&loop);
285 }
286 }
287
StringFromSingleUTF16EncodedCodePoint(TNode<Int32T> codepoint)288 TNode<String> StringBuiltinsAssembler::StringFromSingleUTF16EncodedCodePoint(
289 TNode<Int32T> codepoint) {
290 TVARIABLE(String, var_result, EmptyStringConstant());
291
292 Label if_isword16(this), if_isword32(this), return_result(this);
293
294 Branch(Uint32LessThan(codepoint, Int32Constant(0x10000)), &if_isword16,
295 &if_isword32);
296
297 BIND(&if_isword16);
298 {
299 var_result = StringFromSingleCharCode(codepoint);
300 Goto(&return_result);
301 }
302
303 BIND(&if_isword32);
304 {
305 TNode<String> value = AllocateSeqTwoByteString(2);
306 StoreNoWriteBarrier(
307 MachineRepresentation::kWord32, value,
308 IntPtrConstant(SeqTwoByteString::kHeaderSize - kHeapObjectTag),
309 codepoint);
310 var_result = value;
311 Goto(&return_result);
312 }
313
314 BIND(&return_result);
315 return var_result.value();
316 }
317
AllocateConsString(TNode<Uint32T> length,TNode<String> left,TNode<String> right)318 TNode<String> StringBuiltinsAssembler::AllocateConsString(TNode<Uint32T> length,
319 TNode<String> left,
320 TNode<String> right) {
321 // Added string can be a cons string.
322 Comment("Allocating ConsString");
323 TNode<Int32T> left_instance_type = LoadInstanceType(left);
324 TNode<Int32T> right_instance_type = LoadInstanceType(right);
325
326 // Determine the resulting ConsString map to use depending on whether
327 // any of {left} or {right} has two byte encoding.
328 STATIC_ASSERT(kOneByteStringTag != 0);
329 STATIC_ASSERT(kTwoByteStringTag == 0);
330 TNode<Int32T> combined_instance_type =
331 Word32And(left_instance_type, right_instance_type);
332 TNode<Map> result_map = CAST(Select<Object>(
333 IsSetWord32(combined_instance_type, kStringEncodingMask),
334 [=] { return ConsOneByteStringMapConstant(); },
335 [=] { return ConsStringMapConstant(); }));
336 TNode<HeapObject> result = AllocateInNewSpace(ConsString::kSize);
337 StoreMapNoWriteBarrier(result, result_map);
338 StoreObjectFieldNoWriteBarrier(result, ConsString::kLengthOffset, length);
339 StoreObjectFieldNoWriteBarrier(result, ConsString::kHashFieldOffset,
340 Int32Constant(String::kEmptyHashField));
341 StoreObjectFieldNoWriteBarrier(result, ConsString::kFirstOffset, left);
342 StoreObjectFieldNoWriteBarrier(result, ConsString::kSecondOffset, right);
343 return CAST(result);
344 }
345
StringAdd(TNode<ContextOrEmptyContext> context,TNode<String> left,TNode<String> right)346 TNode<String> StringBuiltinsAssembler::StringAdd(
347 TNode<ContextOrEmptyContext> context, TNode<String> left,
348 TNode<String> right) {
349 CSA_ASSERT(this, IsZeroOrContext(context));
350
351 TVARIABLE(String, result);
352 Label check_right(this), runtime(this, Label::kDeferred), cons(this),
353 done(this, &result), done_native(this, &result);
354 Counters* counters = isolate()->counters();
355
356 TNode<Uint32T> left_length = LoadStringLengthAsWord32(left);
357 GotoIfNot(Word32Equal(left_length, Uint32Constant(0)), &check_right);
358 result = right;
359 Goto(&done_native);
360
361 BIND(&check_right);
362 TNode<Uint32T> right_length = LoadStringLengthAsWord32(right);
363 GotoIfNot(Word32Equal(right_length, Uint32Constant(0)), &cons);
364 result = left;
365 Goto(&done_native);
366
367 BIND(&cons);
368 {
369 TNode<Uint32T> new_length = Uint32Add(left_length, right_length);
370
371 // If new length is greater than String::kMaxLength, goto runtime to
372 // throw. Note: we also need to invalidate the string length protector, so
373 // can't just throw here directly.
374 GotoIf(Uint32GreaterThan(new_length, Uint32Constant(String::kMaxLength)),
375 &runtime);
376
377 TVARIABLE(String, var_left, left);
378 TVARIABLE(String, var_right, right);
379 Label non_cons(this, {&var_left, &var_right});
380 Label slow(this, Label::kDeferred);
381 GotoIf(Uint32LessThan(new_length, Uint32Constant(ConsString::kMinLength)),
382 &non_cons);
383
384 result =
385 AllocateConsString(new_length, var_left.value(), var_right.value());
386 Goto(&done_native);
387
388 BIND(&non_cons);
389
390 Comment("Full string concatenate");
391 TNode<Int32T> left_instance_type = LoadInstanceType(var_left.value());
392 TNode<Int32T> right_instance_type = LoadInstanceType(var_right.value());
393 // Compute intersection and difference of instance types.
394
395 TNode<Int32T> ored_instance_types =
396 Word32Or(left_instance_type, right_instance_type);
397 TNode<Word32T> xored_instance_types =
398 Word32Xor(left_instance_type, right_instance_type);
399
400 // Check if both strings have the same encoding and both are sequential.
401 GotoIf(IsSetWord32(xored_instance_types, kStringEncodingMask), &runtime);
402 GotoIf(IsSetWord32(ored_instance_types, kStringRepresentationMask), &slow);
403
404 TNode<IntPtrT> word_left_length = Signed(ChangeUint32ToWord(left_length));
405 TNode<IntPtrT> word_right_length = Signed(ChangeUint32ToWord(right_length));
406
407 Label two_byte(this);
408 GotoIf(Word32Equal(Word32And(ored_instance_types,
409 Int32Constant(kStringEncodingMask)),
410 Int32Constant(kTwoByteStringTag)),
411 &two_byte);
412 // One-byte sequential string case
413 result = AllocateSeqOneByteString(new_length);
414 CopyStringCharacters(var_left.value(), result.value(), IntPtrConstant(0),
415 IntPtrConstant(0), word_left_length,
416 String::ONE_BYTE_ENCODING, String::ONE_BYTE_ENCODING);
417 CopyStringCharacters(var_right.value(), result.value(), IntPtrConstant(0),
418 word_left_length, word_right_length,
419 String::ONE_BYTE_ENCODING, String::ONE_BYTE_ENCODING);
420 Goto(&done_native);
421
422 BIND(&two_byte);
423 {
424 // Two-byte sequential string case
425 result = AllocateSeqTwoByteString(new_length);
426 CopyStringCharacters(var_left.value(), result.value(), IntPtrConstant(0),
427 IntPtrConstant(0), word_left_length,
428 String::TWO_BYTE_ENCODING,
429 String::TWO_BYTE_ENCODING);
430 CopyStringCharacters(var_right.value(), result.value(), IntPtrConstant(0),
431 word_left_length, word_right_length,
432 String::TWO_BYTE_ENCODING,
433 String::TWO_BYTE_ENCODING);
434 Goto(&done_native);
435 }
436
437 BIND(&slow);
438 {
439 // Try to unwrap indirect strings, restart the above attempt on success.
440 MaybeDerefIndirectStrings(&var_left, left_instance_type, &var_right,
441 right_instance_type, &non_cons);
442 Goto(&runtime);
443 }
444 }
445 BIND(&runtime);
446 {
447 result = CAST(CallRuntime(Runtime::kStringAdd, context, left, right));
448 Goto(&done);
449 }
450
451 BIND(&done_native);
452 {
453 IncrementCounter(counters->string_add_native(), 1);
454 Goto(&done);
455 }
456
457 BIND(&done);
458 return result.value();
459 }
460
BranchIfCanDerefIndirectString(TNode<String> string,TNode<Int32T> instance_type,Label * can_deref,Label * cannot_deref)461 void StringBuiltinsAssembler::BranchIfCanDerefIndirectString(
462 TNode<String> string, TNode<Int32T> instance_type, Label* can_deref,
463 Label* cannot_deref) {
464 TNode<Int32T> representation =
465 Word32And(instance_type, Int32Constant(kStringRepresentationMask));
466 GotoIf(Word32Equal(representation, Int32Constant(kThinStringTag)), can_deref);
467 GotoIf(Word32NotEqual(representation, Int32Constant(kConsStringTag)),
468 cannot_deref);
469 // Cons string.
470 TNode<String> rhs =
471 LoadObjectField<String>(string, ConsString::kSecondOffset);
472 GotoIf(IsEmptyString(rhs), can_deref);
473 Goto(cannot_deref);
474 }
475
DerefIndirectString(TVariable<String> * var_string,TNode<Int32T> instance_type)476 void StringBuiltinsAssembler::DerefIndirectString(TVariable<String>* var_string,
477 TNode<Int32T> instance_type) {
478 #ifdef DEBUG
479 Label can_deref(this), cannot_deref(this);
480 BranchIfCanDerefIndirectString(var_string->value(), instance_type, &can_deref,
481 &cannot_deref);
482 BIND(&cannot_deref);
483 DebugBreak(); // Should be able to dereference string.
484 Goto(&can_deref);
485 BIND(&can_deref);
486 #endif // DEBUG
487
488 STATIC_ASSERT(static_cast<int>(ThinString::kActualOffset) ==
489 static_cast<int>(ConsString::kFirstOffset));
490 *var_string =
491 LoadObjectField<String>(var_string->value(), ThinString::kActualOffset);
492 }
493
MaybeDerefIndirectString(TVariable<String> * var_string,TNode<Int32T> instance_type,Label * did_deref,Label * cannot_deref)494 void StringBuiltinsAssembler::MaybeDerefIndirectString(
495 TVariable<String>* var_string, TNode<Int32T> instance_type,
496 Label* did_deref, Label* cannot_deref) {
497 Label deref(this);
498 BranchIfCanDerefIndirectString(var_string->value(), instance_type, &deref,
499 cannot_deref);
500
501 BIND(&deref);
502 {
503 DerefIndirectString(var_string, instance_type);
504 Goto(did_deref);
505 }
506 }
507
MaybeDerefIndirectStrings(TVariable<String> * var_left,TNode<Int32T> left_instance_type,TVariable<String> * var_right,TNode<Int32T> right_instance_type,Label * did_something)508 void StringBuiltinsAssembler::MaybeDerefIndirectStrings(
509 TVariable<String>* var_left, TNode<Int32T> left_instance_type,
510 TVariable<String>* var_right, TNode<Int32T> right_instance_type,
511 Label* did_something) {
512 Label did_nothing_left(this), did_something_left(this),
513 didnt_do_anything(this);
514 MaybeDerefIndirectString(var_left, left_instance_type, &did_something_left,
515 &did_nothing_left);
516
517 BIND(&did_something_left);
518 {
519 MaybeDerefIndirectString(var_right, right_instance_type, did_something,
520 did_something);
521 }
522
523 BIND(&did_nothing_left);
524 {
525 MaybeDerefIndirectString(var_right, right_instance_type, did_something,
526 &didnt_do_anything);
527 }
528
529 BIND(&didnt_do_anything);
530 // Fall through if neither string was an indirect string.
531 }
532
DerefIndirectString(TNode<String> string,TNode<Int32T> instance_type,Label * cannot_deref)533 TNode<String> StringBuiltinsAssembler::DerefIndirectString(
534 TNode<String> string, TNode<Int32T> instance_type, Label* cannot_deref) {
535 Label deref(this);
536 BranchIfCanDerefIndirectString(string, instance_type, &deref, cannot_deref);
537 BIND(&deref);
538 STATIC_ASSERT(static_cast<int>(ThinString::kActualOffset) ==
539 static_cast<int>(ConsString::kFirstOffset));
540 return LoadObjectField<String>(string, ThinString::kActualOffset);
541 }
542
TF_BUILTIN(StringAdd_CheckNone,StringBuiltinsAssembler)543 TF_BUILTIN(StringAdd_CheckNone, StringBuiltinsAssembler) {
544 auto left = Parameter<String>(Descriptor::kLeft);
545 auto right = Parameter<String>(Descriptor::kRight);
546 TNode<ContextOrEmptyContext> context =
547 UncheckedParameter<ContextOrEmptyContext>(Descriptor::kContext);
548 CSA_ASSERT(this, IsZeroOrContext(context));
549 Return(StringAdd(context, left, right));
550 }
551
TF_BUILTIN(SubString,StringBuiltinsAssembler)552 TF_BUILTIN(SubString, StringBuiltinsAssembler) {
553 auto string = Parameter<String>(Descriptor::kString);
554 auto from = Parameter<Smi>(Descriptor::kFrom);
555 auto to = Parameter<Smi>(Descriptor::kTo);
556 Return(SubString(string, SmiUntag(from), SmiUntag(to)));
557 }
558
GenerateStringRelationalComparison(TNode<String> left,TNode<String> right,Operation op)559 void StringBuiltinsAssembler::GenerateStringRelationalComparison(
560 TNode<String> left, TNode<String> right, Operation op) {
561 TVARIABLE(String, var_left, left);
562 TVARIABLE(String, var_right, right);
563
564 Label if_less(this), if_equal(this), if_greater(this);
565 Label restart(this, {&var_left, &var_right});
566 Goto(&restart);
567 BIND(&restart);
568
569 TNode<String> lhs = var_left.value();
570 TNode<String> rhs = var_right.value();
571 // Fast check to see if {lhs} and {rhs} refer to the same String object.
572 GotoIf(TaggedEqual(lhs, rhs), &if_equal);
573
574 // Load instance types of {lhs} and {rhs}.
575 TNode<Uint16T> lhs_instance_type = LoadInstanceType(lhs);
576 TNode<Uint16T> rhs_instance_type = LoadInstanceType(rhs);
577
578 // Combine the instance types into a single 16-bit value, so we can check
579 // both of them at once.
580 TNode<Int32T> both_instance_types = Word32Or(
581 lhs_instance_type, Word32Shl(rhs_instance_type, Int32Constant(8)));
582
583 // Check that both {lhs} and {rhs} are flat one-byte strings.
584 int const kBothSeqOneByteStringMask =
585 kStringEncodingMask | kStringRepresentationMask |
586 ((kStringEncodingMask | kStringRepresentationMask) << 8);
587 int const kBothSeqOneByteStringTag =
588 kOneByteStringTag | kSeqStringTag |
589 ((kOneByteStringTag | kSeqStringTag) << 8);
590 Label if_bothonebyteseqstrings(this), if_notbothonebyteseqstrings(this);
591 Branch(Word32Equal(Word32And(both_instance_types,
592 Int32Constant(kBothSeqOneByteStringMask)),
593 Int32Constant(kBothSeqOneByteStringTag)),
594 &if_bothonebyteseqstrings, &if_notbothonebyteseqstrings);
595
596 BIND(&if_bothonebyteseqstrings);
597 {
598 // Load the length of {lhs} and {rhs}.
599 TNode<IntPtrT> lhs_length = LoadStringLengthAsWord(lhs);
600 TNode<IntPtrT> rhs_length = LoadStringLengthAsWord(rhs);
601
602 // Determine the minimum length.
603 TNode<IntPtrT> length = IntPtrMin(lhs_length, rhs_length);
604
605 // Compute the effective offset of the first character.
606 TNode<IntPtrT> begin =
607 IntPtrConstant(SeqOneByteString::kHeaderSize - kHeapObjectTag);
608
609 // Compute the first offset after the string from the length.
610 TNode<IntPtrT> end = IntPtrAdd(begin, length);
611
612 // Loop over the {lhs} and {rhs} strings to see if they are equal.
613 TVARIABLE(IntPtrT, var_offset, begin);
614 Label loop(this, &var_offset);
615 Goto(&loop);
616 BIND(&loop);
617 {
618 // Check if {offset} equals {end}.
619 Label if_done(this), if_notdone(this);
620 Branch(WordEqual(var_offset.value(), end), &if_done, &if_notdone);
621
622 BIND(&if_notdone);
623 {
624 // Load the next characters from {lhs} and {rhs}.
625 TNode<Uint8T> lhs_value = Load<Uint8T>(lhs, var_offset.value());
626 TNode<Uint8T> rhs_value = Load<Uint8T>(rhs, var_offset.value());
627
628 // Check if the characters match.
629 Label if_valueissame(this), if_valueisnotsame(this);
630 Branch(Word32Equal(lhs_value, rhs_value), &if_valueissame,
631 &if_valueisnotsame);
632
633 BIND(&if_valueissame);
634 {
635 // Advance to next character.
636 var_offset = IntPtrAdd(var_offset.value(), IntPtrConstant(1));
637 }
638 Goto(&loop);
639
640 BIND(&if_valueisnotsame);
641 Branch(Uint32LessThan(lhs_value, rhs_value), &if_less, &if_greater);
642 }
643
644 BIND(&if_done);
645 {
646 // All characters up to the min length are equal, decide based on
647 // string length.
648 GotoIf(IntPtrEqual(lhs_length, rhs_length), &if_equal);
649 Branch(IntPtrLessThan(lhs_length, rhs_length), &if_less, &if_greater);
650 }
651 }
652 }
653
654 BIND(&if_notbothonebyteseqstrings);
655 {
656 // Try to unwrap indirect strings, restart the above attempt on success.
657 MaybeDerefIndirectStrings(&var_left, lhs_instance_type, &var_right,
658 rhs_instance_type, &restart);
659 // TODO(bmeurer): Add support for two byte string relational comparisons.
660 switch (op) {
661 case Operation::kLessThan:
662 TailCallRuntime(Runtime::kStringLessThan, NoContextConstant(), lhs,
663 rhs);
664 break;
665 case Operation::kLessThanOrEqual:
666 TailCallRuntime(Runtime::kStringLessThanOrEqual, NoContextConstant(),
667 lhs, rhs);
668 break;
669 case Operation::kGreaterThan:
670 TailCallRuntime(Runtime::kStringGreaterThan, NoContextConstant(), lhs,
671 rhs);
672 break;
673 case Operation::kGreaterThanOrEqual:
674 TailCallRuntime(Runtime::kStringGreaterThanOrEqual, NoContextConstant(),
675 lhs, rhs);
676 break;
677 default:
678 UNREACHABLE();
679 }
680 }
681
682 BIND(&if_less);
683 switch (op) {
684 case Operation::kLessThan:
685 case Operation::kLessThanOrEqual:
686 Return(TrueConstant());
687 break;
688
689 case Operation::kGreaterThan:
690 case Operation::kGreaterThanOrEqual:
691 Return(FalseConstant());
692 break;
693 default:
694 UNREACHABLE();
695 }
696
697 BIND(&if_equal);
698 switch (op) {
699 case Operation::kLessThan:
700 case Operation::kGreaterThan:
701 Return(FalseConstant());
702 break;
703
704 case Operation::kLessThanOrEqual:
705 case Operation::kGreaterThanOrEqual:
706 Return(TrueConstant());
707 break;
708 default:
709 UNREACHABLE();
710 }
711
712 BIND(&if_greater);
713 switch (op) {
714 case Operation::kLessThan:
715 case Operation::kLessThanOrEqual:
716 Return(FalseConstant());
717 break;
718
719 case Operation::kGreaterThan:
720 case Operation::kGreaterThanOrEqual:
721 Return(TrueConstant());
722 break;
723 default:
724 UNREACHABLE();
725 }
726 }
727
TF_BUILTIN(StringEqual,StringBuiltinsAssembler)728 TF_BUILTIN(StringEqual, StringBuiltinsAssembler) {
729 auto left = Parameter<String>(Descriptor::kLeft);
730 auto right = Parameter<String>(Descriptor::kRight);
731 GenerateStringEqual(left, right);
732 }
733
TF_BUILTIN(StringLessThan,StringBuiltinsAssembler)734 TF_BUILTIN(StringLessThan, StringBuiltinsAssembler) {
735 auto left = Parameter<String>(Descriptor::kLeft);
736 auto right = Parameter<String>(Descriptor::kRight);
737 GenerateStringRelationalComparison(left, right, Operation::kLessThan);
738 }
739
TF_BUILTIN(StringLessThanOrEqual,StringBuiltinsAssembler)740 TF_BUILTIN(StringLessThanOrEqual, StringBuiltinsAssembler) {
741 auto left = Parameter<String>(Descriptor::kLeft);
742 auto right = Parameter<String>(Descriptor::kRight);
743 GenerateStringRelationalComparison(left, right, Operation::kLessThanOrEqual);
744 }
745
TF_BUILTIN(StringGreaterThan,StringBuiltinsAssembler)746 TF_BUILTIN(StringGreaterThan, StringBuiltinsAssembler) {
747 auto left = Parameter<String>(Descriptor::kLeft);
748 auto right = Parameter<String>(Descriptor::kRight);
749 GenerateStringRelationalComparison(left, right, Operation::kGreaterThan);
750 }
751
TF_BUILTIN(StringGreaterThanOrEqual,StringBuiltinsAssembler)752 TF_BUILTIN(StringGreaterThanOrEqual, StringBuiltinsAssembler) {
753 auto left = Parameter<String>(Descriptor::kLeft);
754 auto right = Parameter<String>(Descriptor::kRight);
755 GenerateStringRelationalComparison(left, right,
756 Operation::kGreaterThanOrEqual);
757 }
758
TF_BUILTIN(StringCodePointAt,StringBuiltinsAssembler)759 TF_BUILTIN(StringCodePointAt, StringBuiltinsAssembler) {
760 auto receiver = Parameter<String>(Descriptor::kReceiver);
761 auto position = UncheckedParameter<IntPtrT>(Descriptor::kPosition);
762
763 // TODO(sigurds) Figure out if passing length as argument pays off.
764 TNode<IntPtrT> length = LoadStringLengthAsWord(receiver);
765 // Load the character code at the {position} from the {receiver}.
766 TNode<Int32T> code =
767 LoadSurrogatePairAt(receiver, length, position, UnicodeEncoding::UTF32);
768 // And return it as TaggedSigned value.
769 // TODO(turbofan): Allow builtins to return values untagged.
770 TNode<Smi> result = SmiFromInt32(code);
771 Return(result);
772 }
773
TF_BUILTIN(StringFromCodePointAt,StringBuiltinsAssembler)774 TF_BUILTIN(StringFromCodePointAt, StringBuiltinsAssembler) {
775 auto receiver = Parameter<String>(Descriptor::kReceiver);
776 auto position = UncheckedParameter<IntPtrT>(Descriptor::kPosition);
777
778 // TODO(sigurds) Figure out if passing length as argument pays off.
779 TNode<IntPtrT> length = LoadStringLengthAsWord(receiver);
780 // Load the character code at the {position} from the {receiver}.
781 TNode<Int32T> code =
782 LoadSurrogatePairAt(receiver, length, position, UnicodeEncoding::UTF16);
783 // Create a String from the UTF16 encoded code point
784 TNode<String> result = StringFromSingleUTF16EncodedCodePoint(code);
785 Return(result);
786 }
787
788 // -----------------------------------------------------------------------------
789 // ES6 section 21.1 String Objects
790
791 // ES6 #sec-string.fromcharcode
TF_BUILTIN(StringFromCharCode,StringBuiltinsAssembler)792 TF_BUILTIN(StringFromCharCode, StringBuiltinsAssembler) {
793 // TODO(ishell): use constants from Descriptor once the JSFunction linkage
794 // arguments are reordered.
795 auto argc = UncheckedParameter<Int32T>(Descriptor::kJSActualArgumentsCount);
796 auto context = Parameter<Context>(Descriptor::kContext);
797
798 CodeStubArguments arguments(this, argc);
799 // Check if we have exactly one argument (plus the implicit receiver), i.e.
800 // if the parent frame is not an arguments adaptor frame.
801 Label if_oneargument(this), if_notoneargument(this);
802 Branch(Word32Equal(argc, Int32Constant(1)), &if_oneargument,
803 &if_notoneargument);
804
805 BIND(&if_oneargument);
806 {
807 // Single argument case, perform fast single character string cache lookup
808 // for one-byte code units, or fall back to creating a single character
809 // string on the fly otherwise.
810 TNode<Object> code = arguments.AtIndex(0);
811 TNode<Word32T> code32 = TruncateTaggedToWord32(context, code);
812 TNode<Int32T> code16 =
813 Signed(Word32And(code32, Int32Constant(String::kMaxUtf16CodeUnit)));
814 TNode<String> result = StringFromSingleCharCode(code16);
815 arguments.PopAndReturn(result);
816 }
817
818 TNode<Word32T> code16;
819 BIND(&if_notoneargument);
820 {
821 Label two_byte(this);
822 // Assume that the resulting string contains only one-byte characters.
823 TNode<String> one_byte_result = AllocateSeqOneByteString(Unsigned(argc));
824
825 TVARIABLE(IntPtrT, var_max_index, IntPtrConstant(0));
826
827 // Iterate over the incoming arguments, converting them to 8-bit character
828 // codes. Stop if any of the conversions generates a code that doesn't fit
829 // in 8 bits.
830 CodeStubAssembler::VariableList vars({&var_max_index}, zone());
831 arguments.ForEach(vars, [&](TNode<Object> arg) {
832 TNode<Word32T> code32 = TruncateTaggedToWord32(context, arg);
833 code16 = Word32And(code32, Int32Constant(String::kMaxUtf16CodeUnit));
834
835 GotoIf(
836 Int32GreaterThan(code16, Int32Constant(String::kMaxOneByteCharCode)),
837 &two_byte);
838
839 // The {code16} fits into the SeqOneByteString {one_byte_result}.
840 TNode<IntPtrT> offset = ElementOffsetFromIndex(
841 var_max_index.value(), UINT8_ELEMENTS,
842 SeqOneByteString::kHeaderSize - kHeapObjectTag);
843 StoreNoWriteBarrier(MachineRepresentation::kWord8, one_byte_result,
844 offset, code16);
845 var_max_index = IntPtrAdd(var_max_index.value(), IntPtrConstant(1));
846 });
847 arguments.PopAndReturn(one_byte_result);
848
849 BIND(&two_byte);
850
851 // At least one of the characters in the string requires a 16-bit
852 // representation. Allocate a SeqTwoByteString to hold the resulting
853 // string.
854 TNode<String> two_byte_result = AllocateSeqTwoByteString(Unsigned(argc));
855
856 // Copy the characters that have already been put in the 8-bit string into
857 // their corresponding positions in the new 16-bit string.
858 TNode<IntPtrT> zero = IntPtrConstant(0);
859 CopyStringCharacters(one_byte_result, two_byte_result, zero, zero,
860 var_max_index.value(), String::ONE_BYTE_ENCODING,
861 String::TWO_BYTE_ENCODING);
862
863 // Write the character that caused the 8-bit to 16-bit fault.
864 TNode<IntPtrT> max_index_offset =
865 ElementOffsetFromIndex(var_max_index.value(), UINT16_ELEMENTS,
866 SeqTwoByteString::kHeaderSize - kHeapObjectTag);
867 StoreNoWriteBarrier(MachineRepresentation::kWord16, two_byte_result,
868 max_index_offset, code16);
869 var_max_index = IntPtrAdd(var_max_index.value(), IntPtrConstant(1));
870
871 // Resume copying the passed-in arguments from the same place where the
872 // 8-bit copy stopped, but this time copying over all of the characters
873 // using a 16-bit representation.
874 arguments.ForEach(
875 vars,
876 [&](TNode<Object> arg) {
877 TNode<Word32T> code32 = TruncateTaggedToWord32(context, arg);
878 TNode<Word32T> code16 =
879 Word32And(code32, Int32Constant(String::kMaxUtf16CodeUnit));
880
881 TNode<IntPtrT> offset = ElementOffsetFromIndex(
882 var_max_index.value(), UINT16_ELEMENTS,
883 SeqTwoByteString::kHeaderSize - kHeapObjectTag);
884 StoreNoWriteBarrier(MachineRepresentation::kWord16, two_byte_result,
885 offset, code16);
886 var_max_index = IntPtrAdd(var_max_index.value(), IntPtrConstant(1));
887 },
888 var_max_index.value());
889
890 arguments.PopAndReturn(two_byte_result);
891 }
892 }
893
StringIndexOf(const TNode<String> subject_string,const TNode<String> search_string,const TNode<Smi> position,const std::function<void (TNode<Smi>)> & f_return)894 void StringBuiltinsAssembler::StringIndexOf(
895 const TNode<String> subject_string, const TNode<String> search_string,
896 const TNode<Smi> position,
897 const std::function<void(TNode<Smi>)>& f_return) {
898 const TNode<IntPtrT> int_zero = IntPtrConstant(0);
899 const TNode<IntPtrT> search_length = LoadStringLengthAsWord(search_string);
900 const TNode<IntPtrT> subject_length = LoadStringLengthAsWord(subject_string);
901 const TNode<IntPtrT> start_position = IntPtrMax(SmiUntag(position), int_zero);
902
903 Label zero_length_needle(this), return_minus_1(this);
904 {
905 GotoIf(IntPtrEqual(int_zero, search_length), &zero_length_needle);
906
907 // Check that the needle fits in the start position.
908 GotoIfNot(IntPtrLessThanOrEqual(search_length,
909 IntPtrSub(subject_length, start_position)),
910 &return_minus_1);
911 }
912
913 // If the string pointers are identical, we can just return 0. Note that this
914 // implies {start_position} == 0 since we've passed the check above.
915 Label return_zero(this);
916 GotoIf(TaggedEqual(subject_string, search_string), &return_zero);
917
918 // Try to unpack subject and search strings. Bail to runtime if either needs
919 // to be flattened.
920 ToDirectStringAssembler subject_to_direct(state(), subject_string);
921 ToDirectStringAssembler search_to_direct(state(), search_string);
922
923 Label call_runtime_unchecked(this, Label::kDeferred);
924
925 subject_to_direct.TryToDirect(&call_runtime_unchecked);
926 search_to_direct.TryToDirect(&call_runtime_unchecked);
927
928 // Load pointers to string data.
929 const TNode<RawPtrT> subject_ptr =
930 subject_to_direct.PointerToData(&call_runtime_unchecked);
931 const TNode<RawPtrT> search_ptr =
932 search_to_direct.PointerToData(&call_runtime_unchecked);
933
934 const TNode<IntPtrT> subject_offset = subject_to_direct.offset();
935 const TNode<IntPtrT> search_offset = search_to_direct.offset();
936
937 // Like String::IndexOf, the actual matching is done by the optimized
938 // SearchString method in string-search.h. Dispatch based on string instance
939 // types, then call straight into C++ for matching.
940
941 CSA_ASSERT(this, IntPtrGreaterThan(search_length, int_zero));
942 CSA_ASSERT(this, IntPtrGreaterThanOrEqual(start_position, int_zero));
943 CSA_ASSERT(this, IntPtrGreaterThanOrEqual(subject_length, start_position));
944 CSA_ASSERT(this,
945 IntPtrLessThanOrEqual(search_length,
946 IntPtrSub(subject_length, start_position)));
947
948 Label one_one(this), one_two(this), two_one(this), two_two(this);
949 DispatchOnStringEncodings(subject_to_direct.instance_type(),
950 search_to_direct.instance_type(), &one_one,
951 &one_two, &two_one, &two_two);
952
953 using onebyte_t = const uint8_t;
954 using twobyte_t = const uc16;
955
956 BIND(&one_one);
957 {
958 const TNode<RawPtrT> adjusted_subject_ptr = PointerToStringDataAtIndex(
959 subject_ptr, subject_offset, String::ONE_BYTE_ENCODING);
960 const TNode<RawPtrT> adjusted_search_ptr = PointerToStringDataAtIndex(
961 search_ptr, search_offset, String::ONE_BYTE_ENCODING);
962
963 Label direct_memchr_call(this), generic_fast_path(this);
964 Branch(IntPtrEqual(search_length, IntPtrConstant(1)), &direct_memchr_call,
965 &generic_fast_path);
966
967 // An additional fast path that calls directly into memchr for 1-length
968 // search strings.
969 BIND(&direct_memchr_call);
970 {
971 const TNode<RawPtrT> string_addr =
972 RawPtrAdd(adjusted_subject_ptr, start_position);
973 const TNode<IntPtrT> search_length =
974 IntPtrSub(subject_length, start_position);
975 const TNode<IntPtrT> search_byte =
976 ChangeInt32ToIntPtr(Load<Uint8T>(adjusted_search_ptr));
977
978 const TNode<ExternalReference> memchr =
979 ExternalConstant(ExternalReference::libc_memchr_function());
980 const TNode<RawPtrT> result_address = UncheckedCast<RawPtrT>(
981 CallCFunction(memchr, MachineType::Pointer(),
982 std::make_pair(MachineType::Pointer(), string_addr),
983 std::make_pair(MachineType::IntPtr(), search_byte),
984 std::make_pair(MachineType::UintPtr(), search_length)));
985 GotoIf(WordEqual(result_address, int_zero), &return_minus_1);
986 const TNode<IntPtrT> result_index =
987 IntPtrAdd(RawPtrSub(result_address, string_addr), start_position);
988 f_return(SmiTag(result_index));
989 }
990
991 BIND(&generic_fast_path);
992 {
993 const TNode<IntPtrT> result = CallSearchStringRaw<onebyte_t, onebyte_t>(
994 adjusted_subject_ptr, subject_length, adjusted_search_ptr,
995 search_length, start_position);
996 f_return(SmiTag(result));
997 }
998 }
999
1000 BIND(&one_two);
1001 {
1002 const TNode<RawPtrT> adjusted_subject_ptr = PointerToStringDataAtIndex(
1003 subject_ptr, subject_offset, String::ONE_BYTE_ENCODING);
1004 const TNode<RawPtrT> adjusted_search_ptr = PointerToStringDataAtIndex(
1005 search_ptr, search_offset, String::TWO_BYTE_ENCODING);
1006
1007 const TNode<IntPtrT> result = CallSearchStringRaw<onebyte_t, twobyte_t>(
1008 adjusted_subject_ptr, subject_length, adjusted_search_ptr,
1009 search_length, start_position);
1010 f_return(SmiTag(result));
1011 }
1012
1013 BIND(&two_one);
1014 {
1015 const TNode<RawPtrT> adjusted_subject_ptr = PointerToStringDataAtIndex(
1016 subject_ptr, subject_offset, String::TWO_BYTE_ENCODING);
1017 const TNode<RawPtrT> adjusted_search_ptr = PointerToStringDataAtIndex(
1018 search_ptr, search_offset, String::ONE_BYTE_ENCODING);
1019
1020 const TNode<IntPtrT> result = CallSearchStringRaw<twobyte_t, onebyte_t>(
1021 adjusted_subject_ptr, subject_length, adjusted_search_ptr,
1022 search_length, start_position);
1023 f_return(SmiTag(result));
1024 }
1025
1026 BIND(&two_two);
1027 {
1028 const TNode<RawPtrT> adjusted_subject_ptr = PointerToStringDataAtIndex(
1029 subject_ptr, subject_offset, String::TWO_BYTE_ENCODING);
1030 const TNode<RawPtrT> adjusted_search_ptr = PointerToStringDataAtIndex(
1031 search_ptr, search_offset, String::TWO_BYTE_ENCODING);
1032
1033 const TNode<IntPtrT> result = CallSearchStringRaw<twobyte_t, twobyte_t>(
1034 adjusted_subject_ptr, subject_length, adjusted_search_ptr,
1035 search_length, start_position);
1036 f_return(SmiTag(result));
1037 }
1038
1039 BIND(&return_minus_1);
1040 f_return(SmiConstant(-1));
1041
1042 BIND(&return_zero);
1043 f_return(SmiConstant(0));
1044
1045 BIND(&zero_length_needle);
1046 {
1047 Comment("0-length search_string");
1048 f_return(SmiTag(IntPtrMin(subject_length, start_position)));
1049 }
1050
1051 BIND(&call_runtime_unchecked);
1052 {
1053 // Simplified version of the runtime call where the types of the arguments
1054 // are already known due to type checks in this stub.
1055 Comment("Call Runtime Unchecked");
1056 TNode<Smi> result =
1057 CAST(CallRuntime(Runtime::kStringIndexOfUnchecked, NoContextConstant(),
1058 subject_string, search_string, position));
1059 f_return(result);
1060 }
1061 }
1062
1063 // ES6 String.prototype.indexOf(searchString [, position])
1064 // #sec-string.prototype.indexof
1065 // Unchecked helper for builtins lowering.
TF_BUILTIN(StringIndexOf,StringBuiltinsAssembler)1066 TF_BUILTIN(StringIndexOf, StringBuiltinsAssembler) {
1067 auto receiver = Parameter<String>(Descriptor::kReceiver);
1068 auto search_string = Parameter<String>(Descriptor::kSearchString);
1069 auto position = Parameter<Smi>(Descriptor::kPosition);
1070 StringIndexOf(receiver, search_string, position,
1071 [this](TNode<Smi> result) { this->Return(result); });
1072 }
1073
1074 // ES6 String.prototype.includes(searchString [, position])
1075 // #sec-string.prototype.includes
TF_BUILTIN(StringPrototypeIncludes,StringIncludesIndexOfAssembler)1076 TF_BUILTIN(StringPrototypeIncludes, StringIncludesIndexOfAssembler) {
1077 TNode<IntPtrT> argc = ChangeInt32ToIntPtr(
1078 UncheckedParameter<Int32T>(Descriptor::kJSActualArgumentsCount));
1079 auto context = Parameter<Context>(Descriptor::kContext);
1080 Generate(kIncludes, argc, context);
1081 }
1082
1083 // ES6 String.prototype.indexOf(searchString [, position])
1084 // #sec-string.prototype.indexof
TF_BUILTIN(StringPrototypeIndexOf,StringIncludesIndexOfAssembler)1085 TF_BUILTIN(StringPrototypeIndexOf, StringIncludesIndexOfAssembler) {
1086 TNode<IntPtrT> argc = ChangeInt32ToIntPtr(
1087 UncheckedParameter<Int32T>(Descriptor::kJSActualArgumentsCount));
1088 auto context = Parameter<Context>(Descriptor::kContext);
1089 Generate(kIndexOf, argc, context);
1090 }
1091
Generate(SearchVariant variant,TNode<IntPtrT> argc,TNode<Context> context)1092 void StringIncludesIndexOfAssembler::Generate(SearchVariant variant,
1093 TNode<IntPtrT> argc,
1094 TNode<Context> context) {
1095 CodeStubArguments arguments(this, argc);
1096 const TNode<Object> receiver = arguments.GetReceiver();
1097
1098 TVARIABLE(Object, var_search_string);
1099 TVARIABLE(Object, var_position);
1100 Label argc_1(this), argc_2(this), call_runtime(this, Label::kDeferred),
1101 fast_path(this);
1102
1103 GotoIf(IntPtrEqual(argc, IntPtrConstant(1)), &argc_1);
1104 GotoIf(IntPtrGreaterThan(argc, IntPtrConstant(1)), &argc_2);
1105 {
1106 Comment("0 Argument case");
1107 CSA_ASSERT(this, IntPtrEqual(argc, IntPtrConstant(0)));
1108 TNode<Oddball> undefined = UndefinedConstant();
1109 var_search_string = undefined;
1110 var_position = undefined;
1111 Goto(&call_runtime);
1112 }
1113 BIND(&argc_1);
1114 {
1115 Comment("1 Argument case");
1116 var_search_string = arguments.AtIndex(0);
1117 var_position = SmiConstant(0);
1118 Goto(&fast_path);
1119 }
1120 BIND(&argc_2);
1121 {
1122 Comment("2 Argument case");
1123 var_search_string = arguments.AtIndex(0);
1124 var_position = arguments.AtIndex(1);
1125 GotoIfNot(TaggedIsSmi(var_position.value()), &call_runtime);
1126 Goto(&fast_path);
1127 }
1128 BIND(&fast_path);
1129 {
1130 Comment("Fast Path");
1131 const TNode<Object> search = var_search_string.value();
1132 const TNode<Smi> position = CAST(var_position.value());
1133 GotoIf(TaggedIsSmi(receiver), &call_runtime);
1134 GotoIf(TaggedIsSmi(search), &call_runtime);
1135 GotoIfNot(IsString(CAST(receiver)), &call_runtime);
1136 GotoIfNot(IsString(CAST(search)), &call_runtime);
1137
1138 StringIndexOf(CAST(receiver), CAST(search), position,
1139 [&](TNode<Smi> result) {
1140 if (variant == kIndexOf) {
1141 arguments.PopAndReturn(result);
1142 } else {
1143 arguments.PopAndReturn(SelectBooleanConstant(
1144 SmiGreaterThanOrEqual(result, SmiConstant(0))));
1145 }
1146 });
1147 }
1148 BIND(&call_runtime);
1149 {
1150 Comment("Call Runtime");
1151 Runtime::FunctionId runtime = variant == kIndexOf
1152 ? Runtime::kStringIndexOf
1153 : Runtime::kStringIncludes;
1154 const TNode<Object> result =
1155 CallRuntime(runtime, context, receiver, var_search_string.value(),
1156 var_position.value());
1157 arguments.PopAndReturn(result);
1158 }
1159 }
1160
MaybeCallFunctionAtSymbol(const TNode<Context> context,const TNode<Object> object,const TNode<Object> maybe_string,Handle<Symbol> symbol,DescriptorIndexNameValue additional_property_to_check,const NodeFunction0 & regexp_call,const NodeFunction1 & generic_call)1161 void StringBuiltinsAssembler::MaybeCallFunctionAtSymbol(
1162 const TNode<Context> context, const TNode<Object> object,
1163 const TNode<Object> maybe_string, Handle<Symbol> symbol,
1164 DescriptorIndexNameValue additional_property_to_check,
1165 const NodeFunction0& regexp_call, const NodeFunction1& generic_call) {
1166 Label out(this);
1167 Label get_property_lookup(this);
1168
1169 // Smis have to go through the GetProperty lookup in case Number.prototype or
1170 // Object.prototype was modified.
1171 GotoIf(TaggedIsSmi(object), &get_property_lookup);
1172
1173 // Take the fast path for RegExps.
1174 // There's two conditions: {object} needs to be a fast regexp, and
1175 // {maybe_string} must be a string (we can't call ToString on the fast path
1176 // since it may mutate {object}).
1177 {
1178 Label stub_call(this), slow_lookup(this);
1179
1180 TNode<HeapObject> heap_object = CAST(object);
1181
1182 GotoIf(TaggedIsSmi(maybe_string), &slow_lookup);
1183 GotoIfNot(IsString(CAST(maybe_string)), &slow_lookup);
1184
1185 // Note we don't run a full (= permissive) check here, because passing the
1186 // check implies calling the fast variants of target builtins, which assume
1187 // we've already made their appropriate fast path checks. This is not the
1188 // case though; e.g.: some of the target builtins access flag getters.
1189 // TODO(jgruber): Handle slow flag accesses on the fast path and make this
1190 // permissive.
1191 RegExpBuiltinsAssembler regexp_asm(state());
1192 regexp_asm.BranchIfFastRegExp(
1193 context, heap_object, LoadMap(heap_object),
1194 PrototypeCheckAssembler::kCheckPrototypePropertyConstness,
1195 additional_property_to_check, &stub_call, &slow_lookup);
1196
1197 BIND(&stub_call);
1198 // TODO(jgruber): Add a no-JS scope once it exists.
1199 regexp_call();
1200
1201 BIND(&slow_lookup);
1202 // Special case null and undefined to skip the property lookup.
1203 Branch(IsNullOrUndefined(heap_object), &out, &get_property_lookup);
1204 }
1205
1206 // Fall back to a slow lookup of {heap_object[symbol]}.
1207 //
1208 // The spec uses GetMethod({heap_object}, {symbol}), which has a few quirks:
1209 // * null values are turned into undefined, and
1210 // * an exception is thrown if the value is not undefined, null, or callable.
1211 // We handle the former by jumping to {out} for null values as well, while
1212 // the latter is already handled by the Call({maybe_func}) operation.
1213
1214 BIND(&get_property_lookup);
1215 const TNode<Object> maybe_func = GetProperty(context, object, symbol);
1216 GotoIf(IsUndefined(maybe_func), &out);
1217 GotoIf(IsNull(maybe_func), &out);
1218
1219 // Attempt to call the function.
1220 generic_call(maybe_func);
1221
1222 BIND(&out);
1223 }
1224
IndexOfDollarChar(const TNode<Context> context,const TNode<String> string)1225 const TNode<Smi> StringBuiltinsAssembler::IndexOfDollarChar(
1226 const TNode<Context> context, const TNode<String> string) {
1227 const TNode<String> dollar_string = HeapConstant(
1228 isolate()->factory()->LookupSingleCharacterStringFromCode('$'));
1229 const TNode<Smi> dollar_ix =
1230 CAST(CallBuiltin(Builtins::kStringIndexOf, context, string, dollar_string,
1231 SmiConstant(0)));
1232 return dollar_ix;
1233 }
1234
GetSubstitution(TNode<Context> context,TNode<String> subject_string,TNode<Smi> match_start_index,TNode<Smi> match_end_index,TNode<String> replace_string)1235 TNode<String> StringBuiltinsAssembler::GetSubstitution(
1236 TNode<Context> context, TNode<String> subject_string,
1237 TNode<Smi> match_start_index, TNode<Smi> match_end_index,
1238 TNode<String> replace_string) {
1239 CSA_ASSERT(this, TaggedIsPositiveSmi(match_start_index));
1240 CSA_ASSERT(this, TaggedIsPositiveSmi(match_end_index));
1241
1242 TVARIABLE(String, var_result, replace_string);
1243 Label runtime(this), out(this);
1244
1245 // In this primitive implementation we simply look for the next '$' char in
1246 // {replace_string}. If it doesn't exist, we can simply return
1247 // {replace_string} itself. If it does, then we delegate to
1248 // String::GetSubstitution, passing in the index of the first '$' to avoid
1249 // repeated scanning work.
1250 // TODO(jgruber): Possibly extend this in the future to handle more complex
1251 // cases without runtime calls.
1252
1253 const TNode<Smi> dollar_index = IndexOfDollarChar(context, replace_string);
1254 Branch(SmiIsNegative(dollar_index), &out, &runtime);
1255
1256 BIND(&runtime);
1257 {
1258 CSA_ASSERT(this, TaggedIsPositiveSmi(dollar_index));
1259
1260 const TNode<Object> matched =
1261 CallBuiltin(Builtins::kStringSubstring, context, subject_string,
1262 SmiUntag(match_start_index), SmiUntag(match_end_index));
1263 const TNode<String> replacement_string = CAST(
1264 CallRuntime(Runtime::kGetSubstitution, context, matched, subject_string,
1265 match_start_index, replace_string, dollar_index));
1266 var_result = replacement_string;
1267
1268 Goto(&out);
1269 }
1270
1271 BIND(&out);
1272 return var_result.value();
1273 }
1274
1275 // ES6 #sec-string.prototype.replace
TF_BUILTIN(StringPrototypeReplace,StringBuiltinsAssembler)1276 TF_BUILTIN(StringPrototypeReplace, StringBuiltinsAssembler) {
1277 Label out(this);
1278
1279 auto receiver = Parameter<Object>(Descriptor::kReceiver);
1280 const auto search = Parameter<Object>(Descriptor::kSearch);
1281 const auto replace = Parameter<Object>(Descriptor::kReplace);
1282 auto context = Parameter<Context>(Descriptor::kContext);
1283
1284 const TNode<Smi> smi_zero = SmiConstant(0);
1285
1286 RequireObjectCoercible(context, receiver, "String.prototype.replace");
1287
1288 // Redirect to replacer method if {search[@@replace]} is not undefined.
1289
1290 MaybeCallFunctionAtSymbol(
1291 context, search, receiver, isolate()->factory()->replace_symbol(),
1292 DescriptorIndexNameValue{JSRegExp::kSymbolReplaceFunctionDescriptorIndex,
1293 RootIndex::kreplace_symbol,
1294 Context::REGEXP_REPLACE_FUNCTION_INDEX},
1295 [=]() {
1296 Return(CallBuiltin(Builtins::kRegExpReplace, context, search, receiver,
1297 replace));
1298 },
1299 [=](TNode<Object> fn) {
1300 Return(Call(context, fn, search, receiver, replace));
1301 });
1302
1303 // Convert {receiver} and {search} to strings.
1304
1305 const TNode<String> subject_string = ToString_Inline(context, receiver);
1306 const TNode<String> search_string = ToString_Inline(context, search);
1307
1308 const TNode<IntPtrT> subject_length = LoadStringLengthAsWord(subject_string);
1309 const TNode<IntPtrT> search_length = LoadStringLengthAsWord(search_string);
1310
1311 // Fast-path single-char {search}, long cons {receiver}, and simple string
1312 // {replace}.
1313 {
1314 Label next(this);
1315
1316 GotoIfNot(WordEqual(search_length, IntPtrConstant(1)), &next);
1317 GotoIfNot(IntPtrGreaterThan(subject_length, IntPtrConstant(0xFF)), &next);
1318 GotoIf(TaggedIsSmi(replace), &next);
1319 GotoIfNot(IsString(CAST(replace)), &next);
1320
1321 TNode<String> replace_string = CAST(replace);
1322 const TNode<Uint16T> subject_instance_type =
1323 LoadInstanceType(subject_string);
1324 GotoIfNot(IsConsStringInstanceType(subject_instance_type), &next);
1325
1326 GotoIf(TaggedIsPositiveSmi(IndexOfDollarChar(context, replace_string)),
1327 &next);
1328
1329 // Searching by traversing a cons string tree and replace with cons of
1330 // slices works only when the replaced string is a single character, being
1331 // replaced by a simple string and only pays off for long strings.
1332 // TODO(jgruber): Reevaluate if this is still beneficial.
1333 // TODO(jgruber): TailCallRuntime when it correctly handles adapter frames.
1334 Return(CallRuntime(Runtime::kStringReplaceOneCharWithString, context,
1335 subject_string, search_string, replace_string));
1336
1337 BIND(&next);
1338 }
1339
1340 // TODO(jgruber): Extend StringIndexOf to handle two-byte strings and
1341 // longer substrings - we can handle up to 8 chars (one-byte) / 4 chars
1342 // (2-byte).
1343
1344 const TNode<Smi> match_start_index =
1345 CAST(CallBuiltin(Builtins::kStringIndexOf, context, subject_string,
1346 search_string, smi_zero));
1347
1348 // Early exit if no match found.
1349 {
1350 Label next(this), return_subject(this);
1351
1352 GotoIfNot(SmiIsNegative(match_start_index), &next);
1353
1354 // The spec requires to perform ToString(replace) if the {replace} is not
1355 // callable even if we are going to exit here.
1356 // Since ToString() being applied to Smi does not have side effects for
1357 // numbers we can skip it.
1358 GotoIf(TaggedIsSmi(replace), &return_subject);
1359 GotoIf(IsCallableMap(LoadMap(CAST(replace))), &return_subject);
1360
1361 // TODO(jgruber): Could introduce ToStringSideeffectsStub which only
1362 // performs observable parts of ToString.
1363 ToString_Inline(context, replace);
1364 Goto(&return_subject);
1365
1366 BIND(&return_subject);
1367 Return(subject_string);
1368
1369 BIND(&next);
1370 }
1371
1372 const TNode<Smi> match_end_index =
1373 SmiAdd(match_start_index, SmiFromIntPtr(search_length));
1374
1375 TVARIABLE(String, var_result, EmptyStringConstant());
1376
1377 // Compute the prefix.
1378 {
1379 Label next(this);
1380
1381 GotoIf(SmiEqual(match_start_index, smi_zero), &next);
1382 const TNode<String> prefix =
1383 CAST(CallBuiltin(Builtins::kStringSubstring, context, subject_string,
1384 IntPtrConstant(0), SmiUntag(match_start_index)));
1385 var_result = prefix;
1386
1387 Goto(&next);
1388 BIND(&next);
1389 }
1390
1391 // Compute the string to replace with.
1392
1393 Label if_iscallablereplace(this), if_notcallablereplace(this);
1394 GotoIf(TaggedIsSmi(replace), &if_notcallablereplace);
1395 Branch(IsCallableMap(LoadMap(CAST(replace))), &if_iscallablereplace,
1396 &if_notcallablereplace);
1397
1398 BIND(&if_iscallablereplace);
1399 {
1400 const TNode<Object> replacement =
1401 Call(context, replace, UndefinedConstant(), search_string,
1402 match_start_index, subject_string);
1403 const TNode<String> replacement_string =
1404 ToString_Inline(context, replacement);
1405 var_result = CAST(CallBuiltin(Builtins::kStringAdd_CheckNone, context,
1406 var_result.value(), replacement_string));
1407 Goto(&out);
1408 }
1409
1410 BIND(&if_notcallablereplace);
1411 {
1412 const TNode<String> replace_string = ToString_Inline(context, replace);
1413 const TNode<Object> replacement =
1414 GetSubstitution(context, subject_string, match_start_index,
1415 match_end_index, replace_string);
1416 var_result = CAST(CallBuiltin(Builtins::kStringAdd_CheckNone, context,
1417 var_result.value(), replacement));
1418 Goto(&out);
1419 }
1420
1421 BIND(&out);
1422 {
1423 const TNode<Object> suffix =
1424 CallBuiltin(Builtins::kStringSubstring, context, subject_string,
1425 SmiUntag(match_end_index), subject_length);
1426 const TNode<Object> result = CallBuiltin(
1427 Builtins::kStringAdd_CheckNone, context, var_result.value(), suffix);
1428 Return(result);
1429 }
1430 }
1431
1432 class StringMatchSearchAssembler : public StringBuiltinsAssembler {
1433 public:
StringMatchSearchAssembler(compiler::CodeAssemblerState * state)1434 explicit StringMatchSearchAssembler(compiler::CodeAssemblerState* state)
1435 : StringBuiltinsAssembler(state) {}
1436
1437 protected:
1438 enum Variant { kMatch, kSearch };
1439
Generate(Variant variant,const char * method_name,TNode<Object> receiver,TNode<Object> maybe_regexp,TNode<Context> context)1440 void Generate(Variant variant, const char* method_name,
1441 TNode<Object> receiver, TNode<Object> maybe_regexp,
1442 TNode<Context> context) {
1443 Label call_regexp_match_search(this);
1444
1445 Builtins::Name builtin;
1446 Handle<Symbol> symbol;
1447 DescriptorIndexNameValue property_to_check;
1448 if (variant == kMatch) {
1449 builtin = Builtins::kRegExpMatchFast;
1450 symbol = isolate()->factory()->match_symbol();
1451 property_to_check = DescriptorIndexNameValue{
1452 JSRegExp::kSymbolMatchFunctionDescriptorIndex,
1453 RootIndex::kmatch_symbol, Context::REGEXP_MATCH_FUNCTION_INDEX};
1454 } else {
1455 builtin = Builtins::kRegExpSearchFast;
1456 symbol = isolate()->factory()->search_symbol();
1457 property_to_check = DescriptorIndexNameValue{
1458 JSRegExp::kSymbolSearchFunctionDescriptorIndex,
1459 RootIndex::ksearch_symbol, Context::REGEXP_SEARCH_FUNCTION_INDEX};
1460 }
1461
1462 RequireObjectCoercible(context, receiver, method_name);
1463
1464 MaybeCallFunctionAtSymbol(
1465 context, maybe_regexp, receiver, symbol, property_to_check,
1466 [=] { Return(CallBuiltin(builtin, context, maybe_regexp, receiver)); },
1467 [=](TNode<Object> fn) {
1468 Return(Call(context, fn, maybe_regexp, receiver));
1469 });
1470
1471 // maybe_regexp is not a RegExp nor has [@@match / @@search] property.
1472 {
1473 RegExpBuiltinsAssembler regexp_asm(state());
1474
1475 TNode<String> receiver_string = ToString_Inline(context, receiver);
1476 TNode<NativeContext> native_context = LoadNativeContext(context);
1477 TNode<HeapObject> regexp_function = CAST(
1478 LoadContextElement(native_context, Context::REGEXP_FUNCTION_INDEX));
1479 TNode<Map> initial_map = CAST(LoadObjectField(
1480 regexp_function, JSFunction::kPrototypeOrInitialMapOffset));
1481 TNode<Object> regexp = regexp_asm.RegExpCreate(
1482 context, initial_map, maybe_regexp, EmptyStringConstant());
1483
1484 // TODO(jgruber): Handle slow flag accesses on the fast path and make this
1485 // permissive.
1486 Label fast_path(this), slow_path(this);
1487 regexp_asm.BranchIfFastRegExp(
1488 context, CAST(regexp), initial_map,
1489 PrototypeCheckAssembler::kCheckPrototypePropertyConstness,
1490 property_to_check, &fast_path, &slow_path);
1491
1492 BIND(&fast_path);
1493 Return(CallBuiltin(builtin, context, regexp, receiver_string));
1494
1495 BIND(&slow_path);
1496 {
1497 TNode<Object> maybe_func = GetProperty(context, regexp, symbol);
1498 Return(Call(context, maybe_func, regexp, receiver_string));
1499 }
1500 }
1501 }
1502 };
1503
1504 // ES6 #sec-string.prototype.match
TF_BUILTIN(StringPrototypeMatch,StringMatchSearchAssembler)1505 TF_BUILTIN(StringPrototypeMatch, StringMatchSearchAssembler) {
1506 auto receiver = Parameter<Object>(Descriptor::kReceiver);
1507 auto maybe_regexp = Parameter<Object>(Descriptor::kRegexp);
1508 auto context = Parameter<Context>(Descriptor::kContext);
1509
1510 Generate(kMatch, "String.prototype.match", receiver, maybe_regexp, context);
1511 }
1512
1513 // ES #sec-string.prototype.matchAll
TF_BUILTIN(StringPrototypeMatchAll,StringBuiltinsAssembler)1514 TF_BUILTIN(StringPrototypeMatchAll, StringBuiltinsAssembler) {
1515 char const* method_name = "String.prototype.matchAll";
1516
1517 auto context = Parameter<Context>(Descriptor::kContext);
1518 auto maybe_regexp = Parameter<Object>(Descriptor::kRegexp);
1519 auto receiver = Parameter<Object>(Descriptor::kReceiver);
1520 TNode<NativeContext> native_context = LoadNativeContext(context);
1521
1522 // 1. Let O be ? RequireObjectCoercible(this value).
1523 RequireObjectCoercible(context, receiver, method_name);
1524
1525 RegExpMatchAllAssembler regexp_asm(state());
1526 {
1527 Label fast(this), slow(this, Label::kDeferred),
1528 throw_exception(this, Label::kDeferred),
1529 throw_flags_exception(this, Label::kDeferred), next(this);
1530
1531 // 2. If regexp is neither undefined nor null, then
1532 // a. Let isRegExp be ? IsRegExp(regexp).
1533 // b. If isRegExp is true, then
1534 // i. Let flags be ? Get(regexp, "flags").
1535 // ii. Perform ? RequireObjectCoercible(flags).
1536 // iii. If ? ToString(flags) does not contain "g", throw a
1537 // TypeError exception.
1538 GotoIf(TaggedIsSmi(maybe_regexp), &next);
1539 TNode<HeapObject> heap_maybe_regexp = CAST(maybe_regexp);
1540 regexp_asm.BranchIfFastRegExp_Strict(context, heap_maybe_regexp, &fast,
1541 &slow);
1542
1543 BIND(&fast);
1544 {
1545 TNode<BoolT> is_global = regexp_asm.FlagGetter(context, heap_maybe_regexp,
1546 JSRegExp::kGlobal, true);
1547 Branch(is_global, &next, &throw_exception);
1548 }
1549
1550 BIND(&slow);
1551 {
1552 GotoIfNot(regexp_asm.IsRegExp(native_context, heap_maybe_regexp), &next);
1553
1554 TNode<Object> flags = GetProperty(context, heap_maybe_regexp,
1555 isolate()->factory()->flags_string());
1556 // TODO(syg): Implement a RequireObjectCoercible with more flexible error
1557 // messages.
1558 GotoIf(IsNullOrUndefined(flags), &throw_flags_exception);
1559
1560 TNode<String> flags_string = ToString_Inline(context, flags);
1561 TNode<String> global_char_string = StringConstant("g");
1562 TNode<Smi> global_ix =
1563 CAST(CallBuiltin(Builtins::kStringIndexOf, context, flags_string,
1564 global_char_string, SmiConstant(0)));
1565 Branch(SmiEqual(global_ix, SmiConstant(-1)), &throw_exception, &next);
1566 }
1567
1568 BIND(&throw_exception);
1569 ThrowTypeError(context, MessageTemplate::kRegExpGlobalInvokedOnNonGlobal,
1570 method_name);
1571
1572 BIND(&throw_flags_exception);
1573 ThrowTypeError(context,
1574 MessageTemplate::kStringMatchAllNullOrUndefinedFlags);
1575
1576 BIND(&next);
1577 }
1578 // a. Let matcher be ? GetMethod(regexp, @@matchAll).
1579 // b. If matcher is not undefined, then
1580 // i. Return ? Call(matcher, regexp, « O »).
1581 auto if_regexp_call = [&] {
1582 // MaybeCallFunctionAtSymbol guarantees fast path is chosen only if
1583 // maybe_regexp is a fast regexp and receiver is a string.
1584 TNode<String> s = CAST(receiver);
1585
1586 Return(
1587 RegExpPrototypeMatchAllImpl(context, native_context, maybe_regexp, s));
1588 };
1589 auto if_generic_call = [=](TNode<Object> fn) {
1590 Return(Call(context, fn, maybe_regexp, receiver));
1591 };
1592 MaybeCallFunctionAtSymbol(
1593 context, maybe_regexp, receiver, isolate()->factory()->match_all_symbol(),
1594 DescriptorIndexNameValue{JSRegExp::kSymbolMatchAllFunctionDescriptorIndex,
1595 RootIndex::kmatch_all_symbol,
1596 Context::REGEXP_MATCH_ALL_FUNCTION_INDEX},
1597 if_regexp_call, if_generic_call);
1598
1599 // 3. Let S be ? ToString(O).
1600 TNode<String> s = ToString_Inline(context, receiver);
1601
1602 // 4. Let rx be ? RegExpCreate(R, "g").
1603 TNode<Object> rx = regexp_asm.RegExpCreate(context, native_context,
1604 maybe_regexp, StringConstant("g"));
1605
1606 // 5. Return ? Invoke(rx, @@matchAll, « S »).
1607 TNode<Object> match_all_func =
1608 GetProperty(context, rx, isolate()->factory()->match_all_symbol());
1609 Return(Call(context, match_all_func, rx, s));
1610 }
1611
1612 // ES6 #sec-string.prototype.search
TF_BUILTIN(StringPrototypeSearch,StringMatchSearchAssembler)1613 TF_BUILTIN(StringPrototypeSearch, StringMatchSearchAssembler) {
1614 auto receiver = Parameter<Object>(Descriptor::kReceiver);
1615 auto maybe_regexp = Parameter<Object>(Descriptor::kRegexp);
1616 auto context = Parameter<Context>(Descriptor::kContext);
1617 Generate(kSearch, "String.prototype.search", receiver, maybe_regexp, context);
1618 }
1619
StringToArray(TNode<NativeContext> context,TNode<String> subject_string,TNode<Smi> subject_length,TNode<Number> limit_number)1620 TNode<JSArray> StringBuiltinsAssembler::StringToArray(
1621 TNode<NativeContext> context, TNode<String> subject_string,
1622 TNode<Smi> subject_length, TNode<Number> limit_number) {
1623 CSA_ASSERT(this, SmiGreaterThan(subject_length, SmiConstant(0)));
1624
1625 Label done(this), call_runtime(this, Label::kDeferred),
1626 fill_thehole_and_call_runtime(this, Label::kDeferred);
1627 TVARIABLE(JSArray, result_array);
1628
1629 TNode<Uint16T> instance_type = LoadInstanceType(subject_string);
1630 GotoIfNot(IsOneByteStringInstanceType(instance_type), &call_runtime);
1631
1632 // Try to use cached one byte characters.
1633 {
1634 TNode<Smi> length_smi =
1635 Select<Smi>(TaggedIsSmi(limit_number),
1636 [=] { return SmiMin(CAST(limit_number), subject_length); },
1637 [=] { return subject_length; });
1638 TNode<IntPtrT> length = SmiToIntPtr(length_smi);
1639
1640 ToDirectStringAssembler to_direct(state(), subject_string);
1641 to_direct.TryToDirect(&call_runtime);
1642
1643 // The extracted direct string may be two-byte even though the wrapping
1644 // string is one-byte.
1645 GotoIfNot(IsOneByteStringInstanceType(to_direct.instance_type()),
1646 &call_runtime);
1647
1648 TNode<FixedArray> elements = CAST(AllocateFixedArray(
1649 PACKED_ELEMENTS, length, AllocationFlag::kAllowLargeObjectAllocation));
1650 // Don't allocate anything while {string_data} is live!
1651 TNode<RawPtrT> string_data =
1652 to_direct.PointerToData(&fill_thehole_and_call_runtime);
1653 TNode<IntPtrT> string_data_offset = to_direct.offset();
1654 TNode<FixedArray> cache = SingleCharacterStringCacheConstant();
1655
1656 BuildFastLoop<IntPtrT>(
1657 IntPtrConstant(0), length,
1658 [&](TNode<IntPtrT> index) {
1659 // TODO(jkummerow): Implement a CSA version of DisallowHeapAllocation
1660 // and use that to guard ToDirectStringAssembler.PointerToData().
1661 CSA_ASSERT(this, WordEqual(to_direct.PointerToData(&call_runtime),
1662 string_data));
1663 TNode<Int32T> char_code =
1664 UncheckedCast<Int32T>(Load(MachineType::Uint8(), string_data,
1665 IntPtrAdd(index, string_data_offset)));
1666 TNode<UintPtrT> code_index = ChangeUint32ToWord(char_code);
1667 TNode<Object> entry = LoadFixedArrayElement(cache, code_index);
1668
1669 // If we cannot find a char in the cache, fill the hole for the fixed
1670 // array, and call runtime.
1671 GotoIf(IsUndefined(entry), &fill_thehole_and_call_runtime);
1672
1673 StoreFixedArrayElement(elements, index, entry);
1674 },
1675 1, IndexAdvanceMode::kPost);
1676
1677 TNode<Map> array_map = LoadJSArrayElementsMap(PACKED_ELEMENTS, context);
1678 result_array = AllocateJSArray(array_map, elements, length_smi);
1679 Goto(&done);
1680
1681 BIND(&fill_thehole_and_call_runtime);
1682 {
1683 FillFixedArrayWithValue(PACKED_ELEMENTS, elements, IntPtrConstant(0),
1684 length, RootIndex::kTheHoleValue);
1685 Goto(&call_runtime);
1686 }
1687 }
1688
1689 BIND(&call_runtime);
1690 {
1691 result_array = CAST(CallRuntime(Runtime::kStringToArray, context,
1692 subject_string, limit_number));
1693 Goto(&done);
1694 }
1695
1696 BIND(&done);
1697 return result_array.value();
1698 }
1699
1700 // ES6 section 21.1.3.19 String.prototype.split ( separator, limit )
TF_BUILTIN(StringPrototypeSplit,StringBuiltinsAssembler)1701 TF_BUILTIN(StringPrototypeSplit, StringBuiltinsAssembler) {
1702 const int kSeparatorArg = 0;
1703 const int kLimitArg = 1;
1704
1705 const TNode<IntPtrT> argc = ChangeInt32ToIntPtr(
1706 UncheckedParameter<Int32T>(Descriptor::kJSActualArgumentsCount));
1707 CodeStubArguments args(this, argc);
1708
1709 TNode<Object> receiver = args.GetReceiver();
1710 const TNode<Object> separator = args.GetOptionalArgumentValue(kSeparatorArg);
1711 const TNode<Object> limit = args.GetOptionalArgumentValue(kLimitArg);
1712 auto context = Parameter<NativeContext>(Descriptor::kContext);
1713
1714 TNode<Smi> smi_zero = SmiConstant(0);
1715
1716 RequireObjectCoercible(context, receiver, "String.prototype.split");
1717
1718 // Redirect to splitter method if {separator[@@split]} is not undefined.
1719
1720 MaybeCallFunctionAtSymbol(
1721 context, separator, receiver, isolate()->factory()->split_symbol(),
1722 DescriptorIndexNameValue{JSRegExp::kSymbolSplitFunctionDescriptorIndex,
1723 RootIndex::ksplit_symbol,
1724 Context::REGEXP_SPLIT_FUNCTION_INDEX},
1725 [&]() {
1726 args.PopAndReturn(CallBuiltin(Builtins::kRegExpSplit, context,
1727 separator, receiver, limit));
1728 },
1729 [&](TNode<Object> fn) {
1730 args.PopAndReturn(Call(context, fn, separator, receiver, limit));
1731 });
1732
1733 // String and integer conversions.
1734
1735 TNode<String> subject_string = ToString_Inline(context, receiver);
1736 TNode<Number> limit_number = Select<Number>(
1737 IsUndefined(limit), [=] { return NumberConstant(kMaxUInt32); },
1738 [=] { return ToUint32(context, limit); });
1739 const TNode<String> separator_string = ToString_Inline(context, separator);
1740
1741 Label return_empty_array(this);
1742
1743 // Shortcut for {limit} == 0.
1744 GotoIf(TaggedEqual(limit_number, smi_zero), &return_empty_array);
1745
1746 // ECMA-262 says that if {separator} is undefined, the result should
1747 // be an array of size 1 containing the entire string.
1748 {
1749 Label next(this);
1750 GotoIfNot(IsUndefined(separator), &next);
1751
1752 const ElementsKind kind = PACKED_ELEMENTS;
1753 const TNode<NativeContext> native_context = LoadNativeContext(context);
1754 TNode<Map> array_map = LoadJSArrayElementsMap(kind, native_context);
1755
1756 TNode<Smi> length = SmiConstant(1);
1757 TNode<IntPtrT> capacity = IntPtrConstant(1);
1758 TNode<JSArray> result = AllocateJSArray(kind, array_map, capacity, length);
1759
1760 TNode<FixedArray> fixed_array = CAST(LoadElements(result));
1761 StoreFixedArrayElement(fixed_array, 0, subject_string);
1762
1763 args.PopAndReturn(result);
1764
1765 BIND(&next);
1766 }
1767
1768 // If the separator string is empty then return the elements in the subject.
1769 {
1770 Label next(this);
1771 GotoIfNot(SmiEqual(LoadStringLengthAsSmi(separator_string), smi_zero),
1772 &next);
1773
1774 TNode<Smi> subject_length = LoadStringLengthAsSmi(subject_string);
1775 GotoIf(SmiEqual(subject_length, smi_zero), &return_empty_array);
1776
1777 args.PopAndReturn(
1778 StringToArray(context, subject_string, subject_length, limit_number));
1779
1780 BIND(&next);
1781 }
1782
1783 const TNode<Object> result =
1784 CallRuntime(Runtime::kStringSplit, context, subject_string,
1785 separator_string, limit_number);
1786 args.PopAndReturn(result);
1787
1788 BIND(&return_empty_array);
1789 {
1790 const ElementsKind kind = PACKED_ELEMENTS;
1791 const TNode<NativeContext> native_context = LoadNativeContext(context);
1792 TNode<Map> array_map = LoadJSArrayElementsMap(kind, native_context);
1793
1794 TNode<Smi> length = smi_zero;
1795 TNode<IntPtrT> capacity = IntPtrConstant(0);
1796 TNode<JSArray> result = AllocateJSArray(kind, array_map, capacity, length);
1797
1798 args.PopAndReturn(result);
1799 }
1800 }
1801
TF_BUILTIN(StringSubstring,StringBuiltinsAssembler)1802 TF_BUILTIN(StringSubstring, StringBuiltinsAssembler) {
1803 auto string = Parameter<String>(Descriptor::kString);
1804 auto from = UncheckedParameter<IntPtrT>(Descriptor::kFrom);
1805 auto to = UncheckedParameter<IntPtrT>(Descriptor::kTo);
1806
1807 Return(SubString(string, from, to));
1808 }
1809
1810
1811 // Return the |word32| codepoint at {index}. Supports SeqStrings and
1812 // ExternalStrings.
1813 // TODO(v8:9880): Use UintPtrT here.
LoadSurrogatePairAt(TNode<String> string,TNode<IntPtrT> length,TNode<IntPtrT> index,UnicodeEncoding encoding)1814 TNode<Int32T> StringBuiltinsAssembler::LoadSurrogatePairAt(
1815 TNode<String> string, TNode<IntPtrT> length, TNode<IntPtrT> index,
1816 UnicodeEncoding encoding) {
1817 Label handle_surrogate_pair(this), return_result(this);
1818 TVARIABLE(Int32T, var_result);
1819 TVARIABLE(Int32T, var_trail);
1820 var_result = StringCharCodeAt(string, Unsigned(index));
1821 var_trail = Int32Constant(0);
1822
1823 GotoIf(Word32NotEqual(Word32And(var_result.value(), Int32Constant(0xFC00)),
1824 Int32Constant(0xD800)),
1825 &return_result);
1826 TNode<IntPtrT> next_index = IntPtrAdd(index, IntPtrConstant(1));
1827
1828 GotoIfNot(IntPtrLessThan(next_index, length), &return_result);
1829 var_trail = StringCharCodeAt(string, Unsigned(next_index));
1830 Branch(Word32Equal(Word32And(var_trail.value(), Int32Constant(0xFC00)),
1831 Int32Constant(0xDC00)),
1832 &handle_surrogate_pair, &return_result);
1833
1834 BIND(&handle_surrogate_pair);
1835 {
1836 TNode<Int32T> lead = var_result.value();
1837 TNode<Int32T> trail = var_trail.value();
1838
1839 // Check that this path is only taken if a surrogate pair is found
1840 CSA_SLOW_ASSERT(this,
1841 Uint32GreaterThanOrEqual(lead, Int32Constant(0xD800)));
1842 CSA_SLOW_ASSERT(this, Uint32LessThan(lead, Int32Constant(0xDC00)));
1843 CSA_SLOW_ASSERT(this,
1844 Uint32GreaterThanOrEqual(trail, Int32Constant(0xDC00)));
1845 CSA_SLOW_ASSERT(this, Uint32LessThan(trail, Int32Constant(0xE000)));
1846
1847 switch (encoding) {
1848 case UnicodeEncoding::UTF16:
1849 var_result = Word32Or(
1850 // Need to swap the order for big-endian platforms
1851 #if V8_TARGET_BIG_ENDIAN
1852 Word32Shl(lead, Int32Constant(16)), trail);
1853 #else
1854 Word32Shl(trail, Int32Constant(16)), lead);
1855 #endif
1856 break;
1857
1858 case UnicodeEncoding::UTF32: {
1859 // Convert UTF16 surrogate pair into |word32| code point, encoded as
1860 // UTF32.
1861 TNode<Int32T> surrogate_offset =
1862 Int32Constant(0x10000 - (0xD800 << 10) - 0xDC00);
1863
1864 // (lead << 10) + trail + SURROGATE_OFFSET
1865 var_result = Int32Add(Word32Shl(lead, Int32Constant(10)),
1866 Int32Add(trail, surrogate_offset));
1867 break;
1868 }
1869 }
1870 Goto(&return_result);
1871 }
1872
1873 BIND(&return_result);
1874 return var_result.value();
1875 }
1876
BranchIfStringPrimitiveWithNoCustomIteration(TNode<Object> object,TNode<Context> context,Label * if_true,Label * if_false)1877 void StringBuiltinsAssembler::BranchIfStringPrimitiveWithNoCustomIteration(
1878 TNode<Object> object, TNode<Context> context, Label* if_true,
1879 Label* if_false) {
1880 GotoIf(TaggedIsSmi(object), if_false);
1881 GotoIfNot(IsString(CAST(object)), if_false);
1882
1883 // Check that the String iterator hasn't been modified in a way that would
1884 // affect iteration.
1885 TNode<PropertyCell> protector_cell = StringIteratorProtectorConstant();
1886 DCHECK(isolate()->heap()->string_iterator_protector().IsPropertyCell());
1887 Branch(
1888 TaggedEqual(LoadObjectField(protector_cell, PropertyCell::kValueOffset),
1889 SmiConstant(Protectors::kProtectorValid)),
1890 if_true, if_false);
1891 }
1892
1893 // Instantiate template due to shared library requirements.
1894 template V8_EXPORT_PRIVATE void StringBuiltinsAssembler::CopyStringCharacters(
1895 TNode<String> from_string, TNode<String> to_string,
1896 TNode<IntPtrT> from_index, TNode<IntPtrT> to_index,
1897 TNode<IntPtrT> character_count, String::Encoding from_encoding,
1898 String::Encoding to_encoding);
1899
1900 template V8_EXPORT_PRIVATE void StringBuiltinsAssembler::CopyStringCharacters(
1901 TNode<RawPtrT> from_string, TNode<String> to_string,
1902 TNode<IntPtrT> from_index, TNode<IntPtrT> to_index,
1903 TNode<IntPtrT> character_count, String::Encoding from_encoding,
1904 String::Encoding to_encoding);
1905
1906 template <typename T>
CopyStringCharacters(TNode<T> from_string,TNode<String> to_string,TNode<IntPtrT> from_index,TNode<IntPtrT> to_index,TNode<IntPtrT> character_count,String::Encoding from_encoding,String::Encoding to_encoding)1907 void StringBuiltinsAssembler::CopyStringCharacters(
1908 TNode<T> from_string, TNode<String> to_string, TNode<IntPtrT> from_index,
1909 TNode<IntPtrT> to_index, TNode<IntPtrT> character_count,
1910 String::Encoding from_encoding, String::Encoding to_encoding) {
1911 // from_string could be either a String or a RawPtrT in the case we pass in
1912 // faked sequential strings when handling external subject strings.
1913 bool from_one_byte = from_encoding == String::ONE_BYTE_ENCODING;
1914 bool to_one_byte = to_encoding == String::ONE_BYTE_ENCODING;
1915 DCHECK_IMPLIES(to_one_byte, from_one_byte);
1916 Comment("CopyStringCharacters ",
1917 from_one_byte ? "ONE_BYTE_ENCODING" : "TWO_BYTE_ENCODING", " -> ",
1918 to_one_byte ? "ONE_BYTE_ENCODING" : "TWO_BYTE_ENCODING");
1919
1920 ElementsKind from_kind = from_one_byte ? UINT8_ELEMENTS : UINT16_ELEMENTS;
1921 ElementsKind to_kind = to_one_byte ? UINT8_ELEMENTS : UINT16_ELEMENTS;
1922 STATIC_ASSERT(SeqOneByteString::kHeaderSize == SeqTwoByteString::kHeaderSize);
1923 int header_size = SeqOneByteString::kHeaderSize - kHeapObjectTag;
1924 TNode<IntPtrT> from_offset =
1925 ElementOffsetFromIndex(from_index, from_kind, header_size);
1926 TNode<IntPtrT> to_offset =
1927 ElementOffsetFromIndex(to_index, to_kind, header_size);
1928 TNode<IntPtrT> byte_count =
1929 ElementOffsetFromIndex(character_count, from_kind);
1930 TNode<IntPtrT> limit_offset = IntPtrAdd(from_offset, byte_count);
1931
1932 // Prepare the fast loop
1933 MachineType type =
1934 from_one_byte ? MachineType::Uint8() : MachineType::Uint16();
1935 MachineRepresentation rep = to_one_byte ? MachineRepresentation::kWord8
1936 : MachineRepresentation::kWord16;
1937 int from_increment = 1 << ElementsKindToShiftSize(from_kind);
1938 int to_increment = 1 << ElementsKindToShiftSize(to_kind);
1939
1940 TVARIABLE(IntPtrT, current_to_offset, to_offset);
1941 VariableList vars({¤t_to_offset}, zone());
1942 int to_index_constant = 0, from_index_constant = 0;
1943 bool index_same = (from_encoding == to_encoding) &&
1944 (from_index == to_index ||
1945 (ToInt32Constant(from_index, &from_index_constant) &&
1946 ToInt32Constant(to_index, &to_index_constant) &&
1947 from_index_constant == to_index_constant));
1948 BuildFastLoop<IntPtrT>(
1949 vars, from_offset, limit_offset,
1950 [&](TNode<IntPtrT> offset) {
1951 StoreNoWriteBarrier(rep, to_string,
1952 index_same ? offset : current_to_offset.value(),
1953 Load(type, from_string, offset));
1954 if (!index_same) {
1955 Increment(¤t_to_offset, to_increment);
1956 }
1957 },
1958 from_increment, IndexAdvanceMode::kPost);
1959 }
1960
1961 // A wrapper around CopyStringCharacters which determines the correct string
1962 // encoding, allocates a corresponding sequential string, and then copies the
1963 // given character range using CopyStringCharacters.
1964 // |from_string| must be a sequential string.
1965 // 0 <= |from_index| <= |from_index| + |character_count| < from_string.length.
1966 template <typename T>
AllocAndCopyStringCharacters(TNode<T> from,TNode<Int32T> from_instance_type,TNode<IntPtrT> from_index,TNode<IntPtrT> character_count)1967 TNode<String> StringBuiltinsAssembler::AllocAndCopyStringCharacters(
1968 TNode<T> from, TNode<Int32T> from_instance_type, TNode<IntPtrT> from_index,
1969 TNode<IntPtrT> character_count) {
1970 Label end(this), one_byte_sequential(this), two_byte_sequential(this);
1971 TVARIABLE(String, var_result);
1972
1973 Branch(IsOneByteStringInstanceType(from_instance_type), &one_byte_sequential,
1974 &two_byte_sequential);
1975
1976 // The subject string is a sequential one-byte string.
1977 BIND(&one_byte_sequential);
1978 {
1979 TNode<String> result = AllocateSeqOneByteString(
1980 Unsigned(TruncateIntPtrToInt32(character_count)));
1981 CopyStringCharacters<T>(from, result, from_index, IntPtrConstant(0),
1982 character_count, String::ONE_BYTE_ENCODING,
1983 String::ONE_BYTE_ENCODING);
1984 var_result = result;
1985 Goto(&end);
1986 }
1987
1988 // The subject string is a sequential two-byte string.
1989 BIND(&two_byte_sequential);
1990 {
1991 TNode<String> result = AllocateSeqTwoByteString(
1992 Unsigned(TruncateIntPtrToInt32(character_count)));
1993 CopyStringCharacters<T>(from, result, from_index, IntPtrConstant(0),
1994 character_count, String::TWO_BYTE_ENCODING,
1995 String::TWO_BYTE_ENCODING);
1996 var_result = result;
1997 Goto(&end);
1998 }
1999
2000 BIND(&end);
2001 return var_result.value();
2002 }
2003
2004 // TODO(v8:9880): Use UintPtrT here.
SubString(TNode<String> string,TNode<IntPtrT> from,TNode<IntPtrT> to)2005 TNode<String> StringBuiltinsAssembler::SubString(TNode<String> string,
2006 TNode<IntPtrT> from,
2007 TNode<IntPtrT> to) {
2008 TVARIABLE(String, var_result);
2009 ToDirectStringAssembler to_direct(state(), string);
2010 Label end(this), runtime(this);
2011
2012 const TNode<IntPtrT> substr_length = IntPtrSub(to, from);
2013 const TNode<IntPtrT> string_length = LoadStringLengthAsWord(string);
2014
2015 // Begin dispatching based on substring length.
2016
2017 Label original_string_or_invalid_length(this);
2018 GotoIf(UintPtrGreaterThanOrEqual(substr_length, string_length),
2019 &original_string_or_invalid_length);
2020
2021 // A real substring (substr_length < string_length).
2022 Label empty(this);
2023 GotoIf(IntPtrEqual(substr_length, IntPtrConstant(0)), &empty);
2024
2025 Label single_char(this);
2026 GotoIf(IntPtrEqual(substr_length, IntPtrConstant(1)), &single_char);
2027
2028 // Deal with different string types: update the index if necessary
2029 // and extract the underlying string.
2030
2031 TNode<String> direct_string = to_direct.TryToDirect(&runtime);
2032 TNode<IntPtrT> offset = IntPtrAdd(from, to_direct.offset());
2033 const TNode<Int32T> instance_type = to_direct.instance_type();
2034
2035 // The subject string can only be external or sequential string of either
2036 // encoding at this point.
2037 Label external_string(this);
2038 {
2039 if (FLAG_string_slices) {
2040 Label next(this);
2041
2042 // Short slice. Copy instead of slicing.
2043 GotoIf(IntPtrLessThan(substr_length,
2044 IntPtrConstant(SlicedString::kMinLength)),
2045 &next);
2046
2047 // Allocate new sliced string.
2048
2049 Counters* counters = isolate()->counters();
2050 IncrementCounter(counters->sub_string_native(), 1);
2051
2052 Label one_byte_slice(this), two_byte_slice(this);
2053 Branch(IsOneByteStringInstanceType(to_direct.instance_type()),
2054 &one_byte_slice, &two_byte_slice);
2055
2056 BIND(&one_byte_slice);
2057 {
2058 var_result = AllocateSlicedOneByteString(
2059 Unsigned(TruncateIntPtrToInt32(substr_length)), direct_string,
2060 SmiTag(offset));
2061 Goto(&end);
2062 }
2063
2064 BIND(&two_byte_slice);
2065 {
2066 var_result = AllocateSlicedTwoByteString(
2067 Unsigned(TruncateIntPtrToInt32(substr_length)), direct_string,
2068 SmiTag(offset));
2069 Goto(&end);
2070 }
2071
2072 BIND(&next);
2073 }
2074
2075 // The subject string can only be external or sequential string of either
2076 // encoding at this point.
2077 GotoIf(to_direct.is_external(), &external_string);
2078
2079 var_result = AllocAndCopyStringCharacters(direct_string, instance_type,
2080 offset, substr_length);
2081
2082 Counters* counters = isolate()->counters();
2083 IncrementCounter(counters->sub_string_native(), 1);
2084
2085 Goto(&end);
2086 }
2087
2088 // Handle external string.
2089 BIND(&external_string);
2090 {
2091 const TNode<RawPtrT> fake_sequential_string =
2092 to_direct.PointerToString(&runtime);
2093
2094 var_result = AllocAndCopyStringCharacters(
2095 fake_sequential_string, instance_type, offset, substr_length);
2096
2097 Counters* counters = isolate()->counters();
2098 IncrementCounter(counters->sub_string_native(), 1);
2099
2100 Goto(&end);
2101 }
2102
2103 BIND(&empty);
2104 {
2105 var_result = EmptyStringConstant();
2106 Goto(&end);
2107 }
2108
2109 // Substrings of length 1 are generated through CharCodeAt and FromCharCode.
2110 BIND(&single_char);
2111 {
2112 TNode<Int32T> char_code = StringCharCodeAt(string, Unsigned(from));
2113 var_result = StringFromSingleCharCode(char_code);
2114 Goto(&end);
2115 }
2116
2117 BIND(&original_string_or_invalid_length);
2118 {
2119 CSA_ASSERT(this, IntPtrEqual(substr_length, string_length));
2120
2121 // Equal length - check if {from, to} == {0, str.length}.
2122 GotoIf(UintPtrGreaterThan(from, IntPtrConstant(0)), &runtime);
2123
2124 // Return the original string (substr_length == string_length).
2125
2126 Counters* counters = isolate()->counters();
2127 IncrementCounter(counters->sub_string_native(), 1);
2128
2129 var_result = string;
2130 Goto(&end);
2131 }
2132
2133 // Fall back to a runtime call.
2134 BIND(&runtime);
2135 {
2136 var_result =
2137 CAST(CallRuntime(Runtime::kStringSubstring, NoContextConstant(), string,
2138 SmiTag(from), SmiTag(to)));
2139 Goto(&end);
2140 }
2141
2142 BIND(&end);
2143 return var_result.value();
2144 }
2145
2146 } // namespace internal
2147 } // namespace v8
2148