1 | // Copyright (c) 2013, the Dart project authors. Please see the AUTHORS file |
2 | // for details. All rights reserved. Use of this source code is governed by a |
3 | // BSD-style license that can be found in the LICENSE file. |
4 | |
5 | #include "vm/globals.h" // Needed here to get TARGET_ARCH_X64. |
6 | #if defined(TARGET_ARCH_X64) |
7 | |
8 | #include "vm/compiler/backend/il.h" |
9 | |
10 | #include "vm/compiler/assembler/assembler.h" |
11 | #include "vm/compiler/backend/flow_graph.h" |
12 | #include "vm/compiler/backend/flow_graph_compiler.h" |
13 | #include "vm/compiler/backend/locations.h" |
14 | #include "vm/compiler/backend/locations_helpers.h" |
15 | #include "vm/compiler/backend/range_analysis.h" |
16 | #include "vm/compiler/ffi/native_calling_convention.h" |
17 | #include "vm/compiler/jit/compiler.h" |
18 | #include "vm/dart_entry.h" |
19 | #include "vm/instructions.h" |
20 | #include "vm/object_store.h" |
21 | #include "vm/parser.h" |
22 | #include "vm/stack_frame.h" |
23 | #include "vm/stub_code.h" |
24 | #include "vm/symbols.h" |
25 | #include "vm/type_testing_stubs.h" |
26 | |
27 | #define __ compiler->assembler()-> |
28 | #define Z (compiler->zone()) |
29 | |
30 | namespace dart { |
31 | |
32 | // Generic summary for call instructions that have all arguments pushed |
33 | // on the stack and return the result in a fixed register RAX (or XMM0 if |
34 | // the return type is double). |
35 | LocationSummary* Instruction::MakeCallSummary(Zone* zone, |
36 | const Instruction* instr, |
37 | LocationSummary* locs) { |
38 | ASSERT(locs == nullptr || locs->always_calls()); |
39 | LocationSummary* result = |
40 | ((locs == nullptr) |
41 | ? (new (zone) LocationSummary(zone, 0, 0, LocationSummary::kCall)) |
42 | : locs); |
43 | const auto representation = instr->representation(); |
44 | switch (representation) { |
45 | case kTagged: |
46 | case kUnboxedInt64: |
47 | result->set_out( |
48 | 0, Location::RegisterLocation(CallingConventions::kReturnReg)); |
49 | break; |
50 | case kUnboxedDouble: |
51 | result->set_out( |
52 | 0, Location::FpuRegisterLocation(CallingConventions::kReturnFpuReg)); |
53 | break; |
54 | default: |
55 | UNREACHABLE(); |
56 | break; |
57 | } |
58 | return result; |
59 | } |
60 | |
61 | LocationSummary* LoadIndexedUnsafeInstr::MakeLocationSummary(Zone* zone, |
62 | bool opt) const { |
63 | const intptr_t kNumInputs = 1; |
64 | const intptr_t kNumTemps = 0; |
65 | LocationSummary* locs = new (zone) |
66 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
67 | |
68 | locs->set_in(0, Location::RequiresRegister()); |
69 | switch (representation()) { |
70 | case kTagged: |
71 | case kUnboxedInt64: |
72 | locs->set_out(0, Location::RequiresRegister()); |
73 | break; |
74 | case kUnboxedDouble: |
75 | locs->set_out(0, Location::RequiresFpuRegister()); |
76 | break; |
77 | default: |
78 | UNREACHABLE(); |
79 | break; |
80 | } |
81 | return locs; |
82 | } |
83 | |
84 | void LoadIndexedUnsafeInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
85 | ASSERT(RequiredInputRepresentation(0) == kTagged); // It is a Smi. |
86 | ASSERT(kSmiTag == 0); |
87 | ASSERT(kSmiTagSize == 1); |
88 | |
89 | const Register index = locs()->in(0).reg(); |
90 | |
91 | switch (representation()) { |
92 | case kTagged: |
93 | case kUnboxedInt64: { |
94 | const auto out = locs()->out(0).reg(); |
95 | __ movq(out, compiler::Address(base_reg(), index, TIMES_4, offset())); |
96 | break; |
97 | } |
98 | case kUnboxedDouble: { |
99 | const auto out = locs()->out(0).fpu_reg(); |
100 | __ movsd(out, compiler::Address(base_reg(), index, TIMES_4, offset())); |
101 | break; |
102 | } |
103 | default: |
104 | UNREACHABLE(); |
105 | break; |
106 | } |
107 | } |
108 | |
109 | DEFINE_BACKEND(StoreIndexedUnsafe, |
110 | (NoLocation, Register index, Register value)) { |
111 | ASSERT(instr->RequiredInputRepresentation( |
112 | StoreIndexedUnsafeInstr::kIndexPos) == kTagged); // It is a Smi. |
113 | __ movq(compiler::Address(instr->base_reg(), index, TIMES_4, instr->offset()), |
114 | value); |
115 | |
116 | ASSERT(kSmiTag == 0); |
117 | ASSERT(kSmiTagSize == 1); |
118 | } |
119 | |
120 | DEFINE_BACKEND(TailCall, (NoLocation, Fixed<Register, ARGS_DESC_REG>)) { |
121 | compiler->EmitTailCallToStub(instr->code()); |
122 | |
123 | // Even though the TailCallInstr will be the last instruction in a basic |
124 | // block, the flow graph compiler will emit native code for other blocks after |
125 | // the one containing this instruction and needs to be able to use the pool. |
126 | // (The `LeaveDartFrame` above disables usages of the pool.) |
127 | __ set_constant_pool_allowed(true); |
128 | } |
129 | |
130 | LocationSummary* MemoryCopyInstr::MakeLocationSummary(Zone* zone, |
131 | bool opt) const { |
132 | const intptr_t kNumInputs = 5; |
133 | const intptr_t kNumTemps = 0; |
134 | LocationSummary* locs = new (zone) |
135 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
136 | locs->set_in(kSrcPos, Location::RegisterLocation(RSI)); |
137 | locs->set_in(kDestPos, Location::RegisterLocation(RDI)); |
138 | locs->set_in(kSrcStartPos, Location::WritableRegister()); |
139 | locs->set_in(kDestStartPos, Location::WritableRegister()); |
140 | locs->set_in(kLengthPos, Location::RegisterLocation(RCX)); |
141 | return locs; |
142 | } |
143 | |
144 | void MemoryCopyInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
145 | const Register src_start_reg = locs()->in(kSrcStartPos).reg(); |
146 | const Register dest_start_reg = locs()->in(kDestStartPos).reg(); |
147 | |
148 | EmitComputeStartPointer(compiler, src_cid_, src_start(), RSI, src_start_reg); |
149 | EmitComputeStartPointer(compiler, dest_cid_, dest_start(), RDI, |
150 | dest_start_reg); |
151 | if (element_size_ <= 8) { |
152 | __ SmiUntag(RCX); |
153 | } |
154 | switch (element_size_) { |
155 | case 1: |
156 | __ rep_movsb(); |
157 | break; |
158 | case 2: |
159 | __ rep_movsw(); |
160 | break; |
161 | case 4: |
162 | __ rep_movsl(); |
163 | break; |
164 | case 8: |
165 | case 16: |
166 | __ rep_movsq(); |
167 | break; |
168 | } |
169 | } |
170 | |
171 | void MemoryCopyInstr::EmitComputeStartPointer(FlowGraphCompiler* compiler, |
172 | classid_t array_cid, |
173 | Value* start, |
174 | Register array_reg, |
175 | Register start_reg) { |
176 | intptr_t offset; |
177 | if (IsTypedDataBaseClassId(array_cid)) { |
178 | __ movq( |
179 | array_reg, |
180 | compiler::FieldAddress( |
181 | array_reg, compiler::target::TypedDataBase::data_field_offset())); |
182 | offset = 0; |
183 | } else { |
184 | switch (array_cid) { |
185 | case kOneByteStringCid: |
186 | offset = |
187 | compiler::target::OneByteString::data_offset() - kHeapObjectTag; |
188 | break; |
189 | case kTwoByteStringCid: |
190 | offset = |
191 | compiler::target::TwoByteString::data_offset() - kHeapObjectTag; |
192 | break; |
193 | case kExternalOneByteStringCid: |
194 | __ movq(array_reg, |
195 | compiler::FieldAddress(array_reg, |
196 | compiler::target::ExternalOneByteString:: |
197 | external_data_offset())); |
198 | offset = 0; |
199 | break; |
200 | case kExternalTwoByteStringCid: |
201 | __ movq(array_reg, |
202 | compiler::FieldAddress(array_reg, |
203 | compiler::target::ExternalTwoByteString:: |
204 | external_data_offset())); |
205 | offset = 0; |
206 | break; |
207 | default: |
208 | UNREACHABLE(); |
209 | break; |
210 | } |
211 | } |
212 | ScaleFactor scale; |
213 | switch (element_size_) { |
214 | case 1: |
215 | __ SmiUntag(start_reg); |
216 | scale = TIMES_1; |
217 | break; |
218 | case 2: |
219 | scale = TIMES_1; |
220 | break; |
221 | case 4: |
222 | scale = TIMES_2; |
223 | break; |
224 | case 8: |
225 | scale = TIMES_4; |
226 | break; |
227 | case 16: |
228 | scale = TIMES_8; |
229 | break; |
230 | default: |
231 | UNREACHABLE(); |
232 | break; |
233 | } |
234 | __ leaq(array_reg, compiler::Address(array_reg, start_reg, scale, offset)); |
235 | } |
236 | |
237 | LocationSummary* PushArgumentInstr::MakeLocationSummary(Zone* zone, |
238 | bool opt) const { |
239 | const intptr_t kNumInputs = 1; |
240 | const intptr_t kNumTemps = 0; |
241 | LocationSummary* locs = new (zone) |
242 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
243 | if (representation() == kUnboxedDouble) { |
244 | locs->set_in(0, Location::RequiresFpuRegister()); |
245 | } else if (representation() == kUnboxedInt64) { |
246 | locs->set_in(0, Location::RequiresRegister()); |
247 | } else { |
248 | locs->set_in(0, LocationAnyOrConstant(value())); |
249 | } |
250 | return locs; |
251 | } |
252 | |
253 | void PushArgumentInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
254 | // In SSA mode, we need an explicit push. Nothing to do in non-SSA mode |
255 | // where arguments are pushed by their definitions. |
256 | if (compiler->is_optimizing()) { |
257 | Location value = locs()->in(0); |
258 | if (value.IsRegister()) { |
259 | __ pushq(value.reg()); |
260 | } else if (value.IsConstant()) { |
261 | __ PushObject(value.constant()); |
262 | } else if (value.IsFpuRegister()) { |
263 | __ AddImmediate(RSP, compiler::Immediate(-kDoubleSize)); |
264 | __ movsd(compiler::Address(RSP, 0), value.fpu_reg()); |
265 | } else { |
266 | ASSERT(value.IsStackSlot()); |
267 | __ pushq(LocationToStackSlotAddress(value)); |
268 | } |
269 | } |
270 | } |
271 | |
272 | LocationSummary* ReturnInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
273 | const intptr_t kNumInputs = 1; |
274 | const intptr_t kNumTemps = 0; |
275 | LocationSummary* locs = new (zone) |
276 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
277 | switch (representation()) { |
278 | case kTagged: |
279 | case kUnboxedInt64: |
280 | locs->set_in(0, |
281 | Location::RegisterLocation(CallingConventions::kReturnReg)); |
282 | break; |
283 | case kUnboxedDouble: |
284 | locs->set_in( |
285 | 0, Location::FpuRegisterLocation(CallingConventions::kReturnFpuReg)); |
286 | break; |
287 | default: |
288 | UNREACHABLE(); |
289 | break; |
290 | } |
291 | return locs; |
292 | } |
293 | |
294 | // Attempt optimized compilation at return instruction instead of at the entry. |
295 | // The entry needs to be patchable, no inlined objects are allowed in the area |
296 | // that will be overwritten by the patch instruction: a jump). |
297 | void ReturnInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
298 | if (locs()->in(0).IsRegister()) { |
299 | const Register result = locs()->in(0).reg(); |
300 | ASSERT(result == CallingConventions::kReturnReg); |
301 | } else { |
302 | ASSERT(locs()->in(0).IsFpuRegister()); |
303 | const FpuRegister result = locs()->in(0).fpu_reg(); |
304 | ASSERT(result == CallingConventions::kReturnFpuReg); |
305 | } |
306 | |
307 | if (compiler->intrinsic_mode()) { |
308 | // Intrinsics don't have a frame. |
309 | __ ret(); |
310 | return; |
311 | } |
312 | |
313 | #if defined(DEBUG) |
314 | __ Comment("Stack Check" ); |
315 | compiler::Label done; |
316 | const intptr_t fp_sp_dist = |
317 | (compiler::target::frame_layout.first_local_from_fp + 1 - |
318 | compiler->StackSize()) * |
319 | kWordSize; |
320 | ASSERT(fp_sp_dist <= 0); |
321 | __ movq(RDI, RSP); |
322 | __ subq(RDI, RBP); |
323 | __ CompareImmediate(RDI, compiler::Immediate(fp_sp_dist)); |
324 | __ j(EQUAL, &done, compiler::Assembler::kNearJump); |
325 | __ int3(); |
326 | __ Bind(&done); |
327 | #endif |
328 | ASSERT(__ constant_pool_allowed()); |
329 | if (yield_index() != PcDescriptorsLayout::kInvalidYieldIndex) { |
330 | compiler->EmitYieldPositionMetadata(token_pos(), yield_index()); |
331 | } |
332 | __ LeaveDartFrame(); // Disallows constant pool use. |
333 | __ ret(); |
334 | // This ReturnInstr may be emitted out of order by the optimizer. The next |
335 | // block may be a target expecting a properly set constant pool pointer. |
336 | __ set_constant_pool_allowed(true); |
337 | } |
338 | |
339 | void NativeReturnInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
340 | EmitReturnMoves(compiler); |
341 | |
342 | __ LeaveDartFrame(); |
343 | |
344 | // Pop dummy return address. |
345 | __ popq(TMP); |
346 | |
347 | // Anything besides the return register. |
348 | const Register vm_tag_reg = RBX; |
349 | const Register old_exit_frame_reg = RCX; |
350 | const Register old_exit_through_ffi_reg = RDI; |
351 | |
352 | __ popq(old_exit_frame_reg); |
353 | |
354 | __ popq(old_exit_through_ffi_reg); |
355 | |
356 | // Restore top_resource. |
357 | __ popq(TMP); |
358 | __ movq( |
359 | compiler::Address(THR, compiler::target::Thread::top_resource_offset()), |
360 | TMP); |
361 | |
362 | __ popq(vm_tag_reg); |
363 | |
364 | // If we were called by a trampoline, it will enter the safepoint on our |
365 | // behalf. |
366 | __ TransitionGeneratedToNative( |
367 | vm_tag_reg, old_exit_frame_reg, old_exit_through_ffi_reg, |
368 | /*enter_safepoint=*/!NativeCallbackTrampolines::Enabled()); |
369 | |
370 | // Restore C++ ABI callee-saved registers. |
371 | __ PopRegisters(CallingConventions::kCalleeSaveCpuRegisters, |
372 | CallingConventions::kCalleeSaveXmmRegisters); |
373 | |
374 | #if defined(TARGET_OS_FUCHSIA) |
375 | UNREACHABLE(); // Fuchsia does not allow dart:ffi. |
376 | #elif defined(USING_SHADOW_CALL_STACK) |
377 | #error Unimplemented |
378 | #endif |
379 | |
380 | // Leave the entry frame. |
381 | __ LeaveFrame(); |
382 | |
383 | // Leave the dummy frame holding the pushed arguments. |
384 | __ LeaveFrame(); |
385 | |
386 | __ ret(); |
387 | |
388 | // For following blocks. |
389 | __ set_constant_pool_allowed(true); |
390 | } |
391 | |
392 | // Detect pattern when one value is zero and another is a power of 2. |
393 | static bool IsPowerOfTwoKind(intptr_t v1, intptr_t v2) { |
394 | return (Utils::IsPowerOfTwo(v1) && (v2 == 0)) || |
395 | (Utils::IsPowerOfTwo(v2) && (v1 == 0)); |
396 | } |
397 | |
398 | LocationSummary* IfThenElseInstr::MakeLocationSummary(Zone* zone, |
399 | bool opt) const { |
400 | comparison()->InitializeLocationSummary(zone, opt); |
401 | // TODO(dartbug.com/30952) support convertion of Register to corresponding |
402 | // least significant byte register (e.g. RAX -> AL, RSI -> SIL, r15 -> r15b). |
403 | comparison()->locs()->set_out(0, Location::RegisterLocation(RDX)); |
404 | return comparison()->locs(); |
405 | } |
406 | |
407 | void IfThenElseInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
408 | ASSERT(locs()->out(0).reg() == RDX); |
409 | |
410 | // Clear upper part of the out register. We are going to use setcc on it |
411 | // which is a byte move. |
412 | __ xorq(RDX, RDX); |
413 | |
414 | // Emit comparison code. This must not overwrite the result register. |
415 | // IfThenElseInstr::Supports() should prevent EmitComparisonCode from using |
416 | // the labels or returning an invalid condition. |
417 | BranchLabels labels = {NULL, NULL, NULL}; |
418 | Condition true_condition = comparison()->EmitComparisonCode(compiler, labels); |
419 | ASSERT(true_condition != kInvalidCondition); |
420 | |
421 | const bool is_power_of_two_kind = IsPowerOfTwoKind(if_true_, if_false_); |
422 | |
423 | intptr_t true_value = if_true_; |
424 | intptr_t false_value = if_false_; |
425 | |
426 | if (is_power_of_two_kind) { |
427 | if (true_value == 0) { |
428 | // We need to have zero in RDX on true_condition. |
429 | true_condition = InvertCondition(true_condition); |
430 | } |
431 | } else { |
432 | if (true_value == 0) { |
433 | // Swap values so that false_value is zero. |
434 | intptr_t temp = true_value; |
435 | true_value = false_value; |
436 | false_value = temp; |
437 | } else { |
438 | true_condition = InvertCondition(true_condition); |
439 | } |
440 | } |
441 | |
442 | __ setcc(true_condition, DL); |
443 | |
444 | if (is_power_of_two_kind) { |
445 | const intptr_t shift = |
446 | Utils::ShiftForPowerOfTwo(Utils::Maximum(true_value, false_value)); |
447 | __ shlq(RDX, compiler::Immediate(shift + kSmiTagSize)); |
448 | } else { |
449 | __ decq(RDX); |
450 | __ AndImmediate(RDX, compiler::Immediate(Smi::RawValue(true_value) - |
451 | Smi::RawValue(false_value))); |
452 | if (false_value != 0) { |
453 | __ AddImmediate(RDX, compiler::Immediate(Smi::RawValue(false_value))); |
454 | } |
455 | } |
456 | } |
457 | |
458 | LocationSummary* LoadLocalInstr::MakeLocationSummary(Zone* zone, |
459 | bool opt) const { |
460 | const intptr_t kNumInputs = 0; |
461 | const intptr_t stack_index = |
462 | compiler::target::frame_layout.FrameSlotForVariable(&local()); |
463 | return LocationSummary::Make(zone, kNumInputs, |
464 | Location::StackSlot(stack_index, FPREG), |
465 | LocationSummary::kNoCall); |
466 | } |
467 | |
468 | void LoadLocalInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
469 | ASSERT(!compiler->is_optimizing()); |
470 | // Nothing to do. |
471 | } |
472 | |
473 | LocationSummary* StoreLocalInstr::MakeLocationSummary(Zone* zone, |
474 | bool opt) const { |
475 | const intptr_t kNumInputs = 1; |
476 | return LocationSummary::Make(zone, kNumInputs, Location::SameAsFirstInput(), |
477 | LocationSummary::kNoCall); |
478 | } |
479 | |
480 | void StoreLocalInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
481 | Register value = locs()->in(0).reg(); |
482 | Register result = locs()->out(0).reg(); |
483 | ASSERT(result == value); // Assert that register assignment is correct. |
484 | __ movq(compiler::Address( |
485 | RBP, compiler::target::FrameOffsetInBytesForVariable(&local())), |
486 | value); |
487 | } |
488 | |
489 | LocationSummary* ConstantInstr::MakeLocationSummary(Zone* zone, |
490 | bool opt) const { |
491 | const intptr_t kNumInputs = 0; |
492 | return LocationSummary::Make(zone, kNumInputs, |
493 | compiler::Assembler::IsSafe(value()) |
494 | ? Location::Constant(this) |
495 | : Location::RequiresRegister(), |
496 | LocationSummary::kNoCall); |
497 | } |
498 | |
499 | void ConstantInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
500 | // The register allocator drops constant definitions that have no uses. |
501 | Location out = locs()->out(0); |
502 | ASSERT(out.IsRegister() || out.IsConstant() || out.IsInvalid()); |
503 | if (out.IsRegister()) { |
504 | Register result = out.reg(); |
505 | __ LoadObject(result, value()); |
506 | } |
507 | } |
508 | |
509 | void ConstantInstr::EmitMoveToLocation(FlowGraphCompiler* compiler, |
510 | const Location& destination, |
511 | Register tmp) { |
512 | if (destination.IsRegister()) { |
513 | if (representation() == kUnboxedInt32 || |
514 | representation() == kUnboxedInt64) { |
515 | const int64_t value = Integer::Cast(value_).AsInt64Value(); |
516 | if (value == 0) { |
517 | __ xorl(destination.reg(), destination.reg()); |
518 | } else { |
519 | __ movq(destination.reg(), compiler::Immediate(value)); |
520 | } |
521 | } else { |
522 | ASSERT(representation() == kTagged); |
523 | __ LoadObject(destination.reg(), value_); |
524 | } |
525 | } else if (destination.IsFpuRegister()) { |
526 | if (Utils::DoublesBitEqual(Double::Cast(value_).value(), 0.0)) { |
527 | __ xorps(destination.fpu_reg(), destination.fpu_reg()); |
528 | } else { |
529 | ASSERT(tmp != kNoRegister); |
530 | __ LoadObject(tmp, value_); |
531 | __ movsd(destination.fpu_reg(), |
532 | compiler::FieldAddress(tmp, Double::value_offset())); |
533 | } |
534 | } else if (destination.IsDoubleStackSlot()) { |
535 | if (Utils::DoublesBitEqual(Double::Cast(value_).value(), 0.0)) { |
536 | __ xorps(FpuTMP, FpuTMP); |
537 | } else { |
538 | ASSERT(tmp != kNoRegister); |
539 | __ LoadObject(tmp, value_); |
540 | __ movsd(FpuTMP, compiler::FieldAddress(tmp, Double::value_offset())); |
541 | } |
542 | __ movsd(LocationToStackSlotAddress(destination), FpuTMP); |
543 | } else { |
544 | ASSERT(destination.IsStackSlot()); |
545 | if (representation() == kUnboxedInt32 || |
546 | representation() == kUnboxedInt64) { |
547 | const int64_t value = Integer::Cast(value_).AsInt64Value(); |
548 | __ movq(LocationToStackSlotAddress(destination), |
549 | compiler::Immediate(value)); |
550 | } else { |
551 | ASSERT(representation() == kTagged); |
552 | __ StoreObject(LocationToStackSlotAddress(destination), value_); |
553 | } |
554 | } |
555 | } |
556 | |
557 | LocationSummary* UnboxedConstantInstr::MakeLocationSummary(Zone* zone, |
558 | bool opt) const { |
559 | const intptr_t kNumInputs = 0; |
560 | const intptr_t kNumTemps = IsUnboxedSignedIntegerConstant() ? 0 : 1; |
561 | LocationSummary* locs = new (zone) |
562 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
563 | switch (representation()) { |
564 | case kUnboxedDouble: |
565 | locs->set_out(0, Location::RequiresFpuRegister()); |
566 | locs->set_temp(0, Location::RequiresRegister()); |
567 | break; |
568 | case kUnboxedInt32: |
569 | case kUnboxedInt64: |
570 | locs->set_out(0, Location::RequiresRegister()); |
571 | break; |
572 | default: |
573 | UNREACHABLE(); |
574 | break; |
575 | } |
576 | return locs; |
577 | } |
578 | |
579 | void UnboxedConstantInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
580 | // The register allocator drops constant definitions that have no uses. |
581 | if (!locs()->out(0).IsInvalid()) { |
582 | const Register scratch = |
583 | IsUnboxedSignedIntegerConstant() ? kNoRegister : locs()->temp(0).reg(); |
584 | EmitMoveToLocation(compiler, locs()->out(0), scratch); |
585 | } |
586 | } |
587 | |
588 | LocationSummary* AssertAssignableInstr::MakeLocationSummary(Zone* zone, |
589 | bool opt) const { |
590 | auto const dst_type_loc = |
591 | LocationFixedRegisterOrConstant(dst_type(), TypeTestABI::kDstTypeReg); |
592 | |
593 | // When using a type testing stub, we want to prevent spilling of the |
594 | // function/instantiator type argument vectors, since stub preserves them. So |
595 | // we make this a `kNoCall` summary, even though most other registers can be |
596 | // modified by the stub. To tell the register allocator about it, we reserve |
597 | // all the other registers as temporary registers. |
598 | // TODO(http://dartbug.com/32788): Simplify this. |
599 | const bool using_stub = dst_type_loc.IsConstant() && |
600 | FlowGraphCompiler::ShouldUseTypeTestingStubFor( |
601 | opt, AbstractType::Cast(dst_type_loc.constant())); |
602 | |
603 | const intptr_t kNonChangeableInputRegs = |
604 | (1 << TypeTestABI::kInstanceReg) | |
605 | ((dst_type_loc.IsRegister() ? 1 : 0) << TypeTestABI::kDstTypeReg) | |
606 | (1 << TypeTestABI::kInstantiatorTypeArgumentsReg) | |
607 | (1 << TypeTestABI::kFunctionTypeArgumentsReg); |
608 | |
609 | const intptr_t kNumInputs = 4; |
610 | |
611 | // We invoke a stub that can potentially clobber any CPU register |
612 | // but can only clobber FPU registers on the slow path when |
613 | // entering runtime. Preserve all FPU registers that are |
614 | // not guarateed to be preserved by the ABI. |
615 | const intptr_t kCpuRegistersToPreserve = |
616 | kDartAvailableCpuRegs & ~kNonChangeableInputRegs; |
617 | const intptr_t kFpuRegistersToPreserve = |
618 | CallingConventions::kVolatileXmmRegisters & ~(1 << FpuTMP); |
619 | |
620 | const intptr_t kNumTemps = |
621 | using_stub ? (Utils::CountOneBits64(kCpuRegistersToPreserve) + |
622 | Utils::CountOneBits64(kFpuRegistersToPreserve)) |
623 | : 0; |
624 | |
625 | LocationSummary* summary = new (zone) LocationSummary( |
626 | zone, kNumInputs, kNumTemps, |
627 | using_stub ? LocationSummary::kCallCalleeSafe : LocationSummary::kCall); |
628 | summary->set_in(0, Location::RegisterLocation(TypeTestABI::kInstanceReg)); |
629 | summary->set_in(1, dst_type_loc); |
630 | summary->set_in(2, Location::RegisterLocation( |
631 | TypeTestABI::kInstantiatorTypeArgumentsReg)); |
632 | summary->set_in( |
633 | 3, Location::RegisterLocation(TypeTestABI::kFunctionTypeArgumentsReg)); |
634 | summary->set_out(0, Location::SameAsFirstInput()); |
635 | |
636 | if (using_stub) { |
637 | // Let's reserve all registers except for the input ones. |
638 | intptr_t next_temp = 0; |
639 | for (intptr_t i = 0; i < kNumberOfCpuRegisters; ++i) { |
640 | const bool should_preserve = ((1 << i) & kCpuRegistersToPreserve) != 0; |
641 | if (should_preserve) { |
642 | summary->set_temp(next_temp++, |
643 | Location::RegisterLocation(static_cast<Register>(i))); |
644 | } |
645 | } |
646 | |
647 | for (intptr_t i = 0; i < kNumberOfFpuRegisters; i++) { |
648 | const bool should_preserve = ((1 << i) & kFpuRegistersToPreserve) != 0; |
649 | if (should_preserve) { |
650 | summary->set_temp(next_temp++, Location::FpuRegisterLocation( |
651 | static_cast<FpuRegister>(i))); |
652 | } |
653 | } |
654 | } |
655 | |
656 | return summary; |
657 | } |
658 | |
659 | static Condition TokenKindToIntCondition(Token::Kind kind) { |
660 | switch (kind) { |
661 | case Token::kEQ: |
662 | return EQUAL; |
663 | case Token::kNE: |
664 | return NOT_EQUAL; |
665 | case Token::kLT: |
666 | return LESS; |
667 | case Token::kGT: |
668 | return GREATER; |
669 | case Token::kLTE: |
670 | return LESS_EQUAL; |
671 | case Token::kGTE: |
672 | return GREATER_EQUAL; |
673 | default: |
674 | UNREACHABLE(); |
675 | return OVERFLOW; |
676 | } |
677 | } |
678 | |
679 | LocationSummary* EqualityCompareInstr::MakeLocationSummary(Zone* zone, |
680 | bool opt) const { |
681 | const intptr_t kNumInputs = 2; |
682 | if (operation_cid() == kDoubleCid) { |
683 | const intptr_t kNumTemps = 0; |
684 | LocationSummary* locs = new (zone) |
685 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
686 | locs->set_in(0, Location::RequiresFpuRegister()); |
687 | locs->set_in(1, Location::RequiresFpuRegister()); |
688 | locs->set_out(0, Location::RequiresRegister()); |
689 | return locs; |
690 | } |
691 | if (operation_cid() == kSmiCid || operation_cid() == kMintCid) { |
692 | const intptr_t kNumTemps = 0; |
693 | LocationSummary* locs = new (zone) |
694 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
695 | locs->set_in(0, LocationRegisterOrConstant(left())); |
696 | // Only one input can be a constant operand. The case of two constant |
697 | // operands should be handled by constant propagation. |
698 | // Only right can be a stack slot. |
699 | locs->set_in(1, locs->in(0).IsConstant() |
700 | ? Location::RequiresRegister() |
701 | : LocationRegisterOrConstant(right())); |
702 | locs->set_out(0, Location::RequiresRegister()); |
703 | return locs; |
704 | } |
705 | UNREACHABLE(); |
706 | return NULL; |
707 | } |
708 | |
709 | static void LoadValueCid(FlowGraphCompiler* compiler, |
710 | Register value_cid_reg, |
711 | Register value_reg, |
712 | compiler::Label* value_is_smi = NULL) { |
713 | compiler::Label done; |
714 | if (value_is_smi == NULL) { |
715 | __ LoadImmediate(value_cid_reg, compiler::Immediate(kSmiCid)); |
716 | } |
717 | __ testq(value_reg, compiler::Immediate(kSmiTagMask)); |
718 | if (value_is_smi == NULL) { |
719 | __ j(ZERO, &done, compiler::Assembler::kNearJump); |
720 | } else { |
721 | __ j(ZERO, value_is_smi); |
722 | } |
723 | __ LoadClassId(value_cid_reg, value_reg); |
724 | __ Bind(&done); |
725 | } |
726 | |
727 | static Condition FlipCondition(Condition condition) { |
728 | switch (condition) { |
729 | case EQUAL: |
730 | return EQUAL; |
731 | case NOT_EQUAL: |
732 | return NOT_EQUAL; |
733 | case LESS: |
734 | return GREATER; |
735 | case LESS_EQUAL: |
736 | return GREATER_EQUAL; |
737 | case GREATER: |
738 | return LESS; |
739 | case GREATER_EQUAL: |
740 | return LESS_EQUAL; |
741 | case BELOW: |
742 | return ABOVE; |
743 | case BELOW_EQUAL: |
744 | return ABOVE_EQUAL; |
745 | case ABOVE: |
746 | return BELOW; |
747 | case ABOVE_EQUAL: |
748 | return BELOW_EQUAL; |
749 | default: |
750 | UNIMPLEMENTED(); |
751 | return EQUAL; |
752 | } |
753 | } |
754 | |
755 | static void EmitBranchOnCondition(FlowGraphCompiler* compiler, |
756 | Condition true_condition, |
757 | BranchLabels labels) { |
758 | if (labels.fall_through == labels.false_label) { |
759 | // If the next block is the false successor, fall through to it. |
760 | __ j(true_condition, labels.true_label); |
761 | } else { |
762 | // If the next block is not the false successor, branch to it. |
763 | Condition false_condition = InvertCondition(true_condition); |
764 | __ j(false_condition, labels.false_label); |
765 | |
766 | // Fall through or jump to the true successor. |
767 | if (labels.fall_through != labels.true_label) { |
768 | __ jmp(labels.true_label); |
769 | } |
770 | } |
771 | } |
772 | |
773 | static Condition EmitInt64ComparisonOp(FlowGraphCompiler* compiler, |
774 | const LocationSummary& locs, |
775 | Token::Kind kind) { |
776 | Location left = locs.in(0); |
777 | Location right = locs.in(1); |
778 | ASSERT(!left.IsConstant() || !right.IsConstant()); |
779 | |
780 | Condition true_condition = TokenKindToIntCondition(kind); |
781 | if (left.IsConstant() || right.IsConstant()) { |
782 | // Ensure constant is on the right. |
783 | ConstantInstr* constant = NULL; |
784 | if (left.IsConstant()) { |
785 | constant = left.constant_instruction(); |
786 | Location tmp = right; |
787 | right = left; |
788 | left = tmp; |
789 | true_condition = FlipCondition(true_condition); |
790 | } else { |
791 | constant = right.constant_instruction(); |
792 | } |
793 | |
794 | if (constant->IsUnboxedSignedIntegerConstant()) { |
795 | __ cmpq(left.reg(), |
796 | compiler::Immediate( |
797 | constant->GetUnboxedSignedIntegerConstantValue())); |
798 | } else { |
799 | ASSERT(constant->representation() == kTagged); |
800 | __ CompareObject(left.reg(), right.constant()); |
801 | } |
802 | } else if (right.IsStackSlot()) { |
803 | __ cmpq(left.reg(), LocationToStackSlotAddress(right)); |
804 | } else { |
805 | __ cmpq(left.reg(), right.reg()); |
806 | } |
807 | return true_condition; |
808 | } |
809 | |
810 | static Condition TokenKindToDoubleCondition(Token::Kind kind) { |
811 | switch (kind) { |
812 | case Token::kEQ: |
813 | return EQUAL; |
814 | case Token::kNE: |
815 | return NOT_EQUAL; |
816 | case Token::kLT: |
817 | return BELOW; |
818 | case Token::kGT: |
819 | return ABOVE; |
820 | case Token::kLTE: |
821 | return BELOW_EQUAL; |
822 | case Token::kGTE: |
823 | return ABOVE_EQUAL; |
824 | default: |
825 | UNREACHABLE(); |
826 | return OVERFLOW; |
827 | } |
828 | } |
829 | |
830 | static Condition EmitDoubleComparisonOp(FlowGraphCompiler* compiler, |
831 | const LocationSummary& locs, |
832 | Token::Kind kind, |
833 | BranchLabels labels) { |
834 | XmmRegister left = locs.in(0).fpu_reg(); |
835 | XmmRegister right = locs.in(1).fpu_reg(); |
836 | |
837 | __ comisd(left, right); |
838 | |
839 | Condition true_condition = TokenKindToDoubleCondition(kind); |
840 | compiler::Label* nan_result = |
841 | (true_condition == NOT_EQUAL) ? labels.true_label : labels.false_label; |
842 | __ j(PARITY_EVEN, nan_result); |
843 | return true_condition; |
844 | } |
845 | |
846 | Condition EqualityCompareInstr::EmitComparisonCode(FlowGraphCompiler* compiler, |
847 | BranchLabels labels) { |
848 | if ((operation_cid() == kSmiCid) || (operation_cid() == kMintCid)) { |
849 | return EmitInt64ComparisonOp(compiler, *locs(), kind()); |
850 | } else { |
851 | ASSERT(operation_cid() == kDoubleCid); |
852 | return EmitDoubleComparisonOp(compiler, *locs(), kind(), labels); |
853 | } |
854 | } |
855 | |
856 | void ComparisonInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
857 | compiler::Label is_true, is_false; |
858 | BranchLabels labels = {&is_true, &is_false, &is_false}; |
859 | Condition true_condition = EmitComparisonCode(compiler, labels); |
860 | if (true_condition != kInvalidCondition) { |
861 | EmitBranchOnCondition(compiler, true_condition, labels); |
862 | } |
863 | |
864 | Register result = locs()->out(0).reg(); |
865 | compiler::Label done; |
866 | __ Bind(&is_false); |
867 | __ LoadObject(result, Bool::False()); |
868 | __ jmp(&done); |
869 | __ Bind(&is_true); |
870 | __ LoadObject(result, Bool::True()); |
871 | __ Bind(&done); |
872 | } |
873 | |
874 | void ComparisonInstr::EmitBranchCode(FlowGraphCompiler* compiler, |
875 | BranchInstr* branch) { |
876 | BranchLabels labels = compiler->CreateBranchLabels(branch); |
877 | Condition true_condition = EmitComparisonCode(compiler, labels); |
878 | if (true_condition != kInvalidCondition) { |
879 | EmitBranchOnCondition(compiler, true_condition, labels); |
880 | } |
881 | } |
882 | |
883 | LocationSummary* TestSmiInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
884 | const intptr_t kNumInputs = 2; |
885 | const intptr_t kNumTemps = 0; |
886 | LocationSummary* locs = new (zone) |
887 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
888 | locs->set_in(0, Location::RequiresRegister()); |
889 | // Only one input can be a constant operand. The case of two constant |
890 | // operands should be handled by constant propagation. |
891 | locs->set_in(1, LocationRegisterOrConstant(right())); |
892 | return locs; |
893 | } |
894 | |
895 | Condition TestSmiInstr::EmitComparisonCode(FlowGraphCompiler* compiler, |
896 | BranchLabels labels) { |
897 | Register left_reg = locs()->in(0).reg(); |
898 | Location right = locs()->in(1); |
899 | if (right.IsConstant()) { |
900 | ASSERT(right.constant().IsSmi()); |
901 | const int64_t imm = static_cast<int64_t>(right.constant().raw()); |
902 | __ TestImmediate(left_reg, compiler::Immediate(imm)); |
903 | } else { |
904 | __ testq(left_reg, right.reg()); |
905 | } |
906 | Condition true_condition = (kind() == Token::kNE) ? NOT_ZERO : ZERO; |
907 | return true_condition; |
908 | } |
909 | |
910 | LocationSummary* TestCidsInstr::MakeLocationSummary(Zone* zone, |
911 | bool opt) const { |
912 | const intptr_t kNumInputs = 1; |
913 | const intptr_t kNumTemps = 1; |
914 | LocationSummary* locs = new (zone) |
915 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
916 | locs->set_in(0, Location::RequiresRegister()); |
917 | locs->set_temp(0, Location::RequiresRegister()); |
918 | locs->set_out(0, Location::RequiresRegister()); |
919 | return locs; |
920 | } |
921 | |
922 | Condition TestCidsInstr::EmitComparisonCode(FlowGraphCompiler* compiler, |
923 | BranchLabels labels) { |
924 | ASSERT((kind() == Token::kIS) || (kind() == Token::kISNOT)); |
925 | Register val_reg = locs()->in(0).reg(); |
926 | Register cid_reg = locs()->temp(0).reg(); |
927 | |
928 | compiler::Label* deopt = |
929 | CanDeoptimize() |
930 | ? compiler->AddDeoptStub(deopt_id(), ICData::kDeoptTestCids, |
931 | licm_hoisted_ ? ICData::kHoisted : 0) |
932 | : NULL; |
933 | |
934 | const intptr_t true_result = (kind() == Token::kIS) ? 1 : 0; |
935 | const ZoneGrowableArray<intptr_t>& data = cid_results(); |
936 | ASSERT(data[0] == kSmiCid); |
937 | bool result = data[1] == true_result; |
938 | __ testq(val_reg, compiler::Immediate(kSmiTagMask)); |
939 | __ j(ZERO, result ? labels.true_label : labels.false_label); |
940 | __ LoadClassId(cid_reg, val_reg); |
941 | for (intptr_t i = 2; i < data.length(); i += 2) { |
942 | const intptr_t test_cid = data[i]; |
943 | ASSERT(test_cid != kSmiCid); |
944 | result = data[i + 1] == true_result; |
945 | __ cmpq(cid_reg, compiler::Immediate(test_cid)); |
946 | __ j(EQUAL, result ? labels.true_label : labels.false_label); |
947 | } |
948 | // No match found, deoptimize or default action. |
949 | if (deopt == NULL) { |
950 | // If the cid is not in the list, jump to the opposite label from the cids |
951 | // that are in the list. These must be all the same (see asserts in the |
952 | // constructor). |
953 | compiler::Label* target = result ? labels.false_label : labels.true_label; |
954 | if (target != labels.fall_through) { |
955 | __ jmp(target); |
956 | } |
957 | } else { |
958 | __ jmp(deopt); |
959 | } |
960 | // Dummy result as this method already did the jump, there's no need |
961 | // for the caller to branch on a condition. |
962 | return kInvalidCondition; |
963 | } |
964 | |
965 | LocationSummary* RelationalOpInstr::MakeLocationSummary(Zone* zone, |
966 | bool opt) const { |
967 | const intptr_t kNumInputs = 2; |
968 | const intptr_t kNumTemps = 0; |
969 | if (operation_cid() == kDoubleCid) { |
970 | LocationSummary* summary = new (zone) |
971 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
972 | summary->set_in(0, Location::RequiresFpuRegister()); |
973 | summary->set_in(1, Location::RequiresFpuRegister()); |
974 | summary->set_out(0, Location::RequiresRegister()); |
975 | return summary; |
976 | } |
977 | if (operation_cid() == kSmiCid || operation_cid() == kMintCid) { |
978 | LocationSummary* summary = new (zone) |
979 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
980 | summary->set_in(0, LocationRegisterOrConstant(left())); |
981 | // Only one input can be a constant operand. The case of two constant |
982 | // operands should be handled by constant propagation. |
983 | summary->set_in(1, summary->in(0).IsConstant() |
984 | ? Location::RequiresRegister() |
985 | : LocationRegisterOrConstant(right())); |
986 | summary->set_out(0, Location::RequiresRegister()); |
987 | return summary; |
988 | } |
989 | UNREACHABLE(); |
990 | return NULL; |
991 | } |
992 | |
993 | Condition RelationalOpInstr::EmitComparisonCode(FlowGraphCompiler* compiler, |
994 | BranchLabels labels) { |
995 | if (operation_cid() == kSmiCid || operation_cid() == kMintCid) { |
996 | return EmitInt64ComparisonOp(compiler, *locs(), kind()); |
997 | } else { |
998 | ASSERT(operation_cid() == kDoubleCid); |
999 | return EmitDoubleComparisonOp(compiler, *locs(), kind(), labels); |
1000 | } |
1001 | } |
1002 | |
1003 | void NativeCallInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1004 | SetupNative(); |
1005 | Register result = locs()->out(0).reg(); |
1006 | const intptr_t argc_tag = NativeArguments::ComputeArgcTag(function()); |
1007 | |
1008 | // All arguments are already @RSP due to preceding PushArgument()s. |
1009 | ASSERT(ArgumentCount() == |
1010 | function().NumParameters() + (function().IsGeneric() ? 1 : 0)); |
1011 | |
1012 | // Push the result place holder initialized to NULL. |
1013 | __ PushObject(Object::null_object()); |
1014 | |
1015 | // Pass a pointer to the first argument in RAX. |
1016 | __ leaq(RAX, compiler::Address(RSP, ArgumentCount() * kWordSize)); |
1017 | |
1018 | __ LoadImmediate(R10, compiler::Immediate(argc_tag)); |
1019 | const Code* stub; |
1020 | if (link_lazily()) { |
1021 | stub = &StubCode::CallBootstrapNative(); |
1022 | compiler::ExternalLabel label(NativeEntry::LinkNativeCallEntry()); |
1023 | __ LoadNativeEntry(RBX, &label, |
1024 | compiler::ObjectPoolBuilderEntry::kPatchable); |
1025 | compiler->GeneratePatchableCall(token_pos(), *stub, |
1026 | PcDescriptorsLayout::kOther, locs()); |
1027 | } else { |
1028 | if (is_bootstrap_native()) { |
1029 | stub = &StubCode::CallBootstrapNative(); |
1030 | } else if (is_auto_scope()) { |
1031 | stub = &StubCode::CallAutoScopeNative(); |
1032 | } else { |
1033 | stub = &StubCode::CallNoScopeNative(); |
1034 | } |
1035 | const compiler::ExternalLabel label( |
1036 | reinterpret_cast<uword>(native_c_function())); |
1037 | __ LoadNativeEntry(RBX, &label, |
1038 | compiler::ObjectPoolBuilderEntry::kNotPatchable); |
1039 | compiler->GenerateStubCall(token_pos(), *stub, PcDescriptorsLayout::kOther, |
1040 | locs()); |
1041 | } |
1042 | __ popq(result); |
1043 | |
1044 | __ Drop(ArgumentCount()); // Drop the arguments. |
1045 | } |
1046 | |
1047 | void FfiCallInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1048 | const Register saved_fp = locs()->temp(0).reg(); |
1049 | const Register target_address = locs()->in(TargetAddressIndex()).reg(); |
1050 | |
1051 | // Save frame pointer because we're going to update it when we enter the exit |
1052 | // frame. |
1053 | __ movq(saved_fp, FPREG); |
1054 | |
1055 | // Make a space to put the return address. |
1056 | __ pushq(compiler::Immediate(0)); |
1057 | |
1058 | // We need to create a dummy "exit frame". It will share the same pool pointer |
1059 | // but have a null code object. |
1060 | __ LoadObject(CODE_REG, Object::null_object()); |
1061 | __ set_constant_pool_allowed(false); |
1062 | __ EnterDartFrame(marshaller_.StackTopInBytes(), PP); |
1063 | |
1064 | // Align frame before entering C++ world. |
1065 | if (OS::ActivationFrameAlignment() > 1) { |
1066 | __ andq(SPREG, compiler::Immediate(~(OS::ActivationFrameAlignment() - 1))); |
1067 | } |
1068 | |
1069 | EmitParamMoves(compiler); |
1070 | |
1071 | // We need to copy a dummy return address up into the dummy stack frame so the |
1072 | // stack walker will know which safepoint to use. RIP points to the *next* |
1073 | // instruction, so 'AddressRIPRelative' loads the address of the following |
1074 | // 'movq'. |
1075 | __ leaq(TMP, compiler::Address::AddressRIPRelative(0)); |
1076 | compiler->EmitCallsiteMetadata(TokenPosition::kNoSource, deopt_id(), |
1077 | PcDescriptorsLayout::Kind::kOther, locs()); |
1078 | __ movq(compiler::Address(FPREG, kSavedCallerPcSlotFromFp * kWordSize), TMP); |
1079 | |
1080 | if (CanExecuteGeneratedCodeInSafepoint()) { |
1081 | // Update information in the thread object and enter a safepoint. |
1082 | __ movq(TMP, |
1083 | compiler::Immediate(compiler::target::Thread::exit_through_ffi())); |
1084 | __ TransitionGeneratedToNative(target_address, FPREG, TMP, |
1085 | /*enter_safepoint=*/true); |
1086 | |
1087 | __ CallCFunction(target_address); |
1088 | |
1089 | // Update information in the thread object and leave the safepoint. |
1090 | __ TransitionNativeToGenerated(/*leave_safepoint=*/true); |
1091 | } else { |
1092 | // We cannot trust that this code will be executable within a safepoint. |
1093 | // Therefore we delegate the responsibility of entering/exiting the |
1094 | // safepoint to a stub which in the VM isolate's heap, which will never lose |
1095 | // execute permission. |
1096 | __ movq(TMP, |
1097 | compiler::Address( |
1098 | THR, compiler::target::Thread:: |
1099 | call_native_through_safepoint_entry_point_offset())); |
1100 | |
1101 | // Calls RBX within a safepoint. |
1102 | ASSERT(saved_fp == RBX); |
1103 | __ movq(RBX, target_address); |
1104 | __ call(TMP); |
1105 | } |
1106 | |
1107 | EmitReturnMoves(compiler); |
1108 | |
1109 | // Although PP is a callee-saved register, it may have been moved by the GC. |
1110 | __ LeaveDartFrame(compiler::kRestoreCallerPP); |
1111 | |
1112 | // Restore the global object pool after returning from runtime (old space is |
1113 | // moving, so the GOP could have been relocated). |
1114 | if (FLAG_precompiled_mode && FLAG_use_bare_instructions) { |
1115 | __ movq(PP, compiler::Address(THR, Thread::global_object_pool_offset())); |
1116 | } |
1117 | |
1118 | __ set_constant_pool_allowed(true); |
1119 | |
1120 | // Instead of returning to the "fake" return address, we just pop it. |
1121 | __ popq(TMP); |
1122 | } |
1123 | |
1124 | void NativeEntryInstr::SaveArgument( |
1125 | FlowGraphCompiler* compiler, |
1126 | const compiler::ffi::NativeLocation& nloc) const { |
1127 | if (nloc.IsStack()) return; |
1128 | |
1129 | if (nloc.IsRegisters()) { |
1130 | const auto& regs_loc = nloc.AsRegisters(); |
1131 | ASSERT(regs_loc.num_regs() == 1); |
1132 | __ pushq(regs_loc.reg_at(0)); |
1133 | } else if (nloc.IsFpuRegisters()) { |
1134 | // TODO(dartbug.com/40469): Reduce code size. |
1135 | __ movq(TMP, nloc.AsFpuRegisters().fpu_reg()); |
1136 | __ pushq(TMP); |
1137 | } else { |
1138 | UNREACHABLE(); |
1139 | } |
1140 | } |
1141 | |
1142 | void NativeEntryInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1143 | __ Bind(compiler->GetJumpLabel(this)); |
1144 | |
1145 | // Create a dummy frame holding the pushed arguments. This simplifies |
1146 | // NativeReturnInstr::EmitNativeCode. |
1147 | __ EnterFrame(0); |
1148 | |
1149 | #if defined(TARGET_OS_FUCHSIA) |
1150 | UNREACHABLE(); // Fuchsia does not allow dart:ffi. |
1151 | #elif defined(USING_SHADOW_CALL_STACK) |
1152 | #error Unimplemented |
1153 | #endif |
1154 | |
1155 | // Save the argument registers, in reverse order. |
1156 | for (intptr_t i = marshaller_.num_args(); i-- > 0;) { |
1157 | SaveArgument(compiler, marshaller_.Location(i)); |
1158 | } |
1159 | |
1160 | // Enter the entry frame. Push a dummy return address for consistency with |
1161 | // EnterFrame on ARM(64). |
1162 | __ PushImmediate(compiler::Immediate(0)); |
1163 | __ EnterFrame(0); |
1164 | |
1165 | // Save a space for the code object. |
1166 | __ PushImmediate(compiler::Immediate(0)); |
1167 | |
1168 | // InvokeDartCodeStub saves the arguments descriptor here. We don't have one, |
1169 | // but we need to follow the same frame layout for the stack walker. |
1170 | __ PushImmediate(compiler::Immediate(0)); |
1171 | |
1172 | // Save ABI callee-saved registers. |
1173 | __ PushRegisters(CallingConventions::kCalleeSaveCpuRegisters, |
1174 | CallingConventions::kCalleeSaveXmmRegisters); |
1175 | |
1176 | // Load the address of DLRT_GetThreadForNativeCallback without using Thread. |
1177 | if (FLAG_precompiled_mode) { |
1178 | compiler->LoadBSSEntry(BSS::Relocation::DRT_GetThreadForNativeCallback, RAX, |
1179 | RCX); |
1180 | } else if (!NativeCallbackTrampolines::Enabled()) { |
1181 | // In JIT mode, we can just paste the address of the runtime entry into the |
1182 | // generated code directly. This is not a problem since we don't save |
1183 | // callbacks into JIT snapshots. |
1184 | __ movq(RAX, compiler::Immediate(reinterpret_cast<intptr_t>( |
1185 | DLRT_GetThreadForNativeCallback))); |
1186 | } |
1187 | |
1188 | // Create another frame to align the frame before continuing in "native" code. |
1189 | // If we were called by a trampoline, it has already loaded the thread. |
1190 | if (!NativeCallbackTrampolines::Enabled()) { |
1191 | __ EnterFrame(0); |
1192 | __ ReserveAlignedFrameSpace(0); |
1193 | |
1194 | COMPILE_ASSERT(RAX != CallingConventions::kArg1Reg); |
1195 | __ movq(CallingConventions::kArg1Reg, compiler::Immediate(callback_id_)); |
1196 | __ CallCFunction(RAX); |
1197 | __ movq(THR, RAX); |
1198 | |
1199 | __ LeaveFrame(); |
1200 | } |
1201 | |
1202 | // Save the current VMTag on the stack. |
1203 | __ movq(RAX, compiler::Assembler::VMTagAddress()); |
1204 | __ pushq(RAX); |
1205 | |
1206 | // Save top resource. |
1207 | __ pushq( |
1208 | compiler::Address(THR, compiler::target::Thread::top_resource_offset())); |
1209 | __ movq( |
1210 | compiler::Address(THR, compiler::target::Thread::top_resource_offset()), |
1211 | compiler::Immediate(0)); |
1212 | |
1213 | __ pushq(compiler::Address( |
1214 | THR, compiler::target::Thread::exit_through_ffi_offset())); |
1215 | |
1216 | // Save top exit frame info. Stack walker expects it to be here. |
1217 | __ pushq(compiler::Address( |
1218 | THR, compiler::target::Thread::top_exit_frame_info_offset())); |
1219 | |
1220 | // In debug mode, verify that we've pushed the top exit frame info at the |
1221 | // correct offset from FP. |
1222 | __ EmitEntryFrameVerification(); |
1223 | |
1224 | // Either DLRT_GetThreadForNativeCallback or the callback trampoline (caller) |
1225 | // will leave the safepoint for us. |
1226 | __ TransitionNativeToGenerated(/*exit_safepoint=*/false); |
1227 | |
1228 | // Load the code object. |
1229 | __ movq(RAX, compiler::Address( |
1230 | THR, compiler::target::Thread::callback_code_offset())); |
1231 | __ movq(RAX, compiler::FieldAddress( |
1232 | RAX, compiler::target::GrowableObjectArray::data_offset())); |
1233 | __ movq(CODE_REG, compiler::FieldAddress( |
1234 | RAX, compiler::target::Array::data_offset() + |
1235 | callback_id_ * compiler::target::kWordSize)); |
1236 | |
1237 | // Put the code object in the reserved slot. |
1238 | __ movq(compiler::Address(FPREG, |
1239 | kPcMarkerSlotFromFp * compiler::target::kWordSize), |
1240 | CODE_REG); |
1241 | |
1242 | if (FLAG_precompiled_mode && FLAG_use_bare_instructions) { |
1243 | __ movq(PP, |
1244 | compiler::Address( |
1245 | THR, compiler::target::Thread::global_object_pool_offset())); |
1246 | } else { |
1247 | __ xorq(PP, PP); // GC-safe value into PP. |
1248 | } |
1249 | |
1250 | // Load a GC-safe value for arguments descriptor (unused but tagged). |
1251 | __ xorq(ARGS_DESC_REG, ARGS_DESC_REG); |
1252 | |
1253 | // Push a dummy return address which suggests that we are inside of |
1254 | // InvokeDartCodeStub. This is how the stack walker detects an entry frame. |
1255 | __ movq(RAX, |
1256 | compiler::Address( |
1257 | THR, compiler::target::Thread::invoke_dart_code_stub_offset())); |
1258 | __ pushq(compiler::FieldAddress( |
1259 | RAX, compiler::target::Code::entry_point_offset())); |
1260 | |
1261 | // Continue with Dart frame setup. |
1262 | FunctionEntryInstr::EmitNativeCode(compiler); |
1263 | } |
1264 | |
1265 | static bool CanBeImmediateIndex(Value* index, intptr_t cid) { |
1266 | if (!index->definition()->IsConstant()) return false; |
1267 | const Object& constant = index->definition()->AsConstant()->value(); |
1268 | if (!constant.IsSmi()) return false; |
1269 | const Smi& smi_const = Smi::Cast(constant); |
1270 | const intptr_t scale = Instance::ElementSizeFor(cid); |
1271 | const intptr_t data_offset = Instance::DataOffsetFor(cid); |
1272 | const int64_t disp = smi_const.AsInt64Value() * scale + data_offset; |
1273 | return Utils::IsInt(32, disp); |
1274 | } |
1275 | |
1276 | LocationSummary* OneByteStringFromCharCodeInstr::MakeLocationSummary( |
1277 | Zone* zone, |
1278 | bool opt) const { |
1279 | const intptr_t kNumInputs = 1; |
1280 | // TODO(fschneider): Allow immediate operands for the char code. |
1281 | return LocationSummary::Make(zone, kNumInputs, Location::RequiresRegister(), |
1282 | LocationSummary::kNoCall); |
1283 | } |
1284 | |
1285 | void OneByteStringFromCharCodeInstr::EmitNativeCode( |
1286 | FlowGraphCompiler* compiler) { |
1287 | ASSERT(compiler->is_optimizing()); |
1288 | Register char_code = locs()->in(0).reg(); |
1289 | Register result = locs()->out(0).reg(); |
1290 | |
1291 | __ movq(result, |
1292 | compiler::Address(THR, Thread::predefined_symbols_address_offset())); |
1293 | __ movq(result, |
1294 | compiler::Address(result, char_code, |
1295 | TIMES_HALF_WORD_SIZE, // Char code is a smi. |
1296 | Symbols::kNullCharCodeSymbolOffset * kWordSize)); |
1297 | } |
1298 | |
1299 | LocationSummary* StringToCharCodeInstr::MakeLocationSummary(Zone* zone, |
1300 | bool opt) const { |
1301 | const intptr_t kNumInputs = 1; |
1302 | return LocationSummary::Make(zone, kNumInputs, Location::RequiresRegister(), |
1303 | LocationSummary::kNoCall); |
1304 | } |
1305 | |
1306 | void StringToCharCodeInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1307 | ASSERT(cid_ == kOneByteStringCid); |
1308 | Register str = locs()->in(0).reg(); |
1309 | Register result = locs()->out(0).reg(); |
1310 | compiler::Label is_one, done; |
1311 | __ movq(result, compiler::FieldAddress(str, String::length_offset())); |
1312 | __ cmpq(result, compiler::Immediate(Smi::RawValue(1))); |
1313 | __ j(EQUAL, &is_one, compiler::Assembler::kNearJump); |
1314 | __ movq(result, compiler::Immediate(Smi::RawValue(-1))); |
1315 | __ jmp(&done); |
1316 | __ Bind(&is_one); |
1317 | __ movzxb(result, compiler::FieldAddress(str, OneByteString::data_offset())); |
1318 | __ SmiTag(result); |
1319 | __ Bind(&done); |
1320 | } |
1321 | |
1322 | LocationSummary* StringInterpolateInstr::MakeLocationSummary(Zone* zone, |
1323 | bool opt) const { |
1324 | const intptr_t kNumInputs = 1; |
1325 | const intptr_t kNumTemps = 0; |
1326 | LocationSummary* summary = new (zone) |
1327 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
1328 | summary->set_in(0, Location::RegisterLocation(RAX)); |
1329 | summary->set_out(0, Location::RegisterLocation(RAX)); |
1330 | return summary; |
1331 | } |
1332 | |
1333 | void StringInterpolateInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1334 | Register array = locs()->in(0).reg(); |
1335 | __ pushq(array); |
1336 | const int kTypeArgsLen = 0; |
1337 | const int kNumberOfArguments = 1; |
1338 | constexpr int kSizeOfArguments = 1; |
1339 | const Array& kNoArgumentNames = Object::null_array(); |
1340 | ArgumentsInfo args_info(kTypeArgsLen, kNumberOfArguments, kSizeOfArguments, |
1341 | kNoArgumentNames); |
1342 | compiler->GenerateStaticCall(deopt_id(), token_pos(), CallFunction(), |
1343 | args_info, locs(), ICData::Handle(), |
1344 | ICData::kStatic); |
1345 | ASSERT(locs()->out(0).reg() == RAX); |
1346 | } |
1347 | |
1348 | LocationSummary* Utf8ScanInstr::MakeLocationSummary(Zone* zone, |
1349 | bool opt) const { |
1350 | const intptr_t kNumInputs = 5; |
1351 | const intptr_t kNumTemps = 1; |
1352 | LocationSummary* summary = new (zone) |
1353 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
1354 | summary->set_in(0, Location::Any()); // decoder |
1355 | summary->set_in(1, Location::WritableRegister()); // bytes |
1356 | summary->set_in(2, Location::WritableRegister()); // start |
1357 | summary->set_in(3, Location::WritableRegister()); // end |
1358 | summary->set_in(4, Location::RequiresRegister()); // table |
1359 | summary->set_temp(0, Location::RequiresRegister()); |
1360 | summary->set_out(0, Location::RequiresRegister()); |
1361 | return summary; |
1362 | } |
1363 | |
1364 | void Utf8ScanInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1365 | const Register bytes_reg = locs()->in(1).reg(); |
1366 | const Register start_reg = locs()->in(2).reg(); |
1367 | const Register end_reg = locs()->in(3).reg(); |
1368 | const Register table_reg = locs()->in(4).reg(); |
1369 | const Register size_reg = locs()->out(0).reg(); |
1370 | |
1371 | const Register bytes_ptr_reg = start_reg; |
1372 | const Register bytes_end_reg = end_reg; |
1373 | const Register bytes_end_minus_16_reg = bytes_reg; |
1374 | const Register flags_reg = locs()->temp(0).reg(); |
1375 | const Register temp_reg = TMP; |
1376 | const XmmRegister vector_reg = FpuTMP; |
1377 | |
1378 | static const intptr_t kSizeMask = 0x03; |
1379 | static const intptr_t kFlagsMask = 0x3C; |
1380 | |
1381 | compiler::Label scan_ascii, ascii_loop, ascii_loop_in, nonascii_loop; |
1382 | compiler::Label rest, rest_loop, rest_loop_in, done; |
1383 | |
1384 | // Address of input bytes. |
1385 | __ movq(bytes_reg, |
1386 | compiler::FieldAddress( |
1387 | bytes_reg, compiler::target::TypedDataBase::data_field_offset())); |
1388 | |
1389 | // Pointers to start, end and end-16. |
1390 | __ leaq(bytes_ptr_reg, compiler::Address(bytes_reg, start_reg, TIMES_1, 0)); |
1391 | __ leaq(bytes_end_reg, compiler::Address(bytes_reg, end_reg, TIMES_1, 0)); |
1392 | __ leaq(bytes_end_minus_16_reg, compiler::Address(bytes_end_reg, -16)); |
1393 | |
1394 | // Initialize size and flags. |
1395 | __ xorq(size_reg, size_reg); |
1396 | __ xorq(flags_reg, flags_reg); |
1397 | |
1398 | __ jmp(&scan_ascii, compiler::Assembler::kNearJump); |
1399 | |
1400 | // Loop scanning through ASCII bytes one 16-byte vector at a time. |
1401 | // While scanning, the size register contains the size as it was at the start |
1402 | // of the current block of ASCII bytes, minus the address of the start of the |
1403 | // block. After the block, the end address of the block is added to update the |
1404 | // size to include the bytes in the block. |
1405 | __ Bind(&ascii_loop); |
1406 | __ addq(bytes_ptr_reg, compiler::Immediate(16)); |
1407 | __ Bind(&ascii_loop_in); |
1408 | |
1409 | // Exit vectorized loop when there are less than 16 bytes left. |
1410 | __ cmpq(bytes_ptr_reg, bytes_end_minus_16_reg); |
1411 | __ j(UNSIGNED_GREATER, &rest, compiler::Assembler::kNearJump); |
1412 | |
1413 | // Find next non-ASCII byte within the next 16 bytes. |
1414 | // Note: In principle, we should use MOVDQU here, since the loaded value is |
1415 | // used as input to an integer instruction. In practice, according to Agner |
1416 | // Fog, there is no penalty for using the wrong kind of load. |
1417 | __ movups(vector_reg, compiler::Address(bytes_ptr_reg, 0)); |
1418 | __ pmovmskb(temp_reg, vector_reg); |
1419 | __ bsfq(temp_reg, temp_reg); |
1420 | __ j(EQUAL, &ascii_loop, compiler::Assembler::kNearJump); |
1421 | |
1422 | // Point to non-ASCII byte and update size. |
1423 | __ addq(bytes_ptr_reg, temp_reg); |
1424 | __ addq(size_reg, bytes_ptr_reg); |
1425 | |
1426 | // Read first non-ASCII byte. |
1427 | __ movzxb(temp_reg, compiler::Address(bytes_ptr_reg, 0)); |
1428 | |
1429 | // Loop over block of non-ASCII bytes. |
1430 | __ Bind(&nonascii_loop); |
1431 | __ addq(bytes_ptr_reg, compiler::Immediate(1)); |
1432 | |
1433 | // Update size and flags based on byte value. |
1434 | __ movzxb(temp_reg, compiler::FieldAddress( |
1435 | table_reg, temp_reg, TIMES_1, |
1436 | compiler::target::OneByteString::data_offset())); |
1437 | __ orq(flags_reg, temp_reg); |
1438 | __ andq(temp_reg, compiler::Immediate(kSizeMask)); |
1439 | __ addq(size_reg, temp_reg); |
1440 | |
1441 | // Stop if end is reached. |
1442 | __ cmpq(bytes_ptr_reg, bytes_end_reg); |
1443 | __ j(UNSIGNED_GREATER_EQUAL, &done, compiler::Assembler::kNearJump); |
1444 | |
1445 | // Go to ASCII scan if next byte is ASCII, otherwise loop. |
1446 | __ movzxb(temp_reg, compiler::Address(bytes_ptr_reg, 0)); |
1447 | __ testq(temp_reg, compiler::Immediate(0x80)); |
1448 | __ j(NOT_EQUAL, &nonascii_loop, compiler::Assembler::kNearJump); |
1449 | |
1450 | // Enter the ASCII scanning loop. |
1451 | __ Bind(&scan_ascii); |
1452 | __ subq(size_reg, bytes_ptr_reg); |
1453 | __ jmp(&ascii_loop_in); |
1454 | |
1455 | // Less than 16 bytes left. Process the remaining bytes individually. |
1456 | __ Bind(&rest); |
1457 | |
1458 | // Update size after ASCII scanning loop. |
1459 | __ addq(size_reg, bytes_ptr_reg); |
1460 | __ jmp(&rest_loop_in, compiler::Assembler::kNearJump); |
1461 | |
1462 | __ Bind(&rest_loop); |
1463 | |
1464 | // Read byte and increment pointer. |
1465 | __ movzxb(temp_reg, compiler::Address(bytes_ptr_reg, 0)); |
1466 | __ addq(bytes_ptr_reg, compiler::Immediate(1)); |
1467 | |
1468 | // Update size and flags based on byte value. |
1469 | __ movzxb(temp_reg, compiler::FieldAddress( |
1470 | table_reg, temp_reg, TIMES_1, |
1471 | compiler::target::OneByteString::data_offset())); |
1472 | __ orq(flags_reg, temp_reg); |
1473 | __ andq(temp_reg, compiler::Immediate(kSizeMask)); |
1474 | __ addq(size_reg, temp_reg); |
1475 | |
1476 | // Stop if end is reached. |
1477 | __ Bind(&rest_loop_in); |
1478 | __ cmpq(bytes_ptr_reg, bytes_end_reg); |
1479 | __ j(UNSIGNED_LESS, &rest_loop, compiler::Assembler::kNearJump); |
1480 | __ Bind(&done); |
1481 | |
1482 | // Write flags to field. |
1483 | __ andq(flags_reg, compiler::Immediate(kFlagsMask)); |
1484 | if (!IsScanFlagsUnboxed()) { |
1485 | __ SmiTag(flags_reg); |
1486 | } |
1487 | Register decoder_reg; |
1488 | const Location decoder_location = locs()->in(0); |
1489 | if (decoder_location.IsStackSlot()) { |
1490 | __ movq(temp_reg, LocationToStackSlotAddress(decoder_location)); |
1491 | decoder_reg = temp_reg; |
1492 | } else { |
1493 | decoder_reg = decoder_location.reg(); |
1494 | } |
1495 | const auto scan_flags_field_offset = scan_flags_field_.offset_in_bytes(); |
1496 | __ orq(compiler::FieldAddress(decoder_reg, scan_flags_field_offset), |
1497 | flags_reg); |
1498 | } |
1499 | |
1500 | LocationSummary* LoadUntaggedInstr::MakeLocationSummary(Zone* zone, |
1501 | bool opt) const { |
1502 | const intptr_t kNumInputs = 1; |
1503 | return LocationSummary::Make(zone, kNumInputs, Location::RequiresRegister(), |
1504 | LocationSummary::kNoCall); |
1505 | } |
1506 | |
1507 | void LoadUntaggedInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1508 | Register obj = locs()->in(0).reg(); |
1509 | Register result = locs()->out(0).reg(); |
1510 | if (object()->definition()->representation() == kUntagged) { |
1511 | __ movq(result, compiler::Address(obj, offset())); |
1512 | } else { |
1513 | ASSERT(object()->definition()->representation() == kTagged); |
1514 | __ movq(result, compiler::FieldAddress(obj, offset())); |
1515 | } |
1516 | } |
1517 | |
1518 | DEFINE_BACKEND(StoreUntagged, (NoLocation, Register obj, Register value)) { |
1519 | __ movq(compiler::Address(obj, instr->offset_from_tagged()), value); |
1520 | } |
1521 | |
1522 | class BoxAllocationSlowPath : public TemplateSlowPathCode<Instruction> { |
1523 | public: |
1524 | BoxAllocationSlowPath(Instruction* instruction, |
1525 | const Class& cls, |
1526 | Register result) |
1527 | : TemplateSlowPathCode(instruction), cls_(cls), result_(result) {} |
1528 | |
1529 | virtual void EmitNativeCode(FlowGraphCompiler* compiler) { |
1530 | if (compiler::Assembler::EmittingComments()) { |
1531 | __ Comment("%s slow path allocation of %s" , instruction()->DebugName(), |
1532 | String::Handle(cls_.ScrubbedName()).ToCString()); |
1533 | } |
1534 | __ Bind(entry_label()); |
1535 | const Code& stub = Code::ZoneHandle( |
1536 | compiler->zone(), StubCode::GetAllocationStubForClass(cls_)); |
1537 | |
1538 | LocationSummary* locs = instruction()->locs(); |
1539 | |
1540 | locs->live_registers()->Remove(Location::RegisterLocation(result_)); |
1541 | |
1542 | compiler->SaveLiveRegisters(locs); |
1543 | compiler->GenerateStubCall(TokenPosition::kNoSource, // No token position. |
1544 | stub, PcDescriptorsLayout::kOther, locs); |
1545 | __ MoveRegister(result_, RAX); |
1546 | compiler->RestoreLiveRegisters(locs); |
1547 | __ jmp(exit_label()); |
1548 | } |
1549 | |
1550 | static void Allocate(FlowGraphCompiler* compiler, |
1551 | Instruction* instruction, |
1552 | const Class& cls, |
1553 | Register result, |
1554 | Register temp) { |
1555 | if (compiler->intrinsic_mode()) { |
1556 | __ TryAllocate(cls, compiler->intrinsic_slow_path_label(), |
1557 | compiler::Assembler::kFarJump, result, temp); |
1558 | } else { |
1559 | BoxAllocationSlowPath* slow_path = |
1560 | new BoxAllocationSlowPath(instruction, cls, result); |
1561 | compiler->AddSlowPathCode(slow_path); |
1562 | |
1563 | __ TryAllocate(cls, slow_path->entry_label(), |
1564 | compiler::Assembler::kFarJump, result, temp); |
1565 | __ Bind(slow_path->exit_label()); |
1566 | } |
1567 | } |
1568 | |
1569 | private: |
1570 | const Class& cls_; |
1571 | const Register result_; |
1572 | }; |
1573 | |
1574 | Representation LoadIndexedInstr::representation() const { |
1575 | switch (class_id_) { |
1576 | case kArrayCid: |
1577 | case kImmutableArrayCid: |
1578 | return kTagged; |
1579 | case kOneByteStringCid: |
1580 | case kTwoByteStringCid: |
1581 | case kTypedDataInt8ArrayCid: |
1582 | case kTypedDataInt16ArrayCid: |
1583 | case kTypedDataUint8ArrayCid: |
1584 | case kTypedDataUint8ClampedArrayCid: |
1585 | case kTypedDataUint16ArrayCid: |
1586 | case kExternalOneByteStringCid: |
1587 | case kExternalTwoByteStringCid: |
1588 | case kExternalTypedDataUint8ArrayCid: |
1589 | case kExternalTypedDataUint8ClampedArrayCid: |
1590 | return kUnboxedIntPtr; |
1591 | case kTypedDataInt32ArrayCid: |
1592 | return kUnboxedInt32; |
1593 | case kTypedDataUint32ArrayCid: |
1594 | return kUnboxedUint32; |
1595 | case kTypedDataInt64ArrayCid: |
1596 | case kTypedDataUint64ArrayCid: |
1597 | return kUnboxedInt64; |
1598 | case kTypedDataFloat32ArrayCid: |
1599 | case kTypedDataFloat64ArrayCid: |
1600 | return kUnboxedDouble; |
1601 | case kTypedDataInt32x4ArrayCid: |
1602 | return kUnboxedInt32x4; |
1603 | case kTypedDataFloat32x4ArrayCid: |
1604 | return kUnboxedFloat32x4; |
1605 | case kTypedDataFloat64x2ArrayCid: |
1606 | return kUnboxedFloat64x2; |
1607 | default: |
1608 | UNIMPLEMENTED(); |
1609 | return kTagged; |
1610 | } |
1611 | } |
1612 | |
1613 | LocationSummary* LoadIndexedInstr::MakeLocationSummary(Zone* zone, |
1614 | bool opt) const { |
1615 | const intptr_t kNumInputs = 2; |
1616 | const intptr_t kNumTemps = 0; |
1617 | LocationSummary* locs = new (zone) |
1618 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
1619 | locs->set_in(0, Location::RequiresRegister()); |
1620 | // For tagged index with index_scale=1 as well as untagged index with |
1621 | // index_scale=16 we need a writable register due to assdressing mode |
1622 | // restrictions on X64. |
1623 | const bool need_writable_index_register = |
1624 | (index_scale() == 1 && !index_unboxed_) || |
1625 | (index_scale() == 16 && index_unboxed_); |
1626 | locs->set_in( |
1627 | 1, CanBeImmediateIndex(index(), class_id()) |
1628 | ? Location::Constant(index()->definition()->AsConstant()) |
1629 | : (need_writable_index_register ? Location::WritableRegister() |
1630 | : Location::RequiresRegister())); |
1631 | if ((representation() == kUnboxedDouble) || |
1632 | (representation() == kUnboxedFloat32x4) || |
1633 | (representation() == kUnboxedInt32x4) || |
1634 | (representation() == kUnboxedFloat64x2)) { |
1635 | locs->set_out(0, Location::RequiresFpuRegister()); |
1636 | } else { |
1637 | locs->set_out(0, Location::RequiresRegister()); |
1638 | } |
1639 | return locs; |
1640 | } |
1641 | |
1642 | void LoadIndexedInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1643 | // The array register points to the backing store for external arrays. |
1644 | const Register array = locs()->in(0).reg(); |
1645 | const Location index = locs()->in(1); |
1646 | |
1647 | intptr_t index_scale = index_scale_; |
1648 | if (index.IsRegister()) { |
1649 | if (index_scale == 1 && !index_unboxed_) { |
1650 | __ SmiUntag(index.reg()); |
1651 | } else if (index_scale == 16 && index_unboxed_) { |
1652 | // X64 does not support addressing mode using TIMES_16. |
1653 | __ SmiTag(index.reg()); |
1654 | index_scale >>= 1; |
1655 | } |
1656 | } else { |
1657 | ASSERT(index.IsConstant()); |
1658 | } |
1659 | |
1660 | compiler::Address element_address = |
1661 | index.IsRegister() ? compiler::Assembler::ElementAddressForRegIndex( |
1662 | IsExternal(), class_id(), index_scale, |
1663 | index_unboxed_, array, index.reg()) |
1664 | : compiler::Assembler::ElementAddressForIntIndex( |
1665 | IsExternal(), class_id(), index_scale, array, |
1666 | Smi::Cast(index.constant()).Value()); |
1667 | |
1668 | if (representation() == kUnboxedDouble || |
1669 | representation() == kUnboxedFloat32x4 || |
1670 | representation() == kUnboxedInt32x4 || |
1671 | representation() == kUnboxedFloat64x2) { |
1672 | XmmRegister result = locs()->out(0).fpu_reg(); |
1673 | if (class_id() == kTypedDataFloat32ArrayCid) { |
1674 | // Load single precision float. |
1675 | __ movss(result, element_address); |
1676 | } else if (class_id() == kTypedDataFloat64ArrayCid) { |
1677 | __ movsd(result, element_address); |
1678 | } else { |
1679 | ASSERT((class_id() == kTypedDataInt32x4ArrayCid) || |
1680 | (class_id() == kTypedDataFloat32x4ArrayCid) || |
1681 | (class_id() == kTypedDataFloat64x2ArrayCid)); |
1682 | __ movups(result, element_address); |
1683 | } |
1684 | return; |
1685 | } |
1686 | |
1687 | Register result = locs()->out(0).reg(); |
1688 | switch (class_id()) { |
1689 | case kTypedDataInt32ArrayCid: |
1690 | ASSERT(representation() == kUnboxedInt32); |
1691 | __ movsxd(result, element_address); |
1692 | break; |
1693 | case kTypedDataUint32ArrayCid: |
1694 | ASSERT(representation() == kUnboxedUint32); |
1695 | __ movl(result, element_address); |
1696 | break; |
1697 | case kTypedDataInt64ArrayCid: |
1698 | case kTypedDataUint64ArrayCid: |
1699 | ASSERT(representation() == kUnboxedInt64); |
1700 | __ movq(result, element_address); |
1701 | break; |
1702 | case kTypedDataInt8ArrayCid: |
1703 | ASSERT(representation() == kUnboxedIntPtr); |
1704 | __ movsxb(result, element_address); |
1705 | break; |
1706 | case kTypedDataUint8ArrayCid: |
1707 | case kTypedDataUint8ClampedArrayCid: |
1708 | case kExternalTypedDataUint8ArrayCid: |
1709 | case kExternalTypedDataUint8ClampedArrayCid: |
1710 | case kOneByteStringCid: |
1711 | case kExternalOneByteStringCid: |
1712 | ASSERT(representation() == kUnboxedIntPtr); |
1713 | __ movzxb(result, element_address); |
1714 | break; |
1715 | case kTypedDataInt16ArrayCid: |
1716 | ASSERT(representation() == kUnboxedIntPtr); |
1717 | __ movsxw(result, element_address); |
1718 | break; |
1719 | case kTypedDataUint16ArrayCid: |
1720 | case kTwoByteStringCid: |
1721 | case kExternalTwoByteStringCid: |
1722 | ASSERT(representation() == kUnboxedIntPtr); |
1723 | __ movzxw(result, element_address); |
1724 | break; |
1725 | default: |
1726 | ASSERT(representation() == kTagged); |
1727 | ASSERT((class_id() == kArrayCid) || (class_id() == kImmutableArrayCid)); |
1728 | __ movq(result, element_address); |
1729 | break; |
1730 | } |
1731 | } |
1732 | |
1733 | LocationSummary* LoadCodeUnitsInstr::MakeLocationSummary(Zone* zone, |
1734 | bool opt) const { |
1735 | const intptr_t kNumInputs = 2; |
1736 | const intptr_t kNumTemps = 0; |
1737 | LocationSummary* summary = new (zone) |
1738 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
1739 | summary->set_in(0, Location::RequiresRegister()); |
1740 | // The smi index is either untagged (element size == 1), or it is left smi |
1741 | // tagged (for all element sizes > 1). |
1742 | summary->set_in(1, index_scale() == 1 ? Location::WritableRegister() |
1743 | : Location::RequiresRegister()); |
1744 | summary->set_out(0, Location::RequiresRegister()); |
1745 | return summary; |
1746 | } |
1747 | |
1748 | void LoadCodeUnitsInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1749 | // The string register points to the backing store for external strings. |
1750 | const Register str = locs()->in(0).reg(); |
1751 | const Location index = locs()->in(1); |
1752 | |
1753 | compiler::Address element_address = |
1754 | compiler::Assembler::ElementAddressForRegIndex( |
1755 | IsExternal(), class_id(), index_scale(), /*index_unboxed=*/false, str, |
1756 | index.reg()); |
1757 | |
1758 | if ((index_scale() == 1)) { |
1759 | __ SmiUntag(index.reg()); |
1760 | } |
1761 | Register result = locs()->out(0).reg(); |
1762 | switch (class_id()) { |
1763 | case kOneByteStringCid: |
1764 | case kExternalOneByteStringCid: |
1765 | switch (element_count()) { |
1766 | case 1: |
1767 | __ movzxb(result, element_address); |
1768 | break; |
1769 | case 2: |
1770 | __ movzxw(result, element_address); |
1771 | break; |
1772 | case 4: |
1773 | __ movl(result, element_address); |
1774 | break; |
1775 | default: |
1776 | UNREACHABLE(); |
1777 | } |
1778 | __ SmiTag(result); |
1779 | break; |
1780 | case kTwoByteStringCid: |
1781 | case kExternalTwoByteStringCid: |
1782 | switch (element_count()) { |
1783 | case 1: |
1784 | __ movzxw(result, element_address); |
1785 | break; |
1786 | case 2: |
1787 | __ movl(result, element_address); |
1788 | break; |
1789 | default: |
1790 | UNREACHABLE(); |
1791 | } |
1792 | __ SmiTag(result); |
1793 | break; |
1794 | default: |
1795 | UNREACHABLE(); |
1796 | break; |
1797 | } |
1798 | } |
1799 | |
1800 | Representation StoreIndexedInstr::RequiredInputRepresentation( |
1801 | intptr_t idx) const { |
1802 | if (idx == 0) return kNoRepresentation; |
1803 | if (idx == 1) { |
1804 | if (index_unboxed_) { |
1805 | return kUnboxedInt64; |
1806 | } else { |
1807 | return kTagged; // Index is a smi. |
1808 | } |
1809 | } |
1810 | ASSERT(idx == 2); |
1811 | switch (class_id_) { |
1812 | case kArrayCid: |
1813 | return kTagged; |
1814 | case kOneByteStringCid: |
1815 | case kTwoByteStringCid: |
1816 | case kTypedDataInt8ArrayCid: |
1817 | case kTypedDataInt16ArrayCid: |
1818 | case kTypedDataUint8ArrayCid: |
1819 | case kTypedDataUint8ClampedArrayCid: |
1820 | case kTypedDataUint16ArrayCid: |
1821 | case kExternalTypedDataUint8ArrayCid: |
1822 | case kExternalTypedDataUint8ClampedArrayCid: |
1823 | return kUnboxedIntPtr; |
1824 | case kTypedDataInt32ArrayCid: |
1825 | return kUnboxedInt32; |
1826 | case kTypedDataUint32ArrayCid: |
1827 | return kUnboxedUint32; |
1828 | case kTypedDataInt64ArrayCid: |
1829 | case kTypedDataUint64ArrayCid: |
1830 | return kUnboxedInt64; |
1831 | case kTypedDataFloat32ArrayCid: |
1832 | case kTypedDataFloat64ArrayCid: |
1833 | return kUnboxedDouble; |
1834 | case kTypedDataFloat32x4ArrayCid: |
1835 | return kUnboxedFloat32x4; |
1836 | case kTypedDataInt32x4ArrayCid: |
1837 | return kUnboxedInt32x4; |
1838 | case kTypedDataFloat64x2ArrayCid: |
1839 | return kUnboxedFloat64x2; |
1840 | default: |
1841 | UNIMPLEMENTED(); |
1842 | return kTagged; |
1843 | } |
1844 | } |
1845 | |
1846 | LocationSummary* StoreIndexedInstr::MakeLocationSummary(Zone* zone, |
1847 | bool opt) const { |
1848 | const intptr_t kNumInputs = 3; |
1849 | const intptr_t kNumTemps = |
1850 | class_id() == kArrayCid && ShouldEmitStoreBarrier() ? 1 : 0; |
1851 | LocationSummary* locs = new (zone) |
1852 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
1853 | locs->set_in(0, Location::RequiresRegister()); |
1854 | // For tagged index with index_scale=1 as well as untagged index with |
1855 | // index_scale=16 we need a writable register due to assdressing mode |
1856 | // restrictions on X64. |
1857 | const bool need_writable_index_register = |
1858 | (index_scale() == 1 && !index_unboxed_) || |
1859 | (index_scale() == 16 && index_unboxed_); |
1860 | locs->set_in( |
1861 | 1, CanBeImmediateIndex(index(), class_id()) |
1862 | ? Location::Constant(index()->definition()->AsConstant()) |
1863 | : (need_writable_index_register ? Location::WritableRegister() |
1864 | : Location::RequiresRegister())); |
1865 | switch (class_id()) { |
1866 | case kArrayCid: |
1867 | locs->set_in(2, ShouldEmitStoreBarrier() |
1868 | ? Location::RegisterLocation(kWriteBarrierValueReg) |
1869 | : LocationRegisterOrConstant(value())); |
1870 | if (ShouldEmitStoreBarrier()) { |
1871 | locs->set_in(0, Location::RegisterLocation(kWriteBarrierObjectReg)); |
1872 | locs->set_temp(0, Location::RegisterLocation(kWriteBarrierSlotReg)); |
1873 | } |
1874 | break; |
1875 | case kExternalTypedDataUint8ArrayCid: |
1876 | case kExternalTypedDataUint8ClampedArrayCid: |
1877 | case kTypedDataInt8ArrayCid: |
1878 | case kTypedDataUint8ArrayCid: |
1879 | case kTypedDataUint8ClampedArrayCid: |
1880 | case kOneByteStringCid: |
1881 | case kTwoByteStringCid: |
1882 | // TODO(fschneider): Add location constraint for byte registers (RAX, |
1883 | // RBX, RCX, RDX) instead of using a fixed register. |
1884 | locs->set_in(2, LocationFixedRegisterOrSmiConstant(value(), RAX)); |
1885 | break; |
1886 | case kTypedDataInt16ArrayCid: |
1887 | case kTypedDataUint16ArrayCid: |
1888 | // Writable register because the value must be untagged before storing. |
1889 | locs->set_in(2, Location::WritableRegister()); |
1890 | break; |
1891 | case kTypedDataInt32ArrayCid: |
1892 | case kTypedDataUint32ArrayCid: |
1893 | case kTypedDataInt64ArrayCid: |
1894 | case kTypedDataUint64ArrayCid: |
1895 | locs->set_in(2, Location::RequiresRegister()); |
1896 | break; |
1897 | case kTypedDataFloat32ArrayCid: |
1898 | case kTypedDataFloat64ArrayCid: |
1899 | // TODO(srdjan): Support Float64 constants. |
1900 | locs->set_in(2, Location::RequiresFpuRegister()); |
1901 | break; |
1902 | case kTypedDataInt32x4ArrayCid: |
1903 | case kTypedDataFloat64x2ArrayCid: |
1904 | case kTypedDataFloat32x4ArrayCid: |
1905 | locs->set_in(2, Location::RequiresFpuRegister()); |
1906 | break; |
1907 | default: |
1908 | UNREACHABLE(); |
1909 | return NULL; |
1910 | } |
1911 | return locs; |
1912 | } |
1913 | |
1914 | void StoreIndexedInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
1915 | // The array register points to the backing store for external arrays. |
1916 | const Register array = locs()->in(0).reg(); |
1917 | const Location index = locs()->in(1); |
1918 | |
1919 | intptr_t index_scale = index_scale_; |
1920 | if (index.IsRegister()) { |
1921 | if (index_scale == 1 && !index_unboxed_) { |
1922 | __ SmiUntag(index.reg()); |
1923 | } else if (index_scale == 16 && index_unboxed_) { |
1924 | // X64 does not support addressing mode using TIMES_16. |
1925 | __ SmiTag(index.reg()); |
1926 | index_scale >>= 1; |
1927 | } |
1928 | } else { |
1929 | ASSERT(index.IsConstant()); |
1930 | } |
1931 | |
1932 | compiler::Address element_address = |
1933 | index.IsRegister() ? compiler::Assembler::ElementAddressForRegIndex( |
1934 | IsExternal(), class_id(), index_scale, |
1935 | index_unboxed_, array, index.reg()) |
1936 | : compiler::Assembler::ElementAddressForIntIndex( |
1937 | IsExternal(), class_id(), index_scale, array, |
1938 | Smi::Cast(index.constant()).Value()); |
1939 | |
1940 | switch (class_id()) { |
1941 | case kArrayCid: |
1942 | if (ShouldEmitStoreBarrier()) { |
1943 | Register value = locs()->in(2).reg(); |
1944 | Register slot = locs()->temp(0).reg(); |
1945 | __ leaq(slot, element_address); |
1946 | __ StoreIntoArray(array, slot, value, CanValueBeSmi()); |
1947 | } else if (locs()->in(2).IsConstant()) { |
1948 | const Object& constant = locs()->in(2).constant(); |
1949 | __ StoreIntoObjectNoBarrier(array, element_address, constant); |
1950 | } else { |
1951 | Register value = locs()->in(2).reg(); |
1952 | __ StoreIntoObjectNoBarrier(array, element_address, value); |
1953 | } |
1954 | break; |
1955 | case kOneByteStringCid: |
1956 | case kTypedDataInt8ArrayCid: |
1957 | case kTypedDataUint8ArrayCid: |
1958 | case kExternalTypedDataUint8ArrayCid: |
1959 | ASSERT(RequiredInputRepresentation(2) == kUnboxedIntPtr); |
1960 | if (locs()->in(2).IsConstant()) { |
1961 | const Smi& constant = Smi::Cast(locs()->in(2).constant()); |
1962 | __ movb(element_address, |
1963 | compiler::Immediate(static_cast<int8_t>(constant.Value()))); |
1964 | } else { |
1965 | ASSERT(locs()->in(2).reg() == RAX); |
1966 | __ movb(element_address, RAX); |
1967 | } |
1968 | break; |
1969 | case kTypedDataUint8ClampedArrayCid: |
1970 | case kExternalTypedDataUint8ClampedArrayCid: { |
1971 | ASSERT(RequiredInputRepresentation(2) == kUnboxedIntPtr); |
1972 | if (locs()->in(2).IsConstant()) { |
1973 | const Smi& constant = Smi::Cast(locs()->in(2).constant()); |
1974 | intptr_t value = constant.Value(); |
1975 | // Clamp to 0x0 or 0xFF respectively. |
1976 | if (value > 0xFF) { |
1977 | value = 0xFF; |
1978 | } else if (value < 0) { |
1979 | value = 0; |
1980 | } |
1981 | __ movb(element_address, |
1982 | compiler::Immediate(static_cast<int8_t>(value))); |
1983 | } else { |
1984 | ASSERT(locs()->in(2).reg() == RAX); |
1985 | compiler::Label store_value, store_0xff; |
1986 | __ CompareImmediate(RAX, compiler::Immediate(0xFF)); |
1987 | __ j(BELOW_EQUAL, &store_value, compiler::Assembler::kNearJump); |
1988 | // Clamp to 0x0 or 0xFF respectively. |
1989 | __ j(GREATER, &store_0xff); |
1990 | __ xorq(RAX, RAX); |
1991 | __ jmp(&store_value, compiler::Assembler::kNearJump); |
1992 | __ Bind(&store_0xff); |
1993 | __ LoadImmediate(RAX, compiler::Immediate(0xFF)); |
1994 | __ Bind(&store_value); |
1995 | __ movb(element_address, RAX); |
1996 | } |
1997 | break; |
1998 | } |
1999 | case kTwoByteStringCid: |
2000 | case kTypedDataInt16ArrayCid: |
2001 | case kTypedDataUint16ArrayCid: { |
2002 | ASSERT(RequiredInputRepresentation(2) == kUnboxedIntPtr); |
2003 | Register value = locs()->in(2).reg(); |
2004 | __ movw(element_address, value); |
2005 | break; |
2006 | } |
2007 | case kTypedDataInt32ArrayCid: |
2008 | case kTypedDataUint32ArrayCid: { |
2009 | Register value = locs()->in(2).reg(); |
2010 | __ movl(element_address, value); |
2011 | break; |
2012 | } |
2013 | case kTypedDataInt64ArrayCid: |
2014 | case kTypedDataUint64ArrayCid: { |
2015 | Register value = locs()->in(2).reg(); |
2016 | __ movq(element_address, value); |
2017 | break; |
2018 | } |
2019 | case kTypedDataFloat32ArrayCid: |
2020 | __ movss(element_address, locs()->in(2).fpu_reg()); |
2021 | break; |
2022 | case kTypedDataFloat64ArrayCid: |
2023 | __ movsd(element_address, locs()->in(2).fpu_reg()); |
2024 | break; |
2025 | case kTypedDataInt32x4ArrayCid: |
2026 | case kTypedDataFloat64x2ArrayCid: |
2027 | case kTypedDataFloat32x4ArrayCid: |
2028 | __ movups(element_address, locs()->in(2).fpu_reg()); |
2029 | break; |
2030 | default: |
2031 | UNREACHABLE(); |
2032 | } |
2033 | } |
2034 | |
2035 | LocationSummary* GuardFieldClassInstr::MakeLocationSummary(Zone* zone, |
2036 | bool opt) const { |
2037 | const intptr_t kNumInputs = 1; |
2038 | |
2039 | const intptr_t value_cid = value()->Type()->ToCid(); |
2040 | const intptr_t field_cid = field().guarded_cid(); |
2041 | |
2042 | const bool emit_full_guard = !opt || (field_cid == kIllegalCid); |
2043 | const bool needs_value_cid_temp_reg = |
2044 | (value_cid == kDynamicCid) && (emit_full_guard || (field_cid != kSmiCid)); |
2045 | const bool needs_field_temp_reg = emit_full_guard; |
2046 | |
2047 | intptr_t num_temps = 0; |
2048 | if (needs_value_cid_temp_reg) { |
2049 | num_temps++; |
2050 | } |
2051 | if (needs_field_temp_reg) { |
2052 | num_temps++; |
2053 | } |
2054 | |
2055 | LocationSummary* summary = new (zone) |
2056 | LocationSummary(zone, kNumInputs, num_temps, LocationSummary::kNoCall); |
2057 | summary->set_in(0, Location::RequiresRegister()); |
2058 | |
2059 | for (intptr_t i = 0; i < num_temps; i++) { |
2060 | summary->set_temp(i, Location::RequiresRegister()); |
2061 | } |
2062 | |
2063 | return summary; |
2064 | } |
2065 | |
2066 | void GuardFieldClassInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2067 | ASSERT(compiler::target::ObjectLayout::kClassIdTagSize == 16); |
2068 | ASSERT(sizeof(FieldLayout::guarded_cid_) == 2); |
2069 | ASSERT(sizeof(FieldLayout::is_nullable_) == 2); |
2070 | |
2071 | const intptr_t value_cid = value()->Type()->ToCid(); |
2072 | const intptr_t field_cid = field().guarded_cid(); |
2073 | const intptr_t nullability = field().is_nullable() ? kNullCid : kIllegalCid; |
2074 | |
2075 | if (field_cid == kDynamicCid) { |
2076 | return; // Nothing to emit. |
2077 | } |
2078 | |
2079 | const bool emit_full_guard = |
2080 | !compiler->is_optimizing() || (field_cid == kIllegalCid); |
2081 | |
2082 | const bool needs_value_cid_temp_reg = |
2083 | (value_cid == kDynamicCid) && (emit_full_guard || (field_cid != kSmiCid)); |
2084 | |
2085 | const bool needs_field_temp_reg = emit_full_guard; |
2086 | |
2087 | const Register value_reg = locs()->in(0).reg(); |
2088 | |
2089 | const Register value_cid_reg = |
2090 | needs_value_cid_temp_reg ? locs()->temp(0).reg() : kNoRegister; |
2091 | |
2092 | const Register field_reg = needs_field_temp_reg |
2093 | ? locs()->temp(locs()->temp_count() - 1).reg() |
2094 | : kNoRegister; |
2095 | |
2096 | compiler::Label ok, fail_label; |
2097 | |
2098 | compiler::Label* deopt = NULL; |
2099 | if (compiler->is_optimizing()) { |
2100 | deopt = compiler->AddDeoptStub(deopt_id(), ICData::kDeoptGuardField); |
2101 | } |
2102 | |
2103 | compiler::Label* fail = (deopt != NULL) ? deopt : &fail_label; |
2104 | |
2105 | if (emit_full_guard) { |
2106 | __ LoadObject(field_reg, Field::ZoneHandle(field().Original())); |
2107 | |
2108 | compiler::FieldAddress field_cid_operand(field_reg, |
2109 | Field::guarded_cid_offset()); |
2110 | compiler::FieldAddress field_nullability_operand( |
2111 | field_reg, Field::is_nullable_offset()); |
2112 | |
2113 | if (value_cid == kDynamicCid) { |
2114 | LoadValueCid(compiler, value_cid_reg, value_reg); |
2115 | |
2116 | __ cmpw(value_cid_reg, field_cid_operand); |
2117 | __ j(EQUAL, &ok); |
2118 | __ cmpw(value_cid_reg, field_nullability_operand); |
2119 | } else if (value_cid == kNullCid) { |
2120 | __ cmpw(field_nullability_operand, compiler::Immediate(value_cid)); |
2121 | } else { |
2122 | __ cmpw(field_cid_operand, compiler::Immediate(value_cid)); |
2123 | } |
2124 | __ j(EQUAL, &ok); |
2125 | |
2126 | // Check if the tracked state of the guarded field can be initialized |
2127 | // inline. If the field needs length check or requires type arguments and |
2128 | // class hierarchy processing for exactness tracking then we fall through |
2129 | // into runtime which is responsible for computing offset of the length |
2130 | // field based on the class id. |
2131 | const bool is_complicated_field = |
2132 | field().needs_length_check() || |
2133 | field().static_type_exactness_state().IsUninitialized(); |
2134 | if (!is_complicated_field) { |
2135 | // Uninitialized field can be handled inline. Check if the |
2136 | // field is still unitialized. |
2137 | __ cmpw(field_cid_operand, compiler::Immediate(kIllegalCid)); |
2138 | __ j(NOT_EQUAL, fail); |
2139 | |
2140 | if (value_cid == kDynamicCid) { |
2141 | __ movw(field_cid_operand, value_cid_reg); |
2142 | __ movw(field_nullability_operand, value_cid_reg); |
2143 | } else { |
2144 | ASSERT(field_reg != kNoRegister); |
2145 | __ movw(field_cid_operand, compiler::Immediate(value_cid)); |
2146 | __ movw(field_nullability_operand, compiler::Immediate(value_cid)); |
2147 | } |
2148 | |
2149 | __ jmp(&ok); |
2150 | } |
2151 | |
2152 | if (deopt == NULL) { |
2153 | ASSERT(!compiler->is_optimizing()); |
2154 | __ Bind(fail); |
2155 | |
2156 | __ cmpw(compiler::FieldAddress(field_reg, Field::guarded_cid_offset()), |
2157 | compiler::Immediate(kDynamicCid)); |
2158 | __ j(EQUAL, &ok); |
2159 | |
2160 | __ pushq(field_reg); |
2161 | __ pushq(value_reg); |
2162 | __ CallRuntime(kUpdateFieldCidRuntimeEntry, 2); |
2163 | __ Drop(2); // Drop the field and the value. |
2164 | } else { |
2165 | __ jmp(fail); |
2166 | } |
2167 | } else { |
2168 | ASSERT(compiler->is_optimizing()); |
2169 | ASSERT(deopt != NULL); |
2170 | |
2171 | // Field guard class has been initialized and is known. |
2172 | if (value_cid == kDynamicCid) { |
2173 | // Value's class id is not known. |
2174 | __ testq(value_reg, compiler::Immediate(kSmiTagMask)); |
2175 | |
2176 | if (field_cid != kSmiCid) { |
2177 | __ j(ZERO, fail); |
2178 | __ LoadClassId(value_cid_reg, value_reg); |
2179 | __ CompareImmediate(value_cid_reg, compiler::Immediate(field_cid)); |
2180 | } |
2181 | |
2182 | if (field().is_nullable() && (field_cid != kNullCid)) { |
2183 | __ j(EQUAL, &ok); |
2184 | __ CompareObject(value_reg, Object::null_object()); |
2185 | } |
2186 | |
2187 | __ j(NOT_EQUAL, fail); |
2188 | } else if (value_cid == field_cid) { |
2189 | // This would normaly be caught by Canonicalize, but RemoveRedefinitions |
2190 | // may sometimes produce the situation after the last Canonicalize pass. |
2191 | } else { |
2192 | // Both value's and field's class id is known. |
2193 | ASSERT(value_cid != nullability); |
2194 | __ jmp(fail); |
2195 | } |
2196 | } |
2197 | __ Bind(&ok); |
2198 | } |
2199 | |
2200 | LocationSummary* GuardFieldLengthInstr::MakeLocationSummary(Zone* zone, |
2201 | bool opt) const { |
2202 | const intptr_t kNumInputs = 1; |
2203 | if (!opt || (field().guarded_list_length() == Field::kUnknownFixedLength)) { |
2204 | const intptr_t kNumTemps = 3; |
2205 | LocationSummary* summary = new (zone) |
2206 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
2207 | summary->set_in(0, Location::RequiresRegister()); |
2208 | // We need temporaries for field object, length offset and expected length. |
2209 | summary->set_temp(0, Location::RequiresRegister()); |
2210 | summary->set_temp(1, Location::RequiresRegister()); |
2211 | summary->set_temp(2, Location::RequiresRegister()); |
2212 | return summary; |
2213 | } else { |
2214 | LocationSummary* summary = new (zone) |
2215 | LocationSummary(zone, kNumInputs, 0, LocationSummary::kNoCall); |
2216 | summary->set_in(0, Location::RequiresRegister()); |
2217 | return summary; |
2218 | } |
2219 | UNREACHABLE(); |
2220 | } |
2221 | |
2222 | void GuardFieldLengthInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2223 | if (field().guarded_list_length() == Field::kNoFixedLength) { |
2224 | return; // Nothing to emit. |
2225 | } |
2226 | |
2227 | compiler::Label* deopt = |
2228 | compiler->is_optimizing() |
2229 | ? compiler->AddDeoptStub(deopt_id(), ICData::kDeoptGuardField) |
2230 | : NULL; |
2231 | |
2232 | const Register value_reg = locs()->in(0).reg(); |
2233 | |
2234 | if (!compiler->is_optimizing() || |
2235 | (field().guarded_list_length() == Field::kUnknownFixedLength)) { |
2236 | const Register field_reg = locs()->temp(0).reg(); |
2237 | const Register offset_reg = locs()->temp(1).reg(); |
2238 | const Register length_reg = locs()->temp(2).reg(); |
2239 | |
2240 | compiler::Label ok; |
2241 | |
2242 | __ LoadObject(field_reg, Field::ZoneHandle(field().Original())); |
2243 | |
2244 | __ movsxb( |
2245 | offset_reg, |
2246 | compiler::FieldAddress( |
2247 | field_reg, Field::guarded_list_length_in_object_offset_offset())); |
2248 | __ movq(length_reg, compiler::FieldAddress( |
2249 | field_reg, Field::guarded_list_length_offset())); |
2250 | |
2251 | __ cmpq(offset_reg, compiler::Immediate(0)); |
2252 | __ j(NEGATIVE, &ok); |
2253 | |
2254 | // Load the length from the value. GuardFieldClass already verified that |
2255 | // value's class matches guarded class id of the field. |
2256 | // offset_reg contains offset already corrected by -kHeapObjectTag that is |
2257 | // why we use Address instead of FieldAddress. |
2258 | __ cmpq(length_reg, compiler::Address(value_reg, offset_reg, TIMES_1, 0)); |
2259 | |
2260 | if (deopt == NULL) { |
2261 | __ j(EQUAL, &ok); |
2262 | |
2263 | __ pushq(field_reg); |
2264 | __ pushq(value_reg); |
2265 | __ CallRuntime(kUpdateFieldCidRuntimeEntry, 2); |
2266 | __ Drop(2); // Drop the field and the value. |
2267 | } else { |
2268 | __ j(NOT_EQUAL, deopt); |
2269 | } |
2270 | |
2271 | __ Bind(&ok); |
2272 | } else { |
2273 | ASSERT(compiler->is_optimizing()); |
2274 | ASSERT(field().guarded_list_length() >= 0); |
2275 | ASSERT(field().guarded_list_length_in_object_offset() != |
2276 | Field::kUnknownLengthOffset); |
2277 | |
2278 | __ CompareImmediate( |
2279 | compiler::FieldAddress(value_reg, |
2280 | field().guarded_list_length_in_object_offset()), |
2281 | compiler::Immediate(Smi::RawValue(field().guarded_list_length()))); |
2282 | __ j(NOT_EQUAL, deopt); |
2283 | } |
2284 | } |
2285 | |
2286 | LocationSummary* GuardFieldTypeInstr::MakeLocationSummary(Zone* zone, |
2287 | bool opt) const { |
2288 | const intptr_t kNumInputs = 1; |
2289 | const intptr_t kNumTemps = 1; |
2290 | LocationSummary* summary = new (zone) |
2291 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
2292 | summary->set_in(0, Location::RequiresRegister()); |
2293 | summary->set_temp(0, Location::RequiresRegister()); |
2294 | return summary; |
2295 | } |
2296 | |
2297 | void GuardFieldTypeInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2298 | // Should never emit GuardFieldType for fields that are marked as NotTracking. |
2299 | ASSERT(field().static_type_exactness_state().IsTracking()); |
2300 | if (!field().static_type_exactness_state().NeedsFieldGuard()) { |
2301 | // Nothing to do: we only need to perform checks for trivially invariant |
2302 | // fields. If optimizing Canonicalize pass should have removed |
2303 | // this instruction. |
2304 | return; |
2305 | } |
2306 | |
2307 | compiler::Label* deopt = |
2308 | compiler->is_optimizing() |
2309 | ? compiler->AddDeoptStub(deopt_id(), ICData::kDeoptGuardField) |
2310 | : NULL; |
2311 | |
2312 | compiler::Label ok; |
2313 | |
2314 | const Register value_reg = locs()->in(0).reg(); |
2315 | const Register temp = locs()->temp(0).reg(); |
2316 | |
2317 | // Skip null values for nullable fields. |
2318 | if (!compiler->is_optimizing() || field().is_nullable()) { |
2319 | __ CompareObject(value_reg, Object::Handle()); |
2320 | __ j(EQUAL, &ok); |
2321 | } |
2322 | |
2323 | // Get the state. |
2324 | const Field& original = |
2325 | Field::ZoneHandle(compiler->zone(), field().Original()); |
2326 | __ LoadObject(temp, original); |
2327 | __ movsxb(temp, compiler::FieldAddress( |
2328 | temp, Field::static_type_exactness_state_offset())); |
2329 | |
2330 | if (!compiler->is_optimizing()) { |
2331 | // Check if field requires checking (it is in unitialized or trivially |
2332 | // exact state). |
2333 | __ cmpq(temp, |
2334 | compiler::Immediate(StaticTypeExactnessState::kUninitialized)); |
2335 | __ j(LESS, &ok); |
2336 | } |
2337 | |
2338 | compiler::Label call_runtime; |
2339 | if (field().static_type_exactness_state().IsUninitialized()) { |
2340 | // Can't initialize the field state inline in optimized code. |
2341 | __ cmpq(temp, |
2342 | compiler::Immediate(StaticTypeExactnessState::kUninitialized)); |
2343 | __ j(EQUAL, compiler->is_optimizing() ? deopt : &call_runtime); |
2344 | } |
2345 | |
2346 | // At this point temp is known to be type arguments offset in words. |
2347 | __ movq(temp, compiler::FieldAddress(value_reg, temp, TIMES_8, 0)); |
2348 | __ CompareObject(temp, TypeArguments::ZoneHandle( |
2349 | compiler->zone(), |
2350 | AbstractType::Handle(field().type()).arguments())); |
2351 | if (deopt != nullptr) { |
2352 | __ j(NOT_EQUAL, deopt); |
2353 | } else { |
2354 | __ j(EQUAL, &ok); |
2355 | |
2356 | __ Bind(&call_runtime); |
2357 | __ PushObject(original); |
2358 | __ pushq(value_reg); |
2359 | __ CallRuntime(kUpdateFieldCidRuntimeEntry, 2); |
2360 | __ Drop(2); |
2361 | } |
2362 | |
2363 | __ Bind(&ok); |
2364 | } |
2365 | |
2366 | LocationSummary* StoreInstanceFieldInstr::MakeLocationSummary(Zone* zone, |
2367 | bool opt) const { |
2368 | const intptr_t kNumInputs = 2; |
2369 | const intptr_t kNumTemps = (IsUnboxedStore() && opt) |
2370 | ? (FLAG_precompiled_mode ? 0 : 2) |
2371 | : (IsPotentialUnboxedStore() ? 3 : 0); |
2372 | LocationSummary* summary = new (zone) |
2373 | LocationSummary(zone, kNumInputs, kNumTemps, |
2374 | (!FLAG_precompiled_mode && |
2375 | ((IsUnboxedStore() && opt && is_initialization()) || |
2376 | IsPotentialUnboxedStore())) |
2377 | ? LocationSummary::kCallOnSlowPath |
2378 | : LocationSummary::kNoCall); |
2379 | |
2380 | summary->set_in(0, Location::RequiresRegister()); |
2381 | if (IsUnboxedStore() && opt) { |
2382 | if (slot().field().is_non_nullable_integer()) { |
2383 | ASSERT(FLAG_precompiled_mode); |
2384 | summary->set_in(1, Location::RequiresRegister()); |
2385 | } else { |
2386 | summary->set_in(1, Location::RequiresFpuRegister()); |
2387 | } |
2388 | if (!FLAG_precompiled_mode) { |
2389 | summary->set_temp(0, Location::RequiresRegister()); |
2390 | summary->set_temp(1, Location::RequiresRegister()); |
2391 | } |
2392 | } else if (IsPotentialUnboxedStore()) { |
2393 | summary->set_in(1, ShouldEmitStoreBarrier() ? Location::WritableRegister() |
2394 | : Location::RequiresRegister()); |
2395 | summary->set_temp(0, Location::RequiresRegister()); |
2396 | summary->set_temp(1, Location::RequiresRegister()); |
2397 | summary->set_temp(2, opt ? Location::RequiresFpuRegister() |
2398 | : Location::FpuRegisterLocation(XMM1)); |
2399 | } else { |
2400 | summary->set_in(1, ShouldEmitStoreBarrier() |
2401 | ? Location::RegisterLocation(kWriteBarrierValueReg) |
2402 | : LocationRegisterOrConstant(value())); |
2403 | } |
2404 | return summary; |
2405 | } |
2406 | |
2407 | static void EnsureMutableBox(FlowGraphCompiler* compiler, |
2408 | StoreInstanceFieldInstr* instruction, |
2409 | Register box_reg, |
2410 | const Class& cls, |
2411 | Register instance_reg, |
2412 | intptr_t offset, |
2413 | Register temp) { |
2414 | compiler::Label done; |
2415 | __ movq(box_reg, compiler::FieldAddress(instance_reg, offset)); |
2416 | __ CompareObject(box_reg, Object::null_object()); |
2417 | __ j(NOT_EQUAL, &done); |
2418 | BoxAllocationSlowPath::Allocate(compiler, instruction, cls, box_reg, temp); |
2419 | __ movq(temp, box_reg); |
2420 | __ StoreIntoObject(instance_reg, compiler::FieldAddress(instance_reg, offset), |
2421 | temp, compiler::Assembler::kValueIsNotSmi); |
2422 | |
2423 | __ Bind(&done); |
2424 | } |
2425 | |
2426 | void StoreInstanceFieldInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2427 | ASSERT(compiler::target::ObjectLayout::kClassIdTagSize == 16); |
2428 | ASSERT(sizeof(FieldLayout::guarded_cid_) == 2); |
2429 | ASSERT(sizeof(FieldLayout::is_nullable_) == 2); |
2430 | |
2431 | compiler::Label skip_store; |
2432 | |
2433 | const Register instance_reg = locs()->in(0).reg(); |
2434 | const intptr_t offset_in_bytes = OffsetInBytes(); |
2435 | ASSERT(offset_in_bytes > 0); // Field is finalized and points after header. |
2436 | |
2437 | if (IsUnboxedStore() && compiler->is_optimizing()) { |
2438 | if (slot().field().is_non_nullable_integer()) { |
2439 | const Register value = locs()->in(1).reg(); |
2440 | __ Comment("UnboxedIntegerStoreInstanceFieldInstr" ); |
2441 | __ movq(compiler::FieldAddress(instance_reg, offset_in_bytes), value); |
2442 | return; |
2443 | } |
2444 | |
2445 | XmmRegister value = locs()->in(1).fpu_reg(); |
2446 | const intptr_t cid = slot().field().UnboxedFieldCid(); |
2447 | |
2448 | // Real unboxed field |
2449 | if (FLAG_precompiled_mode) { |
2450 | switch (cid) { |
2451 | case kDoubleCid: |
2452 | __ Comment("UnboxedDoubleStoreInstanceFieldInstr" ); |
2453 | __ movsd(compiler::FieldAddress(instance_reg, offset_in_bytes), |
2454 | value); |
2455 | return; |
2456 | case kFloat32x4Cid: |
2457 | __ Comment("UnboxedFloat32x4StoreInstanceFieldInstr" ); |
2458 | __ movups(compiler::FieldAddress(instance_reg, offset_in_bytes), |
2459 | value); |
2460 | return; |
2461 | case kFloat64x2Cid: |
2462 | __ Comment("UnboxedFloat64x2StoreInstanceFieldInstr" ); |
2463 | __ movups(compiler::FieldAddress(instance_reg, offset_in_bytes), |
2464 | value); |
2465 | return; |
2466 | default: |
2467 | UNREACHABLE(); |
2468 | } |
2469 | } |
2470 | |
2471 | Register temp = locs()->temp(0).reg(); |
2472 | Register temp2 = locs()->temp(1).reg(); |
2473 | |
2474 | if (is_initialization()) { |
2475 | const Class* cls = NULL; |
2476 | switch (cid) { |
2477 | case kDoubleCid: |
2478 | cls = &compiler->double_class(); |
2479 | break; |
2480 | case kFloat32x4Cid: |
2481 | cls = &compiler->float32x4_class(); |
2482 | break; |
2483 | case kFloat64x2Cid: |
2484 | cls = &compiler->float64x2_class(); |
2485 | break; |
2486 | default: |
2487 | UNREACHABLE(); |
2488 | } |
2489 | |
2490 | BoxAllocationSlowPath::Allocate(compiler, this, *cls, temp, temp2); |
2491 | __ movq(temp2, temp); |
2492 | __ StoreIntoObject(instance_reg, |
2493 | compiler::FieldAddress(instance_reg, offset_in_bytes), |
2494 | temp2, compiler::Assembler::kValueIsNotSmi); |
2495 | } else { |
2496 | __ movq(temp, compiler::FieldAddress(instance_reg, offset_in_bytes)); |
2497 | } |
2498 | switch (cid) { |
2499 | case kDoubleCid: |
2500 | __ Comment("UnboxedDoubleStoreInstanceFieldInstr" ); |
2501 | __ movsd(compiler::FieldAddress(temp, Double::value_offset()), value); |
2502 | break; |
2503 | case kFloat32x4Cid: |
2504 | __ Comment("UnboxedFloat32x4StoreInstanceFieldInstr" ); |
2505 | __ movups(compiler::FieldAddress(temp, Float32x4::value_offset()), |
2506 | value); |
2507 | break; |
2508 | case kFloat64x2Cid: |
2509 | __ Comment("UnboxedFloat64x2StoreInstanceFieldInstr" ); |
2510 | __ movups(compiler::FieldAddress(temp, Float64x2::value_offset()), |
2511 | value); |
2512 | break; |
2513 | default: |
2514 | UNREACHABLE(); |
2515 | } |
2516 | return; |
2517 | } |
2518 | |
2519 | if (IsPotentialUnboxedStore()) { |
2520 | Register value_reg = locs()->in(1).reg(); |
2521 | Register temp = locs()->temp(0).reg(); |
2522 | Register temp2 = locs()->temp(1).reg(); |
2523 | FpuRegister fpu_temp = locs()->temp(2).fpu_reg(); |
2524 | |
2525 | if (ShouldEmitStoreBarrier()) { |
2526 | // Value input is a writable register and should be manually preserved |
2527 | // across allocation slow-path. |
2528 | locs()->live_registers()->Add(locs()->in(1), kTagged); |
2529 | } |
2530 | |
2531 | compiler::Label store_pointer; |
2532 | compiler::Label store_double; |
2533 | compiler::Label store_float32x4; |
2534 | compiler::Label store_float64x2; |
2535 | |
2536 | __ LoadObject(temp, Field::ZoneHandle(Z, slot().field().Original())); |
2537 | |
2538 | __ cmpw(compiler::FieldAddress(temp, Field::is_nullable_offset()), |
2539 | compiler::Immediate(kNullCid)); |
2540 | __ j(EQUAL, &store_pointer); |
2541 | |
2542 | __ movzxb(temp2, compiler::FieldAddress(temp, Field::kind_bits_offset())); |
2543 | __ testq(temp2, compiler::Immediate(1 << Field::kUnboxingCandidateBit)); |
2544 | __ j(ZERO, &store_pointer); |
2545 | |
2546 | __ cmpw(compiler::FieldAddress(temp, Field::guarded_cid_offset()), |
2547 | compiler::Immediate(kDoubleCid)); |
2548 | __ j(EQUAL, &store_double); |
2549 | |
2550 | __ cmpw(compiler::FieldAddress(temp, Field::guarded_cid_offset()), |
2551 | compiler::Immediate(kFloat32x4Cid)); |
2552 | __ j(EQUAL, &store_float32x4); |
2553 | |
2554 | __ cmpw(compiler::FieldAddress(temp, Field::guarded_cid_offset()), |
2555 | compiler::Immediate(kFloat64x2Cid)); |
2556 | __ j(EQUAL, &store_float64x2); |
2557 | |
2558 | // Fall through. |
2559 | __ jmp(&store_pointer); |
2560 | |
2561 | if (!compiler->is_optimizing()) { |
2562 | locs()->live_registers()->Add(locs()->in(0)); |
2563 | locs()->live_registers()->Add(locs()->in(1)); |
2564 | } |
2565 | |
2566 | { |
2567 | __ Bind(&store_double); |
2568 | EnsureMutableBox(compiler, this, temp, compiler->double_class(), |
2569 | instance_reg, offset_in_bytes, temp2); |
2570 | __ movsd(fpu_temp, |
2571 | compiler::FieldAddress(value_reg, Double::value_offset())); |
2572 | __ movsd(compiler::FieldAddress(temp, Double::value_offset()), fpu_temp); |
2573 | __ jmp(&skip_store); |
2574 | } |
2575 | |
2576 | { |
2577 | __ Bind(&store_float32x4); |
2578 | EnsureMutableBox(compiler, this, temp, compiler->float32x4_class(), |
2579 | instance_reg, offset_in_bytes, temp2); |
2580 | __ movups(fpu_temp, |
2581 | compiler::FieldAddress(value_reg, Float32x4::value_offset())); |
2582 | __ movups(compiler::FieldAddress(temp, Float32x4::value_offset()), |
2583 | fpu_temp); |
2584 | __ jmp(&skip_store); |
2585 | } |
2586 | |
2587 | { |
2588 | __ Bind(&store_float64x2); |
2589 | EnsureMutableBox(compiler, this, temp, compiler->float64x2_class(), |
2590 | instance_reg, offset_in_bytes, temp2); |
2591 | __ movups(fpu_temp, |
2592 | compiler::FieldAddress(value_reg, Float64x2::value_offset())); |
2593 | __ movups(compiler::FieldAddress(temp, Float64x2::value_offset()), |
2594 | fpu_temp); |
2595 | __ jmp(&skip_store); |
2596 | } |
2597 | |
2598 | __ Bind(&store_pointer); |
2599 | } |
2600 | |
2601 | if (ShouldEmitStoreBarrier()) { |
2602 | Register value_reg = locs()->in(1).reg(); |
2603 | __ StoreIntoObject(instance_reg, |
2604 | compiler::FieldAddress(instance_reg, offset_in_bytes), |
2605 | value_reg, CanValueBeSmi()); |
2606 | } else { |
2607 | if (locs()->in(1).IsConstant()) { |
2608 | __ StoreIntoObjectNoBarrier( |
2609 | instance_reg, compiler::FieldAddress(instance_reg, offset_in_bytes), |
2610 | locs()->in(1).constant()); |
2611 | } else { |
2612 | Register value_reg = locs()->in(1).reg(); |
2613 | __ StoreIntoObjectNoBarrier( |
2614 | instance_reg, compiler::FieldAddress(instance_reg, offset_in_bytes), |
2615 | value_reg); |
2616 | } |
2617 | } |
2618 | __ Bind(&skip_store); |
2619 | } |
2620 | |
2621 | LocationSummary* StoreStaticFieldInstr::MakeLocationSummary(Zone* zone, |
2622 | bool opt) const { |
2623 | const intptr_t kNumInputs = 1; |
2624 | const intptr_t kNumTemps = 1; |
2625 | LocationSummary* locs = new (zone) |
2626 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
2627 | locs->set_in(0, Location::RegisterLocation(kWriteBarrierValueReg)); |
2628 | locs->set_temp(0, Location::RequiresRegister()); |
2629 | return locs; |
2630 | } |
2631 | |
2632 | void StoreStaticFieldInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2633 | Register value = locs()->in(0).reg(); |
2634 | Register temp = locs()->temp(0).reg(); |
2635 | |
2636 | compiler->used_static_fields().Add(&field()); |
2637 | |
2638 | __ movq(temp, |
2639 | compiler::Address( |
2640 | THR, compiler::target::Thread::field_table_values_offset())); |
2641 | // Note: static fields ids won't be changed by hot-reload. |
2642 | __ movq( |
2643 | compiler::Address(temp, compiler::target::FieldTable::OffsetOf(field())), |
2644 | value); |
2645 | } |
2646 | |
2647 | LocationSummary* InstanceOfInstr::MakeLocationSummary(Zone* zone, |
2648 | bool opt) const { |
2649 | const intptr_t kNumInputs = 3; |
2650 | const intptr_t kNumTemps = 0; |
2651 | LocationSummary* summary = new (zone) |
2652 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
2653 | summary->set_in(0, Location::RegisterLocation(TypeTestABI::kInstanceReg)); |
2654 | summary->set_in(1, Location::RegisterLocation( |
2655 | TypeTestABI::kInstantiatorTypeArgumentsReg)); |
2656 | summary->set_in( |
2657 | 2, Location::RegisterLocation(TypeTestABI::kFunctionTypeArgumentsReg)); |
2658 | summary->set_out(0, Location::RegisterLocation(RAX)); |
2659 | return summary; |
2660 | } |
2661 | |
2662 | void InstanceOfInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2663 | ASSERT(locs()->in(0).reg() == TypeTestABI::kInstanceReg); |
2664 | ASSERT(locs()->in(1).reg() == TypeTestABI::kInstantiatorTypeArgumentsReg); |
2665 | ASSERT(locs()->in(2).reg() == TypeTestABI::kFunctionTypeArgumentsReg); |
2666 | |
2667 | compiler->GenerateInstanceOf(token_pos(), deopt_id(), type(), locs()); |
2668 | ASSERT(locs()->out(0).reg() == RAX); |
2669 | } |
2670 | |
2671 | // TODO(srdjan): In case of constant inputs make CreateArray kNoCall and |
2672 | // use slow path stub. |
2673 | LocationSummary* CreateArrayInstr::MakeLocationSummary(Zone* zone, |
2674 | bool opt) const { |
2675 | const intptr_t kNumInputs = 2; |
2676 | const intptr_t kNumTemps = 0; |
2677 | LocationSummary* locs = new (zone) |
2678 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
2679 | locs->set_in(0, Location::RegisterLocation(RBX)); |
2680 | locs->set_in(1, Location::RegisterLocation(R10)); |
2681 | locs->set_out(0, Location::RegisterLocation(RAX)); |
2682 | return locs; |
2683 | } |
2684 | |
2685 | // Inlines array allocation for known constant values. |
2686 | static void InlineArrayAllocation(FlowGraphCompiler* compiler, |
2687 | intptr_t num_elements, |
2688 | compiler::Label* slow_path, |
2689 | compiler::Label* done) { |
2690 | const int kInlineArraySize = 12; // Same as kInlineInstanceSize. |
2691 | const Register kLengthReg = R10; |
2692 | const Register kElemTypeReg = RBX; |
2693 | const intptr_t instance_size = Array::InstanceSize(num_elements); |
2694 | |
2695 | __ TryAllocateArray(kArrayCid, instance_size, slow_path, |
2696 | compiler::Assembler::kFarJump, |
2697 | RAX, // instance |
2698 | RCX, // end address |
2699 | R13); // temp |
2700 | |
2701 | // RAX: new object start as a tagged pointer. |
2702 | // Store the type argument field. |
2703 | __ StoreIntoObjectNoBarrier( |
2704 | RAX, compiler::FieldAddress(RAX, Array::type_arguments_offset()), |
2705 | kElemTypeReg); |
2706 | |
2707 | // Set the length field. |
2708 | __ StoreIntoObjectNoBarrier( |
2709 | RAX, compiler::FieldAddress(RAX, Array::length_offset()), kLengthReg); |
2710 | |
2711 | // Initialize all array elements to raw_null. |
2712 | // RAX: new object start as a tagged pointer. |
2713 | // RCX: new object end address. |
2714 | // RDI: iterator which initially points to the start of the variable |
2715 | // data area to be initialized. |
2716 | if (num_elements > 0) { |
2717 | const intptr_t array_size = instance_size - sizeof(ArrayLayout); |
2718 | __ LoadObject(R12, Object::null_object()); |
2719 | __ leaq(RDI, compiler::FieldAddress(RAX, sizeof(ArrayLayout))); |
2720 | if (array_size < (kInlineArraySize * kWordSize)) { |
2721 | intptr_t current_offset = 0; |
2722 | while (current_offset < array_size) { |
2723 | __ StoreIntoObjectNoBarrier(RAX, compiler::Address(RDI, current_offset), |
2724 | R12); |
2725 | current_offset += kWordSize; |
2726 | } |
2727 | } else { |
2728 | compiler::Label init_loop; |
2729 | __ Bind(&init_loop); |
2730 | __ StoreIntoObjectNoBarrier(RAX, compiler::Address(RDI, 0), R12); |
2731 | __ addq(RDI, compiler::Immediate(kWordSize)); |
2732 | __ cmpq(RDI, RCX); |
2733 | __ j(BELOW, &init_loop, compiler::Assembler::kNearJump); |
2734 | } |
2735 | } |
2736 | __ jmp(done, compiler::Assembler::kNearJump); |
2737 | } |
2738 | |
2739 | void CreateArrayInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2740 | TypeUsageInfo* type_usage_info = compiler->thread()->type_usage_info(); |
2741 | if (type_usage_info != nullptr) { |
2742 | const Class& list_class = Class::Handle( |
2743 | compiler->thread()->isolate()->class_table()->At(kArrayCid)); |
2744 | RegisterTypeArgumentsUse(compiler->function(), type_usage_info, list_class, |
2745 | element_type()->definition()); |
2746 | } |
2747 | |
2748 | // Allocate the array. R10 = length, RBX = element type. |
2749 | const Register kLengthReg = R10; |
2750 | const Register kElemTypeReg = RBX; |
2751 | const Register kResultReg = RAX; |
2752 | ASSERT(locs()->in(0).reg() == kElemTypeReg); |
2753 | ASSERT(locs()->in(1).reg() == kLengthReg); |
2754 | |
2755 | compiler::Label slow_path, done; |
2756 | if (compiler->is_optimizing() && !FLAG_precompiled_mode && |
2757 | num_elements()->BindsToConstant() && |
2758 | num_elements()->BoundConstant().IsSmi()) { |
2759 | const intptr_t length = Smi::Cast(num_elements()->BoundConstant()).Value(); |
2760 | if (Array::IsValidLength(length)) { |
2761 | InlineArrayAllocation(compiler, length, &slow_path, &done); |
2762 | } |
2763 | } |
2764 | |
2765 | __ Bind(&slow_path); |
2766 | auto object_store = compiler->isolate()->object_store(); |
2767 | const auto& allocate_array_stub = |
2768 | Code::ZoneHandle(compiler->zone(), object_store->allocate_array_stub()); |
2769 | compiler->GenerateStubCall(token_pos(), allocate_array_stub, |
2770 | PcDescriptorsLayout::kOther, locs(), deopt_id()); |
2771 | __ Bind(&done); |
2772 | ASSERT(locs()->out(0).reg() == kResultReg); |
2773 | } |
2774 | |
2775 | LocationSummary* LoadFieldInstr::MakeLocationSummary(Zone* zone, |
2776 | bool opt) const { |
2777 | const intptr_t kNumInputs = 1; |
2778 | const intptr_t kNumTemps = (IsUnboxedLoad() && opt) |
2779 | ? (FLAG_precompiled_mode ? 0 : 1) |
2780 | : (IsPotentialUnboxedLoad() ? 2 : 0); |
2781 | const auto contains_call = |
2782 | (IsUnboxedLoad() && opt) |
2783 | ? LocationSummary::kNoCall |
2784 | : (IsPotentialUnboxedLoad() |
2785 | ? LocationSummary::kCallOnSlowPath |
2786 | : (calls_initializer() ? LocationSummary::kCall |
2787 | : LocationSummary::kNoCall)); |
2788 | |
2789 | LocationSummary* locs = |
2790 | new (zone) LocationSummary(zone, kNumInputs, kNumTemps, contains_call); |
2791 | |
2792 | locs->set_in(0, calls_initializer() ? Location::RegisterLocation( |
2793 | InitInstanceFieldABI::kInstanceReg) |
2794 | : Location::RequiresRegister()); |
2795 | |
2796 | if (IsUnboxedLoad() && opt) { |
2797 | ASSERT(!calls_initializer()); |
2798 | if (!FLAG_precompiled_mode) { |
2799 | locs->set_temp(0, Location::RequiresRegister()); |
2800 | } |
2801 | if (slot().field().is_non_nullable_integer()) { |
2802 | ASSERT(FLAG_precompiled_mode); |
2803 | locs->set_out(0, Location::RequiresRegister()); |
2804 | } else { |
2805 | locs->set_out(0, Location::RequiresFpuRegister()); |
2806 | } |
2807 | } else if (IsPotentialUnboxedLoad()) { |
2808 | ASSERT(!calls_initializer()); |
2809 | locs->set_temp(0, opt ? Location::RequiresFpuRegister() |
2810 | : Location::FpuRegisterLocation(XMM1)); |
2811 | locs->set_temp(1, Location::RequiresRegister()); |
2812 | locs->set_out(0, Location::RequiresRegister()); |
2813 | } else if (calls_initializer()) { |
2814 | locs->set_out(0, |
2815 | Location::RegisterLocation(InitInstanceFieldABI::kResultReg)); |
2816 | } else { |
2817 | locs->set_out(0, Location::RequiresRegister()); |
2818 | } |
2819 | return locs; |
2820 | } |
2821 | |
2822 | void LoadFieldInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2823 | ASSERT(compiler::target::ObjectLayout::kClassIdTagSize == 16); |
2824 | ASSERT(sizeof(FieldLayout::guarded_cid_) == 2); |
2825 | ASSERT(sizeof(FieldLayout::is_nullable_) == 2); |
2826 | |
2827 | Register instance_reg = locs()->in(0).reg(); |
2828 | if (IsUnboxedLoad() && compiler->is_optimizing()) { |
2829 | ASSERT(!calls_initializer()); |
2830 | if (slot().field().is_non_nullable_integer()) { |
2831 | const Register result = locs()->out(0).reg(); |
2832 | __ Comment("UnboxedIntegerLoadFieldInstr" ); |
2833 | __ movq(result, compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2834 | return; |
2835 | } |
2836 | |
2837 | XmmRegister result = locs()->out(0).fpu_reg(); |
2838 | const intptr_t cid = slot().field().UnboxedFieldCid(); |
2839 | |
2840 | // Real unboxed field |
2841 | if (FLAG_precompiled_mode) { |
2842 | switch (cid) { |
2843 | case kDoubleCid: |
2844 | __ Comment("UnboxedDoubleLoadFieldInstr" ); |
2845 | __ movsd(result, |
2846 | compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2847 | break; |
2848 | case kFloat32x4Cid: |
2849 | __ Comment("UnboxedFloat32x4LoadFieldInstr" ); |
2850 | __ movups(result, |
2851 | compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2852 | break; |
2853 | case kFloat64x2Cid: |
2854 | __ Comment("UnboxedFloat64x2LoadFieldInstr" ); |
2855 | __ movups(result, |
2856 | compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2857 | break; |
2858 | default: |
2859 | UNREACHABLE(); |
2860 | } |
2861 | return; |
2862 | } |
2863 | |
2864 | Register temp = locs()->temp(0).reg(); |
2865 | __ movq(temp, compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2866 | switch (cid) { |
2867 | case kDoubleCid: |
2868 | __ Comment("UnboxedDoubleLoadFieldInstr" ); |
2869 | __ movsd(result, compiler::FieldAddress(temp, Double::value_offset())); |
2870 | break; |
2871 | case kFloat32x4Cid: |
2872 | __ Comment("UnboxedFloat32x4LoadFieldInstr" ); |
2873 | __ movups(result, |
2874 | compiler::FieldAddress(temp, Float32x4::value_offset())); |
2875 | break; |
2876 | case kFloat64x2Cid: |
2877 | __ Comment("UnboxedFloat64x2LoadFieldInstr" ); |
2878 | __ movups(result, |
2879 | compiler::FieldAddress(temp, Float64x2::value_offset())); |
2880 | break; |
2881 | default: |
2882 | UNREACHABLE(); |
2883 | } |
2884 | return; |
2885 | } |
2886 | |
2887 | compiler::Label done; |
2888 | Register result = locs()->out(0).reg(); |
2889 | if (IsPotentialUnboxedLoad()) { |
2890 | ASSERT(!calls_initializer()); |
2891 | Register temp = locs()->temp(1).reg(); |
2892 | XmmRegister value = locs()->temp(0).fpu_reg(); |
2893 | |
2894 | compiler::Label load_pointer; |
2895 | compiler::Label load_double; |
2896 | compiler::Label load_float32x4; |
2897 | compiler::Label load_float64x2; |
2898 | |
2899 | __ LoadObject(result, Field::ZoneHandle(slot().field().Original())); |
2900 | |
2901 | compiler::FieldAddress field_cid_operand(result, |
2902 | Field::guarded_cid_offset()); |
2903 | compiler::FieldAddress field_nullability_operand( |
2904 | result, Field::is_nullable_offset()); |
2905 | |
2906 | __ cmpw(field_nullability_operand, compiler::Immediate(kNullCid)); |
2907 | __ j(EQUAL, &load_pointer); |
2908 | |
2909 | __ cmpw(field_cid_operand, compiler::Immediate(kDoubleCid)); |
2910 | __ j(EQUAL, &load_double); |
2911 | |
2912 | __ cmpw(field_cid_operand, compiler::Immediate(kFloat32x4Cid)); |
2913 | __ j(EQUAL, &load_float32x4); |
2914 | |
2915 | __ cmpw(field_cid_operand, compiler::Immediate(kFloat64x2Cid)); |
2916 | __ j(EQUAL, &load_float64x2); |
2917 | |
2918 | // Fall through. |
2919 | __ jmp(&load_pointer); |
2920 | |
2921 | if (!compiler->is_optimizing()) { |
2922 | locs()->live_registers()->Add(locs()->in(0)); |
2923 | } |
2924 | |
2925 | { |
2926 | __ Bind(&load_double); |
2927 | BoxAllocationSlowPath::Allocate(compiler, this, compiler->double_class(), |
2928 | result, temp); |
2929 | __ movq(temp, compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2930 | __ movsd(value, compiler::FieldAddress(temp, Double::value_offset())); |
2931 | __ movsd(compiler::FieldAddress(result, Double::value_offset()), value); |
2932 | __ jmp(&done); |
2933 | } |
2934 | |
2935 | { |
2936 | __ Bind(&load_float32x4); |
2937 | BoxAllocationSlowPath::Allocate( |
2938 | compiler, this, compiler->float32x4_class(), result, temp); |
2939 | __ movq(temp, compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2940 | __ movups(value, compiler::FieldAddress(temp, Float32x4::value_offset())); |
2941 | __ movups(compiler::FieldAddress(result, Float32x4::value_offset()), |
2942 | value); |
2943 | __ jmp(&done); |
2944 | } |
2945 | |
2946 | { |
2947 | __ Bind(&load_float64x2); |
2948 | BoxAllocationSlowPath::Allocate( |
2949 | compiler, this, compiler->float64x2_class(), result, temp); |
2950 | __ movq(temp, compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2951 | __ movups(value, compiler::FieldAddress(temp, Float64x2::value_offset())); |
2952 | __ movups(compiler::FieldAddress(result, Float64x2::value_offset()), |
2953 | value); |
2954 | __ jmp(&done); |
2955 | } |
2956 | |
2957 | __ Bind(&load_pointer); |
2958 | } |
2959 | |
2960 | __ movq(result, compiler::FieldAddress(instance_reg, OffsetInBytes())); |
2961 | |
2962 | if (calls_initializer()) { |
2963 | EmitNativeCodeForInitializerCall(compiler); |
2964 | } |
2965 | |
2966 | __ Bind(&done); |
2967 | } |
2968 | |
2969 | LocationSummary* InstantiateTypeInstr::MakeLocationSummary(Zone* zone, |
2970 | bool opt) const { |
2971 | const intptr_t kNumInputs = 2; |
2972 | const intptr_t kNumTemps = 0; |
2973 | LocationSummary* locs = new (zone) |
2974 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
2975 | locs->set_in(0, Location::RegisterLocation( |
2976 | InstantiationABI::kInstantiatorTypeArgumentsReg)); |
2977 | locs->set_in(1, Location::RegisterLocation( |
2978 | InstantiationABI::kFunctionTypeArgumentsReg)); |
2979 | locs->set_out(0, |
2980 | Location::RegisterLocation(InstantiationABI::kResultTypeReg)); |
2981 | return locs; |
2982 | } |
2983 | |
2984 | void InstantiateTypeInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
2985 | Register instantiator_type_args_reg = locs()->in(0).reg(); |
2986 | Register function_type_args_reg = locs()->in(1).reg(); |
2987 | Register result_reg = locs()->out(0).reg(); |
2988 | |
2989 | // 'instantiator_type_args_reg' is a TypeArguments object (or null). |
2990 | // 'function_type_args_reg' is a TypeArguments object (or null). |
2991 | // A runtime call to instantiate the type is required. |
2992 | __ PushObject(Object::null_object()); // Make room for the result. |
2993 | __ PushObject(type()); |
2994 | __ pushq(instantiator_type_args_reg); // Push instantiator type arguments. |
2995 | __ pushq(function_type_args_reg); // Push function type arguments. |
2996 | compiler->GenerateRuntimeCall(token_pos(), deopt_id(), |
2997 | kInstantiateTypeRuntimeEntry, 3, locs()); |
2998 | __ Drop(3); // Drop 2 type vectors, and uninstantiated type. |
2999 | __ popq(result_reg); // Pop instantiated type. |
3000 | } |
3001 | |
3002 | LocationSummary* InstantiateTypeArgumentsInstr::MakeLocationSummary( |
3003 | Zone* zone, |
3004 | bool opt) const { |
3005 | const intptr_t kNumInputs = 2; |
3006 | const intptr_t kNumTemps = 0; |
3007 | LocationSummary* locs = new (zone) |
3008 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
3009 | locs->set_in(0, Location::RegisterLocation( |
3010 | InstantiationABI::kInstantiatorTypeArgumentsReg)); |
3011 | locs->set_in(1, Location::RegisterLocation( |
3012 | InstantiationABI::kFunctionTypeArgumentsReg)); |
3013 | locs->set_out( |
3014 | 0, Location::RegisterLocation(InstantiationABI::kResultTypeArgumentsReg)); |
3015 | return locs; |
3016 | } |
3017 | |
3018 | void InstantiateTypeArgumentsInstr::EmitNativeCode( |
3019 | FlowGraphCompiler* compiler) { |
3020 | Register instantiator_type_args_reg = locs()->in(0).reg(); |
3021 | Register function_type_args_reg = locs()->in(1).reg(); |
3022 | Register result_reg = locs()->out(0).reg(); |
3023 | |
3024 | // 'instantiator_type_args_reg' is a TypeArguments object (or null). |
3025 | // 'function_type_args_reg' is a TypeArguments object (or null). |
3026 | |
3027 | // If both the instantiator and function type arguments are null and if the |
3028 | // type argument vector instantiated from null becomes a vector of dynamic, |
3029 | // then use null as the type arguments. |
3030 | compiler::Label type_arguments_instantiated; |
3031 | const intptr_t len = type_arguments().Length(); |
3032 | const bool can_function_type_args_be_null = |
3033 | function_type_arguments()->CanBe(Object::null_object()); |
3034 | if (type_arguments().IsRawWhenInstantiatedFromRaw(len) && |
3035 | can_function_type_args_be_null) { |
3036 | compiler::Label non_null_type_args; |
3037 | ASSERT(result_reg != instantiator_type_args_reg && |
3038 | result_reg != function_type_args_reg); |
3039 | __ LoadObject(result_reg, Object::null_object()); |
3040 | __ cmpq(instantiator_type_args_reg, result_reg); |
3041 | if (!function_type_arguments()->BindsToConstant()) { |
3042 | __ j(NOT_EQUAL, &non_null_type_args, compiler::Assembler::kNearJump); |
3043 | __ cmpq(function_type_args_reg, result_reg); |
3044 | } |
3045 | __ j(EQUAL, &type_arguments_instantiated, compiler::Assembler::kNearJump); |
3046 | __ Bind(&non_null_type_args); |
3047 | } |
3048 | __ LoadObject(InstantiationABI::kUninstantiatedTypeArgumentsReg, |
3049 | type_arguments()); |
3050 | compiler->GenerateStubCall(token_pos(), GetStub(), |
3051 | PcDescriptorsLayout::kOther, locs()); |
3052 | __ Bind(&type_arguments_instantiated); |
3053 | } |
3054 | |
3055 | LocationSummary* AllocateUninitializedContextInstr::MakeLocationSummary( |
3056 | Zone* zone, |
3057 | bool opt) const { |
3058 | ASSERT(opt); |
3059 | const intptr_t kNumInputs = 0; |
3060 | const intptr_t kNumTemps = 2; |
3061 | LocationSummary* locs = new (zone) LocationSummary( |
3062 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
3063 | locs->set_temp(0, Location::RegisterLocation(R10)); |
3064 | locs->set_temp(1, Location::RegisterLocation(R13)); |
3065 | locs->set_out(0, Location::RegisterLocation(RAX)); |
3066 | return locs; |
3067 | } |
3068 | |
3069 | class AllocateContextSlowPath |
3070 | : public TemplateSlowPathCode<AllocateUninitializedContextInstr> { |
3071 | public: |
3072 | explicit AllocateContextSlowPath( |
3073 | AllocateUninitializedContextInstr* instruction) |
3074 | : TemplateSlowPathCode(instruction) {} |
3075 | |
3076 | virtual void EmitNativeCode(FlowGraphCompiler* compiler) { |
3077 | __ Comment("AllocateContextSlowPath" ); |
3078 | __ Bind(entry_label()); |
3079 | |
3080 | LocationSummary* locs = instruction()->locs(); |
3081 | locs->live_registers()->Remove(locs->out(0)); |
3082 | |
3083 | compiler->SaveLiveRegisters(locs); |
3084 | |
3085 | auto object_store = compiler->isolate()->object_store(); |
3086 | const auto& allocate_context_stub = Code::ZoneHandle( |
3087 | compiler->zone(), object_store->allocate_context_stub()); |
3088 | |
3089 | __ LoadImmediate( |
3090 | R10, compiler::Immediate(instruction()->num_context_variables())); |
3091 | compiler->GenerateStubCall(instruction()->token_pos(), |
3092 | allocate_context_stub, |
3093 | PcDescriptorsLayout::kOther, locs); |
3094 | ASSERT(instruction()->locs()->out(0).reg() == RAX); |
3095 | compiler->RestoreLiveRegisters(instruction()->locs()); |
3096 | __ jmp(exit_label()); |
3097 | } |
3098 | }; |
3099 | |
3100 | void AllocateUninitializedContextInstr::EmitNativeCode( |
3101 | FlowGraphCompiler* compiler) { |
3102 | ASSERT(compiler->is_optimizing()); |
3103 | Register temp = locs()->temp(0).reg(); |
3104 | Register result = locs()->out(0).reg(); |
3105 | // Try allocate the object. |
3106 | AllocateContextSlowPath* slow_path = new AllocateContextSlowPath(this); |
3107 | compiler->AddSlowPathCode(slow_path); |
3108 | intptr_t instance_size = Context::InstanceSize(num_context_variables()); |
3109 | |
3110 | __ TryAllocateArray(kContextCid, instance_size, slow_path->entry_label(), |
3111 | compiler::Assembler::kFarJump, |
3112 | result, // instance |
3113 | temp, // end address |
3114 | locs()->temp(1).reg()); |
3115 | |
3116 | // Setup up number of context variables field. |
3117 | __ movq(compiler::FieldAddress(result, Context::num_variables_offset()), |
3118 | compiler::Immediate(num_context_variables())); |
3119 | |
3120 | __ Bind(slow_path->exit_label()); |
3121 | } |
3122 | |
3123 | LocationSummary* AllocateContextInstr::MakeLocationSummary(Zone* zone, |
3124 | bool opt) const { |
3125 | const intptr_t kNumInputs = 0; |
3126 | const intptr_t kNumTemps = 1; |
3127 | LocationSummary* locs = new (zone) |
3128 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
3129 | locs->set_temp(0, Location::RegisterLocation(R10)); |
3130 | locs->set_out(0, Location::RegisterLocation(RAX)); |
3131 | return locs; |
3132 | } |
3133 | |
3134 | void AllocateContextInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3135 | ASSERT(locs()->temp(0).reg() == R10); |
3136 | ASSERT(locs()->out(0).reg() == RAX); |
3137 | |
3138 | auto object_store = compiler->isolate()->object_store(); |
3139 | const auto& allocate_context_stub = |
3140 | Code::ZoneHandle(compiler->zone(), object_store->allocate_context_stub()); |
3141 | |
3142 | __ LoadImmediate(R10, compiler::Immediate(num_context_variables())); |
3143 | compiler->GenerateStubCall(token_pos(), allocate_context_stub, |
3144 | PcDescriptorsLayout::kOther, locs()); |
3145 | } |
3146 | |
3147 | LocationSummary* CloneContextInstr::MakeLocationSummary(Zone* zone, |
3148 | bool opt) const { |
3149 | const intptr_t kNumInputs = 1; |
3150 | const intptr_t kNumTemps = 0; |
3151 | LocationSummary* locs = new (zone) |
3152 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
3153 | locs->set_in(0, Location::RegisterLocation(R9)); |
3154 | locs->set_out(0, Location::RegisterLocation(RAX)); |
3155 | return locs; |
3156 | } |
3157 | |
3158 | void CloneContextInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3159 | ASSERT(locs()->in(0).reg() == R9); |
3160 | ASSERT(locs()->out(0).reg() == RAX); |
3161 | |
3162 | auto object_store = compiler->isolate()->object_store(); |
3163 | const auto& clone_context_stub = |
3164 | Code::ZoneHandle(compiler->zone(), object_store->clone_context_stub()); |
3165 | compiler->GenerateStubCall(token_pos(), clone_context_stub, |
3166 | /*kind=*/PcDescriptorsLayout::kOther, locs()); |
3167 | } |
3168 | |
3169 | LocationSummary* CatchBlockEntryInstr::MakeLocationSummary(Zone* zone, |
3170 | bool opt) const { |
3171 | UNREACHABLE(); |
3172 | return NULL; |
3173 | } |
3174 | |
3175 | void CatchBlockEntryInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3176 | __ Bind(compiler->GetJumpLabel(this)); |
3177 | compiler->AddExceptionHandler( |
3178 | catch_try_index(), try_index(), compiler->assembler()->CodeSize(), |
3179 | is_generated(), catch_handler_types_, needs_stacktrace()); |
3180 | if (!FLAG_precompiled_mode) { |
3181 | // On lazy deoptimization we patch the optimized code here to enter the |
3182 | // deoptimization stub. |
3183 | const intptr_t deopt_id = DeoptId::ToDeoptAfter(GetDeoptId()); |
3184 | if (compiler->is_optimizing()) { |
3185 | compiler->AddDeoptIndexAtCall(deopt_id); |
3186 | } else { |
3187 | compiler->AddCurrentDescriptor(PcDescriptorsLayout::kDeopt, deopt_id, |
3188 | TokenPosition::kNoSource); |
3189 | } |
3190 | } |
3191 | if (HasParallelMove()) { |
3192 | compiler->parallel_move_resolver()->EmitNativeCode(parallel_move()); |
3193 | } |
3194 | |
3195 | // Restore RSP from RBP as we are coming from a throw and the code for |
3196 | // popping arguments has not been run. |
3197 | const intptr_t fp_sp_dist = |
3198 | (compiler::target::frame_layout.first_local_from_fp + 1 - |
3199 | compiler->StackSize()) * |
3200 | kWordSize; |
3201 | ASSERT(fp_sp_dist <= 0); |
3202 | __ leaq(RSP, compiler::Address(RBP, fp_sp_dist)); |
3203 | |
3204 | if (!compiler->is_optimizing()) { |
3205 | if (raw_exception_var_ != nullptr) { |
3206 | __ movq(compiler::Address(RBP, |
3207 | compiler::target::FrameOffsetInBytesForVariable( |
3208 | raw_exception_var_)), |
3209 | kExceptionObjectReg); |
3210 | } |
3211 | if (raw_stacktrace_var_ != nullptr) { |
3212 | __ movq(compiler::Address(RBP, |
3213 | compiler::target::FrameOffsetInBytesForVariable( |
3214 | raw_stacktrace_var_)), |
3215 | kStackTraceObjectReg); |
3216 | } |
3217 | } |
3218 | } |
3219 | |
3220 | LocationSummary* CheckStackOverflowInstr::MakeLocationSummary(Zone* zone, |
3221 | bool opt) const { |
3222 | const intptr_t kNumInputs = 0; |
3223 | const intptr_t kNumTemps = 1; |
3224 | const bool using_shared_stub = UseSharedSlowPathStub(opt); |
3225 | LocationSummary* summary = new (zone) |
3226 | LocationSummary(zone, kNumInputs, kNumTemps, |
3227 | using_shared_stub ? LocationSummary::kCallOnSharedSlowPath |
3228 | : LocationSummary::kCallOnSlowPath); |
3229 | summary->set_temp(0, Location::RequiresRegister()); |
3230 | return summary; |
3231 | } |
3232 | |
3233 | class CheckStackOverflowSlowPath |
3234 | : public TemplateSlowPathCode<CheckStackOverflowInstr> { |
3235 | public: |
3236 | static constexpr intptr_t kNumSlowPathArgs = 0; |
3237 | |
3238 | explicit CheckStackOverflowSlowPath(CheckStackOverflowInstr* instruction) |
3239 | : TemplateSlowPathCode(instruction) {} |
3240 | |
3241 | virtual void EmitNativeCode(FlowGraphCompiler* compiler) { |
3242 | if (compiler->isolate()->use_osr() && osr_entry_label()->IsLinked()) { |
3243 | __ Comment("CheckStackOverflowSlowPathOsr" ); |
3244 | __ Bind(osr_entry_label()); |
3245 | __ movq(compiler::Address(THR, Thread::stack_overflow_flags_offset()), |
3246 | compiler::Immediate(Thread::kOsrRequest)); |
3247 | } |
3248 | __ Comment("CheckStackOverflowSlowPath" ); |
3249 | __ Bind(entry_label()); |
3250 | const bool using_shared_stub = |
3251 | instruction()->locs()->call_on_shared_slow_path(); |
3252 | if (!using_shared_stub) { |
3253 | compiler->SaveLiveRegisters(instruction()->locs()); |
3254 | } |
3255 | // pending_deoptimization_env_ is needed to generate a runtime call that |
3256 | // may throw an exception. |
3257 | ASSERT(compiler->pending_deoptimization_env_ == NULL); |
3258 | Environment* env = |
3259 | compiler->SlowPathEnvironmentFor(instruction(), kNumSlowPathArgs); |
3260 | compiler->pending_deoptimization_env_ = env; |
3261 | |
3262 | if (using_shared_stub) { |
3263 | const uword entry_point_offset = |
3264 | Thread::stack_overflow_shared_stub_entry_point_offset( |
3265 | instruction()->locs()->live_registers()->FpuRegisterCount() > 0); |
3266 | __ call(compiler::Address(THR, entry_point_offset)); |
3267 | compiler->RecordSafepoint(instruction()->locs(), kNumSlowPathArgs); |
3268 | compiler->RecordCatchEntryMoves(); |
3269 | compiler->AddDescriptor( |
3270 | PcDescriptorsLayout::kOther, compiler->assembler()->CodeSize(), |
3271 | instruction()->deopt_id(), instruction()->token_pos(), |
3272 | compiler->CurrentTryIndex()); |
3273 | } else { |
3274 | compiler->GenerateRuntimeCall( |
3275 | instruction()->token_pos(), instruction()->deopt_id(), |
3276 | kStackOverflowRuntimeEntry, kNumSlowPathArgs, instruction()->locs()); |
3277 | } |
3278 | |
3279 | if (compiler->isolate()->use_osr() && !compiler->is_optimizing() && |
3280 | instruction()->in_loop()) { |
3281 | // In unoptimized code, record loop stack checks as possible OSR entries. |
3282 | compiler->AddCurrentDescriptor(PcDescriptorsLayout::kOsrEntry, |
3283 | instruction()->deopt_id(), |
3284 | TokenPosition::kNoSource); |
3285 | } |
3286 | compiler->pending_deoptimization_env_ = NULL; |
3287 | if (!using_shared_stub) { |
3288 | compiler->RestoreLiveRegisters(instruction()->locs()); |
3289 | } |
3290 | __ jmp(exit_label()); |
3291 | } |
3292 | |
3293 | compiler::Label* osr_entry_label() { |
3294 | ASSERT(Isolate::Current()->use_osr()); |
3295 | return &osr_entry_label_; |
3296 | } |
3297 | |
3298 | private: |
3299 | compiler::Label osr_entry_label_; |
3300 | }; |
3301 | |
3302 | void CheckStackOverflowInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3303 | CheckStackOverflowSlowPath* slow_path = new CheckStackOverflowSlowPath(this); |
3304 | compiler->AddSlowPathCode(slow_path); |
3305 | |
3306 | Register temp = locs()->temp(0).reg(); |
3307 | // Generate stack overflow check. |
3308 | __ cmpq(RSP, compiler::Address(THR, Thread::stack_limit_offset())); |
3309 | __ j(BELOW_EQUAL, slow_path->entry_label()); |
3310 | if (compiler->CanOSRFunction() && in_loop()) { |
3311 | // In unoptimized code check the usage counter to trigger OSR at loop |
3312 | // stack checks. Use progressively higher thresholds for more deeply |
3313 | // nested loops to attempt to hit outer loops with OSR when possible. |
3314 | __ LoadObject(temp, compiler->parsed_function().function()); |
3315 | int32_t threshold = |
3316 | FLAG_optimization_counter_threshold * (loop_depth() + 1); |
3317 | __ incl(compiler::FieldAddress(temp, Function::usage_counter_offset())); |
3318 | __ cmpl(compiler::FieldAddress(temp, Function::usage_counter_offset()), |
3319 | compiler::Immediate(threshold)); |
3320 | __ j(GREATER_EQUAL, slow_path->osr_entry_label()); |
3321 | } |
3322 | if (compiler->ForceSlowPathForStackOverflow()) { |
3323 | __ jmp(slow_path->entry_label()); |
3324 | } |
3325 | __ Bind(slow_path->exit_label()); |
3326 | } |
3327 | |
3328 | static void EmitSmiShiftLeft(FlowGraphCompiler* compiler, |
3329 | BinarySmiOpInstr* shift_left) { |
3330 | const LocationSummary& locs = *shift_left->locs(); |
3331 | Register left = locs.in(0).reg(); |
3332 | Register result = locs.out(0).reg(); |
3333 | ASSERT(left == result); |
3334 | compiler::Label* deopt = |
3335 | shift_left->CanDeoptimize() |
3336 | ? compiler->AddDeoptStub(shift_left->deopt_id(), |
3337 | ICData::kDeoptBinarySmiOp) |
3338 | : NULL; |
3339 | if (locs.in(1).IsConstant()) { |
3340 | const Object& constant = locs.in(1).constant(); |
3341 | ASSERT(constant.IsSmi()); |
3342 | // shlq operation masks the count to 6 bits. |
3343 | const intptr_t kCountLimit = 0x3F; |
3344 | const intptr_t value = Smi::Cast(constant).Value(); |
3345 | ASSERT((0 < value) && (value < kCountLimit)); |
3346 | if (shift_left->can_overflow()) { |
3347 | if (value == 1) { |
3348 | // Use overflow flag. |
3349 | __ shlq(left, compiler::Immediate(1)); |
3350 | __ j(OVERFLOW, deopt); |
3351 | return; |
3352 | } |
3353 | // Check for overflow. |
3354 | Register temp = locs.temp(0).reg(); |
3355 | __ movq(temp, left); |
3356 | __ shlq(left, compiler::Immediate(value)); |
3357 | __ sarq(left, compiler::Immediate(value)); |
3358 | __ cmpq(left, temp); |
3359 | __ j(NOT_EQUAL, deopt); // Overflow. |
3360 | } |
3361 | // Shift for result now we know there is no overflow. |
3362 | __ shlq(left, compiler::Immediate(value)); |
3363 | return; |
3364 | } |
3365 | |
3366 | // Right (locs.in(1)) is not constant. |
3367 | Register right = locs.in(1).reg(); |
3368 | Range* right_range = shift_left->right_range(); |
3369 | if (shift_left->left()->BindsToConstant() && shift_left->can_overflow()) { |
3370 | // TODO(srdjan): Implement code below for is_truncating(). |
3371 | // If left is constant, we know the maximal allowed size for right. |
3372 | const Object& obj = shift_left->left()->BoundConstant(); |
3373 | if (obj.IsSmi()) { |
3374 | const intptr_t left_int = Smi::Cast(obj).Value(); |
3375 | if (left_int == 0) { |
3376 | __ CompareImmediate(right, compiler::Immediate(0)); |
3377 | __ j(NEGATIVE, deopt); |
3378 | return; |
3379 | } |
3380 | const intptr_t max_right = kSmiBits - Utils::HighestBit(left_int); |
3381 | const bool right_needs_check = |
3382 | !RangeUtils::IsWithin(right_range, 0, max_right - 1); |
3383 | if (right_needs_check) { |
3384 | __ CompareImmediate( |
3385 | right, |
3386 | compiler::Immediate(static_cast<int64_t>(Smi::New(max_right)))); |
3387 | __ j(ABOVE_EQUAL, deopt); |
3388 | } |
3389 | __ SmiUntag(right); |
3390 | __ shlq(left, right); |
3391 | } |
3392 | return; |
3393 | } |
3394 | |
3395 | const bool right_needs_check = |
3396 | !RangeUtils::IsWithin(right_range, 0, (Smi::kBits - 1)); |
3397 | ASSERT(right == RCX); // Count must be in RCX |
3398 | if (!shift_left->can_overflow()) { |
3399 | if (right_needs_check) { |
3400 | const bool right_may_be_negative = |
3401 | (right_range == NULL) || !right_range->IsPositive(); |
3402 | if (right_may_be_negative) { |
3403 | ASSERT(shift_left->CanDeoptimize()); |
3404 | __ CompareImmediate(right, compiler::Immediate(0)); |
3405 | __ j(NEGATIVE, deopt); |
3406 | } |
3407 | compiler::Label done, is_not_zero; |
3408 | __ CompareImmediate( |
3409 | right, |
3410 | compiler::Immediate(static_cast<int64_t>(Smi::New(Smi::kBits)))); |
3411 | __ j(BELOW, &is_not_zero, compiler::Assembler::kNearJump); |
3412 | __ xorq(left, left); |
3413 | __ jmp(&done, compiler::Assembler::kNearJump); |
3414 | __ Bind(&is_not_zero); |
3415 | __ SmiUntag(right); |
3416 | __ shlq(left, right); |
3417 | __ Bind(&done); |
3418 | } else { |
3419 | __ SmiUntag(right); |
3420 | __ shlq(left, right); |
3421 | } |
3422 | } else { |
3423 | if (right_needs_check) { |
3424 | ASSERT(shift_left->CanDeoptimize()); |
3425 | __ CompareImmediate( |
3426 | right, |
3427 | compiler::Immediate(static_cast<int64_t>(Smi::New(Smi::kBits)))); |
3428 | __ j(ABOVE_EQUAL, deopt); |
3429 | } |
3430 | // Left is not a constant. |
3431 | Register temp = locs.temp(0).reg(); |
3432 | // Check if count too large for handling it inlined. |
3433 | __ movq(temp, left); |
3434 | __ SmiUntag(right); |
3435 | // Overflow test (preserve temp and right); |
3436 | __ shlq(left, right); |
3437 | __ sarq(left, right); |
3438 | __ cmpq(left, temp); |
3439 | __ j(NOT_EQUAL, deopt); // Overflow. |
3440 | // Shift for result now we know there is no overflow. |
3441 | __ shlq(left, right); |
3442 | } |
3443 | } |
3444 | |
3445 | class CheckedSmiSlowPath : public TemplateSlowPathCode<CheckedSmiOpInstr> { |
3446 | public: |
3447 | CheckedSmiSlowPath(CheckedSmiOpInstr* instruction, intptr_t try_index) |
3448 | : TemplateSlowPathCode(instruction), try_index_(try_index) {} |
3449 | |
3450 | static constexpr intptr_t kNumSlowPathArgs = 2; |
3451 | |
3452 | virtual void EmitNativeCode(FlowGraphCompiler* compiler) { |
3453 | if (compiler::Assembler::EmittingComments()) { |
3454 | __ Comment("slow path smi operation" ); |
3455 | } |
3456 | __ Bind(entry_label()); |
3457 | LocationSummary* locs = instruction()->locs(); |
3458 | Register result = locs->out(0).reg(); |
3459 | locs->live_registers()->Remove(Location::RegisterLocation(result)); |
3460 | |
3461 | compiler->SaveLiveRegisters(locs); |
3462 | if (instruction()->env() != NULL) { |
3463 | Environment* env = |
3464 | compiler->SlowPathEnvironmentFor(instruction(), kNumSlowPathArgs); |
3465 | compiler->pending_deoptimization_env_ = env; |
3466 | } |
3467 | __ pushq(locs->in(0).reg()); |
3468 | __ pushq(locs->in(1).reg()); |
3469 | const auto& selector = String::Handle(instruction()->call()->Selector()); |
3470 | const auto& arguments_descriptor = |
3471 | Array::Handle(ArgumentsDescriptor::NewBoxed( |
3472 | /*type_args_len=*/0, /*num_arguments=*/2)); |
3473 | compiler->EmitMegamorphicInstanceCall( |
3474 | selector, arguments_descriptor, instruction()->call()->deopt_id(), |
3475 | instruction()->token_pos(), locs, try_index_, kNumSlowPathArgs); |
3476 | __ MoveRegister(result, RAX); |
3477 | compiler->RestoreLiveRegisters(locs); |
3478 | __ jmp(exit_label()); |
3479 | compiler->pending_deoptimization_env_ = NULL; |
3480 | } |
3481 | |
3482 | private: |
3483 | intptr_t try_index_; |
3484 | }; |
3485 | |
3486 | LocationSummary* CheckedSmiOpInstr::MakeLocationSummary(Zone* zone, |
3487 | bool opt) const { |
3488 | bool is_shift = (op_kind() == Token::kSHL) || (op_kind() == Token::kSHR); |
3489 | const intptr_t kNumInputs = 2; |
3490 | const intptr_t kNumTemps = is_shift ? 1 : 0; |
3491 | LocationSummary* summary = new (zone) LocationSummary( |
3492 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
3493 | summary->set_in(0, Location::RequiresRegister()); |
3494 | summary->set_in(1, Location::RequiresRegister()); |
3495 | switch (op_kind()) { |
3496 | case Token::kADD: |
3497 | case Token::kSUB: |
3498 | case Token::kMUL: |
3499 | case Token::kSHL: |
3500 | case Token::kSHR: |
3501 | summary->set_out(0, Location::RequiresRegister()); |
3502 | break; |
3503 | case Token::kBIT_OR: |
3504 | case Token::kBIT_AND: |
3505 | case Token::kBIT_XOR: |
3506 | summary->set_out(0, Location::SameAsFirstInput()); |
3507 | break; |
3508 | default: |
3509 | UNIMPLEMENTED(); |
3510 | } |
3511 | if (is_shift) { |
3512 | summary->set_temp(0, Location::RegisterLocation(RCX)); |
3513 | } |
3514 | return summary; |
3515 | } |
3516 | |
3517 | void CheckedSmiOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3518 | CheckedSmiSlowPath* slow_path = |
3519 | new CheckedSmiSlowPath(this, compiler->CurrentTryIndex()); |
3520 | compiler->AddSlowPathCode(slow_path); |
3521 | // Test operands if necessary. |
3522 | |
3523 | intptr_t left_cid = left()->Type()->ToCid(); |
3524 | intptr_t right_cid = right()->Type()->ToCid(); |
3525 | Register left = locs()->in(0).reg(); |
3526 | Register right = locs()->in(1).reg(); |
3527 | if (this->left()->definition() == this->right()->definition()) { |
3528 | __ testq(left, compiler::Immediate(kSmiTagMask)); |
3529 | } else if (left_cid == kSmiCid) { |
3530 | __ testq(right, compiler::Immediate(kSmiTagMask)); |
3531 | } else if (right_cid == kSmiCid) { |
3532 | __ testq(left, compiler::Immediate(kSmiTagMask)); |
3533 | } else { |
3534 | __ movq(TMP, left); |
3535 | __ orq(TMP, right); |
3536 | __ testq(TMP, compiler::Immediate(kSmiTagMask)); |
3537 | } |
3538 | __ j(NOT_ZERO, slow_path->entry_label()); |
3539 | Register result = locs()->out(0).reg(); |
3540 | switch (op_kind()) { |
3541 | case Token::kADD: |
3542 | __ movq(result, left); |
3543 | __ addq(result, right); |
3544 | __ j(OVERFLOW, slow_path->entry_label()); |
3545 | break; |
3546 | case Token::kSUB: |
3547 | __ movq(result, left); |
3548 | __ subq(result, right); |
3549 | __ j(OVERFLOW, slow_path->entry_label()); |
3550 | break; |
3551 | case Token::kMUL: |
3552 | __ movq(result, left); |
3553 | __ SmiUntag(result); |
3554 | __ imulq(result, right); |
3555 | __ j(OVERFLOW, slow_path->entry_label()); |
3556 | break; |
3557 | case Token::kBIT_OR: |
3558 | ASSERT(left == result); |
3559 | __ orq(result, right); |
3560 | break; |
3561 | case Token::kBIT_AND: |
3562 | ASSERT(left == result); |
3563 | __ andq(result, right); |
3564 | break; |
3565 | case Token::kBIT_XOR: |
3566 | ASSERT(left == result); |
3567 | __ xorq(result, right); |
3568 | break; |
3569 | case Token::kSHL: |
3570 | ASSERT(result != right); |
3571 | ASSERT(locs()->temp(0).reg() == RCX); |
3572 | __ cmpq(right, compiler::Immediate(Smi::RawValue(Smi::kBits))); |
3573 | __ j(ABOVE_EQUAL, slow_path->entry_label()); |
3574 | |
3575 | __ movq(RCX, right); |
3576 | __ SmiUntag(RCX); |
3577 | __ movq(result, left); |
3578 | __ shlq(result, RCX); |
3579 | __ movq(TMP, result); |
3580 | __ sarq(TMP, RCX); |
3581 | __ cmpq(TMP, left); |
3582 | __ j(NOT_EQUAL, slow_path->entry_label()); |
3583 | break; |
3584 | case Token::kSHR: { |
3585 | compiler::Label shift_count_ok; |
3586 | ASSERT(result != right); |
3587 | ASSERT(locs()->temp(0).reg() == RCX); |
3588 | __ cmpq(right, compiler::Immediate(Smi::RawValue(Smi::kBits))); |
3589 | __ j(ABOVE_EQUAL, slow_path->entry_label()); |
3590 | |
3591 | __ movq(RCX, right); |
3592 | __ SmiUntag(RCX); |
3593 | __ movq(result, left); |
3594 | __ SmiUntag(result); |
3595 | __ sarq(result, RCX); |
3596 | __ SmiTag(result); |
3597 | break; |
3598 | } |
3599 | default: |
3600 | UNIMPLEMENTED(); |
3601 | } |
3602 | __ Bind(slow_path->exit_label()); |
3603 | } |
3604 | |
3605 | class CheckedSmiComparisonSlowPath |
3606 | : public TemplateSlowPathCode<CheckedSmiComparisonInstr> { |
3607 | public: |
3608 | static constexpr intptr_t kNumSlowPathArgs = 2; |
3609 | |
3610 | CheckedSmiComparisonSlowPath(CheckedSmiComparisonInstr* instruction, |
3611 | Environment* env, |
3612 | intptr_t try_index, |
3613 | BranchLabels labels, |
3614 | bool merged) |
3615 | : TemplateSlowPathCode(instruction), |
3616 | try_index_(try_index), |
3617 | labels_(labels), |
3618 | merged_(merged), |
3619 | env_(env) { |
3620 | // The environment must either come from the comparison or the environment |
3621 | // was cleared from the comparison (and moved to a branch). |
3622 | ASSERT(env == instruction->env() || |
3623 | (merged && instruction->env() == nullptr)); |
3624 | } |
3625 | |
3626 | virtual void EmitNativeCode(FlowGraphCompiler* compiler) { |
3627 | if (compiler::Assembler::EmittingComments()) { |
3628 | __ Comment("slow path smi comparison" ); |
3629 | } |
3630 | __ Bind(entry_label()); |
3631 | LocationSummary* locs = instruction()->locs(); |
3632 | Register result = merged_ ? locs->temp(0).reg() : locs->out(0).reg(); |
3633 | locs->live_registers()->Remove(Location::RegisterLocation(result)); |
3634 | |
3635 | compiler->SaveLiveRegisters(locs); |
3636 | if (env_ != nullptr) { |
3637 | compiler->pending_deoptimization_env_ = |
3638 | compiler->SlowPathEnvironmentFor(env_, locs, kNumSlowPathArgs); |
3639 | } |
3640 | __ pushq(locs->in(0).reg()); |
3641 | __ pushq(locs->in(1).reg()); |
3642 | |
3643 | const auto& selector = String::Handle(instruction()->call()->Selector()); |
3644 | const auto& arguments_descriptor = |
3645 | Array::Handle(ArgumentsDescriptor::NewBoxed( |
3646 | /*type_args_len=*/0, /*num_arguments=*/2)); |
3647 | |
3648 | compiler->EmitMegamorphicInstanceCall( |
3649 | selector, arguments_descriptor, instruction()->call()->deopt_id(), |
3650 | instruction()->token_pos(), locs, try_index_, kNumSlowPathArgs); |
3651 | __ MoveRegister(result, RAX); |
3652 | compiler->RestoreLiveRegisters(locs); |
3653 | compiler->pending_deoptimization_env_ = nullptr; |
3654 | if (merged_) { |
3655 | __ CompareObject(result, Bool::True()); |
3656 | __ j(EQUAL, instruction()->is_negated() ? labels_.false_label |
3657 | : labels_.true_label); |
3658 | __ jmp(instruction()->is_negated() ? labels_.true_label |
3659 | : labels_.false_label); |
3660 | ASSERT(exit_label()->IsUnused()); |
3661 | } else { |
3662 | ASSERT(!instruction()->is_negated()); |
3663 | __ jmp(exit_label()); |
3664 | } |
3665 | } |
3666 | |
3667 | private: |
3668 | intptr_t try_index_; |
3669 | BranchLabels labels_; |
3670 | bool merged_; |
3671 | Environment* env_; |
3672 | }; |
3673 | |
3674 | LocationSummary* CheckedSmiComparisonInstr::MakeLocationSummary( |
3675 | Zone* zone, |
3676 | bool opt) const { |
3677 | const intptr_t kNumInputs = 2; |
3678 | const intptr_t kNumTemps = 1; |
3679 | LocationSummary* summary = new (zone) LocationSummary( |
3680 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
3681 | summary->set_in(0, Location::RequiresRegister()); |
3682 | summary->set_in(1, Location::RequiresRegister()); |
3683 | summary->set_temp(0, Location::RequiresRegister()); |
3684 | summary->set_out(0, Location::RequiresRegister()); |
3685 | return summary; |
3686 | } |
3687 | |
3688 | Condition CheckedSmiComparisonInstr::EmitComparisonCode( |
3689 | FlowGraphCompiler* compiler, |
3690 | BranchLabels labels) { |
3691 | return EmitInt64ComparisonOp(compiler, *locs(), kind()); |
3692 | } |
3693 | |
3694 | #define EMIT_SMI_CHECK \ |
3695 | intptr_t left_cid = left()->Type()->ToCid(); \ |
3696 | intptr_t right_cid = right()->Type()->ToCid(); \ |
3697 | Register left = locs()->in(0).reg(); \ |
3698 | Register right = locs()->in(1).reg(); \ |
3699 | if (this->left()->definition() == this->right()->definition()) { \ |
3700 | __ testq(left, compiler::Immediate(kSmiTagMask)); \ |
3701 | } else if (left_cid == kSmiCid) { \ |
3702 | __ testq(right, compiler::Immediate(kSmiTagMask)); \ |
3703 | } else if (right_cid == kSmiCid) { \ |
3704 | __ testq(left, compiler::Immediate(kSmiTagMask)); \ |
3705 | } else { \ |
3706 | __ movq(TMP, left); \ |
3707 | __ orq(TMP, right); \ |
3708 | __ testq(TMP, compiler::Immediate(kSmiTagMask)); \ |
3709 | } \ |
3710 | __ j(NOT_ZERO, slow_path->entry_label()) |
3711 | |
3712 | void CheckedSmiComparisonInstr::EmitBranchCode(FlowGraphCompiler* compiler, |
3713 | BranchInstr* branch) { |
3714 | BranchLabels labels = compiler->CreateBranchLabels(branch); |
3715 | CheckedSmiComparisonSlowPath* slow_path = new CheckedSmiComparisonSlowPath( |
3716 | this, branch->env(), compiler->CurrentTryIndex(), labels, |
3717 | /* merged = */ true); |
3718 | compiler->AddSlowPathCode(slow_path); |
3719 | EMIT_SMI_CHECK; |
3720 | Condition true_condition = EmitComparisonCode(compiler, labels); |
3721 | ASSERT(true_condition != kInvalidCondition); |
3722 | EmitBranchOnCondition(compiler, true_condition, labels); |
3723 | // No need to bind slow_path->exit_label() as slow path exits through |
3724 | // true/false branch labels. |
3725 | } |
3726 | |
3727 | void CheckedSmiComparisonInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3728 | // Zone-allocate labels to pass them to slow-path which outlives local scope. |
3729 | compiler::Label* true_label = new (Z) compiler::Label(); |
3730 | compiler::Label* false_label = new (Z) compiler::Label(); |
3731 | compiler::Label done; |
3732 | BranchLabels labels = {true_label, false_label, false_label}; |
3733 | // In case of negated comparison result of a slow path call should be negated. |
3734 | // For this purpose, 'merged' slow path is generated: it tests |
3735 | // result of a call and jumps directly to true or false label. |
3736 | CheckedSmiComparisonSlowPath* slow_path = new CheckedSmiComparisonSlowPath( |
3737 | this, env(), compiler->CurrentTryIndex(), labels, |
3738 | /* merged = */ is_negated()); |
3739 | compiler->AddSlowPathCode(slow_path); |
3740 | EMIT_SMI_CHECK; |
3741 | Condition true_condition = EmitComparisonCode(compiler, labels); |
3742 | ASSERT(true_condition != kInvalidCondition); |
3743 | EmitBranchOnCondition(compiler, true_condition, labels); |
3744 | Register result = locs()->out(0).reg(); |
3745 | __ Bind(false_label); |
3746 | __ LoadObject(result, Bool::False()); |
3747 | __ jmp(&done); |
3748 | __ Bind(true_label); |
3749 | __ LoadObject(result, Bool::True()); |
3750 | __ Bind(&done); |
3751 | // In case of negated comparison slow path exits through true/false labels. |
3752 | if (!is_negated()) { |
3753 | __ Bind(slow_path->exit_label()); |
3754 | } |
3755 | } |
3756 | |
3757 | static bool CanBeImmediate(const Object& constant) { |
3758 | return constant.IsSmi() && |
3759 | compiler::Immediate(static_cast<int64_t>(constant.raw())).is_int32(); |
3760 | } |
3761 | |
3762 | static bool IsSmiValue(const Object& constant, intptr_t value) { |
3763 | return constant.IsSmi() && (Smi::Cast(constant).Value() == value); |
3764 | } |
3765 | |
3766 | LocationSummary* BinarySmiOpInstr::MakeLocationSummary(Zone* zone, |
3767 | bool opt) const { |
3768 | const intptr_t kNumInputs = 2; |
3769 | |
3770 | ConstantInstr* right_constant = right()->definition()->AsConstant(); |
3771 | if ((right_constant != NULL) && (op_kind() != Token::kTRUNCDIV) && |
3772 | (op_kind() != Token::kSHL) && (op_kind() != Token::kMUL) && |
3773 | (op_kind() != Token::kMOD) && CanBeImmediate(right_constant->value())) { |
3774 | const intptr_t kNumTemps = 0; |
3775 | LocationSummary* summary = new (zone) |
3776 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
3777 | summary->set_in(0, Location::RequiresRegister()); |
3778 | summary->set_in(1, Location::Constant(right_constant)); |
3779 | summary->set_out(0, Location::SameAsFirstInput()); |
3780 | return summary; |
3781 | } |
3782 | |
3783 | if (op_kind() == Token::kTRUNCDIV) { |
3784 | const intptr_t kNumTemps = 1; |
3785 | LocationSummary* summary = new (zone) |
3786 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
3787 | if (RightIsPowerOfTwoConstant()) { |
3788 | summary->set_in(0, Location::RequiresRegister()); |
3789 | ConstantInstr* right_constant = right()->definition()->AsConstant(); |
3790 | summary->set_in(1, Location::Constant(right_constant)); |
3791 | summary->set_temp(0, Location::RequiresRegister()); |
3792 | summary->set_out(0, Location::SameAsFirstInput()); |
3793 | } else { |
3794 | // Both inputs must be writable because they will be untagged. |
3795 | summary->set_in(0, Location::RegisterLocation(RAX)); |
3796 | summary->set_in(1, Location::WritableRegister()); |
3797 | summary->set_out(0, Location::SameAsFirstInput()); |
3798 | // Will be used for sign extension and division. |
3799 | summary->set_temp(0, Location::RegisterLocation(RDX)); |
3800 | } |
3801 | return summary; |
3802 | } else if (op_kind() == Token::kMOD) { |
3803 | const intptr_t kNumTemps = 1; |
3804 | LocationSummary* summary = new (zone) |
3805 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
3806 | // Both inputs must be writable because they will be untagged. |
3807 | summary->set_in(0, Location::RegisterLocation(RDX)); |
3808 | summary->set_in(1, Location::WritableRegister()); |
3809 | summary->set_out(0, Location::SameAsFirstInput()); |
3810 | // Will be used for sign extension and division. |
3811 | summary->set_temp(0, Location::RegisterLocation(RAX)); |
3812 | return summary; |
3813 | } else if (op_kind() == Token::kSHR) { |
3814 | const intptr_t kNumTemps = 0; |
3815 | LocationSummary* summary = new (zone) |
3816 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
3817 | summary->set_in(0, Location::RequiresRegister()); |
3818 | summary->set_in(1, LocationFixedRegisterOrSmiConstant(right(), RCX)); |
3819 | summary->set_out(0, Location::SameAsFirstInput()); |
3820 | return summary; |
3821 | } else if (op_kind() == Token::kSHL) { |
3822 | // Shift-by-1 overflow checking can use flags, otherwise we need a temp. |
3823 | const bool shiftBy1 = |
3824 | (right_constant != NULL) && IsSmiValue(right_constant->value(), 1); |
3825 | const intptr_t kNumTemps = (can_overflow() && !shiftBy1) ? 1 : 0; |
3826 | LocationSummary* summary = new (zone) |
3827 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
3828 | summary->set_in(0, Location::RequiresRegister()); |
3829 | summary->set_in(1, LocationFixedRegisterOrSmiConstant(right(), RCX)); |
3830 | if (kNumTemps == 1) { |
3831 | summary->set_temp(0, Location::RequiresRegister()); |
3832 | } |
3833 | summary->set_out(0, Location::SameAsFirstInput()); |
3834 | return summary; |
3835 | } else { |
3836 | const intptr_t kNumTemps = 0; |
3837 | LocationSummary* summary = new (zone) |
3838 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
3839 | summary->set_in(0, Location::RequiresRegister()); |
3840 | ConstantInstr* constant = right()->definition()->AsConstant(); |
3841 | if (constant != NULL) { |
3842 | summary->set_in(1, LocationRegisterOrSmiConstant(right())); |
3843 | } else { |
3844 | summary->set_in(1, Location::PrefersRegister()); |
3845 | } |
3846 | summary->set_out(0, Location::SameAsFirstInput()); |
3847 | return summary; |
3848 | } |
3849 | } |
3850 | |
3851 | void BinarySmiOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
3852 | if (op_kind() == Token::kSHL) { |
3853 | EmitSmiShiftLeft(compiler, this); |
3854 | return; |
3855 | } |
3856 | |
3857 | Register left = locs()->in(0).reg(); |
3858 | Register result = locs()->out(0).reg(); |
3859 | ASSERT(left == result); |
3860 | compiler::Label* deopt = NULL; |
3861 | if (CanDeoptimize()) { |
3862 | deopt = compiler->AddDeoptStub(deopt_id(), ICData::kDeoptBinarySmiOp); |
3863 | } |
3864 | |
3865 | if (locs()->in(1).IsConstant()) { |
3866 | const Object& constant = locs()->in(1).constant(); |
3867 | ASSERT(constant.IsSmi()); |
3868 | const int64_t imm = static_cast<int64_t>(constant.raw()); |
3869 | switch (op_kind()) { |
3870 | case Token::kADD: { |
3871 | __ AddImmediate(left, compiler::Immediate(imm)); |
3872 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3873 | break; |
3874 | } |
3875 | case Token::kSUB: { |
3876 | __ SubImmediate(left, compiler::Immediate(imm)); |
3877 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3878 | break; |
3879 | } |
3880 | case Token::kMUL: { |
3881 | // Keep left value tagged and untag right value. |
3882 | const intptr_t value = Smi::Cast(constant).Value(); |
3883 | __ MulImmediate(left, compiler::Immediate(value)); |
3884 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3885 | break; |
3886 | } |
3887 | case Token::kTRUNCDIV: { |
3888 | const intptr_t value = Smi::Cast(constant).Value(); |
3889 | ASSERT(value != kIntptrMin); |
3890 | ASSERT(Utils::IsPowerOfTwo(Utils::Abs(value))); |
3891 | const intptr_t shift_count = |
3892 | Utils::ShiftForPowerOfTwo(Utils::Abs(value)) + kSmiTagSize; |
3893 | ASSERT(kSmiTagSize == 1); |
3894 | Register temp = locs()->temp(0).reg(); |
3895 | __ movq(temp, left); |
3896 | __ sarq(temp, compiler::Immediate(63)); |
3897 | ASSERT(shift_count > 1); // 1, -1 case handled above. |
3898 | __ shrq(temp, compiler::Immediate(64 - shift_count)); |
3899 | __ addq(left, temp); |
3900 | ASSERT(shift_count > 0); |
3901 | __ sarq(left, compiler::Immediate(shift_count)); |
3902 | if (value < 0) { |
3903 | __ negq(left); |
3904 | } |
3905 | __ SmiTag(left); |
3906 | break; |
3907 | } |
3908 | case Token::kBIT_AND: { |
3909 | // No overflow check. |
3910 | __ AndImmediate(left, compiler::Immediate(imm)); |
3911 | break; |
3912 | } |
3913 | case Token::kBIT_OR: { |
3914 | // No overflow check. |
3915 | __ OrImmediate(left, compiler::Immediate(imm)); |
3916 | break; |
3917 | } |
3918 | case Token::kBIT_XOR: { |
3919 | // No overflow check. |
3920 | __ XorImmediate(left, compiler::Immediate(imm)); |
3921 | break; |
3922 | } |
3923 | |
3924 | case Token::kSHR: { |
3925 | // sarq operation masks the count to 6 bits. |
3926 | const intptr_t kCountLimit = 0x3F; |
3927 | const intptr_t value = Smi::Cast(constant).Value(); |
3928 | __ sarq(left, compiler::Immediate( |
3929 | Utils::Minimum(value + kSmiTagSize, kCountLimit))); |
3930 | __ SmiTag(left); |
3931 | break; |
3932 | } |
3933 | |
3934 | default: |
3935 | UNREACHABLE(); |
3936 | break; |
3937 | } |
3938 | return; |
3939 | } // locs()->in(1).IsConstant(). |
3940 | |
3941 | if (locs()->in(1).IsStackSlot()) { |
3942 | const compiler::Address& right = LocationToStackSlotAddress(locs()->in(1)); |
3943 | switch (op_kind()) { |
3944 | case Token::kADD: { |
3945 | __ addq(left, right); |
3946 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3947 | break; |
3948 | } |
3949 | case Token::kSUB: { |
3950 | __ subq(left, right); |
3951 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3952 | break; |
3953 | } |
3954 | case Token::kMUL: { |
3955 | __ SmiUntag(left); |
3956 | __ imulq(left, right); |
3957 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3958 | break; |
3959 | } |
3960 | case Token::kBIT_AND: { |
3961 | // No overflow check. |
3962 | __ andq(left, right); |
3963 | break; |
3964 | } |
3965 | case Token::kBIT_OR: { |
3966 | // No overflow check. |
3967 | __ orq(left, right); |
3968 | break; |
3969 | } |
3970 | case Token::kBIT_XOR: { |
3971 | // No overflow check. |
3972 | __ xorq(left, right); |
3973 | break; |
3974 | } |
3975 | default: |
3976 | UNREACHABLE(); |
3977 | break; |
3978 | } |
3979 | return; |
3980 | } // locs()->in(1).IsStackSlot(). |
3981 | |
3982 | // if locs()->in(1).IsRegister. |
3983 | Register right = locs()->in(1).reg(); |
3984 | switch (op_kind()) { |
3985 | case Token::kADD: { |
3986 | __ addq(left, right); |
3987 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3988 | break; |
3989 | } |
3990 | case Token::kSUB: { |
3991 | __ subq(left, right); |
3992 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3993 | break; |
3994 | } |
3995 | case Token::kMUL: { |
3996 | __ SmiUntag(left); |
3997 | __ imulq(left, right); |
3998 | if (deopt != NULL) __ j(OVERFLOW, deopt); |
3999 | break; |
4000 | } |
4001 | case Token::kBIT_AND: { |
4002 | // No overflow check. |
4003 | __ andq(left, right); |
4004 | break; |
4005 | } |
4006 | case Token::kBIT_OR: { |
4007 | // No overflow check. |
4008 | __ orq(left, right); |
4009 | break; |
4010 | } |
4011 | case Token::kBIT_XOR: { |
4012 | // No overflow check. |
4013 | __ xorq(left, right); |
4014 | break; |
4015 | } |
4016 | case Token::kTRUNCDIV: { |
4017 | compiler::Label not_32bit, done; |
4018 | |
4019 | Register temp = locs()->temp(0).reg(); |
4020 | ASSERT(left == RAX); |
4021 | ASSERT((right != RDX) && (right != RAX)); |
4022 | ASSERT(temp == RDX); |
4023 | ASSERT(result == RAX); |
4024 | if (RangeUtils::CanBeZero(right_range())) { |
4025 | // Handle divide by zero in runtime. |
4026 | __ testq(right, right); |
4027 | __ j(ZERO, deopt); |
4028 | } |
4029 | // Check if both operands fit into 32bits as idiv with 64bit operands |
4030 | // requires twice as many cycles and has much higher latency. |
4031 | // We are checking this before untagging them to avoid corner case |
4032 | // dividing INT_MAX by -1 that raises exception because quotient is |
4033 | // too large for 32bit register. |
4034 | __ movsxd(temp, left); |
4035 | __ cmpq(temp, left); |
4036 | __ j(NOT_EQUAL, ¬_32bit); |
4037 | __ movsxd(temp, right); |
4038 | __ cmpq(temp, right); |
4039 | __ j(NOT_EQUAL, ¬_32bit); |
4040 | |
4041 | // Both operands are 31bit smis. Divide using 32bit idiv. |
4042 | __ SmiUntag(left); |
4043 | __ SmiUntag(right); |
4044 | __ cdq(); |
4045 | __ idivl(right); |
4046 | __ movsxd(result, result); |
4047 | __ jmp(&done); |
4048 | |
4049 | // Divide using 64bit idiv. |
4050 | __ Bind(¬_32bit); |
4051 | __ SmiUntag(left); |
4052 | __ SmiUntag(right); |
4053 | __ cqo(); // Sign extend RAX -> RDX:RAX. |
4054 | __ idivq(right); // RAX: quotient, RDX: remainder. |
4055 | if (RangeUtils::Overlaps(right_range(), -1, -1)) { |
4056 | // Check the corner case of dividing the 'MIN_SMI' with -1, in which |
4057 | // case we cannot tag the result. |
4058 | __ CompareImmediate(result, compiler::Immediate(0x4000000000000000)); |
4059 | __ j(EQUAL, deopt); |
4060 | } |
4061 | __ Bind(&done); |
4062 | __ SmiTag(result); |
4063 | break; |
4064 | } |
4065 | case Token::kMOD: { |
4066 | compiler::Label not_32bit, div_done; |
4067 | |
4068 | Register temp = locs()->temp(0).reg(); |
4069 | ASSERT(left == RDX); |
4070 | ASSERT((right != RDX) && (right != RAX)); |
4071 | ASSERT(temp == RAX); |
4072 | ASSERT(result == RDX); |
4073 | if (RangeUtils::CanBeZero(right_range())) { |
4074 | // Handle divide by zero in runtime. |
4075 | __ testq(right, right); |
4076 | __ j(ZERO, deopt); |
4077 | } |
4078 | // Check if both operands fit into 32bits as idiv with 64bit operands |
4079 | // requires twice as many cycles and has much higher latency. |
4080 | // We are checking this before untagging them to avoid corner case |
4081 | // dividing INT_MAX by -1 that raises exception because quotient is |
4082 | // too large for 32bit register. |
4083 | __ movsxd(temp, left); |
4084 | __ cmpq(temp, left); |
4085 | __ j(NOT_EQUAL, ¬_32bit); |
4086 | __ movsxd(temp, right); |
4087 | __ cmpq(temp, right); |
4088 | __ j(NOT_EQUAL, ¬_32bit); |
4089 | // Both operands are 31bit smis. Divide using 32bit idiv. |
4090 | __ SmiUntag(left); |
4091 | __ SmiUntag(right); |
4092 | __ movq(RAX, RDX); |
4093 | __ cdq(); |
4094 | __ idivl(right); |
4095 | __ movsxd(result, result); |
4096 | __ jmp(&div_done); |
4097 | |
4098 | // Divide using 64bit idiv. |
4099 | __ Bind(¬_32bit); |
4100 | __ SmiUntag(left); |
4101 | __ SmiUntag(right); |
4102 | __ movq(RAX, RDX); |
4103 | __ cqo(); // Sign extend RAX -> RDX:RAX. |
4104 | __ idivq(right); // RAX: quotient, RDX: remainder. |
4105 | __ Bind(&div_done); |
4106 | // res = left % right; |
4107 | // if (res < 0) { |
4108 | // if (right < 0) { |
4109 | // res = res - right; |
4110 | // } else { |
4111 | // res = res + right; |
4112 | // } |
4113 | // } |
4114 | compiler::Label all_done; |
4115 | __ cmpq(result, compiler::Immediate(0)); |
4116 | __ j(GREATER_EQUAL, &all_done, compiler::Assembler::kNearJump); |
4117 | // Result is negative, adjust it. |
4118 | if (RangeUtils::Overlaps(right_range(), -1, 1)) { |
4119 | compiler::Label subtract; |
4120 | __ cmpq(right, compiler::Immediate(0)); |
4121 | __ j(LESS, &subtract, compiler::Assembler::kNearJump); |
4122 | __ addq(result, right); |
4123 | __ jmp(&all_done, compiler::Assembler::kNearJump); |
4124 | __ Bind(&subtract); |
4125 | __ subq(result, right); |
4126 | } else if (right_range()->IsPositive()) { |
4127 | // Right is positive. |
4128 | __ addq(result, right); |
4129 | } else { |
4130 | // Right is negative. |
4131 | __ subq(result, right); |
4132 | } |
4133 | __ Bind(&all_done); |
4134 | __ SmiTag(result); |
4135 | break; |
4136 | } |
4137 | case Token::kSHR: { |
4138 | if (CanDeoptimize()) { |
4139 | __ CompareImmediate(right, compiler::Immediate(0)); |
4140 | __ j(LESS, deopt); |
4141 | } |
4142 | __ SmiUntag(right); |
4143 | // sarq operation masks the count to 6 bits. |
4144 | const intptr_t kCountLimit = 0x3F; |
4145 | if (!RangeUtils::OnlyLessThanOrEqualTo(right_range(), kCountLimit)) { |
4146 | __ CompareImmediate(right, compiler::Immediate(kCountLimit)); |
4147 | compiler::Label count_ok; |
4148 | __ j(LESS, &count_ok, compiler::Assembler::kNearJump); |
4149 | __ LoadImmediate(right, compiler::Immediate(kCountLimit)); |
4150 | __ Bind(&count_ok); |
4151 | } |
4152 | ASSERT(right == RCX); // Count must be in RCX |
4153 | __ SmiUntag(left); |
4154 | __ sarq(left, right); |
4155 | __ SmiTag(left); |
4156 | break; |
4157 | } |
4158 | case Token::kDIV: { |
4159 | // Dispatches to 'Double./'. |
4160 | // TODO(srdjan): Implement as conversion to double and double division. |
4161 | UNREACHABLE(); |
4162 | break; |
4163 | } |
4164 | case Token::kOR: |
4165 | case Token::kAND: { |
4166 | // Flow graph builder has dissected this operation to guarantee correct |
4167 | // behavior (short-circuit evaluation). |
4168 | UNREACHABLE(); |
4169 | break; |
4170 | } |
4171 | default: |
4172 | UNREACHABLE(); |
4173 | break; |
4174 | } |
4175 | } |
4176 | |
4177 | LocationSummary* CheckEitherNonSmiInstr::MakeLocationSummary(Zone* zone, |
4178 | bool opt) const { |
4179 | intptr_t left_cid = left()->Type()->ToCid(); |
4180 | intptr_t right_cid = right()->Type()->ToCid(); |
4181 | ASSERT((left_cid != kDoubleCid) && (right_cid != kDoubleCid)); |
4182 | const intptr_t kNumInputs = 2; |
4183 | const bool need_temp = (left()->definition() != right()->definition()) && |
4184 | (left_cid != kSmiCid) && (right_cid != kSmiCid); |
4185 | const intptr_t kNumTemps = need_temp ? 1 : 0; |
4186 | LocationSummary* summary = new (zone) |
4187 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
4188 | summary->set_in(0, Location::RequiresRegister()); |
4189 | summary->set_in(1, Location::RequiresRegister()); |
4190 | if (need_temp) summary->set_temp(0, Location::RequiresRegister()); |
4191 | return summary; |
4192 | } |
4193 | |
4194 | void CheckEitherNonSmiInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4195 | compiler::Label* deopt = |
4196 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptBinaryDoubleOp, |
4197 | licm_hoisted_ ? ICData::kHoisted : 0); |
4198 | intptr_t left_cid = left()->Type()->ToCid(); |
4199 | intptr_t right_cid = right()->Type()->ToCid(); |
4200 | Register left = locs()->in(0).reg(); |
4201 | Register right = locs()->in(1).reg(); |
4202 | if (this->left()->definition() == this->right()->definition()) { |
4203 | __ testq(left, compiler::Immediate(kSmiTagMask)); |
4204 | } else if (left_cid == kSmiCid) { |
4205 | __ testq(right, compiler::Immediate(kSmiTagMask)); |
4206 | } else if (right_cid == kSmiCid) { |
4207 | __ testq(left, compiler::Immediate(kSmiTagMask)); |
4208 | } else { |
4209 | Register temp = locs()->temp(0).reg(); |
4210 | __ movq(temp, left); |
4211 | __ orq(temp, right); |
4212 | __ testq(temp, compiler::Immediate(kSmiTagMask)); |
4213 | } |
4214 | __ j(ZERO, deopt); |
4215 | } |
4216 | |
4217 | LocationSummary* BoxInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
4218 | const intptr_t kNumInputs = 1; |
4219 | const intptr_t kNumTemps = 1; |
4220 | LocationSummary* summary = new (zone) LocationSummary( |
4221 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
4222 | summary->set_in(0, Location::RequiresFpuRegister()); |
4223 | summary->set_temp(0, Location::RequiresRegister()); |
4224 | summary->set_out(0, Location::RequiresRegister()); |
4225 | return summary; |
4226 | } |
4227 | |
4228 | void BoxInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4229 | Register out_reg = locs()->out(0).reg(); |
4230 | Register temp = locs()->temp(0).reg(); |
4231 | XmmRegister value = locs()->in(0).fpu_reg(); |
4232 | |
4233 | BoxAllocationSlowPath::Allocate(compiler, this, |
4234 | compiler->BoxClassFor(from_representation()), |
4235 | out_reg, temp); |
4236 | |
4237 | switch (from_representation()) { |
4238 | case kUnboxedDouble: |
4239 | __ movsd(compiler::FieldAddress(out_reg, ValueOffset()), value); |
4240 | break; |
4241 | case kUnboxedFloat: { |
4242 | __ cvtss2sd(FpuTMP, value); |
4243 | __ movsd(compiler::FieldAddress(out_reg, ValueOffset()), FpuTMP); |
4244 | break; |
4245 | } |
4246 | case kUnboxedFloat32x4: |
4247 | case kUnboxedFloat64x2: |
4248 | case kUnboxedInt32x4: |
4249 | __ movups(compiler::FieldAddress(out_reg, ValueOffset()), value); |
4250 | break; |
4251 | default: |
4252 | UNREACHABLE(); |
4253 | break; |
4254 | } |
4255 | } |
4256 | |
4257 | LocationSummary* UnboxInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
4258 | const intptr_t kNumInputs = 1; |
4259 | const intptr_t kNumTemps = 0; |
4260 | const bool needs_writable_input = |
4261 | (representation() != kUnboxedInt64) && |
4262 | (value()->Type()->ToNullableCid() != BoxCid()); |
4263 | LocationSummary* summary = new (zone) |
4264 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
4265 | summary->set_in(0, needs_writable_input ? Location::WritableRegister() |
4266 | : Location::RequiresRegister()); |
4267 | if (representation() == kUnboxedInt64 || representation() == kUnboxedInt32) { |
4268 | summary->set_out(0, Location::SameAsFirstInput()); |
4269 | } else { |
4270 | summary->set_out(0, Location::RequiresFpuRegister()); |
4271 | } |
4272 | return summary; |
4273 | } |
4274 | |
4275 | void UnboxInstr::EmitLoadFromBox(FlowGraphCompiler* compiler) { |
4276 | const Register box = locs()->in(0).reg(); |
4277 | |
4278 | switch (representation()) { |
4279 | case kUnboxedInt64: { |
4280 | const Register result = locs()->out(0).reg(); |
4281 | __ movq(result, compiler::FieldAddress(box, ValueOffset())); |
4282 | break; |
4283 | } |
4284 | |
4285 | case kUnboxedDouble: { |
4286 | const FpuRegister result = locs()->out(0).fpu_reg(); |
4287 | __ movsd(result, compiler::FieldAddress(box, ValueOffset())); |
4288 | break; |
4289 | } |
4290 | |
4291 | case kUnboxedFloat: { |
4292 | const FpuRegister result = locs()->out(0).fpu_reg(); |
4293 | __ movsd(result, compiler::FieldAddress(box, ValueOffset())); |
4294 | __ cvtsd2ss(result, result); |
4295 | break; |
4296 | } |
4297 | |
4298 | case kUnboxedFloat32x4: |
4299 | case kUnboxedFloat64x2: |
4300 | case kUnboxedInt32x4: { |
4301 | const FpuRegister result = locs()->out(0).fpu_reg(); |
4302 | __ movups(result, compiler::FieldAddress(box, ValueOffset())); |
4303 | break; |
4304 | } |
4305 | |
4306 | default: |
4307 | UNREACHABLE(); |
4308 | break; |
4309 | } |
4310 | } |
4311 | |
4312 | void UnboxInstr::EmitSmiConversion(FlowGraphCompiler* compiler) { |
4313 | const Register box = locs()->in(0).reg(); |
4314 | |
4315 | switch (representation()) { |
4316 | case kUnboxedInt64: { |
4317 | const Register result = locs()->out(0).reg(); |
4318 | ASSERT(result == box); |
4319 | __ SmiUntag(box); |
4320 | break; |
4321 | } |
4322 | |
4323 | case kUnboxedDouble: { |
4324 | const FpuRegister result = locs()->out(0).fpu_reg(); |
4325 | __ SmiUntag(box); |
4326 | __ cvtsi2sdq(result, box); |
4327 | break; |
4328 | } |
4329 | |
4330 | default: |
4331 | UNREACHABLE(); |
4332 | break; |
4333 | } |
4334 | } |
4335 | |
4336 | void UnboxInstr::EmitLoadInt32FromBoxOrSmi(FlowGraphCompiler* compiler) { |
4337 | const Register value = locs()->in(0).reg(); |
4338 | const Register result = locs()->out(0).reg(); |
4339 | ASSERT(value == result); |
4340 | compiler::Label done; |
4341 | __ SmiUntag(value); |
4342 | __ j(NOT_CARRY, &done, compiler::Assembler::kNearJump); |
4343 | __ movsxw(result, compiler::Address(value, TIMES_2, Mint::value_offset())); |
4344 | __ Bind(&done); |
4345 | } |
4346 | |
4347 | void UnboxInstr::EmitLoadInt64FromBoxOrSmi(FlowGraphCompiler* compiler) { |
4348 | const Register value = locs()->in(0).reg(); |
4349 | const Register result = locs()->out(0).reg(); |
4350 | ASSERT(value == result); |
4351 | compiler::Label done; |
4352 | __ SmiUntag(value); |
4353 | __ j(NOT_CARRY, &done, compiler::Assembler::kNearJump); |
4354 | __ movq(value, compiler::Address(value, TIMES_2, Mint::value_offset())); |
4355 | __ Bind(&done); |
4356 | } |
4357 | |
4358 | LocationSummary* UnboxInteger32Instr::MakeLocationSummary(Zone* zone, |
4359 | bool opt) const { |
4360 | const intptr_t kNumInputs = 1; |
4361 | const intptr_t kNumTemps = (!is_truncating() && CanDeoptimize()) ? 1 : 0; |
4362 | LocationSummary* summary = new (zone) |
4363 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
4364 | summary->set_in(0, Location::RequiresRegister()); |
4365 | summary->set_out(0, Location::SameAsFirstInput()); |
4366 | if (kNumTemps > 0) { |
4367 | summary->set_temp(0, Location::RequiresRegister()); |
4368 | } |
4369 | return summary; |
4370 | } |
4371 | |
4372 | void UnboxInteger32Instr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4373 | const intptr_t value_cid = value()->Type()->ToCid(); |
4374 | const Register value = locs()->in(0).reg(); |
4375 | compiler::Label* deopt = |
4376 | CanDeoptimize() |
4377 | ? compiler->AddDeoptStub(GetDeoptId(), ICData::kDeoptUnboxInteger) |
4378 | : NULL; |
4379 | ASSERT(value == locs()->out(0).reg()); |
4380 | |
4381 | if (value_cid == kSmiCid) { |
4382 | __ SmiUntag(value); |
4383 | } else if (value_cid == kMintCid) { |
4384 | __ movq(value, compiler::FieldAddress(value, Mint::value_offset())); |
4385 | } else if (!CanDeoptimize()) { |
4386 | // Type information is not conclusive, but range analysis found |
4387 | // the value to be in int64 range. Therefore it must be a smi |
4388 | // or mint value. |
4389 | ASSERT(is_truncating()); |
4390 | compiler::Label done; |
4391 | __ SmiUntag(value); |
4392 | __ j(NOT_CARRY, &done, compiler::Assembler::kNearJump); |
4393 | __ movq(value, compiler::Address(value, TIMES_2, Mint::value_offset())); |
4394 | __ Bind(&done); |
4395 | return; |
4396 | } else { |
4397 | compiler::Label done; |
4398 | // Optimistically untag value. |
4399 | __ SmiUntagOrCheckClass(value, kMintCid, &done); |
4400 | __ j(NOT_EQUAL, deopt); |
4401 | // Undo untagging by multiplying value with 2. |
4402 | __ movq(value, compiler::Address(value, TIMES_2, Mint::value_offset())); |
4403 | __ Bind(&done); |
4404 | } |
4405 | |
4406 | // TODO(vegorov): as it is implemented right now truncating unboxing would |
4407 | // leave "garbage" in the higher word. |
4408 | if (!is_truncating() && (deopt != NULL)) { |
4409 | ASSERT(representation() == kUnboxedInt32); |
4410 | Register temp = locs()->temp(0).reg(); |
4411 | __ movsxd(temp, value); |
4412 | __ cmpq(temp, value); |
4413 | __ j(NOT_EQUAL, deopt); |
4414 | } |
4415 | } |
4416 | |
4417 | LocationSummary* BoxInteger32Instr::MakeLocationSummary(Zone* zone, |
4418 | bool opt) const { |
4419 | ASSERT((from_representation() == kUnboxedInt32) || |
4420 | (from_representation() == kUnboxedUint32)); |
4421 | const intptr_t kNumInputs = 1; |
4422 | const intptr_t kNumTemps = 0; |
4423 | LocationSummary* summary = new (zone) |
4424 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
4425 | summary->set_in(0, Location::RequiresRegister()); |
4426 | summary->set_out(0, Location::RequiresRegister()); |
4427 | return summary; |
4428 | } |
4429 | |
4430 | void BoxInteger32Instr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4431 | const Register value = locs()->in(0).reg(); |
4432 | const Register out = locs()->out(0).reg(); |
4433 | ASSERT(value != out); |
4434 | |
4435 | ASSERT(kSmiTagSize == 1); |
4436 | if (from_representation() == kUnboxedInt32) { |
4437 | __ movsxd(out, value); |
4438 | } else { |
4439 | ASSERT(from_representation() == kUnboxedUint32); |
4440 | __ movl(out, value); |
4441 | } |
4442 | __ SmiTag(out); |
4443 | } |
4444 | |
4445 | LocationSummary* BoxInt64Instr::MakeLocationSummary(Zone* zone, |
4446 | bool opt) const { |
4447 | const intptr_t kNumInputs = 1; |
4448 | const intptr_t kNumTemps = ValueFitsSmi() ? 0 : 1; |
4449 | // Shared slow path is used in BoxInt64Instr::EmitNativeCode in |
4450 | // FLAG_use_bare_instructions mode and only after VM isolate stubs where |
4451 | // replaced with isolate-specific stubs. |
4452 | auto object_store = Isolate::Current()->object_store(); |
4453 | const bool stubs_in_vm_isolate = |
4454 | object_store->allocate_mint_with_fpu_regs_stub() |
4455 | ->ptr() |
4456 | ->InVMIsolateHeap() || |
4457 | object_store->allocate_mint_without_fpu_regs_stub() |
4458 | ->ptr() |
4459 | ->InVMIsolateHeap(); |
4460 | const bool shared_slow_path_call = SlowPathSharingSupported(opt) && |
4461 | FLAG_use_bare_instructions && |
4462 | !stubs_in_vm_isolate; |
4463 | LocationSummary* summary = new (zone) LocationSummary( |
4464 | zone, kNumInputs, kNumTemps, |
4465 | ValueFitsSmi() |
4466 | ? LocationSummary::kNoCall |
4467 | : ((shared_slow_path_call ? LocationSummary::kCallOnSharedSlowPath |
4468 | : LocationSummary::kCallOnSlowPath))); |
4469 | summary->set_in(0, Location::RequiresRegister()); |
4470 | if (ValueFitsSmi()) { |
4471 | summary->set_out(0, Location::RequiresRegister()); |
4472 | } else if (shared_slow_path_call) { |
4473 | summary->set_out(0, |
4474 | Location::RegisterLocation(AllocateMintABI::kResultReg)); |
4475 | summary->set_temp(0, Location::RegisterLocation(AllocateMintABI::kTempReg)); |
4476 | } else { |
4477 | summary->set_out(0, Location::RequiresRegister()); |
4478 | summary->set_temp(0, Location::RequiresRegister()); |
4479 | } |
4480 | return summary; |
4481 | } |
4482 | |
4483 | void BoxInt64Instr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4484 | const Register out = locs()->out(0).reg(); |
4485 | const Register value = locs()->in(0).reg(); |
4486 | __ MoveRegister(out, value); |
4487 | __ SmiTag(out); |
4488 | if (ValueFitsSmi()) { |
4489 | return; |
4490 | } |
4491 | // If the value doesn't fit in a smi, the tagging changes the sign, |
4492 | // which causes the overflow flag to be set. |
4493 | compiler::Label done; |
4494 | __ j(NO_OVERFLOW, &done); |
4495 | |
4496 | const Register temp = locs()->temp(0).reg(); |
4497 | if (compiler->intrinsic_mode()) { |
4498 | __ TryAllocate(compiler->mint_class(), |
4499 | compiler->intrinsic_slow_path_label(), |
4500 | /*near_jump=*/true, out, temp); |
4501 | } else if (locs()->call_on_shared_slow_path()) { |
4502 | auto object_store = compiler->isolate()->object_store(); |
4503 | const bool live_fpu_regs = locs()->live_registers()->FpuRegisterCount() > 0; |
4504 | const auto& stub = Code::ZoneHandle( |
4505 | compiler->zone(), |
4506 | live_fpu_regs ? object_store->allocate_mint_with_fpu_regs_stub() |
4507 | : object_store->allocate_mint_without_fpu_regs_stub()); |
4508 | |
4509 | ASSERT(!locs()->live_registers()->ContainsRegister( |
4510 | AllocateMintABI::kResultReg)); |
4511 | auto extended_env = compiler->SlowPathEnvironmentFor(this, 0); |
4512 | compiler->GenerateStubCall(token_pos(), stub, PcDescriptorsLayout::kOther, |
4513 | locs(), DeoptId::kNone, extended_env); |
4514 | } else { |
4515 | BoxAllocationSlowPath::Allocate(compiler, this, compiler->mint_class(), out, |
4516 | temp); |
4517 | } |
4518 | |
4519 | __ movq(compiler::FieldAddress(out, Mint::value_offset()), value); |
4520 | __ Bind(&done); |
4521 | } |
4522 | |
4523 | LocationSummary* BinaryDoubleOpInstr::MakeLocationSummary(Zone* zone, |
4524 | bool opt) const { |
4525 | const intptr_t kNumInputs = 2; |
4526 | const intptr_t kNumTemps = 0; |
4527 | LocationSummary* summary = new (zone) |
4528 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
4529 | summary->set_in(0, Location::RequiresFpuRegister()); |
4530 | summary->set_in(1, Location::RequiresFpuRegister()); |
4531 | summary->set_out(0, Location::SameAsFirstInput()); |
4532 | return summary; |
4533 | } |
4534 | |
4535 | void BinaryDoubleOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4536 | XmmRegister left = locs()->in(0).fpu_reg(); |
4537 | XmmRegister right = locs()->in(1).fpu_reg(); |
4538 | |
4539 | ASSERT(locs()->out(0).fpu_reg() == left); |
4540 | |
4541 | switch (op_kind()) { |
4542 | case Token::kADD: |
4543 | __ addsd(left, right); |
4544 | break; |
4545 | case Token::kSUB: |
4546 | __ subsd(left, right); |
4547 | break; |
4548 | case Token::kMUL: |
4549 | __ mulsd(left, right); |
4550 | break; |
4551 | case Token::kDIV: |
4552 | __ divsd(left, right); |
4553 | break; |
4554 | default: |
4555 | UNREACHABLE(); |
4556 | } |
4557 | } |
4558 | |
4559 | LocationSummary* DoubleTestOpInstr::MakeLocationSummary(Zone* zone, |
4560 | bool opt) const { |
4561 | const intptr_t kNumInputs = 1; |
4562 | const intptr_t kNumTemps = |
4563 | (op_kind() == MethodRecognizer::kDouble_getIsInfinite) ? 1 : 0; |
4564 | LocationSummary* summary = new (zone) |
4565 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
4566 | summary->set_in(0, Location::RequiresFpuRegister()); |
4567 | if (op_kind() == MethodRecognizer::kDouble_getIsInfinite) { |
4568 | summary->set_temp(0, Location::RequiresRegister()); |
4569 | } |
4570 | summary->set_out(0, Location::RequiresRegister()); |
4571 | return summary; |
4572 | } |
4573 | |
4574 | Condition DoubleTestOpInstr::EmitComparisonCode(FlowGraphCompiler* compiler, |
4575 | BranchLabels labels) { |
4576 | ASSERT(compiler->is_optimizing()); |
4577 | const XmmRegister value = locs()->in(0).fpu_reg(); |
4578 | const bool is_negated = kind() != Token::kEQ; |
4579 | if (op_kind() == MethodRecognizer::kDouble_getIsNaN) { |
4580 | compiler::Label is_nan; |
4581 | __ comisd(value, value); |
4582 | return is_negated ? PARITY_ODD : PARITY_EVEN; |
4583 | } else { |
4584 | ASSERT(op_kind() == MethodRecognizer::kDouble_getIsInfinite); |
4585 | const Register temp = locs()->temp(0).reg(); |
4586 | __ AddImmediate(RSP, compiler::Immediate(-kDoubleSize)); |
4587 | __ movsd(compiler::Address(RSP, 0), value); |
4588 | __ movq(temp, compiler::Address(RSP, 0)); |
4589 | __ AddImmediate(RSP, compiler::Immediate(kDoubleSize)); |
4590 | // Mask off the sign. |
4591 | __ AndImmediate(temp, compiler::Immediate(0x7FFFFFFFFFFFFFFFLL)); |
4592 | // Compare with +infinity. |
4593 | __ CompareImmediate(temp, compiler::Immediate(0x7FF0000000000000LL)); |
4594 | return is_negated ? NOT_EQUAL : EQUAL; |
4595 | } |
4596 | } |
4597 | |
4598 | // SIMD |
4599 | |
4600 | #define DEFINE_EMIT(Name, Args) \ |
4601 | static void Emit##Name(FlowGraphCompiler* compiler, SimdOpInstr* instr, \ |
4602 | PP_APPLY(PP_UNPACK, Args)) |
4603 | |
4604 | #define SIMD_OP_FLOAT_ARITH(V, Name, op) \ |
4605 | V(Float32x4##Name, op##ps) \ |
4606 | V(Float64x2##Name, op##pd) |
4607 | |
4608 | #define SIMD_OP_SIMPLE_BINARY(V) \ |
4609 | SIMD_OP_FLOAT_ARITH(V, Add, add) \ |
4610 | SIMD_OP_FLOAT_ARITH(V, Sub, sub) \ |
4611 | SIMD_OP_FLOAT_ARITH(V, Mul, mul) \ |
4612 | SIMD_OP_FLOAT_ARITH(V, Div, div) \ |
4613 | SIMD_OP_FLOAT_ARITH(V, Min, min) \ |
4614 | SIMD_OP_FLOAT_ARITH(V, Max, max) \ |
4615 | V(Int32x4Add, addpl) \ |
4616 | V(Int32x4Sub, subpl) \ |
4617 | V(Int32x4BitAnd, andps) \ |
4618 | V(Int32x4BitOr, orps) \ |
4619 | V(Int32x4BitXor, xorps) \ |
4620 | V(Float32x4Equal, cmppseq) \ |
4621 | V(Float32x4NotEqual, cmppsneq) \ |
4622 | V(Float32x4GreaterThan, cmppsnle) \ |
4623 | V(Float32x4GreaterThanOrEqual, cmppsnlt) \ |
4624 | V(Float32x4LessThan, cmppslt) \ |
4625 | V(Float32x4LessThanOrEqual, cmppsle) |
4626 | |
4627 | DEFINE_EMIT(SimdBinaryOp, |
4628 | (SameAsFirstInput, XmmRegister left, XmmRegister right)) { |
4629 | switch (instr->kind()) { |
4630 | #define EMIT(Name, op) \ |
4631 | case SimdOpInstr::k##Name: \ |
4632 | __ op(left, right); \ |
4633 | break; |
4634 | SIMD_OP_SIMPLE_BINARY(EMIT) |
4635 | #undef EMIT |
4636 | case SimdOpInstr::kFloat32x4Scale: |
4637 | __ cvtsd2ss(left, left); |
4638 | __ shufps(left, left, compiler::Immediate(0x00)); |
4639 | __ mulps(left, right); |
4640 | break; |
4641 | case SimdOpInstr::kFloat32x4ShuffleMix: |
4642 | case SimdOpInstr::kInt32x4ShuffleMix: |
4643 | __ shufps(left, right, compiler::Immediate(instr->mask())); |
4644 | break; |
4645 | case SimdOpInstr::kFloat64x2FromDoubles: |
4646 | // shufpd mask 0x0 results in: |
4647 | // Lower 64-bits of left = Lower 64-bits of left. |
4648 | // Upper 64-bits of left = Lower 64-bits of right. |
4649 | __ shufpd(left, right, compiler::Immediate(0x0)); |
4650 | break; |
4651 | case SimdOpInstr::kFloat64x2Scale: |
4652 | __ shufpd(right, right, compiler::Immediate(0x00)); |
4653 | __ mulpd(left, right); |
4654 | break; |
4655 | case SimdOpInstr::kFloat64x2WithX: |
4656 | case SimdOpInstr::kFloat64x2WithY: { |
4657 | // TODO(dartbug.com/30949) avoid transfer through memory. |
4658 | COMPILE_ASSERT(SimdOpInstr::kFloat64x2WithY == |
4659 | (SimdOpInstr::kFloat64x2WithX + 1)); |
4660 | const intptr_t lane_index = instr->kind() - SimdOpInstr::kFloat64x2WithX; |
4661 | ASSERT(0 <= lane_index && lane_index < 2); |
4662 | |
4663 | __ SubImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4664 | __ movups(compiler::Address(RSP, 0), left); |
4665 | __ movsd(compiler::Address(RSP, lane_index * kDoubleSize), right); |
4666 | __ movups(left, compiler::Address(RSP, 0)); |
4667 | __ AddImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4668 | break; |
4669 | } |
4670 | case SimdOpInstr::kFloat32x4WithX: |
4671 | case SimdOpInstr::kFloat32x4WithY: |
4672 | case SimdOpInstr::kFloat32x4WithZ: |
4673 | case SimdOpInstr::kFloat32x4WithW: { |
4674 | // TODO(dartbug.com/30949) avoid transfer through memory. SSE4.1 has |
4675 | // insertps. SSE2 these instructions can be implemented via a combination |
4676 | // of shufps/movss/movlhps. |
4677 | COMPILE_ASSERT( |
4678 | SimdOpInstr::kFloat32x4WithY == (SimdOpInstr::kFloat32x4WithX + 1) && |
4679 | SimdOpInstr::kFloat32x4WithZ == (SimdOpInstr::kFloat32x4WithX + 2) && |
4680 | SimdOpInstr::kFloat32x4WithW == (SimdOpInstr::kFloat32x4WithX + 3)); |
4681 | const intptr_t lane_index = instr->kind() - SimdOpInstr::kFloat32x4WithX; |
4682 | ASSERT(0 <= lane_index && lane_index < 4); |
4683 | __ cvtsd2ss(left, left); |
4684 | __ SubImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4685 | __ movups(compiler::Address(RSP, 0), right); |
4686 | __ movss(compiler::Address(RSP, lane_index * kFloatSize), left); |
4687 | __ movups(left, compiler::Address(RSP, 0)); |
4688 | __ AddImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4689 | break; |
4690 | } |
4691 | |
4692 | default: |
4693 | UNREACHABLE(); |
4694 | } |
4695 | } |
4696 | |
4697 | #define SIMD_OP_SIMPLE_UNARY(V) \ |
4698 | SIMD_OP_FLOAT_ARITH(V, Sqrt, sqrt) \ |
4699 | SIMD_OP_FLOAT_ARITH(V, Negate, negate) \ |
4700 | SIMD_OP_FLOAT_ARITH(V, Abs, abs) \ |
4701 | V(Float32x4Reciprocal, rcpps) \ |
4702 | V(Float32x4ReciprocalSqrt, rsqrtps) |
4703 | |
4704 | DEFINE_EMIT(SimdUnaryOp, (SameAsFirstInput, XmmRegister value)) { |
4705 | // TODO(dartbug.com/30949) select better register constraints to avoid |
4706 | // redundant move of input into a different register. |
4707 | switch (instr->kind()) { |
4708 | #define EMIT(Name, op) \ |
4709 | case SimdOpInstr::k##Name: \ |
4710 | __ op(value, value); \ |
4711 | break; |
4712 | SIMD_OP_SIMPLE_UNARY(EMIT) |
4713 | #undef EMIT |
4714 | case SimdOpInstr::kFloat32x4ShuffleX: |
4715 | // Shuffle not necessary. |
4716 | __ cvtss2sd(value, value); |
4717 | break; |
4718 | case SimdOpInstr::kFloat32x4ShuffleY: |
4719 | __ shufps(value, value, compiler::Immediate(0x55)); |
4720 | __ cvtss2sd(value, value); |
4721 | break; |
4722 | case SimdOpInstr::kFloat32x4ShuffleZ: |
4723 | __ shufps(value, value, compiler::Immediate(0xAA)); |
4724 | __ cvtss2sd(value, value); |
4725 | break; |
4726 | case SimdOpInstr::kFloat32x4ShuffleW: |
4727 | __ shufps(value, value, compiler::Immediate(0xFF)); |
4728 | __ cvtss2sd(value, value); |
4729 | break; |
4730 | case SimdOpInstr::kFloat32x4Shuffle: |
4731 | case SimdOpInstr::kInt32x4Shuffle: |
4732 | __ shufps(value, value, compiler::Immediate(instr->mask())); |
4733 | break; |
4734 | case SimdOpInstr::kFloat32x4Splat: |
4735 | // Convert to Float32. |
4736 | __ cvtsd2ss(value, value); |
4737 | // Splat across all lanes. |
4738 | __ shufps(value, value, compiler::Immediate(0x00)); |
4739 | break; |
4740 | case SimdOpInstr::kFloat32x4ToFloat64x2: |
4741 | __ cvtps2pd(value, value); |
4742 | break; |
4743 | case SimdOpInstr::kFloat64x2ToFloat32x4: |
4744 | __ cvtpd2ps(value, value); |
4745 | break; |
4746 | case SimdOpInstr::kInt32x4ToFloat32x4: |
4747 | case SimdOpInstr::kFloat32x4ToInt32x4: |
4748 | // TODO(dartbug.com/30949) these operations are essentially nop and should |
4749 | // not generate any code. They should be removed from the graph before |
4750 | // code generation. |
4751 | break; |
4752 | case SimdOpInstr::kFloat64x2GetX: |
4753 | // NOP. |
4754 | break; |
4755 | case SimdOpInstr::kFloat64x2GetY: |
4756 | __ shufpd(value, value, compiler::Immediate(0x33)); |
4757 | break; |
4758 | case SimdOpInstr::kFloat64x2Splat: |
4759 | __ shufpd(value, value, compiler::Immediate(0x0)); |
4760 | break; |
4761 | default: |
4762 | UNREACHABLE(); |
4763 | break; |
4764 | } |
4765 | } |
4766 | |
4767 | DEFINE_EMIT(SimdGetSignMask, (Register out, XmmRegister value)) { |
4768 | switch (instr->kind()) { |
4769 | case SimdOpInstr::kFloat32x4GetSignMask: |
4770 | case SimdOpInstr::kInt32x4GetSignMask: |
4771 | __ movmskps(out, value); |
4772 | break; |
4773 | case SimdOpInstr::kFloat64x2GetSignMask: |
4774 | __ movmskpd(out, value); |
4775 | break; |
4776 | default: |
4777 | UNREACHABLE(); |
4778 | break; |
4779 | } |
4780 | } |
4781 | |
4782 | DEFINE_EMIT( |
4783 | Float32x4FromDoubles, |
4784 | (SameAsFirstInput, XmmRegister v0, XmmRegister, XmmRegister, XmmRegister)) { |
4785 | // TODO(dartbug.com/30949) avoid transfer through memory. SSE4.1 has |
4786 | // insertps, with SSE2 this instruction can be implemented through unpcklps. |
4787 | const XmmRegister out = v0; |
4788 | __ SubImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4789 | for (intptr_t i = 0; i < 4; i++) { |
4790 | __ cvtsd2ss(out, instr->locs()->in(i).fpu_reg()); |
4791 | __ movss(compiler::Address(RSP, i * kFloatSize), out); |
4792 | } |
4793 | __ movups(out, compiler::Address(RSP, 0)); |
4794 | __ AddImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4795 | } |
4796 | |
4797 | DEFINE_EMIT(Float32x4Zero, (XmmRegister value)) { |
4798 | __ xorps(value, value); |
4799 | } |
4800 | |
4801 | DEFINE_EMIT(Float64x2Zero, (XmmRegister value)) { |
4802 | __ xorpd(value, value); |
4803 | } |
4804 | |
4805 | DEFINE_EMIT(Float32x4Clamp, |
4806 | (SameAsFirstInput, |
4807 | XmmRegister value, |
4808 | XmmRegister lower, |
4809 | XmmRegister upper)) { |
4810 | __ minps(value, upper); |
4811 | __ maxps(value, lower); |
4812 | } |
4813 | |
4814 | DEFINE_EMIT(Int32x4FromInts, |
4815 | (XmmRegister result, Register, Register, Register, Register)) { |
4816 | // TODO(dartbug.com/30949) avoid transfer through memory. |
4817 | __ SubImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4818 | for (intptr_t i = 0; i < 4; i++) { |
4819 | __ movl(compiler::Address(RSP, i * kInt32Size), instr->locs()->in(i).reg()); |
4820 | } |
4821 | __ movups(result, compiler::Address(RSP, 0)); |
4822 | __ AddImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4823 | } |
4824 | |
4825 | DEFINE_EMIT(Int32x4FromBools, |
4826 | (XmmRegister result, |
4827 | Register, |
4828 | Register, |
4829 | Register, |
4830 | Register, |
4831 | Temp<Register> temp)) { |
4832 | // TODO(dartbug.com/30949) avoid transfer through memory. |
4833 | __ SubImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4834 | for (intptr_t i = 0; i < 4; i++) { |
4835 | compiler::Label done, load_false; |
4836 | __ xorq(temp, temp); |
4837 | __ CompareObject(instr->locs()->in(i).reg(), Bool::True()); |
4838 | __ setcc(EQUAL, ByteRegisterOf(temp)); |
4839 | __ negl(temp); // temp = input ? -1 : 0 |
4840 | __ movl(compiler::Address(RSP, kInt32Size * i), temp); |
4841 | } |
4842 | __ movups(result, compiler::Address(RSP, 0)); |
4843 | __ AddImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4844 | } |
4845 | |
4846 | static void EmitToBoolean(FlowGraphCompiler* compiler, Register out) { |
4847 | ASSERT_BOOL_FALSE_FOLLOWS_BOOL_TRUE(); |
4848 | __ testl(out, out); |
4849 | __ setcc(ZERO, ByteRegisterOf(out)); |
4850 | __ movzxb(out, out); |
4851 | __ movq(out, |
4852 | compiler::Address(THR, out, TIMES_8, Thread::bool_true_offset())); |
4853 | } |
4854 | |
4855 | DEFINE_EMIT(Int32x4GetFlagZorW, |
4856 | (Register out, XmmRegister value, Temp<XmmRegister> temp)) { |
4857 | __ movhlps(temp, value); // extract upper half. |
4858 | __ movq(out, temp); |
4859 | if (instr->kind() == SimdOpInstr::kInt32x4GetFlagW) { |
4860 | __ shrq(out, compiler::Immediate(32)); // extract upper 32bits. |
4861 | } |
4862 | EmitToBoolean(compiler, out); |
4863 | } |
4864 | |
4865 | DEFINE_EMIT(Int32x4GetFlagXorY, (Register out, XmmRegister value)) { |
4866 | __ movq(out, value); |
4867 | if (instr->kind() == SimdOpInstr::kInt32x4GetFlagY) { |
4868 | __ shrq(out, compiler::Immediate(32)); // extract upper 32bits. |
4869 | } |
4870 | EmitToBoolean(compiler, out); |
4871 | } |
4872 | |
4873 | DEFINE_EMIT( |
4874 | Int32x4WithFlag, |
4875 | (SameAsFirstInput, XmmRegister mask, Register flag, Temp<Register> temp)) { |
4876 | // TODO(dartbug.com/30949) avoid transfer through memory. |
4877 | COMPILE_ASSERT( |
4878 | SimdOpInstr::kInt32x4WithFlagY == (SimdOpInstr::kInt32x4WithFlagX + 1) && |
4879 | SimdOpInstr::kInt32x4WithFlagZ == (SimdOpInstr::kInt32x4WithFlagX + 2) && |
4880 | SimdOpInstr::kInt32x4WithFlagW == (SimdOpInstr::kInt32x4WithFlagX + 3)); |
4881 | const intptr_t lane_index = instr->kind() - SimdOpInstr::kInt32x4WithFlagX; |
4882 | ASSERT(0 <= lane_index && lane_index < 4); |
4883 | __ SubImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4884 | __ movups(compiler::Address(RSP, 0), mask); |
4885 | |
4886 | // temp = flag == true ? -1 : 0 |
4887 | __ xorq(temp, temp); |
4888 | __ CompareObject(flag, Bool::True()); |
4889 | __ setcc(EQUAL, ByteRegisterOf(temp)); |
4890 | __ negl(temp); |
4891 | |
4892 | __ movl(compiler::Address(RSP, lane_index * kInt32Size), temp); |
4893 | __ movups(mask, compiler::Address(RSP, 0)); |
4894 | __ AddImmediate(RSP, compiler::Immediate(kSimd128Size)); |
4895 | } |
4896 | |
4897 | DEFINE_EMIT(Int32x4Select, |
4898 | (SameAsFirstInput, |
4899 | XmmRegister mask, |
4900 | XmmRegister trueValue, |
4901 | XmmRegister falseValue, |
4902 | Temp<XmmRegister> temp)) { |
4903 | // Copy mask. |
4904 | __ movaps(temp, mask); |
4905 | // Invert it. |
4906 | __ notps(temp, temp); |
4907 | // mask = mask & trueValue. |
4908 | __ andps(mask, trueValue); |
4909 | // temp = temp & falseValue. |
4910 | __ andps(temp, falseValue); |
4911 | // out = mask | temp. |
4912 | __ orps(mask, temp); |
4913 | } |
4914 | |
4915 | // Map SimdOpInstr::Kind-s to corresponding emit functions. Uses the following |
4916 | // format: |
4917 | // |
4918 | // CASE(OpA) CASE(OpB) ____(Emitter) - Emitter is used to emit OpA and OpB. |
4919 | // SIMPLE(OpA) - Emitter with name OpA is used to emit OpA. |
4920 | // |
4921 | #define SIMD_OP_VARIANTS(CASE, ____, SIMPLE) \ |
4922 | SIMD_OP_SIMPLE_BINARY(CASE) \ |
4923 | CASE(Float32x4Scale) \ |
4924 | CASE(Float32x4ShuffleMix) \ |
4925 | CASE(Int32x4ShuffleMix) \ |
4926 | CASE(Float64x2FromDoubles) \ |
4927 | CASE(Float64x2Scale) \ |
4928 | CASE(Float64x2WithX) \ |
4929 | CASE(Float64x2WithY) \ |
4930 | CASE(Float32x4WithX) \ |
4931 | CASE(Float32x4WithY) \ |
4932 | CASE(Float32x4WithZ) \ |
4933 | CASE(Float32x4WithW) \ |
4934 | ____(SimdBinaryOp) \ |
4935 | SIMD_OP_SIMPLE_UNARY(CASE) \ |
4936 | CASE(Float32x4ShuffleX) \ |
4937 | CASE(Float32x4ShuffleY) \ |
4938 | CASE(Float32x4ShuffleZ) \ |
4939 | CASE(Float32x4ShuffleW) \ |
4940 | CASE(Float32x4Shuffle) \ |
4941 | CASE(Int32x4Shuffle) \ |
4942 | CASE(Float32x4Splat) \ |
4943 | CASE(Float32x4ToFloat64x2) \ |
4944 | CASE(Float64x2ToFloat32x4) \ |
4945 | CASE(Int32x4ToFloat32x4) \ |
4946 | CASE(Float32x4ToInt32x4) \ |
4947 | CASE(Float64x2GetX) \ |
4948 | CASE(Float64x2GetY) \ |
4949 | CASE(Float64x2Splat) \ |
4950 | ____(SimdUnaryOp) \ |
4951 | CASE(Float32x4GetSignMask) \ |
4952 | CASE(Int32x4GetSignMask) \ |
4953 | CASE(Float64x2GetSignMask) \ |
4954 | ____(SimdGetSignMask) \ |
4955 | SIMPLE(Float32x4FromDoubles) \ |
4956 | SIMPLE(Int32x4FromInts) \ |
4957 | SIMPLE(Int32x4FromBools) \ |
4958 | SIMPLE(Float32x4Zero) \ |
4959 | SIMPLE(Float64x2Zero) \ |
4960 | SIMPLE(Float32x4Clamp) \ |
4961 | CASE(Int32x4GetFlagX) \ |
4962 | CASE(Int32x4GetFlagY) \ |
4963 | ____(Int32x4GetFlagXorY) \ |
4964 | CASE(Int32x4GetFlagZ) \ |
4965 | CASE(Int32x4GetFlagW) \ |
4966 | ____(Int32x4GetFlagZorW) \ |
4967 | CASE(Int32x4WithFlagX) \ |
4968 | CASE(Int32x4WithFlagY) \ |
4969 | CASE(Int32x4WithFlagZ) \ |
4970 | CASE(Int32x4WithFlagW) \ |
4971 | ____(Int32x4WithFlag) \ |
4972 | SIMPLE(Int32x4Select) |
4973 | |
4974 | LocationSummary* SimdOpInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
4975 | switch (kind()) { |
4976 | #define CASE(Name, ...) case k##Name: |
4977 | #define EMIT(Name) \ |
4978 | return MakeLocationSummaryFromEmitter(zone, this, &Emit##Name); |
4979 | #define SIMPLE(Name) CASE(Name) EMIT(Name) |
4980 | SIMD_OP_VARIANTS(CASE, EMIT, SIMPLE) |
4981 | #undef CASE |
4982 | #undef EMIT |
4983 | #undef SIMPLE |
4984 | case kIllegalSimdOp: |
4985 | break; |
4986 | } |
4987 | UNREACHABLE(); |
4988 | return NULL; |
4989 | } |
4990 | |
4991 | void SimdOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
4992 | switch (kind()) { |
4993 | #define CASE(Name, ...) case k##Name: |
4994 | #define EMIT(Name) \ |
4995 | InvokeEmitter(compiler, this, &Emit##Name); \ |
4996 | break; |
4997 | #define SIMPLE(Name) CASE(Name) EMIT(Name) |
4998 | SIMD_OP_VARIANTS(CASE, EMIT, SIMPLE) |
4999 | #undef CASE |
5000 | #undef EMIT |
5001 | #undef SIMPLE |
5002 | case kIllegalSimdOp: |
5003 | UNREACHABLE(); |
5004 | break; |
5005 | } |
5006 | } |
5007 | |
5008 | #undef DEFINE_EMIT |
5009 | |
5010 | LocationSummary* MathUnaryInstr::MakeLocationSummary(Zone* zone, |
5011 | bool opt) const { |
5012 | ASSERT((kind() == MathUnaryInstr::kSqrt) || |
5013 | (kind() == MathUnaryInstr::kDoubleSquare)); |
5014 | const intptr_t kNumInputs = 1; |
5015 | const intptr_t kNumTemps = 0; |
5016 | LocationSummary* summary = new (zone) |
5017 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5018 | summary->set_in(0, Location::RequiresFpuRegister()); |
5019 | if (kind() == MathUnaryInstr::kDoubleSquare) { |
5020 | summary->set_out(0, Location::SameAsFirstInput()); |
5021 | } else { |
5022 | summary->set_out(0, Location::RequiresFpuRegister()); |
5023 | } |
5024 | return summary; |
5025 | } |
5026 | |
5027 | void MathUnaryInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5028 | if (kind() == MathUnaryInstr::kSqrt) { |
5029 | __ sqrtsd(locs()->out(0).fpu_reg(), locs()->in(0).fpu_reg()); |
5030 | } else if (kind() == MathUnaryInstr::kDoubleSquare) { |
5031 | XmmRegister value_reg = locs()->in(0).fpu_reg(); |
5032 | __ mulsd(value_reg, value_reg); |
5033 | ASSERT(value_reg == locs()->out(0).fpu_reg()); |
5034 | } else { |
5035 | UNREACHABLE(); |
5036 | } |
5037 | } |
5038 | |
5039 | LocationSummary* CaseInsensitiveCompareInstr::MakeLocationSummary( |
5040 | Zone* zone, |
5041 | bool opt) const { |
5042 | const intptr_t kNumTemps = 0; |
5043 | LocationSummary* summary = new (zone) |
5044 | LocationSummary(zone, InputCount(), kNumTemps, LocationSummary::kCall); |
5045 | summary->set_in(0, Location::RegisterLocation(CallingConventions::kArg1Reg)); |
5046 | summary->set_in(1, Location::RegisterLocation(CallingConventions::kArg2Reg)); |
5047 | summary->set_in(2, Location::RegisterLocation(CallingConventions::kArg3Reg)); |
5048 | summary->set_in(3, Location::RegisterLocation(CallingConventions::kArg4Reg)); |
5049 | summary->set_out(0, Location::RegisterLocation(RAX)); |
5050 | return summary; |
5051 | } |
5052 | |
5053 | void CaseInsensitiveCompareInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5054 | // Save RSP. R13 is chosen because it is callee saved so we do not need to |
5055 | // back it up before calling into the runtime. |
5056 | static const Register kSavedSPReg = R13; |
5057 | __ movq(kSavedSPReg, RSP); |
5058 | __ ReserveAlignedFrameSpace(0); |
5059 | |
5060 | // Call the function. Parameters are already in their correct spots. |
5061 | __ CallRuntime(TargetFunction(), TargetFunction().argument_count()); |
5062 | |
5063 | // Restore RSP. |
5064 | __ movq(RSP, kSavedSPReg); |
5065 | } |
5066 | |
5067 | LocationSummary* UnarySmiOpInstr::MakeLocationSummary(Zone* zone, |
5068 | bool opt) const { |
5069 | const intptr_t kNumInputs = 1; |
5070 | return LocationSummary::Make(zone, kNumInputs, Location::SameAsFirstInput(), |
5071 | LocationSummary::kNoCall); |
5072 | } |
5073 | |
5074 | void UnarySmiOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5075 | Register value = locs()->in(0).reg(); |
5076 | ASSERT(value == locs()->out(0).reg()); |
5077 | switch (op_kind()) { |
5078 | case Token::kNEGATE: { |
5079 | compiler::Label* deopt = |
5080 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptUnaryOp); |
5081 | __ negq(value); |
5082 | __ j(OVERFLOW, deopt); |
5083 | break; |
5084 | } |
5085 | case Token::kBIT_NOT: |
5086 | __ notq(value); |
5087 | // Remove inverted smi-tag. |
5088 | __ AndImmediate(value, compiler::Immediate(~kSmiTagMask)); |
5089 | break; |
5090 | default: |
5091 | UNREACHABLE(); |
5092 | } |
5093 | } |
5094 | |
5095 | LocationSummary* UnaryDoubleOpInstr::MakeLocationSummary(Zone* zone, |
5096 | bool opt) const { |
5097 | const intptr_t kNumInputs = 1; |
5098 | const intptr_t kNumTemps = 0; |
5099 | LocationSummary* summary = new (zone) |
5100 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5101 | summary->set_in(0, Location::RequiresFpuRegister()); |
5102 | summary->set_out(0, Location::SameAsFirstInput()); |
5103 | return summary; |
5104 | } |
5105 | |
5106 | void UnaryDoubleOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5107 | XmmRegister value = locs()->in(0).fpu_reg(); |
5108 | ASSERT(locs()->out(0).fpu_reg() == value); |
5109 | __ DoubleNegate(value, value); |
5110 | } |
5111 | |
5112 | LocationSummary* MathMinMaxInstr::MakeLocationSummary(Zone* zone, |
5113 | bool opt) const { |
5114 | if (result_cid() == kDoubleCid) { |
5115 | const intptr_t kNumInputs = 2; |
5116 | const intptr_t kNumTemps = 1; |
5117 | LocationSummary* summary = new (zone) |
5118 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5119 | summary->set_in(0, Location::RequiresFpuRegister()); |
5120 | summary->set_in(1, Location::RequiresFpuRegister()); |
5121 | // Reuse the left register so that code can be made shorter. |
5122 | summary->set_out(0, Location::SameAsFirstInput()); |
5123 | summary->set_temp(0, Location::RequiresRegister()); |
5124 | return summary; |
5125 | } |
5126 | ASSERT(result_cid() == kSmiCid); |
5127 | const intptr_t kNumInputs = 2; |
5128 | const intptr_t kNumTemps = 0; |
5129 | LocationSummary* summary = new (zone) |
5130 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5131 | summary->set_in(0, Location::RequiresRegister()); |
5132 | summary->set_in(1, Location::RequiresRegister()); |
5133 | // Reuse the left register so that code can be made shorter. |
5134 | summary->set_out(0, Location::SameAsFirstInput()); |
5135 | return summary; |
5136 | } |
5137 | |
5138 | void MathMinMaxInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5139 | ASSERT((op_kind() == MethodRecognizer::kMathMin) || |
5140 | (op_kind() == MethodRecognizer::kMathMax)); |
5141 | const bool is_min = op_kind() == MethodRecognizer::kMathMin; |
5142 | if (result_cid() == kDoubleCid) { |
5143 | compiler::Label done, returns_nan, are_equal; |
5144 | XmmRegister left = locs()->in(0).fpu_reg(); |
5145 | XmmRegister right = locs()->in(1).fpu_reg(); |
5146 | XmmRegister result = locs()->out(0).fpu_reg(); |
5147 | Register temp = locs()->temp(0).reg(); |
5148 | __ comisd(left, right); |
5149 | __ j(PARITY_EVEN, &returns_nan, compiler::Assembler::kNearJump); |
5150 | __ j(EQUAL, &are_equal, compiler::Assembler::kNearJump); |
5151 | const Condition double_condition = |
5152 | is_min ? TokenKindToDoubleCondition(Token::kLT) |
5153 | : TokenKindToDoubleCondition(Token::kGT); |
5154 | ASSERT(left == result); |
5155 | __ j(double_condition, &done, compiler::Assembler::kNearJump); |
5156 | __ movsd(result, right); |
5157 | __ jmp(&done, compiler::Assembler::kNearJump); |
5158 | |
5159 | __ Bind(&returns_nan); |
5160 | __ movq(temp, compiler::Address(THR, Thread::double_nan_address_offset())); |
5161 | __ movsd(result, compiler::Address(temp, 0)); |
5162 | __ jmp(&done, compiler::Assembler::kNearJump); |
5163 | |
5164 | __ Bind(&are_equal); |
5165 | compiler::Label left_is_negative; |
5166 | // Check for negative zero: -0.0 is equal 0.0 but min or max must return |
5167 | // -0.0 or 0.0 respectively. |
5168 | // Check for negative left value (get the sign bit): |
5169 | // - min -> left is negative ? left : right. |
5170 | // - max -> left is negative ? right : left |
5171 | // Check the sign bit. |
5172 | __ movmskpd(temp, left); |
5173 | __ testq(temp, compiler::Immediate(1)); |
5174 | if (is_min) { |
5175 | ASSERT(left == result); |
5176 | __ j(NOT_ZERO, &done, |
5177 | compiler::Assembler::kNearJump); // Negative -> return left. |
5178 | } else { |
5179 | ASSERT(left == result); |
5180 | __ j(ZERO, &done, |
5181 | compiler::Assembler::kNearJump); // Positive -> return left. |
5182 | } |
5183 | __ movsd(result, right); |
5184 | __ Bind(&done); |
5185 | return; |
5186 | } |
5187 | |
5188 | ASSERT(result_cid() == kSmiCid); |
5189 | Register left = locs()->in(0).reg(); |
5190 | Register right = locs()->in(1).reg(); |
5191 | Register result = locs()->out(0).reg(); |
5192 | __ cmpq(left, right); |
5193 | ASSERT(result == left); |
5194 | if (is_min) { |
5195 | __ cmovgeq(result, right); |
5196 | } else { |
5197 | __ cmovlq(result, right); |
5198 | } |
5199 | } |
5200 | |
5201 | LocationSummary* Int32ToDoubleInstr::MakeLocationSummary(Zone* zone, |
5202 | bool opt) const { |
5203 | const intptr_t kNumInputs = 1; |
5204 | const intptr_t kNumTemps = 0; |
5205 | LocationSummary* result = new (zone) |
5206 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5207 | result->set_in(0, Location::RequiresRegister()); |
5208 | result->set_out(0, Location::RequiresFpuRegister()); |
5209 | return result; |
5210 | } |
5211 | |
5212 | void Int32ToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5213 | Register value = locs()->in(0).reg(); |
5214 | FpuRegister result = locs()->out(0).fpu_reg(); |
5215 | __ cvtsi2sdl(result, value); |
5216 | } |
5217 | |
5218 | LocationSummary* SmiToDoubleInstr::MakeLocationSummary(Zone* zone, |
5219 | bool opt) const { |
5220 | const intptr_t kNumInputs = 1; |
5221 | const intptr_t kNumTemps = 0; |
5222 | LocationSummary* result = new (zone) |
5223 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5224 | result->set_in(0, Location::WritableRegister()); |
5225 | result->set_out(0, Location::RequiresFpuRegister()); |
5226 | return result; |
5227 | } |
5228 | |
5229 | void SmiToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5230 | Register value = locs()->in(0).reg(); |
5231 | FpuRegister result = locs()->out(0).fpu_reg(); |
5232 | __ SmiUntag(value); |
5233 | __ cvtsi2sdq(result, value); |
5234 | } |
5235 | |
5236 | DEFINE_BACKEND(Int64ToDouble, (FpuRegister result, Register value)) { |
5237 | __ cvtsi2sdq(result, value); |
5238 | } |
5239 | |
5240 | LocationSummary* DoubleToIntegerInstr::MakeLocationSummary(Zone* zone, |
5241 | bool opt) const { |
5242 | const intptr_t kNumInputs = 1; |
5243 | const intptr_t kNumTemps = 1; |
5244 | LocationSummary* result = new (zone) |
5245 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
5246 | result->set_in(0, Location::RegisterLocation(RCX)); |
5247 | result->set_out(0, Location::RegisterLocation(RAX)); |
5248 | result->set_temp(0, Location::RegisterLocation(RBX)); |
5249 | return result; |
5250 | } |
5251 | |
5252 | void DoubleToIntegerInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5253 | Register result = locs()->out(0).reg(); |
5254 | Register value_obj = locs()->in(0).reg(); |
5255 | Register temp = locs()->temp(0).reg(); |
5256 | XmmRegister value_double = FpuTMP; |
5257 | ASSERT(result == RAX); |
5258 | ASSERT(result != value_obj); |
5259 | ASSERT(result != temp); |
5260 | __ movsd(value_double, |
5261 | compiler::FieldAddress(value_obj, Double::value_offset())); |
5262 | __ cvttsd2siq(result, value_double); |
5263 | // Overflow is signalled with minint. |
5264 | compiler::Label do_call, done; |
5265 | // Check for overflow and that it fits into Smi. |
5266 | __ movq(temp, result); |
5267 | __ shlq(temp, compiler::Immediate(1)); |
5268 | __ j(OVERFLOW, &do_call, compiler::Assembler::kNearJump); |
5269 | __ SmiTag(result); |
5270 | __ jmp(&done); |
5271 | __ Bind(&do_call); |
5272 | __ pushq(value_obj); |
5273 | ASSERT(instance_call()->HasICData()); |
5274 | const ICData& ic_data = *instance_call()->ic_data(); |
5275 | ASSERT(ic_data.NumberOfChecksIs(1)); |
5276 | const Function& target = Function::ZoneHandle(ic_data.GetTargetAt(0)); |
5277 | const int kTypeArgsLen = 0; |
5278 | const int kNumberOfArguments = 1; |
5279 | constexpr int kSizeOfArguments = 1; |
5280 | const Array& kNoArgumentNames = Object::null_array(); |
5281 | ArgumentsInfo args_info(kTypeArgsLen, kNumberOfArguments, kSizeOfArguments, |
5282 | kNoArgumentNames); |
5283 | compiler->GenerateStaticCall(deopt_id(), instance_call()->token_pos(), target, |
5284 | args_info, locs(), ICData::Handle(), |
5285 | ICData::kStatic); |
5286 | __ Bind(&done); |
5287 | } |
5288 | |
5289 | LocationSummary* DoubleToSmiInstr::MakeLocationSummary(Zone* zone, |
5290 | bool opt) const { |
5291 | const intptr_t kNumInputs = 1; |
5292 | const intptr_t kNumTemps = 1; |
5293 | LocationSummary* result = new (zone) |
5294 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5295 | result->set_in(0, Location::RequiresFpuRegister()); |
5296 | result->set_out(0, Location::RequiresRegister()); |
5297 | result->set_temp(0, Location::RequiresRegister()); |
5298 | return result; |
5299 | } |
5300 | |
5301 | void DoubleToSmiInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5302 | compiler::Label* deopt = |
5303 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptDoubleToSmi); |
5304 | Register result = locs()->out(0).reg(); |
5305 | XmmRegister value = locs()->in(0).fpu_reg(); |
5306 | Register temp = locs()->temp(0).reg(); |
5307 | |
5308 | __ cvttsd2siq(result, value); |
5309 | // Overflow is signalled with minint. |
5310 | compiler::Label do_call, done; |
5311 | // Check for overflow and that it fits into Smi. |
5312 | __ movq(temp, result); |
5313 | __ shlq(temp, compiler::Immediate(1)); |
5314 | __ j(OVERFLOW, deopt); |
5315 | __ SmiTag(result); |
5316 | } |
5317 | |
5318 | LocationSummary* DoubleToDoubleInstr::MakeLocationSummary(Zone* zone, |
5319 | bool opt) const { |
5320 | const intptr_t kNumInputs = 1; |
5321 | const intptr_t kNumTemps = 0; |
5322 | LocationSummary* result = new (zone) |
5323 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5324 | result->set_in(0, Location::RequiresFpuRegister()); |
5325 | result->set_out(0, Location::RequiresFpuRegister()); |
5326 | return result; |
5327 | } |
5328 | |
5329 | void DoubleToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5330 | XmmRegister value = locs()->in(0).fpu_reg(); |
5331 | XmmRegister result = locs()->out(0).fpu_reg(); |
5332 | switch (recognized_kind()) { |
5333 | case MethodRecognizer::kDoubleTruncate: |
5334 | __ roundsd(result, value, compiler::Assembler::kRoundToZero); |
5335 | break; |
5336 | case MethodRecognizer::kDoubleFloor: |
5337 | __ roundsd(result, value, compiler::Assembler::kRoundDown); |
5338 | break; |
5339 | case MethodRecognizer::kDoubleCeil: |
5340 | __ roundsd(result, value, compiler::Assembler::kRoundUp); |
5341 | break; |
5342 | default: |
5343 | UNREACHABLE(); |
5344 | } |
5345 | } |
5346 | |
5347 | LocationSummary* DoubleToFloatInstr::MakeLocationSummary(Zone* zone, |
5348 | bool opt) const { |
5349 | const intptr_t kNumInputs = 1; |
5350 | const intptr_t kNumTemps = 0; |
5351 | LocationSummary* result = new (zone) |
5352 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5353 | result->set_in(0, Location::RequiresFpuRegister()); |
5354 | result->set_out(0, Location::SameAsFirstInput()); |
5355 | return result; |
5356 | } |
5357 | |
5358 | void DoubleToFloatInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5359 | __ cvtsd2ss(locs()->out(0).fpu_reg(), locs()->in(0).fpu_reg()); |
5360 | } |
5361 | |
5362 | LocationSummary* FloatToDoubleInstr::MakeLocationSummary(Zone* zone, |
5363 | bool opt) const { |
5364 | const intptr_t kNumInputs = 1; |
5365 | const intptr_t kNumTemps = 0; |
5366 | LocationSummary* result = new (zone) |
5367 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5368 | result->set_in(0, Location::RequiresFpuRegister()); |
5369 | result->set_out(0, Location::SameAsFirstInput()); |
5370 | return result; |
5371 | } |
5372 | |
5373 | void FloatToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5374 | __ cvtss2sd(locs()->out(0).fpu_reg(), locs()->in(0).fpu_reg()); |
5375 | } |
5376 | |
5377 | LocationSummary* InvokeMathCFunctionInstr::MakeLocationSummary(Zone* zone, |
5378 | bool opt) const { |
5379 | // Calling convention on x64 uses XMM0 and XMM1 to pass the first two |
5380 | // double arguments and XMM0 to return the result. |
5381 | // |
5382 | // TODO(sjindel): allow XMM0 to be used. Requires refactoring InvokeDoublePow |
5383 | // to allow input 1/output register to be equal. |
5384 | ASSERT((InputCount() == 1) || (InputCount() == 2)); |
5385 | const intptr_t kNumTemps = |
5386 | (recognized_kind() == MethodRecognizer::kMathDoublePow) ? 4 : 1; |
5387 | LocationSummary* result = new (zone) |
5388 | LocationSummary(zone, InputCount(), kNumTemps, LocationSummary::kCall); |
5389 | ASSERT(R13 != CALLEE_SAVED_TEMP); |
5390 | ASSERT(((1 << R13) & CallingConventions::kCalleeSaveCpuRegisters) != 0); |
5391 | result->set_temp(0, Location::RegisterLocation(R13)); |
5392 | result->set_in(0, Location::FpuRegisterLocation(XMM2)); |
5393 | if (InputCount() == 2) { |
5394 | result->set_in(1, Location::FpuRegisterLocation(XMM1)); |
5395 | } |
5396 | if (recognized_kind() == MethodRecognizer::kMathDoublePow) { |
5397 | // Temp index 1. |
5398 | result->set_temp(1, Location::RegisterLocation(RAX)); |
5399 | // Temp index 2. |
5400 | result->set_temp(2, Location::FpuRegisterLocation(XMM4)); |
5401 | // Block XMM0 for the calling convention. |
5402 | result->set_temp(3, Location::FpuRegisterLocation(XMM0)); |
5403 | } |
5404 | result->set_out(0, Location::FpuRegisterLocation(XMM3)); |
5405 | return result; |
5406 | } |
5407 | |
5408 | // Pseudo code: |
5409 | // if (exponent == 0.0) return 1.0; |
5410 | // // Speed up simple cases. |
5411 | // if (exponent == 1.0) return base; |
5412 | // if (exponent == 2.0) return base * base; |
5413 | // if (exponent == 3.0) return base * base * base; |
5414 | // if (base == 1.0) return 1.0; |
5415 | // if (base.isNaN || exponent.isNaN) { |
5416 | // return double.NAN; |
5417 | // } |
5418 | // if (base != -Infinity && exponent == 0.5) { |
5419 | // if (base == 0.0) return 0.0; |
5420 | // return sqrt(value); |
5421 | // } |
5422 | // TODO(srdjan): Move into a stub? |
5423 | static void InvokeDoublePow(FlowGraphCompiler* compiler, |
5424 | InvokeMathCFunctionInstr* instr) { |
5425 | ASSERT(instr->recognized_kind() == MethodRecognizer::kMathDoublePow); |
5426 | const intptr_t kInputCount = 2; |
5427 | ASSERT(instr->InputCount() == kInputCount); |
5428 | LocationSummary* locs = instr->locs(); |
5429 | |
5430 | XmmRegister base = locs->in(0).fpu_reg(); |
5431 | XmmRegister exp = locs->in(1).fpu_reg(); |
5432 | XmmRegister result = locs->out(0).fpu_reg(); |
5433 | Register temp = locs->temp(InvokeMathCFunctionInstr::kObjectTempIndex).reg(); |
5434 | XmmRegister zero_temp = |
5435 | locs->temp(InvokeMathCFunctionInstr::kDoubleTempIndex).fpu_reg(); |
5436 | |
5437 | __ xorps(zero_temp, zero_temp); |
5438 | __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(1))); |
5439 | __ movsd(result, compiler::FieldAddress(temp, Double::value_offset())); |
5440 | |
5441 | compiler::Label check_base, skip_call; |
5442 | // exponent == 0.0 -> return 1.0; |
5443 | __ comisd(exp, zero_temp); |
5444 | __ j(PARITY_EVEN, &check_base, compiler::Assembler::kNearJump); |
5445 | __ j(EQUAL, &skip_call); // 'result' is 1.0. |
5446 | |
5447 | // exponent == 1.0 ? |
5448 | __ comisd(exp, result); |
5449 | compiler::Label return_base; |
5450 | __ j(EQUAL, &return_base, compiler::Assembler::kNearJump); |
5451 | |
5452 | // exponent == 2.0 ? |
5453 | __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(2.0))); |
5454 | __ movsd(XMM0, compiler::FieldAddress(temp, Double::value_offset())); |
5455 | __ comisd(exp, XMM0); |
5456 | compiler::Label return_base_times_2; |
5457 | __ j(EQUAL, &return_base_times_2, compiler::Assembler::kNearJump); |
5458 | |
5459 | // exponent == 3.0 ? |
5460 | __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(3.0))); |
5461 | __ movsd(XMM0, compiler::FieldAddress(temp, Double::value_offset())); |
5462 | __ comisd(exp, XMM0); |
5463 | __ j(NOT_EQUAL, &check_base); |
5464 | |
5465 | // Base times 3. |
5466 | __ movsd(result, base); |
5467 | __ mulsd(result, base); |
5468 | __ mulsd(result, base); |
5469 | __ jmp(&skip_call); |
5470 | |
5471 | __ Bind(&return_base); |
5472 | __ movsd(result, base); |
5473 | __ jmp(&skip_call); |
5474 | |
5475 | __ Bind(&return_base_times_2); |
5476 | __ movsd(result, base); |
5477 | __ mulsd(result, base); |
5478 | __ jmp(&skip_call); |
5479 | |
5480 | __ Bind(&check_base); |
5481 | // Note: 'exp' could be NaN. |
5482 | |
5483 | compiler::Label return_nan; |
5484 | // base == 1.0 -> return 1.0; |
5485 | __ comisd(base, result); |
5486 | __ j(PARITY_EVEN, &return_nan, compiler::Assembler::kNearJump); |
5487 | __ j(EQUAL, &skip_call, compiler::Assembler::kNearJump); |
5488 | // Note: 'base' could be NaN. |
5489 | __ comisd(exp, base); |
5490 | // Neither 'exp' nor 'base' is NaN. |
5491 | compiler::Label try_sqrt; |
5492 | __ j(PARITY_ODD, &try_sqrt, compiler::Assembler::kNearJump); |
5493 | // Return NaN. |
5494 | __ Bind(&return_nan); |
5495 | __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(NAN))); |
5496 | __ movsd(result, compiler::FieldAddress(temp, Double::value_offset())); |
5497 | __ jmp(&skip_call); |
5498 | |
5499 | compiler::Label do_pow, return_zero; |
5500 | __ Bind(&try_sqrt); |
5501 | // Before calling pow, check if we could use sqrt instead of pow. |
5502 | __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(kNegInfinity))); |
5503 | __ movsd(result, compiler::FieldAddress(temp, Double::value_offset())); |
5504 | // base == -Infinity -> call pow; |
5505 | __ comisd(base, result); |
5506 | __ j(EQUAL, &do_pow, compiler::Assembler::kNearJump); |
5507 | |
5508 | // exponent == 0.5 ? |
5509 | __ LoadObject(temp, Double::ZoneHandle(Double::NewCanonical(0.5))); |
5510 | __ movsd(result, compiler::FieldAddress(temp, Double::value_offset())); |
5511 | __ comisd(exp, result); |
5512 | __ j(NOT_EQUAL, &do_pow, compiler::Assembler::kNearJump); |
5513 | |
5514 | // base == 0 -> return 0; |
5515 | __ comisd(base, zero_temp); |
5516 | __ j(EQUAL, &return_zero, compiler::Assembler::kNearJump); |
5517 | |
5518 | __ sqrtsd(result, base); |
5519 | __ jmp(&skip_call, compiler::Assembler::kNearJump); |
5520 | |
5521 | __ Bind(&return_zero); |
5522 | __ movsd(result, zero_temp); |
5523 | __ jmp(&skip_call); |
5524 | |
5525 | __ Bind(&do_pow); |
5526 | |
5527 | // Save RSP. |
5528 | __ movq(locs->temp(InvokeMathCFunctionInstr::kSavedSpTempIndex).reg(), RSP); |
5529 | __ ReserveAlignedFrameSpace(0); |
5530 | __ movaps(XMM0, locs->in(0).fpu_reg()); |
5531 | ASSERT(locs->in(1).fpu_reg() == XMM1); |
5532 | |
5533 | __ CallRuntime(instr->TargetFunction(), kInputCount); |
5534 | __ movaps(locs->out(0).fpu_reg(), XMM0); |
5535 | // Restore RSP. |
5536 | __ movq(RSP, locs->temp(InvokeMathCFunctionInstr::kSavedSpTempIndex).reg()); |
5537 | __ Bind(&skip_call); |
5538 | } |
5539 | |
5540 | void InvokeMathCFunctionInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5541 | if (recognized_kind() == MethodRecognizer::kMathDoublePow) { |
5542 | InvokeDoublePow(compiler, this); |
5543 | return; |
5544 | } |
5545 | // Save RSP. |
5546 | __ movq(locs()->temp(kSavedSpTempIndex).reg(), RSP); |
5547 | __ ReserveAlignedFrameSpace(0); |
5548 | __ movaps(XMM0, locs()->in(0).fpu_reg()); |
5549 | if (InputCount() == 2) { |
5550 | ASSERT(locs()->in(1).fpu_reg() == XMM1); |
5551 | } |
5552 | |
5553 | __ CallRuntime(TargetFunction(), InputCount()); |
5554 | __ movaps(locs()->out(0).fpu_reg(), XMM0); |
5555 | // Restore RSP. |
5556 | __ movq(RSP, locs()->temp(kSavedSpTempIndex).reg()); |
5557 | } |
5558 | |
5559 | LocationSummary* ExtractNthOutputInstr::(Zone* zone, |
5560 | bool opt) const { |
5561 | // Only use this instruction in optimized code. |
5562 | ASSERT(opt); |
5563 | const intptr_t kNumInputs = 1; |
5564 | LocationSummary* summary = |
5565 | new (zone) LocationSummary(zone, kNumInputs, 0, LocationSummary::kNoCall); |
5566 | if (representation() == kUnboxedDouble) { |
5567 | if (index() == 0) { |
5568 | summary->set_in( |
5569 | 0, Location::Pair(Location::RequiresFpuRegister(), Location::Any())); |
5570 | } else { |
5571 | ASSERT(index() == 1); |
5572 | summary->set_in( |
5573 | 0, Location::Pair(Location::Any(), Location::RequiresFpuRegister())); |
5574 | } |
5575 | summary->set_out(0, Location::RequiresFpuRegister()); |
5576 | } else { |
5577 | ASSERT(representation() == kTagged); |
5578 | if (index() == 0) { |
5579 | summary->set_in( |
5580 | 0, Location::Pair(Location::RequiresRegister(), Location::Any())); |
5581 | } else { |
5582 | ASSERT(index() == 1); |
5583 | summary->set_in( |
5584 | 0, Location::Pair(Location::Any(), Location::RequiresRegister())); |
5585 | } |
5586 | summary->set_out(0, Location::RequiresRegister()); |
5587 | } |
5588 | return summary; |
5589 | } |
5590 | |
5591 | void ExtractNthOutputInstr::(FlowGraphCompiler* compiler) { |
5592 | ASSERT(locs()->in(0).IsPairLocation()); |
5593 | PairLocation* pair = locs()->in(0).AsPairLocation(); |
5594 | Location in_loc = pair->At(index()); |
5595 | if (representation() == kUnboxedDouble) { |
5596 | XmmRegister out = locs()->out(0).fpu_reg(); |
5597 | XmmRegister in = in_loc.fpu_reg(); |
5598 | __ movaps(out, in); |
5599 | } else { |
5600 | ASSERT(representation() == kTagged); |
5601 | Register out = locs()->out(0).reg(); |
5602 | Register in = in_loc.reg(); |
5603 | __ movq(out, in); |
5604 | } |
5605 | } |
5606 | |
5607 | LocationSummary* TruncDivModInstr::MakeLocationSummary(Zone* zone, |
5608 | bool opt) const { |
5609 | const intptr_t kNumInputs = 2; |
5610 | const intptr_t kNumTemps = 0; |
5611 | LocationSummary* summary = new (zone) |
5612 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5613 | // Both inputs must be writable because they will be untagged. |
5614 | summary->set_in(0, Location::RegisterLocation(RAX)); |
5615 | summary->set_in(1, Location::WritableRegister()); |
5616 | summary->set_out(0, Location::Pair(Location::RegisterLocation(RAX), |
5617 | Location::RegisterLocation(RDX))); |
5618 | return summary; |
5619 | } |
5620 | |
5621 | void TruncDivModInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5622 | ASSERT(CanDeoptimize()); |
5623 | compiler::Label* deopt = |
5624 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptBinarySmiOp); |
5625 | Register left = locs()->in(0).reg(); |
5626 | Register right = locs()->in(1).reg(); |
5627 | ASSERT(locs()->out(0).IsPairLocation()); |
5628 | PairLocation* pair = locs()->out(0).AsPairLocation(); |
5629 | Register result1 = pair->At(0).reg(); |
5630 | Register result2 = pair->At(1).reg(); |
5631 | compiler::Label not_32bit, done; |
5632 | Register temp = RDX; |
5633 | ASSERT(left == RAX); |
5634 | ASSERT((right != RDX) && (right != RAX)); |
5635 | ASSERT(result1 == RAX); |
5636 | ASSERT(result2 == RDX); |
5637 | if (RangeUtils::CanBeZero(divisor_range())) { |
5638 | // Handle divide by zero in runtime. |
5639 | __ testq(right, right); |
5640 | __ j(ZERO, deopt); |
5641 | } |
5642 | // Check if both operands fit into 32bits as idiv with 64bit operands |
5643 | // requires twice as many cycles and has much higher latency. |
5644 | // We are checking this before untagging them to avoid corner case |
5645 | // dividing INT_MAX by -1 that raises exception because quotient is |
5646 | // too large for 32bit register. |
5647 | __ movsxd(temp, left); |
5648 | __ cmpq(temp, left); |
5649 | __ j(NOT_EQUAL, ¬_32bit); |
5650 | __ movsxd(temp, right); |
5651 | __ cmpq(temp, right); |
5652 | __ j(NOT_EQUAL, ¬_32bit); |
5653 | |
5654 | // Both operands are 31bit smis. Divide using 32bit idiv. |
5655 | __ SmiUntag(left); |
5656 | __ SmiUntag(right); |
5657 | __ cdq(); |
5658 | __ idivl(right); |
5659 | __ movsxd(RAX, RAX); |
5660 | __ movsxd(RDX, RDX); |
5661 | __ jmp(&done); |
5662 | |
5663 | // Divide using 64bit idiv. |
5664 | __ Bind(¬_32bit); |
5665 | __ SmiUntag(left); |
5666 | __ SmiUntag(right); |
5667 | __ cqo(); // Sign extend RAX -> RDX:RAX. |
5668 | __ idivq(right); // RAX: quotient, RDX: remainder. |
5669 | // Check the corner case of dividing the 'MIN_SMI' with -1, in which |
5670 | // case we cannot tag the result. |
5671 | __ CompareImmediate(RAX, compiler::Immediate(0x4000000000000000)); |
5672 | __ j(EQUAL, deopt); |
5673 | __ Bind(&done); |
5674 | |
5675 | // Modulo correction (RDX). |
5676 | // res = left % right; |
5677 | // if (res < 0) { |
5678 | // if (right < 0) { |
5679 | // res = res - right; |
5680 | // } else { |
5681 | // res = res + right; |
5682 | // } |
5683 | // } |
5684 | compiler::Label all_done; |
5685 | __ cmpq(RDX, compiler::Immediate(0)); |
5686 | __ j(GREATER_EQUAL, &all_done, compiler::Assembler::kNearJump); |
5687 | // Result is negative, adjust it. |
5688 | if ((divisor_range() == NULL) || divisor_range()->Overlaps(-1, 1)) { |
5689 | compiler::Label subtract; |
5690 | __ cmpq(right, compiler::Immediate(0)); |
5691 | __ j(LESS, &subtract, compiler::Assembler::kNearJump); |
5692 | __ addq(RDX, right); |
5693 | __ jmp(&all_done, compiler::Assembler::kNearJump); |
5694 | __ Bind(&subtract); |
5695 | __ subq(RDX, right); |
5696 | } else if (divisor_range()->IsPositive()) { |
5697 | // Right is positive. |
5698 | __ addq(RDX, right); |
5699 | } else { |
5700 | // Right is negative. |
5701 | __ subq(RDX, right); |
5702 | } |
5703 | __ Bind(&all_done); |
5704 | |
5705 | __ SmiTag(RAX); |
5706 | __ SmiTag(RDX); |
5707 | // Note that the result of an integer division/modulo of two |
5708 | // in-range arguments, cannot create out-of-range result. |
5709 | } |
5710 | |
5711 | LocationSummary* BranchInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
5712 | comparison()->InitializeLocationSummary(zone, opt); |
5713 | // Branches don't produce a result. |
5714 | comparison()->locs()->set_out(0, Location::NoLocation()); |
5715 | return comparison()->locs(); |
5716 | } |
5717 | |
5718 | void BranchInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5719 | comparison()->EmitBranchCode(compiler, this); |
5720 | } |
5721 | |
5722 | LocationSummary* CheckClassInstr::MakeLocationSummary(Zone* zone, |
5723 | bool opt) const { |
5724 | const intptr_t kNumInputs = 1; |
5725 | const bool need_mask_temp = IsBitTest(); |
5726 | const intptr_t kNumTemps = !IsNullCheck() ? (need_mask_temp ? 2 : 1) : 0; |
5727 | LocationSummary* summary = new (zone) |
5728 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5729 | summary->set_in(0, Location::RequiresRegister()); |
5730 | if (!IsNullCheck()) { |
5731 | summary->set_temp(0, Location::RequiresRegister()); |
5732 | if (need_mask_temp) { |
5733 | summary->set_temp(1, Location::RequiresRegister()); |
5734 | } |
5735 | } |
5736 | return summary; |
5737 | } |
5738 | |
5739 | void CheckClassInstr::EmitNullCheck(FlowGraphCompiler* compiler, |
5740 | compiler::Label* deopt) { |
5741 | __ CompareObject(locs()->in(0).reg(), Object::null_object()); |
5742 | Condition cond = IsDeoptIfNull() ? EQUAL : NOT_EQUAL; |
5743 | __ j(cond, deopt); |
5744 | } |
5745 | |
5746 | void CheckClassInstr::EmitBitTest(FlowGraphCompiler* compiler, |
5747 | intptr_t min, |
5748 | intptr_t max, |
5749 | intptr_t mask, |
5750 | compiler::Label* deopt) { |
5751 | Register biased_cid = locs()->temp(0).reg(); |
5752 | __ subq(biased_cid, compiler::Immediate(min)); |
5753 | __ cmpq(biased_cid, compiler::Immediate(max - min)); |
5754 | __ j(ABOVE, deopt); |
5755 | |
5756 | Register mask_reg = locs()->temp(1).reg(); |
5757 | __ movq(mask_reg, compiler::Immediate(mask)); |
5758 | __ btq(mask_reg, biased_cid); |
5759 | __ j(NOT_CARRY, deopt); |
5760 | } |
5761 | |
5762 | int CheckClassInstr::EmitCheckCid(FlowGraphCompiler* compiler, |
5763 | int bias, |
5764 | intptr_t cid_start, |
5765 | intptr_t cid_end, |
5766 | bool is_last, |
5767 | compiler::Label* is_ok, |
5768 | compiler::Label* deopt, |
5769 | bool use_near_jump) { |
5770 | Register biased_cid = locs()->temp(0).reg(); |
5771 | Condition no_match, match; |
5772 | if (cid_start == cid_end) { |
5773 | __ cmpl(biased_cid, compiler::Immediate(cid_start - bias)); |
5774 | no_match = NOT_EQUAL; |
5775 | match = EQUAL; |
5776 | } else { |
5777 | // For class ID ranges use a subtract followed by an unsigned |
5778 | // comparison to check both ends of the ranges with one comparison. |
5779 | __ addl(biased_cid, compiler::Immediate(bias - cid_start)); |
5780 | bias = cid_start; |
5781 | __ cmpl(biased_cid, compiler::Immediate(cid_end - cid_start)); |
5782 | no_match = ABOVE; |
5783 | match = BELOW_EQUAL; |
5784 | } |
5785 | |
5786 | if (is_last) { |
5787 | __ j(no_match, deopt); |
5788 | } else { |
5789 | if (use_near_jump) { |
5790 | __ j(match, is_ok, compiler::Assembler::kNearJump); |
5791 | } else { |
5792 | __ j(match, is_ok); |
5793 | } |
5794 | } |
5795 | return bias; |
5796 | } |
5797 | |
5798 | LocationSummary* CheckSmiInstr::MakeLocationSummary(Zone* zone, |
5799 | bool opt) const { |
5800 | const intptr_t kNumInputs = 1; |
5801 | const intptr_t kNumTemps = 0; |
5802 | LocationSummary* summary = new (zone) |
5803 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5804 | summary->set_in(0, Location::RequiresRegister()); |
5805 | return summary; |
5806 | } |
5807 | |
5808 | void CheckSmiInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5809 | Register value = locs()->in(0).reg(); |
5810 | compiler::Label* deopt = compiler->AddDeoptStub( |
5811 | deopt_id(), ICData::kDeoptCheckSmi, licm_hoisted_ ? ICData::kHoisted : 0); |
5812 | __ BranchIfNotSmi(value, deopt); |
5813 | } |
5814 | |
5815 | void CheckNullInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5816 | ThrowErrorSlowPathCode* slow_path = |
5817 | new NullErrorSlowPath(this, compiler->CurrentTryIndex()); |
5818 | compiler->AddSlowPathCode(slow_path); |
5819 | |
5820 | Register value_reg = locs()->in(0).reg(); |
5821 | // TODO(dartbug.com/30480): Consider passing `null` literal as an argument |
5822 | // in order to be able to allocate it on register. |
5823 | __ CompareObject(value_reg, Object::null_object()); |
5824 | __ BranchIf(EQUAL, slow_path->entry_label()); |
5825 | } |
5826 | |
5827 | LocationSummary* CheckClassIdInstr::MakeLocationSummary(Zone* zone, |
5828 | bool opt) const { |
5829 | const intptr_t kNumInputs = 1; |
5830 | const intptr_t kNumTemps = 0; |
5831 | LocationSummary* summary = new (zone) |
5832 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5833 | summary->set_in(0, cids_.IsSingleCid() ? Location::RequiresRegister() |
5834 | : Location::WritableRegister()); |
5835 | return summary; |
5836 | } |
5837 | |
5838 | void CheckClassIdInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5839 | Register value = locs()->in(0).reg(); |
5840 | compiler::Label* deopt = |
5841 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptCheckClass); |
5842 | if (cids_.IsSingleCid()) { |
5843 | __ CompareImmediate(value, |
5844 | compiler::Immediate(Smi::RawValue(cids_.cid_start))); |
5845 | __ j(NOT_ZERO, deopt); |
5846 | } else { |
5847 | __ AddImmediate(value, |
5848 | compiler::Immediate(-Smi::RawValue(cids_.cid_start))); |
5849 | __ cmpq(value, compiler::Immediate(Smi::RawValue(cids_.Extent()))); |
5850 | __ j(ABOVE, deopt); |
5851 | } |
5852 | } |
5853 | |
5854 | LocationSummary* CheckConditionInstr::MakeLocationSummary(Zone* zone, |
5855 | bool opt) const { |
5856 | comparison()->InitializeLocationSummary(zone, opt); |
5857 | comparison()->locs()->set_out(0, Location::NoLocation()); |
5858 | return comparison()->locs(); |
5859 | } |
5860 | |
5861 | void CheckConditionInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5862 | compiler::Label if_true; |
5863 | compiler::Label* if_false = |
5864 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptUnknown); |
5865 | BranchLabels labels = {&if_true, if_false, &if_true}; |
5866 | Condition true_condition = comparison()->EmitComparisonCode(compiler, labels); |
5867 | if (true_condition != kInvalidCondition) { |
5868 | __ j(InvertCondition(true_condition), if_false); |
5869 | } |
5870 | __ Bind(&if_true); |
5871 | } |
5872 | |
5873 | LocationSummary* CheckArrayBoundInstr::MakeLocationSummary(Zone* zone, |
5874 | bool opt) const { |
5875 | const intptr_t kNumInputs = 2; |
5876 | const intptr_t kNumTemps = 0; |
5877 | LocationSummary* locs = new (zone) |
5878 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
5879 | locs->set_in(kLengthPos, LocationRegisterOrSmiConstant(length())); |
5880 | locs->set_in(kIndexPos, LocationRegisterOrSmiConstant(index())); |
5881 | return locs; |
5882 | } |
5883 | |
5884 | void CheckArrayBoundInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
5885 | uint32_t flags = generalized_ ? ICData::kGeneralized : 0; |
5886 | flags |= licm_hoisted_ ? ICData::kHoisted : 0; |
5887 | compiler::Label* deopt = |
5888 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptCheckArrayBound, flags); |
5889 | |
5890 | Location length_loc = locs()->in(kLengthPos); |
5891 | Location index_loc = locs()->in(kIndexPos); |
5892 | |
5893 | if (length_loc.IsConstant() && index_loc.IsConstant()) { |
5894 | ASSERT((Smi::Cast(length_loc.constant()).Value() <= |
5895 | Smi::Cast(index_loc.constant()).Value()) || |
5896 | (Smi::Cast(index_loc.constant()).Value() < 0)); |
5897 | // Unconditionally deoptimize for constant bounds checks because they |
5898 | // only occur only when index is out-of-bounds. |
5899 | __ jmp(deopt); |
5900 | return; |
5901 | } |
5902 | |
5903 | const intptr_t index_cid = index()->Type()->ToCid(); |
5904 | if (index_loc.IsConstant()) { |
5905 | Register length = length_loc.reg(); |
5906 | const Smi& index = Smi::Cast(index_loc.constant()); |
5907 | __ CompareImmediate(length, |
5908 | compiler::Immediate(static_cast<int64_t>(index.raw()))); |
5909 | __ j(BELOW_EQUAL, deopt); |
5910 | } else if (length_loc.IsConstant()) { |
5911 | const Smi& length = Smi::Cast(length_loc.constant()); |
5912 | Register index = index_loc.reg(); |
5913 | if (index_cid != kSmiCid) { |
5914 | __ BranchIfNotSmi(index, deopt); |
5915 | } |
5916 | if (length.Value() == Smi::kMaxValue) { |
5917 | __ testq(index, index); |
5918 | __ j(NEGATIVE, deopt); |
5919 | } else { |
5920 | __ CompareImmediate( |
5921 | index, compiler::Immediate(static_cast<int64_t>(length.raw()))); |
5922 | __ j(ABOVE_EQUAL, deopt); |
5923 | } |
5924 | } else { |
5925 | Register length = length_loc.reg(); |
5926 | Register index = index_loc.reg(); |
5927 | if (index_cid != kSmiCid) { |
5928 | __ BranchIfNotSmi(index, deopt); |
5929 | } |
5930 | __ cmpq(index, length); |
5931 | __ j(ABOVE_EQUAL, deopt); |
5932 | } |
5933 | } |
5934 | |
5935 | class Int64DivideSlowPath : public ThrowErrorSlowPathCode { |
5936 | public: |
5937 | static const intptr_t kNumberOfArguments = 0; |
5938 | |
5939 | Int64DivideSlowPath(BinaryInt64OpInstr* instruction, |
5940 | Register divisor, |
5941 | Range* divisor_range, |
5942 | intptr_t try_index) |
5943 | : ThrowErrorSlowPathCode(instruction, |
5944 | kIntegerDivisionByZeroExceptionRuntimeEntry, |
5945 | kNumberOfArguments, |
5946 | try_index), |
5947 | is_mod_(instruction->op_kind() == Token::kMOD), |
5948 | divisor_(divisor), |
5949 | divisor_range_(divisor_range), |
5950 | div_by_minus_one_label_(), |
5951 | adjust_sign_label_() {} |
5952 | |
5953 | void EmitNativeCode(FlowGraphCompiler* compiler) override { |
5954 | // Handle modulo/division by zero, if needed. Use superclass code. |
5955 | if (has_divide_by_zero()) { |
5956 | ThrowErrorSlowPathCode::EmitNativeCode(compiler); |
5957 | } else { |
5958 | __ Bind(entry_label()); // not used, but keeps destructor happy |
5959 | if (compiler::Assembler::EmittingComments()) { |
5960 | __ Comment("slow path %s operation (no throw)" , name()); |
5961 | } |
5962 | } |
5963 | // Handle modulo/division by minus one, if needed. |
5964 | // Note: an exact -1 divisor is best optimized prior to codegen. |
5965 | if (has_divide_by_minus_one()) { |
5966 | __ Bind(div_by_minus_one_label()); |
5967 | if (is_mod_) { |
5968 | __ xorq(RDX, RDX); // x % -1 = 0 |
5969 | } else { |
5970 | __ negq(RAX); // x / -1 = -x |
5971 | } |
5972 | __ jmp(exit_label()); |
5973 | } |
5974 | // Adjust modulo for negative sign, optimized for known ranges. |
5975 | // if (divisor < 0) |
5976 | // out -= divisor; |
5977 | // else |
5978 | // out += divisor; |
5979 | if (has_adjust_sign()) { |
5980 | __ Bind(adjust_sign_label()); |
5981 | if (RangeUtils::Overlaps(divisor_range_, -1, 1)) { |
5982 | // General case. |
5983 | compiler::Label subtract; |
5984 | __ testq(divisor_, divisor_); |
5985 | __ j(LESS, &subtract, compiler::Assembler::kNearJump); |
5986 | __ addq(RDX, divisor_); |
5987 | __ jmp(exit_label()); |
5988 | __ Bind(&subtract); |
5989 | __ subq(RDX, divisor_); |
5990 | } else if (divisor_range_->IsPositive()) { |
5991 | // Always positive. |
5992 | __ addq(RDX, divisor_); |
5993 | } else { |
5994 | // Always negative. |
5995 | __ subq(RDX, divisor_); |
5996 | } |
5997 | __ jmp(exit_label()); |
5998 | } |
5999 | } |
6000 | |
6001 | const char* name() override { return "int64 divide" ; } |
6002 | |
6003 | bool has_divide_by_zero() { return RangeUtils::CanBeZero(divisor_range_); } |
6004 | |
6005 | bool has_divide_by_minus_one() { |
6006 | return RangeUtils::Overlaps(divisor_range_, -1, -1); |
6007 | } |
6008 | |
6009 | bool has_adjust_sign() { return is_mod_; } |
6010 | |
6011 | bool is_needed() { |
6012 | return has_divide_by_zero() || has_divide_by_minus_one() || |
6013 | has_adjust_sign(); |
6014 | } |
6015 | |
6016 | compiler::Label* div_by_minus_one_label() { |
6017 | ASSERT(has_divide_by_minus_one()); |
6018 | return &div_by_minus_one_label_; |
6019 | } |
6020 | |
6021 | compiler::Label* adjust_sign_label() { |
6022 | ASSERT(has_adjust_sign()); |
6023 | return &adjust_sign_label_; |
6024 | } |
6025 | |
6026 | private: |
6027 | bool is_mod_; |
6028 | Register divisor_; |
6029 | Range* divisor_range_; |
6030 | compiler::Label div_by_minus_one_label_; |
6031 | compiler::Label adjust_sign_label_; |
6032 | }; |
6033 | |
6034 | static void EmitInt64ModTruncDiv(FlowGraphCompiler* compiler, |
6035 | BinaryInt64OpInstr* instruction, |
6036 | Token::Kind op_kind, |
6037 | Register left, |
6038 | Register right, |
6039 | Register tmp, |
6040 | Register out) { |
6041 | ASSERT(op_kind == Token::kMOD || op_kind == Token::kTRUNCDIV); |
6042 | |
6043 | // Special case 64-bit div/mod by compile-time constant. Note that various |
6044 | // special constants (such as powers of two) should have been optimized |
6045 | // earlier in the pipeline. Div or mod by zero falls into general code |
6046 | // to implement the exception. |
6047 | if (auto c = instruction->right()->definition()->AsConstant()) { |
6048 | if (c->value().IsInteger()) { |
6049 | const int64_t divisor = Integer::Cast(c->value()).AsInt64Value(); |
6050 | if (divisor <= -2 || divisor >= 2) { |
6051 | // For x DIV c or x MOD c: use magic operations. |
6052 | compiler::Label pos; |
6053 | int64_t magic = 0; |
6054 | int64_t shift = 0; |
6055 | Utils::CalculateMagicAndShiftForDivRem(divisor, &magic, &shift); |
6056 | // RDX:RAX = magic * numerator. |
6057 | ASSERT(left == RAX); |
6058 | __ MoveRegister(TMP, RAX); // save numerator |
6059 | __ LoadImmediate(RAX, compiler::Immediate(magic)); |
6060 | __ imulq(TMP); |
6061 | // RDX +/-= numerator. |
6062 | if (divisor > 0 && magic < 0) { |
6063 | __ addq(RDX, TMP); |
6064 | } else if (divisor < 0 && magic > 0) { |
6065 | __ subq(RDX, TMP); |
6066 | } |
6067 | // Shift if needed. |
6068 | if (shift != 0) { |
6069 | __ sarq(RDX, compiler::Immediate(shift)); |
6070 | } |
6071 | // RDX += 1 if RDX < 0. |
6072 | __ movq(RAX, RDX); |
6073 | __ shrq(RDX, compiler::Immediate(63)); |
6074 | __ addq(RDX, RAX); |
6075 | // Finalize DIV or MOD. |
6076 | if (op_kind == Token::kTRUNCDIV) { |
6077 | ASSERT(out == RAX && tmp == RDX); |
6078 | __ movq(RAX, RDX); |
6079 | } else { |
6080 | ASSERT(out == RDX && tmp == RAX); |
6081 | __ movq(RAX, TMP); |
6082 | __ LoadImmediate(TMP, compiler::Immediate(divisor)); |
6083 | __ imulq(RDX, TMP); |
6084 | __ subq(RAX, RDX); |
6085 | // Compensate for Dart's Euclidean view of MOD. |
6086 | __ testq(RAX, RAX); |
6087 | __ j(GREATER_EQUAL, &pos); |
6088 | if (divisor > 0) { |
6089 | __ addq(RAX, TMP); |
6090 | } else { |
6091 | __ subq(RAX, TMP); |
6092 | } |
6093 | __ Bind(&pos); |
6094 | __ movq(RDX, RAX); |
6095 | } |
6096 | return; |
6097 | } |
6098 | } |
6099 | } |
6100 | |
6101 | // Prepare a slow path. |
6102 | Range* right_range = instruction->right()->definition()->range(); |
6103 | Int64DivideSlowPath* slow_path = new (Z) Int64DivideSlowPath( |
6104 | instruction, right, right_range, compiler->CurrentTryIndex()); |
6105 | |
6106 | // Handle modulo/division by zero exception on slow path. |
6107 | if (slow_path->has_divide_by_zero()) { |
6108 | __ testq(right, right); |
6109 | __ j(EQUAL, slow_path->entry_label()); |
6110 | } |
6111 | |
6112 | // Handle modulo/division by minus one explicitly on slow path |
6113 | // (to avoid arithmetic exception on 0x8000000000000000 / -1). |
6114 | if (slow_path->has_divide_by_minus_one()) { |
6115 | __ cmpq(right, compiler::Immediate(-1)); |
6116 | __ j(EQUAL, slow_path->div_by_minus_one_label()); |
6117 | } |
6118 | |
6119 | // Perform actual operation |
6120 | // out = left % right |
6121 | // or |
6122 | // out = left / right. |
6123 | // |
6124 | // Note that since 64-bit division requires twice as many cycles |
6125 | // and has much higher latency compared to the 32-bit division, |
6126 | // even for this non-speculative 64-bit path we add a "fast path". |
6127 | // Integers are untagged at this stage, so testing if sign extending |
6128 | // the lower half of each operand equals the full operand, effectively |
6129 | // tests if the values fit in 32-bit operands (and the slightly |
6130 | // dangerous division by -1 has been handled above already). |
6131 | ASSERT(left == RAX); |
6132 | ASSERT(right != RDX); // available at this stage |
6133 | compiler::Label div_64; |
6134 | compiler::Label div_merge; |
6135 | __ movsxd(RDX, left); |
6136 | __ cmpq(RDX, left); |
6137 | __ j(NOT_EQUAL, &div_64, compiler::Assembler::kNearJump); |
6138 | __ movsxd(RDX, right); |
6139 | __ cmpq(RDX, right); |
6140 | __ j(NOT_EQUAL, &div_64, compiler::Assembler::kNearJump); |
6141 | __ cdq(); // sign-ext eax into edx:eax |
6142 | __ idivl(right); // quotient eax, remainder edx |
6143 | __ movsxd(out, out); |
6144 | __ jmp(&div_merge, compiler::Assembler::kNearJump); |
6145 | __ Bind(&div_64); |
6146 | __ cqo(); // sign-ext rax into rdx:rax |
6147 | __ idivq(right); // quotient rax, remainder rdx |
6148 | __ Bind(&div_merge); |
6149 | if (op_kind == Token::kMOD) { |
6150 | ASSERT(out == RDX); |
6151 | ASSERT(tmp == RAX); |
6152 | // For the % operator, again the idiv instruction does |
6153 | // not quite do what we want. Adjust for sign on slow path. |
6154 | __ testq(out, out); |
6155 | __ j(LESS, slow_path->adjust_sign_label()); |
6156 | } else { |
6157 | ASSERT(out == RAX); |
6158 | ASSERT(tmp == RDX); |
6159 | } |
6160 | |
6161 | if (slow_path->is_needed()) { |
6162 | __ Bind(slow_path->exit_label()); |
6163 | compiler->AddSlowPathCode(slow_path); |
6164 | } |
6165 | } |
6166 | |
6167 | template <typename OperandType> |
6168 | static void EmitInt64Arithmetic(FlowGraphCompiler* compiler, |
6169 | Token::Kind op_kind, |
6170 | Register left, |
6171 | const OperandType& right) { |
6172 | switch (op_kind) { |
6173 | case Token::kADD: |
6174 | __ addq(left, right); |
6175 | break; |
6176 | case Token::kSUB: |
6177 | __ subq(left, right); |
6178 | break; |
6179 | case Token::kBIT_AND: |
6180 | __ andq(left, right); |
6181 | break; |
6182 | case Token::kBIT_OR: |
6183 | __ orq(left, right); |
6184 | break; |
6185 | case Token::kBIT_XOR: |
6186 | __ xorq(left, right); |
6187 | break; |
6188 | case Token::kMUL: |
6189 | __ imulq(left, right); |
6190 | break; |
6191 | default: |
6192 | UNREACHABLE(); |
6193 | } |
6194 | } |
6195 | |
6196 | LocationSummary* BinaryInt64OpInstr::MakeLocationSummary(Zone* zone, |
6197 | bool opt) const { |
6198 | switch (op_kind()) { |
6199 | case Token::kMOD: |
6200 | case Token::kTRUNCDIV: { |
6201 | const intptr_t kNumInputs = 2; |
6202 | const intptr_t kNumTemps = 1; |
6203 | LocationSummary* summary = new (zone) LocationSummary( |
6204 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
6205 | summary->set_in(0, Location::RegisterLocation(RAX)); |
6206 | summary->set_in(1, Location::RequiresRegister()); |
6207 | // Intel uses rdx:rax with quotient rax and remainder rdx. Pick the |
6208 | // appropriate one for output and reserve the other as temp. |
6209 | summary->set_out( |
6210 | 0, Location::RegisterLocation(op_kind() == Token::kMOD ? RDX : RAX)); |
6211 | summary->set_temp( |
6212 | 0, Location::RegisterLocation(op_kind() == Token::kMOD ? RAX : RDX)); |
6213 | return summary; |
6214 | } |
6215 | default: { |
6216 | const intptr_t kNumInputs = 2; |
6217 | const intptr_t kNumTemps = 0; |
6218 | LocationSummary* summary = new (zone) LocationSummary( |
6219 | zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6220 | summary->set_in(0, Location::RequiresRegister()); |
6221 | summary->set_in(1, LocationRegisterOrConstant(right())); |
6222 | summary->set_out(0, Location::SameAsFirstInput()); |
6223 | return summary; |
6224 | } |
6225 | } |
6226 | } |
6227 | |
6228 | void BinaryInt64OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6229 | const Location left = locs()->in(0); |
6230 | const Location right = locs()->in(1); |
6231 | const Location out = locs()->out(0); |
6232 | ASSERT(!can_overflow()); |
6233 | ASSERT(!CanDeoptimize()); |
6234 | |
6235 | if (op_kind() == Token::kMOD || op_kind() == Token::kTRUNCDIV) { |
6236 | const Location temp = locs()->temp(0); |
6237 | EmitInt64ModTruncDiv(compiler, this, op_kind(), left.reg(), right.reg(), |
6238 | temp.reg(), out.reg()); |
6239 | } else if (right.IsConstant()) { |
6240 | ASSERT(out.reg() == left.reg()); |
6241 | ConstantInstr* constant_instr = right.constant_instruction(); |
6242 | const int64_t value = |
6243 | constant_instr->GetUnboxedSignedIntegerConstantValue(); |
6244 | EmitInt64Arithmetic(compiler, op_kind(), left.reg(), |
6245 | compiler::Immediate(value)); |
6246 | } else { |
6247 | ASSERT(out.reg() == left.reg()); |
6248 | EmitInt64Arithmetic(compiler, op_kind(), left.reg(), right.reg()); |
6249 | } |
6250 | } |
6251 | |
6252 | LocationSummary* UnaryInt64OpInstr::MakeLocationSummary(Zone* zone, |
6253 | bool opt) const { |
6254 | const intptr_t kNumInputs = 1; |
6255 | const intptr_t kNumTemps = 0; |
6256 | LocationSummary* summary = new (zone) |
6257 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6258 | summary->set_in(0, Location::RequiresRegister()); |
6259 | summary->set_out(0, Location::SameAsFirstInput()); |
6260 | return summary; |
6261 | } |
6262 | |
6263 | void UnaryInt64OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6264 | const Register left = locs()->in(0).reg(); |
6265 | const Register out = locs()->out(0).reg(); |
6266 | ASSERT(out == left); |
6267 | switch (op_kind()) { |
6268 | case Token::kBIT_NOT: |
6269 | __ notq(left); |
6270 | break; |
6271 | case Token::kNEGATE: |
6272 | __ negq(left); |
6273 | break; |
6274 | default: |
6275 | UNREACHABLE(); |
6276 | } |
6277 | } |
6278 | |
6279 | static void EmitShiftInt64ByConstant(FlowGraphCompiler* compiler, |
6280 | Token::Kind op_kind, |
6281 | Register left, |
6282 | const Object& right) { |
6283 | const int64_t shift = Integer::Cast(right).AsInt64Value(); |
6284 | ASSERT(shift >= 0); |
6285 | switch (op_kind) { |
6286 | case Token::kSHR: |
6287 | __ sarq(left, compiler::Immediate( |
6288 | Utils::Minimum<int64_t>(shift, kBitsPerWord - 1))); |
6289 | break; |
6290 | case Token::kSHL: { |
6291 | ASSERT(shift < 64); |
6292 | __ shlq(left, compiler::Immediate(shift)); |
6293 | break; |
6294 | } |
6295 | default: |
6296 | UNREACHABLE(); |
6297 | } |
6298 | } |
6299 | |
6300 | static void EmitShiftInt64ByRCX(FlowGraphCompiler* compiler, |
6301 | Token::Kind op_kind, |
6302 | Register left) { |
6303 | switch (op_kind) { |
6304 | case Token::kSHR: { |
6305 | __ sarq(left, RCX); |
6306 | break; |
6307 | } |
6308 | case Token::kSHL: { |
6309 | __ shlq(left, RCX); |
6310 | break; |
6311 | } |
6312 | default: |
6313 | UNREACHABLE(); |
6314 | } |
6315 | } |
6316 | |
6317 | static void EmitShiftUint32ByConstant(FlowGraphCompiler* compiler, |
6318 | Token::Kind op_kind, |
6319 | Register left, |
6320 | const Object& right) { |
6321 | const int64_t shift = Integer::Cast(right).AsInt64Value(); |
6322 | ASSERT(shift >= 0); |
6323 | if (shift >= 32) { |
6324 | __ xorl(left, left); |
6325 | } else { |
6326 | switch (op_kind) { |
6327 | case Token::kSHR: { |
6328 | __ shrl(left, compiler::Immediate(shift)); |
6329 | break; |
6330 | } |
6331 | case Token::kSHL: { |
6332 | __ shll(left, compiler::Immediate(shift)); |
6333 | break; |
6334 | } |
6335 | default: |
6336 | UNREACHABLE(); |
6337 | } |
6338 | } |
6339 | } |
6340 | |
6341 | static void EmitShiftUint32ByRCX(FlowGraphCompiler* compiler, |
6342 | Token::Kind op_kind, |
6343 | Register left) { |
6344 | switch (op_kind) { |
6345 | case Token::kSHR: { |
6346 | __ shrl(left, RCX); |
6347 | break; |
6348 | } |
6349 | case Token::kSHL: { |
6350 | __ shll(left, RCX); |
6351 | break; |
6352 | } |
6353 | default: |
6354 | UNREACHABLE(); |
6355 | } |
6356 | } |
6357 | |
6358 | class ShiftInt64OpSlowPath : public ThrowErrorSlowPathCode { |
6359 | public: |
6360 | static const intptr_t kNumberOfArguments = 0; |
6361 | |
6362 | ShiftInt64OpSlowPath(ShiftInt64OpInstr* instruction, intptr_t try_index) |
6363 | : ThrowErrorSlowPathCode(instruction, |
6364 | kArgumentErrorUnboxedInt64RuntimeEntry, |
6365 | kNumberOfArguments, |
6366 | try_index) {} |
6367 | |
6368 | const char* name() override { return "int64 shift" ; } |
6369 | |
6370 | void EmitCodeAtSlowPathEntry(FlowGraphCompiler* compiler) override { |
6371 | const Register out = instruction()->locs()->out(0).reg(); |
6372 | ASSERT(out == instruction()->locs()->in(0).reg()); |
6373 | |
6374 | compiler::Label throw_error; |
6375 | __ testq(RCX, RCX); |
6376 | __ j(LESS, &throw_error); |
6377 | |
6378 | switch (instruction()->AsShiftInt64Op()->op_kind()) { |
6379 | case Token::kSHR: |
6380 | __ sarq(out, compiler::Immediate(kBitsPerInt64 - 1)); |
6381 | break; |
6382 | case Token::kSHL: |
6383 | __ xorq(out, out); |
6384 | break; |
6385 | default: |
6386 | UNREACHABLE(); |
6387 | } |
6388 | __ jmp(exit_label()); |
6389 | |
6390 | __ Bind(&throw_error); |
6391 | |
6392 | // Can't pass unboxed int64 value directly to runtime call, as all |
6393 | // arguments are expected to be tagged (boxed). |
6394 | // The unboxed int64 argument is passed through a dedicated slot in Thread. |
6395 | // TODO(dartbug.com/33549): Clean this up when unboxed values |
6396 | // could be passed as arguments. |
6397 | __ movq(compiler::Address(THR, Thread::unboxed_int64_runtime_arg_offset()), |
6398 | RCX); |
6399 | } |
6400 | }; |
6401 | |
6402 | LocationSummary* ShiftInt64OpInstr::MakeLocationSummary(Zone* zone, |
6403 | bool opt) const { |
6404 | const intptr_t kNumInputs = 2; |
6405 | const intptr_t kNumTemps = 0; |
6406 | LocationSummary* summary = new (zone) LocationSummary( |
6407 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
6408 | summary->set_in(0, Location::RequiresRegister()); |
6409 | summary->set_in(1, RangeUtils::IsPositive(shift_range()) |
6410 | ? LocationFixedRegisterOrConstant(right(), RCX) |
6411 | : Location::RegisterLocation(RCX)); |
6412 | summary->set_out(0, Location::SameAsFirstInput()); |
6413 | return summary; |
6414 | } |
6415 | |
6416 | void ShiftInt64OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6417 | const Register left = locs()->in(0).reg(); |
6418 | const Register out = locs()->out(0).reg(); |
6419 | ASSERT(left == out); |
6420 | ASSERT(!can_overflow()); |
6421 | |
6422 | if (locs()->in(1).IsConstant()) { |
6423 | EmitShiftInt64ByConstant(compiler, op_kind(), left, |
6424 | locs()->in(1).constant()); |
6425 | } else { |
6426 | // Code for a variable shift amount (or constant that throws). |
6427 | ASSERT(locs()->in(1).reg() == RCX); |
6428 | |
6429 | // Jump to a slow path if shift count is > 63 or negative. |
6430 | ShiftInt64OpSlowPath* slow_path = NULL; |
6431 | if (!IsShiftCountInRange()) { |
6432 | slow_path = |
6433 | new (Z) ShiftInt64OpSlowPath(this, compiler->CurrentTryIndex()); |
6434 | compiler->AddSlowPathCode(slow_path); |
6435 | |
6436 | __ cmpq(RCX, compiler::Immediate(kShiftCountLimit)); |
6437 | __ j(ABOVE, slow_path->entry_label()); |
6438 | } |
6439 | |
6440 | EmitShiftInt64ByRCX(compiler, op_kind(), left); |
6441 | |
6442 | if (slow_path != NULL) { |
6443 | __ Bind(slow_path->exit_label()); |
6444 | } |
6445 | } |
6446 | } |
6447 | |
6448 | LocationSummary* SpeculativeShiftInt64OpInstr::MakeLocationSummary( |
6449 | Zone* zone, |
6450 | bool opt) const { |
6451 | const intptr_t kNumInputs = 2; |
6452 | const intptr_t kNumTemps = 0; |
6453 | LocationSummary* summary = new (zone) |
6454 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6455 | summary->set_in(0, Location::RequiresRegister()); |
6456 | summary->set_in(1, LocationFixedRegisterOrSmiConstant(right(), RCX)); |
6457 | summary->set_out(0, Location::SameAsFirstInput()); |
6458 | return summary; |
6459 | } |
6460 | |
6461 | void SpeculativeShiftInt64OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6462 | const Register left = locs()->in(0).reg(); |
6463 | const Register out = locs()->out(0).reg(); |
6464 | ASSERT(left == out); |
6465 | ASSERT(!can_overflow()); |
6466 | |
6467 | if (locs()->in(1).IsConstant()) { |
6468 | EmitShiftInt64ByConstant(compiler, op_kind(), left, |
6469 | locs()->in(1).constant()); |
6470 | } else { |
6471 | ASSERT(locs()->in(1).reg() == RCX); |
6472 | __ SmiUntag(RCX); |
6473 | |
6474 | // Deoptimize if shift count is > 63 or negative (or not a smi). |
6475 | if (!IsShiftCountInRange()) { |
6476 | ASSERT(CanDeoptimize()); |
6477 | compiler::Label* deopt = |
6478 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptBinaryInt64Op); |
6479 | |
6480 | __ cmpq(RCX, compiler::Immediate(kShiftCountLimit)); |
6481 | __ j(ABOVE, deopt); |
6482 | } |
6483 | |
6484 | EmitShiftInt64ByRCX(compiler, op_kind(), left); |
6485 | } |
6486 | } |
6487 | |
6488 | class ShiftUint32OpSlowPath : public ThrowErrorSlowPathCode { |
6489 | public: |
6490 | static const intptr_t kNumberOfArguments = 0; |
6491 | |
6492 | ShiftUint32OpSlowPath(ShiftUint32OpInstr* instruction, intptr_t try_index) |
6493 | : ThrowErrorSlowPathCode(instruction, |
6494 | kArgumentErrorUnboxedInt64RuntimeEntry, |
6495 | kNumberOfArguments, |
6496 | try_index) {} |
6497 | |
6498 | const char* name() override { return "uint32 shift" ; } |
6499 | |
6500 | void EmitCodeAtSlowPathEntry(FlowGraphCompiler* compiler) override { |
6501 | const Register out = instruction()->locs()->out(0).reg(); |
6502 | ASSERT(out == instruction()->locs()->in(0).reg()); |
6503 | |
6504 | compiler::Label throw_error; |
6505 | __ testq(RCX, RCX); |
6506 | __ j(LESS, &throw_error); |
6507 | |
6508 | __ xorl(out, out); |
6509 | __ jmp(exit_label()); |
6510 | |
6511 | __ Bind(&throw_error); |
6512 | |
6513 | // Can't pass unboxed int64 value directly to runtime call, as all |
6514 | // arguments are expected to be tagged (boxed). |
6515 | // The unboxed int64 argument is passed through a dedicated slot in Thread. |
6516 | // TODO(dartbug.com/33549): Clean this up when unboxed values |
6517 | // could be passed as arguments. |
6518 | __ movq(compiler::Address(THR, Thread::unboxed_int64_runtime_arg_offset()), |
6519 | RCX); |
6520 | } |
6521 | }; |
6522 | |
6523 | LocationSummary* ShiftUint32OpInstr::MakeLocationSummary(Zone* zone, |
6524 | bool opt) const { |
6525 | const intptr_t kNumInputs = 2; |
6526 | const intptr_t kNumTemps = 0; |
6527 | LocationSummary* summary = new (zone) LocationSummary( |
6528 | zone, kNumInputs, kNumTemps, LocationSummary::kCallOnSlowPath); |
6529 | summary->set_in(0, Location::RequiresRegister()); |
6530 | summary->set_in(1, RangeUtils::IsPositive(shift_range()) |
6531 | ? LocationFixedRegisterOrConstant(right(), RCX) |
6532 | : Location::RegisterLocation(RCX)); |
6533 | summary->set_out(0, Location::SameAsFirstInput()); |
6534 | return summary; |
6535 | } |
6536 | |
6537 | void ShiftUint32OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6538 | Register left = locs()->in(0).reg(); |
6539 | Register out = locs()->out(0).reg(); |
6540 | ASSERT(left == out); |
6541 | |
6542 | if (locs()->in(1).IsConstant()) { |
6543 | EmitShiftUint32ByConstant(compiler, op_kind(), left, |
6544 | locs()->in(1).constant()); |
6545 | } else { |
6546 | // Code for a variable shift amount (or constant that throws). |
6547 | ASSERT(locs()->in(1).reg() == RCX); |
6548 | |
6549 | // Jump to a slow path if shift count is > 31 or negative. |
6550 | ShiftUint32OpSlowPath* slow_path = NULL; |
6551 | if (!IsShiftCountInRange(kUint32ShiftCountLimit)) { |
6552 | slow_path = |
6553 | new (Z) ShiftUint32OpSlowPath(this, compiler->CurrentTryIndex()); |
6554 | compiler->AddSlowPathCode(slow_path); |
6555 | |
6556 | __ cmpq(RCX, compiler::Immediate(kUint32ShiftCountLimit)); |
6557 | __ j(ABOVE, slow_path->entry_label()); |
6558 | } |
6559 | |
6560 | EmitShiftUint32ByRCX(compiler, op_kind(), left); |
6561 | |
6562 | if (slow_path != NULL) { |
6563 | __ Bind(slow_path->exit_label()); |
6564 | } |
6565 | } |
6566 | } |
6567 | |
6568 | LocationSummary* SpeculativeShiftUint32OpInstr::MakeLocationSummary( |
6569 | Zone* zone, |
6570 | bool opt) const { |
6571 | const intptr_t kNumInputs = 2; |
6572 | const intptr_t kNumTemps = 0; |
6573 | LocationSummary* summary = new (zone) |
6574 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6575 | summary->set_in(0, Location::RequiresRegister()); |
6576 | summary->set_in(1, LocationFixedRegisterOrSmiConstant(right(), RCX)); |
6577 | summary->set_out(0, Location::SameAsFirstInput()); |
6578 | return summary; |
6579 | } |
6580 | |
6581 | void SpeculativeShiftUint32OpInstr::EmitNativeCode( |
6582 | FlowGraphCompiler* compiler) { |
6583 | Register left = locs()->in(0).reg(); |
6584 | Register out = locs()->out(0).reg(); |
6585 | ASSERT(left == out); |
6586 | |
6587 | if (locs()->in(1).IsConstant()) { |
6588 | EmitShiftUint32ByConstant(compiler, op_kind(), left, |
6589 | locs()->in(1).constant()); |
6590 | } else { |
6591 | ASSERT(locs()->in(1).reg() == RCX); |
6592 | __ SmiUntag(RCX); |
6593 | |
6594 | if (!IsShiftCountInRange(kUint32ShiftCountLimit)) { |
6595 | if (!IsShiftCountInRange()) { |
6596 | // Deoptimize if shift count is negative. |
6597 | ASSERT(CanDeoptimize()); |
6598 | compiler::Label* deopt = |
6599 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptBinaryInt64Op); |
6600 | |
6601 | __ testq(RCX, RCX); |
6602 | __ j(LESS, deopt); |
6603 | } |
6604 | |
6605 | compiler::Label cont; |
6606 | __ cmpq(RCX, compiler::Immediate(kUint32ShiftCountLimit)); |
6607 | __ j(LESS_EQUAL, &cont); |
6608 | |
6609 | __ xorl(left, left); |
6610 | |
6611 | __ Bind(&cont); |
6612 | } |
6613 | |
6614 | EmitShiftUint32ByRCX(compiler, op_kind(), left); |
6615 | } |
6616 | } |
6617 | |
6618 | LocationSummary* BinaryUint32OpInstr::MakeLocationSummary(Zone* zone, |
6619 | bool opt) const { |
6620 | const intptr_t kNumInputs = 2; |
6621 | const intptr_t kNumTemps = 0; |
6622 | LocationSummary* summary = new (zone) |
6623 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6624 | summary->set_in(0, Location::RequiresRegister()); |
6625 | summary->set_in(1, Location::RequiresRegister()); |
6626 | summary->set_out(0, Location::SameAsFirstInput()); |
6627 | return summary; |
6628 | } |
6629 | |
6630 | template <typename OperandType> |
6631 | static void EmitIntegerArithmetic(FlowGraphCompiler* compiler, |
6632 | Token::Kind op_kind, |
6633 | Register left, |
6634 | const OperandType& right) { |
6635 | switch (op_kind) { |
6636 | case Token::kADD: |
6637 | __ addl(left, right); |
6638 | break; |
6639 | case Token::kSUB: |
6640 | __ subl(left, right); |
6641 | break; |
6642 | case Token::kBIT_AND: |
6643 | __ andl(left, right); |
6644 | break; |
6645 | case Token::kBIT_OR: |
6646 | __ orl(left, right); |
6647 | break; |
6648 | case Token::kBIT_XOR: |
6649 | __ xorl(left, right); |
6650 | break; |
6651 | case Token::kMUL: |
6652 | __ imull(left, right); |
6653 | break; |
6654 | default: |
6655 | UNREACHABLE(); |
6656 | } |
6657 | } |
6658 | |
6659 | void BinaryUint32OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6660 | Register left = locs()->in(0).reg(); |
6661 | Register right = locs()->in(1).reg(); |
6662 | Register out = locs()->out(0).reg(); |
6663 | ASSERT(out == left); |
6664 | switch (op_kind()) { |
6665 | case Token::kBIT_AND: |
6666 | case Token::kBIT_OR: |
6667 | case Token::kBIT_XOR: |
6668 | case Token::kADD: |
6669 | case Token::kSUB: |
6670 | case Token::kMUL: |
6671 | EmitIntegerArithmetic(compiler, op_kind(), left, right); |
6672 | return; |
6673 | default: |
6674 | UNREACHABLE(); |
6675 | } |
6676 | } |
6677 | |
6678 | DEFINE_BACKEND(UnaryUint32Op, (SameAsFirstInput, Register value)) { |
6679 | ASSERT(instr->op_kind() == Token::kBIT_NOT); |
6680 | __ notl(value); |
6681 | } |
6682 | |
6683 | DEFINE_UNIMPLEMENTED_INSTRUCTION(BinaryInt32OpInstr) |
6684 | |
6685 | LocationSummary* IntConverterInstr::MakeLocationSummary(Zone* zone, |
6686 | bool opt) const { |
6687 | const intptr_t kNumInputs = 1; |
6688 | const intptr_t kNumTemps = 0; |
6689 | LocationSummary* summary = new (zone) |
6690 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6691 | if (from() == kUntagged || to() == kUntagged) { |
6692 | ASSERT((from() == kUntagged && to() == kUnboxedIntPtr) || |
6693 | (from() == kUnboxedIntPtr && to() == kUntagged)); |
6694 | ASSERT(!CanDeoptimize()); |
6695 | } else if (from() == kUnboxedInt64) { |
6696 | ASSERT(to() == kUnboxedUint32 || to() == kUnboxedInt32); |
6697 | } else if (to() == kUnboxedInt64) { |
6698 | ASSERT(from() == kUnboxedInt32 || from() == kUnboxedUint32); |
6699 | } else { |
6700 | ASSERT(to() == kUnboxedUint32 || to() == kUnboxedInt32); |
6701 | ASSERT(from() == kUnboxedUint32 || from() == kUnboxedInt32); |
6702 | } |
6703 | |
6704 | summary->set_in(0, Location::RequiresRegister()); |
6705 | summary->set_out(0, Location::SameAsFirstInput()); |
6706 | |
6707 | return summary; |
6708 | } |
6709 | |
6710 | void IntConverterInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6711 | const bool is_nop_conversion = |
6712 | (from() == kUntagged && to() == kUnboxedIntPtr) || |
6713 | (from() == kUnboxedIntPtr && to() == kUntagged); |
6714 | if (is_nop_conversion) { |
6715 | ASSERT(locs()->in(0).reg() == locs()->out(0).reg()); |
6716 | return; |
6717 | } |
6718 | |
6719 | if (from() == kUnboxedInt32 && to() == kUnboxedUint32) { |
6720 | const Register value = locs()->in(0).reg(); |
6721 | const Register out = locs()->out(0).reg(); |
6722 | // Representations are bitwise equivalent but we want to normalize |
6723 | // upperbits for safety reasons. |
6724 | // TODO(vegorov) if we ensure that we never use upperbits we could |
6725 | // avoid this. |
6726 | __ movl(out, value); |
6727 | } else if (from() == kUnboxedUint32 && to() == kUnboxedInt32) { |
6728 | // Representations are bitwise equivalent. |
6729 | const Register value = locs()->in(0).reg(); |
6730 | const Register out = locs()->out(0).reg(); |
6731 | __ movsxd(out, value); |
6732 | if (CanDeoptimize()) { |
6733 | compiler::Label* deopt = |
6734 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptUnboxInteger); |
6735 | __ testl(out, out); |
6736 | __ j(NEGATIVE, deopt); |
6737 | } |
6738 | } else if (from() == kUnboxedInt64) { |
6739 | ASSERT(to() == kUnboxedUint32 || to() == kUnboxedInt32); |
6740 | const Register value = locs()->in(0).reg(); |
6741 | const Register out = locs()->out(0).reg(); |
6742 | if (!CanDeoptimize()) { |
6743 | // Copy low. |
6744 | __ movl(out, value); |
6745 | } else { |
6746 | compiler::Label* deopt = |
6747 | compiler->AddDeoptStub(deopt_id(), ICData::kDeoptUnboxInteger); |
6748 | // Sign extend. |
6749 | __ movsxd(out, value); |
6750 | // Compare with original value. |
6751 | __ cmpq(out, value); |
6752 | // Value cannot be held in Int32, deopt. |
6753 | __ j(NOT_EQUAL, deopt); |
6754 | } |
6755 | } else if (to() == kUnboxedInt64) { |
6756 | ASSERT((from() == kUnboxedUint32) || (from() == kUnboxedInt32)); |
6757 | const Register value = locs()->in(0).reg(); |
6758 | const Register out = locs()->out(0).reg(); |
6759 | if (from() == kUnboxedUint32) { |
6760 | // Zero extend. |
6761 | __ movl(out, value); |
6762 | } else { |
6763 | // Sign extend. |
6764 | ASSERT(from() == kUnboxedInt32); |
6765 | __ movsxd(out, value); |
6766 | } |
6767 | } else { |
6768 | UNREACHABLE(); |
6769 | } |
6770 | } |
6771 | |
6772 | LocationSummary* StopInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
6773 | return new (zone) LocationSummary(zone, 0, 0, LocationSummary::kNoCall); |
6774 | } |
6775 | |
6776 | void StopInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6777 | __ Stop(message()); |
6778 | } |
6779 | |
6780 | void GraphEntryInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6781 | BlockEntryInstr* entry = normal_entry(); |
6782 | if (entry != nullptr) { |
6783 | if (!compiler->CanFallThroughTo(entry)) { |
6784 | FATAL("Checked function entry must have no offset" ); |
6785 | } |
6786 | } else { |
6787 | entry = osr_entry(); |
6788 | if (!compiler->CanFallThroughTo(entry)) { |
6789 | __ jmp(compiler->GetJumpLabel(entry)); |
6790 | } |
6791 | } |
6792 | } |
6793 | |
6794 | LocationSummary* GotoInstr::MakeLocationSummary(Zone* zone, bool opt) const { |
6795 | return new (zone) LocationSummary(zone, 0, 0, LocationSummary::kNoCall); |
6796 | } |
6797 | |
6798 | void GotoInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6799 | if (!compiler->is_optimizing()) { |
6800 | if (FLAG_reorder_basic_blocks) { |
6801 | compiler->EmitEdgeCounter(block()->preorder_number()); |
6802 | } |
6803 | // Add a deoptimization descriptor for deoptimizing instructions that |
6804 | // may be inserted before this instruction. |
6805 | compiler->AddCurrentDescriptor(PcDescriptorsLayout::kDeopt, GetDeoptId(), |
6806 | TokenPosition::kNoSource); |
6807 | } |
6808 | if (HasParallelMove()) { |
6809 | compiler->parallel_move_resolver()->EmitNativeCode(parallel_move()); |
6810 | } |
6811 | |
6812 | // We can fall through if the successor is the next block in the list. |
6813 | // Otherwise, we need a jump. |
6814 | if (!compiler->CanFallThroughTo(successor())) { |
6815 | __ jmp(compiler->GetJumpLabel(successor())); |
6816 | } |
6817 | } |
6818 | |
6819 | LocationSummary* IndirectGotoInstr::MakeLocationSummary(Zone* zone, |
6820 | bool opt) const { |
6821 | const intptr_t kNumInputs = 1; |
6822 | const intptr_t kNumTemps = 1; |
6823 | |
6824 | LocationSummary* summary = new (zone) |
6825 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6826 | |
6827 | summary->set_in(0, Location::RequiresRegister()); |
6828 | summary->set_temp(0, Location::RequiresRegister()); |
6829 | |
6830 | return summary; |
6831 | } |
6832 | |
6833 | void IndirectGotoInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6834 | Register offset_reg = locs()->in(0).reg(); |
6835 | Register target_address_reg = locs()->temp(0).reg(); |
6836 | |
6837 | { |
6838 | const intptr_t kRIPRelativeLeaqSize = 7; |
6839 | const intptr_t entry_to_rip_offset = __ CodeSize() + kRIPRelativeLeaqSize; |
6840 | __ leaq(target_address_reg, |
6841 | compiler::Address::AddressRIPRelative(-entry_to_rip_offset)); |
6842 | ASSERT(__ CodeSize() == entry_to_rip_offset); |
6843 | } |
6844 | |
6845 | // Load from FP+compiler::target::frame_layout.code_from_fp. |
6846 | |
6847 | // Calculate the final absolute address. |
6848 | if (offset()->definition()->representation() == kTagged) { |
6849 | __ SmiUntag(offset_reg); |
6850 | } |
6851 | __ addq(target_address_reg, offset_reg); |
6852 | |
6853 | // Jump to the absolute address. |
6854 | __ jmp(target_address_reg); |
6855 | } |
6856 | |
6857 | LocationSummary* StrictCompareInstr::MakeLocationSummary(Zone* zone, |
6858 | bool opt) const { |
6859 | const intptr_t kNumInputs = 2; |
6860 | const intptr_t kNumTemps = 0; |
6861 | if (needs_number_check()) { |
6862 | LocationSummary* locs = new (zone) |
6863 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
6864 | locs->set_in(0, Location::RegisterLocation(RAX)); |
6865 | locs->set_in(1, Location::RegisterLocation(RCX)); |
6866 | locs->set_out(0, Location::RegisterLocation(RAX)); |
6867 | return locs; |
6868 | } |
6869 | LocationSummary* locs = new (zone) |
6870 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall); |
6871 | locs->set_in(0, LocationRegisterOrConstant(left())); |
6872 | // Only one of the inputs can be a constant. Choose register if the first one |
6873 | // is a constant. |
6874 | locs->set_in(1, locs->in(0).IsConstant() |
6875 | ? Location::RequiresRegister() |
6876 | : LocationRegisterOrConstant(right())); |
6877 | locs->set_out(0, Location::RequiresRegister()); |
6878 | return locs; |
6879 | } |
6880 | |
6881 | Condition StrictCompareInstr::EmitComparisonCodeRegConstant( |
6882 | FlowGraphCompiler* compiler, |
6883 | BranchLabels labels, |
6884 | Register reg, |
6885 | const Object& obj) { |
6886 | return compiler->EmitEqualityRegConstCompare(reg, obj, needs_number_check(), |
6887 | token_pos(), deopt_id()); |
6888 | } |
6889 | |
6890 | LocationSummary* DispatchTableCallInstr::MakeLocationSummary(Zone* zone, |
6891 | bool opt) const { |
6892 | const intptr_t kNumInputs = 1; |
6893 | const intptr_t kNumTemps = 0; |
6894 | LocationSummary* summary = new (zone) |
6895 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
6896 | summary->set_in(0, Location::RegisterLocation(RCX)); // ClassId |
6897 | return MakeCallSummary(zone, this, summary); |
6898 | } |
6899 | |
6900 | LocationSummary* ClosureCallInstr::MakeLocationSummary(Zone* zone, |
6901 | bool opt) const { |
6902 | const intptr_t kNumInputs = 1; |
6903 | const intptr_t kNumTemps = 0; |
6904 | LocationSummary* summary = new (zone) |
6905 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
6906 | summary->set_in(0, Location::RegisterLocation(RAX)); // Function. |
6907 | return MakeCallSummary(zone, this, summary); |
6908 | } |
6909 | |
6910 | void ClosureCallInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6911 | // Arguments descriptor is expected in R10. |
6912 | const intptr_t argument_count = ArgumentCount(); // Includes type args. |
6913 | const Array& arguments_descriptor = |
6914 | Array::ZoneHandle(Z, GetArgumentsDescriptor()); |
6915 | __ LoadObject(R10, arguments_descriptor); |
6916 | |
6917 | // Function in RAX. |
6918 | ASSERT(locs()->in(0).reg() == RAX); |
6919 | if (!FLAG_precompiled_mode || !FLAG_use_bare_instructions) { |
6920 | __ movq(CODE_REG, compiler::FieldAddress( |
6921 | RAX, compiler::target::Function::code_offset())); |
6922 | } |
6923 | __ movq( |
6924 | RCX, |
6925 | compiler::FieldAddress( |
6926 | RAX, compiler::target::Function::entry_point_offset(entry_kind()))); |
6927 | |
6928 | // RAX: Function. |
6929 | // R10: Arguments descriptor array. |
6930 | if (!FLAG_precompiled_mode) { |
6931 | // RBX: Smi 0 (no IC data; the lazy-compile stub expects a GC-safe value). |
6932 | __ xorq(RBX, RBX); |
6933 | } |
6934 | __ call(RCX); |
6935 | compiler->EmitCallsiteMetadata(token_pos(), deopt_id(), |
6936 | PcDescriptorsLayout::kOther, locs()); |
6937 | __ Drop(argument_count); |
6938 | } |
6939 | |
6940 | LocationSummary* BooleanNegateInstr::MakeLocationSummary(Zone* zone, |
6941 | bool opt) const { |
6942 | return LocationSummary::Make(zone, 1, |
6943 | value()->Type()->ToCid() == kBoolCid |
6944 | ? Location::SameAsFirstInput() |
6945 | : Location::RequiresRegister(), |
6946 | LocationSummary::kNoCall); |
6947 | } |
6948 | |
6949 | void BooleanNegateInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6950 | Register input = locs()->in(0).reg(); |
6951 | Register result = locs()->out(0).reg(); |
6952 | |
6953 | if (value()->Type()->ToCid() == kBoolCid) { |
6954 | ASSERT(input == result); |
6955 | __ xorq(result, compiler::Immediate( |
6956 | compiler::target::ObjectAlignment::kBoolValueMask)); |
6957 | } else { |
6958 | ASSERT(input != result); |
6959 | compiler::Label done; |
6960 | __ LoadObject(result, Bool::True()); |
6961 | __ CompareRegisters(result, input); |
6962 | __ j(NOT_EQUAL, &done, compiler::Assembler::kNearJump); |
6963 | __ LoadObject(result, Bool::False()); |
6964 | __ Bind(&done); |
6965 | } |
6966 | } |
6967 | |
6968 | LocationSummary* AllocateObjectInstr::MakeLocationSummary(Zone* zone, |
6969 | bool opt) const { |
6970 | const intptr_t kNumInputs = (type_arguments() != nullptr) ? 1 : 0; |
6971 | const intptr_t kNumTemps = 0; |
6972 | LocationSummary* locs = new (zone) |
6973 | LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kCall); |
6974 | if (type_arguments() != nullptr) { |
6975 | locs->set_in(0, |
6976 | Location::RegisterLocation(kAllocationStubTypeArgumentsReg)); |
6977 | } |
6978 | locs->set_out(0, Location::RegisterLocation(RAX)); |
6979 | return locs; |
6980 | } |
6981 | |
6982 | void AllocateObjectInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6983 | if (type_arguments() != nullptr) { |
6984 | TypeUsageInfo* type_usage_info = compiler->thread()->type_usage_info(); |
6985 | if (type_usage_info != nullptr) { |
6986 | RegisterTypeArgumentsUse(compiler->function(), type_usage_info, cls_, |
6987 | type_arguments()->definition()); |
6988 | } |
6989 | } |
6990 | const Code& stub = Code::ZoneHandle( |
6991 | compiler->zone(), StubCode::GetAllocationStubForClass(cls())); |
6992 | compiler->GenerateStubCall(token_pos(), stub, PcDescriptorsLayout::kOther, |
6993 | locs()); |
6994 | } |
6995 | |
6996 | void DebugStepCheckInstr::EmitNativeCode(FlowGraphCompiler* compiler) { |
6997 | #ifdef PRODUCT |
6998 | UNREACHABLE(); |
6999 | #else |
7000 | ASSERT(!compiler->is_optimizing()); |
7001 | __ CallPatchable(StubCode::DebugStepCheck()); |
7002 | compiler->AddCurrentDescriptor(stub_kind_, deopt_id_, token_pos()); |
7003 | compiler->RecordSafepoint(locs()); |
7004 | #endif |
7005 | } |
7006 | |
7007 | } // namespace dart |
7008 | |
7009 | #undef __ |
7010 | |
7011 | #endif // defined(TARGET_ARCH_X64) |
7012 | |